blob: eb9d63a570cd1ce595076096d9b2aea3723a4c38 [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
Joe Perchesafd465032012-03-12 07:03:32 +000023#define pr_fmt(fmt) "IPv4: " fmt
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/slab.h>
Shan Weie9017b52010-01-23 01:57:42 -080038#include <net/route.h>
39#include <net/dst.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <net/sock.h>
41#include <net/ip.h>
42#include <net/icmp.h>
43#include <net/checksum.h>
Herbert Xu89cee8b2005-12-13 23:14:27 -080044#include <net/inetpeer.h>
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070045#include <net/inet_frag.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/inet.h>
49#include <linux/netfilter_ipv4.h>
Eric Dumazet6623e3b2011-01-05 07:52:55 +000050#include <net/inet_ecn.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
54 * as well. Or notify me, at least. --ANK
55 */
56
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -080057static int sysctl_ipfrag_max_dist __read_mostly = 64;
Herbert Xu89cee8b2005-12-13 23:14:27 -080058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059struct ipfrag_skb_cb
60{
61 struct inet_skb_parm h;
62 int offset;
63};
64
Jianjun Kongfd3f8c42008-11-03 02:47:38 -080065#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67/* Describe an entry in the "incomplete datagrams" queue. */
68struct ipq {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070069 struct inet_frag_queue q;
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 u32 user;
Al Viro18277772006-09-26 22:19:02 -070072 __be32 saddr;
73 __be32 daddr;
74 __be16 id;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 u8 protocol;
Eric Dumazet6623e3b2011-01-05 07:52:55 +000076 u8 ecn; /* RFC3168 support */
Herbert Xu89cee8b2005-12-13 23:14:27 -080077 int iif;
78 unsigned int rid;
79 struct inet_peer *peer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080};
81
Eric Dumazet5173cc02011-05-16 08:37:37 +000082/* RFC 3168 support :
83 * We want to check ECN values of all fragments, do detect invalid combinations.
84 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
85 */
David S. Miller1d1652c2011-05-17 17:28:02 -040086#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
87#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
88#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
89#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
Eric Dumazet6623e3b2011-01-05 07:52:55 +000090
91static inline u8 ip4_frag_ecn(u8 tos)
92{
Eric Dumazet5173cc02011-05-16 08:37:37 +000093 return 1 << (tos & INET_ECN_MASK);
Eric Dumazet6623e3b2011-01-05 07:52:55 +000094}
95
Eric Dumazet5173cc02011-05-16 08:37:37 +000096/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
97 * Value : 0xff if frame should be dropped.
98 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
99 */
100static const u8 ip4_frag_ecn_table[16] = {
101 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
103 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
104 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
105
106 /* invalid combinations : drop frame */
107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
112 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
113 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
114};
115
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700116static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800118int ip_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700119{
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800120 return net->ipv4.frags.nqueues;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700121}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800123int ip_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700124{
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800125 return atomic_read(&net->ipv4.frags.mem);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Herbert Xu1706d582007-10-14 00:38:15 -0700128static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
129 struct net_device *dev);
130
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700131struct ip4_create_arg {
132 struct iphdr *iph;
133 u32 user;
134};
135
Al Viro18277772006-09-26 22:19:02 -0700136static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
Al Viro18277772006-09-26 22:19:02 -0700138 return jhash_3words((__force u32)id << 16 | prot,
139 (__force u32)saddr, (__force u32)daddr,
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700140 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700143static unsigned int ip4_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700145 struct ipq *ipq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700147 ipq = container_of(q, struct ipq, q);
148 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Eric Dumazetcbc264c2012-05-18 05:57:13 +0200151static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700152{
153 struct ipq *qp;
154 struct ip4_create_arg *arg = a;
155
156 qp = container_of(q, struct ipq, q);
Eric Dumazeta02cec22010-09-22 20:43:57 +0000157 return qp->id == arg->iph->id &&
Eric Dumazetcbc264c2012-05-18 05:57:13 +0200158 qp->saddr == arg->iph->saddr &&
159 qp->daddr == arg->iph->daddr &&
160 qp->protocol == arg->iph->protocol &&
161 qp->user == arg->user;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700162}
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164/* Memory Tracking Functions. */
Eric Dumazeta95d8c82010-06-13 23:22:43 +0000165static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800167 atomic_sub(skb->truesize, &nf->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 kfree_skb(skb);
169}
170
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700171static void ip4_frag_init(struct inet_frag_queue *q, void *a)
172{
173 struct ipq *qp = container_of(q, struct ipq, q);
Gao feng54db0cc2012-06-08 01:21:40 +0000174 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
175 frags);
176 struct net *net = container_of(ipv4, struct net, ipv4);
177
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700178 struct ip4_create_arg *arg = a;
179
180 qp->protocol = arg->iph->protocol;
181 qp->id = arg->iph->id;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000182 qp->ecn = ip4_frag_ecn(arg->iph->tos);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700183 qp->saddr = arg->iph->saddr;
184 qp->daddr = arg->iph->daddr;
185 qp->user = arg->user;
186 qp->peer = sysctl_ipfrag_max_dist ?
David S. Millerc0efc882012-06-09 19:12:36 -0700187 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700188}
189
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700190static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700192 struct ipq *qp;
193
194 qp = container_of(q, struct ipq, q);
195 if (qp->peer)
196 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200/* Destruction primitives. */
201
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700202static __inline__ void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700204 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
207/* Kill ipq entry. It is not destroyed immediately,
208 * because caller (and someone more) holds reference count.
209 */
210static void ipq_kill(struct ipq *ipq)
211{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700212 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900215/* Memory limiting on fragments. Evictor trashes the oldest
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 * fragment queue until we are back under the threshold.
217 */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800218static void ip_evictor(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700220 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Amerigo Wang6b102862012-09-18 16:50:11 +0000222 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700223 if (evicted)
Pavel Emelyanovc5346fe2008-07-16 20:20:33 -0700224 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
227/*
228 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
229 */
230static void ip_expire(unsigned long arg)
231{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700232 struct ipq *qp;
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700233 struct net *net;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700234
235 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700236 net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700238 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Joe Perchesbc578a52008-03-28 16:35:27 -0700240 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 goto out;
242
243 ipq_kill(qp);
244
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700245 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
246 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Joe Perchesbc578a52008-03-28 16:35:27 -0700248 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700249 struct sk_buff *head = qp->q.fragments;
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000250 const struct iphdr *iph;
251 int err;
Denis V. Lunevcb846632008-03-24 15:31:00 -0700252
Eric Dumazet69df9d52009-11-05 20:59:47 -0800253 rcu_read_lock();
254 head->dev = dev_get_by_index_rcu(net, qp->iif);
Shan Weie9017b52010-01-23 01:57:42 -0800255 if (!head->dev)
256 goto out_rcu_unlock;
257
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000258 /* skb dst is stale, drop it, and perform route lookup again */
259 skb_dst_drop(head);
260 iph = ip_hdr(head);
David S. Millerc6cffba2012-07-26 11:14:38 +0000261 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
262 iph->tos, head->dev);
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000263 if (err)
264 goto out_rcu_unlock;
265
Shan Weie9017b52010-01-23 01:57:42 -0800266 /*
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000267 * Only an end host needs to send an ICMP
268 * "Fragment Reassembly Timeout" message, per RFC792.
Shan Weie9017b52010-01-23 01:57:42 -0800269 */
David S. Miller595fc712011-07-05 01:05:48 -0700270 if (qp->user == IP_DEFRAG_AF_PACKET ||
271 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
272 skb_rtable(head)->rt_type != RTN_LOCAL))
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000273 goto out_rcu_unlock;
Shan Weie9017b52010-01-23 01:57:42 -0800274
Shan Weie9017b52010-01-23 01:57:42 -0800275
276 /* Send an ICMP "Fragment Reassembly Timeout" message. */
277 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
Shan Weie9017b52010-01-23 01:57:42 -0800278out_rcu_unlock:
Patrick McHardyd1c9ae62010-02-02 11:46:50 -0800279 rcu_read_unlock();
280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700282 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700283 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700286/* Find the correct entry in the "incomplete datagrams" queue for
287 * this IP datagram, and create new one, if nothing is found.
288 */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800289static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700291 struct inet_frag_queue *q;
292 struct ip4_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700293 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700295 arg.iph = iph;
296 arg.user = user;
Pavel Emelyanov9a375802008-06-27 20:06:08 -0700297
298 read_lock(&ip4_frags.lock);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700299 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700300
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800301 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700302 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 goto out_nomem;
304
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700305 return container_of(q, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
307out_nomem:
Joe Perchesafd465032012-03-12 07:03:32 +0000308 LIMIT_NETDEBUG(KERN_ERR pr_fmt("ip_frag_create: no memory left !\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 return NULL;
310}
311
Herbert Xu89cee8b2005-12-13 23:14:27 -0800312/* Is the fragment too far ahead to be part of ipq? */
313static inline int ip_frag_too_far(struct ipq *qp)
314{
315 struct inet_peer *peer = qp->peer;
316 unsigned int max = sysctl_ipfrag_max_dist;
317 unsigned int start, end;
318
319 int rc;
320
321 if (!peer || !max)
322 return 0;
323
324 start = qp->rid;
325 end = atomic_inc_return(&peer->rid);
326 qp->rid = end;
327
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700328 rc = qp->q.fragments && (end - start) > max;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800329
330 if (rc) {
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700331 struct net *net;
332
333 net = container_of(qp->q.net, struct net, ipv4.frags);
334 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800335 }
336
337 return rc;
338}
339
340static int ip_frag_reinit(struct ipq *qp)
341{
342 struct sk_buff *fp;
343
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800344 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700345 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800346 return -ETIMEDOUT;
347 }
348
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700349 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800350 do {
351 struct sk_buff *xp = fp->next;
Eric Dumazeta95d8c82010-06-13 23:22:43 +0000352 frag_kfree_skb(qp->q.net, fp);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800353 fp = xp;
354 } while (fp);
355
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700356 qp->q.last_in = 0;
357 qp->q.len = 0;
358 qp->q.meat = 0;
359 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000360 qp->q.fragments_tail = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800361 qp->iif = 0;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000362 qp->ecn = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800363
364 return 0;
365}
366
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700368static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
370 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700371 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 int flags, offset;
373 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700374 int err = -ENOENT;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000375 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Joe Perchesbc578a52008-03-28 16:35:27 -0700377 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 goto err;
379
Herbert Xu89cee8b2005-12-13 23:14:27 -0800380 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700381 unlikely(ip_frag_too_far(qp)) &&
382 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800383 ipq_kill(qp);
384 goto err;
385 }
386
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000387 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700388 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 flags = offset & ~IP_OFFSET;
390 offset &= IP_OFFSET;
391 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300392 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 /* Determine the position of this fragment. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900395 end = offset + skb->len - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700396 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 /* Is this the final fragment? */
399 if ((flags & IP_MF) == 0) {
400 /* If we already have some bits beyond end
Justin P. Mattock42b2aa82011-11-28 20:31:00 -0800401 * or have different end, the segment is corrupted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700403 if (end < qp->q.len ||
Joe Perchesbc578a52008-03-28 16:35:27 -0700404 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 goto err;
Joe Perchesbc578a52008-03-28 16:35:27 -0700406 qp->q.last_in |= INET_FRAG_LAST_IN;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700407 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 } else {
409 if (end&7) {
410 end &= ~7;
411 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
412 skb->ip_summed = CHECKSUM_NONE;
413 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700414 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 /* Some bits beyond end -> corruption. */
Joe Perchesbc578a52008-03-28 16:35:27 -0700416 if (qp->q.last_in & INET_FRAG_LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700418 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 }
420 }
421 if (end == offset)
422 goto err;
423
Herbert Xu1706d582007-10-14 00:38:15 -0700424 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 if (pskb_pull(skb, ihl) == NULL)
426 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700427
428 err = pskb_trim_rcsum(skb, end - offset);
429 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 goto err;
431
432 /* Find out which fragments are in front and at the back of us
433 * in the chain of fragments so far. We must know where to put
434 * this fragment, right?
435 */
Changli Gaod6bebca2010-06-29 04:39:37 +0000436 prev = qp->q.fragments_tail;
437 if (!prev || FRAG_CB(prev)->offset < offset) {
438 next = NULL;
439 goto found;
440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700442 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (FRAG_CB(next)->offset >= offset)
444 break; /* bingo! */
445 prev = next;
446 }
447
Changli Gaod6bebca2010-06-29 04:39:37 +0000448found:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 /* We found where to put this one. Check for overlap with
450 * preceding fragment, and, if needed, align things so that
451 * any overlaps are eliminated.
452 */
453 if (prev) {
454 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
455
456 if (i > 0) {
457 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700458 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (end <= offset)
460 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700461 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 if (!pskb_pull(skb, i))
463 goto err;
464 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
465 skb->ip_summed = CHECKSUM_NONE;
466 }
467 }
468
Herbert Xu1706d582007-10-14 00:38:15 -0700469 err = -ENOMEM;
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 while (next && FRAG_CB(next)->offset < end) {
472 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
473
474 if (i < next->len) {
475 /* Eat head of the next overlapped fragment
476 * and leave the loop. The next ones cannot overlap.
477 */
478 if (!pskb_pull(next, i))
479 goto err;
480 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700481 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 if (next->ip_summed != CHECKSUM_UNNECESSARY)
483 next->ip_summed = CHECKSUM_NONE;
484 break;
485 } else {
486 struct sk_buff *free_it = next;
487
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100488 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 * new one drop it.
490 */
491 next = next->next;
492
493 if (prev)
494 prev->next = next;
495 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700496 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700498 qp->q.meat -= free_it->len;
Eric Dumazeta95d8c82010-06-13 23:22:43 +0000499 frag_kfree_skb(qp->q.net, free_it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
501 }
502
503 FRAG_CB(skb)->offset = offset;
504
505 /* Insert this fragment in the chain of fragments. */
506 skb->next = next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000507 if (!next)
508 qp->q.fragments_tail = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 if (prev)
510 prev->next = skb;
511 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700512 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Herbert Xu1706d582007-10-14 00:38:15 -0700514 dev = skb->dev;
515 if (dev) {
516 qp->iif = dev->ifindex;
517 skb->dev = NULL;
518 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700519 qp->q.stamp = skb->tstamp;
520 qp->q.meat += skb->len;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000521 qp->ecn |= ecn;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800522 atomic_add(skb->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 if (offset == 0)
Joe Perchesbc578a52008-03-28 16:35:27 -0700524 qp->q.last_in |= INET_FRAG_FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200526 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
527 skb->len + ihl > qp->q.max_size)
528 qp->q.max_size = skb->len + ihl;
529
Joe Perchesbc578a52008-03-28 16:35:27 -0700530 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
531 qp->q.meat == qp->q.len)
Herbert Xu1706d582007-10-14 00:38:15 -0700532 return ip_frag_reasm(qp, prev, dev);
533
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700534 write_lock(&ip4_frags.lock);
Pavel Emelyanov3140c252008-01-22 06:11:48 -0800535 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700536 write_unlock(&ip4_frags.lock);
Herbert Xu1706d582007-10-14 00:38:15 -0700537 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538
539err:
540 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700541 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542}
543
544
545/* Build a new IP datagram from all its fragments. */
546
Herbert Xu1706d582007-10-14 00:38:15 -0700547static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
548 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700550 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700552 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 int len;
554 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700555 int err;
Eric Dumazet3cc49492012-05-19 03:02:20 +0000556 int sum_truesize;
Eric Dumazet5173cc02011-05-16 08:37:37 +0000557 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559 ipq_kill(qp);
560
Eric Dumazet5173cc02011-05-16 08:37:37 +0000561 ecn = ip4_frag_ecn_table[qp->ecn];
562 if (unlikely(ecn == 0xff)) {
563 err = -EINVAL;
564 goto out_fail;
565 }
Herbert Xu1706d582007-10-14 00:38:15 -0700566 /* Make the one we just received the head. */
567 if (prev) {
568 head = prev->next;
569 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700570 if (!fp)
571 goto out_nomem;
572
573 fp->next = head->next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000574 if (!fp->next)
575 qp->q.fragments_tail = fp;
Herbert Xu1706d582007-10-14 00:38:15 -0700576 prev->next = fp;
577
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700578 skb_morph(head, qp->q.fragments);
579 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700580
Eric Dumazetcbf8f7b2012-04-19 06:10:26 +0000581 consume_skb(qp->q.fragments);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700582 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700583 }
584
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700585 WARN_ON(head == NULL);
586 WARN_ON(FRAG_CB(head)->offset != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300589 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700590 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Herbert Xu1706d582007-10-14 00:38:15 -0700592 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800593 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 goto out_oversize;
595
596 /* Head of list must not be cloned. */
597 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
598 goto out_nomem;
599
600 /* If the first fragment is fragmented itself, we split
601 * it to two chunks: the first with data and paged part
602 * and the second, holding only fragments. */
David S. Miller21dc3302010-08-23 00:13:46 -0700603 if (skb_has_frag_list(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 struct sk_buff *clone;
605 int i, plen = 0;
606
607 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
608 goto out_nomem;
609 clone->next = head->next;
610 head->next = clone;
611 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700612 skb_frag_list_init(head);
Eric Dumazet9e903e02011-10-18 21:00:24 +0000613 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
614 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 clone->len = clone->data_len = head->data_len - plen;
616 head->data_len -= clone->len;
617 head->len -= clone->len;
618 clone->csum = 0;
619 clone->ip_summed = head->ip_summed;
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800620 atomic_add(clone->truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 }
622
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700623 skb_push(head, head->data - skb_network_header(head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Eric Dumazet3cc49492012-05-19 03:02:20 +0000625 sum_truesize = head->truesize;
626 for (fp = head->next; fp;) {
627 bool headstolen;
628 int delta;
629 struct sk_buff *next = fp->next;
630
631 sum_truesize += fp->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 if (head->ip_summed != fp->ip_summed)
633 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700634 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 head->csum = csum_add(head->csum, fp->csum);
Eric Dumazet3cc49492012-05-19 03:02:20 +0000636
637 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
638 kfree_skb_partial(fp, headstolen);
639 } else {
640 if (!skb_shinfo(head)->frag_list)
641 skb_shinfo(head)->frag_list = fp;
642 head->data_len += fp->len;
643 head->len += fp->len;
644 head->truesize += fp->truesize;
645 }
646 fp = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 }
Eric Dumazet3cc49492012-05-19 03:02:20 +0000648 atomic_sub(sum_truesize, &qp->q.net->mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
650 head->next = NULL;
651 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700652 head->tstamp = qp->q.stamp;
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200653 IPCB(head)->frag_max_size = qp->q.max_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700655 iph = ip_hdr(head);
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200656 /* max_size != 0 implies at least one fragment had IP_DF set */
657 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 iph->tot_len = htons(len);
Eric Dumazet5173cc02011-05-16 08:37:37 +0000659 iph->tos |= ecn;
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700660 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700661 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000662 qp->q.fragments_tail = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700663 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665out_nomem:
Joe Perchesafd465032012-03-12 07:03:32 +0000666 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
667 qp);
David Howells45542472007-10-17 21:37:22 -0700668 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 goto out_fail;
670out_oversize:
Joe Perchese87cc472012-05-13 21:56:26 +0000671 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672out_fail:
David Fordbbf31bf2009-11-29 23:02:22 -0800673 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700674 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
677/* Process an incoming IP datagram fragment. */
Herbert Xu776c7292007-10-14 00:38:32 -0700678int ip_defrag(struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 struct ipq *qp;
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800681 struct net *net;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900682
Eric Dumazetadf30902009-06-02 05:19:30 +0000683 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700684 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
686 /* Start by cleaning up the memory. */
Amerigo Wang6b102862012-09-18 16:50:11 +0000687 ip_evictor(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 /* Lookup (or create) queue header */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800690 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
Herbert Xu1706d582007-10-14 00:38:15 -0700691 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700693 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Herbert Xu1706d582007-10-14 00:38:15 -0700695 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700697 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700698 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700699 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700702 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700704 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000706EXPORT_SYMBOL(ip_defrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Eric Dumazetbc416d92011-10-06 10:28:31 +0000708struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
709{
Johannes Berg1bf37512012-12-09 23:41:06 +0000710 struct iphdr iph;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000711 u32 len;
712
713 if (skb->protocol != htons(ETH_P_IP))
714 return skb;
715
Johannes Berg1bf37512012-12-09 23:41:06 +0000716 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
Eric Dumazetbc416d92011-10-06 10:28:31 +0000717 return skb;
718
Johannes Berg1bf37512012-12-09 23:41:06 +0000719 if (iph.ihl < 5 || iph.version != 4)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000720 return skb;
721
Johannes Berg1bf37512012-12-09 23:41:06 +0000722 len = ntohs(iph.tot_len);
723 if (skb->len < len || len < (iph.ihl * 4))
724 return skb;
725
726 if (ip_is_fragment(&iph)) {
Eric Dumazetbc416d92011-10-06 10:28:31 +0000727 skb = skb_share_check(skb, GFP_ATOMIC);
728 if (skb) {
Johannes Berg1bf37512012-12-09 23:41:06 +0000729 if (!pskb_may_pull(skb, iph.ihl*4))
730 return skb;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000731 if (pskb_trim_rcsum(skb, len))
732 return skb;
733 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
734 if (ip_defrag(skb, user))
735 return NULL;
736 skb->rxhash = 0;
737 }
738 }
739 return skb;
740}
741EXPORT_SYMBOL(ip_check_defrag);
742
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800743#ifdef CONFIG_SYSCTL
744static int zero;
745
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700746static struct ctl_table ip4_frags_ns_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800747 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800748 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800749 .data = &init_net.ipv4.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800750 .maxlen = sizeof(int),
751 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800752 .proc_handler = proc_dointvec
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800753 },
754 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800755 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800756 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800757 .maxlen = sizeof(int),
758 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800759 .proc_handler = proc_dointvec
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800760 },
761 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800762 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800763 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800764 .maxlen = sizeof(int),
765 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800766 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800767 },
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700768 { }
769};
770
771static struct ctl_table ip4_frags_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800772 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800773 .procname = "ipfrag_secret_interval",
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800774 .data = &ip4_frags.secret_interval,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800775 .maxlen = sizeof(int),
776 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800777 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800778 },
779 {
780 .procname = "ipfrag_max_dist",
781 .data = &sysctl_ipfrag_max_dist,
782 .maxlen = sizeof(int),
783 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800784 .proc_handler = proc_dointvec_minmax,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800785 .extra1 = &zero
786 },
787 { }
788};
789
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000790static int __net_init ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800791{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800792 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800793 struct ctl_table_header *hdr;
794
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700795 table = ip4_frags_ns_ctl_table;
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800796 if (!net_eq(net, &init_net)) {
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700797 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800798 if (table == NULL)
799 goto err_alloc;
800
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800801 table[0].data = &net->ipv4.frags.high_thresh;
802 table[1].data = &net->ipv4.frags.low_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800803 table[2].data = &net->ipv4.frags.timeout;
Eric W. Biederman464dc802012-11-16 03:02:59 +0000804
805 /* Don't export sysctls to unprivileged users */
806 if (net->user_ns != &init_user_ns)
807 table[0].procname = NULL;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800808 }
809
Eric W. Biedermanec8f23c2012-04-19 13:44:49 +0000810 hdr = register_net_sysctl(net, "net/ipv4", table);
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800811 if (hdr == NULL)
812 goto err_reg;
813
814 net->ipv4.frags_hdr = hdr;
815 return 0;
816
817err_reg:
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800818 if (!net_eq(net, &init_net))
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800819 kfree(table);
820err_alloc:
821 return -ENOMEM;
822}
823
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000824static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800825{
826 struct ctl_table *table;
827
828 table = net->ipv4.frags_hdr->ctl_table_arg;
829 unregister_net_sysctl_table(net->ipv4.frags_hdr);
830 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800831}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700832
833static void ip4_frags_ctl_register(void)
834{
Eric W. Biederman43444752012-04-19 13:22:55 +0000835 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700836}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800837#else
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700838static inline int ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800839{
840 return 0;
841}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800842
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700843static inline void ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800844{
845}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700846
847static inline void ip4_frags_ctl_register(void)
848{
849}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800850#endif
851
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000852static int __net_init ipv4_frags_init_net(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800853{
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800854 /*
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800855 * Fragment cache limits. We will commit 256K at one time. Should we
856 * cross that limit we will prune down to 192K. This should cope with
857 * even the most extreme cases without allowing an attacker to
858 * measurably harm machine performance.
859 */
860 net->ipv4.frags.high_thresh = 256 * 1024;
861 net->ipv4.frags.low_thresh = 192 * 1024;
862 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800863 * Important NOTE! Fragment queue must be destroyed before MSL expires.
864 * RFC791 is wrong proposing to prolongate timer each fragment arrival
865 * by TTL.
866 */
867 net->ipv4.frags.timeout = IP_FRAG_TIME;
868
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800869 inet_frags_init_net(&net->ipv4.frags);
870
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700871 return ip4_frags_ns_ctl_register(net);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800872}
873
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000874static void __net_exit ipv4_frags_exit_net(struct net *net)
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800875{
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700876 ip4_frags_ns_ctl_unregister(net);
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800877 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
878}
879
880static struct pernet_operations ip4_frags_ops = {
881 .init = ipv4_frags_init_net,
882 .exit = ipv4_frags_exit_net,
883};
884
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700885void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700887 ip4_frags_ctl_register();
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800888 register_pernet_subsys(&ip4_frags_ops);
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700889 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700890 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700891 ip4_frags.destructor = ip4_frag_free;
892 ip4_frags.skb_free = NULL;
893 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700894 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700895 ip4_frags.frag_expire = ip_expire;
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800896 ip4_frags.secret_interval = 10 * 60 * HZ;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700897 inet_frags_init(&ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}