blob: bbe7f72db9c157ba2d6c5292637c2f58ad39a123 [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>
David Ahern385add92015-09-29 20:07:13 -070051#include <net/l3mdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
54 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
55 * as well. Or notify me, at least. --ANK
56 */
Nikolay Aleksandrovd4ad4d22014-08-01 12:29:48 +020057static const char ip_frag_cache_name[] = "ip4-frags";
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 */
Florian Westphald6b915e2015-05-22 16:32:51 +020077 u16 max_df_size; /* largest frag with DF set seen */
Herbert Xu89cee8b2005-12-13 23:14:27 -080078 int iif;
David Ahern385add92015-09-29 20:07:13 -070079 int vif; /* L3 master device index */
Herbert Xu89cee8b2005-12-13 23:14:27 -080080 unsigned int rid;
81 struct inet_peer *peer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082};
83
Fabian Frederickaa1f7312014-11-04 20:44:04 +010084static u8 ip4_frag_ecn(u8 tos)
Eric Dumazet6623e3b2011-01-05 07:52:55 +000085{
Eric Dumazet5173cc02011-05-16 08:37:37 +000086 return 1 << (tos & INET_ECN_MASK);
Eric Dumazet6623e3b2011-01-05 07:52:55 +000087}
88
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070089static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080091int ip_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070092{
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +000093 return sum_frag_mem_limit(&net->ipv4.frags);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070094}
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Herbert Xu1706d582007-10-14 00:38:15 -070096static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
97 struct net_device *dev);
98
Pavel Emelyanovc6fda282007-10-17 19:46:47 -070099struct ip4_create_arg {
100 struct iphdr *iph;
101 u32 user;
David Ahern9972f132015-08-13 14:59:09 -0600102 int vif;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700103};
104
Al Viro18277772006-09-26 22:19:02 -0700105static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
Hannes Frederic Sowae7b519b2013-10-23 11:06:55 +0200107 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
Al Viro18277772006-09-26 22:19:02 -0700108 return jhash_3words((__force u32)id << 16 | prot,
109 (__force u32)saddr, (__force u32)daddr,
Florian Westphalfb3cfe62014-07-24 16:50:30 +0200110 ip4_frags.rnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111}
112
Florian Westphal36c77782014-07-24 16:50:29 +0200113static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
Florian Westphal36c77782014-07-24 16:50:29 +0200115 const struct ipq *ipq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700117 ipq = container_of(q, struct ipq, q);
118 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119}
120
Florian Westphal36c77782014-07-24 16:50:29 +0200121static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700122{
Florian Westphal36c77782014-07-24 16:50:29 +0200123 const struct ipq *qp;
124 const struct ip4_create_arg *arg = a;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700125
126 qp = container_of(q, struct ipq, q);
Eric Dumazeta02cec22010-09-22 20:43:57 +0000127 return qp->id == arg->iph->id &&
Eric Dumazetcbc264c2012-05-18 05:57:13 +0200128 qp->saddr == arg->iph->saddr &&
129 qp->daddr == arg->iph->daddr &&
130 qp->protocol == arg->iph->protocol &&
David Ahern9972f132015-08-13 14:59:09 -0600131 qp->user == arg->user &&
132 qp->vif == arg->vif;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700133}
134
Florian Westphal36c77782014-07-24 16:50:29 +0200135static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700136{
137 struct ipq *qp = container_of(q, struct ipq, q);
Gao feng54db0cc2012-06-08 01:21:40 +0000138 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
139 frags);
140 struct net *net = container_of(ipv4, struct net, ipv4);
141
Florian Westphal36c77782014-07-24 16:50:29 +0200142 const struct ip4_create_arg *arg = a;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700143
144 qp->protocol = arg->iph->protocol;
145 qp->id = arg->iph->id;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000146 qp->ecn = ip4_frag_ecn(arg->iph->tos);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700147 qp->saddr = arg->iph->saddr;
148 qp->daddr = arg->iph->daddr;
David Ahern9972f132015-08-13 14:59:09 -0600149 qp->vif = arg->vif;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700150 qp->user = arg->user;
Nikolay Borisov0fbf4cb2016-02-15 12:11:31 +0200151 qp->peer = q->net->max_dist ?
David Ahern192132b2015-08-27 16:07:03 -0700152 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
153 NULL;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700154}
155
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100156static void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700158 struct ipq *qp;
159
160 qp = container_of(q, struct ipq, q);
161 if (qp->peer)
162 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
166/* Destruction primitives. */
167
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100168static void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700170 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
172
173/* Kill ipq entry. It is not destroyed immediately,
174 * because caller (and someone more) holds reference count.
175 */
176static void ipq_kill(struct ipq *ipq)
177{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700178 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
180
Andy Zhou5cf42282015-05-15 14:15:35 -0700181static bool frag_expire_skip_icmp(u32 user)
182{
183 return user == IP_DEFRAG_AF_PACKET ||
184 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
Andy Zhou8bc04862015-05-15 14:15:36 -0700185 __IP_DEFRAG_CONNTRACK_IN_END) ||
186 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
187 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
Andy Zhou5cf42282015-05-15 14:15:35 -0700188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/*
191 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
192 */
193static void ip_expire(unsigned long arg)
194{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700195 struct ipq *qp;
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700196 struct net *net;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700197
198 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700199 net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700201 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200203 if (qp->q.flags & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 goto out;
205
206 ipq_kill(qp);
Eric Dumazetb45386e2016-04-27 16:44:35 -0700207 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Nikolay Aleksandrovcaaecdd2015-07-23 12:05:40 +0200209 if (!inet_frag_evicting(&qp->q)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700210 struct sk_buff *head = qp->q.fragments;
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000211 const struct iphdr *iph;
212 int err;
Denis V. Lunevcb846632008-03-24 15:31:00 -0700213
Eric Dumazetb45386e2016-04-27 16:44:35 -0700214 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
Nikolay Aleksandrov2e404f62014-08-01 12:29:47 +0200215
216 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
217 goto out;
218
Eric Dumazet69df9d52009-11-05 20:59:47 -0800219 rcu_read_lock();
220 head->dev = dev_get_by_index_rcu(net, qp->iif);
Shan Weie9017b52010-01-23 01:57:42 -0800221 if (!head->dev)
222 goto out_rcu_unlock;
223
Eric Dumazet97599dc2013-04-16 12:55:41 +0000224 /* skb has no dst, perform route lookup again */
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000225 iph = ip_hdr(head);
David S. Millerc6cffba2012-07-26 11:14:38 +0000226 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
227 iph->tos, head->dev);
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000228 if (err)
229 goto out_rcu_unlock;
230
Nikolay Aleksandrov2e404f62014-08-01 12:29:47 +0200231 /* Only an end host needs to send an ICMP
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000232 * "Fragment Reassembly Timeout" message, per RFC792.
Shan Weie9017b52010-01-23 01:57:42 -0800233 */
Andy Zhou5cf42282015-05-15 14:15:35 -0700234 if (frag_expire_skip_icmp(qp->user) &&
235 (skb_rtable(head)->rt_type != RTN_LOCAL))
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000236 goto out_rcu_unlock;
Shan Weie9017b52010-01-23 01:57:42 -0800237
Shan Weie9017b52010-01-23 01:57:42 -0800238 /* Send an ICMP "Fragment Reassembly Timeout" message. */
239 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
Shan Weie9017b52010-01-23 01:57:42 -0800240out_rcu_unlock:
Patrick McHardyd1c9ae62010-02-02 11:46:50 -0800241 rcu_read_unlock();
242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700244 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700245 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700248/* Find the correct entry in the "incomplete datagrams" queue for
249 * this IP datagram, and create new one, if nothing is found.
250 */
David Ahern9972f132015-08-13 14:59:09 -0600251static struct ipq *ip_find(struct net *net, struct iphdr *iph,
252 u32 user, int vif)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700254 struct inet_frag_queue *q;
255 struct ip4_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700256 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700258 arg.iph = iph;
259 arg.user = user;
David Ahern9972f132015-08-13 14:59:09 -0600260 arg.vif = vif;
Pavel Emelyanov9a375802008-06-27 20:06:08 -0700261
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700262 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700263
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800264 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
Hannes Frederic Sowa5a3da1f2013-03-15 11:32:30 +0000265 if (IS_ERR_OR_NULL(q)) {
266 inet_frag_maybe_warn_overflow(q, pr_fmt());
267 return NULL;
268 }
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700269 return container_of(q, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270}
271
Herbert Xu89cee8b2005-12-13 23:14:27 -0800272/* Is the fragment too far ahead to be part of ipq? */
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100273static int ip_frag_too_far(struct ipq *qp)
Herbert Xu89cee8b2005-12-13 23:14:27 -0800274{
275 struct inet_peer *peer = qp->peer;
Nikolay Borisov0fbf4cb2016-02-15 12:11:31 +0200276 unsigned int max = qp->q.net->max_dist;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800277 unsigned int start, end;
278
279 int rc;
280
281 if (!peer || !max)
282 return 0;
283
284 start = qp->rid;
285 end = atomic_inc_return(&peer->rid);
286 qp->rid = end;
287
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700288 rc = qp->q.fragments && (end - start) > max;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800289
290 if (rc) {
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700291 struct net *net;
292
293 net = container_of(qp->q.net, struct net, ipv4.frags);
Eric Dumazetb45386e2016-04-27 16:44:35 -0700294 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800295 }
296
297 return rc;
298}
299
300static int ip_frag_reinit(struct ipq *qp)
301{
302 struct sk_buff *fp;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000303 unsigned int sum_truesize = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800304
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800305 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700306 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800307 return -ETIMEDOUT;
308 }
309
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700310 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800311 do {
312 struct sk_buff *xp = fp->next;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000313
314 sum_truesize += fp->truesize;
315 kfree_skb(fp);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800316 fp = xp;
317 } while (fp);
Florian Westphal0e60d242015-07-23 12:05:38 +0200318 sub_frag_mem_limit(qp->q.net, sum_truesize);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800319
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200320 qp->q.flags = 0;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700321 qp->q.len = 0;
322 qp->q.meat = 0;
323 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000324 qp->q.fragments_tail = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800325 qp->iif = 0;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000326 qp->ecn = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800327
328 return 0;
329}
330
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700332static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
334 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700335 struct net_device *dev;
Florian Westphald6b915e2015-05-22 16:32:51 +0200336 unsigned int fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 int flags, offset;
338 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700339 int err = -ENOENT;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000340 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200342 if (qp->q.flags & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 goto err;
344
Herbert Xu89cee8b2005-12-13 23:14:27 -0800345 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700346 unlikely(ip_frag_too_far(qp)) &&
347 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800348 ipq_kill(qp);
349 goto err;
350 }
351
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000352 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700353 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 flags = offset & ~IP_OFFSET;
355 offset &= IP_OFFSET;
356 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300357 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
359 /* Determine the position of this fragment. */
Edward Hyunkoo Jee0848f642015-07-21 09:43:59 +0200360 end = offset + skb->len - skb_network_offset(skb) - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700361 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 /* Is this the final fragment? */
364 if ((flags & IP_MF) == 0) {
365 /* If we already have some bits beyond end
Justin P. Mattock42b2aa82011-11-28 20:31:00 -0800366 * or have different end, the segment is corrupted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700368 if (end < qp->q.len ||
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200369 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 goto err;
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200371 qp->q.flags |= INET_FRAG_LAST_IN;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700372 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 } else {
374 if (end&7) {
375 end &= ~7;
376 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
377 skb->ip_summed = CHECKSUM_NONE;
378 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700379 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 /* Some bits beyond end -> corruption. */
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200381 if (qp->q.flags & INET_FRAG_LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700383 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 }
385 }
386 if (end == offset)
387 goto err;
388
Herbert Xu1706d582007-10-14 00:38:15 -0700389 err = -ENOMEM;
Edward Hyunkoo Jee0848f642015-07-21 09:43:59 +0200390 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700392
393 err = pskb_trim_rcsum(skb, end - offset);
394 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 goto err;
396
397 /* Find out which fragments are in front and at the back of us
398 * in the chain of fragments so far. We must know where to put
399 * this fragment, right?
400 */
Changli Gaod6bebca2010-06-29 04:39:37 +0000401 prev = qp->q.fragments_tail;
402 if (!prev || FRAG_CB(prev)->offset < offset) {
403 next = NULL;
404 goto found;
405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700407 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (FRAG_CB(next)->offset >= offset)
409 break; /* bingo! */
410 prev = next;
411 }
412
Changli Gaod6bebca2010-06-29 04:39:37 +0000413found:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 /* We found where to put this one. Check for overlap with
415 * preceding fragment, and, if needed, align things so that
416 * any overlaps are eliminated.
417 */
418 if (prev) {
419 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
420
421 if (i > 0) {
422 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700423 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 if (end <= offset)
425 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700426 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (!pskb_pull(skb, i))
428 goto err;
429 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
430 skb->ip_summed = CHECKSUM_NONE;
431 }
432 }
433
Herbert Xu1706d582007-10-14 00:38:15 -0700434 err = -ENOMEM;
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 while (next && FRAG_CB(next)->offset < end) {
437 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
438
439 if (i < next->len) {
440 /* Eat head of the next overlapped fragment
441 * and leave the loop. The next ones cannot overlap.
442 */
443 if (!pskb_pull(next, i))
444 goto err;
445 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700446 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 if (next->ip_summed != CHECKSUM_UNNECESSARY)
448 next->ip_summed = CHECKSUM_NONE;
449 break;
450 } else {
451 struct sk_buff *free_it = next;
452
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100453 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 * new one drop it.
455 */
456 next = next->next;
457
458 if (prev)
459 prev->next = next;
460 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700461 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700463 qp->q.meat -= free_it->len;
Florian Westphal0e60d242015-07-23 12:05:38 +0200464 sub_frag_mem_limit(qp->q.net, free_it->truesize);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000465 kfree_skb(free_it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 }
467 }
468
469 FRAG_CB(skb)->offset = offset;
470
471 /* Insert this fragment in the chain of fragments. */
472 skb->next = next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000473 if (!next)
474 qp->q.fragments_tail = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 if (prev)
476 prev->next = skb;
477 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700478 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Herbert Xu1706d582007-10-14 00:38:15 -0700480 dev = skb->dev;
481 if (dev) {
482 qp->iif = dev->ifindex;
483 skb->dev = NULL;
484 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700485 qp->q.stamp = skb->tstamp;
486 qp->q.meat += skb->len;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000487 qp->ecn |= ecn;
Florian Westphal0e60d242015-07-23 12:05:38 +0200488 add_frag_mem_limit(qp->q.net, skb->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (offset == 0)
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200490 qp->q.flags |= INET_FRAG_FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Florian Westphald6b915e2015-05-22 16:32:51 +0200492 fragsize = skb->len + ihl;
493
494 if (fragsize > qp->q.max_size)
495 qp->q.max_size = fragsize;
496
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200497 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
Florian Westphald6b915e2015-05-22 16:32:51 +0200498 fragsize > qp->max_df_size)
499 qp->max_df_size = fragsize;
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200500
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200501 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
Eric Dumazet97599dc2013-04-16 12:55:41 +0000502 qp->q.meat == qp->q.len) {
503 unsigned long orefdst = skb->_skb_refdst;
Herbert Xu1706d582007-10-14 00:38:15 -0700504
Eric Dumazet97599dc2013-04-16 12:55:41 +0000505 skb->_skb_refdst = 0UL;
506 err = ip_frag_reasm(qp, prev, dev);
507 skb->_skb_refdst = orefdst;
508 return err;
509 }
510
511 skb_dst_drop(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700512 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514err:
515 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700516 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517}
518
519
520/* Build a new IP datagram from all its fragments. */
521
Herbert Xu1706d582007-10-14 00:38:15 -0700522static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
523 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700525 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700527 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 int len;
529 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700530 int err;
Eric Dumazet5173cc02011-05-16 08:37:37 +0000531 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 ipq_kill(qp);
534
Hannes Frederic Sowabe991972013-03-22 08:24:37 +0000535 ecn = ip_frag_ecn_table[qp->ecn];
Eric Dumazet5173cc02011-05-16 08:37:37 +0000536 if (unlikely(ecn == 0xff)) {
537 err = -EINVAL;
538 goto out_fail;
539 }
Herbert Xu1706d582007-10-14 00:38:15 -0700540 /* Make the one we just received the head. */
541 if (prev) {
542 head = prev->next;
543 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700544 if (!fp)
545 goto out_nomem;
546
547 fp->next = head->next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000548 if (!fp->next)
549 qp->q.fragments_tail = fp;
Herbert Xu1706d582007-10-14 00:38:15 -0700550 prev->next = fp;
551
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700552 skb_morph(head, qp->q.fragments);
553 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700554
Eric Dumazetcbf8f7b2012-04-19 06:10:26 +0000555 consume_skb(qp->q.fragments);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700556 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700557 }
558
Ian Morris51456b22015-04-03 09:17:26 +0100559 WARN_ON(!head);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700560 WARN_ON(FRAG_CB(head)->offset != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300563 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700564 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Herbert Xu1706d582007-10-14 00:38:15 -0700566 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800567 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 goto out_oversize;
569
570 /* Head of list must not be cloned. */
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +0000571 if (skb_unclone(head, GFP_ATOMIC))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 goto out_nomem;
573
574 /* If the first fragment is fragmented itself, we split
575 * it to two chunks: the first with data and paged part
576 * and the second, holding only fragments. */
David S. Miller21dc3302010-08-23 00:13:46 -0700577 if (skb_has_frag_list(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 struct sk_buff *clone;
579 int i, plen = 0;
580
Ian Morris51456b22015-04-03 09:17:26 +0100581 clone = alloc_skb(0, GFP_ATOMIC);
582 if (!clone)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 goto out_nomem;
584 clone->next = head->next;
585 head->next = clone;
586 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700587 skb_frag_list_init(head);
Eric Dumazet9e903e02011-10-18 21:00:24 +0000588 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
589 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 clone->len = clone->data_len = head->data_len - plen;
591 head->data_len -= clone->len;
592 head->len -= clone->len;
593 clone->csum = 0;
594 clone->ip_summed = head->ip_summed;
Florian Westphal0e60d242015-07-23 12:05:38 +0200595 add_frag_mem_limit(qp->q.net, clone->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 }
597
Florian Westphal14fe22e2015-07-11 01:37:36 +0200598 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700599 skb_push(head, head->data - skb_network_header(head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Florian Westphal14fe22e2015-07-11 01:37:36 +0200601 for (fp=head->next; fp; fp = fp->next) {
602 head->data_len += fp->len;
603 head->len += fp->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 if (head->ip_summed != fp->ip_summed)
605 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700606 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 head->csum = csum_add(head->csum, fp->csum);
Florian Westphal14fe22e2015-07-11 01:37:36 +0200608 head->truesize += fp->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
David S. Miller5510b3c2015-07-31 23:52:20 -0700610 sub_frag_mem_limit(qp->q.net, head->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
612 head->next = NULL;
613 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700614 head->tstamp = qp->q.stamp;
Florian Westphald6b915e2015-05-22 16:32:51 +0200615 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700617 iph = ip_hdr(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 iph->tot_len = htons(len);
Eric Dumazet5173cc02011-05-16 08:37:37 +0000619 iph->tos |= ecn;
Florian Westphald6b915e2015-05-22 16:32:51 +0200620
621 /* When we set IP_DF on a refragmented skb we must also force a
622 * call to ip_fragment to avoid forwarding a DF-skb of size s while
623 * original sender only sent fragments of size f (where f < s).
624 *
625 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
626 * frag seen to avoid sending tiny DF-fragments in case skb was built
627 * from one very small df-fragment and one large non-df frag.
628 */
629 if (qp->max_df_size == qp->q.max_size) {
630 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
631 iph->frag_off = htons(IP_DF);
632 } else {
633 iph->frag_off = 0;
634 }
635
Edward Hyunkoo Jee0848f642015-07-21 09:43:59 +0200636 ip_send_check(iph);
637
Eric Dumazetb45386e2016-04-27 16:44:35 -0700638 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700639 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000640 qp->q.fragments_tail = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700641 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643out_nomem:
Joe Perchesba7a46f2014-11-11 10:59:17 -0800644 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
David Howells45542472007-10-17 21:37:22 -0700645 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 goto out_fail;
647out_oversize:
Joe Perchese87cc472012-05-13 21:56:26 +0000648 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649out_fail:
Eric Dumazetb45386e2016-04-27 16:44:35 -0700650 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700651 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
653
654/* Process an incoming IP datagram fragment. */
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500655int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
David Ahern9972f132015-08-13 14:59:09 -0600657 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
David Ahern385add92015-09-29 20:07:13 -0700658 int vif = l3mdev_master_ifindex_rcu(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 struct ipq *qp;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900660
Eric Dumazetb45386e2016-04-27 16:44:35 -0700661 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
Joe Stringer8282f272016-01-22 15:49:12 -0800662 skb_orphan(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 /* Lookup (or create) queue header */
David Ahern9972f132015-08-13 14:59:09 -0600665 qp = ip_find(net, ip_hdr(skb), user, vif);
Ian Morris00db4122015-04-03 09:17:27 +0100666 if (qp) {
Herbert Xu1706d582007-10-14 00:38:15 -0700667 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700669 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Herbert Xu1706d582007-10-14 00:38:15 -0700671 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700673 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700674 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700675 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
677
Eric Dumazetb45386e2016-04-27 16:44:35 -0700678 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700680 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000682EXPORT_SYMBOL(ip_defrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500684struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000685{
Johannes Berg1bf37512012-12-09 23:41:06 +0000686 struct iphdr iph;
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300687 int netoff;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000688 u32 len;
689
690 if (skb->protocol != htons(ETH_P_IP))
691 return skb;
692
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300693 netoff = skb_network_offset(skb);
694
695 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000696 return skb;
697
Johannes Berg1bf37512012-12-09 23:41:06 +0000698 if (iph.ihl < 5 || iph.version != 4)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000699 return skb;
700
Johannes Berg1bf37512012-12-09 23:41:06 +0000701 len = ntohs(iph.tot_len);
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300702 if (skb->len < netoff + len || len < (iph.ihl * 4))
Johannes Berg1bf37512012-12-09 23:41:06 +0000703 return skb;
704
705 if (ip_is_fragment(&iph)) {
Eric Dumazetbc416d92011-10-06 10:28:31 +0000706 skb = skb_share_check(skb, GFP_ATOMIC);
707 if (skb) {
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300708 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
Johannes Berg1bf37512012-12-09 23:41:06 +0000709 return skb;
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300710 if (pskb_trim_rcsum(skb, netoff + len))
Eric Dumazetbc416d92011-10-06 10:28:31 +0000711 return skb;
712 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500713 if (ip_defrag(net, skb, user))
Eric Dumazetbc416d92011-10-06 10:28:31 +0000714 return NULL;
Tom Herbert7539fad2013-12-15 22:12:18 -0800715 skb_clear_hash(skb);
Eric Dumazetbc416d92011-10-06 10:28:31 +0000716 }
717 }
718 return skb;
719}
720EXPORT_SYMBOL(ip_check_defrag);
721
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800722#ifdef CONFIG_SYSCTL
723static int zero;
724
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700725static struct ctl_table ip4_frags_ns_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800726 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800727 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800728 .data = &init_net.ipv4.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800729 .maxlen = sizeof(int),
730 .mode = 0644,
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200731 .proc_handler = proc_dointvec_minmax,
732 .extra1 = &init_net.ipv4.frags.low_thresh
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800733 },
734 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800735 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800736 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800737 .maxlen = sizeof(int),
738 .mode = 0644,
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200739 .proc_handler = proc_dointvec_minmax,
740 .extra1 = &zero,
741 .extra2 = &init_net.ipv4.frags.high_thresh
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800742 },
743 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800744 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800745 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800746 .maxlen = sizeof(int),
747 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800748 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800749 },
Nikolay Borisov0fbf4cb2016-02-15 12:11:31 +0200750 {
751 .procname = "ipfrag_max_dist",
752 .data = &init_net.ipv4.frags.max_dist,
753 .maxlen = sizeof(int),
754 .mode = 0644,
755 .proc_handler = proc_dointvec_minmax,
756 .extra1 = &zero
757 },
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700758 { }
759};
760
Florian Westphale3a57d12014-07-24 16:50:35 +0200761/* secret interval has been deprecated */
762static int ip4_frags_secret_interval_unused;
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700763static struct ctl_table ip4_frags_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800764 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800765 .procname = "ipfrag_secret_interval",
Florian Westphale3a57d12014-07-24 16:50:35 +0200766 .data = &ip4_frags_secret_interval_unused,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800767 .maxlen = sizeof(int),
768 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800769 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800770 },
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800771 { }
772};
773
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000774static int __net_init ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800775{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800776 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800777 struct ctl_table_header *hdr;
778
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700779 table = ip4_frags_ns_ctl_table;
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800780 if (!net_eq(net, &init_net)) {
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700781 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
Ian Morris51456b22015-04-03 09:17:26 +0100782 if (!table)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800783 goto err_alloc;
784
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800785 table[0].data = &net->ipv4.frags.high_thresh;
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200786 table[0].extra1 = &net->ipv4.frags.low_thresh;
787 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800788 table[1].data = &net->ipv4.frags.low_thresh;
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200789 table[1].extra2 = &net->ipv4.frags.high_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800790 table[2].data = &net->ipv4.frags.timeout;
Nikolay Borisov0fbf4cb2016-02-15 12:11:31 +0200791 table[3].data = &net->ipv4.frags.max_dist;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800792 }
793
Eric W. Biedermanec8f23c2012-04-19 13:44:49 +0000794 hdr = register_net_sysctl(net, "net/ipv4", table);
Ian Morris51456b22015-04-03 09:17:26 +0100795 if (!hdr)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800796 goto err_reg;
797
798 net->ipv4.frags_hdr = hdr;
799 return 0;
800
801err_reg:
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800802 if (!net_eq(net, &init_net))
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800803 kfree(table);
804err_alloc:
805 return -ENOMEM;
806}
807
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000808static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800809{
810 struct ctl_table *table;
811
812 table = net->ipv4.frags_hdr->ctl_table_arg;
813 unregister_net_sysctl_table(net->ipv4.frags_hdr);
814 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800815}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700816
Fabian Frederick57a02c32014-10-01 19:18:57 +0200817static void __init ip4_frags_ctl_register(void)
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700818{
Eric W. Biederman43444752012-04-19 13:22:55 +0000819 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700820}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800821#else
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100822static int ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800823{
824 return 0;
825}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800826
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100827static void ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800828{
829}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700830
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100831static void __init ip4_frags_ctl_register(void)
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700832{
833}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800834#endif
835
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000836static int __net_init ipv4_frags_init_net(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800837{
Eric Dumazet1d6119b2015-11-02 09:03:11 -0800838 int res;
839
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000840 /* Fragment cache limits.
841 *
842 * The fragment memory accounting code, (tries to) account for
843 * the real memory usage, by measuring both the size of frag
844 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
845 * and the SKB's truesize.
846 *
847 * A 64K fragment consumes 129736 bytes (44*2944)+200
848 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
849 *
850 * We will commit 4MB at one time. Should we cross that limit
851 * we will prune down to 3MB, making room for approx 8 big 64K
852 * fragments 8x128k.
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800853 */
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000854 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
855 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800856 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800857 * Important NOTE! Fragment queue must be destroyed before MSL expires.
858 * RFC791 is wrong proposing to prolongate timer each fragment arrival
859 * by TTL.
860 */
861 net->ipv4.frags.timeout = IP_FRAG_TIME;
862
Nikolay Borisov0fbf4cb2016-02-15 12:11:31 +0200863 net->ipv4.frags.max_dist = 64;
864
Eric Dumazet1d6119b2015-11-02 09:03:11 -0800865 res = inet_frags_init_net(&net->ipv4.frags);
866 if (res)
867 return res;
868 res = ip4_frags_ns_ctl_register(net);
869 if (res)
870 inet_frags_uninit_net(&net->ipv4.frags);
871 return res;
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;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700892 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700893 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700894 ip4_frags.frag_expire = ip_expire;
Nikolay Aleksandrovd4ad4d22014-08-01 12:29:48 +0200895 ip4_frags.frags_cache_name = ip_frag_cache_name;
896 if (inet_frags_init(&ip4_frags))
897 panic("IP: failed to allocate ip4_frags cache\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}