blob: 187c6fcc3027779ba31d0721eb4a7389addaaf8a [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 */
57
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -080058static int sysctl_ipfrag_max_dist __read_mostly = 64;
Nikolay Aleksandrovd4ad4d22014-08-01 12:29:48 +020059static const char ip_frag_cache_name[] = "ip4-frags";
Herbert Xu89cee8b2005-12-13 23:14:27 -080060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061struct ipfrag_skb_cb
62{
63 struct inet_skb_parm h;
64 int offset;
65};
66
Jianjun Kongfd3f8c42008-11-03 02:47:38 -080067#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69/* Describe an entry in the "incomplete datagrams" queue. */
70struct ipq {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070071 struct inet_frag_queue q;
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 u32 user;
Al Viro18277772006-09-26 22:19:02 -070074 __be32 saddr;
75 __be32 daddr;
76 __be16 id;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 u8 protocol;
Eric Dumazet6623e3b2011-01-05 07:52:55 +000078 u8 ecn; /* RFC3168 support */
Florian Westphald6b915e2015-05-22 16:32:51 +020079 u16 max_df_size; /* largest frag with DF set seen */
Herbert Xu89cee8b2005-12-13 23:14:27 -080080 int iif;
David Ahern385add92015-09-29 20:07:13 -070081 int vif; /* L3 master device index */
Herbert Xu89cee8b2005-12-13 23:14:27 -080082 unsigned int rid;
83 struct inet_peer *peer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084};
85
Fabian Frederickaa1f7312014-11-04 20:44:04 +010086static u8 ip4_frag_ecn(u8 tos)
Eric Dumazet6623e3b2011-01-05 07:52:55 +000087{
Eric Dumazet5173cc02011-05-16 08:37:37 +000088 return 1 << (tos & INET_ECN_MASK);
Eric Dumazet6623e3b2011-01-05 07:52:55 +000089}
90
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070091static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080093int ip_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070094{
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +000095 return sum_frag_mem_limit(&net->ipv4.frags);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070096}
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Herbert Xu1706d582007-10-14 00:38:15 -070098static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
99 struct net_device *dev);
100
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700101struct ip4_create_arg {
102 struct iphdr *iph;
103 u32 user;
David Ahern9972f132015-08-13 14:59:09 -0600104 int vif;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700105};
106
Al Viro18277772006-09-26 22:19:02 -0700107static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
Hannes Frederic Sowae7b519b2013-10-23 11:06:55 +0200109 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
Al Viro18277772006-09-26 22:19:02 -0700110 return jhash_3words((__force u32)id << 16 | prot,
111 (__force u32)saddr, (__force u32)daddr,
Florian Westphalfb3cfe62014-07-24 16:50:30 +0200112 ip4_frags.rnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113}
114
Florian Westphal36c77782014-07-24 16:50:29 +0200115static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Florian Westphal36c77782014-07-24 16:50:29 +0200117 const struct ipq *ipq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700119 ipq = container_of(q, struct ipq, q);
120 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121}
122
Florian Westphal36c77782014-07-24 16:50:29 +0200123static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700124{
Florian Westphal36c77782014-07-24 16:50:29 +0200125 const struct ipq *qp;
126 const struct ip4_create_arg *arg = a;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700127
128 qp = container_of(q, struct ipq, q);
Eric Dumazeta02cec22010-09-22 20:43:57 +0000129 return qp->id == arg->iph->id &&
Eric Dumazetcbc264c2012-05-18 05:57:13 +0200130 qp->saddr == arg->iph->saddr &&
131 qp->daddr == arg->iph->daddr &&
132 qp->protocol == arg->iph->protocol &&
David Ahern9972f132015-08-13 14:59:09 -0600133 qp->user == arg->user &&
134 qp->vif == arg->vif;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700135}
136
Florian Westphal36c77782014-07-24 16:50:29 +0200137static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700138{
139 struct ipq *qp = container_of(q, struct ipq, q);
Gao feng54db0cc2012-06-08 01:21:40 +0000140 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
141 frags);
142 struct net *net = container_of(ipv4, struct net, ipv4);
143
Florian Westphal36c77782014-07-24 16:50:29 +0200144 const struct ip4_create_arg *arg = a;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700145
146 qp->protocol = arg->iph->protocol;
147 qp->id = arg->iph->id;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000148 qp->ecn = ip4_frag_ecn(arg->iph->tos);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700149 qp->saddr = arg->iph->saddr;
150 qp->daddr = arg->iph->daddr;
David Ahern9972f132015-08-13 14:59:09 -0600151 qp->vif = arg->vif;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700152 qp->user = arg->user;
153 qp->peer = sysctl_ipfrag_max_dist ?
David Ahern192132b2015-08-27 16:07:03 -0700154 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
155 NULL;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700156}
157
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100158static void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700160 struct ipq *qp;
161
162 qp = container_of(q, struct ipq, q);
163 if (qp->peer)
164 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168/* Destruction primitives. */
169
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100170static void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700172 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
175/* Kill ipq entry. It is not destroyed immediately,
176 * because caller (and someone more) holds reference count.
177 */
178static void ipq_kill(struct ipq *ipq)
179{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700180 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181}
182
Andy Zhou5cf42282015-05-15 14:15:35 -0700183static bool frag_expire_skip_icmp(u32 user)
184{
185 return user == IP_DEFRAG_AF_PACKET ||
186 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
Andy Zhou8bc04862015-05-15 14:15:36 -0700187 __IP_DEFRAG_CONNTRACK_IN_END) ||
188 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
189 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
Andy Zhou5cf42282015-05-15 14:15:35 -0700190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/*
193 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
194 */
195static void ip_expire(unsigned long arg)
196{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700197 struct ipq *qp;
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700198 struct net *net;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700199
200 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700201 net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700203 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200205 if (qp->q.flags & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 goto out;
207
208 ipq_kill(qp);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700209 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Nikolay Aleksandrovcaaecdd2015-07-23 12:05:40 +0200211 if (!inet_frag_evicting(&qp->q)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700212 struct sk_buff *head = qp->q.fragments;
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000213 const struct iphdr *iph;
214 int err;
Denis V. Lunevcb846632008-03-24 15:31:00 -0700215
Nikolay Aleksandrov2e404f62014-08-01 12:29:47 +0200216 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
217
218 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
219 goto out;
220
Eric Dumazet69df9d52009-11-05 20:59:47 -0800221 rcu_read_lock();
222 head->dev = dev_get_by_index_rcu(net, qp->iif);
Shan Weie9017b52010-01-23 01:57:42 -0800223 if (!head->dev)
224 goto out_rcu_unlock;
225
Eric Dumazet97599dc2013-04-16 12:55:41 +0000226 /* skb has no dst, perform route lookup again */
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000227 iph = ip_hdr(head);
David S. Millerc6cffba2012-07-26 11:14:38 +0000228 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
229 iph->tos, head->dev);
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000230 if (err)
231 goto out_rcu_unlock;
232
Nikolay Aleksandrov2e404f62014-08-01 12:29:47 +0200233 /* Only an end host needs to send an ICMP
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000234 * "Fragment Reassembly Timeout" message, per RFC792.
Shan Weie9017b52010-01-23 01:57:42 -0800235 */
Andy Zhou5cf42282015-05-15 14:15:35 -0700236 if (frag_expire_skip_icmp(qp->user) &&
237 (skb_rtable(head)->rt_type != RTN_LOCAL))
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000238 goto out_rcu_unlock;
Shan Weie9017b52010-01-23 01:57:42 -0800239
Shan Weie9017b52010-01-23 01:57:42 -0800240 /* Send an ICMP "Fragment Reassembly Timeout" message. */
241 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
Shan Weie9017b52010-01-23 01:57:42 -0800242out_rcu_unlock:
Patrick McHardyd1c9ae62010-02-02 11:46:50 -0800243 rcu_read_unlock();
244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700246 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700247 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700250/* Find the correct entry in the "incomplete datagrams" queue for
251 * this IP datagram, and create new one, if nothing is found.
252 */
David Ahern9972f132015-08-13 14:59:09 -0600253static struct ipq *ip_find(struct net *net, struct iphdr *iph,
254 u32 user, int vif)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700256 struct inet_frag_queue *q;
257 struct ip4_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700258 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700260 arg.iph = iph;
261 arg.user = user;
David Ahern9972f132015-08-13 14:59:09 -0600262 arg.vif = vif;
Pavel Emelyanov9a375802008-06-27 20:06:08 -0700263
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700264 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700265
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800266 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
Hannes Frederic Sowa5a3da1f2013-03-15 11:32:30 +0000267 if (IS_ERR_OR_NULL(q)) {
268 inet_frag_maybe_warn_overflow(q, pr_fmt());
269 return NULL;
270 }
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700271 return container_of(q, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
Herbert Xu89cee8b2005-12-13 23:14:27 -0800274/* Is the fragment too far ahead to be part of ipq? */
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100275static int ip_frag_too_far(struct ipq *qp)
Herbert Xu89cee8b2005-12-13 23:14:27 -0800276{
277 struct inet_peer *peer = qp->peer;
278 unsigned int max = sysctl_ipfrag_max_dist;
279 unsigned int start, end;
280
281 int rc;
282
283 if (!peer || !max)
284 return 0;
285
286 start = qp->rid;
287 end = atomic_inc_return(&peer->rid);
288 qp->rid = end;
289
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700290 rc = qp->q.fragments && (end - start) > max;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800291
292 if (rc) {
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700293 struct net *net;
294
295 net = container_of(qp->q.net, struct net, ipv4.frags);
296 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800297 }
298
299 return rc;
300}
301
302static int ip_frag_reinit(struct ipq *qp)
303{
304 struct sk_buff *fp;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000305 unsigned int sum_truesize = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800306
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800307 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700308 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800309 return -ETIMEDOUT;
310 }
311
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700312 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800313 do {
314 struct sk_buff *xp = fp->next;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000315
316 sum_truesize += fp->truesize;
317 kfree_skb(fp);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800318 fp = xp;
319 } while (fp);
Florian Westphal0e60d242015-07-23 12:05:38 +0200320 sub_frag_mem_limit(qp->q.net, sum_truesize);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800321
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200322 qp->q.flags = 0;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700323 qp->q.len = 0;
324 qp->q.meat = 0;
325 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000326 qp->q.fragments_tail = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800327 qp->iif = 0;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000328 qp->ecn = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800329
330 return 0;
331}
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700334static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
336 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700337 struct net_device *dev;
Florian Westphald6b915e2015-05-22 16:32:51 +0200338 unsigned int fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 int flags, offset;
340 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700341 int err = -ENOENT;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000342 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200344 if (qp->q.flags & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 goto err;
346
Herbert Xu89cee8b2005-12-13 23:14:27 -0800347 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700348 unlikely(ip_frag_too_far(qp)) &&
349 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800350 ipq_kill(qp);
351 goto err;
352 }
353
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000354 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700355 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 flags = offset & ~IP_OFFSET;
357 offset &= IP_OFFSET;
358 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300359 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
361 /* Determine the position of this fragment. */
Edward Hyunkoo Jee0848f642015-07-21 09:43:59 +0200362 end = offset + skb->len - skb_network_offset(skb) - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700363 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 /* Is this the final fragment? */
366 if ((flags & IP_MF) == 0) {
367 /* If we already have some bits beyond end
Justin P. Mattock42b2aa82011-11-28 20:31:00 -0800368 * or have different end, the segment is corrupted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700370 if (end < qp->q.len ||
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200371 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 goto err;
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200373 qp->q.flags |= INET_FRAG_LAST_IN;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700374 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 } else {
376 if (end&7) {
377 end &= ~7;
378 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
379 skb->ip_summed = CHECKSUM_NONE;
380 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700381 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 /* Some bits beyond end -> corruption. */
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200383 if (qp->q.flags & INET_FRAG_LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700385 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 }
387 }
388 if (end == offset)
389 goto err;
390
Herbert Xu1706d582007-10-14 00:38:15 -0700391 err = -ENOMEM;
Edward Hyunkoo Jee0848f642015-07-21 09:43:59 +0200392 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700394
395 err = pskb_trim_rcsum(skb, end - offset);
396 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 goto err;
398
399 /* Find out which fragments are in front and at the back of us
400 * in the chain of fragments so far. We must know where to put
401 * this fragment, right?
402 */
Changli Gaod6bebca2010-06-29 04:39:37 +0000403 prev = qp->q.fragments_tail;
404 if (!prev || FRAG_CB(prev)->offset < offset) {
405 next = NULL;
406 goto found;
407 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700409 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 if (FRAG_CB(next)->offset >= offset)
411 break; /* bingo! */
412 prev = next;
413 }
414
Changli Gaod6bebca2010-06-29 04:39:37 +0000415found:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 /* We found where to put this one. Check for overlap with
417 * preceding fragment, and, if needed, align things so that
418 * any overlaps are eliminated.
419 */
420 if (prev) {
421 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
422
423 if (i > 0) {
424 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700425 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 if (end <= offset)
427 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700428 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 if (!pskb_pull(skb, i))
430 goto err;
431 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
432 skb->ip_summed = CHECKSUM_NONE;
433 }
434 }
435
Herbert Xu1706d582007-10-14 00:38:15 -0700436 err = -ENOMEM;
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 while (next && FRAG_CB(next)->offset < end) {
439 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
440
441 if (i < next->len) {
442 /* Eat head of the next overlapped fragment
443 * and leave the loop. The next ones cannot overlap.
444 */
445 if (!pskb_pull(next, i))
446 goto err;
447 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700448 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 if (next->ip_summed != CHECKSUM_UNNECESSARY)
450 next->ip_summed = CHECKSUM_NONE;
451 break;
452 } else {
453 struct sk_buff *free_it = next;
454
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100455 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 * new one drop it.
457 */
458 next = next->next;
459
460 if (prev)
461 prev->next = next;
462 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700463 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700465 qp->q.meat -= free_it->len;
Florian Westphal0e60d242015-07-23 12:05:38 +0200466 sub_frag_mem_limit(qp->q.net, free_it->truesize);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000467 kfree_skb(free_it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
469 }
470
471 FRAG_CB(skb)->offset = offset;
472
473 /* Insert this fragment in the chain of fragments. */
474 skb->next = next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000475 if (!next)
476 qp->q.fragments_tail = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 if (prev)
478 prev->next = skb;
479 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700480 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Herbert Xu1706d582007-10-14 00:38:15 -0700482 dev = skb->dev;
483 if (dev) {
484 qp->iif = dev->ifindex;
485 skb->dev = NULL;
486 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700487 qp->q.stamp = skb->tstamp;
488 qp->q.meat += skb->len;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000489 qp->ecn |= ecn;
Florian Westphal0e60d242015-07-23 12:05:38 +0200490 add_frag_mem_limit(qp->q.net, skb->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 if (offset == 0)
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200492 qp->q.flags |= INET_FRAG_FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Florian Westphald6b915e2015-05-22 16:32:51 +0200494 fragsize = skb->len + ihl;
495
496 if (fragsize > qp->q.max_size)
497 qp->q.max_size = fragsize;
498
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200499 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
Florian Westphald6b915e2015-05-22 16:32:51 +0200500 fragsize > qp->max_df_size)
501 qp->max_df_size = fragsize;
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200502
Nikolay Aleksandrov06aa8b82014-08-01 12:29:44 +0200503 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
Eric Dumazet97599dc2013-04-16 12:55:41 +0000504 qp->q.meat == qp->q.len) {
505 unsigned long orefdst = skb->_skb_refdst;
Herbert Xu1706d582007-10-14 00:38:15 -0700506
Eric Dumazet97599dc2013-04-16 12:55:41 +0000507 skb->_skb_refdst = 0UL;
508 err = ip_frag_reasm(qp, prev, dev);
509 skb->_skb_refdst = orefdst;
510 return err;
511 }
512
513 skb_dst_drop(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700514 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
516err:
517 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700518 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
520
521
522/* Build a new IP datagram from all its fragments. */
523
Herbert Xu1706d582007-10-14 00:38:15 -0700524static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
525 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700527 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700529 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 int len;
531 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700532 int err;
Eric Dumazet5173cc02011-05-16 08:37:37 +0000533 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535 ipq_kill(qp);
536
Hannes Frederic Sowabe991972013-03-22 08:24:37 +0000537 ecn = ip_frag_ecn_table[qp->ecn];
Eric Dumazet5173cc02011-05-16 08:37:37 +0000538 if (unlikely(ecn == 0xff)) {
539 err = -EINVAL;
540 goto out_fail;
541 }
Herbert Xu1706d582007-10-14 00:38:15 -0700542 /* Make the one we just received the head. */
543 if (prev) {
544 head = prev->next;
545 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700546 if (!fp)
547 goto out_nomem;
548
549 fp->next = head->next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000550 if (!fp->next)
551 qp->q.fragments_tail = fp;
Herbert Xu1706d582007-10-14 00:38:15 -0700552 prev->next = fp;
553
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700554 skb_morph(head, qp->q.fragments);
555 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700556
Eric Dumazetcbf8f7b2012-04-19 06:10:26 +0000557 consume_skb(qp->q.fragments);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700558 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700559 }
560
Ian Morris51456b22015-04-03 09:17:26 +0100561 WARN_ON(!head);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700562 WARN_ON(FRAG_CB(head)->offset != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300565 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700566 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Herbert Xu1706d582007-10-14 00:38:15 -0700568 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800569 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 goto out_oversize;
571
572 /* Head of list must not be cloned. */
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +0000573 if (skb_unclone(head, GFP_ATOMIC))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 goto out_nomem;
575
576 /* If the first fragment is fragmented itself, we split
577 * it to two chunks: the first with data and paged part
578 * and the second, holding only fragments. */
David S. Miller21dc3302010-08-23 00:13:46 -0700579 if (skb_has_frag_list(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 struct sk_buff *clone;
581 int i, plen = 0;
582
Ian Morris51456b22015-04-03 09:17:26 +0100583 clone = alloc_skb(0, GFP_ATOMIC);
584 if (!clone)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 goto out_nomem;
586 clone->next = head->next;
587 head->next = clone;
588 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700589 skb_frag_list_init(head);
Eric Dumazet9e903e02011-10-18 21:00:24 +0000590 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
591 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 clone->len = clone->data_len = head->data_len - plen;
593 head->data_len -= clone->len;
594 head->len -= clone->len;
595 clone->csum = 0;
596 clone->ip_summed = head->ip_summed;
Florian Westphal0e60d242015-07-23 12:05:38 +0200597 add_frag_mem_limit(qp->q.net, clone->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 }
599
Florian Westphal14fe22e2015-07-11 01:37:36 +0200600 skb_shinfo(head)->frag_list = head->next;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700601 skb_push(head, head->data - skb_network_header(head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Florian Westphal14fe22e2015-07-11 01:37:36 +0200603 for (fp=head->next; fp; fp = fp->next) {
604 head->data_len += fp->len;
605 head->len += fp->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 if (head->ip_summed != fp->ip_summed)
607 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700608 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 head->csum = csum_add(head->csum, fp->csum);
Florian Westphal14fe22e2015-07-11 01:37:36 +0200610 head->truesize += fp->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 }
David S. Miller5510b3c2015-07-31 23:52:20 -0700612 sub_frag_mem_limit(qp->q.net, head->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614 head->next = NULL;
615 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700616 head->tstamp = qp->q.stamp;
Florian Westphald6b915e2015-05-22 16:32:51 +0200617 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700619 iph = ip_hdr(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 iph->tot_len = htons(len);
Eric Dumazet5173cc02011-05-16 08:37:37 +0000621 iph->tos |= ecn;
Florian Westphald6b915e2015-05-22 16:32:51 +0200622
623 /* When we set IP_DF on a refragmented skb we must also force a
624 * call to ip_fragment to avoid forwarding a DF-skb of size s while
625 * original sender only sent fragments of size f (where f < s).
626 *
627 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
628 * frag seen to avoid sending tiny DF-fragments in case skb was built
629 * from one very small df-fragment and one large non-df frag.
630 */
631 if (qp->max_df_size == qp->q.max_size) {
632 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
633 iph->frag_off = htons(IP_DF);
634 } else {
635 iph->frag_off = 0;
636 }
637
Edward Hyunkoo Jee0848f642015-07-21 09:43:59 +0200638 ip_send_check(iph);
639
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700640 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700641 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000642 qp->q.fragments_tail = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700643 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
645out_nomem:
Joe Perchesba7a46f2014-11-11 10:59:17 -0800646 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
David Howells45542472007-10-17 21:37:22 -0700647 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 goto out_fail;
649out_oversize:
Joe Perchese87cc472012-05-13 21:56:26 +0000650 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651out_fail:
David Fordbbf31bf2009-11-29 23:02:22 -0800652 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700653 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654}
655
656/* Process an incoming IP datagram fragment. */
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500657int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
David Ahern9972f132015-08-13 14:59:09 -0600659 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
David Ahern385add92015-09-29 20:07:13 -0700660 int vif = l3mdev_master_ifindex_rcu(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 struct ipq *qp;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900662
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700663 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
Joe Stringer8282f272016-01-22 15:49:12 -0800664 skb_orphan(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 /* Lookup (or create) queue header */
David Ahern9972f132015-08-13 14:59:09 -0600667 qp = ip_find(net, ip_hdr(skb), user, vif);
Ian Morris00db4122015-04-03 09:17:27 +0100668 if (qp) {
Herbert Xu1706d582007-10-14 00:38:15 -0700669 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700671 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Herbert Xu1706d582007-10-14 00:38:15 -0700673 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700675 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700676 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700677 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700680 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700682 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000684EXPORT_SYMBOL(ip_defrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500686struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000687{
Johannes Berg1bf37512012-12-09 23:41:06 +0000688 struct iphdr iph;
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300689 int netoff;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000690 u32 len;
691
692 if (skb->protocol != htons(ETH_P_IP))
693 return skb;
694
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300695 netoff = skb_network_offset(skb);
696
697 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000698 return skb;
699
Johannes Berg1bf37512012-12-09 23:41:06 +0000700 if (iph.ihl < 5 || iph.version != 4)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000701 return skb;
702
Johannes Berg1bf37512012-12-09 23:41:06 +0000703 len = ntohs(iph.tot_len);
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300704 if (skb->len < netoff + len || len < (iph.ihl * 4))
Johannes Berg1bf37512012-12-09 23:41:06 +0000705 return skb;
706
707 if (ip_is_fragment(&iph)) {
Eric Dumazetbc416d92011-10-06 10:28:31 +0000708 skb = skb_share_check(skb, GFP_ATOMIC);
709 if (skb) {
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300710 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
Johannes Berg1bf37512012-12-09 23:41:06 +0000711 return skb;
Alexander Drozdov3e32e732015-03-05 10:29:39 +0300712 if (pskb_trim_rcsum(skb, netoff + len))
Eric Dumazetbc416d92011-10-06 10:28:31 +0000713 return skb;
714 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -0500715 if (ip_defrag(net, skb, user))
Eric Dumazetbc416d92011-10-06 10:28:31 +0000716 return NULL;
Tom Herbert7539fad2013-12-15 22:12:18 -0800717 skb_clear_hash(skb);
Eric Dumazetbc416d92011-10-06 10:28:31 +0000718 }
719 }
720 return skb;
721}
722EXPORT_SYMBOL(ip_check_defrag);
723
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800724#ifdef CONFIG_SYSCTL
725static int zero;
726
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700727static struct ctl_table ip4_frags_ns_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800728 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800729 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800730 .data = &init_net.ipv4.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800731 .maxlen = sizeof(int),
732 .mode = 0644,
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200733 .proc_handler = proc_dointvec_minmax,
734 .extra1 = &init_net.ipv4.frags.low_thresh
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800735 },
736 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800737 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800738 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800739 .maxlen = sizeof(int),
740 .mode = 0644,
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200741 .proc_handler = proc_dointvec_minmax,
742 .extra1 = &zero,
743 .extra2 = &init_net.ipv4.frags.high_thresh
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800744 },
745 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800746 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800747 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800748 .maxlen = sizeof(int),
749 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800750 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800751 },
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700752 { }
753};
754
Florian Westphale3a57d12014-07-24 16:50:35 +0200755/* secret interval has been deprecated */
756static int ip4_frags_secret_interval_unused;
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700757static struct ctl_table ip4_frags_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800758 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800759 .procname = "ipfrag_secret_interval",
Florian Westphale3a57d12014-07-24 16:50:35 +0200760 .data = &ip4_frags_secret_interval_unused,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800761 .maxlen = sizeof(int),
762 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800763 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800764 },
765 {
766 .procname = "ipfrag_max_dist",
767 .data = &sysctl_ipfrag_max_dist,
768 .maxlen = sizeof(int),
769 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800770 .proc_handler = proc_dointvec_minmax,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800771 .extra1 = &zero
772 },
773 { }
774};
775
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000776static int __net_init ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800777{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800778 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800779 struct ctl_table_header *hdr;
780
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700781 table = ip4_frags_ns_ctl_table;
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800782 if (!net_eq(net, &init_net)) {
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700783 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
Ian Morris51456b22015-04-03 09:17:26 +0100784 if (!table)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800785 goto err_alloc;
786
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800787 table[0].data = &net->ipv4.frags.high_thresh;
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200788 table[0].extra1 = &net->ipv4.frags.low_thresh;
789 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800790 table[1].data = &net->ipv4.frags.low_thresh;
Nikolay Aleksandrov1bab4c72014-07-24 16:50:37 +0200791 table[1].extra2 = &net->ipv4.frags.high_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800792 table[2].data = &net->ipv4.frags.timeout;
Eric W. Biederman464dc802012-11-16 03:02:59 +0000793
794 /* Don't export sysctls to unprivileged users */
795 if (net->user_ns != &init_user_ns)
796 table[0].procname = NULL;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800797 }
798
Eric W. Biedermanec8f23c2012-04-19 13:44:49 +0000799 hdr = register_net_sysctl(net, "net/ipv4", table);
Ian Morris51456b22015-04-03 09:17:26 +0100800 if (!hdr)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800801 goto err_reg;
802
803 net->ipv4.frags_hdr = hdr;
804 return 0;
805
806err_reg:
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800807 if (!net_eq(net, &init_net))
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800808 kfree(table);
809err_alloc:
810 return -ENOMEM;
811}
812
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000813static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800814{
815 struct ctl_table *table;
816
817 table = net->ipv4.frags_hdr->ctl_table_arg;
818 unregister_net_sysctl_table(net->ipv4.frags_hdr);
819 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800820}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700821
Fabian Frederick57a02c32014-10-01 19:18:57 +0200822static void __init ip4_frags_ctl_register(void)
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700823{
Eric W. Biederman43444752012-04-19 13:22:55 +0000824 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700825}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800826#else
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100827static int ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800828{
829 return 0;
830}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800831
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100832static void ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800833{
834}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700835
Fabian Frederickaa1f7312014-11-04 20:44:04 +0100836static void __init ip4_frags_ctl_register(void)
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700837{
838}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800839#endif
840
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000841static int __net_init ipv4_frags_init_net(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800842{
Eric Dumazet1d6119b2015-11-02 09:03:11 -0800843 int res;
844
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000845 /* Fragment cache limits.
846 *
847 * The fragment memory accounting code, (tries to) account for
848 * the real memory usage, by measuring both the size of frag
849 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
850 * and the SKB's truesize.
851 *
852 * A 64K fragment consumes 129736 bytes (44*2944)+200
853 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
854 *
855 * We will commit 4MB at one time. Should we cross that limit
856 * we will prune down to 3MB, making room for approx 8 big 64K
857 * fragments 8x128k.
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800858 */
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000859 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
860 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800861 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800862 * Important NOTE! Fragment queue must be destroyed before MSL expires.
863 * RFC791 is wrong proposing to prolongate timer each fragment arrival
864 * by TTL.
865 */
866 net->ipv4.frags.timeout = IP_FRAG_TIME;
867
Eric Dumazet1d6119b2015-11-02 09:03:11 -0800868 res = inet_frags_init_net(&net->ipv4.frags);
869 if (res)
870 return res;
871 res = ip4_frags_ns_ctl_register(net);
872 if (res)
873 inet_frags_uninit_net(&net->ipv4.frags);
874 return res;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800875}
876
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000877static void __net_exit ipv4_frags_exit_net(struct net *net)
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800878{
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700879 ip4_frags_ns_ctl_unregister(net);
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800880 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
881}
882
883static struct pernet_operations ip4_frags_ops = {
884 .init = ipv4_frags_init_net,
885 .exit = ipv4_frags_exit_net,
886};
887
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700888void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700890 ip4_frags_ctl_register();
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800891 register_pernet_subsys(&ip4_frags_ops);
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700892 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700893 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700894 ip4_frags.destructor = ip4_frag_free;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700895 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700896 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700897 ip4_frags.frag_expire = ip_expire;
Nikolay Aleksandrovd4ad4d22014-08-01 12:29:48 +0200898 ip4_frags.frags_cache_name = ip_frag_cache_name;
899 if (inet_frags_init(&ip4_frags))
900 panic("IP: failed to allocate ip4_frags cache\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901}