blob: ed32313e307c43202a4710c6f5b74e14c19a4c20 [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 Dumazet6623e3b2011-01-05 07:52:55 +000082static inline u8 ip4_frag_ecn(u8 tos)
83{
Eric Dumazet5173cc02011-05-16 08:37:37 +000084 return 1 << (tos & INET_ECN_MASK);
Eric Dumazet6623e3b2011-01-05 07:52:55 +000085}
86
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070087static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080089int ip_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070090{
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -080091 return net->ipv4.frags.nqueues;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070092}
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -080094int ip_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070095{
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +000096 return sum_frag_mem_limit(&net->ipv4.frags);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070097}
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Herbert Xu1706d582007-10-14 00:38:15 -070099static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
100 struct net_device *dev);
101
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700102struct ip4_create_arg {
103 struct iphdr *iph;
104 u32 user;
105};
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,
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700112 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113}
114
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700115static unsigned int ip4_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700117 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
Eric Dumazetcbc264c2012-05-18 05:57:13 +0200123static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700124{
125 struct ipq *qp;
126 struct ip4_create_arg *arg = a;
127
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 &&
133 qp->user == arg->user;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700134}
135
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700136static void ip4_frag_init(struct inet_frag_queue *q, void *a)
137{
138 struct ipq *qp = container_of(q, struct ipq, q);
Gao feng54db0cc2012-06-08 01:21:40 +0000139 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
140 frags);
141 struct net *net = container_of(ipv4, struct net, ipv4);
142
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700143 struct ip4_create_arg *arg = a;
144
145 qp->protocol = arg->iph->protocol;
146 qp->id = arg->iph->id;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000147 qp->ecn = ip4_frag_ecn(arg->iph->tos);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700148 qp->saddr = arg->iph->saddr;
149 qp->daddr = arg->iph->daddr;
150 qp->user = arg->user;
151 qp->peer = sysctl_ipfrag_max_dist ?
David S. Millerc0efc882012-06-09 19:12:36 -0700152 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700153}
154
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700155static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700157 struct ipq *qp;
158
159 qp = container_of(q, struct ipq, q);
160 if (qp->peer)
161 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165/* Destruction primitives. */
166
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700167static __inline__ void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700169 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170}
171
172/* Kill ipq entry. It is not destroyed immediately,
173 * because caller (and someone more) holds reference count.
174 */
175static void ipq_kill(struct ipq *ipq)
176{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700177 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900180/* Memory limiting on fragments. Evictor trashes the oldest
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 * fragment queue until we are back under the threshold.
182 */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800183static void ip_evictor(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700185 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Amerigo Wang6b102862012-09-18 16:50:11 +0000187 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700188 if (evicted)
Pavel Emelyanovc5346fe2008-07-16 20:20:33 -0700189 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
191
192/*
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
Joe Perchesbc578a52008-03-28 16:35:27 -0700205 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 goto out;
207
208 ipq_kill(qp);
209
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700210 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
211 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Joe Perchesbc578a52008-03-28 16:35:27 -0700213 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700214 struct sk_buff *head = qp->q.fragments;
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000215 const struct iphdr *iph;
216 int err;
Denis V. Lunevcb846632008-03-24 15:31:00 -0700217
Eric Dumazet69df9d52009-11-05 20:59:47 -0800218 rcu_read_lock();
219 head->dev = dev_get_by_index_rcu(net, qp->iif);
Shan Weie9017b52010-01-23 01:57:42 -0800220 if (!head->dev)
221 goto out_rcu_unlock;
222
Eric Dumazet97599dc2013-04-16 12:55:41 +0000223 /* skb has no dst, perform route lookup again */
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000224 iph = ip_hdr(head);
David S. Millerc6cffba2012-07-26 11:14:38 +0000225 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
226 iph->tos, head->dev);
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000227 if (err)
228 goto out_rcu_unlock;
229
Shan Weie9017b52010-01-23 01:57:42 -0800230 /*
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000231 * Only an end host needs to send an ICMP
232 * "Fragment Reassembly Timeout" message, per RFC792.
Shan Weie9017b52010-01-23 01:57:42 -0800233 */
David S. Miller595fc712011-07-05 01:05:48 -0700234 if (qp->user == IP_DEFRAG_AF_PACKET ||
Vasily Averin7c3d5ab2014-05-03 03:14:04 +0400235 ((qp->user >= IP_DEFRAG_CONNTRACK_IN) &&
236 (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) &&
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
241 /* Send an ICMP "Fragment Reassembly Timeout" message. */
242 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
Shan Weie9017b52010-01-23 01:57:42 -0800243out_rcu_unlock:
Patrick McHardyd1c9ae62010-02-02 11:46:50 -0800244 rcu_read_unlock();
245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700247 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700248 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700251/* Find the correct entry in the "incomplete datagrams" queue for
252 * this IP datagram, and create new one, if nothing is found.
253 */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800254static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
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;
Pavel Emelyanov9a375802008-06-27 20:06:08 -0700262
263 read_lock(&ip4_frags.lock);
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? */
275static inline int ip_frag_too_far(struct ipq *qp)
276{
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);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000320 sub_frag_mem_limit(&qp->q, sum_truesize);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800321
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700322 qp->q.last_in = 0;
323 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 int flags, offset;
339 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700340 int err = -ENOENT;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000341 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Joe Perchesbc578a52008-03-28 16:35:27 -0700343 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 goto err;
345
Herbert Xu89cee8b2005-12-13 23:14:27 -0800346 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700347 unlikely(ip_frag_too_far(qp)) &&
348 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800349 ipq_kill(qp);
350 goto err;
351 }
352
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000353 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700354 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 flags = offset & ~IP_OFFSET;
356 offset &= IP_OFFSET;
357 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300358 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 /* Determine the position of this fragment. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900361 end = offset + skb->len - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700362 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364 /* Is this the final fragment? */
365 if ((flags & IP_MF) == 0) {
366 /* If we already have some bits beyond end
Justin P. Mattock42b2aa82011-11-28 20:31:00 -0800367 * or have different end, the segment is corrupted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700369 if (end < qp->q.len ||
Joe Perchesbc578a52008-03-28 16:35:27 -0700370 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 goto err;
Joe Perchesbc578a52008-03-28 16:35:27 -0700372 qp->q.last_in |= INET_FRAG_LAST_IN;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700373 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 } else {
375 if (end&7) {
376 end &= ~7;
377 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
378 skb->ip_summed = CHECKSUM_NONE;
379 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700380 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /* Some bits beyond end -> corruption. */
Joe Perchesbc578a52008-03-28 16:35:27 -0700382 if (qp->q.last_in & INET_FRAG_LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700384 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
386 }
387 if (end == offset)
388 goto err;
389
Herbert Xu1706d582007-10-14 00:38:15 -0700390 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 if (pskb_pull(skb, ihl) == NULL)
392 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700393
394 err = pskb_trim_rcsum(skb, end - offset);
395 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 goto err;
397
398 /* Find out which fragments are in front and at the back of us
399 * in the chain of fragments so far. We must know where to put
400 * this fragment, right?
401 */
Changli Gaod6bebca2010-06-29 04:39:37 +0000402 prev = qp->q.fragments_tail;
403 if (!prev || FRAG_CB(prev)->offset < offset) {
404 next = NULL;
405 goto found;
406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700408 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 if (FRAG_CB(next)->offset >= offset)
410 break; /* bingo! */
411 prev = next;
412 }
413
Changli Gaod6bebca2010-06-29 04:39:37 +0000414found:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 /* We found where to put this one. Check for overlap with
416 * preceding fragment, and, if needed, align things so that
417 * any overlaps are eliminated.
418 */
419 if (prev) {
420 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
421
422 if (i > 0) {
423 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700424 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 if (end <= offset)
426 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700427 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 if (!pskb_pull(skb, i))
429 goto err;
430 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
431 skb->ip_summed = CHECKSUM_NONE;
432 }
433 }
434
Herbert Xu1706d582007-10-14 00:38:15 -0700435 err = -ENOMEM;
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 while (next && FRAG_CB(next)->offset < end) {
438 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
439
440 if (i < next->len) {
441 /* Eat head of the next overlapped fragment
442 * and leave the loop. The next ones cannot overlap.
443 */
444 if (!pskb_pull(next, i))
445 goto err;
446 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700447 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 if (next->ip_summed != CHECKSUM_UNNECESSARY)
449 next->ip_summed = CHECKSUM_NONE;
450 break;
451 } else {
452 struct sk_buff *free_it = next;
453
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100454 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * new one drop it.
456 */
457 next = next->next;
458
459 if (prev)
460 prev->next = next;
461 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700462 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700464 qp->q.meat -= free_it->len;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000465 sub_frag_mem_limit(&qp->q, free_it->truesize);
466 kfree_skb(free_it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 }
468 }
469
470 FRAG_CB(skb)->offset = offset;
471
472 /* Insert this fragment in the chain of fragments. */
473 skb->next = next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000474 if (!next)
475 qp->q.fragments_tail = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 if (prev)
477 prev->next = skb;
478 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700479 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Herbert Xu1706d582007-10-14 00:38:15 -0700481 dev = skb->dev;
482 if (dev) {
483 qp->iif = dev->ifindex;
484 skb->dev = NULL;
485 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700486 qp->q.stamp = skb->tstamp;
487 qp->q.meat += skb->len;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000488 qp->ecn |= ecn;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000489 add_frag_mem_limit(&qp->q, skb->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 if (offset == 0)
Joe Perchesbc578a52008-03-28 16:35:27 -0700491 qp->q.last_in |= INET_FRAG_FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200493 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
494 skb->len + ihl > qp->q.max_size)
495 qp->q.max_size = skb->len + ihl;
496
Joe Perchesbc578a52008-03-28 16:35:27 -0700497 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
Eric Dumazet97599dc2013-04-16 12:55:41 +0000498 qp->q.meat == qp->q.len) {
499 unsigned long orefdst = skb->_skb_refdst;
Herbert Xu1706d582007-10-14 00:38:15 -0700500
Eric Dumazet97599dc2013-04-16 12:55:41 +0000501 skb->_skb_refdst = 0UL;
502 err = ip_frag_reasm(qp, prev, dev);
503 skb->_skb_refdst = orefdst;
504 return err;
505 }
506
507 skb_dst_drop(skb);
Jesper Dangaard Brouer3ef0eb02013-01-28 23:45:51 +0000508 inet_frag_lru_move(&qp->q);
Herbert Xu1706d582007-10-14 00:38:15 -0700509 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511err:
512 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700513 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}
515
516
517/* Build a new IP datagram from all its fragments. */
518
Herbert Xu1706d582007-10-14 00:38:15 -0700519static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
520 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700522 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700524 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 int len;
526 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700527 int err;
Eric Dumazet3cc49492012-05-19 03:02:20 +0000528 int sum_truesize;
Eric Dumazet5173cc02011-05-16 08:37:37 +0000529 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531 ipq_kill(qp);
532
Hannes Frederic Sowabe991972013-03-22 08:24:37 +0000533 ecn = ip_frag_ecn_table[qp->ecn];
Eric Dumazet5173cc02011-05-16 08:37:37 +0000534 if (unlikely(ecn == 0xff)) {
535 err = -EINVAL;
536 goto out_fail;
537 }
Herbert Xu1706d582007-10-14 00:38:15 -0700538 /* Make the one we just received the head. */
539 if (prev) {
540 head = prev->next;
541 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700542 if (!fp)
543 goto out_nomem;
544
545 fp->next = head->next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000546 if (!fp->next)
547 qp->q.fragments_tail = fp;
Herbert Xu1706d582007-10-14 00:38:15 -0700548 prev->next = fp;
549
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700550 skb_morph(head, qp->q.fragments);
551 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700552
Eric Dumazetcbf8f7b2012-04-19 06:10:26 +0000553 consume_skb(qp->q.fragments);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700554 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700555 }
556
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700557 WARN_ON(head == NULL);
558 WARN_ON(FRAG_CB(head)->offset != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300561 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700562 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Herbert Xu1706d582007-10-14 00:38:15 -0700564 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800565 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 goto out_oversize;
567
568 /* Head of list must not be cloned. */
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +0000569 if (skb_unclone(head, GFP_ATOMIC))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 goto out_nomem;
571
572 /* If the first fragment is fragmented itself, we split
573 * it to two chunks: the first with data and paged part
574 * and the second, holding only fragments. */
David S. Miller21dc3302010-08-23 00:13:46 -0700575 if (skb_has_frag_list(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 struct sk_buff *clone;
577 int i, plen = 0;
578
579 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
580 goto out_nomem;
581 clone->next = head->next;
582 head->next = clone;
583 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700584 skb_frag_list_init(head);
Eric Dumazet9e903e02011-10-18 21:00:24 +0000585 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
586 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 clone->len = clone->data_len = head->data_len - plen;
588 head->data_len -= clone->len;
589 head->len -= clone->len;
590 clone->csum = 0;
591 clone->ip_summed = head->ip_summed;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000592 add_frag_mem_limit(&qp->q, clone->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 }
594
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700595 skb_push(head, head->data - skb_network_header(head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Eric Dumazet3cc49492012-05-19 03:02:20 +0000597 sum_truesize = head->truesize;
598 for (fp = head->next; fp;) {
599 bool headstolen;
600 int delta;
601 struct sk_buff *next = fp->next;
602
603 sum_truesize += fp->truesize;
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);
Eric Dumazet3cc49492012-05-19 03:02:20 +0000608
609 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
610 kfree_skb_partial(fp, headstolen);
611 } else {
612 if (!skb_shinfo(head)->frag_list)
613 skb_shinfo(head)->frag_list = fp;
614 head->data_len += fp->len;
615 head->len += fp->len;
616 head->truesize += fp->truesize;
617 }
618 fp = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 }
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000620 sub_frag_mem_limit(&qp->q, sum_truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622 head->next = NULL;
623 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700624 head->tstamp = qp->q.stamp;
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200625 IPCB(head)->frag_max_size = qp->q.max_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700627 iph = ip_hdr(head);
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200628 /* max_size != 0 implies at least one fragment had IP_DF set */
629 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 iph->tot_len = htons(len);
Eric Dumazet5173cc02011-05-16 08:37:37 +0000631 iph->tos |= ecn;
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700632 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700633 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000634 qp->q.fragments_tail = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700635 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
637out_nomem:
Joe Perchesafd465032012-03-12 07:03:32 +0000638 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
639 qp);
David Howells45542472007-10-17 21:37:22 -0700640 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 goto out_fail;
642out_oversize:
Joe Perchese87cc472012-05-13 21:56:26 +0000643 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644out_fail:
David Fordbbf31bf2009-11-29 23:02:22 -0800645 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700646 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649/* Process an incoming IP datagram fragment. */
Herbert Xu776c7292007-10-14 00:38:32 -0700650int ip_defrag(struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 struct ipq *qp;
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800653 struct net *net;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900654
Eric Dumazetadf30902009-06-02 05:19:30 +0000655 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700656 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658 /* Start by cleaning up the memory. */
Amerigo Wang6b102862012-09-18 16:50:11 +0000659 ip_evictor(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 /* Lookup (or create) queue header */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800662 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
Herbert Xu1706d582007-10-14 00:38:15 -0700663 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700665 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Herbert Xu1706d582007-10-14 00:38:15 -0700667 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700669 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700670 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700671 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
673
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700674 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700676 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000678EXPORT_SYMBOL(ip_defrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Eric Dumazetbc416d92011-10-06 10:28:31 +0000680struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
681{
Johannes Berg1bf37512012-12-09 23:41:06 +0000682 struct iphdr iph;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000683 u32 len;
684
685 if (skb->protocol != htons(ETH_P_IP))
686 return skb;
687
Johannes Berg1bf37512012-12-09 23:41:06 +0000688 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
Eric Dumazetbc416d92011-10-06 10:28:31 +0000689 return skb;
690
Johannes Berg1bf37512012-12-09 23:41:06 +0000691 if (iph.ihl < 5 || iph.version != 4)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000692 return skb;
693
Johannes Berg1bf37512012-12-09 23:41:06 +0000694 len = ntohs(iph.tot_len);
695 if (skb->len < len || len < (iph.ihl * 4))
696 return skb;
697
698 if (ip_is_fragment(&iph)) {
Eric Dumazetbc416d92011-10-06 10:28:31 +0000699 skb = skb_share_check(skb, GFP_ATOMIC);
700 if (skb) {
Johannes Berg1bf37512012-12-09 23:41:06 +0000701 if (!pskb_may_pull(skb, iph.ihl*4))
702 return skb;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000703 if (pskb_trim_rcsum(skb, len))
704 return skb;
705 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
706 if (ip_defrag(skb, user))
707 return NULL;
Tom Herbert7539fad2013-12-15 22:12:18 -0800708 skb_clear_hash(skb);
Eric Dumazetbc416d92011-10-06 10:28:31 +0000709 }
710 }
711 return skb;
712}
713EXPORT_SYMBOL(ip_check_defrag);
714
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800715#ifdef CONFIG_SYSCTL
716static int zero;
717
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700718static struct ctl_table ip4_frags_ns_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800719 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800720 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800721 .data = &init_net.ipv4.frags.high_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800722 .maxlen = sizeof(int),
723 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800724 .proc_handler = proc_dointvec
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800725 },
726 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800727 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800728 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800729 .maxlen = sizeof(int),
730 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800731 .proc_handler = proc_dointvec
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800732 },
733 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800734 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800735 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800736 .maxlen = sizeof(int),
737 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800738 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800739 },
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700740 { }
741};
742
743static struct ctl_table ip4_frags_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800744 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800745 .procname = "ipfrag_secret_interval",
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800746 .data = &ip4_frags.secret_interval,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800747 .maxlen = sizeof(int),
748 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800749 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800750 },
751 {
752 .procname = "ipfrag_max_dist",
753 .data = &sysctl_ipfrag_max_dist,
754 .maxlen = sizeof(int),
755 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800756 .proc_handler = proc_dointvec_minmax,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800757 .extra1 = &zero
758 },
759 { }
760};
761
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000762static int __net_init ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800763{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800764 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800765 struct ctl_table_header *hdr;
766
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700767 table = ip4_frags_ns_ctl_table;
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800768 if (!net_eq(net, &init_net)) {
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700769 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800770 if (table == NULL)
771 goto err_alloc;
772
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800773 table[0].data = &net->ipv4.frags.high_thresh;
774 table[1].data = &net->ipv4.frags.low_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800775 table[2].data = &net->ipv4.frags.timeout;
Eric W. Biederman464dc802012-11-16 03:02:59 +0000776
777 /* Don't export sysctls to unprivileged users */
778 if (net->user_ns != &init_user_ns)
779 table[0].procname = NULL;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800780 }
781
Eric W. Biedermanec8f23c2012-04-19 13:44:49 +0000782 hdr = register_net_sysctl(net, "net/ipv4", table);
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800783 if (hdr == NULL)
784 goto err_reg;
785
786 net->ipv4.frags_hdr = hdr;
787 return 0;
788
789err_reg:
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800790 if (!net_eq(net, &init_net))
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800791 kfree(table);
792err_alloc:
793 return -ENOMEM;
794}
795
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000796static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800797{
798 struct ctl_table *table;
799
800 table = net->ipv4.frags_hdr->ctl_table_arg;
801 unregister_net_sysctl_table(net->ipv4.frags_hdr);
802 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800803}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700804
805static void ip4_frags_ctl_register(void)
806{
Eric W. Biederman43444752012-04-19 13:22:55 +0000807 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700808}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800809#else
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700810static inline int ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800811{
812 return 0;
813}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800814
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700815static inline void ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800816{
817}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700818
819static inline void ip4_frags_ctl_register(void)
820{
821}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800822#endif
823
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000824static int __net_init ipv4_frags_init_net(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800825{
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000826 /* Fragment cache limits.
827 *
828 * The fragment memory accounting code, (tries to) account for
829 * the real memory usage, by measuring both the size of frag
830 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
831 * and the SKB's truesize.
832 *
833 * A 64K fragment consumes 129736 bytes (44*2944)+200
834 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
835 *
836 * We will commit 4MB at one time. Should we cross that limit
837 * we will prune down to 3MB, making room for approx 8 big 64K
838 * fragments 8x128k.
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800839 */
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000840 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
841 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800842 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800843 * Important NOTE! Fragment queue must be destroyed before MSL expires.
844 * RFC791 is wrong proposing to prolongate timer each fragment arrival
845 * by TTL.
846 */
847 net->ipv4.frags.timeout = IP_FRAG_TIME;
848
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800849 inet_frags_init_net(&net->ipv4.frags);
850
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700851 return ip4_frags_ns_ctl_register(net);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800852}
853
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000854static void __net_exit ipv4_frags_exit_net(struct net *net)
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800855{
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700856 ip4_frags_ns_ctl_unregister(net);
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800857 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
858}
859
860static struct pernet_operations ip4_frags_ops = {
861 .init = ipv4_frags_init_net,
862 .exit = ipv4_frags_exit_net,
863};
864
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700865void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700867 ip4_frags_ctl_register();
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800868 register_pernet_subsys(&ip4_frags_ops);
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700869 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700870 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700871 ip4_frags.destructor = ip4_frag_free;
872 ip4_frags.skb_free = NULL;
873 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700874 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700875 ip4_frags.frag_expire = ip_expire;
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800876 ip4_frags.secret_interval = 10 * 60 * HZ;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700877 inet_frags_init(&ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}