blob: b6d30acb600c5e306ba825ed394c252e4fe27a8b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The IP fragmentation functionality.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
Alan Cox113aa832008-10-13 19:01:08 -07009 * Alan Cox <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 *
11 * Fixes:
12 * Alan Cox : Split from ip.c , see ip_input.c for history.
13 * David S. Miller : Begin massive cleanup...
14 * Andi Kleen : Add sysctls.
15 * xxxx : Overlapfrag bug.
16 * Ultima : ip_expire() kernel panic.
17 * Bill Hawes : Frag accounting and evictor fixes.
18 * John McDonald : 0 length frag bug.
19 * Alexey Kuznetsov: SMP races, threading, cleanup.
20 * Patrick McHardy : LRU queue of frag heads for evictor.
21 */
22
Joe Perchesafd465032012-03-12 07:03:32 +000023#define pr_fmt(fmt) "IPv4: " fmt
24
Herbert Xu89cee8b2005-12-13 23:14:27 -080025#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/types.h>
28#include <linux/mm.h>
29#include <linux/jiffies.h>
30#include <linux/skbuff.h>
31#include <linux/list.h>
32#include <linux/ip.h>
33#include <linux/icmp.h>
34#include <linux/netdevice.h>
35#include <linux/jhash.h>
36#include <linux/random.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/slab.h>
Shan Weie9017b52010-01-23 01:57:42 -080038#include <net/route.h>
39#include <net/dst.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <net/sock.h>
41#include <net/ip.h>
42#include <net/icmp.h>
43#include <net/checksum.h>
Herbert Xu89cee8b2005-12-13 23:14:27 -080044#include <net/inetpeer.h>
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070045#include <net/inet_frag.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/inet.h>
49#include <linux/netfilter_ipv4.h>
Eric Dumazet6623e3b2011-01-05 07:52:55 +000050#include <net/inet_ecn.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
54 * as well. Or notify me, at least. --ANK
55 */
56
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -080057static int sysctl_ipfrag_max_dist __read_mostly = 64;
Herbert Xu89cee8b2005-12-13 23:14:27 -080058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059struct ipfrag_skb_cb
60{
61 struct inet_skb_parm h;
62 int offset;
63};
64
Jianjun Kongfd3f8c42008-11-03 02:47:38 -080065#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67/* Describe an entry in the "incomplete datagrams" queue. */
68struct ipq {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070069 struct inet_frag_queue q;
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 u32 user;
Al Viro18277772006-09-26 22:19:02 -070072 __be32 saddr;
73 __be32 daddr;
74 __be16 id;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 u8 protocol;
Eric Dumazet6623e3b2011-01-05 07:52:55 +000076 u8 ecn; /* RFC3168 support */
Herbert Xu89cee8b2005-12-13 23:14:27 -080077 int iif;
78 unsigned int rid;
79 struct inet_peer *peer;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080};
81
Eric Dumazet5173cc02011-05-16 08:37:37 +000082/* RFC 3168 support :
83 * We want to check ECN values of all fragments, do detect invalid combinations.
84 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
85 */
David S. Miller1d1652c2011-05-17 17:28:02 -040086#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
87#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
88#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
89#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
Eric Dumazet6623e3b2011-01-05 07:52:55 +000090
91static inline u8 ip4_frag_ecn(u8 tos)
92{
Eric Dumazet5173cc02011-05-16 08:37:37 +000093 return 1 << (tos & INET_ECN_MASK);
Eric Dumazet6623e3b2011-01-05 07:52:55 +000094}
95
Eric Dumazet5173cc02011-05-16 08:37:37 +000096/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
97 * Value : 0xff if frame should be dropped.
98 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
99 */
100static const u8 ip4_frag_ecn_table[16] = {
101 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
103 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
104 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
105
106 /* invalid combinations : drop frame */
107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
112 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
113 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
114};
115
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700116static struct inet_frags ip4_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800118int ip_frag_nqueues(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700119{
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800120 return net->ipv4.frags.nqueues;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700121}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800123int ip_frag_mem(struct net *net)
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700124{
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000125 return sum_frag_mem_limit(&net->ipv4.frags);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Herbert Xu1706d582007-10-14 00:38:15 -0700128static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
129 struct net_device *dev);
130
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700131struct ip4_create_arg {
132 struct iphdr *iph;
133 u32 user;
134};
135
Al Viro18277772006-09-26 22:19:02 -0700136static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
Al Viro18277772006-09-26 22:19:02 -0700138 return jhash_3words((__force u32)id << 16 | prot,
139 (__force u32)saddr, (__force u32)daddr,
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700140 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700143static unsigned int ip4_hashfn(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700145 struct ipq *ipq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700147 ipq = container_of(q, struct ipq, q);
148 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Eric Dumazetcbc264c2012-05-18 05:57:13 +0200151static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700152{
153 struct ipq *qp;
154 struct ip4_create_arg *arg = a;
155
156 qp = container_of(q, struct ipq, q);
Eric Dumazeta02cec22010-09-22 20:43:57 +0000157 return qp->id == arg->iph->id &&
Eric Dumazetcbc264c2012-05-18 05:57:13 +0200158 qp->saddr == arg->iph->saddr &&
159 qp->daddr == arg->iph->daddr &&
160 qp->protocol == arg->iph->protocol &&
161 qp->user == arg->user;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700162}
163
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700164static void ip4_frag_init(struct inet_frag_queue *q, void *a)
165{
166 struct ipq *qp = container_of(q, struct ipq, q);
Gao feng54db0cc2012-06-08 01:21:40 +0000167 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
168 frags);
169 struct net *net = container_of(ipv4, struct net, ipv4);
170
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700171 struct ip4_create_arg *arg = a;
172
173 qp->protocol = arg->iph->protocol;
174 qp->id = arg->iph->id;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000175 qp->ecn = ip4_frag_ecn(arg->iph->tos);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700176 qp->saddr = arg->iph->saddr;
177 qp->daddr = arg->iph->daddr;
178 qp->user = arg->user;
179 qp->peer = sysctl_ipfrag_max_dist ?
David S. Millerc0efc882012-06-09 19:12:36 -0700180 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700181}
182
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700183static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700185 struct ipq *qp;
186
187 qp = container_of(q, struct ipq, q);
188 if (qp->peer)
189 inet_putpeer(qp->peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
193/* Destruction primitives. */
194
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700195static __inline__ void ipq_put(struct ipq *ipq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700197 inet_frag_put(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198}
199
200/* Kill ipq entry. It is not destroyed immediately,
201 * because caller (and someone more) holds reference count.
202 */
203static void ipq_kill(struct ipq *ipq)
204{
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700205 inet_frag_kill(&ipq->q, &ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900208/* Memory limiting on fragments. Evictor trashes the oldest
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 * fragment queue until we are back under the threshold.
210 */
Pavel Emelyanov6ddc0822008-01-22 06:07:25 -0800211static void ip_evictor(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700213 int evicted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Amerigo Wang6b102862012-09-18 16:50:11 +0000215 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
Pavel Emelyanov8e7999c2007-10-15 02:40:06 -0700216 if (evicted)
Pavel Emelyanovc5346fe2008-07-16 20:20:33 -0700217 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
220/*
221 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
222 */
223static void ip_expire(unsigned long arg)
224{
Pavel Emelyanove521db92007-10-17 19:45:23 -0700225 struct ipq *qp;
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700226 struct net *net;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700227
228 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
Pavel Emelyanov84a3aa02008-07-16 20:19:08 -0700229 net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700231 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Joe Perchesbc578a52008-03-28 16:35:27 -0700233 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 goto out;
235
236 ipq_kill(qp);
237
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700238 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
239 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Joe Perchesbc578a52008-03-28 16:35:27 -0700241 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700242 struct sk_buff *head = qp->q.fragments;
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000243 const struct iphdr *iph;
244 int err;
Denis V. Lunevcb846632008-03-24 15:31:00 -0700245
Eric Dumazet69df9d52009-11-05 20:59:47 -0800246 rcu_read_lock();
247 head->dev = dev_get_by_index_rcu(net, qp->iif);
Shan Weie9017b52010-01-23 01:57:42 -0800248 if (!head->dev)
249 goto out_rcu_unlock;
250
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000251 /* skb dst is stale, drop it, and perform route lookup again */
252 skb_dst_drop(head);
253 iph = ip_hdr(head);
David S. Millerc6cffba2012-07-26 11:14:38 +0000254 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
255 iph->tos, head->dev);
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000256 if (err)
257 goto out_rcu_unlock;
258
Shan Weie9017b52010-01-23 01:57:42 -0800259 /*
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000260 * Only an end host needs to send an ICMP
261 * "Fragment Reassembly Timeout" message, per RFC792.
Shan Weie9017b52010-01-23 01:57:42 -0800262 */
David S. Miller595fc712011-07-05 01:05:48 -0700263 if (qp->user == IP_DEFRAG_AF_PACKET ||
264 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
265 skb_rtable(head)->rt_type != RTN_LOCAL))
Eric Dumazet64f3b9e2011-05-04 10:02:26 +0000266 goto out_rcu_unlock;
Shan Weie9017b52010-01-23 01:57:42 -0800267
Shan Weie9017b52010-01-23 01:57:42 -0800268
269 /* Send an ICMP "Fragment Reassembly Timeout" message. */
270 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
Shan Weie9017b52010-01-23 01:57:42 -0800271out_rcu_unlock:
Patrick McHardyd1c9ae62010-02-02 11:46:50 -0800272 rcu_read_unlock();
273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274out:
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700275 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700276 ipq_put(qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700279/* Find the correct entry in the "incomplete datagrams" queue for
280 * this IP datagram, and create new one, if nothing is found.
281 */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800282static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700284 struct inet_frag_queue *q;
285 struct ip4_create_arg arg;
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700286 unsigned int hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700288 arg.iph = iph;
289 arg.user = user;
Pavel Emelyanov9a375802008-06-27 20:06:08 -0700290
291 read_lock(&ip4_frags.lock);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700292 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700293
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800294 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700295 if (q == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 goto out_nomem;
297
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700298 return container_of(q, struct ipq, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300out_nomem:
Joe Perchesafd465032012-03-12 07:03:32 +0000301 LIMIT_NETDEBUG(KERN_ERR pr_fmt("ip_frag_create: no memory left !\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 return NULL;
303}
304
Herbert Xu89cee8b2005-12-13 23:14:27 -0800305/* Is the fragment too far ahead to be part of ipq? */
306static inline int ip_frag_too_far(struct ipq *qp)
307{
308 struct inet_peer *peer = qp->peer;
309 unsigned int max = sysctl_ipfrag_max_dist;
310 unsigned int start, end;
311
312 int rc;
313
314 if (!peer || !max)
315 return 0;
316
317 start = qp->rid;
318 end = atomic_inc_return(&peer->rid);
319 qp->rid = end;
320
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700321 rc = qp->q.fragments && (end - start) > max;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800322
323 if (rc) {
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700324 struct net *net;
325
326 net = container_of(qp->q.net, struct net, ipv4.frags);
327 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800328 }
329
330 return rc;
331}
332
333static int ip_frag_reinit(struct ipq *qp)
334{
335 struct sk_buff *fp;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000336 unsigned int sum_truesize = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800337
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800338 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700339 atomic_inc(&qp->q.refcnt);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800340 return -ETIMEDOUT;
341 }
342
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700343 fp = qp->q.fragments;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800344 do {
345 struct sk_buff *xp = fp->next;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000346
347 sum_truesize += fp->truesize;
348 kfree_skb(fp);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800349 fp = xp;
350 } while (fp);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000351 sub_frag_mem_limit(&qp->q, sum_truesize);
Herbert Xu89cee8b2005-12-13 23:14:27 -0800352
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700353 qp->q.last_in = 0;
354 qp->q.len = 0;
355 qp->q.meat = 0;
356 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000357 qp->q.fragments_tail = NULL;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800358 qp->iif = 0;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000359 qp->ecn = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800360
361 return 0;
362}
363
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364/* Add new segment to existing queue. */
Herbert Xu1706d582007-10-14 00:38:15 -0700365static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
367 struct sk_buff *prev, *next;
Herbert Xu1706d582007-10-14 00:38:15 -0700368 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 int flags, offset;
370 int ihl, end;
Herbert Xu1706d582007-10-14 00:38:15 -0700371 int err = -ENOENT;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000372 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Joe Perchesbc578a52008-03-28 16:35:27 -0700374 if (qp->q.last_in & INET_FRAG_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 goto err;
376
Herbert Xu89cee8b2005-12-13 23:14:27 -0800377 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
Herbert Xu1706d582007-10-14 00:38:15 -0700378 unlikely(ip_frag_too_far(qp)) &&
379 unlikely(err = ip_frag_reinit(qp))) {
Herbert Xu89cee8b2005-12-13 23:14:27 -0800380 ipq_kill(qp);
381 goto err;
382 }
383
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000384 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700385 offset = ntohs(ip_hdr(skb)->frag_off);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 flags = offset & ~IP_OFFSET;
387 offset &= IP_OFFSET;
388 offset <<= 3; /* offset is in 8-byte chunks */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300389 ihl = ip_hdrlen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
391 /* Determine the position of this fragment. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900392 end = offset + skb->len - ihl;
Herbert Xu1706d582007-10-14 00:38:15 -0700393 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 /* Is this the final fragment? */
396 if ((flags & IP_MF) == 0) {
397 /* If we already have some bits beyond end
Justin P. Mattock42b2aa82011-11-28 20:31:00 -0800398 * or have different end, the segment is corrupted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700400 if (end < qp->q.len ||
Joe Perchesbc578a52008-03-28 16:35:27 -0700401 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 goto err;
Joe Perchesbc578a52008-03-28 16:35:27 -0700403 qp->q.last_in |= INET_FRAG_LAST_IN;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700404 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 } else {
406 if (end&7) {
407 end &= ~7;
408 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
409 skb->ip_summed = CHECKSUM_NONE;
410 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700411 if (end > qp->q.len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /* Some bits beyond end -> corruption. */
Joe Perchesbc578a52008-03-28 16:35:27 -0700413 if (qp->q.last_in & INET_FRAG_LAST_IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 goto err;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700415 qp->q.len = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
417 }
418 if (end == offset)
419 goto err;
420
Herbert Xu1706d582007-10-14 00:38:15 -0700421 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 if (pskb_pull(skb, ihl) == NULL)
423 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700424
425 err = pskb_trim_rcsum(skb, end - offset);
426 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 goto err;
428
429 /* Find out which fragments are in front and at the back of us
430 * in the chain of fragments so far. We must know where to put
431 * this fragment, right?
432 */
Changli Gaod6bebca2010-06-29 04:39:37 +0000433 prev = qp->q.fragments_tail;
434 if (!prev || FRAG_CB(prev)->offset < offset) {
435 next = NULL;
436 goto found;
437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 prev = NULL;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700439 for (next = qp->q.fragments; next != NULL; next = next->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (FRAG_CB(next)->offset >= offset)
441 break; /* bingo! */
442 prev = next;
443 }
444
Changli Gaod6bebca2010-06-29 04:39:37 +0000445found:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 /* We found where to put this one. Check for overlap with
447 * preceding fragment, and, if needed, align things so that
448 * any overlaps are eliminated.
449 */
450 if (prev) {
451 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
452
453 if (i > 0) {
454 offset += i;
Herbert Xu1706d582007-10-14 00:38:15 -0700455 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 if (end <= offset)
457 goto err;
Herbert Xu1706d582007-10-14 00:38:15 -0700458 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (!pskb_pull(skb, i))
460 goto err;
461 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
462 skb->ip_summed = CHECKSUM_NONE;
463 }
464 }
465
Herbert Xu1706d582007-10-14 00:38:15 -0700466 err = -ENOMEM;
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 while (next && FRAG_CB(next)->offset < end) {
469 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
470
471 if (i < next->len) {
472 /* Eat head of the next overlapped fragment
473 * and leave the loop. The next ones cannot overlap.
474 */
475 if (!pskb_pull(next, i))
476 goto err;
477 FRAG_CB(next)->offset += i;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700478 qp->q.meat -= i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (next->ip_summed != CHECKSUM_UNNECESSARY)
480 next->ip_summed = CHECKSUM_NONE;
481 break;
482 } else {
483 struct sk_buff *free_it = next;
484
Peter Zijlstra47c6bf772006-12-12 19:48:59 +0100485 /* Old fragment is completely overridden with
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 * new one drop it.
487 */
488 next = next->next;
489
490 if (prev)
491 prev->next = next;
492 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700493 qp->q.fragments = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700495 qp->q.meat -= free_it->len;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000496 sub_frag_mem_limit(&qp->q, free_it->truesize);
497 kfree_skb(free_it);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 }
499 }
500
501 FRAG_CB(skb)->offset = offset;
502
503 /* Insert this fragment in the chain of fragments. */
504 skb->next = next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000505 if (!next)
506 qp->q.fragments_tail = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (prev)
508 prev->next = skb;
509 else
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700510 qp->q.fragments = skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Herbert Xu1706d582007-10-14 00:38:15 -0700512 dev = skb->dev;
513 if (dev) {
514 qp->iif = dev->ifindex;
515 skb->dev = NULL;
516 }
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700517 qp->q.stamp = skb->tstamp;
518 qp->q.meat += skb->len;
Eric Dumazet6623e3b2011-01-05 07:52:55 +0000519 qp->ecn |= ecn;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000520 add_frag_mem_limit(&qp->q, skb->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if (offset == 0)
Joe Perchesbc578a52008-03-28 16:35:27 -0700522 qp->q.last_in |= INET_FRAG_FIRST_IN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200524 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
525 skb->len + ihl > qp->q.max_size)
526 qp->q.max_size = skb->len + ihl;
527
Joe Perchesbc578a52008-03-28 16:35:27 -0700528 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
529 qp->q.meat == qp->q.len)
Herbert Xu1706d582007-10-14 00:38:15 -0700530 return ip_frag_reasm(qp, prev, dev);
531
Jesper Dangaard Brouer3ef0eb02013-01-28 23:45:51 +0000532 inet_frag_lru_move(&qp->q);
Herbert Xu1706d582007-10-14 00:38:15 -0700533 return -EINPROGRESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535err:
536 kfree_skb(skb);
Herbert Xu1706d582007-10-14 00:38:15 -0700537 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538}
539
540
541/* Build a new IP datagram from all its fragments. */
542
Herbert Xu1706d582007-10-14 00:38:15 -0700543static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
544 struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700546 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 struct iphdr *iph;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700548 struct sk_buff *fp, *head = qp->q.fragments;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 int len;
550 int ihlen;
Herbert Xu1706d582007-10-14 00:38:15 -0700551 int err;
Eric Dumazet3cc49492012-05-19 03:02:20 +0000552 int sum_truesize;
Eric Dumazet5173cc02011-05-16 08:37:37 +0000553 u8 ecn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 ipq_kill(qp);
556
Eric Dumazet5173cc02011-05-16 08:37:37 +0000557 ecn = ip4_frag_ecn_table[qp->ecn];
558 if (unlikely(ecn == 0xff)) {
559 err = -EINVAL;
560 goto out_fail;
561 }
Herbert Xu1706d582007-10-14 00:38:15 -0700562 /* Make the one we just received the head. */
563 if (prev) {
564 head = prev->next;
565 fp = skb_clone(head, GFP_ATOMIC);
Herbert Xu1706d582007-10-14 00:38:15 -0700566 if (!fp)
567 goto out_nomem;
568
569 fp->next = head->next;
Changli Gaod6bebca2010-06-29 04:39:37 +0000570 if (!fp->next)
571 qp->q.fragments_tail = fp;
Herbert Xu1706d582007-10-14 00:38:15 -0700572 prev->next = fp;
573
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700574 skb_morph(head, qp->q.fragments);
575 head->next = qp->q.fragments->next;
Herbert Xu1706d582007-10-14 00:38:15 -0700576
Eric Dumazetcbf8f7b2012-04-19 06:10:26 +0000577 consume_skb(qp->q.fragments);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700578 qp->q.fragments = head;
Herbert Xu1706d582007-10-14 00:38:15 -0700579 }
580
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700581 WARN_ON(head == NULL);
582 WARN_ON(FRAG_CB(head)->offset != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 /* Allocate a new buffer for the datagram. */
Arnaldo Carvalho de Meloc9bdd4b2007-03-12 20:09:15 -0300585 ihlen = ip_hdrlen(head);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700586 len = ihlen + qp->q.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Herbert Xu1706d582007-10-14 00:38:15 -0700588 err = -E2BIG;
Stephen Hemminger132adf52007-03-08 20:44:43 -0800589 if (len > 65535)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 goto out_oversize;
591
592 /* Head of list must not be cloned. */
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +0000593 if (skb_unclone(head, GFP_ATOMIC))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 goto out_nomem;
595
596 /* If the first fragment is fragmented itself, we split
597 * it to two chunks: the first with data and paged part
598 * and the second, holding only fragments. */
David S. Miller21dc3302010-08-23 00:13:46 -0700599 if (skb_has_frag_list(head)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 struct sk_buff *clone;
601 int i, plen = 0;
602
603 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
604 goto out_nomem;
605 clone->next = head->next;
606 head->next = clone;
607 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700608 skb_frag_list_init(head);
Eric Dumazet9e903e02011-10-18 21:00:24 +0000609 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
610 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 clone->len = clone->data_len = head->data_len - plen;
612 head->data_len -= clone->len;
613 head->len -= clone->len;
614 clone->csum = 0;
615 clone->ip_summed = head->ip_summed;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000616 add_frag_mem_limit(&qp->q, clone->truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700619 skb_push(head, head->data - skb_network_header(head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Eric Dumazet3cc49492012-05-19 03:02:20 +0000621 sum_truesize = head->truesize;
622 for (fp = head->next; fp;) {
623 bool headstolen;
624 int delta;
625 struct sk_buff *next = fp->next;
626
627 sum_truesize += fp->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (head->ip_summed != fp->ip_summed)
629 head->ip_summed = CHECKSUM_NONE;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700630 else if (head->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 head->csum = csum_add(head->csum, fp->csum);
Eric Dumazet3cc49492012-05-19 03:02:20 +0000632
633 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
634 kfree_skb_partial(fp, headstolen);
635 } else {
636 if (!skb_shinfo(head)->frag_list)
637 skb_shinfo(head)->frag_list = fp;
638 head->data_len += fp->len;
639 head->len += fp->len;
640 head->truesize += fp->truesize;
641 }
642 fp = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 }
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000644 sub_frag_mem_limit(&qp->q, sum_truesize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 head->next = NULL;
647 head->dev = dev;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700648 head->tstamp = qp->q.stamp;
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200649 IPCB(head)->frag_max_size = qp->q.max_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700651 iph = ip_hdr(head);
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200652 /* max_size != 0 implies at least one fragment had IP_DF set */
653 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 iph->tot_len = htons(len);
Eric Dumazet5173cc02011-05-16 08:37:37 +0000655 iph->tos |= ecn;
Jorge Boncompte [DTI2]2bad35b2009-03-18 23:26:11 -0700656 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700657 qp->q.fragments = NULL;
Changli Gaod6bebca2010-06-29 04:39:37 +0000658 qp->q.fragments_tail = NULL;
Herbert Xu1706d582007-10-14 00:38:15 -0700659 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661out_nomem:
Joe Perchesafd465032012-03-12 07:03:32 +0000662 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
663 qp);
David Howells45542472007-10-17 21:37:22 -0700664 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 goto out_fail;
666out_oversize:
Joe Perchese87cc472012-05-13 21:56:26 +0000667 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668out_fail:
David Fordbbf31bf2009-11-29 23:02:22 -0800669 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Herbert Xu1706d582007-10-14 00:38:15 -0700670 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671}
672
673/* Process an incoming IP datagram fragment. */
Herbert Xu776c7292007-10-14 00:38:32 -0700674int ip_defrag(struct sk_buff *skb, u32 user)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 struct ipq *qp;
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800677 struct net *net;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900678
Eric Dumazetadf30902009-06-02 05:19:30 +0000679 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700680 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 /* Start by cleaning up the memory. */
Amerigo Wang6b102862012-09-18 16:50:11 +0000683 ip_evictor(net);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 /* Lookup (or create) queue header */
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800686 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
Herbert Xu1706d582007-10-14 00:38:15 -0700687 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700689 spin_lock(&qp->q.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Herbert Xu1706d582007-10-14 00:38:15 -0700691 ret = ip_frag_queue(qp, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700693 spin_unlock(&qp->q.lock);
Pavel Emelyanov4b6cb5d2007-10-15 02:41:09 -0700694 ipq_put(qp);
Herbert Xu776c7292007-10-14 00:38:32 -0700695 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
697
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700698 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 kfree_skb(skb);
Herbert Xu776c7292007-10-14 00:38:32 -0700700 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000702EXPORT_SYMBOL(ip_defrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Eric Dumazetbc416d92011-10-06 10:28:31 +0000704struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
705{
Johannes Berg1bf37512012-12-09 23:41:06 +0000706 struct iphdr iph;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000707 u32 len;
708
709 if (skb->protocol != htons(ETH_P_IP))
710 return skb;
711
Johannes Berg1bf37512012-12-09 23:41:06 +0000712 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
Eric Dumazetbc416d92011-10-06 10:28:31 +0000713 return skb;
714
Johannes Berg1bf37512012-12-09 23:41:06 +0000715 if (iph.ihl < 5 || iph.version != 4)
Eric Dumazetbc416d92011-10-06 10:28:31 +0000716 return skb;
717
Johannes Berg1bf37512012-12-09 23:41:06 +0000718 len = ntohs(iph.tot_len);
719 if (skb->len < len || len < (iph.ihl * 4))
720 return skb;
721
722 if (ip_is_fragment(&iph)) {
Eric Dumazetbc416d92011-10-06 10:28:31 +0000723 skb = skb_share_check(skb, GFP_ATOMIC);
724 if (skb) {
Johannes Berg1bf37512012-12-09 23:41:06 +0000725 if (!pskb_may_pull(skb, iph.ihl*4))
726 return skb;
Eric Dumazetbc416d92011-10-06 10:28:31 +0000727 if (pskb_trim_rcsum(skb, len))
728 return skb;
729 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
730 if (ip_defrag(skb, user))
731 return NULL;
732 skb->rxhash = 0;
733 }
734 }
735 return skb;
736}
737EXPORT_SYMBOL(ip_check_defrag);
738
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800739#ifdef CONFIG_SYSCTL
740static int zero;
741
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700742static struct ctl_table ip4_frags_ns_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800743 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800744 .procname = "ipfrag_high_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800745 .data = &init_net.ipv4.frags.high_thresh,
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
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800749 },
750 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800751 .procname = "ipfrag_low_thresh",
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800752 .data = &init_net.ipv4.frags.low_thresh,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800753 .maxlen = sizeof(int),
754 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800755 .proc_handler = proc_dointvec
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800756 },
757 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800758 .procname = "ipfrag_time",
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800759 .data = &init_net.ipv4.frags.timeout,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800760 .maxlen = sizeof(int),
761 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800762 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800763 },
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700764 { }
765};
766
767static struct ctl_table ip4_frags_ctl_table[] = {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800768 {
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800769 .procname = "ipfrag_secret_interval",
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800770 .data = &ip4_frags.secret_interval,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800771 .maxlen = sizeof(int),
772 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800773 .proc_handler = proc_dointvec_jiffies,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800774 },
775 {
776 .procname = "ipfrag_max_dist",
777 .data = &sysctl_ipfrag_max_dist,
778 .maxlen = sizeof(int),
779 .mode = 0644,
Alexey Dobriyan6d9f2392008-11-03 18:21:05 -0800780 .proc_handler = proc_dointvec_minmax,
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800781 .extra1 = &zero
782 },
783 { }
784};
785
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000786static int __net_init ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800787{
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800788 struct ctl_table *table;
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800789 struct ctl_table_header *hdr;
790
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700791 table = ip4_frags_ns_ctl_table;
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800792 if (!net_eq(net, &init_net)) {
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700793 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800794 if (table == NULL)
795 goto err_alloc;
796
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800797 table[0].data = &net->ipv4.frags.high_thresh;
798 table[1].data = &net->ipv4.frags.low_thresh;
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800799 table[2].data = &net->ipv4.frags.timeout;
Eric W. Biederman464dc802012-11-16 03:02:59 +0000800
801 /* Don't export sysctls to unprivileged users */
802 if (net->user_ns != &init_user_ns)
803 table[0].procname = NULL;
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800804 }
805
Eric W. Biedermanec8f23c2012-04-19 13:44:49 +0000806 hdr = register_net_sysctl(net, "net/ipv4", table);
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800807 if (hdr == NULL)
808 goto err_reg;
809
810 net->ipv4.frags_hdr = hdr;
811 return 0;
812
813err_reg:
Octavian Purdila09ad9bc2009-11-25 15:14:13 -0800814 if (!net_eq(net, &init_net))
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800815 kfree(table);
816err_alloc:
817 return -ENOMEM;
818}
819
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000820static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800821{
822 struct ctl_table *table;
823
824 table = net->ipv4.frags_hdr->ctl_table_arg;
825 unregister_net_sysctl_table(net->ipv4.frags_hdr);
826 kfree(table);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800827}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700828
829static void ip4_frags_ctl_register(void)
830{
Eric W. Biederman43444752012-04-19 13:22:55 +0000831 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700832}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800833#else
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700834static inline int ip4_frags_ns_ctl_register(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800835{
836 return 0;
837}
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800838
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700839static inline void ip4_frags_ns_ctl_unregister(struct net *net)
Pavel Emelyanove4a2d5c2008-01-22 06:08:36 -0800840{
841}
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700842
843static inline void ip4_frags_ctl_register(void)
844{
845}
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800846#endif
847
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000848static int __net_init ipv4_frags_init_net(struct net *net)
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800849{
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000850 /* Fragment cache limits.
851 *
852 * The fragment memory accounting code, (tries to) account for
853 * the real memory usage, by measuring both the size of frag
854 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
855 * and the SKB's truesize.
856 *
857 * A 64K fragment consumes 129736 bytes (44*2944)+200
858 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
859 *
860 * We will commit 4MB at one time. Should we cross that limit
861 * we will prune down to 3MB, making room for approx 8 big 64K
862 * fragments 8x128k.
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800863 */
Jesper Dangaard Brouerc2a93662013-01-15 07:16:35 +0000864 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
865 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -0800866 /*
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -0800867 * Important NOTE! Fragment queue must be destroyed before MSL expires.
868 * RFC791 is wrong proposing to prolongate timer each fragment arrival
869 * by TTL.
870 */
871 net->ipv4.frags.timeout = IP_FRAG_TIME;
872
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800873 inet_frags_init_net(&net->ipv4.frags);
874
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700875 return ip4_frags_ns_ctl_register(net);
Pavel Emelyanov8d8354d2008-01-22 05:58:31 -0800876}
877
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +0000878static void __net_exit ipv4_frags_exit_net(struct net *net)
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800879{
Pavel Emelyanov0a64b4b2008-05-19 13:51:29 -0700880 ip4_frags_ns_ctl_unregister(net);
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800881 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
882}
883
884static struct pernet_operations ip4_frags_ops = {
885 .init = ipv4_frags_init_net,
886 .exit = ipv4_frags_exit_net,
887};
888
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -0700889void __init ipfrag_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
Pavel Emelyanov7d291eb2008-05-19 13:53:02 -0700891 ip4_frags_ctl_register();
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800892 register_pernet_subsys(&ip4_frags_ops);
Pavel Emelyanov321a3a92007-10-15 02:38:08 -0700893 ip4_frags.hashfn = ip4_hashfn;
Pavel Emelyanovc6fda282007-10-17 19:46:47 -0700894 ip4_frags.constructor = ip4_frag_init;
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700895 ip4_frags.destructor = ip4_frag_free;
896 ip4_frags.skb_free = NULL;
897 ip4_frags.qsize = sizeof(struct ipq);
Pavel Emelyanovabd65232007-10-17 19:47:21 -0700898 ip4_frags.match = ip4_frag_match;
Pavel Emelyanove521db92007-10-17 19:45:23 -0700899 ip4_frags.frag_expire = ip_expire;
Pavel Emelyanov3b4bc4a2008-01-22 06:11:04 -0800900 ip4_frags.secret_interval = 10 * 60 * HZ;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700901 inet_frags_init(&ip4_frags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902}