]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/ipv4/ip_fragment.c
be2net: FW download for Lancer
[mirror_ubuntu-bionic-kernel.git] / net / ipv4 / ip_fragment.c
CommitLineData
1da177e4
LT
1/*
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.
e905a9ed 7 *
1da177e4 8 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
113aa838 9 * Alan Cox <alan@lxorguk.ukuu.org.uk>
1da177e4
LT
10 *
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
89cee8b1 23#include <linux/compiler.h>
1da177e4
LT
24#include <linux/module.h>
25#include <linux/types.h>
26#include <linux/mm.h>
27#include <linux/jiffies.h>
28#include <linux/skbuff.h>
29#include <linux/list.h>
30#include <linux/ip.h>
31#include <linux/icmp.h>
32#include <linux/netdevice.h>
33#include <linux/jhash.h>
34#include <linux/random.h>
5a0e3ad6 35#include <linux/slab.h>
e9017b55
SW
36#include <net/route.h>
37#include <net/dst.h>
1da177e4
LT
38#include <net/sock.h>
39#include <net/ip.h>
40#include <net/icmp.h>
41#include <net/checksum.h>
89cee8b1 42#include <net/inetpeer.h>
5ab11c98 43#include <net/inet_frag.h>
1da177e4
LT
44#include <linux/tcp.h>
45#include <linux/udp.h>
46#include <linux/inet.h>
47#include <linux/netfilter_ipv4.h>
6623e3b2 48#include <net/inet_ecn.h>
1da177e4
LT
49
50/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
51 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
52 * as well. Or notify me, at least. --ANK
53 */
54
8d8354d2 55static int sysctl_ipfrag_max_dist __read_mostly = 64;
89cee8b1 56
1da177e4
LT
57struct ipfrag_skb_cb
58{
59 struct inet_skb_parm h;
60 int offset;
61};
62
fd3f8c4c 63#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
1da177e4
LT
64
65/* Describe an entry in the "incomplete datagrams" queue. */
66struct ipq {
5ab11c98
PE
67 struct inet_frag_queue q;
68
1da177e4 69 u32 user;
18277770
AV
70 __be32 saddr;
71 __be32 daddr;
72 __be16 id;
1da177e4 73 u8 protocol;
6623e3b2 74 u8 ecn; /* RFC3168 support */
89cee8b1
HX
75 int iif;
76 unsigned int rid;
77 struct inet_peer *peer;
1da177e4
LT
78};
79
6623e3b2
ED
80#define IPFRAG_ECN_CLEAR 0x01 /* one frag had INET_ECN_NOT_ECT */
81#define IPFRAG_ECN_SET_CE 0x04 /* one frag had INET_ECN_CE */
82
83static inline u8 ip4_frag_ecn(u8 tos)
84{
85 tos = (tos & INET_ECN_MASK) + 1;
86 /*
87 * After the last operation we have (in binary):
88 * INET_ECN_NOT_ECT => 001
89 * INET_ECN_ECT_1 => 010
90 * INET_ECN_ECT_0 => 011
91 * INET_ECN_CE => 100
92 */
93 return (tos & 2) ? 0 : tos;
94}
95
7eb95156 96static struct inet_frags ip4_frags;
1da177e4 97
e5a2bb84 98int ip_frag_nqueues(struct net *net)
7eb95156 99{
e5a2bb84 100 return net->ipv4.frags.nqueues;
7eb95156 101}
1da177e4 102
6ddc0822 103int ip_frag_mem(struct net *net)
7eb95156 104{
6ddc0822 105 return atomic_read(&net->ipv4.frags.mem);
7eb95156 106}
1da177e4 107
1706d587
HX
108static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
109 struct net_device *dev);
110
c6fda282
PE
111struct ip4_create_arg {
112 struct iphdr *iph;
113 u32 user;
114};
115
18277770 116static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
1da177e4 117{
18277770
AV
118 return jhash_3words((__force u32)id << 16 | prot,
119 (__force u32)saddr, (__force u32)daddr,
7eb95156 120 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
1da177e4
LT
121}
122
321a3a99 123static unsigned int ip4_hashfn(struct inet_frag_queue *q)
1da177e4 124{
321a3a99 125 struct ipq *ipq;
1da177e4 126
321a3a99
PE
127 ipq = container_of(q, struct ipq, q);
128 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
1da177e4
LT
129}
130
abd6523d
PE
131static int ip4_frag_match(struct inet_frag_queue *q, void *a)
132{
133 struct ipq *qp;
134 struct ip4_create_arg *arg = a;
135
136 qp = container_of(q, struct ipq, q);
a02cec21 137 return qp->id == arg->iph->id &&
abd6523d
PE
138 qp->saddr == arg->iph->saddr &&
139 qp->daddr == arg->iph->daddr &&
140 qp->protocol == arg->iph->protocol &&
a02cec21 141 qp->user == arg->user;
abd6523d
PE
142}
143
1da177e4 144/* Memory Tracking Functions. */
a95d8c88 145static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
1da177e4 146{
6ddc0822 147 atomic_sub(skb->truesize, &nf->mem);
1da177e4
LT
148 kfree_skb(skb);
149}
150
c6fda282
PE
151static void ip4_frag_init(struct inet_frag_queue *q, void *a)
152{
153 struct ipq *qp = container_of(q, struct ipq, q);
154 struct ip4_create_arg *arg = a;
155
156 qp->protocol = arg->iph->protocol;
157 qp->id = arg->iph->id;
6623e3b2 158 qp->ecn = ip4_frag_ecn(arg->iph->tos);
c6fda282
PE
159 qp->saddr = arg->iph->saddr;
160 qp->daddr = arg->iph->daddr;
161 qp->user = arg->user;
162 qp->peer = sysctl_ipfrag_max_dist ?
b534ecf1 163 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
c6fda282
PE
164}
165
1e4b8287 166static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
1da177e4 167{
1e4b8287
PE
168 struct ipq *qp;
169
170 qp = container_of(q, struct ipq, q);
171 if (qp->peer)
172 inet_putpeer(qp->peer);
1da177e4
LT
173}
174
1da177e4
LT
175
176/* Destruction primitives. */
177
4b6cb5d8 178static __inline__ void ipq_put(struct ipq *ipq)
1da177e4 179{
762cc408 180 inet_frag_put(&ipq->q, &ip4_frags);
1da177e4
LT
181}
182
183/* Kill ipq entry. It is not destroyed immediately,
184 * because caller (and someone more) holds reference count.
185 */
186static void ipq_kill(struct ipq *ipq)
187{
277e650d 188 inet_frag_kill(&ipq->q, &ip4_frags);
1da177e4
LT
189}
190
e905a9ed 191/* Memory limiting on fragments. Evictor trashes the oldest
1da177e4
LT
192 * fragment queue until we are back under the threshold.
193 */
6ddc0822 194static void ip_evictor(struct net *net)
1da177e4 195{
8e7999c4
PE
196 int evicted;
197
6ddc0822 198 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
8e7999c4 199 if (evicted)
c5346fe3 200 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
1da177e4
LT
201}
202
203/*
204 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
205 */
206static void ip_expire(unsigned long arg)
207{
e521db9d 208 struct ipq *qp;
84a3aa00 209 struct net *net;
e521db9d
PE
210
211 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
84a3aa00 212 net = container_of(qp->q.net, struct net, ipv4.frags);
1da177e4 213
5ab11c98 214 spin_lock(&qp->q.lock);
1da177e4 215
bc578a54 216 if (qp->q.last_in & INET_FRAG_COMPLETE)
1da177e4
LT
217 goto out;
218
219 ipq_kill(qp);
220
7c73a6fa
PE
221 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
222 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
1da177e4 223
bc578a54 224 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
5ab11c98 225 struct sk_buff *head = qp->q.fragments;
64f3b9e2
ED
226 const struct iphdr *iph;
227 int err;
cb84663e 228
69df9d59
ED
229 rcu_read_lock();
230 head->dev = dev_get_by_index_rcu(net, qp->iif);
e9017b55
SW
231 if (!head->dev)
232 goto out_rcu_unlock;
233
64f3b9e2
ED
234 /* skb dst is stale, drop it, and perform route lookup again */
235 skb_dst_drop(head);
236 iph = ip_hdr(head);
237 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
238 iph->tos, head->dev);
239 if (err)
240 goto out_rcu_unlock;
241
e9017b55 242 /*
64f3b9e2
ED
243 * Only an end host needs to send an ICMP
244 * "Fragment Reassembly Timeout" message, per RFC792.
e9017b55 245 */
64f3b9e2
ED
246 if (qp->user == IP_DEFRAG_CONNTRACK_IN &&
247 skb_rtable(head)->rt_type != RTN_LOCAL)
248 goto out_rcu_unlock;
249
e9017b55
SW
250
251 /* Send an ICMP "Fragment Reassembly Timeout" message. */
252 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
e9017b55 253out_rcu_unlock:
d1c9ae6d
PM
254 rcu_read_unlock();
255 }
1da177e4 256out:
5ab11c98 257 spin_unlock(&qp->q.lock);
4b6cb5d8 258 ipq_put(qp);
1da177e4
LT
259}
260
abd6523d
PE
261/* Find the correct entry in the "incomplete datagrams" queue for
262 * this IP datagram, and create new one, if nothing is found.
263 */
ac18e750 264static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
1da177e4 265{
c6fda282
PE
266 struct inet_frag_queue *q;
267 struct ip4_create_arg arg;
abd6523d 268 unsigned int hash;
1da177e4 269
c6fda282
PE
270 arg.iph = iph;
271 arg.user = user;
9a375803
PE
272
273 read_lock(&ip4_frags.lock);
abd6523d 274 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
1da177e4 275
ac18e750 276 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
c6fda282
PE
277 if (q == NULL)
278 goto out_nomem;
1da177e4 279
c6fda282 280 return container_of(q, struct ipq, q);
1da177e4
LT
281
282out_nomem:
64ce2073 283 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
1da177e4
LT
284 return NULL;
285}
286
89cee8b1
HX
287/* Is the fragment too far ahead to be part of ipq? */
288static inline int ip_frag_too_far(struct ipq *qp)
289{
290 struct inet_peer *peer = qp->peer;
291 unsigned int max = sysctl_ipfrag_max_dist;
292 unsigned int start, end;
293
294 int rc;
295
296 if (!peer || !max)
297 return 0;
298
299 start = qp->rid;
300 end = atomic_inc_return(&peer->rid);
301 qp->rid = end;
302
5ab11c98 303 rc = qp->q.fragments && (end - start) > max;
89cee8b1
HX
304
305 if (rc) {
7c73a6fa
PE
306 struct net *net;
307
308 net = container_of(qp->q.net, struct net, ipv4.frags);
309 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
89cee8b1
HX
310 }
311
312 return rc;
313}
314
315static int ip_frag_reinit(struct ipq *qp)
316{
317 struct sk_buff *fp;
318
b2fd5321 319 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
5ab11c98 320 atomic_inc(&qp->q.refcnt);
89cee8b1
HX
321 return -ETIMEDOUT;
322 }
323
5ab11c98 324 fp = qp->q.fragments;
89cee8b1
HX
325 do {
326 struct sk_buff *xp = fp->next;
a95d8c88 327 frag_kfree_skb(qp->q.net, fp);
89cee8b1
HX
328 fp = xp;
329 } while (fp);
330
5ab11c98
PE
331 qp->q.last_in = 0;
332 qp->q.len = 0;
333 qp->q.meat = 0;
334 qp->q.fragments = NULL;
d6bebca9 335 qp->q.fragments_tail = NULL;
89cee8b1 336 qp->iif = 0;
6623e3b2 337 qp->ecn = 0;
89cee8b1
HX
338
339 return 0;
340}
341
1da177e4 342/* Add new segment to existing queue. */
1706d587 343static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
1da177e4
LT
344{
345 struct sk_buff *prev, *next;
1706d587 346 struct net_device *dev;
1da177e4
LT
347 int flags, offset;
348 int ihl, end;
1706d587 349 int err = -ENOENT;
6623e3b2 350 u8 ecn;
1da177e4 351
bc578a54 352 if (qp->q.last_in & INET_FRAG_COMPLETE)
1da177e4
LT
353 goto err;
354
89cee8b1 355 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
1706d587
HX
356 unlikely(ip_frag_too_far(qp)) &&
357 unlikely(err = ip_frag_reinit(qp))) {
89cee8b1
HX
358 ipq_kill(qp);
359 goto err;
360 }
361
6623e3b2 362 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
eddc9ec5 363 offset = ntohs(ip_hdr(skb)->frag_off);
1da177e4
LT
364 flags = offset & ~IP_OFFSET;
365 offset &= IP_OFFSET;
366 offset <<= 3; /* offset is in 8-byte chunks */
c9bdd4b5 367 ihl = ip_hdrlen(skb);
1da177e4
LT
368
369 /* Determine the position of this fragment. */
e905a9ed 370 end = offset + skb->len - ihl;
1706d587 371 err = -EINVAL;
1da177e4
LT
372
373 /* Is this the final fragment? */
374 if ((flags & IP_MF) == 0) {
375 /* If we already have some bits beyond end
376 * or have different end, the segment is corrrupted.
377 */
5ab11c98 378 if (end < qp->q.len ||
bc578a54 379 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
1da177e4 380 goto err;
bc578a54 381 qp->q.last_in |= INET_FRAG_LAST_IN;
5ab11c98 382 qp->q.len = end;
1da177e4
LT
383 } else {
384 if (end&7) {
385 end &= ~7;
386 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
387 skb->ip_summed = CHECKSUM_NONE;
388 }
5ab11c98 389 if (end > qp->q.len) {
1da177e4 390 /* Some bits beyond end -> corruption. */
bc578a54 391 if (qp->q.last_in & INET_FRAG_LAST_IN)
1da177e4 392 goto err;
5ab11c98 393 qp->q.len = end;
1da177e4
LT
394 }
395 }
396 if (end == offset)
397 goto err;
398
1706d587 399 err = -ENOMEM;
1da177e4
LT
400 if (pskb_pull(skb, ihl) == NULL)
401 goto err;
1706d587
HX
402
403 err = pskb_trim_rcsum(skb, end - offset);
404 if (err)
1da177e4
LT
405 goto err;
406
407 /* Find out which fragments are in front and at the back of us
408 * in the chain of fragments so far. We must know where to put
409 * this fragment, right?
410 */
d6bebca9
CG
411 prev = qp->q.fragments_tail;
412 if (!prev || FRAG_CB(prev)->offset < offset) {
413 next = NULL;
414 goto found;
415 }
1da177e4 416 prev = NULL;
5ab11c98 417 for (next = qp->q.fragments; next != NULL; next = next->next) {
1da177e4
LT
418 if (FRAG_CB(next)->offset >= offset)
419 break; /* bingo! */
420 prev = next;
421 }
422
d6bebca9 423found:
1da177e4
LT
424 /* We found where to put this one. Check for overlap with
425 * preceding fragment, and, if needed, align things so that
426 * any overlaps are eliminated.
427 */
428 if (prev) {
429 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
430
431 if (i > 0) {
432 offset += i;
1706d587 433 err = -EINVAL;
1da177e4
LT
434 if (end <= offset)
435 goto err;
1706d587 436 err = -ENOMEM;
1da177e4
LT
437 if (!pskb_pull(skb, i))
438 goto err;
439 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
440 skb->ip_summed = CHECKSUM_NONE;
441 }
442 }
443
1706d587
HX
444 err = -ENOMEM;
445
1da177e4
LT
446 while (next && FRAG_CB(next)->offset < end) {
447 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
448
449 if (i < next->len) {
450 /* Eat head of the next overlapped fragment
451 * and leave the loop. The next ones cannot overlap.
452 */
453 if (!pskb_pull(next, i))
454 goto err;
455 FRAG_CB(next)->offset += i;
5ab11c98 456 qp->q.meat -= i;
1da177e4
LT
457 if (next->ip_summed != CHECKSUM_UNNECESSARY)
458 next->ip_summed = CHECKSUM_NONE;
459 break;
460 } else {
461 struct sk_buff *free_it = next;
462
47c6bf77 463 /* Old fragment is completely overridden with
1da177e4
LT
464 * new one drop it.
465 */
466 next = next->next;
467
468 if (prev)
469 prev->next = next;
470 else
5ab11c98 471 qp->q.fragments = next;
1da177e4 472
5ab11c98 473 qp->q.meat -= free_it->len;
a95d8c88 474 frag_kfree_skb(qp->q.net, free_it);
1da177e4
LT
475 }
476 }
477
478 FRAG_CB(skb)->offset = offset;
479
480 /* Insert this fragment in the chain of fragments. */
481 skb->next = next;
d6bebca9
CG
482 if (!next)
483 qp->q.fragments_tail = skb;
1da177e4
LT
484 if (prev)
485 prev->next = skb;
486 else
5ab11c98 487 qp->q.fragments = skb;
1da177e4 488
1706d587
HX
489 dev = skb->dev;
490 if (dev) {
491 qp->iif = dev->ifindex;
492 skb->dev = NULL;
493 }
5ab11c98
PE
494 qp->q.stamp = skb->tstamp;
495 qp->q.meat += skb->len;
6623e3b2 496 qp->ecn |= ecn;
6ddc0822 497 atomic_add(skb->truesize, &qp->q.net->mem);
1da177e4 498 if (offset == 0)
bc578a54 499 qp->q.last_in |= INET_FRAG_FIRST_IN;
1da177e4 500
bc578a54
JP
501 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
502 qp->q.meat == qp->q.len)
1706d587
HX
503 return ip_frag_reasm(qp, prev, dev);
504
7eb95156 505 write_lock(&ip4_frags.lock);
3140c25c 506 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
7eb95156 507 write_unlock(&ip4_frags.lock);
1706d587 508 return -EINPROGRESS;
1da177e4
LT
509
510err:
511 kfree_skb(skb);
1706d587 512 return err;
1da177e4
LT
513}
514
515
516/* Build a new IP datagram from all its fragments. */
517
1706d587
HX
518static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
519 struct net_device *dev)
1da177e4 520{
2bad35b7 521 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
1da177e4 522 struct iphdr *iph;
5ab11c98 523 struct sk_buff *fp, *head = qp->q.fragments;
1da177e4
LT
524 int len;
525 int ihlen;
1706d587 526 int err;
1da177e4
LT
527
528 ipq_kill(qp);
529
1706d587
HX
530 /* Make the one we just received the head. */
531 if (prev) {
532 head = prev->next;
533 fp = skb_clone(head, GFP_ATOMIC);
1706d587
HX
534 if (!fp)
535 goto out_nomem;
536
537 fp->next = head->next;
d6bebca9
CG
538 if (!fp->next)
539 qp->q.fragments_tail = fp;
1706d587
HX
540 prev->next = fp;
541
5ab11c98
PE
542 skb_morph(head, qp->q.fragments);
543 head->next = qp->q.fragments->next;
1706d587 544
5ab11c98
PE
545 kfree_skb(qp->q.fragments);
546 qp->q.fragments = head;
1706d587
HX
547 }
548
547b792c
IJ
549 WARN_ON(head == NULL);
550 WARN_ON(FRAG_CB(head)->offset != 0);
1da177e4
LT
551
552 /* Allocate a new buffer for the datagram. */
c9bdd4b5 553 ihlen = ip_hdrlen(head);
5ab11c98 554 len = ihlen + qp->q.len;
1da177e4 555
1706d587 556 err = -E2BIG;
132adf54 557 if (len > 65535)
1da177e4
LT
558 goto out_oversize;
559
560 /* Head of list must not be cloned. */
561 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
562 goto out_nomem;
563
564 /* If the first fragment is fragmented itself, we split
565 * it to two chunks: the first with data and paged part
566 * and the second, holding only fragments. */
21dc3301 567 if (skb_has_frag_list(head)) {
1da177e4
LT
568 struct sk_buff *clone;
569 int i, plen = 0;
570
571 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
572 goto out_nomem;
573 clone->next = head->next;
574 head->next = clone;
575 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
d7fcf1a5 576 skb_frag_list_init(head);
1da177e4
LT
577 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
578 plen += skb_shinfo(head)->frags[i].size;
579 clone->len = clone->data_len = head->data_len - plen;
580 head->data_len -= clone->len;
581 head->len -= clone->len;
582 clone->csum = 0;
583 clone->ip_summed = head->ip_summed;
6ddc0822 584 atomic_add(clone->truesize, &qp->q.net->mem);
1da177e4
LT
585 }
586
587 skb_shinfo(head)->frag_list = head->next;
d56f90a7 588 skb_push(head, head->data - skb_network_header(head));
1da177e4
LT
589
590 for (fp=head->next; fp; fp = fp->next) {
591 head->data_len += fp->len;
592 head->len += fp->len;
593 if (head->ip_summed != fp->ip_summed)
594 head->ip_summed = CHECKSUM_NONE;
84fa7933 595 else if (head->ip_summed == CHECKSUM_COMPLETE)
1da177e4
LT
596 head->csum = csum_add(head->csum, fp->csum);
597 head->truesize += fp->truesize;
1da177e4 598 }
d27f9b35 599 atomic_sub(head->truesize, &qp->q.net->mem);
1da177e4
LT
600
601 head->next = NULL;
602 head->dev = dev;
5ab11c98 603 head->tstamp = qp->q.stamp;
1da177e4 604
eddc9ec5 605 iph = ip_hdr(head);
1da177e4
LT
606 iph->frag_off = 0;
607 iph->tot_len = htons(len);
6623e3b2
ED
608 /* RFC3168 5.3 Fragmentation support
609 * If one fragment had INET_ECN_NOT_ECT,
610 * reassembled frame also has INET_ECN_NOT_ECT
611 * Elif one fragment had INET_ECN_CE
612 * reassembled frame also has INET_ECN_CE
613 */
614 if (qp->ecn & IPFRAG_ECN_CLEAR)
615 iph->tos &= ~INET_ECN_MASK;
616 else if (qp->ecn & IPFRAG_ECN_SET_CE)
617 iph->tos |= INET_ECN_CE;
618
2bad35b7 619 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
5ab11c98 620 qp->q.fragments = NULL;
d6bebca9 621 qp->q.fragments_tail = NULL;
1706d587 622 return 0;
1da177e4
LT
623
624out_nomem:
e905a9ed 625 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
64ce2073 626 "queue %p\n", qp);
45542479 627 err = -ENOMEM;
1da177e4
LT
628 goto out_fail;
629out_oversize:
630 if (net_ratelimit())
673d57e7
HH
631 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
632 &qp->saddr);
1da177e4 633out_fail:
bbf31bf1 634 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
1706d587 635 return err;
1da177e4
LT
636}
637
638/* Process an incoming IP datagram fragment. */
776c729e 639int ip_defrag(struct sk_buff *skb, u32 user)
1da177e4 640{
1da177e4 641 struct ipq *qp;
ac18e750 642 struct net *net;
e905a9ed 643
adf30907 644 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
7c73a6fa 645 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
1da177e4
LT
646
647 /* Start by cleaning up the memory. */
e31e0bdc 648 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
6ddc0822 649 ip_evictor(net);
1da177e4 650
1da177e4 651 /* Lookup (or create) queue header */
ac18e750 652 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
1706d587 653 int ret;
1da177e4 654
5ab11c98 655 spin_lock(&qp->q.lock);
1da177e4 656
1706d587 657 ret = ip_frag_queue(qp, skb);
1da177e4 658
5ab11c98 659 spin_unlock(&qp->q.lock);
4b6cb5d8 660 ipq_put(qp);
776c729e 661 return ret;
1da177e4
LT
662 }
663
7c73a6fa 664 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
1da177e4 665 kfree_skb(skb);
776c729e 666 return -ENOMEM;
1da177e4 667}
4bc2f18b 668EXPORT_SYMBOL(ip_defrag);
1da177e4 669
8d8354d2
PE
670#ifdef CONFIG_SYSCTL
671static int zero;
672
0a64b4b8 673static struct ctl_table ip4_frags_ns_ctl_table[] = {
8d8354d2 674 {
8d8354d2 675 .procname = "ipfrag_high_thresh",
e31e0bdc 676 .data = &init_net.ipv4.frags.high_thresh,
8d8354d2
PE
677 .maxlen = sizeof(int),
678 .mode = 0644,
6d9f239a 679 .proc_handler = proc_dointvec
8d8354d2
PE
680 },
681 {
8d8354d2 682 .procname = "ipfrag_low_thresh",
e31e0bdc 683 .data = &init_net.ipv4.frags.low_thresh,
8d8354d2
PE
684 .maxlen = sizeof(int),
685 .mode = 0644,
6d9f239a 686 .proc_handler = proc_dointvec
8d8354d2
PE
687 },
688 {
8d8354d2 689 .procname = "ipfrag_time",
b2fd5321 690 .data = &init_net.ipv4.frags.timeout,
8d8354d2
PE
691 .maxlen = sizeof(int),
692 .mode = 0644,
6d9f239a 693 .proc_handler = proc_dointvec_jiffies,
8d8354d2 694 },
7d291ebb
PE
695 { }
696};
697
698static struct ctl_table ip4_frags_ctl_table[] = {
8d8354d2 699 {
8d8354d2 700 .procname = "ipfrag_secret_interval",
3b4bc4a2 701 .data = &ip4_frags.secret_interval,
8d8354d2
PE
702 .maxlen = sizeof(int),
703 .mode = 0644,
6d9f239a 704 .proc_handler = proc_dointvec_jiffies,
8d8354d2
PE
705 },
706 {
707 .procname = "ipfrag_max_dist",
708 .data = &sysctl_ipfrag_max_dist,
709 .maxlen = sizeof(int),
710 .mode = 0644,
6d9f239a 711 .proc_handler = proc_dointvec_minmax,
8d8354d2
PE
712 .extra1 = &zero
713 },
714 { }
715};
716
2c8c1e72 717static int __net_init ip4_frags_ns_ctl_register(struct net *net)
8d8354d2 718{
e4a2d5c2 719 struct ctl_table *table;
8d8354d2
PE
720 struct ctl_table_header *hdr;
721
0a64b4b8 722 table = ip4_frags_ns_ctl_table;
09ad9bc7 723 if (!net_eq(net, &init_net)) {
0a64b4b8 724 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
e4a2d5c2
PE
725 if (table == NULL)
726 goto err_alloc;
727
e31e0bdc
PE
728 table[0].data = &net->ipv4.frags.high_thresh;
729 table[1].data = &net->ipv4.frags.low_thresh;
b2fd5321 730 table[2].data = &net->ipv4.frags.timeout;
e4a2d5c2
PE
731 }
732
733 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
734 if (hdr == NULL)
735 goto err_reg;
736
737 net->ipv4.frags_hdr = hdr;
738 return 0;
739
740err_reg:
09ad9bc7 741 if (!net_eq(net, &init_net))
e4a2d5c2
PE
742 kfree(table);
743err_alloc:
744 return -ENOMEM;
745}
746
2c8c1e72 747static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
e4a2d5c2
PE
748{
749 struct ctl_table *table;
750
751 table = net->ipv4.frags_hdr->ctl_table_arg;
752 unregister_net_sysctl_table(net->ipv4.frags_hdr);
753 kfree(table);
8d8354d2 754}
7d291ebb
PE
755
756static void ip4_frags_ctl_register(void)
757{
758 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
759}
8d8354d2 760#else
0a64b4b8 761static inline int ip4_frags_ns_ctl_register(struct net *net)
8d8354d2
PE
762{
763 return 0;
764}
e4a2d5c2 765
0a64b4b8 766static inline void ip4_frags_ns_ctl_unregister(struct net *net)
e4a2d5c2
PE
767{
768}
7d291ebb
PE
769
770static inline void ip4_frags_ctl_register(void)
771{
772}
8d8354d2
PE
773#endif
774
2c8c1e72 775static int __net_init ipv4_frags_init_net(struct net *net)
8d8354d2 776{
e31e0bdc
PE
777 /*
778 * Fragment cache limits. We will commit 256K at one time. Should we
779 * cross that limit we will prune down to 192K. This should cope with
780 * even the most extreme cases without allowing an attacker to
781 * measurably harm machine performance.
782 */
783 net->ipv4.frags.high_thresh = 256 * 1024;
784 net->ipv4.frags.low_thresh = 192 * 1024;
b2fd5321
PE
785 /*
786 * Important NOTE! Fragment queue must be destroyed before MSL expires.
787 * RFC791 is wrong proposing to prolongate timer each fragment arrival
788 * by TTL.
789 */
790 net->ipv4.frags.timeout = IP_FRAG_TIME;
791
e5a2bb84
PE
792 inet_frags_init_net(&net->ipv4.frags);
793
0a64b4b8 794 return ip4_frags_ns_ctl_register(net);
8d8354d2
PE
795}
796
2c8c1e72 797static void __net_exit ipv4_frags_exit_net(struct net *net)
81566e83 798{
0a64b4b8 799 ip4_frags_ns_ctl_unregister(net);
81566e83
PE
800 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
801}
802
803static struct pernet_operations ip4_frags_ops = {
804 .init = ipv4_frags_init_net,
805 .exit = ipv4_frags_exit_net,
806};
807
b7aa0bf7 808void __init ipfrag_init(void)
1da177e4 809{
7d291ebb 810 ip4_frags_ctl_register();
81566e83 811 register_pernet_subsys(&ip4_frags_ops);
321a3a99 812 ip4_frags.hashfn = ip4_hashfn;
c6fda282 813 ip4_frags.constructor = ip4_frag_init;
1e4b8287
PE
814 ip4_frags.destructor = ip4_frag_free;
815 ip4_frags.skb_free = NULL;
816 ip4_frags.qsize = sizeof(struct ipq);
abd6523d 817 ip4_frags.match = ip4_frag_match;
e521db9d 818 ip4_frags.frag_expire = ip_expire;
3b4bc4a2 819 ip4_frags.secret_interval = 10 * 60 * HZ;
7eb95156 820 inet_frags_init(&ip4_frags);
1da177e4 821}