Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * The IP fragmentation functionality.
8 *
9 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox <alan@lxorguk.ukuu.org.uk>
11 *
12 * Fixes:
13 * Alan Cox : Split from ip.c , see ip_input.c for history.
14 * David S. Miller : Begin massive cleanup...
15 * Andi Kleen : Add sysctls.
16 * xxxx : Overlapfrag bug.
17 * Ultima : ip_expire() kernel panic.
18 * Bill Hawes : Frag accounting and evictor fixes.
19 * John McDonald : 0 length frag bug.
20 * Alexey Kuznetsov: SMP races, threading, cleanup.
21 * Patrick McHardy : LRU queue of frag heads for evictor.
22 */
23
24#define pr_fmt(fmt) "IPv4: " fmt
25
26#include <linux/compiler.h>
27#include <linux/module.h>
28#include <linux/types.h>
29#include <linux/mm.h>
30#include <linux/jiffies.h>
31#include <linux/skbuff.h>
32#include <linux/list.h>
33#include <linux/ip.h>
34#include <linux/icmp.h>
35#include <linux/netdevice.h>
36#include <linux/jhash.h>
37#include <linux/random.h>
38#include <linux/slab.h>
39#include <net/route.h>
40#include <net/dst.h>
41#include <net/sock.h>
42#include <net/ip.h>
43#include <net/icmp.h>
44#include <net/checksum.h>
45#include <net/inetpeer.h>
46#include <net/inet_frag.h>
47#include <linux/tcp.h>
48#include <linux/udp.h>
49#include <linux/inet.h>
50#include <linux/netfilter_ipv4.h>
51#include <net/inet_ecn.h>
52#include <net/l3mdev.h>
53
54/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
55 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
56 * as well. Or notify me, at least. --ANK
57 */
58static const char ip_frag_cache_name[] = "ip4-frags";
59
60/* Describe an entry in the "incomplete datagrams" queue. */
61struct ipq {
62 struct inet_frag_queue q;
63
64 u8 ecn; /* RFC3168 support */
65 u16 max_df_size; /* largest frag with DF set seen */
66 int iif;
67 unsigned int rid;
68 struct inet_peer *peer;
69};
70
71static u8 ip4_frag_ecn(u8 tos)
72{
73 return 1 << (tos & INET_ECN_MASK);
74}
75
76static struct inet_frags ip4_frags;
77
78static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
79 struct net_device *dev);
80
81
82static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
83{
84 struct ipq *qp = container_of(q, struct ipq, q);
85 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
86 frags);
87 struct net *net = container_of(ipv4, struct net, ipv4);
88
89 const struct frag_v4_compare_key *key = a;
90
91 q->key.v4 = *key;
92 qp->ecn = 0;
93 qp->peer = q->net->max_dist ?
94 inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
95 NULL;
96}
97
98static void ip4_frag_free(struct inet_frag_queue *q)
99{
100 struct ipq *qp;
101
102 qp = container_of(q, struct ipq, q);
103 if (qp->peer)
104 inet_putpeer(qp->peer);
105}
106
107
108/* Destruction primitives. */
109
110static void ipq_put(struct ipq *ipq)
111{
112 inet_frag_put(&ipq->q);
113}
114
115/* Kill ipq entry. It is not destroyed immediately,
116 * because caller (and someone more) holds reference count.
117 */
118static void ipq_kill(struct ipq *ipq)
119{
120 inet_frag_kill(&ipq->q);
121}
122
123static bool frag_expire_skip_icmp(u32 user)
124{
125 return user == IP_DEFRAG_AF_PACKET ||
126 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
127 __IP_DEFRAG_CONNTRACK_IN_END) ||
128 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
129 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
130}
131
132/*
133 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
134 */
135static void ip_expire(struct timer_list *t)
136{
137 struct inet_frag_queue *frag = from_timer(frag, t, timer);
138 const struct iphdr *iph;
139 struct sk_buff *head;
140 struct net *net;
141 struct ipq *qp;
142 int err;
143
144 qp = container_of(frag, struct ipq, q);
145 net = container_of(qp->q.net, struct net, ipv4.frags);
146
147 rcu_read_lock();
148 spin_lock(&qp->q.lock);
149
150 if (qp->q.flags & INET_FRAG_COMPLETE)
151 goto out;
152
153 ipq_kill(qp);
154 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
155
156 head = qp->q.fragments;
157
158 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
159
160 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !head)
161 goto out;
162
163 head->dev = dev_get_by_index_rcu(net, qp->iif);
164 if (!head->dev)
165 goto out;
166
167
168 /* skb has no dst, perform route lookup again */
169 iph = ip_hdr(head);
170 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
171 iph->tos, head->dev);
172 if (err)
173 goto out;
174
175 /* Only an end host needs to send an ICMP
176 * "Fragment Reassembly Timeout" message, per RFC792.
177 */
178 if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
179 (skb_rtable(head)->rt_type != RTN_LOCAL))
180 goto out;
181
182 skb_get(head);
183 spin_unlock(&qp->q.lock);
184 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
185 kfree_skb(head);
186 goto out_rcu_unlock;
187
188out:
189 spin_unlock(&qp->q.lock);
190out_rcu_unlock:
191 rcu_read_unlock();
192 ipq_put(qp);
193}
194
195/* Find the correct entry in the "incomplete datagrams" queue for
196 * this IP datagram, and create new one, if nothing is found.
197 */
198static struct ipq *ip_find(struct net *net, struct iphdr *iph,
199 u32 user, int vif)
200{
201 struct frag_v4_compare_key key = {
202 .saddr = iph->saddr,
203 .daddr = iph->daddr,
204 .user = user,
205 .vif = vif,
206 .id = iph->id,
207 .protocol = iph->protocol,
208 };
209 struct inet_frag_queue *q;
210
211 q = inet_frag_find(&net->ipv4.frags, &key);
212 if (!q)
213 return NULL;
214
215 return container_of(q, struct ipq, q);
216}
217
218/* Is the fragment too far ahead to be part of ipq? */
219static int ip_frag_too_far(struct ipq *qp)
220{
221 struct inet_peer *peer = qp->peer;
222 unsigned int max = qp->q.net->max_dist;
223 unsigned int start, end;
224
225 int rc;
226
227 if (!peer || !max)
228 return 0;
229
230 start = qp->rid;
231 end = atomic_inc_return(&peer->rid);
232 qp->rid = end;
233
234 rc = qp->q.fragments && (end - start) > max;
235
236 if (rc) {
237 struct net *net;
238
239 net = container_of(qp->q.net, struct net, ipv4.frags);
240 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
241 }
242
243 return rc;
244}
245
246static int ip_frag_reinit(struct ipq *qp)
247{
248 struct sk_buff *fp;
249 unsigned int sum_truesize = 0;
250
251 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
252 refcount_inc(&qp->q.refcnt);
253 return -ETIMEDOUT;
254 }
255
256 fp = qp->q.fragments;
257 do {
258 struct sk_buff *xp = fp->next;
259
260 sum_truesize += fp->truesize;
261 kfree_skb(fp);
262 fp = xp;
263 } while (fp);
264 sub_frag_mem_limit(qp->q.net, sum_truesize);
265
266 qp->q.flags = 0;
267 qp->q.len = 0;
268 qp->q.meat = 0;
269 qp->q.fragments = NULL;
270 qp->q.fragments_tail = NULL;
271 qp->iif = 0;
272 qp->ecn = 0;
273
274 return 0;
275}
276
277/* Add new segment to existing queue. */
278static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
279{
280 struct sk_buff *prev, *next;
281 struct net_device *dev;
282 unsigned int fragsize;
283 int flags, offset;
284 int ihl, end;
285 int err = -ENOENT;
286 u8 ecn;
287
288 if (qp->q.flags & INET_FRAG_COMPLETE)
289 goto err;
290
291 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
292 unlikely(ip_frag_too_far(qp)) &&
293 unlikely(err = ip_frag_reinit(qp))) {
294 ipq_kill(qp);
295 goto err;
296 }
297
298 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
299 offset = ntohs(ip_hdr(skb)->frag_off);
300 flags = offset & ~IP_OFFSET;
301 offset &= IP_OFFSET;
302 offset <<= 3; /* offset is in 8-byte chunks */
303 ihl = ip_hdrlen(skb);
304
305 /* Determine the position of this fragment. */
306 end = offset + skb->len - skb_network_offset(skb) - ihl;
307 err = -EINVAL;
308
309 /* Is this the final fragment? */
310 if ((flags & IP_MF) == 0) {
311 /* If we already have some bits beyond end
312 * or have different end, the segment is corrupted.
313 */
314 if (end < qp->q.len ||
315 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
316 goto err;
317 qp->q.flags |= INET_FRAG_LAST_IN;
318 qp->q.len = end;
319 } else {
320 if (end&7) {
321 end &= ~7;
322 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
323 skb->ip_summed = CHECKSUM_NONE;
324 }
325 if (end > qp->q.len) {
326 /* Some bits beyond end -> corruption. */
327 if (qp->q.flags & INET_FRAG_LAST_IN)
328 goto err;
329 qp->q.len = end;
330 }
331 }
332 if (end == offset)
333 goto err;
334
335 err = -ENOMEM;
336 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
337 goto err;
338
339 err = pskb_trim_rcsum(skb, end - offset);
340 if (err)
341 goto err;
342
343 /* Find out which fragments are in front and at the back of us
344 * in the chain of fragments so far. We must know where to put
345 * this fragment, right?
346 */
347 prev = qp->q.fragments_tail;
348 if (!prev || prev->ip_defrag_offset < offset) {
349 next = NULL;
350 goto found;
351 }
352 prev = NULL;
353 for (next = qp->q.fragments; next != NULL; next = next->next) {
354 if (next->ip_defrag_offset >= offset)
355 break; /* bingo! */
356 prev = next;
357 }
358
359found:
360 /* We found where to put this one. Check for overlap with
361 * preceding fragment, and, if needed, align things so that
362 * any overlaps are eliminated.
363 */
364 if (prev) {
365 int i = (prev->ip_defrag_offset + prev->len) - offset;
366
367 if (i > 0) {
368 offset += i;
369 err = -EINVAL;
370 if (end <= offset)
371 goto err;
372 err = -ENOMEM;
373 if (!pskb_pull(skb, i))
374 goto err;
375 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
376 skb->ip_summed = CHECKSUM_NONE;
377 }
378 }
379
380 err = -ENOMEM;
381
382 while (next && next->ip_defrag_offset < end) {
383 int i = end - next->ip_defrag_offset; /* overlap is 'i' bytes */
384
385 if (i < next->len) {
386 /* Eat head of the next overlapped fragment
387 * and leave the loop. The next ones cannot overlap.
388 */
389 if (!pskb_pull(next, i))
390 goto err;
391 next->ip_defrag_offset += i;
392 qp->q.meat -= i;
393 if (next->ip_summed != CHECKSUM_UNNECESSARY)
394 next->ip_summed = CHECKSUM_NONE;
395 break;
396 } else {
397 struct sk_buff *free_it = next;
398
399 /* Old fragment is completely overridden with
400 * new one drop it.
401 */
402 next = next->next;
403
404 if (prev)
405 prev->next = next;
406 else
407 qp->q.fragments = next;
408
409 qp->q.meat -= free_it->len;
410 sub_frag_mem_limit(qp->q.net, free_it->truesize);
411 kfree_skb(free_it);
412 }
413 }
414
415 /* Note : skb->ip_defrag_offset and skb->dev share the same location */
416 dev = skb->dev;
417 if (dev)
418 qp->iif = dev->ifindex;
419 /* Makes sure compiler wont do silly aliasing games */
420 barrier();
421 skb->ip_defrag_offset = offset;
422
423 /* Insert this fragment in the chain of fragments. */
424 skb->next = next;
425 if (!next)
426 qp->q.fragments_tail = skb;
427 if (prev)
428 prev->next = skb;
429 else
430 qp->q.fragments = skb;
431
432 qp->q.stamp = skb->tstamp;
433 qp->q.meat += skb->len;
434 qp->ecn |= ecn;
435 add_frag_mem_limit(qp->q.net, skb->truesize);
436 if (offset == 0)
437 qp->q.flags |= INET_FRAG_FIRST_IN;
438
439 fragsize = skb->len + ihl;
440
441 if (fragsize > qp->q.max_size)
442 qp->q.max_size = fragsize;
443
444 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
445 fragsize > qp->max_df_size)
446 qp->max_df_size = fragsize;
447
448 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
449 qp->q.meat == qp->q.len) {
450 unsigned long orefdst = skb->_skb_refdst;
451
452 skb->_skb_refdst = 0UL;
453 err = ip_frag_reasm(qp, prev, dev);
454 skb->_skb_refdst = orefdst;
455 return err;
456 }
457
458 skb_dst_drop(skb);
459 return -EINPROGRESS;
460
461err:
462 kfree_skb(skb);
463 return err;
464}
465
466
467/* Build a new IP datagram from all its fragments. */
468
469static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
470 struct net_device *dev)
471{
472 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
473 struct iphdr *iph;
474 struct sk_buff *fp, *head = qp->q.fragments;
475 int len;
476 int ihlen;
477 int err;
478 u8 ecn;
479
480 ipq_kill(qp);
481
482 ecn = ip_frag_ecn_table[qp->ecn];
483 if (unlikely(ecn == 0xff)) {
484 err = -EINVAL;
485 goto out_fail;
486 }
487 /* Make the one we just received the head. */
488 if (prev) {
489 head = prev->next;
490 fp = skb_clone(head, GFP_ATOMIC);
491 if (!fp)
492 goto out_nomem;
493
494 fp->next = head->next;
495 if (!fp->next)
496 qp->q.fragments_tail = fp;
497 prev->next = fp;
498
499 skb_morph(head, qp->q.fragments);
500 head->next = qp->q.fragments->next;
501
502 consume_skb(qp->q.fragments);
503 qp->q.fragments = head;
504 }
505
506 WARN_ON(!head);
507 WARN_ON(head->ip_defrag_offset != 0);
508
509 /* Allocate a new buffer for the datagram. */
510 ihlen = ip_hdrlen(head);
511 len = ihlen + qp->q.len;
512
513 err = -E2BIG;
514 if (len > 65535)
515 goto out_oversize;
516
517 /* Head of list must not be cloned. */
518 if (skb_unclone(head, GFP_ATOMIC))
519 goto out_nomem;
520
521 /* If the first fragment is fragmented itself, we split
522 * it to two chunks: the first with data and paged part
523 * and the second, holding only fragments. */
524 if (skb_has_frag_list(head)) {
525 struct sk_buff *clone;
526 int i, plen = 0;
527
528 clone = alloc_skb(0, GFP_ATOMIC);
529 if (!clone)
530 goto out_nomem;
531 clone->next = head->next;
532 head->next = clone;
533 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
534 skb_frag_list_init(head);
535 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
536 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
537 clone->len = clone->data_len = head->data_len - plen;
538 head->data_len -= clone->len;
539 head->len -= clone->len;
540 clone->csum = 0;
541 clone->ip_summed = head->ip_summed;
542 add_frag_mem_limit(qp->q.net, clone->truesize);
543 }
544
545 skb_shinfo(head)->frag_list = head->next;
546 skb_push(head, head->data - skb_network_header(head));
547
548 for (fp=head->next; fp; fp = fp->next) {
549 head->data_len += fp->len;
550 head->len += fp->len;
551 if (head->ip_summed != fp->ip_summed)
552 head->ip_summed = CHECKSUM_NONE;
553 else if (head->ip_summed == CHECKSUM_COMPLETE)
554 head->csum = csum_add(head->csum, fp->csum);
555 head->truesize += fp->truesize;
556 }
557 sub_frag_mem_limit(qp->q.net, head->truesize);
558
559 head->next = NULL;
560 head->dev = dev;
561 head->tstamp = qp->q.stamp;
562 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
563
564 iph = ip_hdr(head);
565 iph->tot_len = htons(len);
566 iph->tos |= ecn;
567
568 /* When we set IP_DF on a refragmented skb we must also force a
569 * call to ip_fragment to avoid forwarding a DF-skb of size s while
570 * original sender only sent fragments of size f (where f < s).
571 *
572 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
573 * frag seen to avoid sending tiny DF-fragments in case skb was built
574 * from one very small df-fragment and one large non-df frag.
575 */
576 if (qp->max_df_size == qp->q.max_size) {
577 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
578 iph->frag_off = htons(IP_DF);
579 } else {
580 iph->frag_off = 0;
581 }
582
583 ip_send_check(iph);
584
585 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
586 qp->q.fragments = NULL;
587 qp->q.fragments_tail = NULL;
588 return 0;
589
590out_nomem:
591 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
592 err = -ENOMEM;
593 goto out_fail;
594out_oversize:
595 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
596out_fail:
597 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
598 return err;
599}
600
601/* Process an incoming IP datagram fragment. */
602int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
603{
604 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
605 int vif = l3mdev_master_ifindex_rcu(dev);
606 struct ipq *qp;
607
608 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
609 skb_orphan(skb);
610
611 /* Lookup (or create) queue header */
612 qp = ip_find(net, ip_hdr(skb), user, vif);
613 if (qp) {
614 int ret;
615
616 spin_lock(&qp->q.lock);
617
618 ret = ip_frag_queue(qp, skb);
619
620 spin_unlock(&qp->q.lock);
621 ipq_put(qp);
622 return ret;
623 }
624
625 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
626 kfree_skb(skb);
627 return -ENOMEM;
628}
629EXPORT_SYMBOL(ip_defrag);
630
631struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
632{
633 struct iphdr iph;
634 int netoff;
635 u32 len;
636
637 if (skb->protocol != htons(ETH_P_IP))
638 return skb;
639
640 netoff = skb_network_offset(skb);
641
642 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
643 return skb;
644
645 if (iph.ihl < 5 || iph.version != 4)
646 return skb;
647
648 len = ntohs(iph.tot_len);
649 if (skb->len < netoff + len || len < (iph.ihl * 4))
650 return skb;
651
652 if (ip_is_fragment(&iph)) {
653 skb = skb_share_check(skb, GFP_ATOMIC);
654 if (skb) {
655 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
656 return skb;
657 if (pskb_trim_rcsum(skb, netoff + len))
658 return skb;
659 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
660 if (ip_defrag(net, skb, user))
661 return NULL;
662 skb_clear_hash(skb);
663 }
664 }
665 return skb;
666}
667EXPORT_SYMBOL(ip_check_defrag);
668
669#ifdef CONFIG_SYSCTL
670static int dist_min;
671
672static struct ctl_table ip4_frags_ns_ctl_table[] = {
673 {
674 .procname = "ipfrag_high_thresh",
675 .data = &init_net.ipv4.frags.high_thresh,
676 .maxlen = sizeof(unsigned long),
677 .mode = 0644,
678 .proc_handler = proc_doulongvec_minmax,
679 .extra1 = &init_net.ipv4.frags.low_thresh
680 },
681 {
682 .procname = "ipfrag_low_thresh",
683 .data = &init_net.ipv4.frags.low_thresh,
684 .maxlen = sizeof(unsigned long),
685 .mode = 0644,
686 .proc_handler = proc_doulongvec_minmax,
687 .extra2 = &init_net.ipv4.frags.high_thresh
688 },
689 {
690 .procname = "ipfrag_time",
691 .data = &init_net.ipv4.frags.timeout,
692 .maxlen = sizeof(int),
693 .mode = 0644,
694 .proc_handler = proc_dointvec_jiffies,
695 },
696 {
697 .procname = "ipfrag_max_dist",
698 .data = &init_net.ipv4.frags.max_dist,
699 .maxlen = sizeof(int),
700 .mode = 0644,
701 .proc_handler = proc_dointvec_minmax,
702 .extra1 = &dist_min,
703 },
704 { }
705};
706
707/* secret interval has been deprecated */
708static int ip4_frags_secret_interval_unused;
709static struct ctl_table ip4_frags_ctl_table[] = {
710 {
711 .procname = "ipfrag_secret_interval",
712 .data = &ip4_frags_secret_interval_unused,
713 .maxlen = sizeof(int),
714 .mode = 0644,
715 .proc_handler = proc_dointvec_jiffies,
716 },
717 { }
718};
719
720static int __net_init ip4_frags_ns_ctl_register(struct net *net)
721{
722 struct ctl_table *table;
723 struct ctl_table_header *hdr;
724
725 table = ip4_frags_ns_ctl_table;
726 if (!net_eq(net, &init_net)) {
727 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
728 if (!table)
729 goto err_alloc;
730
731 table[0].data = &net->ipv4.frags.high_thresh;
732 table[0].extra1 = &net->ipv4.frags.low_thresh;
733 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
734 table[1].data = &net->ipv4.frags.low_thresh;
735 table[1].extra2 = &net->ipv4.frags.high_thresh;
736 table[2].data = &net->ipv4.frags.timeout;
737 table[3].data = &net->ipv4.frags.max_dist;
738 }
739
740 hdr = register_net_sysctl(net, "net/ipv4", table);
741 if (!hdr)
742 goto err_reg;
743
744 net->ipv4.frags_hdr = hdr;
745 return 0;
746
747err_reg:
748 if (!net_eq(net, &init_net))
749 kfree(table);
750err_alloc:
751 return -ENOMEM;
752}
753
754static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
755{
756 struct ctl_table *table;
757
758 table = net->ipv4.frags_hdr->ctl_table_arg;
759 unregister_net_sysctl_table(net->ipv4.frags_hdr);
760 kfree(table);
761}
762
763static void __init ip4_frags_ctl_register(void)
764{
765 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
766}
767#else
768static int ip4_frags_ns_ctl_register(struct net *net)
769{
770 return 0;
771}
772
773static void ip4_frags_ns_ctl_unregister(struct net *net)
774{
775}
776
777static void __init ip4_frags_ctl_register(void)
778{
779}
780#endif
781
782static int __net_init ipv4_frags_init_net(struct net *net)
783{
784 int res;
785
786 /* Fragment cache limits.
787 *
788 * The fragment memory accounting code, (tries to) account for
789 * the real memory usage, by measuring both the size of frag
790 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
791 * and the SKB's truesize.
792 *
793 * A 64K fragment consumes 129736 bytes (44*2944)+200
794 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
795 *
796 * We will commit 4MB at one time. Should we cross that limit
797 * we will prune down to 3MB, making room for approx 8 big 64K
798 * fragments 8x128k.
799 */
800 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
801 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
802 /*
803 * Important NOTE! Fragment queue must be destroyed before MSL expires.
804 * RFC791 is wrong proposing to prolongate timer each fragment arrival
805 * by TTL.
806 */
807 net->ipv4.frags.timeout = IP_FRAG_TIME;
808
809 net->ipv4.frags.max_dist = 64;
810 net->ipv4.frags.f = &ip4_frags;
811
812 res = inet_frags_init_net(&net->ipv4.frags);
813 if (res < 0)
814 return res;
815 res = ip4_frags_ns_ctl_register(net);
816 if (res < 0)
817 inet_frags_exit_net(&net->ipv4.frags);
818 return res;
819}
820
821static void __net_exit ipv4_frags_exit_net(struct net *net)
822{
823 ip4_frags_ns_ctl_unregister(net);
824 inet_frags_exit_net(&net->ipv4.frags);
825}
826
827static struct pernet_operations ip4_frags_ops = {
828 .init = ipv4_frags_init_net,
829 .exit = ipv4_frags_exit_net,
830};
831
832
833static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
834{
835 return jhash2(data,
836 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
837}
838
839static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
840{
841 const struct inet_frag_queue *fq = data;
842
843 return jhash2((const u32 *)&fq->key.v4,
844 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
845}
846
847static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
848{
849 const struct frag_v4_compare_key *key = arg->key;
850 const struct inet_frag_queue *fq = ptr;
851
852 return !!memcmp(&fq->key, key, sizeof(*key));
853}
854
855static const struct rhashtable_params ip4_rhash_params = {
856 .head_offset = offsetof(struct inet_frag_queue, node),
857 .key_offset = offsetof(struct inet_frag_queue, key),
858 .key_len = sizeof(struct frag_v4_compare_key),
859 .hashfn = ip4_key_hashfn,
860 .obj_hashfn = ip4_obj_hashfn,
861 .obj_cmpfn = ip4_obj_cmpfn,
862 .automatic_shrinking = true,
863};
864
865void __init ipfrag_init(void)
866{
867 ip4_frags.constructor = ip4_frag_init;
868 ip4_frags.destructor = ip4_frag_free;
869 ip4_frags.qsize = sizeof(struct ipq);
870 ip4_frags.frag_expire = ip_expire;
871 ip4_frags.frags_cache_name = ip_frag_cache_name;
872 ip4_frags.rhash_params = ip4_rhash_params;
873 if (inet_frags_init(&ip4_frags))
874 panic("IP: failed to allocate ip4_frags cache\n");
875 ip4_frags_ctl_register();
876 register_pernet_subsys(&ip4_frags_ops);
877}
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.
7 *
8 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
9 * Alan Cox <alan@lxorguk.ukuu.org.uk>
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
23#include <linux/compiler.h>
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>
35#include <linux/slab.h>
36#include <net/route.h>
37#include <net/dst.h>
38#include <net/sock.h>
39#include <net/ip.h>
40#include <net/icmp.h>
41#include <net/checksum.h>
42#include <net/inetpeer.h>
43#include <net/inet_frag.h>
44#include <linux/tcp.h>
45#include <linux/udp.h>
46#include <linux/inet.h>
47#include <linux/netfilter_ipv4.h>
48#include <net/inet_ecn.h>
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
55static int sysctl_ipfrag_max_dist __read_mostly = 64;
56
57struct ipfrag_skb_cb
58{
59 struct inet_skb_parm h;
60 int offset;
61};
62
63#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
64
65/* Describe an entry in the "incomplete datagrams" queue. */
66struct ipq {
67 struct inet_frag_queue q;
68
69 u32 user;
70 __be32 saddr;
71 __be32 daddr;
72 __be16 id;
73 u8 protocol;
74 u8 ecn; /* RFC3168 support */
75 int iif;
76 unsigned int rid;
77 struct inet_peer *peer;
78};
79
80/* RFC 3168 support :
81 * We want to check ECN values of all fragments, do detect invalid combinations.
82 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
83 */
84#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
85#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
86#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
87#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
88
89static inline u8 ip4_frag_ecn(u8 tos)
90{
91 return 1 << (tos & INET_ECN_MASK);
92}
93
94/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
95 * Value : 0xff if frame should be dropped.
96 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
97 */
98static const u8 ip4_frag_ecn_table[16] = {
99 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
100 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
101 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
103
104 /* invalid combinations : drop frame */
105 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
106 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
112};
113
114static struct inet_frags ip4_frags;
115
116int ip_frag_nqueues(struct net *net)
117{
118 return net->ipv4.frags.nqueues;
119}
120
121int ip_frag_mem(struct net *net)
122{
123 return atomic_read(&net->ipv4.frags.mem);
124}
125
126static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
127 struct net_device *dev);
128
129struct ip4_create_arg {
130 struct iphdr *iph;
131 u32 user;
132};
133
134static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
135{
136 return jhash_3words((__force u32)id << 16 | prot,
137 (__force u32)saddr, (__force u32)daddr,
138 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
139}
140
141static unsigned int ip4_hashfn(struct inet_frag_queue *q)
142{
143 struct ipq *ipq;
144
145 ipq = container_of(q, struct ipq, q);
146 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
147}
148
149static int ip4_frag_match(struct inet_frag_queue *q, void *a)
150{
151 struct ipq *qp;
152 struct ip4_create_arg *arg = a;
153
154 qp = container_of(q, struct ipq, q);
155 return qp->id == arg->iph->id &&
156 qp->saddr == arg->iph->saddr &&
157 qp->daddr == arg->iph->daddr &&
158 qp->protocol == arg->iph->protocol &&
159 qp->user == arg->user;
160}
161
162/* Memory Tracking Functions. */
163static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
164{
165 atomic_sub(skb->truesize, &nf->mem);
166 kfree_skb(skb);
167}
168
169static void ip4_frag_init(struct inet_frag_queue *q, void *a)
170{
171 struct ipq *qp = container_of(q, struct ipq, q);
172 struct ip4_create_arg *arg = a;
173
174 qp->protocol = arg->iph->protocol;
175 qp->id = arg->iph->id;
176 qp->ecn = ip4_frag_ecn(arg->iph->tos);
177 qp->saddr = arg->iph->saddr;
178 qp->daddr = arg->iph->daddr;
179 qp->user = arg->user;
180 qp->peer = sysctl_ipfrag_max_dist ?
181 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
182}
183
184static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
185{
186 struct ipq *qp;
187
188 qp = container_of(q, struct ipq, q);
189 if (qp->peer)
190 inet_putpeer(qp->peer);
191}
192
193
194/* Destruction primitives. */
195
196static __inline__ void ipq_put(struct ipq *ipq)
197{
198 inet_frag_put(&ipq->q, &ip4_frags);
199}
200
201/* Kill ipq entry. It is not destroyed immediately,
202 * because caller (and someone more) holds reference count.
203 */
204static void ipq_kill(struct ipq *ipq)
205{
206 inet_frag_kill(&ipq->q, &ip4_frags);
207}
208
209/* Memory limiting on fragments. Evictor trashes the oldest
210 * fragment queue until we are back under the threshold.
211 */
212static void ip_evictor(struct net *net)
213{
214 int evicted;
215
216 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
217 if (evicted)
218 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
219}
220
221/*
222 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
223 */
224static void ip_expire(unsigned long arg)
225{
226 struct ipq *qp;
227 struct net *net;
228
229 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
230 net = container_of(qp->q.net, struct net, ipv4.frags);
231
232 spin_lock(&qp->q.lock);
233
234 if (qp->q.last_in & INET_FRAG_COMPLETE)
235 goto out;
236
237 ipq_kill(qp);
238
239 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
240 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
241
242 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
243 struct sk_buff *head = qp->q.fragments;
244 const struct iphdr *iph;
245 int err;
246
247 rcu_read_lock();
248 head->dev = dev_get_by_index_rcu(net, qp->iif);
249 if (!head->dev)
250 goto out_rcu_unlock;
251
252 /* skb dst is stale, drop it, and perform route lookup again */
253 skb_dst_drop(head);
254 iph = ip_hdr(head);
255 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
256 iph->tos, head->dev);
257 if (err)
258 goto out_rcu_unlock;
259
260 /*
261 * Only an end host needs to send an ICMP
262 * "Fragment Reassembly Timeout" message, per RFC792.
263 */
264 if (qp->user == IP_DEFRAG_AF_PACKET ||
265 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
266 skb_rtable(head)->rt_type != RTN_LOCAL))
267 goto out_rcu_unlock;
268
269
270 /* Send an ICMP "Fragment Reassembly Timeout" message. */
271 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
272out_rcu_unlock:
273 rcu_read_unlock();
274 }
275out:
276 spin_unlock(&qp->q.lock);
277 ipq_put(qp);
278}
279
280/* Find the correct entry in the "incomplete datagrams" queue for
281 * this IP datagram, and create new one, if nothing is found.
282 */
283static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
284{
285 struct inet_frag_queue *q;
286 struct ip4_create_arg arg;
287 unsigned int hash;
288
289 arg.iph = iph;
290 arg.user = user;
291
292 read_lock(&ip4_frags.lock);
293 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
294
295 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
296 if (q == NULL)
297 goto out_nomem;
298
299 return container_of(q, struct ipq, q);
300
301out_nomem:
302 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
303 return NULL;
304}
305
306/* Is the fragment too far ahead to be part of ipq? */
307static inline int ip_frag_too_far(struct ipq *qp)
308{
309 struct inet_peer *peer = qp->peer;
310 unsigned int max = sysctl_ipfrag_max_dist;
311 unsigned int start, end;
312
313 int rc;
314
315 if (!peer || !max)
316 return 0;
317
318 start = qp->rid;
319 end = atomic_inc_return(&peer->rid);
320 qp->rid = end;
321
322 rc = qp->q.fragments && (end - start) > max;
323
324 if (rc) {
325 struct net *net;
326
327 net = container_of(qp->q.net, struct net, ipv4.frags);
328 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
329 }
330
331 return rc;
332}
333
334static int ip_frag_reinit(struct ipq *qp)
335{
336 struct sk_buff *fp;
337
338 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
339 atomic_inc(&qp->q.refcnt);
340 return -ETIMEDOUT;
341 }
342
343 fp = qp->q.fragments;
344 do {
345 struct sk_buff *xp = fp->next;
346 frag_kfree_skb(qp->q.net, fp);
347 fp = xp;
348 } while (fp);
349
350 qp->q.last_in = 0;
351 qp->q.len = 0;
352 qp->q.meat = 0;
353 qp->q.fragments = NULL;
354 qp->q.fragments_tail = NULL;
355 qp->iif = 0;
356 qp->ecn = 0;
357
358 return 0;
359}
360
361/* Add new segment to existing queue. */
362static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
363{
364 struct sk_buff *prev, *next;
365 struct net_device *dev;
366 int flags, offset;
367 int ihl, end;
368 int err = -ENOENT;
369 u8 ecn;
370
371 if (qp->q.last_in & INET_FRAG_COMPLETE)
372 goto err;
373
374 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
375 unlikely(ip_frag_too_far(qp)) &&
376 unlikely(err = ip_frag_reinit(qp))) {
377 ipq_kill(qp);
378 goto err;
379 }
380
381 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
382 offset = ntohs(ip_hdr(skb)->frag_off);
383 flags = offset & ~IP_OFFSET;
384 offset &= IP_OFFSET;
385 offset <<= 3; /* offset is in 8-byte chunks */
386 ihl = ip_hdrlen(skb);
387
388 /* Determine the position of this fragment. */
389 end = offset + skb->len - ihl;
390 err = -EINVAL;
391
392 /* Is this the final fragment? */
393 if ((flags & IP_MF) == 0) {
394 /* If we already have some bits beyond end
395 * or have different end, the segment is corrrupted.
396 */
397 if (end < qp->q.len ||
398 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
399 goto err;
400 qp->q.last_in |= INET_FRAG_LAST_IN;
401 qp->q.len = end;
402 } else {
403 if (end&7) {
404 end &= ~7;
405 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
406 skb->ip_summed = CHECKSUM_NONE;
407 }
408 if (end > qp->q.len) {
409 /* Some bits beyond end -> corruption. */
410 if (qp->q.last_in & INET_FRAG_LAST_IN)
411 goto err;
412 qp->q.len = end;
413 }
414 }
415 if (end == offset)
416 goto err;
417
418 err = -ENOMEM;
419 if (pskb_pull(skb, ihl) == NULL)
420 goto err;
421
422 err = pskb_trim_rcsum(skb, end - offset);
423 if (err)
424 goto err;
425
426 /* Find out which fragments are in front and at the back of us
427 * in the chain of fragments so far. We must know where to put
428 * this fragment, right?
429 */
430 prev = qp->q.fragments_tail;
431 if (!prev || FRAG_CB(prev)->offset < offset) {
432 next = NULL;
433 goto found;
434 }
435 prev = NULL;
436 for (next = qp->q.fragments; next != NULL; next = next->next) {
437 if (FRAG_CB(next)->offset >= offset)
438 break; /* bingo! */
439 prev = next;
440 }
441
442found:
443 /* We found where to put this one. Check for overlap with
444 * preceding fragment, and, if needed, align things so that
445 * any overlaps are eliminated.
446 */
447 if (prev) {
448 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
449
450 if (i > 0) {
451 offset += i;
452 err = -EINVAL;
453 if (end <= offset)
454 goto err;
455 err = -ENOMEM;
456 if (!pskb_pull(skb, i))
457 goto err;
458 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
459 skb->ip_summed = CHECKSUM_NONE;
460 }
461 }
462
463 err = -ENOMEM;
464
465 while (next && FRAG_CB(next)->offset < end) {
466 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
467
468 if (i < next->len) {
469 /* Eat head of the next overlapped fragment
470 * and leave the loop. The next ones cannot overlap.
471 */
472 if (!pskb_pull(next, i))
473 goto err;
474 FRAG_CB(next)->offset += i;
475 qp->q.meat -= i;
476 if (next->ip_summed != CHECKSUM_UNNECESSARY)
477 next->ip_summed = CHECKSUM_NONE;
478 break;
479 } else {
480 struct sk_buff *free_it = next;
481
482 /* Old fragment is completely overridden with
483 * new one drop it.
484 */
485 next = next->next;
486
487 if (prev)
488 prev->next = next;
489 else
490 qp->q.fragments = next;
491
492 qp->q.meat -= free_it->len;
493 frag_kfree_skb(qp->q.net, free_it);
494 }
495 }
496
497 FRAG_CB(skb)->offset = offset;
498
499 /* Insert this fragment in the chain of fragments. */
500 skb->next = next;
501 if (!next)
502 qp->q.fragments_tail = skb;
503 if (prev)
504 prev->next = skb;
505 else
506 qp->q.fragments = skb;
507
508 dev = skb->dev;
509 if (dev) {
510 qp->iif = dev->ifindex;
511 skb->dev = NULL;
512 }
513 qp->q.stamp = skb->tstamp;
514 qp->q.meat += skb->len;
515 qp->ecn |= ecn;
516 atomic_add(skb->truesize, &qp->q.net->mem);
517 if (offset == 0)
518 qp->q.last_in |= INET_FRAG_FIRST_IN;
519
520 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
521 qp->q.meat == qp->q.len)
522 return ip_frag_reasm(qp, prev, dev);
523
524 write_lock(&ip4_frags.lock);
525 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
526 write_unlock(&ip4_frags.lock);
527 return -EINPROGRESS;
528
529err:
530 kfree_skb(skb);
531 return err;
532}
533
534
535/* Build a new IP datagram from all its fragments. */
536
537static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
538 struct net_device *dev)
539{
540 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
541 struct iphdr *iph;
542 struct sk_buff *fp, *head = qp->q.fragments;
543 int len;
544 int ihlen;
545 int err;
546 u8 ecn;
547
548 ipq_kill(qp);
549
550 ecn = ip4_frag_ecn_table[qp->ecn];
551 if (unlikely(ecn == 0xff)) {
552 err = -EINVAL;
553 goto out_fail;
554 }
555 /* Make the one we just received the head. */
556 if (prev) {
557 head = prev->next;
558 fp = skb_clone(head, GFP_ATOMIC);
559 if (!fp)
560 goto out_nomem;
561
562 fp->next = head->next;
563 if (!fp->next)
564 qp->q.fragments_tail = fp;
565 prev->next = fp;
566
567 skb_morph(head, qp->q.fragments);
568 head->next = qp->q.fragments->next;
569
570 kfree_skb(qp->q.fragments);
571 qp->q.fragments = head;
572 }
573
574 WARN_ON(head == NULL);
575 WARN_ON(FRAG_CB(head)->offset != 0);
576
577 /* Allocate a new buffer for the datagram. */
578 ihlen = ip_hdrlen(head);
579 len = ihlen + qp->q.len;
580
581 err = -E2BIG;
582 if (len > 65535)
583 goto out_oversize;
584
585 /* Head of list must not be cloned. */
586 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
587 goto out_nomem;
588
589 /* If the first fragment is fragmented itself, we split
590 * it to two chunks: the first with data and paged part
591 * and the second, holding only fragments. */
592 if (skb_has_frag_list(head)) {
593 struct sk_buff *clone;
594 int i, plen = 0;
595
596 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
597 goto out_nomem;
598 clone->next = head->next;
599 head->next = clone;
600 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
601 skb_frag_list_init(head);
602 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
603 plen += skb_shinfo(head)->frags[i].size;
604 clone->len = clone->data_len = head->data_len - plen;
605 head->data_len -= clone->len;
606 head->len -= clone->len;
607 clone->csum = 0;
608 clone->ip_summed = head->ip_summed;
609 atomic_add(clone->truesize, &qp->q.net->mem);
610 }
611
612 skb_shinfo(head)->frag_list = head->next;
613 skb_push(head, head->data - skb_network_header(head));
614
615 for (fp=head->next; fp; fp = fp->next) {
616 head->data_len += fp->len;
617 head->len += fp->len;
618 if (head->ip_summed != fp->ip_summed)
619 head->ip_summed = CHECKSUM_NONE;
620 else if (head->ip_summed == CHECKSUM_COMPLETE)
621 head->csum = csum_add(head->csum, fp->csum);
622 head->truesize += fp->truesize;
623 }
624 atomic_sub(head->truesize, &qp->q.net->mem);
625
626 head->next = NULL;
627 head->dev = dev;
628 head->tstamp = qp->q.stamp;
629
630 iph = ip_hdr(head);
631 iph->frag_off = 0;
632 iph->tot_len = htons(len);
633 iph->tos |= ecn;
634 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
635 qp->q.fragments = NULL;
636 qp->q.fragments_tail = NULL;
637 return 0;
638
639out_nomem:
640 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
641 "queue %p\n", qp);
642 err = -ENOMEM;
643 goto out_fail;
644out_oversize:
645 if (net_ratelimit())
646 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
647 &qp->saddr);
648out_fail:
649 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
650 return err;
651}
652
653/* Process an incoming IP datagram fragment. */
654int ip_defrag(struct sk_buff *skb, u32 user)
655{
656 struct ipq *qp;
657 struct net *net;
658
659 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
660 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
661
662 /* Start by cleaning up the memory. */
663 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
664 ip_evictor(net);
665
666 /* Lookup (or create) queue header */
667 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
668 int ret;
669
670 spin_lock(&qp->q.lock);
671
672 ret = ip_frag_queue(qp, skb);
673
674 spin_unlock(&qp->q.lock);
675 ipq_put(qp);
676 return ret;
677 }
678
679 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
680 kfree_skb(skb);
681 return -ENOMEM;
682}
683EXPORT_SYMBOL(ip_defrag);
684
685#ifdef CONFIG_SYSCTL
686static int zero;
687
688static struct ctl_table ip4_frags_ns_ctl_table[] = {
689 {
690 .procname = "ipfrag_high_thresh",
691 .data = &init_net.ipv4.frags.high_thresh,
692 .maxlen = sizeof(int),
693 .mode = 0644,
694 .proc_handler = proc_dointvec
695 },
696 {
697 .procname = "ipfrag_low_thresh",
698 .data = &init_net.ipv4.frags.low_thresh,
699 .maxlen = sizeof(int),
700 .mode = 0644,
701 .proc_handler = proc_dointvec
702 },
703 {
704 .procname = "ipfrag_time",
705 .data = &init_net.ipv4.frags.timeout,
706 .maxlen = sizeof(int),
707 .mode = 0644,
708 .proc_handler = proc_dointvec_jiffies,
709 },
710 { }
711};
712
713static struct ctl_table ip4_frags_ctl_table[] = {
714 {
715 .procname = "ipfrag_secret_interval",
716 .data = &ip4_frags.secret_interval,
717 .maxlen = sizeof(int),
718 .mode = 0644,
719 .proc_handler = proc_dointvec_jiffies,
720 },
721 {
722 .procname = "ipfrag_max_dist",
723 .data = &sysctl_ipfrag_max_dist,
724 .maxlen = sizeof(int),
725 .mode = 0644,
726 .proc_handler = proc_dointvec_minmax,
727 .extra1 = &zero
728 },
729 { }
730};
731
732static int __net_init ip4_frags_ns_ctl_register(struct net *net)
733{
734 struct ctl_table *table;
735 struct ctl_table_header *hdr;
736
737 table = ip4_frags_ns_ctl_table;
738 if (!net_eq(net, &init_net)) {
739 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
740 if (table == NULL)
741 goto err_alloc;
742
743 table[0].data = &net->ipv4.frags.high_thresh;
744 table[1].data = &net->ipv4.frags.low_thresh;
745 table[2].data = &net->ipv4.frags.timeout;
746 }
747
748 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
749 if (hdr == NULL)
750 goto err_reg;
751
752 net->ipv4.frags_hdr = hdr;
753 return 0;
754
755err_reg:
756 if (!net_eq(net, &init_net))
757 kfree(table);
758err_alloc:
759 return -ENOMEM;
760}
761
762static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
763{
764 struct ctl_table *table;
765
766 table = net->ipv4.frags_hdr->ctl_table_arg;
767 unregister_net_sysctl_table(net->ipv4.frags_hdr);
768 kfree(table);
769}
770
771static void ip4_frags_ctl_register(void)
772{
773 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
774}
775#else
776static inline int ip4_frags_ns_ctl_register(struct net *net)
777{
778 return 0;
779}
780
781static inline void ip4_frags_ns_ctl_unregister(struct net *net)
782{
783}
784
785static inline void ip4_frags_ctl_register(void)
786{
787}
788#endif
789
790static int __net_init ipv4_frags_init_net(struct net *net)
791{
792 /*
793 * Fragment cache limits. We will commit 256K at one time. Should we
794 * cross that limit we will prune down to 192K. This should cope with
795 * even the most extreme cases without allowing an attacker to
796 * measurably harm machine performance.
797 */
798 net->ipv4.frags.high_thresh = 256 * 1024;
799 net->ipv4.frags.low_thresh = 192 * 1024;
800 /*
801 * Important NOTE! Fragment queue must be destroyed before MSL expires.
802 * RFC791 is wrong proposing to prolongate timer each fragment arrival
803 * by TTL.
804 */
805 net->ipv4.frags.timeout = IP_FRAG_TIME;
806
807 inet_frags_init_net(&net->ipv4.frags);
808
809 return ip4_frags_ns_ctl_register(net);
810}
811
812static void __net_exit ipv4_frags_exit_net(struct net *net)
813{
814 ip4_frags_ns_ctl_unregister(net);
815 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
816}
817
818static struct pernet_operations ip4_frags_ops = {
819 .init = ipv4_frags_init_net,
820 .exit = ipv4_frags_exit_net,
821};
822
823void __init ipfrag_init(void)
824{
825 ip4_frags_ctl_register();
826 register_pernet_subsys(&ip4_frags_ops);
827 ip4_frags.hashfn = ip4_hashfn;
828 ip4_frags.constructor = ip4_frag_init;
829 ip4_frags.destructor = ip4_frag_free;
830 ip4_frags.skb_free = NULL;
831 ip4_frags.qsize = sizeof(struct ipq);
832 ip4_frags.match = ip4_frag_match;
833 ip4_frags.frag_expire = ip_expire;
834 ip4_frags.secret_interval = 10 * 60 * HZ;
835 inet_frags_init(&ip4_frags);
836}