Loading...
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#define pr_fmt(fmt) "IPv4: " fmt
24
25#include <linux/compiler.h>
26#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>
37#include <linux/slab.h>
38#include <net/route.h>
39#include <net/dst.h>
40#include <net/sock.h>
41#include <net/ip.h>
42#include <net/icmp.h>
43#include <net/checksum.h>
44#include <net/inetpeer.h>
45#include <net/inet_frag.h>
46#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/inet.h>
49#include <linux/netfilter_ipv4.h>
50#include <net/inet_ecn.h>
51
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
57static int sysctl_ipfrag_max_dist __read_mostly = 64;
58
59struct ipfrag_skb_cb
60{
61 struct inet_skb_parm h;
62 int offset;
63};
64
65#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
66
67/* Describe an entry in the "incomplete datagrams" queue. */
68struct ipq {
69 struct inet_frag_queue q;
70
71 u32 user;
72 __be32 saddr;
73 __be32 daddr;
74 __be16 id;
75 u8 protocol;
76 u8 ecn; /* RFC3168 support */
77 int iif;
78 unsigned int rid;
79 struct inet_peer *peer;
80};
81
82static inline u8 ip4_frag_ecn(u8 tos)
83{
84 return 1 << (tos & INET_ECN_MASK);
85}
86
87static struct inet_frags ip4_frags;
88
89int ip_frag_nqueues(struct net *net)
90{
91 return net->ipv4.frags.nqueues;
92}
93
94int ip_frag_mem(struct net *net)
95{
96 return sum_frag_mem_limit(&net->ipv4.frags);
97}
98
99static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
100 struct net_device *dev);
101
102struct ip4_create_arg {
103 struct iphdr *iph;
104 u32 user;
105};
106
107static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
108{
109 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
110 return jhash_3words((__force u32)id << 16 | prot,
111 (__force u32)saddr, (__force u32)daddr,
112 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
113}
114
115static unsigned int ip4_hashfn(struct inet_frag_queue *q)
116{
117 struct ipq *ipq;
118
119 ipq = container_of(q, struct ipq, q);
120 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
121}
122
123static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
124{
125 struct ipq *qp;
126 struct ip4_create_arg *arg = a;
127
128 qp = container_of(q, struct ipq, q);
129 return qp->id == arg->iph->id &&
130 qp->saddr == arg->iph->saddr &&
131 qp->daddr == arg->iph->daddr &&
132 qp->protocol == arg->iph->protocol &&
133 qp->user == arg->user;
134}
135
136static void ip4_frag_init(struct inet_frag_queue *q, void *a)
137{
138 struct ipq *qp = container_of(q, struct ipq, q);
139 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
140 frags);
141 struct net *net = container_of(ipv4, struct net, ipv4);
142
143 struct ip4_create_arg *arg = a;
144
145 qp->protocol = arg->iph->protocol;
146 qp->id = arg->iph->id;
147 qp->ecn = ip4_frag_ecn(arg->iph->tos);
148 qp->saddr = arg->iph->saddr;
149 qp->daddr = arg->iph->daddr;
150 qp->user = arg->user;
151 qp->peer = sysctl_ipfrag_max_dist ?
152 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
153}
154
155static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
156{
157 struct ipq *qp;
158
159 qp = container_of(q, struct ipq, q);
160 if (qp->peer)
161 inet_putpeer(qp->peer);
162}
163
164
165/* Destruction primitives. */
166
167static __inline__ void ipq_put(struct ipq *ipq)
168{
169 inet_frag_put(&ipq->q, &ip4_frags);
170}
171
172/* Kill ipq entry. It is not destroyed immediately,
173 * because caller (and someone more) holds reference count.
174 */
175static void ipq_kill(struct ipq *ipq)
176{
177 inet_frag_kill(&ipq->q, &ip4_frags);
178}
179
180/* Memory limiting on fragments. Evictor trashes the oldest
181 * fragment queue until we are back under the threshold.
182 */
183static void ip_evictor(struct net *net)
184{
185 int evicted;
186
187 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
188 if (evicted)
189 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
190}
191
192/*
193 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
194 */
195static void ip_expire(unsigned long arg)
196{
197 struct ipq *qp;
198 struct net *net;
199
200 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
201 net = container_of(qp->q.net, struct net, ipv4.frags);
202
203 spin_lock(&qp->q.lock);
204
205 if (qp->q.last_in & INET_FRAG_COMPLETE)
206 goto out;
207
208 ipq_kill(qp);
209
210 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
211 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
212
213 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
214 struct sk_buff *head = qp->q.fragments;
215 const struct iphdr *iph;
216 int err;
217
218 rcu_read_lock();
219 head->dev = dev_get_by_index_rcu(net, qp->iif);
220 if (!head->dev)
221 goto out_rcu_unlock;
222
223 /* skb has no dst, perform route lookup again */
224 iph = ip_hdr(head);
225 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
226 iph->tos, head->dev);
227 if (err)
228 goto out_rcu_unlock;
229
230 /*
231 * Only an end host needs to send an ICMP
232 * "Fragment Reassembly Timeout" message, per RFC792.
233 */
234 if (qp->user == IP_DEFRAG_AF_PACKET ||
235 ((qp->user >= IP_DEFRAG_CONNTRACK_IN) &&
236 (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) &&
237 (skb_rtable(head)->rt_type != RTN_LOCAL)))
238 goto out_rcu_unlock;
239
240
241 /* Send an ICMP "Fragment Reassembly Timeout" message. */
242 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
243out_rcu_unlock:
244 rcu_read_unlock();
245 }
246out:
247 spin_unlock(&qp->q.lock);
248 ipq_put(qp);
249}
250
251/* Find the correct entry in the "incomplete datagrams" queue for
252 * this IP datagram, and create new one, if nothing is found.
253 */
254static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
255{
256 struct inet_frag_queue *q;
257 struct ip4_create_arg arg;
258 unsigned int hash;
259
260 arg.iph = iph;
261 arg.user = user;
262
263 read_lock(&ip4_frags.lock);
264 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
265
266 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
267 if (IS_ERR_OR_NULL(q)) {
268 inet_frag_maybe_warn_overflow(q, pr_fmt());
269 return NULL;
270 }
271 return container_of(q, struct ipq, q);
272}
273
274/* Is the fragment too far ahead to be part of ipq? */
275static inline int ip_frag_too_far(struct ipq *qp)
276{
277 struct inet_peer *peer = qp->peer;
278 unsigned int max = sysctl_ipfrag_max_dist;
279 unsigned int start, end;
280
281 int rc;
282
283 if (!peer || !max)
284 return 0;
285
286 start = qp->rid;
287 end = atomic_inc_return(&peer->rid);
288 qp->rid = end;
289
290 rc = qp->q.fragments && (end - start) > max;
291
292 if (rc) {
293 struct net *net;
294
295 net = container_of(qp->q.net, struct net, ipv4.frags);
296 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
297 }
298
299 return rc;
300}
301
302static int ip_frag_reinit(struct ipq *qp)
303{
304 struct sk_buff *fp;
305 unsigned int sum_truesize = 0;
306
307 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
308 atomic_inc(&qp->q.refcnt);
309 return -ETIMEDOUT;
310 }
311
312 fp = qp->q.fragments;
313 do {
314 struct sk_buff *xp = fp->next;
315
316 sum_truesize += fp->truesize;
317 kfree_skb(fp);
318 fp = xp;
319 } while (fp);
320 sub_frag_mem_limit(&qp->q, sum_truesize);
321
322 qp->q.last_in = 0;
323 qp->q.len = 0;
324 qp->q.meat = 0;
325 qp->q.fragments = NULL;
326 qp->q.fragments_tail = NULL;
327 qp->iif = 0;
328 qp->ecn = 0;
329
330 return 0;
331}
332
333/* Add new segment to existing queue. */
334static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
335{
336 struct sk_buff *prev, *next;
337 struct net_device *dev;
338 int flags, offset;
339 int ihl, end;
340 int err = -ENOENT;
341 u8 ecn;
342
343 if (qp->q.last_in & INET_FRAG_COMPLETE)
344 goto err;
345
346 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
347 unlikely(ip_frag_too_far(qp)) &&
348 unlikely(err = ip_frag_reinit(qp))) {
349 ipq_kill(qp);
350 goto err;
351 }
352
353 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
354 offset = ntohs(ip_hdr(skb)->frag_off);
355 flags = offset & ~IP_OFFSET;
356 offset &= IP_OFFSET;
357 offset <<= 3; /* offset is in 8-byte chunks */
358 ihl = ip_hdrlen(skb);
359
360 /* Determine the position of this fragment. */
361 end = offset + skb->len - ihl;
362 err = -EINVAL;
363
364 /* Is this the final fragment? */
365 if ((flags & IP_MF) == 0) {
366 /* If we already have some bits beyond end
367 * or have different end, the segment is corrupted.
368 */
369 if (end < qp->q.len ||
370 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
371 goto err;
372 qp->q.last_in |= INET_FRAG_LAST_IN;
373 qp->q.len = end;
374 } else {
375 if (end&7) {
376 end &= ~7;
377 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
378 skb->ip_summed = CHECKSUM_NONE;
379 }
380 if (end > qp->q.len) {
381 /* Some bits beyond end -> corruption. */
382 if (qp->q.last_in & INET_FRAG_LAST_IN)
383 goto err;
384 qp->q.len = end;
385 }
386 }
387 if (end == offset)
388 goto err;
389
390 err = -ENOMEM;
391 if (pskb_pull(skb, ihl) == NULL)
392 goto err;
393
394 err = pskb_trim_rcsum(skb, end - offset);
395 if (err)
396 goto err;
397
398 /* Find out which fragments are in front and at the back of us
399 * in the chain of fragments so far. We must know where to put
400 * this fragment, right?
401 */
402 prev = qp->q.fragments_tail;
403 if (!prev || FRAG_CB(prev)->offset < offset) {
404 next = NULL;
405 goto found;
406 }
407 prev = NULL;
408 for (next = qp->q.fragments; next != NULL; next = next->next) {
409 if (FRAG_CB(next)->offset >= offset)
410 break; /* bingo! */
411 prev = next;
412 }
413
414found:
415 /* We found where to put this one. Check for overlap with
416 * preceding fragment, and, if needed, align things so that
417 * any overlaps are eliminated.
418 */
419 if (prev) {
420 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
421
422 if (i > 0) {
423 offset += i;
424 err = -EINVAL;
425 if (end <= offset)
426 goto err;
427 err = -ENOMEM;
428 if (!pskb_pull(skb, i))
429 goto err;
430 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
431 skb->ip_summed = CHECKSUM_NONE;
432 }
433 }
434
435 err = -ENOMEM;
436
437 while (next && FRAG_CB(next)->offset < end) {
438 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
439
440 if (i < next->len) {
441 /* Eat head of the next overlapped fragment
442 * and leave the loop. The next ones cannot overlap.
443 */
444 if (!pskb_pull(next, i))
445 goto err;
446 FRAG_CB(next)->offset += i;
447 qp->q.meat -= i;
448 if (next->ip_summed != CHECKSUM_UNNECESSARY)
449 next->ip_summed = CHECKSUM_NONE;
450 break;
451 } else {
452 struct sk_buff *free_it = next;
453
454 /* Old fragment is completely overridden with
455 * new one drop it.
456 */
457 next = next->next;
458
459 if (prev)
460 prev->next = next;
461 else
462 qp->q.fragments = next;
463
464 qp->q.meat -= free_it->len;
465 sub_frag_mem_limit(&qp->q, free_it->truesize);
466 kfree_skb(free_it);
467 }
468 }
469
470 FRAG_CB(skb)->offset = offset;
471
472 /* Insert this fragment in the chain of fragments. */
473 skb->next = next;
474 if (!next)
475 qp->q.fragments_tail = skb;
476 if (prev)
477 prev->next = skb;
478 else
479 qp->q.fragments = skb;
480
481 dev = skb->dev;
482 if (dev) {
483 qp->iif = dev->ifindex;
484 skb->dev = NULL;
485 }
486 qp->q.stamp = skb->tstamp;
487 qp->q.meat += skb->len;
488 qp->ecn |= ecn;
489 add_frag_mem_limit(&qp->q, skb->truesize);
490 if (offset == 0)
491 qp->q.last_in |= INET_FRAG_FIRST_IN;
492
493 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
494 skb->len + ihl > qp->q.max_size)
495 qp->q.max_size = skb->len + ihl;
496
497 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
498 qp->q.meat == qp->q.len) {
499 unsigned long orefdst = skb->_skb_refdst;
500
501 skb->_skb_refdst = 0UL;
502 err = ip_frag_reasm(qp, prev, dev);
503 skb->_skb_refdst = orefdst;
504 return err;
505 }
506
507 skb_dst_drop(skb);
508 inet_frag_lru_move(&qp->q);
509 return -EINPROGRESS;
510
511err:
512 kfree_skb(skb);
513 return err;
514}
515
516
517/* Build a new IP datagram from all its fragments. */
518
519static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
520 struct net_device *dev)
521{
522 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
523 struct iphdr *iph;
524 struct sk_buff *fp, *head = qp->q.fragments;
525 int len;
526 int ihlen;
527 int err;
528 int sum_truesize;
529 u8 ecn;
530
531 ipq_kill(qp);
532
533 ecn = ip_frag_ecn_table[qp->ecn];
534 if (unlikely(ecn == 0xff)) {
535 err = -EINVAL;
536 goto out_fail;
537 }
538 /* Make the one we just received the head. */
539 if (prev) {
540 head = prev->next;
541 fp = skb_clone(head, GFP_ATOMIC);
542 if (!fp)
543 goto out_nomem;
544
545 fp->next = head->next;
546 if (!fp->next)
547 qp->q.fragments_tail = fp;
548 prev->next = fp;
549
550 skb_morph(head, qp->q.fragments);
551 head->next = qp->q.fragments->next;
552
553 consume_skb(qp->q.fragments);
554 qp->q.fragments = head;
555 }
556
557 WARN_ON(head == NULL);
558 WARN_ON(FRAG_CB(head)->offset != 0);
559
560 /* Allocate a new buffer for the datagram. */
561 ihlen = ip_hdrlen(head);
562 len = ihlen + qp->q.len;
563
564 err = -E2BIG;
565 if (len > 65535)
566 goto out_oversize;
567
568 /* Head of list must not be cloned. */
569 if (skb_unclone(head, GFP_ATOMIC))
570 goto out_nomem;
571
572 /* If the first fragment is fragmented itself, we split
573 * it to two chunks: the first with data and paged part
574 * and the second, holding only fragments. */
575 if (skb_has_frag_list(head)) {
576 struct sk_buff *clone;
577 int i, plen = 0;
578
579 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
580 goto out_nomem;
581 clone->next = head->next;
582 head->next = clone;
583 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
584 skb_frag_list_init(head);
585 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
586 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
587 clone->len = clone->data_len = head->data_len - plen;
588 head->data_len -= clone->len;
589 head->len -= clone->len;
590 clone->csum = 0;
591 clone->ip_summed = head->ip_summed;
592 add_frag_mem_limit(&qp->q, clone->truesize);
593 }
594
595 skb_push(head, head->data - skb_network_header(head));
596
597 sum_truesize = head->truesize;
598 for (fp = head->next; fp;) {
599 bool headstolen;
600 int delta;
601 struct sk_buff *next = fp->next;
602
603 sum_truesize += fp->truesize;
604 if (head->ip_summed != fp->ip_summed)
605 head->ip_summed = CHECKSUM_NONE;
606 else if (head->ip_summed == CHECKSUM_COMPLETE)
607 head->csum = csum_add(head->csum, fp->csum);
608
609 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
610 kfree_skb_partial(fp, headstolen);
611 } else {
612 if (!skb_shinfo(head)->frag_list)
613 skb_shinfo(head)->frag_list = fp;
614 head->data_len += fp->len;
615 head->len += fp->len;
616 head->truesize += fp->truesize;
617 }
618 fp = next;
619 }
620 sub_frag_mem_limit(&qp->q, sum_truesize);
621
622 head->next = NULL;
623 head->dev = dev;
624 head->tstamp = qp->q.stamp;
625 IPCB(head)->frag_max_size = qp->q.max_size;
626
627 iph = ip_hdr(head);
628 /* max_size != 0 implies at least one fragment had IP_DF set */
629 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
630 iph->tot_len = htons(len);
631 iph->tos |= ecn;
632 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
633 qp->q.fragments = NULL;
634 qp->q.fragments_tail = NULL;
635 return 0;
636
637out_nomem:
638 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
639 qp);
640 err = -ENOMEM;
641 goto out_fail;
642out_oversize:
643 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
644out_fail:
645 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
646 return err;
647}
648
649/* Process an incoming IP datagram fragment. */
650int ip_defrag(struct sk_buff *skb, u32 user)
651{
652 struct ipq *qp;
653 struct net *net;
654
655 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
656 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
657
658 /* Start by cleaning up the memory. */
659 ip_evictor(net);
660
661 /* Lookup (or create) queue header */
662 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
663 int ret;
664
665 spin_lock(&qp->q.lock);
666
667 ret = ip_frag_queue(qp, skb);
668
669 spin_unlock(&qp->q.lock);
670 ipq_put(qp);
671 return ret;
672 }
673
674 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
675 kfree_skb(skb);
676 return -ENOMEM;
677}
678EXPORT_SYMBOL(ip_defrag);
679
680struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
681{
682 struct iphdr iph;
683 u32 len;
684
685 if (skb->protocol != htons(ETH_P_IP))
686 return skb;
687
688 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
689 return skb;
690
691 if (iph.ihl < 5 || iph.version != 4)
692 return skb;
693
694 len = ntohs(iph.tot_len);
695 if (skb->len < len || len < (iph.ihl * 4))
696 return skb;
697
698 if (ip_is_fragment(&iph)) {
699 skb = skb_share_check(skb, GFP_ATOMIC);
700 if (skb) {
701 if (!pskb_may_pull(skb, iph.ihl*4))
702 return skb;
703 if (pskb_trim_rcsum(skb, len))
704 return skb;
705 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
706 if (ip_defrag(skb, user))
707 return NULL;
708 skb_clear_hash(skb);
709 }
710 }
711 return skb;
712}
713EXPORT_SYMBOL(ip_check_defrag);
714
715#ifdef CONFIG_SYSCTL
716static int zero;
717
718static struct ctl_table ip4_frags_ns_ctl_table[] = {
719 {
720 .procname = "ipfrag_high_thresh",
721 .data = &init_net.ipv4.frags.high_thresh,
722 .maxlen = sizeof(int),
723 .mode = 0644,
724 .proc_handler = proc_dointvec
725 },
726 {
727 .procname = "ipfrag_low_thresh",
728 .data = &init_net.ipv4.frags.low_thresh,
729 .maxlen = sizeof(int),
730 .mode = 0644,
731 .proc_handler = proc_dointvec
732 },
733 {
734 .procname = "ipfrag_time",
735 .data = &init_net.ipv4.frags.timeout,
736 .maxlen = sizeof(int),
737 .mode = 0644,
738 .proc_handler = proc_dointvec_jiffies,
739 },
740 { }
741};
742
743static struct ctl_table ip4_frags_ctl_table[] = {
744 {
745 .procname = "ipfrag_secret_interval",
746 .data = &ip4_frags.secret_interval,
747 .maxlen = sizeof(int),
748 .mode = 0644,
749 .proc_handler = proc_dointvec_jiffies,
750 },
751 {
752 .procname = "ipfrag_max_dist",
753 .data = &sysctl_ipfrag_max_dist,
754 .maxlen = sizeof(int),
755 .mode = 0644,
756 .proc_handler = proc_dointvec_minmax,
757 .extra1 = &zero
758 },
759 { }
760};
761
762static int __net_init ip4_frags_ns_ctl_register(struct net *net)
763{
764 struct ctl_table *table;
765 struct ctl_table_header *hdr;
766
767 table = ip4_frags_ns_ctl_table;
768 if (!net_eq(net, &init_net)) {
769 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
770 if (table == NULL)
771 goto err_alloc;
772
773 table[0].data = &net->ipv4.frags.high_thresh;
774 table[1].data = &net->ipv4.frags.low_thresh;
775 table[2].data = &net->ipv4.frags.timeout;
776
777 /* Don't export sysctls to unprivileged users */
778 if (net->user_ns != &init_user_ns)
779 table[0].procname = NULL;
780 }
781
782 hdr = register_net_sysctl(net, "net/ipv4", table);
783 if (hdr == NULL)
784 goto err_reg;
785
786 net->ipv4.frags_hdr = hdr;
787 return 0;
788
789err_reg:
790 if (!net_eq(net, &init_net))
791 kfree(table);
792err_alloc:
793 return -ENOMEM;
794}
795
796static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
797{
798 struct ctl_table *table;
799
800 table = net->ipv4.frags_hdr->ctl_table_arg;
801 unregister_net_sysctl_table(net->ipv4.frags_hdr);
802 kfree(table);
803}
804
805static void ip4_frags_ctl_register(void)
806{
807 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
808}
809#else
810static inline int ip4_frags_ns_ctl_register(struct net *net)
811{
812 return 0;
813}
814
815static inline void ip4_frags_ns_ctl_unregister(struct net *net)
816{
817}
818
819static inline void ip4_frags_ctl_register(void)
820{
821}
822#endif
823
824static int __net_init ipv4_frags_init_net(struct net *net)
825{
826 /* Fragment cache limits.
827 *
828 * The fragment memory accounting code, (tries to) account for
829 * the real memory usage, by measuring both the size of frag
830 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
831 * and the SKB's truesize.
832 *
833 * A 64K fragment consumes 129736 bytes (44*2944)+200
834 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
835 *
836 * We will commit 4MB at one time. Should we cross that limit
837 * we will prune down to 3MB, making room for approx 8 big 64K
838 * fragments 8x128k.
839 */
840 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
841 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
842 /*
843 * Important NOTE! Fragment queue must be destroyed before MSL expires.
844 * RFC791 is wrong proposing to prolongate timer each fragment arrival
845 * by TTL.
846 */
847 net->ipv4.frags.timeout = IP_FRAG_TIME;
848
849 inet_frags_init_net(&net->ipv4.frags);
850
851 return ip4_frags_ns_ctl_register(net);
852}
853
854static void __net_exit ipv4_frags_exit_net(struct net *net)
855{
856 ip4_frags_ns_ctl_unregister(net);
857 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
858}
859
860static struct pernet_operations ip4_frags_ops = {
861 .init = ipv4_frags_init_net,
862 .exit = ipv4_frags_exit_net,
863};
864
865void __init ipfrag_init(void)
866{
867 ip4_frags_ctl_register();
868 register_pernet_subsys(&ip4_frags_ops);
869 ip4_frags.hashfn = ip4_hashfn;
870 ip4_frags.constructor = ip4_frag_init;
871 ip4_frags.destructor = ip4_frag_free;
872 ip4_frags.skb_free = NULL;
873 ip4_frags.qsize = sizeof(struct ipq);
874 ip4_frags.match = ip4_frag_match;
875 ip4_frags.frag_expire = ip_expire;
876 ip4_frags.secret_interval = 10 * 60 * HZ;
877 inet_frags_init(&ip4_frags);
878}
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#define pr_fmt(fmt) "IPv4: " fmt
24
25#include <linux/compiler.h>
26#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>
37#include <linux/slab.h>
38#include <net/route.h>
39#include <net/dst.h>
40#include <net/sock.h>
41#include <net/ip.h>
42#include <net/icmp.h>
43#include <net/checksum.h>
44#include <net/inetpeer.h>
45#include <net/inet_frag.h>
46#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/inet.h>
49#include <linux/netfilter_ipv4.h>
50#include <net/inet_ecn.h>
51
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
57static int sysctl_ipfrag_max_dist __read_mostly = 64;
58
59struct ipfrag_skb_cb
60{
61 struct inet_skb_parm h;
62 int offset;
63};
64
65#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
66
67/* Describe an entry in the "incomplete datagrams" queue. */
68struct ipq {
69 struct inet_frag_queue q;
70
71 u32 user;
72 __be32 saddr;
73 __be32 daddr;
74 __be16 id;
75 u8 protocol;
76 u8 ecn; /* RFC3168 support */
77 int iif;
78 unsigned int rid;
79 struct inet_peer *peer;
80};
81
82/* 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 */
86#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 */
90
91static inline u8 ip4_frag_ecn(u8 tos)
92{
93 return 1 << (tos & INET_ECN_MASK);
94}
95
96/* 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
116static struct inet_frags ip4_frags;
117
118int ip_frag_nqueues(struct net *net)
119{
120 return net->ipv4.frags.nqueues;
121}
122
123int ip_frag_mem(struct net *net)
124{
125 return atomic_read(&net->ipv4.frags.mem);
126}
127
128static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
129 struct net_device *dev);
130
131struct ip4_create_arg {
132 struct iphdr *iph;
133 u32 user;
134};
135
136static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
137{
138 return jhash_3words((__force u32)id << 16 | prot,
139 (__force u32)saddr, (__force u32)daddr,
140 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
141}
142
143static unsigned int ip4_hashfn(struct inet_frag_queue *q)
144{
145 struct ipq *ipq;
146
147 ipq = container_of(q, struct ipq, q);
148 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
149}
150
151static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
152{
153 struct ipq *qp;
154 struct ip4_create_arg *arg = a;
155
156 qp = container_of(q, struct ipq, q);
157 return qp->id == arg->iph->id &&
158 qp->saddr == arg->iph->saddr &&
159 qp->daddr == arg->iph->daddr &&
160 qp->protocol == arg->iph->protocol &&
161 qp->user == arg->user;
162}
163
164/* Memory Tracking Functions. */
165static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
166{
167 atomic_sub(skb->truesize, &nf->mem);
168 kfree_skb(skb);
169}
170
171static void ip4_frag_init(struct inet_frag_queue *q, void *a)
172{
173 struct ipq *qp = container_of(q, struct ipq, q);
174 struct ip4_create_arg *arg = a;
175
176 qp->protocol = arg->iph->protocol;
177 qp->id = arg->iph->id;
178 qp->ecn = ip4_frag_ecn(arg->iph->tos);
179 qp->saddr = arg->iph->saddr;
180 qp->daddr = arg->iph->daddr;
181 qp->user = arg->user;
182 qp->peer = sysctl_ipfrag_max_dist ?
183 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
184}
185
186static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
187{
188 struct ipq *qp;
189
190 qp = container_of(q, struct ipq, q);
191 if (qp->peer)
192 inet_putpeer(qp->peer);
193}
194
195
196/* Destruction primitives. */
197
198static __inline__ void ipq_put(struct ipq *ipq)
199{
200 inet_frag_put(&ipq->q, &ip4_frags);
201}
202
203/* Kill ipq entry. It is not destroyed immediately,
204 * because caller (and someone more) holds reference count.
205 */
206static void ipq_kill(struct ipq *ipq)
207{
208 inet_frag_kill(&ipq->q, &ip4_frags);
209}
210
211/* Memory limiting on fragments. Evictor trashes the oldest
212 * fragment queue until we are back under the threshold.
213 */
214static void ip_evictor(struct net *net)
215{
216 int evicted;
217
218 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
219 if (evicted)
220 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
221}
222
223/*
224 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
225 */
226static void ip_expire(unsigned long arg)
227{
228 struct ipq *qp;
229 struct net *net;
230
231 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
232 net = container_of(qp->q.net, struct net, ipv4.frags);
233
234 spin_lock(&qp->q.lock);
235
236 if (qp->q.last_in & INET_FRAG_COMPLETE)
237 goto out;
238
239 ipq_kill(qp);
240
241 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
242 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
243
244 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
245 struct sk_buff *head = qp->q.fragments;
246 const struct iphdr *iph;
247 int err;
248
249 rcu_read_lock();
250 head->dev = dev_get_by_index_rcu(net, qp->iif);
251 if (!head->dev)
252 goto out_rcu_unlock;
253
254 /* skb dst is stale, drop it, and perform route lookup again */
255 skb_dst_drop(head);
256 iph = ip_hdr(head);
257 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
258 iph->tos, head->dev);
259 if (err)
260 goto out_rcu_unlock;
261
262 /*
263 * Only an end host needs to send an ICMP
264 * "Fragment Reassembly Timeout" message, per RFC792.
265 */
266 if (qp->user == IP_DEFRAG_AF_PACKET ||
267 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
268 skb_rtable(head)->rt_type != RTN_LOCAL))
269 goto out_rcu_unlock;
270
271
272 /* Send an ICMP "Fragment Reassembly Timeout" message. */
273 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
274out_rcu_unlock:
275 rcu_read_unlock();
276 }
277out:
278 spin_unlock(&qp->q.lock);
279 ipq_put(qp);
280}
281
282/* Find the correct entry in the "incomplete datagrams" queue for
283 * this IP datagram, and create new one, if nothing is found.
284 */
285static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
286{
287 struct inet_frag_queue *q;
288 struct ip4_create_arg arg;
289 unsigned int hash;
290
291 arg.iph = iph;
292 arg.user = user;
293
294 read_lock(&ip4_frags.lock);
295 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
296
297 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
298 if (q == NULL)
299 goto out_nomem;
300
301 return container_of(q, struct ipq, q);
302
303out_nomem:
304 LIMIT_NETDEBUG(KERN_ERR pr_fmt("ip_frag_create: no memory left !\n"));
305 return NULL;
306}
307
308/* Is the fragment too far ahead to be part of ipq? */
309static inline int ip_frag_too_far(struct ipq *qp)
310{
311 struct inet_peer *peer = qp->peer;
312 unsigned int max = sysctl_ipfrag_max_dist;
313 unsigned int start, end;
314
315 int rc;
316
317 if (!peer || !max)
318 return 0;
319
320 start = qp->rid;
321 end = atomic_inc_return(&peer->rid);
322 qp->rid = end;
323
324 rc = qp->q.fragments && (end - start) > max;
325
326 if (rc) {
327 struct net *net;
328
329 net = container_of(qp->q.net, struct net, ipv4.frags);
330 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
331 }
332
333 return rc;
334}
335
336static int ip_frag_reinit(struct ipq *qp)
337{
338 struct sk_buff *fp;
339
340 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
341 atomic_inc(&qp->q.refcnt);
342 return -ETIMEDOUT;
343 }
344
345 fp = qp->q.fragments;
346 do {
347 struct sk_buff *xp = fp->next;
348 frag_kfree_skb(qp->q.net, fp);
349 fp = xp;
350 } while (fp);
351
352 qp->q.last_in = 0;
353 qp->q.len = 0;
354 qp->q.meat = 0;
355 qp->q.fragments = NULL;
356 qp->q.fragments_tail = NULL;
357 qp->iif = 0;
358 qp->ecn = 0;
359
360 return 0;
361}
362
363/* Add new segment to existing queue. */
364static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
365{
366 struct sk_buff *prev, *next;
367 struct net_device *dev;
368 int flags, offset;
369 int ihl, end;
370 int err = -ENOENT;
371 u8 ecn;
372
373 if (qp->q.last_in & INET_FRAG_COMPLETE)
374 goto err;
375
376 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
377 unlikely(ip_frag_too_far(qp)) &&
378 unlikely(err = ip_frag_reinit(qp))) {
379 ipq_kill(qp);
380 goto err;
381 }
382
383 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
384 offset = ntohs(ip_hdr(skb)->frag_off);
385 flags = offset & ~IP_OFFSET;
386 offset &= IP_OFFSET;
387 offset <<= 3; /* offset is in 8-byte chunks */
388 ihl = ip_hdrlen(skb);
389
390 /* Determine the position of this fragment. */
391 end = offset + skb->len - ihl;
392 err = -EINVAL;
393
394 /* Is this the final fragment? */
395 if ((flags & IP_MF) == 0) {
396 /* If we already have some bits beyond end
397 * or have different end, the segment is corrupted.
398 */
399 if (end < qp->q.len ||
400 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
401 goto err;
402 qp->q.last_in |= INET_FRAG_LAST_IN;
403 qp->q.len = end;
404 } else {
405 if (end&7) {
406 end &= ~7;
407 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
408 skb->ip_summed = CHECKSUM_NONE;
409 }
410 if (end > qp->q.len) {
411 /* Some bits beyond end -> corruption. */
412 if (qp->q.last_in & INET_FRAG_LAST_IN)
413 goto err;
414 qp->q.len = end;
415 }
416 }
417 if (end == offset)
418 goto err;
419
420 err = -ENOMEM;
421 if (pskb_pull(skb, ihl) == NULL)
422 goto err;
423
424 err = pskb_trim_rcsum(skb, end - offset);
425 if (err)
426 goto err;
427
428 /* Find out which fragments are in front and at the back of us
429 * in the chain of fragments so far. We must know where to put
430 * this fragment, right?
431 */
432 prev = qp->q.fragments_tail;
433 if (!prev || FRAG_CB(prev)->offset < offset) {
434 next = NULL;
435 goto found;
436 }
437 prev = NULL;
438 for (next = qp->q.fragments; next != NULL; next = next->next) {
439 if (FRAG_CB(next)->offset >= offset)
440 break; /* bingo! */
441 prev = next;
442 }
443
444found:
445 /* We found where to put this one. Check for overlap with
446 * preceding fragment, and, if needed, align things so that
447 * any overlaps are eliminated.
448 */
449 if (prev) {
450 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
451
452 if (i > 0) {
453 offset += i;
454 err = -EINVAL;
455 if (end <= offset)
456 goto err;
457 err = -ENOMEM;
458 if (!pskb_pull(skb, i))
459 goto err;
460 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
461 skb->ip_summed = CHECKSUM_NONE;
462 }
463 }
464
465 err = -ENOMEM;
466
467 while (next && FRAG_CB(next)->offset < end) {
468 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
469
470 if (i < next->len) {
471 /* Eat head of the next overlapped fragment
472 * and leave the loop. The next ones cannot overlap.
473 */
474 if (!pskb_pull(next, i))
475 goto err;
476 FRAG_CB(next)->offset += i;
477 qp->q.meat -= i;
478 if (next->ip_summed != CHECKSUM_UNNECESSARY)
479 next->ip_summed = CHECKSUM_NONE;
480 break;
481 } else {
482 struct sk_buff *free_it = next;
483
484 /* Old fragment is completely overridden with
485 * new one drop it.
486 */
487 next = next->next;
488
489 if (prev)
490 prev->next = next;
491 else
492 qp->q.fragments = next;
493
494 qp->q.meat -= free_it->len;
495 frag_kfree_skb(qp->q.net, free_it);
496 }
497 }
498
499 FRAG_CB(skb)->offset = offset;
500
501 /* Insert this fragment in the chain of fragments. */
502 skb->next = next;
503 if (!next)
504 qp->q.fragments_tail = skb;
505 if (prev)
506 prev->next = skb;
507 else
508 qp->q.fragments = skb;
509
510 dev = skb->dev;
511 if (dev) {
512 qp->iif = dev->ifindex;
513 skb->dev = NULL;
514 }
515 qp->q.stamp = skb->tstamp;
516 qp->q.meat += skb->len;
517 qp->ecn |= ecn;
518 atomic_add(skb->truesize, &qp->q.net->mem);
519 if (offset == 0)
520 qp->q.last_in |= INET_FRAG_FIRST_IN;
521
522 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
523 qp->q.meat == qp->q.len)
524 return ip_frag_reasm(qp, prev, dev);
525
526 write_lock(&ip4_frags.lock);
527 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
528 write_unlock(&ip4_frags.lock);
529 return -EINPROGRESS;
530
531err:
532 kfree_skb(skb);
533 return err;
534}
535
536
537/* Build a new IP datagram from all its fragments. */
538
539static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
540 struct net_device *dev)
541{
542 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
543 struct iphdr *iph;
544 struct sk_buff *fp, *head = qp->q.fragments;
545 int len;
546 int ihlen;
547 int err;
548 int sum_truesize;
549 u8 ecn;
550
551 ipq_kill(qp);
552
553 ecn = ip4_frag_ecn_table[qp->ecn];
554 if (unlikely(ecn == 0xff)) {
555 err = -EINVAL;
556 goto out_fail;
557 }
558 /* Make the one we just received the head. */
559 if (prev) {
560 head = prev->next;
561 fp = skb_clone(head, GFP_ATOMIC);
562 if (!fp)
563 goto out_nomem;
564
565 fp->next = head->next;
566 if (!fp->next)
567 qp->q.fragments_tail = fp;
568 prev->next = fp;
569
570 skb_morph(head, qp->q.fragments);
571 head->next = qp->q.fragments->next;
572
573 consume_skb(qp->q.fragments);
574 qp->q.fragments = head;
575 }
576
577 WARN_ON(head == NULL);
578 WARN_ON(FRAG_CB(head)->offset != 0);
579
580 /* Allocate a new buffer for the datagram. */
581 ihlen = ip_hdrlen(head);
582 len = ihlen + qp->q.len;
583
584 err = -E2BIG;
585 if (len > 65535)
586 goto out_oversize;
587
588 /* Head of list must not be cloned. */
589 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
590 goto out_nomem;
591
592 /* If the first fragment is fragmented itself, we split
593 * it to two chunks: the first with data and paged part
594 * and the second, holding only fragments. */
595 if (skb_has_frag_list(head)) {
596 struct sk_buff *clone;
597 int i, plen = 0;
598
599 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
600 goto out_nomem;
601 clone->next = head->next;
602 head->next = clone;
603 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
604 skb_frag_list_init(head);
605 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
606 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
607 clone->len = clone->data_len = head->data_len - plen;
608 head->data_len -= clone->len;
609 head->len -= clone->len;
610 clone->csum = 0;
611 clone->ip_summed = head->ip_summed;
612 atomic_add(clone->truesize, &qp->q.net->mem);
613 }
614
615 skb_push(head, head->data - skb_network_header(head));
616
617 sum_truesize = head->truesize;
618 for (fp = head->next; fp;) {
619 bool headstolen;
620 int delta;
621 struct sk_buff *next = fp->next;
622
623 sum_truesize += fp->truesize;
624 if (head->ip_summed != fp->ip_summed)
625 head->ip_summed = CHECKSUM_NONE;
626 else if (head->ip_summed == CHECKSUM_COMPLETE)
627 head->csum = csum_add(head->csum, fp->csum);
628
629 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
630 kfree_skb_partial(fp, headstolen);
631 } else {
632 if (!skb_shinfo(head)->frag_list)
633 skb_shinfo(head)->frag_list = fp;
634 head->data_len += fp->len;
635 head->len += fp->len;
636 head->truesize += fp->truesize;
637 }
638 fp = next;
639 }
640 atomic_sub(sum_truesize, &qp->q.net->mem);
641
642 head->next = NULL;
643 head->dev = dev;
644 head->tstamp = qp->q.stamp;
645
646 iph = ip_hdr(head);
647 iph->frag_off = 0;
648 iph->tot_len = htons(len);
649 iph->tos |= ecn;
650 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
651 qp->q.fragments = NULL;
652 qp->q.fragments_tail = NULL;
653 return 0;
654
655out_nomem:
656 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
657 qp);
658 err = -ENOMEM;
659 goto out_fail;
660out_oversize:
661 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
662out_fail:
663 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
664 return err;
665}
666
667/* Process an incoming IP datagram fragment. */
668int ip_defrag(struct sk_buff *skb, u32 user)
669{
670 struct ipq *qp;
671 struct net *net;
672
673 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
674 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
675
676 /* Start by cleaning up the memory. */
677 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
678 ip_evictor(net);
679
680 /* Lookup (or create) queue header */
681 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
682 int ret;
683
684 spin_lock(&qp->q.lock);
685
686 ret = ip_frag_queue(qp, skb);
687
688 spin_unlock(&qp->q.lock);
689 ipq_put(qp);
690 return ret;
691 }
692
693 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
694 kfree_skb(skb);
695 return -ENOMEM;
696}
697EXPORT_SYMBOL(ip_defrag);
698
699struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
700{
701 const struct iphdr *iph;
702 u32 len;
703
704 if (skb->protocol != htons(ETH_P_IP))
705 return skb;
706
707 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
708 return skb;
709
710 iph = ip_hdr(skb);
711 if (iph->ihl < 5 || iph->version != 4)
712 return skb;
713 if (!pskb_may_pull(skb, iph->ihl*4))
714 return skb;
715 iph = ip_hdr(skb);
716 len = ntohs(iph->tot_len);
717 if (skb->len < len || len < (iph->ihl * 4))
718 return skb;
719
720 if (ip_is_fragment(ip_hdr(skb))) {
721 skb = skb_share_check(skb, GFP_ATOMIC);
722 if (skb) {
723 if (pskb_trim_rcsum(skb, len))
724 return skb;
725 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
726 if (ip_defrag(skb, user))
727 return NULL;
728 skb->rxhash = 0;
729 }
730 }
731 return skb;
732}
733EXPORT_SYMBOL(ip_check_defrag);
734
735#ifdef CONFIG_SYSCTL
736static int zero;
737
738static struct ctl_table ip4_frags_ns_ctl_table[] = {
739 {
740 .procname = "ipfrag_high_thresh",
741 .data = &init_net.ipv4.frags.high_thresh,
742 .maxlen = sizeof(int),
743 .mode = 0644,
744 .proc_handler = proc_dointvec
745 },
746 {
747 .procname = "ipfrag_low_thresh",
748 .data = &init_net.ipv4.frags.low_thresh,
749 .maxlen = sizeof(int),
750 .mode = 0644,
751 .proc_handler = proc_dointvec
752 },
753 {
754 .procname = "ipfrag_time",
755 .data = &init_net.ipv4.frags.timeout,
756 .maxlen = sizeof(int),
757 .mode = 0644,
758 .proc_handler = proc_dointvec_jiffies,
759 },
760 { }
761};
762
763static struct ctl_table ip4_frags_ctl_table[] = {
764 {
765 .procname = "ipfrag_secret_interval",
766 .data = &ip4_frags.secret_interval,
767 .maxlen = sizeof(int),
768 .mode = 0644,
769 .proc_handler = proc_dointvec_jiffies,
770 },
771 {
772 .procname = "ipfrag_max_dist",
773 .data = &sysctl_ipfrag_max_dist,
774 .maxlen = sizeof(int),
775 .mode = 0644,
776 .proc_handler = proc_dointvec_minmax,
777 .extra1 = &zero
778 },
779 { }
780};
781
782static int __net_init ip4_frags_ns_ctl_register(struct net *net)
783{
784 struct ctl_table *table;
785 struct ctl_table_header *hdr;
786
787 table = ip4_frags_ns_ctl_table;
788 if (!net_eq(net, &init_net)) {
789 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
790 if (table == NULL)
791 goto err_alloc;
792
793 table[0].data = &net->ipv4.frags.high_thresh;
794 table[1].data = &net->ipv4.frags.low_thresh;
795 table[2].data = &net->ipv4.frags.timeout;
796 }
797
798 hdr = register_net_sysctl(net, "net/ipv4", table);
799 if (hdr == NULL)
800 goto err_reg;
801
802 net->ipv4.frags_hdr = hdr;
803 return 0;
804
805err_reg:
806 if (!net_eq(net, &init_net))
807 kfree(table);
808err_alloc:
809 return -ENOMEM;
810}
811
812static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
813{
814 struct ctl_table *table;
815
816 table = net->ipv4.frags_hdr->ctl_table_arg;
817 unregister_net_sysctl_table(net->ipv4.frags_hdr);
818 kfree(table);
819}
820
821static void ip4_frags_ctl_register(void)
822{
823 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
824}
825#else
826static inline int ip4_frags_ns_ctl_register(struct net *net)
827{
828 return 0;
829}
830
831static inline void ip4_frags_ns_ctl_unregister(struct net *net)
832{
833}
834
835static inline void ip4_frags_ctl_register(void)
836{
837}
838#endif
839
840static int __net_init ipv4_frags_init_net(struct net *net)
841{
842 /*
843 * Fragment cache limits. We will commit 256K at one time. Should we
844 * cross that limit we will prune down to 192K. This should cope with
845 * even the most extreme cases without allowing an attacker to
846 * measurably harm machine performance.
847 */
848 net->ipv4.frags.high_thresh = 256 * 1024;
849 net->ipv4.frags.low_thresh = 192 * 1024;
850 /*
851 * Important NOTE! Fragment queue must be destroyed before MSL expires.
852 * RFC791 is wrong proposing to prolongate timer each fragment arrival
853 * by TTL.
854 */
855 net->ipv4.frags.timeout = IP_FRAG_TIME;
856
857 inet_frags_init_net(&net->ipv4.frags);
858
859 return ip4_frags_ns_ctl_register(net);
860}
861
862static void __net_exit ipv4_frags_exit_net(struct net *net)
863{
864 ip4_frags_ns_ctl_unregister(net);
865 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
866}
867
868static struct pernet_operations ip4_frags_ops = {
869 .init = ipv4_frags_init_net,
870 .exit = ipv4_frags_exit_net,
871};
872
873void __init ipfrag_init(void)
874{
875 ip4_frags_ctl_register();
876 register_pernet_subsys(&ip4_frags_ops);
877 ip4_frags.hashfn = ip4_hashfn;
878 ip4_frags.constructor = ip4_frag_init;
879 ip4_frags.destructor = ip4_frag_free;
880 ip4_frags.skb_free = NULL;
881 ip4_frags.qsize = sizeof(struct ipq);
882 ip4_frags.match = ip4_frag_match;
883 ip4_frags.frag_expire = ip_expire;
884 ip4_frags.secret_interval = 10 * 60 * HZ;
885 inet_frags_init(&ip4_frags);
886}