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 *skb,
79 struct sk_buff *prev_tail, 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 const struct frag_v4_compare_key *key = a;
86 struct net *net = q->fqdir->net;
87 struct inet_peer *p = NULL;
88
89 q->key.v4 = *key;
90 qp->ecn = 0;
91 if (q->fqdir->max_dist) {
92 rcu_read_lock();
93 p = inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif);
94 if (p && !refcount_inc_not_zero(&p->refcnt))
95 p = NULL;
96 rcu_read_unlock();
97 }
98 qp->peer = p;
99}
100
101static void ip4_frag_free(struct inet_frag_queue *q)
102{
103 struct ipq *qp;
104
105 qp = container_of(q, struct ipq, q);
106 if (qp->peer)
107 inet_putpeer(qp->peer);
108}
109
110
111/* Destruction primitives. */
112
113static void ipq_put(struct ipq *ipq)
114{
115 inet_frag_put(&ipq->q);
116}
117
118/* Kill ipq entry. It is not destroyed immediately,
119 * because caller (and someone more) holds reference count.
120 */
121static void ipq_kill(struct ipq *ipq)
122{
123 inet_frag_kill(&ipq->q);
124}
125
126static bool frag_expire_skip_icmp(u32 user)
127{
128 return user == IP_DEFRAG_AF_PACKET ||
129 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
130 __IP_DEFRAG_CONNTRACK_IN_END) ||
131 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
132 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
133}
134
135/*
136 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
137 */
138static void ip_expire(struct timer_list *t)
139{
140 enum skb_drop_reason reason = SKB_DROP_REASON_FRAG_REASM_TIMEOUT;
141 struct inet_frag_queue *frag = from_timer(frag, t, timer);
142 const struct iphdr *iph;
143 struct sk_buff *head = NULL;
144 struct net *net;
145 struct ipq *qp;
146
147 qp = container_of(frag, struct ipq, q);
148 net = qp->q.fqdir->net;
149
150 rcu_read_lock();
151
152 /* Paired with WRITE_ONCE() in fqdir_pre_exit(). */
153 if (READ_ONCE(qp->q.fqdir->dead))
154 goto out_rcu_unlock;
155
156 spin_lock(&qp->q.lock);
157
158 if (qp->q.flags & INET_FRAG_COMPLETE)
159 goto out;
160
161 qp->q.flags |= INET_FRAG_DROP;
162 ipq_kill(qp);
163 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
164 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
165
166 if (!(qp->q.flags & INET_FRAG_FIRST_IN))
167 goto out;
168
169 /* sk_buff::dev and sk_buff::rbnode are unionized. So we
170 * pull the head out of the tree in order to be able to
171 * deal with head->dev.
172 */
173 head = inet_frag_pull_head(&qp->q);
174 if (!head)
175 goto out;
176 head->dev = dev_get_by_index_rcu(net, qp->iif);
177 if (!head->dev)
178 goto out;
179
180
181 /* skb has no dst, perform route lookup again */
182 iph = ip_hdr(head);
183 reason = ip_route_input_noref(head, iph->daddr, iph->saddr,
184 ip4h_dscp(iph), head->dev);
185 if (reason)
186 goto out;
187
188 /* Only an end host needs to send an ICMP
189 * "Fragment Reassembly Timeout" message, per RFC792.
190 */
191 reason = SKB_DROP_REASON_FRAG_REASM_TIMEOUT;
192 if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
193 (skb_rtable(head)->rt_type != RTN_LOCAL))
194 goto out;
195
196 spin_unlock(&qp->q.lock);
197 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
198 goto out_rcu_unlock;
199
200out:
201 spin_unlock(&qp->q.lock);
202out_rcu_unlock:
203 rcu_read_unlock();
204 kfree_skb_reason(head, reason);
205 ipq_put(qp);
206}
207
208/* Find the correct entry in the "incomplete datagrams" queue for
209 * this IP datagram, and create new one, if nothing is found.
210 */
211static struct ipq *ip_find(struct net *net, struct iphdr *iph,
212 u32 user, int vif)
213{
214 struct frag_v4_compare_key key = {
215 .saddr = iph->saddr,
216 .daddr = iph->daddr,
217 .user = user,
218 .vif = vif,
219 .id = iph->id,
220 .protocol = iph->protocol,
221 };
222 struct inet_frag_queue *q;
223
224 q = inet_frag_find(net->ipv4.fqdir, &key);
225 if (!q)
226 return NULL;
227
228 return container_of(q, struct ipq, q);
229}
230
231/* Is the fragment too far ahead to be part of ipq? */
232static int ip_frag_too_far(struct ipq *qp)
233{
234 struct inet_peer *peer = qp->peer;
235 unsigned int max = qp->q.fqdir->max_dist;
236 unsigned int start, end;
237
238 int rc;
239
240 if (!peer || !max)
241 return 0;
242
243 start = qp->rid;
244 end = atomic_inc_return(&peer->rid);
245 qp->rid = end;
246
247 rc = qp->q.fragments_tail && (end - start) > max;
248
249 if (rc)
250 __IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
251
252 return rc;
253}
254
255static int ip_frag_reinit(struct ipq *qp)
256{
257 unsigned int sum_truesize = 0;
258
259 if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
260 refcount_inc(&qp->q.refcnt);
261 return -ETIMEDOUT;
262 }
263
264 sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments,
265 SKB_DROP_REASON_FRAG_TOO_FAR);
266 sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
267
268 qp->q.flags = 0;
269 qp->q.len = 0;
270 qp->q.meat = 0;
271 qp->q.rb_fragments = RB_ROOT;
272 qp->q.fragments_tail = NULL;
273 qp->q.last_run_head = NULL;
274 qp->iif = 0;
275 qp->ecn = 0;
276
277 return 0;
278}
279
280/* Add new segment to existing queue. */
281static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
282{
283 struct net *net = qp->q.fqdir->net;
284 int ihl, end, flags, offset;
285 struct sk_buff *prev_tail;
286 struct net_device *dev;
287 unsigned int fragsize;
288 int err = -ENOENT;
289 SKB_DR(reason);
290 u8 ecn;
291
292 /* If reassembly is already done, @skb must be a duplicate frag. */
293 if (qp->q.flags & INET_FRAG_COMPLETE) {
294 SKB_DR_SET(reason, DUP_FRAG);
295 goto err;
296 }
297
298 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
299 unlikely(ip_frag_too_far(qp)) &&
300 unlikely(err = ip_frag_reinit(qp))) {
301 ipq_kill(qp);
302 goto err;
303 }
304
305 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
306 offset = ntohs(ip_hdr(skb)->frag_off);
307 flags = offset & ~IP_OFFSET;
308 offset &= IP_OFFSET;
309 offset <<= 3; /* offset is in 8-byte chunks */
310 ihl = ip_hdrlen(skb);
311
312 /* Determine the position of this fragment. */
313 end = offset + skb->len - skb_network_offset(skb) - ihl;
314 err = -EINVAL;
315
316 /* Is this the final fragment? */
317 if ((flags & IP_MF) == 0) {
318 /* If we already have some bits beyond end
319 * or have different end, the segment is corrupted.
320 */
321 if (end < qp->q.len ||
322 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
323 goto discard_qp;
324 qp->q.flags |= INET_FRAG_LAST_IN;
325 qp->q.len = end;
326 } else {
327 if (end&7) {
328 end &= ~7;
329 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
330 skb->ip_summed = CHECKSUM_NONE;
331 }
332 if (end > qp->q.len) {
333 /* Some bits beyond end -> corruption. */
334 if (qp->q.flags & INET_FRAG_LAST_IN)
335 goto discard_qp;
336 qp->q.len = end;
337 }
338 }
339 if (end == offset)
340 goto discard_qp;
341
342 err = -ENOMEM;
343 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
344 goto discard_qp;
345
346 err = pskb_trim_rcsum(skb, end - offset);
347 if (err)
348 goto discard_qp;
349
350 /* Note : skb->rbnode and skb->dev share the same location. */
351 dev = skb->dev;
352 /* Makes sure compiler wont do silly aliasing games */
353 barrier();
354
355 prev_tail = qp->q.fragments_tail;
356 err = inet_frag_queue_insert(&qp->q, skb, offset, end);
357 if (err)
358 goto insert_error;
359
360 if (dev)
361 qp->iif = dev->ifindex;
362
363 qp->q.stamp = skb->tstamp;
364 qp->q.tstamp_type = skb->tstamp_type;
365 qp->q.meat += skb->len;
366 qp->ecn |= ecn;
367 add_frag_mem_limit(qp->q.fqdir, skb->truesize);
368 if (offset == 0)
369 qp->q.flags |= INET_FRAG_FIRST_IN;
370
371 fragsize = skb->len + ihl;
372
373 if (fragsize > qp->q.max_size)
374 qp->q.max_size = fragsize;
375
376 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
377 fragsize > qp->max_df_size)
378 qp->max_df_size = fragsize;
379
380 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
381 qp->q.meat == qp->q.len) {
382 unsigned long orefdst = skb->_skb_refdst;
383
384 skb->_skb_refdst = 0UL;
385 err = ip_frag_reasm(qp, skb, prev_tail, dev);
386 skb->_skb_refdst = orefdst;
387 if (err)
388 inet_frag_kill(&qp->q);
389 return err;
390 }
391
392 skb_dst_drop(skb);
393 skb_orphan(skb);
394 return -EINPROGRESS;
395
396insert_error:
397 if (err == IPFRAG_DUP) {
398 SKB_DR_SET(reason, DUP_FRAG);
399 err = -EINVAL;
400 goto err;
401 }
402 err = -EINVAL;
403 __IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
404discard_qp:
405 inet_frag_kill(&qp->q);
406 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
407err:
408 kfree_skb_reason(skb, reason);
409 return err;
410}
411
412static bool ip_frag_coalesce_ok(const struct ipq *qp)
413{
414 return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
415}
416
417/* Build a new IP datagram from all its fragments. */
418static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
419 struct sk_buff *prev_tail, struct net_device *dev)
420{
421 struct net *net = qp->q.fqdir->net;
422 struct iphdr *iph;
423 void *reasm_data;
424 int len, err;
425 u8 ecn;
426
427 ipq_kill(qp);
428
429 ecn = ip_frag_ecn_table[qp->ecn];
430 if (unlikely(ecn == 0xff)) {
431 err = -EINVAL;
432 goto out_fail;
433 }
434
435 /* Make the one we just received the head. */
436 reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
437 if (!reasm_data)
438 goto out_nomem;
439
440 len = ip_hdrlen(skb) + qp->q.len;
441 err = -E2BIG;
442 if (len > 65535)
443 goto out_oversize;
444
445 inet_frag_reasm_finish(&qp->q, skb, reasm_data,
446 ip_frag_coalesce_ok(qp));
447
448 skb->dev = dev;
449 IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
450
451 iph = ip_hdr(skb);
452 iph->tot_len = htons(len);
453 iph->tos |= ecn;
454
455 /* When we set IP_DF on a refragmented skb we must also force a
456 * call to ip_fragment to avoid forwarding a DF-skb of size s while
457 * original sender only sent fragments of size f (where f < s).
458 *
459 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
460 * frag seen to avoid sending tiny DF-fragments in case skb was built
461 * from one very small df-fragment and one large non-df frag.
462 */
463 if (qp->max_df_size == qp->q.max_size) {
464 IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
465 iph->frag_off = htons(IP_DF);
466 } else {
467 iph->frag_off = 0;
468 }
469
470 ip_send_check(iph);
471
472 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
473 qp->q.rb_fragments = RB_ROOT;
474 qp->q.fragments_tail = NULL;
475 qp->q.last_run_head = NULL;
476 return 0;
477
478out_nomem:
479 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
480 err = -ENOMEM;
481 goto out_fail;
482out_oversize:
483 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
484out_fail:
485 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
486 return err;
487}
488
489/* Process an incoming IP datagram fragment. */
490int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
491{
492 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
493 int vif = l3mdev_master_ifindex_rcu(dev);
494 struct ipq *qp;
495
496 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
497
498 /* Lookup (or create) queue header */
499 qp = ip_find(net, ip_hdr(skb), user, vif);
500 if (qp) {
501 int ret;
502
503 spin_lock(&qp->q.lock);
504
505 ret = ip_frag_queue(qp, skb);
506
507 spin_unlock(&qp->q.lock);
508 ipq_put(qp);
509 return ret;
510 }
511
512 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
513 kfree_skb(skb);
514 return -ENOMEM;
515}
516EXPORT_SYMBOL(ip_defrag);
517
518struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
519{
520 struct iphdr iph;
521 int netoff;
522 u32 len;
523
524 if (skb->protocol != htons(ETH_P_IP))
525 return skb;
526
527 netoff = skb_network_offset(skb);
528
529 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
530 return skb;
531
532 if (iph.ihl < 5 || iph.version != 4)
533 return skb;
534
535 len = ntohs(iph.tot_len);
536 if (skb->len < netoff + len || len < (iph.ihl * 4))
537 return skb;
538
539 if (ip_is_fragment(&iph)) {
540 skb = skb_share_check(skb, GFP_ATOMIC);
541 if (skb) {
542 if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
543 kfree_skb(skb);
544 return NULL;
545 }
546 if (pskb_trim_rcsum(skb, netoff + len)) {
547 kfree_skb(skb);
548 return NULL;
549 }
550 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
551 if (ip_defrag(net, skb, user))
552 return NULL;
553 skb_clear_hash(skb);
554 }
555 }
556 return skb;
557}
558EXPORT_SYMBOL(ip_check_defrag);
559
560#ifdef CONFIG_SYSCTL
561static int dist_min;
562
563static struct ctl_table ip4_frags_ns_ctl_table[] = {
564 {
565 .procname = "ipfrag_high_thresh",
566 .maxlen = sizeof(unsigned long),
567 .mode = 0644,
568 .proc_handler = proc_doulongvec_minmax,
569 },
570 {
571 .procname = "ipfrag_low_thresh",
572 .maxlen = sizeof(unsigned long),
573 .mode = 0644,
574 .proc_handler = proc_doulongvec_minmax,
575 },
576 {
577 .procname = "ipfrag_time",
578 .maxlen = sizeof(int),
579 .mode = 0644,
580 .proc_handler = proc_dointvec_jiffies,
581 },
582 {
583 .procname = "ipfrag_max_dist",
584 .maxlen = sizeof(int),
585 .mode = 0644,
586 .proc_handler = proc_dointvec_minmax,
587 .extra1 = &dist_min,
588 },
589};
590
591/* secret interval has been deprecated */
592static int ip4_frags_secret_interval_unused;
593static struct ctl_table ip4_frags_ctl_table[] = {
594 {
595 .procname = "ipfrag_secret_interval",
596 .data = &ip4_frags_secret_interval_unused,
597 .maxlen = sizeof(int),
598 .mode = 0644,
599 .proc_handler = proc_dointvec_jiffies,
600 },
601};
602
603static int __net_init ip4_frags_ns_ctl_register(struct net *net)
604{
605 struct ctl_table *table;
606 struct ctl_table_header *hdr;
607
608 table = ip4_frags_ns_ctl_table;
609 if (!net_eq(net, &init_net)) {
610 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
611 if (!table)
612 goto err_alloc;
613
614 }
615 table[0].data = &net->ipv4.fqdir->high_thresh;
616 table[0].extra1 = &net->ipv4.fqdir->low_thresh;
617 table[1].data = &net->ipv4.fqdir->low_thresh;
618 table[1].extra2 = &net->ipv4.fqdir->high_thresh;
619 table[2].data = &net->ipv4.fqdir->timeout;
620 table[3].data = &net->ipv4.fqdir->max_dist;
621
622 hdr = register_net_sysctl_sz(net, "net/ipv4", table,
623 ARRAY_SIZE(ip4_frags_ns_ctl_table));
624 if (!hdr)
625 goto err_reg;
626
627 net->ipv4.frags_hdr = hdr;
628 return 0;
629
630err_reg:
631 if (!net_eq(net, &init_net))
632 kfree(table);
633err_alloc:
634 return -ENOMEM;
635}
636
637static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
638{
639 const struct ctl_table *table;
640
641 table = net->ipv4.frags_hdr->ctl_table_arg;
642 unregister_net_sysctl_table(net->ipv4.frags_hdr);
643 kfree(table);
644}
645
646static void __init ip4_frags_ctl_register(void)
647{
648 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
649}
650#else
651static int ip4_frags_ns_ctl_register(struct net *net)
652{
653 return 0;
654}
655
656static void ip4_frags_ns_ctl_unregister(struct net *net)
657{
658}
659
660static void __init ip4_frags_ctl_register(void)
661{
662}
663#endif
664
665static int __net_init ipv4_frags_init_net(struct net *net)
666{
667 int res;
668
669 res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
670 if (res < 0)
671 return res;
672 /* Fragment cache limits.
673 *
674 * The fragment memory accounting code, (tries to) account for
675 * the real memory usage, by measuring both the size of frag
676 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
677 * and the SKB's truesize.
678 *
679 * A 64K fragment consumes 129736 bytes (44*2944)+200
680 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
681 *
682 * We will commit 4MB at one time. Should we cross that limit
683 * we will prune down to 3MB, making room for approx 8 big 64K
684 * fragments 8x128k.
685 */
686 net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
687 net->ipv4.fqdir->low_thresh = 3 * 1024 * 1024;
688 /*
689 * Important NOTE! Fragment queue must be destroyed before MSL expires.
690 * RFC791 is wrong proposing to prolongate timer each fragment arrival
691 * by TTL.
692 */
693 net->ipv4.fqdir->timeout = IP_FRAG_TIME;
694
695 net->ipv4.fqdir->max_dist = 64;
696
697 res = ip4_frags_ns_ctl_register(net);
698 if (res < 0)
699 fqdir_exit(net->ipv4.fqdir);
700 return res;
701}
702
703static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
704{
705 fqdir_pre_exit(net->ipv4.fqdir);
706}
707
708static void __net_exit ipv4_frags_exit_net(struct net *net)
709{
710 ip4_frags_ns_ctl_unregister(net);
711 fqdir_exit(net->ipv4.fqdir);
712}
713
714static struct pernet_operations ip4_frags_ops = {
715 .init = ipv4_frags_init_net,
716 .pre_exit = ipv4_frags_pre_exit_net,
717 .exit = ipv4_frags_exit_net,
718};
719
720
721static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
722{
723 return jhash2(data,
724 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
725}
726
727static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
728{
729 const struct inet_frag_queue *fq = data;
730
731 return jhash2((const u32 *)&fq->key.v4,
732 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
733}
734
735static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
736{
737 const struct frag_v4_compare_key *key = arg->key;
738 const struct inet_frag_queue *fq = ptr;
739
740 return !!memcmp(&fq->key, key, sizeof(*key));
741}
742
743static const struct rhashtable_params ip4_rhash_params = {
744 .head_offset = offsetof(struct inet_frag_queue, node),
745 .key_offset = offsetof(struct inet_frag_queue, key),
746 .key_len = sizeof(struct frag_v4_compare_key),
747 .hashfn = ip4_key_hashfn,
748 .obj_hashfn = ip4_obj_hashfn,
749 .obj_cmpfn = ip4_obj_cmpfn,
750 .automatic_shrinking = true,
751};
752
753void __init ipfrag_init(void)
754{
755 ip4_frags.constructor = ip4_frag_init;
756 ip4_frags.destructor = ip4_frag_free;
757 ip4_frags.qsize = sizeof(struct ipq);
758 ip4_frags.frag_expire = ip_expire;
759 ip4_frags.frags_cache_name = ip_frag_cache_name;
760 ip4_frags.rhash_params = ip4_rhash_params;
761 if (inet_frags_init(&ip4_frags))
762 panic("IP: failed to allocate ip4_frags cache\n");
763 ip4_frags_ctl_register();
764 register_pernet_subsys(&ip4_frags_ops);
765}
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}