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#include <net/l3mdev.h>
52
53/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
54 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
55 * as well. Or notify me, at least. --ANK
56 */
57static const char ip_frag_cache_name[] = "ip4-frags";
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 u16 max_df_size; /* largest frag with DF set seen */
78 int iif;
79 int vif; /* L3 master device index */
80 unsigned int rid;
81 struct inet_peer *peer;
82};
83
84static u8 ip4_frag_ecn(u8 tos)
85{
86 return 1 << (tos & INET_ECN_MASK);
87}
88
89static struct inet_frags ip4_frags;
90
91int ip_frag_mem(struct net *net)
92{
93 return sum_frag_mem_limit(&net->ipv4.frags);
94}
95
96static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
97 struct net_device *dev);
98
99struct ip4_create_arg {
100 struct iphdr *iph;
101 u32 user;
102 int vif;
103};
104
105static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
106{
107 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
108 return jhash_3words((__force u32)id << 16 | prot,
109 (__force u32)saddr, (__force u32)daddr,
110 ip4_frags.rnd);
111}
112
113static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
114{
115 const struct ipq *ipq;
116
117 ipq = container_of(q, struct ipq, q);
118 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
119}
120
121static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
122{
123 const struct ipq *qp;
124 const struct ip4_create_arg *arg = a;
125
126 qp = container_of(q, struct ipq, q);
127 return qp->id == arg->iph->id &&
128 qp->saddr == arg->iph->saddr &&
129 qp->daddr == arg->iph->daddr &&
130 qp->protocol == arg->iph->protocol &&
131 qp->user == arg->user &&
132 qp->vif == arg->vif;
133}
134
135static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
136{
137 struct ipq *qp = container_of(q, struct ipq, q);
138 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
139 frags);
140 struct net *net = container_of(ipv4, struct net, ipv4);
141
142 const struct ip4_create_arg *arg = a;
143
144 qp->protocol = arg->iph->protocol;
145 qp->id = arg->iph->id;
146 qp->ecn = ip4_frag_ecn(arg->iph->tos);
147 qp->saddr = arg->iph->saddr;
148 qp->daddr = arg->iph->daddr;
149 qp->vif = arg->vif;
150 qp->user = arg->user;
151 qp->peer = q->net->max_dist ?
152 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
153 NULL;
154}
155
156static void ip4_frag_free(struct inet_frag_queue *q)
157{
158 struct ipq *qp;
159
160 qp = container_of(q, struct ipq, q);
161 if (qp->peer)
162 inet_putpeer(qp->peer);
163}
164
165
166/* Destruction primitives. */
167
168static void ipq_put(struct ipq *ipq)
169{
170 inet_frag_put(&ipq->q, &ip4_frags);
171}
172
173/* Kill ipq entry. It is not destroyed immediately,
174 * because caller (and someone more) holds reference count.
175 */
176static void ipq_kill(struct ipq *ipq)
177{
178 inet_frag_kill(&ipq->q, &ip4_frags);
179}
180
181static bool frag_expire_skip_icmp(u32 user)
182{
183 return user == IP_DEFRAG_AF_PACKET ||
184 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
185 __IP_DEFRAG_CONNTRACK_IN_END) ||
186 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
187 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
188}
189
190/*
191 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
192 */
193static void ip_expire(unsigned long arg)
194{
195 struct ipq *qp;
196 struct net *net;
197
198 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
199 net = container_of(qp->q.net, struct net, ipv4.frags);
200
201 spin_lock(&qp->q.lock);
202
203 if (qp->q.flags & INET_FRAG_COMPLETE)
204 goto out;
205
206 ipq_kill(qp);
207 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
208
209 if (!inet_frag_evicting(&qp->q)) {
210 struct sk_buff *head = qp->q.fragments;
211 const struct iphdr *iph;
212 int err;
213
214 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
215
216 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
217 goto out;
218
219 rcu_read_lock();
220 head->dev = dev_get_by_index_rcu(net, qp->iif);
221 if (!head->dev)
222 goto out_rcu_unlock;
223
224 /* skb has no dst, perform route lookup again */
225 iph = ip_hdr(head);
226 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
227 iph->tos, head->dev);
228 if (err)
229 goto out_rcu_unlock;
230
231 /* Only an end host needs to send an ICMP
232 * "Fragment Reassembly Timeout" message, per RFC792.
233 */
234 if (frag_expire_skip_icmp(qp->user) &&
235 (skb_rtable(head)->rt_type != RTN_LOCAL))
236 goto out_rcu_unlock;
237
238 /* Send an ICMP "Fragment Reassembly Timeout" message. */
239 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
240out_rcu_unlock:
241 rcu_read_unlock();
242 }
243out:
244 spin_unlock(&qp->q.lock);
245 ipq_put(qp);
246}
247
248/* Find the correct entry in the "incomplete datagrams" queue for
249 * this IP datagram, and create new one, if nothing is found.
250 */
251static struct ipq *ip_find(struct net *net, struct iphdr *iph,
252 u32 user, int vif)
253{
254 struct inet_frag_queue *q;
255 struct ip4_create_arg arg;
256 unsigned int hash;
257
258 arg.iph = iph;
259 arg.user = user;
260 arg.vif = vif;
261
262 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
263
264 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
265 if (IS_ERR_OR_NULL(q)) {
266 inet_frag_maybe_warn_overflow(q, pr_fmt());
267 return NULL;
268 }
269 return container_of(q, struct ipq, q);
270}
271
272/* Is the fragment too far ahead to be part of ipq? */
273static int ip_frag_too_far(struct ipq *qp)
274{
275 struct inet_peer *peer = qp->peer;
276 unsigned int max = qp->q.net->max_dist;
277 unsigned int start, end;
278
279 int rc;
280
281 if (!peer || !max)
282 return 0;
283
284 start = qp->rid;
285 end = atomic_inc_return(&peer->rid);
286 qp->rid = end;
287
288 rc = qp->q.fragments && (end - start) > max;
289
290 if (rc) {
291 struct net *net;
292
293 net = container_of(qp->q.net, struct net, ipv4.frags);
294 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
295 }
296
297 return rc;
298}
299
300static int ip_frag_reinit(struct ipq *qp)
301{
302 struct sk_buff *fp;
303 unsigned int sum_truesize = 0;
304
305 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
306 atomic_inc(&qp->q.refcnt);
307 return -ETIMEDOUT;
308 }
309
310 fp = qp->q.fragments;
311 do {
312 struct sk_buff *xp = fp->next;
313
314 sum_truesize += fp->truesize;
315 kfree_skb(fp);
316 fp = xp;
317 } while (fp);
318 sub_frag_mem_limit(qp->q.net, sum_truesize);
319
320 qp->q.flags = 0;
321 qp->q.len = 0;
322 qp->q.meat = 0;
323 qp->q.fragments = NULL;
324 qp->q.fragments_tail = NULL;
325 qp->iif = 0;
326 qp->ecn = 0;
327
328 return 0;
329}
330
331/* Add new segment to existing queue. */
332static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
333{
334 struct sk_buff *prev, *next;
335 struct net_device *dev;
336 unsigned int fragsize;
337 int flags, offset;
338 int ihl, end;
339 int err = -ENOENT;
340 u8 ecn;
341
342 if (qp->q.flags & INET_FRAG_COMPLETE)
343 goto err;
344
345 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
346 unlikely(ip_frag_too_far(qp)) &&
347 unlikely(err = ip_frag_reinit(qp))) {
348 ipq_kill(qp);
349 goto err;
350 }
351
352 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
353 offset = ntohs(ip_hdr(skb)->frag_off);
354 flags = offset & ~IP_OFFSET;
355 offset &= IP_OFFSET;
356 offset <<= 3; /* offset is in 8-byte chunks */
357 ihl = ip_hdrlen(skb);
358
359 /* Determine the position of this fragment. */
360 end = offset + skb->len - skb_network_offset(skb) - ihl;
361 err = -EINVAL;
362
363 /* Is this the final fragment? */
364 if ((flags & IP_MF) == 0) {
365 /* If we already have some bits beyond end
366 * or have different end, the segment is corrupted.
367 */
368 if (end < qp->q.len ||
369 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
370 goto err;
371 qp->q.flags |= INET_FRAG_LAST_IN;
372 qp->q.len = end;
373 } else {
374 if (end&7) {
375 end &= ~7;
376 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
377 skb->ip_summed = CHECKSUM_NONE;
378 }
379 if (end > qp->q.len) {
380 /* Some bits beyond end -> corruption. */
381 if (qp->q.flags & INET_FRAG_LAST_IN)
382 goto err;
383 qp->q.len = end;
384 }
385 }
386 if (end == offset)
387 goto err;
388
389 err = -ENOMEM;
390 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
391 goto err;
392
393 err = pskb_trim_rcsum(skb, end - offset);
394 if (err)
395 goto err;
396
397 /* Find out which fragments are in front and at the back of us
398 * in the chain of fragments so far. We must know where to put
399 * this fragment, right?
400 */
401 prev = qp->q.fragments_tail;
402 if (!prev || FRAG_CB(prev)->offset < offset) {
403 next = NULL;
404 goto found;
405 }
406 prev = NULL;
407 for (next = qp->q.fragments; next != NULL; next = next->next) {
408 if (FRAG_CB(next)->offset >= offset)
409 break; /* bingo! */
410 prev = next;
411 }
412
413found:
414 /* We found where to put this one. Check for overlap with
415 * preceding fragment, and, if needed, align things so that
416 * any overlaps are eliminated.
417 */
418 if (prev) {
419 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
420
421 if (i > 0) {
422 offset += i;
423 err = -EINVAL;
424 if (end <= offset)
425 goto err;
426 err = -ENOMEM;
427 if (!pskb_pull(skb, i))
428 goto err;
429 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
430 skb->ip_summed = CHECKSUM_NONE;
431 }
432 }
433
434 err = -ENOMEM;
435
436 while (next && FRAG_CB(next)->offset < end) {
437 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
438
439 if (i < next->len) {
440 /* Eat head of the next overlapped fragment
441 * and leave the loop. The next ones cannot overlap.
442 */
443 if (!pskb_pull(next, i))
444 goto err;
445 FRAG_CB(next)->offset += i;
446 qp->q.meat -= i;
447 if (next->ip_summed != CHECKSUM_UNNECESSARY)
448 next->ip_summed = CHECKSUM_NONE;
449 break;
450 } else {
451 struct sk_buff *free_it = next;
452
453 /* Old fragment is completely overridden with
454 * new one drop it.
455 */
456 next = next->next;
457
458 if (prev)
459 prev->next = next;
460 else
461 qp->q.fragments = next;
462
463 qp->q.meat -= free_it->len;
464 sub_frag_mem_limit(qp->q.net, free_it->truesize);
465 kfree_skb(free_it);
466 }
467 }
468
469 FRAG_CB(skb)->offset = offset;
470
471 /* Insert this fragment in the chain of fragments. */
472 skb->next = next;
473 if (!next)
474 qp->q.fragments_tail = skb;
475 if (prev)
476 prev->next = skb;
477 else
478 qp->q.fragments = skb;
479
480 dev = skb->dev;
481 if (dev) {
482 qp->iif = dev->ifindex;
483 skb->dev = NULL;
484 }
485 qp->q.stamp = skb->tstamp;
486 qp->q.meat += skb->len;
487 qp->ecn |= ecn;
488 add_frag_mem_limit(qp->q.net, skb->truesize);
489 if (offset == 0)
490 qp->q.flags |= INET_FRAG_FIRST_IN;
491
492 fragsize = skb->len + ihl;
493
494 if (fragsize > qp->q.max_size)
495 qp->q.max_size = fragsize;
496
497 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
498 fragsize > qp->max_df_size)
499 qp->max_df_size = fragsize;
500
501 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
502 qp->q.meat == qp->q.len) {
503 unsigned long orefdst = skb->_skb_refdst;
504
505 skb->_skb_refdst = 0UL;
506 err = ip_frag_reasm(qp, prev, dev);
507 skb->_skb_refdst = orefdst;
508 return err;
509 }
510
511 skb_dst_drop(skb);
512 return -EINPROGRESS;
513
514err:
515 kfree_skb(skb);
516 return err;
517}
518
519
520/* Build a new IP datagram from all its fragments. */
521
522static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
523 struct net_device *dev)
524{
525 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
526 struct iphdr *iph;
527 struct sk_buff *fp, *head = qp->q.fragments;
528 int len;
529 int ihlen;
530 int err;
531 u8 ecn;
532
533 ipq_kill(qp);
534
535 ecn = ip_frag_ecn_table[qp->ecn];
536 if (unlikely(ecn == 0xff)) {
537 err = -EINVAL;
538 goto out_fail;
539 }
540 /* Make the one we just received the head. */
541 if (prev) {
542 head = prev->next;
543 fp = skb_clone(head, GFP_ATOMIC);
544 if (!fp)
545 goto out_nomem;
546
547 fp->next = head->next;
548 if (!fp->next)
549 qp->q.fragments_tail = fp;
550 prev->next = fp;
551
552 skb_morph(head, qp->q.fragments);
553 head->next = qp->q.fragments->next;
554
555 consume_skb(qp->q.fragments);
556 qp->q.fragments = head;
557 }
558
559 WARN_ON(!head);
560 WARN_ON(FRAG_CB(head)->offset != 0);
561
562 /* Allocate a new buffer for the datagram. */
563 ihlen = ip_hdrlen(head);
564 len = ihlen + qp->q.len;
565
566 err = -E2BIG;
567 if (len > 65535)
568 goto out_oversize;
569
570 /* Head of list must not be cloned. */
571 if (skb_unclone(head, GFP_ATOMIC))
572 goto out_nomem;
573
574 /* If the first fragment is fragmented itself, we split
575 * it to two chunks: the first with data and paged part
576 * and the second, holding only fragments. */
577 if (skb_has_frag_list(head)) {
578 struct sk_buff *clone;
579 int i, plen = 0;
580
581 clone = alloc_skb(0, GFP_ATOMIC);
582 if (!clone)
583 goto out_nomem;
584 clone->next = head->next;
585 head->next = clone;
586 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
587 skb_frag_list_init(head);
588 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
589 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
590 clone->len = clone->data_len = head->data_len - plen;
591 head->data_len -= clone->len;
592 head->len -= clone->len;
593 clone->csum = 0;
594 clone->ip_summed = head->ip_summed;
595 add_frag_mem_limit(qp->q.net, clone->truesize);
596 }
597
598 skb_shinfo(head)->frag_list = head->next;
599 skb_push(head, head->data - skb_network_header(head));
600
601 for (fp=head->next; fp; fp = fp->next) {
602 head->data_len += fp->len;
603 head->len += fp->len;
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 head->truesize += fp->truesize;
609 }
610 sub_frag_mem_limit(qp->q.net, head->truesize);
611
612 head->next = NULL;
613 head->dev = dev;
614 head->tstamp = qp->q.stamp;
615 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
616
617 iph = ip_hdr(head);
618 iph->tot_len = htons(len);
619 iph->tos |= ecn;
620
621 /* When we set IP_DF on a refragmented skb we must also force a
622 * call to ip_fragment to avoid forwarding a DF-skb of size s while
623 * original sender only sent fragments of size f (where f < s).
624 *
625 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
626 * frag seen to avoid sending tiny DF-fragments in case skb was built
627 * from one very small df-fragment and one large non-df frag.
628 */
629 if (qp->max_df_size == qp->q.max_size) {
630 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
631 iph->frag_off = htons(IP_DF);
632 } else {
633 iph->frag_off = 0;
634 }
635
636 ip_send_check(iph);
637
638 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
639 qp->q.fragments = NULL;
640 qp->q.fragments_tail = NULL;
641 return 0;
642
643out_nomem:
644 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
645 err = -ENOMEM;
646 goto out_fail;
647out_oversize:
648 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
649out_fail:
650 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
651 return err;
652}
653
654/* Process an incoming IP datagram fragment. */
655int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
656{
657 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
658 int vif = l3mdev_master_ifindex_rcu(dev);
659 struct ipq *qp;
660
661 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
662 skb_orphan(skb);
663
664 /* Lookup (or create) queue header */
665 qp = ip_find(net, ip_hdr(skb), user, vif);
666 if (qp) {
667 int ret;
668
669 spin_lock(&qp->q.lock);
670
671 ret = ip_frag_queue(qp, skb);
672
673 spin_unlock(&qp->q.lock);
674 ipq_put(qp);
675 return ret;
676 }
677
678 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
679 kfree_skb(skb);
680 return -ENOMEM;
681}
682EXPORT_SYMBOL(ip_defrag);
683
684struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
685{
686 struct iphdr iph;
687 int netoff;
688 u32 len;
689
690 if (skb->protocol != htons(ETH_P_IP))
691 return skb;
692
693 netoff = skb_network_offset(skb);
694
695 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
696 return skb;
697
698 if (iph.ihl < 5 || iph.version != 4)
699 return skb;
700
701 len = ntohs(iph.tot_len);
702 if (skb->len < netoff + len || len < (iph.ihl * 4))
703 return skb;
704
705 if (ip_is_fragment(&iph)) {
706 skb = skb_share_check(skb, GFP_ATOMIC);
707 if (skb) {
708 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
709 return skb;
710 if (pskb_trim_rcsum(skb, netoff + len))
711 return skb;
712 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
713 if (ip_defrag(net, skb, user))
714 return NULL;
715 skb_clear_hash(skb);
716 }
717 }
718 return skb;
719}
720EXPORT_SYMBOL(ip_check_defrag);
721
722#ifdef CONFIG_SYSCTL
723static int zero;
724
725static struct ctl_table ip4_frags_ns_ctl_table[] = {
726 {
727 .procname = "ipfrag_high_thresh",
728 .data = &init_net.ipv4.frags.high_thresh,
729 .maxlen = sizeof(int),
730 .mode = 0644,
731 .proc_handler = proc_dointvec_minmax,
732 .extra1 = &init_net.ipv4.frags.low_thresh
733 },
734 {
735 .procname = "ipfrag_low_thresh",
736 .data = &init_net.ipv4.frags.low_thresh,
737 .maxlen = sizeof(int),
738 .mode = 0644,
739 .proc_handler = proc_dointvec_minmax,
740 .extra1 = &zero,
741 .extra2 = &init_net.ipv4.frags.high_thresh
742 },
743 {
744 .procname = "ipfrag_time",
745 .data = &init_net.ipv4.frags.timeout,
746 .maxlen = sizeof(int),
747 .mode = 0644,
748 .proc_handler = proc_dointvec_jiffies,
749 },
750 {
751 .procname = "ipfrag_max_dist",
752 .data = &init_net.ipv4.frags.max_dist,
753 .maxlen = sizeof(int),
754 .mode = 0644,
755 .proc_handler = proc_dointvec_minmax,
756 .extra1 = &zero
757 },
758 { }
759};
760
761/* secret interval has been deprecated */
762static int ip4_frags_secret_interval_unused;
763static struct ctl_table ip4_frags_ctl_table[] = {
764 {
765 .procname = "ipfrag_secret_interval",
766 .data = &ip4_frags_secret_interval_unused,
767 .maxlen = sizeof(int),
768 .mode = 0644,
769 .proc_handler = proc_dointvec_jiffies,
770 },
771 { }
772};
773
774static int __net_init ip4_frags_ns_ctl_register(struct net *net)
775{
776 struct ctl_table *table;
777 struct ctl_table_header *hdr;
778
779 table = ip4_frags_ns_ctl_table;
780 if (!net_eq(net, &init_net)) {
781 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
782 if (!table)
783 goto err_alloc;
784
785 table[0].data = &net->ipv4.frags.high_thresh;
786 table[0].extra1 = &net->ipv4.frags.low_thresh;
787 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
788 table[1].data = &net->ipv4.frags.low_thresh;
789 table[1].extra2 = &net->ipv4.frags.high_thresh;
790 table[2].data = &net->ipv4.frags.timeout;
791 table[3].data = &net->ipv4.frags.max_dist;
792 }
793
794 hdr = register_net_sysctl(net, "net/ipv4", table);
795 if (!hdr)
796 goto err_reg;
797
798 net->ipv4.frags_hdr = hdr;
799 return 0;
800
801err_reg:
802 if (!net_eq(net, &init_net))
803 kfree(table);
804err_alloc:
805 return -ENOMEM;
806}
807
808static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
809{
810 struct ctl_table *table;
811
812 table = net->ipv4.frags_hdr->ctl_table_arg;
813 unregister_net_sysctl_table(net->ipv4.frags_hdr);
814 kfree(table);
815}
816
817static void __init ip4_frags_ctl_register(void)
818{
819 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
820}
821#else
822static int ip4_frags_ns_ctl_register(struct net *net)
823{
824 return 0;
825}
826
827static void ip4_frags_ns_ctl_unregister(struct net *net)
828{
829}
830
831static void __init ip4_frags_ctl_register(void)
832{
833}
834#endif
835
836static int __net_init ipv4_frags_init_net(struct net *net)
837{
838 int res;
839
840 /* Fragment cache limits.
841 *
842 * The fragment memory accounting code, (tries to) account for
843 * the real memory usage, by measuring both the size of frag
844 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
845 * and the SKB's truesize.
846 *
847 * A 64K fragment consumes 129736 bytes (44*2944)+200
848 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
849 *
850 * We will commit 4MB at one time. Should we cross that limit
851 * we will prune down to 3MB, making room for approx 8 big 64K
852 * fragments 8x128k.
853 */
854 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
855 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
856 /*
857 * Important NOTE! Fragment queue must be destroyed before MSL expires.
858 * RFC791 is wrong proposing to prolongate timer each fragment arrival
859 * by TTL.
860 */
861 net->ipv4.frags.timeout = IP_FRAG_TIME;
862
863 net->ipv4.frags.max_dist = 64;
864
865 res = inet_frags_init_net(&net->ipv4.frags);
866 if (res)
867 return res;
868 res = ip4_frags_ns_ctl_register(net);
869 if (res)
870 inet_frags_uninit_net(&net->ipv4.frags);
871 return res;
872}
873
874static void __net_exit ipv4_frags_exit_net(struct net *net)
875{
876 ip4_frags_ns_ctl_unregister(net);
877 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
878}
879
880static struct pernet_operations ip4_frags_ops = {
881 .init = ipv4_frags_init_net,
882 .exit = ipv4_frags_exit_net,
883};
884
885void __init ipfrag_init(void)
886{
887 ip4_frags_ctl_register();
888 register_pernet_subsys(&ip4_frags_ops);
889 ip4_frags.hashfn = ip4_hashfn;
890 ip4_frags.constructor = ip4_frag_init;
891 ip4_frags.destructor = ip4_frag_free;
892 ip4_frags.qsize = sizeof(struct ipq);
893 ip4_frags.match = ip4_frag_match;
894 ip4_frags.frag_expire = ip_expire;
895 ip4_frags.frags_cache_name = ip_frag_cache_name;
896 if (inet_frags_init(&ip4_frags))
897 panic("IP: failed to allocate ip4_frags cache\n");
898}