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 struct net *net = q->fqdir->net;
86
87 const struct frag_v4_compare_key *key = a;
88
89 q->key.v4 = *key;
90 qp->ecn = 0;
91 qp->peer = q->fqdir->max_dist ?
92 inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
93 NULL;
94}
95
96static void ip4_frag_free(struct inet_frag_queue *q)
97{
98 struct ipq *qp;
99
100 qp = container_of(q, struct ipq, q);
101 if (qp->peer)
102 inet_putpeer(qp->peer);
103}
104
105
106/* Destruction primitives. */
107
108static void ipq_put(struct ipq *ipq)
109{
110 inet_frag_put(&ipq->q);
111}
112
113/* Kill ipq entry. It is not destroyed immediately,
114 * because caller (and someone more) holds reference count.
115 */
116static void ipq_kill(struct ipq *ipq)
117{
118 inet_frag_kill(&ipq->q);
119}
120
121static bool frag_expire_skip_icmp(u32 user)
122{
123 return user == IP_DEFRAG_AF_PACKET ||
124 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
125 __IP_DEFRAG_CONNTRACK_IN_END) ||
126 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
127 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
128}
129
130/*
131 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
132 */
133static void ip_expire(struct timer_list *t)
134{
135 struct inet_frag_queue *frag = from_timer(frag, t, timer);
136 const struct iphdr *iph;
137 struct sk_buff *head = NULL;
138 struct net *net;
139 struct ipq *qp;
140 int err;
141
142 qp = container_of(frag, struct ipq, q);
143 net = qp->q.fqdir->net;
144
145 rcu_read_lock();
146
147 /* Paired with WRITE_ONCE() in fqdir_pre_exit(). */
148 if (READ_ONCE(qp->q.fqdir->dead))
149 goto out_rcu_unlock;
150
151 spin_lock(&qp->q.lock);
152
153 if (qp->q.flags & INET_FRAG_COMPLETE)
154 goto out;
155
156 qp->q.flags |= INET_FRAG_DROP;
157 ipq_kill(qp);
158 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
159 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
160
161 if (!(qp->q.flags & INET_FRAG_FIRST_IN))
162 goto out;
163
164 /* sk_buff::dev and sk_buff::rbnode are unionized. So we
165 * pull the head out of the tree in order to be able to
166 * deal with head->dev.
167 */
168 head = inet_frag_pull_head(&qp->q);
169 if (!head)
170 goto out;
171 head->dev = dev_get_by_index_rcu(net, qp->iif);
172 if (!head->dev)
173 goto out;
174
175
176 /* skb has no dst, perform route lookup again */
177 iph = ip_hdr(head);
178 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
179 iph->tos, head->dev);
180 if (err)
181 goto out;
182
183 /* Only an end host needs to send an ICMP
184 * "Fragment Reassembly Timeout" message, per RFC792.
185 */
186 if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
187 (skb_rtable(head)->rt_type != RTN_LOCAL))
188 goto out;
189
190 spin_unlock(&qp->q.lock);
191 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
192 goto out_rcu_unlock;
193
194out:
195 spin_unlock(&qp->q.lock);
196out_rcu_unlock:
197 rcu_read_unlock();
198 kfree_skb_reason(head, SKB_DROP_REASON_FRAG_REASM_TIMEOUT);
199 ipq_put(qp);
200}
201
202/* Find the correct entry in the "incomplete datagrams" queue for
203 * this IP datagram, and create new one, if nothing is found.
204 */
205static struct ipq *ip_find(struct net *net, struct iphdr *iph,
206 u32 user, int vif)
207{
208 struct frag_v4_compare_key key = {
209 .saddr = iph->saddr,
210 .daddr = iph->daddr,
211 .user = user,
212 .vif = vif,
213 .id = iph->id,
214 .protocol = iph->protocol,
215 };
216 struct inet_frag_queue *q;
217
218 q = inet_frag_find(net->ipv4.fqdir, &key);
219 if (!q)
220 return NULL;
221
222 return container_of(q, struct ipq, q);
223}
224
225/* Is the fragment too far ahead to be part of ipq? */
226static int ip_frag_too_far(struct ipq *qp)
227{
228 struct inet_peer *peer = qp->peer;
229 unsigned int max = qp->q.fqdir->max_dist;
230 unsigned int start, end;
231
232 int rc;
233
234 if (!peer || !max)
235 return 0;
236
237 start = qp->rid;
238 end = atomic_inc_return(&peer->rid);
239 qp->rid = end;
240
241 rc = qp->q.fragments_tail && (end - start) > max;
242
243 if (rc)
244 __IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
245
246 return rc;
247}
248
249static int ip_frag_reinit(struct ipq *qp)
250{
251 unsigned int sum_truesize = 0;
252
253 if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
254 refcount_inc(&qp->q.refcnt);
255 return -ETIMEDOUT;
256 }
257
258 sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments,
259 SKB_DROP_REASON_FRAG_TOO_FAR);
260 sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
261
262 qp->q.flags = 0;
263 qp->q.len = 0;
264 qp->q.meat = 0;
265 qp->q.rb_fragments = RB_ROOT;
266 qp->q.fragments_tail = NULL;
267 qp->q.last_run_head = NULL;
268 qp->iif = 0;
269 qp->ecn = 0;
270
271 return 0;
272}
273
274/* Add new segment to existing queue. */
275static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
276{
277 struct net *net = qp->q.fqdir->net;
278 int ihl, end, flags, offset;
279 struct sk_buff *prev_tail;
280 struct net_device *dev;
281 unsigned int fragsize;
282 int err = -ENOENT;
283 SKB_DR(reason);
284 u8 ecn;
285
286 /* If reassembly is already done, @skb must be a duplicate frag. */
287 if (qp->q.flags & INET_FRAG_COMPLETE) {
288 SKB_DR_SET(reason, DUP_FRAG);
289 goto err;
290 }
291
292 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
293 unlikely(ip_frag_too_far(qp)) &&
294 unlikely(err = ip_frag_reinit(qp))) {
295 ipq_kill(qp);
296 goto err;
297 }
298
299 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
300 offset = ntohs(ip_hdr(skb)->frag_off);
301 flags = offset & ~IP_OFFSET;
302 offset &= IP_OFFSET;
303 offset <<= 3; /* offset is in 8-byte chunks */
304 ihl = ip_hdrlen(skb);
305
306 /* Determine the position of this fragment. */
307 end = offset + skb->len - skb_network_offset(skb) - ihl;
308 err = -EINVAL;
309
310 /* Is this the final fragment? */
311 if ((flags & IP_MF) == 0) {
312 /* If we already have some bits beyond end
313 * or have different end, the segment is corrupted.
314 */
315 if (end < qp->q.len ||
316 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
317 goto discard_qp;
318 qp->q.flags |= INET_FRAG_LAST_IN;
319 qp->q.len = end;
320 } else {
321 if (end&7) {
322 end &= ~7;
323 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
324 skb->ip_summed = CHECKSUM_NONE;
325 }
326 if (end > qp->q.len) {
327 /* Some bits beyond end -> corruption. */
328 if (qp->q.flags & INET_FRAG_LAST_IN)
329 goto discard_qp;
330 qp->q.len = end;
331 }
332 }
333 if (end == offset)
334 goto discard_qp;
335
336 err = -ENOMEM;
337 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
338 goto discard_qp;
339
340 err = pskb_trim_rcsum(skb, end - offset);
341 if (err)
342 goto discard_qp;
343
344 /* Note : skb->rbnode and skb->dev share the same location. */
345 dev = skb->dev;
346 /* Makes sure compiler wont do silly aliasing games */
347 barrier();
348
349 prev_tail = qp->q.fragments_tail;
350 err = inet_frag_queue_insert(&qp->q, skb, offset, end);
351 if (err)
352 goto insert_error;
353
354 if (dev)
355 qp->iif = dev->ifindex;
356
357 qp->q.stamp = skb->tstamp;
358 qp->q.mono_delivery_time = skb->mono_delivery_time;
359 qp->q.meat += skb->len;
360 qp->ecn |= ecn;
361 add_frag_mem_limit(qp->q.fqdir, skb->truesize);
362 if (offset == 0)
363 qp->q.flags |= INET_FRAG_FIRST_IN;
364
365 fragsize = skb->len + ihl;
366
367 if (fragsize > qp->q.max_size)
368 qp->q.max_size = fragsize;
369
370 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
371 fragsize > qp->max_df_size)
372 qp->max_df_size = fragsize;
373
374 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
375 qp->q.meat == qp->q.len) {
376 unsigned long orefdst = skb->_skb_refdst;
377
378 skb->_skb_refdst = 0UL;
379 err = ip_frag_reasm(qp, skb, prev_tail, dev);
380 skb->_skb_refdst = orefdst;
381 if (err)
382 inet_frag_kill(&qp->q);
383 return err;
384 }
385
386 skb_dst_drop(skb);
387 skb_orphan(skb);
388 return -EINPROGRESS;
389
390insert_error:
391 if (err == IPFRAG_DUP) {
392 SKB_DR_SET(reason, DUP_FRAG);
393 err = -EINVAL;
394 goto err;
395 }
396 err = -EINVAL;
397 __IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
398discard_qp:
399 inet_frag_kill(&qp->q);
400 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
401err:
402 kfree_skb_reason(skb, reason);
403 return err;
404}
405
406static bool ip_frag_coalesce_ok(const struct ipq *qp)
407{
408 return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
409}
410
411/* Build a new IP datagram from all its fragments. */
412static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
413 struct sk_buff *prev_tail, struct net_device *dev)
414{
415 struct net *net = qp->q.fqdir->net;
416 struct iphdr *iph;
417 void *reasm_data;
418 int len, err;
419 u8 ecn;
420
421 ipq_kill(qp);
422
423 ecn = ip_frag_ecn_table[qp->ecn];
424 if (unlikely(ecn == 0xff)) {
425 err = -EINVAL;
426 goto out_fail;
427 }
428
429 /* Make the one we just received the head. */
430 reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
431 if (!reasm_data)
432 goto out_nomem;
433
434 len = ip_hdrlen(skb) + qp->q.len;
435 err = -E2BIG;
436 if (len > 65535)
437 goto out_oversize;
438
439 inet_frag_reasm_finish(&qp->q, skb, reasm_data,
440 ip_frag_coalesce_ok(qp));
441
442 skb->dev = dev;
443 IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
444
445 iph = ip_hdr(skb);
446 iph->tot_len = htons(len);
447 iph->tos |= ecn;
448
449 /* When we set IP_DF on a refragmented skb we must also force a
450 * call to ip_fragment to avoid forwarding a DF-skb of size s while
451 * original sender only sent fragments of size f (where f < s).
452 *
453 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
454 * frag seen to avoid sending tiny DF-fragments in case skb was built
455 * from one very small df-fragment and one large non-df frag.
456 */
457 if (qp->max_df_size == qp->q.max_size) {
458 IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
459 iph->frag_off = htons(IP_DF);
460 } else {
461 iph->frag_off = 0;
462 }
463
464 ip_send_check(iph);
465
466 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
467 qp->q.rb_fragments = RB_ROOT;
468 qp->q.fragments_tail = NULL;
469 qp->q.last_run_head = NULL;
470 return 0;
471
472out_nomem:
473 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
474 err = -ENOMEM;
475 goto out_fail;
476out_oversize:
477 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
478out_fail:
479 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
480 return err;
481}
482
483/* Process an incoming IP datagram fragment. */
484int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
485{
486 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
487 int vif = l3mdev_master_ifindex_rcu(dev);
488 struct ipq *qp;
489
490 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
491
492 /* Lookup (or create) queue header */
493 qp = ip_find(net, ip_hdr(skb), user, vif);
494 if (qp) {
495 int ret;
496
497 spin_lock(&qp->q.lock);
498
499 ret = ip_frag_queue(qp, skb);
500
501 spin_unlock(&qp->q.lock);
502 ipq_put(qp);
503 return ret;
504 }
505
506 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
507 kfree_skb(skb);
508 return -ENOMEM;
509}
510EXPORT_SYMBOL(ip_defrag);
511
512struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
513{
514 struct iphdr iph;
515 int netoff;
516 u32 len;
517
518 if (skb->protocol != htons(ETH_P_IP))
519 return skb;
520
521 netoff = skb_network_offset(skb);
522
523 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
524 return skb;
525
526 if (iph.ihl < 5 || iph.version != 4)
527 return skb;
528
529 len = ntohs(iph.tot_len);
530 if (skb->len < netoff + len || len < (iph.ihl * 4))
531 return skb;
532
533 if (ip_is_fragment(&iph)) {
534 skb = skb_share_check(skb, GFP_ATOMIC);
535 if (skb) {
536 if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
537 kfree_skb(skb);
538 return NULL;
539 }
540 if (pskb_trim_rcsum(skb, netoff + len)) {
541 kfree_skb(skb);
542 return NULL;
543 }
544 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
545 if (ip_defrag(net, skb, user))
546 return NULL;
547 skb_clear_hash(skb);
548 }
549 }
550 return skb;
551}
552EXPORT_SYMBOL(ip_check_defrag);
553
554#ifdef CONFIG_SYSCTL
555static int dist_min;
556
557static struct ctl_table ip4_frags_ns_ctl_table[] = {
558 {
559 .procname = "ipfrag_high_thresh",
560 .maxlen = sizeof(unsigned long),
561 .mode = 0644,
562 .proc_handler = proc_doulongvec_minmax,
563 },
564 {
565 .procname = "ipfrag_low_thresh",
566 .maxlen = sizeof(unsigned long),
567 .mode = 0644,
568 .proc_handler = proc_doulongvec_minmax,
569 },
570 {
571 .procname = "ipfrag_time",
572 .maxlen = sizeof(int),
573 .mode = 0644,
574 .proc_handler = proc_dointvec_jiffies,
575 },
576 {
577 .procname = "ipfrag_max_dist",
578 .maxlen = sizeof(int),
579 .mode = 0644,
580 .proc_handler = proc_dointvec_minmax,
581 .extra1 = &dist_min,
582 },
583 { }
584};
585
586/* secret interval has been deprecated */
587static int ip4_frags_secret_interval_unused;
588static struct ctl_table ip4_frags_ctl_table[] = {
589 {
590 .procname = "ipfrag_secret_interval",
591 .data = &ip4_frags_secret_interval_unused,
592 .maxlen = sizeof(int),
593 .mode = 0644,
594 .proc_handler = proc_dointvec_jiffies,
595 },
596 { }
597};
598
599static int __net_init ip4_frags_ns_ctl_register(struct net *net)
600{
601 struct ctl_table *table;
602 struct ctl_table_header *hdr;
603
604 table = ip4_frags_ns_ctl_table;
605 if (!net_eq(net, &init_net)) {
606 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
607 if (!table)
608 goto err_alloc;
609
610 }
611 table[0].data = &net->ipv4.fqdir->high_thresh;
612 table[0].extra1 = &net->ipv4.fqdir->low_thresh;
613 table[1].data = &net->ipv4.fqdir->low_thresh;
614 table[1].extra2 = &net->ipv4.fqdir->high_thresh;
615 table[2].data = &net->ipv4.fqdir->timeout;
616 table[3].data = &net->ipv4.fqdir->max_dist;
617
618 hdr = register_net_sysctl_sz(net, "net/ipv4", table,
619 ARRAY_SIZE(ip4_frags_ns_ctl_table));
620 if (!hdr)
621 goto err_reg;
622
623 net->ipv4.frags_hdr = hdr;
624 return 0;
625
626err_reg:
627 if (!net_eq(net, &init_net))
628 kfree(table);
629err_alloc:
630 return -ENOMEM;
631}
632
633static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
634{
635 struct ctl_table *table;
636
637 table = net->ipv4.frags_hdr->ctl_table_arg;
638 unregister_net_sysctl_table(net->ipv4.frags_hdr);
639 kfree(table);
640}
641
642static void __init ip4_frags_ctl_register(void)
643{
644 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
645}
646#else
647static int ip4_frags_ns_ctl_register(struct net *net)
648{
649 return 0;
650}
651
652static void ip4_frags_ns_ctl_unregister(struct net *net)
653{
654}
655
656static void __init ip4_frags_ctl_register(void)
657{
658}
659#endif
660
661static int __net_init ipv4_frags_init_net(struct net *net)
662{
663 int res;
664
665 res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
666 if (res < 0)
667 return res;
668 /* Fragment cache limits.
669 *
670 * The fragment memory accounting code, (tries to) account for
671 * the real memory usage, by measuring both the size of frag
672 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
673 * and the SKB's truesize.
674 *
675 * A 64K fragment consumes 129736 bytes (44*2944)+200
676 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
677 *
678 * We will commit 4MB at one time. Should we cross that limit
679 * we will prune down to 3MB, making room for approx 8 big 64K
680 * fragments 8x128k.
681 */
682 net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
683 net->ipv4.fqdir->low_thresh = 3 * 1024 * 1024;
684 /*
685 * Important NOTE! Fragment queue must be destroyed before MSL expires.
686 * RFC791 is wrong proposing to prolongate timer each fragment arrival
687 * by TTL.
688 */
689 net->ipv4.fqdir->timeout = IP_FRAG_TIME;
690
691 net->ipv4.fqdir->max_dist = 64;
692
693 res = ip4_frags_ns_ctl_register(net);
694 if (res < 0)
695 fqdir_exit(net->ipv4.fqdir);
696 return res;
697}
698
699static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
700{
701 fqdir_pre_exit(net->ipv4.fqdir);
702}
703
704static void __net_exit ipv4_frags_exit_net(struct net *net)
705{
706 ip4_frags_ns_ctl_unregister(net);
707 fqdir_exit(net->ipv4.fqdir);
708}
709
710static struct pernet_operations ip4_frags_ops = {
711 .init = ipv4_frags_init_net,
712 .pre_exit = ipv4_frags_pre_exit_net,
713 .exit = ipv4_frags_exit_net,
714};
715
716
717static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
718{
719 return jhash2(data,
720 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
721}
722
723static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
724{
725 const struct inet_frag_queue *fq = data;
726
727 return jhash2((const u32 *)&fq->key.v4,
728 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
729}
730
731static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
732{
733 const struct frag_v4_compare_key *key = arg->key;
734 const struct inet_frag_queue *fq = ptr;
735
736 return !!memcmp(&fq->key, key, sizeof(*key));
737}
738
739static const struct rhashtable_params ip4_rhash_params = {
740 .head_offset = offsetof(struct inet_frag_queue, node),
741 .key_offset = offsetof(struct inet_frag_queue, key),
742 .key_len = sizeof(struct frag_v4_compare_key),
743 .hashfn = ip4_key_hashfn,
744 .obj_hashfn = ip4_obj_hashfn,
745 .obj_cmpfn = ip4_obj_cmpfn,
746 .automatic_shrinking = true,
747};
748
749void __init ipfrag_init(void)
750{
751 ip4_frags.constructor = ip4_frag_init;
752 ip4_frags.destructor = ip4_frag_free;
753 ip4_frags.qsize = sizeof(struct ipq);
754 ip4_frags.frag_expire = ip_expire;
755 ip4_frags.frags_cache_name = ip_frag_cache_name;
756 ip4_frags.rhash_params = ip4_rhash_params;
757 if (inet_frags_init(&ip4_frags))
758 panic("IP: failed to allocate ip4_frags cache\n");
759 ip4_frags_ctl_register();
760 register_pernet_subsys(&ip4_frags_ops);
761}
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The IP fragmentation functionality.
7 *
8 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
9 * Alan Cox <alan@lxorguk.ukuu.org.uk>
10 *
11 * Fixes:
12 * Alan Cox : Split from ip.c , see ip_input.c for history.
13 * David S. Miller : Begin massive cleanup...
14 * Andi Kleen : Add sysctls.
15 * xxxx : Overlapfrag bug.
16 * Ultima : ip_expire() kernel panic.
17 * Bill Hawes : Frag accounting and evictor fixes.
18 * John McDonald : 0 length frag bug.
19 * Alexey Kuznetsov: SMP races, threading, cleanup.
20 * Patrick McHardy : LRU queue of frag heads for evictor.
21 */
22
23#include <linux/compiler.h>
24#include <linux/module.h>
25#include <linux/types.h>
26#include <linux/mm.h>
27#include <linux/jiffies.h>
28#include <linux/skbuff.h>
29#include <linux/list.h>
30#include <linux/ip.h>
31#include <linux/icmp.h>
32#include <linux/netdevice.h>
33#include <linux/jhash.h>
34#include <linux/random.h>
35#include <linux/slab.h>
36#include <net/route.h>
37#include <net/dst.h>
38#include <net/sock.h>
39#include <net/ip.h>
40#include <net/icmp.h>
41#include <net/checksum.h>
42#include <net/inetpeer.h>
43#include <net/inet_frag.h>
44#include <linux/tcp.h>
45#include <linux/udp.h>
46#include <linux/inet.h>
47#include <linux/netfilter_ipv4.h>
48#include <net/inet_ecn.h>
49
50/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
51 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
52 * as well. Or notify me, at least. --ANK
53 */
54
55static int sysctl_ipfrag_max_dist __read_mostly = 64;
56
57struct ipfrag_skb_cb
58{
59 struct inet_skb_parm h;
60 int offset;
61};
62
63#define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
64
65/* Describe an entry in the "incomplete datagrams" queue. */
66struct ipq {
67 struct inet_frag_queue q;
68
69 u32 user;
70 __be32 saddr;
71 __be32 daddr;
72 __be16 id;
73 u8 protocol;
74 u8 ecn; /* RFC3168 support */
75 int iif;
76 unsigned int rid;
77 struct inet_peer *peer;
78};
79
80/* RFC 3168 support :
81 * We want to check ECN values of all fragments, do detect invalid combinations.
82 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
83 */
84#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
85#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
86#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
87#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
88
89static inline u8 ip4_frag_ecn(u8 tos)
90{
91 return 1 << (tos & INET_ECN_MASK);
92}
93
94/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
95 * Value : 0xff if frame should be dropped.
96 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
97 */
98static const u8 ip4_frag_ecn_table[16] = {
99 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
100 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
101 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
103
104 /* invalid combinations : drop frame */
105 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
106 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
112};
113
114static struct inet_frags ip4_frags;
115
116int ip_frag_nqueues(struct net *net)
117{
118 return net->ipv4.frags.nqueues;
119}
120
121int ip_frag_mem(struct net *net)
122{
123 return atomic_read(&net->ipv4.frags.mem);
124}
125
126static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
127 struct net_device *dev);
128
129struct ip4_create_arg {
130 struct iphdr *iph;
131 u32 user;
132};
133
134static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
135{
136 return jhash_3words((__force u32)id << 16 | prot,
137 (__force u32)saddr, (__force u32)daddr,
138 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
139}
140
141static unsigned int ip4_hashfn(struct inet_frag_queue *q)
142{
143 struct ipq *ipq;
144
145 ipq = container_of(q, struct ipq, q);
146 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
147}
148
149static int ip4_frag_match(struct inet_frag_queue *q, void *a)
150{
151 struct ipq *qp;
152 struct ip4_create_arg *arg = a;
153
154 qp = container_of(q, struct ipq, q);
155 return qp->id == arg->iph->id &&
156 qp->saddr == arg->iph->saddr &&
157 qp->daddr == arg->iph->daddr &&
158 qp->protocol == arg->iph->protocol &&
159 qp->user == arg->user;
160}
161
162/* Memory Tracking Functions. */
163static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
164{
165 atomic_sub(skb->truesize, &nf->mem);
166 kfree_skb(skb);
167}
168
169static void ip4_frag_init(struct inet_frag_queue *q, void *a)
170{
171 struct ipq *qp = container_of(q, struct ipq, q);
172 struct ip4_create_arg *arg = a;
173
174 qp->protocol = arg->iph->protocol;
175 qp->id = arg->iph->id;
176 qp->ecn = ip4_frag_ecn(arg->iph->tos);
177 qp->saddr = arg->iph->saddr;
178 qp->daddr = arg->iph->daddr;
179 qp->user = arg->user;
180 qp->peer = sysctl_ipfrag_max_dist ?
181 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
182}
183
184static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
185{
186 struct ipq *qp;
187
188 qp = container_of(q, struct ipq, q);
189 if (qp->peer)
190 inet_putpeer(qp->peer);
191}
192
193
194/* Destruction primitives. */
195
196static __inline__ void ipq_put(struct ipq *ipq)
197{
198 inet_frag_put(&ipq->q, &ip4_frags);
199}
200
201/* Kill ipq entry. It is not destroyed immediately,
202 * because caller (and someone more) holds reference count.
203 */
204static void ipq_kill(struct ipq *ipq)
205{
206 inet_frag_kill(&ipq->q, &ip4_frags);
207}
208
209/* Memory limiting on fragments. Evictor trashes the oldest
210 * fragment queue until we are back under the threshold.
211 */
212static void ip_evictor(struct net *net)
213{
214 int evicted;
215
216 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
217 if (evicted)
218 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
219}
220
221/*
222 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
223 */
224static void ip_expire(unsigned long arg)
225{
226 struct ipq *qp;
227 struct net *net;
228
229 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
230 net = container_of(qp->q.net, struct net, ipv4.frags);
231
232 spin_lock(&qp->q.lock);
233
234 if (qp->q.last_in & INET_FRAG_COMPLETE)
235 goto out;
236
237 ipq_kill(qp);
238
239 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
240 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
241
242 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
243 struct sk_buff *head = qp->q.fragments;
244 const struct iphdr *iph;
245 int err;
246
247 rcu_read_lock();
248 head->dev = dev_get_by_index_rcu(net, qp->iif);
249 if (!head->dev)
250 goto out_rcu_unlock;
251
252 /* skb dst is stale, drop it, and perform route lookup again */
253 skb_dst_drop(head);
254 iph = ip_hdr(head);
255 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
256 iph->tos, head->dev);
257 if (err)
258 goto out_rcu_unlock;
259
260 /*
261 * Only an end host needs to send an ICMP
262 * "Fragment Reassembly Timeout" message, per RFC792.
263 */
264 if (qp->user == IP_DEFRAG_AF_PACKET ||
265 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
266 skb_rtable(head)->rt_type != RTN_LOCAL))
267 goto out_rcu_unlock;
268
269
270 /* Send an ICMP "Fragment Reassembly Timeout" message. */
271 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
272out_rcu_unlock:
273 rcu_read_unlock();
274 }
275out:
276 spin_unlock(&qp->q.lock);
277 ipq_put(qp);
278}
279
280/* Find the correct entry in the "incomplete datagrams" queue for
281 * this IP datagram, and create new one, if nothing is found.
282 */
283static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
284{
285 struct inet_frag_queue *q;
286 struct ip4_create_arg arg;
287 unsigned int hash;
288
289 arg.iph = iph;
290 arg.user = user;
291
292 read_lock(&ip4_frags.lock);
293 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
294
295 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
296 if (q == NULL)
297 goto out_nomem;
298
299 return container_of(q, struct ipq, q);
300
301out_nomem:
302 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
303 return NULL;
304}
305
306/* Is the fragment too far ahead to be part of ipq? */
307static inline int ip_frag_too_far(struct ipq *qp)
308{
309 struct inet_peer *peer = qp->peer;
310 unsigned int max = sysctl_ipfrag_max_dist;
311 unsigned int start, end;
312
313 int rc;
314
315 if (!peer || !max)
316 return 0;
317
318 start = qp->rid;
319 end = atomic_inc_return(&peer->rid);
320 qp->rid = end;
321
322 rc = qp->q.fragments && (end - start) > max;
323
324 if (rc) {
325 struct net *net;
326
327 net = container_of(qp->q.net, struct net, ipv4.frags);
328 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
329 }
330
331 return rc;
332}
333
334static int ip_frag_reinit(struct ipq *qp)
335{
336 struct sk_buff *fp;
337
338 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
339 atomic_inc(&qp->q.refcnt);
340 return -ETIMEDOUT;
341 }
342
343 fp = qp->q.fragments;
344 do {
345 struct sk_buff *xp = fp->next;
346 frag_kfree_skb(qp->q.net, fp);
347 fp = xp;
348 } while (fp);
349
350 qp->q.last_in = 0;
351 qp->q.len = 0;
352 qp->q.meat = 0;
353 qp->q.fragments = NULL;
354 qp->q.fragments_tail = NULL;
355 qp->iif = 0;
356 qp->ecn = 0;
357
358 return 0;
359}
360
361/* Add new segment to existing queue. */
362static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
363{
364 struct sk_buff *prev, *next;
365 struct net_device *dev;
366 int flags, offset;
367 int ihl, end;
368 int err = -ENOENT;
369 u8 ecn;
370
371 if (qp->q.last_in & INET_FRAG_COMPLETE)
372 goto err;
373
374 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
375 unlikely(ip_frag_too_far(qp)) &&
376 unlikely(err = ip_frag_reinit(qp))) {
377 ipq_kill(qp);
378 goto err;
379 }
380
381 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
382 offset = ntohs(ip_hdr(skb)->frag_off);
383 flags = offset & ~IP_OFFSET;
384 offset &= IP_OFFSET;
385 offset <<= 3; /* offset is in 8-byte chunks */
386 ihl = ip_hdrlen(skb);
387
388 /* Determine the position of this fragment. */
389 end = offset + skb->len - ihl;
390 err = -EINVAL;
391
392 /* Is this the final fragment? */
393 if ((flags & IP_MF) == 0) {
394 /* If we already have some bits beyond end
395 * or have different end, the segment is corrrupted.
396 */
397 if (end < qp->q.len ||
398 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
399 goto err;
400 qp->q.last_in |= INET_FRAG_LAST_IN;
401 qp->q.len = end;
402 } else {
403 if (end&7) {
404 end &= ~7;
405 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
406 skb->ip_summed = CHECKSUM_NONE;
407 }
408 if (end > qp->q.len) {
409 /* Some bits beyond end -> corruption. */
410 if (qp->q.last_in & INET_FRAG_LAST_IN)
411 goto err;
412 qp->q.len = end;
413 }
414 }
415 if (end == offset)
416 goto err;
417
418 err = -ENOMEM;
419 if (pskb_pull(skb, ihl) == NULL)
420 goto err;
421
422 err = pskb_trim_rcsum(skb, end - offset);
423 if (err)
424 goto err;
425
426 /* Find out which fragments are in front and at the back of us
427 * in the chain of fragments so far. We must know where to put
428 * this fragment, right?
429 */
430 prev = qp->q.fragments_tail;
431 if (!prev || FRAG_CB(prev)->offset < offset) {
432 next = NULL;
433 goto found;
434 }
435 prev = NULL;
436 for (next = qp->q.fragments; next != NULL; next = next->next) {
437 if (FRAG_CB(next)->offset >= offset)
438 break; /* bingo! */
439 prev = next;
440 }
441
442found:
443 /* We found where to put this one. Check for overlap with
444 * preceding fragment, and, if needed, align things so that
445 * any overlaps are eliminated.
446 */
447 if (prev) {
448 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
449
450 if (i > 0) {
451 offset += i;
452 err = -EINVAL;
453 if (end <= offset)
454 goto err;
455 err = -ENOMEM;
456 if (!pskb_pull(skb, i))
457 goto err;
458 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
459 skb->ip_summed = CHECKSUM_NONE;
460 }
461 }
462
463 err = -ENOMEM;
464
465 while (next && FRAG_CB(next)->offset < end) {
466 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
467
468 if (i < next->len) {
469 /* Eat head of the next overlapped fragment
470 * and leave the loop. The next ones cannot overlap.
471 */
472 if (!pskb_pull(next, i))
473 goto err;
474 FRAG_CB(next)->offset += i;
475 qp->q.meat -= i;
476 if (next->ip_summed != CHECKSUM_UNNECESSARY)
477 next->ip_summed = CHECKSUM_NONE;
478 break;
479 } else {
480 struct sk_buff *free_it = next;
481
482 /* Old fragment is completely overridden with
483 * new one drop it.
484 */
485 next = next->next;
486
487 if (prev)
488 prev->next = next;
489 else
490 qp->q.fragments = next;
491
492 qp->q.meat -= free_it->len;
493 frag_kfree_skb(qp->q.net, free_it);
494 }
495 }
496
497 FRAG_CB(skb)->offset = offset;
498
499 /* Insert this fragment in the chain of fragments. */
500 skb->next = next;
501 if (!next)
502 qp->q.fragments_tail = skb;
503 if (prev)
504 prev->next = skb;
505 else
506 qp->q.fragments = skb;
507
508 dev = skb->dev;
509 if (dev) {
510 qp->iif = dev->ifindex;
511 skb->dev = NULL;
512 }
513 qp->q.stamp = skb->tstamp;
514 qp->q.meat += skb->len;
515 qp->ecn |= ecn;
516 atomic_add(skb->truesize, &qp->q.net->mem);
517 if (offset == 0)
518 qp->q.last_in |= INET_FRAG_FIRST_IN;
519
520 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
521 qp->q.meat == qp->q.len)
522 return ip_frag_reasm(qp, prev, dev);
523
524 write_lock(&ip4_frags.lock);
525 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
526 write_unlock(&ip4_frags.lock);
527 return -EINPROGRESS;
528
529err:
530 kfree_skb(skb);
531 return err;
532}
533
534
535/* Build a new IP datagram from all its fragments. */
536
537static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
538 struct net_device *dev)
539{
540 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
541 struct iphdr *iph;
542 struct sk_buff *fp, *head = qp->q.fragments;
543 int len;
544 int ihlen;
545 int err;
546 u8 ecn;
547
548 ipq_kill(qp);
549
550 ecn = ip4_frag_ecn_table[qp->ecn];
551 if (unlikely(ecn == 0xff)) {
552 err = -EINVAL;
553 goto out_fail;
554 }
555 /* Make the one we just received the head. */
556 if (prev) {
557 head = prev->next;
558 fp = skb_clone(head, GFP_ATOMIC);
559 if (!fp)
560 goto out_nomem;
561
562 fp->next = head->next;
563 if (!fp->next)
564 qp->q.fragments_tail = fp;
565 prev->next = fp;
566
567 skb_morph(head, qp->q.fragments);
568 head->next = qp->q.fragments->next;
569
570 kfree_skb(qp->q.fragments);
571 qp->q.fragments = head;
572 }
573
574 WARN_ON(head == NULL);
575 WARN_ON(FRAG_CB(head)->offset != 0);
576
577 /* Allocate a new buffer for the datagram. */
578 ihlen = ip_hdrlen(head);
579 len = ihlen + qp->q.len;
580
581 err = -E2BIG;
582 if (len > 65535)
583 goto out_oversize;
584
585 /* Head of list must not be cloned. */
586 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
587 goto out_nomem;
588
589 /* If the first fragment is fragmented itself, we split
590 * it to two chunks: the first with data and paged part
591 * and the second, holding only fragments. */
592 if (skb_has_frag_list(head)) {
593 struct sk_buff *clone;
594 int i, plen = 0;
595
596 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
597 goto out_nomem;
598 clone->next = head->next;
599 head->next = clone;
600 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
601 skb_frag_list_init(head);
602 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
603 plen += skb_shinfo(head)->frags[i].size;
604 clone->len = clone->data_len = head->data_len - plen;
605 head->data_len -= clone->len;
606 head->len -= clone->len;
607 clone->csum = 0;
608 clone->ip_summed = head->ip_summed;
609 atomic_add(clone->truesize, &qp->q.net->mem);
610 }
611
612 skb_shinfo(head)->frag_list = head->next;
613 skb_push(head, head->data - skb_network_header(head));
614
615 for (fp=head->next; fp; fp = fp->next) {
616 head->data_len += fp->len;
617 head->len += fp->len;
618 if (head->ip_summed != fp->ip_summed)
619 head->ip_summed = CHECKSUM_NONE;
620 else if (head->ip_summed == CHECKSUM_COMPLETE)
621 head->csum = csum_add(head->csum, fp->csum);
622 head->truesize += fp->truesize;
623 }
624 atomic_sub(head->truesize, &qp->q.net->mem);
625
626 head->next = NULL;
627 head->dev = dev;
628 head->tstamp = qp->q.stamp;
629
630 iph = ip_hdr(head);
631 iph->frag_off = 0;
632 iph->tot_len = htons(len);
633 iph->tos |= ecn;
634 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
635 qp->q.fragments = NULL;
636 qp->q.fragments_tail = NULL;
637 return 0;
638
639out_nomem:
640 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
641 "queue %p\n", qp);
642 err = -ENOMEM;
643 goto out_fail;
644out_oversize:
645 if (net_ratelimit())
646 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
647 &qp->saddr);
648out_fail:
649 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
650 return err;
651}
652
653/* Process an incoming IP datagram fragment. */
654int ip_defrag(struct sk_buff *skb, u32 user)
655{
656 struct ipq *qp;
657 struct net *net;
658
659 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
660 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
661
662 /* Start by cleaning up the memory. */
663 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
664 ip_evictor(net);
665
666 /* Lookup (or create) queue header */
667 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
668 int ret;
669
670 spin_lock(&qp->q.lock);
671
672 ret = ip_frag_queue(qp, skb);
673
674 spin_unlock(&qp->q.lock);
675 ipq_put(qp);
676 return ret;
677 }
678
679 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
680 kfree_skb(skb);
681 return -ENOMEM;
682}
683EXPORT_SYMBOL(ip_defrag);
684
685#ifdef CONFIG_SYSCTL
686static int zero;
687
688static struct ctl_table ip4_frags_ns_ctl_table[] = {
689 {
690 .procname = "ipfrag_high_thresh",
691 .data = &init_net.ipv4.frags.high_thresh,
692 .maxlen = sizeof(int),
693 .mode = 0644,
694 .proc_handler = proc_dointvec
695 },
696 {
697 .procname = "ipfrag_low_thresh",
698 .data = &init_net.ipv4.frags.low_thresh,
699 .maxlen = sizeof(int),
700 .mode = 0644,
701 .proc_handler = proc_dointvec
702 },
703 {
704 .procname = "ipfrag_time",
705 .data = &init_net.ipv4.frags.timeout,
706 .maxlen = sizeof(int),
707 .mode = 0644,
708 .proc_handler = proc_dointvec_jiffies,
709 },
710 { }
711};
712
713static struct ctl_table ip4_frags_ctl_table[] = {
714 {
715 .procname = "ipfrag_secret_interval",
716 .data = &ip4_frags.secret_interval,
717 .maxlen = sizeof(int),
718 .mode = 0644,
719 .proc_handler = proc_dointvec_jiffies,
720 },
721 {
722 .procname = "ipfrag_max_dist",
723 .data = &sysctl_ipfrag_max_dist,
724 .maxlen = sizeof(int),
725 .mode = 0644,
726 .proc_handler = proc_dointvec_minmax,
727 .extra1 = &zero
728 },
729 { }
730};
731
732static int __net_init ip4_frags_ns_ctl_register(struct net *net)
733{
734 struct ctl_table *table;
735 struct ctl_table_header *hdr;
736
737 table = ip4_frags_ns_ctl_table;
738 if (!net_eq(net, &init_net)) {
739 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
740 if (table == NULL)
741 goto err_alloc;
742
743 table[0].data = &net->ipv4.frags.high_thresh;
744 table[1].data = &net->ipv4.frags.low_thresh;
745 table[2].data = &net->ipv4.frags.timeout;
746 }
747
748 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
749 if (hdr == NULL)
750 goto err_reg;
751
752 net->ipv4.frags_hdr = hdr;
753 return 0;
754
755err_reg:
756 if (!net_eq(net, &init_net))
757 kfree(table);
758err_alloc:
759 return -ENOMEM;
760}
761
762static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
763{
764 struct ctl_table *table;
765
766 table = net->ipv4.frags_hdr->ctl_table_arg;
767 unregister_net_sysctl_table(net->ipv4.frags_hdr);
768 kfree(table);
769}
770
771static void ip4_frags_ctl_register(void)
772{
773 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
774}
775#else
776static inline int ip4_frags_ns_ctl_register(struct net *net)
777{
778 return 0;
779}
780
781static inline void ip4_frags_ns_ctl_unregister(struct net *net)
782{
783}
784
785static inline void ip4_frags_ctl_register(void)
786{
787}
788#endif
789
790static int __net_init ipv4_frags_init_net(struct net *net)
791{
792 /*
793 * Fragment cache limits. We will commit 256K at one time. Should we
794 * cross that limit we will prune down to 192K. This should cope with
795 * even the most extreme cases without allowing an attacker to
796 * measurably harm machine performance.
797 */
798 net->ipv4.frags.high_thresh = 256 * 1024;
799 net->ipv4.frags.low_thresh = 192 * 1024;
800 /*
801 * Important NOTE! Fragment queue must be destroyed before MSL expires.
802 * RFC791 is wrong proposing to prolongate timer each fragment arrival
803 * by TTL.
804 */
805 net->ipv4.frags.timeout = IP_FRAG_TIME;
806
807 inet_frags_init_net(&net->ipv4.frags);
808
809 return ip4_frags_ns_ctl_register(net);
810}
811
812static void __net_exit ipv4_frags_exit_net(struct net *net)
813{
814 ip4_frags_ns_ctl_unregister(net);
815 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
816}
817
818static struct pernet_operations ip4_frags_ops = {
819 .init = ipv4_frags_init_net,
820 .exit = ipv4_frags_exit_net,
821};
822
823void __init ipfrag_init(void)
824{
825 ip4_frags_ctl_register();
826 register_pernet_subsys(&ip4_frags_ops);
827 ip4_frags.hashfn = ip4_hashfn;
828 ip4_frags.constructor = ip4_frag_init;
829 ip4_frags.destructor = ip4_frag_free;
830 ip4_frags.skb_free = NULL;
831 ip4_frags.qsize = sizeof(struct ipq);
832 ip4_frags.match = ip4_frag_match;
833 ip4_frags.frag_expire = ip_expire;
834 ip4_frags.secret_interval = 10 * 60 * HZ;
835 inet_frags_init(&ip4_frags);
836}