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