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 <net/sock.h>
36 #include <net/ip.h>
37 #include <net/icmp.h>
38 #include <net/checksum.h>
39 #include <net/inetpeer.h>
40 #include <net/inet_frag.h>
41 #include <linux/tcp.h>
42 #include <linux/udp.h>
43 #include <linux/inet.h>
44 #include <linux/netfilter_ipv4.h>
45
46 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
47 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
48 * as well. Or notify me, at least. --ANK
49 */
50
51 static int sysctl_ipfrag_max_dist __read_mostly = 64;
52
53 struct ipfrag_skb_cb
54 {
55 struct inet_skb_parm h;
56 int offset;
57 };
58
59 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
60
61 /* Describe an entry in the "incomplete datagrams" queue. */
62 struct ipq {
63 struct inet_frag_queue q;
64
65 u32 user;
66 __be32 saddr;
67 __be32 daddr;
68 __be16 id;
69 u8 protocol;
70 int iif;
71 unsigned int rid;
72 struct inet_peer *peer;
73 };
74
75 static struct inet_frags ip4_frags;
76
ip_frag_nqueues(struct net * net)77 int ip_frag_nqueues(struct net *net)
78 {
79 return net->ipv4.frags.nqueues;
80 }
81
ip_frag_mem(struct net * net)82 int ip_frag_mem(struct net *net)
83 {
84 return atomic_read(&net->ipv4.frags.mem);
85 }
86
87 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
88 struct net_device *dev);
89
90 struct ip4_create_arg {
91 struct iphdr *iph;
92 u32 user;
93 };
94
ipqhashfn(__be16 id,__be32 saddr,__be32 daddr,u8 prot)95 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
96 {
97 return jhash_3words((__force u32)id << 16 | prot,
98 (__force u32)saddr, (__force u32)daddr,
99 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
100 }
101
ip4_hashfn(struct inet_frag_queue * q)102 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
103 {
104 struct ipq *ipq;
105
106 ipq = container_of(q, struct ipq, q);
107 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
108 }
109
ip4_frag_match(struct inet_frag_queue * q,void * a)110 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
111 {
112 struct ipq *qp;
113 struct ip4_create_arg *arg = a;
114
115 qp = container_of(q, struct ipq, q);
116 return (qp->id == arg->iph->id &&
117 qp->saddr == arg->iph->saddr &&
118 qp->daddr == arg->iph->daddr &&
119 qp->protocol == arg->iph->protocol &&
120 qp->user == arg->user);
121 }
122
123 /* Memory Tracking Functions. */
frag_kfree_skb(struct netns_frags * nf,struct sk_buff * skb,int * work)124 static __inline__ void frag_kfree_skb(struct netns_frags *nf,
125 struct sk_buff *skb, int *work)
126 {
127 if (work)
128 *work -= skb->truesize;
129 atomic_sub(skb->truesize, &nf->mem);
130 kfree_skb(skb);
131 }
132
ip4_frag_init(struct inet_frag_queue * q,void * a)133 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
134 {
135 struct ipq *qp = container_of(q, struct ipq, q);
136 struct ip4_create_arg *arg = a;
137
138 qp->protocol = arg->iph->protocol;
139 qp->id = arg->iph->id;
140 qp->saddr = arg->iph->saddr;
141 qp->daddr = arg->iph->daddr;
142 qp->user = arg->user;
143 qp->peer = sysctl_ipfrag_max_dist ?
144 inet_getpeer(arg->iph->saddr, 1) : NULL;
145 }
146
ip4_frag_free(struct inet_frag_queue * q)147 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
148 {
149 struct ipq *qp;
150
151 qp = container_of(q, struct ipq, q);
152 if (qp->peer)
153 inet_putpeer(qp->peer);
154 }
155
156
157 /* Destruction primitives. */
158
ipq_put(struct ipq * ipq)159 static __inline__ void ipq_put(struct ipq *ipq)
160 {
161 inet_frag_put(&ipq->q, &ip4_frags);
162 }
163
164 /* Kill ipq entry. It is not destroyed immediately,
165 * because caller (and someone more) holds reference count.
166 */
ipq_kill(struct ipq * ipq)167 static void ipq_kill(struct ipq *ipq)
168 {
169 inet_frag_kill(&ipq->q, &ip4_frags);
170 }
171
172 /* Memory limiting on fragments. Evictor trashes the oldest
173 * fragment queue until we are back under the threshold.
174 */
ip_evictor(struct net * net)175 static void ip_evictor(struct net *net)
176 {
177 int evicted;
178
179 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
180 if (evicted)
181 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
182 }
183
184 /*
185 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
186 */
ip_expire(unsigned long arg)187 static void ip_expire(unsigned long arg)
188 {
189 struct ipq *qp;
190 struct net *net;
191
192 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
193 net = container_of(qp->q.net, struct net, ipv4.frags);
194
195 spin_lock(&qp->q.lock);
196
197 if (qp->q.last_in & INET_FRAG_COMPLETE)
198 goto out;
199
200 ipq_kill(qp);
201
202 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
203 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
204
205 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
206 struct sk_buff *head = qp->q.fragments;
207
208 /* Send an ICMP "Fragment Reassembly Timeout" message. */
209 if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) {
210 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
211 dev_put(head->dev);
212 }
213 }
214 out:
215 spin_unlock(&qp->q.lock);
216 ipq_put(qp);
217 }
218
219 /* Find the correct entry in the "incomplete datagrams" queue for
220 * this IP datagram, and create new one, if nothing is found.
221 */
ip_find(struct net * net,struct iphdr * iph,u32 user)222 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
223 {
224 struct inet_frag_queue *q;
225 struct ip4_create_arg arg;
226 unsigned int hash;
227
228 arg.iph = iph;
229 arg.user = user;
230
231 read_lock(&ip4_frags.lock);
232 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
233
234 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
235 if (q == NULL)
236 goto out_nomem;
237
238 return container_of(q, struct ipq, q);
239
240 out_nomem:
241 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
242 return NULL;
243 }
244
245 /* Is the fragment too far ahead to be part of ipq? */
ip_frag_too_far(struct ipq * qp)246 static inline int ip_frag_too_far(struct ipq *qp)
247 {
248 struct inet_peer *peer = qp->peer;
249 unsigned int max = sysctl_ipfrag_max_dist;
250 unsigned int start, end;
251
252 int rc;
253
254 if (!peer || !max)
255 return 0;
256
257 start = qp->rid;
258 end = atomic_inc_return(&peer->rid);
259 qp->rid = end;
260
261 rc = qp->q.fragments && (end - start) > max;
262
263 if (rc) {
264 struct net *net;
265
266 net = container_of(qp->q.net, struct net, ipv4.frags);
267 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
268 }
269
270 return rc;
271 }
272
ip_frag_reinit(struct ipq * qp)273 static int ip_frag_reinit(struct ipq *qp)
274 {
275 struct sk_buff *fp;
276
277 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
278 atomic_inc(&qp->q.refcnt);
279 return -ETIMEDOUT;
280 }
281
282 fp = qp->q.fragments;
283 do {
284 struct sk_buff *xp = fp->next;
285 frag_kfree_skb(qp->q.net, fp, NULL);
286 fp = xp;
287 } while (fp);
288
289 qp->q.last_in = 0;
290 qp->q.len = 0;
291 qp->q.meat = 0;
292 qp->q.fragments = NULL;
293 qp->iif = 0;
294
295 return 0;
296 }
297
298 /* Add new segment to existing queue. */
ip_frag_queue(struct ipq * qp,struct sk_buff * skb)299 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
300 {
301 struct sk_buff *prev, *next;
302 struct net_device *dev;
303 int flags, offset;
304 int ihl, end;
305 int err = -ENOENT;
306
307 if (qp->q.last_in & INET_FRAG_COMPLETE)
308 goto err;
309
310 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
311 unlikely(ip_frag_too_far(qp)) &&
312 unlikely(err = ip_frag_reinit(qp))) {
313 ipq_kill(qp);
314 goto err;
315 }
316
317 offset = ntohs(ip_hdr(skb)->frag_off);
318 flags = offset & ~IP_OFFSET;
319 offset &= IP_OFFSET;
320 offset <<= 3; /* offset is in 8-byte chunks */
321 ihl = ip_hdrlen(skb);
322
323 /* Determine the position of this fragment. */
324 end = offset + skb->len - ihl;
325 err = -EINVAL;
326
327 /* Is this the final fragment? */
328 if ((flags & IP_MF) == 0) {
329 /* If we already have some bits beyond end
330 * or have different end, the segment is corrrupted.
331 */
332 if (end < qp->q.len ||
333 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
334 goto err;
335 qp->q.last_in |= INET_FRAG_LAST_IN;
336 qp->q.len = end;
337 } else {
338 if (end&7) {
339 end &= ~7;
340 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
341 skb->ip_summed = CHECKSUM_NONE;
342 }
343 if (end > qp->q.len) {
344 /* Some bits beyond end -> corruption. */
345 if (qp->q.last_in & INET_FRAG_LAST_IN)
346 goto err;
347 qp->q.len = end;
348 }
349 }
350 if (end == offset)
351 goto err;
352
353 err = -ENOMEM;
354 if (pskb_pull(skb, ihl) == NULL)
355 goto err;
356
357 err = pskb_trim_rcsum(skb, end - offset);
358 if (err)
359 goto err;
360
361 /* Find out which fragments are in front and at the back of us
362 * in the chain of fragments so far. We must know where to put
363 * this fragment, right?
364 */
365 prev = NULL;
366 for (next = qp->q.fragments; next != NULL; next = next->next) {
367 if (FRAG_CB(next)->offset >= offset)
368 break; /* bingo! */
369 prev = next;
370 }
371
372 /* We found where to put this one. Check for overlap with
373 * preceding fragment, and, if needed, align things so that
374 * any overlaps are eliminated.
375 */
376 if (prev) {
377 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
378
379 if (i > 0) {
380 offset += i;
381 err = -EINVAL;
382 if (end <= offset)
383 goto err;
384 err = -ENOMEM;
385 if (!pskb_pull(skb, i))
386 goto err;
387 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
388 skb->ip_summed = CHECKSUM_NONE;
389 }
390 }
391
392 err = -ENOMEM;
393
394 while (next && FRAG_CB(next)->offset < end) {
395 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
396
397 if (i < next->len) {
398 /* Eat head of the next overlapped fragment
399 * and leave the loop. The next ones cannot overlap.
400 */
401 if (!pskb_pull(next, i))
402 goto err;
403 FRAG_CB(next)->offset += i;
404 qp->q.meat -= i;
405 if (next->ip_summed != CHECKSUM_UNNECESSARY)
406 next->ip_summed = CHECKSUM_NONE;
407 break;
408 } else {
409 struct sk_buff *free_it = next;
410
411 /* Old fragment is completely overridden with
412 * new one drop it.
413 */
414 next = next->next;
415
416 if (prev)
417 prev->next = next;
418 else
419 qp->q.fragments = next;
420
421 qp->q.meat -= free_it->len;
422 frag_kfree_skb(qp->q.net, free_it, NULL);
423 }
424 }
425
426 FRAG_CB(skb)->offset = offset;
427
428 /* Insert this fragment in the chain of fragments. */
429 skb->next = next;
430 if (prev)
431 prev->next = skb;
432 else
433 qp->q.fragments = skb;
434
435 dev = skb->dev;
436 if (dev) {
437 qp->iif = dev->ifindex;
438 skb->dev = NULL;
439 }
440 qp->q.stamp = skb->tstamp;
441 qp->q.meat += skb->len;
442 atomic_add(skb->truesize, &qp->q.net->mem);
443 if (offset == 0)
444 qp->q.last_in |= INET_FRAG_FIRST_IN;
445
446 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
447 qp->q.meat == qp->q.len)
448 return ip_frag_reasm(qp, prev, dev);
449
450 write_lock(&ip4_frags.lock);
451 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
452 write_unlock(&ip4_frags.lock);
453 return -EINPROGRESS;
454
455 err:
456 kfree_skb(skb);
457 return err;
458 }
459
460
461 /* Build a new IP datagram from all its fragments. */
462
ip_frag_reasm(struct ipq * qp,struct sk_buff * prev,struct net_device * dev)463 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
464 struct net_device *dev)
465 {
466 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
467 struct iphdr *iph;
468 struct sk_buff *fp, *head = qp->q.fragments;
469 int len;
470 int ihlen;
471 int err;
472
473 ipq_kill(qp);
474
475 /* Make the one we just received the head. */
476 if (prev) {
477 head = prev->next;
478 fp = skb_clone(head, GFP_ATOMIC);
479 if (!fp)
480 goto out_nomem;
481
482 fp->next = head->next;
483 prev->next = fp;
484
485 skb_morph(head, qp->q.fragments);
486 head->next = qp->q.fragments->next;
487
488 kfree_skb(qp->q.fragments);
489 qp->q.fragments = head;
490 }
491
492 WARN_ON(head == NULL);
493 WARN_ON(FRAG_CB(head)->offset != 0);
494
495 /* Allocate a new buffer for the datagram. */
496 ihlen = ip_hdrlen(head);
497 len = ihlen + qp->q.len;
498
499 err = -E2BIG;
500 if (len > 65535)
501 goto out_oversize;
502
503 /* Head of list must not be cloned. */
504 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
505 goto out_nomem;
506
507 /* If the first fragment is fragmented itself, we split
508 * it to two chunks: the first with data and paged part
509 * and the second, holding only fragments. */
510 if (skb_shinfo(head)->frag_list) {
511 struct sk_buff *clone;
512 int i, plen = 0;
513
514 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
515 goto out_nomem;
516 clone->next = head->next;
517 head->next = clone;
518 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
519 skb_shinfo(head)->frag_list = NULL;
520 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
521 plen += skb_shinfo(head)->frags[i].size;
522 clone->len = clone->data_len = head->data_len - plen;
523 head->data_len -= clone->len;
524 head->len -= clone->len;
525 clone->csum = 0;
526 clone->ip_summed = head->ip_summed;
527 atomic_add(clone->truesize, &qp->q.net->mem);
528 }
529
530 skb_shinfo(head)->frag_list = head->next;
531 skb_push(head, head->data - skb_network_header(head));
532 atomic_sub(head->truesize, &qp->q.net->mem);
533
534 for (fp=head->next; fp; fp = fp->next) {
535 head->data_len += fp->len;
536 head->len += fp->len;
537 if (head->ip_summed != fp->ip_summed)
538 head->ip_summed = CHECKSUM_NONE;
539 else if (head->ip_summed == CHECKSUM_COMPLETE)
540 head->csum = csum_add(head->csum, fp->csum);
541 head->truesize += fp->truesize;
542 atomic_sub(fp->truesize, &qp->q.net->mem);
543 }
544
545 head->next = NULL;
546 head->dev = dev;
547 head->tstamp = qp->q.stamp;
548
549 iph = ip_hdr(head);
550 iph->frag_off = 0;
551 iph->tot_len = htons(len);
552 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
553 qp->q.fragments = NULL;
554 return 0;
555
556 out_nomem:
557 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
558 "queue %p\n", qp);
559 err = -ENOMEM;
560 goto out_fail;
561 out_oversize:
562 if (net_ratelimit())
563 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
564 &qp->saddr);
565 out_fail:
566 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_REASMFAILS);
567 return err;
568 }
569
570 /* Process an incoming IP datagram fragment. */
ip_defrag(struct sk_buff * skb,u32 user)571 int ip_defrag(struct sk_buff *skb, u32 user)
572 {
573 struct ipq *qp;
574 struct net *net;
575
576 net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev);
577 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
578
579 /* Start by cleaning up the memory. */
580 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
581 ip_evictor(net);
582
583 /* Lookup (or create) queue header */
584 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
585 int ret;
586
587 spin_lock(&qp->q.lock);
588
589 ret = ip_frag_queue(qp, skb);
590
591 spin_unlock(&qp->q.lock);
592 ipq_put(qp);
593 return ret;
594 }
595
596 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
597 kfree_skb(skb);
598 return -ENOMEM;
599 }
600
601 #ifdef CONFIG_SYSCTL
602 static int zero;
603
604 static struct ctl_table ip4_frags_ns_ctl_table[] = {
605 {
606 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
607 .procname = "ipfrag_high_thresh",
608 .data = &init_net.ipv4.frags.high_thresh,
609 .maxlen = sizeof(int),
610 .mode = 0644,
611 .proc_handler = proc_dointvec
612 },
613 {
614 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
615 .procname = "ipfrag_low_thresh",
616 .data = &init_net.ipv4.frags.low_thresh,
617 .maxlen = sizeof(int),
618 .mode = 0644,
619 .proc_handler = proc_dointvec
620 },
621 {
622 .ctl_name = NET_IPV4_IPFRAG_TIME,
623 .procname = "ipfrag_time",
624 .data = &init_net.ipv4.frags.timeout,
625 .maxlen = sizeof(int),
626 .mode = 0644,
627 .proc_handler = proc_dointvec_jiffies,
628 .strategy = sysctl_jiffies
629 },
630 { }
631 };
632
633 static struct ctl_table ip4_frags_ctl_table[] = {
634 {
635 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
636 .procname = "ipfrag_secret_interval",
637 .data = &ip4_frags.secret_interval,
638 .maxlen = sizeof(int),
639 .mode = 0644,
640 .proc_handler = proc_dointvec_jiffies,
641 .strategy = sysctl_jiffies
642 },
643 {
644 .procname = "ipfrag_max_dist",
645 .data = &sysctl_ipfrag_max_dist,
646 .maxlen = sizeof(int),
647 .mode = 0644,
648 .proc_handler = proc_dointvec_minmax,
649 .extra1 = &zero
650 },
651 { }
652 };
653
ip4_frags_ns_ctl_register(struct net * net)654 static int ip4_frags_ns_ctl_register(struct net *net)
655 {
656 struct ctl_table *table;
657 struct ctl_table_header *hdr;
658
659 table = ip4_frags_ns_ctl_table;
660 if (net != &init_net) {
661 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
662 if (table == NULL)
663 goto err_alloc;
664
665 table[0].data = &net->ipv4.frags.high_thresh;
666 table[1].data = &net->ipv4.frags.low_thresh;
667 table[2].data = &net->ipv4.frags.timeout;
668 }
669
670 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
671 if (hdr == NULL)
672 goto err_reg;
673
674 net->ipv4.frags_hdr = hdr;
675 return 0;
676
677 err_reg:
678 if (net != &init_net)
679 kfree(table);
680 err_alloc:
681 return -ENOMEM;
682 }
683
ip4_frags_ns_ctl_unregister(struct net * net)684 static void ip4_frags_ns_ctl_unregister(struct net *net)
685 {
686 struct ctl_table *table;
687
688 table = net->ipv4.frags_hdr->ctl_table_arg;
689 unregister_net_sysctl_table(net->ipv4.frags_hdr);
690 kfree(table);
691 }
692
ip4_frags_ctl_register(void)693 static void ip4_frags_ctl_register(void)
694 {
695 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
696 }
697 #else
ip4_frags_ns_ctl_register(struct net * net)698 static inline int ip4_frags_ns_ctl_register(struct net *net)
699 {
700 return 0;
701 }
702
ip4_frags_ns_ctl_unregister(struct net * net)703 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
704 {
705 }
706
ip4_frags_ctl_register(void)707 static inline void ip4_frags_ctl_register(void)
708 {
709 }
710 #endif
711
ipv4_frags_init_net(struct net * net)712 static int ipv4_frags_init_net(struct net *net)
713 {
714 /*
715 * Fragment cache limits. We will commit 256K at one time. Should we
716 * cross that limit we will prune down to 192K. This should cope with
717 * even the most extreme cases without allowing an attacker to
718 * measurably harm machine performance.
719 */
720 net->ipv4.frags.high_thresh = 256 * 1024;
721 net->ipv4.frags.low_thresh = 192 * 1024;
722 /*
723 * Important NOTE! Fragment queue must be destroyed before MSL expires.
724 * RFC791 is wrong proposing to prolongate timer each fragment arrival
725 * by TTL.
726 */
727 net->ipv4.frags.timeout = IP_FRAG_TIME;
728
729 inet_frags_init_net(&net->ipv4.frags);
730
731 return ip4_frags_ns_ctl_register(net);
732 }
733
ipv4_frags_exit_net(struct net * net)734 static void ipv4_frags_exit_net(struct net *net)
735 {
736 ip4_frags_ns_ctl_unregister(net);
737 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
738 }
739
740 static struct pernet_operations ip4_frags_ops = {
741 .init = ipv4_frags_init_net,
742 .exit = ipv4_frags_exit_net,
743 };
744
ipfrag_init(void)745 void __init ipfrag_init(void)
746 {
747 ip4_frags_ctl_register();
748 register_pernet_subsys(&ip4_frags_ops);
749 ip4_frags.hashfn = ip4_hashfn;
750 ip4_frags.constructor = ip4_frag_init;
751 ip4_frags.destructor = ip4_frag_free;
752 ip4_frags.skb_free = NULL;
753 ip4_frags.qsize = sizeof(struct ipq);
754 ip4_frags.match = ip4_frag_match;
755 ip4_frags.frag_expire = ip_expire;
756 ip4_frags.secret_interval = 10 * 60 * HZ;
757 inet_frags_init(&ip4_frags);
758 }
759
760 EXPORT_SYMBOL(ip_defrag);
761