1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The IP fragmentation functionality.
7 *
8 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
9 * Alan Cox <alan@lxorguk.ukuu.org.uk>
10 *
11 * Fixes:
12 * Alan Cox : Split from ip.c , see ip_input.c for history.
13 * David S. Miller : Begin massive cleanup...
14 * Andi Kleen : Add sysctls.
15 * xxxx : Overlapfrag bug.
16 * Ultima : ip_expire() kernel panic.
17 * Bill Hawes : Frag accounting and evictor fixes.
18 * John McDonald : 0 length frag bug.
19 * Alexey Kuznetsov: SMP races, threading, cleanup.
20 * Patrick McHardy : LRU queue of frag heads for evictor.
21 */
22
23 #define pr_fmt(fmt) "IPv4: " fmt
24
25 #include <linux/compiler.h>
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/mm.h>
29 #include <linux/jiffies.h>
30 #include <linux/skbuff.h>
31 #include <linux/list.h>
32 #include <linux/ip.h>
33 #include <linux/icmp.h>
34 #include <linux/netdevice.h>
35 #include <linux/jhash.h>
36 #include <linux/random.h>
37 #include <linux/slab.h>
38 #include <net/route.h>
39 #include <net/dst.h>
40 #include <net/sock.h>
41 #include <net/ip.h>
42 #include <net/icmp.h>
43 #include <net/checksum.h>
44 #include <net/inetpeer.h>
45 #include <net/inet_frag.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/inet.h>
49 #include <linux/netfilter_ipv4.h>
50 #include <net/inet_ecn.h>
51
52 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
54 * as well. Or notify me, at least. --ANK
55 */
56
57 static int sysctl_ipfrag_max_dist __read_mostly = 64;
58
59 struct ipfrag_skb_cb
60 {
61 struct inet_skb_parm h;
62 int offset;
63 };
64
65 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
66
67 /* Describe an entry in the "incomplete datagrams" queue. */
68 struct ipq {
69 struct inet_frag_queue q;
70
71 u32 user;
72 __be32 saddr;
73 __be32 daddr;
74 __be16 id;
75 u8 protocol;
76 u8 ecn; /* RFC3168 support */
77 int iif;
78 unsigned int rid;
79 struct inet_peer *peer;
80 };
81
82 /* RFC 3168 support :
83 * We want to check ECN values of all fragments, do detect invalid combinations.
84 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
85 */
86 #define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
87 #define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
88 #define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
89 #define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
90
ip4_frag_ecn(u8 tos)91 static inline u8 ip4_frag_ecn(u8 tos)
92 {
93 return 1 << (tos & INET_ECN_MASK);
94 }
95
96 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
97 * Value : 0xff if frame should be dropped.
98 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
99 */
100 static const u8 ip4_frag_ecn_table[16] = {
101 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
103 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
104 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
105
106 /* invalid combinations : drop frame */
107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
112 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
113 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
114 };
115
116 static struct inet_frags ip4_frags;
117
ip_frag_nqueues(struct net * net)118 int ip_frag_nqueues(struct net *net)
119 {
120 return net->ipv4.frags.nqueues;
121 }
122
ip_frag_mem(struct net * net)123 int ip_frag_mem(struct net *net)
124 {
125 return atomic_read(&net->ipv4.frags.mem);
126 }
127
128 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
129 struct net_device *dev);
130
131 struct ip4_create_arg {
132 struct iphdr *iph;
133 u32 user;
134 };
135
ipqhashfn(__be16 id,__be32 saddr,__be32 daddr,u8 prot)136 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
137 {
138 return jhash_3words((__force u32)id << 16 | prot,
139 (__force u32)saddr, (__force u32)daddr,
140 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
141 }
142
ip4_hashfn(struct inet_frag_queue * q)143 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
144 {
145 struct ipq *ipq;
146
147 ipq = container_of(q, struct ipq, q);
148 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
149 }
150
ip4_frag_match(struct inet_frag_queue * q,void * a)151 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
152 {
153 struct ipq *qp;
154 struct ip4_create_arg *arg = a;
155
156 qp = container_of(q, struct ipq, q);
157 return qp->id == arg->iph->id &&
158 qp->saddr == arg->iph->saddr &&
159 qp->daddr == arg->iph->daddr &&
160 qp->protocol == arg->iph->protocol &&
161 qp->user == arg->user;
162 }
163
164 /* Memory Tracking Functions. */
frag_kfree_skb(struct netns_frags * nf,struct sk_buff * skb)165 static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
166 {
167 atomic_sub(skb->truesize, &nf->mem);
168 kfree_skb(skb);
169 }
170
ip4_frag_init(struct inet_frag_queue * q,void * a)171 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
172 {
173 struct ipq *qp = container_of(q, struct ipq, q);
174 struct ip4_create_arg *arg = a;
175
176 qp->protocol = arg->iph->protocol;
177 qp->id = arg->iph->id;
178 qp->ecn = ip4_frag_ecn(arg->iph->tos);
179 qp->saddr = arg->iph->saddr;
180 qp->daddr = arg->iph->daddr;
181 qp->user = arg->user;
182 qp->peer = sysctl_ipfrag_max_dist ?
183 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
184 }
185
ip4_frag_free(struct inet_frag_queue * q)186 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
187 {
188 struct ipq *qp;
189
190 qp = container_of(q, struct ipq, q);
191 if (qp->peer)
192 inet_putpeer(qp->peer);
193 }
194
195
196 /* Destruction primitives. */
197
ipq_put(struct ipq * ipq)198 static __inline__ void ipq_put(struct ipq *ipq)
199 {
200 inet_frag_put(&ipq->q, &ip4_frags);
201 }
202
203 /* Kill ipq entry. It is not destroyed immediately,
204 * because caller (and someone more) holds reference count.
205 */
ipq_kill(struct ipq * ipq)206 static void ipq_kill(struct ipq *ipq)
207 {
208 inet_frag_kill(&ipq->q, &ip4_frags);
209 }
210
211 /* Memory limiting on fragments. Evictor trashes the oldest
212 * fragment queue until we are back under the threshold.
213 */
ip_evictor(struct net * net)214 static void ip_evictor(struct net *net)
215 {
216 int evicted;
217
218 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
219 if (evicted)
220 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
221 }
222
223 /*
224 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
225 */
ip_expire(unsigned long arg)226 static void ip_expire(unsigned long arg)
227 {
228 struct ipq *qp;
229 struct net *net;
230
231 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
232 net = container_of(qp->q.net, struct net, ipv4.frags);
233
234 spin_lock(&qp->q.lock);
235
236 if (qp->q.last_in & INET_FRAG_COMPLETE)
237 goto out;
238
239 ipq_kill(qp);
240
241 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
242 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
243
244 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
245 struct sk_buff *head = qp->q.fragments;
246 const struct iphdr *iph;
247 int err;
248
249 rcu_read_lock();
250 head->dev = dev_get_by_index_rcu(net, qp->iif);
251 if (!head->dev)
252 goto out_rcu_unlock;
253
254 /* skb has no dst, perform route lookup again */
255 iph = ip_hdr(head);
256 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
257 iph->tos, head->dev);
258 if (err)
259 goto out_rcu_unlock;
260
261 /*
262 * Only an end host needs to send an ICMP
263 * "Fragment Reassembly Timeout" message, per RFC792.
264 */
265 if (qp->user == IP_DEFRAG_AF_PACKET ||
266 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
267 skb_rtable(head)->rt_type != RTN_LOCAL))
268 goto out_rcu_unlock;
269
270
271 /* Send an ICMP "Fragment Reassembly Timeout" message. */
272 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
273 out_rcu_unlock:
274 rcu_read_unlock();
275 }
276 out:
277 spin_unlock(&qp->q.lock);
278 ipq_put(qp);
279 }
280
281 /* Find the correct entry in the "incomplete datagrams" queue for
282 * this IP datagram, and create new one, if nothing is found.
283 */
ip_find(struct net * net,struct iphdr * iph,u32 user)284 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
285 {
286 struct inet_frag_queue *q;
287 struct ip4_create_arg arg;
288 unsigned int hash;
289
290 arg.iph = iph;
291 arg.user = user;
292
293 read_lock(&ip4_frags.lock);
294 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
295
296 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
297 if (IS_ERR_OR_NULL(q)) {
298 inet_frag_maybe_warn_overflow(q, pr_fmt());
299 return NULL;
300 }
301 return container_of(q, struct ipq, q);
302 }
303
304 /* Is the fragment too far ahead to be part of ipq? */
ip_frag_too_far(struct ipq * qp)305 static inline int ip_frag_too_far(struct ipq *qp)
306 {
307 struct inet_peer *peer = qp->peer;
308 unsigned int max = sysctl_ipfrag_max_dist;
309 unsigned int start, end;
310
311 int rc;
312
313 if (!peer || !max)
314 return 0;
315
316 start = qp->rid;
317 end = atomic_inc_return(&peer->rid);
318 qp->rid = end;
319
320 rc = qp->q.fragments && (end - start) > max;
321
322 if (rc) {
323 struct net *net;
324
325 net = container_of(qp->q.net, struct net, ipv4.frags);
326 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
327 }
328
329 return rc;
330 }
331
ip_frag_reinit(struct ipq * qp)332 static int ip_frag_reinit(struct ipq *qp)
333 {
334 struct sk_buff *fp;
335
336 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
337 atomic_inc(&qp->q.refcnt);
338 return -ETIMEDOUT;
339 }
340
341 fp = qp->q.fragments;
342 do {
343 struct sk_buff *xp = fp->next;
344 frag_kfree_skb(qp->q.net, fp);
345 fp = xp;
346 } while (fp);
347
348 qp->q.last_in = 0;
349 qp->q.len = 0;
350 qp->q.meat = 0;
351 qp->q.fragments = NULL;
352 qp->q.fragments_tail = NULL;
353 qp->iif = 0;
354 qp->ecn = 0;
355
356 return 0;
357 }
358
359 /* Add new segment to existing queue. */
ip_frag_queue(struct ipq * qp,struct sk_buff * skb)360 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
361 {
362 struct sk_buff *prev, *next;
363 struct net_device *dev;
364 int flags, offset;
365 int ihl, end;
366 int err = -ENOENT;
367 u8 ecn;
368
369 if (qp->q.last_in & INET_FRAG_COMPLETE)
370 goto err;
371
372 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
373 unlikely(ip_frag_too_far(qp)) &&
374 unlikely(err = ip_frag_reinit(qp))) {
375 ipq_kill(qp);
376 goto err;
377 }
378
379 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
380 offset = ntohs(ip_hdr(skb)->frag_off);
381 flags = offset & ~IP_OFFSET;
382 offset &= IP_OFFSET;
383 offset <<= 3; /* offset is in 8-byte chunks */
384 ihl = ip_hdrlen(skb);
385
386 /* Determine the position of this fragment. */
387 end = offset + skb->len - ihl;
388 err = -EINVAL;
389
390 /* Is this the final fragment? */
391 if ((flags & IP_MF) == 0) {
392 /* If we already have some bits beyond end
393 * or have different end, the segment is corrupted.
394 */
395 if (end < qp->q.len ||
396 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
397 goto err;
398 qp->q.last_in |= INET_FRAG_LAST_IN;
399 qp->q.len = end;
400 } else {
401 if (end&7) {
402 end &= ~7;
403 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
404 skb->ip_summed = CHECKSUM_NONE;
405 }
406 if (end > qp->q.len) {
407 /* Some bits beyond end -> corruption. */
408 if (qp->q.last_in & INET_FRAG_LAST_IN)
409 goto err;
410 qp->q.len = end;
411 }
412 }
413 if (end == offset)
414 goto err;
415
416 err = -ENOMEM;
417 if (pskb_pull(skb, ihl) == NULL)
418 goto err;
419
420 err = pskb_trim_rcsum(skb, end - offset);
421 if (err)
422 goto err;
423
424 /* Find out which fragments are in front and at the back of us
425 * in the chain of fragments so far. We must know where to put
426 * this fragment, right?
427 */
428 prev = qp->q.fragments_tail;
429 if (!prev || FRAG_CB(prev)->offset < offset) {
430 next = NULL;
431 goto found;
432 }
433 prev = NULL;
434 for (next = qp->q.fragments; next != NULL; next = next->next) {
435 if (FRAG_CB(next)->offset >= offset)
436 break; /* bingo! */
437 prev = next;
438 }
439
440 found:
441 /* We found where to put this one. Check for overlap with
442 * preceding fragment, and, if needed, align things so that
443 * any overlaps are eliminated.
444 */
445 if (prev) {
446 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
447
448 if (i > 0) {
449 offset += i;
450 err = -EINVAL;
451 if (end <= offset)
452 goto err;
453 err = -ENOMEM;
454 if (!pskb_pull(skb, i))
455 goto err;
456 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
457 skb->ip_summed = CHECKSUM_NONE;
458 }
459 }
460
461 err = -ENOMEM;
462
463 while (next && FRAG_CB(next)->offset < end) {
464 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
465
466 if (i < next->len) {
467 /* Eat head of the next overlapped fragment
468 * and leave the loop. The next ones cannot overlap.
469 */
470 if (!pskb_pull(next, i))
471 goto err;
472 FRAG_CB(next)->offset += i;
473 qp->q.meat -= i;
474 if (next->ip_summed != CHECKSUM_UNNECESSARY)
475 next->ip_summed = CHECKSUM_NONE;
476 break;
477 } else {
478 struct sk_buff *free_it = next;
479
480 /* Old fragment is completely overridden with
481 * new one drop it.
482 */
483 next = next->next;
484
485 if (prev)
486 prev->next = next;
487 else
488 qp->q.fragments = next;
489
490 qp->q.meat -= free_it->len;
491 frag_kfree_skb(qp->q.net, free_it);
492 }
493 }
494
495 FRAG_CB(skb)->offset = offset;
496
497 /* Insert this fragment in the chain of fragments. */
498 skb->next = next;
499 if (!next)
500 qp->q.fragments_tail = skb;
501 if (prev)
502 prev->next = skb;
503 else
504 qp->q.fragments = skb;
505
506 dev = skb->dev;
507 if (dev) {
508 qp->iif = dev->ifindex;
509 skb->dev = NULL;
510 }
511 qp->q.stamp = skb->tstamp;
512 qp->q.meat += skb->len;
513 qp->ecn |= ecn;
514 atomic_add(skb->truesize, &qp->q.net->mem);
515 if (offset == 0)
516 qp->q.last_in |= INET_FRAG_FIRST_IN;
517
518 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
519 qp->q.meat == qp->q.len) {
520 unsigned long orefdst = skb->_skb_refdst;
521
522 skb->_skb_refdst = 0UL;
523 err = ip_frag_reasm(qp, prev, dev);
524 skb->_skb_refdst = orefdst;
525 return err;
526 }
527
528 skb_dst_drop(skb);
529
530 write_lock(&ip4_frags.lock);
531 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
532 write_unlock(&ip4_frags.lock);
533 return -EINPROGRESS;
534
535 err:
536 kfree_skb(skb);
537 return err;
538 }
539
540
541 /* Build a new IP datagram from all its fragments. */
542
ip_frag_reasm(struct ipq * qp,struct sk_buff * prev,struct net_device * dev)543 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
544 struct net_device *dev)
545 {
546 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
547 struct iphdr *iph;
548 struct sk_buff *fp, *head = qp->q.fragments;
549 int len;
550 int ihlen;
551 int err;
552 u8 ecn;
553
554 ipq_kill(qp);
555
556 ecn = ip4_frag_ecn_table[qp->ecn];
557 if (unlikely(ecn == 0xff)) {
558 err = -EINVAL;
559 goto out_fail;
560 }
561 /* Make the one we just received the head. */
562 if (prev) {
563 head = prev->next;
564 fp = skb_clone(head, GFP_ATOMIC);
565 if (!fp)
566 goto out_nomem;
567
568 fp->next = head->next;
569 if (!fp->next)
570 qp->q.fragments_tail = fp;
571 prev->next = fp;
572
573 skb_morph(head, qp->q.fragments);
574 head->next = qp->q.fragments->next;
575
576 kfree_skb(qp->q.fragments);
577 qp->q.fragments = head;
578 }
579
580 WARN_ON(head == NULL);
581 WARN_ON(FRAG_CB(head)->offset != 0);
582
583 /* Allocate a new buffer for the datagram. */
584 ihlen = ip_hdrlen(head);
585 len = ihlen + qp->q.len;
586
587 err = -E2BIG;
588 if (len > 65535)
589 goto out_oversize;
590
591 /* Head of list must not be cloned. */
592 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
593 goto out_nomem;
594
595 /* If the first fragment is fragmented itself, we split
596 * it to two chunks: the first with data and paged part
597 * and the second, holding only fragments. */
598 if (skb_has_frag_list(head)) {
599 struct sk_buff *clone;
600 int i, plen = 0;
601
602 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
603 goto out_nomem;
604 clone->next = head->next;
605 head->next = clone;
606 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
607 skb_frag_list_init(head);
608 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
609 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
610 clone->len = clone->data_len = head->data_len - plen;
611 head->data_len -= clone->len;
612 head->len -= clone->len;
613 clone->csum = 0;
614 clone->ip_summed = head->ip_summed;
615 atomic_add(clone->truesize, &qp->q.net->mem);
616 }
617
618 skb_shinfo(head)->frag_list = head->next;
619 skb_push(head, head->data - skb_network_header(head));
620
621 for (fp=head->next; fp; fp = fp->next) {
622 head->data_len += fp->len;
623 head->len += fp->len;
624 if (head->ip_summed != fp->ip_summed)
625 head->ip_summed = CHECKSUM_NONE;
626 else if (head->ip_summed == CHECKSUM_COMPLETE)
627 head->csum = csum_add(head->csum, fp->csum);
628 head->truesize += fp->truesize;
629 }
630 atomic_sub(head->truesize, &qp->q.net->mem);
631
632 head->next = NULL;
633 head->dev = dev;
634 head->tstamp = qp->q.stamp;
635
636 iph = ip_hdr(head);
637 iph->frag_off = 0;
638 iph->tot_len = htons(len);
639 iph->tos |= ecn;
640 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
641 qp->q.fragments = NULL;
642 qp->q.fragments_tail = NULL;
643 return 0;
644
645 out_nomem:
646 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
647 qp);
648 err = -ENOMEM;
649 goto out_fail;
650 out_oversize:
651 if (net_ratelimit())
652 pr_info("Oversized IP packet from %pI4\n", &qp->saddr);
653 out_fail:
654 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
655 return err;
656 }
657
658 /* Process an incoming IP datagram fragment. */
ip_defrag(struct sk_buff * skb,u32 user)659 int ip_defrag(struct sk_buff *skb, u32 user)
660 {
661 struct ipq *qp;
662 struct net *net;
663
664 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
665 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
666
667 /* Start by cleaning up the memory. */
668 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
669 ip_evictor(net);
670
671 /* Lookup (or create) queue header */
672 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
673 int ret;
674
675 spin_lock(&qp->q.lock);
676
677 ret = ip_frag_queue(qp, skb);
678
679 spin_unlock(&qp->q.lock);
680 ipq_put(qp);
681 return ret;
682 }
683
684 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
685 kfree_skb(skb);
686 return -ENOMEM;
687 }
688 EXPORT_SYMBOL(ip_defrag);
689
ip_check_defrag(struct sk_buff * skb,u32 user)690 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
691 {
692 struct iphdr iph;
693 u32 len;
694
695 if (skb->protocol != htons(ETH_P_IP))
696 return skb;
697
698 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
699 return skb;
700
701 if (iph.ihl < 5 || iph.version != 4)
702 return skb;
703
704 len = ntohs(iph.tot_len);
705 if (skb->len < len || len < (iph.ihl * 4))
706 return skb;
707
708 if (ip_is_fragment(&iph)) {
709 skb = skb_share_check(skb, GFP_ATOMIC);
710 if (skb) {
711 if (!pskb_may_pull(skb, iph.ihl*4))
712 return skb;
713 if (pskb_trim_rcsum(skb, len))
714 return skb;
715 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
716 if (ip_defrag(skb, user))
717 return NULL;
718 skb->rxhash = 0;
719 }
720 }
721 return skb;
722 }
723 EXPORT_SYMBOL(ip_check_defrag);
724
725 #ifdef CONFIG_SYSCTL
726 static int zero;
727
728 static struct ctl_table ip4_frags_ns_ctl_table[] = {
729 {
730 .procname = "ipfrag_high_thresh",
731 .data = &init_net.ipv4.frags.high_thresh,
732 .maxlen = sizeof(int),
733 .mode = 0644,
734 .proc_handler = proc_dointvec
735 },
736 {
737 .procname = "ipfrag_low_thresh",
738 .data = &init_net.ipv4.frags.low_thresh,
739 .maxlen = sizeof(int),
740 .mode = 0644,
741 .proc_handler = proc_dointvec
742 },
743 {
744 .procname = "ipfrag_time",
745 .data = &init_net.ipv4.frags.timeout,
746 .maxlen = sizeof(int),
747 .mode = 0644,
748 .proc_handler = proc_dointvec_jiffies,
749 },
750 { }
751 };
752
753 static struct ctl_table ip4_frags_ctl_table[] = {
754 {
755 .procname = "ipfrag_secret_interval",
756 .data = &ip4_frags.secret_interval,
757 .maxlen = sizeof(int),
758 .mode = 0644,
759 .proc_handler = proc_dointvec_jiffies,
760 },
761 {
762 .procname = "ipfrag_max_dist",
763 .data = &sysctl_ipfrag_max_dist,
764 .maxlen = sizeof(int),
765 .mode = 0644,
766 .proc_handler = proc_dointvec_minmax,
767 .extra1 = &zero
768 },
769 { }
770 };
771
ip4_frags_ns_ctl_register(struct net * net)772 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
773 {
774 struct ctl_table *table;
775 struct ctl_table_header *hdr;
776
777 table = ip4_frags_ns_ctl_table;
778 if (!net_eq(net, &init_net)) {
779 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
780 if (table == NULL)
781 goto err_alloc;
782
783 table[0].data = &net->ipv4.frags.high_thresh;
784 table[1].data = &net->ipv4.frags.low_thresh;
785 table[2].data = &net->ipv4.frags.timeout;
786 }
787
788 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
789 if (hdr == NULL)
790 goto err_reg;
791
792 net->ipv4.frags_hdr = hdr;
793 return 0;
794
795 err_reg:
796 if (!net_eq(net, &init_net))
797 kfree(table);
798 err_alloc:
799 return -ENOMEM;
800 }
801
ip4_frags_ns_ctl_unregister(struct net * net)802 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
803 {
804 struct ctl_table *table;
805
806 table = net->ipv4.frags_hdr->ctl_table_arg;
807 unregister_net_sysctl_table(net->ipv4.frags_hdr);
808 kfree(table);
809 }
810
ip4_frags_ctl_register(void)811 static void ip4_frags_ctl_register(void)
812 {
813 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
814 }
815 #else
ip4_frags_ns_ctl_register(struct net * net)816 static inline int ip4_frags_ns_ctl_register(struct net *net)
817 {
818 return 0;
819 }
820
ip4_frags_ns_ctl_unregister(struct net * net)821 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
822 {
823 }
824
ip4_frags_ctl_register(void)825 static inline void ip4_frags_ctl_register(void)
826 {
827 }
828 #endif
829
ipv4_frags_init_net(struct net * net)830 static int __net_init ipv4_frags_init_net(struct net *net)
831 {
832 /*
833 * Fragment cache limits. We will commit 256K at one time. Should we
834 * cross that limit we will prune down to 192K. This should cope with
835 * even the most extreme cases without allowing an attacker to
836 * measurably harm machine performance.
837 */
838 net->ipv4.frags.high_thresh = 256 * 1024;
839 net->ipv4.frags.low_thresh = 192 * 1024;
840 /*
841 * Important NOTE! Fragment queue must be destroyed before MSL expires.
842 * RFC791 is wrong proposing to prolongate timer each fragment arrival
843 * by TTL.
844 */
845 net->ipv4.frags.timeout = IP_FRAG_TIME;
846
847 inet_frags_init_net(&net->ipv4.frags);
848
849 return ip4_frags_ns_ctl_register(net);
850 }
851
ipv4_frags_exit_net(struct net * net)852 static void __net_exit ipv4_frags_exit_net(struct net *net)
853 {
854 ip4_frags_ns_ctl_unregister(net);
855 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
856 }
857
858 static struct pernet_operations ip4_frags_ops = {
859 .init = ipv4_frags_init_net,
860 .exit = ipv4_frags_exit_net,
861 };
862
ipfrag_init(void)863 void __init ipfrag_init(void)
864 {
865 ip4_frags_ctl_register();
866 register_pernet_subsys(&ip4_frags_ops);
867 ip4_frags.hashfn = ip4_hashfn;
868 ip4_frags.constructor = ip4_frag_init;
869 ip4_frags.destructor = ip4_frag_free;
870 ip4_frags.skb_free = NULL;
871 ip4_frags.qsize = sizeof(struct ipq);
872 ip4_frags.match = ip4_frag_match;
873 ip4_frags.frag_expire = ip_expire;
874 ip4_frags.secret_interval = 10 * 60 * HZ;
875 inet_frags_init(&ip4_frags);
876 }
877