blob: 2a8adda15e1199c3d849d4e51ad899eabfeb43c3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 * Version: $Id: ip_fragment.c,v 1.59 2002/01/12 07:54:56 davem Exp $
9 *
10 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox <Alan.Cox@linux.org>
12 *
13 * Fixes:
14 * Alan Cox : Split from ip.c , see ip_input.c for history.
15 * David S. Miller : Begin massive cleanup...
16 * Andi Kleen : Add sysctls.
17 * xxxx : Overlapfrag bug.
18 * Ultima : ip_expire() kernel panic.
19 * Bill Hawes : Frag accounting and evictor fixes.
20 * John McDonald : 0 length frag bug.
21 * Alexey Kuznetsov: SMP races, threading, cleanup.
22 * Patrick McHardy : LRU queue of frag heads for evictor.
23 */
24
Herbert Xu89cee8b2005-12-13 23:14:27 -080025#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/config.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 <net/sock.h>
39#include <net/ip.h>
40#include <net/icmp.h>
41#include <net/checksum.h>
Herbert Xu89cee8b2005-12-13 23:14:27 -080042#include <net/inetpeer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/tcp.h>
44#include <linux/udp.h>
45#include <linux/inet.h>
46#include <linux/netfilter_ipv4.h>
47
48/* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
49 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
50 * as well. Or notify me, at least. --ANK
51 */
52
53/* Fragment cache limits. We will commit 256K at one time. Should we
54 * cross that limit we will prune down to 192K. This should cope with
55 * even the most extreme cases without allowing an attacker to measurably
56 * harm machine performance.
57 */
58int sysctl_ipfrag_high_thresh = 256*1024;
59int sysctl_ipfrag_low_thresh = 192*1024;
60
Herbert Xu89cee8b2005-12-13 23:14:27 -080061int sysctl_ipfrag_max_dist = 64;
62
Linus Torvalds1da177e2005-04-16 15:20:36 -070063/* Important NOTE! Fragment queue must be destroyed before MSL expires.
64 * RFC791 is wrong proposing to prolongate timer each fragment arrival by TTL.
65 */
66int sysctl_ipfrag_time = IP_FRAG_TIME;
67
68struct ipfrag_skb_cb
69{
70 struct inet_skb_parm h;
71 int offset;
72};
73
74#define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb))
75
76/* Describe an entry in the "incomplete datagrams" queue. */
77struct ipq {
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -080078 struct hlist_node list;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 struct list_head lru_list; /* lru list member */
80 u32 user;
81 u32 saddr;
82 u32 daddr;
83 u16 id;
84 u8 protocol;
85 u8 last_in;
86#define COMPLETE 4
87#define FIRST_IN 2
88#define LAST_IN 1
89
90 struct sk_buff *fragments; /* linked list of received fragments */
91 int len; /* total length of original datagram */
92 int meat;
93 spinlock_t lock;
94 atomic_t refcnt;
95 struct timer_list timer; /* when will this queue expire? */
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 struct timeval stamp;
Herbert Xu89cee8b2005-12-13 23:14:27 -080097 int iif;
98 unsigned int rid;
99 struct inet_peer *peer;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100};
101
102/* Hash table. */
103
104#define IPQ_HASHSZ 64
105
106/* Per-bucket lock is easy to add now. */
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800107static struct hlist_head ipq_hash[IPQ_HASHSZ];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108static DEFINE_RWLOCK(ipfrag_lock);
109static u32 ipfrag_hash_rnd;
110static LIST_HEAD(ipq_lru_list);
111int ip_frag_nqueues = 0;
112
113static __inline__ void __ipq_unlink(struct ipq *qp)
114{
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800115 hlist_del(&qp->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 list_del(&qp->lru_list);
117 ip_frag_nqueues--;
118}
119
120static __inline__ void ipq_unlink(struct ipq *ipq)
121{
122 write_lock(&ipfrag_lock);
123 __ipq_unlink(ipq);
124 write_unlock(&ipfrag_lock);
125}
126
127static unsigned int ipqhashfn(u16 id, u32 saddr, u32 daddr, u8 prot)
128{
129 return jhash_3words((u32)id << 16 | prot, saddr, daddr,
130 ipfrag_hash_rnd) & (IPQ_HASHSZ - 1);
131}
132
133static struct timer_list ipfrag_secret_timer;
134int sysctl_ipfrag_secret_interval = 10 * 60 * HZ;
135
136static void ipfrag_secret_rebuild(unsigned long dummy)
137{
138 unsigned long now = jiffies;
139 int i;
140
141 write_lock(&ipfrag_lock);
142 get_random_bytes(&ipfrag_hash_rnd, sizeof(u32));
143 for (i = 0; i < IPQ_HASHSZ; i++) {
144 struct ipq *q;
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800145 struct hlist_node *p, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800147 hlist_for_each_entry_safe(q, p, n, &ipq_hash[i], list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 unsigned int hval = ipqhashfn(q->id, q->saddr,
149 q->daddr, q->protocol);
150
151 if (hval != i) {
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800152 hlist_del(&q->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154 /* Relink to new hash chain. */
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800155 hlist_add_head(&q->list, &ipq_hash[hval]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
158 }
159 write_unlock(&ipfrag_lock);
160
161 mod_timer(&ipfrag_secret_timer, now + sysctl_ipfrag_secret_interval);
162}
163
164atomic_t ip_frag_mem = ATOMIC_INIT(0); /* Memory used for fragments */
165
166/* Memory Tracking Functions. */
167static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work)
168{
169 if (work)
170 *work -= skb->truesize;
171 atomic_sub(skb->truesize, &ip_frag_mem);
172 kfree_skb(skb);
173}
174
175static __inline__ void frag_free_queue(struct ipq *qp, int *work)
176{
177 if (work)
178 *work -= sizeof(struct ipq);
179 atomic_sub(sizeof(struct ipq), &ip_frag_mem);
180 kfree(qp);
181}
182
183static __inline__ struct ipq *frag_alloc_queue(void)
184{
185 struct ipq *qp = kmalloc(sizeof(struct ipq), GFP_ATOMIC);
186
187 if(!qp)
188 return NULL;
189 atomic_add(sizeof(struct ipq), &ip_frag_mem);
190 return qp;
191}
192
193
194/* Destruction primitives. */
195
196/* Complete destruction of ipq. */
197static void ip_frag_destroy(struct ipq *qp, int *work)
198{
199 struct sk_buff *fp;
200
201 BUG_TRAP(qp->last_in&COMPLETE);
202 BUG_TRAP(del_timer(&qp->timer) == 0);
203
Herbert Xu89cee8b2005-12-13 23:14:27 -0800204 if (qp->peer)
205 inet_putpeer(qp->peer);
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 /* Release all fragment data. */
208 fp = qp->fragments;
209 while (fp) {
210 struct sk_buff *xp = fp->next;
211
212 frag_kfree_skb(fp, work);
213 fp = xp;
214 }
215
216 /* Finally, release the queue descriptor itself. */
217 frag_free_queue(qp, work);
218}
219
220static __inline__ void ipq_put(struct ipq *ipq, int *work)
221{
222 if (atomic_dec_and_test(&ipq->refcnt))
223 ip_frag_destroy(ipq, work);
224}
225
226/* Kill ipq entry. It is not destroyed immediately,
227 * because caller (and someone more) holds reference count.
228 */
229static void ipq_kill(struct ipq *ipq)
230{
231 if (del_timer(&ipq->timer))
232 atomic_dec(&ipq->refcnt);
233
234 if (!(ipq->last_in & COMPLETE)) {
235 ipq_unlink(ipq);
236 atomic_dec(&ipq->refcnt);
237 ipq->last_in |= COMPLETE;
238 }
239}
240
241/* Memory limiting on fragments. Evictor trashes the oldest
242 * fragment queue until we are back under the threshold.
243 */
244static void ip_evictor(void)
245{
246 struct ipq *qp;
247 struct list_head *tmp;
248 int work;
249
250 work = atomic_read(&ip_frag_mem) - sysctl_ipfrag_low_thresh;
251 if (work <= 0)
252 return;
253
254 while (work > 0) {
255 read_lock(&ipfrag_lock);
256 if (list_empty(&ipq_lru_list)) {
257 read_unlock(&ipfrag_lock);
258 return;
259 }
260 tmp = ipq_lru_list.next;
261 qp = list_entry(tmp, struct ipq, lru_list);
262 atomic_inc(&qp->refcnt);
263 read_unlock(&ipfrag_lock);
264
265 spin_lock(&qp->lock);
266 if (!(qp->last_in&COMPLETE))
267 ipq_kill(qp);
268 spin_unlock(&qp->lock);
269
270 ipq_put(qp, &work);
271 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
272 }
273}
274
275/*
276 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
277 */
278static void ip_expire(unsigned long arg)
279{
280 struct ipq *qp = (struct ipq *) arg;
281
282 spin_lock(&qp->lock);
283
284 if (qp->last_in & COMPLETE)
285 goto out;
286
287 ipq_kill(qp);
288
289 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
290 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
291
292 if ((qp->last_in&FIRST_IN) && qp->fragments != NULL) {
293 struct sk_buff *head = qp->fragments;
294 /* Send an ICMP "Fragment Reassembly Timeout" message. */
295 if ((head->dev = dev_get_by_index(qp->iif)) != NULL) {
296 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
297 dev_put(head->dev);
298 }
299 }
300out:
301 spin_unlock(&qp->lock);
302 ipq_put(qp, NULL);
303}
304
305/* Creation primitives. */
306
307static struct ipq *ip_frag_intern(unsigned int hash, struct ipq *qp_in)
308{
309 struct ipq *qp;
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800310#ifdef CONFIG_SMP
311 struct hlist_node *n;
312#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 write_lock(&ipfrag_lock);
314#ifdef CONFIG_SMP
315 /* With SMP race we have to recheck hash table, because
316 * such entry could be created on other cpu, while we
317 * promoted read lock to write lock.
318 */
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800319 hlist_for_each_entry(qp, n, &ipq_hash[hash], list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 if(qp->id == qp_in->id &&
321 qp->saddr == qp_in->saddr &&
322 qp->daddr == qp_in->daddr &&
323 qp->protocol == qp_in->protocol &&
324 qp->user == qp_in->user) {
325 atomic_inc(&qp->refcnt);
326 write_unlock(&ipfrag_lock);
327 qp_in->last_in |= COMPLETE;
328 ipq_put(qp_in, NULL);
329 return qp;
330 }
331 }
332#endif
333 qp = qp_in;
334
335 if (!mod_timer(&qp->timer, jiffies + sysctl_ipfrag_time))
336 atomic_inc(&qp->refcnt);
337
338 atomic_inc(&qp->refcnt);
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800339 hlist_add_head(&qp->list, &ipq_hash[hash]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 INIT_LIST_HEAD(&qp->lru_list);
341 list_add_tail(&qp->lru_list, &ipq_lru_list);
342 ip_frag_nqueues++;
343 write_unlock(&ipfrag_lock);
344 return qp;
345}
346
347/* Add an entry to the 'ipq' queue for a newly received IP datagram. */
348static struct ipq *ip_frag_create(unsigned hash, struct iphdr *iph, u32 user)
349{
350 struct ipq *qp;
351
352 if ((qp = frag_alloc_queue()) == NULL)
353 goto out_nomem;
354
355 qp->protocol = iph->protocol;
356 qp->last_in = 0;
357 qp->id = iph->id;
358 qp->saddr = iph->saddr;
359 qp->daddr = iph->daddr;
360 qp->user = user;
361 qp->len = 0;
362 qp->meat = 0;
363 qp->fragments = NULL;
364 qp->iif = 0;
Herbert Xu89cee8b2005-12-13 23:14:27 -0800365 qp->peer = sysctl_ipfrag_max_dist ? inet_getpeer(iph->saddr, 1) : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 /* Initialize a timer for this entry. */
368 init_timer(&qp->timer);
369 qp->timer.data = (unsigned long) qp; /* pointer to queue */
370 qp->timer.function = ip_expire; /* expire function */
371 spin_lock_init(&qp->lock);
372 atomic_set(&qp->refcnt, 1);
373
374 return ip_frag_intern(hash, qp);
375
376out_nomem:
Patrick McHardy64ce2072005-08-09 20:50:53 -0700377 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return NULL;
379}
380
381/* Find the correct entry in the "incomplete datagrams" queue for
382 * this IP datagram, and create new one, if nothing is found.
383 */
384static inline struct ipq *ip_find(struct iphdr *iph, u32 user)
385{
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800386 __be16 id = iph->id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 __u32 saddr = iph->saddr;
388 __u32 daddr = iph->daddr;
389 __u8 protocol = iph->protocol;
390 unsigned int hash = ipqhashfn(id, saddr, daddr, protocol);
391 struct ipq *qp;
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800392 struct hlist_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 read_lock(&ipfrag_lock);
Yasuyuki Kozakaie7c8a412005-11-16 12:55:37 -0800395 hlist_for_each_entry(qp, n, &ipq_hash[hash], list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 if(qp->id == id &&
397 qp->saddr == saddr &&
398 qp->daddr == daddr &&
399 qp->protocol == protocol &&
400 qp->user == user) {
401 atomic_inc(&qp->refcnt);
402 read_unlock(&ipfrag_lock);
403 return qp;
404 }
405 }
406 read_unlock(&ipfrag_lock);
407
408 return ip_frag_create(hash, iph, user);
409}
410
Herbert Xu89cee8b2005-12-13 23:14:27 -0800411/* Is the fragment too far ahead to be part of ipq? */
412static inline int ip_frag_too_far(struct ipq *qp)
413{
414 struct inet_peer *peer = qp->peer;
415 unsigned int max = sysctl_ipfrag_max_dist;
416 unsigned int start, end;
417
418 int rc;
419
420 if (!peer || !max)
421 return 0;
422
423 start = qp->rid;
424 end = atomic_inc_return(&peer->rid);
425 qp->rid = end;
426
427 rc = qp->fragments && (end - start) > max;
428
429 if (rc) {
430 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
431 }
432
433 return rc;
434}
435
436static int ip_frag_reinit(struct ipq *qp)
437{
438 struct sk_buff *fp;
439
440 if (!mod_timer(&qp->timer, jiffies + sysctl_ipfrag_time)) {
441 atomic_inc(&qp->refcnt);
442 return -ETIMEDOUT;
443 }
444
445 fp = qp->fragments;
446 do {
447 struct sk_buff *xp = fp->next;
448 frag_kfree_skb(fp, NULL);
449 fp = xp;
450 } while (fp);
451
452 qp->last_in = 0;
453 qp->len = 0;
454 qp->meat = 0;
455 qp->fragments = NULL;
456 qp->iif = 0;
457
458 return 0;
459}
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461/* Add new segment to existing queue. */
462static void ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
463{
464 struct sk_buff *prev, *next;
465 int flags, offset;
466 int ihl, end;
467
468 if (qp->last_in & COMPLETE)
469 goto err;
470
Herbert Xu89cee8b2005-12-13 23:14:27 -0800471 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
472 unlikely(ip_frag_too_far(qp)) && unlikely(ip_frag_reinit(qp))) {
473 ipq_kill(qp);
474 goto err;
475 }
476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 offset = ntohs(skb->nh.iph->frag_off);
478 flags = offset & ~IP_OFFSET;
479 offset &= IP_OFFSET;
480 offset <<= 3; /* offset is in 8-byte chunks */
481 ihl = skb->nh.iph->ihl * 4;
482
483 /* Determine the position of this fragment. */
484 end = offset + skb->len - ihl;
485
486 /* Is this the final fragment? */
487 if ((flags & IP_MF) == 0) {
488 /* If we already have some bits beyond end
489 * or have different end, the segment is corrrupted.
490 */
491 if (end < qp->len ||
492 ((qp->last_in & LAST_IN) && end != qp->len))
493 goto err;
494 qp->last_in |= LAST_IN;
495 qp->len = end;
496 } else {
497 if (end&7) {
498 end &= ~7;
499 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
500 skb->ip_summed = CHECKSUM_NONE;
501 }
502 if (end > qp->len) {
503 /* Some bits beyond end -> corruption. */
504 if (qp->last_in & LAST_IN)
505 goto err;
506 qp->len = end;
507 }
508 }
509 if (end == offset)
510 goto err;
511
512 if (pskb_pull(skb, ihl) == NULL)
513 goto err;
Stephen Hemminger48bc41a2005-09-06 15:51:48 -0700514 if (pskb_trim_rcsum(skb, end-offset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 goto err;
516
517 /* Find out which fragments are in front and at the back of us
518 * in the chain of fragments so far. We must know where to put
519 * this fragment, right?
520 */
521 prev = NULL;
522 for(next = qp->fragments; next != NULL; next = next->next) {
523 if (FRAG_CB(next)->offset >= offset)
524 break; /* bingo! */
525 prev = next;
526 }
527
528 /* We found where to put this one. Check for overlap with
529 * preceding fragment, and, if needed, align things so that
530 * any overlaps are eliminated.
531 */
532 if (prev) {
533 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
534
535 if (i > 0) {
536 offset += i;
537 if (end <= offset)
538 goto err;
539 if (!pskb_pull(skb, i))
540 goto err;
541 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
542 skb->ip_summed = CHECKSUM_NONE;
543 }
544 }
545
546 while (next && FRAG_CB(next)->offset < end) {
547 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
548
549 if (i < next->len) {
550 /* Eat head of the next overlapped fragment
551 * and leave the loop. The next ones cannot overlap.
552 */
553 if (!pskb_pull(next, i))
554 goto err;
555 FRAG_CB(next)->offset += i;
556 qp->meat -= i;
557 if (next->ip_summed != CHECKSUM_UNNECESSARY)
558 next->ip_summed = CHECKSUM_NONE;
559 break;
560 } else {
561 struct sk_buff *free_it = next;
562
563 /* Old fragmnet is completely overridden with
564 * new one drop it.
565 */
566 next = next->next;
567
568 if (prev)
569 prev->next = next;
570 else
571 qp->fragments = next;
572
573 qp->meat -= free_it->len;
574 frag_kfree_skb(free_it, NULL);
575 }
576 }
577
578 FRAG_CB(skb)->offset = offset;
579
580 /* Insert this fragment in the chain of fragments. */
581 skb->next = next;
582 if (prev)
583 prev->next = skb;
584 else
585 qp->fragments = skb;
586
587 if (skb->dev)
588 qp->iif = skb->dev->ifindex;
589 skb->dev = NULL;
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700590 skb_get_timestamp(skb, &qp->stamp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 qp->meat += skb->len;
592 atomic_add(skb->truesize, &ip_frag_mem);
593 if (offset == 0)
594 qp->last_in |= FIRST_IN;
595
596 write_lock(&ipfrag_lock);
597 list_move_tail(&qp->lru_list, &ipq_lru_list);
598 write_unlock(&ipfrag_lock);
599
600 return;
601
602err:
603 kfree_skb(skb);
604}
605
606
607/* Build a new IP datagram from all its fragments. */
608
609static struct sk_buff *ip_frag_reasm(struct ipq *qp, struct net_device *dev)
610{
611 struct iphdr *iph;
612 struct sk_buff *fp, *head = qp->fragments;
613 int len;
614 int ihlen;
615
616 ipq_kill(qp);
617
618 BUG_TRAP(head != NULL);
619 BUG_TRAP(FRAG_CB(head)->offset == 0);
620
621 /* Allocate a new buffer for the datagram. */
622 ihlen = head->nh.iph->ihl*4;
623 len = ihlen + qp->len;
624
625 if(len > 65535)
626 goto out_oversize;
627
628 /* Head of list must not be cloned. */
629 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
630 goto out_nomem;
631
632 /* If the first fragment is fragmented itself, we split
633 * it to two chunks: the first with data and paged part
634 * and the second, holding only fragments. */
635 if (skb_shinfo(head)->frag_list) {
636 struct sk_buff *clone;
637 int i, plen = 0;
638
639 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
640 goto out_nomem;
641 clone->next = head->next;
642 head->next = clone;
643 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
644 skb_shinfo(head)->frag_list = NULL;
645 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
646 plen += skb_shinfo(head)->frags[i].size;
647 clone->len = clone->data_len = head->data_len - plen;
648 head->data_len -= clone->len;
649 head->len -= clone->len;
650 clone->csum = 0;
651 clone->ip_summed = head->ip_summed;
652 atomic_add(clone->truesize, &ip_frag_mem);
653 }
654
655 skb_shinfo(head)->frag_list = head->next;
656 skb_push(head, head->data - head->nh.raw);
657 atomic_sub(head->truesize, &ip_frag_mem);
658
659 for (fp=head->next; fp; fp = fp->next) {
660 head->data_len += fp->len;
661 head->len += fp->len;
662 if (head->ip_summed != fp->ip_summed)
663 head->ip_summed = CHECKSUM_NONE;
664 else if (head->ip_summed == CHECKSUM_HW)
665 head->csum = csum_add(head->csum, fp->csum);
666 head->truesize += fp->truesize;
667 atomic_sub(fp->truesize, &ip_frag_mem);
668 }
669
670 head->next = NULL;
671 head->dev = dev;
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700672 skb_set_timestamp(head, &qp->stamp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
674 iph = head->nh.iph;
675 iph->frag_off = 0;
676 iph->tot_len = htons(len);
677 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
678 qp->fragments = NULL;
679 return head;
680
681out_nomem:
Patrick McHardy64ce2072005-08-09 20:50:53 -0700682 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
683 "queue %p\n", qp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 goto out_fail;
685out_oversize:
686 if (net_ratelimit())
687 printk(KERN_INFO
688 "Oversized IP packet from %d.%d.%d.%d.\n",
689 NIPQUAD(qp->saddr));
690out_fail:
691 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
692 return NULL;
693}
694
695/* Process an incoming IP datagram fragment. */
696struct sk_buff *ip_defrag(struct sk_buff *skb, u32 user)
697{
698 struct iphdr *iph = skb->nh.iph;
699 struct ipq *qp;
700 struct net_device *dev;
701
702 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
703
704 /* Start by cleaning up the memory. */
705 if (atomic_read(&ip_frag_mem) > sysctl_ipfrag_high_thresh)
706 ip_evictor();
707
708 dev = skb->dev;
709
710 /* Lookup (or create) queue header */
711 if ((qp = ip_find(iph, user)) != NULL) {
712 struct sk_buff *ret = NULL;
713
714 spin_lock(&qp->lock);
715
716 ip_frag_queue(qp, skb);
717
718 if (qp->last_in == (FIRST_IN|LAST_IN) &&
719 qp->meat == qp->len)
720 ret = ip_frag_reasm(qp, dev);
721
722 spin_unlock(&qp->lock);
723 ipq_put(qp, NULL);
724 return ret;
725 }
726
727 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
728 kfree_skb(skb);
729 return NULL;
730}
731
732void ipfrag_init(void)
733{
734 ipfrag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
735 (jiffies ^ (jiffies >> 6)));
736
737 init_timer(&ipfrag_secret_timer);
738 ipfrag_secret_timer.function = ipfrag_secret_rebuild;
739 ipfrag_secret_timer.expires = jiffies + sysctl_ipfrag_secret_interval;
740 add_timer(&ipfrag_secret_timer);
741}
742
743EXPORT_SYMBOL(ip_defrag);