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.
6 * The IP fragmentation functionality.
8 * Version: $Id: ip_fragment.c,v 1.59 2002/01/12 07:54:56 davem Exp $
10 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox <Alan.Cox@linux.org>
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.
25 #include <linux/compiler.h>
26 #include <linux/module.h>
27 #include <linux/types.h>
29 #include <linux/jiffies.h>
30 #include <linux/skbuff.h>
31 #include <linux/list.h>
33 #include <linux/icmp.h>
34 #include <linux/netdevice.h>
35 #include <linux/jhash.h>
36 #include <linux/random.h>
40 #include <net/checksum.h>
41 #include <net/inetpeer.h>
42 #include <net/inet_frag.h>
43 #include <linux/tcp.h>
44 #include <linux/udp.h>
45 #include <linux/inet.h>
46 #include <linux/netfilter_ipv4.h>
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
53 int sysctl_ipfrag_max_dist __read_mostly = 64;
57 struct inet_skb_parm h;
61 #define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb))
63 /* Describe an entry in the "incomplete datagrams" queue. */
65 struct inet_frag_queue q;
74 struct inet_peer *peer;
77 struct inet_frags_ctl ip4_frags_ctl __read_mostly = {
79 * Fragment cache limits. We will commit 256K at one time. Should we
80 * cross that limit we will prune down to 192K. This should cope with
81 * even the most extreme cases without allowing an attacker to
82 * measurably harm machine performance.
84 .high_thresh = 256 * 1024,
85 .low_thresh = 192 * 1024,
88 * Important NOTE! Fragment queue must be destroyed before MSL expires.
89 * RFC791 is wrong proposing to prolongate timer each fragment arrival
92 .timeout = IP_FRAG_TIME,
93 .secret_interval = 10 * 60 * HZ,
96 static struct inet_frags ip4_frags;
98 int ip_frag_nqueues(void)
100 return ip4_frags.nqueues;
103 int ip_frag_mem(void)
105 return atomic_read(&ip4_frags.mem);
108 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
109 struct net_device *dev);
111 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
113 return jhash_3words((__force u32)id << 16 | prot,
114 (__force u32)saddr, (__force u32)daddr,
115 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
118 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
122 ipq = container_of(q, struct ipq, q);
123 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
126 static int ip4_frag_equal(struct inet_frag_queue *q1,
127 struct inet_frag_queue *q2)
129 struct ipq *qp1, *qp2;
131 qp1 = container_of(q1, struct ipq, q);
132 qp2 = container_of(q2, struct ipq, q);
133 return (qp1->id == qp2->id &&
134 qp1->saddr == qp2->saddr &&
135 qp1->daddr == qp2->daddr &&
136 qp1->protocol == qp2->protocol &&
137 qp1->user == qp2->user);
140 /* Memory Tracking Functions. */
141 static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work)
144 *work -= skb->truesize;
145 atomic_sub(skb->truesize, &ip4_frags.mem);
149 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
153 qp = container_of(q, struct ipq, q);
155 inet_putpeer(qp->peer);
159 static __inline__ struct ipq *frag_alloc_queue(void)
161 struct inet_frag_queue *q;
163 q = inet_frag_alloc(&ip4_frags);
164 return q ? container_of(q, struct ipq, q) : NULL;
168 /* Destruction primitives. */
170 static __inline__ void ipq_put(struct ipq *ipq)
172 inet_frag_put(&ipq->q, &ip4_frags);
175 /* Kill ipq entry. It is not destroyed immediately,
176 * because caller (and someone more) holds reference count.
178 static void ipq_kill(struct ipq *ipq)
180 inet_frag_kill(&ipq->q, &ip4_frags);
183 /* Memory limiting on fragments. Evictor trashes the oldest
184 * fragment queue until we are back under the threshold.
186 static void ip_evictor(void)
190 evicted = inet_frag_evictor(&ip4_frags);
192 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
196 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
198 static void ip_expire(unsigned long arg)
202 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
204 spin_lock(&qp->q.lock);
206 if (qp->q.last_in & COMPLETE)
211 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
212 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
214 if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
215 struct sk_buff *head = qp->q.fragments;
216 /* Send an ICMP "Fragment Reassembly Timeout" message. */
217 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
218 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
223 spin_unlock(&qp->q.lock);
227 /* Creation primitives. */
229 static struct ipq *ip_frag_intern(struct ipq *qp_in, unsigned int hash)
231 struct inet_frag_queue *q;
233 q = inet_frag_intern(&qp_in->q, &ip4_frags, hash);
234 return container_of(q, struct ipq, q);
237 /* Add an entry to the 'ipq' queue for a newly received IP datagram. */
238 static struct ipq *ip_frag_create(struct iphdr *iph, u32 user, unsigned int h)
242 if ((qp = frag_alloc_queue()) == NULL)
245 qp->protocol = iph->protocol;
247 qp->saddr = iph->saddr;
248 qp->daddr = iph->daddr;
250 qp->peer = sysctl_ipfrag_max_dist ? inet_getpeer(iph->saddr, 1) : NULL;
252 return ip_frag_intern(qp, h);
255 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
259 /* Find the correct entry in the "incomplete datagrams" queue for
260 * this IP datagram, and create new one, if nothing is found.
262 static inline struct ipq *ip_find(struct iphdr *iph, u32 user)
265 __be32 saddr = iph->saddr;
266 __be32 daddr = iph->daddr;
267 __u8 protocol = iph->protocol;
270 struct hlist_node *n;
272 read_lock(&ip4_frags.lock);
273 hash = ipqhashfn(id, saddr, daddr, protocol);
274 hlist_for_each_entry(qp, n, &ip4_frags.hash[hash], q.list) {
276 qp->saddr == saddr &&
277 qp->daddr == daddr &&
278 qp->protocol == protocol &&
280 atomic_inc(&qp->q.refcnt);
281 read_unlock(&ip4_frags.lock);
285 read_unlock(&ip4_frags.lock);
287 return ip_frag_create(iph, user, hash);
290 /* Is the fragment too far ahead to be part of ipq? */
291 static inline int ip_frag_too_far(struct ipq *qp)
293 struct inet_peer *peer = qp->peer;
294 unsigned int max = sysctl_ipfrag_max_dist;
295 unsigned int start, end;
303 end = atomic_inc_return(&peer->rid);
306 rc = qp->q.fragments && (end - start) > max;
309 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
315 static int ip_frag_reinit(struct ipq *qp)
319 if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) {
320 atomic_inc(&qp->q.refcnt);
324 fp = qp->q.fragments;
326 struct sk_buff *xp = fp->next;
327 frag_kfree_skb(fp, NULL);
334 qp->q.fragments = NULL;
340 /* Add new segment to existing queue. */
341 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
343 struct sk_buff *prev, *next;
344 struct net_device *dev;
349 if (qp->q.last_in & COMPLETE)
352 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
353 unlikely(ip_frag_too_far(qp)) &&
354 unlikely(err = ip_frag_reinit(qp))) {
359 offset = ntohs(ip_hdr(skb)->frag_off);
360 flags = offset & ~IP_OFFSET;
362 offset <<= 3; /* offset is in 8-byte chunks */
363 ihl = ip_hdrlen(skb);
365 /* Determine the position of this fragment. */
366 end = offset + skb->len - ihl;
369 /* Is this the final fragment? */
370 if ((flags & IP_MF) == 0) {
371 /* If we already have some bits beyond end
372 * or have different end, the segment is corrrupted.
374 if (end < qp->q.len ||
375 ((qp->q.last_in & LAST_IN) && end != qp->q.len))
377 qp->q.last_in |= LAST_IN;
382 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
383 skb->ip_summed = CHECKSUM_NONE;
385 if (end > qp->q.len) {
386 /* Some bits beyond end -> corruption. */
387 if (qp->q.last_in & LAST_IN)
396 if (pskb_pull(skb, ihl) == NULL)
399 err = pskb_trim_rcsum(skb, end - offset);
403 /* Find out which fragments are in front and at the back of us
404 * in the chain of fragments so far. We must know where to put
405 * this fragment, right?
408 for (next = qp->q.fragments; next != NULL; next = next->next) {
409 if (FRAG_CB(next)->offset >= offset)
414 /* We found where to put this one. Check for overlap with
415 * preceding fragment, and, if needed, align things so that
416 * any overlaps are eliminated.
419 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
427 if (!pskb_pull(skb, i))
429 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
430 skb->ip_summed = CHECKSUM_NONE;
436 while (next && FRAG_CB(next)->offset < end) {
437 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
440 /* Eat head of the next overlapped fragment
441 * and leave the loop. The next ones cannot overlap.
443 if (!pskb_pull(next, i))
445 FRAG_CB(next)->offset += i;
447 if (next->ip_summed != CHECKSUM_UNNECESSARY)
448 next->ip_summed = CHECKSUM_NONE;
451 struct sk_buff *free_it = next;
453 /* Old fragment is completely overridden with
461 qp->q.fragments = next;
463 qp->q.meat -= free_it->len;
464 frag_kfree_skb(free_it, NULL);
468 FRAG_CB(skb)->offset = offset;
470 /* Insert this fragment in the chain of fragments. */
475 qp->q.fragments = skb;
479 qp->iif = dev->ifindex;
482 qp->q.stamp = skb->tstamp;
483 qp->q.meat += skb->len;
484 atomic_add(skb->truesize, &ip4_frags.mem);
486 qp->q.last_in |= FIRST_IN;
488 if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
489 return ip_frag_reasm(qp, prev, dev);
491 write_lock(&ip4_frags.lock);
492 list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
493 write_unlock(&ip4_frags.lock);
502 /* Build a new IP datagram from all its fragments. */
504 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
505 struct net_device *dev)
508 struct sk_buff *fp, *head = qp->q.fragments;
515 /* Make the one we just received the head. */
518 fp = skb_clone(head, GFP_ATOMIC);
523 fp->next = head->next;
526 skb_morph(head, qp->q.fragments);
527 head->next = qp->q.fragments->next;
529 kfree_skb(qp->q.fragments);
530 qp->q.fragments = head;
533 BUG_TRAP(head != NULL);
534 BUG_TRAP(FRAG_CB(head)->offset == 0);
536 /* Allocate a new buffer for the datagram. */
537 ihlen = ip_hdrlen(head);
538 len = ihlen + qp->q.len;
544 /* Head of list must not be cloned. */
546 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
549 /* If the first fragment is fragmented itself, we split
550 * it to two chunks: the first with data and paged part
551 * and the second, holding only fragments. */
552 if (skb_shinfo(head)->frag_list) {
553 struct sk_buff *clone;
556 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
558 clone->next = head->next;
560 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
561 skb_shinfo(head)->frag_list = NULL;
562 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
563 plen += skb_shinfo(head)->frags[i].size;
564 clone->len = clone->data_len = head->data_len - plen;
565 head->data_len -= clone->len;
566 head->len -= clone->len;
568 clone->ip_summed = head->ip_summed;
569 atomic_add(clone->truesize, &ip4_frags.mem);
572 skb_shinfo(head)->frag_list = head->next;
573 skb_push(head, head->data - skb_network_header(head));
574 atomic_sub(head->truesize, &ip4_frags.mem);
576 for (fp=head->next; fp; fp = fp->next) {
577 head->data_len += fp->len;
578 head->len += fp->len;
579 if (head->ip_summed != fp->ip_summed)
580 head->ip_summed = CHECKSUM_NONE;
581 else if (head->ip_summed == CHECKSUM_COMPLETE)
582 head->csum = csum_add(head->csum, fp->csum);
583 head->truesize += fp->truesize;
584 atomic_sub(fp->truesize, &ip4_frags.mem);
589 head->tstamp = qp->q.stamp;
593 iph->tot_len = htons(len);
594 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
595 qp->q.fragments = NULL;
599 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
605 "Oversized IP packet from %d.%d.%d.%d.\n",
608 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
612 /* Process an incoming IP datagram fragment. */
613 int ip_defrag(struct sk_buff *skb, u32 user)
617 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
619 /* Start by cleaning up the memory. */
620 if (atomic_read(&ip4_frags.mem) > ip4_frags_ctl.high_thresh)
623 /* Lookup (or create) queue header */
624 if ((qp = ip_find(ip_hdr(skb), user)) != NULL) {
627 spin_lock(&qp->q.lock);
629 ret = ip_frag_queue(qp, skb);
631 spin_unlock(&qp->q.lock);
636 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
641 void __init ipfrag_init(void)
643 ip4_frags.ctl = &ip4_frags_ctl;
644 ip4_frags.hashfn = ip4_hashfn;
645 ip4_frags.destructor = ip4_frag_free;
646 ip4_frags.skb_free = NULL;
647 ip4_frags.qsize = sizeof(struct ipq);
648 ip4_frags.equal = ip4_frag_equal;
649 ip4_frags.frag_expire = ip_expire;
650 inet_frags_init(&ip4_frags);
653 EXPORT_SYMBOL(ip_defrag);