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 struct ip4_create_arg {
116 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
118 return jhash_3words((__force u32)id << 16 | prot,
119 (__force u32)saddr, (__force u32)daddr,
120 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
123 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
127 ipq = container_of(q, struct ipq, q);
128 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
131 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
134 struct ip4_create_arg *arg = a;
136 qp = container_of(q, struct ipq, q);
137 return (qp->id == arg->iph->id &&
138 qp->saddr == arg->iph->saddr &&
139 qp->daddr == arg->iph->daddr &&
140 qp->protocol == arg->iph->protocol &&
141 qp->user == arg->user);
144 /* Memory Tracking Functions. */
145 static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work)
148 *work -= skb->truesize;
149 atomic_sub(skb->truesize, &ip4_frags.mem);
153 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
155 struct ipq *qp = container_of(q, struct ipq, q);
156 struct ip4_create_arg *arg = a;
158 qp->protocol = arg->iph->protocol;
159 qp->id = arg->iph->id;
160 qp->saddr = arg->iph->saddr;
161 qp->daddr = arg->iph->daddr;
162 qp->user = arg->user;
163 qp->peer = sysctl_ipfrag_max_dist ?
164 inet_getpeer(arg->iph->saddr, 1) : NULL;
167 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
171 qp = container_of(q, struct ipq, q);
173 inet_putpeer(qp->peer);
177 /* Destruction primitives. */
179 static __inline__ void ipq_put(struct ipq *ipq)
181 inet_frag_put(&ipq->q, &ip4_frags);
184 /* Kill ipq entry. It is not destroyed immediately,
185 * because caller (and someone more) holds reference count.
187 static void ipq_kill(struct ipq *ipq)
189 inet_frag_kill(&ipq->q, &ip4_frags);
192 /* Memory limiting on fragments. Evictor trashes the oldest
193 * fragment queue until we are back under the threshold.
195 static void ip_evictor(void)
199 evicted = inet_frag_evictor(&ip4_frags);
201 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
205 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
207 static void ip_expire(unsigned long arg)
211 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
213 spin_lock(&qp->q.lock);
215 if (qp->q.last_in & COMPLETE)
220 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
221 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
223 if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
224 struct sk_buff *head = qp->q.fragments;
225 /* Send an ICMP "Fragment Reassembly Timeout" message. */
226 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
227 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
232 spin_unlock(&qp->q.lock);
236 /* Find the correct entry in the "incomplete datagrams" queue for
237 * this IP datagram, and create new one, if nothing is found.
239 static inline struct ipq *ip_find(struct iphdr *iph, u32 user)
241 struct inet_frag_queue *q;
242 struct ip4_create_arg arg;
247 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
249 q = inet_frag_find(&ip4_frags, &arg, hash);
253 return container_of(q, struct ipq, q);
256 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
260 /* Is the fragment too far ahead to be part of ipq? */
261 static inline int ip_frag_too_far(struct ipq *qp)
263 struct inet_peer *peer = qp->peer;
264 unsigned int max = sysctl_ipfrag_max_dist;
265 unsigned int start, end;
273 end = atomic_inc_return(&peer->rid);
276 rc = qp->q.fragments && (end - start) > max;
279 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
285 static int ip_frag_reinit(struct ipq *qp)
289 if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) {
290 atomic_inc(&qp->q.refcnt);
294 fp = qp->q.fragments;
296 struct sk_buff *xp = fp->next;
297 frag_kfree_skb(fp, NULL);
304 qp->q.fragments = NULL;
310 /* Add new segment to existing queue. */
311 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
313 struct sk_buff *prev, *next;
314 struct net_device *dev;
319 if (qp->q.last_in & COMPLETE)
322 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
323 unlikely(ip_frag_too_far(qp)) &&
324 unlikely(err = ip_frag_reinit(qp))) {
329 offset = ntohs(ip_hdr(skb)->frag_off);
330 flags = offset & ~IP_OFFSET;
332 offset <<= 3; /* offset is in 8-byte chunks */
333 ihl = ip_hdrlen(skb);
335 /* Determine the position of this fragment. */
336 end = offset + skb->len - ihl;
339 /* Is this the final fragment? */
340 if ((flags & IP_MF) == 0) {
341 /* If we already have some bits beyond end
342 * or have different end, the segment is corrrupted.
344 if (end < qp->q.len ||
345 ((qp->q.last_in & LAST_IN) && end != qp->q.len))
347 qp->q.last_in |= LAST_IN;
352 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
353 skb->ip_summed = CHECKSUM_NONE;
355 if (end > qp->q.len) {
356 /* Some bits beyond end -> corruption. */
357 if (qp->q.last_in & LAST_IN)
366 if (pskb_pull(skb, ihl) == NULL)
369 err = pskb_trim_rcsum(skb, end - offset);
373 /* Find out which fragments are in front and at the back of us
374 * in the chain of fragments so far. We must know where to put
375 * this fragment, right?
378 for (next = qp->q.fragments; next != NULL; next = next->next) {
379 if (FRAG_CB(next)->offset >= offset)
384 /* We found where to put this one. Check for overlap with
385 * preceding fragment, and, if needed, align things so that
386 * any overlaps are eliminated.
389 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
397 if (!pskb_pull(skb, i))
399 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
400 skb->ip_summed = CHECKSUM_NONE;
406 while (next && FRAG_CB(next)->offset < end) {
407 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
410 /* Eat head of the next overlapped fragment
411 * and leave the loop. The next ones cannot overlap.
413 if (!pskb_pull(next, i))
415 FRAG_CB(next)->offset += i;
417 if (next->ip_summed != CHECKSUM_UNNECESSARY)
418 next->ip_summed = CHECKSUM_NONE;
421 struct sk_buff *free_it = next;
423 /* Old fragment is completely overridden with
431 qp->q.fragments = next;
433 qp->q.meat -= free_it->len;
434 frag_kfree_skb(free_it, NULL);
438 FRAG_CB(skb)->offset = offset;
440 /* Insert this fragment in the chain of fragments. */
445 qp->q.fragments = skb;
449 qp->iif = dev->ifindex;
452 qp->q.stamp = skb->tstamp;
453 qp->q.meat += skb->len;
454 atomic_add(skb->truesize, &ip4_frags.mem);
456 qp->q.last_in |= FIRST_IN;
458 if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
459 return ip_frag_reasm(qp, prev, dev);
461 write_lock(&ip4_frags.lock);
462 list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
463 write_unlock(&ip4_frags.lock);
472 /* Build a new IP datagram from all its fragments. */
474 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
475 struct net_device *dev)
478 struct sk_buff *fp, *head = qp->q.fragments;
485 /* Make the one we just received the head. */
488 fp = skb_clone(head, GFP_ATOMIC);
492 fp->next = head->next;
495 skb_morph(head, qp->q.fragments);
496 head->next = qp->q.fragments->next;
498 kfree_skb(qp->q.fragments);
499 qp->q.fragments = head;
502 BUG_TRAP(head != NULL);
503 BUG_TRAP(FRAG_CB(head)->offset == 0);
505 /* Allocate a new buffer for the datagram. */
506 ihlen = ip_hdrlen(head);
507 len = ihlen + qp->q.len;
513 /* Head of list must not be cloned. */
514 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
517 /* If the first fragment is fragmented itself, we split
518 * it to two chunks: the first with data and paged part
519 * and the second, holding only fragments. */
520 if (skb_shinfo(head)->frag_list) {
521 struct sk_buff *clone;
524 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
526 clone->next = head->next;
528 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
529 skb_shinfo(head)->frag_list = NULL;
530 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
531 plen += skb_shinfo(head)->frags[i].size;
532 clone->len = clone->data_len = head->data_len - plen;
533 head->data_len -= clone->len;
534 head->len -= clone->len;
536 clone->ip_summed = head->ip_summed;
537 atomic_add(clone->truesize, &ip4_frags.mem);
540 skb_shinfo(head)->frag_list = head->next;
541 skb_push(head, head->data - skb_network_header(head));
542 atomic_sub(head->truesize, &ip4_frags.mem);
544 for (fp=head->next; fp; fp = fp->next) {
545 head->data_len += fp->len;
546 head->len += fp->len;
547 if (head->ip_summed != fp->ip_summed)
548 head->ip_summed = CHECKSUM_NONE;
549 else if (head->ip_summed == CHECKSUM_COMPLETE)
550 head->csum = csum_add(head->csum, fp->csum);
551 head->truesize += fp->truesize;
552 atomic_sub(fp->truesize, &ip4_frags.mem);
557 head->tstamp = qp->q.stamp;
561 iph->tot_len = htons(len);
562 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
563 qp->q.fragments = NULL;
567 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
574 "Oversized IP packet from %d.%d.%d.%d.\n",
577 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
581 /* Process an incoming IP datagram fragment. */
582 int ip_defrag(struct sk_buff *skb, u32 user)
586 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
588 /* Start by cleaning up the memory. */
589 if (atomic_read(&ip4_frags.mem) > ip4_frags_ctl.high_thresh)
592 /* Lookup (or create) queue header */
593 if ((qp = ip_find(ip_hdr(skb), user)) != NULL) {
596 spin_lock(&qp->q.lock);
598 ret = ip_frag_queue(qp, skb);
600 spin_unlock(&qp->q.lock);
605 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
610 void __init ipfrag_init(void)
612 ip4_frags.ctl = &ip4_frags_ctl;
613 ip4_frags.hashfn = ip4_hashfn;
614 ip4_frags.constructor = ip4_frag_init;
615 ip4_frags.destructor = ip4_frag_free;
616 ip4_frags.skb_free = NULL;
617 ip4_frags.qsize = sizeof(struct ipq);
618 ip4_frags.match = ip4_frag_match;
619 ip4_frags.frag_expire = ip_expire;
620 inet_frags_init(&ip4_frags);
623 EXPORT_SYMBOL(ip_defrag);