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 static 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 static struct inet_frags_ctl ip4_frags_ctl __read_mostly = {
78 .secret_interval = 10 * 60 * HZ,
81 static struct inet_frags ip4_frags;
83 int ip_frag_nqueues(struct net *net)
85 return net->ipv4.frags.nqueues;
88 int ip_frag_mem(struct net *net)
90 return atomic_read(&net->ipv4.frags.mem);
93 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
94 struct net_device *dev);
96 struct ip4_create_arg {
101 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
103 return jhash_3words((__force u32)id << 16 | prot,
104 (__force u32)saddr, (__force u32)daddr,
105 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
108 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
112 ipq = container_of(q, struct ipq, q);
113 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
116 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
119 struct ip4_create_arg *arg = a;
121 qp = container_of(q, struct ipq, q);
122 return (qp->id == arg->iph->id &&
123 qp->saddr == arg->iph->saddr &&
124 qp->daddr == arg->iph->daddr &&
125 qp->protocol == arg->iph->protocol &&
126 qp->user == arg->user);
129 /* Memory Tracking Functions. */
130 static __inline__ void frag_kfree_skb(struct netns_frags *nf,
131 struct sk_buff *skb, int *work)
134 *work -= skb->truesize;
135 atomic_sub(skb->truesize, &nf->mem);
139 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
141 struct ipq *qp = container_of(q, struct ipq, q);
142 struct ip4_create_arg *arg = a;
144 qp->protocol = arg->iph->protocol;
145 qp->id = arg->iph->id;
146 qp->saddr = arg->iph->saddr;
147 qp->daddr = arg->iph->daddr;
148 qp->user = arg->user;
149 qp->peer = sysctl_ipfrag_max_dist ?
150 inet_getpeer(arg->iph->saddr, 1) : NULL;
153 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
157 qp = container_of(q, struct ipq, q);
159 inet_putpeer(qp->peer);
163 /* Destruction primitives. */
165 static __inline__ void ipq_put(struct ipq *ipq)
167 inet_frag_put(&ipq->q, &ip4_frags);
170 /* Kill ipq entry. It is not destroyed immediately,
171 * because caller (and someone more) holds reference count.
173 static void ipq_kill(struct ipq *ipq)
175 inet_frag_kill(&ipq->q, &ip4_frags);
178 /* Memory limiting on fragments. Evictor trashes the oldest
179 * fragment queue until we are back under the threshold.
181 static void ip_evictor(struct net *net)
185 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
187 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
191 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
193 static void ip_expire(unsigned long arg)
197 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
199 spin_lock(&qp->q.lock);
201 if (qp->q.last_in & COMPLETE)
206 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
207 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
209 if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
210 struct sk_buff *head = qp->q.fragments;
211 /* Send an ICMP "Fragment Reassembly Timeout" message. */
212 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
213 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
218 spin_unlock(&qp->q.lock);
222 /* Find the correct entry in the "incomplete datagrams" queue for
223 * this IP datagram, and create new one, if nothing is found.
225 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
227 struct inet_frag_queue *q;
228 struct ip4_create_arg arg;
233 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
235 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
239 return container_of(q, struct ipq, q);
242 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
246 /* Is the fragment too far ahead to be part of ipq? */
247 static inline int ip_frag_too_far(struct ipq *qp)
249 struct inet_peer *peer = qp->peer;
250 unsigned int max = sysctl_ipfrag_max_dist;
251 unsigned int start, end;
259 end = atomic_inc_return(&peer->rid);
262 rc = qp->q.fragments && (end - start) > max;
265 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
271 static int ip_frag_reinit(struct ipq *qp)
275 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
276 atomic_inc(&qp->q.refcnt);
280 fp = qp->q.fragments;
282 struct sk_buff *xp = fp->next;
283 frag_kfree_skb(qp->q.net, fp, NULL);
290 qp->q.fragments = NULL;
296 /* Add new segment to existing queue. */
297 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
299 struct sk_buff *prev, *next;
300 struct net_device *dev;
305 if (qp->q.last_in & COMPLETE)
308 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
309 unlikely(ip_frag_too_far(qp)) &&
310 unlikely(err = ip_frag_reinit(qp))) {
315 offset = ntohs(ip_hdr(skb)->frag_off);
316 flags = offset & ~IP_OFFSET;
318 offset <<= 3; /* offset is in 8-byte chunks */
319 ihl = ip_hdrlen(skb);
321 /* Determine the position of this fragment. */
322 end = offset + skb->len - ihl;
325 /* Is this the final fragment? */
326 if ((flags & IP_MF) == 0) {
327 /* If we already have some bits beyond end
328 * or have different end, the segment is corrrupted.
330 if (end < qp->q.len ||
331 ((qp->q.last_in & LAST_IN) && end != qp->q.len))
333 qp->q.last_in |= LAST_IN;
338 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
339 skb->ip_summed = CHECKSUM_NONE;
341 if (end > qp->q.len) {
342 /* Some bits beyond end -> corruption. */
343 if (qp->q.last_in & LAST_IN)
352 if (pskb_pull(skb, ihl) == NULL)
355 err = pskb_trim_rcsum(skb, end - offset);
359 /* Find out which fragments are in front and at the back of us
360 * in the chain of fragments so far. We must know where to put
361 * this fragment, right?
364 for (next = qp->q.fragments; next != NULL; next = next->next) {
365 if (FRAG_CB(next)->offset >= offset)
370 /* We found where to put this one. Check for overlap with
371 * preceding fragment, and, if needed, align things so that
372 * any overlaps are eliminated.
375 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
383 if (!pskb_pull(skb, i))
385 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
386 skb->ip_summed = CHECKSUM_NONE;
392 while (next && FRAG_CB(next)->offset < end) {
393 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
396 /* Eat head of the next overlapped fragment
397 * and leave the loop. The next ones cannot overlap.
399 if (!pskb_pull(next, i))
401 FRAG_CB(next)->offset += i;
403 if (next->ip_summed != CHECKSUM_UNNECESSARY)
404 next->ip_summed = CHECKSUM_NONE;
407 struct sk_buff *free_it = next;
409 /* Old fragment is completely overridden with
417 qp->q.fragments = next;
419 qp->q.meat -= free_it->len;
420 frag_kfree_skb(qp->q.net, free_it, NULL);
424 FRAG_CB(skb)->offset = offset;
426 /* Insert this fragment in the chain of fragments. */
431 qp->q.fragments = skb;
435 qp->iif = dev->ifindex;
438 qp->q.stamp = skb->tstamp;
439 qp->q.meat += skb->len;
440 atomic_add(skb->truesize, &qp->q.net->mem);
442 qp->q.last_in |= FIRST_IN;
444 if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
445 return ip_frag_reasm(qp, prev, dev);
447 write_lock(&ip4_frags.lock);
448 list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
449 write_unlock(&ip4_frags.lock);
458 /* Build a new IP datagram from all its fragments. */
460 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
461 struct net_device *dev)
464 struct sk_buff *fp, *head = qp->q.fragments;
471 /* Make the one we just received the head. */
474 fp = skb_clone(head, GFP_ATOMIC);
478 fp->next = head->next;
481 skb_morph(head, qp->q.fragments);
482 head->next = qp->q.fragments->next;
484 kfree_skb(qp->q.fragments);
485 qp->q.fragments = head;
488 BUG_TRAP(head != NULL);
489 BUG_TRAP(FRAG_CB(head)->offset == 0);
491 /* Allocate a new buffer for the datagram. */
492 ihlen = ip_hdrlen(head);
493 len = ihlen + qp->q.len;
499 /* Head of list must not be cloned. */
500 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
503 /* If the first fragment is fragmented itself, we split
504 * it to two chunks: the first with data and paged part
505 * and the second, holding only fragments. */
506 if (skb_shinfo(head)->frag_list) {
507 struct sk_buff *clone;
510 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
512 clone->next = head->next;
514 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
515 skb_shinfo(head)->frag_list = NULL;
516 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
517 plen += skb_shinfo(head)->frags[i].size;
518 clone->len = clone->data_len = head->data_len - plen;
519 head->data_len -= clone->len;
520 head->len -= clone->len;
522 clone->ip_summed = head->ip_summed;
523 atomic_add(clone->truesize, &qp->q.net->mem);
526 skb_shinfo(head)->frag_list = head->next;
527 skb_push(head, head->data - skb_network_header(head));
528 atomic_sub(head->truesize, &qp->q.net->mem);
530 for (fp=head->next; fp; fp = fp->next) {
531 head->data_len += fp->len;
532 head->len += fp->len;
533 if (head->ip_summed != fp->ip_summed)
534 head->ip_summed = CHECKSUM_NONE;
535 else if (head->ip_summed == CHECKSUM_COMPLETE)
536 head->csum = csum_add(head->csum, fp->csum);
537 head->truesize += fp->truesize;
538 atomic_sub(fp->truesize, &qp->q.net->mem);
543 head->tstamp = qp->q.stamp;
547 iph->tot_len = htons(len);
548 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
549 qp->q.fragments = NULL;
553 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
560 "Oversized IP packet from %d.%d.%d.%d.\n",
563 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
567 /* Process an incoming IP datagram fragment. */
568 int ip_defrag(struct sk_buff *skb, u32 user)
573 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
575 net = skb->dev->nd_net;
576 /* Start by cleaning up the memory. */
577 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
580 /* Lookup (or create) queue header */
581 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
584 spin_lock(&qp->q.lock);
586 ret = ip_frag_queue(qp, skb);
588 spin_unlock(&qp->q.lock);
593 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
601 static struct ctl_table ip4_frags_ctl_table[] = {
603 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
604 .procname = "ipfrag_high_thresh",
605 .data = &init_net.ipv4.frags.high_thresh,
606 .maxlen = sizeof(int),
608 .proc_handler = &proc_dointvec
611 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
612 .procname = "ipfrag_low_thresh",
613 .data = &init_net.ipv4.frags.low_thresh,
614 .maxlen = sizeof(int),
616 .proc_handler = &proc_dointvec
619 .ctl_name = NET_IPV4_IPFRAG_TIME,
620 .procname = "ipfrag_time",
621 .data = &init_net.ipv4.frags.timeout,
622 .maxlen = sizeof(int),
624 .proc_handler = &proc_dointvec_jiffies,
625 .strategy = &sysctl_jiffies
628 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
629 .procname = "ipfrag_secret_interval",
630 .data = &ip4_frags_ctl.secret_interval,
631 .maxlen = sizeof(int),
633 .proc_handler = &proc_dointvec_jiffies,
634 .strategy = &sysctl_jiffies
637 .procname = "ipfrag_max_dist",
638 .data = &sysctl_ipfrag_max_dist,
639 .maxlen = sizeof(int),
641 .proc_handler = &proc_dointvec_minmax,
647 static int ip4_frags_ctl_register(struct net *net)
649 struct ctl_table *table;
650 struct ctl_table_header *hdr;
652 table = ip4_frags_ctl_table;
653 if (net != &init_net) {
654 table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL);
658 table[0].data = &net->ipv4.frags.high_thresh;
659 table[1].data = &net->ipv4.frags.low_thresh;
660 table[2].data = &net->ipv4.frags.timeout;
661 table[3].mode &= ~0222;
662 table[4].mode &= ~0222;
665 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
669 net->ipv4.frags_hdr = hdr;
673 if (net != &init_net)
679 static void ip4_frags_ctl_unregister(struct net *net)
681 struct ctl_table *table;
683 table = net->ipv4.frags_hdr->ctl_table_arg;
684 unregister_net_sysctl_table(net->ipv4.frags_hdr);
688 static inline int ip4_frags_ctl_register(struct net *net)
693 static inline void ip4_frags_ctl_unregister(struct net *net)
698 static int ipv4_frags_init_net(struct net *net)
701 * Fragment cache limits. We will commit 256K at one time. Should we
702 * cross that limit we will prune down to 192K. This should cope with
703 * even the most extreme cases without allowing an attacker to
704 * measurably harm machine performance.
706 net->ipv4.frags.high_thresh = 256 * 1024;
707 net->ipv4.frags.low_thresh = 192 * 1024;
709 * Important NOTE! Fragment queue must be destroyed before MSL expires.
710 * RFC791 is wrong proposing to prolongate timer each fragment arrival
713 net->ipv4.frags.timeout = IP_FRAG_TIME;
715 inet_frags_init_net(&net->ipv4.frags);
717 return ip4_frags_ctl_register(net);
720 void __init ipfrag_init(void)
722 ipv4_frags_init_net(&init_net);
723 ip4_frags.ctl = &ip4_frags_ctl;
724 ip4_frags.hashfn = ip4_hashfn;
725 ip4_frags.constructor = ip4_frag_init;
726 ip4_frags.destructor = ip4_frag_free;
727 ip4_frags.skb_free = NULL;
728 ip4_frags.qsize = sizeof(struct ipq);
729 ip4_frags.match = ip4_frag_match;
730 ip4_frags.frag_expire = ip_expire;
731 inet_frags_init(&ip4_frags);
734 EXPORT_SYMBOL(ip_defrag);