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 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
9 * Alan Cox <alan@lxorguk.ukuu.org.uk>
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.
23 #include <linux/compiler.h>
24 #include <linux/module.h>
25 #include <linux/types.h>
27 #include <linux/jiffies.h>
28 #include <linux/skbuff.h>
29 #include <linux/list.h>
31 #include <linux/icmp.h>
32 #include <linux/netdevice.h>
33 #include <linux/jhash.h>
34 #include <linux/random.h>
35 #include <net/route.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 ip4_frags;
79 int ip_frag_nqueues(struct net *net)
81 return net->ipv4.frags.nqueues;
84 int ip_frag_mem(struct net *net)
86 return atomic_read(&net->ipv4.frags.mem);
89 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
90 struct net_device *dev);
92 struct ip4_create_arg {
97 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
99 return jhash_3words((__force u32)id << 16 | prot,
100 (__force u32)saddr, (__force u32)daddr,
101 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
104 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
108 ipq = container_of(q, struct ipq, q);
109 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
112 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
115 struct ip4_create_arg *arg = a;
117 qp = container_of(q, struct ipq, q);
118 return (qp->id == arg->iph->id &&
119 qp->saddr == arg->iph->saddr &&
120 qp->daddr == arg->iph->daddr &&
121 qp->protocol == arg->iph->protocol &&
122 qp->user == arg->user);
125 /* Memory Tracking Functions. */
126 static __inline__ void frag_kfree_skb(struct netns_frags *nf,
127 struct sk_buff *skb, int *work)
130 *work -= skb->truesize;
131 atomic_sub(skb->truesize, &nf->mem);
135 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
137 struct ipq *qp = container_of(q, struct ipq, q);
138 struct ip4_create_arg *arg = a;
140 qp->protocol = arg->iph->protocol;
141 qp->id = arg->iph->id;
142 qp->saddr = arg->iph->saddr;
143 qp->daddr = arg->iph->daddr;
144 qp->user = arg->user;
145 qp->peer = sysctl_ipfrag_max_dist ?
146 inet_getpeer(arg->iph->saddr, 1) : NULL;
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 /* Destruction primitives. */
161 static __inline__ void ipq_put(struct ipq *ipq)
163 inet_frag_put(&ipq->q, &ip4_frags);
166 /* Kill ipq entry. It is not destroyed immediately,
167 * because caller (and someone more) holds reference count.
169 static void ipq_kill(struct ipq *ipq)
171 inet_frag_kill(&ipq->q, &ip4_frags);
174 /* Memory limiting on fragments. Evictor trashes the oldest
175 * fragment queue until we are back under the threshold.
177 static void ip_evictor(struct net *net)
181 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
183 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
187 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
189 static void ip_expire(unsigned long arg)
194 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
195 net = container_of(qp->q.net, struct net, ipv4.frags);
197 spin_lock(&qp->q.lock);
199 if (qp->q.last_in & INET_FRAG_COMPLETE)
204 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
205 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
207 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
208 struct sk_buff *head = qp->q.fragments;
211 head->dev = dev_get_by_index_rcu(net, qp->iif);
216 * Only search router table for the head fragment,
217 * when defraging timeout at PRE_ROUTING HOOK.
219 if (qp->user == IP_DEFRAG_CONNTRACK_IN && !skb_dst(head)) {
220 const struct iphdr *iph = ip_hdr(head);
221 int err = ip_route_input(head, iph->daddr, iph->saddr,
222 iph->tos, head->dev);
227 * Only an end host needs to send an ICMP
228 * "Fragment Reassembly Timeout" message, per RFC792.
230 if (skb_rtable(head)->rt_type != RTN_LOCAL)
235 /* Send an ICMP "Fragment Reassembly Timeout" message. */
236 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
241 spin_unlock(&qp->q.lock);
245 /* Find the correct entry in the "incomplete datagrams" queue for
246 * this IP datagram, and create new one, if nothing is found.
248 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
250 struct inet_frag_queue *q;
251 struct ip4_create_arg arg;
257 read_lock(&ip4_frags.lock);
258 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
260 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
264 return container_of(q, struct ipq, q);
267 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
271 /* Is the fragment too far ahead to be part of ipq? */
272 static inline int ip_frag_too_far(struct ipq *qp)
274 struct inet_peer *peer = qp->peer;
275 unsigned int max = sysctl_ipfrag_max_dist;
276 unsigned int start, end;
284 end = atomic_inc_return(&peer->rid);
287 rc = qp->q.fragments && (end - start) > max;
292 net = container_of(qp->q.net, struct net, ipv4.frags);
293 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
299 static int ip_frag_reinit(struct ipq *qp)
303 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
304 atomic_inc(&qp->q.refcnt);
308 fp = qp->q.fragments;
310 struct sk_buff *xp = fp->next;
311 frag_kfree_skb(qp->q.net, fp, NULL);
318 qp->q.fragments = NULL;
324 /* Add new segment to existing queue. */
325 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
327 struct sk_buff *prev, *next;
328 struct net_device *dev;
333 if (qp->q.last_in & INET_FRAG_COMPLETE)
336 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
337 unlikely(ip_frag_too_far(qp)) &&
338 unlikely(err = ip_frag_reinit(qp))) {
343 offset = ntohs(ip_hdr(skb)->frag_off);
344 flags = offset & ~IP_OFFSET;
346 offset <<= 3; /* offset is in 8-byte chunks */
347 ihl = ip_hdrlen(skb);
349 /* Determine the position of this fragment. */
350 end = offset + skb->len - ihl;
353 /* Is this the final fragment? */
354 if ((flags & IP_MF) == 0) {
355 /* If we already have some bits beyond end
356 * or have different end, the segment is corrrupted.
358 if (end < qp->q.len ||
359 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
361 qp->q.last_in |= INET_FRAG_LAST_IN;
366 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
367 skb->ip_summed = CHECKSUM_NONE;
369 if (end > qp->q.len) {
370 /* Some bits beyond end -> corruption. */
371 if (qp->q.last_in & INET_FRAG_LAST_IN)
380 if (pskb_pull(skb, ihl) == NULL)
383 err = pskb_trim_rcsum(skb, end - offset);
387 /* Find out which fragments are in front and at the back of us
388 * in the chain of fragments so far. We must know where to put
389 * this fragment, right?
392 for (next = qp->q.fragments; next != NULL; next = next->next) {
393 if (FRAG_CB(next)->offset >= offset)
398 /* We found where to put this one. Check for overlap with
399 * preceding fragment, and, if needed, align things so that
400 * any overlaps are eliminated.
403 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
411 if (!pskb_pull(skb, i))
413 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
414 skb->ip_summed = CHECKSUM_NONE;
420 while (next && FRAG_CB(next)->offset < end) {
421 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
424 /* Eat head of the next overlapped fragment
425 * and leave the loop. The next ones cannot overlap.
427 if (!pskb_pull(next, i))
429 FRAG_CB(next)->offset += i;
431 if (next->ip_summed != CHECKSUM_UNNECESSARY)
432 next->ip_summed = CHECKSUM_NONE;
435 struct sk_buff *free_it = next;
437 /* Old fragment is completely overridden with
445 qp->q.fragments = next;
447 qp->q.meat -= free_it->len;
448 frag_kfree_skb(qp->q.net, free_it, NULL);
452 FRAG_CB(skb)->offset = offset;
454 /* Insert this fragment in the chain of fragments. */
459 qp->q.fragments = skb;
463 qp->iif = dev->ifindex;
466 qp->q.stamp = skb->tstamp;
467 qp->q.meat += skb->len;
468 atomic_add(skb->truesize, &qp->q.net->mem);
470 qp->q.last_in |= INET_FRAG_FIRST_IN;
472 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
473 qp->q.meat == qp->q.len)
474 return ip_frag_reasm(qp, prev, dev);
476 write_lock(&ip4_frags.lock);
477 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
478 write_unlock(&ip4_frags.lock);
487 /* Build a new IP datagram from all its fragments. */
489 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
490 struct net_device *dev)
492 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
494 struct sk_buff *fp, *head = qp->q.fragments;
501 /* Make the one we just received the head. */
504 fp = skb_clone(head, GFP_ATOMIC);
508 fp->next = head->next;
511 skb_morph(head, qp->q.fragments);
512 head->next = qp->q.fragments->next;
514 kfree_skb(qp->q.fragments);
515 qp->q.fragments = head;
518 WARN_ON(head == NULL);
519 WARN_ON(FRAG_CB(head)->offset != 0);
521 /* Allocate a new buffer for the datagram. */
522 ihlen = ip_hdrlen(head);
523 len = ihlen + qp->q.len;
529 /* Head of list must not be cloned. */
530 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
533 /* If the first fragment is fragmented itself, we split
534 * it to two chunks: the first with data and paged part
535 * and the second, holding only fragments. */
536 if (skb_has_frags(head)) {
537 struct sk_buff *clone;
540 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
542 clone->next = head->next;
544 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
545 skb_frag_list_init(head);
546 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
547 plen += skb_shinfo(head)->frags[i].size;
548 clone->len = clone->data_len = head->data_len - plen;
549 head->data_len -= clone->len;
550 head->len -= clone->len;
552 clone->ip_summed = head->ip_summed;
553 atomic_add(clone->truesize, &qp->q.net->mem);
556 skb_shinfo(head)->frag_list = head->next;
557 skb_push(head, head->data - skb_network_header(head));
558 atomic_sub(head->truesize, &qp->q.net->mem);
560 for (fp=head->next; fp; fp = fp->next) {
561 head->data_len += fp->len;
562 head->len += fp->len;
563 if (head->ip_summed != fp->ip_summed)
564 head->ip_summed = CHECKSUM_NONE;
565 else if (head->ip_summed == CHECKSUM_COMPLETE)
566 head->csum = csum_add(head->csum, fp->csum);
567 head->truesize += fp->truesize;
568 atomic_sub(fp->truesize, &qp->q.net->mem);
573 head->tstamp = qp->q.stamp;
577 iph->tot_len = htons(len);
578 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
579 qp->q.fragments = NULL;
583 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
589 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
592 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
596 /* Process an incoming IP datagram fragment. */
597 int ip_defrag(struct sk_buff *skb, u32 user)
602 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
603 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
605 /* Start by cleaning up the memory. */
606 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
609 /* Lookup (or create) queue header */
610 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
613 spin_lock(&qp->q.lock);
615 ret = ip_frag_queue(qp, skb);
617 spin_unlock(&qp->q.lock);
622 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
630 static struct ctl_table ip4_frags_ns_ctl_table[] = {
632 .procname = "ipfrag_high_thresh",
633 .data = &init_net.ipv4.frags.high_thresh,
634 .maxlen = sizeof(int),
636 .proc_handler = proc_dointvec
639 .procname = "ipfrag_low_thresh",
640 .data = &init_net.ipv4.frags.low_thresh,
641 .maxlen = sizeof(int),
643 .proc_handler = proc_dointvec
646 .procname = "ipfrag_time",
647 .data = &init_net.ipv4.frags.timeout,
648 .maxlen = sizeof(int),
650 .proc_handler = proc_dointvec_jiffies,
655 static struct ctl_table ip4_frags_ctl_table[] = {
657 .procname = "ipfrag_secret_interval",
658 .data = &ip4_frags.secret_interval,
659 .maxlen = sizeof(int),
661 .proc_handler = proc_dointvec_jiffies,
664 .procname = "ipfrag_max_dist",
665 .data = &sysctl_ipfrag_max_dist,
666 .maxlen = sizeof(int),
668 .proc_handler = proc_dointvec_minmax,
674 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
676 struct ctl_table *table;
677 struct ctl_table_header *hdr;
679 table = ip4_frags_ns_ctl_table;
680 if (!net_eq(net, &init_net)) {
681 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
685 table[0].data = &net->ipv4.frags.high_thresh;
686 table[1].data = &net->ipv4.frags.low_thresh;
687 table[2].data = &net->ipv4.frags.timeout;
690 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
694 net->ipv4.frags_hdr = hdr;
698 if (!net_eq(net, &init_net))
704 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
706 struct ctl_table *table;
708 table = net->ipv4.frags_hdr->ctl_table_arg;
709 unregister_net_sysctl_table(net->ipv4.frags_hdr);
713 static void ip4_frags_ctl_register(void)
715 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
718 static inline int ip4_frags_ns_ctl_register(struct net *net)
723 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
727 static inline void ip4_frags_ctl_register(void)
732 static int __net_init ipv4_frags_init_net(struct net *net)
735 * Fragment cache limits. We will commit 256K at one time. Should we
736 * cross that limit we will prune down to 192K. This should cope with
737 * even the most extreme cases without allowing an attacker to
738 * measurably harm machine performance.
740 net->ipv4.frags.high_thresh = 256 * 1024;
741 net->ipv4.frags.low_thresh = 192 * 1024;
743 * Important NOTE! Fragment queue must be destroyed before MSL expires.
744 * RFC791 is wrong proposing to prolongate timer each fragment arrival
747 net->ipv4.frags.timeout = IP_FRAG_TIME;
749 inet_frags_init_net(&net->ipv4.frags);
751 return ip4_frags_ns_ctl_register(net);
754 static void __net_exit ipv4_frags_exit_net(struct net *net)
756 ip4_frags_ns_ctl_unregister(net);
757 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
760 static struct pernet_operations ip4_frags_ops = {
761 .init = ipv4_frags_init_net,
762 .exit = ipv4_frags_exit_net,
765 void __init ipfrag_init(void)
767 ip4_frags_ctl_register();
768 register_pernet_subsys(&ip4_frags_ops);
769 ip4_frags.hashfn = ip4_hashfn;
770 ip4_frags.constructor = ip4_frag_init;
771 ip4_frags.destructor = ip4_frag_free;
772 ip4_frags.skb_free = NULL;
773 ip4_frags.qsize = sizeof(struct ipq);
774 ip4_frags.match = ip4_frag_match;
775 ip4_frags.frag_expire = ip_expire;
776 ip4_frags.secret_interval = 10 * 60 * HZ;
777 inet_frags_init(&ip4_frags);
780 EXPORT_SYMBOL(ip_defrag);