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 <linux/slab.h>
36 #include <net/route.h>
41 #include <net/checksum.h>
42 #include <net/inetpeer.h>
43 #include <net/inet_frag.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/inet.h>
47 #include <linux/netfilter_ipv4.h>
49 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
50 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
51 * as well. Or notify me, at least. --ANK
54 static int sysctl_ipfrag_max_dist __read_mostly = 64;
58 struct inet_skb_parm h;
62 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
64 /* Describe an entry in the "incomplete datagrams" queue. */
66 struct inet_frag_queue q;
75 struct inet_peer *peer;
78 static struct inet_frags ip4_frags;
80 int ip_frag_nqueues(struct net *net)
82 return net->ipv4.frags.nqueues;
85 int ip_frag_mem(struct net *net)
87 return atomic_read(&net->ipv4.frags.mem);
90 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
91 struct net_device *dev);
93 struct ip4_create_arg {
98 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
100 return jhash_3words((__force u32)id << 16 | prot,
101 (__force u32)saddr, (__force u32)daddr,
102 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
105 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
109 ipq = container_of(q, struct ipq, q);
110 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
113 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
116 struct ip4_create_arg *arg = a;
118 qp = container_of(q, struct ipq, q);
119 return (qp->id == arg->iph->id &&
120 qp->saddr == arg->iph->saddr &&
121 qp->daddr == arg->iph->daddr &&
122 qp->protocol == arg->iph->protocol &&
123 qp->user == arg->user);
126 /* Memory Tracking Functions. */
127 static __inline__ void frag_kfree_skb(struct netns_frags *nf,
128 struct sk_buff *skb, int *work)
131 *work -= skb->truesize;
132 atomic_sub(skb->truesize, &nf->mem);
136 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
138 struct ipq *qp = container_of(q, struct ipq, q);
139 struct ip4_create_arg *arg = a;
141 qp->protocol = arg->iph->protocol;
142 qp->id = arg->iph->id;
143 qp->saddr = arg->iph->saddr;
144 qp->daddr = arg->iph->daddr;
145 qp->user = arg->user;
146 qp->peer = sysctl_ipfrag_max_dist ?
147 inet_getpeer(arg->iph->saddr, 1) : NULL;
150 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
154 qp = container_of(q, struct ipq, q);
156 inet_putpeer(qp->peer);
160 /* Destruction primitives. */
162 static __inline__ void ipq_put(struct ipq *ipq)
164 inet_frag_put(&ipq->q, &ip4_frags);
167 /* Kill ipq entry. It is not destroyed immediately,
168 * because caller (and someone more) holds reference count.
170 static void ipq_kill(struct ipq *ipq)
172 inet_frag_kill(&ipq->q, &ip4_frags);
175 /* Memory limiting on fragments. Evictor trashes the oldest
176 * fragment queue until we are back under the threshold.
178 static void ip_evictor(struct net *net)
182 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
184 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
188 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
190 static void ip_expire(unsigned long arg)
195 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
196 net = container_of(qp->q.net, struct net, ipv4.frags);
198 spin_lock(&qp->q.lock);
200 if (qp->q.last_in & INET_FRAG_COMPLETE)
205 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
206 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
208 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
209 struct sk_buff *head = qp->q.fragments;
212 head->dev = dev_get_by_index_rcu(net, qp->iif);
217 * Only search router table for the head fragment,
218 * when defraging timeout at PRE_ROUTING HOOK.
220 if (qp->user == IP_DEFRAG_CONNTRACK_IN && !skb_dst(head)) {
221 const struct iphdr *iph = ip_hdr(head);
222 int err = ip_route_input(head, iph->daddr, iph->saddr,
223 iph->tos, head->dev);
228 * Only an end host needs to send an ICMP
229 * "Fragment Reassembly Timeout" message, per RFC792.
231 if (skb_rtable(head)->rt_type != RTN_LOCAL)
236 /* Send an ICMP "Fragment Reassembly Timeout" message. */
237 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
242 spin_unlock(&qp->q.lock);
246 /* Find the correct entry in the "incomplete datagrams" queue for
247 * this IP datagram, and create new one, if nothing is found.
249 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
251 struct inet_frag_queue *q;
252 struct ip4_create_arg arg;
258 read_lock(&ip4_frags.lock);
259 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
261 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
265 return container_of(q, struct ipq, q);
268 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
272 /* Is the fragment too far ahead to be part of ipq? */
273 static inline int ip_frag_too_far(struct ipq *qp)
275 struct inet_peer *peer = qp->peer;
276 unsigned int max = sysctl_ipfrag_max_dist;
277 unsigned int start, end;
285 end = atomic_inc_return(&peer->rid);
288 rc = qp->q.fragments && (end - start) > max;
293 net = container_of(qp->q.net, struct net, ipv4.frags);
294 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
300 static int ip_frag_reinit(struct ipq *qp)
304 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
305 atomic_inc(&qp->q.refcnt);
309 fp = qp->q.fragments;
311 struct sk_buff *xp = fp->next;
312 frag_kfree_skb(qp->q.net, fp, NULL);
319 qp->q.fragments = NULL;
325 /* Add new segment to existing queue. */
326 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
328 struct sk_buff *prev, *next;
329 struct net_device *dev;
334 if (qp->q.last_in & INET_FRAG_COMPLETE)
337 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
338 unlikely(ip_frag_too_far(qp)) &&
339 unlikely(err = ip_frag_reinit(qp))) {
344 offset = ntohs(ip_hdr(skb)->frag_off);
345 flags = offset & ~IP_OFFSET;
347 offset <<= 3; /* offset is in 8-byte chunks */
348 ihl = ip_hdrlen(skb);
350 /* Determine the position of this fragment. */
351 end = offset + skb->len - ihl;
354 /* Is this the final fragment? */
355 if ((flags & IP_MF) == 0) {
356 /* If we already have some bits beyond end
357 * or have different end, the segment is corrrupted.
359 if (end < qp->q.len ||
360 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
362 qp->q.last_in |= INET_FRAG_LAST_IN;
367 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
368 skb->ip_summed = CHECKSUM_NONE;
370 if (end > qp->q.len) {
371 /* Some bits beyond end -> corruption. */
372 if (qp->q.last_in & INET_FRAG_LAST_IN)
381 if (pskb_pull(skb, ihl) == NULL)
384 err = pskb_trim_rcsum(skb, end - offset);
388 /* Find out which fragments are in front and at the back of us
389 * in the chain of fragments so far. We must know where to put
390 * this fragment, right?
393 for (next = qp->q.fragments; next != NULL; next = next->next) {
394 if (FRAG_CB(next)->offset >= offset)
399 /* We found where to put this one. Check for overlap with
400 * preceding fragment, and, if needed, align things so that
401 * any overlaps are eliminated.
404 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
412 if (!pskb_pull(skb, i))
414 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
415 skb->ip_summed = CHECKSUM_NONE;
421 while (next && FRAG_CB(next)->offset < end) {
422 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
425 /* Eat head of the next overlapped fragment
426 * and leave the loop. The next ones cannot overlap.
428 if (!pskb_pull(next, i))
430 FRAG_CB(next)->offset += i;
432 if (next->ip_summed != CHECKSUM_UNNECESSARY)
433 next->ip_summed = CHECKSUM_NONE;
436 struct sk_buff *free_it = next;
438 /* Old fragment is completely overridden with
446 qp->q.fragments = next;
448 qp->q.meat -= free_it->len;
449 frag_kfree_skb(qp->q.net, free_it, NULL);
453 FRAG_CB(skb)->offset = offset;
455 /* Insert this fragment in the chain of fragments. */
460 qp->q.fragments = skb;
464 qp->iif = dev->ifindex;
467 qp->q.stamp = skb->tstamp;
468 qp->q.meat += skb->len;
469 atomic_add(skb->truesize, &qp->q.net->mem);
471 qp->q.last_in |= INET_FRAG_FIRST_IN;
473 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
474 qp->q.meat == qp->q.len)
475 return ip_frag_reasm(qp, prev, dev);
477 write_lock(&ip4_frags.lock);
478 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
479 write_unlock(&ip4_frags.lock);
488 /* Build a new IP datagram from all its fragments. */
490 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
491 struct net_device *dev)
493 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
495 struct sk_buff *fp, *head = qp->q.fragments;
502 /* Make the one we just received the head. */
505 fp = skb_clone(head, GFP_ATOMIC);
509 fp->next = head->next;
512 skb_morph(head, qp->q.fragments);
513 head->next = qp->q.fragments->next;
515 kfree_skb(qp->q.fragments);
516 qp->q.fragments = head;
519 WARN_ON(head == NULL);
520 WARN_ON(FRAG_CB(head)->offset != 0);
522 /* Allocate a new buffer for the datagram. */
523 ihlen = ip_hdrlen(head);
524 len = ihlen + qp->q.len;
530 /* Head of list must not be cloned. */
531 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
534 /* If the first fragment is fragmented itself, we split
535 * it to two chunks: the first with data and paged part
536 * and the second, holding only fragments. */
537 if (skb_has_frags(head)) {
538 struct sk_buff *clone;
541 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
543 clone->next = head->next;
545 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
546 skb_frag_list_init(head);
547 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
548 plen += skb_shinfo(head)->frags[i].size;
549 clone->len = clone->data_len = head->data_len - plen;
550 head->data_len -= clone->len;
551 head->len -= clone->len;
553 clone->ip_summed = head->ip_summed;
554 atomic_add(clone->truesize, &qp->q.net->mem);
557 skb_shinfo(head)->frag_list = head->next;
558 skb_push(head, head->data - skb_network_header(head));
559 atomic_sub(head->truesize, &qp->q.net->mem);
561 for (fp=head->next; fp; fp = fp->next) {
562 head->data_len += fp->len;
563 head->len += fp->len;
564 if (head->ip_summed != fp->ip_summed)
565 head->ip_summed = CHECKSUM_NONE;
566 else if (head->ip_summed == CHECKSUM_COMPLETE)
567 head->csum = csum_add(head->csum, fp->csum);
568 head->truesize += fp->truesize;
569 atomic_sub(fp->truesize, &qp->q.net->mem);
574 head->tstamp = qp->q.stamp;
578 iph->tot_len = htons(len);
579 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
580 qp->q.fragments = NULL;
584 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
590 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
593 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
597 /* Process an incoming IP datagram fragment. */
598 int ip_defrag(struct sk_buff *skb, u32 user)
603 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
604 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
606 /* Start by cleaning up the memory. */
607 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
610 /* Lookup (or create) queue header */
611 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
614 spin_lock(&qp->q.lock);
616 ret = ip_frag_queue(qp, skb);
618 spin_unlock(&qp->q.lock);
623 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
631 static struct ctl_table ip4_frags_ns_ctl_table[] = {
633 .procname = "ipfrag_high_thresh",
634 .data = &init_net.ipv4.frags.high_thresh,
635 .maxlen = sizeof(int),
637 .proc_handler = proc_dointvec
640 .procname = "ipfrag_low_thresh",
641 .data = &init_net.ipv4.frags.low_thresh,
642 .maxlen = sizeof(int),
644 .proc_handler = proc_dointvec
647 .procname = "ipfrag_time",
648 .data = &init_net.ipv4.frags.timeout,
649 .maxlen = sizeof(int),
651 .proc_handler = proc_dointvec_jiffies,
656 static struct ctl_table ip4_frags_ctl_table[] = {
658 .procname = "ipfrag_secret_interval",
659 .data = &ip4_frags.secret_interval,
660 .maxlen = sizeof(int),
662 .proc_handler = proc_dointvec_jiffies,
665 .procname = "ipfrag_max_dist",
666 .data = &sysctl_ipfrag_max_dist,
667 .maxlen = sizeof(int),
669 .proc_handler = proc_dointvec_minmax,
675 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
677 struct ctl_table *table;
678 struct ctl_table_header *hdr;
680 table = ip4_frags_ns_ctl_table;
681 if (!net_eq(net, &init_net)) {
682 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
686 table[0].data = &net->ipv4.frags.high_thresh;
687 table[1].data = &net->ipv4.frags.low_thresh;
688 table[2].data = &net->ipv4.frags.timeout;
691 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
695 net->ipv4.frags_hdr = hdr;
699 if (!net_eq(net, &init_net))
705 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
707 struct ctl_table *table;
709 table = net->ipv4.frags_hdr->ctl_table_arg;
710 unregister_net_sysctl_table(net->ipv4.frags_hdr);
714 static void ip4_frags_ctl_register(void)
716 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
719 static inline int ip4_frags_ns_ctl_register(struct net *net)
724 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
728 static inline void ip4_frags_ctl_register(void)
733 static int __net_init ipv4_frags_init_net(struct net *net)
736 * Fragment cache limits. We will commit 256K at one time. Should we
737 * cross that limit we will prune down to 192K. This should cope with
738 * even the most extreme cases without allowing an attacker to
739 * measurably harm machine performance.
741 net->ipv4.frags.high_thresh = 256 * 1024;
742 net->ipv4.frags.low_thresh = 192 * 1024;
744 * Important NOTE! Fragment queue must be destroyed before MSL expires.
745 * RFC791 is wrong proposing to prolongate timer each fragment arrival
748 net->ipv4.frags.timeout = IP_FRAG_TIME;
750 inet_frags_init_net(&net->ipv4.frags);
752 return ip4_frags_ns_ctl_register(net);
755 static void __net_exit ipv4_frags_exit_net(struct net *net)
757 ip4_frags_ns_ctl_unregister(net);
758 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
761 static struct pernet_operations ip4_frags_ops = {
762 .init = ipv4_frags_init_net,
763 .exit = ipv4_frags_exit_net,
766 void __init ipfrag_init(void)
768 ip4_frags_ctl_register();
769 register_pernet_subsys(&ip4_frags_ops);
770 ip4_frags.hashfn = ip4_hashfn;
771 ip4_frags.constructor = ip4_frag_init;
772 ip4_frags.destructor = ip4_frag_free;
773 ip4_frags.skb_free = NULL;
774 ip4_frags.qsize = sizeof(struct ipq);
775 ip4_frags.match = ip4_frag_match;
776 ip4_frags.frag_expire = ip_expire;
777 ip4_frags.secret_interval = 10 * 60 * HZ;
778 inet_frags_init(&ip4_frags);
781 EXPORT_SYMBOL(ip_defrag);