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 void frag_kfree_skb(struct netns_frags
*nf
, struct sk_buff
*skb
)
129 atomic_sub(skb
->truesize
, &nf
->mem
);
133 static void ip4_frag_init(struct inet_frag_queue
*q
, void *a
)
135 struct ipq
*qp
= container_of(q
, struct ipq
, q
);
136 struct ip4_create_arg
*arg
= a
;
138 qp
->protocol
= arg
->iph
->protocol
;
139 qp
->id
= arg
->iph
->id
;
140 qp
->saddr
= arg
->iph
->saddr
;
141 qp
->daddr
= arg
->iph
->daddr
;
142 qp
->user
= arg
->user
;
143 qp
->peer
= sysctl_ipfrag_max_dist
?
144 inet_getpeer(arg
->iph
->saddr
, 1) : NULL
;
147 static __inline__
void ip4_frag_free(struct inet_frag_queue
*q
)
151 qp
= container_of(q
, struct ipq
, q
);
153 inet_putpeer(qp
->peer
);
157 /* Destruction primitives. */
159 static __inline__
void ipq_put(struct ipq
*ipq
)
161 inet_frag_put(&ipq
->q
, &ip4_frags
);
164 /* Kill ipq entry. It is not destroyed immediately,
165 * because caller (and someone more) holds reference count.
167 static void ipq_kill(struct ipq
*ipq
)
169 inet_frag_kill(&ipq
->q
, &ip4_frags
);
172 /* Memory limiting on fragments. Evictor trashes the oldest
173 * fragment queue until we are back under the threshold.
175 static void ip_evictor(struct net
*net
)
179 evicted
= inet_frag_evictor(&net
->ipv4
.frags
, &ip4_frags
);
181 IP_ADD_STATS_BH(net
, IPSTATS_MIB_REASMFAILS
, evicted
);
185 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
187 static void ip_expire(unsigned long arg
)
192 qp
= container_of((struct inet_frag_queue
*) arg
, struct ipq
, q
);
193 net
= container_of(qp
->q
.net
, struct net
, ipv4
.frags
);
195 spin_lock(&qp
->q
.lock
);
197 if (qp
->q
.last_in
& INET_FRAG_COMPLETE
)
202 IP_INC_STATS_BH(net
, IPSTATS_MIB_REASMTIMEOUT
);
203 IP_INC_STATS_BH(net
, IPSTATS_MIB_REASMFAILS
);
205 if ((qp
->q
.last_in
& INET_FRAG_FIRST_IN
) && qp
->q
.fragments
!= NULL
) {
206 struct sk_buff
*head
= qp
->q
.fragments
;
209 head
->dev
= dev_get_by_index_rcu(net
, qp
->iif
);
214 * Only search router table for the head fragment,
215 * when defraging timeout at PRE_ROUTING HOOK.
217 if (qp
->user
== IP_DEFRAG_CONNTRACK_IN
&& !skb_dst(head
)) {
218 const struct iphdr
*iph
= ip_hdr(head
);
219 int err
= ip_route_input(head
, iph
->daddr
, iph
->saddr
,
220 iph
->tos
, head
->dev
);
225 * Only an end host needs to send an ICMP
226 * "Fragment Reassembly Timeout" message, per RFC792.
228 if (skb_rtable(head
)->rt_type
!= RTN_LOCAL
)
233 /* Send an ICMP "Fragment Reassembly Timeout" message. */
234 icmp_send(head
, ICMP_TIME_EXCEEDED
, ICMP_EXC_FRAGTIME
, 0);
239 spin_unlock(&qp
->q
.lock
);
243 /* Find the correct entry in the "incomplete datagrams" queue for
244 * this IP datagram, and create new one, if nothing is found.
246 static inline struct ipq
*ip_find(struct net
*net
, struct iphdr
*iph
, u32 user
)
248 struct inet_frag_queue
*q
;
249 struct ip4_create_arg arg
;
255 read_lock(&ip4_frags
.lock
);
256 hash
= ipqhashfn(iph
->id
, iph
->saddr
, iph
->daddr
, iph
->protocol
);
258 q
= inet_frag_find(&net
->ipv4
.frags
, &ip4_frags
, &arg
, hash
);
262 return container_of(q
, struct ipq
, q
);
265 LIMIT_NETDEBUG(KERN_ERR
"ip_frag_create: no memory left !\n");
269 /* Is the fragment too far ahead to be part of ipq? */
270 static inline int ip_frag_too_far(struct ipq
*qp
)
272 struct inet_peer
*peer
= qp
->peer
;
273 unsigned int max
= sysctl_ipfrag_max_dist
;
274 unsigned int start
, end
;
282 end
= atomic_inc_return(&peer
->rid
);
285 rc
= qp
->q
.fragments
&& (end
- start
) > max
;
290 net
= container_of(qp
->q
.net
, struct net
, ipv4
.frags
);
291 IP_INC_STATS_BH(net
, IPSTATS_MIB_REASMFAILS
);
297 static int ip_frag_reinit(struct ipq
*qp
)
301 if (!mod_timer(&qp
->q
.timer
, jiffies
+ qp
->q
.net
->timeout
)) {
302 atomic_inc(&qp
->q
.refcnt
);
306 fp
= qp
->q
.fragments
;
308 struct sk_buff
*xp
= fp
->next
;
309 frag_kfree_skb(qp
->q
.net
, fp
);
316 qp
->q
.fragments
= NULL
;
322 /* Add new segment to existing queue. */
323 static int ip_frag_queue(struct ipq
*qp
, struct sk_buff
*skb
)
325 struct sk_buff
*prev
, *next
;
326 struct net_device
*dev
;
331 if (qp
->q
.last_in
& INET_FRAG_COMPLETE
)
334 if (!(IPCB(skb
)->flags
& IPSKB_FRAG_COMPLETE
) &&
335 unlikely(ip_frag_too_far(qp
)) &&
336 unlikely(err
= ip_frag_reinit(qp
))) {
341 offset
= ntohs(ip_hdr(skb
)->frag_off
);
342 flags
= offset
& ~IP_OFFSET
;
344 offset
<<= 3; /* offset is in 8-byte chunks */
345 ihl
= ip_hdrlen(skb
);
347 /* Determine the position of this fragment. */
348 end
= offset
+ skb
->len
- ihl
;
351 /* Is this the final fragment? */
352 if ((flags
& IP_MF
) == 0) {
353 /* If we already have some bits beyond end
354 * or have different end, the segment is corrrupted.
356 if (end
< qp
->q
.len
||
357 ((qp
->q
.last_in
& INET_FRAG_LAST_IN
) && end
!= qp
->q
.len
))
359 qp
->q
.last_in
|= INET_FRAG_LAST_IN
;
364 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
365 skb
->ip_summed
= CHECKSUM_NONE
;
367 if (end
> qp
->q
.len
) {
368 /* Some bits beyond end -> corruption. */
369 if (qp
->q
.last_in
& INET_FRAG_LAST_IN
)
378 if (pskb_pull(skb
, ihl
) == NULL
)
381 err
= pskb_trim_rcsum(skb
, end
- offset
);
385 /* Find out which fragments are in front and at the back of us
386 * in the chain of fragments so far. We must know where to put
387 * this fragment, right?
390 for (next
= qp
->q
.fragments
; next
!= NULL
; next
= next
->next
) {
391 if (FRAG_CB(next
)->offset
>= offset
)
396 /* We found where to put this one. Check for overlap with
397 * preceding fragment, and, if needed, align things so that
398 * any overlaps are eliminated.
401 int i
= (FRAG_CB(prev
)->offset
+ prev
->len
) - offset
;
409 if (!pskb_pull(skb
, i
))
411 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
412 skb
->ip_summed
= CHECKSUM_NONE
;
418 while (next
&& FRAG_CB(next
)->offset
< end
) {
419 int i
= end
- FRAG_CB(next
)->offset
; /* overlap is 'i' bytes */
422 /* Eat head of the next overlapped fragment
423 * and leave the loop. The next ones cannot overlap.
425 if (!pskb_pull(next
, i
))
427 FRAG_CB(next
)->offset
+= i
;
429 if (next
->ip_summed
!= CHECKSUM_UNNECESSARY
)
430 next
->ip_summed
= CHECKSUM_NONE
;
433 struct sk_buff
*free_it
= next
;
435 /* Old fragment is completely overridden with
443 qp
->q
.fragments
= next
;
445 qp
->q
.meat
-= free_it
->len
;
446 frag_kfree_skb(qp
->q
.net
, free_it
);
450 FRAG_CB(skb
)->offset
= offset
;
452 /* Insert this fragment in the chain of fragments. */
457 qp
->q
.fragments
= skb
;
461 qp
->iif
= dev
->ifindex
;
464 qp
->q
.stamp
= skb
->tstamp
;
465 qp
->q
.meat
+= skb
->len
;
466 atomic_add(skb
->truesize
, &qp
->q
.net
->mem
);
468 qp
->q
.last_in
|= INET_FRAG_FIRST_IN
;
470 if (qp
->q
.last_in
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
471 qp
->q
.meat
== qp
->q
.len
)
472 return ip_frag_reasm(qp
, prev
, dev
);
474 write_lock(&ip4_frags
.lock
);
475 list_move_tail(&qp
->q
.lru_list
, &qp
->q
.net
->lru_list
);
476 write_unlock(&ip4_frags
.lock
);
485 /* Build a new IP datagram from all its fragments. */
487 static int ip_frag_reasm(struct ipq
*qp
, struct sk_buff
*prev
,
488 struct net_device
*dev
)
490 struct net
*net
= container_of(qp
->q
.net
, struct net
, ipv4
.frags
);
492 struct sk_buff
*fp
, *head
= qp
->q
.fragments
;
499 /* Make the one we just received the head. */
502 fp
= skb_clone(head
, GFP_ATOMIC
);
506 fp
->next
= head
->next
;
509 skb_morph(head
, qp
->q
.fragments
);
510 head
->next
= qp
->q
.fragments
->next
;
512 kfree_skb(qp
->q
.fragments
);
513 qp
->q
.fragments
= head
;
516 WARN_ON(head
== NULL
);
517 WARN_ON(FRAG_CB(head
)->offset
!= 0);
519 /* Allocate a new buffer for the datagram. */
520 ihlen
= ip_hdrlen(head
);
521 len
= ihlen
+ qp
->q
.len
;
527 /* Head of list must not be cloned. */
528 if (skb_cloned(head
) && pskb_expand_head(head
, 0, 0, GFP_ATOMIC
))
531 /* If the first fragment is fragmented itself, we split
532 * it to two chunks: the first with data and paged part
533 * and the second, holding only fragments. */
534 if (skb_has_frags(head
)) {
535 struct sk_buff
*clone
;
538 if ((clone
= alloc_skb(0, GFP_ATOMIC
)) == NULL
)
540 clone
->next
= head
->next
;
542 skb_shinfo(clone
)->frag_list
= skb_shinfo(head
)->frag_list
;
543 skb_frag_list_init(head
);
544 for (i
=0; i
<skb_shinfo(head
)->nr_frags
; i
++)
545 plen
+= skb_shinfo(head
)->frags
[i
].size
;
546 clone
->len
= clone
->data_len
= head
->data_len
- plen
;
547 head
->data_len
-= clone
->len
;
548 head
->len
-= clone
->len
;
550 clone
->ip_summed
= head
->ip_summed
;
551 atomic_add(clone
->truesize
, &qp
->q
.net
->mem
);
554 skb_shinfo(head
)->frag_list
= head
->next
;
555 skb_push(head
, head
->data
- skb_network_header(head
));
557 for (fp
=head
->next
; fp
; fp
= fp
->next
) {
558 head
->data_len
+= fp
->len
;
559 head
->len
+= fp
->len
;
560 if (head
->ip_summed
!= fp
->ip_summed
)
561 head
->ip_summed
= CHECKSUM_NONE
;
562 else if (head
->ip_summed
== CHECKSUM_COMPLETE
)
563 head
->csum
= csum_add(head
->csum
, fp
->csum
);
564 head
->truesize
+= fp
->truesize
;
566 atomic_sub(head
->truesize
, &qp
->q
.net
->mem
);
570 head
->tstamp
= qp
->q
.stamp
;
574 iph
->tot_len
= htons(len
);
575 IP_INC_STATS_BH(net
, IPSTATS_MIB_REASMOKS
);
576 qp
->q
.fragments
= NULL
;
580 LIMIT_NETDEBUG(KERN_ERR
"IP: queue_glue: no memory for gluing "
586 printk(KERN_INFO
"Oversized IP packet from %pI4.\n",
589 IP_INC_STATS_BH(net
, IPSTATS_MIB_REASMFAILS
);
593 /* Process an incoming IP datagram fragment. */
594 int ip_defrag(struct sk_buff
*skb
, u32 user
)
599 net
= skb
->dev
? dev_net(skb
->dev
) : dev_net(skb_dst(skb
)->dev
);
600 IP_INC_STATS_BH(net
, IPSTATS_MIB_REASMREQDS
);
602 /* Start by cleaning up the memory. */
603 if (atomic_read(&net
->ipv4
.frags
.mem
) > net
->ipv4
.frags
.high_thresh
)
606 /* Lookup (or create) queue header */
607 if ((qp
= ip_find(net
, ip_hdr(skb
), user
)) != NULL
) {
610 spin_lock(&qp
->q
.lock
);
612 ret
= ip_frag_queue(qp
, skb
);
614 spin_unlock(&qp
->q
.lock
);
619 IP_INC_STATS_BH(net
, IPSTATS_MIB_REASMFAILS
);
627 static struct ctl_table ip4_frags_ns_ctl_table
[] = {
629 .procname
= "ipfrag_high_thresh",
630 .data
= &init_net
.ipv4
.frags
.high_thresh
,
631 .maxlen
= sizeof(int),
633 .proc_handler
= proc_dointvec
636 .procname
= "ipfrag_low_thresh",
637 .data
= &init_net
.ipv4
.frags
.low_thresh
,
638 .maxlen
= sizeof(int),
640 .proc_handler
= proc_dointvec
643 .procname
= "ipfrag_time",
644 .data
= &init_net
.ipv4
.frags
.timeout
,
645 .maxlen
= sizeof(int),
647 .proc_handler
= proc_dointvec_jiffies
,
652 static struct ctl_table ip4_frags_ctl_table
[] = {
654 .procname
= "ipfrag_secret_interval",
655 .data
= &ip4_frags
.secret_interval
,
656 .maxlen
= sizeof(int),
658 .proc_handler
= proc_dointvec_jiffies
,
661 .procname
= "ipfrag_max_dist",
662 .data
= &sysctl_ipfrag_max_dist
,
663 .maxlen
= sizeof(int),
665 .proc_handler
= proc_dointvec_minmax
,
671 static int __net_init
ip4_frags_ns_ctl_register(struct net
*net
)
673 struct ctl_table
*table
;
674 struct ctl_table_header
*hdr
;
676 table
= ip4_frags_ns_ctl_table
;
677 if (!net_eq(net
, &init_net
)) {
678 table
= kmemdup(table
, sizeof(ip4_frags_ns_ctl_table
), GFP_KERNEL
);
682 table
[0].data
= &net
->ipv4
.frags
.high_thresh
;
683 table
[1].data
= &net
->ipv4
.frags
.low_thresh
;
684 table
[2].data
= &net
->ipv4
.frags
.timeout
;
687 hdr
= register_net_sysctl_table(net
, net_ipv4_ctl_path
, table
);
691 net
->ipv4
.frags_hdr
= hdr
;
695 if (!net_eq(net
, &init_net
))
701 static void __net_exit
ip4_frags_ns_ctl_unregister(struct net
*net
)
703 struct ctl_table
*table
;
705 table
= net
->ipv4
.frags_hdr
->ctl_table_arg
;
706 unregister_net_sysctl_table(net
->ipv4
.frags_hdr
);
710 static void ip4_frags_ctl_register(void)
712 register_net_sysctl_rotable(net_ipv4_ctl_path
, ip4_frags_ctl_table
);
715 static inline int ip4_frags_ns_ctl_register(struct net
*net
)
720 static inline void ip4_frags_ns_ctl_unregister(struct net
*net
)
724 static inline void ip4_frags_ctl_register(void)
729 static int __net_init
ipv4_frags_init_net(struct net
*net
)
732 * Fragment cache limits. We will commit 256K at one time. Should we
733 * cross that limit we will prune down to 192K. This should cope with
734 * even the most extreme cases without allowing an attacker to
735 * measurably harm machine performance.
737 net
->ipv4
.frags
.high_thresh
= 256 * 1024;
738 net
->ipv4
.frags
.low_thresh
= 192 * 1024;
740 * Important NOTE! Fragment queue must be destroyed before MSL expires.
741 * RFC791 is wrong proposing to prolongate timer each fragment arrival
744 net
->ipv4
.frags
.timeout
= IP_FRAG_TIME
;
746 inet_frags_init_net(&net
->ipv4
.frags
);
748 return ip4_frags_ns_ctl_register(net
);
751 static void __net_exit
ipv4_frags_exit_net(struct net
*net
)
753 ip4_frags_ns_ctl_unregister(net
);
754 inet_frags_exit_net(&net
->ipv4
.frags
, &ip4_frags
);
757 static struct pernet_operations ip4_frags_ops
= {
758 .init
= ipv4_frags_init_net
,
759 .exit
= ipv4_frags_exit_net
,
762 void __init
ipfrag_init(void)
764 ip4_frags_ctl_register();
765 register_pernet_subsys(&ip4_frags_ops
);
766 ip4_frags
.hashfn
= ip4_hashfn
;
767 ip4_frags
.constructor
= ip4_frag_init
;
768 ip4_frags
.destructor
= ip4_frag_free
;
769 ip4_frags
.skb_free
= NULL
;
770 ip4_frags
.qsize
= sizeof(struct ipq
);
771 ip4_frags
.match
= ip4_frag_match
;
772 ip4_frags
.frag_expire
= ip_expire
;
773 ip4_frags
.secret_interval
= 10 * 60 * HZ
;
774 inet_frags_init(&ip4_frags
);
777 EXPORT_SYMBOL(ip_defrag
);