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 #define pr_fmt(fmt) "IPv4: " fmt
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>
37 #include <linux/slab.h>
38 #include <net/route.h>
43 #include <net/checksum.h>
44 #include <net/inetpeer.h>
45 #include <net/inet_frag.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/inet.h>
49 #include <linux/netfilter_ipv4.h>
50 #include <net/inet_ecn.h>
51 #include <net/l3mdev.h>
53 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
54 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
55 * as well. Or notify me, at least. --ANK
57 static const char ip_frag_cache_name
[] = "ip4-frags";
59 /* Describe an entry in the "incomplete datagrams" queue. */
61 struct inet_frag_queue q
;
63 u8 ecn
; /* RFC3168 support */
64 u16 max_df_size
; /* largest frag with DF set seen */
67 struct inet_peer
*peer
;
70 static u8
ip4_frag_ecn(u8 tos
)
72 return 1 << (tos
& INET_ECN_MASK
);
75 static struct inet_frags ip4_frags
;
77 static int ip_frag_reasm(struct ipq
*qp
, struct sk_buff
*skb
,
78 struct sk_buff
*prev_tail
, struct net_device
*dev
);
81 static void ip4_frag_init(struct inet_frag_queue
*q
, const void *a
)
83 struct ipq
*qp
= container_of(q
, struct ipq
, q
);
84 struct netns_ipv4
*ipv4
= container_of(q
->net
, struct netns_ipv4
,
86 struct net
*net
= container_of(ipv4
, struct net
, ipv4
);
88 const struct frag_v4_compare_key
*key
= a
;
92 qp
->peer
= q
->net
->max_dist
?
93 inet_getpeer_v4(net
->ipv4
.peers
, key
->saddr
, key
->vif
, 1) :
97 static void ip4_frag_free(struct inet_frag_queue
*q
)
101 qp
= container_of(q
, struct ipq
, q
);
103 inet_putpeer(qp
->peer
);
107 /* Destruction primitives. */
109 static void ipq_put(struct ipq
*ipq
)
111 inet_frag_put(&ipq
->q
);
114 /* Kill ipq entry. It is not destroyed immediately,
115 * because caller (and someone more) holds reference count.
117 static void ipq_kill(struct ipq
*ipq
)
119 inet_frag_kill(&ipq
->q
);
122 static bool frag_expire_skip_icmp(u32 user
)
124 return user
== IP_DEFRAG_AF_PACKET
||
125 ip_defrag_user_in_between(user
, IP_DEFRAG_CONNTRACK_IN
,
126 __IP_DEFRAG_CONNTRACK_IN_END
) ||
127 ip_defrag_user_in_between(user
, IP_DEFRAG_CONNTRACK_BRIDGE_IN
,
128 __IP_DEFRAG_CONNTRACK_BRIDGE_IN
);
132 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
134 static void ip_expire(unsigned long arg
)
136 const struct iphdr
*iph
;
137 struct sk_buff
*head
= NULL
;
142 qp
= container_of((struct inet_frag_queue
*) arg
, struct ipq
, q
);
143 net
= container_of(qp
->q
.net
, struct net
, ipv4
.frags
);
146 spin_lock(&qp
->q
.lock
);
148 if (qp
->q
.flags
& INET_FRAG_COMPLETE
)
152 __IP_INC_STATS(net
, IPSTATS_MIB_REASMFAILS
);
153 __IP_INC_STATS(net
, IPSTATS_MIB_REASMTIMEOUT
);
155 if (!(qp
->q
.flags
& INET_FRAG_FIRST_IN
))
158 /* sk_buff::dev and sk_buff::rbnode are unionized. So we
159 * pull the head out of the tree in order to be able to
160 * deal with head->dev.
162 head
= inet_frag_pull_head(&qp
->q
);
165 head
->dev
= dev_get_by_index_rcu(net
, qp
->iif
);
170 /* skb has no dst, perform route lookup again */
172 err
= ip_route_input_noref(head
, iph
->daddr
, iph
->saddr
,
173 iph
->tos
, head
->dev
);
177 /* Only an end host needs to send an ICMP
178 * "Fragment Reassembly Timeout" message, per RFC792.
180 if (frag_expire_skip_icmp(qp
->q
.key
.v4
.user
) &&
181 (skb_rtable(head
)->rt_type
!= RTN_LOCAL
))
184 spin_unlock(&qp
->q
.lock
);
185 icmp_send(head
, ICMP_TIME_EXCEEDED
, ICMP_EXC_FRAGTIME
, 0);
189 spin_unlock(&qp
->q
.lock
);
197 /* Find the correct entry in the "incomplete datagrams" queue for
198 * this IP datagram, and create new one, if nothing is found.
200 static struct ipq
*ip_find(struct net
*net
, struct iphdr
*iph
,
203 struct frag_v4_compare_key key
= {
209 .protocol
= iph
->protocol
,
211 struct inet_frag_queue
*q
;
213 q
= inet_frag_find(&net
->ipv4
.frags
, &key
);
217 return container_of(q
, struct ipq
, q
);
220 /* Is the fragment too far ahead to be part of ipq? */
221 static int ip_frag_too_far(struct ipq
*qp
)
223 struct inet_peer
*peer
= qp
->peer
;
224 unsigned int max
= qp
->q
.net
->max_dist
;
225 unsigned int start
, end
;
233 end
= atomic_inc_return(&peer
->rid
);
236 rc
= qp
->q
.fragments_tail
&& (end
- start
) > max
;
241 net
= container_of(qp
->q
.net
, struct net
, ipv4
.frags
);
242 __IP_INC_STATS(net
, IPSTATS_MIB_REASMFAILS
);
248 static int ip_frag_reinit(struct ipq
*qp
)
250 unsigned int sum_truesize
= 0;
252 if (!mod_timer(&qp
->q
.timer
, jiffies
+ qp
->q
.net
->timeout
)) {
253 atomic_inc(&qp
->q
.refcnt
);
257 sum_truesize
= inet_frag_rbtree_purge(&qp
->q
.rb_fragments
);
258 sub_frag_mem_limit(qp
->q
.net
, sum_truesize
);
263 qp
->q
.fragments
= NULL
;
264 qp
->q
.rb_fragments
= RB_ROOT
;
265 qp
->q
.fragments_tail
= NULL
;
266 qp
->q
.last_run_head
= NULL
;
273 /* Add new segment to existing queue. */
274 static int ip_frag_queue(struct ipq
*qp
, struct sk_buff
*skb
)
276 struct net
*net
= container_of(qp
->q
.net
, struct net
, ipv4
.frags
);
277 int ihl
, end
, flags
, offset
;
278 struct sk_buff
*prev_tail
;
279 struct net_device
*dev
;
280 unsigned int fragsize
;
284 if (qp
->q
.flags
& INET_FRAG_COMPLETE
)
287 if (!(IPCB(skb
)->flags
& IPSKB_FRAG_COMPLETE
) &&
288 unlikely(ip_frag_too_far(qp
)) &&
289 unlikely(err
= ip_frag_reinit(qp
))) {
294 ecn
= ip4_frag_ecn(ip_hdr(skb
)->tos
);
295 offset
= ntohs(ip_hdr(skb
)->frag_off
);
296 flags
= offset
& ~IP_OFFSET
;
298 offset
<<= 3; /* offset is in 8-byte chunks */
299 ihl
= ip_hdrlen(skb
);
301 /* Determine the position of this fragment. */
302 end
= offset
+ skb
->len
- skb_network_offset(skb
) - ihl
;
305 /* Is this the final fragment? */
306 if ((flags
& IP_MF
) == 0) {
307 /* If we already have some bits beyond end
308 * or have different end, the segment is corrupted.
310 if (end
< qp
->q
.len
||
311 ((qp
->q
.flags
& INET_FRAG_LAST_IN
) && end
!= qp
->q
.len
))
313 qp
->q
.flags
|= INET_FRAG_LAST_IN
;
318 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
)
319 skb
->ip_summed
= CHECKSUM_NONE
;
321 if (end
> qp
->q
.len
) {
322 /* Some bits beyond end -> corruption. */
323 if (qp
->q
.flags
& INET_FRAG_LAST_IN
)
332 if (!pskb_pull(skb
, skb_network_offset(skb
) + ihl
))
335 err
= pskb_trim_rcsum(skb
, end
- offset
);
339 /* Note : skb->rbnode and skb->dev share the same location. */
341 /* Makes sure compiler wont do silly aliasing games */
344 prev_tail
= qp
->q
.fragments_tail
;
345 err
= inet_frag_queue_insert(&qp
->q
, skb
, offset
, end
);
350 qp
->iif
= dev
->ifindex
;
352 qp
->q
.stamp
= skb
->tstamp
;
353 qp
->q
.meat
+= skb
->len
;
355 add_frag_mem_limit(qp
->q
.net
, skb
->truesize
);
357 qp
->q
.flags
|= INET_FRAG_FIRST_IN
;
359 fragsize
= skb
->len
+ ihl
;
361 if (fragsize
> qp
->q
.max_size
)
362 qp
->q
.max_size
= fragsize
;
364 if (ip_hdr(skb
)->frag_off
& htons(IP_DF
) &&
365 fragsize
> qp
->max_df_size
)
366 qp
->max_df_size
= fragsize
;
368 if (qp
->q
.flags
== (INET_FRAG_FIRST_IN
| INET_FRAG_LAST_IN
) &&
369 qp
->q
.meat
== qp
->q
.len
) {
370 unsigned long orefdst
= skb
->_skb_refdst
;
372 skb
->_skb_refdst
= 0UL;
373 err
= ip_frag_reasm(qp
, skb
, prev_tail
, dev
);
374 skb
->_skb_refdst
= orefdst
;
376 inet_frag_kill(&qp
->q
);
384 if (err
== IPFRAG_DUP
) {
389 __IP_INC_STATS(net
, IPSTATS_MIB_REASM_OVERLAPS
);
391 inet_frag_kill(&qp
->q
);
392 __IP_INC_STATS(net
, IPSTATS_MIB_REASMFAILS
);
398 /* Build a new IP datagram from all its fragments. */
399 static int ip_frag_reasm(struct ipq
*qp
, struct sk_buff
*skb
,
400 struct sk_buff
*prev_tail
, struct net_device
*dev
)
402 struct net
*net
= container_of(qp
->q
.net
, struct net
, ipv4
.frags
);
410 ecn
= ip_frag_ecn_table
[qp
->ecn
];
411 if (unlikely(ecn
== 0xff)) {
416 /* Make the one we just received the head. */
417 reasm_data
= inet_frag_reasm_prepare(&qp
->q
, skb
, prev_tail
);
421 len
= ip_hdrlen(skb
) + qp
->q
.len
;
426 inet_frag_reasm_finish(&qp
->q
, skb
, reasm_data
);
429 IPCB(skb
)->frag_max_size
= max(qp
->max_df_size
, qp
->q
.max_size
);
432 iph
->tot_len
= htons(len
);
435 /* When we set IP_DF on a refragmented skb we must also force a
436 * call to ip_fragment to avoid forwarding a DF-skb of size s while
437 * original sender only sent fragments of size f (where f < s).
439 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
440 * frag seen to avoid sending tiny DF-fragments in case skb was built
441 * from one very small df-fragment and one large non-df frag.
443 if (qp
->max_df_size
== qp
->q
.max_size
) {
444 IPCB(skb
)->flags
|= IPSKB_FRAG_PMTU
;
445 iph
->frag_off
= htons(IP_DF
);
452 __IP_INC_STATS(net
, IPSTATS_MIB_REASMOKS
);
453 qp
->q
.fragments
= NULL
;
454 qp
->q
.rb_fragments
= RB_ROOT
;
455 qp
->q
.fragments_tail
= NULL
;
456 qp
->q
.last_run_head
= NULL
;
460 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp
);
464 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp
->q
.key
.v4
.saddr
);
466 __IP_INC_STATS(net
, IPSTATS_MIB_REASMFAILS
);
470 /* Process an incoming IP datagram fragment. */
471 int ip_defrag(struct net
*net
, struct sk_buff
*skb
, u32 user
)
473 struct net_device
*dev
= skb
->dev
? : skb_dst(skb
)->dev
;
474 int vif
= l3mdev_master_ifindex_rcu(dev
);
477 __IP_INC_STATS(net
, IPSTATS_MIB_REASMREQDS
);
480 /* Lookup (or create) queue header */
481 qp
= ip_find(net
, ip_hdr(skb
), user
, vif
);
485 spin_lock(&qp
->q
.lock
);
487 ret
= ip_frag_queue(qp
, skb
);
489 spin_unlock(&qp
->q
.lock
);
494 __IP_INC_STATS(net
, IPSTATS_MIB_REASMFAILS
);
498 EXPORT_SYMBOL(ip_defrag
);
500 struct sk_buff
*ip_check_defrag(struct net
*net
, struct sk_buff
*skb
, u32 user
)
506 if (skb
->protocol
!= htons(ETH_P_IP
))
509 netoff
= skb_network_offset(skb
);
511 if (skb_copy_bits(skb
, netoff
, &iph
, sizeof(iph
)) < 0)
514 if (iph
.ihl
< 5 || iph
.version
!= 4)
517 len
= ntohs(iph
.tot_len
);
518 if (skb
->len
< netoff
+ len
|| len
< (iph
.ihl
* 4))
521 if (ip_is_fragment(&iph
)) {
522 skb
= skb_share_check(skb
, GFP_ATOMIC
);
524 if (!pskb_may_pull(skb
, netoff
+ iph
.ihl
* 4)) {
528 if (pskb_trim_rcsum(skb
, netoff
+ len
)) {
532 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
533 if (ip_defrag(net
, skb
, user
))
540 EXPORT_SYMBOL(ip_check_defrag
);
545 static struct ctl_table ip4_frags_ns_ctl_table
[] = {
547 .procname
= "ipfrag_high_thresh",
548 .data
= &init_net
.ipv4
.frags
.high_thresh
,
549 .maxlen
= sizeof(unsigned long),
551 .proc_handler
= proc_doulongvec_minmax
,
552 .extra1
= &init_net
.ipv4
.frags
.low_thresh
555 .procname
= "ipfrag_low_thresh",
556 .data
= &init_net
.ipv4
.frags
.low_thresh
,
557 .maxlen
= sizeof(unsigned long),
559 .proc_handler
= proc_doulongvec_minmax
,
560 .extra2
= &init_net
.ipv4
.frags
.high_thresh
563 .procname
= "ipfrag_time",
564 .data
= &init_net
.ipv4
.frags
.timeout
,
565 .maxlen
= sizeof(int),
567 .proc_handler
= proc_dointvec_jiffies
,
570 .procname
= "ipfrag_max_dist",
571 .data
= &init_net
.ipv4
.frags
.max_dist
,
572 .maxlen
= sizeof(int),
574 .proc_handler
= proc_dointvec_minmax
,
580 /* secret interval has been deprecated */
581 static int ip4_frags_secret_interval_unused
;
582 static struct ctl_table ip4_frags_ctl_table
[] = {
584 .procname
= "ipfrag_secret_interval",
585 .data
= &ip4_frags_secret_interval_unused
,
586 .maxlen
= sizeof(int),
588 .proc_handler
= proc_dointvec_jiffies
,
593 static int __net_init
ip4_frags_ns_ctl_register(struct net
*net
)
595 struct ctl_table
*table
;
596 struct ctl_table_header
*hdr
;
598 table
= ip4_frags_ns_ctl_table
;
599 if (!net_eq(net
, &init_net
)) {
600 table
= kmemdup(table
, sizeof(ip4_frags_ns_ctl_table
), GFP_KERNEL
);
604 table
[0].data
= &net
->ipv4
.frags
.high_thresh
;
605 table
[0].extra1
= &net
->ipv4
.frags
.low_thresh
;
606 table
[0].extra2
= &init_net
.ipv4
.frags
.high_thresh
;
607 table
[1].data
= &net
->ipv4
.frags
.low_thresh
;
608 table
[1].extra2
= &net
->ipv4
.frags
.high_thresh
;
609 table
[2].data
= &net
->ipv4
.frags
.timeout
;
610 table
[3].data
= &net
->ipv4
.frags
.max_dist
;
613 hdr
= register_net_sysctl(net
, "net/ipv4", table
);
617 net
->ipv4
.frags_hdr
= hdr
;
621 if (!net_eq(net
, &init_net
))
627 static void __net_exit
ip4_frags_ns_ctl_unregister(struct net
*net
)
629 struct ctl_table
*table
;
631 table
= net
->ipv4
.frags_hdr
->ctl_table_arg
;
632 unregister_net_sysctl_table(net
->ipv4
.frags_hdr
);
636 static void __init
ip4_frags_ctl_register(void)
638 register_net_sysctl(&init_net
, "net/ipv4", ip4_frags_ctl_table
);
641 static int ip4_frags_ns_ctl_register(struct net
*net
)
646 static void ip4_frags_ns_ctl_unregister(struct net
*net
)
650 static void __init
ip4_frags_ctl_register(void)
655 static int __net_init
ipv4_frags_init_net(struct net
*net
)
659 /* Fragment cache limits.
661 * The fragment memory accounting code, (tries to) account for
662 * the real memory usage, by measuring both the size of frag
663 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
664 * and the SKB's truesize.
666 * A 64K fragment consumes 129736 bytes (44*2944)+200
667 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
669 * We will commit 4MB at one time. Should we cross that limit
670 * we will prune down to 3MB, making room for approx 8 big 64K
673 net
->ipv4
.frags
.high_thresh
= 4 * 1024 * 1024;
674 net
->ipv4
.frags
.low_thresh
= 3 * 1024 * 1024;
676 * Important NOTE! Fragment queue must be destroyed before MSL expires.
677 * RFC791 is wrong proposing to prolongate timer each fragment arrival
680 net
->ipv4
.frags
.timeout
= IP_FRAG_TIME
;
682 net
->ipv4
.frags
.max_dist
= 64;
683 net
->ipv4
.frags
.f
= &ip4_frags
;
685 res
= inet_frags_init_net(&net
->ipv4
.frags
);
688 res
= ip4_frags_ns_ctl_register(net
);
690 inet_frags_exit_net(&net
->ipv4
.frags
);
694 static void __net_exit
ipv4_frags_exit_net(struct net
*net
)
696 ip4_frags_ns_ctl_unregister(net
);
697 inet_frags_exit_net(&net
->ipv4
.frags
);
700 static struct pernet_operations ip4_frags_ops
= {
701 .init
= ipv4_frags_init_net
,
702 .exit
= ipv4_frags_exit_net
,
706 static u32
ip4_key_hashfn(const void *data
, u32 len
, u32 seed
)
709 sizeof(struct frag_v4_compare_key
) / sizeof(u32
), seed
);
712 static u32
ip4_obj_hashfn(const void *data
, u32 len
, u32 seed
)
714 const struct inet_frag_queue
*fq
= data
;
716 return jhash2((const u32
*)&fq
->key
.v4
,
717 sizeof(struct frag_v4_compare_key
) / sizeof(u32
), seed
);
720 static int ip4_obj_cmpfn(struct rhashtable_compare_arg
*arg
, const void *ptr
)
722 const struct frag_v4_compare_key
*key
= arg
->key
;
723 const struct inet_frag_queue
*fq
= ptr
;
725 return !!memcmp(&fq
->key
, key
, sizeof(*key
));
728 static const struct rhashtable_params ip4_rhash_params
= {
729 .head_offset
= offsetof(struct inet_frag_queue
, node
),
730 .key_offset
= offsetof(struct inet_frag_queue
, key
),
731 .key_len
= sizeof(struct frag_v4_compare_key
),
732 .hashfn
= ip4_key_hashfn
,
733 .obj_hashfn
= ip4_obj_hashfn
,
734 .obj_cmpfn
= ip4_obj_cmpfn
,
735 .automatic_shrinking
= true,
738 void __init
ipfrag_init(void)
740 ip4_frags
.constructor
= ip4_frag_init
;
741 ip4_frags
.destructor
= ip4_frag_free
;
742 ip4_frags
.qsize
= sizeof(struct ipq
);
743 ip4_frags
.frag_expire
= ip_expire
;
744 ip4_frags
.frags_cache_name
= ip_frag_cache_name
;
745 ip4_frags
.rhash_params
= ip4_rhash_params
;
746 if (inet_frags_init(&ip4_frags
))
747 panic("IP: failed to allocate ip4_frags cache\n");
748 ip4_frags_ctl_register();
749 register_pernet_subsys(&ip4_frags_ops
);