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";
61 struct inet_skb_parm h;
65 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
67 /* Describe an entry in the "incomplete datagrams" queue. */
69 struct inet_frag_queue q;
76 u8 ecn; /* RFC3168 support */
77 u16 max_df_size; /* largest frag with DF set seen */
79 int vif; /* L3 master device index */
81 struct inet_peer *peer;
84 static u8 ip4_frag_ecn(u8 tos)
86 return 1 << (tos & INET_ECN_MASK);
89 static struct inet_frags ip4_frags;
91 int ip_frag_mem(struct net *net)
93 return sum_frag_mem_limit(&net->ipv4.frags);
96 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
97 struct net_device *dev);
99 struct ip4_create_arg {
105 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
107 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
108 return jhash_3words((__force u32)id << 16 | prot,
109 (__force u32)saddr, (__force u32)daddr,
113 static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
115 const struct ipq *ipq;
117 ipq = container_of(q, struct ipq, q);
118 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
121 static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
123 const struct ipq *qp;
124 const struct ip4_create_arg *arg = a;
126 qp = container_of(q, struct ipq, q);
127 return qp->id == arg->iph->id &&
128 qp->saddr == arg->iph->saddr &&
129 qp->daddr == arg->iph->daddr &&
130 qp->protocol == arg->iph->protocol &&
131 qp->user == arg->user &&
135 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
137 struct ipq *qp = container_of(q, struct ipq, q);
138 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
140 struct net *net = container_of(ipv4, struct net, ipv4);
142 const struct ip4_create_arg *arg = a;
144 qp->protocol = arg->iph->protocol;
145 qp->id = arg->iph->id;
146 qp->ecn = ip4_frag_ecn(arg->iph->tos);
147 qp->saddr = arg->iph->saddr;
148 qp->daddr = arg->iph->daddr;
150 qp->user = arg->user;
151 qp->peer = q->net->max_dist ?
152 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
156 static void ip4_frag_free(struct inet_frag_queue *q)
160 qp = container_of(q, struct ipq, q);
162 inet_putpeer(qp->peer);
166 /* Destruction primitives. */
168 static void ipq_put(struct ipq *ipq)
170 inet_frag_put(&ipq->q, &ip4_frags);
173 /* Kill ipq entry. It is not destroyed immediately,
174 * because caller (and someone more) holds reference count.
176 static void ipq_kill(struct ipq *ipq)
178 inet_frag_kill(&ipq->q, &ip4_frags);
181 static bool frag_expire_skip_icmp(u32 user)
183 return user == IP_DEFRAG_AF_PACKET ||
184 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
185 __IP_DEFRAG_CONNTRACK_IN_END) ||
186 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
187 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
191 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
193 static void ip_expire(unsigned long arg)
198 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
199 net = container_of(qp->q.net, struct net, ipv4.frags);
202 spin_lock(&qp->q.lock);
204 if (qp->q.flags & INET_FRAG_COMPLETE)
208 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
210 if (!inet_frag_evicting(&qp->q)) {
211 struct sk_buff *clone, *head = qp->q.fragments;
212 const struct iphdr *iph;
215 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
217 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
220 head->dev = dev_get_by_index_rcu(net, qp->iif);
225 /* skb has no dst, perform route lookup again */
227 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
228 iph->tos, head->dev);
232 /* Only an end host needs to send an ICMP
233 * "Fragment Reassembly Timeout" message, per RFC792.
235 if (frag_expire_skip_icmp(qp->user) &&
236 (skb_rtable(head)->rt_type != RTN_LOCAL))
239 clone = skb_clone(head, GFP_ATOMIC);
241 /* Send an ICMP "Fragment Reassembly Timeout" message. */
243 spin_unlock(&qp->q.lock);
244 icmp_send(clone, ICMP_TIME_EXCEEDED,
245 ICMP_EXC_FRAGTIME, 0);
251 spin_unlock(&qp->q.lock);
257 /* Find the correct entry in the "incomplete datagrams" queue for
258 * this IP datagram, and create new one, if nothing is found.
260 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
263 struct inet_frag_queue *q;
264 struct ip4_create_arg arg;
271 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
273 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
274 if (IS_ERR_OR_NULL(q)) {
275 inet_frag_maybe_warn_overflow(q, pr_fmt());
278 return container_of(q, struct ipq, q);
281 /* Is the fragment too far ahead to be part of ipq? */
282 static int ip_frag_too_far(struct ipq *qp)
284 struct inet_peer *peer = qp->peer;
285 unsigned int max = qp->q.net->max_dist;
286 unsigned int start, end;
294 end = atomic_inc_return(&peer->rid);
297 rc = qp->q.fragments && (end - start) > max;
302 net = container_of(qp->q.net, struct net, ipv4.frags);
303 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
309 static int ip_frag_reinit(struct ipq *qp)
312 unsigned int sum_truesize = 0;
314 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
315 atomic_inc(&qp->q.refcnt);
319 fp = qp->q.fragments;
321 struct sk_buff *xp = fp->next;
323 sum_truesize += fp->truesize;
327 sub_frag_mem_limit(qp->q.net, sum_truesize);
332 qp->q.fragments = NULL;
333 qp->q.fragments_tail = NULL;
340 /* Add new segment to existing queue. */
341 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
343 struct sk_buff *prev, *next;
344 struct net_device *dev;
345 unsigned int fragsize;
351 if (qp->q.flags & INET_FRAG_COMPLETE)
354 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
355 unlikely(ip_frag_too_far(qp)) &&
356 unlikely(err = ip_frag_reinit(qp))) {
361 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
362 offset = ntohs(ip_hdr(skb)->frag_off);
363 flags = offset & ~IP_OFFSET;
365 offset <<= 3; /* offset is in 8-byte chunks */
366 ihl = ip_hdrlen(skb);
368 /* Determine the position of this fragment. */
369 end = offset + skb->len - skb_network_offset(skb) - ihl;
372 /* Is this the final fragment? */
373 if ((flags & IP_MF) == 0) {
374 /* If we already have some bits beyond end
375 * or have different end, the segment is corrupted.
377 if (end < qp->q.len ||
378 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
380 qp->q.flags |= INET_FRAG_LAST_IN;
385 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
386 skb->ip_summed = CHECKSUM_NONE;
388 if (end > qp->q.len) {
389 /* Some bits beyond end -> corruption. */
390 if (qp->q.flags & INET_FRAG_LAST_IN)
399 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
402 err = pskb_trim_rcsum(skb, end - offset);
406 /* Find out which fragments are in front and at the back of us
407 * in the chain of fragments so far. We must know where to put
408 * this fragment, right?
410 prev = qp->q.fragments_tail;
411 if (!prev || FRAG_CB(prev)->offset < offset) {
416 for (next = qp->q.fragments; next != NULL; next = next->next) {
417 if (FRAG_CB(next)->offset >= offset)
423 /* We found where to put this one. Check for overlap with
424 * preceding fragment, and, if needed, align things so that
425 * any overlaps are eliminated.
428 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
436 if (!pskb_pull(skb, i))
438 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
439 skb->ip_summed = CHECKSUM_NONE;
445 while (next && FRAG_CB(next)->offset < end) {
446 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
449 /* Eat head of the next overlapped fragment
450 * and leave the loop. The next ones cannot overlap.
452 if (!pskb_pull(next, i))
454 FRAG_CB(next)->offset += i;
456 if (next->ip_summed != CHECKSUM_UNNECESSARY)
457 next->ip_summed = CHECKSUM_NONE;
460 struct sk_buff *free_it = next;
462 /* Old fragment is completely overridden with
470 qp->q.fragments = next;
472 qp->q.meat -= free_it->len;
473 sub_frag_mem_limit(qp->q.net, free_it->truesize);
478 FRAG_CB(skb)->offset = offset;
480 /* Insert this fragment in the chain of fragments. */
483 qp->q.fragments_tail = skb;
487 qp->q.fragments = skb;
491 qp->iif = dev->ifindex;
494 qp->q.stamp = skb->tstamp;
495 qp->q.meat += skb->len;
497 add_frag_mem_limit(qp->q.net, skb->truesize);
499 qp->q.flags |= INET_FRAG_FIRST_IN;
501 fragsize = skb->len + ihl;
503 if (fragsize > qp->q.max_size)
504 qp->q.max_size = fragsize;
506 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
507 fragsize > qp->max_df_size)
508 qp->max_df_size = fragsize;
510 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
511 qp->q.meat == qp->q.len) {
512 unsigned long orefdst = skb->_skb_refdst;
514 skb->_skb_refdst = 0UL;
515 err = ip_frag_reasm(qp, prev, dev);
516 skb->_skb_refdst = orefdst;
529 /* Build a new IP datagram from all its fragments. */
531 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
532 struct net_device *dev)
534 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
536 struct sk_buff *fp, *head = qp->q.fragments;
544 ecn = ip_frag_ecn_table[qp->ecn];
545 if (unlikely(ecn == 0xff)) {
549 /* Make the one we just received the head. */
552 fp = skb_clone(head, GFP_ATOMIC);
556 fp->next = head->next;
558 qp->q.fragments_tail = fp;
561 skb_morph(head, qp->q.fragments);
562 head->next = qp->q.fragments->next;
564 consume_skb(qp->q.fragments);
565 qp->q.fragments = head;
569 WARN_ON(FRAG_CB(head)->offset != 0);
571 /* Allocate a new buffer for the datagram. */
572 ihlen = ip_hdrlen(head);
573 len = ihlen + qp->q.len;
579 /* Head of list must not be cloned. */
580 if (skb_unclone(head, GFP_ATOMIC))
583 /* If the first fragment is fragmented itself, we split
584 * it to two chunks: the first with data and paged part
585 * and the second, holding only fragments. */
586 if (skb_has_frag_list(head)) {
587 struct sk_buff *clone;
590 clone = alloc_skb(0, GFP_ATOMIC);
593 clone->next = head->next;
595 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
596 skb_frag_list_init(head);
597 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
598 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
599 clone->len = clone->data_len = head->data_len - plen;
600 head->data_len -= clone->len;
601 head->len -= clone->len;
603 clone->ip_summed = head->ip_summed;
604 add_frag_mem_limit(qp->q.net, clone->truesize);
607 skb_shinfo(head)->frag_list = head->next;
608 skb_push(head, head->data - skb_network_header(head));
610 for (fp=head->next; fp; fp = fp->next) {
611 head->data_len += fp->len;
612 head->len += fp->len;
613 if (head->ip_summed != fp->ip_summed)
614 head->ip_summed = CHECKSUM_NONE;
615 else if (head->ip_summed == CHECKSUM_COMPLETE)
616 head->csum = csum_add(head->csum, fp->csum);
617 head->truesize += fp->truesize;
619 sub_frag_mem_limit(qp->q.net, head->truesize);
623 head->tstamp = qp->q.stamp;
624 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
627 iph->tot_len = htons(len);
630 /* When we set IP_DF on a refragmented skb we must also force a
631 * call to ip_fragment to avoid forwarding a DF-skb of size s while
632 * original sender only sent fragments of size f (where f < s).
634 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
635 * frag seen to avoid sending tiny DF-fragments in case skb was built
636 * from one very small df-fragment and one large non-df frag.
638 if (qp->max_df_size == qp->q.max_size) {
639 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
640 iph->frag_off = htons(IP_DF);
647 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
648 qp->q.fragments = NULL;
649 qp->q.fragments_tail = NULL;
653 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
657 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
659 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
663 /* Process an incoming IP datagram fragment. */
664 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
666 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
667 int vif = l3mdev_master_ifindex_rcu(dev);
670 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
673 /* Lookup (or create) queue header */
674 qp = ip_find(net, ip_hdr(skb), user, vif);
678 spin_lock(&qp->q.lock);
680 ret = ip_frag_queue(qp, skb);
682 spin_unlock(&qp->q.lock);
687 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
691 EXPORT_SYMBOL(ip_defrag);
693 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
699 if (skb->protocol != htons(ETH_P_IP))
702 netoff = skb_network_offset(skb);
704 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
707 if (iph.ihl < 5 || iph.version != 4)
710 len = ntohs(iph.tot_len);
711 if (skb->len < netoff + len || len < (iph.ihl * 4))
714 if (ip_is_fragment(&iph)) {
715 skb = skb_share_check(skb, GFP_ATOMIC);
717 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
719 if (pskb_trim_rcsum(skb, netoff + len))
721 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
722 if (ip_defrag(net, skb, user))
729 EXPORT_SYMBOL(ip_check_defrag);
734 static struct ctl_table ip4_frags_ns_ctl_table[] = {
736 .procname = "ipfrag_high_thresh",
737 .data = &init_net.ipv4.frags.high_thresh,
738 .maxlen = sizeof(int),
740 .proc_handler = proc_dointvec_minmax,
741 .extra1 = &init_net.ipv4.frags.low_thresh
744 .procname = "ipfrag_low_thresh",
745 .data = &init_net.ipv4.frags.low_thresh,
746 .maxlen = sizeof(int),
748 .proc_handler = proc_dointvec_minmax,
750 .extra2 = &init_net.ipv4.frags.high_thresh
753 .procname = "ipfrag_time",
754 .data = &init_net.ipv4.frags.timeout,
755 .maxlen = sizeof(int),
757 .proc_handler = proc_dointvec_jiffies,
760 .procname = "ipfrag_max_dist",
761 .data = &init_net.ipv4.frags.max_dist,
762 .maxlen = sizeof(int),
764 .proc_handler = proc_dointvec_minmax,
770 /* secret interval has been deprecated */
771 static int ip4_frags_secret_interval_unused;
772 static struct ctl_table ip4_frags_ctl_table[] = {
774 .procname = "ipfrag_secret_interval",
775 .data = &ip4_frags_secret_interval_unused,
776 .maxlen = sizeof(int),
778 .proc_handler = proc_dointvec_jiffies,
783 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
785 struct ctl_table *table;
786 struct ctl_table_header *hdr;
788 table = ip4_frags_ns_ctl_table;
789 if (!net_eq(net, &init_net)) {
790 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
794 table[0].data = &net->ipv4.frags.high_thresh;
795 table[0].extra1 = &net->ipv4.frags.low_thresh;
796 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
797 table[1].data = &net->ipv4.frags.low_thresh;
798 table[1].extra2 = &net->ipv4.frags.high_thresh;
799 table[2].data = &net->ipv4.frags.timeout;
800 table[3].data = &net->ipv4.frags.max_dist;
803 hdr = register_net_sysctl(net, "net/ipv4", table);
807 net->ipv4.frags_hdr = hdr;
811 if (!net_eq(net, &init_net))
817 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
819 struct ctl_table *table;
821 table = net->ipv4.frags_hdr->ctl_table_arg;
822 unregister_net_sysctl_table(net->ipv4.frags_hdr);
826 static void __init ip4_frags_ctl_register(void)
828 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
831 static int ip4_frags_ns_ctl_register(struct net *net)
836 static void ip4_frags_ns_ctl_unregister(struct net *net)
840 static void __init ip4_frags_ctl_register(void)
845 static int __net_init ipv4_frags_init_net(struct net *net)
849 /* Fragment cache limits.
851 * The fragment memory accounting code, (tries to) account for
852 * the real memory usage, by measuring both the size of frag
853 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
854 * and the SKB's truesize.
856 * A 64K fragment consumes 129736 bytes (44*2944)+200
857 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
859 * We will commit 4MB at one time. Should we cross that limit
860 * we will prune down to 3MB, making room for approx 8 big 64K
863 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
864 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
866 * Important NOTE! Fragment queue must be destroyed before MSL expires.
867 * RFC791 is wrong proposing to prolongate timer each fragment arrival
870 net->ipv4.frags.timeout = IP_FRAG_TIME;
872 net->ipv4.frags.max_dist = 64;
874 res = inet_frags_init_net(&net->ipv4.frags);
877 res = ip4_frags_ns_ctl_register(net);
879 inet_frags_uninit_net(&net->ipv4.frags);
883 static void __net_exit ipv4_frags_exit_net(struct net *net)
885 ip4_frags_ns_ctl_unregister(net);
886 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
889 static struct pernet_operations ip4_frags_ops = {
890 .init = ipv4_frags_init_net,
891 .exit = ipv4_frags_exit_net,
894 void __init ipfrag_init(void)
896 ip4_frags_ctl_register();
897 register_pernet_subsys(&ip4_frags_ops);
898 ip4_frags.hashfn = ip4_hashfn;
899 ip4_frags.constructor = ip4_frag_init;
900 ip4_frags.destructor = ip4_frag_free;
901 ip4_frags.qsize = sizeof(struct ipq);
902 ip4_frags.match = ip4_frag_match;
903 ip4_frags.frag_expire = ip_expire;
904 ip4_frags.frags_cache_name = ip_frag_cache_name;
905 if (inet_frags_init(&ip4_frags))
906 panic("IP: failed to allocate ip4_frags cache\n");