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
58 static int sysctl_ipfrag_max_dist __read_mostly = 64;
59 static const char ip_frag_cache_name[] = "ip4-frags";
63 struct inet_skb_parm h;
67 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
69 /* Describe an entry in the "incomplete datagrams" queue. */
71 struct inet_frag_queue q;
78 u8 ecn; /* RFC3168 support */
79 u16 max_df_size; /* largest frag with DF set seen */
81 int vif; /* L3 master device index */
83 struct inet_peer *peer;
86 static u8 ip4_frag_ecn(u8 tos)
88 return 1 << (tos & INET_ECN_MASK);
91 static struct inet_frags ip4_frags;
93 int ip_frag_mem(struct net *net)
95 return sum_frag_mem_limit(&net->ipv4.frags);
98 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
99 struct net_device *dev);
101 struct ip4_create_arg {
107 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
109 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
110 return jhash_3words((__force u32)id << 16 | prot,
111 (__force u32)saddr, (__force u32)daddr,
115 static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
117 const struct ipq *ipq;
119 ipq = container_of(q, struct ipq, q);
120 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
123 static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
125 const struct ipq *qp;
126 const struct ip4_create_arg *arg = a;
128 qp = container_of(q, struct ipq, q);
129 return qp->id == arg->iph->id &&
130 qp->saddr == arg->iph->saddr &&
131 qp->daddr == arg->iph->daddr &&
132 qp->protocol == arg->iph->protocol &&
133 qp->user == arg->user &&
137 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
139 struct ipq *qp = container_of(q, struct ipq, q);
140 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
142 struct net *net = container_of(ipv4, struct net, ipv4);
144 const struct ip4_create_arg *arg = a;
146 qp->protocol = arg->iph->protocol;
147 qp->id = arg->iph->id;
148 qp->ecn = ip4_frag_ecn(arg->iph->tos);
149 qp->saddr = arg->iph->saddr;
150 qp->daddr = arg->iph->daddr;
152 qp->user = arg->user;
153 qp->peer = sysctl_ipfrag_max_dist ?
154 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, arg->vif, 1) :
158 static void ip4_frag_free(struct inet_frag_queue *q)
162 qp = container_of(q, struct ipq, q);
164 inet_putpeer(qp->peer);
168 /* Destruction primitives. */
170 static void ipq_put(struct ipq *ipq)
172 inet_frag_put(&ipq->q, &ip4_frags);
175 /* Kill ipq entry. It is not destroyed immediately,
176 * because caller (and someone more) holds reference count.
178 static void ipq_kill(struct ipq *ipq)
180 inet_frag_kill(&ipq->q, &ip4_frags);
183 static bool frag_expire_skip_icmp(u32 user)
185 return user == IP_DEFRAG_AF_PACKET ||
186 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
187 __IP_DEFRAG_CONNTRACK_IN_END) ||
188 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
189 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
193 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
195 static void ip_expire(unsigned long arg)
200 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
201 net = container_of(qp->q.net, struct net, ipv4.frags);
204 spin_lock(&qp->q.lock);
206 if (qp->q.flags & INET_FRAG_COMPLETE)
210 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
212 if (!inet_frag_evicting(&qp->q)) {
213 struct sk_buff *clone, *head = qp->q.fragments;
214 const struct iphdr *iph;
217 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
219 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
222 head->dev = dev_get_by_index_rcu(net, qp->iif);
227 /* skb has no dst, perform route lookup again */
229 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
230 iph->tos, head->dev);
234 /* Only an end host needs to send an ICMP
235 * "Fragment Reassembly Timeout" message, per RFC792.
237 if (frag_expire_skip_icmp(qp->user) &&
238 (skb_rtable(head)->rt_type != RTN_LOCAL))
241 clone = skb_clone(head, GFP_ATOMIC);
243 /* Send an ICMP "Fragment Reassembly Timeout" message. */
245 spin_unlock(&qp->q.lock);
246 icmp_send(clone, ICMP_TIME_EXCEEDED,
247 ICMP_EXC_FRAGTIME, 0);
253 spin_unlock(&qp->q.lock);
259 /* Find the correct entry in the "incomplete datagrams" queue for
260 * this IP datagram, and create new one, if nothing is found.
262 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
265 struct inet_frag_queue *q;
266 struct ip4_create_arg arg;
273 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
275 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
276 if (IS_ERR_OR_NULL(q)) {
277 inet_frag_maybe_warn_overflow(q, pr_fmt());
280 return container_of(q, struct ipq, q);
283 /* Is the fragment too far ahead to be part of ipq? */
284 static int ip_frag_too_far(struct ipq *qp)
286 struct inet_peer *peer = qp->peer;
287 unsigned int max = sysctl_ipfrag_max_dist;
288 unsigned int start, end;
296 end = atomic_inc_return(&peer->rid);
299 rc = qp->q.fragments && (end - start) > max;
304 net = container_of(qp->q.net, struct net, ipv4.frags);
305 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
311 static int ip_frag_reinit(struct ipq *qp)
314 unsigned int sum_truesize = 0;
316 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
317 atomic_inc(&qp->q.refcnt);
321 fp = qp->q.fragments;
323 struct sk_buff *xp = fp->next;
325 sum_truesize += fp->truesize;
329 sub_frag_mem_limit(qp->q.net, sum_truesize);
334 qp->q.fragments = NULL;
335 qp->q.fragments_tail = NULL;
342 /* Add new segment to existing queue. */
343 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
345 struct sk_buff *prev, *next;
346 struct net_device *dev;
347 unsigned int fragsize;
353 if (qp->q.flags & INET_FRAG_COMPLETE)
356 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
357 unlikely(ip_frag_too_far(qp)) &&
358 unlikely(err = ip_frag_reinit(qp))) {
363 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
364 offset = ntohs(ip_hdr(skb)->frag_off);
365 flags = offset & ~IP_OFFSET;
367 offset <<= 3; /* offset is in 8-byte chunks */
368 ihl = ip_hdrlen(skb);
370 /* Determine the position of this fragment. */
371 end = offset + skb->len - skb_network_offset(skb) - ihl;
374 /* Is this the final fragment? */
375 if ((flags & IP_MF) == 0) {
376 /* If we already have some bits beyond end
377 * or have different end, the segment is corrupted.
379 if (end < qp->q.len ||
380 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
382 qp->q.flags |= INET_FRAG_LAST_IN;
387 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
388 skb->ip_summed = CHECKSUM_NONE;
390 if (end > qp->q.len) {
391 /* Some bits beyond end -> corruption. */
392 if (qp->q.flags & INET_FRAG_LAST_IN)
401 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
404 err = pskb_trim_rcsum(skb, end - offset);
408 /* Find out which fragments are in front and at the back of us
409 * in the chain of fragments so far. We must know where to put
410 * this fragment, right?
412 prev = qp->q.fragments_tail;
413 if (!prev || FRAG_CB(prev)->offset < offset) {
418 for (next = qp->q.fragments; next != NULL; next = next->next) {
419 if (FRAG_CB(next)->offset >= offset)
425 /* We found where to put this one. Check for overlap with
426 * preceding fragment, and, if needed, align things so that
427 * any overlaps are eliminated.
430 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
438 if (!pskb_pull(skb, i))
440 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
441 skb->ip_summed = CHECKSUM_NONE;
447 while (next && FRAG_CB(next)->offset < end) {
448 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
451 /* Eat head of the next overlapped fragment
452 * and leave the loop. The next ones cannot overlap.
454 if (!pskb_pull(next, i))
456 FRAG_CB(next)->offset += i;
458 if (next->ip_summed != CHECKSUM_UNNECESSARY)
459 next->ip_summed = CHECKSUM_NONE;
462 struct sk_buff *free_it = next;
464 /* Old fragment is completely overridden with
472 qp->q.fragments = next;
474 qp->q.meat -= free_it->len;
475 sub_frag_mem_limit(qp->q.net, free_it->truesize);
480 FRAG_CB(skb)->offset = offset;
482 /* Insert this fragment in the chain of fragments. */
485 qp->q.fragments_tail = skb;
489 qp->q.fragments = skb;
493 qp->iif = dev->ifindex;
496 qp->q.stamp = skb->tstamp;
497 qp->q.meat += skb->len;
499 add_frag_mem_limit(qp->q.net, skb->truesize);
501 qp->q.flags |= INET_FRAG_FIRST_IN;
503 fragsize = skb->len + ihl;
505 if (fragsize > qp->q.max_size)
506 qp->q.max_size = fragsize;
508 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
509 fragsize > qp->max_df_size)
510 qp->max_df_size = fragsize;
512 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
513 qp->q.meat == qp->q.len) {
514 unsigned long orefdst = skb->_skb_refdst;
516 skb->_skb_refdst = 0UL;
517 err = ip_frag_reasm(qp, prev, dev);
518 skb->_skb_refdst = orefdst;
531 /* Build a new IP datagram from all its fragments. */
533 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
534 struct net_device *dev)
536 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
538 struct sk_buff *fp, *head = qp->q.fragments;
546 ecn = ip_frag_ecn_table[qp->ecn];
547 if (unlikely(ecn == 0xff)) {
551 /* Make the one we just received the head. */
554 fp = skb_clone(head, GFP_ATOMIC);
558 fp->next = head->next;
560 qp->q.fragments_tail = fp;
563 skb_morph(head, qp->q.fragments);
564 head->next = qp->q.fragments->next;
566 consume_skb(qp->q.fragments);
567 qp->q.fragments = head;
571 WARN_ON(FRAG_CB(head)->offset != 0);
573 /* Allocate a new buffer for the datagram. */
574 ihlen = ip_hdrlen(head);
575 len = ihlen + qp->q.len;
581 /* Head of list must not be cloned. */
582 if (skb_unclone(head, GFP_ATOMIC))
585 /* If the first fragment is fragmented itself, we split
586 * it to two chunks: the first with data and paged part
587 * and the second, holding only fragments. */
588 if (skb_has_frag_list(head)) {
589 struct sk_buff *clone;
592 clone = alloc_skb(0, GFP_ATOMIC);
595 clone->next = head->next;
597 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
598 skb_frag_list_init(head);
599 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
600 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
601 clone->len = clone->data_len = head->data_len - plen;
602 head->data_len -= clone->len;
603 head->len -= clone->len;
605 clone->ip_summed = head->ip_summed;
606 add_frag_mem_limit(qp->q.net, clone->truesize);
609 skb_shinfo(head)->frag_list = head->next;
610 skb_push(head, head->data - skb_network_header(head));
612 for (fp=head->next; fp; fp = fp->next) {
613 head->data_len += fp->len;
614 head->len += fp->len;
615 if (head->ip_summed != fp->ip_summed)
616 head->ip_summed = CHECKSUM_NONE;
617 else if (head->ip_summed == CHECKSUM_COMPLETE)
618 head->csum = csum_add(head->csum, fp->csum);
619 head->truesize += fp->truesize;
621 sub_frag_mem_limit(qp->q.net, head->truesize);
625 head->tstamp = qp->q.stamp;
626 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
629 iph->tot_len = htons(len);
632 /* When we set IP_DF on a refragmented skb we must also force a
633 * call to ip_fragment to avoid forwarding a DF-skb of size s while
634 * original sender only sent fragments of size f (where f < s).
636 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
637 * frag seen to avoid sending tiny DF-fragments in case skb was built
638 * from one very small df-fragment and one large non-df frag.
640 if (qp->max_df_size == qp->q.max_size) {
641 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
642 iph->frag_off = htons(IP_DF);
649 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
650 qp->q.fragments = NULL;
651 qp->q.fragments_tail = NULL;
655 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
659 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
661 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
665 /* Process an incoming IP datagram fragment. */
666 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
668 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
669 int vif = l3mdev_master_ifindex_rcu(dev);
672 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
675 /* Lookup (or create) queue header */
676 qp = ip_find(net, ip_hdr(skb), user, vif);
680 spin_lock(&qp->q.lock);
682 ret = ip_frag_queue(qp, skb);
684 spin_unlock(&qp->q.lock);
689 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
693 EXPORT_SYMBOL(ip_defrag);
695 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
701 if (skb->protocol != htons(ETH_P_IP))
704 netoff = skb_network_offset(skb);
706 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
709 if (iph.ihl < 5 || iph.version != 4)
712 len = ntohs(iph.tot_len);
713 if (skb->len < netoff + len || len < (iph.ihl * 4))
716 if (ip_is_fragment(&iph)) {
717 skb = skb_share_check(skb, GFP_ATOMIC);
719 if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
723 if (pskb_trim_rcsum(skb, netoff + len)) {
727 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
728 if (ip_defrag(net, skb, user))
735 EXPORT_SYMBOL(ip_check_defrag);
740 static struct ctl_table ip4_frags_ns_ctl_table[] = {
742 .procname = "ipfrag_high_thresh",
743 .data = &init_net.ipv4.frags.high_thresh,
744 .maxlen = sizeof(int),
746 .proc_handler = proc_dointvec_minmax,
747 .extra1 = &init_net.ipv4.frags.low_thresh
750 .procname = "ipfrag_low_thresh",
751 .data = &init_net.ipv4.frags.low_thresh,
752 .maxlen = sizeof(int),
754 .proc_handler = proc_dointvec_minmax,
756 .extra2 = &init_net.ipv4.frags.high_thresh
759 .procname = "ipfrag_time",
760 .data = &init_net.ipv4.frags.timeout,
761 .maxlen = sizeof(int),
763 .proc_handler = proc_dointvec_jiffies,
768 /* secret interval has been deprecated */
769 static int ip4_frags_secret_interval_unused;
770 static struct ctl_table ip4_frags_ctl_table[] = {
772 .procname = "ipfrag_secret_interval",
773 .data = &ip4_frags_secret_interval_unused,
774 .maxlen = sizeof(int),
776 .proc_handler = proc_dointvec_jiffies,
779 .procname = "ipfrag_max_dist",
780 .data = &sysctl_ipfrag_max_dist,
781 .maxlen = sizeof(int),
783 .proc_handler = proc_dointvec_minmax,
789 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
791 struct ctl_table *table;
792 struct ctl_table_header *hdr;
794 table = ip4_frags_ns_ctl_table;
795 if (!net_eq(net, &init_net)) {
796 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
800 table[0].data = &net->ipv4.frags.high_thresh;
801 table[0].extra1 = &net->ipv4.frags.low_thresh;
802 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
803 table[1].data = &net->ipv4.frags.low_thresh;
804 table[1].extra2 = &net->ipv4.frags.high_thresh;
805 table[2].data = &net->ipv4.frags.timeout;
807 /* Don't export sysctls to unprivileged users */
808 if (net->user_ns != &init_user_ns)
809 table[0].procname = NULL;
812 hdr = register_net_sysctl(net, "net/ipv4", table);
816 net->ipv4.frags_hdr = hdr;
820 if (!net_eq(net, &init_net))
826 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
828 struct ctl_table *table;
830 table = net->ipv4.frags_hdr->ctl_table_arg;
831 unregister_net_sysctl_table(net->ipv4.frags_hdr);
835 static void __init ip4_frags_ctl_register(void)
837 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
840 static int ip4_frags_ns_ctl_register(struct net *net)
845 static void ip4_frags_ns_ctl_unregister(struct net *net)
849 static void __init ip4_frags_ctl_register(void)
854 static int __net_init ipv4_frags_init_net(struct net *net)
858 /* Fragment cache limits.
860 * The fragment memory accounting code, (tries to) account for
861 * the real memory usage, by measuring both the size of frag
862 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
863 * and the SKB's truesize.
865 * A 64K fragment consumes 129736 bytes (44*2944)+200
866 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
868 * We will commit 4MB at one time. Should we cross that limit
869 * we will prune down to 3MB, making room for approx 8 big 64K
872 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
873 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
875 * Important NOTE! Fragment queue must be destroyed before MSL expires.
876 * RFC791 is wrong proposing to prolongate timer each fragment arrival
879 net->ipv4.frags.timeout = IP_FRAG_TIME;
881 res = inet_frags_init_net(&net->ipv4.frags);
884 res = ip4_frags_ns_ctl_register(net);
886 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
890 static void __net_exit ipv4_frags_exit_net(struct net *net)
892 ip4_frags_ns_ctl_unregister(net);
893 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
896 static struct pernet_operations ip4_frags_ops = {
897 .init = ipv4_frags_init_net,
898 .exit = ipv4_frags_exit_net,
901 void __init ipfrag_init(void)
903 ip4_frags_ctl_register();
904 register_pernet_subsys(&ip4_frags_ops);
905 ip4_frags.hashfn = ip4_hashfn;
906 ip4_frags.constructor = ip4_frag_init;
907 ip4_frags.destructor = ip4_frag_free;
908 ip4_frags.skb_free = NULL;
909 ip4_frags.qsize = sizeof(struct ipq);
910 ip4_frags.match = ip4_frag_match;
911 ip4_frags.frag_expire = ip_expire;
912 ip4_frags.frags_cache_name = ip_frag_cache_name;
913 if (inet_frags_init(&ip4_frags))
914 panic("IP: failed to allocate ip4_frags cache\n");