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>
52 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
54 * as well. Or notify me, at least. --ANK
57 static int sysctl_ipfrag_max_dist __read_mostly = 64;
58 static const char ip_frag_cache_name[] = "ip4-frags";
62 struct inet_skb_parm h;
66 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
68 /* Describe an entry in the "incomplete datagrams" queue. */
70 struct inet_frag_queue q;
77 u8 ecn; /* RFC3168 support */
78 u16 max_df_size; /* largest frag with DF set seen */
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 {
104 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
106 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
107 return jhash_3words((__force u32)id << 16 | prot,
108 (__force u32)saddr, (__force u32)daddr,
112 static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
114 const struct ipq *ipq;
116 ipq = container_of(q, struct ipq, q);
117 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
120 static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
122 const struct ipq *qp;
123 const struct ip4_create_arg *arg = a;
125 qp = container_of(q, struct ipq, q);
126 return qp->id == arg->iph->id &&
127 qp->saddr == arg->iph->saddr &&
128 qp->daddr == arg->iph->daddr &&
129 qp->protocol == arg->iph->protocol &&
130 qp->user == arg->user;
133 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
135 struct ipq *qp = container_of(q, struct ipq, q);
136 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
138 struct net *net = container_of(ipv4, struct net, ipv4);
140 const struct ip4_create_arg *arg = a;
142 qp->protocol = arg->iph->protocol;
143 qp->id = arg->iph->id;
144 qp->ecn = ip4_frag_ecn(arg->iph->tos);
145 qp->saddr = arg->iph->saddr;
146 qp->daddr = arg->iph->daddr;
147 qp->user = arg->user;
148 qp->peer = sysctl_ipfrag_max_dist ?
149 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
152 static void ip4_frag_free(struct inet_frag_queue *q)
156 qp = container_of(q, struct ipq, q);
158 inet_putpeer(qp->peer);
162 /* Destruction primitives. */
164 static void ipq_put(struct ipq *ipq)
166 inet_frag_put(&ipq->q, &ip4_frags);
169 /* Kill ipq entry. It is not destroyed immediately,
170 * because caller (and someone more) holds reference count.
172 static void ipq_kill(struct ipq *ipq)
174 inet_frag_kill(&ipq->q, &ip4_frags);
177 static bool frag_expire_skip_icmp(u32 user)
179 return user == IP_DEFRAG_AF_PACKET ||
180 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
181 __IP_DEFRAG_CONNTRACK_IN_END) ||
182 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
183 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
187 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
189 static void ip_expire(unsigned long arg)
194 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
195 net = container_of(qp->q.net, struct net, ipv4.frags);
197 spin_lock(&qp->q.lock);
199 if (qp->q.flags & INET_FRAG_COMPLETE)
203 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
205 if (!inet_frag_evicting(&qp->q)) {
206 struct sk_buff *head = qp->q.fragments;
207 const struct iphdr *iph;
210 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
212 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
216 head->dev = dev_get_by_index_rcu(net, qp->iif);
220 /* skb has no dst, perform route lookup again */
222 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
223 iph->tos, head->dev);
227 /* Only an end host needs to send an ICMP
228 * "Fragment Reassembly Timeout" message, per RFC792.
230 if (frag_expire_skip_icmp(qp->user) &&
231 (skb_rtable(head)->rt_type != RTN_LOCAL))
234 /* Send an ICMP "Fragment Reassembly Timeout" message. */
235 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
240 spin_unlock(&qp->q.lock);
244 /* Find the correct entry in the "incomplete datagrams" queue for
245 * this IP datagram, and create new one, if nothing is found.
247 static struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
249 struct inet_frag_queue *q;
250 struct ip4_create_arg arg;
256 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
258 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
259 if (IS_ERR_OR_NULL(q)) {
260 inet_frag_maybe_warn_overflow(q, pr_fmt());
263 return container_of(q, struct ipq, q);
266 /* Is the fragment too far ahead to be part of ipq? */
267 static int ip_frag_too_far(struct ipq *qp)
269 struct inet_peer *peer = qp->peer;
270 unsigned int max = sysctl_ipfrag_max_dist;
271 unsigned int start, end;
279 end = atomic_inc_return(&peer->rid);
282 rc = qp->q.fragments && (end - start) > max;
287 net = container_of(qp->q.net, struct net, ipv4.frags);
288 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
294 static int ip_frag_reinit(struct ipq *qp)
297 unsigned int sum_truesize = 0;
299 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
300 atomic_inc(&qp->q.refcnt);
304 fp = qp->q.fragments;
306 struct sk_buff *xp = fp->next;
308 sum_truesize += fp->truesize;
312 sub_frag_mem_limit(qp->q.net, sum_truesize);
317 qp->q.fragments = NULL;
318 qp->q.fragments_tail = 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;
330 unsigned int fragsize;
336 if (qp->q.flags & INET_FRAG_COMPLETE)
339 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
340 unlikely(ip_frag_too_far(qp)) &&
341 unlikely(err = ip_frag_reinit(qp))) {
346 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
347 offset = ntohs(ip_hdr(skb)->frag_off);
348 flags = offset & ~IP_OFFSET;
350 offset <<= 3; /* offset is in 8-byte chunks */
351 ihl = ip_hdrlen(skb);
353 /* Determine the position of this fragment. */
354 end = offset + skb->len - skb_network_offset(skb) - ihl;
357 /* Is this the final fragment? */
358 if ((flags & IP_MF) == 0) {
359 /* If we already have some bits beyond end
360 * or have different end, the segment is corrupted.
362 if (end < qp->q.len ||
363 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
365 qp->q.flags |= INET_FRAG_LAST_IN;
370 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
371 skb->ip_summed = CHECKSUM_NONE;
373 if (end > qp->q.len) {
374 /* Some bits beyond end -> corruption. */
375 if (qp->q.flags & INET_FRAG_LAST_IN)
384 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
387 err = pskb_trim_rcsum(skb, end - offset);
391 /* Find out which fragments are in front and at the back of us
392 * in the chain of fragments so far. We must know where to put
393 * this fragment, right?
395 prev = qp->q.fragments_tail;
396 if (!prev || FRAG_CB(prev)->offset < offset) {
401 for (next = qp->q.fragments; next != NULL; next = next->next) {
402 if (FRAG_CB(next)->offset >= offset)
408 /* We found where to put this one. Check for overlap with
409 * preceding fragment, and, if needed, align things so that
410 * any overlaps are eliminated.
413 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
421 if (!pskb_pull(skb, i))
423 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
424 skb->ip_summed = CHECKSUM_NONE;
430 while (next && FRAG_CB(next)->offset < end) {
431 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
434 /* Eat head of the next overlapped fragment
435 * and leave the loop. The next ones cannot overlap.
437 if (!pskb_pull(next, i))
439 FRAG_CB(next)->offset += i;
441 if (next->ip_summed != CHECKSUM_UNNECESSARY)
442 next->ip_summed = CHECKSUM_NONE;
445 struct sk_buff *free_it = next;
447 /* Old fragment is completely overridden with
455 qp->q.fragments = next;
457 qp->q.meat -= free_it->len;
458 sub_frag_mem_limit(qp->q.net, free_it->truesize);
463 FRAG_CB(skb)->offset = offset;
465 /* Insert this fragment in the chain of fragments. */
468 qp->q.fragments_tail = skb;
472 qp->q.fragments = skb;
476 qp->iif = dev->ifindex;
479 qp->q.stamp = skb->tstamp;
480 qp->q.meat += skb->len;
482 add_frag_mem_limit(qp->q.net, skb->truesize);
484 qp->q.flags |= INET_FRAG_FIRST_IN;
486 fragsize = skb->len + ihl;
488 if (fragsize > qp->q.max_size)
489 qp->q.max_size = fragsize;
491 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
492 fragsize > qp->max_df_size)
493 qp->max_df_size = fragsize;
495 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
496 qp->q.meat == qp->q.len) {
497 unsigned long orefdst = skb->_skb_refdst;
499 skb->_skb_refdst = 0UL;
500 err = ip_frag_reasm(qp, prev, dev);
501 skb->_skb_refdst = orefdst;
514 /* Build a new IP datagram from all its fragments. */
516 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
517 struct net_device *dev)
519 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
521 struct sk_buff *fp, *head = qp->q.fragments;
530 ecn = ip_frag_ecn_table[qp->ecn];
531 if (unlikely(ecn == 0xff)) {
535 /* Make the one we just received the head. */
538 fp = skb_clone(head, GFP_ATOMIC);
542 fp->next = head->next;
544 qp->q.fragments_tail = fp;
547 skb_morph(head, qp->q.fragments);
548 head->next = qp->q.fragments->next;
550 consume_skb(qp->q.fragments);
551 qp->q.fragments = head;
555 WARN_ON(FRAG_CB(head)->offset != 0);
557 /* Allocate a new buffer for the datagram. */
558 ihlen = ip_hdrlen(head);
559 len = ihlen + qp->q.len;
565 /* Head of list must not be cloned. */
566 if (skb_unclone(head, GFP_ATOMIC))
569 /* If the first fragment is fragmented itself, we split
570 * it to two chunks: the first with data and paged part
571 * and the second, holding only fragments. */
572 if (skb_has_frag_list(head)) {
573 struct sk_buff *clone;
576 clone = alloc_skb(0, GFP_ATOMIC);
579 clone->next = head->next;
581 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
582 skb_frag_list_init(head);
583 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
584 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
585 clone->len = clone->data_len = head->data_len - plen;
586 head->data_len -= clone->len;
587 head->len -= clone->len;
589 clone->ip_summed = head->ip_summed;
590 add_frag_mem_limit(qp->q.net, clone->truesize);
593 skb_push(head, head->data - skb_network_header(head));
595 sum_truesize = head->truesize;
596 for (fp = head->next; fp;) {
599 struct sk_buff *next = fp->next;
601 sum_truesize += fp->truesize;
602 if (head->ip_summed != fp->ip_summed)
603 head->ip_summed = CHECKSUM_NONE;
604 else if (head->ip_summed == CHECKSUM_COMPLETE)
605 head->csum = csum_add(head->csum, fp->csum);
607 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
608 kfree_skb_partial(fp, headstolen);
610 if (!skb_shinfo(head)->frag_list)
611 skb_shinfo(head)->frag_list = fp;
612 head->data_len += fp->len;
613 head->len += fp->len;
614 head->truesize += fp->truesize;
618 sub_frag_mem_limit(qp->q.net, sum_truesize);
622 head->tstamp = qp->q.stamp;
623 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
626 iph->tot_len = htons(len);
629 /* When we set IP_DF on a refragmented skb we must also force a
630 * call to ip_fragment to avoid forwarding a DF-skb of size s while
631 * original sender only sent fragments of size f (where f < s).
633 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
634 * frag seen to avoid sending tiny DF-fragments in case skb was built
635 * from one very small df-fragment and one large non-df frag.
637 if (qp->max_df_size == qp->q.max_size) {
638 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
639 iph->frag_off = htons(IP_DF);
646 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
647 qp->q.fragments = NULL;
648 qp->q.fragments_tail = NULL;
652 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
656 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
658 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
662 /* Process an incoming IP datagram fragment. */
663 int ip_defrag(struct sk_buff *skb, u32 user)
668 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
669 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
671 /* Lookup (or create) queue header */
672 qp = ip_find(net, ip_hdr(skb), user);
676 spin_lock(&qp->q.lock);
678 ret = ip_frag_queue(qp, skb);
680 spin_unlock(&qp->q.lock);
685 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
689 EXPORT_SYMBOL(ip_defrag);
691 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
697 if (skb->protocol != htons(ETH_P_IP))
700 netoff = skb_network_offset(skb);
702 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
705 if (iph.ihl < 5 || iph.version != 4)
708 len = ntohs(iph.tot_len);
709 if (skb->len < netoff + len || len < (iph.ihl * 4))
712 if (ip_is_fragment(&iph)) {
713 skb = skb_share_check(skb, GFP_ATOMIC);
715 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
717 if (pskb_trim_rcsum(skb, netoff + len))
719 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
720 if (ip_defrag(skb, user))
727 EXPORT_SYMBOL(ip_check_defrag);
732 static struct ctl_table ip4_frags_ns_ctl_table[] = {
734 .procname = "ipfrag_high_thresh",
735 .data = &init_net.ipv4.frags.high_thresh,
736 .maxlen = sizeof(int),
738 .proc_handler = proc_dointvec_minmax,
739 .extra1 = &init_net.ipv4.frags.low_thresh
742 .procname = "ipfrag_low_thresh",
743 .data = &init_net.ipv4.frags.low_thresh,
744 .maxlen = sizeof(int),
746 .proc_handler = proc_dointvec_minmax,
748 .extra2 = &init_net.ipv4.frags.high_thresh
751 .procname = "ipfrag_time",
752 .data = &init_net.ipv4.frags.timeout,
753 .maxlen = sizeof(int),
755 .proc_handler = proc_dointvec_jiffies,
760 /* secret interval has been deprecated */
761 static int ip4_frags_secret_interval_unused;
762 static struct ctl_table ip4_frags_ctl_table[] = {
764 .procname = "ipfrag_secret_interval",
765 .data = &ip4_frags_secret_interval_unused,
766 .maxlen = sizeof(int),
768 .proc_handler = proc_dointvec_jiffies,
771 .procname = "ipfrag_max_dist",
772 .data = &sysctl_ipfrag_max_dist,
773 .maxlen = sizeof(int),
775 .proc_handler = proc_dointvec_minmax,
781 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
783 struct ctl_table *table;
784 struct ctl_table_header *hdr;
786 table = ip4_frags_ns_ctl_table;
787 if (!net_eq(net, &init_net)) {
788 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
792 table[0].data = &net->ipv4.frags.high_thresh;
793 table[0].extra1 = &net->ipv4.frags.low_thresh;
794 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
795 table[1].data = &net->ipv4.frags.low_thresh;
796 table[1].extra2 = &net->ipv4.frags.high_thresh;
797 table[2].data = &net->ipv4.frags.timeout;
799 /* Don't export sysctls to unprivileged users */
800 if (net->user_ns != &init_user_ns)
801 table[0].procname = NULL;
804 hdr = register_net_sysctl(net, "net/ipv4", table);
808 net->ipv4.frags_hdr = hdr;
812 if (!net_eq(net, &init_net))
818 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
820 struct ctl_table *table;
822 table = net->ipv4.frags_hdr->ctl_table_arg;
823 unregister_net_sysctl_table(net->ipv4.frags_hdr);
827 static void __init ip4_frags_ctl_register(void)
829 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
832 static int ip4_frags_ns_ctl_register(struct net *net)
837 static void ip4_frags_ns_ctl_unregister(struct net *net)
841 static void __init ip4_frags_ctl_register(void)
846 static int __net_init ipv4_frags_init_net(struct net *net)
848 /* Fragment cache limits.
850 * The fragment memory accounting code, (tries to) account for
851 * the real memory usage, by measuring both the size of frag
852 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
853 * and the SKB's truesize.
855 * A 64K fragment consumes 129736 bytes (44*2944)+200
856 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
858 * We will commit 4MB at one time. Should we cross that limit
859 * we will prune down to 3MB, making room for approx 8 big 64K
862 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
863 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
865 * Important NOTE! Fragment queue must be destroyed before MSL expires.
866 * RFC791 is wrong proposing to prolongate timer each fragment arrival
869 net->ipv4.frags.timeout = IP_FRAG_TIME;
871 inet_frags_init_net(&net->ipv4.frags);
873 return ip4_frags_ns_ctl_register(net);
876 static void __net_exit ipv4_frags_exit_net(struct net *net)
878 ip4_frags_ns_ctl_unregister(net);
879 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
882 static struct pernet_operations ip4_frags_ops = {
883 .init = ipv4_frags_init_net,
884 .exit = ipv4_frags_exit_net,
887 void __init ipfrag_init(void)
889 ip4_frags_ctl_register();
890 register_pernet_subsys(&ip4_frags_ops);
891 ip4_frags.hashfn = ip4_hashfn;
892 ip4_frags.constructor = ip4_frag_init;
893 ip4_frags.destructor = ip4_frag_free;
894 ip4_frags.skb_free = NULL;
895 ip4_frags.qsize = sizeof(struct ipq);
896 ip4_frags.match = ip4_frag_match;
897 ip4_frags.frag_expire = ip_expire;
898 ip4_frags.frags_cache_name = ip_frag_cache_name;
899 if (inet_frags_init(&ip4_frags))
900 panic("IP: failed to allocate ip4_frags cache\n");