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;
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 */
79 struct inet_peer *peer;
82 static inline u8 ip4_frag_ecn(u8 tos)
84 return 1 << (tos & INET_ECN_MASK);
87 static struct inet_frags ip4_frags;
89 int ip_frag_nqueues(struct net *net)
91 return net->ipv4.frags.nqueues;
94 int ip_frag_mem(struct net *net)
96 return sum_frag_mem_limit(&net->ipv4.frags);
99 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
100 struct net_device *dev);
102 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,
112 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
115 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
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(struct inet_frag_queue *q, void *a)
126 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;
136 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
138 struct ipq *qp = container_of(q, struct ipq, q);
139 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
141 struct net *net = container_of(ipv4, struct net, ipv4);
143 struct ip4_create_arg *arg = a;
145 qp->protocol = arg->iph->protocol;
146 qp->id = arg->iph->id;
147 qp->ecn = ip4_frag_ecn(arg->iph->tos);
148 qp->saddr = arg->iph->saddr;
149 qp->daddr = arg->iph->daddr;
150 qp->user = arg->user;
151 qp->peer = sysctl_ipfrag_max_dist ?
152 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
155 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
159 qp = container_of(q, struct ipq, q);
161 inet_putpeer(qp->peer);
165 /* Destruction primitives. */
167 static __inline__ void ipq_put(struct ipq *ipq)
169 inet_frag_put(&ipq->q, &ip4_frags);
172 /* Kill ipq entry. It is not destroyed immediately,
173 * because caller (and someone more) holds reference count.
175 static void ipq_kill(struct ipq *ipq)
177 inet_frag_kill(&ipq->q, &ip4_frags);
180 /* Memory limiting on fragments. Evictor trashes the oldest
181 * fragment queue until we are back under the threshold.
183 static void ip_evictor(struct net *net)
187 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
189 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
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);
203 spin_lock(&qp->q.lock);
205 if (qp->q.last_in & INET_FRAG_COMPLETE)
210 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
211 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
213 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
214 struct sk_buff *head = qp->q.fragments;
215 const struct iphdr *iph;
219 head->dev = dev_get_by_index_rcu(net, qp->iif);
223 /* skb has no dst, perform route lookup again */
225 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
226 iph->tos, head->dev);
231 * Only an end host needs to send an ICMP
232 * "Fragment Reassembly Timeout" message, per RFC792.
234 if (qp->user == IP_DEFRAG_AF_PACKET ||
235 ((qp->user >= IP_DEFRAG_CONNTRACK_IN) &&
236 (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) &&
237 (skb_rtable(head)->rt_type != RTN_LOCAL)))
241 /* Send an ICMP "Fragment Reassembly Timeout" message. */
242 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
247 spin_unlock(&qp->q.lock);
251 /* Find the correct entry in the "incomplete datagrams" queue for
252 * this IP datagram, and create new one, if nothing is found.
254 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
256 struct inet_frag_queue *q;
257 struct ip4_create_arg arg;
263 read_lock(&ip4_frags.lock);
264 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
266 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
267 if (IS_ERR_OR_NULL(q)) {
268 inet_frag_maybe_warn_overflow(q, pr_fmt());
271 return container_of(q, struct ipq, q);
274 /* Is the fragment too far ahead to be part of ipq? */
275 static inline int ip_frag_too_far(struct ipq *qp)
277 struct inet_peer *peer = qp->peer;
278 unsigned int max = sysctl_ipfrag_max_dist;
279 unsigned int start, end;
287 end = atomic_inc_return(&peer->rid);
290 rc = qp->q.fragments && (end - start) > max;
295 net = container_of(qp->q.net, struct net, ipv4.frags);
296 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
302 static int ip_frag_reinit(struct ipq *qp)
305 unsigned int sum_truesize = 0;
307 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
308 atomic_inc(&qp->q.refcnt);
312 fp = qp->q.fragments;
314 struct sk_buff *xp = fp->next;
316 sum_truesize += fp->truesize;
320 sub_frag_mem_limit(&qp->q, sum_truesize);
325 qp->q.fragments = NULL;
326 qp->q.fragments_tail = NULL;
333 /* Add new segment to existing queue. */
334 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
336 struct sk_buff *prev, *next;
337 struct net_device *dev;
343 if (qp->q.last_in & INET_FRAG_COMPLETE)
346 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
347 unlikely(ip_frag_too_far(qp)) &&
348 unlikely(err = ip_frag_reinit(qp))) {
353 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
354 offset = ntohs(ip_hdr(skb)->frag_off);
355 flags = offset & ~IP_OFFSET;
357 offset <<= 3; /* offset is in 8-byte chunks */
358 ihl = ip_hdrlen(skb);
360 /* Determine the position of this fragment. */
361 end = offset + skb->len - ihl;
364 /* Is this the final fragment? */
365 if ((flags & IP_MF) == 0) {
366 /* If we already have some bits beyond end
367 * or have different end, the segment is corrupted.
369 if (end < qp->q.len ||
370 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
372 qp->q.last_in |= INET_FRAG_LAST_IN;
377 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
378 skb->ip_summed = CHECKSUM_NONE;
380 if (end > qp->q.len) {
381 /* Some bits beyond end -> corruption. */
382 if (qp->q.last_in & INET_FRAG_LAST_IN)
391 if (pskb_pull(skb, ihl) == NULL)
394 err = pskb_trim_rcsum(skb, end - offset);
398 /* Find out which fragments are in front and at the back of us
399 * in the chain of fragments so far. We must know where to put
400 * this fragment, right?
402 prev = qp->q.fragments_tail;
403 if (!prev || FRAG_CB(prev)->offset < offset) {
408 for (next = qp->q.fragments; next != NULL; next = next->next) {
409 if (FRAG_CB(next)->offset >= offset)
415 /* We found where to put this one. Check for overlap with
416 * preceding fragment, and, if needed, align things so that
417 * any overlaps are eliminated.
420 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
428 if (!pskb_pull(skb, i))
430 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
431 skb->ip_summed = CHECKSUM_NONE;
437 while (next && FRAG_CB(next)->offset < end) {
438 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
441 /* Eat head of the next overlapped fragment
442 * and leave the loop. The next ones cannot overlap.
444 if (!pskb_pull(next, i))
446 FRAG_CB(next)->offset += i;
448 if (next->ip_summed != CHECKSUM_UNNECESSARY)
449 next->ip_summed = CHECKSUM_NONE;
452 struct sk_buff *free_it = next;
454 /* Old fragment is completely overridden with
462 qp->q.fragments = next;
464 qp->q.meat -= free_it->len;
465 sub_frag_mem_limit(&qp->q, free_it->truesize);
470 FRAG_CB(skb)->offset = offset;
472 /* Insert this fragment in the chain of fragments. */
475 qp->q.fragments_tail = skb;
479 qp->q.fragments = skb;
483 qp->iif = dev->ifindex;
486 qp->q.stamp = skb->tstamp;
487 qp->q.meat += skb->len;
489 add_frag_mem_limit(&qp->q, skb->truesize);
491 qp->q.last_in |= INET_FRAG_FIRST_IN;
493 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
494 skb->len + ihl > qp->q.max_size)
495 qp->q.max_size = skb->len + ihl;
497 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
498 qp->q.meat == qp->q.len) {
499 unsigned long orefdst = skb->_skb_refdst;
501 skb->_skb_refdst = 0UL;
502 err = ip_frag_reasm(qp, prev, dev);
503 skb->_skb_refdst = orefdst;
508 inet_frag_lru_move(&qp->q);
517 /* Build a new IP datagram from all its fragments. */
519 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
520 struct net_device *dev)
522 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
524 struct sk_buff *fp, *head = qp->q.fragments;
533 ecn = ip_frag_ecn_table[qp->ecn];
534 if (unlikely(ecn == 0xff)) {
538 /* Make the one we just received the head. */
541 fp = skb_clone(head, GFP_ATOMIC);
545 fp->next = head->next;
547 qp->q.fragments_tail = fp;
550 skb_morph(head, qp->q.fragments);
551 head->next = qp->q.fragments->next;
553 consume_skb(qp->q.fragments);
554 qp->q.fragments = head;
557 WARN_ON(head == NULL);
558 WARN_ON(FRAG_CB(head)->offset != 0);
560 /* Allocate a new buffer for the datagram. */
561 ihlen = ip_hdrlen(head);
562 len = ihlen + qp->q.len;
568 /* Head of list must not be cloned. */
569 if (skb_unclone(head, GFP_ATOMIC))
572 /* If the first fragment is fragmented itself, we split
573 * it to two chunks: the first with data and paged part
574 * and the second, holding only fragments. */
575 if (skb_has_frag_list(head)) {
576 struct sk_buff *clone;
579 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
581 clone->next = head->next;
583 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
584 skb_frag_list_init(head);
585 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
586 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
587 clone->len = clone->data_len = head->data_len - plen;
588 head->data_len -= clone->len;
589 head->len -= clone->len;
591 clone->ip_summed = head->ip_summed;
592 add_frag_mem_limit(&qp->q, clone->truesize);
595 skb_push(head, head->data - skb_network_header(head));
597 sum_truesize = head->truesize;
598 for (fp = head->next; fp;) {
601 struct sk_buff *next = fp->next;
603 sum_truesize += fp->truesize;
604 if (head->ip_summed != fp->ip_summed)
605 head->ip_summed = CHECKSUM_NONE;
606 else if (head->ip_summed == CHECKSUM_COMPLETE)
607 head->csum = csum_add(head->csum, fp->csum);
609 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
610 kfree_skb_partial(fp, headstolen);
612 if (!skb_shinfo(head)->frag_list)
613 skb_shinfo(head)->frag_list = fp;
614 head->data_len += fp->len;
615 head->len += fp->len;
616 head->truesize += fp->truesize;
620 sub_frag_mem_limit(&qp->q, sum_truesize);
624 head->tstamp = qp->q.stamp;
625 IPCB(head)->frag_max_size = qp->q.max_size;
628 /* max_size != 0 implies at least one fragment had IP_DF set */
629 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
630 iph->tot_len = htons(len);
632 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
633 qp->q.fragments = NULL;
634 qp->q.fragments_tail = NULL;
638 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
643 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
645 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
649 /* Process an incoming IP datagram fragment. */
650 int ip_defrag(struct sk_buff *skb, u32 user)
655 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
656 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
658 /* Start by cleaning up the memory. */
661 /* Lookup (or create) queue header */
662 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
665 spin_lock(&qp->q.lock);
667 ret = ip_frag_queue(qp, skb);
669 spin_unlock(&qp->q.lock);
674 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
678 EXPORT_SYMBOL(ip_defrag);
680 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
685 if (skb->protocol != htons(ETH_P_IP))
688 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
691 if (iph.ihl < 5 || iph.version != 4)
694 len = ntohs(iph.tot_len);
695 if (skb->len < len || len < (iph.ihl * 4))
698 if (ip_is_fragment(&iph)) {
699 skb = skb_share_check(skb, GFP_ATOMIC);
701 if (!pskb_may_pull(skb, iph.ihl*4))
703 if (pskb_trim_rcsum(skb, len))
705 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
706 if (ip_defrag(skb, user))
713 EXPORT_SYMBOL(ip_check_defrag);
718 static struct ctl_table ip4_frags_ns_ctl_table[] = {
720 .procname = "ipfrag_high_thresh",
721 .data = &init_net.ipv4.frags.high_thresh,
722 .maxlen = sizeof(int),
724 .proc_handler = proc_dointvec
727 .procname = "ipfrag_low_thresh",
728 .data = &init_net.ipv4.frags.low_thresh,
729 .maxlen = sizeof(int),
731 .proc_handler = proc_dointvec
734 .procname = "ipfrag_time",
735 .data = &init_net.ipv4.frags.timeout,
736 .maxlen = sizeof(int),
738 .proc_handler = proc_dointvec_jiffies,
743 static struct ctl_table ip4_frags_ctl_table[] = {
745 .procname = "ipfrag_secret_interval",
746 .data = &ip4_frags.secret_interval,
747 .maxlen = sizeof(int),
749 .proc_handler = proc_dointvec_jiffies,
752 .procname = "ipfrag_max_dist",
753 .data = &sysctl_ipfrag_max_dist,
754 .maxlen = sizeof(int),
756 .proc_handler = proc_dointvec_minmax,
762 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
764 struct ctl_table *table;
765 struct ctl_table_header *hdr;
767 table = ip4_frags_ns_ctl_table;
768 if (!net_eq(net, &init_net)) {
769 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
773 table[0].data = &net->ipv4.frags.high_thresh;
774 table[1].data = &net->ipv4.frags.low_thresh;
775 table[2].data = &net->ipv4.frags.timeout;
777 /* Don't export sysctls to unprivileged users */
778 if (net->user_ns != &init_user_ns)
779 table[0].procname = NULL;
782 hdr = register_net_sysctl(net, "net/ipv4", table);
786 net->ipv4.frags_hdr = hdr;
790 if (!net_eq(net, &init_net))
796 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
798 struct ctl_table *table;
800 table = net->ipv4.frags_hdr->ctl_table_arg;
801 unregister_net_sysctl_table(net->ipv4.frags_hdr);
805 static void ip4_frags_ctl_register(void)
807 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
810 static inline int ip4_frags_ns_ctl_register(struct net *net)
815 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
819 static inline void ip4_frags_ctl_register(void)
824 static int __net_init ipv4_frags_init_net(struct net *net)
826 /* Fragment cache limits.
828 * The fragment memory accounting code, (tries to) account for
829 * the real memory usage, by measuring both the size of frag
830 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
831 * and the SKB's truesize.
833 * A 64K fragment consumes 129736 bytes (44*2944)+200
834 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
836 * We will commit 4MB at one time. Should we cross that limit
837 * we will prune down to 3MB, making room for approx 8 big 64K
840 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
841 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
843 * Important NOTE! Fragment queue must be destroyed before MSL expires.
844 * RFC791 is wrong proposing to prolongate timer each fragment arrival
847 net->ipv4.frags.timeout = IP_FRAG_TIME;
849 inet_frags_init_net(&net->ipv4.frags);
851 return ip4_frags_ns_ctl_register(net);
854 static void __net_exit ipv4_frags_exit_net(struct net *net)
856 ip4_frags_ns_ctl_unregister(net);
857 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
860 static struct pernet_operations ip4_frags_ops = {
861 .init = ipv4_frags_init_net,
862 .exit = ipv4_frags_exit_net,
865 void __init ipfrag_init(void)
867 ip4_frags_ctl_register();
868 register_pernet_subsys(&ip4_frags_ops);
869 ip4_frags.hashfn = ip4_hashfn;
870 ip4_frags.constructor = ip4_frag_init;
871 ip4_frags.destructor = ip4_frag_free;
872 ip4_frags.skb_free = NULL;
873 ip4_frags.qsize = sizeof(struct ipq);
874 ip4_frags.match = ip4_frag_match;
875 ip4_frags.frag_expire = ip_expire;
876 ip4_frags.secret_interval = 10 * 60 * HZ;
877 inet_frags_init(&ip4_frags);