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;
83 * We want to check ECN values of all fragments, do detect invalid combinations.
84 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
86 #define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
87 #define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
88 #define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
89 #define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
91 static inline u8 ip4_frag_ecn(u8 tos)
93 return 1 << (tos & INET_ECN_MASK);
96 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
97 * Value : 0xff if frame should be dropped.
98 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
100 static const u8 ip4_frag_ecn_table[16] = {
101 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
103 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
104 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
106 /* invalid combinations : drop frame */
107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
112 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
113 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
116 static struct inet_frags ip4_frags;
118 int ip_frag_nqueues(struct net *net)
120 return net->ipv4.frags.nqueues;
123 int ip_frag_mem(struct net *net)
125 return atomic_read(&net->ipv4.frags.mem);
128 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
129 struct net_device *dev);
131 struct ip4_create_arg {
136 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
138 return jhash_3words((__force u32)id << 16 | prot,
139 (__force u32)saddr, (__force u32)daddr,
140 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
143 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
147 ipq = container_of(q, struct ipq, q);
148 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
151 static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
154 struct ip4_create_arg *arg = a;
156 qp = container_of(q, struct ipq, q);
157 return qp->id == arg->iph->id &&
158 qp->saddr == arg->iph->saddr &&
159 qp->daddr == arg->iph->daddr &&
160 qp->protocol == arg->iph->protocol &&
161 qp->user == arg->user;
164 /* Memory Tracking Functions. */
165 static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
167 atomic_sub(skb->truesize, &nf->mem);
171 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
173 struct ipq *qp = container_of(q, struct ipq, q);
174 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
176 struct net *net = container_of(ipv4, struct net, ipv4);
178 struct ip4_create_arg *arg = a;
180 qp->protocol = arg->iph->protocol;
181 qp->id = arg->iph->id;
182 qp->ecn = ip4_frag_ecn(arg->iph->tos);
183 qp->saddr = arg->iph->saddr;
184 qp->daddr = arg->iph->daddr;
185 qp->user = arg->user;
186 qp->peer = sysctl_ipfrag_max_dist ?
187 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
190 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
194 qp = container_of(q, struct ipq, q);
196 inet_putpeer(qp->peer);
200 /* Destruction primitives. */
202 static __inline__ void ipq_put(struct ipq *ipq)
204 inet_frag_put(&ipq->q, &ip4_frags);
207 /* Kill ipq entry. It is not destroyed immediately,
208 * because caller (and someone more) holds reference count.
210 static void ipq_kill(struct ipq *ipq)
212 inet_frag_kill(&ipq->q, &ip4_frags);
215 /* Memory limiting on fragments. Evictor trashes the oldest
216 * fragment queue until we are back under the threshold.
218 static void ip_evictor(struct net *net)
222 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
224 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
228 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
230 static void ip_expire(unsigned long arg)
235 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
236 net = container_of(qp->q.net, struct net, ipv4.frags);
238 spin_lock(&qp->q.lock);
240 if (qp->q.last_in & INET_FRAG_COMPLETE)
245 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
246 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
248 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
249 struct sk_buff *head = qp->q.fragments;
250 const struct iphdr *iph;
254 head->dev = dev_get_by_index_rcu(net, qp->iif);
258 /* skb dst is stale, drop it, and perform route lookup again */
261 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
262 iph->tos, head->dev);
267 * Only an end host needs to send an ICMP
268 * "Fragment Reassembly Timeout" message, per RFC792.
270 if (qp->user == IP_DEFRAG_AF_PACKET ||
271 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
272 skb_rtable(head)->rt_type != RTN_LOCAL))
276 /* Send an ICMP "Fragment Reassembly Timeout" message. */
277 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
282 spin_unlock(&qp->q.lock);
286 /* Find the correct entry in the "incomplete datagrams" queue for
287 * this IP datagram, and create new one, if nothing is found.
289 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
291 struct inet_frag_queue *q;
292 struct ip4_create_arg arg;
298 read_lock(&ip4_frags.lock);
299 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
301 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
305 return container_of(q, struct ipq, q);
308 LIMIT_NETDEBUG(KERN_ERR pr_fmt("ip_frag_create: no memory left !\n"));
312 /* Is the fragment too far ahead to be part of ipq? */
313 static inline int ip_frag_too_far(struct ipq *qp)
315 struct inet_peer *peer = qp->peer;
316 unsigned int max = sysctl_ipfrag_max_dist;
317 unsigned int start, end;
325 end = atomic_inc_return(&peer->rid);
328 rc = qp->q.fragments && (end - start) > max;
333 net = container_of(qp->q.net, struct net, ipv4.frags);
334 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
340 static int ip_frag_reinit(struct ipq *qp)
344 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
345 atomic_inc(&qp->q.refcnt);
349 fp = qp->q.fragments;
351 struct sk_buff *xp = fp->next;
352 frag_kfree_skb(qp->q.net, fp);
359 qp->q.fragments = NULL;
360 qp->q.fragments_tail = NULL;
367 /* Add new segment to existing queue. */
368 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
370 struct sk_buff *prev, *next;
371 struct net_device *dev;
377 if (qp->q.last_in & INET_FRAG_COMPLETE)
380 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
381 unlikely(ip_frag_too_far(qp)) &&
382 unlikely(err = ip_frag_reinit(qp))) {
387 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
388 offset = ntohs(ip_hdr(skb)->frag_off);
389 flags = offset & ~IP_OFFSET;
391 offset <<= 3; /* offset is in 8-byte chunks */
392 ihl = ip_hdrlen(skb);
394 /* Determine the position of this fragment. */
395 end = offset + skb->len - ihl;
398 /* Is this the final fragment? */
399 if ((flags & IP_MF) == 0) {
400 /* If we already have some bits beyond end
401 * or have different end, the segment is corrupted.
403 if (end < qp->q.len ||
404 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
406 qp->q.last_in |= INET_FRAG_LAST_IN;
411 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
412 skb->ip_summed = CHECKSUM_NONE;
414 if (end > qp->q.len) {
415 /* Some bits beyond end -> corruption. */
416 if (qp->q.last_in & INET_FRAG_LAST_IN)
425 if (pskb_pull(skb, ihl) == NULL)
428 err = pskb_trim_rcsum(skb, end - offset);
432 /* Find out which fragments are in front and at the back of us
433 * in the chain of fragments so far. We must know where to put
434 * this fragment, right?
436 prev = qp->q.fragments_tail;
437 if (!prev || FRAG_CB(prev)->offset < offset) {
442 for (next = qp->q.fragments; next != NULL; next = next->next) {
443 if (FRAG_CB(next)->offset >= offset)
449 /* We found where to put this one. Check for overlap with
450 * preceding fragment, and, if needed, align things so that
451 * any overlaps are eliminated.
454 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
462 if (!pskb_pull(skb, i))
464 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
465 skb->ip_summed = CHECKSUM_NONE;
471 while (next && FRAG_CB(next)->offset < end) {
472 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
475 /* Eat head of the next overlapped fragment
476 * and leave the loop. The next ones cannot overlap.
478 if (!pskb_pull(next, i))
480 FRAG_CB(next)->offset += i;
482 if (next->ip_summed != CHECKSUM_UNNECESSARY)
483 next->ip_summed = CHECKSUM_NONE;
486 struct sk_buff *free_it = next;
488 /* Old fragment is completely overridden with
496 qp->q.fragments = next;
498 qp->q.meat -= free_it->len;
499 frag_kfree_skb(qp->q.net, free_it);
503 FRAG_CB(skb)->offset = offset;
505 /* Insert this fragment in the chain of fragments. */
508 qp->q.fragments_tail = skb;
512 qp->q.fragments = skb;
516 qp->iif = dev->ifindex;
519 qp->q.stamp = skb->tstamp;
520 qp->q.meat += skb->len;
522 atomic_add(skb->truesize, &qp->q.net->mem);
524 qp->q.last_in |= INET_FRAG_FIRST_IN;
526 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
527 skb->len + ihl > qp->q.max_size)
528 qp->q.max_size = skb->len + ihl;
530 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
531 qp->q.meat == qp->q.len)
532 return ip_frag_reasm(qp, prev, dev);
534 write_lock(&ip4_frags.lock);
535 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
536 write_unlock(&ip4_frags.lock);
545 /* Build a new IP datagram from all its fragments. */
547 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
548 struct net_device *dev)
550 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
552 struct sk_buff *fp, *head = qp->q.fragments;
561 ecn = ip4_frag_ecn_table[qp->ecn];
562 if (unlikely(ecn == 0xff)) {
566 /* Make the one we just received the head. */
569 fp = skb_clone(head, GFP_ATOMIC);
573 fp->next = head->next;
575 qp->q.fragments_tail = fp;
578 skb_morph(head, qp->q.fragments);
579 head->next = qp->q.fragments->next;
581 consume_skb(qp->q.fragments);
582 qp->q.fragments = head;
585 WARN_ON(head == NULL);
586 WARN_ON(FRAG_CB(head)->offset != 0);
588 /* Allocate a new buffer for the datagram. */
589 ihlen = ip_hdrlen(head);
590 len = ihlen + qp->q.len;
596 /* Head of list must not be cloned. */
597 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
600 /* If the first fragment is fragmented itself, we split
601 * it to two chunks: the first with data and paged part
602 * and the second, holding only fragments. */
603 if (skb_has_frag_list(head)) {
604 struct sk_buff *clone;
607 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
609 clone->next = head->next;
611 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
612 skb_frag_list_init(head);
613 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
614 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
615 clone->len = clone->data_len = head->data_len - plen;
616 head->data_len -= clone->len;
617 head->len -= clone->len;
619 clone->ip_summed = head->ip_summed;
620 atomic_add(clone->truesize, &qp->q.net->mem);
623 skb_push(head, head->data - skb_network_header(head));
625 sum_truesize = head->truesize;
626 for (fp = head->next; fp;) {
629 struct sk_buff *next = fp->next;
631 sum_truesize += fp->truesize;
632 if (head->ip_summed != fp->ip_summed)
633 head->ip_summed = CHECKSUM_NONE;
634 else if (head->ip_summed == CHECKSUM_COMPLETE)
635 head->csum = csum_add(head->csum, fp->csum);
637 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
638 kfree_skb_partial(fp, headstolen);
640 if (!skb_shinfo(head)->frag_list)
641 skb_shinfo(head)->frag_list = fp;
642 head->data_len += fp->len;
643 head->len += fp->len;
644 head->truesize += fp->truesize;
648 atomic_sub(sum_truesize, &qp->q.net->mem);
652 head->tstamp = qp->q.stamp;
653 IPCB(head)->frag_max_size = qp->q.max_size;
656 /* max_size != 0 implies at least one fragment had IP_DF set */
657 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
658 iph->tot_len = htons(len);
660 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
661 qp->q.fragments = NULL;
662 qp->q.fragments_tail = NULL;
666 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
671 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
673 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
677 /* Process an incoming IP datagram fragment. */
678 int ip_defrag(struct sk_buff *skb, u32 user)
683 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
684 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
686 /* Start by cleaning up the memory. */
689 /* Lookup (or create) queue header */
690 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
693 spin_lock(&qp->q.lock);
695 ret = ip_frag_queue(qp, skb);
697 spin_unlock(&qp->q.lock);
702 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
706 EXPORT_SYMBOL(ip_defrag);
708 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
713 if (skb->protocol != htons(ETH_P_IP))
716 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
719 if (iph.ihl < 5 || iph.version != 4)
722 len = ntohs(iph.tot_len);
723 if (skb->len < len || len < (iph.ihl * 4))
726 if (ip_is_fragment(&iph)) {
727 skb = skb_share_check(skb, GFP_ATOMIC);
729 if (!pskb_may_pull(skb, iph.ihl*4))
731 if (pskb_trim_rcsum(skb, len))
733 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
734 if (ip_defrag(skb, user))
741 EXPORT_SYMBOL(ip_check_defrag);
746 static struct ctl_table ip4_frags_ns_ctl_table[] = {
748 .procname = "ipfrag_high_thresh",
749 .data = &init_net.ipv4.frags.high_thresh,
750 .maxlen = sizeof(int),
752 .proc_handler = proc_dointvec
755 .procname = "ipfrag_low_thresh",
756 .data = &init_net.ipv4.frags.low_thresh,
757 .maxlen = sizeof(int),
759 .proc_handler = proc_dointvec
762 .procname = "ipfrag_time",
763 .data = &init_net.ipv4.frags.timeout,
764 .maxlen = sizeof(int),
766 .proc_handler = proc_dointvec_jiffies,
771 static struct ctl_table ip4_frags_ctl_table[] = {
773 .procname = "ipfrag_secret_interval",
774 .data = &ip4_frags.secret_interval,
775 .maxlen = sizeof(int),
777 .proc_handler = proc_dointvec_jiffies,
780 .procname = "ipfrag_max_dist",
781 .data = &sysctl_ipfrag_max_dist,
782 .maxlen = sizeof(int),
784 .proc_handler = proc_dointvec_minmax,
790 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
792 struct ctl_table *table;
793 struct ctl_table_header *hdr;
795 table = ip4_frags_ns_ctl_table;
796 if (!net_eq(net, &init_net)) {
797 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
801 table[0].data = &net->ipv4.frags.high_thresh;
802 table[1].data = &net->ipv4.frags.low_thresh;
803 table[2].data = &net->ipv4.frags.timeout;
805 /* Don't export sysctls to unprivileged users */
806 if (net->user_ns != &init_user_ns)
807 table[0].procname = NULL;
810 hdr = register_net_sysctl(net, "net/ipv4", table);
814 net->ipv4.frags_hdr = hdr;
818 if (!net_eq(net, &init_net))
824 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
826 struct ctl_table *table;
828 table = net->ipv4.frags_hdr->ctl_table_arg;
829 unregister_net_sysctl_table(net->ipv4.frags_hdr);
833 static void ip4_frags_ctl_register(void)
835 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
838 static inline int ip4_frags_ns_ctl_register(struct net *net)
843 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
847 static inline void ip4_frags_ctl_register(void)
852 static int __net_init ipv4_frags_init_net(struct net *net)
855 * Fragment cache limits. We will commit 256K at one time. Should we
856 * cross that limit we will prune down to 192K. This should cope with
857 * even the most extreme cases without allowing an attacker to
858 * measurably harm machine performance.
860 net->ipv4.frags.high_thresh = 256 * 1024;
861 net->ipv4.frags.low_thresh = 192 * 1024;
863 * Important NOTE! Fragment queue must be destroyed before MSL expires.
864 * RFC791 is wrong proposing to prolongate timer each fragment arrival
867 net->ipv4.frags.timeout = IP_FRAG_TIME;
869 inet_frags_init_net(&net->ipv4.frags);
871 return ip4_frags_ns_ctl_register(net);
874 static void __net_exit ipv4_frags_exit_net(struct net *net)
876 ip4_frags_ns_ctl_unregister(net);
877 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
880 static struct pernet_operations ip4_frags_ops = {
881 .init = ipv4_frags_init_net,
882 .exit = ipv4_frags_exit_net,
885 void __init ipfrag_init(void)
887 ip4_frags_ctl_register();
888 register_pernet_subsys(&ip4_frags_ops);
889 ip4_frags.hashfn = ip4_hashfn;
890 ip4_frags.constructor = ip4_frag_init;
891 ip4_frags.destructor = ip4_frag_free;
892 ip4_frags.skb_free = NULL;
893 ip4_frags.qsize = sizeof(struct ipq);
894 ip4_frags.match = ip4_frag_match;
895 ip4_frags.frag_expire = ip_expire;
896 ip4_frags.secret_interval = 10 * 60 * HZ;
897 inet_frags_init(&ip4_frags);