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 #include <linux/compiler.h>
24 #include <linux/module.h>
25 #include <linux/types.h>
27 #include <linux/jiffies.h>
28 #include <linux/skbuff.h>
29 #include <linux/list.h>
31 #include <linux/icmp.h>
32 #include <linux/netdevice.h>
33 #include <linux/jhash.h>
34 #include <linux/random.h>
35 #include <linux/slab.h>
36 #include <net/route.h>
41 #include <net/checksum.h>
42 #include <net/inetpeer.h>
43 #include <net/inet_frag.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/inet.h>
47 #include <linux/netfilter_ipv4.h>
48 #include <net/inet_ecn.h>
50 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
51 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
52 * as well. Or notify me, at least. --ANK
55 static int sysctl_ipfrag_max_dist __read_mostly = 64;
59 struct inet_skb_parm h;
63 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
65 /* Describe an entry in the "incomplete datagrams" queue. */
67 struct inet_frag_queue q;
74 u8 ecn; /* RFC3168 support */
77 struct inet_peer *peer;
81 * We want to check ECN values of all fragments, do detect invalid combinations.
82 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
84 #define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
85 #define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
86 #define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
87 #define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
89 static inline u8 ip4_frag_ecn(u8 tos)
91 return 1 << (tos & INET_ECN_MASK);
94 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
95 * Value : 0xff if frame should be dropped.
96 * 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
98 static const u8 ip4_frag_ecn_table[16] = {
99 /* at least one fragment had CE, and others ECT_0 or ECT_1 */
100 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
101 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
102 [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
104 /* invalid combinations : drop frame */
105 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
106 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
107 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
108 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
109 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
110 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
111 [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
114 static struct inet_frags ip4_frags;
116 int ip_frag_nqueues(struct net *net)
118 return net->ipv4.frags.nqueues;
121 int ip_frag_mem(struct net *net)
123 return atomic_read(&net->ipv4.frags.mem);
126 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
127 struct net_device *dev);
129 struct ip4_create_arg {
134 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
136 return jhash_3words((__force u32)id << 16 | prot,
137 (__force u32)saddr, (__force u32)daddr,
138 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
141 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
145 ipq = container_of(q, struct ipq, q);
146 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
149 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
152 struct ip4_create_arg *arg = a;
154 qp = container_of(q, struct ipq, q);
155 return qp->id == arg->iph->id &&
156 qp->saddr == arg->iph->saddr &&
157 qp->daddr == arg->iph->daddr &&
158 qp->protocol == arg->iph->protocol &&
159 qp->user == arg->user;
162 /* Memory Tracking Functions. */
163 static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
165 atomic_sub(skb->truesize, &nf->mem);
169 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
171 struct ipq *qp = container_of(q, struct ipq, q);
172 struct ip4_create_arg *arg = a;
174 qp->protocol = arg->iph->protocol;
175 qp->id = arg->iph->id;
176 qp->ecn = ip4_frag_ecn(arg->iph->tos);
177 qp->saddr = arg->iph->saddr;
178 qp->daddr = arg->iph->daddr;
179 qp->user = arg->user;
180 qp->peer = sysctl_ipfrag_max_dist ?
181 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
184 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
188 qp = container_of(q, struct ipq, q);
190 inet_putpeer(qp->peer);
194 /* Destruction primitives. */
196 static __inline__ void ipq_put(struct ipq *ipq)
198 inet_frag_put(&ipq->q, &ip4_frags);
201 /* Kill ipq entry. It is not destroyed immediately,
202 * because caller (and someone more) holds reference count.
204 static void ipq_kill(struct ipq *ipq)
206 inet_frag_kill(&ipq->q, &ip4_frags);
209 /* Memory limiting on fragments. Evictor trashes the oldest
210 * fragment queue until we are back under the threshold.
212 static void ip_evictor(struct net *net)
216 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
218 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
222 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
224 static void ip_expire(unsigned long arg)
229 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
230 net = container_of(qp->q.net, struct net, ipv4.frags);
232 spin_lock(&qp->q.lock);
234 if (qp->q.last_in & INET_FRAG_COMPLETE)
239 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
240 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
242 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
243 struct sk_buff *head = qp->q.fragments;
244 const struct iphdr *iph;
248 head->dev = dev_get_by_index_rcu(net, qp->iif);
252 /* skb dst is stale, drop it, and perform route lookup again */
255 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
256 iph->tos, head->dev);
261 * Only an end host needs to send an ICMP
262 * "Fragment Reassembly Timeout" message, per RFC792.
264 if (qp->user == IP_DEFRAG_AF_PACKET ||
265 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
266 skb_rtable(head)->rt_type != RTN_LOCAL))
270 /* Send an ICMP "Fragment Reassembly Timeout" message. */
271 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
276 spin_unlock(&qp->q.lock);
280 /* Find the correct entry in the "incomplete datagrams" queue for
281 * this IP datagram, and create new one, if nothing is found.
283 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
285 struct inet_frag_queue *q;
286 struct ip4_create_arg arg;
292 read_lock(&ip4_frags.lock);
293 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
295 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
299 return container_of(q, struct ipq, q);
302 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
306 /* Is the fragment too far ahead to be part of ipq? */
307 static inline int ip_frag_too_far(struct ipq *qp)
309 struct inet_peer *peer = qp->peer;
310 unsigned int max = sysctl_ipfrag_max_dist;
311 unsigned int start, end;
319 end = atomic_inc_return(&peer->rid);
322 rc = qp->q.fragments && (end - start) > max;
327 net = container_of(qp->q.net, struct net, ipv4.frags);
328 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
334 static int ip_frag_reinit(struct ipq *qp)
338 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
339 atomic_inc(&qp->q.refcnt);
343 fp = qp->q.fragments;
345 struct sk_buff *xp = fp->next;
346 frag_kfree_skb(qp->q.net, fp);
353 qp->q.fragments = NULL;
354 qp->q.fragments_tail = NULL;
361 /* Add new segment to existing queue. */
362 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
364 struct sk_buff *prev, *next;
365 struct net_device *dev;
371 if (qp->q.last_in & INET_FRAG_COMPLETE)
374 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
375 unlikely(ip_frag_too_far(qp)) &&
376 unlikely(err = ip_frag_reinit(qp))) {
381 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
382 offset = ntohs(ip_hdr(skb)->frag_off);
383 flags = offset & ~IP_OFFSET;
385 offset <<= 3; /* offset is in 8-byte chunks */
386 ihl = ip_hdrlen(skb);
388 /* Determine the position of this fragment. */
389 end = offset + skb->len - ihl;
392 /* Is this the final fragment? */
393 if ((flags & IP_MF) == 0) {
394 /* If we already have some bits beyond end
395 * or have different end, the segment is corrupted.
397 if (end < qp->q.len ||
398 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
400 qp->q.last_in |= INET_FRAG_LAST_IN;
405 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
406 skb->ip_summed = CHECKSUM_NONE;
408 if (end > qp->q.len) {
409 /* Some bits beyond end -> corruption. */
410 if (qp->q.last_in & INET_FRAG_LAST_IN)
419 if (pskb_pull(skb, ihl) == NULL)
422 err = pskb_trim_rcsum(skb, end - offset);
426 /* Find out which fragments are in front and at the back of us
427 * in the chain of fragments so far. We must know where to put
428 * this fragment, right?
430 prev = qp->q.fragments_tail;
431 if (!prev || FRAG_CB(prev)->offset < offset) {
436 for (next = qp->q.fragments; next != NULL; next = next->next) {
437 if (FRAG_CB(next)->offset >= offset)
443 /* We found where to put this one. Check for overlap with
444 * preceding fragment, and, if needed, align things so that
445 * any overlaps are eliminated.
448 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
456 if (!pskb_pull(skb, i))
458 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
459 skb->ip_summed = CHECKSUM_NONE;
465 while (next && FRAG_CB(next)->offset < end) {
466 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
469 /* Eat head of the next overlapped fragment
470 * and leave the loop. The next ones cannot overlap.
472 if (!pskb_pull(next, i))
474 FRAG_CB(next)->offset += i;
476 if (next->ip_summed != CHECKSUM_UNNECESSARY)
477 next->ip_summed = CHECKSUM_NONE;
480 struct sk_buff *free_it = next;
482 /* Old fragment is completely overridden with
490 qp->q.fragments = next;
492 qp->q.meat -= free_it->len;
493 frag_kfree_skb(qp->q.net, free_it);
497 FRAG_CB(skb)->offset = offset;
499 /* Insert this fragment in the chain of fragments. */
502 qp->q.fragments_tail = skb;
506 qp->q.fragments = skb;
510 qp->iif = dev->ifindex;
513 qp->q.stamp = skb->tstamp;
514 qp->q.meat += skb->len;
516 atomic_add(skb->truesize, &qp->q.net->mem);
518 qp->q.last_in |= INET_FRAG_FIRST_IN;
520 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
521 qp->q.meat == qp->q.len)
522 return ip_frag_reasm(qp, prev, dev);
524 write_lock(&ip4_frags.lock);
525 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
526 write_unlock(&ip4_frags.lock);
535 /* Build a new IP datagram from all its fragments. */
537 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
538 struct net_device *dev)
540 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
542 struct sk_buff *fp, *head = qp->q.fragments;
550 ecn = ip4_frag_ecn_table[qp->ecn];
551 if (unlikely(ecn == 0xff)) {
555 /* Make the one we just received the head. */
558 fp = skb_clone(head, GFP_ATOMIC);
562 fp->next = head->next;
564 qp->q.fragments_tail = fp;
567 skb_morph(head, qp->q.fragments);
568 head->next = qp->q.fragments->next;
570 kfree_skb(qp->q.fragments);
571 qp->q.fragments = head;
574 WARN_ON(head == NULL);
575 WARN_ON(FRAG_CB(head)->offset != 0);
577 /* Allocate a new buffer for the datagram. */
578 ihlen = ip_hdrlen(head);
579 len = ihlen + qp->q.len;
585 /* Head of list must not be cloned. */
586 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
589 /* If the first fragment is fragmented itself, we split
590 * it to two chunks: the first with data and paged part
591 * and the second, holding only fragments. */
592 if (skb_has_frag_list(head)) {
593 struct sk_buff *clone;
596 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
598 clone->next = head->next;
600 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
601 skb_frag_list_init(head);
602 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
603 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
604 clone->len = clone->data_len = head->data_len - plen;
605 head->data_len -= clone->len;
606 head->len -= clone->len;
608 clone->ip_summed = head->ip_summed;
609 atomic_add(clone->truesize, &qp->q.net->mem);
612 skb_shinfo(head)->frag_list = head->next;
613 skb_push(head, head->data - skb_network_header(head));
615 for (fp=head->next; fp; fp = fp->next) {
616 head->data_len += fp->len;
617 head->len += fp->len;
618 if (head->ip_summed != fp->ip_summed)
619 head->ip_summed = CHECKSUM_NONE;
620 else if (head->ip_summed == CHECKSUM_COMPLETE)
621 head->csum = csum_add(head->csum, fp->csum);
622 head->truesize += fp->truesize;
624 atomic_sub(head->truesize, &qp->q.net->mem);
628 head->tstamp = qp->q.stamp;
632 iph->tot_len = htons(len);
634 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
635 qp->q.fragments = NULL;
636 qp->q.fragments_tail = NULL;
640 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
646 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
649 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
653 /* Process an incoming IP datagram fragment. */
654 int ip_defrag(struct sk_buff *skb, u32 user)
659 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
660 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
662 /* Start by cleaning up the memory. */
663 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
666 /* Lookup (or create) queue header */
667 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
670 spin_lock(&qp->q.lock);
672 ret = ip_frag_queue(qp, skb);
674 spin_unlock(&qp->q.lock);
679 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
683 EXPORT_SYMBOL(ip_defrag);
685 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
687 const struct iphdr *iph;
690 if (skb->protocol != htons(ETH_P_IP))
693 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
697 if (iph->ihl < 5 || iph->version != 4)
699 if (!pskb_may_pull(skb, iph->ihl*4))
702 len = ntohs(iph->tot_len);
703 if (skb->len < len || len < (iph->ihl * 4))
706 if (ip_is_fragment(ip_hdr(skb))) {
707 skb = skb_share_check(skb, GFP_ATOMIC);
709 if (pskb_trim_rcsum(skb, len))
711 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
712 if (ip_defrag(skb, user))
719 EXPORT_SYMBOL(ip_check_defrag);
724 static struct ctl_table ip4_frags_ns_ctl_table[] = {
726 .procname = "ipfrag_high_thresh",
727 .data = &init_net.ipv4.frags.high_thresh,
728 .maxlen = sizeof(int),
730 .proc_handler = proc_dointvec
733 .procname = "ipfrag_low_thresh",
734 .data = &init_net.ipv4.frags.low_thresh,
735 .maxlen = sizeof(int),
737 .proc_handler = proc_dointvec
740 .procname = "ipfrag_time",
741 .data = &init_net.ipv4.frags.timeout,
742 .maxlen = sizeof(int),
744 .proc_handler = proc_dointvec_jiffies,
749 static struct ctl_table ip4_frags_ctl_table[] = {
751 .procname = "ipfrag_secret_interval",
752 .data = &ip4_frags.secret_interval,
753 .maxlen = sizeof(int),
755 .proc_handler = proc_dointvec_jiffies,
758 .procname = "ipfrag_max_dist",
759 .data = &sysctl_ipfrag_max_dist,
760 .maxlen = sizeof(int),
762 .proc_handler = proc_dointvec_minmax,
768 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
770 struct ctl_table *table;
771 struct ctl_table_header *hdr;
773 table = ip4_frags_ns_ctl_table;
774 if (!net_eq(net, &init_net)) {
775 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
779 table[0].data = &net->ipv4.frags.high_thresh;
780 table[1].data = &net->ipv4.frags.low_thresh;
781 table[2].data = &net->ipv4.frags.timeout;
784 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
788 net->ipv4.frags_hdr = hdr;
792 if (!net_eq(net, &init_net))
798 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
800 struct ctl_table *table;
802 table = net->ipv4.frags_hdr->ctl_table_arg;
803 unregister_net_sysctl_table(net->ipv4.frags_hdr);
807 static void ip4_frags_ctl_register(void)
809 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
812 static inline int ip4_frags_ns_ctl_register(struct net *net)
817 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
821 static inline void ip4_frags_ctl_register(void)
826 static int __net_init ipv4_frags_init_net(struct net *net)
829 * Fragment cache limits. We will commit 256K at one time. Should we
830 * cross that limit we will prune down to 192K. This should cope with
831 * even the most extreme cases without allowing an attacker to
832 * measurably harm machine performance.
834 net->ipv4.frags.high_thresh = 256 * 1024;
835 net->ipv4.frags.low_thresh = 192 * 1024;
837 * Important NOTE! Fragment queue must be destroyed before MSL expires.
838 * RFC791 is wrong proposing to prolongate timer each fragment arrival
841 net->ipv4.frags.timeout = IP_FRAG_TIME;
843 inet_frags_init_net(&net->ipv4.frags);
845 return ip4_frags_ns_ctl_register(net);
848 static void __net_exit ipv4_frags_exit_net(struct net *net)
850 ip4_frags_ns_ctl_unregister(net);
851 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
854 static struct pernet_operations ip4_frags_ops = {
855 .init = ipv4_frags_init_net,
856 .exit = ipv4_frags_exit_net,
859 void __init ipfrag_init(void)
861 ip4_frags_ctl_register();
862 register_pernet_subsys(&ip4_frags_ops);
863 ip4_frags.hashfn = ip4_hashfn;
864 ip4_frags.constructor = ip4_frag_init;
865 ip4_frags.destructor = ip4_frag_free;
866 ip4_frags.skb_free = NULL;
867 ip4_frags.qsize = sizeof(struct ipq);
868 ip4_frags.match = ip4_frag_match;
869 ip4_frags.frag_expire = ip_expire;
870 ip4_frags.secret_interval = 10 * 60 * HZ;
871 inet_frags_init(&ip4_frags);