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_CONNTRACK_IN &&
265 skb_rtable(head)->rt_type != RTN_LOCAL)
269 /* Send an ICMP "Fragment Reassembly Timeout" message. */
270 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
275 spin_unlock(&qp->q.lock);
279 /* Find the correct entry in the "incomplete datagrams" queue for
280 * this IP datagram, and create new one, if nothing is found.
282 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
284 struct inet_frag_queue *q;
285 struct ip4_create_arg arg;
291 read_lock(&ip4_frags.lock);
292 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
294 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
298 return container_of(q, struct ipq, q);
301 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
305 /* Is the fragment too far ahead to be part of ipq? */
306 static inline int ip_frag_too_far(struct ipq *qp)
308 struct inet_peer *peer = qp->peer;
309 unsigned int max = sysctl_ipfrag_max_dist;
310 unsigned int start, end;
318 end = atomic_inc_return(&peer->rid);
321 rc = qp->q.fragments && (end - start) > max;
326 net = container_of(qp->q.net, struct net, ipv4.frags);
327 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
333 static int ip_frag_reinit(struct ipq *qp)
337 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
338 atomic_inc(&qp->q.refcnt);
342 fp = qp->q.fragments;
344 struct sk_buff *xp = fp->next;
345 frag_kfree_skb(qp->q.net, fp);
352 qp->q.fragments = NULL;
353 qp->q.fragments_tail = NULL;
360 /* Add new segment to existing queue. */
361 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
363 struct sk_buff *prev, *next;
364 struct net_device *dev;
370 if (qp->q.last_in & INET_FRAG_COMPLETE)
373 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
374 unlikely(ip_frag_too_far(qp)) &&
375 unlikely(err = ip_frag_reinit(qp))) {
380 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
381 offset = ntohs(ip_hdr(skb)->frag_off);
382 flags = offset & ~IP_OFFSET;
384 offset <<= 3; /* offset is in 8-byte chunks */
385 ihl = ip_hdrlen(skb);
387 /* Determine the position of this fragment. */
388 end = offset + skb->len - ihl;
391 /* Is this the final fragment? */
392 if ((flags & IP_MF) == 0) {
393 /* If we already have some bits beyond end
394 * or have different end, the segment is corrrupted.
396 if (end < qp->q.len ||
397 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
399 qp->q.last_in |= INET_FRAG_LAST_IN;
404 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
405 skb->ip_summed = CHECKSUM_NONE;
407 if (end > qp->q.len) {
408 /* Some bits beyond end -> corruption. */
409 if (qp->q.last_in & INET_FRAG_LAST_IN)
418 if (pskb_pull(skb, ihl) == NULL)
421 err = pskb_trim_rcsum(skb, end - offset);
425 /* Find out which fragments are in front and at the back of us
426 * in the chain of fragments so far. We must know where to put
427 * this fragment, right?
429 prev = qp->q.fragments_tail;
430 if (!prev || FRAG_CB(prev)->offset < offset) {
435 for (next = qp->q.fragments; next != NULL; next = next->next) {
436 if (FRAG_CB(next)->offset >= offset)
442 /* We found where to put this one. Check for overlap with
443 * preceding fragment, and, if needed, align things so that
444 * any overlaps are eliminated.
447 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
455 if (!pskb_pull(skb, i))
457 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
458 skb->ip_summed = CHECKSUM_NONE;
464 while (next && FRAG_CB(next)->offset < end) {
465 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
468 /* Eat head of the next overlapped fragment
469 * and leave the loop. The next ones cannot overlap.
471 if (!pskb_pull(next, i))
473 FRAG_CB(next)->offset += i;
475 if (next->ip_summed != CHECKSUM_UNNECESSARY)
476 next->ip_summed = CHECKSUM_NONE;
479 struct sk_buff *free_it = next;
481 /* Old fragment is completely overridden with
489 qp->q.fragments = next;
491 qp->q.meat -= free_it->len;
492 frag_kfree_skb(qp->q.net, free_it);
496 FRAG_CB(skb)->offset = offset;
498 /* Insert this fragment in the chain of fragments. */
501 qp->q.fragments_tail = skb;
505 qp->q.fragments = skb;
509 qp->iif = dev->ifindex;
512 qp->q.stamp = skb->tstamp;
513 qp->q.meat += skb->len;
515 atomic_add(skb->truesize, &qp->q.net->mem);
517 qp->q.last_in |= INET_FRAG_FIRST_IN;
519 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
520 qp->q.meat == qp->q.len)
521 return ip_frag_reasm(qp, prev, dev);
523 write_lock(&ip4_frags.lock);
524 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
525 write_unlock(&ip4_frags.lock);
534 /* Build a new IP datagram from all its fragments. */
536 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
537 struct net_device *dev)
539 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
541 struct sk_buff *fp, *head = qp->q.fragments;
549 ecn = ip4_frag_ecn_table[qp->ecn];
550 if (unlikely(ecn == 0xff)) {
554 /* Make the one we just received the head. */
557 fp = skb_clone(head, GFP_ATOMIC);
561 fp->next = head->next;
563 qp->q.fragments_tail = fp;
566 skb_morph(head, qp->q.fragments);
567 head->next = qp->q.fragments->next;
569 kfree_skb(qp->q.fragments);
570 qp->q.fragments = head;
573 WARN_ON(head == NULL);
574 WARN_ON(FRAG_CB(head)->offset != 0);
576 /* Allocate a new buffer for the datagram. */
577 ihlen = ip_hdrlen(head);
578 len = ihlen + qp->q.len;
584 /* Head of list must not be cloned. */
585 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
588 /* If the first fragment is fragmented itself, we split
589 * it to two chunks: the first with data and paged part
590 * and the second, holding only fragments. */
591 if (skb_has_frag_list(head)) {
592 struct sk_buff *clone;
595 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
597 clone->next = head->next;
599 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
600 skb_frag_list_init(head);
601 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
602 plen += skb_shinfo(head)->frags[i].size;
603 clone->len = clone->data_len = head->data_len - plen;
604 head->data_len -= clone->len;
605 head->len -= clone->len;
607 clone->ip_summed = head->ip_summed;
608 atomic_add(clone->truesize, &qp->q.net->mem);
611 skb_shinfo(head)->frag_list = head->next;
612 skb_push(head, head->data - skb_network_header(head));
614 for (fp=head->next; fp; fp = fp->next) {
615 head->data_len += fp->len;
616 head->len += fp->len;
617 if (head->ip_summed != fp->ip_summed)
618 head->ip_summed = CHECKSUM_NONE;
619 else if (head->ip_summed == CHECKSUM_COMPLETE)
620 head->csum = csum_add(head->csum, fp->csum);
621 head->truesize += fp->truesize;
623 atomic_sub(head->truesize, &qp->q.net->mem);
627 head->tstamp = qp->q.stamp;
631 iph->tot_len = htons(len);
633 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
634 qp->q.fragments = NULL;
635 qp->q.fragments_tail = NULL;
639 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
645 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
648 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
652 /* Process an incoming IP datagram fragment. */
653 int ip_defrag(struct sk_buff *skb, u32 user)
658 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
659 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
661 /* Start by cleaning up the memory. */
662 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
665 /* Lookup (or create) queue header */
666 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
669 spin_lock(&qp->q.lock);
671 ret = ip_frag_queue(qp, skb);
673 spin_unlock(&qp->q.lock);
678 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
682 EXPORT_SYMBOL(ip_defrag);
687 static struct ctl_table ip4_frags_ns_ctl_table[] = {
689 .procname = "ipfrag_high_thresh",
690 .data = &init_net.ipv4.frags.high_thresh,
691 .maxlen = sizeof(int),
693 .proc_handler = proc_dointvec
696 .procname = "ipfrag_low_thresh",
697 .data = &init_net.ipv4.frags.low_thresh,
698 .maxlen = sizeof(int),
700 .proc_handler = proc_dointvec
703 .procname = "ipfrag_time",
704 .data = &init_net.ipv4.frags.timeout,
705 .maxlen = sizeof(int),
707 .proc_handler = proc_dointvec_jiffies,
712 static struct ctl_table ip4_frags_ctl_table[] = {
714 .procname = "ipfrag_secret_interval",
715 .data = &ip4_frags.secret_interval,
716 .maxlen = sizeof(int),
718 .proc_handler = proc_dointvec_jiffies,
721 .procname = "ipfrag_max_dist",
722 .data = &sysctl_ipfrag_max_dist,
723 .maxlen = sizeof(int),
725 .proc_handler = proc_dointvec_minmax,
731 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
733 struct ctl_table *table;
734 struct ctl_table_header *hdr;
736 table = ip4_frags_ns_ctl_table;
737 if (!net_eq(net, &init_net)) {
738 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
742 table[0].data = &net->ipv4.frags.high_thresh;
743 table[1].data = &net->ipv4.frags.low_thresh;
744 table[2].data = &net->ipv4.frags.timeout;
747 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
751 net->ipv4.frags_hdr = hdr;
755 if (!net_eq(net, &init_net))
761 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
763 struct ctl_table *table;
765 table = net->ipv4.frags_hdr->ctl_table_arg;
766 unregister_net_sysctl_table(net->ipv4.frags_hdr);
770 static void ip4_frags_ctl_register(void)
772 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
775 static inline int ip4_frags_ns_ctl_register(struct net *net)
780 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
784 static inline void ip4_frags_ctl_register(void)
789 static int __net_init ipv4_frags_init_net(struct net *net)
792 * Fragment cache limits. We will commit 256K at one time. Should we
793 * cross that limit we will prune down to 192K. This should cope with
794 * even the most extreme cases without allowing an attacker to
795 * measurably harm machine performance.
797 net->ipv4.frags.high_thresh = 256 * 1024;
798 net->ipv4.frags.low_thresh = 192 * 1024;
800 * Important NOTE! Fragment queue must be destroyed before MSL expires.
801 * RFC791 is wrong proposing to prolongate timer each fragment arrival
804 net->ipv4.frags.timeout = IP_FRAG_TIME;
806 inet_frags_init_net(&net->ipv4.frags);
808 return ip4_frags_ns_ctl_register(net);
811 static void __net_exit ipv4_frags_exit_net(struct net *net)
813 ip4_frags_ns_ctl_unregister(net);
814 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
817 static struct pernet_operations ip4_frags_ops = {
818 .init = ipv4_frags_init_net,
819 .exit = ipv4_frags_exit_net,
822 void __init ipfrag_init(void)
824 ip4_frags_ctl_register();
825 register_pernet_subsys(&ip4_frags_ops);
826 ip4_frags.hashfn = ip4_hashfn;
827 ip4_frags.constructor = ip4_frag_init;
828 ip4_frags.destructor = ip4_frag_free;
829 ip4_frags.skb_free = NULL;
830 ip4_frags.qsize = sizeof(struct ipq);
831 ip4_frags.match = ip4_frag_match;
832 ip4_frags.frag_expire = ip_expire;
833 ip4_frags.secret_interval = 10 * 60 * HZ;
834 inet_frags_init(&ip4_frags);