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 int 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 ip4_create_arg *arg = a;
176 qp->protocol = arg->iph->protocol;
177 qp->id = arg->iph->id;
178 qp->ecn = ip4_frag_ecn(arg->iph->tos);
179 qp->saddr = arg->iph->saddr;
180 qp->daddr = arg->iph->daddr;
181 qp->user = arg->user;
182 qp->peer = sysctl_ipfrag_max_dist ?
183 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
186 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
190 qp = container_of(q, struct ipq, q);
192 inet_putpeer(qp->peer);
196 /* Destruction primitives. */
198 static __inline__ void ipq_put(struct ipq *ipq)
200 inet_frag_put(&ipq->q, &ip4_frags);
203 /* Kill ipq entry. It is not destroyed immediately,
204 * because caller (and someone more) holds reference count.
206 static void ipq_kill(struct ipq *ipq)
208 inet_frag_kill(&ipq->q, &ip4_frags);
211 /* Memory limiting on fragments. Evictor trashes the oldest
212 * fragment queue until we are back under the threshold.
214 static void ip_evictor(struct net *net)
218 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
220 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
224 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
226 static void ip_expire(unsigned long arg)
231 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
232 net = container_of(qp->q.net, struct net, ipv4.frags);
234 spin_lock(&qp->q.lock);
236 if (qp->q.last_in & INET_FRAG_COMPLETE)
241 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
242 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
244 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
245 struct sk_buff *head = qp->q.fragments;
246 const struct iphdr *iph;
250 head->dev = dev_get_by_index_rcu(net, qp->iif);
254 /* skb has no dst, perform route lookup again */
256 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
257 iph->tos, head->dev);
262 * Only an end host needs to send an ICMP
263 * "Fragment Reassembly Timeout" message, per RFC792.
265 if (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);
296 if (IS_ERR_OR_NULL(q)) {
297 inet_frag_maybe_warn_overflow(q, pr_fmt());
301 return container_of(q, struct ipq, q);
304 /* Is the fragment too far ahead to be part of ipq? */
305 static inline int ip_frag_too_far(struct ipq *qp)
307 struct inet_peer *peer = qp->peer;
308 unsigned int max = sysctl_ipfrag_max_dist;
309 unsigned int start, end;
317 end = atomic_inc_return(&peer->rid);
320 rc = qp->q.fragments && (end - start) > max;
325 net = container_of(qp->q.net, struct net, ipv4.frags);
326 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
332 static int ip_frag_reinit(struct ipq *qp)
336 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
337 atomic_inc(&qp->q.refcnt);
341 fp = qp->q.fragments;
343 struct sk_buff *xp = fp->next;
344 frag_kfree_skb(qp->q.net, fp);
351 qp->q.fragments = NULL;
352 qp->q.fragments_tail = NULL;
359 /* Add new segment to existing queue. */
360 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
362 struct sk_buff *prev, *next;
363 struct net_device *dev;
369 if (qp->q.last_in & INET_FRAG_COMPLETE)
372 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
373 unlikely(ip_frag_too_far(qp)) &&
374 unlikely(err = ip_frag_reinit(qp))) {
379 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
380 offset = ntohs(ip_hdr(skb)->frag_off);
381 flags = offset & ~IP_OFFSET;
383 offset <<= 3; /* offset is in 8-byte chunks */
384 ihl = ip_hdrlen(skb);
386 /* Determine the position of this fragment. */
387 end = offset + skb->len - ihl;
390 /* Is this the final fragment? */
391 if ((flags & IP_MF) == 0) {
392 /* If we already have some bits beyond end
393 * or have different end, the segment is corrrupted.
395 if (end < qp->q.len ||
396 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
398 qp->q.last_in |= INET_FRAG_LAST_IN;
403 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
404 skb->ip_summed = CHECKSUM_NONE;
406 if (end > qp->q.len) {
407 /* Some bits beyond end -> corruption. */
408 if (qp->q.last_in & INET_FRAG_LAST_IN)
417 if (pskb_pull(skb, ihl) == NULL)
420 err = pskb_trim_rcsum(skb, end - offset);
424 /* Find out which fragments are in front and at the back of us
425 * in the chain of fragments so far. We must know where to put
426 * this fragment, right?
428 prev = qp->q.fragments_tail;
429 if (!prev || FRAG_CB(prev)->offset < offset) {
434 for (next = qp->q.fragments; next != NULL; next = next->next) {
435 if (FRAG_CB(next)->offset >= offset)
441 /* We found where to put this one. Check for overlap with
442 * preceding fragment, and, if needed, align things so that
443 * any overlaps are eliminated.
446 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
454 if (!pskb_pull(skb, i))
456 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
457 skb->ip_summed = CHECKSUM_NONE;
463 while (next && FRAG_CB(next)->offset < end) {
464 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
467 /* Eat head of the next overlapped fragment
468 * and leave the loop. The next ones cannot overlap.
470 if (!pskb_pull(next, i))
472 FRAG_CB(next)->offset += i;
474 if (next->ip_summed != CHECKSUM_UNNECESSARY)
475 next->ip_summed = CHECKSUM_NONE;
478 struct sk_buff *free_it = next;
480 /* Old fragment is completely overridden with
488 qp->q.fragments = next;
490 qp->q.meat -= free_it->len;
491 frag_kfree_skb(qp->q.net, free_it);
495 FRAG_CB(skb)->offset = offset;
497 /* Insert this fragment in the chain of fragments. */
500 qp->q.fragments_tail = skb;
504 qp->q.fragments = skb;
508 qp->iif = dev->ifindex;
511 qp->q.stamp = skb->tstamp;
512 qp->q.meat += skb->len;
514 atomic_add(skb->truesize, &qp->q.net->mem);
516 qp->q.last_in |= INET_FRAG_FIRST_IN;
518 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
519 qp->q.meat == qp->q.len) {
520 unsigned long orefdst = skb->_skb_refdst;
522 skb->_skb_refdst = 0UL;
523 err = ip_frag_reasm(qp, prev, dev);
524 skb->_skb_refdst = orefdst;
530 write_lock(&ip4_frags.lock);
531 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
532 write_unlock(&ip4_frags.lock);
541 /* Build a new IP datagram from all its fragments. */
543 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
544 struct net_device *dev)
546 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
548 struct sk_buff *fp, *head = qp->q.fragments;
556 ecn = ip4_frag_ecn_table[qp->ecn];
557 if (unlikely(ecn == 0xff)) {
561 /* Make the one we just received the head. */
564 fp = skb_clone(head, GFP_ATOMIC);
568 fp->next = head->next;
570 qp->q.fragments_tail = fp;
573 skb_morph(head, qp->q.fragments);
574 head->next = qp->q.fragments->next;
576 kfree_skb(qp->q.fragments);
577 qp->q.fragments = head;
580 WARN_ON(head == NULL);
581 WARN_ON(FRAG_CB(head)->offset != 0);
583 /* Allocate a new buffer for the datagram. */
584 ihlen = ip_hdrlen(head);
585 len = ihlen + qp->q.len;
591 /* Head of list must not be cloned. */
592 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
595 /* If the first fragment is fragmented itself, we split
596 * it to two chunks: the first with data and paged part
597 * and the second, holding only fragments. */
598 if (skb_has_frag_list(head)) {
599 struct sk_buff *clone;
602 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
604 clone->next = head->next;
606 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
607 skb_frag_list_init(head);
608 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
609 plen += skb_shinfo(head)->frags[i].size;
610 clone->len = clone->data_len = head->data_len - plen;
611 head->data_len -= clone->len;
612 head->len -= clone->len;
614 clone->ip_summed = head->ip_summed;
615 atomic_add(clone->truesize, &qp->q.net->mem);
618 skb_shinfo(head)->frag_list = head->next;
619 skb_push(head, head->data - skb_network_header(head));
621 for (fp=head->next; fp; fp = fp->next) {
622 head->data_len += fp->len;
623 head->len += fp->len;
624 if (head->ip_summed != fp->ip_summed)
625 head->ip_summed = CHECKSUM_NONE;
626 else if (head->ip_summed == CHECKSUM_COMPLETE)
627 head->csum = csum_add(head->csum, fp->csum);
628 head->truesize += fp->truesize;
630 atomic_sub(head->truesize, &qp->q.net->mem);
634 head->tstamp = qp->q.stamp;
638 iph->tot_len = htons(len);
640 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
641 qp->q.fragments = NULL;
642 qp->q.fragments_tail = NULL;
646 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
652 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
655 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
659 /* Process an incoming IP datagram fragment. */
660 int ip_defrag(struct sk_buff *skb, u32 user)
665 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
666 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
668 /* Start by cleaning up the memory. */
669 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
672 /* Lookup (or create) queue header */
673 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
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);
694 static struct ctl_table ip4_frags_ns_ctl_table[] = {
696 .procname = "ipfrag_high_thresh",
697 .data = &init_net.ipv4.frags.high_thresh,
698 .maxlen = sizeof(int),
700 .proc_handler = proc_dointvec
703 .procname = "ipfrag_low_thresh",
704 .data = &init_net.ipv4.frags.low_thresh,
705 .maxlen = sizeof(int),
707 .proc_handler = proc_dointvec
710 .procname = "ipfrag_time",
711 .data = &init_net.ipv4.frags.timeout,
712 .maxlen = sizeof(int),
714 .proc_handler = proc_dointvec_jiffies,
719 static struct ctl_table ip4_frags_ctl_table[] = {
721 .procname = "ipfrag_secret_interval",
722 .data = &ip4_frags.secret_interval,
723 .maxlen = sizeof(int),
725 .proc_handler = proc_dointvec_jiffies,
728 .procname = "ipfrag_max_dist",
729 .data = &sysctl_ipfrag_max_dist,
730 .maxlen = sizeof(int),
732 .proc_handler = proc_dointvec_minmax,
738 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
740 struct ctl_table *table;
741 struct ctl_table_header *hdr;
743 table = ip4_frags_ns_ctl_table;
744 if (!net_eq(net, &init_net)) {
745 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
749 table[0].data = &net->ipv4.frags.high_thresh;
750 table[1].data = &net->ipv4.frags.low_thresh;
751 table[2].data = &net->ipv4.frags.timeout;
754 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
758 net->ipv4.frags_hdr = hdr;
762 if (!net_eq(net, &init_net))
768 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
770 struct ctl_table *table;
772 table = net->ipv4.frags_hdr->ctl_table_arg;
773 unregister_net_sysctl_table(net->ipv4.frags_hdr);
777 static void ip4_frags_ctl_register(void)
779 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
782 static inline int ip4_frags_ns_ctl_register(struct net *net)
787 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
791 static inline void ip4_frags_ctl_register(void)
796 static int __net_init ipv4_frags_init_net(struct net *net)
799 * Fragment cache limits. We will commit 256K at one time. Should we
800 * cross that limit we will prune down to 192K. This should cope with
801 * even the most extreme cases without allowing an attacker to
802 * measurably harm machine performance.
804 net->ipv4.frags.high_thresh = 256 * 1024;
805 net->ipv4.frags.low_thresh = 192 * 1024;
807 * Important NOTE! Fragment queue must be destroyed before MSL expires.
808 * RFC791 is wrong proposing to prolongate timer each fragment arrival
811 net->ipv4.frags.timeout = IP_FRAG_TIME;
813 inet_frags_init_net(&net->ipv4.frags);
815 return ip4_frags_ns_ctl_register(net);
818 static void __net_exit ipv4_frags_exit_net(struct net *net)
820 ip4_frags_ns_ctl_unregister(net);
821 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
824 static struct pernet_operations ip4_frags_ops = {
825 .init = ipv4_frags_init_net,
826 .exit = ipv4_frags_exit_net,
829 void __init ipfrag_init(void)
831 ip4_frags_ctl_register();
832 register_pernet_subsys(&ip4_frags_ops);
833 ip4_frags.hashfn = ip4_hashfn;
834 ip4_frags.constructor = ip4_frag_init;
835 ip4_frags.destructor = ip4_frag_free;
836 ip4_frags.skb_free = NULL;
837 ip4_frags.qsize = sizeof(struct ipq);
838 ip4_frags.match = ip4_frag_match;
839 ip4_frags.frag_expire = ip_expire;
840 ip4_frags.secret_interval = 10 * 60 * HZ;
841 inet_frags_init(&ip4_frags);