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.
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;
58 static const char ip_frag_cache_name[] = "ip4-frags";
62 struct inet_skb_parm h;
66 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
68 /* Describe an entry in the "incomplete datagrams" queue. */
70 struct inet_frag_queue q;
77 u8 ecn; /* RFC3168 support */
80 struct inet_peer *peer;
83 static u8 ip4_frag_ecn(u8 tos)
85 return 1 << (tos & INET_ECN_MASK);
88 static struct inet_frags ip4_frags;
90 int ip_frag_mem(struct net *net)
92 return sum_frag_mem_limit(&net->ipv4.frags);
95 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
96 struct net_device *dev);
98 struct ip4_create_arg {
103 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
105 net_get_random_once(&ip4_frags.rnd, sizeof(ip4_frags.rnd));
106 return jhash_3words((__force u32)id << 16 | prot,
107 (__force u32)saddr, (__force u32)daddr,
111 static unsigned int ip4_hashfn(const struct inet_frag_queue *q)
113 const struct ipq *ipq;
115 ipq = container_of(q, struct ipq, q);
116 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
119 static bool ip4_frag_match(const struct inet_frag_queue *q, const void *a)
121 const struct ipq *qp;
122 const struct ip4_create_arg *arg = a;
124 qp = container_of(q, struct ipq, q);
125 return qp->id == arg->iph->id &&
126 qp->saddr == arg->iph->saddr &&
127 qp->daddr == arg->iph->daddr &&
128 qp->protocol == arg->iph->protocol &&
129 qp->user == arg->user;
132 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
134 struct ipq *qp = container_of(q, struct ipq, q);
135 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
137 struct net *net = container_of(ipv4, struct net, ipv4);
139 const struct ip4_create_arg *arg = a;
141 qp->protocol = arg->iph->protocol;
142 qp->id = arg->iph->id;
143 qp->ecn = ip4_frag_ecn(arg->iph->tos);
144 qp->saddr = arg->iph->saddr;
145 qp->daddr = arg->iph->daddr;
146 qp->user = arg->user;
147 qp->peer = sysctl_ipfrag_max_dist ?
148 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
151 static void ip4_frag_free(struct inet_frag_queue *q)
155 qp = container_of(q, struct ipq, q);
157 inet_putpeer(qp->peer);
161 /* Destruction primitives. */
163 static void ipq_put(struct ipq *ipq)
165 inet_frag_put(&ipq->q, &ip4_frags);
168 /* Kill ipq entry. It is not destroyed immediately,
169 * because caller (and someone more) holds reference count.
171 static void ipq_kill(struct ipq *ipq)
173 inet_frag_kill(&ipq->q, &ip4_frags);
177 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
179 static void ip_expire(unsigned long arg)
184 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
185 net = container_of(qp->q.net, struct net, ipv4.frags);
187 spin_lock(&qp->q.lock);
189 if (qp->q.flags & INET_FRAG_COMPLETE)
193 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
195 if (!(qp->q.flags & INET_FRAG_EVICTED)) {
196 struct sk_buff *head = qp->q.fragments;
197 const struct iphdr *iph;
200 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
202 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
206 head->dev = dev_get_by_index_rcu(net, qp->iif);
210 /* skb has no dst, perform route lookup again */
212 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
213 iph->tos, head->dev);
217 /* Only an end host needs to send an ICMP
218 * "Fragment Reassembly Timeout" message, per RFC792.
220 if (qp->user == IP_DEFRAG_AF_PACKET ||
221 ((qp->user >= IP_DEFRAG_CONNTRACK_IN) &&
222 (qp->user <= __IP_DEFRAG_CONNTRACK_IN_END) &&
223 (skb_rtable(head)->rt_type != RTN_LOCAL)))
226 /* Send an ICMP "Fragment Reassembly Timeout" message. */
227 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
232 spin_unlock(&qp->q.lock);
236 /* Find the correct entry in the "incomplete datagrams" queue for
237 * this IP datagram, and create new one, if nothing is found.
239 static struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
241 struct inet_frag_queue *q;
242 struct ip4_create_arg arg;
248 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
250 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
251 if (IS_ERR_OR_NULL(q)) {
252 inet_frag_maybe_warn_overflow(q, pr_fmt());
255 return container_of(q, struct ipq, q);
258 /* Is the fragment too far ahead to be part of ipq? */
259 static int ip_frag_too_far(struct ipq *qp)
261 struct inet_peer *peer = qp->peer;
262 unsigned int max = sysctl_ipfrag_max_dist;
263 unsigned int start, end;
271 end = atomic_inc_return(&peer->rid);
274 rc = qp->q.fragments && (end - start) > max;
279 net = container_of(qp->q.net, struct net, ipv4.frags);
280 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
286 static int ip_frag_reinit(struct ipq *qp)
289 unsigned int sum_truesize = 0;
291 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
292 atomic_inc(&qp->q.refcnt);
296 fp = qp->q.fragments;
298 struct sk_buff *xp = fp->next;
300 sum_truesize += fp->truesize;
304 sub_frag_mem_limit(&qp->q, sum_truesize);
309 qp->q.fragments = NULL;
310 qp->q.fragments_tail = NULL;
317 /* Add new segment to existing queue. */
318 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
320 struct sk_buff *prev, *next;
321 struct net_device *dev;
327 if (qp->q.flags & INET_FRAG_COMPLETE)
330 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
331 unlikely(ip_frag_too_far(qp)) &&
332 unlikely(err = ip_frag_reinit(qp))) {
337 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
338 offset = ntohs(ip_hdr(skb)->frag_off);
339 flags = offset & ~IP_OFFSET;
341 offset <<= 3; /* offset is in 8-byte chunks */
342 ihl = ip_hdrlen(skb);
344 /* Determine the position of this fragment. */
345 end = offset + skb->len - ihl;
348 /* Is this the final fragment? */
349 if ((flags & IP_MF) == 0) {
350 /* If we already have some bits beyond end
351 * or have different end, the segment is corrupted.
353 if (end < qp->q.len ||
354 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
356 qp->q.flags |= INET_FRAG_LAST_IN;
361 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
362 skb->ip_summed = CHECKSUM_NONE;
364 if (end > qp->q.len) {
365 /* Some bits beyond end -> corruption. */
366 if (qp->q.flags & INET_FRAG_LAST_IN)
375 if (!pskb_pull(skb, ihl))
378 err = pskb_trim_rcsum(skb, end - offset);
382 /* Find out which fragments are in front and at the back of us
383 * in the chain of fragments so far. We must know where to put
384 * this fragment, right?
386 prev = qp->q.fragments_tail;
387 if (!prev || FRAG_CB(prev)->offset < offset) {
392 for (next = qp->q.fragments; next != NULL; next = next->next) {
393 if (FRAG_CB(next)->offset >= offset)
399 /* We found where to put this one. Check for overlap with
400 * preceding fragment, and, if needed, align things so that
401 * any overlaps are eliminated.
404 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
412 if (!pskb_pull(skb, i))
414 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
415 skb->ip_summed = CHECKSUM_NONE;
421 while (next && FRAG_CB(next)->offset < end) {
422 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
425 /* Eat head of the next overlapped fragment
426 * and leave the loop. The next ones cannot overlap.
428 if (!pskb_pull(next, i))
430 FRAG_CB(next)->offset += i;
432 if (next->ip_summed != CHECKSUM_UNNECESSARY)
433 next->ip_summed = CHECKSUM_NONE;
436 struct sk_buff *free_it = next;
438 /* Old fragment is completely overridden with
446 qp->q.fragments = next;
448 qp->q.meat -= free_it->len;
449 sub_frag_mem_limit(&qp->q, free_it->truesize);
454 FRAG_CB(skb)->offset = offset;
456 /* Insert this fragment in the chain of fragments. */
459 qp->q.fragments_tail = skb;
463 qp->q.fragments = skb;
467 qp->iif = dev->ifindex;
470 qp->q.stamp = skb->tstamp;
471 qp->q.meat += skb->len;
473 add_frag_mem_limit(&qp->q, skb->truesize);
475 qp->q.flags |= INET_FRAG_FIRST_IN;
477 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
478 skb->len + ihl > qp->q.max_size)
479 qp->q.max_size = skb->len + ihl;
481 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
482 qp->q.meat == qp->q.len) {
483 unsigned long orefdst = skb->_skb_refdst;
485 skb->_skb_refdst = 0UL;
486 err = ip_frag_reasm(qp, prev, dev);
487 skb->_skb_refdst = orefdst;
500 /* Build a new IP datagram from all its fragments. */
502 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
503 struct net_device *dev)
505 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
507 struct sk_buff *fp, *head = qp->q.fragments;
516 ecn = ip_frag_ecn_table[qp->ecn];
517 if (unlikely(ecn == 0xff)) {
521 /* Make the one we just received the head. */
524 fp = skb_clone(head, GFP_ATOMIC);
528 fp->next = head->next;
530 qp->q.fragments_tail = fp;
533 skb_morph(head, qp->q.fragments);
534 head->next = qp->q.fragments->next;
536 consume_skb(qp->q.fragments);
537 qp->q.fragments = head;
541 WARN_ON(FRAG_CB(head)->offset != 0);
543 /* Allocate a new buffer for the datagram. */
544 ihlen = ip_hdrlen(head);
545 len = ihlen + qp->q.len;
551 /* Head of list must not be cloned. */
552 if (skb_unclone(head, GFP_ATOMIC))
555 /* If the first fragment is fragmented itself, we split
556 * it to two chunks: the first with data and paged part
557 * and the second, holding only fragments. */
558 if (skb_has_frag_list(head)) {
559 struct sk_buff *clone;
562 clone = alloc_skb(0, GFP_ATOMIC);
565 clone->next = head->next;
567 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
568 skb_frag_list_init(head);
569 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
570 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
571 clone->len = clone->data_len = head->data_len - plen;
572 head->data_len -= clone->len;
573 head->len -= clone->len;
575 clone->ip_summed = head->ip_summed;
576 add_frag_mem_limit(&qp->q, clone->truesize);
579 skb_push(head, head->data - skb_network_header(head));
581 sum_truesize = head->truesize;
582 for (fp = head->next; fp;) {
585 struct sk_buff *next = fp->next;
587 sum_truesize += fp->truesize;
588 if (head->ip_summed != fp->ip_summed)
589 head->ip_summed = CHECKSUM_NONE;
590 else if (head->ip_summed == CHECKSUM_COMPLETE)
591 head->csum = csum_add(head->csum, fp->csum);
593 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
594 kfree_skb_partial(fp, headstolen);
596 if (!skb_shinfo(head)->frag_list)
597 skb_shinfo(head)->frag_list = fp;
598 head->data_len += fp->len;
599 head->len += fp->len;
600 head->truesize += fp->truesize;
604 sub_frag_mem_limit(&qp->q, sum_truesize);
608 head->tstamp = qp->q.stamp;
609 IPCB(head)->frag_max_size = qp->q.max_size;
612 /* max_size != 0 implies at least one fragment had IP_DF set */
613 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
614 iph->tot_len = htons(len);
616 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
617 qp->q.fragments = NULL;
618 qp->q.fragments_tail = NULL;
622 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
626 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
628 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
632 /* Process an incoming IP datagram fragment. */
633 int ip_defrag(struct sk_buff *skb, u32 user)
638 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
639 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
641 /* Lookup (or create) queue header */
642 qp = ip_find(net, ip_hdr(skb), user);
646 spin_lock(&qp->q.lock);
648 ret = ip_frag_queue(qp, skb);
650 spin_unlock(&qp->q.lock);
655 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
659 EXPORT_SYMBOL(ip_defrag);
661 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
667 if (skb->protocol != htons(ETH_P_IP))
670 netoff = skb_network_offset(skb);
672 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
675 if (iph.ihl < 5 || iph.version != 4)
678 len = ntohs(iph.tot_len);
679 if (skb->len < netoff + len || len < (iph.ihl * 4))
682 if (ip_is_fragment(&iph)) {
683 skb = skb_share_check(skb, GFP_ATOMIC);
685 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
687 if (pskb_trim_rcsum(skb, netoff + len))
689 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
690 if (ip_defrag(skb, user))
697 EXPORT_SYMBOL(ip_check_defrag);
702 static struct ctl_table ip4_frags_ns_ctl_table[] = {
704 .procname = "ipfrag_high_thresh",
705 .data = &init_net.ipv4.frags.high_thresh,
706 .maxlen = sizeof(int),
708 .proc_handler = proc_dointvec_minmax,
709 .extra1 = &init_net.ipv4.frags.low_thresh
712 .procname = "ipfrag_low_thresh",
713 .data = &init_net.ipv4.frags.low_thresh,
714 .maxlen = sizeof(int),
716 .proc_handler = proc_dointvec_minmax,
718 .extra2 = &init_net.ipv4.frags.high_thresh
721 .procname = "ipfrag_time",
722 .data = &init_net.ipv4.frags.timeout,
723 .maxlen = sizeof(int),
725 .proc_handler = proc_dointvec_jiffies,
730 /* secret interval has been deprecated */
731 static int ip4_frags_secret_interval_unused;
732 static struct ctl_table ip4_frags_ctl_table[] = {
734 .procname = "ipfrag_secret_interval",
735 .data = &ip4_frags_secret_interval_unused,
736 .maxlen = sizeof(int),
738 .proc_handler = proc_dointvec_jiffies,
741 .procname = "ipfrag_max_dist",
742 .data = &sysctl_ipfrag_max_dist,
743 .maxlen = sizeof(int),
745 .proc_handler = proc_dointvec_minmax,
751 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
753 struct ctl_table *table;
754 struct ctl_table_header *hdr;
756 table = ip4_frags_ns_ctl_table;
757 if (!net_eq(net, &init_net)) {
758 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
762 table[0].data = &net->ipv4.frags.high_thresh;
763 table[0].extra1 = &net->ipv4.frags.low_thresh;
764 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
765 table[1].data = &net->ipv4.frags.low_thresh;
766 table[1].extra2 = &net->ipv4.frags.high_thresh;
767 table[2].data = &net->ipv4.frags.timeout;
769 /* Don't export sysctls to unprivileged users */
770 if (net->user_ns != &init_user_ns)
771 table[0].procname = NULL;
774 hdr = register_net_sysctl(net, "net/ipv4", table);
778 net->ipv4.frags_hdr = hdr;
782 if (!net_eq(net, &init_net))
788 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
790 struct ctl_table *table;
792 table = net->ipv4.frags_hdr->ctl_table_arg;
793 unregister_net_sysctl_table(net->ipv4.frags_hdr);
797 static void __init ip4_frags_ctl_register(void)
799 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
802 static int ip4_frags_ns_ctl_register(struct net *net)
807 static void ip4_frags_ns_ctl_unregister(struct net *net)
811 static void __init ip4_frags_ctl_register(void)
816 static int __net_init ipv4_frags_init_net(struct net *net)
818 /* Fragment cache limits.
820 * The fragment memory accounting code, (tries to) account for
821 * the real memory usage, by measuring both the size of frag
822 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
823 * and the SKB's truesize.
825 * A 64K fragment consumes 129736 bytes (44*2944)+200
826 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
828 * We will commit 4MB at one time. Should we cross that limit
829 * we will prune down to 3MB, making room for approx 8 big 64K
832 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
833 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
835 * Important NOTE! Fragment queue must be destroyed before MSL expires.
836 * RFC791 is wrong proposing to prolongate timer each fragment arrival
839 net->ipv4.frags.timeout = IP_FRAG_TIME;
841 inet_frags_init_net(&net->ipv4.frags);
843 return ip4_frags_ns_ctl_register(net);
846 static void __net_exit ipv4_frags_exit_net(struct net *net)
848 ip4_frags_ns_ctl_unregister(net);
849 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
852 static struct pernet_operations ip4_frags_ops = {
853 .init = ipv4_frags_init_net,
854 .exit = ipv4_frags_exit_net,
857 void __init ipfrag_init(void)
859 ip4_frags_ctl_register();
860 register_pernet_subsys(&ip4_frags_ops);
861 ip4_frags.hashfn = ip4_hashfn;
862 ip4_frags.constructor = ip4_frag_init;
863 ip4_frags.destructor = ip4_frag_free;
864 ip4_frags.skb_free = NULL;
865 ip4_frags.qsize = sizeof(struct ipq);
866 ip4_frags.match = ip4_frag_match;
867 ip4_frags.frag_expire = ip_expire;
868 ip4_frags.frags_cache_name = ip_frag_cache_name;
869 if (inet_frags_init(&ip4_frags))
870 panic("IP: failed to allocate ip4_frags cache\n");