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;
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);
176 static bool frag_expire_skip_icmp(u32 user)
178 return user == IP_DEFRAG_AF_PACKET ||
179 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
180 __IP_DEFRAG_CONNTRACK_IN_END) ||
181 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
182 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
186 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
188 static void ip_expire(unsigned long arg)
193 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
194 net = container_of(qp->q.net, struct net, ipv4.frags);
196 spin_lock(&qp->q.lock);
198 if (qp->q.flags & INET_FRAG_COMPLETE)
202 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
204 if (!(qp->q.flags & INET_FRAG_EVICTED)) {
205 struct sk_buff *head = qp->q.fragments;
206 const struct iphdr *iph;
209 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
211 if (!(qp->q.flags & INET_FRAG_FIRST_IN) || !qp->q.fragments)
215 head->dev = dev_get_by_index_rcu(net, qp->iif);
219 /* skb has no dst, perform route lookup again */
221 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
222 iph->tos, head->dev);
226 /* Only an end host needs to send an ICMP
227 * "Fragment Reassembly Timeout" message, per RFC792.
229 if (frag_expire_skip_icmp(qp->user) &&
230 (skb_rtable(head)->rt_type != RTN_LOCAL))
233 /* Send an ICMP "Fragment Reassembly Timeout" message. */
234 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
239 spin_unlock(&qp->q.lock);
243 /* Find the correct entry in the "incomplete datagrams" queue for
244 * this IP datagram, and create new one, if nothing is found.
246 static struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
248 struct inet_frag_queue *q;
249 struct ip4_create_arg arg;
255 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
257 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
258 if (IS_ERR_OR_NULL(q)) {
259 inet_frag_maybe_warn_overflow(q, pr_fmt());
262 return container_of(q, struct ipq, q);
265 /* Is the fragment too far ahead to be part of ipq? */
266 static int ip_frag_too_far(struct ipq *qp)
268 struct inet_peer *peer = qp->peer;
269 unsigned int max = sysctl_ipfrag_max_dist;
270 unsigned int start, end;
278 end = atomic_inc_return(&peer->rid);
281 rc = qp->q.fragments && (end - start) > max;
286 net = container_of(qp->q.net, struct net, ipv4.frags);
287 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
293 static int ip_frag_reinit(struct ipq *qp)
296 unsigned int sum_truesize = 0;
298 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
299 atomic_inc(&qp->q.refcnt);
303 fp = qp->q.fragments;
305 struct sk_buff *xp = fp->next;
307 sum_truesize += fp->truesize;
311 sub_frag_mem_limit(&qp->q, sum_truesize);
316 qp->q.fragments = NULL;
317 qp->q.fragments_tail = NULL;
324 /* Add new segment to existing queue. */
325 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
327 struct sk_buff *prev, *next;
328 struct net_device *dev;
334 if (qp->q.flags & INET_FRAG_COMPLETE)
337 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
338 unlikely(ip_frag_too_far(qp)) &&
339 unlikely(err = ip_frag_reinit(qp))) {
344 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
345 offset = ntohs(ip_hdr(skb)->frag_off);
346 flags = offset & ~IP_OFFSET;
348 offset <<= 3; /* offset is in 8-byte chunks */
349 ihl = ip_hdrlen(skb);
351 /* Determine the position of this fragment. */
352 end = offset + skb->len - ihl;
355 /* Is this the final fragment? */
356 if ((flags & IP_MF) == 0) {
357 /* If we already have some bits beyond end
358 * or have different end, the segment is corrupted.
360 if (end < qp->q.len ||
361 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
363 qp->q.flags |= INET_FRAG_LAST_IN;
368 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
369 skb->ip_summed = CHECKSUM_NONE;
371 if (end > qp->q.len) {
372 /* Some bits beyond end -> corruption. */
373 if (qp->q.flags & INET_FRAG_LAST_IN)
382 if (!pskb_pull(skb, ihl))
385 err = pskb_trim_rcsum(skb, end - offset);
389 /* Find out which fragments are in front and at the back of us
390 * in the chain of fragments so far. We must know where to put
391 * this fragment, right?
393 prev = qp->q.fragments_tail;
394 if (!prev || FRAG_CB(prev)->offset < offset) {
399 for (next = qp->q.fragments; next != NULL; next = next->next) {
400 if (FRAG_CB(next)->offset >= offset)
406 /* We found where to put this one. Check for overlap with
407 * preceding fragment, and, if needed, align things so that
408 * any overlaps are eliminated.
411 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
419 if (!pskb_pull(skb, i))
421 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
422 skb->ip_summed = CHECKSUM_NONE;
428 while (next && FRAG_CB(next)->offset < end) {
429 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
432 /* Eat head of the next overlapped fragment
433 * and leave the loop. The next ones cannot overlap.
435 if (!pskb_pull(next, i))
437 FRAG_CB(next)->offset += i;
439 if (next->ip_summed != CHECKSUM_UNNECESSARY)
440 next->ip_summed = CHECKSUM_NONE;
443 struct sk_buff *free_it = next;
445 /* Old fragment is completely overridden with
453 qp->q.fragments = next;
455 qp->q.meat -= free_it->len;
456 sub_frag_mem_limit(&qp->q, free_it->truesize);
461 FRAG_CB(skb)->offset = offset;
463 /* Insert this fragment in the chain of fragments. */
466 qp->q.fragments_tail = skb;
470 qp->q.fragments = skb;
474 qp->iif = dev->ifindex;
477 qp->q.stamp = skb->tstamp;
478 qp->q.meat += skb->len;
480 add_frag_mem_limit(&qp->q, skb->truesize);
482 qp->q.flags |= INET_FRAG_FIRST_IN;
484 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
485 skb->len + ihl > qp->q.max_size)
486 qp->q.max_size = skb->len + ihl;
488 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
489 qp->q.meat == qp->q.len) {
490 unsigned long orefdst = skb->_skb_refdst;
492 skb->_skb_refdst = 0UL;
493 err = ip_frag_reasm(qp, prev, dev);
494 skb->_skb_refdst = orefdst;
507 /* Build a new IP datagram from all its fragments. */
509 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
510 struct net_device *dev)
512 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
514 struct sk_buff *fp, *head = qp->q.fragments;
523 ecn = ip_frag_ecn_table[qp->ecn];
524 if (unlikely(ecn == 0xff)) {
528 /* Make the one we just received the head. */
531 fp = skb_clone(head, GFP_ATOMIC);
535 fp->next = head->next;
537 qp->q.fragments_tail = fp;
540 skb_morph(head, qp->q.fragments);
541 head->next = qp->q.fragments->next;
543 consume_skb(qp->q.fragments);
544 qp->q.fragments = head;
548 WARN_ON(FRAG_CB(head)->offset != 0);
550 /* Allocate a new buffer for the datagram. */
551 ihlen = ip_hdrlen(head);
552 len = ihlen + qp->q.len;
558 /* Head of list must not be cloned. */
559 if (skb_unclone(head, GFP_ATOMIC))
562 /* If the first fragment is fragmented itself, we split
563 * it to two chunks: the first with data and paged part
564 * and the second, holding only fragments. */
565 if (skb_has_frag_list(head)) {
566 struct sk_buff *clone;
569 clone = alloc_skb(0, GFP_ATOMIC);
572 clone->next = head->next;
574 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
575 skb_frag_list_init(head);
576 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
577 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
578 clone->len = clone->data_len = head->data_len - plen;
579 head->data_len -= clone->len;
580 head->len -= clone->len;
582 clone->ip_summed = head->ip_summed;
583 add_frag_mem_limit(&qp->q, clone->truesize);
586 skb_push(head, head->data - skb_network_header(head));
588 sum_truesize = head->truesize;
589 for (fp = head->next; fp;) {
592 struct sk_buff *next = fp->next;
594 sum_truesize += fp->truesize;
595 if (head->ip_summed != fp->ip_summed)
596 head->ip_summed = CHECKSUM_NONE;
597 else if (head->ip_summed == CHECKSUM_COMPLETE)
598 head->csum = csum_add(head->csum, fp->csum);
600 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
601 kfree_skb_partial(fp, headstolen);
603 if (!skb_shinfo(head)->frag_list)
604 skb_shinfo(head)->frag_list = fp;
605 head->data_len += fp->len;
606 head->len += fp->len;
607 head->truesize += fp->truesize;
611 sub_frag_mem_limit(&qp->q, sum_truesize);
615 head->tstamp = qp->q.stamp;
616 IPCB(head)->frag_max_size = qp->q.max_size;
619 /* max_size != 0 implies at least one fragment had IP_DF set */
620 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
621 iph->tot_len = htons(len);
623 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
624 qp->q.fragments = NULL;
625 qp->q.fragments_tail = NULL;
629 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
633 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
635 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
639 /* Process an incoming IP datagram fragment. */
640 int ip_defrag(struct sk_buff *skb, u32 user)
645 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
646 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
648 /* Lookup (or create) queue header */
649 qp = ip_find(net, ip_hdr(skb), user);
653 spin_lock(&qp->q.lock);
655 ret = ip_frag_queue(qp, skb);
657 spin_unlock(&qp->q.lock);
662 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
666 EXPORT_SYMBOL(ip_defrag);
668 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
674 if (skb->protocol != htons(ETH_P_IP))
677 netoff = skb_network_offset(skb);
679 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
682 if (iph.ihl < 5 || iph.version != 4)
685 len = ntohs(iph.tot_len);
686 if (skb->len < netoff + len || len < (iph.ihl * 4))
689 if (ip_is_fragment(&iph)) {
690 skb = skb_share_check(skb, GFP_ATOMIC);
692 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
694 if (pskb_trim_rcsum(skb, netoff + len))
696 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
697 if (ip_defrag(skb, user))
704 EXPORT_SYMBOL(ip_check_defrag);
709 static struct ctl_table ip4_frags_ns_ctl_table[] = {
711 .procname = "ipfrag_high_thresh",
712 .data = &init_net.ipv4.frags.high_thresh,
713 .maxlen = sizeof(int),
715 .proc_handler = proc_dointvec_minmax,
716 .extra1 = &init_net.ipv4.frags.low_thresh
719 .procname = "ipfrag_low_thresh",
720 .data = &init_net.ipv4.frags.low_thresh,
721 .maxlen = sizeof(int),
723 .proc_handler = proc_dointvec_minmax,
725 .extra2 = &init_net.ipv4.frags.high_thresh
728 .procname = "ipfrag_time",
729 .data = &init_net.ipv4.frags.timeout,
730 .maxlen = sizeof(int),
732 .proc_handler = proc_dointvec_jiffies,
737 /* secret interval has been deprecated */
738 static int ip4_frags_secret_interval_unused;
739 static struct ctl_table ip4_frags_ctl_table[] = {
741 .procname = "ipfrag_secret_interval",
742 .data = &ip4_frags_secret_interval_unused,
743 .maxlen = sizeof(int),
745 .proc_handler = proc_dointvec_jiffies,
748 .procname = "ipfrag_max_dist",
749 .data = &sysctl_ipfrag_max_dist,
750 .maxlen = sizeof(int),
752 .proc_handler = proc_dointvec_minmax,
758 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
760 struct ctl_table *table;
761 struct ctl_table_header *hdr;
763 table = ip4_frags_ns_ctl_table;
764 if (!net_eq(net, &init_net)) {
765 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
769 table[0].data = &net->ipv4.frags.high_thresh;
770 table[0].extra1 = &net->ipv4.frags.low_thresh;
771 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
772 table[1].data = &net->ipv4.frags.low_thresh;
773 table[1].extra2 = &net->ipv4.frags.high_thresh;
774 table[2].data = &net->ipv4.frags.timeout;
776 /* Don't export sysctls to unprivileged users */
777 if (net->user_ns != &init_user_ns)
778 table[0].procname = NULL;
781 hdr = register_net_sysctl(net, "net/ipv4", table);
785 net->ipv4.frags_hdr = hdr;
789 if (!net_eq(net, &init_net))
795 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
797 struct ctl_table *table;
799 table = net->ipv4.frags_hdr->ctl_table_arg;
800 unregister_net_sysctl_table(net->ipv4.frags_hdr);
804 static void __init ip4_frags_ctl_register(void)
806 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
809 static int ip4_frags_ns_ctl_register(struct net *net)
814 static void ip4_frags_ns_ctl_unregister(struct net *net)
818 static void __init ip4_frags_ctl_register(void)
823 static int __net_init ipv4_frags_init_net(struct net *net)
825 /* Fragment cache limits.
827 * The fragment memory accounting code, (tries to) account for
828 * the real memory usage, by measuring both the size of frag
829 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
830 * and the SKB's truesize.
832 * A 64K fragment consumes 129736 bytes (44*2944)+200
833 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
835 * We will commit 4MB at one time. Should we cross that limit
836 * we will prune down to 3MB, making room for approx 8 big 64K
839 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
840 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
842 * Important NOTE! Fragment queue must be destroyed before MSL expires.
843 * RFC791 is wrong proposing to prolongate timer each fragment arrival
846 net->ipv4.frags.timeout = IP_FRAG_TIME;
848 inet_frags_init_net(&net->ipv4.frags);
850 return ip4_frags_ns_ctl_register(net);
853 static void __net_exit ipv4_frags_exit_net(struct net *net)
855 ip4_frags_ns_ctl_unregister(net);
856 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
859 static struct pernet_operations ip4_frags_ops = {
860 .init = ipv4_frags_init_net,
861 .exit = ipv4_frags_exit_net,
864 void __init ipfrag_init(void)
866 ip4_frags_ctl_register();
867 register_pernet_subsys(&ip4_frags_ops);
868 ip4_frags.hashfn = ip4_hashfn;
869 ip4_frags.constructor = ip4_frag_init;
870 ip4_frags.destructor = ip4_frag_free;
871 ip4_frags.skb_free = NULL;
872 ip4_frags.qsize = sizeof(struct ipq);
873 ip4_frags.match = ip4_frag_match;
874 ip4_frags.frag_expire = ip_expire;
875 ip4_frags.frags_cache_name = ip_frag_cache_name;
876 if (inet_frags_init(&ip4_frags))
877 panic("IP: failed to allocate ip4_frags cache\n");