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 sum_frag_mem_limit(&net->ipv4.frags);
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 bool 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 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
166 struct ipq *qp = container_of(q, struct ipq, q);
167 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
169 struct net *net = container_of(ipv4, struct net, ipv4);
171 struct ip4_create_arg *arg = a;
173 qp->protocol = arg->iph->protocol;
174 qp->id = arg->iph->id;
175 qp->ecn = ip4_frag_ecn(arg->iph->tos);
176 qp->saddr = arg->iph->saddr;
177 qp->daddr = arg->iph->daddr;
178 qp->user = arg->user;
179 qp->peer = sysctl_ipfrag_max_dist ?
180 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
183 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
187 qp = container_of(q, struct ipq, q);
189 inet_putpeer(qp->peer);
193 /* Destruction primitives. */
195 static __inline__ void ipq_put(struct ipq *ipq)
197 inet_frag_put(&ipq->q, &ip4_frags);
200 /* Kill ipq entry. It is not destroyed immediately,
201 * because caller (and someone more) holds reference count.
203 static void ipq_kill(struct ipq *ipq)
205 inet_frag_kill(&ipq->q, &ip4_frags);
208 /* Memory limiting on fragments. Evictor trashes the oldest
209 * fragment queue until we are back under the threshold.
211 static void ip_evictor(struct net *net)
215 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
217 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
221 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
223 static void ip_expire(unsigned long arg)
228 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
229 net = container_of(qp->q.net, struct net, ipv4.frags);
231 spin_lock(&qp->q.lock);
233 if (qp->q.last_in & INET_FRAG_COMPLETE)
238 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
239 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
241 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
242 struct sk_buff *head = qp->q.fragments;
243 const struct iphdr *iph;
247 head->dev = dev_get_by_index_rcu(net, qp->iif);
251 /* skb dst is stale, drop it, and perform route lookup again */
254 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
255 iph->tos, head->dev);
260 * Only an end host needs to send an ICMP
261 * "Fragment Reassembly Timeout" message, per RFC792.
263 if (qp->user == IP_DEFRAG_AF_PACKET ||
264 (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);
295 if (IS_ERR_OR_NULL(q)) {
296 inet_frag_maybe_warn_overflow(q, pr_fmt());
299 return container_of(q, struct ipq, q);
302 /* Is the fragment too far ahead to be part of ipq? */
303 static inline int ip_frag_too_far(struct ipq *qp)
305 struct inet_peer *peer = qp->peer;
306 unsigned int max = sysctl_ipfrag_max_dist;
307 unsigned int start, end;
315 end = atomic_inc_return(&peer->rid);
318 rc = qp->q.fragments && (end - start) > max;
323 net = container_of(qp->q.net, struct net, ipv4.frags);
324 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
330 static int ip_frag_reinit(struct ipq *qp)
333 unsigned int sum_truesize = 0;
335 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
336 atomic_inc(&qp->q.refcnt);
340 fp = qp->q.fragments;
342 struct sk_buff *xp = fp->next;
344 sum_truesize += fp->truesize;
348 sub_frag_mem_limit(&qp->q, sum_truesize);
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 sub_frag_mem_limit(&qp->q, free_it->truesize);
498 FRAG_CB(skb)->offset = offset;
500 /* Insert this fragment in the chain of fragments. */
503 qp->q.fragments_tail = skb;
507 qp->q.fragments = skb;
511 qp->iif = dev->ifindex;
514 qp->q.stamp = skb->tstamp;
515 qp->q.meat += skb->len;
517 add_frag_mem_limit(&qp->q, skb->truesize);
519 qp->q.last_in |= INET_FRAG_FIRST_IN;
521 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
522 skb->len + ihl > qp->q.max_size)
523 qp->q.max_size = skb->len + ihl;
525 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
526 qp->q.meat == qp->q.len)
527 return ip_frag_reasm(qp, prev, dev);
529 inet_frag_lru_move(&qp->q);
538 /* Build a new IP datagram from all its fragments. */
540 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
541 struct net_device *dev)
543 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
545 struct sk_buff *fp, *head = qp->q.fragments;
554 ecn = ip4_frag_ecn_table[qp->ecn];
555 if (unlikely(ecn == 0xff)) {
559 /* Make the one we just received the head. */
562 fp = skb_clone(head, GFP_ATOMIC);
566 fp->next = head->next;
568 qp->q.fragments_tail = fp;
571 skb_morph(head, qp->q.fragments);
572 head->next = qp->q.fragments->next;
574 consume_skb(qp->q.fragments);
575 qp->q.fragments = head;
578 WARN_ON(head == NULL);
579 WARN_ON(FRAG_CB(head)->offset != 0);
581 /* Allocate a new buffer for the datagram. */
582 ihlen = ip_hdrlen(head);
583 len = ihlen + qp->q.len;
589 /* Head of list must not be cloned. */
590 if (skb_unclone(head, GFP_ATOMIC))
593 /* If the first fragment is fragmented itself, we split
594 * it to two chunks: the first with data and paged part
595 * and the second, holding only fragments. */
596 if (skb_has_frag_list(head)) {
597 struct sk_buff *clone;
600 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
602 clone->next = head->next;
604 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
605 skb_frag_list_init(head);
606 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
607 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
608 clone->len = clone->data_len = head->data_len - plen;
609 head->data_len -= clone->len;
610 head->len -= clone->len;
612 clone->ip_summed = head->ip_summed;
613 add_frag_mem_limit(&qp->q, clone->truesize);
616 skb_push(head, head->data - skb_network_header(head));
618 sum_truesize = head->truesize;
619 for (fp = head->next; fp;) {
622 struct sk_buff *next = fp->next;
624 sum_truesize += fp->truesize;
625 if (head->ip_summed != fp->ip_summed)
626 head->ip_summed = CHECKSUM_NONE;
627 else if (head->ip_summed == CHECKSUM_COMPLETE)
628 head->csum = csum_add(head->csum, fp->csum);
630 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
631 kfree_skb_partial(fp, headstolen);
633 if (!skb_shinfo(head)->frag_list)
634 skb_shinfo(head)->frag_list = fp;
635 head->data_len += fp->len;
636 head->len += fp->len;
637 head->truesize += fp->truesize;
641 sub_frag_mem_limit(&qp->q, sum_truesize);
645 head->tstamp = qp->q.stamp;
646 IPCB(head)->frag_max_size = qp->q.max_size;
649 /* max_size != 0 implies at least one fragment had IP_DF set */
650 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
651 iph->tot_len = htons(len);
653 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
654 qp->q.fragments = NULL;
655 qp->q.fragments_tail = NULL;
659 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
664 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
666 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
670 /* Process an incoming IP datagram fragment. */
671 int ip_defrag(struct sk_buff *skb, u32 user)
676 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
677 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
679 /* Start by cleaning up the memory. */
682 /* Lookup (or create) queue header */
683 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
686 spin_lock(&qp->q.lock);
688 ret = ip_frag_queue(qp, skb);
690 spin_unlock(&qp->q.lock);
695 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
699 EXPORT_SYMBOL(ip_defrag);
701 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
706 if (skb->protocol != htons(ETH_P_IP))
709 if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
712 if (iph.ihl < 5 || iph.version != 4)
715 len = ntohs(iph.tot_len);
716 if (skb->len < len || len < (iph.ihl * 4))
719 if (ip_is_fragment(&iph)) {
720 skb = skb_share_check(skb, GFP_ATOMIC);
722 if (!pskb_may_pull(skb, iph.ihl*4))
724 if (pskb_trim_rcsum(skb, len))
726 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
727 if (ip_defrag(skb, user))
734 EXPORT_SYMBOL(ip_check_defrag);
739 static struct ctl_table ip4_frags_ns_ctl_table[] = {
741 .procname = "ipfrag_high_thresh",
742 .data = &init_net.ipv4.frags.high_thresh,
743 .maxlen = sizeof(int),
745 .proc_handler = proc_dointvec
748 .procname = "ipfrag_low_thresh",
749 .data = &init_net.ipv4.frags.low_thresh,
750 .maxlen = sizeof(int),
752 .proc_handler = proc_dointvec
755 .procname = "ipfrag_time",
756 .data = &init_net.ipv4.frags.timeout,
757 .maxlen = sizeof(int),
759 .proc_handler = proc_dointvec_jiffies,
764 static struct ctl_table ip4_frags_ctl_table[] = {
766 .procname = "ipfrag_secret_interval",
767 .data = &ip4_frags.secret_interval,
768 .maxlen = sizeof(int),
770 .proc_handler = proc_dointvec_jiffies,
773 .procname = "ipfrag_max_dist",
774 .data = &sysctl_ipfrag_max_dist,
775 .maxlen = sizeof(int),
777 .proc_handler = proc_dointvec_minmax,
783 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
785 struct ctl_table *table;
786 struct ctl_table_header *hdr;
788 table = ip4_frags_ns_ctl_table;
789 if (!net_eq(net, &init_net)) {
790 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
794 table[0].data = &net->ipv4.frags.high_thresh;
795 table[1].data = &net->ipv4.frags.low_thresh;
796 table[2].data = &net->ipv4.frags.timeout;
798 /* Don't export sysctls to unprivileged users */
799 if (net->user_ns != &init_user_ns)
800 table[0].procname = NULL;
803 hdr = register_net_sysctl(net, "net/ipv4", table);
807 net->ipv4.frags_hdr = hdr;
811 if (!net_eq(net, &init_net))
817 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
819 struct ctl_table *table;
821 table = net->ipv4.frags_hdr->ctl_table_arg;
822 unregister_net_sysctl_table(net->ipv4.frags_hdr);
826 static void ip4_frags_ctl_register(void)
828 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
831 static inline int ip4_frags_ns_ctl_register(struct net *net)
836 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
840 static inline void ip4_frags_ctl_register(void)
845 static int __net_init ipv4_frags_init_net(struct net *net)
847 /* Fragment cache limits.
849 * The fragment memory accounting code, (tries to) account for
850 * the real memory usage, by measuring both the size of frag
851 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
852 * and the SKB's truesize.
854 * A 64K fragment consumes 129736 bytes (44*2944)+200
855 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
857 * We will commit 4MB at one time. Should we cross that limit
858 * we will prune down to 3MB, making room for approx 8 big 64K
861 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
862 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
864 * Important NOTE! Fragment queue must be destroyed before MSL expires.
865 * RFC791 is wrong proposing to prolongate timer each fragment arrival
868 net->ipv4.frags.timeout = IP_FRAG_TIME;
870 inet_frags_init_net(&net->ipv4.frags);
872 return ip4_frags_ns_ctl_register(net);
875 static void __net_exit ipv4_frags_exit_net(struct net *net)
877 ip4_frags_ns_ctl_unregister(net);
878 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
881 static struct pernet_operations ip4_frags_ops = {
882 .init = ipv4_frags_init_net,
883 .exit = ipv4_frags_exit_net,
886 void __init ipfrag_init(void)
888 ip4_frags_ctl_register();
889 register_pernet_subsys(&ip4_frags_ops);
890 ip4_frags.hashfn = ip4_hashfn;
891 ip4_frags.constructor = ip4_frag_init;
892 ip4_frags.destructor = ip4_frag_free;
893 ip4_frags.skb_free = NULL;
894 ip4_frags.qsize = sizeof(struct ipq);
895 ip4_frags.match = ip4_frag_match;
896 ip4_frags.frag_expire = ip_expire;
897 ip4_frags.secret_interval = 10 * 60 * HZ;
898 inet_frags_init(&ip4_frags);