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>
38 #include <net/checksum.h>
39 #include <net/inetpeer.h>
40 #include <net/inet_frag.h>
41 #include <linux/tcp.h>
42 #include <linux/udp.h>
43 #include <linux/inet.h>
44 #include <linux/netfilter_ipv4.h>
46 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
47 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
48 * as well. Or notify me, at least. --ANK
51 static int sysctl_ipfrag_max_dist __read_mostly = 64;
55 struct inet_skb_parm h;
59 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
61 /* Describe an entry in the "incomplete datagrams" queue. */
63 struct inet_frag_queue q;
72 struct inet_peer *peer;
75 static struct inet_frags ip4_frags;
77 int ip_frag_nqueues(struct net *net)
79 return net->ipv4.frags.nqueues;
82 int ip_frag_mem(struct net *net)
84 return atomic_read(&net->ipv4.frags.mem);
87 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
88 struct net_device *dev);
90 struct ip4_create_arg {
95 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
97 return jhash_3words((__force u32)id << 16 | prot,
98 (__force u32)saddr, (__force u32)daddr,
99 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
102 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
106 ipq = container_of(q, struct ipq, q);
107 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
110 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
113 struct ip4_create_arg *arg = a;
115 qp = container_of(q, struct ipq, q);
116 return (qp->id == arg->iph->id &&
117 qp->saddr == arg->iph->saddr &&
118 qp->daddr == arg->iph->daddr &&
119 qp->protocol == arg->iph->protocol &&
120 qp->user == arg->user);
123 /* Memory Tracking Functions. */
124 static __inline__ void frag_kfree_skb(struct netns_frags *nf,
125 struct sk_buff *skb, int *work)
128 *work -= skb->truesize;
129 atomic_sub(skb->truesize, &nf->mem);
133 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
135 struct ipq *qp = container_of(q, struct ipq, q);
136 struct ip4_create_arg *arg = a;
138 qp->protocol = arg->iph->protocol;
139 qp->id = arg->iph->id;
140 qp->saddr = arg->iph->saddr;
141 qp->daddr = arg->iph->daddr;
142 qp->user = arg->user;
143 qp->peer = sysctl_ipfrag_max_dist ?
144 inet_getpeer(arg->iph->saddr, 1) : NULL;
147 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
151 qp = container_of(q, struct ipq, q);
153 inet_putpeer(qp->peer);
157 /* Destruction primitives. */
159 static __inline__ void ipq_put(struct ipq *ipq)
161 inet_frag_put(&ipq->q, &ip4_frags);
164 /* Kill ipq entry. It is not destroyed immediately,
165 * because caller (and someone more) holds reference count.
167 static void ipq_kill(struct ipq *ipq)
169 inet_frag_kill(&ipq->q, &ip4_frags);
172 /* Memory limiting on fragments. Evictor trashes the oldest
173 * fragment queue until we are back under the threshold.
175 static void ip_evictor(struct net *net)
179 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
181 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
185 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
187 static void ip_expire(unsigned long arg)
192 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
193 net = container_of(qp->q.net, struct net, ipv4.frags);
195 spin_lock(&qp->q.lock);
197 if (qp->q.last_in & INET_FRAG_COMPLETE)
202 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
203 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
205 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
206 struct sk_buff *head = qp->q.fragments;
208 /* Send an ICMP "Fragment Reassembly Timeout" message. */
209 if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) {
210 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
215 spin_unlock(&qp->q.lock);
219 /* Find the correct entry in the "incomplete datagrams" queue for
220 * this IP datagram, and create new one, if nothing is found.
222 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
224 struct inet_frag_queue *q;
225 struct ip4_create_arg arg;
231 read_lock(&ip4_frags.lock);
232 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
234 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
238 return container_of(q, struct ipq, q);
241 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
245 /* Is the fragment too far ahead to be part of ipq? */
246 static inline int ip_frag_too_far(struct ipq *qp)
248 struct inet_peer *peer = qp->peer;
249 unsigned int max = sysctl_ipfrag_max_dist;
250 unsigned int start, end;
258 end = atomic_inc_return(&peer->rid);
261 rc = qp->q.fragments && (end - start) > max;
266 net = container_of(qp->q.net, struct net, ipv4.frags);
267 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
273 static int ip_frag_reinit(struct ipq *qp)
277 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
278 atomic_inc(&qp->q.refcnt);
282 fp = qp->q.fragments;
284 struct sk_buff *xp = fp->next;
285 frag_kfree_skb(qp->q.net, fp, NULL);
292 qp->q.fragments = NULL;
298 /* Add new segment to existing queue. */
299 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
301 struct sk_buff *prev, *next;
302 struct net_device *dev;
307 if (qp->q.last_in & INET_FRAG_COMPLETE)
310 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
311 unlikely(ip_frag_too_far(qp)) &&
312 unlikely(err = ip_frag_reinit(qp))) {
317 offset = ntohs(ip_hdr(skb)->frag_off);
318 flags = offset & ~IP_OFFSET;
320 offset <<= 3; /* offset is in 8-byte chunks */
321 ihl = ip_hdrlen(skb);
323 /* Determine the position of this fragment. */
324 end = offset + skb->len - ihl;
327 /* Is this the final fragment? */
328 if ((flags & IP_MF) == 0) {
329 /* If we already have some bits beyond end
330 * or have different end, the segment is corrrupted.
332 if (end < qp->q.len ||
333 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
335 qp->q.last_in |= INET_FRAG_LAST_IN;
340 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
341 skb->ip_summed = CHECKSUM_NONE;
343 if (end > qp->q.len) {
344 /* Some bits beyond end -> corruption. */
345 if (qp->q.last_in & INET_FRAG_LAST_IN)
354 if (pskb_pull(skb, ihl) == NULL)
357 err = pskb_trim_rcsum(skb, end - offset);
361 /* Find out which fragments are in front and at the back of us
362 * in the chain of fragments so far. We must know where to put
363 * this fragment, right?
366 for (next = qp->q.fragments; next != NULL; next = next->next) {
367 if (FRAG_CB(next)->offset >= offset)
372 /* We found where to put this one. Check for overlap with
373 * preceding fragment, and, if needed, align things so that
374 * any overlaps are eliminated.
377 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
385 if (!pskb_pull(skb, i))
387 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
388 skb->ip_summed = CHECKSUM_NONE;
394 while (next && FRAG_CB(next)->offset < end) {
395 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
398 /* Eat head of the next overlapped fragment
399 * and leave the loop. The next ones cannot overlap.
401 if (!pskb_pull(next, i))
403 FRAG_CB(next)->offset += i;
405 if (next->ip_summed != CHECKSUM_UNNECESSARY)
406 next->ip_summed = CHECKSUM_NONE;
409 struct sk_buff *free_it = next;
411 /* Old fragment is completely overridden with
419 qp->q.fragments = next;
421 qp->q.meat -= free_it->len;
422 frag_kfree_skb(qp->q.net, free_it, NULL);
426 FRAG_CB(skb)->offset = offset;
428 /* Insert this fragment in the chain of fragments. */
433 qp->q.fragments = skb;
437 qp->iif = dev->ifindex;
440 qp->q.stamp = skb->tstamp;
441 qp->q.meat += skb->len;
442 atomic_add(skb->truesize, &qp->q.net->mem);
444 qp->q.last_in |= INET_FRAG_FIRST_IN;
446 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
447 qp->q.meat == qp->q.len)
448 return ip_frag_reasm(qp, prev, dev);
450 write_lock(&ip4_frags.lock);
451 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
452 write_unlock(&ip4_frags.lock);
461 /* Build a new IP datagram from all its fragments. */
463 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
464 struct net_device *dev)
466 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
468 struct sk_buff *fp, *head = qp->q.fragments;
475 /* Make the one we just received the head. */
478 fp = skb_clone(head, GFP_ATOMIC);
482 fp->next = head->next;
485 skb_morph(head, qp->q.fragments);
486 head->next = qp->q.fragments->next;
488 kfree_skb(qp->q.fragments);
489 qp->q.fragments = head;
492 WARN_ON(head == NULL);
493 WARN_ON(FRAG_CB(head)->offset != 0);
495 /* Allocate a new buffer for the datagram. */
496 ihlen = ip_hdrlen(head);
497 len = ihlen + qp->q.len;
503 /* Head of list must not be cloned. */
504 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
507 /* If the first fragment is fragmented itself, we split
508 * it to two chunks: the first with data and paged part
509 * and the second, holding only fragments. */
510 if (skb_has_frags(head)) {
511 struct sk_buff *clone;
514 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
516 clone->next = head->next;
518 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
519 skb_frag_list_init(head);
520 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
521 plen += skb_shinfo(head)->frags[i].size;
522 clone->len = clone->data_len = head->data_len - plen;
523 head->data_len -= clone->len;
524 head->len -= clone->len;
526 clone->ip_summed = head->ip_summed;
527 atomic_add(clone->truesize, &qp->q.net->mem);
530 skb_shinfo(head)->frag_list = head->next;
531 skb_push(head, head->data - skb_network_header(head));
532 atomic_sub(head->truesize, &qp->q.net->mem);
534 for (fp=head->next; fp; fp = fp->next) {
535 head->data_len += fp->len;
536 head->len += fp->len;
537 if (head->ip_summed != fp->ip_summed)
538 head->ip_summed = CHECKSUM_NONE;
539 else if (head->ip_summed == CHECKSUM_COMPLETE)
540 head->csum = csum_add(head->csum, fp->csum);
541 head->truesize += fp->truesize;
542 atomic_sub(fp->truesize, &qp->q.net->mem);
547 head->tstamp = qp->q.stamp;
551 iph->tot_len = htons(len);
552 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
553 qp->q.fragments = NULL;
557 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
563 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
566 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_REASMFAILS);
570 /* Process an incoming IP datagram fragment. */
571 int ip_defrag(struct sk_buff *skb, u32 user)
576 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
577 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
579 /* Start by cleaning up the memory. */
580 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
583 /* Lookup (or create) queue header */
584 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
587 spin_lock(&qp->q.lock);
589 ret = ip_frag_queue(qp, skb);
591 spin_unlock(&qp->q.lock);
596 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
604 static struct ctl_table ip4_frags_ns_ctl_table[] = {
606 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
607 .procname = "ipfrag_high_thresh",
608 .data = &init_net.ipv4.frags.high_thresh,
609 .maxlen = sizeof(int),
611 .proc_handler = proc_dointvec
614 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
615 .procname = "ipfrag_low_thresh",
616 .data = &init_net.ipv4.frags.low_thresh,
617 .maxlen = sizeof(int),
619 .proc_handler = proc_dointvec
622 .ctl_name = NET_IPV4_IPFRAG_TIME,
623 .procname = "ipfrag_time",
624 .data = &init_net.ipv4.frags.timeout,
625 .maxlen = sizeof(int),
627 .proc_handler = proc_dointvec_jiffies,
628 .strategy = sysctl_jiffies
633 static struct ctl_table ip4_frags_ctl_table[] = {
635 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
636 .procname = "ipfrag_secret_interval",
637 .data = &ip4_frags.secret_interval,
638 .maxlen = sizeof(int),
640 .proc_handler = proc_dointvec_jiffies,
641 .strategy = sysctl_jiffies
644 .procname = "ipfrag_max_dist",
645 .data = &sysctl_ipfrag_max_dist,
646 .maxlen = sizeof(int),
648 .proc_handler = proc_dointvec_minmax,
654 static int ip4_frags_ns_ctl_register(struct net *net)
656 struct ctl_table *table;
657 struct ctl_table_header *hdr;
659 table = ip4_frags_ns_ctl_table;
660 if (net != &init_net) {
661 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
665 table[0].data = &net->ipv4.frags.high_thresh;
666 table[1].data = &net->ipv4.frags.low_thresh;
667 table[2].data = &net->ipv4.frags.timeout;
670 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
674 net->ipv4.frags_hdr = hdr;
678 if (net != &init_net)
684 static void ip4_frags_ns_ctl_unregister(struct net *net)
686 struct ctl_table *table;
688 table = net->ipv4.frags_hdr->ctl_table_arg;
689 unregister_net_sysctl_table(net->ipv4.frags_hdr);
693 static void ip4_frags_ctl_register(void)
695 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
698 static inline int ip4_frags_ns_ctl_register(struct net *net)
703 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
707 static inline void ip4_frags_ctl_register(void)
712 static int ipv4_frags_init_net(struct net *net)
715 * Fragment cache limits. We will commit 256K at one time. Should we
716 * cross that limit we will prune down to 192K. This should cope with
717 * even the most extreme cases without allowing an attacker to
718 * measurably harm machine performance.
720 net->ipv4.frags.high_thresh = 256 * 1024;
721 net->ipv4.frags.low_thresh = 192 * 1024;
723 * Important NOTE! Fragment queue must be destroyed before MSL expires.
724 * RFC791 is wrong proposing to prolongate timer each fragment arrival
727 net->ipv4.frags.timeout = IP_FRAG_TIME;
729 inet_frags_init_net(&net->ipv4.frags);
731 return ip4_frags_ns_ctl_register(net);
734 static void ipv4_frags_exit_net(struct net *net)
736 ip4_frags_ns_ctl_unregister(net);
737 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
740 static struct pernet_operations ip4_frags_ops = {
741 .init = ipv4_frags_init_net,
742 .exit = ipv4_frags_exit_net,
745 void __init ipfrag_init(void)
747 ip4_frags_ctl_register();
748 register_pernet_subsys(&ip4_frags_ops);
749 ip4_frags.hashfn = ip4_hashfn;
750 ip4_frags.constructor = ip4_frag_init;
751 ip4_frags.destructor = ip4_frag_free;
752 ip4_frags.skb_free = NULL;
753 ip4_frags.qsize = sizeof(struct ipq);
754 ip4_frags.match = ip4_frag_match;
755 ip4_frags.frag_expire = ip_expire;
756 ip4_frags.secret_interval = 10 * 60 * HZ;
757 inet_frags_init(&ip4_frags);
760 EXPORT_SYMBOL(ip_defrag);