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 * Version: $Id: ip_fragment.c,v 1.59 2002/01/12 07:54:56 davem Exp $
10 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox <Alan.Cox@linux.org>
14 * Alan Cox : Split from ip.c , see ip_input.c for history.
15 * David S. Miller : Begin massive cleanup...
16 * Andi Kleen : Add sysctls.
17 * xxxx : Overlapfrag bug.
18 * Ultima : ip_expire() kernel panic.
19 * Bill Hawes : Frag accounting and evictor fixes.
20 * John McDonald : 0 length frag bug.
21 * Alexey Kuznetsov: SMP races, threading, cleanup.
22 * Patrick McHardy : LRU queue of frag heads for evictor.
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>
40 #include <net/checksum.h>
41 #include <net/inetpeer.h>
42 #include <net/inet_frag.h>
43 #include <linux/tcp.h>
44 #include <linux/udp.h>
45 #include <linux/inet.h>
46 #include <linux/netfilter_ipv4.h>
48 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
49 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
50 * as well. Or notify me, at least. --ANK
53 static int sysctl_ipfrag_max_dist __read_mostly = 64;
57 struct inet_skb_parm h;
61 #define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb))
63 /* Describe an entry in the "incomplete datagrams" queue. */
65 struct inet_frag_queue q;
74 struct inet_peer *peer;
77 static struct inet_frags ip4_frags;
79 int ip_frag_nqueues(struct net *net)
81 return net->ipv4.frags.nqueues;
84 int ip_frag_mem(struct net *net)
86 return atomic_read(&net->ipv4.frags.mem);
89 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
90 struct net_device *dev);
92 struct ip4_create_arg {
97 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
99 return jhash_3words((__force u32)id << 16 | prot,
100 (__force u32)saddr, (__force u32)daddr,
101 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
104 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
108 ipq = container_of(q, struct ipq, q);
109 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
112 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
115 struct ip4_create_arg *arg = a;
117 qp = container_of(q, struct ipq, q);
118 return (qp->id == arg->iph->id &&
119 qp->saddr == arg->iph->saddr &&
120 qp->daddr == arg->iph->daddr &&
121 qp->protocol == arg->iph->protocol &&
122 qp->user == arg->user);
125 /* Memory Tracking Functions. */
126 static __inline__ void frag_kfree_skb(struct netns_frags *nf,
127 struct sk_buff *skb, int *work)
130 *work -= skb->truesize;
131 atomic_sub(skb->truesize, &nf->mem);
135 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
137 struct ipq *qp = container_of(q, struct ipq, q);
138 struct ip4_create_arg *arg = a;
140 qp->protocol = arg->iph->protocol;
141 qp->id = arg->iph->id;
142 qp->saddr = arg->iph->saddr;
143 qp->daddr = arg->iph->daddr;
144 qp->user = arg->user;
145 qp->peer = sysctl_ipfrag_max_dist ?
146 inet_getpeer(arg->iph->saddr, 1) : NULL;
149 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
153 qp = container_of(q, struct ipq, q);
155 inet_putpeer(qp->peer);
159 /* Destruction primitives. */
161 static __inline__ void ipq_put(struct ipq *ipq)
163 inet_frag_put(&ipq->q, &ip4_frags);
166 /* Kill ipq entry. It is not destroyed immediately,
167 * because caller (and someone more) holds reference count.
169 static void ipq_kill(struct ipq *ipq)
171 inet_frag_kill(&ipq->q, &ip4_frags);
174 /* Memory limiting on fragments. Evictor trashes the oldest
175 * fragment queue until we are back under the threshold.
177 static void ip_evictor(struct net *net)
181 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
183 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
187 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
189 static void ip_expire(unsigned long arg)
193 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
195 spin_lock(&qp->q.lock);
197 if (qp->q.last_in & INET_FRAG_COMPLETE)
202 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
203 IP_INC_STATS_BH(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;
209 net = container_of(qp->q.net, struct net, ipv4.frags);
210 /* Send an ICMP "Fragment Reassembly Timeout" message. */
211 if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) {
212 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
217 spin_unlock(&qp->q.lock);
221 /* Find the correct entry in the "incomplete datagrams" queue for
222 * this IP datagram, and create new one, if nothing is found.
224 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
226 struct inet_frag_queue *q;
227 struct ip4_create_arg arg;
233 read_lock(&ip4_frags.lock);
234 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
236 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
240 return container_of(q, struct ipq, q);
243 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
247 /* Is the fragment too far ahead to be part of ipq? */
248 static inline int ip_frag_too_far(struct ipq *qp)
250 struct inet_peer *peer = qp->peer;
251 unsigned int max = sysctl_ipfrag_max_dist;
252 unsigned int start, end;
260 end = atomic_inc_return(&peer->rid);
263 rc = qp->q.fragments && (end - start) > max;
266 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
272 static int ip_frag_reinit(struct ipq *qp)
276 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
277 atomic_inc(&qp->q.refcnt);
281 fp = qp->q.fragments;
283 struct sk_buff *xp = fp->next;
284 frag_kfree_skb(qp->q.net, fp, NULL);
291 qp->q.fragments = NULL;
297 /* Add new segment to existing queue. */
298 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
300 struct sk_buff *prev, *next;
301 struct net_device *dev;
306 if (qp->q.last_in & INET_FRAG_COMPLETE)
309 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
310 unlikely(ip_frag_too_far(qp)) &&
311 unlikely(err = ip_frag_reinit(qp))) {
316 offset = ntohs(ip_hdr(skb)->frag_off);
317 flags = offset & ~IP_OFFSET;
319 offset <<= 3; /* offset is in 8-byte chunks */
320 ihl = ip_hdrlen(skb);
322 /* Determine the position of this fragment. */
323 end = offset + skb->len - ihl;
326 /* Is this the final fragment? */
327 if ((flags & IP_MF) == 0) {
328 /* If we already have some bits beyond end
329 * or have different end, the segment is corrrupted.
331 if (end < qp->q.len ||
332 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
334 qp->q.last_in |= INET_FRAG_LAST_IN;
339 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
340 skb->ip_summed = CHECKSUM_NONE;
342 if (end > qp->q.len) {
343 /* Some bits beyond end -> corruption. */
344 if (qp->q.last_in & INET_FRAG_LAST_IN)
353 if (pskb_pull(skb, ihl) == NULL)
356 err = pskb_trim_rcsum(skb, end - offset);
360 /* Find out which fragments are in front and at the back of us
361 * in the chain of fragments so far. We must know where to put
362 * this fragment, right?
365 for (next = qp->q.fragments; next != NULL; next = next->next) {
366 if (FRAG_CB(next)->offset >= offset)
371 /* We found where to put this one. Check for overlap with
372 * preceding fragment, and, if needed, align things so that
373 * any overlaps are eliminated.
376 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
384 if (!pskb_pull(skb, i))
386 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
387 skb->ip_summed = CHECKSUM_NONE;
393 while (next && FRAG_CB(next)->offset < end) {
394 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
397 /* Eat head of the next overlapped fragment
398 * and leave the loop. The next ones cannot overlap.
400 if (!pskb_pull(next, i))
402 FRAG_CB(next)->offset += i;
404 if (next->ip_summed != CHECKSUM_UNNECESSARY)
405 next->ip_summed = CHECKSUM_NONE;
408 struct sk_buff *free_it = next;
410 /* Old fragment is completely overridden with
418 qp->q.fragments = next;
420 qp->q.meat -= free_it->len;
421 frag_kfree_skb(qp->q.net, free_it, NULL);
425 FRAG_CB(skb)->offset = offset;
427 /* Insert this fragment in the chain of fragments. */
432 qp->q.fragments = skb;
436 qp->iif = dev->ifindex;
439 qp->q.stamp = skb->tstamp;
440 qp->q.meat += skb->len;
441 atomic_add(skb->truesize, &qp->q.net->mem);
443 qp->q.last_in |= INET_FRAG_FIRST_IN;
445 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
446 qp->q.meat == qp->q.len)
447 return ip_frag_reasm(qp, prev, dev);
449 write_lock(&ip4_frags.lock);
450 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
451 write_unlock(&ip4_frags.lock);
460 /* Build a new IP datagram from all its fragments. */
462 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
463 struct net_device *dev)
466 struct sk_buff *fp, *head = qp->q.fragments;
473 /* Make the one we just received the head. */
476 fp = skb_clone(head, GFP_ATOMIC);
480 fp->next = head->next;
483 skb_morph(head, qp->q.fragments);
484 head->next = qp->q.fragments->next;
486 kfree_skb(qp->q.fragments);
487 qp->q.fragments = head;
490 BUG_TRAP(head != NULL);
491 BUG_TRAP(FRAG_CB(head)->offset == 0);
493 /* Allocate a new buffer for the datagram. */
494 ihlen = ip_hdrlen(head);
495 len = ihlen + qp->q.len;
501 /* Head of list must not be cloned. */
502 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
505 /* If the first fragment is fragmented itself, we split
506 * it to two chunks: the first with data and paged part
507 * and the second, holding only fragments. */
508 if (skb_shinfo(head)->frag_list) {
509 struct sk_buff *clone;
512 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
514 clone->next = head->next;
516 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
517 skb_shinfo(head)->frag_list = NULL;
518 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
519 plen += skb_shinfo(head)->frags[i].size;
520 clone->len = clone->data_len = head->data_len - plen;
521 head->data_len -= clone->len;
522 head->len -= clone->len;
524 clone->ip_summed = head->ip_summed;
525 atomic_add(clone->truesize, &qp->q.net->mem);
528 skb_shinfo(head)->frag_list = head->next;
529 skb_push(head, head->data - skb_network_header(head));
530 atomic_sub(head->truesize, &qp->q.net->mem);
532 for (fp=head->next; fp; fp = fp->next) {
533 head->data_len += fp->len;
534 head->len += fp->len;
535 if (head->ip_summed != fp->ip_summed)
536 head->ip_summed = CHECKSUM_NONE;
537 else if (head->ip_summed == CHECKSUM_COMPLETE)
538 head->csum = csum_add(head->csum, fp->csum);
539 head->truesize += fp->truesize;
540 atomic_sub(fp->truesize, &qp->q.net->mem);
545 head->tstamp = qp->q.stamp;
549 iph->tot_len = htons(len);
550 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
551 qp->q.fragments = NULL;
555 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
562 "Oversized IP packet from " NIPQUAD_FMT ".\n",
565 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
569 /* Process an incoming IP datagram fragment. */
570 int ip_defrag(struct sk_buff *skb, u32 user)
575 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
577 net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev);
578 /* Start by cleaning up the memory. */
579 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
582 /* Lookup (or create) queue header */
583 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
586 spin_lock(&qp->q.lock);
588 ret = ip_frag_queue(qp, skb);
590 spin_unlock(&qp->q.lock);
595 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
603 static struct ctl_table ip4_frags_ctl_table[] = {
605 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
606 .procname = "ipfrag_high_thresh",
607 .data = &init_net.ipv4.frags.high_thresh,
608 .maxlen = sizeof(int),
610 .proc_handler = &proc_dointvec
613 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
614 .procname = "ipfrag_low_thresh",
615 .data = &init_net.ipv4.frags.low_thresh,
616 .maxlen = sizeof(int),
618 .proc_handler = &proc_dointvec
621 .ctl_name = NET_IPV4_IPFRAG_TIME,
622 .procname = "ipfrag_time",
623 .data = &init_net.ipv4.frags.timeout,
624 .maxlen = sizeof(int),
626 .proc_handler = &proc_dointvec_jiffies,
627 .strategy = &sysctl_jiffies
630 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
631 .procname = "ipfrag_secret_interval",
632 .data = &ip4_frags.secret_interval,
633 .maxlen = sizeof(int),
635 .proc_handler = &proc_dointvec_jiffies,
636 .strategy = &sysctl_jiffies
639 .procname = "ipfrag_max_dist",
640 .data = &sysctl_ipfrag_max_dist,
641 .maxlen = sizeof(int),
643 .proc_handler = &proc_dointvec_minmax,
649 static int ip4_frags_ctl_register(struct net *net)
651 struct ctl_table *table;
652 struct ctl_table_header *hdr;
654 table = ip4_frags_ctl_table;
655 if (net != &init_net) {
656 table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL);
660 table[0].data = &net->ipv4.frags.high_thresh;
661 table[1].data = &net->ipv4.frags.low_thresh;
662 table[2].data = &net->ipv4.frags.timeout;
663 table[3].mode &= ~0222;
664 table[4].mode &= ~0222;
667 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
671 net->ipv4.frags_hdr = hdr;
675 if (net != &init_net)
681 static void ip4_frags_ctl_unregister(struct net *net)
683 struct ctl_table *table;
685 table = net->ipv4.frags_hdr->ctl_table_arg;
686 unregister_net_sysctl_table(net->ipv4.frags_hdr);
690 static inline int ip4_frags_ctl_register(struct net *net)
695 static inline void ip4_frags_ctl_unregister(struct net *net)
700 static int ipv4_frags_init_net(struct net *net)
703 * Fragment cache limits. We will commit 256K at one time. Should we
704 * cross that limit we will prune down to 192K. This should cope with
705 * even the most extreme cases without allowing an attacker to
706 * measurably harm machine performance.
708 net->ipv4.frags.high_thresh = 256 * 1024;
709 net->ipv4.frags.low_thresh = 192 * 1024;
711 * Important NOTE! Fragment queue must be destroyed before MSL expires.
712 * RFC791 is wrong proposing to prolongate timer each fragment arrival
715 net->ipv4.frags.timeout = IP_FRAG_TIME;
717 inet_frags_init_net(&net->ipv4.frags);
719 return ip4_frags_ctl_register(net);
722 static void ipv4_frags_exit_net(struct net *net)
724 ip4_frags_ctl_unregister(net);
725 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
728 static struct pernet_operations ip4_frags_ops = {
729 .init = ipv4_frags_init_net,
730 .exit = ipv4_frags_exit_net,
733 void __init ipfrag_init(void)
735 register_pernet_subsys(&ip4_frags_ops);
736 ip4_frags.hashfn = ip4_hashfn;
737 ip4_frags.constructor = ip4_frag_init;
738 ip4_frags.destructor = ip4_frag_free;
739 ip4_frags.skb_free = NULL;
740 ip4_frags.qsize = sizeof(struct ipq);
741 ip4_frags.match = ip4_frag_match;
742 ip4_frags.frag_expire = ip_expire;
743 ip4_frags.secret_interval = 10 * 60 * HZ;
744 inet_frags_init(&ip4_frags);
747 EXPORT_SYMBOL(ip_defrag);