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 & COMPLETE)
202 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
203 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
205 if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
206 struct sk_buff *head = qp->q.fragments;
207 /* Send an ICMP "Fragment Reassembly Timeout" message. */
208 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
209 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
214 spin_unlock(&qp->q.lock);
218 /* Find the correct entry in the "incomplete datagrams" queue for
219 * this IP datagram, and create new one, if nothing is found.
221 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
223 struct inet_frag_queue *q;
224 struct ip4_create_arg arg;
229 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
231 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
235 return container_of(q, struct ipq, q);
238 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
242 /* Is the fragment too far ahead to be part of ipq? */
243 static inline int ip_frag_too_far(struct ipq *qp)
245 struct inet_peer *peer = qp->peer;
246 unsigned int max = sysctl_ipfrag_max_dist;
247 unsigned int start, end;
255 end = atomic_inc_return(&peer->rid);
258 rc = qp->q.fragments && (end - start) > max;
261 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
267 static int ip_frag_reinit(struct ipq *qp)
271 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
272 atomic_inc(&qp->q.refcnt);
276 fp = qp->q.fragments;
278 struct sk_buff *xp = fp->next;
279 frag_kfree_skb(qp->q.net, fp, NULL);
286 qp->q.fragments = NULL;
292 /* Add new segment to existing queue. */
293 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
295 struct sk_buff *prev, *next;
296 struct net_device *dev;
301 if (qp->q.last_in & COMPLETE)
304 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
305 unlikely(ip_frag_too_far(qp)) &&
306 unlikely(err = ip_frag_reinit(qp))) {
311 offset = ntohs(ip_hdr(skb)->frag_off);
312 flags = offset & ~IP_OFFSET;
314 offset <<= 3; /* offset is in 8-byte chunks */
315 ihl = ip_hdrlen(skb);
317 /* Determine the position of this fragment. */
318 end = offset + skb->len - ihl;
321 /* Is this the final fragment? */
322 if ((flags & IP_MF) == 0) {
323 /* If we already have some bits beyond end
324 * or have different end, the segment is corrrupted.
326 if (end < qp->q.len ||
327 ((qp->q.last_in & LAST_IN) && end != qp->q.len))
329 qp->q.last_in |= LAST_IN;
334 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
335 skb->ip_summed = CHECKSUM_NONE;
337 if (end > qp->q.len) {
338 /* Some bits beyond end -> corruption. */
339 if (qp->q.last_in & LAST_IN)
348 if (pskb_pull(skb, ihl) == NULL)
351 err = pskb_trim_rcsum(skb, end - offset);
355 /* Find out which fragments are in front and at the back of us
356 * in the chain of fragments so far. We must know where to put
357 * this fragment, right?
360 for (next = qp->q.fragments; next != NULL; next = next->next) {
361 if (FRAG_CB(next)->offset >= offset)
366 /* We found where to put this one. Check for overlap with
367 * preceding fragment, and, if needed, align things so that
368 * any overlaps are eliminated.
371 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
379 if (!pskb_pull(skb, i))
381 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
382 skb->ip_summed = CHECKSUM_NONE;
388 while (next && FRAG_CB(next)->offset < end) {
389 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
392 /* Eat head of the next overlapped fragment
393 * and leave the loop. The next ones cannot overlap.
395 if (!pskb_pull(next, i))
397 FRAG_CB(next)->offset += i;
399 if (next->ip_summed != CHECKSUM_UNNECESSARY)
400 next->ip_summed = CHECKSUM_NONE;
403 struct sk_buff *free_it = next;
405 /* Old fragment is completely overridden with
413 qp->q.fragments = next;
415 qp->q.meat -= free_it->len;
416 frag_kfree_skb(qp->q.net, free_it, NULL);
420 FRAG_CB(skb)->offset = offset;
422 /* Insert this fragment in the chain of fragments. */
427 qp->q.fragments = skb;
431 qp->iif = dev->ifindex;
434 qp->q.stamp = skb->tstamp;
435 qp->q.meat += skb->len;
436 atomic_add(skb->truesize, &qp->q.net->mem);
438 qp->q.last_in |= FIRST_IN;
440 if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
441 return ip_frag_reasm(qp, prev, dev);
443 write_lock(&ip4_frags.lock);
444 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
445 write_unlock(&ip4_frags.lock);
454 /* Build a new IP datagram from all its fragments. */
456 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
457 struct net_device *dev)
460 struct sk_buff *fp, *head = qp->q.fragments;
467 /* Make the one we just received the head. */
470 fp = skb_clone(head, GFP_ATOMIC);
474 fp->next = head->next;
477 skb_morph(head, qp->q.fragments);
478 head->next = qp->q.fragments->next;
480 kfree_skb(qp->q.fragments);
481 qp->q.fragments = head;
484 BUG_TRAP(head != NULL);
485 BUG_TRAP(FRAG_CB(head)->offset == 0);
487 /* Allocate a new buffer for the datagram. */
488 ihlen = ip_hdrlen(head);
489 len = ihlen + qp->q.len;
495 /* Head of list must not be cloned. */
496 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
499 /* If the first fragment is fragmented itself, we split
500 * it to two chunks: the first with data and paged part
501 * and the second, holding only fragments. */
502 if (skb_shinfo(head)->frag_list) {
503 struct sk_buff *clone;
506 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
508 clone->next = head->next;
510 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
511 skb_shinfo(head)->frag_list = NULL;
512 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
513 plen += skb_shinfo(head)->frags[i].size;
514 clone->len = clone->data_len = head->data_len - plen;
515 head->data_len -= clone->len;
516 head->len -= clone->len;
518 clone->ip_summed = head->ip_summed;
519 atomic_add(clone->truesize, &qp->q.net->mem);
522 skb_shinfo(head)->frag_list = head->next;
523 skb_push(head, head->data - skb_network_header(head));
524 atomic_sub(head->truesize, &qp->q.net->mem);
526 for (fp=head->next; fp; fp = fp->next) {
527 head->data_len += fp->len;
528 head->len += fp->len;
529 if (head->ip_summed != fp->ip_summed)
530 head->ip_summed = CHECKSUM_NONE;
531 else if (head->ip_summed == CHECKSUM_COMPLETE)
532 head->csum = csum_add(head->csum, fp->csum);
533 head->truesize += fp->truesize;
534 atomic_sub(fp->truesize, &qp->q.net->mem);
539 head->tstamp = qp->q.stamp;
543 iph->tot_len = htons(len);
544 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
545 qp->q.fragments = NULL;
549 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
556 "Oversized IP packet from %d.%d.%d.%d.\n",
559 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
563 /* Process an incoming IP datagram fragment. */
564 int ip_defrag(struct sk_buff *skb, u32 user)
569 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
571 net = skb->dev ? skb->dev->nd_net : skb->dst->dev->nd_net;
572 /* Start by cleaning up the memory. */
573 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
576 /* Lookup (or create) queue header */
577 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
580 spin_lock(&qp->q.lock);
582 ret = ip_frag_queue(qp, skb);
584 spin_unlock(&qp->q.lock);
589 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
597 static struct ctl_table ip4_frags_ctl_table[] = {
599 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
600 .procname = "ipfrag_high_thresh",
601 .data = &init_net.ipv4.frags.high_thresh,
602 .maxlen = sizeof(int),
604 .proc_handler = &proc_dointvec
607 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
608 .procname = "ipfrag_low_thresh",
609 .data = &init_net.ipv4.frags.low_thresh,
610 .maxlen = sizeof(int),
612 .proc_handler = &proc_dointvec
615 .ctl_name = NET_IPV4_IPFRAG_TIME,
616 .procname = "ipfrag_time",
617 .data = &init_net.ipv4.frags.timeout,
618 .maxlen = sizeof(int),
620 .proc_handler = &proc_dointvec_jiffies,
621 .strategy = &sysctl_jiffies
624 .ctl_name = NET_IPV4_IPFRAG_SECRET_INTERVAL,
625 .procname = "ipfrag_secret_interval",
626 .data = &ip4_frags.secret_interval,
627 .maxlen = sizeof(int),
629 .proc_handler = &proc_dointvec_jiffies,
630 .strategy = &sysctl_jiffies
633 .procname = "ipfrag_max_dist",
634 .data = &sysctl_ipfrag_max_dist,
635 .maxlen = sizeof(int),
637 .proc_handler = &proc_dointvec_minmax,
643 static int ip4_frags_ctl_register(struct net *net)
645 struct ctl_table *table;
646 struct ctl_table_header *hdr;
648 table = ip4_frags_ctl_table;
649 if (net != &init_net) {
650 table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL);
654 table[0].data = &net->ipv4.frags.high_thresh;
655 table[1].data = &net->ipv4.frags.low_thresh;
656 table[2].data = &net->ipv4.frags.timeout;
657 table[3].mode &= ~0222;
658 table[4].mode &= ~0222;
661 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
665 net->ipv4.frags_hdr = hdr;
669 if (net != &init_net)
675 static void ip4_frags_ctl_unregister(struct net *net)
677 struct ctl_table *table;
679 table = net->ipv4.frags_hdr->ctl_table_arg;
680 unregister_net_sysctl_table(net->ipv4.frags_hdr);
684 static inline int ip4_frags_ctl_register(struct net *net)
689 static inline void ip4_frags_ctl_unregister(struct net *net)
694 static int ipv4_frags_init_net(struct net *net)
697 * Fragment cache limits. We will commit 256K at one time. Should we
698 * cross that limit we will prune down to 192K. This should cope with
699 * even the most extreme cases without allowing an attacker to
700 * measurably harm machine performance.
702 net->ipv4.frags.high_thresh = 256 * 1024;
703 net->ipv4.frags.low_thresh = 192 * 1024;
705 * Important NOTE! Fragment queue must be destroyed before MSL expires.
706 * RFC791 is wrong proposing to prolongate timer each fragment arrival
709 net->ipv4.frags.timeout = IP_FRAG_TIME;
711 inet_frags_init_net(&net->ipv4.frags);
713 return ip4_frags_ctl_register(net);
716 static void ipv4_frags_exit_net(struct net *net)
718 ip4_frags_ctl_unregister(net);
719 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
722 static struct pernet_operations ip4_frags_ops = {
723 .init = ipv4_frags_init_net,
724 .exit = ipv4_frags_exit_net,
727 void __init ipfrag_init(void)
729 register_pernet_subsys(&ip4_frags_ops);
730 ip4_frags.hashfn = ip4_hashfn;
731 ip4_frags.constructor = ip4_frag_init;
732 ip4_frags.destructor = ip4_frag_free;
733 ip4_frags.skb_free = NULL;
734 ip4_frags.qsize = sizeof(struct ipq);
735 ip4_frags.match = ip4_frag_match;
736 ip4_frags.frag_expire = ip_expire;
737 ip4_frags.secret_interval = 10 * 60 * HZ;
738 inet_frags_init(&ip4_frags);
741 EXPORT_SYMBOL(ip_defrag);