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.Cox@linux.org>
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(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)
191 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
193 spin_lock(&qp->q.lock);
195 if (qp->q.last_in & INET_FRAG_COMPLETE)
200 IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
201 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
203 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
204 struct sk_buff *head = qp->q.fragments;
207 net = container_of(qp->q.net, struct net, ipv4.frags);
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;
230 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
232 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
236 return container_of(q, struct ipq, q);
239 LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
243 /* Is the fragment too far ahead to be part of ipq? */
244 static inline int ip_frag_too_far(struct ipq *qp)
246 struct inet_peer *peer = qp->peer;
247 unsigned int max = sysctl_ipfrag_max_dist;
248 unsigned int start, end;
256 end = atomic_inc_return(&peer->rid);
259 rc = qp->q.fragments && (end - start) > max;
262 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
268 static int ip_frag_reinit(struct ipq *qp)
272 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
273 atomic_inc(&qp->q.refcnt);
277 fp = qp->q.fragments;
279 struct sk_buff *xp = fp->next;
280 frag_kfree_skb(qp->q.net, fp, NULL);
287 qp->q.fragments = NULL;
293 /* Add new segment to existing queue. */
294 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
296 struct sk_buff *prev, *next;
297 struct net_device *dev;
302 if (qp->q.last_in & INET_FRAG_COMPLETE)
305 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
306 unlikely(ip_frag_too_far(qp)) &&
307 unlikely(err = ip_frag_reinit(qp))) {
312 offset = ntohs(ip_hdr(skb)->frag_off);
313 flags = offset & ~IP_OFFSET;
315 offset <<= 3; /* offset is in 8-byte chunks */
316 ihl = ip_hdrlen(skb);
318 /* Determine the position of this fragment. */
319 end = offset + skb->len - ihl;
322 /* Is this the final fragment? */
323 if ((flags & IP_MF) == 0) {
324 /* If we already have some bits beyond end
325 * or have different end, the segment is corrrupted.
327 if (end < qp->q.len ||
328 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
330 qp->q.last_in |= INET_FRAG_LAST_IN;
335 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
336 skb->ip_summed = CHECKSUM_NONE;
338 if (end > qp->q.len) {
339 /* Some bits beyond end -> corruption. */
340 if (qp->q.last_in & INET_FRAG_LAST_IN)
349 if (pskb_pull(skb, ihl) == NULL)
352 err = pskb_trim_rcsum(skb, end - offset);
356 /* Find out which fragments are in front and at the back of us
357 * in the chain of fragments so far. We must know where to put
358 * this fragment, right?
361 for (next = qp->q.fragments; next != NULL; next = next->next) {
362 if (FRAG_CB(next)->offset >= offset)
367 /* We found where to put this one. Check for overlap with
368 * preceding fragment, and, if needed, align things so that
369 * any overlaps are eliminated.
372 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
380 if (!pskb_pull(skb, i))
382 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
383 skb->ip_summed = CHECKSUM_NONE;
389 while (next && FRAG_CB(next)->offset < end) {
390 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
393 /* Eat head of the next overlapped fragment
394 * and leave the loop. The next ones cannot overlap.
396 if (!pskb_pull(next, i))
398 FRAG_CB(next)->offset += i;
400 if (next->ip_summed != CHECKSUM_UNNECESSARY)
401 next->ip_summed = CHECKSUM_NONE;
404 struct sk_buff *free_it = next;
406 /* Old fragment is completely overridden with
414 qp->q.fragments = next;
416 qp->q.meat -= free_it->len;
417 frag_kfree_skb(qp->q.net, free_it, NULL);
421 FRAG_CB(skb)->offset = offset;
423 /* Insert this fragment in the chain of fragments. */
428 qp->q.fragments = skb;
432 qp->iif = dev->ifindex;
435 qp->q.stamp = skb->tstamp;
436 qp->q.meat += skb->len;
437 atomic_add(skb->truesize, &qp->q.net->mem);
439 qp->q.last_in |= INET_FRAG_FIRST_IN;
441 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
442 qp->q.meat == qp->q.len)
443 return ip_frag_reasm(qp, prev, dev);
445 write_lock(&ip4_frags.lock);
446 list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
447 write_unlock(&ip4_frags.lock);
456 /* Build a new IP datagram from all its fragments. */
458 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
459 struct net_device *dev)
462 struct sk_buff *fp, *head = qp->q.fragments;
469 /* Make the one we just received the head. */
472 fp = skb_clone(head, GFP_ATOMIC);
476 fp->next = head->next;
479 skb_morph(head, qp->q.fragments);
480 head->next = qp->q.fragments->next;
482 kfree_skb(qp->q.fragments);
483 qp->q.fragments = head;
486 BUG_TRAP(head != NULL);
487 BUG_TRAP(FRAG_CB(head)->offset == 0);
489 /* Allocate a new buffer for the datagram. */
490 ihlen = ip_hdrlen(head);
491 len = ihlen + qp->q.len;
497 /* Head of list must not be cloned. */
498 if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
501 /* If the first fragment is fragmented itself, we split
502 * it to two chunks: the first with data and paged part
503 * and the second, holding only fragments. */
504 if (skb_shinfo(head)->frag_list) {
505 struct sk_buff *clone;
508 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
510 clone->next = head->next;
512 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
513 skb_shinfo(head)->frag_list = NULL;
514 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
515 plen += skb_shinfo(head)->frags[i].size;
516 clone->len = clone->data_len = head->data_len - plen;
517 head->data_len -= clone->len;
518 head->len -= clone->len;
520 clone->ip_summed = head->ip_summed;
521 atomic_add(clone->truesize, &qp->q.net->mem);
524 skb_shinfo(head)->frag_list = head->next;
525 skb_push(head, head->data - skb_network_header(head));
526 atomic_sub(head->truesize, &qp->q.net->mem);
528 for (fp=head->next; fp; fp = fp->next) {
529 head->data_len += fp->len;
530 head->len += fp->len;
531 if (head->ip_summed != fp->ip_summed)
532 head->ip_summed = CHECKSUM_NONE;
533 else if (head->ip_summed == CHECKSUM_COMPLETE)
534 head->csum = csum_add(head->csum, fp->csum);
535 head->truesize += fp->truesize;
536 atomic_sub(fp->truesize, &qp->q.net->mem);
541 head->tstamp = qp->q.stamp;
545 iph->tot_len = htons(len);
546 IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
547 qp->q.fragments = NULL;
551 LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
558 "Oversized IP packet from " NIPQUAD_FMT ".\n",
561 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
565 /* Process an incoming IP datagram fragment. */
566 int ip_defrag(struct sk_buff *skb, u32 user)
571 IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
573 net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev);
574 /* Start by cleaning up the memory. */
575 if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
578 /* Lookup (or create) queue header */
579 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
582 spin_lock(&qp->q.lock);
584 ret = ip_frag_queue(qp, skb);
586 spin_unlock(&qp->q.lock);
591 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
599 static struct ctl_table ip4_frags_ns_ctl_table[] = {
601 .ctl_name = NET_IPV4_IPFRAG_HIGH_THRESH,
602 .procname = "ipfrag_high_thresh",
603 .data = &init_net.ipv4.frags.high_thresh,
604 .maxlen = sizeof(int),
606 .proc_handler = &proc_dointvec
609 .ctl_name = NET_IPV4_IPFRAG_LOW_THRESH,
610 .procname = "ipfrag_low_thresh",
611 .data = &init_net.ipv4.frags.low_thresh,
612 .maxlen = sizeof(int),
614 .proc_handler = &proc_dointvec
617 .ctl_name = NET_IPV4_IPFRAG_TIME,
618 .procname = "ipfrag_time",
619 .data = &init_net.ipv4.frags.timeout,
620 .maxlen = sizeof(int),
622 .proc_handler = &proc_dointvec_jiffies,
623 .strategy = &sysctl_jiffies
628 static struct ctl_table ip4_frags_ctl_table[] = {
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_ns_ctl_register(struct net *net)
651 struct ctl_table *table;
652 struct ctl_table_header *hdr;
654 table = ip4_frags_ns_ctl_table;
655 if (net != &init_net) {
656 table = kmemdup(table, sizeof(ip4_frags_ns_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;
665 hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
669 net->ipv4.frags_hdr = hdr;
673 if (net != &init_net)
679 static void ip4_frags_ns_ctl_unregister(struct net *net)
681 struct ctl_table *table;
683 table = net->ipv4.frags_hdr->ctl_table_arg;
684 unregister_net_sysctl_table(net->ipv4.frags_hdr);
688 static void ip4_frags_ctl_register(void)
690 register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
693 static inline int ip4_frags_ns_ctl_register(struct net *net)
698 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
702 static inline void ip4_frags_ctl_register(void)
707 static int ipv4_frags_init_net(struct net *net)
710 * Fragment cache limits. We will commit 256K at one time. Should we
711 * cross that limit we will prune down to 192K. This should cope with
712 * even the most extreme cases without allowing an attacker to
713 * measurably harm machine performance.
715 net->ipv4.frags.high_thresh = 256 * 1024;
716 net->ipv4.frags.low_thresh = 192 * 1024;
718 * Important NOTE! Fragment queue must be destroyed before MSL expires.
719 * RFC791 is wrong proposing to prolongate timer each fragment arrival
722 net->ipv4.frags.timeout = IP_FRAG_TIME;
724 inet_frags_init_net(&net->ipv4.frags);
726 return ip4_frags_ns_ctl_register(net);
729 static void ipv4_frags_exit_net(struct net *net)
731 ip4_frags_ns_ctl_unregister(net);
732 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
735 static struct pernet_operations ip4_frags_ops = {
736 .init = ipv4_frags_init_net,
737 .exit = ipv4_frags_exit_net,
740 void __init ipfrag_init(void)
742 ip4_frags_ctl_register();
743 register_pernet_subsys(&ip4_frags_ops);
744 ip4_frags.hashfn = ip4_hashfn;
745 ip4_frags.constructor = ip4_frag_init;
746 ip4_frags.destructor = ip4_frag_free;
747 ip4_frags.skb_free = NULL;
748 ip4_frags.qsize = sizeof(struct ipq);
749 ip4_frags.match = ip4_frag_match;
750 ip4_frags.frag_expire = ip_expire;
751 ip4_frags.secret_interval = 10 * 60 * HZ;
752 inet_frags_init(&ip4_frags);
755 EXPORT_SYMBOL(ip_defrag);