1 // SPDX-License-Identifier: GPL-2.0
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * The IP fragmentation functionality.
9 * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox <alan@lxorguk.ukuu.org.uk>
13 * Alan Cox : Split from ip.c , see ip_input.c for history.
14 * David S. Miller : Begin massive cleanup...
15 * Andi Kleen : Add sysctls.
16 * xxxx : Overlapfrag bug.
17 * Ultima : ip_expire() kernel panic.
18 * Bill Hawes : Frag accounting and evictor fixes.
19 * John McDonald : 0 length frag bug.
20 * Alexey Kuznetsov: SMP races, threading, cleanup.
21 * Patrick McHardy : LRU queue of frag heads for evictor.
24 #define pr_fmt(fmt) "IPv4: " fmt
26 #include <linux/compiler.h>
27 #include <linux/module.h>
28 #include <linux/types.h>
30 #include <linux/jiffies.h>
31 #include <linux/skbuff.h>
32 #include <linux/list.h>
34 #include <linux/icmp.h>
35 #include <linux/netdevice.h>
36 #include <linux/jhash.h>
37 #include <linux/random.h>
38 #include <linux/slab.h>
39 #include <net/route.h>
44 #include <net/checksum.h>
45 #include <net/inetpeer.h>
46 #include <net/inet_frag.h>
47 #include <linux/tcp.h>
48 #include <linux/udp.h>
49 #include <linux/inet.h>
50 #include <linux/netfilter_ipv4.h>
51 #include <net/inet_ecn.h>
52 #include <net/l3mdev.h>
54 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
55 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
56 * as well. Or notify me, at least. --ANK
58 static const char ip_frag_cache_name[] = "ip4-frags";
60 /* Describe an entry in the "incomplete datagrams" queue. */
62 struct inet_frag_queue q;
64 u8 ecn; /* RFC3168 support */
65 u16 max_df_size; /* largest frag with DF set seen */
68 struct inet_peer *peer;
71 static u8 ip4_frag_ecn(u8 tos)
73 return 1 << (tos & INET_ECN_MASK);
76 static struct inet_frags ip4_frags;
78 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
79 struct sk_buff *prev_tail, struct net_device *dev);
82 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
84 struct ipq *qp = container_of(q, struct ipq, q);
85 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
87 struct net *net = container_of(ipv4, struct net, ipv4);
89 const struct frag_v4_compare_key *key = a;
93 qp->peer = q->net->max_dist ?
94 inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
98 static void ip4_frag_free(struct inet_frag_queue *q)
102 qp = container_of(q, struct ipq, q);
104 inet_putpeer(qp->peer);
108 /* Destruction primitives. */
110 static void ipq_put(struct ipq *ipq)
112 inet_frag_put(&ipq->q);
115 /* Kill ipq entry. It is not destroyed immediately,
116 * because caller (and someone more) holds reference count.
118 static void ipq_kill(struct ipq *ipq)
120 inet_frag_kill(&ipq->q);
123 static bool frag_expire_skip_icmp(u32 user)
125 return user == IP_DEFRAG_AF_PACKET ||
126 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
127 __IP_DEFRAG_CONNTRACK_IN_END) ||
128 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
129 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
133 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
135 static void ip_expire(struct timer_list *t)
137 struct inet_frag_queue *frag = from_timer(frag, t, timer);
138 const struct iphdr *iph;
139 struct sk_buff *head = NULL;
144 qp = container_of(frag, struct ipq, q);
145 net = container_of(qp->q.net, struct net, ipv4.frags);
148 spin_lock(&qp->q.lock);
150 if (qp->q.flags & INET_FRAG_COMPLETE)
154 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
155 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
157 if (!(qp->q.flags & INET_FRAG_FIRST_IN))
160 /* sk_buff::dev and sk_buff::rbnode are unionized. So we
161 * pull the head out of the tree in order to be able to
162 * deal with head->dev.
164 head = inet_frag_pull_head(&qp->q);
167 head->dev = dev_get_by_index_rcu(net, qp->iif);
172 /* skb has no dst, perform route lookup again */
174 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
175 iph->tos, head->dev);
179 /* Only an end host needs to send an ICMP
180 * "Fragment Reassembly Timeout" message, per RFC792.
182 if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
183 (skb_rtable(head)->rt_type != RTN_LOCAL))
186 spin_unlock(&qp->q.lock);
187 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
191 spin_unlock(&qp->q.lock);
199 /* Find the correct entry in the "incomplete datagrams" queue for
200 * this IP datagram, and create new one, if nothing is found.
202 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
205 struct frag_v4_compare_key key = {
211 .protocol = iph->protocol,
213 struct inet_frag_queue *q;
215 q = inet_frag_find(&net->ipv4.frags, &key);
219 return container_of(q, struct ipq, q);
222 /* Is the fragment too far ahead to be part of ipq? */
223 static int ip_frag_too_far(struct ipq *qp)
225 struct inet_peer *peer = qp->peer;
226 unsigned int max = qp->q.net->max_dist;
227 unsigned int start, end;
235 end = atomic_inc_return(&peer->rid);
238 rc = qp->q.fragments_tail && (end - start) > max;
243 net = container_of(qp->q.net, struct net, ipv4.frags);
244 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
250 static int ip_frag_reinit(struct ipq *qp)
252 unsigned int sum_truesize = 0;
254 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
255 refcount_inc(&qp->q.refcnt);
259 sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments);
260 sub_frag_mem_limit(qp->q.net, sum_truesize);
265 qp->q.fragments = NULL;
266 qp->q.rb_fragments = RB_ROOT;
267 qp->q.fragments_tail = NULL;
268 qp->q.last_run_head = NULL;
275 /* Add new segment to existing queue. */
276 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
278 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
279 int ihl, end, flags, offset;
280 struct sk_buff *prev_tail;
281 struct net_device *dev;
282 unsigned int fragsize;
286 if (qp->q.flags & INET_FRAG_COMPLETE)
289 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
290 unlikely(ip_frag_too_far(qp)) &&
291 unlikely(err = ip_frag_reinit(qp))) {
296 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
297 offset = ntohs(ip_hdr(skb)->frag_off);
298 flags = offset & ~IP_OFFSET;
300 offset <<= 3; /* offset is in 8-byte chunks */
301 ihl = ip_hdrlen(skb);
303 /* Determine the position of this fragment. */
304 end = offset + skb->len - skb_network_offset(skb) - ihl;
307 /* Is this the final fragment? */
308 if ((flags & IP_MF) == 0) {
309 /* If we already have some bits beyond end
310 * or have different end, the segment is corrupted.
312 if (end < qp->q.len ||
313 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
315 qp->q.flags |= INET_FRAG_LAST_IN;
320 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
321 skb->ip_summed = CHECKSUM_NONE;
323 if (end > qp->q.len) {
324 /* Some bits beyond end -> corruption. */
325 if (qp->q.flags & INET_FRAG_LAST_IN)
334 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
337 err = pskb_trim_rcsum(skb, end - offset);
341 /* Note : skb->rbnode and skb->dev share the same location. */
343 /* Makes sure compiler wont do silly aliasing games */
346 prev_tail = qp->q.fragments_tail;
347 err = inet_frag_queue_insert(&qp->q, skb, offset, end);
352 qp->iif = dev->ifindex;
354 qp->q.stamp = skb->tstamp;
355 qp->q.meat += skb->len;
357 add_frag_mem_limit(qp->q.net, skb->truesize);
359 qp->q.flags |= INET_FRAG_FIRST_IN;
361 fragsize = skb->len + ihl;
363 if (fragsize > qp->q.max_size)
364 qp->q.max_size = fragsize;
366 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
367 fragsize > qp->max_df_size)
368 qp->max_df_size = fragsize;
370 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
371 qp->q.meat == qp->q.len) {
372 unsigned long orefdst = skb->_skb_refdst;
374 skb->_skb_refdst = 0UL;
375 err = ip_frag_reasm(qp, skb, prev_tail, dev);
376 skb->_skb_refdst = orefdst;
378 inet_frag_kill(&qp->q);
386 if (err == IPFRAG_DUP) {
391 __IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
393 inet_frag_kill(&qp->q);
394 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
400 /* Build a new IP datagram from all its fragments. */
401 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
402 struct sk_buff *prev_tail, struct net_device *dev)
404 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
412 ecn = ip_frag_ecn_table[qp->ecn];
413 if (unlikely(ecn == 0xff)) {
418 /* Make the one we just received the head. */
419 reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
423 len = ip_hdrlen(skb) + qp->q.len;
428 inet_frag_reasm_finish(&qp->q, skb, reasm_data);
431 IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
434 iph->tot_len = htons(len);
437 /* When we set IP_DF on a refragmented skb we must also force a
438 * call to ip_fragment to avoid forwarding a DF-skb of size s while
439 * original sender only sent fragments of size f (where f < s).
441 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
442 * frag seen to avoid sending tiny DF-fragments in case skb was built
443 * from one very small df-fragment and one large non-df frag.
445 if (qp->max_df_size == qp->q.max_size) {
446 IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
447 iph->frag_off = htons(IP_DF);
454 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
455 qp->q.fragments = NULL;
456 qp->q.rb_fragments = RB_ROOT;
457 qp->q.fragments_tail = NULL;
458 qp->q.last_run_head = NULL;
462 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
466 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
468 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
472 /* Process an incoming IP datagram fragment. */
473 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
475 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
476 int vif = l3mdev_master_ifindex_rcu(dev);
479 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
482 /* Lookup (or create) queue header */
483 qp = ip_find(net, ip_hdr(skb), user, vif);
487 spin_lock(&qp->q.lock);
489 ret = ip_frag_queue(qp, skb);
491 spin_unlock(&qp->q.lock);
496 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
500 EXPORT_SYMBOL(ip_defrag);
502 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
508 if (skb->protocol != htons(ETH_P_IP))
511 netoff = skb_network_offset(skb);
513 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
516 if (iph.ihl < 5 || iph.version != 4)
519 len = ntohs(iph.tot_len);
520 if (skb->len < netoff + len || len < (iph.ihl * 4))
523 if (ip_is_fragment(&iph)) {
524 skb = skb_share_check(skb, GFP_ATOMIC);
526 if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
530 if (pskb_trim_rcsum(skb, netoff + len)) {
534 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
535 if (ip_defrag(net, skb, user))
542 EXPORT_SYMBOL(ip_check_defrag);
547 static struct ctl_table ip4_frags_ns_ctl_table[] = {
549 .procname = "ipfrag_high_thresh",
550 .data = &init_net.ipv4.frags.high_thresh,
551 .maxlen = sizeof(unsigned long),
553 .proc_handler = proc_doulongvec_minmax,
554 .extra1 = &init_net.ipv4.frags.low_thresh
557 .procname = "ipfrag_low_thresh",
558 .data = &init_net.ipv4.frags.low_thresh,
559 .maxlen = sizeof(unsigned long),
561 .proc_handler = proc_doulongvec_minmax,
562 .extra2 = &init_net.ipv4.frags.high_thresh
565 .procname = "ipfrag_time",
566 .data = &init_net.ipv4.frags.timeout,
567 .maxlen = sizeof(int),
569 .proc_handler = proc_dointvec_jiffies,
572 .procname = "ipfrag_max_dist",
573 .data = &init_net.ipv4.frags.max_dist,
574 .maxlen = sizeof(int),
576 .proc_handler = proc_dointvec_minmax,
582 /* secret interval has been deprecated */
583 static int ip4_frags_secret_interval_unused;
584 static struct ctl_table ip4_frags_ctl_table[] = {
586 .procname = "ipfrag_secret_interval",
587 .data = &ip4_frags_secret_interval_unused,
588 .maxlen = sizeof(int),
590 .proc_handler = proc_dointvec_jiffies,
595 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
597 struct ctl_table *table;
598 struct ctl_table_header *hdr;
600 table = ip4_frags_ns_ctl_table;
601 if (!net_eq(net, &init_net)) {
602 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
606 table[0].data = &net->ipv4.frags.high_thresh;
607 table[0].extra1 = &net->ipv4.frags.low_thresh;
608 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
609 table[1].data = &net->ipv4.frags.low_thresh;
610 table[1].extra2 = &net->ipv4.frags.high_thresh;
611 table[2].data = &net->ipv4.frags.timeout;
612 table[3].data = &net->ipv4.frags.max_dist;
615 hdr = register_net_sysctl(net, "net/ipv4", table);
619 net->ipv4.frags_hdr = hdr;
623 if (!net_eq(net, &init_net))
629 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
631 struct ctl_table *table;
633 table = net->ipv4.frags_hdr->ctl_table_arg;
634 unregister_net_sysctl_table(net->ipv4.frags_hdr);
638 static void __init ip4_frags_ctl_register(void)
640 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
643 static int ip4_frags_ns_ctl_register(struct net *net)
648 static void ip4_frags_ns_ctl_unregister(struct net *net)
652 static void __init ip4_frags_ctl_register(void)
657 static int __net_init ipv4_frags_init_net(struct net *net)
661 /* Fragment cache limits.
663 * The fragment memory accounting code, (tries to) account for
664 * the real memory usage, by measuring both the size of frag
665 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
666 * and the SKB's truesize.
668 * A 64K fragment consumes 129736 bytes (44*2944)+200
669 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
671 * We will commit 4MB at one time. Should we cross that limit
672 * we will prune down to 3MB, making room for approx 8 big 64K
675 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
676 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
678 * Important NOTE! Fragment queue must be destroyed before MSL expires.
679 * RFC791 is wrong proposing to prolongate timer each fragment arrival
682 net->ipv4.frags.timeout = IP_FRAG_TIME;
684 net->ipv4.frags.max_dist = 64;
685 net->ipv4.frags.f = &ip4_frags;
687 res = inet_frags_init_net(&net->ipv4.frags);
690 res = ip4_frags_ns_ctl_register(net);
692 inet_frags_exit_net(&net->ipv4.frags);
696 static void __net_exit ipv4_frags_exit_net(struct net *net)
698 ip4_frags_ns_ctl_unregister(net);
699 inet_frags_exit_net(&net->ipv4.frags);
702 static struct pernet_operations ip4_frags_ops = {
703 .init = ipv4_frags_init_net,
704 .exit = ipv4_frags_exit_net,
708 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
711 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
714 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
716 const struct inet_frag_queue *fq = data;
718 return jhash2((const u32 *)&fq->key.v4,
719 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
722 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
724 const struct frag_v4_compare_key *key = arg->key;
725 const struct inet_frag_queue *fq = ptr;
727 return !!memcmp(&fq->key, key, sizeof(*key));
730 static const struct rhashtable_params ip4_rhash_params = {
731 .head_offset = offsetof(struct inet_frag_queue, node),
732 .key_offset = offsetof(struct inet_frag_queue, key),
733 .key_len = sizeof(struct frag_v4_compare_key),
734 .hashfn = ip4_key_hashfn,
735 .obj_hashfn = ip4_obj_hashfn,
736 .obj_cmpfn = ip4_obj_cmpfn,
737 .automatic_shrinking = true,
740 void __init ipfrag_init(void)
742 ip4_frags.constructor = ip4_frag_init;
743 ip4_frags.destructor = ip4_frag_free;
744 ip4_frags.qsize = sizeof(struct ipq);
745 ip4_frags.frag_expire = ip_expire;
746 ip4_frags.frags_cache_name = ip_frag_cache_name;
747 ip4_frags.rhash_params = ip4_rhash_params;
748 if (inet_frags_init(&ip4_frags))
749 panic("IP: failed to allocate ip4_frags cache\n");
750 ip4_frags_ctl_register();
751 register_pernet_subsys(&ip4_frags_ops);