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 /* Use skb->cb to track consecutive/adjacent fragments coming at
61 * the end of the queue. Nodes in the rb-tree queue will
62 * contain "runs" of one or more adjacent fragments.
65 * - next_frag is NULL at the tail of a "run";
66 * - the head of a "run" has the sum of all fragment lengths in frag_run_len.
68 struct ipfrag_skb_cb {
69 struct inet_skb_parm h;
70 struct sk_buff *next_frag;
74 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
76 static void ip4_frag_init_run(struct sk_buff *skb)
78 BUILD_BUG_ON(sizeof(struct ipfrag_skb_cb) > sizeof(skb->cb));
80 FRAG_CB(skb)->next_frag = NULL;
81 FRAG_CB(skb)->frag_run_len = skb->len;
84 /* Append skb to the last "run". */
85 static void ip4_frag_append_to_last_run(struct inet_frag_queue *q,
88 RB_CLEAR_NODE(&skb->rbnode);
89 FRAG_CB(skb)->next_frag = NULL;
91 FRAG_CB(q->last_run_head)->frag_run_len += skb->len;
92 FRAG_CB(q->fragments_tail)->next_frag = skb;
93 q->fragments_tail = skb;
96 /* Create a new "run" with the skb. */
97 static void ip4_frag_create_run(struct inet_frag_queue *q, struct sk_buff *skb)
100 rb_link_node(&skb->rbnode, &q->last_run_head->rbnode,
101 &q->last_run_head->rbnode.rb_right);
103 rb_link_node(&skb->rbnode, NULL, &q->rb_fragments.rb_node);
104 rb_insert_color(&skb->rbnode, &q->rb_fragments);
106 ip4_frag_init_run(skb);
107 q->fragments_tail = skb;
108 q->last_run_head = skb;
111 /* Describe an entry in the "incomplete datagrams" queue. */
113 struct inet_frag_queue q;
115 u8 ecn; /* RFC3168 support */
116 u16 max_df_size; /* largest frag with DF set seen */
119 struct inet_peer *peer;
122 static u8 ip4_frag_ecn(u8 tos)
124 return 1 << (tos & INET_ECN_MASK);
127 static struct inet_frags ip4_frags;
129 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
130 struct sk_buff *prev_tail, struct net_device *dev);
133 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
135 struct ipq *qp = container_of(q, struct ipq, q);
136 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
138 struct net *net = container_of(ipv4, struct net, ipv4);
140 const struct frag_v4_compare_key *key = a;
144 qp->peer = q->net->max_dist ?
145 inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
149 static 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 void ipq_put(struct ipq *ipq)
163 inet_frag_put(&ipq->q);
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);
174 static bool frag_expire_skip_icmp(u32 user)
176 return user == IP_DEFRAG_AF_PACKET ||
177 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
178 __IP_DEFRAG_CONNTRACK_IN_END) ||
179 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
180 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
184 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
186 static void ip_expire(struct timer_list *t)
188 struct inet_frag_queue *frag = from_timer(frag, t, timer);
189 const struct iphdr *iph;
190 struct sk_buff *head = NULL;
195 qp = container_of(frag, struct ipq, q);
196 net = container_of(qp->q.net, struct net, ipv4.frags);
199 spin_lock(&qp->q.lock);
201 if (qp->q.flags & INET_FRAG_COMPLETE)
205 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
206 __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
208 if (!(qp->q.flags & INET_FRAG_FIRST_IN))
211 /* sk_buff::dev and sk_buff::rbnode are unionized. So we
212 * pull the head out of the tree in order to be able to
213 * deal with head->dev.
215 if (qp->q.fragments) {
216 head = qp->q.fragments;
217 qp->q.fragments = head->next;
219 head = skb_rb_first(&qp->q.rb_fragments);
222 if (FRAG_CB(head)->next_frag)
223 rb_replace_node(&head->rbnode,
224 &FRAG_CB(head)->next_frag->rbnode,
225 &qp->q.rb_fragments);
227 rb_erase(&head->rbnode, &qp->q.rb_fragments);
228 memset(&head->rbnode, 0, sizeof(head->rbnode));
231 if (head == qp->q.fragments_tail)
232 qp->q.fragments_tail = NULL;
234 sub_frag_mem_limit(qp->q.net, head->truesize);
236 head->dev = dev_get_by_index_rcu(net, qp->iif);
241 /* skb has no dst, perform route lookup again */
243 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
244 iph->tos, head->dev);
248 /* Only an end host needs to send an ICMP
249 * "Fragment Reassembly Timeout" message, per RFC792.
251 if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
252 (skb_rtable(head)->rt_type != RTN_LOCAL))
255 spin_unlock(&qp->q.lock);
256 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
260 spin_unlock(&qp->q.lock);
267 /* Find the correct entry in the "incomplete datagrams" queue for
268 * this IP datagram, and create new one, if nothing is found.
270 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
273 struct frag_v4_compare_key key = {
279 .protocol = iph->protocol,
281 struct inet_frag_queue *q;
283 q = inet_frag_find(&net->ipv4.frags, &key);
287 return container_of(q, struct ipq, q);
290 /* Is the fragment too far ahead to be part of ipq? */
291 static int ip_frag_too_far(struct ipq *qp)
293 struct inet_peer *peer = qp->peer;
294 unsigned int max = qp->q.net->max_dist;
295 unsigned int start, end;
303 end = atomic_inc_return(&peer->rid);
306 rc = qp->q.fragments_tail && (end - start) > max;
311 net = container_of(qp->q.net, struct net, ipv4.frags);
312 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
318 static int ip_frag_reinit(struct ipq *qp)
320 unsigned int sum_truesize = 0;
322 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
323 refcount_inc(&qp->q.refcnt);
327 sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments);
328 sub_frag_mem_limit(qp->q.net, sum_truesize);
333 qp->q.fragments = NULL;
334 qp->q.rb_fragments = RB_ROOT;
335 qp->q.fragments_tail = NULL;
336 qp->q.last_run_head = NULL;
343 /* Add new segment to existing queue. */
344 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
346 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
347 struct rb_node **rbn, *parent;
348 struct sk_buff *skb1, *prev_tail;
349 int ihl, end, skb1_run_end;
350 struct net_device *dev;
351 unsigned int fragsize;
356 if (qp->q.flags & INET_FRAG_COMPLETE)
359 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
360 unlikely(ip_frag_too_far(qp)) &&
361 unlikely(err = ip_frag_reinit(qp))) {
366 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
367 offset = ntohs(ip_hdr(skb)->frag_off);
368 flags = offset & ~IP_OFFSET;
370 offset <<= 3; /* offset is in 8-byte chunks */
371 ihl = ip_hdrlen(skb);
373 /* Determine the position of this fragment. */
374 end = offset + skb->len - skb_network_offset(skb) - ihl;
377 /* Is this the final fragment? */
378 if ((flags & IP_MF) == 0) {
379 /* If we already have some bits beyond end
380 * or have different end, the segment is corrupted.
382 if (end < qp->q.len ||
383 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
385 qp->q.flags |= INET_FRAG_LAST_IN;
390 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
391 skb->ip_summed = CHECKSUM_NONE;
393 if (end > qp->q.len) {
394 /* Some bits beyond end -> corruption. */
395 if (qp->q.flags & INET_FRAG_LAST_IN)
404 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
407 err = pskb_trim_rcsum(skb, end - offset);
411 /* Note : skb->rbnode and skb->dev share the same location. */
413 /* Makes sure compiler wont do silly aliasing games */
416 /* RFC5722, Section 4, amended by Errata ID : 3089
417 * When reassembling an IPv6 datagram, if
418 * one or more its constituent fragments is determined to be an
419 * overlapping fragment, the entire datagram (and any constituent
420 * fragments) MUST be silently discarded.
422 * We do the same here for IPv4 (and increment an snmp counter) but
423 * we do not want to drop the whole queue in response to a duplicate
428 /* Find out where to put this fragment. */
429 prev_tail = qp->q.fragments_tail;
431 ip4_frag_create_run(&qp->q, skb); /* First fragment. */
432 else if (prev_tail->ip_defrag_offset + prev_tail->len < end) {
433 /* This is the common case: skb goes to the end. */
434 /* Detect and discard overlaps. */
435 if (offset < prev_tail->ip_defrag_offset + prev_tail->len)
437 if (offset == prev_tail->ip_defrag_offset + prev_tail->len)
438 ip4_frag_append_to_last_run(&qp->q, skb);
440 ip4_frag_create_run(&qp->q, skb);
442 /* Binary search. Note that skb can become the first fragment,
443 * but not the last (covered above).
445 rbn = &qp->q.rb_fragments.rb_node;
448 skb1 = rb_to_skb(parent);
449 skb1_run_end = skb1->ip_defrag_offset +
450 FRAG_CB(skb1)->frag_run_len;
451 if (end <= skb1->ip_defrag_offset)
452 rbn = &parent->rb_left;
453 else if (offset >= skb1_run_end)
454 rbn = &parent->rb_right;
455 else if (offset >= skb1->ip_defrag_offset &&
457 goto err; /* No new data, potential duplicate */
459 goto overlap; /* Found an overlap */
461 /* Here we have parent properly set, and rbn pointing to
462 * one of its NULL left/right children. Insert skb.
464 ip4_frag_init_run(skb);
465 rb_link_node(&skb->rbnode, parent, rbn);
466 rb_insert_color(&skb->rbnode, &qp->q.rb_fragments);
470 qp->iif = dev->ifindex;
471 skb->ip_defrag_offset = offset;
473 qp->q.stamp = skb->tstamp;
474 qp->q.meat += skb->len;
476 add_frag_mem_limit(qp->q.net, skb->truesize);
478 qp->q.flags |= INET_FRAG_FIRST_IN;
480 fragsize = skb->len + ihl;
482 if (fragsize > qp->q.max_size)
483 qp->q.max_size = fragsize;
485 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
486 fragsize > qp->max_df_size)
487 qp->max_df_size = fragsize;
489 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
490 qp->q.meat == qp->q.len) {
491 unsigned long orefdst = skb->_skb_refdst;
493 skb->_skb_refdst = 0UL;
494 err = ip_frag_reasm(qp, skb, prev_tail, dev);
495 skb->_skb_refdst = orefdst;
497 inet_frag_kill(&qp->q);
505 __IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
507 inet_frag_kill(&qp->q);
513 /* Build a new IP datagram from all its fragments. */
514 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
515 struct sk_buff *prev_tail, struct net_device *dev)
517 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
519 struct sk_buff *fp, *head = skb_rb_first(&qp->q.rb_fragments);
520 struct sk_buff **nextp; /* To build frag_list. */
530 ecn = ip_frag_ecn_table[qp->ecn];
531 if (unlikely(ecn == 0xff)) {
535 /* Make the one we just received the head. */
537 fp = skb_clone(skb, GFP_ATOMIC);
540 FRAG_CB(fp)->next_frag = FRAG_CB(skb)->next_frag;
541 if (RB_EMPTY_NODE(&skb->rbnode))
542 FRAG_CB(prev_tail)->next_frag = fp;
544 rb_replace_node(&skb->rbnode, &fp->rbnode,
545 &qp->q.rb_fragments);
546 if (qp->q.fragments_tail == skb)
547 qp->q.fragments_tail = fp;
548 skb_morph(skb, head);
549 FRAG_CB(skb)->next_frag = FRAG_CB(head)->next_frag;
550 rb_replace_node(&head->rbnode, &skb->rbnode,
551 &qp->q.rb_fragments);
556 WARN_ON(head->ip_defrag_offset != 0);
558 /* Allocate a new buffer for the datagram. */
559 ihlen = ip_hdrlen(head);
560 len = ihlen + qp->q.len;
566 delta = - head->truesize;
568 /* Head of list must not be cloned. */
569 if (skb_unclone(head, GFP_ATOMIC))
572 delta += head->truesize;
574 add_frag_mem_limit(qp->q.net, delta);
576 /* If the first fragment is fragmented itself, we split
577 * it to two chunks: the first with data and paged part
578 * and the second, holding only fragments. */
579 if (skb_has_frag_list(head)) {
580 struct sk_buff *clone;
583 clone = alloc_skb(0, GFP_ATOMIC);
586 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
587 skb_frag_list_init(head);
588 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
589 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
590 clone->len = clone->data_len = head->data_len - plen;
591 head->truesize += clone->truesize;
593 clone->ip_summed = head->ip_summed;
594 add_frag_mem_limit(qp->q.net, clone->truesize);
595 skb_shinfo(head)->frag_list = clone;
596 nextp = &clone->next;
598 nextp = &skb_shinfo(head)->frag_list;
601 skb_push(head, head->data - skb_network_header(head));
603 /* Traverse the tree in order, to build frag_list. */
604 fp = FRAG_CB(head)->next_frag;
605 rbn = rb_next(&head->rbnode);
606 rb_erase(&head->rbnode, &qp->q.rb_fragments);
608 /* fp points to the next sk_buff in the current run;
609 * rbn points to the next run.
611 /* Go through the current run. */
616 memset(&fp->rbnode, 0, sizeof(fp->rbnode));
618 head->data_len += fp->len;
619 head->len += fp->len;
620 if (head->ip_summed != fp->ip_summed)
621 head->ip_summed = CHECKSUM_NONE;
622 else if (head->ip_summed == CHECKSUM_COMPLETE)
623 head->csum = csum_add(head->csum, fp->csum);
624 head->truesize += fp->truesize;
625 fp = FRAG_CB(fp)->next_frag;
627 /* Move to the next run. */
629 struct rb_node *rbnext = rb_next(rbn);
632 rb_erase(rbn, &qp->q.rb_fragments);
636 sub_frag_mem_limit(qp->q.net, head->truesize);
639 skb_mark_not_on_list(head);
642 head->tstamp = qp->q.stamp;
643 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
646 iph->tot_len = htons(len);
649 /* When we set IP_DF on a refragmented skb we must also force a
650 * call to ip_fragment to avoid forwarding a DF-skb of size s while
651 * original sender only sent fragments of size f (where f < s).
653 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
654 * frag seen to avoid sending tiny DF-fragments in case skb was built
655 * from one very small df-fragment and one large non-df frag.
657 if (qp->max_df_size == qp->q.max_size) {
658 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
659 iph->frag_off = htons(IP_DF);
666 __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
667 qp->q.fragments = NULL;
668 qp->q.rb_fragments = RB_ROOT;
669 qp->q.fragments_tail = NULL;
670 qp->q.last_run_head = NULL;
674 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
678 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
680 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
684 /* Process an incoming IP datagram fragment. */
685 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
687 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
688 int vif = l3mdev_master_ifindex_rcu(dev);
691 __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
694 /* Lookup (or create) queue header */
695 qp = ip_find(net, ip_hdr(skb), user, vif);
699 spin_lock(&qp->q.lock);
701 ret = ip_frag_queue(qp, skb);
703 spin_unlock(&qp->q.lock);
708 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
712 EXPORT_SYMBOL(ip_defrag);
714 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
720 if (skb->protocol != htons(ETH_P_IP))
723 netoff = skb_network_offset(skb);
725 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
728 if (iph.ihl < 5 || iph.version != 4)
731 len = ntohs(iph.tot_len);
732 if (skb->len < netoff + len || len < (iph.ihl * 4))
735 if (ip_is_fragment(&iph)) {
736 skb = skb_share_check(skb, GFP_ATOMIC);
738 if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
742 if (pskb_trim_rcsum(skb, netoff + len)) {
746 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
747 if (ip_defrag(net, skb, user))
754 EXPORT_SYMBOL(ip_check_defrag);
756 unsigned int inet_frag_rbtree_purge(struct rb_root *root)
758 struct rb_node *p = rb_first(root);
759 unsigned int sum = 0;
762 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
765 rb_erase(&skb->rbnode, root);
767 struct sk_buff *next = FRAG_CB(skb)->next_frag;
769 sum += skb->truesize;
776 EXPORT_SYMBOL(inet_frag_rbtree_purge);
781 static struct ctl_table ip4_frags_ns_ctl_table[] = {
783 .procname = "ipfrag_high_thresh",
784 .data = &init_net.ipv4.frags.high_thresh,
785 .maxlen = sizeof(unsigned long),
787 .proc_handler = proc_doulongvec_minmax,
788 .extra1 = &init_net.ipv4.frags.low_thresh
791 .procname = "ipfrag_low_thresh",
792 .data = &init_net.ipv4.frags.low_thresh,
793 .maxlen = sizeof(unsigned long),
795 .proc_handler = proc_doulongvec_minmax,
796 .extra2 = &init_net.ipv4.frags.high_thresh
799 .procname = "ipfrag_time",
800 .data = &init_net.ipv4.frags.timeout,
801 .maxlen = sizeof(int),
803 .proc_handler = proc_dointvec_jiffies,
806 .procname = "ipfrag_max_dist",
807 .data = &init_net.ipv4.frags.max_dist,
808 .maxlen = sizeof(int),
810 .proc_handler = proc_dointvec_minmax,
816 /* secret interval has been deprecated */
817 static int ip4_frags_secret_interval_unused;
818 static struct ctl_table ip4_frags_ctl_table[] = {
820 .procname = "ipfrag_secret_interval",
821 .data = &ip4_frags_secret_interval_unused,
822 .maxlen = sizeof(int),
824 .proc_handler = proc_dointvec_jiffies,
829 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
831 struct ctl_table *table;
832 struct ctl_table_header *hdr;
834 table = ip4_frags_ns_ctl_table;
835 if (!net_eq(net, &init_net)) {
836 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
840 table[0].data = &net->ipv4.frags.high_thresh;
841 table[0].extra1 = &net->ipv4.frags.low_thresh;
842 table[1].data = &net->ipv4.frags.low_thresh;
843 table[1].extra2 = &net->ipv4.frags.high_thresh;
844 table[2].data = &net->ipv4.frags.timeout;
845 table[3].data = &net->ipv4.frags.max_dist;
848 hdr = register_net_sysctl(net, "net/ipv4", table);
852 net->ipv4.frags_hdr = hdr;
856 if (!net_eq(net, &init_net))
862 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
864 struct ctl_table *table;
866 table = net->ipv4.frags_hdr->ctl_table_arg;
867 unregister_net_sysctl_table(net->ipv4.frags_hdr);
871 static void __init ip4_frags_ctl_register(void)
873 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
876 static int ip4_frags_ns_ctl_register(struct net *net)
881 static void ip4_frags_ns_ctl_unregister(struct net *net)
885 static void __init ip4_frags_ctl_register(void)
890 static int __net_init ipv4_frags_init_net(struct net *net)
894 /* Fragment cache limits.
896 * The fragment memory accounting code, (tries to) account for
897 * the real memory usage, by measuring both the size of frag
898 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
899 * and the SKB's truesize.
901 * A 64K fragment consumes 129736 bytes (44*2944)+200
902 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
904 * We will commit 4MB at one time. Should we cross that limit
905 * we will prune down to 3MB, making room for approx 8 big 64K
908 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
909 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
911 * Important NOTE! Fragment queue must be destroyed before MSL expires.
912 * RFC791 is wrong proposing to prolongate timer each fragment arrival
915 net->ipv4.frags.timeout = IP_FRAG_TIME;
917 net->ipv4.frags.max_dist = 64;
918 net->ipv4.frags.f = &ip4_frags;
920 res = inet_frags_init_net(&net->ipv4.frags);
923 res = ip4_frags_ns_ctl_register(net);
925 inet_frags_exit_net(&net->ipv4.frags);
929 static void __net_exit ipv4_frags_exit_net(struct net *net)
931 ip4_frags_ns_ctl_unregister(net);
932 inet_frags_exit_net(&net->ipv4.frags);
935 static struct pernet_operations ip4_frags_ops = {
936 .init = ipv4_frags_init_net,
937 .exit = ipv4_frags_exit_net,
941 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
944 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
947 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
949 const struct inet_frag_queue *fq = data;
951 return jhash2((const u32 *)&fq->key.v4,
952 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
955 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
957 const struct frag_v4_compare_key *key = arg->key;
958 const struct inet_frag_queue *fq = ptr;
960 return !!memcmp(&fq->key, key, sizeof(*key));
963 static const struct rhashtable_params ip4_rhash_params = {
964 .head_offset = offsetof(struct inet_frag_queue, node),
965 .key_offset = offsetof(struct inet_frag_queue, key),
966 .key_len = sizeof(struct frag_v4_compare_key),
967 .hashfn = ip4_key_hashfn,
968 .obj_hashfn = ip4_obj_hashfn,
969 .obj_cmpfn = ip4_obj_cmpfn,
970 .automatic_shrinking = true,
973 void __init ipfrag_init(void)
975 ip4_frags.constructor = ip4_frag_init;
976 ip4_frags.destructor = ip4_frag_free;
977 ip4_frags.qsize = sizeof(struct ipq);
978 ip4_frags.frag_expire = ip_expire;
979 ip4_frags.frags_cache_name = ip_frag_cache_name;
980 ip4_frags.rhash_params = ip4_rhash_params;
981 if (inet_frags_init(&ip4_frags))
982 panic("IP: failed to allocate ip4_frags cache\n");
983 ip4_frags_ctl_register();
984 register_pernet_subsys(&ip4_frags_ops);