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@lxorguk.ukuu.org.uk>
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 #define pr_fmt(fmt) "IPv4: " fmt
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>
37 #include <linux/slab.h>
38 #include <net/route.h>
43 #include <net/checksum.h>
44 #include <net/inetpeer.h>
45 #include <net/inet_frag.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/inet.h>
49 #include <linux/netfilter_ipv4.h>
50 #include <net/inet_ecn.h>
51 #include <net/l3mdev.h>
53 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
54 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
55 * as well. Or notify me, at least. --ANK
58 static int sysctl_ipfrag_max_dist __read_mostly = 64;
59 static const char ip_frag_cache_name[] = "ip4-frags";
61 /* Use skb->cb to track consecutive/adjacent fragments coming at
62 * the end of the queue. Nodes in the rb-tree queue will
63 * contain "runs" of one or more adjacent fragments.
66 * - next_frag is NULL at the tail of a "run";
67 * - the head of a "run" has the sum of all fragment lengths in frag_run_len.
69 struct ipfrag_skb_cb {
70 struct inet_skb_parm h;
71 struct sk_buff *next_frag;
75 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
77 static void ip4_frag_init_run(struct sk_buff *skb)
79 BUILD_BUG_ON(sizeof(struct ipfrag_skb_cb) > sizeof(skb->cb));
81 FRAG_CB(skb)->next_frag = NULL;
82 FRAG_CB(skb)->frag_run_len = skb->len;
85 /* Append skb to the last "run". */
86 static void ip4_frag_append_to_last_run(struct inet_frag_queue *q,
89 RB_CLEAR_NODE(&skb->rbnode);
90 FRAG_CB(skb)->next_frag = NULL;
92 FRAG_CB(q->last_run_head)->frag_run_len += skb->len;
93 FRAG_CB(q->fragments_tail)->next_frag = skb;
94 q->fragments_tail = skb;
97 /* Create a new "run" with the skb. */
98 static void ip4_frag_create_run(struct inet_frag_queue *q, struct sk_buff *skb)
100 if (q->last_run_head)
101 rb_link_node(&skb->rbnode, &q->last_run_head->rbnode,
102 &q->last_run_head->rbnode.rb_right);
104 rb_link_node(&skb->rbnode, NULL, &q->rb_fragments.rb_node);
105 rb_insert_color(&skb->rbnode, &q->rb_fragments);
107 ip4_frag_init_run(skb);
108 q->fragments_tail = skb;
109 q->last_run_head = skb;
112 /* Describe an entry in the "incomplete datagrams" queue. */
114 struct inet_frag_queue q;
116 u8 ecn; /* RFC3168 support */
117 u16 max_df_size; /* largest frag with DF set seen */
120 struct inet_peer *peer;
123 static u8 ip4_frag_ecn(u8 tos)
125 return 1 << (tos & INET_ECN_MASK);
128 static struct inet_frags ip4_frags;
130 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
131 struct sk_buff *prev_tail, struct net_device *dev);
134 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
136 struct ipq *qp = container_of(q, struct ipq, q);
137 struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
139 struct net *net = container_of(ipv4, struct net, ipv4);
141 const struct frag_v4_compare_key *key = a;
145 qp->peer = sysctl_ipfrag_max_dist ?
146 inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
150 static void ip4_frag_free(struct inet_frag_queue *q)
154 qp = container_of(q, struct ipq, q);
156 inet_putpeer(qp->peer);
160 /* Destruction primitives. */
162 static void ipq_put(struct ipq *ipq)
164 inet_frag_put(&ipq->q);
167 /* Kill ipq entry. It is not destroyed immediately,
168 * because caller (and someone more) holds reference count.
170 static void ipq_kill(struct ipq *ipq)
172 inet_frag_kill(&ipq->q);
175 static bool frag_expire_skip_icmp(u32 user)
177 return user == IP_DEFRAG_AF_PACKET ||
178 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
179 __IP_DEFRAG_CONNTRACK_IN_END) ||
180 ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
181 __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
185 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
187 static void ip_expire(unsigned long arg)
189 const struct iphdr *iph;
190 struct sk_buff *head = NULL;
195 qp = container_of((struct inet_frag_queue *) arg, 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_BH(net, IPSTATS_MIB_REASMFAILS);
206 IP_INC_STATS_BH(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);
268 /* Find the correct entry in the "incomplete datagrams" queue for
269 * this IP datagram, and create new one, if nothing is found.
271 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
274 struct frag_v4_compare_key key = {
280 .protocol = iph->protocol,
282 struct inet_frag_queue *q;
284 q = inet_frag_find(&net->ipv4.frags, &key);
288 return container_of(q, struct ipq, q);
291 /* Is the fragment too far ahead to be part of ipq? */
292 static int ip_frag_too_far(struct ipq *qp)
294 struct inet_peer *peer = qp->peer;
295 unsigned int max = sysctl_ipfrag_max_dist;
296 unsigned int start, end;
304 end = atomic_inc_return(&peer->rid);
307 rc = qp->q.fragments_tail && (end - start) > max;
312 net = container_of(qp->q.net, struct net, ipv4.frags);
313 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
319 static int ip_frag_reinit(struct ipq *qp)
321 unsigned int sum_truesize = 0;
323 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
324 atomic_inc(&qp->q.refcnt);
328 sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments);
329 sub_frag_mem_limit(qp->q.net, sum_truesize);
334 qp->q.fragments = NULL;
335 qp->q.rb_fragments = RB_ROOT;
336 qp->q.fragments_tail = NULL;
337 qp->q.last_run_head = NULL;
344 /* Add new segment to existing queue. */
345 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
347 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
348 struct rb_node **rbn, *parent;
349 struct sk_buff *skb1, *prev_tail;
350 int ihl, end, skb1_run_end;
351 struct net_device *dev;
352 unsigned int fragsize;
357 if (qp->q.flags & INET_FRAG_COMPLETE)
360 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
361 unlikely(ip_frag_too_far(qp)) &&
362 unlikely(err = ip_frag_reinit(qp))) {
367 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
368 offset = ntohs(ip_hdr(skb)->frag_off);
369 flags = offset & ~IP_OFFSET;
371 offset <<= 3; /* offset is in 8-byte chunks */
372 ihl = ip_hdrlen(skb);
374 /* Determine the position of this fragment. */
375 end = offset + skb->len - skb_network_offset(skb) - ihl;
378 /* Is this the final fragment? */
379 if ((flags & IP_MF) == 0) {
380 /* If we already have some bits beyond end
381 * or have different end, the segment is corrupted.
383 if (end < qp->q.len ||
384 ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
386 qp->q.flags |= INET_FRAG_LAST_IN;
391 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
392 skb->ip_summed = CHECKSUM_NONE;
394 if (end > qp->q.len) {
395 /* Some bits beyond end -> corruption. */
396 if (qp->q.flags & INET_FRAG_LAST_IN)
405 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
408 err = pskb_trim_rcsum(skb, end - offset);
412 /* Note : skb->rbnode and skb->dev share the same location. */
414 /* Makes sure compiler wont do silly aliasing games */
417 /* RFC5722, Section 4, amended by Errata ID : 3089
418 * When reassembling an IPv6 datagram, if
419 * one or more its constituent fragments is determined to be an
420 * overlapping fragment, the entire datagram (and any constituent
421 * fragments) MUST be silently discarded.
423 * We do the same here for IPv4 (and increment an snmp counter) but
424 * we do not want to drop the whole queue in response to a duplicate
429 /* Find out where to put this fragment. */
430 prev_tail = qp->q.fragments_tail;
432 ip4_frag_create_run(&qp->q, skb); /* First fragment. */
433 else if (prev_tail->ip_defrag_offset + prev_tail->len < end) {
434 /* This is the common case: skb goes to the end. */
435 /* Detect and discard overlaps. */
436 if (offset < prev_tail->ip_defrag_offset + prev_tail->len)
438 if (offset == prev_tail->ip_defrag_offset + prev_tail->len)
439 ip4_frag_append_to_last_run(&qp->q, skb);
441 ip4_frag_create_run(&qp->q, skb);
443 /* Binary search. Note that skb can become the first fragment,
444 * but not the last (covered above).
446 rbn = &qp->q.rb_fragments.rb_node;
449 skb1 = rb_to_skb(parent);
450 skb1_run_end = skb1->ip_defrag_offset +
451 FRAG_CB(skb1)->frag_run_len;
452 if (end <= skb1->ip_defrag_offset)
453 rbn = &parent->rb_left;
454 else if (offset >= skb1_run_end)
455 rbn = &parent->rb_right;
456 else if (offset >= skb1->ip_defrag_offset &&
458 goto err; /* No new data, potential duplicate */
460 goto discard_qp; /* Found an overlap */
462 /* Here we have parent properly set, and rbn pointing to
463 * one of its NULL left/right children. Insert skb.
465 ip4_frag_init_run(skb);
466 rb_link_node(&skb->rbnode, parent, rbn);
467 rb_insert_color(&skb->rbnode, &qp->q.rb_fragments);
471 qp->iif = dev->ifindex;
472 skb->ip_defrag_offset = offset;
474 qp->q.stamp = skb->tstamp;
475 qp->q.meat += skb->len;
477 add_frag_mem_limit(qp->q.net, skb->truesize);
479 qp->q.flags |= INET_FRAG_FIRST_IN;
481 fragsize = skb->len + ihl;
483 if (fragsize > qp->q.max_size)
484 qp->q.max_size = fragsize;
486 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
487 fragsize > qp->max_df_size)
488 qp->max_df_size = fragsize;
490 if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
491 qp->q.meat == qp->q.len) {
492 unsigned long orefdst = skb->_skb_refdst;
494 skb->_skb_refdst = 0UL;
495 err = ip_frag_reasm(qp, skb, prev_tail, dev);
496 skb->_skb_refdst = orefdst;
504 inet_frag_kill(&qp->q);
505 IP_INC_STATS_BH(net, IPSTATS_MIB_REASM_OVERLAPS);
511 /* Build a new IP datagram from all its fragments. */
512 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
513 struct sk_buff *prev_tail, struct net_device *dev)
515 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
517 struct sk_buff *fp, *head = skb_rb_first(&qp->q.rb_fragments);
518 struct sk_buff **nextp; /* To build frag_list. */
527 ecn = ip_frag_ecn_table[qp->ecn];
528 if (unlikely(ecn == 0xff)) {
532 /* Make the one we just received the head. */
534 fp = skb_clone(skb, GFP_ATOMIC);
537 FRAG_CB(fp)->next_frag = FRAG_CB(skb)->next_frag;
538 if (RB_EMPTY_NODE(&skb->rbnode))
539 FRAG_CB(prev_tail)->next_frag = fp;
541 rb_replace_node(&skb->rbnode, &fp->rbnode,
542 &qp->q.rb_fragments);
543 if (qp->q.fragments_tail == skb)
544 qp->q.fragments_tail = fp;
545 skb_morph(skb, head);
546 FRAG_CB(skb)->next_frag = FRAG_CB(head)->next_frag;
547 rb_replace_node(&head->rbnode, &skb->rbnode,
548 &qp->q.rb_fragments);
553 WARN_ON(head->ip_defrag_offset != 0);
555 /* Allocate a new buffer for the datagram. */
556 ihlen = ip_hdrlen(head);
557 len = ihlen + qp->q.len;
563 /* Head of list must not be cloned. */
564 if (skb_unclone(head, GFP_ATOMIC))
567 /* If the first fragment is fragmented itself, we split
568 * it to two chunks: the first with data and paged part
569 * and the second, holding only fragments. */
570 if (skb_has_frag_list(head)) {
571 struct sk_buff *clone;
574 clone = alloc_skb(0, GFP_ATOMIC);
577 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
578 skb_frag_list_init(head);
579 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
580 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
581 clone->len = clone->data_len = head->data_len - plen;
582 head->truesize += clone->truesize;
584 clone->ip_summed = head->ip_summed;
585 add_frag_mem_limit(qp->q.net, clone->truesize);
586 skb_shinfo(head)->frag_list = clone;
587 nextp = &clone->next;
589 nextp = &skb_shinfo(head)->frag_list;
592 skb_push(head, head->data - skb_network_header(head));
594 /* Traverse the tree in order, to build frag_list. */
595 fp = FRAG_CB(head)->next_frag;
596 rbn = rb_next(&head->rbnode);
597 rb_erase(&head->rbnode, &qp->q.rb_fragments);
599 /* fp points to the next sk_buff in the current run;
600 * rbn points to the next run.
602 /* Go through the current run. */
607 memset(&fp->rbnode, 0, sizeof(fp->rbnode));
609 head->data_len += fp->len;
610 head->len += fp->len;
611 if (head->ip_summed != fp->ip_summed)
612 head->ip_summed = CHECKSUM_NONE;
613 else if (head->ip_summed == CHECKSUM_COMPLETE)
614 head->csum = csum_add(head->csum, fp->csum);
615 head->truesize += fp->truesize;
616 fp = FRAG_CB(fp)->next_frag;
618 /* Move to the next run. */
620 struct rb_node *rbnext = rb_next(rbn);
623 rb_erase(rbn, &qp->q.rb_fragments);
627 sub_frag_mem_limit(qp->q.net, head->truesize);
633 head->tstamp = qp->q.stamp;
634 IPCB(head)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
637 iph->tot_len = htons(len);
640 /* When we set IP_DF on a refragmented skb we must also force a
641 * call to ip_fragment to avoid forwarding a DF-skb of size s while
642 * original sender only sent fragments of size f (where f < s).
644 * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
645 * frag seen to avoid sending tiny DF-fragments in case skb was built
646 * from one very small df-fragment and one large non-df frag.
648 if (qp->max_df_size == qp->q.max_size) {
649 IPCB(head)->flags |= IPSKB_FRAG_PMTU;
650 iph->frag_off = htons(IP_DF);
657 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
658 qp->q.fragments = NULL;
659 qp->q.rb_fragments = RB_ROOT;
660 qp->q.fragments_tail = NULL;
661 qp->q.last_run_head = NULL;
665 net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
669 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
671 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
675 /* Process an incoming IP datagram fragment. */
676 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
678 struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
679 int vif = l3mdev_master_ifindex_rcu(dev);
682 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
685 /* Lookup (or create) queue header */
686 qp = ip_find(net, ip_hdr(skb), user, vif);
690 spin_lock(&qp->q.lock);
692 ret = ip_frag_queue(qp, skb);
694 spin_unlock(&qp->q.lock);
699 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
703 EXPORT_SYMBOL(ip_defrag);
705 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
711 if (skb->protocol != htons(ETH_P_IP))
714 netoff = skb_network_offset(skb);
716 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
719 if (iph.ihl < 5 || iph.version != 4)
722 len = ntohs(iph.tot_len);
723 if (skb->len < netoff + len || len < (iph.ihl * 4))
726 if (ip_is_fragment(&iph)) {
727 skb = skb_share_check(skb, GFP_ATOMIC);
729 if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
733 if (pskb_trim_rcsum(skb, netoff + len)) {
737 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
738 if (ip_defrag(net, skb, user))
745 EXPORT_SYMBOL(ip_check_defrag);
747 unsigned int inet_frag_rbtree_purge(struct rb_root *root)
749 struct rb_node *p = rb_first(root);
750 unsigned int sum = 0;
753 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
756 rb_erase(&skb->rbnode, root);
758 struct sk_buff *next = FRAG_CB(skb)->next_frag;
760 sum += skb->truesize;
767 EXPORT_SYMBOL(inet_frag_rbtree_purge);
772 static struct ctl_table ip4_frags_ns_ctl_table[] = {
774 .procname = "ipfrag_high_thresh",
775 .data = &init_net.ipv4.frags.high_thresh,
776 .maxlen = sizeof(unsigned long),
778 .proc_handler = proc_doulongvec_minmax,
779 .extra1 = &init_net.ipv4.frags.low_thresh
782 .procname = "ipfrag_low_thresh",
783 .data = &init_net.ipv4.frags.low_thresh,
784 .maxlen = sizeof(unsigned long),
786 .proc_handler = proc_doulongvec_minmax,
787 .extra2 = &init_net.ipv4.frags.high_thresh
790 .procname = "ipfrag_time",
791 .data = &init_net.ipv4.frags.timeout,
792 .maxlen = sizeof(int),
794 .proc_handler = proc_dointvec_jiffies,
799 /* secret interval has been deprecated */
800 static int ip4_frags_secret_interval_unused;
801 static struct ctl_table ip4_frags_ctl_table[] = {
803 .procname = "ipfrag_secret_interval",
804 .data = &ip4_frags_secret_interval_unused,
805 .maxlen = sizeof(int),
807 .proc_handler = proc_dointvec_jiffies,
810 .procname = "ipfrag_max_dist",
811 .data = &sysctl_ipfrag_max_dist,
812 .maxlen = sizeof(int),
814 .proc_handler = proc_dointvec_minmax,
820 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
822 struct ctl_table *table;
823 struct ctl_table_header *hdr;
825 table = ip4_frags_ns_ctl_table;
826 if (!net_eq(net, &init_net)) {
827 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
831 table[0].data = &net->ipv4.frags.high_thresh;
832 table[0].extra1 = &net->ipv4.frags.low_thresh;
833 table[0].extra2 = &init_net.ipv4.frags.high_thresh;
834 table[1].data = &net->ipv4.frags.low_thresh;
835 table[1].extra2 = &net->ipv4.frags.high_thresh;
836 table[2].data = &net->ipv4.frags.timeout;
838 /* Don't export sysctls to unprivileged users */
839 if (net->user_ns != &init_user_ns)
840 table[0].procname = NULL;
843 hdr = register_net_sysctl(net, "net/ipv4", table);
847 net->ipv4.frags_hdr = hdr;
851 if (!net_eq(net, &init_net))
857 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
859 struct ctl_table *table;
861 table = net->ipv4.frags_hdr->ctl_table_arg;
862 unregister_net_sysctl_table(net->ipv4.frags_hdr);
866 static void __init ip4_frags_ctl_register(void)
868 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
871 static int ip4_frags_ns_ctl_register(struct net *net)
876 static void ip4_frags_ns_ctl_unregister(struct net *net)
880 static void __init ip4_frags_ctl_register(void)
885 static int __net_init ipv4_frags_init_net(struct net *net)
889 /* Fragment cache limits.
891 * The fragment memory accounting code, (tries to) account for
892 * the real memory usage, by measuring both the size of frag
893 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
894 * and the SKB's truesize.
896 * A 64K fragment consumes 129736 bytes (44*2944)+200
897 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
899 * We will commit 4MB at one time. Should we cross that limit
900 * we will prune down to 3MB, making room for approx 8 big 64K
903 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
904 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
906 * Important NOTE! Fragment queue must be destroyed before MSL expires.
907 * RFC791 is wrong proposing to prolongate timer each fragment arrival
910 net->ipv4.frags.timeout = IP_FRAG_TIME;
912 net->ipv4.frags.f = &ip4_frags;
914 res = inet_frags_init_net(&net->ipv4.frags);
917 res = ip4_frags_ns_ctl_register(net);
919 inet_frags_exit_net(&net->ipv4.frags);
923 static void __net_exit ipv4_frags_exit_net(struct net *net)
925 ip4_frags_ns_ctl_unregister(net);
926 inet_frags_exit_net(&net->ipv4.frags);
929 static struct pernet_operations ip4_frags_ops = {
930 .init = ipv4_frags_init_net,
931 .exit = ipv4_frags_exit_net,
935 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
938 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
941 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
943 const struct inet_frag_queue *fq = data;
945 return jhash2((const u32 *)&fq->key.v4,
946 sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
949 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
951 const struct frag_v4_compare_key *key = arg->key;
952 const struct inet_frag_queue *fq = ptr;
954 return !!memcmp(&fq->key, key, sizeof(*key));
957 static const struct rhashtable_params ip4_rhash_params = {
958 .head_offset = offsetof(struct inet_frag_queue, node),
959 .key_offset = offsetof(struct inet_frag_queue, key),
960 .key_len = sizeof(struct frag_v4_compare_key),
961 .hashfn = ip4_key_hashfn,
962 .obj_hashfn = ip4_obj_hashfn,
963 .obj_cmpfn = ip4_obj_cmpfn,
964 .automatic_shrinking = true,
967 void __init ipfrag_init(void)
969 ip4_frags.constructor = ip4_frag_init;
970 ip4_frags.destructor = ip4_frag_free;
971 ip4_frags.skb_free = NULL;
972 ip4_frags.qsize = sizeof(struct ipq);
973 ip4_frags.frag_expire = ip_expire;
974 ip4_frags.frags_cache_name = ip_frag_cache_name;
975 ip4_frags.rhash_params = ip4_rhash_params;
976 if (inet_frags_init(&ip4_frags))
977 panic("IP: failed to allocate ip4_frags cache\n");
978 ip4_frags_ctl_register();
979 register_pernet_subsys(&ip4_frags_ops);