mfd: wm8350-i2c: Make sure the i2c regmap functions are compiled
[linux/fpc-iii.git] / net / ipv4 / ip_fragment.c
blob04c7e4618008732e63cdaa983108b49bb29cc087
1 /*
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>
11 * Fixes:
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>
28 #include <linux/mm.h>
29 #include <linux/jiffies.h>
30 #include <linux/skbuff.h>
31 #include <linux/list.h>
32 #include <linux/ip.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>
39 #include <net/dst.h>
40 #include <net/sock.h>
41 #include <net/ip.h>
42 #include <net/icmp.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>
52 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
53 * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
54 * as well. Or notify me, at least. --ANK
57 static int sysctl_ipfrag_max_dist __read_mostly = 64;
59 struct ipfrag_skb_cb
61 struct inet_skb_parm h;
62 int offset;
65 #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb))
67 /* Describe an entry in the "incomplete datagrams" queue. */
68 struct ipq {
69 struct inet_frag_queue q;
71 u32 user;
72 __be32 saddr;
73 __be32 daddr;
74 __be16 id;
75 u8 protocol;
76 u8 ecn; /* RFC3168 support */
77 int iif;
78 unsigned int rid;
79 struct inet_peer *peer;
82 static inline u8 ip4_frag_ecn(u8 tos)
84 return 1 << (tos & INET_ECN_MASK);
87 static struct inet_frags ip4_frags;
89 int ip_frag_nqueues(struct net *net)
91 return net->ipv4.frags.nqueues;
94 int ip_frag_mem(struct net *net)
96 return sum_frag_mem_limit(&net->ipv4.frags);
99 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
100 struct net_device *dev);
102 struct ip4_create_arg {
103 struct iphdr *iph;
104 u32 user;
107 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
109 return jhash_3words((__force u32)id << 16 | prot,
110 (__force u32)saddr, (__force u32)daddr,
111 ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
114 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
116 struct ipq *ipq;
118 ipq = container_of(q, struct ipq, q);
119 return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
122 static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
124 struct ipq *qp;
125 struct ip4_create_arg *arg = a;
127 qp = container_of(q, struct ipq, q);
128 return qp->id == arg->iph->id &&
129 qp->saddr == arg->iph->saddr &&
130 qp->daddr == arg->iph->daddr &&
131 qp->protocol == arg->iph->protocol &&
132 qp->user == arg->user;
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 netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
139 frags);
140 struct net *net = container_of(ipv4, struct net, ipv4);
142 struct ip4_create_arg *arg = a;
144 qp->protocol = arg->iph->protocol;
145 qp->id = arg->iph->id;
146 qp->ecn = ip4_frag_ecn(arg->iph->tos);
147 qp->saddr = arg->iph->saddr;
148 qp->daddr = arg->iph->daddr;
149 qp->user = arg->user;
150 qp->peer = sysctl_ipfrag_max_dist ?
151 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
154 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
156 struct ipq *qp;
158 qp = container_of(q, struct ipq, q);
159 if (qp->peer)
160 inet_putpeer(qp->peer);
164 /* Destruction primitives. */
166 static __inline__ void ipq_put(struct ipq *ipq)
168 inet_frag_put(&ipq->q, &ip4_frags);
171 /* Kill ipq entry. It is not destroyed immediately,
172 * because caller (and someone more) holds reference count.
174 static void ipq_kill(struct ipq *ipq)
176 inet_frag_kill(&ipq->q, &ip4_frags);
179 /* Memory limiting on fragments. Evictor trashes the oldest
180 * fragment queue until we are back under the threshold.
182 static void ip_evictor(struct net *net)
184 int evicted;
186 evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
187 if (evicted)
188 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
192 * Oops, a fragment queue timed out. Kill it and send an ICMP reply.
194 static void ip_expire(unsigned long arg)
196 struct ipq *qp;
197 struct net *net;
199 qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
200 net = container_of(qp->q.net, struct net, ipv4.frags);
202 spin_lock(&qp->q.lock);
204 if (qp->q.last_in & INET_FRAG_COMPLETE)
205 goto out;
207 ipq_kill(qp);
209 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
210 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
212 if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
213 struct sk_buff *head = qp->q.fragments;
214 const struct iphdr *iph;
215 int err;
217 rcu_read_lock();
218 head->dev = dev_get_by_index_rcu(net, qp->iif);
219 if (!head->dev)
220 goto out_rcu_unlock;
222 /* skb has no dst, perform route lookup again */
223 iph = ip_hdr(head);
224 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
225 iph->tos, head->dev);
226 if (err)
227 goto out_rcu_unlock;
230 * Only an end host needs to send an ICMP
231 * "Fragment Reassembly Timeout" message, per RFC792.
233 if (qp->user == IP_DEFRAG_AF_PACKET ||
234 (qp->user == IP_DEFRAG_CONNTRACK_IN &&
235 skb_rtable(head)->rt_type != RTN_LOCAL))
236 goto out_rcu_unlock;
239 /* Send an ICMP "Fragment Reassembly Timeout" message. */
240 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
241 out_rcu_unlock:
242 rcu_read_unlock();
244 out:
245 spin_unlock(&qp->q.lock);
246 ipq_put(qp);
249 /* Find the correct entry in the "incomplete datagrams" queue for
250 * this IP datagram, and create new one, if nothing is found.
252 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
254 struct inet_frag_queue *q;
255 struct ip4_create_arg arg;
256 unsigned int hash;
258 arg.iph = iph;
259 arg.user = user;
261 read_lock(&ip4_frags.lock);
262 hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
264 q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
265 if (IS_ERR_OR_NULL(q)) {
266 inet_frag_maybe_warn_overflow(q, pr_fmt());
267 return NULL;
269 return container_of(q, struct ipq, q);
272 /* Is the fragment too far ahead to be part of ipq? */
273 static inline int ip_frag_too_far(struct ipq *qp)
275 struct inet_peer *peer = qp->peer;
276 unsigned int max = sysctl_ipfrag_max_dist;
277 unsigned int start, end;
279 int rc;
281 if (!peer || !max)
282 return 0;
284 start = qp->rid;
285 end = atomic_inc_return(&peer->rid);
286 qp->rid = end;
288 rc = qp->q.fragments && (end - start) > max;
290 if (rc) {
291 struct net *net;
293 net = container_of(qp->q.net, struct net, ipv4.frags);
294 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
297 return rc;
300 static int ip_frag_reinit(struct ipq *qp)
302 struct sk_buff *fp;
303 unsigned int sum_truesize = 0;
305 if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
306 atomic_inc(&qp->q.refcnt);
307 return -ETIMEDOUT;
310 fp = qp->q.fragments;
311 do {
312 struct sk_buff *xp = fp->next;
314 sum_truesize += fp->truesize;
315 kfree_skb(fp);
316 fp = xp;
317 } while (fp);
318 sub_frag_mem_limit(&qp->q, sum_truesize);
320 qp->q.last_in = 0;
321 qp->q.len = 0;
322 qp->q.meat = 0;
323 qp->q.fragments = NULL;
324 qp->q.fragments_tail = NULL;
325 qp->iif = 0;
326 qp->ecn = 0;
328 return 0;
331 /* Add new segment to existing queue. */
332 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
334 struct sk_buff *prev, *next;
335 struct net_device *dev;
336 int flags, offset;
337 int ihl, end;
338 int err = -ENOENT;
339 u8 ecn;
341 if (qp->q.last_in & INET_FRAG_COMPLETE)
342 goto err;
344 if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
345 unlikely(ip_frag_too_far(qp)) &&
346 unlikely(err = ip_frag_reinit(qp))) {
347 ipq_kill(qp);
348 goto err;
351 ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
352 offset = ntohs(ip_hdr(skb)->frag_off);
353 flags = offset & ~IP_OFFSET;
354 offset &= IP_OFFSET;
355 offset <<= 3; /* offset is in 8-byte chunks */
356 ihl = ip_hdrlen(skb);
358 /* Determine the position of this fragment. */
359 end = offset + skb->len - skb_network_offset(skb) - ihl;
360 err = -EINVAL;
362 /* Is this the final fragment? */
363 if ((flags & IP_MF) == 0) {
364 /* If we already have some bits beyond end
365 * or have different end, the segment is corrupted.
367 if (end < qp->q.len ||
368 ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
369 goto err;
370 qp->q.last_in |= INET_FRAG_LAST_IN;
371 qp->q.len = end;
372 } else {
373 if (end&7) {
374 end &= ~7;
375 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
376 skb->ip_summed = CHECKSUM_NONE;
378 if (end > qp->q.len) {
379 /* Some bits beyond end -> corruption. */
380 if (qp->q.last_in & INET_FRAG_LAST_IN)
381 goto err;
382 qp->q.len = end;
385 if (end == offset)
386 goto err;
388 err = -ENOMEM;
389 if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
390 goto err;
392 err = pskb_trim_rcsum(skb, end - offset);
393 if (err)
394 goto err;
396 /* Find out which fragments are in front and at the back of us
397 * in the chain of fragments so far. We must know where to put
398 * this fragment, right?
400 prev = qp->q.fragments_tail;
401 if (!prev || FRAG_CB(prev)->offset < offset) {
402 next = NULL;
403 goto found;
405 prev = NULL;
406 for (next = qp->q.fragments; next != NULL; next = next->next) {
407 if (FRAG_CB(next)->offset >= offset)
408 break; /* bingo! */
409 prev = next;
412 found:
413 /* We found where to put this one. Check for overlap with
414 * preceding fragment, and, if needed, align things so that
415 * any overlaps are eliminated.
417 if (prev) {
418 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
420 if (i > 0) {
421 offset += i;
422 err = -EINVAL;
423 if (end <= offset)
424 goto err;
425 err = -ENOMEM;
426 if (!pskb_pull(skb, i))
427 goto err;
428 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
429 skb->ip_summed = CHECKSUM_NONE;
433 err = -ENOMEM;
435 while (next && FRAG_CB(next)->offset < end) {
436 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
438 if (i < next->len) {
439 /* Eat head of the next overlapped fragment
440 * and leave the loop. The next ones cannot overlap.
442 if (!pskb_pull(next, i))
443 goto err;
444 FRAG_CB(next)->offset += i;
445 qp->q.meat -= i;
446 if (next->ip_summed != CHECKSUM_UNNECESSARY)
447 next->ip_summed = CHECKSUM_NONE;
448 break;
449 } else {
450 struct sk_buff *free_it = next;
452 /* Old fragment is completely overridden with
453 * new one drop it.
455 next = next->next;
457 if (prev)
458 prev->next = next;
459 else
460 qp->q.fragments = next;
462 qp->q.meat -= free_it->len;
463 sub_frag_mem_limit(&qp->q, free_it->truesize);
464 kfree_skb(free_it);
468 FRAG_CB(skb)->offset = offset;
470 /* Insert this fragment in the chain of fragments. */
471 skb->next = next;
472 if (!next)
473 qp->q.fragments_tail = skb;
474 if (prev)
475 prev->next = skb;
476 else
477 qp->q.fragments = skb;
479 dev = skb->dev;
480 if (dev) {
481 qp->iif = dev->ifindex;
482 skb->dev = NULL;
484 qp->q.stamp = skb->tstamp;
485 qp->q.meat += skb->len;
486 qp->ecn |= ecn;
487 add_frag_mem_limit(&qp->q, skb->truesize);
488 if (offset == 0)
489 qp->q.last_in |= INET_FRAG_FIRST_IN;
491 if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
492 skb->len + ihl > qp->q.max_size)
493 qp->q.max_size = skb->len + ihl;
495 if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
496 qp->q.meat == qp->q.len) {
497 unsigned long orefdst = skb->_skb_refdst;
499 skb->_skb_refdst = 0UL;
500 err = ip_frag_reasm(qp, prev, dev);
501 skb->_skb_refdst = orefdst;
502 return err;
505 skb_dst_drop(skb);
506 inet_frag_lru_move(&qp->q);
507 return -EINPROGRESS;
509 err:
510 kfree_skb(skb);
511 return err;
515 /* Build a new IP datagram from all its fragments. */
517 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
518 struct net_device *dev)
520 struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
521 struct iphdr *iph;
522 struct sk_buff *fp, *head = qp->q.fragments;
523 int len;
524 int ihlen;
525 int err;
526 int sum_truesize;
527 u8 ecn;
529 ipq_kill(qp);
531 ecn = ip_frag_ecn_table[qp->ecn];
532 if (unlikely(ecn == 0xff)) {
533 err = -EINVAL;
534 goto out_fail;
536 /* Make the one we just received the head. */
537 if (prev) {
538 head = prev->next;
539 fp = skb_clone(head, GFP_ATOMIC);
540 if (!fp)
541 goto out_nomem;
543 fp->next = head->next;
544 if (!fp->next)
545 qp->q.fragments_tail = fp;
546 prev->next = fp;
548 skb_morph(head, qp->q.fragments);
549 head->next = qp->q.fragments->next;
551 consume_skb(qp->q.fragments);
552 qp->q.fragments = head;
555 WARN_ON(head == NULL);
556 WARN_ON(FRAG_CB(head)->offset != 0);
558 /* Allocate a new buffer for the datagram. */
559 ihlen = ip_hdrlen(head);
560 len = ihlen + qp->q.len;
562 err = -E2BIG;
563 if (len > 65535)
564 goto out_oversize;
566 /* Head of list must not be cloned. */
567 if (skb_unclone(head, GFP_ATOMIC))
568 goto out_nomem;
570 /* If the first fragment is fragmented itself, we split
571 * it to two chunks: the first with data and paged part
572 * and the second, holding only fragments. */
573 if (skb_has_frag_list(head)) {
574 struct sk_buff *clone;
575 int i, plen = 0;
577 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
578 goto out_nomem;
579 clone->next = head->next;
580 head->next = clone;
581 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
582 skb_frag_list_init(head);
583 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
584 plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
585 clone->len = clone->data_len = head->data_len - plen;
586 head->data_len -= clone->len;
587 head->len -= clone->len;
588 clone->csum = 0;
589 clone->ip_summed = head->ip_summed;
590 add_frag_mem_limit(&qp->q, clone->truesize);
593 skb_push(head, head->data - skb_network_header(head));
595 sum_truesize = head->truesize;
596 for (fp = head->next; fp;) {
597 bool headstolen;
598 int delta;
599 struct sk_buff *next = fp->next;
601 sum_truesize += fp->truesize;
602 if (head->ip_summed != fp->ip_summed)
603 head->ip_summed = CHECKSUM_NONE;
604 else if (head->ip_summed == CHECKSUM_COMPLETE)
605 head->csum = csum_add(head->csum, fp->csum);
607 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
608 kfree_skb_partial(fp, headstolen);
609 } else {
610 if (!skb_shinfo(head)->frag_list)
611 skb_shinfo(head)->frag_list = fp;
612 head->data_len += fp->len;
613 head->len += fp->len;
614 head->truesize += fp->truesize;
616 fp = next;
618 sub_frag_mem_limit(&qp->q, sum_truesize);
620 head->next = NULL;
621 head->dev = dev;
622 head->tstamp = qp->q.stamp;
623 IPCB(head)->frag_max_size = qp->q.max_size;
625 iph = ip_hdr(head);
626 /* max_size != 0 implies at least one fragment had IP_DF set */
627 iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
628 iph->tot_len = htons(len);
629 iph->tos |= ecn;
631 ip_send_check(iph);
633 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
634 qp->q.fragments = NULL;
635 qp->q.fragments_tail = NULL;
636 return 0;
638 out_nomem:
639 LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
640 qp);
641 err = -ENOMEM;
642 goto out_fail;
643 out_oversize:
644 net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
645 out_fail:
646 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
647 return err;
650 /* Process an incoming IP datagram fragment. */
651 int ip_defrag(struct sk_buff *skb, u32 user)
653 struct ipq *qp;
654 struct net *net;
656 net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
657 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
659 if (!net->ipv4.frags.high_thresh)
660 goto fail;
662 /* Start by cleaning up the memory. */
663 ip_evictor(net);
665 /* Lookup (or create) queue header */
666 if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
667 int ret;
669 spin_lock(&qp->q.lock);
671 ret = ip_frag_queue(qp, skb);
673 spin_unlock(&qp->q.lock);
674 ipq_put(qp);
675 return ret;
678 fail:
679 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
680 kfree_skb(skb);
681 return -ENOMEM;
683 EXPORT_SYMBOL(ip_defrag);
685 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
687 struct iphdr iph;
688 int netoff;
689 u32 len;
691 if (skb->protocol != htons(ETH_P_IP))
692 return skb;
694 netoff = skb_network_offset(skb);
696 if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
697 return skb;
699 if (iph.ihl < 5 || iph.version != 4)
700 return skb;
702 len = ntohs(iph.tot_len);
703 if (skb->len < netoff + len || len < (iph.ihl * 4))
704 return skb;
706 if (ip_is_fragment(&iph)) {
707 skb = skb_share_check(skb, GFP_ATOMIC);
708 if (skb) {
709 if (!pskb_may_pull(skb, netoff + iph.ihl * 4))
710 return skb;
711 if (pskb_trim_rcsum(skb, netoff + len))
712 return skb;
713 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
714 if (ip_defrag(skb, user))
715 return NULL;
716 skb->rxhash = 0;
719 return skb;
721 EXPORT_SYMBOL(ip_check_defrag);
723 #ifdef CONFIG_SYSCTL
724 static int zero;
726 static struct ctl_table ip4_frags_ns_ctl_table[] = {
728 .procname = "ipfrag_high_thresh",
729 .data = &init_net.ipv4.frags.high_thresh,
730 .maxlen = sizeof(int),
731 .mode = 0644,
732 .proc_handler = proc_dointvec
735 .procname = "ipfrag_low_thresh",
736 .data = &init_net.ipv4.frags.low_thresh,
737 .maxlen = sizeof(int),
738 .mode = 0644,
739 .proc_handler = proc_dointvec
742 .procname = "ipfrag_time",
743 .data = &init_net.ipv4.frags.timeout,
744 .maxlen = sizeof(int),
745 .mode = 0644,
746 .proc_handler = proc_dointvec_jiffies,
751 static struct ctl_table ip4_frags_ctl_table[] = {
753 .procname = "ipfrag_secret_interval",
754 .data = &ip4_frags.secret_interval,
755 .maxlen = sizeof(int),
756 .mode = 0644,
757 .proc_handler = proc_dointvec_jiffies,
760 .procname = "ipfrag_max_dist",
761 .data = &sysctl_ipfrag_max_dist,
762 .maxlen = sizeof(int),
763 .mode = 0644,
764 .proc_handler = proc_dointvec_minmax,
765 .extra1 = &zero
770 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
772 struct ctl_table *table;
773 struct ctl_table_header *hdr;
775 table = ip4_frags_ns_ctl_table;
776 if (!net_eq(net, &init_net)) {
777 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
778 if (table == NULL)
779 goto err_alloc;
781 table[0].data = &net->ipv4.frags.high_thresh;
782 table[1].data = &net->ipv4.frags.low_thresh;
783 table[2].data = &net->ipv4.frags.timeout;
785 /* Don't export sysctls to unprivileged users */
786 if (net->user_ns != &init_user_ns)
787 table[0].procname = NULL;
790 hdr = register_net_sysctl(net, "net/ipv4", table);
791 if (hdr == NULL)
792 goto err_reg;
794 net->ipv4.frags_hdr = hdr;
795 return 0;
797 err_reg:
798 if (!net_eq(net, &init_net))
799 kfree(table);
800 err_alloc:
801 return -ENOMEM;
804 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
806 struct ctl_table *table;
808 table = net->ipv4.frags_hdr->ctl_table_arg;
809 unregister_net_sysctl_table(net->ipv4.frags_hdr);
810 kfree(table);
813 static void ip4_frags_ctl_register(void)
815 register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
817 #else
818 static inline int ip4_frags_ns_ctl_register(struct net *net)
820 return 0;
823 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
827 static inline void ip4_frags_ctl_register(void)
830 #endif
832 static int __net_init ipv4_frags_init_net(struct net *net)
834 /* Fragment cache limits.
836 * The fragment memory accounting code, (tries to) account for
837 * the real memory usage, by measuring both the size of frag
838 * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
839 * and the SKB's truesize.
841 * A 64K fragment consumes 129736 bytes (44*2944)+200
842 * (1500 truesize == 2944, sizeof(struct ipq) == 200)
844 * We will commit 4MB at one time. Should we cross that limit
845 * we will prune down to 3MB, making room for approx 8 big 64K
846 * fragments 8x128k.
848 net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
849 net->ipv4.frags.low_thresh = 3 * 1024 * 1024;
851 * Important NOTE! Fragment queue must be destroyed before MSL expires.
852 * RFC791 is wrong proposing to prolongate timer each fragment arrival
853 * by TTL.
855 net->ipv4.frags.timeout = IP_FRAG_TIME;
857 inet_frags_init_net(&net->ipv4.frags);
859 return ip4_frags_ns_ctl_register(net);
862 static void __net_exit ipv4_frags_exit_net(struct net *net)
864 ip4_frags_ns_ctl_unregister(net);
865 inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
868 static struct pernet_operations ip4_frags_ops = {
869 .init = ipv4_frags_init_net,
870 .exit = ipv4_frags_exit_net,
873 void __init ipfrag_init(void)
875 ip4_frags_ctl_register();
876 register_pernet_subsys(&ip4_frags_ops);
877 ip4_frags.hashfn = ip4_hashfn;
878 ip4_frags.constructor = ip4_frag_init;
879 ip4_frags.destructor = ip4_frag_free;
880 ip4_frags.skb_free = NULL;
881 ip4_frags.qsize = sizeof(struct ipq);
882 ip4_frags.match = ip4_frag_match;
883 ip4_frags.frag_expire = ip_expire;
884 ip4_frags.secret_interval = 10 * 60 * HZ;
885 inet_frags_init(&ip4_frags);