net/ipv4/ip_fragment.c

   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.
   5  *
   6  *              The IP fragmentation functionality.
   7  *
   8  * Version:     $Id: ip_fragment.c,v 1.59 2002/01/12 07:54:56 davem Exp $
   9  *
  10  * Authors:     Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Alan Cox <Alan.Cox@linux.org>
  12  *
  13  * Fixes:
  14  *              Alan Cox        :       Split from ip.c , see ip_input.c for history.
  15  *              David S. Miller :       Begin massive cleanup...
  16  *              Andi Kleen      :       Add sysctls.
  17  *              xxxx            :       Overlapfrag bug.
  18  *              Ultima          :       ip_expire() kernel panic.
  19  *              Bill Hawes      :       Frag accounting and evictor fixes.
  20  *              John McDonald   :       0 length frag bug.
  21  *              Alexey Kuznetsov:       SMP races, threading, cleanup.
  22  *              Patrick McHardy :       LRU queue of frag heads for evictor.
  23  */
  24
  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 <net/sock.h>
  38 #include <net/ip.h>
  39 #include <net/icmp.h>
  40 #include <net/checksum.h>
  41 #include <net/inetpeer.h>
  42 #include <net/inet_frag.h>
  43 #include <linux/tcp.h>
  44 #include <linux/udp.h>
  45 #include <linux/inet.h>
  46 #include <linux/netfilter_ipv4.h>
  47
  48 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
  49  * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
  50  * as well. Or notify me, at least. --ANK
  51  */
  52
  53 static int sysctl_ipfrag_max_dist __read_mostly = 64;
  54
  55 struct ipfrag_skb_cb
  56 {
  57         struct inet_skb_parm    h;
  58         int                     offset;
  59 };
  60
  61 #define FRAG_CB(skb)    ((struct ipfrag_skb_cb*)((skb)->cb))
  62
  63 /* Describe an entry in the "incomplete datagrams" queue. */
  64 struct ipq {
  65         struct inet_frag_queue q;
  66
  67         u32             user;
  68         __be32          saddr;
  69         __be32          daddr;
  70         __be16          id;
  71         u8              protocol;
  72         int             iif;
  73         unsigned int    rid;
  74         struct inet_peer *peer;
  75 };
  76
  77 static struct inet_frags_ctl ip4_frags_ctl __read_mostly = {
  78         /*
  79          * Fragment cache limits. We will commit 256K at one time. Should we
  80          * cross that limit we will prune down to 192K. This should cope with
  81          * even the most extreme cases without allowing an attacker to
  82          * measurably harm machine performance.
  83          */
  84         .high_thresh     = 256 * 1024,
  85         .low_thresh      = 192 * 1024,
  86         .secret_interval = 10 * 60 * HZ,
  87 };
  88
  89 static struct inet_frags ip4_frags;
  90
  91 int ip_frag_nqueues(struct net *net)
  92 {
  93         return net->ipv4.frags.nqueues;
  94 }
  95
  96 int ip_frag_mem(struct net *net)
  97 {
  98         return atomic_read(&net->ipv4.frags.mem);
  99 }
 100
 101 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 102                          struct net_device *dev);
 103
 104 struct ip4_create_arg {
 105         struct iphdr *iph;
 106         u32 user;
 107 };
 108
 109 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
 110 {
 111         return jhash_3words((__force u32)id << 16 | prot,
 112                             (__force u32)saddr, (__force u32)daddr,
 113                             ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
 114 }
 115
 116 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
 117 {
 118         struct ipq *ipq;
 119
 120         ipq = container_of(q, struct ipq, q);
 121         return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
 122 }
 123
 124 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
 125 {
 126         struct ipq *qp;
 127         struct ip4_create_arg *arg = a;
 128
 129         qp = container_of(q, struct ipq, q);
 130         return (qp->id == arg->iph->id &&
 131                         qp->saddr == arg->iph->saddr &&
 132                         qp->daddr == arg->iph->daddr &&
 133                         qp->protocol == arg->iph->protocol &&
 134                         qp->user == arg->user);
 135 }
 136
 137 /* Memory Tracking Functions. */
 138 static __inline__ void frag_kfree_skb(struct netns_frags *nf,
 139                 struct sk_buff *skb, int *work)
 140 {
 141         if (work)
 142                 *work -= skb->truesize;
 143         atomic_sub(skb->truesize, &nf->mem);
 144         kfree_skb(skb);
 145 }
 146
 147 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
 148 {
 149         struct ipq *qp = container_of(q, struct ipq, q);
 150         struct ip4_create_arg *arg = a;
 151
 152         qp->protocol = arg->iph->protocol;
 153         qp->id = arg->iph->id;
 154         qp->saddr = arg->iph->saddr;
 155         qp->daddr = arg->iph->daddr;
 156         qp->user = arg->user;
 157         qp->peer = sysctl_ipfrag_max_dist ?
 158                 inet_getpeer(arg->iph->saddr, 1) : NULL;
 159 }
 160
 161 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
 162 {
 163         struct ipq *qp;
 164
 165         qp = container_of(q, struct ipq, q);
 166         if (qp->peer)
 167                 inet_putpeer(qp->peer);
 168 }
 169
 170
 171 /* Destruction primitives. */
 172
 173 static __inline__ void ipq_put(struct ipq *ipq)
 174 {
 175         inet_frag_put(&ipq->q, &ip4_frags);
 176 }
 177
 178 /* Kill ipq entry. It is not destroyed immediately,
 179  * because caller (and someone more) holds reference count.
 180  */
 181 static void ipq_kill(struct ipq *ipq)
 182 {
 183         inet_frag_kill(&ipq->q, &ip4_frags);
 184 }
 185
 186 /* Memory limiting on fragments.  Evictor trashes the oldest
 187  * fragment queue until we are back under the threshold.
 188  */
 189 static void ip_evictor(struct net *net)
 190 {
 191         int evicted;
 192
 193         evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
 194         if (evicted)
 195                 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
 196 }
 197
 198 /*
 199  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
 200  */
 201 static void ip_expire(unsigned long arg)
 202 {
 203         struct ipq *qp;
 204
 205         qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
 206
 207         spin_lock(&qp->q.lock);
 208
 209         if (qp->q.last_in & COMPLETE)
 210                 goto out;
 211
 212         ipq_kill(qp);
 213
 214         IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
 215         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 216
 217         if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
 218                 struct sk_buff *head = qp->q.fragments;
 219                 /* Send an ICMP "Fragment Reassembly Timeout" message. */
 220                 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
 221                         icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
 222                         dev_put(head->dev);
 223                 }
 224         }
 225 out:
 226         spin_unlock(&qp->q.lock);
 227         ipq_put(qp);
 228 }
 229
 230 /* Find the correct entry in the "incomplete datagrams" queue for
 231  * this IP datagram, and create new one, if nothing is found.
 232  */
 233 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
 234 {
 235         struct inet_frag_queue *q;
 236         struct ip4_create_arg arg;
 237         unsigned int hash;
 238
 239         arg.iph = iph;
 240         arg.user = user;
 241         hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
 242
 243         q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
 244         if (q == NULL)
 245                 goto out_nomem;
 246
 247         return container_of(q, struct ipq, q);
 248
 249 out_nomem:
 250         LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
 251         return NULL;
 252 }
 253
 254 /* Is the fragment too far ahead to be part of ipq? */
 255 static inline int ip_frag_too_far(struct ipq *qp)
 256 {
 257         struct inet_peer *peer = qp->peer;
 258         unsigned int max = sysctl_ipfrag_max_dist;
 259         unsigned int start, end;
 260
 261         int rc;
 262
 263         if (!peer || !max)
 264                 return 0;
 265
 266         start = qp->rid;
 267         end = atomic_inc_return(&peer->rid);
 268         qp->rid = end;
 269
 270         rc = qp->q.fragments && (end - start) > max;
 271
 272         if (rc) {
 273                 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 274         }
 275
 276         return rc;
 277 }
 278
 279 static int ip_frag_reinit(struct ipq *qp)
 280 {
 281         struct sk_buff *fp;
 282
 283         if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
 284                 atomic_inc(&qp->q.refcnt);
 285                 return -ETIMEDOUT;
 286         }
 287
 288         fp = qp->q.fragments;
 289         do {
 290                 struct sk_buff *xp = fp->next;
 291                 frag_kfree_skb(qp->q.net, fp, NULL);
 292                 fp = xp;
 293         } while (fp);
 294
 295         qp->q.last_in = 0;
 296         qp->q.len = 0;
 297         qp->q.meat = 0;
 298         qp->q.fragments = NULL;
 299         qp->iif = 0;
 300
 301         return 0;
 302 }
 303
 304 /* Add new segment to existing queue. */
 305 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
 306 {
 307         struct sk_buff *prev, *next;
 308         struct net_device *dev;
 309         int flags, offset;
 310         int ihl, end;
 311         int err = -ENOENT;
 312
 313         if (qp->q.last_in & COMPLETE)
 314                 goto err;
 315
 316         if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
 317             unlikely(ip_frag_too_far(qp)) &&
 318             unlikely(err = ip_frag_reinit(qp))) {
 319                 ipq_kill(qp);
 320                 goto err;
 321         }
 322
 323         offset = ntohs(ip_hdr(skb)->frag_off);
 324         flags = offset & ~IP_OFFSET;
 325         offset &= IP_OFFSET;
 326         offset <<= 3;           /* offset is in 8-byte chunks */
 327         ihl = ip_hdrlen(skb);
 328
 329         /* Determine the position of this fragment. */
 330         end = offset + skb->len - ihl;
 331         err = -EINVAL;
 332
 333         /* Is this the final fragment? */
 334         if ((flags & IP_MF) == 0) {
 335                 /* If we already have some bits beyond end
 336                  * or have different end, the segment is corrrupted.
 337                  */
 338                 if (end < qp->q.len ||
 339                     ((qp->q.last_in & LAST_IN) && end != qp->q.len))
 340                         goto err;
 341                 qp->q.last_in |= LAST_IN;
 342                 qp->q.len = end;
 343         } else {
 344                 if (end&7) {
 345                         end &= ~7;
 346                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 347                                 skb->ip_summed = CHECKSUM_NONE;
 348                 }
 349                 if (end > qp->q.len) {
 350                         /* Some bits beyond end -> corruption. */
 351                         if (qp->q.last_in & LAST_IN)
 352                                 goto err;
 353                         qp->q.len = end;
 354                 }
 355         }
 356         if (end == offset)
 357                 goto err;
 358
 359         err = -ENOMEM;
 360         if (pskb_pull(skb, ihl) == NULL)
 361                 goto err;
 362
 363         err = pskb_trim_rcsum(skb, end - offset);
 364         if (err)
 365                 goto err;
 366
 367         /* Find out which fragments are in front and at the back of us
 368          * in the chain of fragments so far.  We must know where to put
 369          * this fragment, right?
 370          */
 371         prev = NULL;
 372         for (next = qp->q.fragments; next != NULL; next = next->next) {
 373                 if (FRAG_CB(next)->offset >= offset)
 374                         break;  /* bingo! */
 375                 prev = next;
 376         }
 377
 378         /* We found where to put this one.  Check for overlap with
 379          * preceding fragment, and, if needed, align things so that
 380          * any overlaps are eliminated.
 381          */
 382         if (prev) {
 383                 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
 384
 385                 if (i > 0) {
 386                         offset += i;
 387                         err = -EINVAL;
 388                         if (end <= offset)
 389                                 goto err;
 390                         err = -ENOMEM;
 391                         if (!pskb_pull(skb, i))
 392                                 goto err;
 393                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 394                                 skb->ip_summed = CHECKSUM_NONE;
 395                 }
 396         }
 397
 398         err = -ENOMEM;
 399
 400         while (next && FRAG_CB(next)->offset < end) {
 401                 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
 402
 403                 if (i < next->len) {
 404                         /* Eat head of the next overlapped fragment
 405                          * and leave the loop. The next ones cannot overlap.
 406                          */
 407                         if (!pskb_pull(next, i))
 408                                 goto err;
 409                         FRAG_CB(next)->offset += i;
 410                         qp->q.meat -= i;
 411                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 412                                 next->ip_summed = CHECKSUM_NONE;
 413                         break;
 414                 } else {
 415                         struct sk_buff *free_it = next;
 416
 417                         /* Old fragment is completely overridden with
 418                          * new one drop it.
 419                          */
 420                         next = next->next;
 421
 422                         if (prev)
 423                                 prev->next = next;
 424                         else
 425                                 qp->q.fragments = next;
 426
 427                         qp->q.meat -= free_it->len;
 428                         frag_kfree_skb(qp->q.net, free_it, NULL);
 429                 }
 430         }
 431
 432         FRAG_CB(skb)->offset = offset;
 433
 434         /* Insert this fragment in the chain of fragments. */
 435         skb->next = next;
 436         if (prev)
 437                 prev->next = skb;
 438         else
 439                 qp->q.fragments = skb;
 440
 441         dev = skb->dev;
 442         if (dev) {
 443                 qp->iif = dev->ifindex;
 444                 skb->dev = NULL;
 445         }
 446         qp->q.stamp = skb->tstamp;
 447         qp->q.meat += skb->len;
 448         atomic_add(skb->truesize, &qp->q.net->mem);
 449         if (offset == 0)
 450                 qp->q.last_in |= FIRST_IN;
 451
 452         if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
 453                 return ip_frag_reasm(qp, prev, dev);
 454
 455         write_lock(&ip4_frags.lock);
 456         list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
 457         write_unlock(&ip4_frags.lock);
 458         return -EINPROGRESS;
 459
 460 err:
 461         kfree_skb(skb);
 462         return err;
 463 }
 464
 465
 466 /* Build a new IP datagram from all its fragments. */
 467
 468 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 469                          struct net_device *dev)
 470 {
 471         struct iphdr *iph;
 472         struct sk_buff *fp, *head = qp->q.fragments;
 473         int len;
 474         int ihlen;
 475         int err;
 476
 477         ipq_kill(qp);
 478
 479         /* Make the one we just received the head. */
 480         if (prev) {
 481                 head = prev->next;
 482                 fp = skb_clone(head, GFP_ATOMIC);
 483                 if (!fp)
 484                         goto out_nomem;
 485
 486                 fp->next = head->next;
 487                 prev->next = fp;
 488
 489                 skb_morph(head, qp->q.fragments);
 490                 head->next = qp->q.fragments->next;
 491
 492                 kfree_skb(qp->q.fragments);
 493                 qp->q.fragments = head;
 494         }
 495
 496         BUG_TRAP(head != NULL);
 497         BUG_TRAP(FRAG_CB(head)->offset == 0);
 498
 499         /* Allocate a new buffer for the datagram. */
 500         ihlen = ip_hdrlen(head);
 501         len = ihlen + qp->q.len;
 502
 503         err = -E2BIG;
 504         if (len > 65535)
 505                 goto out_oversize;
 506
 507         /* Head of list must not be cloned. */
 508         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 509                 goto out_nomem;
 510
 511         /* If the first fragment is fragmented itself, we split
 512          * it to two chunks: the first with data and paged part
 513          * and the second, holding only fragments. */
 514         if (skb_shinfo(head)->frag_list) {
 515                 struct sk_buff *clone;
 516                 int i, plen = 0;
 517
 518                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 519                         goto out_nomem;
 520                 clone->next = head->next;
 521                 head->next = clone;
 522                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 523                 skb_shinfo(head)->frag_list = NULL;
 524                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 525                         plen += skb_shinfo(head)->frags[i].size;
 526                 clone->len = clone->data_len = head->data_len - plen;
 527                 head->data_len -= clone->len;
 528                 head->len -= clone->len;
 529                 clone->csum = 0;
 530                 clone->ip_summed = head->ip_summed;
 531                 atomic_add(clone->truesize, &qp->q.net->mem);
 532         }
 533
 534         skb_shinfo(head)->frag_list = head->next;
 535         skb_push(head, head->data - skb_network_header(head));
 536         atomic_sub(head->truesize, &qp->q.net->mem);
 537
 538         for (fp=head->next; fp; fp = fp->next) {
 539                 head->data_len += fp->len;
 540                 head->len += fp->len;
 541                 if (head->ip_summed != fp->ip_summed)
 542                         head->ip_summed = CHECKSUM_NONE;
 543                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 544                         head->csum = csum_add(head->csum, fp->csum);
 545                 head->truesize += fp->truesize;
 546                 atomic_sub(fp->truesize, &qp->q.net->mem);
 547         }
 548
 549         head->next = NULL;
 550         head->dev = dev;
 551         head->tstamp = qp->q.stamp;
 552
 553         iph = ip_hdr(head);
 554         iph->frag_off = 0;
 555         iph->tot_len = htons(len);
 556         IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
 557         qp->q.fragments = NULL;
 558         return 0;
 559
 560 out_nomem:
 561         LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
 562                               "queue %p\n", qp);
 563         err = -ENOMEM;
 564         goto out_fail;
 565 out_oversize:
 566         if (net_ratelimit())
 567                 printk(KERN_INFO
 568                         "Oversized IP packet from %d.%d.%d.%d.\n",
 569                         NIPQUAD(qp->saddr));
 570 out_fail:
 571         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 572         return err;
 573 }
 574
 575 /* Process an incoming IP datagram fragment. */
 576 int ip_defrag(struct sk_buff *skb, u32 user)
 577 {
 578         struct ipq *qp;
 579         struct net *net;
 580
 581         IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
 582
 583         net = skb->dev->nd_net;
 584         /* Start by cleaning up the memory. */
 585         if (atomic_read(&net->ipv4.frags.mem) > ip4_frags_ctl.high_thresh)
 586                 ip_evictor(net);
 587
 588         /* Lookup (or create) queue header */
 589         if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
 590                 int ret;
 591
 592                 spin_lock(&qp->q.lock);
 593
 594                 ret = ip_frag_queue(qp, skb);
 595
 596                 spin_unlock(&qp->q.lock);
 597                 ipq_put(qp);
 598                 return ret;
 599         }
 600
 601         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 602         kfree_skb(skb);
 603         return -ENOMEM;
 604 }
 605
 606 #ifdef CONFIG_SYSCTL
 607 static int zero;
 608
 609 static struct ctl_table ip4_frags_ctl_table[] = {
 610         {
 611                 .ctl_name       = NET_IPV4_IPFRAG_HIGH_THRESH,
 612                 .procname       = "ipfrag_high_thresh",
 613                 .data           = &ip4_frags_ctl.high_thresh,
 614                 .maxlen         = sizeof(int),
 615                 .mode           = 0644,
 616                 .proc_handler   = &proc_dointvec
 617         },
 618         {
 619                 .ctl_name       = NET_IPV4_IPFRAG_LOW_THRESH,
 620                 .procname       = "ipfrag_low_thresh",
 621                 .data           = &ip4_frags_ctl.low_thresh,
 622                 .maxlen         = sizeof(int),
 623                 .mode           = 0644,
 624                 .proc_handler   = &proc_dointvec
 625         },
 626         {
 627                 .ctl_name       = NET_IPV4_IPFRAG_TIME,
 628                 .procname       = "ipfrag_time",
 629                 .data           = &init_net.ipv4.frags.timeout,
 630                 .maxlen         = sizeof(int),
 631                 .mode           = 0644,
 632                 .proc_handler   = &proc_dointvec_jiffies,
 633                 .strategy       = &sysctl_jiffies
 634         },
 635         {
 636                 .ctl_name       = NET_IPV4_IPFRAG_SECRET_INTERVAL,
 637                 .procname       = "ipfrag_secret_interval",
 638                 .data           = &ip4_frags_ctl.secret_interval,
 639                 .maxlen         = sizeof(int),
 640                 .mode           = 0644,
 641                 .proc_handler   = &proc_dointvec_jiffies,
 642                 .strategy       = &sysctl_jiffies
 643         },
 644         {
 645                 .procname       = "ipfrag_max_dist",
 646                 .data           = &sysctl_ipfrag_max_dist,
 647                 .maxlen         = sizeof(int),
 648                 .mode           = 0644,
 649                 .proc_handler   = &proc_dointvec_minmax,
 650                 .extra1         = &zero
 651         },
 652         { }
 653 };
 654
 655 static int ip4_frags_ctl_register(struct net *net)
 656 {
 657         struct ctl_table *table;
 658         struct ctl_table_header *hdr;
 659
 660         table = ip4_frags_ctl_table;
 661         if (net != &init_net) {
 662                 table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL);
 663                 if (table == NULL)
 664                         goto err_alloc;
 665
 666                 table[0].mode &= ~0222;
 667                 table[1].mode &= ~0222;
 668                 table[2].data = &net->ipv4.frags.timeout;
 669                 table[3].mode &= ~0222;
 670                 table[4].mode &= ~0222;
 671         }
 672
 673         hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
 674         if (hdr == NULL)
 675                 goto err_reg;
 676
 677         net->ipv4.frags_hdr = hdr;
 678         return 0;
 679
 680 err_reg:
 681         if (net != &init_net)
 682                 kfree(table);
 683 err_alloc:
 684         return -ENOMEM;
 685 }
 686
 687 static void ip4_frags_ctl_unregister(struct net *net)
 688 {
 689         struct ctl_table *table;
 690
 691         table = net->ipv4.frags_hdr->ctl_table_arg;
 692         unregister_net_sysctl_table(net->ipv4.frags_hdr);
 693         kfree(table);
 694 }
 695 #else
 696 static inline int ip4_frags_ctl_register(struct net *net)
 697 {
 698         return 0;
 699 }
 700
 701 static inline void ip4_frags_ctl_unregister(struct net *net)
 702 {
 703 }
 704 #endif
 705
 706 static int ipv4_frags_init_net(struct net *net)
 707 {
 708         /*
 709          * Important NOTE! Fragment queue must be destroyed before MSL expires.
 710          * RFC791 is wrong proposing to prolongate timer each fragment arrival
 711          * by TTL.
 712          */
 713         net->ipv4.frags.timeout = IP_FRAG_TIME;
 714
 715         inet_frags_init_net(&net->ipv4.frags);
 716
 717         return ip4_frags_ctl_register(net);
 718 }
 719
 720 void __init ipfrag_init(void)
 721 {
 722         ipv4_frags_init_net(&init_net);
 723         ip4_frags.ctl = &ip4_frags_ctl;
 724         ip4_frags.hashfn = ip4_hashfn;
 725         ip4_frags.constructor = ip4_frag_init;
 726         ip4_frags.destructor = ip4_frag_free;
 727         ip4_frags.skb_free = NULL;
 728         ip4_frags.qsize = sizeof(struct ipq);
 729         ip4_frags.match = ip4_frag_match;
 730         ip4_frags.frag_expire = ip_expire;
 731         inet_frags_init(&ip4_frags);
 732 }
 733
 734 EXPORT_SYMBOL(ip_defrag);