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  * Authors:     Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
   9  *              Alan Cox <alan@lxorguk.ukuu.org.uk>
  10  *
  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.
  21  */
  22
  23 #include <linux/compiler.h>
  24 #include <linux/module.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/jiffies.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/list.h>
  30 #include <linux/ip.h>
  31 #include <linux/icmp.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/jhash.h>
  34 #include <linux/random.h>
  35 #include <linux/slab.h>
  36 #include <net/route.h>
  37 #include <net/dst.h>
  38 #include <net/sock.h>
  39 #include <net/ip.h>
  40 #include <net/icmp.h>
  41 #include <net/checksum.h>
  42 #include <net/inetpeer.h>
  43 #include <net/inet_frag.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/inet.h>
  47 #include <linux/netfilter_ipv4.h>
  48 #include <net/inet_ecn.h>
  49
  50 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
  51  * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
  52  * as well. Or notify me, at least. --ANK
  53  */
  54
  55 static int sysctl_ipfrag_max_dist __read_mostly = 64;
  56
  57 struct ipfrag_skb_cb
  58 {
  59         struct inet_skb_parm    h;
  60         int                     offset;
  61 };
  62
  63 #define FRAG_CB(skb)    ((struct ipfrag_skb_cb *)((skb)->cb))
  64
  65 /* Describe an entry in the "incomplete datagrams" queue. */
  66 struct ipq {
  67         struct inet_frag_queue q;
  68
  69         u32             user;
  70         __be32          saddr;
  71         __be32          daddr;
  72         __be16          id;
  73         u8              protocol;
  74         u8              ecn; /* RFC3168 support */
  75         int             iif;
  76         unsigned int    rid;
  77         struct inet_peer *peer;
  78 };
  79
  80 /* RFC 3168 support :
  81  * We want to check ECN values of all fragments, do detect invalid combinations.
  82  * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
  83  */
  84 #define IPFRAG_ECN_NOT_ECT      0x01 /* one frag had ECN_NOT_ECT */
  85 #define IPFRAG_ECN_ECT_1        0x02 /* one frag had ECN_ECT_1 */
  86 #define IPFRAG_ECN_ECT_0        0x04 /* one frag had ECN_ECT_0 */
  87 #define IPFRAG_ECN_CE           0x08 /* one frag had ECN_CE */
  88
  89 static inline u8 ip4_frag_ecn(u8 tos)
  90 {
  91         return 1 << (tos & INET_ECN_MASK);
  92 }
  93
  94 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
  95  * Value : 0xff if frame should be dropped.
  96  *         0 or INET_ECN_CE value, to be ORed in to final iph->tos field
  97  */
  98 static const u8 ip4_frag_ecn_table[16] = {
  99         /* at least one fragment had CE, and others ECT_0 or ECT_1 */
 100         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0]                      = INET_ECN_CE,
 101         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1]                      = INET_ECN_CE,
 102         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1]   = INET_ECN_CE,
 103
 104         /* invalid combinations : drop frame */
 105         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
 106         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
 107         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
 108         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
 109         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
 110         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
 111         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
 112 };
 113
 114 static struct inet_frags ip4_frags;
 115
 116 int ip_frag_nqueues(struct net *net)
 117 {
 118         return net->ipv4.frags.nqueues;
 119 }
 120
 121 int ip_frag_mem(struct net *net)
 122 {
 123         return atomic_read(&net->ipv4.frags.mem);
 124 }
 125
 126 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 127                          struct net_device *dev);
 128
 129 struct ip4_create_arg {
 130         struct iphdr *iph;
 131         u32 user;
 132 };
 133
 134 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
 135 {
 136         return jhash_3words((__force u32)id << 16 | prot,
 137                             (__force u32)saddr, (__force u32)daddr,
 138                             ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
 139 }
 140
 141 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
 142 {
 143         struct ipq *ipq;
 144
 145         ipq = container_of(q, struct ipq, q);
 146         return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
 147 }
 148
 149 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
 150 {
 151         struct ipq *qp;
 152         struct ip4_create_arg *arg = a;
 153
 154         qp = container_of(q, struct ipq, q);
 155         return  qp->id == arg->iph->id &&
 156                         qp->saddr == arg->iph->saddr &&
 157                         qp->daddr == arg->iph->daddr &&
 158                         qp->protocol == arg->iph->protocol &&
 159                         qp->user == arg->user;
 160 }
 161
 162 /* Memory Tracking Functions. */
 163 static void frag_kfree_skb(struct netns_frags *nf, struct sk_buff *skb)
 164 {
 165         atomic_sub(skb->truesize, &nf->mem);
 166         kfree_skb(skb);
 167 }
 168
 169 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
 170 {
 171         struct ipq *qp = container_of(q, struct ipq, q);
 172         struct ip4_create_arg *arg = a;
 173
 174         qp->protocol = arg->iph->protocol;
 175         qp->id = arg->iph->id;
 176         qp->ecn = ip4_frag_ecn(arg->iph->tos);
 177         qp->saddr = arg->iph->saddr;
 178         qp->daddr = arg->iph->daddr;
 179         qp->user = arg->user;
 180         qp->peer = sysctl_ipfrag_max_dist ?
 181                 inet_getpeer_v4(arg->iph->saddr, 1) : NULL;
 182 }
 183
 184 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
 185 {
 186         struct ipq *qp;
 187
 188         qp = container_of(q, struct ipq, q);
 189         if (qp->peer)
 190                 inet_putpeer(qp->peer);
 191 }
 192
 193
 194 /* Destruction primitives. */
 195
 196 static __inline__ void ipq_put(struct ipq *ipq)
 197 {
 198         inet_frag_put(&ipq->q, &ip4_frags);
 199 }
 200
 201 /* Kill ipq entry. It is not destroyed immediately,
 202  * because caller (and someone more) holds reference count.
 203  */
 204 static void ipq_kill(struct ipq *ipq)
 205 {
 206         inet_frag_kill(&ipq->q, &ip4_frags);
 207 }
 208
 209 /* Memory limiting on fragments.  Evictor trashes the oldest
 210  * fragment queue until we are back under the threshold.
 211  */
 212 static void ip_evictor(struct net *net)
 213 {
 214         int evicted;
 215
 216         evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
 217         if (evicted)
 218                 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
 219 }
 220
 221 /*
 222  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
 223  */
 224 static void ip_expire(unsigned long arg)
 225 {
 226         struct ipq *qp;
 227         struct net *net;
 228
 229         qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
 230         net = container_of(qp->q.net, struct net, ipv4.frags);
 231
 232         spin_lock(&qp->q.lock);
 233
 234         if (qp->q.last_in & INET_FRAG_COMPLETE)
 235                 goto out;
 236
 237         ipq_kill(qp);
 238
 239         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
 240         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 241
 242         if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
 243                 struct sk_buff *head = qp->q.fragments;
 244                 const struct iphdr *iph;
 245                 int err;
 246
 247                 rcu_read_lock();
 248                 head->dev = dev_get_by_index_rcu(net, qp->iif);
 249                 if (!head->dev)
 250                         goto out_rcu_unlock;
 251
 252                 /* skb dst is stale, drop it, and perform route lookup again */
 253                 skb_dst_drop(head);
 254                 iph = ip_hdr(head);
 255                 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
 256                                            iph->tos, head->dev);
 257                 if (err)
 258                         goto out_rcu_unlock;
 259
 260                 /*
 261                  * Only an end host needs to send an ICMP
 262                  * "Fragment Reassembly Timeout" message, per RFC792.
 263                  */
 264                 if (qp->user == IP_DEFRAG_AF_PACKET ||
 265                     (qp->user == IP_DEFRAG_CONNTRACK_IN &&
 266                      skb_rtable(head)->rt_type != RTN_LOCAL))
 267                         goto out_rcu_unlock;
 268
 269
 270                 /* Send an ICMP "Fragment Reassembly Timeout" message. */
 271                 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
 272 out_rcu_unlock:
 273                 rcu_read_unlock();
 274         }
 275 out:
 276         spin_unlock(&qp->q.lock);
 277         ipq_put(qp);
 278 }
 279
 280 /* Find the correct entry in the "incomplete datagrams" queue for
 281  * this IP datagram, and create new one, if nothing is found.
 282  */
 283 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
 284 {
 285         struct inet_frag_queue *q;
 286         struct ip4_create_arg arg;
 287         unsigned int hash;
 288
 289         arg.iph = iph;
 290         arg.user = user;
 291
 292         read_lock(&ip4_frags.lock);
 293         hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
 294
 295         q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
 296         if (q == NULL)
 297                 goto out_nomem;
 298
 299         return container_of(q, struct ipq, q);
 300
 301 out_nomem:
 302         LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
 303         return NULL;
 304 }
 305
 306 /* Is the fragment too far ahead to be part of ipq? */
 307 static inline int ip_frag_too_far(struct ipq *qp)
 308 {
 309         struct inet_peer *peer = qp->peer;
 310         unsigned int max = sysctl_ipfrag_max_dist;
 311         unsigned int start, end;
 312
 313         int rc;
 314
 315         if (!peer || !max)
 316                 return 0;
 317
 318         start = qp->rid;
 319         end = atomic_inc_return(&peer->rid);
 320         qp->rid = end;
 321
 322         rc = qp->q.fragments && (end - start) > max;
 323
 324         if (rc) {
 325                 struct net *net;
 326
 327                 net = container_of(qp->q.net, struct net, ipv4.frags);
 328                 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 329         }
 330
 331         return rc;
 332 }
 333
 334 static int ip_frag_reinit(struct ipq *qp)
 335 {
 336         struct sk_buff *fp;
 337
 338         if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
 339                 atomic_inc(&qp->q.refcnt);
 340                 return -ETIMEDOUT;
 341         }
 342
 343         fp = qp->q.fragments;
 344         do {
 345                 struct sk_buff *xp = fp->next;
 346                 frag_kfree_skb(qp->q.net, fp);
 347                 fp = xp;
 348         } while (fp);
 349
 350         qp->q.last_in = 0;
 351         qp->q.len = 0;
 352         qp->q.meat = 0;
 353         qp->q.fragments = NULL;
 354         qp->q.fragments_tail = NULL;
 355         qp->iif = 0;
 356         qp->ecn = 0;
 357
 358         return 0;
 359 }
 360
 361 /* Add new segment to existing queue. */
 362 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
 363 {
 364         struct sk_buff *prev, *next;
 365         struct net_device *dev;
 366         int flags, offset;
 367         int ihl, end;
 368         int err = -ENOENT;
 369         u8 ecn;
 370
 371         if (qp->q.last_in & INET_FRAG_COMPLETE)
 372                 goto err;
 373
 374         if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
 375             unlikely(ip_frag_too_far(qp)) &&
 376             unlikely(err = ip_frag_reinit(qp))) {
 377                 ipq_kill(qp);
 378                 goto err;
 379         }
 380
 381         ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
 382         offset = ntohs(ip_hdr(skb)->frag_off);
 383         flags = offset & ~IP_OFFSET;
 384         offset &= IP_OFFSET;
 385         offset <<= 3;           /* offset is in 8-byte chunks */
 386         ihl = ip_hdrlen(skb);
 387
 388         /* Determine the position of this fragment. */
 389         end = offset + skb->len - ihl;
 390         err = -EINVAL;
 391
 392         /* Is this the final fragment? */
 393         if ((flags & IP_MF) == 0) {
 394                 /* If we already have some bits beyond end
 395                  * or have different end, the segment is corrrupted.
 396                  */
 397                 if (end < qp->q.len ||
 398                     ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
 399                         goto err;
 400                 qp->q.last_in |= INET_FRAG_LAST_IN;
 401                 qp->q.len = end;
 402         } else {
 403                 if (end&7) {
 404                         end &= ~7;
 405                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 406                                 skb->ip_summed = CHECKSUM_NONE;
 407                 }
 408                 if (end > qp->q.len) {
 409                         /* Some bits beyond end -> corruption. */
 410                         if (qp->q.last_in & INET_FRAG_LAST_IN)
 411                                 goto err;
 412                         qp->q.len = end;
 413                 }
 414         }
 415         if (end == offset)
 416                 goto err;
 417
 418         err = -ENOMEM;
 419         if (pskb_pull(skb, ihl) == NULL)
 420                 goto err;
 421
 422         err = pskb_trim_rcsum(skb, end - offset);
 423         if (err)
 424                 goto err;
 425
 426         /* Find out which fragments are in front and at the back of us
 427          * in the chain of fragments so far.  We must know where to put
 428          * this fragment, right?
 429          */
 430         prev = qp->q.fragments_tail;
 431         if (!prev || FRAG_CB(prev)->offset < offset) {
 432                 next = NULL;
 433                 goto found;
 434         }
 435         prev = NULL;
 436         for (next = qp->q.fragments; next != NULL; next = next->next) {
 437                 if (FRAG_CB(next)->offset >= offset)
 438                         break;  /* bingo! */
 439                 prev = next;
 440         }
 441
 442 found:
 443         /* We found where to put this one.  Check for overlap with
 444          * preceding fragment, and, if needed, align things so that
 445          * any overlaps are eliminated.
 446          */
 447         if (prev) {
 448                 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
 449
 450                 if (i > 0) {
 451                         offset += i;
 452                         err = -EINVAL;
 453                         if (end <= offset)
 454                                 goto err;
 455                         err = -ENOMEM;
 456                         if (!pskb_pull(skb, i))
 457                                 goto err;
 458                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 459                                 skb->ip_summed = CHECKSUM_NONE;
 460                 }
 461         }
 462
 463         err = -ENOMEM;
 464
 465         while (next && FRAG_CB(next)->offset < end) {
 466                 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
 467
 468                 if (i < next->len) {
 469                         /* Eat head of the next overlapped fragment
 470                          * and leave the loop. The next ones cannot overlap.
 471                          */
 472                         if (!pskb_pull(next, i))
 473                                 goto err;
 474                         FRAG_CB(next)->offset += i;
 475                         qp->q.meat -= i;
 476                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 477                                 next->ip_summed = CHECKSUM_NONE;
 478                         break;
 479                 } else {
 480                         struct sk_buff *free_it = next;
 481
 482                         /* Old fragment is completely overridden with
 483                          * new one drop it.
 484                          */
 485                         next = next->next;
 486
 487                         if (prev)
 488                                 prev->next = next;
 489                         else
 490                                 qp->q.fragments = next;
 491
 492                         qp->q.meat -= free_it->len;
 493                         frag_kfree_skb(qp->q.net, free_it);
 494                 }
 495         }
 496
 497         FRAG_CB(skb)->offset = offset;
 498
 499         /* Insert this fragment in the chain of fragments. */
 500         skb->next = next;
 501         if (!next)
 502                 qp->q.fragments_tail = skb;
 503         if (prev)
 504                 prev->next = skb;
 505         else
 506                 qp->q.fragments = skb;
 507
 508         dev = skb->dev;
 509         if (dev) {
 510                 qp->iif = dev->ifindex;
 511                 skb->dev = NULL;
 512         }
 513         qp->q.stamp = skb->tstamp;
 514         qp->q.meat += skb->len;
 515         qp->ecn |= ecn;
 516         atomic_add(skb->truesize, &qp->q.net->mem);
 517         if (offset == 0)
 518                 qp->q.last_in |= INET_FRAG_FIRST_IN;
 519
 520         if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 521             qp->q.meat == qp->q.len)
 522                 return ip_frag_reasm(qp, prev, dev);
 523
 524         write_lock(&ip4_frags.lock);
 525         list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
 526         write_unlock(&ip4_frags.lock);
 527         return -EINPROGRESS;
 528
 529 err:
 530         kfree_skb(skb);
 531         return err;
 532 }
 533
 534
 535 /* Build a new IP datagram from all its fragments. */
 536
 537 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 538                          struct net_device *dev)
 539 {
 540         struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
 541         struct iphdr *iph;
 542         struct sk_buff *fp, *head = qp->q.fragments;
 543         int len;
 544         int ihlen;
 545         int err;
 546         u8 ecn;
 547
 548         ipq_kill(qp);
 549
 550         ecn = ip4_frag_ecn_table[qp->ecn];
 551         if (unlikely(ecn == 0xff)) {
 552                 err = -EINVAL;
 553                 goto out_fail;
 554         }
 555         /* Make the one we just received the head. */
 556         if (prev) {
 557                 head = prev->next;
 558                 fp = skb_clone(head, GFP_ATOMIC);
 559                 if (!fp)
 560                         goto out_nomem;
 561
 562                 fp->next = head->next;
 563                 if (!fp->next)
 564                         qp->q.fragments_tail = fp;
 565                 prev->next = fp;
 566
 567                 skb_morph(head, qp->q.fragments);
 568                 head->next = qp->q.fragments->next;
 569
 570                 kfree_skb(qp->q.fragments);
 571                 qp->q.fragments = head;
 572         }
 573
 574         WARN_ON(head == NULL);
 575         WARN_ON(FRAG_CB(head)->offset != 0);
 576
 577         /* Allocate a new buffer for the datagram. */
 578         ihlen = ip_hdrlen(head);
 579         len = ihlen + qp->q.len;
 580
 581         err = -E2BIG;
 582         if (len > 65535)
 583                 goto out_oversize;
 584
 585         /* Head of list must not be cloned. */
 586         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 587                 goto out_nomem;
 588
 589         /* If the first fragment is fragmented itself, we split
 590          * it to two chunks: the first with data and paged part
 591          * and the second, holding only fragments. */
 592         if (skb_has_frag_list(head)) {
 593                 struct sk_buff *clone;
 594                 int i, plen = 0;
 595
 596                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 597                         goto out_nomem;
 598                 clone->next = head->next;
 599                 head->next = clone;
 600                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 601                 skb_frag_list_init(head);
 602                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 603                         plen += skb_shinfo(head)->frags[i].size;
 604                 clone->len = clone->data_len = head->data_len - plen;
 605                 head->data_len -= clone->len;
 606                 head->len -= clone->len;
 607                 clone->csum = 0;
 608                 clone->ip_summed = head->ip_summed;
 609                 atomic_add(clone->truesize, &qp->q.net->mem);
 610         }
 611
 612         skb_shinfo(head)->frag_list = head->next;
 613         skb_push(head, head->data - skb_network_header(head));
 614
 615         for (fp=head->next; fp; fp = fp->next) {
 616                 head->data_len += fp->len;
 617                 head->len += fp->len;
 618                 if (head->ip_summed != fp->ip_summed)
 619                         head->ip_summed = CHECKSUM_NONE;
 620                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 621                         head->csum = csum_add(head->csum, fp->csum);
 622                 head->truesize += fp->truesize;
 623         }
 624         atomic_sub(head->truesize, &qp->q.net->mem);
 625
 626         head->next = NULL;
 627         head->dev = dev;
 628         head->tstamp = qp->q.stamp;
 629
 630         iph = ip_hdr(head);
 631         iph->frag_off = 0;
 632         iph->tot_len = htons(len);
 633         iph->tos |= ecn;
 634         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
 635         qp->q.fragments = NULL;
 636         qp->q.fragments_tail = NULL;
 637         return 0;
 638
 639 out_nomem:
 640         LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
 641                               "queue %p\n", qp);
 642         err = -ENOMEM;
 643         goto out_fail;
 644 out_oversize:
 645         if (net_ratelimit())
 646                 printk(KERN_INFO "Oversized IP packet from %pI4.\n",
 647                         &qp->saddr);
 648 out_fail:
 649         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 650         return err;
 651 }
 652
 653 /* Process an incoming IP datagram fragment. */
 654 int ip_defrag(struct sk_buff *skb, u32 user)
 655 {
 656         struct ipq *qp;
 657         struct net *net;
 658
 659         net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
 660         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
 661
 662         /* Start by cleaning up the memory. */
 663         if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
 664                 ip_evictor(net);
 665
 666         /* Lookup (or create) queue header */
 667         if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
 668                 int ret;
 669
 670                 spin_lock(&qp->q.lock);
 671
 672                 ret = ip_frag_queue(qp, skb);
 673
 674                 spin_unlock(&qp->q.lock);
 675                 ipq_put(qp);
 676                 return ret;
 677         }
 678
 679         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 680         kfree_skb(skb);
 681         return -ENOMEM;
 682 }
 683 EXPORT_SYMBOL(ip_defrag);
 684
 685 #ifdef CONFIG_SYSCTL
 686 static int zero;
 687
 688 static struct ctl_table ip4_frags_ns_ctl_table[] = {
 689         {
 690                 .procname       = "ipfrag_high_thresh",
 691                 .data           = &init_net.ipv4.frags.high_thresh,
 692                 .maxlen         = sizeof(int),
 693                 .mode           = 0644,
 694                 .proc_handler   = proc_dointvec
 695         },
 696         {
 697                 .procname       = "ipfrag_low_thresh",
 698                 .data           = &init_net.ipv4.frags.low_thresh,
 699                 .maxlen         = sizeof(int),
 700                 .mode           = 0644,
 701                 .proc_handler   = proc_dointvec
 702         },
 703         {
 704                 .procname       = "ipfrag_time",
 705                 .data           = &init_net.ipv4.frags.timeout,
 706                 .maxlen         = sizeof(int),
 707                 .mode           = 0644,
 708                 .proc_handler   = proc_dointvec_jiffies,
 709         },
 710         { }
 711 };
 712
 713 static struct ctl_table ip4_frags_ctl_table[] = {
 714         {
 715                 .procname       = "ipfrag_secret_interval",
 716                 .data           = &ip4_frags.secret_interval,
 717                 .maxlen         = sizeof(int),
 718                 .mode           = 0644,
 719                 .proc_handler   = proc_dointvec_jiffies,
 720         },
 721         {
 722                 .procname       = "ipfrag_max_dist",
 723                 .data           = &sysctl_ipfrag_max_dist,
 724                 .maxlen         = sizeof(int),
 725                 .mode           = 0644,
 726                 .proc_handler   = proc_dointvec_minmax,
 727                 .extra1         = &zero
 728         },
 729         { }
 730 };
 731
 732 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
 733 {
 734         struct ctl_table *table;
 735         struct ctl_table_header *hdr;
 736
 737         table = ip4_frags_ns_ctl_table;
 738         if (!net_eq(net, &init_net)) {
 739                 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
 740                 if (table == NULL)
 741                         goto err_alloc;
 742
 743                 table[0].data = &net->ipv4.frags.high_thresh;
 744                 table[1].data = &net->ipv4.frags.low_thresh;
 745                 table[2].data = &net->ipv4.frags.timeout;
 746         }
 747
 748         hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
 749         if (hdr == NULL)
 750                 goto err_reg;
 751
 752         net->ipv4.frags_hdr = hdr;
 753         return 0;
 754
 755 err_reg:
 756         if (!net_eq(net, &init_net))
 757                 kfree(table);
 758 err_alloc:
 759         return -ENOMEM;
 760 }
 761
 762 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
 763 {
 764         struct ctl_table *table;
 765
 766         table = net->ipv4.frags_hdr->ctl_table_arg;
 767         unregister_net_sysctl_table(net->ipv4.frags_hdr);
 768         kfree(table);
 769 }
 770
 771 static void ip4_frags_ctl_register(void)
 772 {
 773         register_net_sysctl_rotable(net_ipv4_ctl_path, ip4_frags_ctl_table);
 774 }
 775 #else
 776 static inline int ip4_frags_ns_ctl_register(struct net *net)
 777 {
 778         return 0;
 779 }
 780
 781 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
 782 {
 783 }
 784
 785 static inline void ip4_frags_ctl_register(void)
 786 {
 787 }
 788 #endif
 789
 790 static int __net_init ipv4_frags_init_net(struct net *net)
 791 {
 792         /*
 793          * Fragment cache limits. We will commit 256K at one time. Should we
 794          * cross that limit we will prune down to 192K. This should cope with
 795          * even the most extreme cases without allowing an attacker to
 796          * measurably harm machine performance.
 797          */
 798         net->ipv4.frags.high_thresh = 256 * 1024;
 799         net->ipv4.frags.low_thresh = 192 * 1024;
 800         /*
 801          * Important NOTE! Fragment queue must be destroyed before MSL expires.
 802          * RFC791 is wrong proposing to prolongate timer each fragment arrival
 803          * by TTL.
 804          */
 805         net->ipv4.frags.timeout = IP_FRAG_TIME;
 806
 807         inet_frags_init_net(&net->ipv4.frags);
 808
 809         return ip4_frags_ns_ctl_register(net);
 810 }
 811
 812 static void __net_exit ipv4_frags_exit_net(struct net *net)
 813 {
 814         ip4_frags_ns_ctl_unregister(net);
 815         inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
 816 }
 817
 818 static struct pernet_operations ip4_frags_ops = {
 819         .init = ipv4_frags_init_net,
 820         .exit = ipv4_frags_exit_net,
 821 };
 822
 823 void __init ipfrag_init(void)
 824 {
 825         ip4_frags_ctl_register();
 826         register_pernet_subsys(&ip4_frags_ops);
 827         ip4_frags.hashfn = ip4_hashfn;
 828         ip4_frags.constructor = ip4_frag_init;
 829         ip4_frags.destructor = ip4_frag_free;
 830         ip4_frags.skb_free = NULL;
 831         ip4_frags.qsize = sizeof(struct ipq);
 832         ip4_frags.match = ip4_frag_match;
 833         ip4_frags.frag_expire = ip_expire;
 834         ip4_frags.secret_interval = 10 * 60 * HZ;
 835         inet_frags_init(&ip4_frags);
 836 }