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 #define pr_fmt(fmt) "IPv4: " fmt
  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 <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>
  51
  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
  55  */
  56
  57 static int sysctl_ipfrag_max_dist __read_mostly = 64;
  58
  59 struct ipfrag_skb_cb
  60 {
  61         struct inet_skb_parm    h;
  62         int                     offset;
  63 };
  64
  65 #define FRAG_CB(skb)    ((struct ipfrag_skb_cb *)((skb)->cb))
  66
  67 /* Describe an entry in the "incomplete datagrams" queue. */
  68 struct ipq {
  69         struct inet_frag_queue q;
  70
  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;
  80 };
  81
  82 /* RFC 3168 support :
  83  * We want to check ECN values of all fragments, do detect invalid combinations.
  84  * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
  85  */
  86 #define IPFRAG_ECN_NOT_ECT      0x01 /* one frag had ECN_NOT_ECT */
  87 #define IPFRAG_ECN_ECT_1        0x02 /* one frag had ECN_ECT_1 */
  88 #define IPFRAG_ECN_ECT_0        0x04 /* one frag had ECN_ECT_0 */
  89 #define IPFRAG_ECN_CE           0x08 /* one frag had ECN_CE */
  90
  91 static inline u8 ip4_frag_ecn(u8 tos)
  92 {
  93         return 1 << (tos & INET_ECN_MASK);
  94 }
  95
  96 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
  97  * Value : 0xff if frame should be dropped.
  98  *         0 or INET_ECN_CE value, to be ORed in to final iph->tos field
  99  */
 100 static const u8 ip4_frag_ecn_table[16] = {
 101         /* at least one fragment had CE, and others ECT_0 or ECT_1 */
 102         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0]                      = INET_ECN_CE,
 103         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1]                      = INET_ECN_CE,
 104         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1]   = INET_ECN_CE,
 105
 106         /* invalid combinations : drop frame */
 107         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
 108         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
 109         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
 110         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
 111         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
 112         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
 113         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
 114 };
 115
 116 static struct inet_frags ip4_frags;
 117
 118 int ip_frag_nqueues(struct net *net)
 119 {
 120         return net->ipv4.frags.nqueues;
 121 }
 122
 123 int ip_frag_mem(struct net *net)
 124 {
 125         return sum_frag_mem_limit(&net->ipv4.frags);
 126 }
 127
 128 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 129                          struct net_device *dev);
 130
 131 struct ip4_create_arg {
 132         struct iphdr *iph;
 133         u32 user;
 134 };
 135
 136 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
 137 {
 138         return jhash_3words((__force u32)id << 16 | prot,
 139                             (__force u32)saddr, (__force u32)daddr,
 140                             ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
 141 }
 142
 143 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
 144 {
 145         struct ipq *ipq;
 146
 147         ipq = container_of(q, struct ipq, q);
 148         return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
 149 }
 150
 151 static bool ip4_frag_match(struct inet_frag_queue *q, void *a)
 152 {
 153         struct ipq *qp;
 154         struct ip4_create_arg *arg = a;
 155
 156         qp = container_of(q, struct ipq, q);
 157         return  qp->id == arg->iph->id &&
 158                 qp->saddr == arg->iph->saddr &&
 159                 qp->daddr == arg->iph->daddr &&
 160                 qp->protocol == arg->iph->protocol &&
 161                 qp->user == arg->user;
 162 }
 163
 164 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
 165 {
 166         struct ipq *qp = container_of(q, struct ipq, q);
 167         struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
 168                                                frags);
 169         struct net *net = container_of(ipv4, struct net, ipv4);
 170
 171         struct ip4_create_arg *arg = a;
 172
 173         qp->protocol = arg->iph->protocol;
 174         qp->id = arg->iph->id;
 175         qp->ecn = ip4_frag_ecn(arg->iph->tos);
 176         qp->saddr = arg->iph->saddr;
 177         qp->daddr = arg->iph->daddr;
 178         qp->user = arg->user;
 179         qp->peer = sysctl_ipfrag_max_dist ?
 180                 inet_getpeer_v4(net->ipv4.peers, arg->iph->saddr, 1) : NULL;
 181 }
 182
 183 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
 184 {
 185         struct ipq *qp;
 186
 187         qp = container_of(q, struct ipq, q);
 188         if (qp->peer)
 189                 inet_putpeer(qp->peer);
 190 }
 191
 192
 193 /* Destruction primitives. */
 194
 195 static __inline__ void ipq_put(struct ipq *ipq)
 196 {
 197         inet_frag_put(&ipq->q, &ip4_frags);
 198 }
 199
 200 /* Kill ipq entry. It is not destroyed immediately,
 201  * because caller (and someone more) holds reference count.
 202  */
 203 static void ipq_kill(struct ipq *ipq)
 204 {
 205         inet_frag_kill(&ipq->q, &ip4_frags);
 206 }
 207
 208 /* Memory limiting on fragments.  Evictor trashes the oldest
 209  * fragment queue until we are back under the threshold.
 210  */
 211 static void ip_evictor(struct net *net)
 212 {
 213         int evicted;
 214
 215         evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags, false);
 216         if (evicted)
 217                 IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
 218 }
 219
 220 /*
 221  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
 222  */
 223 static void ip_expire(unsigned long arg)
 224 {
 225         struct ipq *qp;
 226         struct net *net;
 227
 228         qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
 229         net = container_of(qp->q.net, struct net, ipv4.frags);
 230
 231         spin_lock(&qp->q.lock);
 232
 233         if (qp->q.last_in & INET_FRAG_COMPLETE)
 234                 goto out;
 235
 236         ipq_kill(qp);
 237
 238         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
 239         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 240
 241         if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
 242                 struct sk_buff *head = qp->q.fragments;
 243                 const struct iphdr *iph;
 244                 int err;
 245
 246                 rcu_read_lock();
 247                 head->dev = dev_get_by_index_rcu(net, qp->iif);
 248                 if (!head->dev)
 249                         goto out_rcu_unlock;
 250
 251                 /* skb dst is stale, drop it, and perform route lookup again */
 252                 skb_dst_drop(head);
 253                 iph = ip_hdr(head);
 254                 err = ip_route_input_noref(head, iph->daddr, iph->saddr,
 255                                            iph->tos, head->dev);
 256                 if (err)
 257                         goto out_rcu_unlock;
 258
 259                 /*
 260                  * Only an end host needs to send an ICMP
 261                  * "Fragment Reassembly Timeout" message, per RFC792.
 262                  */
 263                 if (qp->user == IP_DEFRAG_AF_PACKET ||
 264                     (qp->user == IP_DEFRAG_CONNTRACK_IN &&
 265                      skb_rtable(head)->rt_type != RTN_LOCAL))
 266                         goto out_rcu_unlock;
 267
 268
 269                 /* Send an ICMP "Fragment Reassembly Timeout" message. */
 270                 icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
 271 out_rcu_unlock:
 272                 rcu_read_unlock();
 273         }
 274 out:
 275         spin_unlock(&qp->q.lock);
 276         ipq_put(qp);
 277 }
 278
 279 /* Find the correct entry in the "incomplete datagrams" queue for
 280  * this IP datagram, and create new one, if nothing is found.
 281  */
 282 static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
 283 {
 284         struct inet_frag_queue *q;
 285         struct ip4_create_arg arg;
 286         unsigned int hash;
 287
 288         arg.iph = iph;
 289         arg.user = user;
 290
 291         read_lock(&ip4_frags.lock);
 292         hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
 293
 294         q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
 295         if (q == NULL)
 296                 goto out_nomem;
 297
 298         return container_of(q, struct ipq, q);
 299
 300 out_nomem:
 301         LIMIT_NETDEBUG(KERN_ERR pr_fmt("ip_frag_create: no memory left !\n"));
 302         return NULL;
 303 }
 304
 305 /* Is the fragment too far ahead to be part of ipq? */
 306 static inline int ip_frag_too_far(struct ipq *qp)
 307 {
 308         struct inet_peer *peer = qp->peer;
 309         unsigned int max = sysctl_ipfrag_max_dist;
 310         unsigned int start, end;
 311
 312         int rc;
 313
 314         if (!peer || !max)
 315                 return 0;
 316
 317         start = qp->rid;
 318         end = atomic_inc_return(&peer->rid);
 319         qp->rid = end;
 320
 321         rc = qp->q.fragments && (end - start) > max;
 322
 323         if (rc) {
 324                 struct net *net;
 325
 326                 net = container_of(qp->q.net, struct net, ipv4.frags);
 327                 IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 328         }
 329
 330         return rc;
 331 }
 332
 333 static int ip_frag_reinit(struct ipq *qp)
 334 {
 335         struct sk_buff *fp;
 336         unsigned int sum_truesize = 0;
 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
 347                 sum_truesize += fp->truesize;
 348                 kfree_skb(fp);
 349                 fp = xp;
 350         } while (fp);
 351         sub_frag_mem_limit(&qp->q, sum_truesize);
 352
 353         qp->q.last_in = 0;
 354         qp->q.len = 0;
 355         qp->q.meat = 0;
 356         qp->q.fragments = NULL;
 357         qp->q.fragments_tail = NULL;
 358         qp->iif = 0;
 359         qp->ecn = 0;
 360
 361         return 0;
 362 }
 363
 364 /* Add new segment to existing queue. */
 365 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
 366 {
 367         struct sk_buff *prev, *next;
 368         struct net_device *dev;
 369         int flags, offset;
 370         int ihl, end;
 371         int err = -ENOENT;
 372         u8 ecn;
 373
 374         if (qp->q.last_in & INET_FRAG_COMPLETE)
 375                 goto err;
 376
 377         if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
 378             unlikely(ip_frag_too_far(qp)) &&
 379             unlikely(err = ip_frag_reinit(qp))) {
 380                 ipq_kill(qp);
 381                 goto err;
 382         }
 383
 384         ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
 385         offset = ntohs(ip_hdr(skb)->frag_off);
 386         flags = offset & ~IP_OFFSET;
 387         offset &= IP_OFFSET;
 388         offset <<= 3;           /* offset is in 8-byte chunks */
 389         ihl = ip_hdrlen(skb);
 390
 391         /* Determine the position of this fragment. */
 392         end = offset + skb->len - ihl;
 393         err = -EINVAL;
 394
 395         /* Is this the final fragment? */
 396         if ((flags & IP_MF) == 0) {
 397                 /* If we already have some bits beyond end
 398                  * or have different end, the segment is corrupted.
 399                  */
 400                 if (end < qp->q.len ||
 401                     ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
 402                         goto err;
 403                 qp->q.last_in |= INET_FRAG_LAST_IN;
 404                 qp->q.len = end;
 405         } else {
 406                 if (end&7) {
 407                         end &= ~7;
 408                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 409                                 skb->ip_summed = CHECKSUM_NONE;
 410                 }
 411                 if (end > qp->q.len) {
 412                         /* Some bits beyond end -> corruption. */
 413                         if (qp->q.last_in & INET_FRAG_LAST_IN)
 414                                 goto err;
 415                         qp->q.len = end;
 416                 }
 417         }
 418         if (end == offset)
 419                 goto err;
 420
 421         err = -ENOMEM;
 422         if (pskb_pull(skb, ihl) == NULL)
 423                 goto err;
 424
 425         err = pskb_trim_rcsum(skb, end - offset);
 426         if (err)
 427                 goto err;
 428
 429         /* Find out which fragments are in front and at the back of us
 430          * in the chain of fragments so far.  We must know where to put
 431          * this fragment, right?
 432          */
 433         prev = qp->q.fragments_tail;
 434         if (!prev || FRAG_CB(prev)->offset < offset) {
 435                 next = NULL;
 436                 goto found;
 437         }
 438         prev = NULL;
 439         for (next = qp->q.fragments; next != NULL; next = next->next) {
 440                 if (FRAG_CB(next)->offset >= offset)
 441                         break;  /* bingo! */
 442                 prev = next;
 443         }
 444
 445 found:
 446         /* We found where to put this one.  Check for overlap with
 447          * preceding fragment, and, if needed, align things so that
 448          * any overlaps are eliminated.
 449          */
 450         if (prev) {
 451                 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
 452
 453                 if (i > 0) {
 454                         offset += i;
 455                         err = -EINVAL;
 456                         if (end <= offset)
 457                                 goto err;
 458                         err = -ENOMEM;
 459                         if (!pskb_pull(skb, i))
 460                                 goto err;
 461                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 462                                 skb->ip_summed = CHECKSUM_NONE;
 463                 }
 464         }
 465
 466         err = -ENOMEM;
 467
 468         while (next && FRAG_CB(next)->offset < end) {
 469                 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
 470
 471                 if (i < next->len) {
 472                         /* Eat head of the next overlapped fragment
 473                          * and leave the loop. The next ones cannot overlap.
 474                          */
 475                         if (!pskb_pull(next, i))
 476                                 goto err;
 477                         FRAG_CB(next)->offset += i;
 478                         qp->q.meat -= i;
 479                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 480                                 next->ip_summed = CHECKSUM_NONE;
 481                         break;
 482                 } else {
 483                         struct sk_buff *free_it = next;
 484
 485                         /* Old fragment is completely overridden with
 486                          * new one drop it.
 487                          */
 488                         next = next->next;
 489
 490                         if (prev)
 491                                 prev->next = next;
 492                         else
 493                                 qp->q.fragments = next;
 494
 495                         qp->q.meat -= free_it->len;
 496                         sub_frag_mem_limit(&qp->q, free_it->truesize);
 497                         kfree_skb(free_it);
 498                 }
 499         }
 500
 501         FRAG_CB(skb)->offset = offset;
 502
 503         /* Insert this fragment in the chain of fragments. */
 504         skb->next = next;
 505         if (!next)
 506                 qp->q.fragments_tail = skb;
 507         if (prev)
 508                 prev->next = skb;
 509         else
 510                 qp->q.fragments = skb;
 511
 512         dev = skb->dev;
 513         if (dev) {
 514                 qp->iif = dev->ifindex;
 515                 skb->dev = NULL;
 516         }
 517         qp->q.stamp = skb->tstamp;
 518         qp->q.meat += skb->len;
 519         qp->ecn |= ecn;
 520         add_frag_mem_limit(&qp->q, skb->truesize);
 521         if (offset == 0)
 522                 qp->q.last_in |= INET_FRAG_FIRST_IN;
 523
 524         if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
 525             skb->len + ihl > qp->q.max_size)
 526                 qp->q.max_size = skb->len + ihl;
 527
 528         if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 529             qp->q.meat == qp->q.len)
 530                 return ip_frag_reasm(qp, prev, dev);
 531
 532         write_lock(&ip4_frags.lock);
 533         list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
 534         write_unlock(&ip4_frags.lock);
 535         return -EINPROGRESS;
 536
 537 err:
 538         kfree_skb(skb);
 539         return err;
 540 }
 541
 542
 543 /* Build a new IP datagram from all its fragments. */
 544
 545 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 546                          struct net_device *dev)
 547 {
 548         struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
 549         struct iphdr *iph;
 550         struct sk_buff *fp, *head = qp->q.fragments;
 551         int len;
 552         int ihlen;
 553         int err;
 554         int sum_truesize;
 555         u8 ecn;
 556
 557         ipq_kill(qp);
 558
 559         ecn = ip4_frag_ecn_table[qp->ecn];
 560         if (unlikely(ecn == 0xff)) {
 561                 err = -EINVAL;
 562                 goto out_fail;
 563         }
 564         /* Make the one we just received the head. */
 565         if (prev) {
 566                 head = prev->next;
 567                 fp = skb_clone(head, GFP_ATOMIC);
 568                 if (!fp)
 569                         goto out_nomem;
 570
 571                 fp->next = head->next;
 572                 if (!fp->next)
 573                         qp->q.fragments_tail = fp;
 574                 prev->next = fp;
 575
 576                 skb_morph(head, qp->q.fragments);
 577                 head->next = qp->q.fragments->next;
 578
 579                 consume_skb(qp->q.fragments);
 580                 qp->q.fragments = head;
 581         }
 582
 583         WARN_ON(head == NULL);
 584         WARN_ON(FRAG_CB(head)->offset != 0);
 585
 586         /* Allocate a new buffer for the datagram. */
 587         ihlen = ip_hdrlen(head);
 588         len = ihlen + qp->q.len;
 589
 590         err = -E2BIG;
 591         if (len > 65535)
 592                 goto out_oversize;
 593
 594         /* Head of list must not be cloned. */
 595         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 596                 goto out_nomem;
 597
 598         /* If the first fragment is fragmented itself, we split
 599          * it to two chunks: the first with data and paged part
 600          * and the second, holding only fragments. */
 601         if (skb_has_frag_list(head)) {
 602                 struct sk_buff *clone;
 603                 int i, plen = 0;
 604
 605                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 606                         goto out_nomem;
 607                 clone->next = head->next;
 608                 head->next = clone;
 609                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 610                 skb_frag_list_init(head);
 611                 for (i = 0; i < skb_shinfo(head)->nr_frags; i++)
 612                         plen += skb_frag_size(&skb_shinfo(head)->frags[i]);
 613                 clone->len = clone->data_len = head->data_len - plen;
 614                 head->data_len -= clone->len;
 615                 head->len -= clone->len;
 616                 clone->csum = 0;
 617                 clone->ip_summed = head->ip_summed;
 618                 add_frag_mem_limit(&qp->q, clone->truesize);
 619         }
 620
 621         skb_push(head, head->data - skb_network_header(head));
 622
 623         sum_truesize = head->truesize;
 624         for (fp = head->next; fp;) {
 625                 bool headstolen;
 626                 int delta;
 627                 struct sk_buff *next = fp->next;
 628
 629                 sum_truesize += fp->truesize;
 630                 if (head->ip_summed != fp->ip_summed)
 631                         head->ip_summed = CHECKSUM_NONE;
 632                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 633                         head->csum = csum_add(head->csum, fp->csum);
 634
 635                 if (skb_try_coalesce(head, fp, &headstolen, &delta)) {
 636                         kfree_skb_partial(fp, headstolen);
 637                 } else {
 638                         if (!skb_shinfo(head)->frag_list)
 639                                 skb_shinfo(head)->frag_list = fp;
 640                         head->data_len += fp->len;
 641                         head->len += fp->len;
 642                         head->truesize += fp->truesize;
 643                 }
 644                 fp = next;
 645         }
 646         sub_frag_mem_limit(&qp->q, sum_truesize);
 647
 648         head->next = NULL;
 649         head->dev = dev;
 650         head->tstamp = qp->q.stamp;
 651         IPCB(head)->frag_max_size = qp->q.max_size;
 652
 653         iph = ip_hdr(head);
 654         /* max_size != 0 implies at least one fragment had IP_DF set */
 655         iph->frag_off = qp->q.max_size ? htons(IP_DF) : 0;
 656         iph->tot_len = htons(len);
 657         iph->tos |= ecn;
 658         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
 659         qp->q.fragments = NULL;
 660         qp->q.fragments_tail = NULL;
 661         return 0;
 662
 663 out_nomem:
 664         LIMIT_NETDEBUG(KERN_ERR pr_fmt("queue_glue: no memory for gluing queue %p\n"),
 665                        qp);
 666         err = -ENOMEM;
 667         goto out_fail;
 668 out_oversize:
 669         net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->saddr);
 670 out_fail:
 671         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 672         return err;
 673 }
 674
 675 /* Process an incoming IP datagram fragment. */
 676 int ip_defrag(struct sk_buff *skb, u32 user)
 677 {
 678         struct ipq *qp;
 679         struct net *net;
 680
 681         net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
 682         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
 683
 684         /* Start by cleaning up the memory. */
 685         ip_evictor(net);
 686
 687         /* Lookup (or create) queue header */
 688         if ((qp = ip_find(net, ip_hdr(skb), user)) != NULL) {
 689                 int ret;
 690
 691                 spin_lock(&qp->q.lock);
 692
 693                 ret = ip_frag_queue(qp, skb);
 694
 695                 spin_unlock(&qp->q.lock);
 696                 ipq_put(qp);
 697                 return ret;
 698         }
 699
 700         IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 701         kfree_skb(skb);
 702         return -ENOMEM;
 703 }
 704 EXPORT_SYMBOL(ip_defrag);
 705
 706 struct sk_buff *ip_check_defrag(struct sk_buff *skb, u32 user)
 707 {
 708         struct iphdr iph;
 709         u32 len;
 710
 711         if (skb->protocol != htons(ETH_P_IP))
 712                 return skb;
 713
 714         if (!skb_copy_bits(skb, 0, &iph, sizeof(iph)))
 715                 return skb;
 716
 717         if (iph.ihl < 5 || iph.version != 4)
 718                 return skb;
 719
 720         len = ntohs(iph.tot_len);
 721         if (skb->len < len || len < (iph.ihl * 4))
 722                 return skb;
 723
 724         if (ip_is_fragment(&iph)) {
 725                 skb = skb_share_check(skb, GFP_ATOMIC);
 726                 if (skb) {
 727                         if (!pskb_may_pull(skb, iph.ihl*4))
 728                                 return skb;
 729                         if (pskb_trim_rcsum(skb, len))
 730                                 return skb;
 731                         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 732                         if (ip_defrag(skb, user))
 733                                 return NULL;
 734                         skb->rxhash = 0;
 735                 }
 736         }
 737         return skb;
 738 }
 739 EXPORT_SYMBOL(ip_check_defrag);
 740
 741 #ifdef CONFIG_SYSCTL
 742 static int zero;
 743
 744 static struct ctl_table ip4_frags_ns_ctl_table[] = {
 745         {
 746                 .procname       = "ipfrag_high_thresh",
 747                 .data           = &init_net.ipv4.frags.high_thresh,
 748                 .maxlen         = sizeof(int),
 749                 .mode           = 0644,
 750                 .proc_handler   = proc_dointvec
 751         },
 752         {
 753                 .procname       = "ipfrag_low_thresh",
 754                 .data           = &init_net.ipv4.frags.low_thresh,
 755                 .maxlen         = sizeof(int),
 756                 .mode           = 0644,
 757                 .proc_handler   = proc_dointvec
 758         },
 759         {
 760                 .procname       = "ipfrag_time",
 761                 .data           = &init_net.ipv4.frags.timeout,
 762                 .maxlen         = sizeof(int),
 763                 .mode           = 0644,
 764                 .proc_handler   = proc_dointvec_jiffies,
 765         },
 766         { }
 767 };
 768
 769 static struct ctl_table ip4_frags_ctl_table[] = {
 770         {
 771                 .procname       = "ipfrag_secret_interval",
 772                 .data           = &ip4_frags.secret_interval,
 773                 .maxlen         = sizeof(int),
 774                 .mode           = 0644,
 775                 .proc_handler   = proc_dointvec_jiffies,
 776         },
 777         {
 778                 .procname       = "ipfrag_max_dist",
 779                 .data           = &sysctl_ipfrag_max_dist,
 780                 .maxlen         = sizeof(int),
 781                 .mode           = 0644,
 782                 .proc_handler   = proc_dointvec_minmax,
 783                 .extra1         = &zero
 784         },
 785         { }
 786 };
 787
 788 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
 789 {
 790         struct ctl_table *table;
 791         struct ctl_table_header *hdr;
 792
 793         table = ip4_frags_ns_ctl_table;
 794         if (!net_eq(net, &init_net)) {
 795                 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
 796                 if (table == NULL)
 797                         goto err_alloc;
 798
 799                 table[0].data = &net->ipv4.frags.high_thresh;
 800                 table[1].data = &net->ipv4.frags.low_thresh;
 801                 table[2].data = &net->ipv4.frags.timeout;
 802
 803                 /* Don't export sysctls to unprivileged users */
 804                 if (net->user_ns != &init_user_ns)
 805                         table[0].procname = NULL;
 806         }
 807
 808         hdr = register_net_sysctl(net, "net/ipv4", table);
 809         if (hdr == NULL)
 810                 goto err_reg;
 811
 812         net->ipv4.frags_hdr = hdr;
 813         return 0;
 814
 815 err_reg:
 816         if (!net_eq(net, &init_net))
 817                 kfree(table);
 818 err_alloc:
 819         return -ENOMEM;
 820 }
 821
 822 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
 823 {
 824         struct ctl_table *table;
 825
 826         table = net->ipv4.frags_hdr->ctl_table_arg;
 827         unregister_net_sysctl_table(net->ipv4.frags_hdr);
 828         kfree(table);
 829 }
 830
 831 static void ip4_frags_ctl_register(void)
 832 {
 833         register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
 834 }
 835 #else
 836 static inline int ip4_frags_ns_ctl_register(struct net *net)
 837 {
 838         return 0;
 839 }
 840
 841 static inline void ip4_frags_ns_ctl_unregister(struct net *net)
 842 {
 843 }
 844
 845 static inline void ip4_frags_ctl_register(void)
 846 {
 847 }
 848 #endif
 849
 850 static int __net_init ipv4_frags_init_net(struct net *net)
 851 {
 852         /* Fragment cache limits.
 853          *
 854          * The fragment memory accounting code, (tries to) account for
 855          * the real memory usage, by measuring both the size of frag
 856          * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
 857          * and the SKB's truesize.
 858          *
 859          * A 64K fragment consumes 129736 bytes (44*2944)+200
 860          * (1500 truesize == 2944, sizeof(struct ipq) == 200)
 861          *
 862          * We will commit 4MB at one time. Should we cross that limit
 863          * we will prune down to 3MB, making room for approx 8 big 64K
 864          * fragments 8x128k.
 865          */
 866         net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
 867         net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
 868         /*
 869          * Important NOTE! Fragment queue must be destroyed before MSL expires.
 870          * RFC791 is wrong proposing to prolongate timer each fragment arrival
 871          * by TTL.
 872          */
 873         net->ipv4.frags.timeout = IP_FRAG_TIME;
 874
 875         inet_frags_init_net(&net->ipv4.frags);
 876
 877         return ip4_frags_ns_ctl_register(net);
 878 }
 879
 880 static void __net_exit ipv4_frags_exit_net(struct net *net)
 881 {
 882         ip4_frags_ns_ctl_unregister(net);
 883         inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
 884 }
 885
 886 static struct pernet_operations ip4_frags_ops = {
 887         .init = ipv4_frags_init_net,
 888         .exit = ipv4_frags_exit_net,
 889 };
 890
 891 void __init ipfrag_init(void)
 892 {
 893         ip4_frags_ctl_register();
 894         register_pernet_subsys(&ip4_frags_ops);
 895         ip4_frags.hashfn = ip4_hashfn;
 896         ip4_frags.constructor = ip4_frag_init;
 897         ip4_frags.destructor = ip4_frag_free;
 898         ip4_frags.skb_free = NULL;
 899         ip4_frags.qsize = sizeof(struct ipq);
 900         ip4_frags.match = ip4_frag_match;
 901         ip4_frags.frag_expire = ip_expire;
 902         ip4_frags.secret_interval = 10 * 60 * HZ;
 903         inet_frags_init(&ip4_frags);
 904 }