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 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 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
  87         /*
  88          * Important NOTE! Fragment queue must be destroyed before MSL expires.
  89          * RFC791 is wrong proposing to prolongate timer each fragment arrival
  90          * by TTL.
  91          */
  92         .timeout         = IP_FRAG_TIME,
  93         .secret_interval = 10 * 60 * HZ,
  94 };
  95
  96 static struct inet_frags ip4_frags;
  97
  98 int ip_frag_nqueues(void)
  99 {
 100         return ip4_frags.nqueues;
 101 }
 102
 103 int ip_frag_mem(void)
 104 {
 105         return atomic_read(&ip4_frags.mem);
 106 }
 107
 108 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 109                          struct net_device *dev);
 110
 111 struct ip4_create_arg {
 112         struct iphdr *iph;
 113         u32 user;
 114 };
 115
 116 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
 117 {
 118         return jhash_3words((__force u32)id << 16 | prot,
 119                             (__force u32)saddr, (__force u32)daddr,
 120                             ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
 121 }
 122
 123 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
 124 {
 125         struct ipq *ipq;
 126
 127         ipq = container_of(q, struct ipq, q);
 128         return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
 129 }
 130
 131 static int ip4_frag_match(struct inet_frag_queue *q, void *a)
 132 {
 133         struct ipq *qp;
 134         struct ip4_create_arg *arg = a;
 135
 136         qp = container_of(q, struct ipq, q);
 137         return (qp->id == arg->iph->id &&
 138                         qp->saddr == arg->iph->saddr &&
 139                         qp->daddr == arg->iph->daddr &&
 140                         qp->protocol == arg->iph->protocol &&
 141                         qp->user == arg->user);
 142 }
 143
 144 /* Memory Tracking Functions. */
 145 static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work)
 146 {
 147         if (work)
 148                 *work -= skb->truesize;
 149         atomic_sub(skb->truesize, &ip4_frags.mem);
 150         kfree_skb(skb);
 151 }
 152
 153 static void ip4_frag_init(struct inet_frag_queue *q, void *a)
 154 {
 155         struct ipq *qp = container_of(q, struct ipq, q);
 156         struct ip4_create_arg *arg = a;
 157
 158         qp->protocol = arg->iph->protocol;
 159         qp->id = arg->iph->id;
 160         qp->saddr = arg->iph->saddr;
 161         qp->daddr = arg->iph->daddr;
 162         qp->user = arg->user;
 163         qp->peer = sysctl_ipfrag_max_dist ?
 164                 inet_getpeer(arg->iph->saddr, 1) : NULL;
 165 }
 166
 167 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
 168 {
 169         struct ipq *qp;
 170
 171         qp = container_of(q, struct ipq, q);
 172         if (qp->peer)
 173                 inet_putpeer(qp->peer);
 174 }
 175
 176
 177 /* Destruction primitives. */
 178
 179 static __inline__ void ipq_put(struct ipq *ipq)
 180 {
 181         inet_frag_put(&ipq->q, &ip4_frags);
 182 }
 183
 184 /* Kill ipq entry. It is not destroyed immediately,
 185  * because caller (and someone more) holds reference count.
 186  */
 187 static void ipq_kill(struct ipq *ipq)
 188 {
 189         inet_frag_kill(&ipq->q, &ip4_frags);
 190 }
 191
 192 /* Memory limiting on fragments.  Evictor trashes the oldest
 193  * fragment queue until we are back under the threshold.
 194  */
 195 static void ip_evictor(void)
 196 {
 197         int evicted;
 198
 199         evicted = inet_frag_evictor(&ip4_frags);
 200         if (evicted)
 201                 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
 202 }
 203
 204 /*
 205  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
 206  */
 207 static void ip_expire(unsigned long arg)
 208 {
 209         struct ipq *qp;
 210
 211         qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
 212
 213         spin_lock(&qp->q.lock);
 214
 215         if (qp->q.last_in & COMPLETE)
 216                 goto out;
 217
 218         ipq_kill(qp);
 219
 220         IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
 221         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 222
 223         if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
 224                 struct sk_buff *head = qp->q.fragments;
 225                 /* Send an ICMP "Fragment Reassembly Timeout" message. */
 226                 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
 227                         icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
 228                         dev_put(head->dev);
 229                 }
 230         }
 231 out:
 232         spin_unlock(&qp->q.lock);
 233         ipq_put(qp);
 234 }
 235
 236 /* Find the correct entry in the "incomplete datagrams" queue for
 237  * this IP datagram, and create new one, if nothing is found.
 238  */
 239 static inline struct ipq *ip_find(struct iphdr *iph, u32 user)
 240 {
 241         struct inet_frag_queue *q;
 242         struct ip4_create_arg arg;
 243         unsigned int hash;
 244
 245         arg.iph = iph;
 246         arg.user = user;
 247         hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
 248
 249         q = inet_frag_find(&ip4_frags, &arg, hash);
 250         if (q == NULL)
 251                 goto out_nomem;
 252
 253         return container_of(q, struct ipq, q);
 254
 255 out_nomem:
 256         LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
 257         return NULL;
 258 }
 259
 260 /* Is the fragment too far ahead to be part of ipq? */
 261 static inline int ip_frag_too_far(struct ipq *qp)
 262 {
 263         struct inet_peer *peer = qp->peer;
 264         unsigned int max = sysctl_ipfrag_max_dist;
 265         unsigned int start, end;
 266
 267         int rc;
 268
 269         if (!peer || !max)
 270                 return 0;
 271
 272         start = qp->rid;
 273         end = atomic_inc_return(&peer->rid);
 274         qp->rid = end;
 275
 276         rc = qp->q.fragments && (end - start) > max;
 277
 278         if (rc) {
 279                 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 280         }
 281
 282         return rc;
 283 }
 284
 285 static int ip_frag_reinit(struct ipq *qp)
 286 {
 287         struct sk_buff *fp;
 288
 289         if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) {
 290                 atomic_inc(&qp->q.refcnt);
 291                 return -ETIMEDOUT;
 292         }
 293
 294         fp = qp->q.fragments;
 295         do {
 296                 struct sk_buff *xp = fp->next;
 297                 frag_kfree_skb(fp, NULL);
 298                 fp = xp;
 299         } while (fp);
 300
 301         qp->q.last_in = 0;
 302         qp->q.len = 0;
 303         qp->q.meat = 0;
 304         qp->q.fragments = NULL;
 305         qp->iif = 0;
 306
 307         return 0;
 308 }
 309
 310 /* Add new segment to existing queue. */
 311 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
 312 {
 313         struct sk_buff *prev, *next;
 314         struct net_device *dev;
 315         int flags, offset;
 316         int ihl, end;
 317         int err = -ENOENT;
 318
 319         if (qp->q.last_in & COMPLETE)
 320                 goto err;
 321
 322         if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
 323             unlikely(ip_frag_too_far(qp)) &&
 324             unlikely(err = ip_frag_reinit(qp))) {
 325                 ipq_kill(qp);
 326                 goto err;
 327         }
 328
 329         offset = ntohs(ip_hdr(skb)->frag_off);
 330         flags = offset & ~IP_OFFSET;
 331         offset &= IP_OFFSET;
 332         offset <<= 3;           /* offset is in 8-byte chunks */
 333         ihl = ip_hdrlen(skb);
 334
 335         /* Determine the position of this fragment. */
 336         end = offset + skb->len - ihl;
 337         err = -EINVAL;
 338
 339         /* Is this the final fragment? */
 340         if ((flags & IP_MF) == 0) {
 341                 /* If we already have some bits beyond end
 342                  * or have different end, the segment is corrrupted.
 343                  */
 344                 if (end < qp->q.len ||
 345                     ((qp->q.last_in & LAST_IN) && end != qp->q.len))
 346                         goto err;
 347                 qp->q.last_in |= LAST_IN;
 348                 qp->q.len = end;
 349         } else {
 350                 if (end&7) {
 351                         end &= ~7;
 352                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 353                                 skb->ip_summed = CHECKSUM_NONE;
 354                 }
 355                 if (end > qp->q.len) {
 356                         /* Some bits beyond end -> corruption. */
 357                         if (qp->q.last_in & LAST_IN)
 358                                 goto err;
 359                         qp->q.len = end;
 360                 }
 361         }
 362         if (end == offset)
 363                 goto err;
 364
 365         err = -ENOMEM;
 366         if (pskb_pull(skb, ihl) == NULL)
 367                 goto err;
 368
 369         err = pskb_trim_rcsum(skb, end - offset);
 370         if (err)
 371                 goto err;
 372
 373         /* Find out which fragments are in front and at the back of us
 374          * in the chain of fragments so far.  We must know where to put
 375          * this fragment, right?
 376          */
 377         prev = NULL;
 378         for (next = qp->q.fragments; next != NULL; next = next->next) {
 379                 if (FRAG_CB(next)->offset >= offset)
 380                         break;  /* bingo! */
 381                 prev = next;
 382         }
 383
 384         /* We found where to put this one.  Check for overlap with
 385          * preceding fragment, and, if needed, align things so that
 386          * any overlaps are eliminated.
 387          */
 388         if (prev) {
 389                 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
 390
 391                 if (i > 0) {
 392                         offset += i;
 393                         err = -EINVAL;
 394                         if (end <= offset)
 395                                 goto err;
 396                         err = -ENOMEM;
 397                         if (!pskb_pull(skb, i))
 398                                 goto err;
 399                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 400                                 skb->ip_summed = CHECKSUM_NONE;
 401                 }
 402         }
 403
 404         err = -ENOMEM;
 405
 406         while (next && FRAG_CB(next)->offset < end) {
 407                 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
 408
 409                 if (i < next->len) {
 410                         /* Eat head of the next overlapped fragment
 411                          * and leave the loop. The next ones cannot overlap.
 412                          */
 413                         if (!pskb_pull(next, i))
 414                                 goto err;
 415                         FRAG_CB(next)->offset += i;
 416                         qp->q.meat -= i;
 417                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 418                                 next->ip_summed = CHECKSUM_NONE;
 419                         break;
 420                 } else {
 421                         struct sk_buff *free_it = next;
 422
 423                         /* Old fragment is completely overridden with
 424                          * new one drop it.
 425                          */
 426                         next = next->next;
 427
 428                         if (prev)
 429                                 prev->next = next;
 430                         else
 431                                 qp->q.fragments = next;
 432
 433                         qp->q.meat -= free_it->len;
 434                         frag_kfree_skb(free_it, NULL);
 435                 }
 436         }
 437
 438         FRAG_CB(skb)->offset = offset;
 439
 440         /* Insert this fragment in the chain of fragments. */
 441         skb->next = next;
 442         if (prev)
 443                 prev->next = skb;
 444         else
 445                 qp->q.fragments = skb;
 446
 447         dev = skb->dev;
 448         if (dev) {
 449                 qp->iif = dev->ifindex;
 450                 skb->dev = NULL;
 451         }
 452         qp->q.stamp = skb->tstamp;
 453         qp->q.meat += skb->len;
 454         atomic_add(skb->truesize, &ip4_frags.mem);
 455         if (offset == 0)
 456                 qp->q.last_in |= FIRST_IN;
 457
 458         if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
 459                 return ip_frag_reasm(qp, prev, dev);
 460
 461         write_lock(&ip4_frags.lock);
 462         list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
 463         write_unlock(&ip4_frags.lock);
 464         return -EINPROGRESS;
 465
 466 err:
 467         kfree_skb(skb);
 468         return err;
 469 }
 470
 471
 472 /* Build a new IP datagram from all its fragments. */
 473
 474 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 475                          struct net_device *dev)
 476 {
 477         struct iphdr *iph;
 478         struct sk_buff *fp, *head = qp->q.fragments;
 479         int len;
 480         int ihlen;
 481         int err;
 482
 483         ipq_kill(qp);
 484
 485         /* Make the one we just received the head. */
 486         if (prev) {
 487                 head = prev->next;
 488                 fp = skb_clone(head, GFP_ATOMIC);
 489
 490                 if (!fp)
 491                         goto out_nomem;
 492
 493                 fp->next = head->next;
 494                 prev->next = fp;
 495
 496                 skb_morph(head, qp->q.fragments);
 497                 head->next = qp->q.fragments->next;
 498
 499                 kfree_skb(qp->q.fragments);
 500                 qp->q.fragments = head;
 501         }
 502
 503         BUG_TRAP(head != NULL);
 504         BUG_TRAP(FRAG_CB(head)->offset == 0);
 505
 506         /* Allocate a new buffer for the datagram. */
 507         ihlen = ip_hdrlen(head);
 508         len = ihlen + qp->q.len;
 509
 510         err = -E2BIG;
 511         if (len > 65535)
 512                 goto out_oversize;
 513
 514         /* Head of list must not be cloned. */
 515         err = -ENOMEM;
 516         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 517                 goto out_nomem;
 518
 519         /* If the first fragment is fragmented itself, we split
 520          * it to two chunks: the first with data and paged part
 521          * and the second, holding only fragments. */
 522         if (skb_shinfo(head)->frag_list) {
 523                 struct sk_buff *clone;
 524                 int i, plen = 0;
 525
 526                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 527                         goto out_nomem;
 528                 clone->next = head->next;
 529                 head->next = clone;
 530                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 531                 skb_shinfo(head)->frag_list = NULL;
 532                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 533                         plen += skb_shinfo(head)->frags[i].size;
 534                 clone->len = clone->data_len = head->data_len - plen;
 535                 head->data_len -= clone->len;
 536                 head->len -= clone->len;
 537                 clone->csum = 0;
 538                 clone->ip_summed = head->ip_summed;
 539                 atomic_add(clone->truesize, &ip4_frags.mem);
 540         }
 541
 542         skb_shinfo(head)->frag_list = head->next;
 543         skb_push(head, head->data - skb_network_header(head));
 544         atomic_sub(head->truesize, &ip4_frags.mem);
 545
 546         for (fp=head->next; fp; fp = fp->next) {
 547                 head->data_len += fp->len;
 548                 head->len += fp->len;
 549                 if (head->ip_summed != fp->ip_summed)
 550                         head->ip_summed = CHECKSUM_NONE;
 551                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 552                         head->csum = csum_add(head->csum, fp->csum);
 553                 head->truesize += fp->truesize;
 554                 atomic_sub(fp->truesize, &ip4_frags.mem);
 555         }
 556
 557         head->next = NULL;
 558         head->dev = dev;
 559         head->tstamp = qp->q.stamp;
 560
 561         iph = ip_hdr(head);
 562         iph->frag_off = 0;
 563         iph->tot_len = htons(len);
 564         IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
 565         qp->q.fragments = NULL;
 566         return 0;
 567
 568 out_nomem:
 569         LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
 570                               "queue %p\n", qp);
 571         goto out_fail;
 572 out_oversize:
 573         if (net_ratelimit())
 574                 printk(KERN_INFO
 575                         "Oversized IP packet from %d.%d.%d.%d.\n",
 576                         NIPQUAD(qp->saddr));
 577 out_fail:
 578         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 579         return err;
 580 }
 581
 582 /* Process an incoming IP datagram fragment. */
 583 int ip_defrag(struct sk_buff *skb, u32 user)
 584 {
 585         struct ipq *qp;
 586
 587         IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
 588
 589         /* Start by cleaning up the memory. */
 590         if (atomic_read(&ip4_frags.mem) > ip4_frags_ctl.high_thresh)
 591                 ip_evictor();
 592
 593         /* Lookup (or create) queue header */
 594         if ((qp = ip_find(ip_hdr(skb), user)) != NULL) {
 595                 int ret;
 596
 597                 spin_lock(&qp->q.lock);
 598
 599                 ret = ip_frag_queue(qp, skb);
 600
 601                 spin_unlock(&qp->q.lock);
 602                 ipq_put(qp);
 603                 return ret;
 604         }
 605
 606         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 607         kfree_skb(skb);
 608         return -ENOMEM;
 609 }
 610
 611 void __init ipfrag_init(void)
 612 {
 613         ip4_frags.ctl = &ip4_frags_ctl;
 614         ip4_frags.hashfn = ip4_hashfn;
 615         ip4_frags.constructor = ip4_frag_init;
 616         ip4_frags.destructor = ip4_frag_free;
 617         ip4_frags.skb_free = NULL;
 618         ip4_frags.qsize = sizeof(struct ipq);
 619         ip4_frags.match = ip4_frag_match;
 620         ip4_frags.frag_expire = ip_expire;
 621         inet_frags_init(&ip4_frags);
 622 }
 623
 624 EXPORT_SYMBOL(ip_defrag);