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 static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot)
 112 {
 113         return jhash_3words((__force u32)id << 16 | prot,
 114                             (__force u32)saddr, (__force u32)daddr,
 115                             ip4_frags.rnd) & (INETFRAGS_HASHSZ - 1);
 116 }
 117
 118 static unsigned int ip4_hashfn(struct inet_frag_queue *q)
 119 {
 120         struct ipq *ipq;
 121
 122         ipq = container_of(q, struct ipq, q);
 123         return ipqhashfn(ipq->id, ipq->saddr, ipq->daddr, ipq->protocol);
 124 }
 125
 126 static int ip4_frag_equal(struct inet_frag_queue *q1,
 127                 struct inet_frag_queue *q2)
 128 {
 129         struct ipq *qp1, *qp2;
 130
 131         qp1 = container_of(q1, struct ipq, q);
 132         qp2 = container_of(q2, struct ipq, q);
 133         return (qp1->id == qp2->id &&
 134                         qp1->saddr == qp2->saddr &&
 135                         qp1->daddr == qp2->daddr &&
 136                         qp1->protocol == qp2->protocol &&
 137                         qp1->user == qp2->user);
 138 }
 139
 140 /* Memory Tracking Functions. */
 141 static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work)
 142 {
 143         if (work)
 144                 *work -= skb->truesize;
 145         atomic_sub(skb->truesize, &ip4_frags.mem);
 146         kfree_skb(skb);
 147 }
 148
 149 static __inline__ void ip4_frag_free(struct inet_frag_queue *q)
 150 {
 151         struct ipq *qp;
 152
 153         qp = container_of(q, struct ipq, q);
 154         if (qp->peer)
 155                 inet_putpeer(qp->peer);
 156         kfree(qp);
 157 }
 158
 159 static __inline__ struct ipq *frag_alloc_queue(void)
 160 {
 161         struct ipq *qp = kzalloc(sizeof(struct ipq), GFP_ATOMIC);
 162
 163         if (!qp)
 164                 return NULL;
 165         atomic_add(sizeof(struct ipq), &ip4_frags.mem);
 166         return qp;
 167 }
 168
 169
 170 /* Destruction primitives. */
 171
 172 static __inline__ void ipq_put(struct ipq *ipq)
 173 {
 174         inet_frag_put(&ipq->q, &ip4_frags);
 175 }
 176
 177 /* Kill ipq entry. It is not destroyed immediately,
 178  * because caller (and someone more) holds reference count.
 179  */
 180 static void ipq_kill(struct ipq *ipq)
 181 {
 182         inet_frag_kill(&ipq->q, &ip4_frags);
 183 }
 184
 185 /* Memory limiting on fragments.  Evictor trashes the oldest
 186  * fragment queue until we are back under the threshold.
 187  */
 188 static void ip_evictor(void)
 189 {
 190         int evicted;
 191
 192         evicted = inet_frag_evictor(&ip4_frags);
 193         if (evicted)
 194                 IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
 195 }
 196
 197 /*
 198  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
 199  */
 200 static void ip_expire(unsigned long arg)
 201 {
 202         struct ipq *qp = (struct ipq *) arg;
 203
 204         spin_lock(&qp->q.lock);
 205
 206         if (qp->q.last_in & COMPLETE)
 207                 goto out;
 208
 209         ipq_kill(qp);
 210
 211         IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
 212         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 213
 214         if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
 215                 struct sk_buff *head = qp->q.fragments;
 216                 /* Send an ICMP "Fragment Reassembly Timeout" message. */
 217                 if ((head->dev = dev_get_by_index(&init_net, qp->iif)) != NULL) {
 218                         icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
 219                         dev_put(head->dev);
 220                 }
 221         }
 222 out:
 223         spin_unlock(&qp->q.lock);
 224         ipq_put(qp);
 225 }
 226
 227 /* Creation primitives. */
 228
 229 static struct ipq *ip_frag_intern(struct ipq *qp_in, unsigned int hash)
 230 {
 231         struct inet_frag_queue *q;
 232
 233         q = inet_frag_intern(&qp_in->q, &ip4_frags, hash);
 234         return container_of(q, struct ipq, q);
 235 }
 236
 237 /* Add an entry to the 'ipq' queue for a newly received IP datagram. */
 238 static struct ipq *ip_frag_create(struct iphdr *iph, u32 user, unsigned int h)
 239 {
 240         struct ipq *qp;
 241
 242         if ((qp = frag_alloc_queue()) == NULL)
 243                 goto out_nomem;
 244
 245         qp->protocol = iph->protocol;
 246         qp->id = iph->id;
 247         qp->saddr = iph->saddr;
 248         qp->daddr = iph->daddr;
 249         qp->user = user;
 250         qp->peer = sysctl_ipfrag_max_dist ? inet_getpeer(iph->saddr, 1) : NULL;
 251
 252         /* Initialize a timer for this entry. */
 253         init_timer(&qp->q.timer);
 254         qp->q.timer.data = (unsigned long) qp;  /* pointer to queue     */
 255         qp->q.timer.function = ip_expire;               /* expire function      */
 256         spin_lock_init(&qp->q.lock);
 257         atomic_set(&qp->q.refcnt, 1);
 258
 259         return ip_frag_intern(qp, h);
 260
 261 out_nomem:
 262         LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n");
 263         return NULL;
 264 }
 265
 266 /* Find the correct entry in the "incomplete datagrams" queue for
 267  * this IP datagram, and create new one, if nothing is found.
 268  */
 269 static inline struct ipq *ip_find(struct iphdr *iph, u32 user)
 270 {
 271         __be16 id = iph->id;
 272         __be32 saddr = iph->saddr;
 273         __be32 daddr = iph->daddr;
 274         __u8 protocol = iph->protocol;
 275         unsigned int hash;
 276         struct ipq *qp;
 277         struct hlist_node *n;
 278
 279         read_lock(&ip4_frags.lock);
 280         hash = ipqhashfn(id, saddr, daddr, protocol);
 281         hlist_for_each_entry(qp, n, &ip4_frags.hash[hash], q.list) {
 282                 if (qp->id == id                &&
 283                     qp->saddr == saddr  &&
 284                     qp->daddr == daddr  &&
 285                     qp->protocol == protocol &&
 286                     qp->user == user) {
 287                         atomic_inc(&qp->q.refcnt);
 288                         read_unlock(&ip4_frags.lock);
 289                         return qp;
 290                 }
 291         }
 292         read_unlock(&ip4_frags.lock);
 293
 294         return ip_frag_create(iph, user, hash);
 295 }
 296
 297 /* Is the fragment too far ahead to be part of ipq? */
 298 static inline int ip_frag_too_far(struct ipq *qp)
 299 {
 300         struct inet_peer *peer = qp->peer;
 301         unsigned int max = sysctl_ipfrag_max_dist;
 302         unsigned int start, end;
 303
 304         int rc;
 305
 306         if (!peer || !max)
 307                 return 0;
 308
 309         start = qp->rid;
 310         end = atomic_inc_return(&peer->rid);
 311         qp->rid = end;
 312
 313         rc = qp->q.fragments && (end - start) > max;
 314
 315         if (rc) {
 316                 IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 317         }
 318
 319         return rc;
 320 }
 321
 322 static int ip_frag_reinit(struct ipq *qp)
 323 {
 324         struct sk_buff *fp;
 325
 326         if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) {
 327                 atomic_inc(&qp->q.refcnt);
 328                 return -ETIMEDOUT;
 329         }
 330
 331         fp = qp->q.fragments;
 332         do {
 333                 struct sk_buff *xp = fp->next;
 334                 frag_kfree_skb(fp, NULL);
 335                 fp = xp;
 336         } while (fp);
 337
 338         qp->q.last_in = 0;
 339         qp->q.len = 0;
 340         qp->q.meat = 0;
 341         qp->q.fragments = NULL;
 342         qp->iif = 0;
 343
 344         return 0;
 345 }
 346
 347 /* Add new segment to existing queue. */
 348 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
 349 {
 350         struct sk_buff *prev, *next;
 351         struct net_device *dev;
 352         int flags, offset;
 353         int ihl, end;
 354         int err = -ENOENT;
 355
 356         if (qp->q.last_in & COMPLETE)
 357                 goto err;
 358
 359         if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
 360             unlikely(ip_frag_too_far(qp)) &&
 361             unlikely(err = ip_frag_reinit(qp))) {
 362                 ipq_kill(qp);
 363                 goto err;
 364         }
 365
 366         offset = ntohs(ip_hdr(skb)->frag_off);
 367         flags = offset & ~IP_OFFSET;
 368         offset &= IP_OFFSET;
 369         offset <<= 3;           /* offset is in 8-byte chunks */
 370         ihl = ip_hdrlen(skb);
 371
 372         /* Determine the position of this fragment. */
 373         end = offset + skb->len - ihl;
 374         err = -EINVAL;
 375
 376         /* Is this the final fragment? */
 377         if ((flags & IP_MF) == 0) {
 378                 /* If we already have some bits beyond end
 379                  * or have different end, the segment is corrrupted.
 380                  */
 381                 if (end < qp->q.len ||
 382                     ((qp->q.last_in & LAST_IN) && end != qp->q.len))
 383                         goto err;
 384                 qp->q.last_in |= LAST_IN;
 385                 qp->q.len = end;
 386         } else {
 387                 if (end&7) {
 388                         end &= ~7;
 389                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 390                                 skb->ip_summed = CHECKSUM_NONE;
 391                 }
 392                 if (end > qp->q.len) {
 393                         /* Some bits beyond end -> corruption. */
 394                         if (qp->q.last_in & LAST_IN)
 395                                 goto err;
 396                         qp->q.len = end;
 397                 }
 398         }
 399         if (end == offset)
 400                 goto err;
 401
 402         err = -ENOMEM;
 403         if (pskb_pull(skb, ihl) == NULL)
 404                 goto err;
 405
 406         err = pskb_trim_rcsum(skb, end - offset);
 407         if (err)
 408                 goto err;
 409
 410         /* Find out which fragments are in front and at the back of us
 411          * in the chain of fragments so far.  We must know where to put
 412          * this fragment, right?
 413          */
 414         prev = NULL;
 415         for (next = qp->q.fragments; next != NULL; next = next->next) {
 416                 if (FRAG_CB(next)->offset >= offset)
 417                         break;  /* bingo! */
 418                 prev = next;
 419         }
 420
 421         /* We found where to put this one.  Check for overlap with
 422          * preceding fragment, and, if needed, align things so that
 423          * any overlaps are eliminated.
 424          */
 425         if (prev) {
 426                 int i = (FRAG_CB(prev)->offset + prev->len) - offset;
 427
 428                 if (i > 0) {
 429                         offset += i;
 430                         err = -EINVAL;
 431                         if (end <= offset)
 432                                 goto err;
 433                         err = -ENOMEM;
 434                         if (!pskb_pull(skb, i))
 435                                 goto err;
 436                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 437                                 skb->ip_summed = CHECKSUM_NONE;
 438                 }
 439         }
 440
 441         err = -ENOMEM;
 442
 443         while (next && FRAG_CB(next)->offset < end) {
 444                 int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */
 445
 446                 if (i < next->len) {
 447                         /* Eat head of the next overlapped fragment
 448                          * and leave the loop. The next ones cannot overlap.
 449                          */
 450                         if (!pskb_pull(next, i))
 451                                 goto err;
 452                         FRAG_CB(next)->offset += i;
 453                         qp->q.meat -= i;
 454                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
 455                                 next->ip_summed = CHECKSUM_NONE;
 456                         break;
 457                 } else {
 458                         struct sk_buff *free_it = next;
 459
 460                         /* Old fragment is completely overridden with
 461                          * new one drop it.
 462                          */
 463                         next = next->next;
 464
 465                         if (prev)
 466                                 prev->next = next;
 467                         else
 468                                 qp->q.fragments = next;
 469
 470                         qp->q.meat -= free_it->len;
 471                         frag_kfree_skb(free_it, NULL);
 472                 }
 473         }
 474
 475         FRAG_CB(skb)->offset = offset;
 476
 477         /* Insert this fragment in the chain of fragments. */
 478         skb->next = next;
 479         if (prev)
 480                 prev->next = skb;
 481         else
 482                 qp->q.fragments = skb;
 483
 484         dev = skb->dev;
 485         if (dev) {
 486                 qp->iif = dev->ifindex;
 487                 skb->dev = NULL;
 488         }
 489         qp->q.stamp = skb->tstamp;
 490         qp->q.meat += skb->len;
 491         atomic_add(skb->truesize, &ip4_frags.mem);
 492         if (offset == 0)
 493                 qp->q.last_in |= FIRST_IN;
 494
 495         if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
 496                 return ip_frag_reasm(qp, prev, dev);
 497
 498         write_lock(&ip4_frags.lock);
 499         list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
 500         write_unlock(&ip4_frags.lock);
 501         return -EINPROGRESS;
 502
 503 err:
 504         kfree_skb(skb);
 505         return err;
 506 }
 507
 508
 509 /* Build a new IP datagram from all its fragments. */
 510
 511 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
 512                          struct net_device *dev)
 513 {
 514         struct iphdr *iph;
 515         struct sk_buff *fp, *head = qp->q.fragments;
 516         int len;
 517         int ihlen;
 518         int err;
 519
 520         ipq_kill(qp);
 521
 522         /* Make the one we just received the head. */
 523         if (prev) {
 524                 head = prev->next;
 525                 fp = skb_clone(head, GFP_ATOMIC);
 526
 527                 if (!fp)
 528                         goto out_nomem;
 529
 530                 fp->next = head->next;
 531                 prev->next = fp;
 532
 533                 skb_morph(head, qp->q.fragments);
 534                 head->next = qp->q.fragments->next;
 535
 536                 kfree_skb(qp->q.fragments);
 537                 qp->q.fragments = head;
 538         }
 539
 540         BUG_TRAP(head != NULL);
 541         BUG_TRAP(FRAG_CB(head)->offset == 0);
 542
 543         /* Allocate a new buffer for the datagram. */
 544         ihlen = ip_hdrlen(head);
 545         len = ihlen + qp->q.len;
 546
 547         err = -E2BIG;
 548         if (len > 65535)
 549                 goto out_oversize;
 550
 551         /* Head of list must not be cloned. */
 552         err = -ENOMEM;
 553         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC))
 554                 goto out_nomem;
 555
 556         /* If the first fragment is fragmented itself, we split
 557          * it to two chunks: the first with data and paged part
 558          * and the second, holding only fragments. */
 559         if (skb_shinfo(head)->frag_list) {
 560                 struct sk_buff *clone;
 561                 int i, plen = 0;
 562
 563                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL)
 564                         goto out_nomem;
 565                 clone->next = head->next;
 566                 head->next = clone;
 567                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
 568                 skb_shinfo(head)->frag_list = NULL;
 569                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
 570                         plen += skb_shinfo(head)->frags[i].size;
 571                 clone->len = clone->data_len = head->data_len - plen;
 572                 head->data_len -= clone->len;
 573                 head->len -= clone->len;
 574                 clone->csum = 0;
 575                 clone->ip_summed = head->ip_summed;
 576                 atomic_add(clone->truesize, &ip4_frags.mem);
 577         }
 578
 579         skb_shinfo(head)->frag_list = head->next;
 580         skb_push(head, head->data - skb_network_header(head));
 581         atomic_sub(head->truesize, &ip4_frags.mem);
 582
 583         for (fp=head->next; fp; fp = fp->next) {
 584                 head->data_len += fp->len;
 585                 head->len += fp->len;
 586                 if (head->ip_summed != fp->ip_summed)
 587                         head->ip_summed = CHECKSUM_NONE;
 588                 else if (head->ip_summed == CHECKSUM_COMPLETE)
 589                         head->csum = csum_add(head->csum, fp->csum);
 590                 head->truesize += fp->truesize;
 591                 atomic_sub(fp->truesize, &ip4_frags.mem);
 592         }
 593
 594         head->next = NULL;
 595         head->dev = dev;
 596         head->tstamp = qp->q.stamp;
 597
 598         iph = ip_hdr(head);
 599         iph->frag_off = 0;
 600         iph->tot_len = htons(len);
 601         IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
 602         qp->q.fragments = NULL;
 603         return 0;
 604
 605 out_nomem:
 606         LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing "
 607                               "queue %p\n", qp);
 608         goto out_fail;
 609 out_oversize:
 610         if (net_ratelimit())
 611                 printk(KERN_INFO
 612                         "Oversized IP packet from %d.%d.%d.%d.\n",
 613                         NIPQUAD(qp->saddr));
 614 out_fail:
 615         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 616         return err;
 617 }
 618
 619 /* Process an incoming IP datagram fragment. */
 620 int ip_defrag(struct sk_buff *skb, u32 user)
 621 {
 622         struct ipq *qp;
 623
 624         IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
 625
 626         /* Start by cleaning up the memory. */
 627         if (atomic_read(&ip4_frags.mem) > ip4_frags_ctl.high_thresh)
 628                 ip_evictor();
 629
 630         /* Lookup (or create) queue header */
 631         if ((qp = ip_find(ip_hdr(skb), user)) != NULL) {
 632                 int ret;
 633
 634                 spin_lock(&qp->q.lock);
 635
 636                 ret = ip_frag_queue(qp, skb);
 637
 638                 spin_unlock(&qp->q.lock);
 639                 ipq_put(qp);
 640                 return ret;
 641         }
 642
 643         IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 644         kfree_skb(skb);
 645         return -ENOMEM;
 646 }
 647
 648 void __init ipfrag_init(void)
 649 {
 650         ip4_frags.ctl = &ip4_frags_ctl;
 651         ip4_frags.hashfn = ip4_hashfn;
 652         ip4_frags.destructor = ip4_frag_free;
 653         ip4_frags.skb_free = NULL;
 654         ip4_frags.qsize = sizeof(struct ipq);
 655         ip4_frags.equal = ip4_frag_equal;
 656         inet_frags_init(&ip4_frags);
 657 }
 658
 659 EXPORT_SYMBOL(ip_defrag);