net/ipv4/ip_fragment.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              The IP fragmentation functionality.
   8  *
   9  * Authors:     Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox <alan@lxorguk.ukuu.org.uk>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       Split from ip.c , see ip_input.c for history.
  14  *              David S. Miller :       Begin massive cleanup...
  15  *              Andi Kleen      :       Add sysctls.
  16  *              xxxx            :       Overlapfrag bug.
  17  *              Ultima          :       ip_expire() kernel panic.
  18  *              Bill Hawes      :       Frag accounting and evictor fixes.
  19  *              John McDonald   :       0 length frag bug.
  20  *              Alexey Kuznetsov:       SMP races, threading, cleanup.
  21  *              Patrick McHardy :       LRU queue of frag heads for evictor.
  22  */
  23
  24 #define pr_fmt(fmt) "IPv4: " fmt
  25
  26 #include <linux/compiler.h>
  27 #include <linux/module.h>
  28 #include <linux/types.h>
  29 #include <linux/mm.h>
  30 #include <linux/jiffies.h>
  31 #include <linux/skbuff.h>
  32 #include <linux/list.h>
  33 #include <linux/ip.h>
  34 #include <linux/icmp.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/jhash.h>
  37 #include <linux/random.h>
  38 #include <linux/slab.h>
  39 #include <net/route.h>
  40 #include <net/dst.h>
  41 #include <net/sock.h>
  42 #include <net/ip.h>
  43 #include <net/icmp.h>
  44 #include <net/checksum.h>
  45 #include <net/inetpeer.h>
  46 #include <net/inet_frag.h>
  47 #include <linux/tcp.h>
  48 #include <linux/udp.h>
  49 #include <linux/inet.h>
  50 #include <linux/netfilter_ipv4.h>
  51 #include <net/inet_ecn.h>
  52 #include <net/l3mdev.h>
  53
  54 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
  55  * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
  56  * as well. Or notify me, at least. --ANK
  57  */
  58 static const char ip_frag_cache_name[] = "ip4-frags";
  59
  60 /* Describe an entry in the "incomplete datagrams" queue. */
  61 struct ipq {
  62         struct inet_frag_queue q;
  63
  64         u8              ecn; /* RFC3168 support */
  65         u16             max_df_size; /* largest frag with DF set seen */
  66         int             iif;
  67         unsigned int    rid;
  68         struct inet_peer *peer;
  69 };
  70
  71 static u8 ip4_frag_ecn(u8 tos)
  72 {
  73         return 1 << (tos & INET_ECN_MASK);
  74 }
  75
  76 static struct inet_frags ip4_frags;
  77
  78 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
  79                          struct sk_buff *prev_tail, struct net_device *dev);
  80
  81
  82 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
  83 {
  84         struct ipq *qp = container_of(q, struct ipq, q);
  85         struct netns_ipv4 *ipv4 = container_of(q->net, struct netns_ipv4,
  86                                                frags);
  87         struct net *net = container_of(ipv4, struct net, ipv4);
  88
  89         const struct frag_v4_compare_key *key = a;
  90
  91         q->key.v4 = *key;
  92         qp->ecn = 0;
  93         qp->peer = q->net->max_dist ?
  94                 inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
  95                 NULL;
  96 }
  97
  98 static void ip4_frag_free(struct inet_frag_queue *q)
  99 {
 100         struct ipq *qp;
 101
 102         qp = container_of(q, struct ipq, q);
 103         if (qp->peer)
 104                 inet_putpeer(qp->peer);
 105 }
 106
 107
 108 /* Destruction primitives. */
 109
 110 static void ipq_put(struct ipq *ipq)
 111 {
 112         inet_frag_put(&ipq->q);
 113 }
 114
 115 /* Kill ipq entry. It is not destroyed immediately,
 116  * because caller (and someone more) holds reference count.
 117  */
 118 static void ipq_kill(struct ipq *ipq)
 119 {
 120         inet_frag_kill(&ipq->q);
 121 }
 122
 123 static bool frag_expire_skip_icmp(u32 user)
 124 {
 125         return user == IP_DEFRAG_AF_PACKET ||
 126                ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
 127                                          __IP_DEFRAG_CONNTRACK_IN_END) ||
 128                ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
 129                                          __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
 130 }
 131
 132 /*
 133  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
 134  */
 135 static void ip_expire(struct timer_list *t)
 136 {
 137         struct inet_frag_queue *frag = from_timer(frag, t, timer);
 138         const struct iphdr *iph;
 139         struct sk_buff *head = NULL;
 140         struct net *net;
 141         struct ipq *qp;
 142         int err;
 143
 144         qp = container_of(frag, struct ipq, q);
 145         net = container_of(qp->q.net, struct net, ipv4.frags);
 146
 147         rcu_read_lock();
 148         spin_lock(&qp->q.lock);
 149
 150         if (qp->q.flags & INET_FRAG_COMPLETE)
 151                 goto out;
 152
 153         ipq_kill(qp);
 154         __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
 155         __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
 156
 157         if (!(qp->q.flags & INET_FRAG_FIRST_IN))
 158                 goto out;
 159
 160         /* sk_buff::dev and sk_buff::rbnode are unionized. So we
 161          * pull the head out of the tree in order to be able to
 162          * deal with head->dev.
 163          */
 164         head = inet_frag_pull_head(&qp->q);
 165         if (!head)
 166                 goto out;
 167         head->dev = dev_get_by_index_rcu(net, qp->iif);
 168         if (!head->dev)
 169                 goto out;
 170
 171
 172         /* skb has no dst, perform route lookup again */
 173         iph = ip_hdr(head);
 174         err = ip_route_input_noref(head, iph->daddr, iph->saddr,
 175                                            iph->tos, head->dev);
 176         if (err)
 177                 goto out;
 178
 179         /* Only an end host needs to send an ICMP
 180          * "Fragment Reassembly Timeout" message, per RFC792.
 181          */
 182         if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
 183             (skb_rtable(head)->rt_type != RTN_LOCAL))
 184                 goto out;
 185
 186         spin_unlock(&qp->q.lock);
 187         icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
 188         goto out_rcu_unlock;
 189
 190 out:
 191         spin_unlock(&qp->q.lock);
 192 out_rcu_unlock:
 193         rcu_read_unlock();
 194         kfree_skb(head);
 195         ipq_put(qp);
 196 }
 197
 198 /* Find the correct entry in the "incomplete datagrams" queue for
 199  * this IP datagram, and create new one, if nothing is found.
 200  */
 201 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
 202                            u32 user, int vif)
 203 {
 204         struct frag_v4_compare_key key = {
 205                 .saddr = iph->saddr,
 206                 .daddr = iph->daddr,
 207                 .user = user,
 208                 .vif = vif,
 209                 .id = iph->id,
 210                 .protocol = iph->protocol,
 211         };
 212         struct inet_frag_queue *q;
 213
 214         q = inet_frag_find(&net->ipv4.frags, &key);
 215         if (!q)
 216                 return NULL;
 217
 218         return container_of(q, struct ipq, q);
 219 }
 220
 221 /* Is the fragment too far ahead to be part of ipq? */
 222 static int ip_frag_too_far(struct ipq *qp)
 223 {
 224         struct inet_peer *peer = qp->peer;
 225         unsigned int max = qp->q.net->max_dist;
 226         unsigned int start, end;
 227
 228         int rc;
 229
 230         if (!peer || !max)
 231                 return 0;
 232
 233         start = qp->rid;
 234         end = atomic_inc_return(&peer->rid);
 235         qp->rid = end;
 236
 237         rc = qp->q.fragments_tail && (end - start) > max;
 238
 239         if (rc) {
 240                 struct net *net;
 241
 242                 net = container_of(qp->q.net, struct net, ipv4.frags);
 243                 __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
 244         }
 245
 246         return rc;
 247 }
 248
 249 static int ip_frag_reinit(struct ipq *qp)
 250 {
 251         unsigned int sum_truesize = 0;
 252
 253         if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
 254                 refcount_inc(&qp->q.refcnt);
 255                 return -ETIMEDOUT;
 256         }
 257
 258         sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments);
 259         sub_frag_mem_limit(qp->q.net, sum_truesize);
 260
 261         qp->q.flags = 0;
 262         qp->q.len = 0;
 263         qp->q.meat = 0;
 264         qp->q.rb_fragments = RB_ROOT;
 265         qp->q.fragments_tail = NULL;
 266         qp->q.last_run_head = NULL;
 267         qp->iif = 0;
 268         qp->ecn = 0;
 269
 270         return 0;
 271 }
 272
 273 /* Add new segment to existing queue. */
 274 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
 275 {
 276         struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
 277         int ihl, end, flags, offset;
 278         struct sk_buff *prev_tail;
 279         struct net_device *dev;
 280         unsigned int fragsize;
 281         int err = -ENOENT;
 282         u8 ecn;
 283
 284         if (qp->q.flags & INET_FRAG_COMPLETE)
 285                 goto err;
 286
 287         if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
 288             unlikely(ip_frag_too_far(qp)) &&
 289             unlikely(err = ip_frag_reinit(qp))) {
 290                 ipq_kill(qp);
 291                 goto err;
 292         }
 293
 294         ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
 295         offset = ntohs(ip_hdr(skb)->frag_off);
 296         flags = offset & ~IP_OFFSET;
 297         offset &= IP_OFFSET;
 298         offset <<= 3;           /* offset is in 8-byte chunks */
 299         ihl = ip_hdrlen(skb);
 300
 301         /* Determine the position of this fragment. */
 302         end = offset + skb->len - skb_network_offset(skb) - ihl;
 303         err = -EINVAL;
 304
 305         /* Is this the final fragment? */
 306         if ((flags & IP_MF) == 0) {
 307                 /* If we already have some bits beyond end
 308                  * or have different end, the segment is corrupted.
 309                  */
 310                 if (end < qp->q.len ||
 311                     ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
 312                         goto discard_qp;
 313                 qp->q.flags |= INET_FRAG_LAST_IN;
 314                 qp->q.len = end;
 315         } else {
 316                 if (end&7) {
 317                         end &= ~7;
 318                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 319                                 skb->ip_summed = CHECKSUM_NONE;
 320                 }
 321                 if (end > qp->q.len) {
 322                         /* Some bits beyond end -> corruption. */
 323                         if (qp->q.flags & INET_FRAG_LAST_IN)
 324                                 goto discard_qp;
 325                         qp->q.len = end;
 326                 }
 327         }
 328         if (end == offset)
 329                 goto discard_qp;
 330
 331         err = -ENOMEM;
 332         if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
 333                 goto discard_qp;
 334
 335         err = pskb_trim_rcsum(skb, end - offset);
 336         if (err)
 337                 goto discard_qp;
 338
 339         /* Note : skb->rbnode and skb->dev share the same location. */
 340         dev = skb->dev;
 341         /* Makes sure compiler wont do silly aliasing games */
 342         barrier();
 343
 344         prev_tail = qp->q.fragments_tail;
 345         err = inet_frag_queue_insert(&qp->q, skb, offset, end);
 346         if (err)
 347                 goto insert_error;
 348
 349         if (dev)
 350                 qp->iif = dev->ifindex;
 351
 352         qp->q.stamp = skb->tstamp;
 353         qp->q.meat += skb->len;
 354         qp->ecn |= ecn;
 355         add_frag_mem_limit(qp->q.net, skb->truesize);
 356         if (offset == 0)
 357                 qp->q.flags |= INET_FRAG_FIRST_IN;
 358
 359         fragsize = skb->len + ihl;
 360
 361         if (fragsize > qp->q.max_size)
 362                 qp->q.max_size = fragsize;
 363
 364         if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
 365             fragsize > qp->max_df_size)
 366                 qp->max_df_size = fragsize;
 367
 368         if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
 369             qp->q.meat == qp->q.len) {
 370                 unsigned long orefdst = skb->_skb_refdst;
 371
 372                 skb->_skb_refdst = 0UL;
 373                 err = ip_frag_reasm(qp, skb, prev_tail, dev);
 374                 skb->_skb_refdst = orefdst;
 375                 if (err)
 376                         inet_frag_kill(&qp->q);
 377                 return err;
 378         }
 379
 380         skb_dst_drop(skb);
 381         return -EINPROGRESS;
 382
 383 insert_error:
 384         if (err == IPFRAG_DUP) {
 385                 kfree_skb(skb);
 386                 return -EINVAL;
 387         }
 388         err = -EINVAL;
 389         __IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
 390 discard_qp:
 391         inet_frag_kill(&qp->q);
 392         __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
 393 err:
 394         kfree_skb(skb);
 395         return err;
 396 }
 397
 398 /* Build a new IP datagram from all its fragments. */
 399 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
 400                          struct sk_buff *prev_tail, struct net_device *dev)
 401 {
 402         struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
 403         struct iphdr *iph;
 404         void *reasm_data;
 405         int len, err;
 406         u8 ecn;
 407
 408         ipq_kill(qp);
 409
 410         ecn = ip_frag_ecn_table[qp->ecn];
 411         if (unlikely(ecn == 0xff)) {
 412                 err = -EINVAL;
 413                 goto out_fail;
 414         }
 415
 416         /* Make the one we just received the head. */
 417         reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
 418         if (!reasm_data)
 419                 goto out_nomem;
 420
 421         len = ip_hdrlen(skb) + qp->q.len;
 422         err = -E2BIG;
 423         if (len > 65535)
 424                 goto out_oversize;
 425
 426         inet_frag_reasm_finish(&qp->q, skb, reasm_data);
 427
 428         skb->dev = dev;
 429         IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
 430
 431         iph = ip_hdr(skb);
 432         iph->tot_len = htons(len);
 433         iph->tos |= ecn;
 434
 435         /* When we set IP_DF on a refragmented skb we must also force a
 436          * call to ip_fragment to avoid forwarding a DF-skb of size s while
 437          * original sender only sent fragments of size f (where f < s).
 438          *
 439          * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
 440          * frag seen to avoid sending tiny DF-fragments in case skb was built
 441          * from one very small df-fragment and one large non-df frag.
 442          */
 443         if (qp->max_df_size == qp->q.max_size) {
 444                 IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
 445                 iph->frag_off = htons(IP_DF);
 446         } else {
 447                 iph->frag_off = 0;
 448         }
 449
 450         ip_send_check(iph);
 451
 452         __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
 453         qp->q.rb_fragments = RB_ROOT;
 454         qp->q.fragments_tail = NULL;
 455         qp->q.last_run_head = NULL;
 456         return 0;
 457
 458 out_nomem:
 459         net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
 460         err = -ENOMEM;
 461         goto out_fail;
 462 out_oversize:
 463         net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
 464 out_fail:
 465         __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
 466         return err;
 467 }
 468
 469 /* Process an incoming IP datagram fragment. */
 470 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
 471 {
 472         struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
 473         int vif = l3mdev_master_ifindex_rcu(dev);
 474         struct ipq *qp;
 475
 476         __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
 477         skb_orphan(skb);
 478
 479         /* Lookup (or create) queue header */
 480         qp = ip_find(net, ip_hdr(skb), user, vif);
 481         if (qp) {
 482                 int ret;
 483
 484                 spin_lock(&qp->q.lock);
 485
 486                 ret = ip_frag_queue(qp, skb);
 487
 488                 spin_unlock(&qp->q.lock);
 489                 ipq_put(qp);
 490                 return ret;
 491         }
 492
 493         __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
 494         kfree_skb(skb);
 495         return -ENOMEM;
 496 }
 497 EXPORT_SYMBOL(ip_defrag);
 498
 499 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
 500 {
 501         struct iphdr iph;
 502         int netoff;
 503         u32 len;
 504
 505         if (skb->protocol != htons(ETH_P_IP))
 506                 return skb;
 507
 508         netoff = skb_network_offset(skb);
 509
 510         if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
 511                 return skb;
 512
 513         if (iph.ihl < 5 || iph.version != 4)
 514                 return skb;
 515
 516         len = ntohs(iph.tot_len);
 517         if (skb->len < netoff + len || len < (iph.ihl * 4))
 518                 return skb;
 519
 520         if (ip_is_fragment(&iph)) {
 521                 skb = skb_share_check(skb, GFP_ATOMIC);
 522                 if (skb) {
 523                         if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
 524                                 kfree_skb(skb);
 525                                 return NULL;
 526                         }
 527                         if (pskb_trim_rcsum(skb, netoff + len)) {
 528                                 kfree_skb(skb);
 529                                 return NULL;
 530                         }
 531                         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 532                         if (ip_defrag(net, skb, user))
 533                                 return NULL;
 534                         skb_clear_hash(skb);
 535                 }
 536         }
 537         return skb;
 538 }
 539 EXPORT_SYMBOL(ip_check_defrag);
 540
 541 #ifdef CONFIG_SYSCTL
 542 static int dist_min;
 543
 544 static struct ctl_table ip4_frags_ns_ctl_table[] = {
 545         {
 546                 .procname       = "ipfrag_high_thresh",
 547                 .data           = &init_net.ipv4.frags.high_thresh,
 548                 .maxlen         = sizeof(unsigned long),
 549                 .mode           = 0644,
 550                 .proc_handler   = proc_doulongvec_minmax,
 551                 .extra1         = &init_net.ipv4.frags.low_thresh
 552         },
 553         {
 554                 .procname       = "ipfrag_low_thresh",
 555                 .data           = &init_net.ipv4.frags.low_thresh,
 556                 .maxlen         = sizeof(unsigned long),
 557                 .mode           = 0644,
 558                 .proc_handler   = proc_doulongvec_minmax,
 559                 .extra2         = &init_net.ipv4.frags.high_thresh
 560         },
 561         {
 562                 .procname       = "ipfrag_time",
 563                 .data           = &init_net.ipv4.frags.timeout,
 564                 .maxlen         = sizeof(int),
 565                 .mode           = 0644,
 566                 .proc_handler   = proc_dointvec_jiffies,
 567         },
 568         {
 569                 .procname       = "ipfrag_max_dist",
 570                 .data           = &init_net.ipv4.frags.max_dist,
 571                 .maxlen         = sizeof(int),
 572                 .mode           = 0644,
 573                 .proc_handler   = proc_dointvec_minmax,
 574                 .extra1         = &dist_min,
 575         },
 576         { }
 577 };
 578
 579 /* secret interval has been deprecated */
 580 static int ip4_frags_secret_interval_unused;
 581 static struct ctl_table ip4_frags_ctl_table[] = {
 582         {
 583                 .procname       = "ipfrag_secret_interval",
 584                 .data           = &ip4_frags_secret_interval_unused,
 585                 .maxlen         = sizeof(int),
 586                 .mode           = 0644,
 587                 .proc_handler   = proc_dointvec_jiffies,
 588         },
 589         { }
 590 };
 591
 592 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
 593 {
 594         struct ctl_table *table;
 595         struct ctl_table_header *hdr;
 596
 597         table = ip4_frags_ns_ctl_table;
 598         if (!net_eq(net, &init_net)) {
 599                 table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
 600                 if (!table)
 601                         goto err_alloc;
 602
 603                 table[0].data = &net->ipv4.frags.high_thresh;
 604                 table[0].extra1 = &net->ipv4.frags.low_thresh;
 605                 table[1].data = &net->ipv4.frags.low_thresh;
 606                 table[1].extra2 = &net->ipv4.frags.high_thresh;
 607                 table[2].data = &net->ipv4.frags.timeout;
 608                 table[3].data = &net->ipv4.frags.max_dist;
 609         }
 610
 611         hdr = register_net_sysctl(net, "net/ipv4", table);
 612         if (!hdr)
 613                 goto err_reg;
 614
 615         net->ipv4.frags_hdr = hdr;
 616         return 0;
 617
 618 err_reg:
 619         if (!net_eq(net, &init_net))
 620                 kfree(table);
 621 err_alloc:
 622         return -ENOMEM;
 623 }
 624
 625 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
 626 {
 627         struct ctl_table *table;
 628
 629         table = net->ipv4.frags_hdr->ctl_table_arg;
 630         unregister_net_sysctl_table(net->ipv4.frags_hdr);
 631         kfree(table);
 632 }
 633
 634 static void __init ip4_frags_ctl_register(void)
 635 {
 636         register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
 637 }
 638 #else
 639 static int ip4_frags_ns_ctl_register(struct net *net)
 640 {
 641         return 0;
 642 }
 643
 644 static void ip4_frags_ns_ctl_unregister(struct net *net)
 645 {
 646 }
 647
 648 static void __init ip4_frags_ctl_register(void)
 649 {
 650 }
 651 #endif
 652
 653 static int __net_init ipv4_frags_init_net(struct net *net)
 654 {
 655         int res;
 656
 657         /* Fragment cache limits.
 658          *
 659          * The fragment memory accounting code, (tries to) account for
 660          * the real memory usage, by measuring both the size of frag
 661          * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
 662          * and the SKB's truesize.
 663          *
 664          * A 64K fragment consumes 129736 bytes (44*2944)+200
 665          * (1500 truesize == 2944, sizeof(struct ipq) == 200)
 666          *
 667          * We will commit 4MB at one time. Should we cross that limit
 668          * we will prune down to 3MB, making room for approx 8 big 64K
 669          * fragments 8x128k.
 670          */
 671         net->ipv4.frags.high_thresh = 4 * 1024 * 1024;
 672         net->ipv4.frags.low_thresh  = 3 * 1024 * 1024;
 673         /*
 674          * Important NOTE! Fragment queue must be destroyed before MSL expires.
 675          * RFC791 is wrong proposing to prolongate timer each fragment arrival
 676          * by TTL.
 677          */
 678         net->ipv4.frags.timeout = IP_FRAG_TIME;
 679
 680         net->ipv4.frags.max_dist = 64;
 681         net->ipv4.frags.f = &ip4_frags;
 682
 683         res = inet_frags_init_net(&net->ipv4.frags);
 684         if (res < 0)
 685                 return res;
 686         res = ip4_frags_ns_ctl_register(net);
 687         if (res < 0)
 688                 inet_frags_exit_net(&net->ipv4.frags);
 689         return res;
 690 }
 691
 692 static void __net_exit ipv4_frags_exit_net(struct net *net)
 693 {
 694         ip4_frags_ns_ctl_unregister(net);
 695         inet_frags_exit_net(&net->ipv4.frags);
 696 }
 697
 698 static struct pernet_operations ip4_frags_ops = {
 699         .init = ipv4_frags_init_net,
 700         .exit = ipv4_frags_exit_net,
 701 };
 702
 703
 704 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
 705 {
 706         return jhash2(data,
 707                       sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
 708 }
 709
 710 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
 711 {
 712         const struct inet_frag_queue *fq = data;
 713
 714         return jhash2((const u32 *)&fq->key.v4,
 715                       sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
 716 }
 717
 718 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
 719 {
 720         const struct frag_v4_compare_key *key = arg->key;
 721         const struct inet_frag_queue *fq = ptr;
 722
 723         return !!memcmp(&fq->key, key, sizeof(*key));
 724 }
 725
 726 static const struct rhashtable_params ip4_rhash_params = {
 727         .head_offset            = offsetof(struct inet_frag_queue, node),
 728         .key_offset             = offsetof(struct inet_frag_queue, key),
 729         .key_len                = sizeof(struct frag_v4_compare_key),
 730         .hashfn                 = ip4_key_hashfn,
 731         .obj_hashfn             = ip4_obj_hashfn,
 732         .obj_cmpfn              = ip4_obj_cmpfn,
 733         .automatic_shrinking    = true,
 734 };
 735
 736 void __init ipfrag_init(void)
 737 {
 738         ip4_frags.constructor = ip4_frag_init;
 739         ip4_frags.destructor = ip4_frag_free;
 740         ip4_frags.qsize = sizeof(struct ipq);
 741         ip4_frags.frag_expire = ip_expire;
 742         ip4_frags.frags_cache_name = ip_frag_cache_name;
 743         ip4_frags.rhash_params = ip4_rhash_params;
 744         if (inet_frags_init(&ip4_frags))
 745                 panic("IP: failed to allocate ip4_frags cache\n");
 746         ip4_frags_ctl_register();
 747         register_pernet_subsys(&ip4_frags_ops);
 748 }