net/batman-adv/fragmentation.c

   1 /* Copyright (C) 2013-2016  B.A.T.M.A.N. contributors:
   2  *
   3  * Martin Hundebøll <martin@hundeboll.net>
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of version 2 of the GNU General Public
   7  * License as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  * General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  16  */
  17
  18 #include "fragmentation.h"
  19 #include "main.h"
  20
  21 #include <linux/atomic.h>
  22 #include <linux/byteorder/generic.h>
  23 #include <linux/etherdevice.h>
  24 #include <linux/fs.h>
  25 #include <linux/if_ether.h>
  26 #include <linux/jiffies.h>
  27 #include <linux/kernel.h>
  28 #include <linux/lockdep.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/pkt_sched.h>
  31 #include <linux/skbuff.h>
  32 #include <linux/slab.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/string.h>
  35
  36 #include "hard-interface.h"
  37 #include "originator.h"
  38 #include "packet.h"
  39 #include "routing.h"
  40 #include "send.h"
  41 #include "soft-interface.h"
  42
  43 /**
  44  * batadv_frag_clear_chain - delete entries in the fragment buffer chain
  45  * @head: head of chain with entries.
  46  *
  47  * Free fragments in the passed hlist. Should be called with appropriate lock.
  48  */
  49 static void batadv_frag_clear_chain(struct hlist_head *head)
  50 {
  51         struct batadv_frag_list_entry *entry;
  52         struct hlist_node *node;
  53
  54         hlist_for_each_entry_safe(entry, node, head, list) {
  55                 hlist_del(&entry->list);
  56                 kfree_skb(entry->skb);
  57                 kfree(entry);
  58         }
  59 }
  60
  61 /**
  62  * batadv_frag_purge_orig - free fragments associated to an orig
  63  * @orig_node: originator to free fragments from
  64  * @check_cb: optional function to tell if an entry should be purged
  65  */
  66 void batadv_frag_purge_orig(struct batadv_orig_node *orig_node,
  67                             bool (*check_cb)(struct batadv_frag_table_entry *))
  68 {
  69         struct batadv_frag_table_entry *chain;
  70         u8 i;
  71
  72         for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) {
  73                 chain = &orig_node->fragments[i];
  74                 spin_lock_bh(&chain->lock);
  75
  76                 if (!check_cb || check_cb(chain)) {
  77                         batadv_frag_clear_chain(&chain->head);
  78                         chain->size = 0;
  79                 }
  80
  81                 spin_unlock_bh(&chain->lock);
  82         }
  83 }
  84
  85 /**
  86  * batadv_frag_size_limit - maximum possible size of packet to be fragmented
  87  *
  88  * Return: the maximum size of payload that can be fragmented.
  89  */
  90 static int batadv_frag_size_limit(void)
  91 {
  92         int limit = BATADV_FRAG_MAX_FRAG_SIZE;
  93
  94         limit -= sizeof(struct batadv_frag_packet);
  95         limit *= BATADV_FRAG_MAX_FRAGMENTS;
  96
  97         return limit;
  98 }
  99
 100 /**
 101  * batadv_frag_init_chain - check and prepare fragment chain for new fragment
 102  * @chain: chain in fragments table to init
 103  * @seqno: sequence number of the received fragment
 104  *
 105  * Make chain ready for a fragment with sequence number "seqno". Delete existing
 106  * entries if they have an "old" sequence number.
 107  *
 108  * Caller must hold chain->lock.
 109  *
 110  * Return: true if chain is empty and caller can just insert the new fragment
 111  * without searching for the right position.
 112  */
 113 static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain,
 114                                    u16 seqno)
 115 {
 116         lockdep_assert_held(&chain->lock);
 117
 118         if (chain->seqno == seqno)
 119                 return false;
 120
 121         if (!hlist_empty(&chain->head))
 122                 batadv_frag_clear_chain(&chain->head);
 123
 124         chain->size = 0;
 125         chain->seqno = seqno;
 126
 127         return true;
 128 }
 129
 130 /**
 131  * batadv_frag_insert_packet - insert a fragment into a fragment chain
 132  * @orig_node: originator that the fragment was received from
 133  * @skb: skb to insert
 134  * @chain_out: list head to attach complete chains of fragments to
 135  *
 136  * Insert a new fragment into the reverse ordered chain in the right table
 137  * entry. The hash table entry is cleared if "old" fragments exist in it.
 138  *
 139  * Return: true if skb is buffered, false on error. If the chain has all the
 140  * fragments needed to merge the packet, the chain is moved to the passed head
 141  * to avoid locking the chain in the table.
 142  */
 143 static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node,
 144                                       struct sk_buff *skb,
 145                                       struct hlist_head *chain_out)
 146 {
 147         struct batadv_frag_table_entry *chain;
 148         struct batadv_frag_list_entry *frag_entry_new = NULL, *frag_entry_curr;
 149         struct batadv_frag_list_entry *frag_entry_last = NULL;
 150         struct batadv_frag_packet *frag_packet;
 151         u8 bucket;
 152         u16 seqno, hdr_size = sizeof(struct batadv_frag_packet);
 153         bool ret = false;
 154
 155         /* Linearize packet to avoid linearizing 16 packets in a row when doing
 156          * the later merge. Non-linear merge should be added to remove this
 157          * linearization.
 158          */
 159         if (skb_linearize(skb) < 0)
 160                 goto err;
 161
 162         frag_packet = (struct batadv_frag_packet *)skb->data;
 163         seqno = ntohs(frag_packet->seqno);
 164         bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
 165
 166         frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
 167         if (!frag_entry_new)
 168                 goto err;
 169
 170         frag_entry_new->skb = skb;
 171         frag_entry_new->no = frag_packet->no;
 172
 173         /* Select entry in the "chain table" and delete any prior fragments
 174          * with another sequence number. batadv_frag_init_chain() returns true,
 175          * if the list is empty at return.
 176          */
 177         chain = &orig_node->fragments[bucket];
 178         spin_lock_bh(&chain->lock);
 179         if (batadv_frag_init_chain(chain, seqno)) {
 180                 hlist_add_head(&frag_entry_new->list, &chain->head);
 181                 chain->size = skb->len - hdr_size;
 182                 chain->timestamp = jiffies;
 183                 chain->total_size = ntohs(frag_packet->total_size);
 184                 ret = true;
 185                 goto out;
 186         }
 187
 188         /* Find the position for the new fragment. */
 189         hlist_for_each_entry(frag_entry_curr, &chain->head, list) {
 190                 /* Drop packet if fragment already exists. */
 191                 if (frag_entry_curr->no == frag_entry_new->no)
 192                         goto err_unlock;
 193
 194                 /* Order fragments from highest to lowest. */
 195                 if (frag_entry_curr->no < frag_entry_new->no) {
 196                         hlist_add_before(&frag_entry_new->list,
 197                                          &frag_entry_curr->list);
 198                         chain->size += skb->len - hdr_size;
 199                         chain->timestamp = jiffies;
 200                         ret = true;
 201                         goto out;
 202                 }
 203
 204                 /* store current entry because it could be the last in list */
 205                 frag_entry_last = frag_entry_curr;
 206         }
 207
 208         /* Reached the end of the list, so insert after 'frag_entry_last'. */
 209         if (likely(frag_entry_last)) {
 210                 hlist_add_behind(&frag_entry_new->list, &frag_entry_last->list);
 211                 chain->size += skb->len - hdr_size;
 212                 chain->timestamp = jiffies;
 213                 ret = true;
 214         }
 215
 216 out:
 217         if (chain->size > batadv_frag_size_limit() ||
 218             chain->total_size != ntohs(frag_packet->total_size) ||
 219             chain->total_size > batadv_frag_size_limit()) {
 220                 /* Clear chain if total size of either the list or the packet
 221                  * exceeds the maximum size of one merged packet. Don't allow
 222                  * packets to have different total_size.
 223                  */
 224                 batadv_frag_clear_chain(&chain->head);
 225                 chain->size = 0;
 226         } else if (ntohs(frag_packet->total_size) == chain->size) {
 227                 /* All fragments received. Hand over chain to caller. */
 228                 hlist_move_list(&chain->head, chain_out);
 229                 chain->size = 0;
 230         }
 231
 232 err_unlock:
 233         spin_unlock_bh(&chain->lock);
 234
 235 err:
 236         if (!ret)
 237                 kfree(frag_entry_new);
 238
 239         return ret;
 240 }
 241
 242 /**
 243  * batadv_frag_merge_packets - merge a chain of fragments
 244  * @chain: head of chain with fragments
 245  *
 246  * Expand the first skb in the chain and copy the content of the remaining
 247  * skb's into the expanded one. After doing so, clear the chain.
 248  *
 249  * Return: the merged skb or NULL on error.
 250  */
 251 static struct sk_buff *
 252 batadv_frag_merge_packets(struct hlist_head *chain)
 253 {
 254         struct batadv_frag_packet *packet;
 255         struct batadv_frag_list_entry *entry;
 256         struct sk_buff *skb_out = NULL;
 257         int size, hdr_size = sizeof(struct batadv_frag_packet);
 258
 259         /* Remove first entry, as this is the destination for the rest of the
 260          * fragments.
 261          */
 262         entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
 263         hlist_del(&entry->list);
 264         skb_out = entry->skb;
 265         kfree(entry);
 266
 267         packet = (struct batadv_frag_packet *)skb_out->data;
 268         size = ntohs(packet->total_size);
 269
 270         /* Make room for the rest of the fragments. */
 271         if (pskb_expand_head(skb_out, 0, size - skb_out->len, GFP_ATOMIC) < 0) {
 272                 kfree_skb(skb_out);
 273                 skb_out = NULL;
 274                 goto free;
 275         }
 276
 277         /* Move the existing MAC header to just before the payload. (Override
 278          * the fragment header.)
 279          */
 280         skb_pull_rcsum(skb_out, hdr_size);
 281         memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN);
 282         skb_set_mac_header(skb_out, -ETH_HLEN);
 283         skb_reset_network_header(skb_out);
 284         skb_reset_transport_header(skb_out);
 285
 286         /* Copy the payload of the each fragment into the last skb */
 287         hlist_for_each_entry(entry, chain, list) {
 288                 size = entry->skb->len - hdr_size;
 289                 memcpy(skb_put(skb_out, size), entry->skb->data + hdr_size,
 290                        size);
 291         }
 292
 293 free:
 294         /* Locking is not needed, because 'chain' is not part of any orig. */
 295         batadv_frag_clear_chain(chain);
 296         return skb_out;
 297 }
 298
 299 /**
 300  * batadv_frag_skb_buffer - buffer fragment for later merge
 301  * @skb: skb to buffer
 302  * @orig_node_src: originator that the skb is received from
 303  *
 304  * Add fragment to buffer and merge fragments if possible.
 305  *
 306  * There are three possible outcomes: 1) Packet is merged: Return true and
 307  * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
 308  * to NULL; 3) Error: Return false and leave skb as is.
 309  *
 310  * Return: true when packet is merged or buffered, false when skb is not not
 311  * used.
 312  */
 313 bool batadv_frag_skb_buffer(struct sk_buff **skb,
 314                             struct batadv_orig_node *orig_node_src)
 315 {
 316         struct sk_buff *skb_out = NULL;
 317         struct hlist_head head = HLIST_HEAD_INIT;
 318         bool ret = false;
 319
 320         /* Add packet to buffer and table entry if merge is possible. */
 321         if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
 322                 goto out_err;
 323
 324         /* Leave if more fragments are needed to merge. */
 325         if (hlist_empty(&head))
 326                 goto out;
 327
 328         skb_out = batadv_frag_merge_packets(&head);
 329         if (!skb_out)
 330                 goto out_err;
 331
 332 out:
 333         *skb = skb_out;
 334         ret = true;
 335 out_err:
 336         return ret;
 337 }
 338
 339 /**
 340  * batadv_frag_skb_fwd - forward fragments that would exceed MTU when merged
 341  * @skb: skb to forward
 342  * @recv_if: interface that the skb is received on
 343  * @orig_node_src: originator that the skb is received from
 344  *
 345  * Look up the next-hop of the fragments payload and check if the merged packet
 346  * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
 347  * without merging it.
 348  *
 349  * Return: true if the fragment is consumed/forwarded, false otherwise.
 350  */
 351 bool batadv_frag_skb_fwd(struct sk_buff *skb,
 352                          struct batadv_hard_iface *recv_if,
 353                          struct batadv_orig_node *orig_node_src)
 354 {
 355         struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
 356         struct batadv_orig_node *orig_node_dst = NULL;
 357         struct batadv_neigh_node *neigh_node = NULL;
 358         struct batadv_frag_packet *packet;
 359         u16 total_size;
 360         bool ret = false;
 361
 362         packet = (struct batadv_frag_packet *)skb->data;
 363         orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
 364         if (!orig_node_dst)
 365                 goto out;
 366
 367         neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
 368         if (!neigh_node)
 369                 goto out;
 370
 371         /* Forward the fragment, if the merged packet would be too big to
 372          * be assembled.
 373          */
 374         total_size = ntohs(packet->total_size);
 375         if (total_size > neigh_node->if_incoming->net_dev->mtu) {
 376                 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
 377                 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
 378                                    skb->len + ETH_HLEN);
 379
 380                 packet->ttl--;
 381                 batadv_send_unicast_skb(skb, neigh_node);
 382                 ret = true;
 383         }
 384
 385 out:
 386         if (orig_node_dst)
 387                 batadv_orig_node_put(orig_node_dst);
 388         if (neigh_node)
 389                 batadv_neigh_node_put(neigh_node);
 390         return ret;
 391 }
 392
 393 /**
 394  * batadv_frag_create - create a fragment from skb
 395  * @skb: skb to create fragment from
 396  * @frag_head: header to use in new fragment
 397  * @mtu: size of new fragment
 398  *
 399  * Split the passed skb into two fragments: A new one with size matching the
 400  * passed mtu and the old one with the rest. The new skb contains data from the
 401  * tail of the old skb.
 402  *
 403  * Return: the new fragment, NULL on error.
 404  */
 405 static struct sk_buff *batadv_frag_create(struct sk_buff *skb,
 406                                           struct batadv_frag_packet *frag_head,
 407                                           unsigned int mtu)
 408 {
 409         struct sk_buff *skb_fragment;
 410         unsigned header_size = sizeof(*frag_head);
 411         unsigned fragment_size = mtu - header_size;
 412
 413         skb_fragment = netdev_alloc_skb(NULL, mtu + ETH_HLEN);
 414         if (!skb_fragment)
 415                 goto err;
 416
 417         skb->priority = TC_PRIO_CONTROL;
 418
 419         /* Eat the last mtu-bytes of the skb */
 420         skb_reserve(skb_fragment, header_size + ETH_HLEN);
 421         skb_split(skb, skb_fragment, skb->len - fragment_size);
 422
 423         /* Add the header */
 424         skb_push(skb_fragment, header_size);
 425         memcpy(skb_fragment->data, frag_head, header_size);
 426
 427 err:
 428         return skb_fragment;
 429 }
 430
 431 /**
 432  * batadv_frag_send_packet - create up to 16 fragments from the passed skb
 433  * @skb: skb to create fragments from
 434  * @orig_node: final destination of the created fragments
 435  * @neigh_node: next-hop of the created fragments
 436  *
 437  * Return: true on success, false otherwise.
 438  */
 439 bool batadv_frag_send_packet(struct sk_buff *skb,
 440                              struct batadv_orig_node *orig_node,
 441                              struct batadv_neigh_node *neigh_node)
 442 {
 443         struct batadv_priv *bat_priv;
 444         struct batadv_hard_iface *primary_if = NULL;
 445         struct batadv_frag_packet frag_header;
 446         struct sk_buff *skb_fragment;
 447         unsigned mtu = neigh_node->if_incoming->net_dev->mtu;
 448         unsigned header_size = sizeof(frag_header);
 449         unsigned max_fragment_size, max_packet_size;
 450         bool ret = false;
 451
 452         /* To avoid merge and refragmentation at next-hops we never send
 453          * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
 454          */
 455         mtu = min_t(unsigned, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
 456         max_fragment_size = mtu - header_size;
 457         max_packet_size = max_fragment_size * BATADV_FRAG_MAX_FRAGMENTS;
 458
 459         /* Don't even try to fragment, if we need more than 16 fragments */
 460         if (skb->len > max_packet_size)
 461                 goto out_err;
 462
 463         bat_priv = orig_node->bat_priv;
 464         primary_if = batadv_primary_if_get_selected(bat_priv);
 465         if (!primary_if)
 466                 goto out_err;
 467
 468         /* Create one header to be copied to all fragments */
 469         frag_header.packet_type = BATADV_UNICAST_FRAG;
 470         frag_header.version = BATADV_COMPAT_VERSION;
 471         frag_header.ttl = BATADV_TTL;
 472         frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
 473         frag_header.reserved = 0;
 474         frag_header.no = 0;
 475         frag_header.total_size = htons(skb->len);
 476         ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr);
 477         ether_addr_copy(frag_header.dest, orig_node->orig);
 478
 479         /* Eat and send fragments from the tail of skb */
 480         while (skb->len > max_fragment_size) {
 481                 skb_fragment = batadv_frag_create(skb, &frag_header, mtu);
 482                 if (!skb_fragment)
 483                         goto out_err;
 484
 485                 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
 486                 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
 487                                    skb_fragment->len + ETH_HLEN);
 488                 batadv_send_unicast_skb(skb_fragment, neigh_node);
 489                 frag_header.no++;
 490
 491                 /* The initial check in this function should cover this case */
 492                 if (frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1)
 493                         goto out_err;
 494         }
 495
 496         /* Make room for the fragment header. */
 497         if (batadv_skb_head_push(skb, header_size) < 0 ||
 498             pskb_expand_head(skb, header_size + ETH_HLEN, 0, GFP_ATOMIC) < 0)
 499                 goto out_err;
 500
 501         memcpy(skb->data, &frag_header, header_size);
 502
 503         /* Send the last fragment */
 504         batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
 505         batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
 506                            skb->len + ETH_HLEN);
 507         batadv_send_unicast_skb(skb, neigh_node);
 508
 509         ret = true;
 510
 511 out_err:
 512         if (primary_if)
 513                 batadv_hardif_put(primary_if);
 514
 515         return ret;
 516 }