net/batman-adv/fragmentation.c

   1 /* Copyright (C) 2013-2014 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 "main.h"
  19 #include "fragmentation.h"
  20 #include "send.h"
  21 #include "originator.h"
  22 #include "routing.h"
  23 #include "hard-interface.h"
  24 #include "soft-interface.h"
  25
  26
  27 /**
  28  * batadv_frag_clear_chain - delete entries in the fragment buffer chain
  29  * @head: head of chain with entries.
  30  *
  31  * Free fragments in the passed hlist. Should be called with appropriate lock.
  32  */
  33 static void batadv_frag_clear_chain(struct hlist_head *head)
  34 {
  35         struct batadv_frag_list_entry *entry;
  36         struct hlist_node *node;
  37
  38         hlist_for_each_entry_safe(entry, node, head, list) {
  39                 hlist_del(&entry->list);
  40                 kfree_skb(entry->skb);
  41                 kfree(entry);
  42         }
  43 }
  44
  45 /**
  46  * batadv_frag_purge_orig - free fragments associated to an orig
  47  * @orig_node: originator to free fragments from
  48  * @check_cb: optional function to tell if an entry should be purged
  49  */
  50 void batadv_frag_purge_orig(struct batadv_orig_node *orig_node,
  51                             bool (*check_cb)(struct batadv_frag_table_entry *))
  52 {
  53         struct batadv_frag_table_entry *chain;
  54         uint8_t i;
  55
  56         for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) {
  57                 chain = &orig_node->fragments[i];
  58                 spin_lock_bh(&orig_node->fragments[i].lock);
  59
  60                 if (!check_cb || check_cb(chain)) {
  61                         batadv_frag_clear_chain(&orig_node->fragments[i].head);
  62                         orig_node->fragments[i].size = 0;
  63                 }
  64
  65                 spin_unlock_bh(&orig_node->fragments[i].lock);
  66         }
  67 }
  68
  69 /**
  70  * batadv_frag_size_limit - maximum possible size of packet to be fragmented
  71  *
  72  * Returns the maximum size of payload that can be fragmented.
  73  */
  74 static int batadv_frag_size_limit(void)
  75 {
  76         int limit = BATADV_FRAG_MAX_FRAG_SIZE;
  77
  78         limit -= sizeof(struct batadv_frag_packet);
  79         limit *= BATADV_FRAG_MAX_FRAGMENTS;
  80
  81         return limit;
  82 }
  83
  84 /**
  85  * batadv_frag_init_chain - check and prepare fragment chain for new fragment
  86  * @chain: chain in fragments table to init
  87  * @seqno: sequence number of the received fragment
  88  *
  89  * Make chain ready for a fragment with sequence number "seqno". Delete existing
  90  * entries if they have an "old" sequence number.
  91  *
  92  * Caller must hold chain->lock.
  93  *
  94  * Returns true if chain is empty and caller can just insert the new fragment
  95  * without searching for the right position.
  96  */
  97 static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain,
  98                                    uint16_t seqno)
  99 {
 100         if (chain->seqno == seqno)
 101                 return false;
 102
 103         if (!hlist_empty(&chain->head))
 104                 batadv_frag_clear_chain(&chain->head);
 105
 106         chain->size = 0;
 107         chain->seqno = seqno;
 108
 109         return true;
 110 }
 111
 112 /**
 113  * batadv_frag_insert_packet - insert a fragment into a fragment chain
 114  * @orig_node: originator that the fragment was received from
 115  * @skb: skb to insert
 116  * @chain_out: list head to attach complete chains of fragments to
 117  *
 118  * Insert a new fragment into the reverse ordered chain in the right table
 119  * entry. The hash table entry is cleared if "old" fragments exist in it.
 120  *
 121  * Returns true if skb is buffered, false on error. If the chain has all the
 122  * fragments needed to merge the packet, the chain is moved to the passed head
 123  * to avoid locking the chain in the table.
 124  */
 125 static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node,
 126                                       struct sk_buff *skb,
 127                                       struct hlist_head *chain_out)
 128 {
 129         struct batadv_frag_table_entry *chain;
 130         struct batadv_frag_list_entry *frag_entry_new = NULL, *frag_entry_curr;
 131         struct batadv_frag_list_entry *frag_entry_last = NULL;
 132         struct batadv_frag_packet *frag_packet;
 133         uint8_t bucket;
 134         uint16_t seqno, hdr_size = sizeof(struct batadv_frag_packet);
 135         bool ret = false;
 136
 137         /* Linearize packet to avoid linearizing 16 packets in a row when doing
 138          * the later merge. Non-linear merge should be added to remove this
 139          * linearization.
 140          */
 141         if (skb_linearize(skb) < 0)
 142                 goto err;
 143
 144         frag_packet = (struct batadv_frag_packet *)skb->data;
 145         seqno = ntohs(frag_packet->seqno);
 146         bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
 147
 148         frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
 149         if (!frag_entry_new)
 150                 goto err;
 151
 152         frag_entry_new->skb = skb;
 153         frag_entry_new->no = frag_packet->no;
 154
 155         /* Select entry in the "chain table" and delete any prior fragments
 156          * with another sequence number. batadv_frag_init_chain() returns true,
 157          * if the list is empty at return.
 158          */
 159         chain = &orig_node->fragments[bucket];
 160         spin_lock_bh(&chain->lock);
 161         if (batadv_frag_init_chain(chain, seqno)) {
 162                 hlist_add_head(&frag_entry_new->list, &chain->head);
 163                 chain->size = skb->len - hdr_size;
 164                 chain->timestamp = jiffies;
 165                 ret = true;
 166                 goto out;
 167         }
 168
 169         /* Find the position for the new fragment. */
 170         hlist_for_each_entry(frag_entry_curr, &chain->head, list) {
 171                 /* Drop packet if fragment already exists. */
 172                 if (frag_entry_curr->no == frag_entry_new->no)
 173                         goto err_unlock;
 174
 175                 /* Order fragments from highest to lowest. */
 176                 if (frag_entry_curr->no < frag_entry_new->no) {
 177                         hlist_add_before(&frag_entry_new->list,
 178                                          &frag_entry_curr->list);
 179                         chain->size += skb->len - hdr_size;
 180                         chain->timestamp = jiffies;
 181                         ret = true;
 182                         goto out;
 183                 }
 184
 185                 /* store current entry because it could be the last in list */
 186                 frag_entry_last = frag_entry_curr;
 187         }
 188
 189         /* Reached the end of the list, so insert after 'frag_entry_last'. */
 190         if (likely(frag_entry_last)) {
 191                 hlist_add_behind(&frag_entry_last->list, &frag_entry_new->list);
 192                 chain->size += skb->len - hdr_size;
 193                 chain->timestamp = jiffies;
 194                 ret = true;
 195         }
 196
 197 out:
 198         if (chain->size > batadv_frag_size_limit() ||
 199             ntohs(frag_packet->total_size) > batadv_frag_size_limit()) {
 200                 /* Clear chain if total size of either the list or the packet
 201                  * exceeds the maximum size of one merged packet.
 202                  */
 203                 batadv_frag_clear_chain(&chain->head);
 204                 chain->size = 0;
 205         } else if (ntohs(frag_packet->total_size) == chain->size) {
 206                 /* All fragments received. Hand over chain to caller. */
 207                 hlist_move_list(&chain->head, chain_out);
 208                 chain->size = 0;
 209         }
 210
 211 err_unlock:
 212         spin_unlock_bh(&chain->lock);
 213
 214 err:
 215         if (!ret)
 216                 kfree(frag_entry_new);
 217
 218         return ret;
 219 }
 220
 221 /**
 222  * batadv_frag_merge_packets - merge a chain of fragments
 223  * @chain: head of chain with fragments
 224  * @skb: packet with total size of skb after merging
 225  *
 226  * Expand the first skb in the chain and copy the content of the remaining
 227  * skb's into the expanded one. After doing so, clear the chain.
 228  *
 229  * Returns the merged skb or NULL on error.
 230  */
 231 static struct sk_buff *
 232 batadv_frag_merge_packets(struct hlist_head *chain, struct sk_buff *skb)
 233 {
 234         struct batadv_frag_packet *packet;
 235         struct batadv_frag_list_entry *entry;
 236         struct sk_buff *skb_out = NULL;
 237         int size, hdr_size = sizeof(struct batadv_frag_packet);
 238
 239         /* Make sure incoming skb has non-bogus data. */
 240         packet = (struct batadv_frag_packet *)skb->data;
 241         size = ntohs(packet->total_size);
 242         if (size > batadv_frag_size_limit())
 243                 goto free;
 244
 245         /* Remove first entry, as this is the destination for the rest of the
 246          * fragments.
 247          */
 248         entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
 249         hlist_del(&entry->list);
 250         skb_out = entry->skb;
 251         kfree(entry);
 252
 253         /* Make room for the rest of the fragments. */
 254         if (pskb_expand_head(skb_out, 0, size - skb->len, GFP_ATOMIC) < 0) {
 255                 kfree_skb(skb_out);
 256                 skb_out = NULL;
 257                 goto free;
 258         }
 259
 260         /* Move the existing MAC header to just before the payload. (Override
 261          * the fragment header.)
 262          */
 263         skb_pull_rcsum(skb_out, hdr_size);
 264         memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN);
 265         skb_set_mac_header(skb_out, -ETH_HLEN);
 266         skb_reset_network_header(skb_out);
 267         skb_reset_transport_header(skb_out);
 268
 269         /* Copy the payload of the each fragment into the last skb */
 270         hlist_for_each_entry(entry, chain, list) {
 271                 size = entry->skb->len - hdr_size;
 272                 memcpy(skb_put(skb_out, size), entry->skb->data + hdr_size,
 273                        size);
 274         }
 275
 276 free:
 277         /* Locking is not needed, because 'chain' is not part of any orig. */
 278         batadv_frag_clear_chain(chain);
 279         return skb_out;
 280 }
 281
 282 /**
 283  * batadv_frag_skb_buffer - buffer fragment for later merge
 284  * @skb: skb to buffer
 285  * @orig_node_src: originator that the skb is received from
 286  *
 287  * Add fragment to buffer and merge fragments if possible.
 288  *
 289  * There are three possible outcomes: 1) Packet is merged: Return true and
 290  * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
 291  * to NULL; 3) Error: Return false and leave skb as is.
 292  */
 293 bool batadv_frag_skb_buffer(struct sk_buff **skb,
 294                             struct batadv_orig_node *orig_node_src)
 295 {
 296         struct sk_buff *skb_out = NULL;
 297         struct hlist_head head = HLIST_HEAD_INIT;
 298         bool ret = false;
 299
 300         /* Add packet to buffer and table entry if merge is possible. */
 301         if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
 302                 goto out_err;
 303
 304         /* Leave if more fragments are needed to merge. */
 305         if (hlist_empty(&head))
 306                 goto out;
 307
 308         skb_out = batadv_frag_merge_packets(&head, *skb);
 309         if (!skb_out)
 310                 goto out_err;
 311
 312 out:
 313         *skb = skb_out;
 314         ret = true;
 315 out_err:
 316         return ret;
 317 }
 318
 319 /**
 320  * batadv_frag_skb_fwd - forward fragments that would exceed MTU when merged
 321  * @skb: skb to forward
 322  * @recv_if: interface that the skb is received on
 323  * @orig_node_src: originator that the skb is received from
 324  *
 325  * Look up the next-hop of the fragments payload and check if the merged packet
 326  * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
 327  * without merging it.
 328  *
 329  * Returns true if the fragment is consumed/forwarded, false otherwise.
 330  */
 331 bool batadv_frag_skb_fwd(struct sk_buff *skb,
 332                          struct batadv_hard_iface *recv_if,
 333                          struct batadv_orig_node *orig_node_src)
 334 {
 335         struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
 336         struct batadv_orig_node *orig_node_dst = NULL;
 337         struct batadv_neigh_node *neigh_node = NULL;
 338         struct batadv_frag_packet *packet;
 339         uint16_t total_size;
 340         bool ret = false;
 341
 342         packet = (struct batadv_frag_packet *)skb->data;
 343         orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
 344         if (!orig_node_dst)
 345                 goto out;
 346
 347         neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
 348         if (!neigh_node)
 349                 goto out;
 350
 351         /* Forward the fragment, if the merged packet would be too big to
 352          * be assembled.
 353          */
 354         total_size = ntohs(packet->total_size);
 355         if (total_size > neigh_node->if_incoming->net_dev->mtu) {
 356                 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
 357                 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
 358                                    skb->len + ETH_HLEN);
 359
 360                 packet->ttl--;
 361                 batadv_send_skb_packet(skb, neigh_node->if_incoming,
 362                                        neigh_node->addr);
 363                 ret = true;
 364         }
 365
 366 out:
 367         if (orig_node_dst)
 368                 batadv_orig_node_free_ref(orig_node_dst);
 369         if (neigh_node)
 370                 batadv_neigh_node_free_ref(neigh_node);
 371         return ret;
 372 }
 373
 374 /**
 375  * batadv_frag_create - create a fragment from skb
 376  * @skb: skb to create fragment from
 377  * @frag_head: header to use in new fragment
 378  * @mtu: size of new fragment
 379  *
 380  * Split the passed skb into two fragments: A new one with size matching the
 381  * passed mtu and the old one with the rest. The new skb contains data from the
 382  * tail of the old skb.
 383  *
 384  * Returns the new fragment, NULL on error.
 385  */
 386 static struct sk_buff *batadv_frag_create(struct sk_buff *skb,
 387                                           struct batadv_frag_packet *frag_head,
 388                                           unsigned int mtu)
 389 {
 390         struct sk_buff *skb_fragment;
 391         unsigned header_size = sizeof(*frag_head);
 392         unsigned fragment_size = mtu - header_size;
 393
 394         skb_fragment = netdev_alloc_skb(NULL, mtu + ETH_HLEN);
 395         if (!skb_fragment)
 396                 goto err;
 397
 398         skb->priority = TC_PRIO_CONTROL;
 399
 400         /* Eat the last mtu-bytes of the skb */
 401         skb_reserve(skb_fragment, header_size + ETH_HLEN);
 402         skb_split(skb, skb_fragment, skb->len - fragment_size);
 403
 404         /* Add the header */
 405         skb_push(skb_fragment, header_size);
 406         memcpy(skb_fragment->data, frag_head, header_size);
 407
 408 err:
 409         return skb_fragment;
 410 }
 411
 412 /**
 413  * batadv_frag_send_packet - create up to 16 fragments from the passed skb
 414  * @skb: skb to create fragments from
 415  * @orig_node: final destination of the created fragments
 416  * @neigh_node: next-hop of the created fragments
 417  *
 418  * Returns true on success, false otherwise.
 419  */
 420 bool batadv_frag_send_packet(struct sk_buff *skb,
 421                              struct batadv_orig_node *orig_node,
 422                              struct batadv_neigh_node *neigh_node)
 423 {
 424         struct batadv_priv *bat_priv;
 425         struct batadv_hard_iface *primary_if = NULL;
 426         struct batadv_frag_packet frag_header;
 427         struct sk_buff *skb_fragment;
 428         unsigned mtu = neigh_node->if_incoming->net_dev->mtu;
 429         unsigned header_size = sizeof(frag_header);
 430         unsigned max_fragment_size, max_packet_size;
 431         bool ret = false;
 432
 433         /* To avoid merge and refragmentation at next-hops we never send
 434          * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
 435          */
 436         mtu = min_t(unsigned, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
 437         max_fragment_size = (mtu - header_size - ETH_HLEN);
 438         max_packet_size = max_fragment_size * BATADV_FRAG_MAX_FRAGMENTS;
 439
 440         /* Don't even try to fragment, if we need more than 16 fragments */
 441         if (skb->len > max_packet_size)
 442                 goto out_err;
 443
 444         bat_priv = orig_node->bat_priv;
 445         primary_if = batadv_primary_if_get_selected(bat_priv);
 446         if (!primary_if)
 447                 goto out_err;
 448
 449         /* Create one header to be copied to all fragments */
 450         frag_header.packet_type = BATADV_UNICAST_FRAG;
 451         frag_header.version = BATADV_COMPAT_VERSION;
 452         frag_header.ttl = BATADV_TTL;
 453         frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
 454         frag_header.reserved = 0;
 455         frag_header.no = 0;
 456         frag_header.total_size = htons(skb->len);
 457         ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr);
 458         ether_addr_copy(frag_header.dest, orig_node->orig);
 459
 460         /* Eat and send fragments from the tail of skb */
 461         while (skb->len > max_fragment_size) {
 462                 skb_fragment = batadv_frag_create(skb, &frag_header, mtu);
 463                 if (!skb_fragment)
 464                         goto out_err;
 465
 466                 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
 467                 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
 468                                    skb_fragment->len + ETH_HLEN);
 469                 batadv_send_skb_packet(skb_fragment, neigh_node->if_incoming,
 470                                        neigh_node->addr);
 471                 frag_header.no++;
 472
 473                 /* The initial check in this function should cover this case */
 474                 if (frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1)
 475                         goto out_err;
 476         }
 477
 478         /* Make room for the fragment header. */
 479         if (batadv_skb_head_push(skb, header_size) < 0 ||
 480             pskb_expand_head(skb, header_size + ETH_HLEN, 0, GFP_ATOMIC) < 0)
 481                 goto out_err;
 482
 483         memcpy(skb->data, &frag_header, header_size);
 484
 485         /* Send the last fragment */
 486         batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
 487         batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
 488                            skb->len + ETH_HLEN);
 489         batadv_send_skb_packet(skb, neigh_node->if_incoming, neigh_node->addr);
 490
 491         ret = true;
 492
 493 out_err:
 494         if (primary_if)
 495                 batadv_hardif_free_ref(primary_if);
 496
 497         return ret;
 498 }