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/skbuff.h> 31 #include <linux/slab.h> 32 #include <linux/spinlock.h> 33 #include <linux/string.h> 34 35 #include "hard-interface.h" 36 #include "originator.h" 37 #include "packet.h" 38 #include "routing.h" 39 #include "send.h" 40 #include "soft-interface.h" 41 42 /** 43 * batadv_frag_clear_chain - delete entries in the fragment buffer chain 44 * @head: head of chain with entries. 45 * 46 * Free fragments in the passed hlist. Should be called with appropriate lock. 47 */ 48 static void batadv_frag_clear_chain(struct hlist_head *head) 49 { 50 struct batadv_frag_list_entry *entry; 51 struct hlist_node *node; 52 53 hlist_for_each_entry_safe(entry, node, head, list) { 54 hlist_del(&entry->list); 55 kfree_skb(entry->skb); 56 kfree(entry); 57 } 58 } 59 60 /** 61 * batadv_frag_purge_orig - free fragments associated to an orig 62 * @orig_node: originator to free fragments from 63 * @check_cb: optional function to tell if an entry should be purged 64 */ 65 void batadv_frag_purge_orig(struct batadv_orig_node *orig_node, 66 bool (*check_cb)(struct batadv_frag_table_entry *)) 67 { 68 struct batadv_frag_table_entry *chain; 69 u8 i; 70 71 for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) { 72 chain = &orig_node->fragments[i]; 73 spin_lock_bh(&chain->lock); 74 75 if (!check_cb || check_cb(chain)) { 76 batadv_frag_clear_chain(&chain->head); 77 chain->size = 0; 78 } 79 80 spin_unlock_bh(&chain->lock); 81 } 82 } 83 84 /** 85 * batadv_frag_size_limit - maximum possible size of packet to be fragmented 86 * 87 * Return: the maximum size of payload that can be fragmented. 88 */ 89 static int batadv_frag_size_limit(void) 90 { 91 int limit = BATADV_FRAG_MAX_FRAG_SIZE; 92 93 limit -= sizeof(struct batadv_frag_packet); 94 limit *= BATADV_FRAG_MAX_FRAGMENTS; 95 96 return limit; 97 } 98 99 /** 100 * batadv_frag_init_chain - check and prepare fragment chain for new fragment 101 * @chain: chain in fragments table to init 102 * @seqno: sequence number of the received fragment 103 * 104 * Make chain ready for a fragment with sequence number "seqno". Delete existing 105 * entries if they have an "old" sequence number. 106 * 107 * Caller must hold chain->lock. 108 * 109 * Return: true if chain is empty and caller can just insert the new fragment 110 * without searching for the right position. 111 */ 112 static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain, 113 u16 seqno) 114 { 115 lockdep_assert_held(&chain->lock); 116 117 if (chain->seqno == seqno) 118 return false; 119 120 if (!hlist_empty(&chain->head)) 121 batadv_frag_clear_chain(&chain->head); 122 123 chain->size = 0; 124 chain->seqno = seqno; 125 126 return true; 127 } 128 129 /** 130 * batadv_frag_insert_packet - insert a fragment into a fragment chain 131 * @orig_node: originator that the fragment was received from 132 * @skb: skb to insert 133 * @chain_out: list head to attach complete chains of fragments to 134 * 135 * Insert a new fragment into the reverse ordered chain in the right table 136 * entry. The hash table entry is cleared if "old" fragments exist in it. 137 * 138 * Return: true if skb is buffered, false on error. If the chain has all the 139 * fragments needed to merge the packet, the chain is moved to the passed head 140 * to avoid locking the chain in the table. 141 */ 142 static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node, 143 struct sk_buff *skb, 144 struct hlist_head *chain_out) 145 { 146 struct batadv_frag_table_entry *chain; 147 struct batadv_frag_list_entry *frag_entry_new = NULL, *frag_entry_curr; 148 struct batadv_frag_list_entry *frag_entry_last = NULL; 149 struct batadv_frag_packet *frag_packet; 150 u8 bucket; 151 u16 seqno, hdr_size = sizeof(struct batadv_frag_packet); 152 bool ret = false; 153 154 /* Linearize packet to avoid linearizing 16 packets in a row when doing 155 * the later merge. Non-linear merge should be added to remove this 156 * linearization. 157 */ 158 if (skb_linearize(skb) < 0) 159 goto err; 160 161 frag_packet = (struct batadv_frag_packet *)skb->data; 162 seqno = ntohs(frag_packet->seqno); 163 bucket = seqno % BATADV_FRAG_BUFFER_COUNT; 164 165 frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC); 166 if (!frag_entry_new) 167 goto err; 168 169 frag_entry_new->skb = skb; 170 frag_entry_new->no = frag_packet->no; 171 172 /* Select entry in the "chain table" and delete any prior fragments 173 * with another sequence number. batadv_frag_init_chain() returns true, 174 * if the list is empty at return. 175 */ 176 chain = &orig_node->fragments[bucket]; 177 spin_lock_bh(&chain->lock); 178 if (batadv_frag_init_chain(chain, seqno)) { 179 hlist_add_head(&frag_entry_new->list, &chain->head); 180 chain->size = skb->len - hdr_size; 181 chain->timestamp = jiffies; 182 chain->total_size = ntohs(frag_packet->total_size); 183 ret = true; 184 goto out; 185 } 186 187 /* Find the position for the new fragment. */ 188 hlist_for_each_entry(frag_entry_curr, &chain->head, list) { 189 /* Drop packet if fragment already exists. */ 190 if (frag_entry_curr->no == frag_entry_new->no) 191 goto err_unlock; 192 193 /* Order fragments from highest to lowest. */ 194 if (frag_entry_curr->no < frag_entry_new->no) { 195 hlist_add_before(&frag_entry_new->list, 196 &frag_entry_curr->list); 197 chain->size += skb->len - hdr_size; 198 chain->timestamp = jiffies; 199 ret = true; 200 goto out; 201 } 202 203 /* store current entry because it could be the last in list */ 204 frag_entry_last = frag_entry_curr; 205 } 206 207 /* Reached the end of the list, so insert after 'frag_entry_last'. */ 208 if (likely(frag_entry_last)) { 209 hlist_add_behind(&frag_entry_new->list, &frag_entry_last->list); 210 chain->size += skb->len - hdr_size; 211 chain->timestamp = jiffies; 212 ret = true; 213 } 214 215 out: 216 if (chain->size > batadv_frag_size_limit() || 217 chain->total_size != ntohs(frag_packet->total_size) || 218 chain->total_size > batadv_frag_size_limit()) { 219 /* Clear chain if total size of either the list or the packet 220 * exceeds the maximum size of one merged packet. Don't allow 221 * packets to have different total_size. 222 */ 223 batadv_frag_clear_chain(&chain->head); 224 chain->size = 0; 225 } else if (ntohs(frag_packet->total_size) == chain->size) { 226 /* All fragments received. Hand over chain to caller. */ 227 hlist_move_list(&chain->head, chain_out); 228 chain->size = 0; 229 } 230 231 err_unlock: 232 spin_unlock_bh(&chain->lock); 233 234 err: 235 if (!ret) 236 kfree(frag_entry_new); 237 238 return ret; 239 } 240 241 /** 242 * batadv_frag_merge_packets - merge a chain of fragments 243 * @chain: head of chain with fragments 244 * 245 * Expand the first skb in the chain and copy the content of the remaining 246 * skb's into the expanded one. After doing so, clear the chain. 247 * 248 * Return: the merged skb or NULL on error. 249 */ 250 static struct sk_buff * 251 batadv_frag_merge_packets(struct hlist_head *chain) 252 { 253 struct batadv_frag_packet *packet; 254 struct batadv_frag_list_entry *entry; 255 struct sk_buff *skb_out = NULL; 256 int size, hdr_size = sizeof(struct batadv_frag_packet); 257 258 /* Remove first entry, as this is the destination for the rest of the 259 * fragments. 260 */ 261 entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list); 262 hlist_del(&entry->list); 263 skb_out = entry->skb; 264 kfree(entry); 265 266 packet = (struct batadv_frag_packet *)skb_out->data; 267 size = ntohs(packet->total_size); 268 269 /* Make room for the rest of the fragments. */ 270 if (pskb_expand_head(skb_out, 0, size - skb_out->len, GFP_ATOMIC) < 0) { 271 kfree_skb(skb_out); 272 skb_out = NULL; 273 goto free; 274 } 275 276 /* Move the existing MAC header to just before the payload. (Override 277 * the fragment header.) 278 */ 279 skb_pull_rcsum(skb_out, hdr_size); 280 memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN); 281 skb_set_mac_header(skb_out, -ETH_HLEN); 282 skb_reset_network_header(skb_out); 283 skb_reset_transport_header(skb_out); 284 285 /* Copy the payload of the each fragment into the last skb */ 286 hlist_for_each_entry(entry, chain, list) { 287 size = entry->skb->len - hdr_size; 288 memcpy(skb_put(skb_out, size), entry->skb->data + hdr_size, 289 size); 290 } 291 292 free: 293 /* Locking is not needed, because 'chain' is not part of any orig. */ 294 batadv_frag_clear_chain(chain); 295 return skb_out; 296 } 297 298 /** 299 * batadv_frag_skb_buffer - buffer fragment for later merge 300 * @skb: skb to buffer 301 * @orig_node_src: originator that the skb is received from 302 * 303 * Add fragment to buffer and merge fragments if possible. 304 * 305 * There are three possible outcomes: 1) Packet is merged: Return true and 306 * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb 307 * to NULL; 3) Error: Return false and leave skb as is. 308 * 309 * Return: true when packet is merged or buffered, false when skb is not not 310 * used. 311 */ 312 bool batadv_frag_skb_buffer(struct sk_buff **skb, 313 struct batadv_orig_node *orig_node_src) 314 { 315 struct sk_buff *skb_out = NULL; 316 struct hlist_head head = HLIST_HEAD_INIT; 317 bool ret = false; 318 319 /* Add packet to buffer and table entry if merge is possible. */ 320 if (!batadv_frag_insert_packet(orig_node_src, *skb, &head)) 321 goto out_err; 322 323 /* Leave if more fragments are needed to merge. */ 324 if (hlist_empty(&head)) 325 goto out; 326 327 skb_out = batadv_frag_merge_packets(&head); 328 if (!skb_out) 329 goto out_err; 330 331 out: 332 *skb = skb_out; 333 ret = true; 334 out_err: 335 return ret; 336 } 337 338 /** 339 * batadv_frag_skb_fwd - forward fragments that would exceed MTU when merged 340 * @skb: skb to forward 341 * @recv_if: interface that the skb is received on 342 * @orig_node_src: originator that the skb is received from 343 * 344 * Look up the next-hop of the fragments payload and check if the merged packet 345 * will exceed the MTU towards the next-hop. If so, the fragment is forwarded 346 * without merging it. 347 * 348 * Return: true if the fragment is consumed/forwarded, false otherwise. 349 */ 350 bool batadv_frag_skb_fwd(struct sk_buff *skb, 351 struct batadv_hard_iface *recv_if, 352 struct batadv_orig_node *orig_node_src) 353 { 354 struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface); 355 struct batadv_orig_node *orig_node_dst = NULL; 356 struct batadv_neigh_node *neigh_node = NULL; 357 struct batadv_frag_packet *packet; 358 u16 total_size; 359 bool ret = false; 360 361 packet = (struct batadv_frag_packet *)skb->data; 362 orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest); 363 if (!orig_node_dst) 364 goto out; 365 366 neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if); 367 if (!neigh_node) 368 goto out; 369 370 /* Forward the fragment, if the merged packet would be too big to 371 * be assembled. 372 */ 373 total_size = ntohs(packet->total_size); 374 if (total_size > neigh_node->if_incoming->net_dev->mtu) { 375 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD); 376 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES, 377 skb->len + ETH_HLEN); 378 379 packet->ttl--; 380 batadv_send_unicast_skb(skb, neigh_node); 381 ret = true; 382 } 383 384 out: 385 if (orig_node_dst) 386 batadv_orig_node_put(orig_node_dst); 387 if (neigh_node) 388 batadv_neigh_node_put(neigh_node); 389 return ret; 390 } 391 392 /** 393 * batadv_frag_create - create a fragment from skb 394 * @skb: skb to create fragment from 395 * @frag_head: header to use in new fragment 396 * @mtu: size of new fragment 397 * 398 * Split the passed skb into two fragments: A new one with size matching the 399 * passed mtu and the old one with the rest. The new skb contains data from the 400 * tail of the old skb. 401 * 402 * Return: the new fragment, NULL on error. 403 */ 404 static struct sk_buff *batadv_frag_create(struct sk_buff *skb, 405 struct batadv_frag_packet *frag_head, 406 unsigned int mtu) 407 { 408 struct sk_buff *skb_fragment; 409 unsigned int header_size = sizeof(*frag_head); 410 unsigned int fragment_size = mtu - header_size; 411 412 skb_fragment = netdev_alloc_skb(NULL, mtu + ETH_HLEN); 413 if (!skb_fragment) 414 goto err; 415 416 skb_fragment->priority = skb->priority; 417 418 /* Eat the last mtu-bytes of the skb */ 419 skb_reserve(skb_fragment, header_size + ETH_HLEN); 420 skb_split(skb, skb_fragment, skb->len - fragment_size); 421 422 /* Add the header */ 423 skb_push(skb_fragment, header_size); 424 memcpy(skb_fragment->data, frag_head, header_size); 425 426 err: 427 return skb_fragment; 428 } 429 430 /** 431 * batadv_frag_send_packet - create up to 16 fragments from the passed skb 432 * @skb: skb to create fragments from 433 * @orig_node: final destination of the created fragments 434 * @neigh_node: next-hop of the created fragments 435 * 436 * Return: the netdev tx status or -1 in case of error. 437 * When -1 is returned the skb is not consumed. 438 */ 439 int 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 int mtu = neigh_node->if_incoming->net_dev->mtu; 448 unsigned int header_size = sizeof(frag_header); 449 unsigned int max_fragment_size, max_packet_size; 450 int ret = -1; 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 int, 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; 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; 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); 476 477 /* skb->priority values from 256->263 are magic values to 478 * directly indicate a specific 802.1d priority. This is used 479 * to allow 802.1d priority to be passed directly in from VLAN 480 * tags, etc. 481 */ 482 if (skb->priority >= 256 && skb->priority <= 263) 483 frag_header.priority = skb->priority - 256; 484 485 ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr); 486 ether_addr_copy(frag_header.dest, orig_node->orig); 487 488 /* Eat and send fragments from the tail of skb */ 489 while (skb->len > max_fragment_size) { 490 skb_fragment = batadv_frag_create(skb, &frag_header, mtu); 491 if (!skb_fragment) 492 goto out; 493 494 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX); 495 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES, 496 skb_fragment->len + ETH_HLEN); 497 ret = batadv_send_unicast_skb(skb_fragment, neigh_node); 498 if (ret != NET_XMIT_SUCCESS) { 499 /* return -1 so that the caller can free the original 500 * skb 501 */ 502 ret = -1; 503 goto out; 504 } 505 506 frag_header.no++; 507 508 /* The initial check in this function should cover this case */ 509 if (frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1) { 510 ret = -1; 511 goto out; 512 } 513 } 514 515 /* Make room for the fragment header. */ 516 if (batadv_skb_head_push(skb, header_size) < 0 || 517 pskb_expand_head(skb, header_size + ETH_HLEN, 0, GFP_ATOMIC) < 0) 518 goto out; 519 520 memcpy(skb->data, &frag_header, header_size); 521 522 /* Send the last fragment */ 523 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX); 524 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES, 525 skb->len + ETH_HLEN); 526 ret = batadv_send_unicast_skb(skb, neigh_node); 527 528 out: 529 if (primary_if) 530 batadv_hardif_put(primary_if); 531 532 return ret; 533 } 534