1 /* Copyright (C) 2013-2016 B.A.T.M.A.N. contributors:
3 * Martin Hundebøll <martin@hundeboll.net>
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.
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.
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/>.
18 #include "fragmentation.h"
21 #include <linux/atomic.h>
22 #include <linux/byteorder/generic.h>
23 #include <linux/etherdevice.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>
35 #include "hard-interface.h"
36 #include "originator.h"
40 #include "soft-interface.h"
43 * batadv_frag_clear_chain - delete entries in the fragment buffer chain
44 * @head: head of chain with entries.
46 * Free fragments in the passed hlist. Should be called with appropriate lock.
48 static void batadv_frag_clear_chain(struct hlist_head *head)
50 struct batadv_frag_list_entry *entry;
51 struct hlist_node *node;
53 hlist_for_each_entry_safe(entry, node, head, list) {
54 hlist_del(&entry->list);
55 kfree_skb(entry->skb);
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
65 void batadv_frag_purge_orig(struct batadv_orig_node *orig_node,
66 bool (*check_cb)(struct batadv_frag_table_entry *))
68 struct batadv_frag_table_entry *chain;
71 for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) {
72 chain = &orig_node->fragments[i];
73 spin_lock_bh(&chain->lock);
75 if (!check_cb || check_cb(chain)) {
76 batadv_frag_clear_chain(&chain->head);
80 spin_unlock_bh(&chain->lock);
85 * batadv_frag_size_limit - maximum possible size of packet to be fragmented
87 * Return: the maximum size of payload that can be fragmented.
89 static int batadv_frag_size_limit(void)
91 int limit = BATADV_FRAG_MAX_FRAG_SIZE;
93 limit -= sizeof(struct batadv_frag_packet);
94 limit *= BATADV_FRAG_MAX_FRAGMENTS;
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
104 * Make chain ready for a fragment with sequence number "seqno". Delete existing
105 * entries if they have an "old" sequence number.
107 * Caller must hold chain->lock.
109 * Return: true if chain is empty and caller can just insert the new fragment
110 * without searching for the right position.
112 static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain,
115 lockdep_assert_held(&chain->lock);
117 if (chain->seqno == seqno)
120 if (!hlist_empty(&chain->head))
121 batadv_frag_clear_chain(&chain->head);
124 chain->seqno = seqno;
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
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.
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.
142 static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node,
144 struct hlist_head *chain_out)
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;
151 u16 seqno, hdr_size = sizeof(struct batadv_frag_packet);
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
158 if (skb_linearize(skb) < 0)
161 frag_packet = (struct batadv_frag_packet *)skb->data;
162 seqno = ntohs(frag_packet->seqno);
163 bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
165 frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
169 frag_entry_new->skb = skb;
170 frag_entry_new->no = frag_packet->no;
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.
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);
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)
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;
203 /* store current entry because it could be the last in list */
204 frag_entry_last = frag_entry_curr;
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;
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.
223 batadv_frag_clear_chain(&chain->head);
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);
232 spin_unlock_bh(&chain->lock);
236 kfree(frag_entry_new);
244 * batadv_frag_merge_packets - merge a chain of fragments
245 * @chain: head of chain with fragments
247 * Expand the first skb in the chain and copy the content of the remaining
248 * skb's into the expanded one. After doing so, clear the chain.
250 * Return: the merged skb or NULL on error.
252 static struct sk_buff *
253 batadv_frag_merge_packets(struct hlist_head *chain)
255 struct batadv_frag_packet *packet;
256 struct batadv_frag_list_entry *entry;
257 struct sk_buff *skb_out = NULL;
258 int size, hdr_size = sizeof(struct batadv_frag_packet);
260 /* Remove first entry, as this is the destination for the rest of the
263 entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
264 hlist_del(&entry->list);
265 skb_out = entry->skb;
268 packet = (struct batadv_frag_packet *)skb_out->data;
269 size = ntohs(packet->total_size) + hdr_size;
271 /* Make room for the rest of the fragments. */
272 if (pskb_expand_head(skb_out, 0, size - skb_out->len, GFP_ATOMIC) < 0) {
278 /* Move the existing MAC header to just before the payload. (Override
279 * the fragment header.)
281 skb_pull(skb_out, hdr_size);
282 skb_out->ip_summed = CHECKSUM_NONE;
283 memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN);
284 skb_set_mac_header(skb_out, -ETH_HLEN);
285 skb_reset_network_header(skb_out);
286 skb_reset_transport_header(skb_out);
288 /* Copy the payload of the each fragment into the last skb */
289 hlist_for_each_entry(entry, chain, list) {
290 size = entry->skb->len - hdr_size;
291 memcpy(skb_put(skb_out, size), entry->skb->data + hdr_size,
296 /* Locking is not needed, because 'chain' is not part of any orig. */
297 batadv_frag_clear_chain(chain);
302 * batadv_frag_skb_buffer - buffer fragment for later merge
303 * @skb: skb to buffer
304 * @orig_node_src: originator that the skb is received from
306 * Add fragment to buffer and merge fragments if possible.
308 * There are three possible outcomes: 1) Packet is merged: Return true and
309 * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
310 * to NULL; 3) Error: Return false and free skb.
312 * Return: true when packet is merged or buffered, false when skb is not not
315 bool batadv_frag_skb_buffer(struct sk_buff **skb,
316 struct batadv_orig_node *orig_node_src)
318 struct sk_buff *skb_out = NULL;
319 struct hlist_head head = HLIST_HEAD_INIT;
322 /* Add packet to buffer and table entry if merge is possible. */
323 if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
326 /* Leave if more fragments are needed to merge. */
327 if (hlist_empty(&head))
330 skb_out = batadv_frag_merge_packets(&head);
342 * batadv_frag_skb_fwd - forward fragments that would exceed MTU when merged
343 * @skb: skb to forward
344 * @recv_if: interface that the skb is received on
345 * @orig_node_src: originator that the skb is received from
347 * Look up the next-hop of the fragments payload and check if the merged packet
348 * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
349 * without merging it.
351 * Return: true if the fragment is consumed/forwarded, false otherwise.
353 bool batadv_frag_skb_fwd(struct sk_buff *skb,
354 struct batadv_hard_iface *recv_if,
355 struct batadv_orig_node *orig_node_src)
357 struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
358 struct batadv_orig_node *orig_node_dst = NULL;
359 struct batadv_neigh_node *neigh_node = NULL;
360 struct batadv_frag_packet *packet;
364 packet = (struct batadv_frag_packet *)skb->data;
365 orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
369 neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
373 /* Forward the fragment, if the merged packet would be too big to
376 total_size = ntohs(packet->total_size);
377 if (total_size > neigh_node->if_incoming->net_dev->mtu) {
378 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
379 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
380 skb->len + ETH_HLEN);
383 batadv_send_unicast_skb(skb, neigh_node);
389 batadv_orig_node_put(orig_node_dst);
391 batadv_neigh_node_put(neigh_node);
396 * batadv_frag_create - create a fragment from skb
397 * @net_dev: outgoing device for fragment
398 * @skb: skb to create fragment from
399 * @frag_head: header to use in new fragment
400 * @fragment_size: size of new fragment
402 * Split the passed skb into two fragments: A new one with size matching the
403 * passed mtu and the old one with the rest. The new skb contains data from the
404 * tail of the old skb.
406 * Return: the new fragment, NULL on error.
408 static struct sk_buff *batadv_frag_create(struct net_device *net_dev,
410 struct batadv_frag_packet *frag_head,
411 unsigned int fragment_size)
413 unsigned int ll_reserved = LL_RESERVED_SPACE(net_dev);
414 unsigned int tailroom = net_dev->needed_tailroom;
415 struct sk_buff *skb_fragment;
416 unsigned int header_size = sizeof(*frag_head);
417 unsigned int mtu = fragment_size + header_size;
419 skb_fragment = dev_alloc_skb(ll_reserved + mtu + tailroom);
423 skb_fragment->priority = skb->priority;
425 /* Eat the last mtu-bytes of the skb */
426 skb_reserve(skb_fragment, ll_reserved + header_size);
427 skb_split(skb, skb_fragment, skb->len - fragment_size);
430 skb_push(skb_fragment, header_size);
431 memcpy(skb_fragment->data, frag_head, header_size);
438 * batadv_frag_send_packet - create up to 16 fragments from the passed skb
439 * @skb: skb to create fragments from
440 * @orig_node: final destination of the created fragments
441 * @neigh_node: next-hop of the created fragments
443 * Return: the netdev tx status or -1 in case of error.
444 * When -1 is returned the skb is not consumed.
446 int batadv_frag_send_packet(struct sk_buff *skb,
447 struct batadv_orig_node *orig_node,
448 struct batadv_neigh_node *neigh_node)
450 struct net_device *net_dev = neigh_node->if_incoming->net_dev;
451 struct batadv_priv *bat_priv;
452 struct batadv_hard_iface *primary_if = NULL;
453 struct batadv_frag_packet frag_header;
454 struct sk_buff *skb_fragment;
455 unsigned int mtu = net_dev->mtu;
456 unsigned int header_size = sizeof(frag_header);
457 unsigned int max_fragment_size, num_fragments;
460 /* To avoid merge and refragmentation at next-hops we never send
461 * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
463 mtu = min_t(unsigned int, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
464 max_fragment_size = mtu - header_size;
466 if (skb->len == 0 || max_fragment_size == 0)
469 num_fragments = (skb->len - 1) / max_fragment_size + 1;
470 max_fragment_size = (skb->len - 1) / num_fragments + 1;
472 /* Don't even try to fragment, if we need more than 16 fragments */
473 if (num_fragments > BATADV_FRAG_MAX_FRAGMENTS)
476 bat_priv = orig_node->bat_priv;
477 primary_if = batadv_primary_if_get_selected(bat_priv);
481 /* Create one header to be copied to all fragments */
482 frag_header.packet_type = BATADV_UNICAST_FRAG;
483 frag_header.version = BATADV_COMPAT_VERSION;
484 frag_header.ttl = BATADV_TTL;
485 frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
486 frag_header.reserved = 0;
488 frag_header.total_size = htons(skb->len);
490 /* skb->priority values from 256->263 are magic values to
491 * directly indicate a specific 802.1d priority. This is used
492 * to allow 802.1d priority to be passed directly in from VLAN
495 if (skb->priority >= 256 && skb->priority <= 263)
496 frag_header.priority = skb->priority - 256;
498 frag_header.priority = 0;
500 ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr);
501 ether_addr_copy(frag_header.dest, orig_node->orig);
503 /* Eat and send fragments from the tail of skb */
504 while (skb->len > max_fragment_size) {
505 /* The initial check in this function should cover this case */
506 if (frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1) {
511 skb_fragment = batadv_frag_create(net_dev, skb, &frag_header,
516 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
517 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
518 skb_fragment->len + ETH_HLEN);
519 ret = batadv_send_unicast_skb(skb_fragment, neigh_node);
520 if (ret != NET_XMIT_SUCCESS) {
521 /* return -1 so that the caller can free the original
531 /* make sure that there is at least enough head for the fragmentation
532 * and ethernet headers
534 ret = skb_cow_head(skb, ETH_HLEN + header_size);
538 skb_push(skb, header_size);
539 memcpy(skb->data, &frag_header, header_size);
541 /* Send the last fragment */
542 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
543 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
544 skb->len + ETH_HLEN);
545 ret = batadv_send_unicast_skb(skb, neigh_node);
549 batadv_hardif_put(primary_if);