blob: 43185ca9da170ca3e2ac6153d36bbf25594aceb9 [file] [log] [blame]
Martin Hundebøll610bfc6bc2013-05-23 16:53:02 +02001/* Copyright (C) 2013 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
Antonio Quartulliebf38fb2013-11-03 20:40:48 +010015 * along with this program; if not, see <http://www.gnu.org/licenses/>.
Martin Hundebøll610bfc6bc2013-05-23 16:53:02 +020016 */
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 */
33static 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 */
50void 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 */
74static 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 */
97static 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 */
125static 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_packet *frag_packet;
132 uint8_t bucket;
133 uint16_t seqno, hdr_size = sizeof(struct batadv_frag_packet);
134 bool ret = false;
135
136 /* Linearize packet to avoid linearizing 16 packets in a row when doing
137 * the later merge. Non-linear merge should be added to remove this
138 * linearization.
139 */
140 if (skb_linearize(skb) < 0)
141 goto err;
142
143 frag_packet = (struct batadv_frag_packet *)skb->data;
144 seqno = ntohs(frag_packet->seqno);
145 bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
146
147 frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
148 if (!frag_entry_new)
149 goto err;
150
151 frag_entry_new->skb = skb;
152 frag_entry_new->no = frag_packet->no;
153
154 /* Select entry in the "chain table" and delete any prior fragments
155 * with another sequence number. batadv_frag_init_chain() returns true,
156 * if the list is empty at return.
157 */
158 chain = &orig_node->fragments[bucket];
159 spin_lock_bh(&chain->lock);
160 if (batadv_frag_init_chain(chain, seqno)) {
161 hlist_add_head(&frag_entry_new->list, &chain->head);
162 chain->size = skb->len - hdr_size;
163 chain->timestamp = jiffies;
164 ret = true;
165 goto out;
166 }
167
168 /* Find the position for the new fragment. */
169 hlist_for_each_entry(frag_entry_curr, &chain->head, list) {
170 /* Drop packet if fragment already exists. */
171 if (frag_entry_curr->no == frag_entry_new->no)
172 goto err_unlock;
173
174 /* Order fragments from highest to lowest. */
175 if (frag_entry_curr->no < frag_entry_new->no) {
176 hlist_add_before(&frag_entry_new->list,
177 &frag_entry_curr->list);
178 chain->size += skb->len - hdr_size;
179 chain->timestamp = jiffies;
180 ret = true;
181 goto out;
182 }
183 }
184
185 /* Reached the end of the list, so insert after 'frag_entry_curr'. */
186 if (likely(frag_entry_curr)) {
187 hlist_add_after(&frag_entry_curr->list, &frag_entry_new->list);
188 chain->size += skb->len - hdr_size;
189 chain->timestamp = jiffies;
190 ret = true;
191 }
192
193out:
194 if (chain->size > batadv_frag_size_limit() ||
195 ntohs(frag_packet->total_size) > batadv_frag_size_limit()) {
196 /* Clear chain if total size of either the list or the packet
197 * exceeds the maximum size of one merged packet.
198 */
199 batadv_frag_clear_chain(&chain->head);
200 chain->size = 0;
201 } else if (ntohs(frag_packet->total_size) == chain->size) {
202 /* All fragments received. Hand over chain to caller. */
203 hlist_move_list(&chain->head, chain_out);
204 chain->size = 0;
205 }
206
207err_unlock:
208 spin_unlock_bh(&chain->lock);
209
210err:
211 if (!ret)
212 kfree(frag_entry_new);
213
214 return ret;
215}
216
217/**
218 * batadv_frag_merge_packets - merge a chain of fragments
219 * @chain: head of chain with fragments
220 * @skb: packet with total size of skb after merging
221 *
222 * Expand the first skb in the chain and copy the content of the remaining
223 * skb's into the expanded one. After doing so, clear the chain.
224 *
225 * Returns the merged skb or NULL on error.
226 */
227static struct sk_buff *
228batadv_frag_merge_packets(struct hlist_head *chain, struct sk_buff *skb)
229{
230 struct batadv_frag_packet *packet;
231 struct batadv_frag_list_entry *entry;
232 struct sk_buff *skb_out = NULL;
233 int size, hdr_size = sizeof(struct batadv_frag_packet);
234
235 /* Make sure incoming skb has non-bogus data. */
236 packet = (struct batadv_frag_packet *)skb->data;
237 size = ntohs(packet->total_size);
238 if (size > batadv_frag_size_limit())
239 goto free;
240
241 /* Remove first entry, as this is the destination for the rest of the
242 * fragments.
243 */
244 entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
245 hlist_del(&entry->list);
246 skb_out = entry->skb;
247 kfree(entry);
248
249 /* Make room for the rest of the fragments. */
250 if (pskb_expand_head(skb_out, 0, size - skb->len, GFP_ATOMIC) < 0) {
251 kfree_skb(skb_out);
252 skb_out = NULL;
253 goto free;
254 }
255
256 /* Move the existing MAC header to just before the payload. (Override
257 * the fragment header.)
258 */
259 skb_pull_rcsum(skb_out, hdr_size);
260 memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN);
261 skb_set_mac_header(skb_out, -ETH_HLEN);
262 skb_reset_network_header(skb_out);
263 skb_reset_transport_header(skb_out);
264
265 /* Copy the payload of the each fragment into the last skb */
266 hlist_for_each_entry(entry, chain, list) {
267 size = entry->skb->len - hdr_size;
268 memcpy(skb_put(skb_out, size), entry->skb->data + hdr_size,
269 size);
270 }
271
272free:
273 /* Locking is not needed, because 'chain' is not part of any orig. */
274 batadv_frag_clear_chain(chain);
275 return skb_out;
276}
277
278/**
279 * batadv_frag_skb_buffer - buffer fragment for later merge
280 * @skb: skb to buffer
281 * @orig_node_src: originator that the skb is received from
282 *
283 * Add fragment to buffer and merge fragments if possible.
284 *
285 * There are three possible outcomes: 1) Packet is merged: Return true and
286 * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
287 * to NULL; 3) Error: Return false and leave skb as is.
288 */
289bool batadv_frag_skb_buffer(struct sk_buff **skb,
290 struct batadv_orig_node *orig_node_src)
291{
292 struct sk_buff *skb_out = NULL;
293 struct hlist_head head = HLIST_HEAD_INIT;
294 bool ret = false;
295
296 /* Add packet to buffer and table entry if merge is possible. */
297 if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
298 goto out_err;
299
300 /* Leave if more fragments are needed to merge. */
301 if (hlist_empty(&head))
302 goto out;
303
304 skb_out = batadv_frag_merge_packets(&head, *skb);
305 if (!skb_out)
306 goto out_err;
307
308out:
309 *skb = skb_out;
310 ret = true;
311out_err:
312 return ret;
313}
314
315/**
316 * batadv_frag_skb_fwd - forward fragments that would exceed MTU when merged
317 * @skb: skb to forward
318 * @recv_if: interface that the skb is received on
319 * @orig_node_src: originator that the skb is received from
320 *
321 * Look up the next-hop of the fragments payload and check if the merged packet
322 * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
323 * without merging it.
324 *
325 * Returns true if the fragment is consumed/forwarded, false otherwise.
326 */
327bool batadv_frag_skb_fwd(struct sk_buff *skb,
328 struct batadv_hard_iface *recv_if,
329 struct batadv_orig_node *orig_node_src)
330{
331 struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
332 struct batadv_orig_node *orig_node_dst = NULL;
333 struct batadv_neigh_node *neigh_node = NULL;
334 struct batadv_frag_packet *packet;
335 uint16_t total_size;
336 bool ret = false;
337
338 packet = (struct batadv_frag_packet *)skb->data;
339 orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
340 if (!orig_node_dst)
341 goto out;
342
343 neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
344 if (!neigh_node)
345 goto out;
346
347 /* Forward the fragment, if the merged packet would be too big to
348 * be assembled.
349 */
350 total_size = ntohs(packet->total_size);
351 if (total_size > neigh_node->if_incoming->net_dev->mtu) {
352 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
353 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
354 skb->len + ETH_HLEN);
355
Simon Wunderlicha40d9b02013-12-02 20:38:31 +0100356 packet->ttl--;
Martin Hundebøll610bfc6bc2013-05-23 16:53:02 +0200357 batadv_send_skb_packet(skb, neigh_node->if_incoming,
358 neigh_node->addr);
359 ret = true;
360 }
361
362out:
363 if (orig_node_dst)
364 batadv_orig_node_free_ref(orig_node_dst);
365 if (neigh_node)
366 batadv_neigh_node_free_ref(neigh_node);
367 return ret;
368}
Martin Hundebøllee75ed82013-05-23 16:53:03 +0200369
370/**
371 * batadv_frag_create - create a fragment from skb
372 * @skb: skb to create fragment from
373 * @frag_head: header to use in new fragment
374 * @mtu: size of new fragment
375 *
376 * Split the passed skb into two fragments: A new one with size matching the
377 * passed mtu and the old one with the rest. The new skb contains data from the
378 * tail of the old skb.
379 *
380 * Returns the new fragment, NULL on error.
381 */
382static struct sk_buff *batadv_frag_create(struct sk_buff *skb,
383 struct batadv_frag_packet *frag_head,
384 unsigned int mtu)
385{
386 struct sk_buff *skb_fragment;
387 unsigned header_size = sizeof(*frag_head);
388 unsigned fragment_size = mtu - header_size;
389
390 skb_fragment = netdev_alloc_skb(NULL, mtu + ETH_HLEN);
391 if (!skb_fragment)
392 goto err;
393
394 skb->priority = TC_PRIO_CONTROL;
395
396 /* Eat the last mtu-bytes of the skb */
397 skb_reserve(skb_fragment, header_size + ETH_HLEN);
398 skb_split(skb, skb_fragment, skb->len - fragment_size);
399
400 /* Add the header */
401 skb_push(skb_fragment, header_size);
402 memcpy(skb_fragment->data, frag_head, header_size);
403
404err:
405 return skb_fragment;
406}
407
408/**
409 * batadv_frag_send_packet - create up to 16 fragments from the passed skb
410 * @skb: skb to create fragments from
411 * @orig_node: final destination of the created fragments
412 * @neigh_node: next-hop of the created fragments
413 *
414 * Returns true on success, false otherwise.
415 */
416bool batadv_frag_send_packet(struct sk_buff *skb,
417 struct batadv_orig_node *orig_node,
418 struct batadv_neigh_node *neigh_node)
419{
420 struct batadv_priv *bat_priv;
421 struct batadv_hard_iface *primary_if;
422 struct batadv_frag_packet frag_header;
423 struct sk_buff *skb_fragment;
424 unsigned mtu = neigh_node->if_incoming->net_dev->mtu;
425 unsigned header_size = sizeof(frag_header);
426 unsigned max_fragment_size, max_packet_size;
427
428 /* To avoid merge and refragmentation at next-hops we never send
429 * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
430 */
431 mtu = min_t(unsigned, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
432 max_fragment_size = (mtu - header_size - ETH_HLEN);
433 max_packet_size = max_fragment_size * BATADV_FRAG_MAX_FRAGMENTS;
434
435 /* Don't even try to fragment, if we need more than 16 fragments */
436 if (skb->len > max_packet_size)
437 goto out_err;
438
439 bat_priv = orig_node->bat_priv;
440 primary_if = batadv_primary_if_get_selected(bat_priv);
441 if (!primary_if)
442 goto out_err;
443
444 /* Create one header to be copied to all fragments */
Simon Wunderlicha40d9b02013-12-02 20:38:31 +0100445 frag_header.packet_type = BATADV_UNICAST_FRAG;
446 frag_header.version = BATADV_COMPAT_VERSION;
447 frag_header.ttl = BATADV_TTL;
Martin Hundebøllee75ed82013-05-23 16:53:03 +0200448 frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
449 frag_header.reserved = 0;
450 frag_header.no = 0;
451 frag_header.total_size = htons(skb->len);
452 memcpy(frag_header.orig, primary_if->net_dev->dev_addr, ETH_ALEN);
453 memcpy(frag_header.dest, orig_node->orig, ETH_ALEN);
454
455 /* Eat and send fragments from the tail of skb */
456 while (skb->len > max_fragment_size) {
457 skb_fragment = batadv_frag_create(skb, &frag_header, mtu);
458 if (!skb_fragment)
459 goto out_err;
460
461 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
462 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
463 skb_fragment->len + ETH_HLEN);
464 batadv_send_skb_packet(skb_fragment, neigh_node->if_incoming,
465 neigh_node->addr);
466 frag_header.no++;
467
468 /* The initial check in this function should cover this case */
469 if (frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1)
470 goto out_err;
471 }
472
473 /* Make room for the fragment header. */
474 if (batadv_skb_head_push(skb, header_size) < 0 ||
475 pskb_expand_head(skb, header_size + ETH_HLEN, 0, GFP_ATOMIC) < 0)
476 goto out_err;
477
478 memcpy(skb->data, &frag_header, header_size);
479
480 /* Send the last fragment */
481 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
482 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
483 skb->len + ETH_HLEN);
484 batadv_send_skb_packet(skb, neigh_node->if_incoming, neigh_node->addr);
485
486 return true;
487out_err:
488 return false;
489}