xref: /linux/net/batman-adv/fragmentation.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /* Copyright (C) 2013-2017  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/errno.h>
24 #include <linux/etherdevice.h>
25 #include <linux/fs.h>
26 #include <linux/if_ether.h>
27 #include <linux/jiffies.h>
28 #include <linux/kernel.h>
29 #include <linux/lockdep.h>
30 #include <linux/netdevice.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  * @dropped: whether the chain is cleared because all fragments are dropped
47  *
48  * Free fragments in the passed hlist. Should be called with appropriate lock.
49  */
50 static void batadv_frag_clear_chain(struct hlist_head *head, bool dropped)
51 {
52 	struct batadv_frag_list_entry *entry;
53 	struct hlist_node *node;
54 
55 	hlist_for_each_entry_safe(entry, node, head, list) {
56 		hlist_del(&entry->list);
57 
58 		if (dropped)
59 			kfree_skb(entry->skb);
60 		else
61 			consume_skb(entry->skb);
62 
63 		kfree(entry);
64 	}
65 }
66 
67 /**
68  * batadv_frag_purge_orig - free fragments associated to an orig
69  * @orig_node: originator to free fragments from
70  * @check_cb: optional function to tell if an entry should be purged
71  */
72 void batadv_frag_purge_orig(struct batadv_orig_node *orig_node,
73 			    bool (*check_cb)(struct batadv_frag_table_entry *))
74 {
75 	struct batadv_frag_table_entry *chain;
76 	u8 i;
77 
78 	for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) {
79 		chain = &orig_node->fragments[i];
80 		spin_lock_bh(&chain->lock);
81 
82 		if (!check_cb || check_cb(chain)) {
83 			batadv_frag_clear_chain(&chain->fragment_list, true);
84 			chain->size = 0;
85 		}
86 
87 		spin_unlock_bh(&chain->lock);
88 	}
89 }
90 
91 /**
92  * batadv_frag_size_limit - maximum possible size of packet to be fragmented
93  *
94  * Return: the maximum size of payload that can be fragmented.
95  */
96 static int batadv_frag_size_limit(void)
97 {
98 	int limit = BATADV_FRAG_MAX_FRAG_SIZE;
99 
100 	limit -= sizeof(struct batadv_frag_packet);
101 	limit *= BATADV_FRAG_MAX_FRAGMENTS;
102 
103 	return limit;
104 }
105 
106 /**
107  * batadv_frag_init_chain - check and prepare fragment chain for new fragment
108  * @chain: chain in fragments table to init
109  * @seqno: sequence number of the received fragment
110  *
111  * Make chain ready for a fragment with sequence number "seqno". Delete existing
112  * entries if they have an "old" sequence number.
113  *
114  * Caller must hold chain->lock.
115  *
116  * Return: true if chain is empty and caller can just insert the new fragment
117  * without searching for the right position.
118  */
119 static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain,
120 				   u16 seqno)
121 {
122 	lockdep_assert_held(&chain->lock);
123 
124 	if (chain->seqno == seqno)
125 		return false;
126 
127 	if (!hlist_empty(&chain->fragment_list))
128 		batadv_frag_clear_chain(&chain->fragment_list, true);
129 
130 	chain->size = 0;
131 	chain->seqno = seqno;
132 
133 	return true;
134 }
135 
136 /**
137  * batadv_frag_insert_packet - insert a fragment into a fragment chain
138  * @orig_node: originator that the fragment was received from
139  * @skb: skb to insert
140  * @chain_out: list head to attach complete chains of fragments to
141  *
142  * Insert a new fragment into the reverse ordered chain in the right table
143  * entry. The hash table entry is cleared if "old" fragments exist in it.
144  *
145  * Return: true if skb is buffered, false on error. If the chain has all the
146  * fragments needed to merge the packet, the chain is moved to the passed head
147  * to avoid locking the chain in the table.
148  */
149 static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node,
150 				      struct sk_buff *skb,
151 				      struct hlist_head *chain_out)
152 {
153 	struct batadv_frag_table_entry *chain;
154 	struct batadv_frag_list_entry *frag_entry_new = NULL, *frag_entry_curr;
155 	struct batadv_frag_list_entry *frag_entry_last = NULL;
156 	struct batadv_frag_packet *frag_packet;
157 	u8 bucket;
158 	u16 seqno, hdr_size = sizeof(struct batadv_frag_packet);
159 	bool ret = false;
160 
161 	/* Linearize packet to avoid linearizing 16 packets in a row when doing
162 	 * the later merge. Non-linear merge should be added to remove this
163 	 * linearization.
164 	 */
165 	if (skb_linearize(skb) < 0)
166 		goto err;
167 
168 	frag_packet = (struct batadv_frag_packet *)skb->data;
169 	seqno = ntohs(frag_packet->seqno);
170 	bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
171 
172 	frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
173 	if (!frag_entry_new)
174 		goto err;
175 
176 	frag_entry_new->skb = skb;
177 	frag_entry_new->no = frag_packet->no;
178 
179 	/* Select entry in the "chain table" and delete any prior fragments
180 	 * with another sequence number. batadv_frag_init_chain() returns true,
181 	 * if the list is empty at return.
182 	 */
183 	chain = &orig_node->fragments[bucket];
184 	spin_lock_bh(&chain->lock);
185 	if (batadv_frag_init_chain(chain, seqno)) {
186 		hlist_add_head(&frag_entry_new->list, &chain->fragment_list);
187 		chain->size = skb->len - hdr_size;
188 		chain->timestamp = jiffies;
189 		chain->total_size = ntohs(frag_packet->total_size);
190 		ret = true;
191 		goto out;
192 	}
193 
194 	/* Find the position for the new fragment. */
195 	hlist_for_each_entry(frag_entry_curr, &chain->fragment_list, list) {
196 		/* Drop packet if fragment already exists. */
197 		if (frag_entry_curr->no == frag_entry_new->no)
198 			goto err_unlock;
199 
200 		/* Order fragments from highest to lowest. */
201 		if (frag_entry_curr->no < frag_entry_new->no) {
202 			hlist_add_before(&frag_entry_new->list,
203 					 &frag_entry_curr->list);
204 			chain->size += skb->len - hdr_size;
205 			chain->timestamp = jiffies;
206 			ret = true;
207 			goto out;
208 		}
209 
210 		/* store current entry because it could be the last in list */
211 		frag_entry_last = frag_entry_curr;
212 	}
213 
214 	/* Reached the end of the list, so insert after 'frag_entry_last'. */
215 	if (likely(frag_entry_last)) {
216 		hlist_add_behind(&frag_entry_new->list, &frag_entry_last->list);
217 		chain->size += skb->len - hdr_size;
218 		chain->timestamp = jiffies;
219 		ret = true;
220 	}
221 
222 out:
223 	if (chain->size > batadv_frag_size_limit() ||
224 	    chain->total_size != ntohs(frag_packet->total_size) ||
225 	    chain->total_size > batadv_frag_size_limit()) {
226 		/* Clear chain if total size of either the list or the packet
227 		 * exceeds the maximum size of one merged packet. Don't allow
228 		 * packets to have different total_size.
229 		 */
230 		batadv_frag_clear_chain(&chain->fragment_list, true);
231 		chain->size = 0;
232 	} else if (ntohs(frag_packet->total_size) == chain->size) {
233 		/* All fragments received. Hand over chain to caller. */
234 		hlist_move_list(&chain->fragment_list, chain_out);
235 		chain->size = 0;
236 	}
237 
238 err_unlock:
239 	spin_unlock_bh(&chain->lock);
240 
241 err:
242 	if (!ret) {
243 		kfree(frag_entry_new);
244 		kfree_skb(skb);
245 	}
246 
247 	return ret;
248 }
249 
250 /**
251  * batadv_frag_merge_packets - merge a chain of fragments
252  * @chain: head of chain with fragments
253  *
254  * Expand the first skb in the chain and copy the content of the remaining
255  * skb's into the expanded one. After doing so, clear the chain.
256  *
257  * Return: the merged skb or NULL on error.
258  */
259 static struct sk_buff *
260 batadv_frag_merge_packets(struct hlist_head *chain)
261 {
262 	struct batadv_frag_packet *packet;
263 	struct batadv_frag_list_entry *entry;
264 	struct sk_buff *skb_out;
265 	int size, hdr_size = sizeof(struct batadv_frag_packet);
266 	bool dropped = false;
267 
268 	/* Remove first entry, as this is the destination for the rest of the
269 	 * fragments.
270 	 */
271 	entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
272 	hlist_del(&entry->list);
273 	skb_out = entry->skb;
274 	kfree(entry);
275 
276 	packet = (struct batadv_frag_packet *)skb_out->data;
277 	size = ntohs(packet->total_size);
278 
279 	/* Make room for the rest of the fragments. */
280 	if (pskb_expand_head(skb_out, 0, size - skb_out->len, GFP_ATOMIC) < 0) {
281 		kfree_skb(skb_out);
282 		skb_out = NULL;
283 		dropped = true;
284 		goto free;
285 	}
286 
287 	/* Move the existing MAC header to just before the payload. (Override
288 	 * the fragment header.)
289 	 */
290 	skb_pull_rcsum(skb_out, hdr_size);
291 	memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN);
292 	skb_set_mac_header(skb_out, -ETH_HLEN);
293 	skb_reset_network_header(skb_out);
294 	skb_reset_transport_header(skb_out);
295 
296 	/* Copy the payload of the each fragment into the last skb */
297 	hlist_for_each_entry(entry, chain, list) {
298 		size = entry->skb->len - hdr_size;
299 		memcpy(skb_put(skb_out, size), entry->skb->data + hdr_size,
300 		       size);
301 	}
302 
303 free:
304 	/* Locking is not needed, because 'chain' is not part of any orig. */
305 	batadv_frag_clear_chain(chain, dropped);
306 	return skb_out;
307 }
308 
309 /**
310  * batadv_frag_skb_buffer - buffer fragment for later merge
311  * @skb: skb to buffer
312  * @orig_node_src: originator that the skb is received from
313  *
314  * Add fragment to buffer and merge fragments if possible.
315  *
316  * There are three possible outcomes: 1) Packet is merged: Return true and
317  * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
318  * to NULL; 3) Error: Return false and free skb.
319  *
320  * Return: true when packet is merged or buffered, false when skb is not not
321  * used.
322  */
323 bool batadv_frag_skb_buffer(struct sk_buff **skb,
324 			    struct batadv_orig_node *orig_node_src)
325 {
326 	struct sk_buff *skb_out = NULL;
327 	struct hlist_head head = HLIST_HEAD_INIT;
328 	bool ret = false;
329 
330 	/* Add packet to buffer and table entry if merge is possible. */
331 	if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
332 		goto out_err;
333 
334 	/* Leave if more fragments are needed to merge. */
335 	if (hlist_empty(&head))
336 		goto out;
337 
338 	skb_out = batadv_frag_merge_packets(&head);
339 	if (!skb_out)
340 		goto out_err;
341 
342 out:
343 	ret = true;
344 out_err:
345 	*skb = skb_out;
346 	return ret;
347 }
348 
349 /**
350  * batadv_frag_skb_fwd - forward fragments that would exceed MTU when merged
351  * @skb: skb to forward
352  * @recv_if: interface that the skb is received on
353  * @orig_node_src: originator that the skb is received from
354  *
355  * Look up the next-hop of the fragments payload and check if the merged packet
356  * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
357  * without merging it.
358  *
359  * Return: true if the fragment is consumed/forwarded, false otherwise.
360  */
361 bool batadv_frag_skb_fwd(struct sk_buff *skb,
362 			 struct batadv_hard_iface *recv_if,
363 			 struct batadv_orig_node *orig_node_src)
364 {
365 	struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
366 	struct batadv_orig_node *orig_node_dst;
367 	struct batadv_neigh_node *neigh_node = NULL;
368 	struct batadv_frag_packet *packet;
369 	u16 total_size;
370 	bool ret = false;
371 
372 	packet = (struct batadv_frag_packet *)skb->data;
373 	orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
374 	if (!orig_node_dst)
375 		goto out;
376 
377 	neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
378 	if (!neigh_node)
379 		goto out;
380 
381 	/* Forward the fragment, if the merged packet would be too big to
382 	 * be assembled.
383 	 */
384 	total_size = ntohs(packet->total_size);
385 	if (total_size > neigh_node->if_incoming->net_dev->mtu) {
386 		batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
387 		batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
388 				   skb->len + ETH_HLEN);
389 
390 		packet->ttl--;
391 		batadv_send_unicast_skb(skb, neigh_node);
392 		ret = true;
393 	}
394 
395 out:
396 	if (orig_node_dst)
397 		batadv_orig_node_put(orig_node_dst);
398 	if (neigh_node)
399 		batadv_neigh_node_put(neigh_node);
400 	return ret;
401 }
402 
403 /**
404  * batadv_frag_create - create a fragment from skb
405  * @skb: skb to create fragment from
406  * @frag_head: header to use in new fragment
407  * @fragment_size: size of new fragment
408  *
409  * Split the passed skb into two fragments: A new one with size matching the
410  * passed mtu and the old one with the rest. The new skb contains data from the
411  * tail of the old skb.
412  *
413  * Return: the new fragment, NULL on error.
414  */
415 static struct sk_buff *batadv_frag_create(struct sk_buff *skb,
416 					  struct batadv_frag_packet *frag_head,
417 					  unsigned int fragment_size)
418 {
419 	struct sk_buff *skb_fragment;
420 	unsigned int header_size = sizeof(*frag_head);
421 	unsigned int mtu = fragment_size + header_size;
422 
423 	skb_fragment = netdev_alloc_skb(NULL, mtu + ETH_HLEN);
424 	if (!skb_fragment)
425 		goto err;
426 
427 	skb_fragment->priority = skb->priority;
428 
429 	/* Eat the last mtu-bytes of the skb */
430 	skb_reserve(skb_fragment, header_size + ETH_HLEN);
431 	skb_split(skb, skb_fragment, skb->len - fragment_size);
432 
433 	/* Add the header */
434 	skb_push(skb_fragment, header_size);
435 	memcpy(skb_fragment->data, frag_head, header_size);
436 
437 err:
438 	return skb_fragment;
439 }
440 
441 /**
442  * batadv_frag_send_packet - create up to 16 fragments from the passed skb
443  * @skb: skb to create fragments from
444  * @orig_node: final destination of the created fragments
445  * @neigh_node: next-hop of the created fragments
446  *
447  * Return: the netdev tx status or a negative errno code on a failure
448  */
449 int batadv_frag_send_packet(struct sk_buff *skb,
450 			    struct batadv_orig_node *orig_node,
451 			    struct batadv_neigh_node *neigh_node)
452 {
453 	struct batadv_priv *bat_priv;
454 	struct batadv_hard_iface *primary_if = NULL;
455 	struct batadv_frag_packet frag_header;
456 	struct sk_buff *skb_fragment;
457 	unsigned int mtu = neigh_node->if_incoming->net_dev->mtu;
458 	unsigned int header_size = sizeof(frag_header);
459 	unsigned int max_fragment_size, num_fragments;
460 	int ret;
461 
462 	/* To avoid merge and refragmentation at next-hops we never send
463 	 * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
464 	 */
465 	mtu = min_t(unsigned int, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
466 	max_fragment_size = mtu - header_size;
467 
468 	if (skb->len == 0 || max_fragment_size == 0)
469 		return -EINVAL;
470 
471 	num_fragments = (skb->len - 1) / max_fragment_size + 1;
472 	max_fragment_size = (skb->len - 1) / num_fragments + 1;
473 
474 	/* Don't even try to fragment, if we need more than 16 fragments */
475 	if (num_fragments > BATADV_FRAG_MAX_FRAGMENTS) {
476 		ret = -EAGAIN;
477 		goto free_skb;
478 	}
479 
480 	bat_priv = orig_node->bat_priv;
481 	primary_if = batadv_primary_if_get_selected(bat_priv);
482 	if (!primary_if) {
483 		ret = -EINVAL;
484 		goto free_skb;
485 	}
486 
487 	/* Create one header to be copied to all fragments */
488 	frag_header.packet_type = BATADV_UNICAST_FRAG;
489 	frag_header.version = BATADV_COMPAT_VERSION;
490 	frag_header.ttl = BATADV_TTL;
491 	frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
492 	frag_header.reserved = 0;
493 	frag_header.no = 0;
494 	frag_header.total_size = htons(skb->len);
495 
496 	/* skb->priority values from 256->263 are magic values to
497 	 * directly indicate a specific 802.1d priority.  This is used
498 	 * to allow 802.1d priority to be passed directly in from VLAN
499 	 * tags, etc.
500 	 */
501 	if (skb->priority >= 256 && skb->priority <= 263)
502 		frag_header.priority = skb->priority - 256;
503 
504 	ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr);
505 	ether_addr_copy(frag_header.dest, orig_node->orig);
506 
507 	/* Eat and send fragments from the tail of skb */
508 	while (skb->len > max_fragment_size) {
509 		/* The initial check in this function should cover this case */
510 		if (unlikely(frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1)) {
511 			ret = -EINVAL;
512 			goto put_primary_if;
513 		}
514 
515 		skb_fragment = batadv_frag_create(skb, &frag_header,
516 						  max_fragment_size);
517 		if (!skb_fragment) {
518 			ret = -ENOMEM;
519 			goto put_primary_if;
520 		}
521 
522 		batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
523 		batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
524 				   skb_fragment->len + ETH_HLEN);
525 		ret = batadv_send_unicast_skb(skb_fragment, neigh_node);
526 		if (ret != NET_XMIT_SUCCESS) {
527 			ret = NET_XMIT_DROP;
528 			goto put_primary_if;
529 		}
530 
531 		frag_header.no++;
532 	}
533 
534 	/* Make room for the fragment header. */
535 	if (batadv_skb_head_push(skb, header_size) < 0 ||
536 	    pskb_expand_head(skb, header_size + ETH_HLEN, 0, GFP_ATOMIC) < 0) {
537 		ret = -ENOMEM;
538 		goto put_primary_if;
539 	}
540 
541 	memcpy(skb->data, &frag_header, header_size);
542 
543 	/* Send the last fragment */
544 	batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
545 	batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
546 			   skb->len + ETH_HLEN);
547 	ret = batadv_send_unicast_skb(skb, neigh_node);
548 	/* skb was consumed */
549 	skb = NULL;
550 
551 put_primary_if:
552 	batadv_hardif_put(primary_if);
553 free_skb:
554 	kfree_skb(skb);
555 
556 	return ret;
557 }
558