1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (C) 2013-2018 B.A.T.M.A.N. contributors:
4 * Martin Hundebøll <martin@hundeboll.net>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of version 2 of the GNU General Public
8 * License as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include "fragmentation.h"
22 #include <linux/atomic.h>
23 #include <linux/byteorder/generic.h>
24 #include <linux/errno.h>
25 #include <linux/etherdevice.h>
26 #include <linux/gfp.h>
27 #include <linux/if_ether.h>
28 #include <linux/jiffies.h>
29 #include <linux/kernel.h>
30 #include <linux/lockdep.h>
31 #include <linux/netdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/slab.h>
34 #include <linux/spinlock.h>
35 #include <linux/string.h>
36 #include <uapi/linux/batadv_packet.h>
38 #include "hard-interface.h"
39 #include "originator.h"
42 #include "soft-interface.h"
45 * batadv_frag_clear_chain() - delete entries in the fragment buffer chain
46 * @head: head of chain with entries.
47 * @dropped: whether the chain is cleared because all fragments are dropped
49 * Free fragments in the passed hlist. Should be called with appropriate lock.
51 static void batadv_frag_clear_chain(struct hlist_head *head, bool dropped)
53 struct batadv_frag_list_entry *entry;
54 struct hlist_node *node;
56 hlist_for_each_entry_safe(entry, node, head, list) {
57 hlist_del(&entry->list);
60 kfree_skb(entry->skb);
62 consume_skb(entry->skb);
69 * batadv_frag_purge_orig() - free fragments associated to an orig
70 * @orig_node: originator to free fragments from
71 * @check_cb: optional function to tell if an entry should be purged
73 void batadv_frag_purge_orig(struct batadv_orig_node *orig_node,
74 bool (*check_cb)(struct batadv_frag_table_entry *))
76 struct batadv_frag_table_entry *chain;
79 for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) {
80 chain = &orig_node->fragments[i];
81 spin_lock_bh(&chain->lock);
83 if (!check_cb || check_cb(chain)) {
84 batadv_frag_clear_chain(&chain->fragment_list, true);
88 spin_unlock_bh(&chain->lock);
93 * batadv_frag_size_limit() - maximum possible size of packet to be fragmented
95 * Return: the maximum size of payload that can be fragmented.
97 static int batadv_frag_size_limit(void)
99 int limit = BATADV_FRAG_MAX_FRAG_SIZE;
101 limit -= sizeof(struct batadv_frag_packet);
102 limit *= BATADV_FRAG_MAX_FRAGMENTS;
108 * batadv_frag_init_chain() - check and prepare fragment chain for new fragment
109 * @chain: chain in fragments table to init
110 * @seqno: sequence number of the received fragment
112 * Make chain ready for a fragment with sequence number "seqno". Delete existing
113 * entries if they have an "old" sequence number.
115 * Caller must hold chain->lock.
117 * Return: true if chain is empty and caller can just insert the new fragment
118 * without searching for the right position.
120 static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain,
123 lockdep_assert_held(&chain->lock);
125 if (chain->seqno == seqno)
128 if (!hlist_empty(&chain->fragment_list))
129 batadv_frag_clear_chain(&chain->fragment_list, true);
132 chain->seqno = seqno;
138 * batadv_frag_insert_packet() - insert a fragment into a fragment chain
139 * @orig_node: originator that the fragment was received from
140 * @skb: skb to insert
141 * @chain_out: list head to attach complete chains of fragments to
143 * Insert a new fragment into the reverse ordered chain in the right table
144 * entry. The hash table entry is cleared if "old" fragments exist in it.
146 * Return: true if skb is buffered, false on error. If the chain has all the
147 * fragments needed to merge the packet, the chain is moved to the passed head
148 * to avoid locking the chain in the table.
150 static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node,
152 struct hlist_head *chain_out)
154 struct batadv_frag_table_entry *chain;
155 struct batadv_frag_list_entry *frag_entry_new = NULL, *frag_entry_curr;
156 struct batadv_frag_list_entry *frag_entry_last = NULL;
157 struct batadv_frag_packet *frag_packet;
159 u16 seqno, hdr_size = sizeof(struct batadv_frag_packet);
162 /* Linearize packet to avoid linearizing 16 packets in a row when doing
163 * the later merge. Non-linear merge should be added to remove this
166 if (skb_linearize(skb) < 0)
169 frag_packet = (struct batadv_frag_packet *)skb->data;
170 seqno = ntohs(frag_packet->seqno);
171 bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
173 frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
177 frag_entry_new->skb = skb;
178 frag_entry_new->no = frag_packet->no;
180 /* Select entry in the "chain table" and delete any prior fragments
181 * with another sequence number. batadv_frag_init_chain() returns true,
182 * if the list is empty at return.
184 chain = &orig_node->fragments[bucket];
185 spin_lock_bh(&chain->lock);
186 if (batadv_frag_init_chain(chain, seqno)) {
187 hlist_add_head(&frag_entry_new->list, &chain->fragment_list);
188 chain->size = skb->len - hdr_size;
189 chain->timestamp = jiffies;
190 chain->total_size = ntohs(frag_packet->total_size);
195 /* Find the position for the new fragment. */
196 hlist_for_each_entry(frag_entry_curr, &chain->fragment_list, list) {
197 /* Drop packet if fragment already exists. */
198 if (frag_entry_curr->no == frag_entry_new->no)
201 /* Order fragments from highest to lowest. */
202 if (frag_entry_curr->no < frag_entry_new->no) {
203 hlist_add_before(&frag_entry_new->list,
204 &frag_entry_curr->list);
205 chain->size += skb->len - hdr_size;
206 chain->timestamp = jiffies;
211 /* store current entry because it could be the last in list */
212 frag_entry_last = frag_entry_curr;
215 /* Reached the end of the list, so insert after 'frag_entry_last'. */
216 if (likely(frag_entry_last)) {
217 hlist_add_behind(&frag_entry_new->list, &frag_entry_last->list);
218 chain->size += skb->len - hdr_size;
219 chain->timestamp = jiffies;
224 if (chain->size > batadv_frag_size_limit() ||
225 chain->total_size != ntohs(frag_packet->total_size) ||
226 chain->total_size > batadv_frag_size_limit()) {
227 /* Clear chain if total size of either the list or the packet
228 * exceeds the maximum size of one merged packet. Don't allow
229 * packets to have different total_size.
231 batadv_frag_clear_chain(&chain->fragment_list, true);
233 } else if (ntohs(frag_packet->total_size) == chain->size) {
234 /* All fragments received. Hand over chain to caller. */
235 hlist_move_list(&chain->fragment_list, chain_out);
240 spin_unlock_bh(&chain->lock);
244 kfree(frag_entry_new);
252 * batadv_frag_merge_packets() - merge a chain of fragments
253 * @chain: head of chain with fragments
255 * Expand the first skb in the chain and copy the content of the remaining
256 * skb's into the expanded one. After doing so, clear the chain.
258 * Return: the merged skb or NULL on error.
260 static struct sk_buff *
261 batadv_frag_merge_packets(struct hlist_head *chain)
263 struct batadv_frag_packet *packet;
264 struct batadv_frag_list_entry *entry;
265 struct sk_buff *skb_out;
266 int size, hdr_size = sizeof(struct batadv_frag_packet);
267 bool dropped = false;
269 /* Remove first entry, as this is the destination for the rest of the
272 entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
273 hlist_del(&entry->list);
274 skb_out = entry->skb;
277 packet = (struct batadv_frag_packet *)skb_out->data;
278 size = ntohs(packet->total_size) + hdr_size;
280 /* Make room for the rest of the fragments. */
281 if (pskb_expand_head(skb_out, 0, size - skb_out->len, GFP_ATOMIC) < 0) {
288 /* Move the existing MAC header to just before the payload. (Override
289 * the fragment header.)
291 skb_pull(skb_out, hdr_size);
292 skb_out->ip_summed = CHECKSUM_NONE;
293 memmove(skb_out->data - ETH_HLEN, skb_mac_header(skb_out), ETH_HLEN);
294 skb_set_mac_header(skb_out, -ETH_HLEN);
295 skb_reset_network_header(skb_out);
296 skb_reset_transport_header(skb_out);
298 /* Copy the payload of the each fragment into the last skb */
299 hlist_for_each_entry(entry, chain, list) {
300 size = entry->skb->len - hdr_size;
301 skb_put_data(skb_out, entry->skb->data + hdr_size, size);
305 /* Locking is not needed, because 'chain' is not part of any orig. */
306 batadv_frag_clear_chain(chain, dropped);
311 * batadv_frag_skb_buffer() - buffer fragment for later merge
312 * @skb: skb to buffer
313 * @orig_node_src: originator that the skb is received from
315 * Add fragment to buffer and merge fragments if possible.
317 * There are three possible outcomes: 1) Packet is merged: Return true and
318 * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
319 * to NULL; 3) Error: Return false and free skb.
321 * Return: true when packet is merged or buffered, false when skb is not not
324 bool batadv_frag_skb_buffer(struct sk_buff **skb,
325 struct batadv_orig_node *orig_node_src)
327 struct sk_buff *skb_out = NULL;
328 struct hlist_head head = HLIST_HEAD_INIT;
331 /* Add packet to buffer and table entry if merge is possible. */
332 if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
335 /* Leave if more fragments are needed to merge. */
336 if (hlist_empty(&head))
339 skb_out = batadv_frag_merge_packets(&head);
351 * batadv_frag_skb_fwd() - forward fragments that would exceed MTU when merged
352 * @skb: skb to forward
353 * @recv_if: interface that the skb is received on
354 * @orig_node_src: originator that the skb is received from
356 * Look up the next-hop of the fragments payload and check if the merged packet
357 * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
358 * without merging it.
360 * Return: true if the fragment is consumed/forwarded, false otherwise.
362 bool batadv_frag_skb_fwd(struct sk_buff *skb,
363 struct batadv_hard_iface *recv_if,
364 struct batadv_orig_node *orig_node_src)
366 struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
367 struct batadv_orig_node *orig_node_dst;
368 struct batadv_neigh_node *neigh_node = NULL;
369 struct batadv_frag_packet *packet;
373 packet = (struct batadv_frag_packet *)skb->data;
374 orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
378 neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
382 /* Forward the fragment, if the merged packet would be too big to
385 total_size = ntohs(packet->total_size);
386 if (total_size > neigh_node->if_incoming->net_dev->mtu) {
387 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
388 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
389 skb->len + ETH_HLEN);
392 batadv_send_unicast_skb(skb, neigh_node);
398 batadv_orig_node_put(orig_node_dst);
400 batadv_neigh_node_put(neigh_node);
405 * batadv_frag_create() - create a fragment from skb
406 * @skb: skb to create fragment from
407 * @frag_head: header to use in new fragment
408 * @fragment_size: size of new fragment
410 * Split the passed skb into two fragments: A new one with size matching the
411 * passed mtu and the old one with the rest. The new skb contains data from the
412 * tail of the old skb.
414 * Return: the new fragment, NULL on error.
416 static struct sk_buff *batadv_frag_create(struct sk_buff *skb,
417 struct batadv_frag_packet *frag_head,
418 unsigned int fragment_size)
420 struct sk_buff *skb_fragment;
421 unsigned int header_size = sizeof(*frag_head);
422 unsigned int mtu = fragment_size + header_size;
424 skb_fragment = netdev_alloc_skb(NULL, mtu + ETH_HLEN);
428 skb_fragment->priority = skb->priority;
430 /* Eat the last mtu-bytes of the skb */
431 skb_reserve(skb_fragment, header_size + ETH_HLEN);
432 skb_split(skb, skb_fragment, skb->len - fragment_size);
435 skb_push(skb_fragment, header_size);
436 memcpy(skb_fragment->data, frag_head, header_size);
443 * batadv_frag_send_packet() - create up to 16 fragments from the passed skb
444 * @skb: skb to create fragments from
445 * @orig_node: final destination of the created fragments
446 * @neigh_node: next-hop of the created fragments
448 * Return: the netdev tx status or a negative errno code on a failure
450 int batadv_frag_send_packet(struct sk_buff *skb,
451 struct batadv_orig_node *orig_node,
452 struct batadv_neigh_node *neigh_node)
454 struct batadv_priv *bat_priv;
455 struct batadv_hard_iface *primary_if = NULL;
456 struct batadv_frag_packet frag_header;
457 struct sk_buff *skb_fragment;
458 unsigned int mtu = neigh_node->if_incoming->net_dev->mtu;
459 unsigned int header_size = sizeof(frag_header);
460 unsigned int max_fragment_size, num_fragments;
463 /* To avoid merge and refragmentation at next-hops we never send
464 * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
466 mtu = min_t(unsigned int, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
467 max_fragment_size = mtu - header_size;
469 if (skb->len == 0 || max_fragment_size == 0)
472 num_fragments = (skb->len - 1) / max_fragment_size + 1;
473 max_fragment_size = (skb->len - 1) / num_fragments + 1;
475 /* Don't even try to fragment, if we need more than 16 fragments */
476 if (num_fragments > BATADV_FRAG_MAX_FRAGMENTS) {
481 bat_priv = orig_node->bat_priv;
482 primary_if = batadv_primary_if_get_selected(bat_priv);
488 /* Create one header to be copied to all fragments */
489 frag_header.packet_type = BATADV_UNICAST_FRAG;
490 frag_header.version = BATADV_COMPAT_VERSION;
491 frag_header.ttl = BATADV_TTL;
492 frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
493 frag_header.reserved = 0;
495 frag_header.total_size = htons(skb->len);
497 /* skb->priority values from 256->263 are magic values to
498 * directly indicate a specific 802.1d priority. This is used
499 * to allow 802.1d priority to be passed directly in from VLAN
502 if (skb->priority >= 256 && skb->priority <= 263)
503 frag_header.priority = skb->priority - 256;
505 frag_header.priority = 0;
507 ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr);
508 ether_addr_copy(frag_header.dest, orig_node->orig);
510 /* Eat and send fragments from the tail of skb */
511 while (skb->len > max_fragment_size) {
512 /* The initial check in this function should cover this case */
513 if (unlikely(frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1)) {
518 skb_fragment = batadv_frag_create(skb, &frag_header,
525 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
526 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
527 skb_fragment->len + ETH_HLEN);
528 ret = batadv_send_unicast_skb(skb_fragment, neigh_node);
529 if (ret != NET_XMIT_SUCCESS) {
537 /* Make room for the fragment header. */
538 if (batadv_skb_head_push(skb, header_size) < 0 ||
539 pskb_expand_head(skb, header_size + ETH_HLEN, 0, GFP_ATOMIC) < 0) {
544 memcpy(skb->data, &frag_header, header_size);
546 /* Send the last fragment */
547 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
548 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
549 skb->len + ETH_HLEN);
550 ret = batadv_send_unicast_skb(skb, neigh_node);
551 /* skb was consumed */
555 batadv_hardif_put(primary_if);