1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (C) B.A.T.M.A.N. contributors:
4 * Martin Hundebøll <martin@hundeboll.net>
7 #include "fragmentation.h"
10 #include <linux/atomic.h>
11 #include <linux/byteorder/generic.h>
12 #include <linux/errno.h>
13 #include <linux/etherdevice.h>
14 #include <linux/gfp.h>
15 #include <linux/if_ether.h>
16 #include <linux/jiffies.h>
17 #include <linux/lockdep.h>
18 #include <linux/minmax.h>
19 #include <linux/netdevice.h>
20 #include <linux/skbuff.h>
21 #include <linux/slab.h>
22 #include <linux/spinlock.h>
23 #include <linux/string.h>
24 #include <uapi/linux/batadv_packet.h>
26 #include "hard-interface.h"
27 #include "originator.h"
32 * batadv_frag_clear_chain() - delete entries in the fragment buffer chain
33 * @head: head of chain with entries.
34 * @dropped: whether the chain is cleared because all fragments are dropped
36 * Free fragments in the passed hlist. Should be called with appropriate lock.
38 static void batadv_frag_clear_chain(struct hlist_head *head, bool dropped)
40 struct batadv_frag_list_entry *entry;
41 struct hlist_node *node;
43 hlist_for_each_entry_safe(entry, node, head, list) {
44 hlist_del(&entry->list);
47 kfree_skb(entry->skb);
49 consume_skb(entry->skb);
56 * batadv_frag_purge_orig() - free fragments associated to an orig
57 * @orig_node: originator to free fragments from
58 * @check_cb: optional function to tell if an entry should be purged
60 void batadv_frag_purge_orig(struct batadv_orig_node *orig_node,
61 bool (*check_cb)(struct batadv_frag_table_entry *))
63 struct batadv_frag_table_entry *chain;
66 for (i = 0; i < BATADV_FRAG_BUFFER_COUNT; i++) {
67 chain = &orig_node->fragments[i];
68 spin_lock_bh(&chain->lock);
70 if (!check_cb || check_cb(chain)) {
71 batadv_frag_clear_chain(&chain->fragment_list, true);
75 spin_unlock_bh(&chain->lock);
80 * batadv_frag_size_limit() - maximum possible size of packet to be fragmented
82 * Return: the maximum size of payload that can be fragmented.
84 static int batadv_frag_size_limit(void)
86 int limit = BATADV_FRAG_MAX_FRAG_SIZE;
88 limit -= sizeof(struct batadv_frag_packet);
89 limit *= BATADV_FRAG_MAX_FRAGMENTS;
95 * batadv_frag_init_chain() - check and prepare fragment chain for new fragment
96 * @chain: chain in fragments table to init
97 * @seqno: sequence number of the received fragment
99 * Make chain ready for a fragment with sequence number "seqno". Delete existing
100 * entries if they have an "old" sequence number.
102 * Caller must hold chain->lock.
104 * Return: true if chain is empty and the caller can just insert the new
105 * fragment without searching for the right position.
107 static bool batadv_frag_init_chain(struct batadv_frag_table_entry *chain,
110 lockdep_assert_held(&chain->lock);
112 if (chain->seqno == seqno)
115 if (!hlist_empty(&chain->fragment_list))
116 batadv_frag_clear_chain(&chain->fragment_list, true);
119 chain->seqno = seqno;
125 * batadv_frag_insert_packet() - insert a fragment into a fragment chain
126 * @orig_node: originator that the fragment was received from
127 * @skb: skb to insert
128 * @chain_out: list head to attach complete chains of fragments to
130 * Insert a new fragment into the reverse ordered chain in the right table
131 * entry. The hash table entry is cleared if "old" fragments exist in it.
133 * Return: true if skb is buffered, false on error. If the chain has all the
134 * fragments needed to merge the packet, the chain is moved to the passed head
135 * to avoid locking the chain in the table.
137 static bool batadv_frag_insert_packet(struct batadv_orig_node *orig_node,
139 struct hlist_head *chain_out)
141 struct batadv_frag_table_entry *chain;
142 struct batadv_frag_list_entry *frag_entry_new = NULL, *frag_entry_curr;
143 struct batadv_frag_list_entry *frag_entry_last = NULL;
144 struct batadv_frag_packet *frag_packet;
146 u16 seqno, hdr_size = sizeof(struct batadv_frag_packet);
149 /* Linearize packet to avoid linearizing 16 packets in a row when doing
150 * the later merge. Non-linear merge should be added to remove this
153 if (skb_linearize(skb) < 0)
156 frag_packet = (struct batadv_frag_packet *)skb->data;
157 seqno = ntohs(frag_packet->seqno);
158 bucket = seqno % BATADV_FRAG_BUFFER_COUNT;
160 frag_entry_new = kmalloc(sizeof(*frag_entry_new), GFP_ATOMIC);
164 frag_entry_new->skb = skb;
165 frag_entry_new->no = frag_packet->no;
167 /* Select entry in the "chain table" and delete any prior fragments
168 * with another sequence number. batadv_frag_init_chain() returns true,
169 * if the list is empty at return.
171 chain = &orig_node->fragments[bucket];
172 spin_lock_bh(&chain->lock);
173 if (batadv_frag_init_chain(chain, seqno)) {
174 hlist_add_head(&frag_entry_new->list, &chain->fragment_list);
175 chain->size = skb->len - hdr_size;
176 chain->timestamp = jiffies;
177 chain->total_size = ntohs(frag_packet->total_size);
182 /* Find the position for the new fragment. */
183 hlist_for_each_entry(frag_entry_curr, &chain->fragment_list, list) {
184 /* Drop packet if fragment already exists. */
185 if (frag_entry_curr->no == frag_entry_new->no)
188 /* Order fragments from highest to lowest. */
189 if (frag_entry_curr->no < frag_entry_new->no) {
190 hlist_add_before(&frag_entry_new->list,
191 &frag_entry_curr->list);
192 chain->size += skb->len - hdr_size;
193 chain->timestamp = jiffies;
198 /* store current entry because it could be the last in list */
199 frag_entry_last = frag_entry_curr;
202 /* Reached the end of the list, so insert after 'frag_entry_last'. */
203 if (likely(frag_entry_last)) {
204 hlist_add_behind(&frag_entry_new->list, &frag_entry_last->list);
205 chain->size += skb->len - hdr_size;
206 chain->timestamp = jiffies;
211 if (chain->size > batadv_frag_size_limit() ||
212 chain->total_size != ntohs(frag_packet->total_size) ||
213 chain->total_size > batadv_frag_size_limit()) {
214 /* Clear chain if total size of either the list or the packet
215 * exceeds the maximum size of one merged packet. Don't allow
216 * packets to have different total_size.
218 batadv_frag_clear_chain(&chain->fragment_list, true);
220 } else if (ntohs(frag_packet->total_size) == chain->size) {
221 /* All fragments received. Hand over chain to caller. */
222 hlist_move_list(&chain->fragment_list, chain_out);
227 spin_unlock_bh(&chain->lock);
231 kfree(frag_entry_new);
239 * batadv_frag_merge_packets() - merge a chain of fragments
240 * @chain: head of chain with fragments
242 * Expand the first skb in the chain and copy the content of the remaining
243 * skb's into the expanded one. After doing so, clear the chain.
245 * Return: the merged skb or NULL on error.
247 static struct sk_buff *
248 batadv_frag_merge_packets(struct hlist_head *chain)
250 struct batadv_frag_packet *packet;
251 struct batadv_frag_list_entry *entry;
252 struct sk_buff *skb_out;
253 int size, hdr_size = sizeof(struct batadv_frag_packet);
254 bool dropped = false;
256 /* Remove first entry, as this is the destination for the rest of the
259 entry = hlist_entry(chain->first, struct batadv_frag_list_entry, list);
260 hlist_del(&entry->list);
261 skb_out = entry->skb;
264 packet = (struct batadv_frag_packet *)skb_out->data;
265 size = ntohs(packet->total_size) + hdr_size;
267 /* Make room for the rest of the fragments. */
268 if (pskb_expand_head(skb_out, 0, size - skb_out->len, GFP_ATOMIC) < 0) {
275 /* Move the existing MAC header to just before the payload. (Override
276 * the fragment header.)
278 skb_pull(skb_out, hdr_size);
279 skb_out->ip_summed = CHECKSUM_NONE;
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);
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 skb_put_data(skb_out, entry->skb->data + hdr_size, size);
292 /* Locking is not needed, because 'chain' is not part of any orig. */
293 batadv_frag_clear_chain(chain, dropped);
298 * batadv_frag_skb_buffer() - buffer fragment for later merge
299 * @skb: skb to buffer
300 * @orig_node_src: originator that the skb is received from
302 * Add fragment to buffer and merge fragments if possible.
304 * There are three possible outcomes: 1) Packet is merged: Return true and
305 * set *skb to merged packet; 2) Packet is buffered: Return true and set *skb
306 * to NULL; 3) Error: Return false and free skb.
308 * Return: true when the packet is merged or buffered, false when skb is not
311 bool batadv_frag_skb_buffer(struct sk_buff **skb,
312 struct batadv_orig_node *orig_node_src)
314 struct sk_buff *skb_out = NULL;
315 struct hlist_head head = HLIST_HEAD_INIT;
318 /* Add packet to buffer and table entry if merge is possible. */
319 if (!batadv_frag_insert_packet(orig_node_src, *skb, &head))
322 /* Leave if more fragments are needed to merge. */
323 if (hlist_empty(&head))
326 skb_out = batadv_frag_merge_packets(&head);
338 * batadv_frag_skb_fwd() - forward fragments that would exceed MTU when merged
339 * @skb: skb to forward
340 * @recv_if: interface that the skb is received on
341 * @orig_node_src: originator that the skb is received from
343 * Look up the next-hop of the fragments payload and check if the merged packet
344 * will exceed the MTU towards the next-hop. If so, the fragment is forwarded
345 * without merging it.
347 * Return: true if the fragment is consumed/forwarded, false otherwise.
349 bool batadv_frag_skb_fwd(struct sk_buff *skb,
350 struct batadv_hard_iface *recv_if,
351 struct batadv_orig_node *orig_node_src)
353 struct batadv_priv *bat_priv = netdev_priv(recv_if->soft_iface);
354 struct batadv_orig_node *orig_node_dst;
355 struct batadv_neigh_node *neigh_node = NULL;
356 struct batadv_frag_packet *packet;
360 packet = (struct batadv_frag_packet *)skb->data;
361 orig_node_dst = batadv_orig_hash_find(bat_priv, packet->dest);
365 neigh_node = batadv_find_router(bat_priv, orig_node_dst, recv_if);
369 /* Forward the fragment, if the merged packet would be too big to
372 total_size = ntohs(packet->total_size);
373 if (total_size > neigh_node->if_incoming->net_dev->mtu) {
374 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_FWD);
375 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_FWD_BYTES,
376 skb->len + ETH_HLEN);
379 batadv_send_unicast_skb(skb, neigh_node);
384 batadv_orig_node_put(orig_node_dst);
385 batadv_neigh_node_put(neigh_node);
390 * batadv_frag_create() - create a fragment from skb
391 * @net_dev: outgoing device for fragment
392 * @skb: skb to create fragment from
393 * @frag_head: header to use in new fragment
394 * @fragment_size: size of new fragment
396 * Split the passed skb into two fragments: A new one with size matching the
397 * passed mtu and the old one with the rest. The new skb contains data from the
398 * tail of the old skb.
400 * Return: the new fragment, NULL on error.
402 static struct sk_buff *batadv_frag_create(struct net_device *net_dev,
404 struct batadv_frag_packet *frag_head,
405 unsigned int fragment_size)
407 unsigned int ll_reserved = LL_RESERVED_SPACE(net_dev);
408 unsigned int tailroom = net_dev->needed_tailroom;
409 struct sk_buff *skb_fragment;
410 unsigned int header_size = sizeof(*frag_head);
411 unsigned int mtu = fragment_size + header_size;
413 skb_fragment = dev_alloc_skb(ll_reserved + mtu + tailroom);
417 skb_fragment->priority = skb->priority;
419 /* Eat the last mtu-bytes of the skb */
420 skb_reserve(skb_fragment, ll_reserved + header_size);
421 skb_split(skb, skb_fragment, skb->len - fragment_size);
424 skb_push(skb_fragment, header_size);
425 memcpy(skb_fragment->data, frag_head, header_size);
432 * batadv_frag_send_packet() - create up to 16 fragments from the passed skb
433 * @skb: skb to create fragments from
434 * @orig_node: final destination of the created fragments
435 * @neigh_node: next-hop of the created fragments
437 * Return: the netdev tx status or a negative errno code on a failure
439 int batadv_frag_send_packet(struct sk_buff *skb,
440 struct batadv_orig_node *orig_node,
441 struct batadv_neigh_node *neigh_node)
443 struct net_device *net_dev = neigh_node->if_incoming->net_dev;
444 struct batadv_priv *bat_priv;
445 struct batadv_hard_iface *primary_if = NULL;
446 struct batadv_frag_packet frag_header;
447 struct sk_buff *skb_fragment;
448 unsigned int mtu = net_dev->mtu;
449 unsigned int header_size = sizeof(frag_header);
450 unsigned int max_fragment_size, num_fragments;
453 /* To avoid merge and refragmentation at next-hops we never send
454 * fragments larger than BATADV_FRAG_MAX_FRAG_SIZE
456 mtu = min_t(unsigned int, mtu, BATADV_FRAG_MAX_FRAG_SIZE);
457 max_fragment_size = mtu - header_size;
459 if (skb->len == 0 || max_fragment_size == 0)
462 num_fragments = (skb->len - 1) / max_fragment_size + 1;
463 max_fragment_size = (skb->len - 1) / num_fragments + 1;
465 /* Don't even try to fragment, if we need more than 16 fragments */
466 if (num_fragments > BATADV_FRAG_MAX_FRAGMENTS) {
471 bat_priv = orig_node->bat_priv;
472 primary_if = batadv_primary_if_get_selected(bat_priv);
478 /* Create one header to be copied to all fragments */
479 frag_header.packet_type = BATADV_UNICAST_FRAG;
480 frag_header.version = BATADV_COMPAT_VERSION;
481 frag_header.ttl = BATADV_TTL;
482 frag_header.seqno = htons(atomic_inc_return(&bat_priv->frag_seqno));
483 frag_header.reserved = 0;
485 frag_header.total_size = htons(skb->len);
487 /* skb->priority values from 256->263 are magic values to
488 * directly indicate a specific 802.1d priority. This is used
489 * to allow 802.1d priority to be passed directly in from VLAN
492 if (skb->priority >= 256 && skb->priority <= 263)
493 frag_header.priority = skb->priority - 256;
495 frag_header.priority = 0;
497 ether_addr_copy(frag_header.orig, primary_if->net_dev->dev_addr);
498 ether_addr_copy(frag_header.dest, orig_node->orig);
500 /* Eat and send fragments from the tail of skb */
501 while (skb->len > max_fragment_size) {
502 /* The initial check in this function should cover this case */
503 if (unlikely(frag_header.no == BATADV_FRAG_MAX_FRAGMENTS - 1)) {
508 skb_fragment = batadv_frag_create(net_dev, skb, &frag_header,
515 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
516 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
517 skb_fragment->len + ETH_HLEN);
518 ret = batadv_send_unicast_skb(skb_fragment, neigh_node);
519 if (ret != NET_XMIT_SUCCESS) {
527 /* make sure that there is at least enough head for the fragmentation
528 * and ethernet headers
530 ret = skb_cow_head(skb, ETH_HLEN + header_size);
534 skb_push(skb, header_size);
535 memcpy(skb->data, &frag_header, header_size);
537 /* Send the last fragment */
538 batadv_inc_counter(bat_priv, BATADV_CNT_FRAG_TX);
539 batadv_add_counter(bat_priv, BATADV_CNT_FRAG_TX_BYTES,
540 skb->len + ETH_HLEN);
541 ret = batadv_send_unicast_skb(skb, neigh_node);
542 /* skb was consumed */
546 batadv_hardif_put(primary_if);