xref: /openbmc/linux/net/mac80211/mesh_pathtbl.c (revision 3b64b188)
1 /*
2  * Copyright (c) 2008, 2009 open80211s Ltd.
3  * Author:     Luis Carlos Cobo <luisca@cozybit.com>
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License version 2 as
7  * published by the Free Software Foundation.
8  */
9 
10 #include <linux/etherdevice.h>
11 #include <linux/list.h>
12 #include <linux/random.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/string.h>
16 #include <net/mac80211.h>
17 #include "wme.h"
18 #include "ieee80211_i.h"
19 #include "mesh.h"
20 
21 /* There will be initially 2^INIT_PATHS_SIZE_ORDER buckets */
22 #define INIT_PATHS_SIZE_ORDER	2
23 
24 /* Keep the mean chain length below this constant */
25 #define MEAN_CHAIN_LEN		2
26 
27 #define MPATH_EXPIRED(mpath) ((mpath->flags & MESH_PATH_ACTIVE) && \
28 				time_after(jiffies, mpath->exp_time) && \
29 				!(mpath->flags & MESH_PATH_FIXED))
30 
31 struct mpath_node {
32 	struct hlist_node list;
33 	struct rcu_head rcu;
34 	/* This indirection allows two different tables to point to the same
35 	 * mesh_path structure, useful when resizing
36 	 */
37 	struct mesh_path *mpath;
38 };
39 
40 static struct mesh_table __rcu *mesh_paths;
41 static struct mesh_table __rcu *mpp_paths; /* Store paths for MPP&MAP */
42 
43 int mesh_paths_generation;
44 
45 /* This lock will have the grow table function as writer and add / delete nodes
46  * as readers. RCU provides sufficient protection only when reading the table
47  * (i.e. doing lookups).  Adding or adding or removing nodes requires we take
48  * the read lock or we risk operating on an old table.  The write lock is only
49  * needed when modifying the number of buckets a table.
50  */
51 static DEFINE_RWLOCK(pathtbl_resize_lock);
52 
53 
54 static inline struct mesh_table *resize_dereference_mesh_paths(void)
55 {
56 	return rcu_dereference_protected(mesh_paths,
57 		lockdep_is_held(&pathtbl_resize_lock));
58 }
59 
60 static inline struct mesh_table *resize_dereference_mpp_paths(void)
61 {
62 	return rcu_dereference_protected(mpp_paths,
63 		lockdep_is_held(&pathtbl_resize_lock));
64 }
65 
66 /*
67  * CAREFUL -- "tbl" must not be an expression,
68  * in particular not an rcu_dereference(), since
69  * it's used twice. So it is illegal to do
70  *	for_each_mesh_entry(rcu_dereference(...), ...)
71  */
72 #define for_each_mesh_entry(tbl, p, node, i) \
73 	for (i = 0; i <= tbl->hash_mask; i++) \
74 		hlist_for_each_entry_rcu(node, p, &tbl->hash_buckets[i], list)
75 
76 
77 static struct mesh_table *mesh_table_alloc(int size_order)
78 {
79 	int i;
80 	struct mesh_table *newtbl;
81 
82 	newtbl = kmalloc(sizeof(struct mesh_table), GFP_ATOMIC);
83 	if (!newtbl)
84 		return NULL;
85 
86 	newtbl->hash_buckets = kzalloc(sizeof(struct hlist_head) *
87 			(1 << size_order), GFP_ATOMIC);
88 
89 	if (!newtbl->hash_buckets) {
90 		kfree(newtbl);
91 		return NULL;
92 	}
93 
94 	newtbl->hashwlock = kmalloc(sizeof(spinlock_t) *
95 			(1 << size_order), GFP_ATOMIC);
96 	if (!newtbl->hashwlock) {
97 		kfree(newtbl->hash_buckets);
98 		kfree(newtbl);
99 		return NULL;
100 	}
101 
102 	newtbl->size_order = size_order;
103 	newtbl->hash_mask = (1 << size_order) - 1;
104 	atomic_set(&newtbl->entries,  0);
105 	get_random_bytes(&newtbl->hash_rnd,
106 			sizeof(newtbl->hash_rnd));
107 	for (i = 0; i <= newtbl->hash_mask; i++)
108 		spin_lock_init(&newtbl->hashwlock[i]);
109 	spin_lock_init(&newtbl->gates_lock);
110 
111 	return newtbl;
112 }
113 
114 static void __mesh_table_free(struct mesh_table *tbl)
115 {
116 	kfree(tbl->hash_buckets);
117 	kfree(tbl->hashwlock);
118 	kfree(tbl);
119 }
120 
121 static void mesh_table_free(struct mesh_table *tbl, bool free_leafs)
122 {
123 	struct hlist_head *mesh_hash;
124 	struct hlist_node *p, *q;
125 	struct mpath_node *gate;
126 	int i;
127 
128 	mesh_hash = tbl->hash_buckets;
129 	for (i = 0; i <= tbl->hash_mask; i++) {
130 		spin_lock_bh(&tbl->hashwlock[i]);
131 		hlist_for_each_safe(p, q, &mesh_hash[i]) {
132 			tbl->free_node(p, free_leafs);
133 			atomic_dec(&tbl->entries);
134 		}
135 		spin_unlock_bh(&tbl->hashwlock[i]);
136 	}
137 	if (free_leafs) {
138 		spin_lock_bh(&tbl->gates_lock);
139 		hlist_for_each_entry_safe(gate, p, q,
140 					 tbl->known_gates, list) {
141 			hlist_del(&gate->list);
142 			kfree(gate);
143 		}
144 		kfree(tbl->known_gates);
145 		spin_unlock_bh(&tbl->gates_lock);
146 	}
147 
148 	__mesh_table_free(tbl);
149 }
150 
151 static int mesh_table_grow(struct mesh_table *oldtbl,
152 			   struct mesh_table *newtbl)
153 {
154 	struct hlist_head *oldhash;
155 	struct hlist_node *p, *q;
156 	int i;
157 
158 	if (atomic_read(&oldtbl->entries)
159 			< oldtbl->mean_chain_len * (oldtbl->hash_mask + 1))
160 		return -EAGAIN;
161 
162 	newtbl->free_node = oldtbl->free_node;
163 	newtbl->mean_chain_len = oldtbl->mean_chain_len;
164 	newtbl->copy_node = oldtbl->copy_node;
165 	newtbl->known_gates = oldtbl->known_gates;
166 	atomic_set(&newtbl->entries, atomic_read(&oldtbl->entries));
167 
168 	oldhash = oldtbl->hash_buckets;
169 	for (i = 0; i <= oldtbl->hash_mask; i++)
170 		hlist_for_each(p, &oldhash[i])
171 			if (oldtbl->copy_node(p, newtbl) < 0)
172 				goto errcopy;
173 
174 	return 0;
175 
176 errcopy:
177 	for (i = 0; i <= newtbl->hash_mask; i++) {
178 		hlist_for_each_safe(p, q, &newtbl->hash_buckets[i])
179 			oldtbl->free_node(p, 0);
180 	}
181 	return -ENOMEM;
182 }
183 
184 static u32 mesh_table_hash(u8 *addr, struct ieee80211_sub_if_data *sdata,
185 			   struct mesh_table *tbl)
186 {
187 	/* Use last four bytes of hw addr and interface index as hash index */
188 	return jhash_2words(*(u32 *)(addr+2), sdata->dev->ifindex, tbl->hash_rnd)
189 		& tbl->hash_mask;
190 }
191 
192 
193 /**
194  *
195  * mesh_path_assign_nexthop - update mesh path next hop
196  *
197  * @mpath: mesh path to update
198  * @sta: next hop to assign
199  *
200  * Locking: mpath->state_lock must be held when calling this function
201  */
202 void mesh_path_assign_nexthop(struct mesh_path *mpath, struct sta_info *sta)
203 {
204 	struct sk_buff *skb;
205 	struct ieee80211_hdr *hdr;
206 	unsigned long flags;
207 
208 	rcu_assign_pointer(mpath->next_hop, sta);
209 
210 	spin_lock_irqsave(&mpath->frame_queue.lock, flags);
211 	skb_queue_walk(&mpath->frame_queue, skb) {
212 		hdr = (struct ieee80211_hdr *) skb->data;
213 		memcpy(hdr->addr1, sta->sta.addr, ETH_ALEN);
214 		memcpy(hdr->addr2, mpath->sdata->vif.addr, ETH_ALEN);
215 	}
216 
217 	spin_unlock_irqrestore(&mpath->frame_queue.lock, flags);
218 }
219 
220 static void prepare_for_gate(struct sk_buff *skb, char *dst_addr,
221 			     struct mesh_path *gate_mpath)
222 {
223 	struct ieee80211_hdr *hdr;
224 	struct ieee80211s_hdr *mshdr;
225 	int mesh_hdrlen, hdrlen;
226 	char *next_hop;
227 
228 	hdr = (struct ieee80211_hdr *) skb->data;
229 	hdrlen = ieee80211_hdrlen(hdr->frame_control);
230 	mshdr = (struct ieee80211s_hdr *) (skb->data + hdrlen);
231 
232 	if (!(mshdr->flags & MESH_FLAGS_AE)) {
233 		/* size of the fixed part of the mesh header */
234 		mesh_hdrlen = 6;
235 
236 		/* make room for the two extended addresses */
237 		skb_push(skb, 2 * ETH_ALEN);
238 		memmove(skb->data, hdr, hdrlen + mesh_hdrlen);
239 
240 		hdr = (struct ieee80211_hdr *) skb->data;
241 
242 		/* we preserve the previous mesh header and only add
243 		 * the new addreses */
244 		mshdr = (struct ieee80211s_hdr *) (skb->data + hdrlen);
245 		mshdr->flags = MESH_FLAGS_AE_A5_A6;
246 		memcpy(mshdr->eaddr1, hdr->addr3, ETH_ALEN);
247 		memcpy(mshdr->eaddr2, hdr->addr4, ETH_ALEN);
248 	}
249 
250 	/* update next hop */
251 	hdr = (struct ieee80211_hdr *) skb->data;
252 	rcu_read_lock();
253 	next_hop = rcu_dereference(gate_mpath->next_hop)->sta.addr;
254 	memcpy(hdr->addr1, next_hop, ETH_ALEN);
255 	rcu_read_unlock();
256 	memcpy(hdr->addr2, gate_mpath->sdata->vif.addr, ETH_ALEN);
257 	memcpy(hdr->addr3, dst_addr, ETH_ALEN);
258 }
259 
260 /**
261  *
262  * mesh_path_move_to_queue - Move or copy frames from one mpath queue to another
263  *
264  * This function is used to transfer or copy frames from an unresolved mpath to
265  * a gate mpath.  The function also adds the Address Extension field and
266  * updates the next hop.
267  *
268  * If a frame already has an Address Extension field, only the next hop and
269  * destination addresses are updated.
270  *
271  * The gate mpath must be an active mpath with a valid mpath->next_hop.
272  *
273  * @mpath: An active mpath the frames will be sent to (i.e. the gate)
274  * @from_mpath: The failed mpath
275  * @copy: When true, copy all the frames to the new mpath queue.  When false,
276  * move them.
277  */
278 static void mesh_path_move_to_queue(struct mesh_path *gate_mpath,
279 				    struct mesh_path *from_mpath,
280 				    bool copy)
281 {
282 	struct sk_buff *skb, *fskb, *tmp;
283 	struct sk_buff_head failq;
284 	unsigned long flags;
285 
286 	BUG_ON(gate_mpath == from_mpath);
287 	BUG_ON(!gate_mpath->next_hop);
288 
289 	__skb_queue_head_init(&failq);
290 
291 	spin_lock_irqsave(&from_mpath->frame_queue.lock, flags);
292 	skb_queue_splice_init(&from_mpath->frame_queue, &failq);
293 	spin_unlock_irqrestore(&from_mpath->frame_queue.lock, flags);
294 
295 	skb_queue_walk_safe(&failq, fskb, tmp) {
296 		if (skb_queue_len(&gate_mpath->frame_queue) >=
297 				  MESH_FRAME_QUEUE_LEN) {
298 			mpath_dbg(gate_mpath->sdata, "mpath queue full!\n");
299 			break;
300 		}
301 
302 		skb = skb_copy(fskb, GFP_ATOMIC);
303 		if (WARN_ON(!skb))
304 			break;
305 
306 		prepare_for_gate(skb, gate_mpath->dst, gate_mpath);
307 		skb_queue_tail(&gate_mpath->frame_queue, skb);
308 
309 		if (copy)
310 			continue;
311 
312 		__skb_unlink(fskb, &failq);
313 		kfree_skb(fskb);
314 	}
315 
316 	mpath_dbg(gate_mpath->sdata, "Mpath queue for gate %pM has %d frames\n",
317 		  gate_mpath->dst, skb_queue_len(&gate_mpath->frame_queue));
318 
319 	if (!copy)
320 		return;
321 
322 	spin_lock_irqsave(&from_mpath->frame_queue.lock, flags);
323 	skb_queue_splice(&failq, &from_mpath->frame_queue);
324 	spin_unlock_irqrestore(&from_mpath->frame_queue.lock, flags);
325 }
326 
327 
328 static struct mesh_path *mpath_lookup(struct mesh_table *tbl, u8 *dst,
329 					  struct ieee80211_sub_if_data *sdata)
330 {
331 	struct mesh_path *mpath;
332 	struct hlist_node *n;
333 	struct hlist_head *bucket;
334 	struct mpath_node *node;
335 
336 	bucket = &tbl->hash_buckets[mesh_table_hash(dst, sdata, tbl)];
337 	hlist_for_each_entry_rcu(node, n, bucket, list) {
338 		mpath = node->mpath;
339 		if (mpath->sdata == sdata &&
340 		    ether_addr_equal(dst, mpath->dst)) {
341 			if (MPATH_EXPIRED(mpath)) {
342 				spin_lock_bh(&mpath->state_lock);
343 				mpath->flags &= ~MESH_PATH_ACTIVE;
344 				spin_unlock_bh(&mpath->state_lock);
345 			}
346 			return mpath;
347 		}
348 	}
349 	return NULL;
350 }
351 
352 /**
353  * mesh_path_lookup - look up a path in the mesh path table
354  * @dst: hardware address (ETH_ALEN length) of destination
355  * @sdata: local subif
356  *
357  * Returns: pointer to the mesh path structure, or NULL if not found
358  *
359  * Locking: must be called within a read rcu section.
360  */
361 struct mesh_path *mesh_path_lookup(u8 *dst, struct ieee80211_sub_if_data *sdata)
362 {
363 	return mpath_lookup(rcu_dereference(mesh_paths), dst, sdata);
364 }
365 
366 struct mesh_path *mpp_path_lookup(u8 *dst, struct ieee80211_sub_if_data *sdata)
367 {
368 	return mpath_lookup(rcu_dereference(mpp_paths), dst, sdata);
369 }
370 
371 
372 /**
373  * mesh_path_lookup_by_idx - look up a path in the mesh path table by its index
374  * @idx: index
375  * @sdata: local subif, or NULL for all entries
376  *
377  * Returns: pointer to the mesh path structure, or NULL if not found.
378  *
379  * Locking: must be called within a read rcu section.
380  */
381 struct mesh_path *mesh_path_lookup_by_idx(int idx, struct ieee80211_sub_if_data *sdata)
382 {
383 	struct mesh_table *tbl = rcu_dereference(mesh_paths);
384 	struct mpath_node *node;
385 	struct hlist_node *p;
386 	int i;
387 	int j = 0;
388 
389 	for_each_mesh_entry(tbl, p, node, i) {
390 		if (sdata && node->mpath->sdata != sdata)
391 			continue;
392 		if (j++ == idx) {
393 			if (MPATH_EXPIRED(node->mpath)) {
394 				spin_lock_bh(&node->mpath->state_lock);
395 				node->mpath->flags &= ~MESH_PATH_ACTIVE;
396 				spin_unlock_bh(&node->mpath->state_lock);
397 			}
398 			return node->mpath;
399 		}
400 	}
401 
402 	return NULL;
403 }
404 
405 /**
406  * mesh_path_add_gate - add the given mpath to a mesh gate to our path table
407  * @mpath: gate path to add to table
408  */
409 int mesh_path_add_gate(struct mesh_path *mpath)
410 {
411 	struct mesh_table *tbl;
412 	struct mpath_node *gate, *new_gate;
413 	struct hlist_node *n;
414 	int err;
415 
416 	rcu_read_lock();
417 	tbl = rcu_dereference(mesh_paths);
418 
419 	hlist_for_each_entry_rcu(gate, n, tbl->known_gates, list)
420 		if (gate->mpath == mpath) {
421 			err = -EEXIST;
422 			goto err_rcu;
423 		}
424 
425 	new_gate = kzalloc(sizeof(struct mpath_node), GFP_ATOMIC);
426 	if (!new_gate) {
427 		err = -ENOMEM;
428 		goto err_rcu;
429 	}
430 
431 	mpath->is_gate = true;
432 	mpath->sdata->u.mesh.num_gates++;
433 	new_gate->mpath = mpath;
434 	spin_lock_bh(&tbl->gates_lock);
435 	hlist_add_head_rcu(&new_gate->list, tbl->known_gates);
436 	spin_unlock_bh(&tbl->gates_lock);
437 	rcu_read_unlock();
438 	mpath_dbg(mpath->sdata,
439 		  "Mesh path: Recorded new gate: %pM. %d known gates\n",
440 		  mpath->dst, mpath->sdata->u.mesh.num_gates);
441 	return 0;
442 err_rcu:
443 	rcu_read_unlock();
444 	return err;
445 }
446 
447 /**
448  * mesh_gate_del - remove a mesh gate from the list of known gates
449  * @tbl: table which holds our list of known gates
450  * @mpath: gate mpath
451  *
452  * Returns: 0 on success
453  *
454  * Locking: must be called inside rcu_read_lock() section
455  */
456 static int mesh_gate_del(struct mesh_table *tbl, struct mesh_path *mpath)
457 {
458 	struct mpath_node *gate;
459 	struct hlist_node *p, *q;
460 
461 	hlist_for_each_entry_safe(gate, p, q, tbl->known_gates, list)
462 		if (gate->mpath == mpath) {
463 			spin_lock_bh(&tbl->gates_lock);
464 			hlist_del_rcu(&gate->list);
465 			kfree_rcu(gate, rcu);
466 			spin_unlock_bh(&tbl->gates_lock);
467 			mpath->sdata->u.mesh.num_gates--;
468 			mpath->is_gate = false;
469 			mpath_dbg(mpath->sdata,
470 				  "Mesh path: Deleted gate: %pM. %d known gates\n",
471 				  mpath->dst, mpath->sdata->u.mesh.num_gates);
472 			break;
473 		}
474 
475 	return 0;
476 }
477 
478 /**
479  * mesh_gate_num - number of gates known to this interface
480  * @sdata: subif data
481  */
482 int mesh_gate_num(struct ieee80211_sub_if_data *sdata)
483 {
484 	return sdata->u.mesh.num_gates;
485 }
486 
487 /**
488  * mesh_path_add - allocate and add a new path to the mesh path table
489  * @addr: destination address of the path (ETH_ALEN length)
490  * @sdata: local subif
491  *
492  * Returns: 0 on success
493  *
494  * State: the initial state of the new path is set to 0
495  */
496 int mesh_path_add(u8 *dst, struct ieee80211_sub_if_data *sdata)
497 {
498 	struct ieee80211_if_mesh *ifmsh = &sdata->u.mesh;
499 	struct ieee80211_local *local = sdata->local;
500 	struct mesh_table *tbl;
501 	struct mesh_path *mpath, *new_mpath;
502 	struct mpath_node *node, *new_node;
503 	struct hlist_head *bucket;
504 	struct hlist_node *n;
505 	int grow = 0;
506 	int err = 0;
507 	u32 hash_idx;
508 
509 	if (ether_addr_equal(dst, sdata->vif.addr))
510 		/* never add ourselves as neighbours */
511 		return -ENOTSUPP;
512 
513 	if (is_multicast_ether_addr(dst))
514 		return -ENOTSUPP;
515 
516 	if (atomic_add_unless(&sdata->u.mesh.mpaths, 1, MESH_MAX_MPATHS) == 0)
517 		return -ENOSPC;
518 
519 	err = -ENOMEM;
520 	new_mpath = kzalloc(sizeof(struct mesh_path), GFP_ATOMIC);
521 	if (!new_mpath)
522 		goto err_path_alloc;
523 
524 	new_node = kmalloc(sizeof(struct mpath_node), GFP_ATOMIC);
525 	if (!new_node)
526 		goto err_node_alloc;
527 
528 	read_lock_bh(&pathtbl_resize_lock);
529 	memcpy(new_mpath->dst, dst, ETH_ALEN);
530 	eth_broadcast_addr(new_mpath->rann_snd_addr);
531 	new_mpath->is_root = false;
532 	new_mpath->sdata = sdata;
533 	new_mpath->flags = 0;
534 	skb_queue_head_init(&new_mpath->frame_queue);
535 	new_node->mpath = new_mpath;
536 	new_mpath->timer.data = (unsigned long) new_mpath;
537 	new_mpath->timer.function = mesh_path_timer;
538 	new_mpath->exp_time = jiffies;
539 	spin_lock_init(&new_mpath->state_lock);
540 	init_timer(&new_mpath->timer);
541 
542 	tbl = resize_dereference_mesh_paths();
543 
544 	hash_idx = mesh_table_hash(dst, sdata, tbl);
545 	bucket = &tbl->hash_buckets[hash_idx];
546 
547 	spin_lock(&tbl->hashwlock[hash_idx]);
548 
549 	err = -EEXIST;
550 	hlist_for_each_entry(node, n, bucket, list) {
551 		mpath = node->mpath;
552 		if (mpath->sdata == sdata &&
553 		    ether_addr_equal(dst, mpath->dst))
554 			goto err_exists;
555 	}
556 
557 	hlist_add_head_rcu(&new_node->list, bucket);
558 	if (atomic_inc_return(&tbl->entries) >=
559 	    tbl->mean_chain_len * (tbl->hash_mask + 1))
560 		grow = 1;
561 
562 	mesh_paths_generation++;
563 
564 	spin_unlock(&tbl->hashwlock[hash_idx]);
565 	read_unlock_bh(&pathtbl_resize_lock);
566 	if (grow) {
567 		set_bit(MESH_WORK_GROW_MPATH_TABLE,  &ifmsh->wrkq_flags);
568 		ieee80211_queue_work(&local->hw, &sdata->work);
569 	}
570 	return 0;
571 
572 err_exists:
573 	spin_unlock(&tbl->hashwlock[hash_idx]);
574 	read_unlock_bh(&pathtbl_resize_lock);
575 	kfree(new_node);
576 err_node_alloc:
577 	kfree(new_mpath);
578 err_path_alloc:
579 	atomic_dec(&sdata->u.mesh.mpaths);
580 	return err;
581 }
582 
583 static void mesh_table_free_rcu(struct rcu_head *rcu)
584 {
585 	struct mesh_table *tbl = container_of(rcu, struct mesh_table, rcu_head);
586 
587 	mesh_table_free(tbl, false);
588 }
589 
590 void mesh_mpath_table_grow(void)
591 {
592 	struct mesh_table *oldtbl, *newtbl;
593 
594 	write_lock_bh(&pathtbl_resize_lock);
595 	oldtbl = resize_dereference_mesh_paths();
596 	newtbl = mesh_table_alloc(oldtbl->size_order + 1);
597 	if (!newtbl)
598 		goto out;
599 	if (mesh_table_grow(oldtbl, newtbl) < 0) {
600 		__mesh_table_free(newtbl);
601 		goto out;
602 	}
603 	rcu_assign_pointer(mesh_paths, newtbl);
604 
605 	call_rcu(&oldtbl->rcu_head, mesh_table_free_rcu);
606 
607  out:
608 	write_unlock_bh(&pathtbl_resize_lock);
609 }
610 
611 void mesh_mpp_table_grow(void)
612 {
613 	struct mesh_table *oldtbl, *newtbl;
614 
615 	write_lock_bh(&pathtbl_resize_lock);
616 	oldtbl = resize_dereference_mpp_paths();
617 	newtbl = mesh_table_alloc(oldtbl->size_order + 1);
618 	if (!newtbl)
619 		goto out;
620 	if (mesh_table_grow(oldtbl, newtbl) < 0) {
621 		__mesh_table_free(newtbl);
622 		goto out;
623 	}
624 	rcu_assign_pointer(mpp_paths, newtbl);
625 	call_rcu(&oldtbl->rcu_head, mesh_table_free_rcu);
626 
627  out:
628 	write_unlock_bh(&pathtbl_resize_lock);
629 }
630 
631 int mpp_path_add(u8 *dst, u8 *mpp, struct ieee80211_sub_if_data *sdata)
632 {
633 	struct ieee80211_if_mesh *ifmsh = &sdata->u.mesh;
634 	struct ieee80211_local *local = sdata->local;
635 	struct mesh_table *tbl;
636 	struct mesh_path *mpath, *new_mpath;
637 	struct mpath_node *node, *new_node;
638 	struct hlist_head *bucket;
639 	struct hlist_node *n;
640 	int grow = 0;
641 	int err = 0;
642 	u32 hash_idx;
643 
644 	if (ether_addr_equal(dst, sdata->vif.addr))
645 		/* never add ourselves as neighbours */
646 		return -ENOTSUPP;
647 
648 	if (is_multicast_ether_addr(dst))
649 		return -ENOTSUPP;
650 
651 	err = -ENOMEM;
652 	new_mpath = kzalloc(sizeof(struct mesh_path), GFP_ATOMIC);
653 	if (!new_mpath)
654 		goto err_path_alloc;
655 
656 	new_node = kmalloc(sizeof(struct mpath_node), GFP_ATOMIC);
657 	if (!new_node)
658 		goto err_node_alloc;
659 
660 	read_lock_bh(&pathtbl_resize_lock);
661 	memcpy(new_mpath->dst, dst, ETH_ALEN);
662 	memcpy(new_mpath->mpp, mpp, ETH_ALEN);
663 	new_mpath->sdata = sdata;
664 	new_mpath->flags = 0;
665 	skb_queue_head_init(&new_mpath->frame_queue);
666 	new_node->mpath = new_mpath;
667 	init_timer(&new_mpath->timer);
668 	new_mpath->exp_time = jiffies;
669 	spin_lock_init(&new_mpath->state_lock);
670 
671 	tbl = resize_dereference_mpp_paths();
672 
673 	hash_idx = mesh_table_hash(dst, sdata, tbl);
674 	bucket = &tbl->hash_buckets[hash_idx];
675 
676 	spin_lock(&tbl->hashwlock[hash_idx]);
677 
678 	err = -EEXIST;
679 	hlist_for_each_entry(node, n, bucket, list) {
680 		mpath = node->mpath;
681 		if (mpath->sdata == sdata &&
682 		    ether_addr_equal(dst, mpath->dst))
683 			goto err_exists;
684 	}
685 
686 	hlist_add_head_rcu(&new_node->list, bucket);
687 	if (atomic_inc_return(&tbl->entries) >=
688 	    tbl->mean_chain_len * (tbl->hash_mask + 1))
689 		grow = 1;
690 
691 	spin_unlock(&tbl->hashwlock[hash_idx]);
692 	read_unlock_bh(&pathtbl_resize_lock);
693 	if (grow) {
694 		set_bit(MESH_WORK_GROW_MPP_TABLE,  &ifmsh->wrkq_flags);
695 		ieee80211_queue_work(&local->hw, &sdata->work);
696 	}
697 	return 0;
698 
699 err_exists:
700 	spin_unlock(&tbl->hashwlock[hash_idx]);
701 	read_unlock_bh(&pathtbl_resize_lock);
702 	kfree(new_node);
703 err_node_alloc:
704 	kfree(new_mpath);
705 err_path_alloc:
706 	return err;
707 }
708 
709 
710 /**
711  * mesh_plink_broken - deactivates paths and sends perr when a link breaks
712  *
713  * @sta: broken peer link
714  *
715  * This function must be called from the rate control algorithm if enough
716  * delivery errors suggest that a peer link is no longer usable.
717  */
718 void mesh_plink_broken(struct sta_info *sta)
719 {
720 	struct mesh_table *tbl;
721 	static const u8 bcast[ETH_ALEN] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
722 	struct mesh_path *mpath;
723 	struct mpath_node *node;
724 	struct hlist_node *p;
725 	struct ieee80211_sub_if_data *sdata = sta->sdata;
726 	int i;
727 	__le16 reason = cpu_to_le16(WLAN_REASON_MESH_PATH_DEST_UNREACHABLE);
728 
729 	rcu_read_lock();
730 	tbl = rcu_dereference(mesh_paths);
731 	for_each_mesh_entry(tbl, p, node, i) {
732 		mpath = node->mpath;
733 		if (rcu_dereference(mpath->next_hop) == sta &&
734 		    mpath->flags & MESH_PATH_ACTIVE &&
735 		    !(mpath->flags & MESH_PATH_FIXED)) {
736 			spin_lock_bh(&mpath->state_lock);
737 			mpath->flags &= ~MESH_PATH_ACTIVE;
738 			++mpath->sn;
739 			spin_unlock_bh(&mpath->state_lock);
740 			mesh_path_error_tx(sdata->u.mesh.mshcfg.element_ttl,
741 					mpath->dst, cpu_to_le32(mpath->sn),
742 					reason, bcast, sdata);
743 		}
744 	}
745 	rcu_read_unlock();
746 }
747 
748 static void mesh_path_node_reclaim(struct rcu_head *rp)
749 {
750 	struct mpath_node *node = container_of(rp, struct mpath_node, rcu);
751 	struct ieee80211_sub_if_data *sdata = node->mpath->sdata;
752 
753 	del_timer_sync(&node->mpath->timer);
754 	atomic_dec(&sdata->u.mesh.mpaths);
755 	kfree(node->mpath);
756 	kfree(node);
757 }
758 
759 /* needs to be called with the corresponding hashwlock taken */
760 static void __mesh_path_del(struct mesh_table *tbl, struct mpath_node *node)
761 {
762 	struct mesh_path *mpath;
763 	mpath = node->mpath;
764 	spin_lock(&mpath->state_lock);
765 	mpath->flags |= MESH_PATH_RESOLVING;
766 	if (mpath->is_gate)
767 		mesh_gate_del(tbl, mpath);
768 	hlist_del_rcu(&node->list);
769 	call_rcu(&node->rcu, mesh_path_node_reclaim);
770 	spin_unlock(&mpath->state_lock);
771 	atomic_dec(&tbl->entries);
772 }
773 
774 /**
775  * mesh_path_flush_by_nexthop - Deletes mesh paths if their next hop matches
776  *
777  * @sta: mesh peer to match
778  *
779  * RCU notes: this function is called when a mesh plink transitions from
780  * PLINK_ESTAB to any other state, since PLINK_ESTAB state is the only one that
781  * allows path creation. This will happen before the sta can be freed (because
782  * sta_info_destroy() calls this) so any reader in a rcu read block will be
783  * protected against the plink disappearing.
784  */
785 void mesh_path_flush_by_nexthop(struct sta_info *sta)
786 {
787 	struct mesh_table *tbl;
788 	struct mesh_path *mpath;
789 	struct mpath_node *node;
790 	struct hlist_node *p;
791 	int i;
792 
793 	rcu_read_lock();
794 	read_lock_bh(&pathtbl_resize_lock);
795 	tbl = resize_dereference_mesh_paths();
796 	for_each_mesh_entry(tbl, p, node, i) {
797 		mpath = node->mpath;
798 		if (rcu_dereference(mpath->next_hop) == sta) {
799 			spin_lock(&tbl->hashwlock[i]);
800 			__mesh_path_del(tbl, node);
801 			spin_unlock(&tbl->hashwlock[i]);
802 		}
803 	}
804 	read_unlock_bh(&pathtbl_resize_lock);
805 	rcu_read_unlock();
806 }
807 
808 static void table_flush_by_iface(struct mesh_table *tbl,
809 				 struct ieee80211_sub_if_data *sdata)
810 {
811 	struct mesh_path *mpath;
812 	struct mpath_node *node;
813 	struct hlist_node *p;
814 	int i;
815 
816 	WARN_ON(!rcu_read_lock_held());
817 	for_each_mesh_entry(tbl, p, node, i) {
818 		mpath = node->mpath;
819 		if (mpath->sdata != sdata)
820 			continue;
821 		spin_lock_bh(&tbl->hashwlock[i]);
822 		__mesh_path_del(tbl, node);
823 		spin_unlock_bh(&tbl->hashwlock[i]);
824 	}
825 }
826 
827 /**
828  * mesh_path_flush_by_iface - Deletes all mesh paths associated with a given iface
829  *
830  * This function deletes both mesh paths as well as mesh portal paths.
831  *
832  * @sdata: interface data to match
833  *
834  */
835 void mesh_path_flush_by_iface(struct ieee80211_sub_if_data *sdata)
836 {
837 	struct mesh_table *tbl;
838 
839 	rcu_read_lock();
840 	read_lock_bh(&pathtbl_resize_lock);
841 	tbl = resize_dereference_mesh_paths();
842 	table_flush_by_iface(tbl, sdata);
843 	tbl = resize_dereference_mpp_paths();
844 	table_flush_by_iface(tbl, sdata);
845 	read_unlock_bh(&pathtbl_resize_lock);
846 	rcu_read_unlock();
847 }
848 
849 /**
850  * mesh_path_del - delete a mesh path from the table
851  *
852  * @addr: dst address (ETH_ALEN length)
853  * @sdata: local subif
854  *
855  * Returns: 0 if successful
856  */
857 int mesh_path_del(u8 *addr, struct ieee80211_sub_if_data *sdata)
858 {
859 	struct mesh_table *tbl;
860 	struct mesh_path *mpath;
861 	struct mpath_node *node;
862 	struct hlist_head *bucket;
863 	struct hlist_node *n;
864 	int hash_idx;
865 	int err = 0;
866 
867 	read_lock_bh(&pathtbl_resize_lock);
868 	tbl = resize_dereference_mesh_paths();
869 	hash_idx = mesh_table_hash(addr, sdata, tbl);
870 	bucket = &tbl->hash_buckets[hash_idx];
871 
872 	spin_lock(&tbl->hashwlock[hash_idx]);
873 	hlist_for_each_entry(node, n, bucket, list) {
874 		mpath = node->mpath;
875 		if (mpath->sdata == sdata &&
876 		    ether_addr_equal(addr, mpath->dst)) {
877 			__mesh_path_del(tbl, node);
878 			goto enddel;
879 		}
880 	}
881 
882 	err = -ENXIO;
883 enddel:
884 	mesh_paths_generation++;
885 	spin_unlock(&tbl->hashwlock[hash_idx]);
886 	read_unlock_bh(&pathtbl_resize_lock);
887 	return err;
888 }
889 
890 /**
891  * mesh_path_tx_pending - sends pending frames in a mesh path queue
892  *
893  * @mpath: mesh path to activate
894  *
895  * Locking: the state_lock of the mpath structure must NOT be held when calling
896  * this function.
897  */
898 void mesh_path_tx_pending(struct mesh_path *mpath)
899 {
900 	if (mpath->flags & MESH_PATH_ACTIVE)
901 		ieee80211_add_pending_skbs(mpath->sdata->local,
902 				&mpath->frame_queue);
903 }
904 
905 /**
906  * mesh_path_send_to_gates - sends pending frames to all known mesh gates
907  *
908  * @mpath: mesh path whose queue will be emptied
909  *
910  * If there is only one gate, the frames are transferred from the failed mpath
911  * queue to that gate's queue.  If there are more than one gates, the frames
912  * are copied from each gate to the next.  After frames are copied, the
913  * mpath queues are emptied onto the transmission queue.
914  */
915 int mesh_path_send_to_gates(struct mesh_path *mpath)
916 {
917 	struct ieee80211_sub_if_data *sdata = mpath->sdata;
918 	struct hlist_node *n;
919 	struct mesh_table *tbl;
920 	struct mesh_path *from_mpath = mpath;
921 	struct mpath_node *gate = NULL;
922 	bool copy = false;
923 	struct hlist_head *known_gates;
924 
925 	rcu_read_lock();
926 	tbl = rcu_dereference(mesh_paths);
927 	known_gates = tbl->known_gates;
928 	rcu_read_unlock();
929 
930 	if (!known_gates)
931 		return -EHOSTUNREACH;
932 
933 	hlist_for_each_entry_rcu(gate, n, known_gates, list) {
934 		if (gate->mpath->sdata != sdata)
935 			continue;
936 
937 		if (gate->mpath->flags & MESH_PATH_ACTIVE) {
938 			mpath_dbg(sdata, "Forwarding to %pM\n", gate->mpath->dst);
939 			mesh_path_move_to_queue(gate->mpath, from_mpath, copy);
940 			from_mpath = gate->mpath;
941 			copy = true;
942 		} else {
943 			mpath_dbg(sdata,
944 				  "Not forwarding %p (flags %#x)\n",
945 				  gate->mpath, gate->mpath->flags);
946 		}
947 	}
948 
949 	hlist_for_each_entry_rcu(gate, n, known_gates, list)
950 		if (gate->mpath->sdata == sdata) {
951 			mpath_dbg(sdata, "Sending to %pM\n", gate->mpath->dst);
952 			mesh_path_tx_pending(gate->mpath);
953 		}
954 
955 	return (from_mpath == mpath) ? -EHOSTUNREACH : 0;
956 }
957 
958 /**
959  * mesh_path_discard_frame - discard a frame whose path could not be resolved
960  *
961  * @skb: frame to discard
962  * @sdata: network subif the frame was to be sent through
963  *
964  * Locking: the function must me called within a rcu_read_lock region
965  */
966 void mesh_path_discard_frame(struct sk_buff *skb,
967 			     struct ieee80211_sub_if_data *sdata)
968 {
969 	kfree_skb(skb);
970 	sdata->u.mesh.mshstats.dropped_frames_no_route++;
971 }
972 
973 /**
974  * mesh_path_flush_pending - free the pending queue of a mesh path
975  *
976  * @mpath: mesh path whose queue has to be freed
977  *
978  * Locking: the function must me called within a rcu_read_lock region
979  */
980 void mesh_path_flush_pending(struct mesh_path *mpath)
981 {
982 	struct sk_buff *skb;
983 
984 	while ((skb = skb_dequeue(&mpath->frame_queue)) != NULL)
985 		mesh_path_discard_frame(skb, mpath->sdata);
986 }
987 
988 /**
989  * mesh_path_fix_nexthop - force a specific next hop for a mesh path
990  *
991  * @mpath: the mesh path to modify
992  * @next_hop: the next hop to force
993  *
994  * Locking: this function must be called holding mpath->state_lock
995  */
996 void mesh_path_fix_nexthop(struct mesh_path *mpath, struct sta_info *next_hop)
997 {
998 	spin_lock_bh(&mpath->state_lock);
999 	mesh_path_assign_nexthop(mpath, next_hop);
1000 	mpath->sn = 0xffff;
1001 	mpath->metric = 0;
1002 	mpath->hop_count = 0;
1003 	mpath->exp_time = 0;
1004 	mpath->flags |= MESH_PATH_FIXED;
1005 	mesh_path_activate(mpath);
1006 	spin_unlock_bh(&mpath->state_lock);
1007 	mesh_path_tx_pending(mpath);
1008 }
1009 
1010 static void mesh_path_node_free(struct hlist_node *p, bool free_leafs)
1011 {
1012 	struct mesh_path *mpath;
1013 	struct mpath_node *node = hlist_entry(p, struct mpath_node, list);
1014 	mpath = node->mpath;
1015 	hlist_del_rcu(p);
1016 	if (free_leafs) {
1017 		del_timer_sync(&mpath->timer);
1018 		kfree(mpath);
1019 	}
1020 	kfree(node);
1021 }
1022 
1023 static int mesh_path_node_copy(struct hlist_node *p, struct mesh_table *newtbl)
1024 {
1025 	struct mesh_path *mpath;
1026 	struct mpath_node *node, *new_node;
1027 	u32 hash_idx;
1028 
1029 	new_node = kmalloc(sizeof(struct mpath_node), GFP_ATOMIC);
1030 	if (new_node == NULL)
1031 		return -ENOMEM;
1032 
1033 	node = hlist_entry(p, struct mpath_node, list);
1034 	mpath = node->mpath;
1035 	new_node->mpath = mpath;
1036 	hash_idx = mesh_table_hash(mpath->dst, mpath->sdata, newtbl);
1037 	hlist_add_head(&new_node->list,
1038 			&newtbl->hash_buckets[hash_idx]);
1039 	return 0;
1040 }
1041 
1042 int mesh_pathtbl_init(void)
1043 {
1044 	struct mesh_table *tbl_path, *tbl_mpp;
1045 	int ret;
1046 
1047 	tbl_path = mesh_table_alloc(INIT_PATHS_SIZE_ORDER);
1048 	if (!tbl_path)
1049 		return -ENOMEM;
1050 	tbl_path->free_node = &mesh_path_node_free;
1051 	tbl_path->copy_node = &mesh_path_node_copy;
1052 	tbl_path->mean_chain_len = MEAN_CHAIN_LEN;
1053 	tbl_path->known_gates = kzalloc(sizeof(struct hlist_head), GFP_ATOMIC);
1054 	if (!tbl_path->known_gates) {
1055 		ret = -ENOMEM;
1056 		goto free_path;
1057 	}
1058 	INIT_HLIST_HEAD(tbl_path->known_gates);
1059 
1060 
1061 	tbl_mpp = mesh_table_alloc(INIT_PATHS_SIZE_ORDER);
1062 	if (!tbl_mpp) {
1063 		ret = -ENOMEM;
1064 		goto free_path;
1065 	}
1066 	tbl_mpp->free_node = &mesh_path_node_free;
1067 	tbl_mpp->copy_node = &mesh_path_node_copy;
1068 	tbl_mpp->mean_chain_len = MEAN_CHAIN_LEN;
1069 	tbl_mpp->known_gates = kzalloc(sizeof(struct hlist_head), GFP_ATOMIC);
1070 	if (!tbl_mpp->known_gates) {
1071 		ret = -ENOMEM;
1072 		goto free_mpp;
1073 	}
1074 	INIT_HLIST_HEAD(tbl_mpp->known_gates);
1075 
1076 	/* Need no locking since this is during init */
1077 	RCU_INIT_POINTER(mesh_paths, tbl_path);
1078 	RCU_INIT_POINTER(mpp_paths, tbl_mpp);
1079 
1080 	return 0;
1081 
1082 free_mpp:
1083 	mesh_table_free(tbl_mpp, true);
1084 free_path:
1085 	mesh_table_free(tbl_path, true);
1086 	return ret;
1087 }
1088 
1089 void mesh_path_expire(struct ieee80211_sub_if_data *sdata)
1090 {
1091 	struct mesh_table *tbl;
1092 	struct mesh_path *mpath;
1093 	struct mpath_node *node;
1094 	struct hlist_node *p;
1095 	int i;
1096 
1097 	rcu_read_lock();
1098 	tbl = rcu_dereference(mesh_paths);
1099 	for_each_mesh_entry(tbl, p, node, i) {
1100 		if (node->mpath->sdata != sdata)
1101 			continue;
1102 		mpath = node->mpath;
1103 		if ((!(mpath->flags & MESH_PATH_RESOLVING)) &&
1104 		    (!(mpath->flags & MESH_PATH_FIXED)) &&
1105 		     time_after(jiffies, mpath->exp_time + MESH_PATH_EXPIRE))
1106 			mesh_path_del(mpath->dst, mpath->sdata);
1107 	}
1108 	rcu_read_unlock();
1109 }
1110 
1111 void mesh_pathtbl_unregister(void)
1112 {
1113 	/* no need for locking during exit path */
1114 	mesh_table_free(rcu_dereference_protected(mesh_paths, 1), true);
1115 	mesh_table_free(rcu_dereference_protected(mpp_paths, 1), true);
1116 }
1117