mac80211: use put_unaligned_le in mesh when necessary
[pandora-kernel.git] / net / mac80211 / mesh_pathtbl.c
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 static inline bool mpath_expired(struct mesh_path *mpath)
28 {
29         return (mpath->flags & MESH_PATH_ACTIVE) &&
30                time_after(jiffies, mpath->exp_time) &&
31                !(mpath->flags & MESH_PATH_FIXED);
32 }
33
34 struct mpath_node {
35         struct hlist_node list;
36         struct rcu_head rcu;
37         /* This indirection allows two different tables to point to the same
38          * mesh_path structure, useful when resizing
39          */
40         struct mesh_path *mpath;
41 };
42
43 static struct mesh_table __rcu *mesh_paths;
44 static struct mesh_table __rcu *mpp_paths; /* Store paths for MPP&MAP */
45
46 int mesh_paths_generation;
47
48 /* This lock will have the grow table function as writer and add / delete nodes
49  * as readers. RCU provides sufficient protection only when reading the table
50  * (i.e. doing lookups).  Adding or adding or removing nodes requires we take
51  * the read lock or we risk operating on an old table.  The write lock is only
52  * needed when modifying the number of buckets a table.
53  */
54 static DEFINE_RWLOCK(pathtbl_resize_lock);
55
56
57 static inline struct mesh_table *resize_dereference_mesh_paths(void)
58 {
59         return rcu_dereference_protected(mesh_paths,
60                 lockdep_is_held(&pathtbl_resize_lock));
61 }
62
63 static inline struct mesh_table *resize_dereference_mpp_paths(void)
64 {
65         return rcu_dereference_protected(mpp_paths,
66                 lockdep_is_held(&pathtbl_resize_lock));
67 }
68
69 /*
70  * CAREFUL -- "tbl" must not be an expression,
71  * in particular not an rcu_dereference(), since
72  * it's used twice. So it is illegal to do
73  *      for_each_mesh_entry(rcu_dereference(...), ...)
74  */
75 #define for_each_mesh_entry(tbl, node, i) \
76         for (i = 0; i <= tbl->hash_mask; i++) \
77                 hlist_for_each_entry_rcu(node, &tbl->hash_buckets[i], list)
78
79
80 static struct mesh_table *mesh_table_alloc(int size_order)
81 {
82         int i;
83         struct mesh_table *newtbl;
84
85         newtbl = kmalloc(sizeof(struct mesh_table), GFP_ATOMIC);
86         if (!newtbl)
87                 return NULL;
88
89         newtbl->hash_buckets = kzalloc(sizeof(struct hlist_head) *
90                         (1 << size_order), GFP_ATOMIC);
91
92         if (!newtbl->hash_buckets) {
93                 kfree(newtbl);
94                 return NULL;
95         }
96
97         newtbl->hashwlock = kmalloc(sizeof(spinlock_t) *
98                         (1 << size_order), GFP_ATOMIC);
99         if (!newtbl->hashwlock) {
100                 kfree(newtbl->hash_buckets);
101                 kfree(newtbl);
102                 return NULL;
103         }
104
105         newtbl->size_order = size_order;
106         newtbl->hash_mask = (1 << size_order) - 1;
107         atomic_set(&newtbl->entries,  0);
108         get_random_bytes(&newtbl->hash_rnd,
109                         sizeof(newtbl->hash_rnd));
110         for (i = 0; i <= newtbl->hash_mask; i++)
111                 spin_lock_init(&newtbl->hashwlock[i]);
112         spin_lock_init(&newtbl->gates_lock);
113
114         return newtbl;
115 }
116
117 static void __mesh_table_free(struct mesh_table *tbl)
118 {
119         kfree(tbl->hash_buckets);
120         kfree(tbl->hashwlock);
121         kfree(tbl);
122 }
123
124 static void mesh_table_free(struct mesh_table *tbl, bool free_leafs)
125 {
126         struct hlist_head *mesh_hash;
127         struct hlist_node *p, *q;
128         struct mpath_node *gate;
129         int i;
130
131         mesh_hash = tbl->hash_buckets;
132         for (i = 0; i <= tbl->hash_mask; i++) {
133                 spin_lock_bh(&tbl->hashwlock[i]);
134                 hlist_for_each_safe(p, q, &mesh_hash[i]) {
135                         tbl->free_node(p, free_leafs);
136                         atomic_dec(&tbl->entries);
137                 }
138                 spin_unlock_bh(&tbl->hashwlock[i]);
139         }
140         if (free_leafs) {
141                 spin_lock_bh(&tbl->gates_lock);
142                 hlist_for_each_entry_safe(gate, q,
143                                          tbl->known_gates, list) {
144                         hlist_del(&gate->list);
145                         kfree(gate);
146                 }
147                 kfree(tbl->known_gates);
148                 spin_unlock_bh(&tbl->gates_lock);
149         }
150
151         __mesh_table_free(tbl);
152 }
153
154 static int mesh_table_grow(struct mesh_table *oldtbl,
155                            struct mesh_table *newtbl)
156 {
157         struct hlist_head *oldhash;
158         struct hlist_node *p, *q;
159         int i;
160
161         if (atomic_read(&oldtbl->entries)
162                         < oldtbl->mean_chain_len * (oldtbl->hash_mask + 1))
163                 return -EAGAIN;
164
165         newtbl->free_node = oldtbl->free_node;
166         newtbl->mean_chain_len = oldtbl->mean_chain_len;
167         newtbl->copy_node = oldtbl->copy_node;
168         newtbl->known_gates = oldtbl->known_gates;
169         atomic_set(&newtbl->entries, atomic_read(&oldtbl->entries));
170
171         oldhash = oldtbl->hash_buckets;
172         for (i = 0; i <= oldtbl->hash_mask; i++)
173                 hlist_for_each(p, &oldhash[i])
174                         if (oldtbl->copy_node(p, newtbl) < 0)
175                                 goto errcopy;
176
177         return 0;
178
179 errcopy:
180         for (i = 0; i <= newtbl->hash_mask; i++) {
181                 hlist_for_each_safe(p, q, &newtbl->hash_buckets[i])
182                         oldtbl->free_node(p, 0);
183         }
184         return -ENOMEM;
185 }
186
187 static u32 mesh_table_hash(const u8 *addr, struct ieee80211_sub_if_data *sdata,
188                            struct mesh_table *tbl)
189 {
190         /* Use last four bytes of hw addr and interface index as hash index */
191         return jhash_2words(*(u32 *)(addr+2), sdata->dev->ifindex,
192                             tbl->hash_rnd) & tbl->hash_mask;
193 }
194
195
196 /**
197  *
198  * mesh_path_assign_nexthop - update mesh path next hop
199  *
200  * @mpath: mesh path to update
201  * @sta: next hop to assign
202  *
203  * Locking: mpath->state_lock must be held when calling this function
204  */
205 void mesh_path_assign_nexthop(struct mesh_path *mpath, struct sta_info *sta)
206 {
207         struct sk_buff *skb;
208         struct ieee80211_hdr *hdr;
209         unsigned long flags;
210
211         rcu_assign_pointer(mpath->next_hop, sta);
212
213         spin_lock_irqsave(&mpath->frame_queue.lock, flags);
214         skb_queue_walk(&mpath->frame_queue, skb) {
215                 hdr = (struct ieee80211_hdr *) skb->data;
216                 memcpy(hdr->addr1, sta->sta.addr, ETH_ALEN);
217                 memcpy(hdr->addr2, mpath->sdata->vif.addr, ETH_ALEN);
218                 ieee80211_mps_set_frame_flags(sta->sdata, sta, hdr);
219         }
220
221         spin_unlock_irqrestore(&mpath->frame_queue.lock, flags);
222 }
223
224 static void prepare_for_gate(struct sk_buff *skb, char *dst_addr,
225                              struct mesh_path *gate_mpath)
226 {
227         struct ieee80211_hdr *hdr;
228         struct ieee80211s_hdr *mshdr;
229         int mesh_hdrlen, hdrlen;
230         char *next_hop;
231
232         hdr = (struct ieee80211_hdr *) skb->data;
233         hdrlen = ieee80211_hdrlen(hdr->frame_control);
234         mshdr = (struct ieee80211s_hdr *) (skb->data + hdrlen);
235
236         if (!(mshdr->flags & MESH_FLAGS_AE)) {
237                 /* size of the fixed part of the mesh header */
238                 mesh_hdrlen = 6;
239
240                 /* make room for the two extended addresses */
241                 skb_push(skb, 2 * ETH_ALEN);
242                 memmove(skb->data, hdr, hdrlen + mesh_hdrlen);
243
244                 hdr = (struct ieee80211_hdr *) skb->data;
245
246                 /* we preserve the previous mesh header and only add
247                  * the new addreses */
248                 mshdr = (struct ieee80211s_hdr *) (skb->data + hdrlen);
249                 mshdr->flags = MESH_FLAGS_AE_A5_A6;
250                 memcpy(mshdr->eaddr1, hdr->addr3, ETH_ALEN);
251                 memcpy(mshdr->eaddr2, hdr->addr4, ETH_ALEN);
252         }
253
254         /* update next hop */
255         hdr = (struct ieee80211_hdr *) skb->data;
256         rcu_read_lock();
257         next_hop = rcu_dereference(gate_mpath->next_hop)->sta.addr;
258         memcpy(hdr->addr1, next_hop, ETH_ALEN);
259         rcu_read_unlock();
260         memcpy(hdr->addr2, gate_mpath->sdata->vif.addr, ETH_ALEN);
261         memcpy(hdr->addr3, dst_addr, ETH_ALEN);
262 }
263
264 /**
265  *
266  * mesh_path_move_to_queue - Move or copy frames from one mpath queue to another
267  *
268  * This function is used to transfer or copy frames from an unresolved mpath to
269  * a gate mpath.  The function also adds the Address Extension field and
270  * updates the next hop.
271  *
272  * If a frame already has an Address Extension field, only the next hop and
273  * destination addresses are updated.
274  *
275  * The gate mpath must be an active mpath with a valid mpath->next_hop.
276  *
277  * @mpath: An active mpath the frames will be sent to (i.e. the gate)
278  * @from_mpath: The failed mpath
279  * @copy: When true, copy all the frames to the new mpath queue.  When false,
280  * move them.
281  */
282 static void mesh_path_move_to_queue(struct mesh_path *gate_mpath,
283                                     struct mesh_path *from_mpath,
284                                     bool copy)
285 {
286         struct sk_buff *skb, *fskb, *tmp;
287         struct sk_buff_head failq;
288         unsigned long flags;
289
290         BUG_ON(gate_mpath == from_mpath);
291         BUG_ON(!gate_mpath->next_hop);
292
293         __skb_queue_head_init(&failq);
294
295         spin_lock_irqsave(&from_mpath->frame_queue.lock, flags);
296         skb_queue_splice_init(&from_mpath->frame_queue, &failq);
297         spin_unlock_irqrestore(&from_mpath->frame_queue.lock, flags);
298
299         skb_queue_walk_safe(&failq, fskb, tmp) {
300                 if (skb_queue_len(&gate_mpath->frame_queue) >=
301                                   MESH_FRAME_QUEUE_LEN) {
302                         mpath_dbg(gate_mpath->sdata, "mpath queue full!\n");
303                         break;
304                 }
305
306                 skb = skb_copy(fskb, GFP_ATOMIC);
307                 if (WARN_ON(!skb))
308                         break;
309
310                 prepare_for_gate(skb, gate_mpath->dst, gate_mpath);
311                 skb_queue_tail(&gate_mpath->frame_queue, skb);
312
313                 if (copy)
314                         continue;
315
316                 __skb_unlink(fskb, &failq);
317                 kfree_skb(fskb);
318         }
319
320         mpath_dbg(gate_mpath->sdata, "Mpath queue for gate %pM has %d frames\n",
321                   gate_mpath->dst, skb_queue_len(&gate_mpath->frame_queue));
322
323         if (!copy)
324                 return;
325
326         spin_lock_irqsave(&from_mpath->frame_queue.lock, flags);
327         skb_queue_splice(&failq, &from_mpath->frame_queue);
328         spin_unlock_irqrestore(&from_mpath->frame_queue.lock, flags);
329 }
330
331
332 static struct mesh_path *mpath_lookup(struct mesh_table *tbl, const u8 *dst,
333                                       struct ieee80211_sub_if_data *sdata)
334 {
335         struct mesh_path *mpath;
336         struct hlist_head *bucket;
337         struct mpath_node *node;
338
339         bucket = &tbl->hash_buckets[mesh_table_hash(dst, sdata, tbl)];
340         hlist_for_each_entry_rcu(node, bucket, list) {
341                 mpath = node->mpath;
342                 if (mpath->sdata == sdata &&
343                     ether_addr_equal(dst, mpath->dst)) {
344                         if (mpath_expired(mpath)) {
345                                 spin_lock_bh(&mpath->state_lock);
346                                 mpath->flags &= ~MESH_PATH_ACTIVE;
347                                 spin_unlock_bh(&mpath->state_lock);
348                         }
349                         return mpath;
350                 }
351         }
352         return NULL;
353 }
354
355 /**
356  * mesh_path_lookup - look up a path in the mesh path table
357  * @sdata: local subif
358  * @dst: hardware address (ETH_ALEN length) of destination
359  *
360  * Returns: pointer to the mesh path structure, or NULL if not found
361  *
362  * Locking: must be called within a read rcu section.
363  */
364 struct mesh_path *
365 mesh_path_lookup(struct ieee80211_sub_if_data *sdata, const u8 *dst)
366 {
367         return mpath_lookup(rcu_dereference(mesh_paths), dst, sdata);
368 }
369
370 struct mesh_path *
371 mpp_path_lookup(struct ieee80211_sub_if_data *sdata, const u8 *dst)
372 {
373         return mpath_lookup(rcu_dereference(mpp_paths), dst, sdata);
374 }
375
376
377 /**
378  * mesh_path_lookup_by_idx - look up a path in the mesh path table by its index
379  * @idx: index
380  * @sdata: local subif, or NULL for all entries
381  *
382  * Returns: pointer to the mesh path structure, or NULL if not found.
383  *
384  * Locking: must be called within a read rcu section.
385  */
386 struct mesh_path *
387 mesh_path_lookup_by_idx(struct ieee80211_sub_if_data *sdata, int idx)
388 {
389         struct mesh_table *tbl = rcu_dereference(mesh_paths);
390         struct mpath_node *node;
391         int i;
392         int j = 0;
393
394         for_each_mesh_entry(tbl, node, i) {
395                 if (sdata && node->mpath->sdata != sdata)
396                         continue;
397                 if (j++ == idx) {
398                         if (mpath_expired(node->mpath)) {
399                                 spin_lock_bh(&node->mpath->state_lock);
400                                 node->mpath->flags &= ~MESH_PATH_ACTIVE;
401                                 spin_unlock_bh(&node->mpath->state_lock);
402                         }
403                         return node->mpath;
404                 }
405         }
406
407         return NULL;
408 }
409
410 /**
411  * mesh_path_add_gate - add the given mpath to a mesh gate to our path table
412  * @mpath: gate path to add to table
413  */
414 int mesh_path_add_gate(struct mesh_path *mpath)
415 {
416         struct mesh_table *tbl;
417         struct mpath_node *gate, *new_gate;
418         int err;
419
420         rcu_read_lock();
421         tbl = rcu_dereference(mesh_paths);
422
423         hlist_for_each_entry_rcu(gate, tbl->known_gates, list)
424                 if (gate->mpath == mpath) {
425                         err = -EEXIST;
426                         goto err_rcu;
427                 }
428
429         new_gate = kzalloc(sizeof(struct mpath_node), GFP_ATOMIC);
430         if (!new_gate) {
431                 err = -ENOMEM;
432                 goto err_rcu;
433         }
434
435         mpath->is_gate = true;
436         mpath->sdata->u.mesh.num_gates++;
437         new_gate->mpath = mpath;
438         spin_lock_bh(&tbl->gates_lock);
439         hlist_add_head_rcu(&new_gate->list, tbl->known_gates);
440         spin_unlock_bh(&tbl->gates_lock);
441         mpath_dbg(mpath->sdata,
442                   "Mesh path: Recorded new gate: %pM. %d known gates\n",
443                   mpath->dst, mpath->sdata->u.mesh.num_gates);
444         err = 0;
445 err_rcu:
446         rcu_read_unlock();
447         return err;
448 }
449
450 /**
451  * mesh_gate_del - remove a mesh gate from the list of known gates
452  * @tbl: table which holds our list of known gates
453  * @mpath: gate mpath
454  *
455  * Locking: must be called inside rcu_read_lock() section
456  */
457 static void mesh_gate_del(struct mesh_table *tbl, struct mesh_path *mpath)
458 {
459         struct mpath_node *gate;
460         struct hlist_node *q;
461
462         hlist_for_each_entry_safe(gate, q, tbl->known_gates, list) {
463                 if (gate->mpath != mpath)
464                         continue;
465                 spin_lock_bh(&tbl->gates_lock);
466                 hlist_del_rcu(&gate->list);
467                 kfree_rcu(gate, rcu);
468                 spin_unlock_bh(&tbl->gates_lock);
469                 mpath->sdata->u.mesh.num_gates--;
470                 mpath->is_gate = false;
471                 mpath_dbg(mpath->sdata,
472                           "Mesh path: Deleted gate: %pM. %d known gates\n",
473                           mpath->dst, mpath->sdata->u.mesh.num_gates);
474                 break;
475         }
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  * @dst: 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 struct mesh_path *mesh_path_add(struct ieee80211_sub_if_data *sdata,
497                                 const u8 *dst)
498 {
499         struct ieee80211_if_mesh *ifmsh = &sdata->u.mesh;
500         struct ieee80211_local *local = sdata->local;
501         struct mesh_table *tbl;
502         struct mesh_path *mpath, *new_mpath;
503         struct mpath_node *node, *new_node;
504         struct hlist_head *bucket;
505         int grow = 0;
506         int err;
507         u32 hash_idx;
508
509         if (ether_addr_equal(dst, sdata->vif.addr))
510                 /* never add ourselves as neighbours */
511                 return ERR_PTR(-ENOTSUPP);
512
513         if (is_multicast_ether_addr(dst))
514                 return ERR_PTR(-ENOTSUPP);
515
516         if (atomic_add_unless(&sdata->u.mesh.mpaths, 1, MESH_MAX_MPATHS) == 0)
517                 return ERR_PTR(-ENOSPC);
518
519         read_lock_bh(&pathtbl_resize_lock);
520         tbl = resize_dereference_mesh_paths();
521
522         hash_idx = mesh_table_hash(dst, sdata, tbl);
523         bucket = &tbl->hash_buckets[hash_idx];
524
525         spin_lock(&tbl->hashwlock[hash_idx]);
526
527         hlist_for_each_entry(node, bucket, list) {
528                 mpath = node->mpath;
529                 if (mpath->sdata == sdata &&
530                     ether_addr_equal(dst, mpath->dst))
531                         goto found;
532         }
533
534         err = -ENOMEM;
535         new_mpath = kzalloc(sizeof(struct mesh_path), GFP_ATOMIC);
536         if (!new_mpath)
537                 goto err_path_alloc;
538
539         new_node = kmalloc(sizeof(struct mpath_node), GFP_ATOMIC);
540         if (!new_node)
541                 goto err_node_alloc;
542
543         memcpy(new_mpath->dst, dst, ETH_ALEN);
544         eth_broadcast_addr(new_mpath->rann_snd_addr);
545         new_mpath->is_root = false;
546         new_mpath->sdata = sdata;
547         new_mpath->flags = 0;
548         skb_queue_head_init(&new_mpath->frame_queue);
549         new_node->mpath = new_mpath;
550         new_mpath->timer.data = (unsigned long) new_mpath;
551         new_mpath->timer.function = mesh_path_timer;
552         new_mpath->exp_time = jiffies;
553         spin_lock_init(&new_mpath->state_lock);
554         init_timer(&new_mpath->timer);
555
556         hlist_add_head_rcu(&new_node->list, bucket);
557         if (atomic_inc_return(&tbl->entries) >=
558             tbl->mean_chain_len * (tbl->hash_mask + 1))
559                 grow = 1;
560
561         mesh_paths_generation++;
562
563         if (grow) {
564                 set_bit(MESH_WORK_GROW_MPATH_TABLE,  &ifmsh->wrkq_flags);
565                 ieee80211_queue_work(&local->hw, &sdata->work);
566         }
567         mpath = new_mpath;
568 found:
569         spin_unlock(&tbl->hashwlock[hash_idx]);
570         read_unlock_bh(&pathtbl_resize_lock);
571         return mpath;
572
573 err_node_alloc:
574         kfree(new_mpath);
575 err_path_alloc:
576         atomic_dec(&sdata->u.mesh.mpaths);
577         spin_unlock(&tbl->hashwlock[hash_idx]);
578         read_unlock_bh(&pathtbl_resize_lock);
579         return ERR_PTR(err);
580 }
581
582 static void mesh_table_free_rcu(struct rcu_head *rcu)
583 {
584         struct mesh_table *tbl = container_of(rcu, struct mesh_table, rcu_head);
585
586         mesh_table_free(tbl, false);
587 }
588
589 void mesh_mpath_table_grow(void)
590 {
591         struct mesh_table *oldtbl, *newtbl;
592
593         write_lock_bh(&pathtbl_resize_lock);
594         oldtbl = resize_dereference_mesh_paths();
595         newtbl = mesh_table_alloc(oldtbl->size_order + 1);
596         if (!newtbl)
597                 goto out;
598         if (mesh_table_grow(oldtbl, newtbl) < 0) {
599                 __mesh_table_free(newtbl);
600                 goto out;
601         }
602         rcu_assign_pointer(mesh_paths, newtbl);
603
604         call_rcu(&oldtbl->rcu_head, mesh_table_free_rcu);
605
606  out:
607         write_unlock_bh(&pathtbl_resize_lock);
608 }
609
610 void mesh_mpp_table_grow(void)
611 {
612         struct mesh_table *oldtbl, *newtbl;
613
614         write_lock_bh(&pathtbl_resize_lock);
615         oldtbl = resize_dereference_mpp_paths();
616         newtbl = mesh_table_alloc(oldtbl->size_order + 1);
617         if (!newtbl)
618                 goto out;
619         if (mesh_table_grow(oldtbl, newtbl) < 0) {
620                 __mesh_table_free(newtbl);
621                 goto out;
622         }
623         rcu_assign_pointer(mpp_paths, newtbl);
624         call_rcu(&oldtbl->rcu_head, mesh_table_free_rcu);
625
626  out:
627         write_unlock_bh(&pathtbl_resize_lock);
628 }
629
630 int mpp_path_add(struct ieee80211_sub_if_data *sdata,
631                  const u8 *dst, const u8 *mpp)
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         int grow = 0;
640         int err = 0;
641         u32 hash_idx;
642
643         if (ether_addr_equal(dst, sdata->vif.addr))
644                 /* never add ourselves as neighbours */
645                 return -ENOTSUPP;
646
647         if (is_multicast_ether_addr(dst))
648                 return -ENOTSUPP;
649
650         err = -ENOMEM;
651         new_mpath = kzalloc(sizeof(struct mesh_path), GFP_ATOMIC);
652         if (!new_mpath)
653                 goto err_path_alloc;
654
655         new_node = kmalloc(sizeof(struct mpath_node), GFP_ATOMIC);
656         if (!new_node)
657                 goto err_node_alloc;
658
659         read_lock_bh(&pathtbl_resize_lock);
660         memcpy(new_mpath->dst, dst, ETH_ALEN);
661         memcpy(new_mpath->mpp, mpp, ETH_ALEN);
662         new_mpath->sdata = sdata;
663         new_mpath->flags = 0;
664         skb_queue_head_init(&new_mpath->frame_queue);
665         new_node->mpath = new_mpath;
666         init_timer(&new_mpath->timer);
667         new_mpath->exp_time = jiffies;
668         spin_lock_init(&new_mpath->state_lock);
669
670         tbl = resize_dereference_mpp_paths();
671
672         hash_idx = mesh_table_hash(dst, sdata, tbl);
673         bucket = &tbl->hash_buckets[hash_idx];
674
675         spin_lock(&tbl->hashwlock[hash_idx]);
676
677         err = -EEXIST;
678         hlist_for_each_entry(node, bucket, list) {
679                 mpath = node->mpath;
680                 if (mpath->sdata == sdata &&
681                     ether_addr_equal(dst, mpath->dst))
682                         goto err_exists;
683         }
684
685         hlist_add_head_rcu(&new_node->list, bucket);
686         if (atomic_inc_return(&tbl->entries) >=
687             tbl->mean_chain_len * (tbl->hash_mask + 1))
688                 grow = 1;
689
690         spin_unlock(&tbl->hashwlock[hash_idx]);
691         read_unlock_bh(&pathtbl_resize_lock);
692         if (grow) {
693                 set_bit(MESH_WORK_GROW_MPP_TABLE,  &ifmsh->wrkq_flags);
694                 ieee80211_queue_work(&local->hw, &sdata->work);
695         }
696         return 0;
697
698 err_exists:
699         spin_unlock(&tbl->hashwlock[hash_idx]);
700         read_unlock_bh(&pathtbl_resize_lock);
701         kfree(new_node);
702 err_node_alloc:
703         kfree(new_mpath);
704 err_path_alloc:
705         return err;
706 }
707
708
709 /**
710  * mesh_plink_broken - deactivates paths and sends perr when a link breaks
711  *
712  * @sta: broken peer link
713  *
714  * This function must be called from the rate control algorithm if enough
715  * delivery errors suggest that a peer link is no longer usable.
716  */
717 void mesh_plink_broken(struct sta_info *sta)
718 {
719         struct mesh_table *tbl;
720         static const u8 bcast[ETH_ALEN] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
721         struct mesh_path *mpath;
722         struct mpath_node *node;
723         struct ieee80211_sub_if_data *sdata = sta->sdata;
724         int i;
725
726         rcu_read_lock();
727         tbl = rcu_dereference(mesh_paths);
728         for_each_mesh_entry(tbl, node, i) {
729                 mpath = node->mpath;
730                 if (rcu_dereference(mpath->next_hop) == sta &&
731                     mpath->flags & MESH_PATH_ACTIVE &&
732                     !(mpath->flags & MESH_PATH_FIXED)) {
733                         spin_lock_bh(&mpath->state_lock);
734                         mpath->flags &= ~MESH_PATH_ACTIVE;
735                         ++mpath->sn;
736                         spin_unlock_bh(&mpath->state_lock);
737                         mesh_path_error_tx(sdata,
738                                 sdata->u.mesh.mshcfg.element_ttl,
739                                 mpath->dst, mpath->sn,
740                                 WLAN_REASON_MESH_PATH_DEST_UNREACHABLE, bcast);
741                 }
742         }
743         rcu_read_unlock();
744 }
745
746 static void mesh_path_node_reclaim(struct rcu_head *rp)
747 {
748         struct mpath_node *node = container_of(rp, struct mpath_node, rcu);
749         struct ieee80211_sub_if_data *sdata = node->mpath->sdata;
750
751         del_timer_sync(&node->mpath->timer);
752         atomic_dec(&sdata->u.mesh.mpaths);
753         kfree(node->mpath);
754         kfree(node);
755 }
756
757 /* needs to be called with the corresponding hashwlock taken */
758 static void __mesh_path_del(struct mesh_table *tbl, struct mpath_node *node)
759 {
760         struct mesh_path *mpath;
761         mpath = node->mpath;
762         spin_lock(&mpath->state_lock);
763         mpath->flags |= MESH_PATH_RESOLVING;
764         if (mpath->is_gate)
765                 mesh_gate_del(tbl, mpath);
766         hlist_del_rcu(&node->list);
767         call_rcu(&node->rcu, mesh_path_node_reclaim);
768         spin_unlock(&mpath->state_lock);
769         atomic_dec(&tbl->entries);
770 }
771
772 /**
773  * mesh_path_flush_by_nexthop - Deletes mesh paths if their next hop matches
774  *
775  * @sta: mesh peer to match
776  *
777  * RCU notes: this function is called when a mesh plink transitions from
778  * PLINK_ESTAB to any other state, since PLINK_ESTAB state is the only one that
779  * allows path creation. This will happen before the sta can be freed (because
780  * sta_info_destroy() calls this) so any reader in a rcu read block will be
781  * protected against the plink disappearing.
782  */
783 void mesh_path_flush_by_nexthop(struct sta_info *sta)
784 {
785         struct mesh_table *tbl;
786         struct mesh_path *mpath;
787         struct mpath_node *node;
788         int i;
789
790         rcu_read_lock();
791         read_lock_bh(&pathtbl_resize_lock);
792         tbl = resize_dereference_mesh_paths();
793         for_each_mesh_entry(tbl, node, i) {
794                 mpath = node->mpath;
795                 if (rcu_dereference(mpath->next_hop) == sta) {
796                         spin_lock(&tbl->hashwlock[i]);
797                         __mesh_path_del(tbl, node);
798                         spin_unlock(&tbl->hashwlock[i]);
799                 }
800         }
801         read_unlock_bh(&pathtbl_resize_lock);
802         rcu_read_unlock();
803 }
804
805 static void table_flush_by_iface(struct mesh_table *tbl,
806                                  struct ieee80211_sub_if_data *sdata)
807 {
808         struct mesh_path *mpath;
809         struct mpath_node *node;
810         int i;
811
812         WARN_ON(!rcu_read_lock_held());
813         for_each_mesh_entry(tbl, node, i) {
814                 mpath = node->mpath;
815                 if (mpath->sdata != sdata)
816                         continue;
817                 spin_lock_bh(&tbl->hashwlock[i]);
818                 __mesh_path_del(tbl, node);
819                 spin_unlock_bh(&tbl->hashwlock[i]);
820         }
821 }
822
823 /**
824  * mesh_path_flush_by_iface - Deletes all mesh paths associated with a given iface
825  *
826  * This function deletes both mesh paths as well as mesh portal paths.
827  *
828  * @sdata: interface data to match
829  *
830  */
831 void mesh_path_flush_by_iface(struct ieee80211_sub_if_data *sdata)
832 {
833         struct mesh_table *tbl;
834
835         rcu_read_lock();
836         read_lock_bh(&pathtbl_resize_lock);
837         tbl = resize_dereference_mesh_paths();
838         table_flush_by_iface(tbl, sdata);
839         tbl = resize_dereference_mpp_paths();
840         table_flush_by_iface(tbl, sdata);
841         read_unlock_bh(&pathtbl_resize_lock);
842         rcu_read_unlock();
843 }
844
845 /**
846  * mesh_path_del - delete a mesh path from the table
847  *
848  * @addr: dst address (ETH_ALEN length)
849  * @sdata: local subif
850  *
851  * Returns: 0 if successful
852  */
853 int mesh_path_del(struct ieee80211_sub_if_data *sdata, const u8 *addr)
854 {
855         struct mesh_table *tbl;
856         struct mesh_path *mpath;
857         struct mpath_node *node;
858         struct hlist_head *bucket;
859         int hash_idx;
860         int err = 0;
861
862         read_lock_bh(&pathtbl_resize_lock);
863         tbl = resize_dereference_mesh_paths();
864         hash_idx = mesh_table_hash(addr, sdata, tbl);
865         bucket = &tbl->hash_buckets[hash_idx];
866
867         spin_lock(&tbl->hashwlock[hash_idx]);
868         hlist_for_each_entry(node, bucket, list) {
869                 mpath = node->mpath;
870                 if (mpath->sdata == sdata &&
871                     ether_addr_equal(addr, mpath->dst)) {
872                         __mesh_path_del(tbl, node);
873                         goto enddel;
874                 }
875         }
876
877         err = -ENXIO;
878 enddel:
879         mesh_paths_generation++;
880         spin_unlock(&tbl->hashwlock[hash_idx]);
881         read_unlock_bh(&pathtbl_resize_lock);
882         return err;
883 }
884
885 /**
886  * mesh_path_tx_pending - sends pending frames in a mesh path queue
887  *
888  * @mpath: mesh path to activate
889  *
890  * Locking: the state_lock of the mpath structure must NOT be held when calling
891  * this function.
892  */
893 void mesh_path_tx_pending(struct mesh_path *mpath)
894 {
895         if (mpath->flags & MESH_PATH_ACTIVE)
896                 ieee80211_add_pending_skbs(mpath->sdata->local,
897                                 &mpath->frame_queue);
898 }
899
900 /**
901  * mesh_path_send_to_gates - sends pending frames to all known mesh gates
902  *
903  * @mpath: mesh path whose queue will be emptied
904  *
905  * If there is only one gate, the frames are transferred from the failed mpath
906  * queue to that gate's queue.  If there are more than one gates, the frames
907  * are copied from each gate to the next.  After frames are copied, the
908  * mpath queues are emptied onto the transmission queue.
909  */
910 int mesh_path_send_to_gates(struct mesh_path *mpath)
911 {
912         struct ieee80211_sub_if_data *sdata = mpath->sdata;
913         struct mesh_table *tbl;
914         struct mesh_path *from_mpath = mpath;
915         struct mpath_node *gate = NULL;
916         bool copy = false;
917         struct hlist_head *known_gates;
918
919         rcu_read_lock();
920         tbl = rcu_dereference(mesh_paths);
921         known_gates = tbl->known_gates;
922         rcu_read_unlock();
923
924         if (!known_gates)
925                 return -EHOSTUNREACH;
926
927         hlist_for_each_entry_rcu(gate, known_gates, list) {
928                 if (gate->mpath->sdata != sdata)
929                         continue;
930
931                 if (gate->mpath->flags & MESH_PATH_ACTIVE) {
932                         mpath_dbg(sdata, "Forwarding to %pM\n", gate->mpath->dst);
933                         mesh_path_move_to_queue(gate->mpath, from_mpath, copy);
934                         from_mpath = gate->mpath;
935                         copy = true;
936                 } else {
937                         mpath_dbg(sdata,
938                                   "Not forwarding %p (flags %#x)\n",
939                                   gate->mpath, gate->mpath->flags);
940                 }
941         }
942
943         hlist_for_each_entry_rcu(gate, known_gates, list)
944                 if (gate->mpath->sdata == sdata) {
945                         mpath_dbg(sdata, "Sending to %pM\n", gate->mpath->dst);
946                         mesh_path_tx_pending(gate->mpath);
947                 }
948
949         return (from_mpath == mpath) ? -EHOSTUNREACH : 0;
950 }
951
952 /**
953  * mesh_path_discard_frame - discard a frame whose path could not be resolved
954  *
955  * @skb: frame to discard
956  * @sdata: network subif the frame was to be sent through
957  *
958  * Locking: the function must me called within a rcu_read_lock region
959  */
960 void mesh_path_discard_frame(struct ieee80211_sub_if_data *sdata,
961                              struct sk_buff *skb)
962 {
963         kfree_skb(skb);
964         sdata->u.mesh.mshstats.dropped_frames_no_route++;
965 }
966
967 /**
968  * mesh_path_flush_pending - free the pending queue of a mesh path
969  *
970  * @mpath: mesh path whose queue has to be freed
971  *
972  * Locking: the function must me called within a rcu_read_lock region
973  */
974 void mesh_path_flush_pending(struct mesh_path *mpath)
975 {
976         struct sk_buff *skb;
977
978         while ((skb = skb_dequeue(&mpath->frame_queue)) != NULL)
979                 mesh_path_discard_frame(mpath->sdata, skb);
980 }
981
982 /**
983  * mesh_path_fix_nexthop - force a specific next hop for a mesh path
984  *
985  * @mpath: the mesh path to modify
986  * @next_hop: the next hop to force
987  *
988  * Locking: this function must be called holding mpath->state_lock
989  */
990 void mesh_path_fix_nexthop(struct mesh_path *mpath, struct sta_info *next_hop)
991 {
992         spin_lock_bh(&mpath->state_lock);
993         mesh_path_assign_nexthop(mpath, next_hop);
994         mpath->sn = 0xffff;
995         mpath->metric = 0;
996         mpath->hop_count = 0;
997         mpath->exp_time = 0;
998         mpath->flags |= MESH_PATH_FIXED;
999         mesh_path_activate(mpath);
1000         spin_unlock_bh(&mpath->state_lock);
1001         mesh_path_tx_pending(mpath);
1002 }
1003
1004 static void mesh_path_node_free(struct hlist_node *p, bool free_leafs)
1005 {
1006         struct mesh_path *mpath;
1007         struct mpath_node *node = hlist_entry(p, struct mpath_node, list);
1008         mpath = node->mpath;
1009         hlist_del_rcu(p);
1010         if (free_leafs) {
1011                 del_timer_sync(&mpath->timer);
1012                 kfree(mpath);
1013         }
1014         kfree(node);
1015 }
1016
1017 static int mesh_path_node_copy(struct hlist_node *p, struct mesh_table *newtbl)
1018 {
1019         struct mesh_path *mpath;
1020         struct mpath_node *node, *new_node;
1021         u32 hash_idx;
1022
1023         new_node = kmalloc(sizeof(struct mpath_node), GFP_ATOMIC);
1024         if (new_node == NULL)
1025                 return -ENOMEM;
1026
1027         node = hlist_entry(p, struct mpath_node, list);
1028         mpath = node->mpath;
1029         new_node->mpath = mpath;
1030         hash_idx = mesh_table_hash(mpath->dst, mpath->sdata, newtbl);
1031         hlist_add_head(&new_node->list,
1032                         &newtbl->hash_buckets[hash_idx]);
1033         return 0;
1034 }
1035
1036 int mesh_pathtbl_init(void)
1037 {
1038         struct mesh_table *tbl_path, *tbl_mpp;
1039         int ret;
1040
1041         tbl_path = mesh_table_alloc(INIT_PATHS_SIZE_ORDER);
1042         if (!tbl_path)
1043                 return -ENOMEM;
1044         tbl_path->free_node = &mesh_path_node_free;
1045         tbl_path->copy_node = &mesh_path_node_copy;
1046         tbl_path->mean_chain_len = MEAN_CHAIN_LEN;
1047         tbl_path->known_gates = kzalloc(sizeof(struct hlist_head), GFP_ATOMIC);
1048         if (!tbl_path->known_gates) {
1049                 ret = -ENOMEM;
1050                 goto free_path;
1051         }
1052         INIT_HLIST_HEAD(tbl_path->known_gates);
1053
1054
1055         tbl_mpp = mesh_table_alloc(INIT_PATHS_SIZE_ORDER);
1056         if (!tbl_mpp) {
1057                 ret = -ENOMEM;
1058                 goto free_path;
1059         }
1060         tbl_mpp->free_node = &mesh_path_node_free;
1061         tbl_mpp->copy_node = &mesh_path_node_copy;
1062         tbl_mpp->mean_chain_len = MEAN_CHAIN_LEN;
1063         tbl_mpp->known_gates = kzalloc(sizeof(struct hlist_head), GFP_ATOMIC);
1064         if (!tbl_mpp->known_gates) {
1065                 ret = -ENOMEM;
1066                 goto free_mpp;
1067         }
1068         INIT_HLIST_HEAD(tbl_mpp->known_gates);
1069
1070         /* Need no locking since this is during init */
1071         RCU_INIT_POINTER(mesh_paths, tbl_path);
1072         RCU_INIT_POINTER(mpp_paths, tbl_mpp);
1073
1074         return 0;
1075
1076 free_mpp:
1077         mesh_table_free(tbl_mpp, true);
1078 free_path:
1079         mesh_table_free(tbl_path, true);
1080         return ret;
1081 }
1082
1083 void mesh_path_expire(struct ieee80211_sub_if_data *sdata)
1084 {
1085         struct mesh_table *tbl;
1086         struct mesh_path *mpath;
1087         struct mpath_node *node;
1088         int i;
1089
1090         rcu_read_lock();
1091         tbl = rcu_dereference(mesh_paths);
1092         for_each_mesh_entry(tbl, node, i) {
1093                 if (node->mpath->sdata != sdata)
1094                         continue;
1095                 mpath = node->mpath;
1096                 if ((!(mpath->flags & MESH_PATH_RESOLVING)) &&
1097                     (!(mpath->flags & MESH_PATH_FIXED)) &&
1098                      time_after(jiffies, mpath->exp_time + MESH_PATH_EXPIRE))
1099                         mesh_path_del(mpath->sdata, mpath->dst);
1100         }
1101         rcu_read_unlock();
1102 }
1103
1104 void mesh_pathtbl_unregister(void)
1105 {
1106         /* no need for locking during exit path */
1107         mesh_table_free(rcu_dereference_protected(mesh_paths, 1), true);
1108         mesh_table_free(rcu_dereference_protected(mpp_paths, 1), true);
1109 }