blob: 6b9670d6bbf9e669c5e833514ccb41cb83f014da [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
Christoph Lametercde53532008-07-04 09:59:22 -07004 * Copyright (C) 2005 SGI, Christoph Lameter
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08005 * Copyright (C) 2006 Nick Piggin
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License as
9 * published by the Free Software Foundation; either version 2, or (at
10 * your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 */
21
22#include <linux/errno.h>
23#include <linux/init.h>
24#include <linux/kernel.h>
25#include <linux/module.h>
26#include <linux/radix-tree.h>
27#include <linux/percpu.h>
28#include <linux/slab.h>
29#include <linux/notifier.h>
30#include <linux/cpu.h>
31#include <linux/gfp.h>
32#include <linux/string.h>
33#include <linux/bitops.h>
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080034#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36
37#ifdef __KERNEL__
Nick Piggincfd9b7d2006-06-23 02:03:22 -070038#define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#else
40#define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */
41#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43#define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
44#define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
45
46#define RADIX_TREE_TAG_LONGS \
47 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
48
49struct radix_tree_node {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080050 unsigned int height; /* Height from the bottom */
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 unsigned int count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080052 struct rcu_head rcu_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 void *slots[RADIX_TREE_MAP_SIZE];
Jonathan Corbetdaff89f2006-03-25 03:08:05 -080054 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -070055};
56
57struct radix_tree_path {
Christoph Lameter201b6262005-09-06 15:16:46 -070058 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 int offset;
60};
61
62#define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
Jeff Moyer26fb1582007-10-16 01:24:49 -070063#define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
64 RADIX_TREE_MAP_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Jeff Moyer26fb1582007-10-16 01:24:49 -070066/*
67 * The height_to_maxindex array needs to be one deeper than the maximum
68 * path as height 0 holds only 1 entry.
69 */
70static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
72/*
73 * Radix tree node cache.
74 */
Christoph Lametere18b8902006-12-06 20:33:20 -080075static struct kmem_cache *radix_tree_node_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77/*
78 * Per-cpu pool of preloaded nodes
79 */
80struct radix_tree_preload {
81 int nr;
82 struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH];
83};
Harvey Harrison8cef7d52009-01-06 14:40:50 -080084static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Nick Piggin612d6c12006-06-23 02:03:22 -070086static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
87{
88 return root->gfp_mask & __GFP_BITS_MASK;
89}
90
Nick Piggin643b52b2008-06-12 15:21:52 -070091static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
92 int offset)
93{
94 __set_bit(offset, node->tags[tag]);
95}
96
97static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
98 int offset)
99{
100 __clear_bit(offset, node->tags[tag]);
101}
102
103static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
104 int offset)
105{
106 return test_bit(offset, node->tags[tag]);
107}
108
109static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
110{
111 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
112}
113
114static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
115{
116 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
117}
118
119static inline void root_tag_clear_all(struct radix_tree_root *root)
120{
121 root->gfp_mask &= __GFP_BITS_MASK;
122}
123
124static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
125{
126 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
127}
128
129/*
130 * Returns 1 if any slot in the node has this tag set.
131 * Otherwise returns 0.
132 */
133static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
134{
135 int idx;
136 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
137 if (node->tags[tag][idx])
138 return 1;
139 }
140 return 0;
141}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142/*
143 * This assumes that the caller has performed appropriate preallocation, and
144 * that the caller has pinned this thread of control to the current CPU.
145 */
146static struct radix_tree_node *
147radix_tree_node_alloc(struct radix_tree_root *root)
148{
Nick Piggine2848a02008-02-04 22:29:10 -0800149 struct radix_tree_node *ret = NULL;
Nick Piggin612d6c12006-06-23 02:03:22 -0700150 gfp_t gfp_mask = root_gfp_mask(root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Nick Piggine2848a02008-02-04 22:29:10 -0800152 if (!(gfp_mask & __GFP_WAIT)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 struct radix_tree_preload *rtp;
154
Nick Piggine2848a02008-02-04 22:29:10 -0800155 /*
156 * Provided the caller has preloaded here, we will always
157 * succeed in getting a node here (and never reach
158 * kmem_cache_alloc)
159 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 rtp = &__get_cpu_var(radix_tree_preloads);
161 if (rtp->nr) {
162 ret = rtp->nodes[rtp->nr - 1];
163 rtp->nodes[rtp->nr - 1] = NULL;
164 rtp->nr--;
165 }
166 }
Nick Piggine2848a02008-02-04 22:29:10 -0800167 if (ret == NULL)
Christoph Lameter488514d2008-04-28 02:12:05 -0700168 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Nick Piggine2848a02008-02-04 22:29:10 -0800169
Nick Pigginc0bc9872007-10-16 01:24:42 -0700170 BUG_ON(radix_tree_is_indirect_ptr(ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 return ret;
172}
173
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800174static void radix_tree_node_rcu_free(struct rcu_head *head)
175{
176 struct radix_tree_node *node =
177 container_of(head, struct radix_tree_node, rcu_head);
Nick Piggin643b52b2008-06-12 15:21:52 -0700178
179 /*
180 * must only free zeroed nodes into the slab. radix_tree_shrink
181 * can leave us with a non-NULL entry in the first slot, so clear
182 * that here to make sure.
183 */
184 tag_clear(node, 0, 0);
185 tag_clear(node, 1, 0);
186 node->slots[0] = NULL;
187 node->count = 0;
188
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800189 kmem_cache_free(radix_tree_node_cachep, node);
190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192static inline void
193radix_tree_node_free(struct radix_tree_node *node)
194{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800195 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196}
197
198/*
199 * Load up this CPU's radix_tree_node buffer with sufficient objects to
200 * ensure that the addition of a single element in the tree cannot fail. On
201 * success, return zero, with preemption disabled. On error, return -ENOMEM
202 * with preemption not disabled.
David Howellsb34df792009-11-19 18:11:14 +0000203 *
204 * To make use of this facility, the radix tree must be initialised without
205 * __GFP_WAIT being passed to INIT_RADIX_TREE().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 */
Al Virodd0fc662005-10-07 07:46:04 +0100207int radix_tree_preload(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208{
209 struct radix_tree_preload *rtp;
210 struct radix_tree_node *node;
211 int ret = -ENOMEM;
212
213 preempt_disable();
214 rtp = &__get_cpu_var(radix_tree_preloads);
215 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
216 preempt_enable();
Christoph Lameter488514d2008-04-28 02:12:05 -0700217 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 if (node == NULL)
219 goto out;
220 preempt_disable();
221 rtp = &__get_cpu_var(radix_tree_preloads);
222 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
223 rtp->nodes[rtp->nr++] = node;
224 else
225 kmem_cache_free(radix_tree_node_cachep, node);
226 }
227 ret = 0;
228out:
229 return ret;
230}
David Chinnerd7f09232007-07-14 16:05:04 +1000231EXPORT_SYMBOL(radix_tree_preload);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Nick Piggin6e954b92006-01-08 01:01:40 -0800233/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 * Return the maximum key which can be store into a
235 * radix tree with height HEIGHT.
236 */
237static inline unsigned long radix_tree_maxindex(unsigned int height)
238{
239 return height_to_maxindex[height];
240}
241
242/*
243 * Extend a radix tree so it can store key @index.
244 */
245static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
246{
247 struct radix_tree_node *node;
248 unsigned int height;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 int tag;
250
251 /* Figure out what the height should be. */
252 height = root->height + 1;
253 while (index > radix_tree_maxindex(height))
254 height++;
255
256 if (root->rnode == NULL) {
257 root->height = height;
258 goto out;
259 }
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 do {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800262 unsigned int newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 if (!(node = radix_tree_node_alloc(root)))
264 return -ENOMEM;
265
266 /* Increase the height. */
Nick Pigginc0bc9872007-10-16 01:24:42 -0700267 node->slots[0] = radix_tree_indirect_to_ptr(root->rnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269 /* Propagate the aggregated tag info into the new root */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800270 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -0700271 if (root_tag_get(root, tag))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 tag_set(node, tag, 0);
273 }
274
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800275 newheight = root->height+1;
276 node->height = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 node->count = 1;
Nick Pigginc0bc9872007-10-16 01:24:42 -0700278 node = radix_tree_ptr_to_indirect(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800279 rcu_assign_pointer(root->rnode, node);
280 root->height = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 } while (height > root->height);
282out:
283 return 0;
284}
285
286/**
287 * radix_tree_insert - insert into a radix tree
288 * @root: radix tree root
289 * @index: index key
290 * @item: item to insert
291 *
292 * Insert an item into the radix tree at position @index.
293 */
294int radix_tree_insert(struct radix_tree_root *root,
295 unsigned long index, void *item)
296{
Christoph Lameter201b6262005-09-06 15:16:46 -0700297 struct radix_tree_node *node = NULL, *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 unsigned int height, shift;
299 int offset;
300 int error;
301
Nick Pigginc0bc9872007-10-16 01:24:42 -0700302 BUG_ON(radix_tree_is_indirect_ptr(item));
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800303
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 /* Make sure the tree is high enough. */
Nick Piggin612d6c12006-06-23 02:03:22 -0700305 if (index > radix_tree_maxindex(root->height)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 error = radix_tree_extend(root, index);
307 if (error)
308 return error;
309 }
310
Nick Pigginc0bc9872007-10-16 01:24:42 -0700311 slot = radix_tree_indirect_to_ptr(root->rnode);
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 height = root->height;
314 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
315
316 offset = 0; /* uninitialised var warning */
Nick Piggin612d6c12006-06-23 02:03:22 -0700317 while (height > 0) {
Christoph Lameter201b6262005-09-06 15:16:46 -0700318 if (slot == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 /* Have to add a child node. */
Christoph Lameter201b6262005-09-06 15:16:46 -0700320 if (!(slot = radix_tree_node_alloc(root)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 return -ENOMEM;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800322 slot->height = height;
Christoph Lameter201b6262005-09-06 15:16:46 -0700323 if (node) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800324 rcu_assign_pointer(node->slots[offset], slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 node->count++;
Christoph Lameter201b6262005-09-06 15:16:46 -0700326 } else
Nick Pigginc0bc9872007-10-16 01:24:42 -0700327 rcu_assign_pointer(root->rnode,
328 radix_tree_ptr_to_indirect(slot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
330
331 /* Go a level down */
332 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Christoph Lameter201b6262005-09-06 15:16:46 -0700333 node = slot;
334 slot = node->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 shift -= RADIX_TREE_MAP_SHIFT;
336 height--;
Nick Piggin612d6c12006-06-23 02:03:22 -0700337 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Christoph Lameter201b6262005-09-06 15:16:46 -0700339 if (slot != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 return -EEXIST;
Christoph Lameter201b6262005-09-06 15:16:46 -0700341
Nick Piggin612d6c12006-06-23 02:03:22 -0700342 if (node) {
343 node->count++;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800344 rcu_assign_pointer(node->slots[offset], item);
Nick Piggin612d6c12006-06-23 02:03:22 -0700345 BUG_ON(tag_get(node, 0, offset));
346 BUG_ON(tag_get(node, 1, offset));
347 } else {
Nick Pigginc0bc9872007-10-16 01:24:42 -0700348 rcu_assign_pointer(root->rnode, item);
Nick Piggin612d6c12006-06-23 02:03:22 -0700349 BUG_ON(root_tag_get(root, 0));
350 BUG_ON(root_tag_get(root, 1));
351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return 0;
354}
355EXPORT_SYMBOL(radix_tree_insert);
356
Huang Shijieb72b71c2009-06-16 15:33:42 -0700357/*
358 * is_slot == 1 : search for the slot.
359 * is_slot == 0 : search for the node.
Hans Reisera4331362005-11-07 00:59:29 -0800360 */
Huang Shijieb72b71c2009-06-16 15:33:42 -0700361static void *radix_tree_lookup_element(struct radix_tree_root *root,
362 unsigned long index, int is_slot)
Hans Reisera4331362005-11-07 00:59:29 -0800363{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800364 unsigned int height, shift;
365 struct radix_tree_node *node, **slot;
366
Paul E. McKenney2676a582010-02-22 17:04:54 -0800367 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800368 if (node == NULL)
369 return NULL;
370
Nick Pigginc0bc9872007-10-16 01:24:42 -0700371 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800372 if (index > 0)
373 return NULL;
Huang Shijieb72b71c2009-06-16 15:33:42 -0700374 return is_slot ? (void *)&root->rnode : node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800375 }
Nick Pigginc0bc9872007-10-16 01:24:42 -0700376 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800377
378 height = node->height;
379 if (index > radix_tree_maxindex(height))
380 return NULL;
381
382 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
383
384 do {
385 slot = (struct radix_tree_node **)
386 (node->slots + ((index>>shift) & RADIX_TREE_MAP_MASK));
Paul E. McKenney2676a582010-02-22 17:04:54 -0800387 node = rcu_dereference_raw(*slot);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800388 if (node == NULL)
389 return NULL;
390
391 shift -= RADIX_TREE_MAP_SHIFT;
392 height--;
393 } while (height > 0);
394
Huang Shijieb72b71c2009-06-16 15:33:42 -0700395 return is_slot ? (void *)slot:node;
396}
397
398/**
399 * radix_tree_lookup_slot - lookup a slot in a radix tree
400 * @root: radix tree root
401 * @index: index key
402 *
403 * Returns: the slot corresponding to the position @index in the
404 * radix tree @root. This is useful for update-if-exists operations.
405 *
406 * This function can be called under rcu_read_lock iff the slot is not
407 * modified by radix_tree_replace_slot, otherwise it must be called
408 * exclusive from other writers. Any dereference of the slot must be done
409 * using radix_tree_deref_slot.
410 */
411void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
412{
413 return (void **)radix_tree_lookup_element(root, index, 1);
Hans Reisera4331362005-11-07 00:59:29 -0800414}
415EXPORT_SYMBOL(radix_tree_lookup_slot);
416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417/**
418 * radix_tree_lookup - perform lookup operation on a radix tree
419 * @root: radix tree root
420 * @index: index key
421 *
422 * Lookup the item at the position @index in the radix tree @root.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800423 *
424 * This function can be called under rcu_read_lock, however the caller
425 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
426 * them safely). No RCU barriers are required to access or modify the
427 * returned item, however.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 */
429void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
430{
Huang Shijieb72b71c2009-06-16 15:33:42 -0700431 return radix_tree_lookup_element(root, index, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433EXPORT_SYMBOL(radix_tree_lookup);
434
435/**
436 * radix_tree_tag_set - set a tag on a radix tree node
437 * @root: radix tree root
438 * @index: index key
439 * @tag: tag index
440 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800441 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
442 * corresponding to @index in the radix tree. From
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 * the root all the way down to the leaf node.
444 *
445 * Returns the address of the tagged item. Setting a tag on a not-present
446 * item is a bug.
447 */
448void *radix_tree_tag_set(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800449 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
451 unsigned int height, shift;
Christoph Lameter201b6262005-09-06 15:16:46 -0700452 struct radix_tree_node *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 height = root->height;
Peter Zijlstra4c91c362006-06-23 02:03:25 -0700455 BUG_ON(index > radix_tree_maxindex(height));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
Nick Pigginc0bc9872007-10-16 01:24:42 -0700457 slot = radix_tree_indirect_to_ptr(root->rnode);
Nick Piggin612d6c12006-06-23 02:03:22 -0700458 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
460 while (height > 0) {
461 int offset;
462
463 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Nick Piggind5274262006-01-08 01:01:41 -0800464 if (!tag_get(slot, tag, offset))
465 tag_set(slot, tag, offset);
Christoph Lameter201b6262005-09-06 15:16:46 -0700466 slot = slot->slots[offset];
467 BUG_ON(slot == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 shift -= RADIX_TREE_MAP_SHIFT;
469 height--;
470 }
471
Nick Piggin612d6c12006-06-23 02:03:22 -0700472 /* set the root's tag bit */
473 if (slot && !root_tag_get(root, tag))
474 root_tag_set(root, tag);
475
Christoph Lameter201b6262005-09-06 15:16:46 -0700476 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478EXPORT_SYMBOL(radix_tree_tag_set);
479
480/**
481 * radix_tree_tag_clear - clear a tag on a radix tree node
482 * @root: radix tree root
483 * @index: index key
484 * @tag: tag index
485 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800486 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
487 * corresponding to @index in the radix tree. If
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 * this causes the leaf node to have no tags set then clear the tag in the
489 * next-to-leaf node, etc.
490 *
491 * Returns the address of the tagged item on success, else NULL. ie:
492 * has the same return value and semantics as radix_tree_lookup().
493 */
494void *radix_tree_tag_clear(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800495 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Jeff Moyer26fb1582007-10-16 01:24:49 -0700497 /*
498 * The radix tree path needs to be one longer than the maximum path
499 * since the "list" is null terminated.
500 */
501 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
Nick Piggin612d6c12006-06-23 02:03:22 -0700502 struct radix_tree_node *slot = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 unsigned int height, shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
505 height = root->height;
506 if (index > radix_tree_maxindex(height))
507 goto out;
508
509 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
510 pathp->node = NULL;
Nick Pigginc0bc9872007-10-16 01:24:42 -0700511 slot = radix_tree_indirect_to_ptr(root->rnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513 while (height > 0) {
514 int offset;
515
Christoph Lameter201b6262005-09-06 15:16:46 -0700516 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 goto out;
518
519 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
520 pathp[1].offset = offset;
Christoph Lameter201b6262005-09-06 15:16:46 -0700521 pathp[1].node = slot;
522 slot = slot->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 pathp++;
524 shift -= RADIX_TREE_MAP_SHIFT;
525 height--;
526 }
527
Nick Piggin612d6c12006-06-23 02:03:22 -0700528 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 goto out;
530
Nick Piggin612d6c12006-06-23 02:03:22 -0700531 while (pathp->node) {
Nick Piggind5274262006-01-08 01:01:41 -0800532 if (!tag_get(pathp->node, tag, pathp->offset))
533 goto out;
Christoph Lameter201b6262005-09-06 15:16:46 -0700534 tag_clear(pathp->node, tag, pathp->offset);
Nick Piggin6e954b92006-01-08 01:01:40 -0800535 if (any_tag_set(pathp->node, tag))
536 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 pathp--;
Nick Piggin612d6c12006-06-23 02:03:22 -0700538 }
539
540 /* clear the root's tag bit */
541 if (root_tag_get(root, tag))
542 root_tag_clear(root, tag);
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544out:
Nick Piggin612d6c12006-06-23 02:03:22 -0700545 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547EXPORT_SYMBOL(radix_tree_tag_clear);
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549/**
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700550 * radix_tree_tag_get - get a tag on a radix tree node
551 * @root: radix tree root
552 * @index: index key
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800553 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 *
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700555 * Return values:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 *
Nick Piggin612d6c12006-06-23 02:03:22 -0700557 * 0: tag not present or not set
558 * 1: tag set
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 */
560int radix_tree_tag_get(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800561 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
563 unsigned int height, shift;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800564 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 int saw_unset_tag = 0;
566
Nick Piggin612d6c12006-06-23 02:03:22 -0700567 /* check the root's tag bit */
568 if (!root_tag_get(root, tag))
569 return 0;
570
Paul E. McKenney2676a582010-02-22 17:04:54 -0800571 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800572 if (node == NULL)
573 return 0;
574
Nick Pigginc0bc9872007-10-16 01:24:42 -0700575 if (!radix_tree_is_indirect_ptr(node))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800576 return (index == 0);
Nick Pigginc0bc9872007-10-16 01:24:42 -0700577 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800578
579 height = node->height;
580 if (index > radix_tree_maxindex(height))
581 return 0;
Nick Piggin612d6c12006-06-23 02:03:22 -0700582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 for ( ; ; ) {
586 int offset;
587
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800588 if (node == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 return 0;
590
591 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
592
593 /*
594 * This is just a debug check. Later, we can bale as soon as
595 * we see an unset tag.
596 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800597 if (!tag_get(node, tag, offset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 saw_unset_tag = 1;
599 if (height == 1) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800600 int ret = tag_get(node, tag, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
602 BUG_ON(ret && saw_unset_tag);
Wu Fengguange5dcd902006-06-25 05:48:14 -0700603 return !!ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 }
Paul E. McKenney2676a582010-02-22 17:04:54 -0800605 node = rcu_dereference_raw(node->slots[offset]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 shift -= RADIX_TREE_MAP_SHIFT;
607 height--;
608 }
609}
610EXPORT_SYMBOL(radix_tree_tag_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700612/**
613 * radix_tree_next_hole - find the next hole (not-present entry)
614 * @root: tree root
615 * @index: index key
616 * @max_scan: maximum range to search
617 *
618 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the lowest
619 * indexed hole.
620 *
621 * Returns: the index of the hole if found, otherwise returns an index
622 * outside of the set specified (in which case 'return - index >= max_scan'
Wu Fengguang8e6bdb72008-11-27 11:42:22 +0100623 * will be true). In rare cases of index wrap-around, 0 will be returned.
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700624 *
625 * radix_tree_next_hole may be called under rcu_read_lock. However, like
Wu Fengguang8e6bdb72008-11-27 11:42:22 +0100626 * radix_tree_gang_lookup, this will not atomically search a snapshot of
627 * the tree at a single point in time. For example, if a hole is created
628 * at index 5, then subsequently a hole is created at index 10,
629 * radix_tree_next_hole covering both indexes may return 10 if called
630 * under rcu_read_lock.
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700631 */
632unsigned long radix_tree_next_hole(struct radix_tree_root *root,
633 unsigned long index, unsigned long max_scan)
634{
635 unsigned long i;
636
637 for (i = 0; i < max_scan; i++) {
638 if (!radix_tree_lookup(root, index))
639 break;
640 index++;
641 if (index == 0)
642 break;
643 }
644
645 return index;
646}
647EXPORT_SYMBOL(radix_tree_next_hole);
648
Wu Fengguangdc566122009-06-16 15:31:32 -0700649/**
650 * radix_tree_prev_hole - find the prev hole (not-present entry)
651 * @root: tree root
652 * @index: index key
653 * @max_scan: maximum range to search
654 *
655 * Search backwards in the range [max(index-max_scan+1, 0), index]
656 * for the first hole.
657 *
658 * Returns: the index of the hole if found, otherwise returns an index
659 * outside of the set specified (in which case 'index - return >= max_scan'
660 * will be true). In rare cases of wrap-around, LONG_MAX will be returned.
661 *
662 * radix_tree_next_hole may be called under rcu_read_lock. However, like
663 * radix_tree_gang_lookup, this will not atomically search a snapshot of
664 * the tree at a single point in time. For example, if a hole is created
665 * at index 10, then subsequently a hole is created at index 5,
666 * radix_tree_prev_hole covering both indexes may return 5 if called under
667 * rcu_read_lock.
668 */
669unsigned long radix_tree_prev_hole(struct radix_tree_root *root,
670 unsigned long index, unsigned long max_scan)
671{
672 unsigned long i;
673
674 for (i = 0; i < max_scan; i++) {
675 if (!radix_tree_lookup(root, index))
676 break;
677 index--;
678 if (index == LONG_MAX)
679 break;
680 }
681
682 return index;
683}
684EXPORT_SYMBOL(radix_tree_prev_hole);
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686static unsigned int
Nick Piggin47feff22008-07-25 19:45:29 -0700687__lookup(struct radix_tree_node *slot, void ***results, unsigned long index,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 unsigned int max_items, unsigned long *next_index)
689{
690 unsigned int nr_found = 0;
Christoph Lameter201b6262005-09-06 15:16:46 -0700691 unsigned int shift, height;
Christoph Lameter201b6262005-09-06 15:16:46 -0700692 unsigned long i;
693
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800694 height = slot->height;
695 if (height == 0)
Christoph Lameter201b6262005-09-06 15:16:46 -0700696 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Christoph Lameter201b6262005-09-06 15:16:46 -0700699 for ( ; height > 1; height--) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800700 i = (index >> shift) & RADIX_TREE_MAP_MASK;
701 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (slot->slots[i] != NULL)
703 break;
704 index &= ~((1UL << shift) - 1);
705 index += 1UL << shift;
706 if (index == 0)
707 goto out; /* 32-bit wraparound */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800708 i++;
709 if (i == RADIX_TREE_MAP_SIZE)
710 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 shift -= RADIX_TREE_MAP_SHIFT;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800714 slot = rcu_dereference_raw(slot->slots[i]);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800715 if (slot == NULL)
716 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 }
Christoph Lameter201b6262005-09-06 15:16:46 -0700718
719 /* Bottom level: grab some items */
720 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
721 index++;
Nick Piggin47feff22008-07-25 19:45:29 -0700722 if (slot->slots[i]) {
723 results[nr_found++] = &(slot->slots[i]);
Christoph Lameter201b6262005-09-06 15:16:46 -0700724 if (nr_found == max_items)
725 goto out;
726 }
727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728out:
729 *next_index = index;
730 return nr_found;
731}
732
733/**
734 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
735 * @root: radix tree root
736 * @results: where the results of the lookup are placed
737 * @first_index: start the lookup from this key
738 * @max_items: place up to this many items at *results
739 *
740 * Performs an index-ascending scan of the tree for present items. Places
741 * them at *@results and returns the number of items which were placed at
742 * *@results.
743 *
744 * The implementation is naive.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800745 *
746 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
747 * rcu_read_lock. In this case, rather than the returned results being
748 * an atomic snapshot of the tree at a single point in time, the semantics
749 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
750 * have been issued in individual locks, and results stored in 'results'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 */
752unsigned int
753radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
754 unsigned long first_index, unsigned int max_items)
755{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800756 unsigned long max_index;
757 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 unsigned long cur_index = first_index;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800759 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Paul E. McKenney2676a582010-02-22 17:04:54 -0800761 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800762 if (!node)
763 return 0;
764
Nick Pigginc0bc9872007-10-16 01:24:42 -0700765 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800766 if (first_index > 0)
767 return 0;
Nick Pigginc0bc9872007-10-16 01:24:42 -0700768 results[0] = node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800769 return 1;
770 }
Nick Pigginc0bc9872007-10-16 01:24:42 -0700771 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800772
773 max_index = radix_tree_maxindex(node->height);
774
775 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 while (ret < max_items) {
Nick Piggin47feff22008-07-25 19:45:29 -0700777 unsigned int nr_found, slots_found, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 unsigned long next_index; /* Index of next search */
779
780 if (cur_index > max_index)
781 break;
Nick Piggin47feff22008-07-25 19:45:29 -0700782 slots_found = __lookup(node, (void ***)results + ret, cur_index,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 max_items - ret, &next_index);
Nick Piggin47feff22008-07-25 19:45:29 -0700784 nr_found = 0;
785 for (i = 0; i < slots_found; i++) {
786 struct radix_tree_node *slot;
787 slot = *(((void ***)results)[ret + i]);
788 if (!slot)
789 continue;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800790 results[ret + nr_found] = rcu_dereference_raw(slot);
Nick Piggin47feff22008-07-25 19:45:29 -0700791 nr_found++;
792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 ret += nr_found;
794 if (next_index == 0)
795 break;
796 cur_index = next_index;
797 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return ret;
800}
801EXPORT_SYMBOL(radix_tree_gang_lookup);
802
Nick Piggin47feff22008-07-25 19:45:29 -0700803/**
804 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
805 * @root: radix tree root
806 * @results: where the results of the lookup are placed
807 * @first_index: start the lookup from this key
808 * @max_items: place up to this many items at *results
809 *
810 * Performs an index-ascending scan of the tree for present items. Places
811 * their slots at *@results and returns the number of items which were
812 * placed at *@results.
813 *
814 * The implementation is naive.
815 *
816 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
817 * be dereferenced with radix_tree_deref_slot, and if using only RCU
818 * protection, radix_tree_deref_slot may fail requiring a retry.
819 */
820unsigned int
821radix_tree_gang_lookup_slot(struct radix_tree_root *root, void ***results,
822 unsigned long first_index, unsigned int max_items)
823{
824 unsigned long max_index;
825 struct radix_tree_node *node;
826 unsigned long cur_index = first_index;
827 unsigned int ret;
828
Paul E. McKenney2676a582010-02-22 17:04:54 -0800829 node = rcu_dereference_raw(root->rnode);
Nick Piggin47feff22008-07-25 19:45:29 -0700830 if (!node)
831 return 0;
832
833 if (!radix_tree_is_indirect_ptr(node)) {
834 if (first_index > 0)
835 return 0;
836 results[0] = (void **)&root->rnode;
837 return 1;
838 }
839 node = radix_tree_indirect_to_ptr(node);
840
841 max_index = radix_tree_maxindex(node->height);
842
843 ret = 0;
844 while (ret < max_items) {
845 unsigned int slots_found;
846 unsigned long next_index; /* Index of next search */
847
848 if (cur_index > max_index)
849 break;
850 slots_found = __lookup(node, results + ret, cur_index,
851 max_items - ret, &next_index);
852 ret += slots_found;
853 if (next_index == 0)
854 break;
855 cur_index = next_index;
856 }
857
858 return ret;
859}
860EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862/*
863 * FIXME: the two tag_get()s here should use find_next_bit() instead of
864 * open-coding the search.
865 */
866static unsigned int
Nick Piggin47feff22008-07-25 19:45:29 -0700867__lookup_tag(struct radix_tree_node *slot, void ***results, unsigned long index,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800868 unsigned int max_items, unsigned long *next_index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
870 unsigned int nr_found = 0;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800871 unsigned int shift, height;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800873 height = slot->height;
874 if (height == 0)
Nick Piggin612d6c12006-06-23 02:03:22 -0700875 goto out;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800876 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
Nick Piggin612d6c12006-06-23 02:03:22 -0700877
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800878 while (height > 0) {
879 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800881 for (;;) {
882 if (tag_get(slot, tag, i))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 index &= ~((1UL << shift) - 1);
885 index += 1UL << shift;
886 if (index == 0)
887 goto out; /* 32-bit wraparound */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800888 i++;
889 if (i == RADIX_TREE_MAP_SIZE)
890 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 height--;
893 if (height == 0) { /* Bottom level: grab some items */
894 unsigned long j = index & RADIX_TREE_MAP_MASK;
895
896 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
897 index++;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800898 if (!tag_get(slot, tag, j))
899 continue;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800900 /*
901 * Even though the tag was found set, we need to
902 * recheck that we have a non-NULL node, because
903 * if this lookup is lockless, it may have been
904 * subsequently deleted.
905 *
906 * Similar care must be taken in any place that
907 * lookup ->slots[x] without a lock (ie. can't
908 * rely on its value remaining the same).
909 */
Nick Piggin47feff22008-07-25 19:45:29 -0700910 if (slot->slots[j]) {
911 results[nr_found++] = &(slot->slots[j]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (nr_found == max_items)
913 goto out;
914 }
915 }
916 }
917 shift -= RADIX_TREE_MAP_SHIFT;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800918 slot = rcu_dereference_raw(slot->slots[i]);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800919 if (slot == NULL)
920 break;
921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922out:
923 *next_index = index;
924 return nr_found;
925}
926
927/**
928 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
929 * based on a tag
930 * @root: radix tree root
931 * @results: where the results of the lookup are placed
932 * @first_index: start the lookup from this key
933 * @max_items: place up to this many items at *results
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800934 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 *
936 * Performs an index-ascending scan of the tree for present items which
937 * have the tag indexed by @tag set. Places the items at *@results and
938 * returns the number of items which were placed at *@results.
939 */
940unsigned int
941radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800942 unsigned long first_index, unsigned int max_items,
943 unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800945 struct radix_tree_node *node;
946 unsigned long max_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 unsigned long cur_index = first_index;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800948 unsigned int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
Nick Piggin612d6c12006-06-23 02:03:22 -0700950 /* check the root's tag bit */
951 if (!root_tag_get(root, tag))
952 return 0;
953
Paul E. McKenney2676a582010-02-22 17:04:54 -0800954 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800955 if (!node)
956 return 0;
957
Nick Pigginc0bc9872007-10-16 01:24:42 -0700958 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800959 if (first_index > 0)
960 return 0;
Nick Pigginc0bc9872007-10-16 01:24:42 -0700961 results[0] = node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800962 return 1;
963 }
Nick Pigginc0bc9872007-10-16 01:24:42 -0700964 node = radix_tree_indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800965
966 max_index = radix_tree_maxindex(node->height);
967
968 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 while (ret < max_items) {
Nick Piggin47feff22008-07-25 19:45:29 -0700970 unsigned int nr_found, slots_found, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 unsigned long next_index; /* Index of next search */
972
973 if (cur_index > max_index)
974 break;
Nick Piggin47feff22008-07-25 19:45:29 -0700975 slots_found = __lookup_tag(node, (void ***)results + ret,
976 cur_index, max_items - ret, &next_index, tag);
977 nr_found = 0;
978 for (i = 0; i < slots_found; i++) {
979 struct radix_tree_node *slot;
980 slot = *(((void ***)results)[ret + i]);
981 if (!slot)
982 continue;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800983 results[ret + nr_found] = rcu_dereference_raw(slot);
Nick Piggin47feff22008-07-25 19:45:29 -0700984 nr_found++;
985 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 ret += nr_found;
987 if (next_index == 0)
988 break;
989 cur_index = next_index;
990 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 return ret;
993}
994EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
995
996/**
Nick Piggin47feff22008-07-25 19:45:29 -0700997 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
998 * radix tree based on a tag
999 * @root: radix tree root
1000 * @results: where the results of the lookup are placed
1001 * @first_index: start the lookup from this key
1002 * @max_items: place up to this many items at *results
1003 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1004 *
1005 * Performs an index-ascending scan of the tree for present items which
1006 * have the tag indexed by @tag set. Places the slots at *@results and
1007 * returns the number of slots which were placed at *@results.
1008 */
1009unsigned int
1010radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1011 unsigned long first_index, unsigned int max_items,
1012 unsigned int tag)
1013{
1014 struct radix_tree_node *node;
1015 unsigned long max_index;
1016 unsigned long cur_index = first_index;
1017 unsigned int ret;
1018
1019 /* check the root's tag bit */
1020 if (!root_tag_get(root, tag))
1021 return 0;
1022
Paul E. McKenney2676a582010-02-22 17:04:54 -08001023 node = rcu_dereference_raw(root->rnode);
Nick Piggin47feff22008-07-25 19:45:29 -07001024 if (!node)
1025 return 0;
1026
1027 if (!radix_tree_is_indirect_ptr(node)) {
1028 if (first_index > 0)
1029 return 0;
1030 results[0] = (void **)&root->rnode;
1031 return 1;
1032 }
1033 node = radix_tree_indirect_to_ptr(node);
1034
1035 max_index = radix_tree_maxindex(node->height);
1036
1037 ret = 0;
1038 while (ret < max_items) {
1039 unsigned int slots_found;
1040 unsigned long next_index; /* Index of next search */
1041
1042 if (cur_index > max_index)
1043 break;
1044 slots_found = __lookup_tag(node, results + ret,
1045 cur_index, max_items - ret, &next_index, tag);
1046 ret += slots_found;
1047 if (next_index == 0)
1048 break;
1049 cur_index = next_index;
1050 }
1051
1052 return ret;
1053}
1054EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1055
1056
1057/**
Nick Piggina5f51c92006-01-08 01:01:41 -08001058 * radix_tree_shrink - shrink height of a radix tree to minimal
1059 * @root radix tree root
1060 */
1061static inline void radix_tree_shrink(struct radix_tree_root *root)
1062{
1063 /* try to shrink tree height */
Nick Pigginc0bc9872007-10-16 01:24:42 -07001064 while (root->height > 0) {
Nick Piggina5f51c92006-01-08 01:01:41 -08001065 struct radix_tree_node *to_free = root->rnode;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001066 void *newptr;
Nick Piggina5f51c92006-01-08 01:01:41 -08001067
Nick Pigginc0bc9872007-10-16 01:24:42 -07001068 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1069 to_free = radix_tree_indirect_to_ptr(to_free);
1070
1071 /*
1072 * The candidate node has more than one child, or its child
1073 * is not at the leftmost slot, we cannot shrink.
1074 */
1075 if (to_free->count != 1)
1076 break;
1077 if (!to_free->slots[0])
1078 break;
1079
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001080 /*
1081 * We don't need rcu_assign_pointer(), since we are simply
1082 * moving the node from one part of the tree to another. If
1083 * it was safe to dereference the old pointer to it
1084 * (to_free->slots[0]), it will be safe to dereference the new
1085 * one (root->rnode).
1086 */
1087 newptr = to_free->slots[0];
Nick Pigginc0bc9872007-10-16 01:24:42 -07001088 if (root->height > 1)
1089 newptr = radix_tree_ptr_to_indirect(newptr);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001090 root->rnode = newptr;
Nick Piggina5f51c92006-01-08 01:01:41 -08001091 root->height--;
Nick Piggina5f51c92006-01-08 01:01:41 -08001092 radix_tree_node_free(to_free);
1093 }
1094}
1095
1096/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 * radix_tree_delete - delete an item from a radix tree
1098 * @root: radix tree root
1099 * @index: index key
1100 *
1101 * Remove the item at @index from the radix tree rooted at @root.
1102 *
1103 * Returns the address of the deleted item, or NULL if it was not present.
1104 */
1105void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1106{
Jeff Moyer26fb1582007-10-16 01:24:49 -07001107 /*
1108 * The radix tree path needs to be one longer than the maximum path
1109 * since the "list" is null terminated.
1110 */
1111 struct radix_tree_path path[RADIX_TREE_MAX_PATH + 1], *pathp = path;
Nick Piggin612d6c12006-06-23 02:03:22 -07001112 struct radix_tree_node *slot = NULL;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001113 struct radix_tree_node *to_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 unsigned int height, shift;
Nick Piggind5274262006-01-08 01:01:41 -08001115 int tag;
1116 int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
1118 height = root->height;
1119 if (index > radix_tree_maxindex(height))
1120 goto out;
1121
Nick Piggin612d6c12006-06-23 02:03:22 -07001122 slot = root->rnode;
Nick Pigginc0bc9872007-10-16 01:24:42 -07001123 if (height == 0) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001124 root_tag_clear_all(root);
1125 root->rnode = NULL;
1126 goto out;
1127 }
Nick Pigginc0bc9872007-10-16 01:24:42 -07001128 slot = radix_tree_indirect_to_ptr(slot);
Nick Piggin612d6c12006-06-23 02:03:22 -07001129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
1131 pathp->node = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Nick Piggin612d6c12006-06-23 02:03:22 -07001133 do {
Christoph Lameter201b6262005-09-06 15:16:46 -07001134 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 goto out;
1136
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 pathp++;
Nick Piggind5274262006-01-08 01:01:41 -08001138 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1139 pathp->offset = offset;
1140 pathp->node = slot;
1141 slot = slot->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 shift -= RADIX_TREE_MAP_SHIFT;
Nick Piggin612d6c12006-06-23 02:03:22 -07001143 height--;
1144 } while (height > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Nick Piggin612d6c12006-06-23 02:03:22 -07001146 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 goto out;
1148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 /*
1150 * Clear all tags associated with the just-deleted item
1151 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001152 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001153 if (tag_get(pathp->node, tag, pathp->offset))
1154 radix_tree_tag_clear(root, index, tag);
Nick Piggind5274262006-01-08 01:01:41 -08001155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001157 to_free = NULL;
Christoph Lameter201b6262005-09-06 15:16:46 -07001158 /* Now free the nodes we do not need anymore */
Nick Piggin612d6c12006-06-23 02:03:22 -07001159 while (pathp->node) {
Christoph Lameter201b6262005-09-06 15:16:46 -07001160 pathp->node->slots[pathp->offset] = NULL;
Nick Piggina5f51c92006-01-08 01:01:41 -08001161 pathp->node->count--;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001162 /*
1163 * Queue the node for deferred freeing after the
1164 * last reference to it disappears (set NULL, above).
1165 */
1166 if (to_free)
1167 radix_tree_node_free(to_free);
Nick Piggina5f51c92006-01-08 01:01:41 -08001168
1169 if (pathp->node->count) {
Nick Pigginc0bc9872007-10-16 01:24:42 -07001170 if (pathp->node ==
1171 radix_tree_indirect_to_ptr(root->rnode))
Nick Piggina5f51c92006-01-08 01:01:41 -08001172 radix_tree_shrink(root);
Christoph Lameter201b6262005-09-06 15:16:46 -07001173 goto out;
Nick Piggina5f51c92006-01-08 01:01:41 -08001174 }
Christoph Lameter201b6262005-09-06 15:16:46 -07001175
1176 /* Node with zero slots in use so free it */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001177 to_free = pathp->node;
Nick Piggin612d6c12006-06-23 02:03:22 -07001178 pathp--;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001179
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 }
Nick Piggin612d6c12006-06-23 02:03:22 -07001181 root_tag_clear_all(root);
Christoph Lameter201b6262005-09-06 15:16:46 -07001182 root->height = 0;
Nick Piggin612d6c12006-06-23 02:03:22 -07001183 root->rnode = NULL;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001184 if (to_free)
1185 radix_tree_node_free(to_free);
Nick Piggin612d6c12006-06-23 02:03:22 -07001186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187out:
Nick Piggin612d6c12006-06-23 02:03:22 -07001188 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189}
1190EXPORT_SYMBOL(radix_tree_delete);
1191
1192/**
1193 * radix_tree_tagged - test whether any items in the tree are tagged
1194 * @root: radix tree root
1195 * @tag: tag to test
1196 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001197int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
Nick Piggin612d6c12006-06-23 02:03:22 -07001199 return root_tag_get(root, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
1201EXPORT_SYMBOL(radix_tree_tagged);
1202
1203static void
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001204radix_tree_node_ctor(void *node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
1206 memset(node, 0, sizeof(struct radix_tree_node));
1207}
1208
1209static __init unsigned long __maxindex(unsigned int height)
1210{
Peter Lund430d2752007-10-16 23:29:35 -07001211 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1212 int shift = RADIX_TREE_INDEX_BITS - width;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Peter Lund430d2752007-10-16 23:29:35 -07001214 if (shift < 0)
1215 return ~0UL;
1216 if (shift >= BITS_PER_LONG)
1217 return 0UL;
1218 return ~0UL >> shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
1221static __init void radix_tree_init_maxindex(void)
1222{
1223 unsigned int i;
1224
1225 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1226 height_to_maxindex[i] = __maxindex(i);
1227}
1228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229static int radix_tree_callback(struct notifier_block *nfb,
1230 unsigned long action,
1231 void *hcpu)
1232{
1233 int cpu = (long)hcpu;
1234 struct radix_tree_preload *rtp;
1235
1236 /* Free per-cpu pool of perloaded nodes */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001237 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 rtp = &per_cpu(radix_tree_preloads, cpu);
1239 while (rtp->nr) {
1240 kmem_cache_free(radix_tree_node_cachep,
1241 rtp->nodes[rtp->nr-1]);
1242 rtp->nodes[rtp->nr-1] = NULL;
1243 rtp->nr--;
1244 }
1245 }
1246 return NOTIFY_OK;
1247}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249void __init radix_tree_init(void)
1250{
1251 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1252 sizeof(struct radix_tree_node), 0,
Christoph Lameter488514d2008-04-28 02:12:05 -07001253 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1254 radix_tree_node_ctor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 radix_tree_init_maxindex();
1256 hotcpu_notifier(radix_tree_callback, 0);
1257}