blob: 6b79e9026e24894000a2bdf4180db80f8190b357 [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
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -07006 * Copyright (C) 2012 Konstantin Khlebnikov
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as
10 * published by the Free Software Foundation; either version 2, or (at
11 * your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 */
22
23#include <linux/errno.h>
24#include <linux/init.h>
25#include <linux/kernel.h>
Paul Gortmaker8bc3bcc2011-11-16 21:29:17 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/radix-tree.h>
28#include <linux/percpu.h>
29#include <linux/slab.h>
Catalin Marinasce80b062014-06-06 14:38:18 -070030#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/notifier.h>
32#include <linux/cpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/string.h>
34#include <linux/bitops.h>
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080035#include <linux/rcupdate.h>
Frederic Weisbecker92cf2112015-05-12 16:41:46 +020036#include <linux/preempt.h> /* in_interrupt() */
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38
Jeff Moyer26fb1582007-10-16 01:24:49 -070039/*
40 * The height_to_maxindex array needs to be one deeper than the maximum
41 * path as height 0 holds only 1 entry.
42 */
43static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45/*
46 * Radix tree node cache.
47 */
Christoph Lametere18b8902006-12-06 20:33:20 -080048static struct kmem_cache *radix_tree_node_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50/*
Nick Piggin55368052012-05-29 15:07:34 -070051 * The radix tree is variable-height, so an insert operation not only has
52 * to build the branch to its corresponding item, it also has to build the
53 * branch to existing items if the size has to be increased (by
54 * radix_tree_extend).
55 *
56 * The worst case is a zero height tree with just a single item at index 0,
57 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
58 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
59 * Hence:
60 */
61#define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
62
63/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * Per-cpu pool of preloaded nodes
65 */
66struct radix_tree_preload {
67 int nr;
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -070068 /* nodes->private_data points to next preallocated node */
69 struct radix_tree_node *nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070};
Harvey Harrison8cef7d52009-01-06 14:40:50 -080071static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Nick Piggin27d20fd2010-11-11 14:05:19 -080073static inline void *ptr_to_indirect(void *ptr)
74{
75 return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR);
76}
77
78static inline void *indirect_to_ptr(void *ptr)
79{
80 return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR);
81}
82
Nick Piggin612d6c12006-06-23 02:03:22 -070083static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
84{
85 return root->gfp_mask & __GFP_BITS_MASK;
86}
87
Nick Piggin643b52b2008-06-12 15:21:52 -070088static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
89 int offset)
90{
91 __set_bit(offset, node->tags[tag]);
92}
93
94static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
95 int offset)
96{
97 __clear_bit(offset, node->tags[tag]);
98}
99
100static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
101 int offset)
102{
103 return test_bit(offset, node->tags[tag]);
104}
105
106static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
107{
108 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
109}
110
111static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
112{
113 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
114}
115
116static inline void root_tag_clear_all(struct radix_tree_root *root)
117{
118 root->gfp_mask &= __GFP_BITS_MASK;
119}
120
121static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
122{
123 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
124}
125
126/*
127 * Returns 1 if any slot in the node has this tag set.
128 * Otherwise returns 0.
129 */
130static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
131{
132 int idx;
133 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
134 if (node->tags[tag][idx])
135 return 1;
136 }
137 return 0;
138}
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700139
140/**
141 * radix_tree_find_next_bit - find the next set bit in a memory region
142 *
143 * @addr: The address to base the search on
144 * @size: The bitmap size in bits
145 * @offset: The bitnumber to start searching at
146 *
147 * Unrollable variant of find_next_bit() for constant size arrays.
148 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
149 * Returns next bit offset, or size if nothing found.
150 */
151static __always_inline unsigned long
152radix_tree_find_next_bit(const unsigned long *addr,
153 unsigned long size, unsigned long offset)
154{
155 if (!__builtin_constant_p(size))
156 return find_next_bit(addr, size, offset);
157
158 if (offset < size) {
159 unsigned long tmp;
160
161 addr += offset / BITS_PER_LONG;
162 tmp = *addr >> (offset % BITS_PER_LONG);
163 if (tmp)
164 return __ffs(tmp) + offset;
165 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
166 while (offset < size) {
167 tmp = *++addr;
168 if (tmp)
169 return __ffs(tmp) + offset;
170 offset += BITS_PER_LONG;
171 }
172 }
173 return size;
174}
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176/*
177 * This assumes that the caller has performed appropriate preallocation, and
178 * that the caller has pinned this thread of control to the current CPU.
179 */
180static struct radix_tree_node *
181radix_tree_node_alloc(struct radix_tree_root *root)
182{
Nick Piggine2848a02008-02-04 22:29:10 -0800183 struct radix_tree_node *ret = NULL;
Nick Piggin612d6c12006-06-23 02:03:22 -0700184 gfp_t gfp_mask = root_gfp_mask(root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Jan Kara5e4c0d972013-09-11 14:26:05 -0700186 /*
187 * Preload code isn't irq safe and it doesn't make sence to use
188 * preloading in the interrupt anyway as all the allocations have to
189 * be atomic. So just do normal allocation when in interrupt.
190 */
Mel Gormand0164ad2015-11-06 16:28:21 -0800191 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 struct radix_tree_preload *rtp;
193
Nick Piggine2848a02008-02-04 22:29:10 -0800194 /*
195 * Provided the caller has preloaded here, we will always
196 * succeed in getting a node here (and never reach
197 * kmem_cache_alloc)
198 */
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700199 rtp = this_cpu_ptr(&radix_tree_preloads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700201 ret = rtp->nodes;
202 rtp->nodes = ret->private_data;
203 ret->private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 rtp->nr--;
205 }
Catalin Marinasce80b062014-06-06 14:38:18 -0700206 /*
207 * Update the allocation stack trace as this is more useful
208 * for debugging.
209 */
210 kmemleak_update_trace(ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 }
Nick Piggine2848a02008-02-04 22:29:10 -0800212 if (ret == NULL)
Christoph Lameter488514d2008-04-28 02:12:05 -0700213 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Nick Piggine2848a02008-02-04 22:29:10 -0800214
Nick Pigginc0bc9872007-10-16 01:24:42 -0700215 BUG_ON(radix_tree_is_indirect_ptr(ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 return ret;
217}
218
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800219static void radix_tree_node_rcu_free(struct rcu_head *head)
220{
221 struct radix_tree_node *node =
222 container_of(head, struct radix_tree_node, rcu_head);
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000223 int i;
Nick Piggin643b52b2008-06-12 15:21:52 -0700224
225 /*
226 * must only free zeroed nodes into the slab. radix_tree_shrink
227 * can leave us with a non-NULL entry in the first slot, so clear
228 * that here to make sure.
229 */
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000230 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
231 tag_clear(node, i, 0);
232
Nick Piggin643b52b2008-06-12 15:21:52 -0700233 node->slots[0] = NULL;
234 node->count = 0;
235
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800236 kmem_cache_free(radix_tree_node_cachep, node);
237}
238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239static inline void
240radix_tree_node_free(struct radix_tree_node *node)
241{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800242 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
245/*
246 * Load up this CPU's radix_tree_node buffer with sufficient objects to
247 * ensure that the addition of a single element in the tree cannot fail. On
248 * success, return zero, with preemption disabled. On error, return -ENOMEM
249 * with preemption not disabled.
David Howellsb34df792009-11-19 18:11:14 +0000250 *
251 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800252 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 */
Jan Kara5e4c0d972013-09-11 14:26:05 -0700254static int __radix_tree_preload(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 struct radix_tree_preload *rtp;
257 struct radix_tree_node *node;
258 int ret = -ENOMEM;
259
260 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700261 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700262 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 preempt_enable();
Christoph Lameter488514d2008-04-28 02:12:05 -0700264 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 if (node == NULL)
266 goto out;
267 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700268 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700269 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
270 node->private_data = rtp->nodes;
271 rtp->nodes = node;
272 rtp->nr++;
273 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 kmem_cache_free(radix_tree_node_cachep, node);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
277 ret = 0;
278out:
279 return ret;
280}
Jan Kara5e4c0d972013-09-11 14:26:05 -0700281
282/*
283 * Load up this CPU's radix_tree_node buffer with sufficient objects to
284 * ensure that the addition of a single element in the tree cannot fail. On
285 * success, return zero, with preemption disabled. On error, return -ENOMEM
286 * with preemption not disabled.
287 *
288 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800289 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Jan Kara5e4c0d972013-09-11 14:26:05 -0700290 */
291int radix_tree_preload(gfp_t gfp_mask)
292{
293 /* Warn on non-sensical use... */
Mel Gormand0164ad2015-11-06 16:28:21 -0800294 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
Jan Kara5e4c0d972013-09-11 14:26:05 -0700295 return __radix_tree_preload(gfp_mask);
296}
David Chinnerd7f09232007-07-14 16:05:04 +1000297EXPORT_SYMBOL(radix_tree_preload);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Nick Piggin6e954b92006-01-08 01:01:40 -0800299/*
Jan Kara5e4c0d972013-09-11 14:26:05 -0700300 * The same as above function, except we don't guarantee preloading happens.
301 * We do it, if we decide it helps. On success, return zero with preemption
302 * disabled. On error, return -ENOMEM with preemption not disabled.
303 */
304int radix_tree_maybe_preload(gfp_t gfp_mask)
305{
Mel Gormand0164ad2015-11-06 16:28:21 -0800306 if (gfpflags_allow_blocking(gfp_mask))
Jan Kara5e4c0d972013-09-11 14:26:05 -0700307 return __radix_tree_preload(gfp_mask);
308 /* Preloading doesn't help anything with this gfp mask, skip it */
309 preempt_disable();
310 return 0;
311}
312EXPORT_SYMBOL(radix_tree_maybe_preload);
313
314/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * Return the maximum key which can be store into a
316 * radix tree with height HEIGHT.
317 */
318static inline unsigned long radix_tree_maxindex(unsigned int height)
319{
320 return height_to_maxindex[height];
321}
322
323/*
324 * Extend a radix tree so it can store key @index.
325 */
326static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
327{
328 struct radix_tree_node *node;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800329 struct radix_tree_node *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 unsigned int height;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 int tag;
332
333 /* Figure out what the height should be. */
334 height = root->height + 1;
335 while (index > radix_tree_maxindex(height))
336 height++;
337
338 if (root->rnode == NULL) {
339 root->height = height;
340 goto out;
341 }
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 do {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800344 unsigned int newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (!(node = radix_tree_node_alloc(root)))
346 return -ENOMEM;
347
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 /* Propagate the aggregated tag info into the new root */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800349 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -0700350 if (root_tag_get(root, tag))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 tag_set(node, tag, 0);
352 }
353
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800354 /* Increase the height. */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800355 newheight = root->height+1;
Johannes Weiner449dd692014-04-03 14:47:56 -0700356 BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK);
357 node->path = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 node->count = 1;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800359 node->parent = NULL;
360 slot = root->rnode;
361 if (newheight > 1) {
362 slot = indirect_to_ptr(slot);
363 slot->parent = node;
364 }
365 node->slots[0] = slot;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800366 node = ptr_to_indirect(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800367 rcu_assign_pointer(root->rnode, node);
368 root->height = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 } while (height > root->height);
370out:
371 return 0;
372}
373
374/**
Johannes Weiner139e5612014-04-03 14:47:54 -0700375 * __radix_tree_create - create a slot in a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 * @root: radix tree root
377 * @index: index key
Johannes Weiner139e5612014-04-03 14:47:54 -0700378 * @nodep: returns node
379 * @slotp: returns slot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 *
Johannes Weiner139e5612014-04-03 14:47:54 -0700381 * Create, if necessary, and return the node and slot for an item
382 * at position @index in the radix tree @root.
383 *
384 * Until there is more than one item in the tree, no nodes are
385 * allocated and @root->rnode is used as a direct slot instead of
386 * pointing to a node, in which case *@nodep will be NULL.
387 *
388 * Returns -ENOMEM, or 0 for success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700390int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
391 struct radix_tree_node **nodep, void ***slotp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Christoph Lameter201b6262005-09-06 15:16:46 -0700393 struct radix_tree_node *node = NULL, *slot;
Johannes Weiner139e5612014-04-03 14:47:54 -0700394 unsigned int height, shift, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 int error;
396
397 /* Make sure the tree is high enough. */
Nick Piggin612d6c12006-06-23 02:03:22 -0700398 if (index > radix_tree_maxindex(root->height)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 error = radix_tree_extend(root, index);
400 if (error)
401 return error;
402 }
403
Nick Piggin27d20fd2010-11-11 14:05:19 -0800404 slot = indirect_to_ptr(root->rnode);
Nick Pigginc0bc9872007-10-16 01:24:42 -0700405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 height = root->height;
407 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
408
409 offset = 0; /* uninitialised var warning */
Nick Piggin612d6c12006-06-23 02:03:22 -0700410 while (height > 0) {
Christoph Lameter201b6262005-09-06 15:16:46 -0700411 if (slot == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /* Have to add a child node. */
Christoph Lameter201b6262005-09-06 15:16:46 -0700413 if (!(slot = radix_tree_node_alloc(root)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 return -ENOMEM;
Johannes Weiner449dd692014-04-03 14:47:56 -0700415 slot->path = height;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800416 slot->parent = node;
Christoph Lameter201b6262005-09-06 15:16:46 -0700417 if (node) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800418 rcu_assign_pointer(node->slots[offset], slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 node->count++;
Johannes Weiner449dd692014-04-03 14:47:56 -0700420 slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT;
Christoph Lameter201b6262005-09-06 15:16:46 -0700421 } else
Nick Piggin27d20fd2010-11-11 14:05:19 -0800422 rcu_assign_pointer(root->rnode, ptr_to_indirect(slot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
424
425 /* Go a level down */
426 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Christoph Lameter201b6262005-09-06 15:16:46 -0700427 node = slot;
428 slot = node->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 shift -= RADIX_TREE_MAP_SHIFT;
430 height--;
Nick Piggin612d6c12006-06-23 02:03:22 -0700431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Johannes Weiner139e5612014-04-03 14:47:54 -0700433 if (nodep)
434 *nodep = node;
435 if (slotp)
436 *slotp = node ? node->slots + offset : (void **)&root->rnode;
437 return 0;
438}
439
440/**
441 * radix_tree_insert - insert into a radix tree
442 * @root: radix tree root
443 * @index: index key
444 * @item: item to insert
445 *
446 * Insert an item into the radix tree at position @index.
447 */
448int radix_tree_insert(struct radix_tree_root *root,
449 unsigned long index, void *item)
450{
451 struct radix_tree_node *node;
452 void **slot;
453 int error;
454
455 BUG_ON(radix_tree_is_indirect_ptr(item));
456
457 error = __radix_tree_create(root, index, &node, &slot);
458 if (error)
459 return error;
460 if (*slot != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 return -EEXIST;
Johannes Weiner139e5612014-04-03 14:47:54 -0700462 rcu_assign_pointer(*slot, item);
Christoph Lameter201b6262005-09-06 15:16:46 -0700463
Nick Piggin612d6c12006-06-23 02:03:22 -0700464 if (node) {
465 node->count++;
Johannes Weiner139e5612014-04-03 14:47:54 -0700466 BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK));
467 BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK));
Nick Piggin612d6c12006-06-23 02:03:22 -0700468 } else {
Nick Piggin612d6c12006-06-23 02:03:22 -0700469 BUG_ON(root_tag_get(root, 0));
470 BUG_ON(root_tag_get(root, 1));
471 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 return 0;
474}
475EXPORT_SYMBOL(radix_tree_insert);
476
Johannes Weiner139e5612014-04-03 14:47:54 -0700477/**
478 * __radix_tree_lookup - lookup an item in a radix tree
479 * @root: radix tree root
480 * @index: index key
481 * @nodep: returns node
482 * @slotp: returns slot
483 *
484 * Lookup and return the item at position @index in the radix
485 * tree @root.
486 *
487 * Until there is more than one item in the tree, no nodes are
488 * allocated and @root->rnode is used as a direct slot instead of
489 * pointing to a node, in which case *@nodep will be NULL.
Hans Reisera4331362005-11-07 00:59:29 -0800490 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700491void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
492 struct radix_tree_node **nodep, void ***slotp)
Hans Reisera4331362005-11-07 00:59:29 -0800493{
Johannes Weiner139e5612014-04-03 14:47:54 -0700494 struct radix_tree_node *node, *parent;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800495 unsigned int height, shift;
Johannes Weiner139e5612014-04-03 14:47:54 -0700496 void **slot;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800497
Paul E. McKenney2676a582010-02-22 17:04:54 -0800498 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800499 if (node == NULL)
500 return NULL;
501
Nick Pigginc0bc9872007-10-16 01:24:42 -0700502 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800503 if (index > 0)
504 return NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -0700505
506 if (nodep)
507 *nodep = NULL;
508 if (slotp)
509 *slotp = (void **)&root->rnode;
510 return node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800511 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800512 node = indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800513
Johannes Weiner449dd692014-04-03 14:47:56 -0700514 height = node->path & RADIX_TREE_HEIGHT_MASK;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800515 if (index > radix_tree_maxindex(height))
516 return NULL;
517
518 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
519
520 do {
Johannes Weiner139e5612014-04-03 14:47:54 -0700521 parent = node;
522 slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK);
Paul E. McKenney2676a582010-02-22 17:04:54 -0800523 node = rcu_dereference_raw(*slot);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800524 if (node == NULL)
525 return NULL;
526
527 shift -= RADIX_TREE_MAP_SHIFT;
528 height--;
529 } while (height > 0);
530
Johannes Weiner139e5612014-04-03 14:47:54 -0700531 if (nodep)
532 *nodep = parent;
533 if (slotp)
534 *slotp = slot;
535 return node;
Huang Shijieb72b71c2009-06-16 15:33:42 -0700536}
537
538/**
539 * radix_tree_lookup_slot - lookup a slot in a radix tree
540 * @root: radix tree root
541 * @index: index key
542 *
543 * Returns: the slot corresponding to the position @index in the
544 * radix tree @root. This is useful for update-if-exists operations.
545 *
546 * This function can be called under rcu_read_lock iff the slot is not
547 * modified by radix_tree_replace_slot, otherwise it must be called
548 * exclusive from other writers. Any dereference of the slot must be done
549 * using radix_tree_deref_slot.
550 */
551void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
552{
Johannes Weiner139e5612014-04-03 14:47:54 -0700553 void **slot;
554
555 if (!__radix_tree_lookup(root, index, NULL, &slot))
556 return NULL;
557 return slot;
Hans Reisera4331362005-11-07 00:59:29 -0800558}
559EXPORT_SYMBOL(radix_tree_lookup_slot);
560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561/**
562 * radix_tree_lookup - perform lookup operation on a radix tree
563 * @root: radix tree root
564 * @index: index key
565 *
566 * Lookup the item at the position @index in the radix tree @root.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800567 *
568 * This function can be called under rcu_read_lock, however the caller
569 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
570 * them safely). No RCU barriers are required to access or modify the
571 * returned item, however.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 */
573void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
574{
Johannes Weiner139e5612014-04-03 14:47:54 -0700575 return __radix_tree_lookup(root, index, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576}
577EXPORT_SYMBOL(radix_tree_lookup);
578
579/**
580 * radix_tree_tag_set - set a tag on a radix tree node
581 * @root: radix tree root
582 * @index: index key
583 * @tag: tag index
584 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800585 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
586 * corresponding to @index in the radix tree. From
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 * the root all the way down to the leaf node.
588 *
589 * Returns the address of the tagged item. Setting a tag on a not-present
590 * item is a bug.
591 */
592void *radix_tree_tag_set(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800593 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
595 unsigned int height, shift;
Christoph Lameter201b6262005-09-06 15:16:46 -0700596 struct radix_tree_node *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598 height = root->height;
Peter Zijlstra4c91c362006-06-23 02:03:25 -0700599 BUG_ON(index > radix_tree_maxindex(height));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Nick Piggin27d20fd2010-11-11 14:05:19 -0800601 slot = indirect_to_ptr(root->rnode);
Nick Piggin612d6c12006-06-23 02:03:22 -0700602 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604 while (height > 0) {
605 int offset;
606
607 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Nick Piggind5274262006-01-08 01:01:41 -0800608 if (!tag_get(slot, tag, offset))
609 tag_set(slot, tag, offset);
Christoph Lameter201b6262005-09-06 15:16:46 -0700610 slot = slot->slots[offset];
611 BUG_ON(slot == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 shift -= RADIX_TREE_MAP_SHIFT;
613 height--;
614 }
615
Nick Piggin612d6c12006-06-23 02:03:22 -0700616 /* set the root's tag bit */
617 if (slot && !root_tag_get(root, tag))
618 root_tag_set(root, tag);
619
Christoph Lameter201b6262005-09-06 15:16:46 -0700620 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
622EXPORT_SYMBOL(radix_tree_tag_set);
623
624/**
625 * radix_tree_tag_clear - clear a tag on a radix tree node
626 * @root: radix tree root
627 * @index: index key
628 * @tag: tag index
629 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800630 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
631 * corresponding to @index in the radix tree. If
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 * this causes the leaf node to have no tags set then clear the tag in the
633 * next-to-leaf node, etc.
634 *
635 * Returns the address of the tagged item on success, else NULL. ie:
636 * has the same return value and semantics as radix_tree_lookup().
637 */
638void *radix_tree_tag_clear(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800639 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800641 struct radix_tree_node *node = NULL;
Nick Piggin612d6c12006-06-23 02:03:22 -0700642 struct radix_tree_node *slot = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 unsigned int height, shift;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800644 int uninitialized_var(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 height = root->height;
647 if (index > radix_tree_maxindex(height))
648 goto out;
649
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800650 shift = height * RADIX_TREE_MAP_SHIFT;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800651 slot = indirect_to_ptr(root->rnode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800653 while (shift) {
Christoph Lameter201b6262005-09-06 15:16:46 -0700654 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 goto out;
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 shift -= RADIX_TREE_MAP_SHIFT;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800658 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
659 node = slot;
660 slot = slot->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 }
662
Nick Piggin612d6c12006-06-23 02:03:22 -0700663 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 goto out;
665
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800666 while (node) {
667 if (!tag_get(node, tag, offset))
Nick Piggind5274262006-01-08 01:01:41 -0800668 goto out;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800669 tag_clear(node, tag, offset);
670 if (any_tag_set(node, tag))
Nick Piggin6e954b92006-01-08 01:01:40 -0800671 goto out;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800672
673 index >>= RADIX_TREE_MAP_SHIFT;
674 offset = index & RADIX_TREE_MAP_MASK;
675 node = node->parent;
Nick Piggin612d6c12006-06-23 02:03:22 -0700676 }
677
678 /* clear the root's tag bit */
679 if (root_tag_get(root, tag))
680 root_tag_clear(root, tag);
681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682out:
Nick Piggin612d6c12006-06-23 02:03:22 -0700683 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685EXPORT_SYMBOL(radix_tree_tag_clear);
686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687/**
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700688 * radix_tree_tag_get - get a tag on a radix tree node
689 * @root: radix tree root
690 * @index: index key
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800691 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 *
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700693 * Return values:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 *
Nick Piggin612d6c12006-06-23 02:03:22 -0700695 * 0: tag not present or not set
696 * 1: tag set
David Howellsce826532010-04-06 22:36:20 +0100697 *
698 * Note that the return value of this function may not be relied on, even if
699 * the RCU lock is held, unless tag modification and node deletion are excluded
700 * from concurrency.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 */
702int radix_tree_tag_get(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800703 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
705 unsigned int height, shift;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800706 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Nick Piggin612d6c12006-06-23 02:03:22 -0700708 /* check the root's tag bit */
709 if (!root_tag_get(root, tag))
710 return 0;
711
Paul E. McKenney2676a582010-02-22 17:04:54 -0800712 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800713 if (node == NULL)
714 return 0;
715
Nick Pigginc0bc9872007-10-16 01:24:42 -0700716 if (!radix_tree_is_indirect_ptr(node))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800717 return (index == 0);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800718 node = indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800719
Johannes Weiner449dd692014-04-03 14:47:56 -0700720 height = node->path & RADIX_TREE_HEIGHT_MASK;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800721 if (index > radix_tree_maxindex(height))
722 return 0;
Nick Piggin612d6c12006-06-23 02:03:22 -0700723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 for ( ; ; ) {
727 int offset;
728
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800729 if (node == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 return 0;
731
732 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800733 if (!tag_get(node, tag, offset))
Hugh Dickins3fa36ac2011-10-31 17:07:02 -0700734 return 0;
David Howellsce826532010-04-06 22:36:20 +0100735 if (height == 1)
Hugh Dickins3fa36ac2011-10-31 17:07:02 -0700736 return 1;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800737 node = rcu_dereference_raw(node->slots[offset]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 shift -= RADIX_TREE_MAP_SHIFT;
739 height--;
740 }
741}
742EXPORT_SYMBOL(radix_tree_tag_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700744/**
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700745 * radix_tree_next_chunk - find next chunk of slots for iteration
746 *
747 * @root: radix tree root
748 * @iter: iterator state
749 * @flags: RADIX_TREE_ITER_* flags and tag index
750 * Returns: pointer to chunk first slot, or NULL if iteration is over
751 */
752void **radix_tree_next_chunk(struct radix_tree_root *root,
753 struct radix_tree_iter *iter, unsigned flags)
754{
755 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
756 struct radix_tree_node *rnode, *node;
Johannes Weiner449dd692014-04-03 14:47:56 -0700757 unsigned long index, offset, height;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700758
759 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
760 return NULL;
761
762 /*
763 * Catch next_index overflow after ~0UL. iter->index never overflows
764 * during iterating; it can be zero only at the beginning.
765 * And we cannot overflow iter->next_index in a single step,
766 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
Konstantin Khlebnikovfffaee32012-06-05 21:36:33 +0400767 *
768 * This condition also used by radix_tree_next_slot() to stop
769 * contiguous iterating, and forbid swithing to the next chunk.
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700770 */
771 index = iter->next_index;
772 if (!index && iter->index)
773 return NULL;
774
775 rnode = rcu_dereference_raw(root->rnode);
776 if (radix_tree_is_indirect_ptr(rnode)) {
777 rnode = indirect_to_ptr(rnode);
778 } else if (rnode && !index) {
779 /* Single-slot tree */
780 iter->index = 0;
781 iter->next_index = 1;
782 iter->tags = 1;
783 return (void **)&root->rnode;
784 } else
785 return NULL;
786
787restart:
Johannes Weiner449dd692014-04-03 14:47:56 -0700788 height = rnode->path & RADIX_TREE_HEIGHT_MASK;
789 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700790 offset = index >> shift;
791
792 /* Index outside of the tree */
793 if (offset >= RADIX_TREE_MAP_SIZE)
794 return NULL;
795
796 node = rnode;
797 while (1) {
798 if ((flags & RADIX_TREE_ITER_TAGGED) ?
799 !test_bit(offset, node->tags[tag]) :
800 !node->slots[offset]) {
801 /* Hole detected */
802 if (flags & RADIX_TREE_ITER_CONTIG)
803 return NULL;
804
805 if (flags & RADIX_TREE_ITER_TAGGED)
806 offset = radix_tree_find_next_bit(
807 node->tags[tag],
808 RADIX_TREE_MAP_SIZE,
809 offset + 1);
810 else
811 while (++offset < RADIX_TREE_MAP_SIZE) {
812 if (node->slots[offset])
813 break;
814 }
815 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
816 index += offset << shift;
817 /* Overflow after ~0UL */
818 if (!index)
819 return NULL;
820 if (offset == RADIX_TREE_MAP_SIZE)
821 goto restart;
822 }
823
824 /* This is leaf-node */
825 if (!shift)
826 break;
827
828 node = rcu_dereference_raw(node->slots[offset]);
829 if (node == NULL)
830 goto restart;
831 shift -= RADIX_TREE_MAP_SHIFT;
832 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
833 }
834
835 /* Update the iterator state */
836 iter->index = index;
837 iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1;
838
839 /* Construct iter->tags bit-mask from node->tags[tag] array */
840 if (flags & RADIX_TREE_ITER_TAGGED) {
841 unsigned tag_long, tag_bit;
842
843 tag_long = offset / BITS_PER_LONG;
844 tag_bit = offset % BITS_PER_LONG;
845 iter->tags = node->tags[tag][tag_long] >> tag_bit;
846 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
847 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
848 /* Pick tags from next element */
849 if (tag_bit)
850 iter->tags |= node->tags[tag][tag_long + 1] <<
851 (BITS_PER_LONG - tag_bit);
852 /* Clip chunk size, here only BITS_PER_LONG tags */
853 iter->next_index = index + BITS_PER_LONG;
854 }
855 }
856
857 return node->slots + offset;
858}
859EXPORT_SYMBOL(radix_tree_next_chunk);
860
861/**
Jan Karaebf8aa42010-08-09 17:19:11 -0700862 * radix_tree_range_tag_if_tagged - for each item in given range set given
863 * tag if item has another tag set
864 * @root: radix tree root
865 * @first_indexp: pointer to a starting index of a range to scan
866 * @last_index: last index of a range to scan
867 * @nr_to_tag: maximum number items to tag
868 * @iftag: tag index to test
869 * @settag: tag index to set if tested tag is set
870 *
871 * This function scans range of radix tree from first_index to last_index
872 * (inclusive). For each item in the range if iftag is set, the function sets
873 * also settag. The function stops either after tagging nr_to_tag items or
874 * after reaching last_index.
875 *
Dave Chinner144dcfc2010-08-23 10:33:53 +1000876 * The tags must be set from the leaf level only and propagated back up the
877 * path to the root. We must do this so that we resolve the full path before
878 * setting any tags on intermediate nodes. If we set tags as we descend, then
879 * we can get to the leaf node and find that the index that has the iftag
880 * set is outside the range we are scanning. This reults in dangling tags and
881 * can lead to problems with later tag operations (e.g. livelocks on lookups).
882 *
Jan Karaebf8aa42010-08-09 17:19:11 -0700883 * The function returns number of leaves where the tag was set and sets
884 * *first_indexp to the first unscanned index.
Jan Karad5ed3a42010-08-19 14:13:33 -0700885 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
886 * be prepared to handle that.
Jan Karaebf8aa42010-08-09 17:19:11 -0700887 */
888unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
889 unsigned long *first_indexp, unsigned long last_index,
890 unsigned long nr_to_tag,
891 unsigned int iftag, unsigned int settag)
892{
Dave Chinner144dcfc2010-08-23 10:33:53 +1000893 unsigned int height = root->height;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800894 struct radix_tree_node *node = NULL;
Dave Chinner144dcfc2010-08-23 10:33:53 +1000895 struct radix_tree_node *slot;
896 unsigned int shift;
897 unsigned long tagged = 0;
898 unsigned long index = *first_indexp;
Jan Karaebf8aa42010-08-09 17:19:11 -0700899
900 last_index = min(last_index, radix_tree_maxindex(height));
901 if (index > last_index)
902 return 0;
903 if (!nr_to_tag)
904 return 0;
905 if (!root_tag_get(root, iftag)) {
906 *first_indexp = last_index + 1;
907 return 0;
908 }
909 if (height == 0) {
910 *first_indexp = last_index + 1;
911 root_tag_set(root, settag);
912 return 1;
913 }
914
915 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800916 slot = indirect_to_ptr(root->rnode);
Jan Karaebf8aa42010-08-09 17:19:11 -0700917
918 for (;;) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800919 unsigned long upindex;
Jan Karaebf8aa42010-08-09 17:19:11 -0700920 int offset;
921
922 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
923 if (!slot->slots[offset])
924 goto next;
925 if (!tag_get(slot, iftag, offset))
926 goto next;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800927 if (shift) {
Dave Chinner144dcfc2010-08-23 10:33:53 +1000928 /* Go down one level */
Dave Chinner144dcfc2010-08-23 10:33:53 +1000929 shift -= RADIX_TREE_MAP_SHIFT;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800930 node = slot;
Dave Chinner144dcfc2010-08-23 10:33:53 +1000931 slot = slot->slots[offset];
932 continue;
Jan Karaebf8aa42010-08-09 17:19:11 -0700933 }
Dave Chinner144dcfc2010-08-23 10:33:53 +1000934
935 /* tag the leaf */
936 tagged++;
937 tag_set(slot, settag, offset);
938
939 /* walk back up the path tagging interior nodes */
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800940 upindex = index;
941 while (node) {
942 upindex >>= RADIX_TREE_MAP_SHIFT;
943 offset = upindex & RADIX_TREE_MAP_MASK;
944
Dave Chinner144dcfc2010-08-23 10:33:53 +1000945 /* stop if we find a node with the tag already set */
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800946 if (tag_get(node, settag, offset))
Dave Chinner144dcfc2010-08-23 10:33:53 +1000947 break;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800948 tag_set(node, settag, offset);
949 node = node->parent;
Dave Chinner144dcfc2010-08-23 10:33:53 +1000950 }
951
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800952 /*
953 * Small optimization: now clear that node pointer.
954 * Since all of this slot's ancestors now have the tag set
955 * from setting it above, we have no further need to walk
956 * back up the tree setting tags, until we update slot to
957 * point to another radix_tree_node.
958 */
959 node = NULL;
960
Jan Karaebf8aa42010-08-09 17:19:11 -0700961next:
962 /* Go to next item at level determined by 'shift' */
963 index = ((index >> shift) + 1) << shift;
Jan Karad5ed3a42010-08-19 14:13:33 -0700964 /* Overflow can happen when last_index is ~0UL... */
965 if (index > last_index || !index)
Jan Karaebf8aa42010-08-09 17:19:11 -0700966 break;
967 if (tagged >= nr_to_tag)
968 break;
969 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
970 /*
971 * We've fully scanned this node. Go up. Because
972 * last_index is guaranteed to be in the tree, what
973 * we do below cannot wander astray.
974 */
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800975 slot = slot->parent;
Jan Karaebf8aa42010-08-09 17:19:11 -0700976 shift += RADIX_TREE_MAP_SHIFT;
977 }
978 }
979 /*
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -0800980 * We need not to tag the root tag if there is no tag which is set with
981 * settag within the range from *first_indexp to last_index.
Jan Karaebf8aa42010-08-09 17:19:11 -0700982 */
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -0800983 if (tagged > 0)
984 root_tag_set(root, settag);
Jan Karaebf8aa42010-08-09 17:19:11 -0700985 *first_indexp = index;
986
987 return tagged;
988}
989EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991/**
992 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
993 * @root: radix tree root
994 * @results: where the results of the lookup are placed
995 * @first_index: start the lookup from this key
996 * @max_items: place up to this many items at *results
997 *
998 * Performs an index-ascending scan of the tree for present items. Places
999 * them at *@results and returns the number of items which were placed at
1000 * *@results.
1001 *
1002 * The implementation is naive.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001003 *
1004 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1005 * rcu_read_lock. In this case, rather than the returned results being
1006 * an atomic snapshot of the tree at a single point in time, the semantics
1007 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1008 * have been issued in individual locks, and results stored in 'results'.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 */
1010unsigned int
1011radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1012 unsigned long first_index, unsigned int max_items)
1013{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001014 struct radix_tree_iter iter;
1015 void **slot;
1016 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001018 if (unlikely(!max_items))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001019 return 0;
1020
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001021 radix_tree_for_each_slot(slot, root, &iter, first_index) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001022 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001023 if (!results[ret])
1024 continue;
Matthew Wilcox46437f92016-02-02 16:57:52 -08001025 if (radix_tree_is_indirect_ptr(results[ret])) {
1026 slot = radix_tree_iter_retry(&iter);
1027 continue;
1028 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001029 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return ret;
1034}
1035EXPORT_SYMBOL(radix_tree_gang_lookup);
1036
Nick Piggin47feff22008-07-25 19:45:29 -07001037/**
1038 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1039 * @root: radix tree root
1040 * @results: where the results of the lookup are placed
Hugh Dickins63286502011-08-03 16:21:18 -07001041 * @indices: where their indices should be placed (but usually NULL)
Nick Piggin47feff22008-07-25 19:45:29 -07001042 * @first_index: start the lookup from this key
1043 * @max_items: place up to this many items at *results
1044 *
1045 * Performs an index-ascending scan of the tree for present items. Places
1046 * their slots at *@results and returns the number of items which were
1047 * placed at *@results.
1048 *
1049 * The implementation is naive.
1050 *
1051 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1052 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1053 * protection, radix_tree_deref_slot may fail requiring a retry.
1054 */
1055unsigned int
Hugh Dickins63286502011-08-03 16:21:18 -07001056radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1057 void ***results, unsigned long *indices,
Nick Piggin47feff22008-07-25 19:45:29 -07001058 unsigned long first_index, unsigned int max_items)
1059{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001060 struct radix_tree_iter iter;
1061 void **slot;
1062 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001063
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001064 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001065 return 0;
1066
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001067 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1068 results[ret] = slot;
Hugh Dickins63286502011-08-03 16:21:18 -07001069 if (indices)
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001070 indices[ret] = iter.index;
1071 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001072 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001073 }
1074
1075 return ret;
1076}
1077EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079/**
1080 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1081 * based on a tag
1082 * @root: radix tree root
1083 * @results: where the results of the lookup are placed
1084 * @first_index: start the lookup from this key
1085 * @max_items: place up to this many items at *results
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001086 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 *
1088 * Performs an index-ascending scan of the tree for present items which
1089 * have the tag indexed by @tag set. Places the items at *@results and
1090 * returns the number of items which were placed at *@results.
1091 */
1092unsigned int
1093radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001094 unsigned long first_index, unsigned int max_items,
1095 unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001097 struct radix_tree_iter iter;
1098 void **slot;
1099 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001101 if (unlikely(!max_items))
Nick Piggin612d6c12006-06-23 02:03:22 -07001102 return 0;
1103
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001104 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001105 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001106 if (!results[ret])
1107 continue;
Matthew Wilcox46437f92016-02-02 16:57:52 -08001108 if (radix_tree_is_indirect_ptr(results[ret])) {
1109 slot = radix_tree_iter_retry(&iter);
1110 continue;
1111 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001112 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 return ret;
1117}
1118EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1119
1120/**
Nick Piggin47feff22008-07-25 19:45:29 -07001121 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1122 * radix tree based on a tag
1123 * @root: radix tree root
1124 * @results: where the results of the lookup are placed
1125 * @first_index: start the lookup from this key
1126 * @max_items: place up to this many items at *results
1127 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1128 *
1129 * Performs an index-ascending scan of the tree for present items which
1130 * have the tag indexed by @tag set. Places the slots at *@results and
1131 * returns the number of slots which were placed at *@results.
1132 */
1133unsigned int
1134radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1135 unsigned long first_index, unsigned int max_items,
1136 unsigned int tag)
1137{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001138 struct radix_tree_iter iter;
1139 void **slot;
1140 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001141
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001142 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001143 return 0;
1144
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001145 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1146 results[ret] = slot;
1147 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001148 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001149 }
1150
1151 return ret;
1152}
1153EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1154
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001155#if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1156#include <linux/sched.h> /* for cond_resched() */
1157
1158/*
1159 * This linear search is at present only useful to shmem_unuse_inode().
1160 */
1161static unsigned long __locate(struct radix_tree_node *slot, void *item,
1162 unsigned long index, unsigned long *found_index)
1163{
1164 unsigned int shift, height;
1165 unsigned long i;
1166
Johannes Weiner449dd692014-04-03 14:47:56 -07001167 height = slot->path & RADIX_TREE_HEIGHT_MASK;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001168 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1169
1170 for ( ; height > 1; height--) {
1171 i = (index >> shift) & RADIX_TREE_MAP_MASK;
1172 for (;;) {
1173 if (slot->slots[i] != NULL)
1174 break;
1175 index &= ~((1UL << shift) - 1);
1176 index += 1UL << shift;
1177 if (index == 0)
1178 goto out; /* 32-bit wraparound */
1179 i++;
1180 if (i == RADIX_TREE_MAP_SIZE)
1181 goto out;
1182 }
1183
1184 shift -= RADIX_TREE_MAP_SHIFT;
1185 slot = rcu_dereference_raw(slot->slots[i]);
1186 if (slot == NULL)
1187 goto out;
1188 }
1189
1190 /* Bottom level: check items */
1191 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
1192 if (slot->slots[i] == item) {
1193 *found_index = index + i;
1194 index = 0;
1195 goto out;
1196 }
1197 }
1198 index += RADIX_TREE_MAP_SIZE;
1199out:
1200 return index;
1201}
1202
1203/**
1204 * radix_tree_locate_item - search through radix tree for item
1205 * @root: radix tree root
1206 * @item: item to be found
1207 *
1208 * Returns index where item was found, or -1 if not found.
1209 * Caller must hold no lock (since this time-consuming function needs
1210 * to be preemptible), and must check afterwards if item is still there.
1211 */
1212unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1213{
1214 struct radix_tree_node *node;
1215 unsigned long max_index;
1216 unsigned long cur_index = 0;
1217 unsigned long found_index = -1;
1218
1219 do {
1220 rcu_read_lock();
1221 node = rcu_dereference_raw(root->rnode);
1222 if (!radix_tree_is_indirect_ptr(node)) {
1223 rcu_read_unlock();
1224 if (node == item)
1225 found_index = 0;
1226 break;
1227 }
1228
1229 node = indirect_to_ptr(node);
Johannes Weiner449dd692014-04-03 14:47:56 -07001230 max_index = radix_tree_maxindex(node->path &
1231 RADIX_TREE_HEIGHT_MASK);
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001232 if (cur_index > max_index) {
1233 rcu_read_unlock();
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001234 break;
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001235 }
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001236
1237 cur_index = __locate(node, item, cur_index, &found_index);
1238 rcu_read_unlock();
1239 cond_resched();
1240 } while (cur_index != 0 && cur_index <= max_index);
1241
1242 return found_index;
1243}
1244#else
1245unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1246{
1247 return -1;
1248}
1249#endif /* CONFIG_SHMEM && CONFIG_SWAP */
Nick Piggin47feff22008-07-25 19:45:29 -07001250
1251/**
Nick Piggina5f51c92006-01-08 01:01:41 -08001252 * radix_tree_shrink - shrink height of a radix tree to minimal
1253 * @root radix tree root
1254 */
1255static inline void radix_tree_shrink(struct radix_tree_root *root)
1256{
1257 /* try to shrink tree height */
Nick Pigginc0bc9872007-10-16 01:24:42 -07001258 while (root->height > 0) {
Nick Piggina5f51c92006-01-08 01:01:41 -08001259 struct radix_tree_node *to_free = root->rnode;
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001260 struct radix_tree_node *slot;
Nick Piggina5f51c92006-01-08 01:01:41 -08001261
Nick Pigginc0bc9872007-10-16 01:24:42 -07001262 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
Nick Piggin27d20fd2010-11-11 14:05:19 -08001263 to_free = indirect_to_ptr(to_free);
Nick Pigginc0bc9872007-10-16 01:24:42 -07001264
1265 /*
1266 * The candidate node has more than one child, or its child
1267 * is not at the leftmost slot, we cannot shrink.
1268 */
1269 if (to_free->count != 1)
1270 break;
1271 if (!to_free->slots[0])
1272 break;
1273
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001274 /*
1275 * We don't need rcu_assign_pointer(), since we are simply
Nick Piggin27d20fd2010-11-11 14:05:19 -08001276 * moving the node from one part of the tree to another: if it
1277 * was safe to dereference the old pointer to it
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001278 * (to_free->slots[0]), it will be safe to dereference the new
Nick Piggin27d20fd2010-11-11 14:05:19 -08001279 * one (root->rnode) as far as dependent read barriers go.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001280 */
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001281 slot = to_free->slots[0];
1282 if (root->height > 1) {
1283 slot->parent = NULL;
1284 slot = ptr_to_indirect(slot);
1285 }
1286 root->rnode = slot;
Nick Piggina5f51c92006-01-08 01:01:41 -08001287 root->height--;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001288
1289 /*
1290 * We have a dilemma here. The node's slot[0] must not be
1291 * NULLed in case there are concurrent lookups expecting to
1292 * find the item. However if this was a bottom-level node,
1293 * then it may be subject to the slot pointer being visible
1294 * to callers dereferencing it. If item corresponding to
1295 * slot[0] is subsequently deleted, these callers would expect
1296 * their slot to become empty sooner or later.
1297 *
1298 * For example, lockless pagecache will look up a slot, deref
1299 * the page pointer, and if the page is 0 refcount it means it
1300 * was concurrently deleted from pagecache so try the deref
1301 * again. Fortunately there is already a requirement for logic
1302 * to retry the entire slot lookup -- the indirect pointer
1303 * problem (replacing direct root node with an indirect pointer
1304 * also results in a stale slot). So tag the slot as indirect
1305 * to force callers to retry.
1306 */
1307 if (root->height == 0)
1308 *((unsigned long *)&to_free->slots[0]) |=
1309 RADIX_TREE_INDIRECT_PTR;
1310
Nick Piggina5f51c92006-01-08 01:01:41 -08001311 radix_tree_node_free(to_free);
1312 }
1313}
1314
1315/**
Johannes Weiner139e5612014-04-03 14:47:54 -07001316 * __radix_tree_delete_node - try to free node after clearing a slot
1317 * @root: radix tree root
Johannes Weiner139e5612014-04-03 14:47:54 -07001318 * @node: node containing @index
1319 *
1320 * After clearing the slot at @index in @node from radix tree
1321 * rooted at @root, call this function to attempt freeing the
1322 * node and shrinking the tree.
1323 *
1324 * Returns %true if @node was freed, %false otherwise.
1325 */
Johannes Weiner449dd692014-04-03 14:47:56 -07001326bool __radix_tree_delete_node(struct radix_tree_root *root,
Johannes Weiner139e5612014-04-03 14:47:54 -07001327 struct radix_tree_node *node)
1328{
1329 bool deleted = false;
1330
1331 do {
1332 struct radix_tree_node *parent;
1333
1334 if (node->count) {
1335 if (node == indirect_to_ptr(root->rnode)) {
1336 radix_tree_shrink(root);
1337 if (root->height == 0)
1338 deleted = true;
1339 }
1340 return deleted;
1341 }
1342
1343 parent = node->parent;
1344 if (parent) {
Johannes Weiner449dd692014-04-03 14:47:56 -07001345 unsigned int offset;
Johannes Weiner139e5612014-04-03 14:47:54 -07001346
Johannes Weiner449dd692014-04-03 14:47:56 -07001347 offset = node->path >> RADIX_TREE_HEIGHT_SHIFT;
1348 parent->slots[offset] = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001349 parent->count--;
1350 } else {
1351 root_tag_clear_all(root);
1352 root->height = 0;
1353 root->rnode = NULL;
1354 }
1355
1356 radix_tree_node_free(node);
1357 deleted = true;
1358
1359 node = parent;
1360 } while (node);
1361
1362 return deleted;
1363}
1364
1365/**
Johannes Weiner53c59f22014-04-03 14:47:39 -07001366 * radix_tree_delete_item - delete an item from a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 * @root: radix tree root
1368 * @index: index key
Johannes Weiner53c59f22014-04-03 14:47:39 -07001369 * @item: expected item
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001371 * Remove @item at @index from the radix tree rooted at @root.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001373 * Returns the address of the deleted item, or NULL if it was not present
1374 * or the entry at the given @index was not @item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 */
Johannes Weiner53c59f22014-04-03 14:47:39 -07001376void *radix_tree_delete_item(struct radix_tree_root *root,
1377 unsigned long index, void *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378{
Johannes Weiner139e5612014-04-03 14:47:54 -07001379 struct radix_tree_node *node;
1380 unsigned int offset;
1381 void **slot;
1382 void *entry;
Nick Piggind5274262006-01-08 01:01:41 -08001383 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Johannes Weiner139e5612014-04-03 14:47:54 -07001385 entry = __radix_tree_lookup(root, index, &node, &slot);
1386 if (!entry)
1387 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Johannes Weiner139e5612014-04-03 14:47:54 -07001389 if (item && entry != item)
1390 return NULL;
1391
1392 if (!node) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001393 root_tag_clear_all(root);
1394 root->rnode = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001395 return entry;
Nick Piggin612d6c12006-06-23 02:03:22 -07001396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
Johannes Weiner139e5612014-04-03 14:47:54 -07001398 offset = index & RADIX_TREE_MAP_MASK;
Johannes Weiner53c59f22014-04-03 14:47:39 -07001399
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 /*
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001401 * Clear all tags associated with the item to be deleted.
1402 * This way of doing it would be inefficient, but seldom is any set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001404 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001405 if (tag_get(node, tag, offset))
Nick Piggin612d6c12006-06-23 02:03:22 -07001406 radix_tree_tag_clear(root, index, tag);
Nick Piggind5274262006-01-08 01:01:41 -08001407 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Johannes Weiner139e5612014-04-03 14:47:54 -07001409 node->slots[offset] = NULL;
1410 node->count--;
Nick Piggina5f51c92006-01-08 01:01:41 -08001411
Johannes Weiner449dd692014-04-03 14:47:56 -07001412 __radix_tree_delete_node(root, node);
Christoph Lameter201b6262005-09-06 15:16:46 -07001413
Johannes Weiner139e5612014-04-03 14:47:54 -07001414 return entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
Johannes Weiner53c59f22014-04-03 14:47:39 -07001416EXPORT_SYMBOL(radix_tree_delete_item);
1417
1418/**
1419 * radix_tree_delete - delete an item from a radix tree
1420 * @root: radix tree root
1421 * @index: index key
1422 *
1423 * Remove the item at @index from the radix tree rooted at @root.
1424 *
1425 * Returns the address of the deleted item, or NULL if it was not present.
1426 */
1427void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1428{
1429 return radix_tree_delete_item(root, index, NULL);
1430}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431EXPORT_SYMBOL(radix_tree_delete);
1432
1433/**
1434 * radix_tree_tagged - test whether any items in the tree are tagged
1435 * @root: radix tree root
1436 * @tag: tag to test
1437 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001438int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439{
Nick Piggin612d6c12006-06-23 02:03:22 -07001440 return root_tag_get(root, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441}
1442EXPORT_SYMBOL(radix_tree_tagged);
1443
1444static void
Johannes Weiner449dd692014-04-03 14:47:56 -07001445radix_tree_node_ctor(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
Johannes Weiner449dd692014-04-03 14:47:56 -07001447 struct radix_tree_node *node = arg;
1448
1449 memset(node, 0, sizeof(*node));
1450 INIT_LIST_HEAD(&node->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451}
1452
1453static __init unsigned long __maxindex(unsigned int height)
1454{
Peter Lund430d2752007-10-16 23:29:35 -07001455 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1456 int shift = RADIX_TREE_INDEX_BITS - width;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Peter Lund430d2752007-10-16 23:29:35 -07001458 if (shift < 0)
1459 return ~0UL;
1460 if (shift >= BITS_PER_LONG)
1461 return 0UL;
1462 return ~0UL >> shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464
1465static __init void radix_tree_init_maxindex(void)
1466{
1467 unsigned int i;
1468
1469 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1470 height_to_maxindex[i] = __maxindex(i);
1471}
1472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473static int radix_tree_callback(struct notifier_block *nfb,
1474 unsigned long action,
1475 void *hcpu)
1476{
1477 int cpu = (long)hcpu;
1478 struct radix_tree_preload *rtp;
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -07001479 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
1481 /* Free per-cpu pool of perloaded nodes */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001482 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 rtp = &per_cpu(radix_tree_preloads, cpu);
1484 while (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -07001485 node = rtp->nodes;
1486 rtp->nodes = node->private_data;
1487 kmem_cache_free(radix_tree_node_cachep, node);
1488 rtp->nr--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 }
1490 }
1491 return NOTIFY_OK;
1492}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494void __init radix_tree_init(void)
1495{
1496 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1497 sizeof(struct radix_tree_node), 0,
Christoph Lameter488514d2008-04-28 02:12:05 -07001498 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1499 radix_tree_node_ctor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 radix_tree_init_maxindex();
1501 hotcpu_notifier(radix_tree_callback, 0);
1502}