blob: 9bb440f317c91a97ce0c0e678c8f5735bee8a9e1 [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 /*
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700195 * Even if the caller has preloaded, try to allocate from the
196 * cache first for the new node to get accounted.
197 */
198 ret = kmem_cache_alloc(radix_tree_node_cachep,
199 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
200 if (ret)
201 goto out;
202
203 /*
Nick Piggine2848a02008-02-04 22:29:10 -0800204 * Provided the caller has preloaded here, we will always
205 * succeed in getting a node here (and never reach
206 * kmem_cache_alloc)
207 */
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700208 rtp = this_cpu_ptr(&radix_tree_preloads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 if (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700210 ret = rtp->nodes;
211 rtp->nodes = ret->private_data;
212 ret->private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 rtp->nr--;
214 }
Catalin Marinasce80b062014-06-06 14:38:18 -0700215 /*
216 * Update the allocation stack trace as this is more useful
217 * for debugging.
218 */
219 kmemleak_update_trace(ret);
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700220 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 }
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700222 ret = kmem_cache_alloc(radix_tree_node_cachep,
223 gfp_mask | __GFP_ACCOUNT);
224out:
Nick Pigginc0bc9872007-10-16 01:24:42 -0700225 BUG_ON(radix_tree_is_indirect_ptr(ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 return ret;
227}
228
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800229static void radix_tree_node_rcu_free(struct rcu_head *head)
230{
231 struct radix_tree_node *node =
232 container_of(head, struct radix_tree_node, rcu_head);
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000233 int i;
Nick Piggin643b52b2008-06-12 15:21:52 -0700234
235 /*
236 * must only free zeroed nodes into the slab. radix_tree_shrink
237 * can leave us with a non-NULL entry in the first slot, so clear
238 * that here to make sure.
239 */
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000240 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
241 tag_clear(node, i, 0);
242
Nick Piggin643b52b2008-06-12 15:21:52 -0700243 node->slots[0] = NULL;
244 node->count = 0;
245
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800246 kmem_cache_free(radix_tree_node_cachep, node);
247}
248
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249static inline void
250radix_tree_node_free(struct radix_tree_node *node)
251{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800252 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
255/*
256 * Load up this CPU's radix_tree_node buffer with sufficient objects to
257 * ensure that the addition of a single element in the tree cannot fail. On
258 * success, return zero, with preemption disabled. On error, return -ENOMEM
259 * with preemption not disabled.
David Howellsb34df792009-11-19 18:11:14 +0000260 *
261 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800262 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
Jan Kara5e4c0d972013-09-11 14:26:05 -0700264static int __radix_tree_preload(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
266 struct radix_tree_preload *rtp;
267 struct radix_tree_node *node;
268 int ret = -ENOMEM;
269
270 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700271 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700272 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 preempt_enable();
Christoph Lameter488514d2008-04-28 02:12:05 -0700274 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (node == NULL)
276 goto out;
277 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700278 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700279 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
280 node->private_data = rtp->nodes;
281 rtp->nodes = node;
282 rtp->nr++;
283 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 kmem_cache_free(radix_tree_node_cachep, node);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 }
287 ret = 0;
288out:
289 return ret;
290}
Jan Kara5e4c0d972013-09-11 14:26:05 -0700291
292/*
293 * Load up this CPU's radix_tree_node buffer with sufficient objects to
294 * ensure that the addition of a single element in the tree cannot fail. On
295 * success, return zero, with preemption disabled. On error, return -ENOMEM
296 * with preemption not disabled.
297 *
298 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800299 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Jan Kara5e4c0d972013-09-11 14:26:05 -0700300 */
301int radix_tree_preload(gfp_t gfp_mask)
302{
303 /* Warn on non-sensical use... */
Mel Gormand0164ad2015-11-06 16:28:21 -0800304 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
Jan Kara5e4c0d972013-09-11 14:26:05 -0700305 return __radix_tree_preload(gfp_mask);
306}
David Chinnerd7f09232007-07-14 16:05:04 +1000307EXPORT_SYMBOL(radix_tree_preload);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Nick Piggin6e954b92006-01-08 01:01:40 -0800309/*
Jan Kara5e4c0d972013-09-11 14:26:05 -0700310 * The same as above function, except we don't guarantee preloading happens.
311 * We do it, if we decide it helps. On success, return zero with preemption
312 * disabled. On error, return -ENOMEM with preemption not disabled.
313 */
314int radix_tree_maybe_preload(gfp_t gfp_mask)
315{
Mel Gormand0164ad2015-11-06 16:28:21 -0800316 if (gfpflags_allow_blocking(gfp_mask))
Jan Kara5e4c0d972013-09-11 14:26:05 -0700317 return __radix_tree_preload(gfp_mask);
318 /* Preloading doesn't help anything with this gfp mask, skip it */
319 preempt_disable();
320 return 0;
321}
322EXPORT_SYMBOL(radix_tree_maybe_preload);
323
324/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 * Return the maximum key which can be store into a
326 * radix tree with height HEIGHT.
327 */
328static inline unsigned long radix_tree_maxindex(unsigned int height)
329{
330 return height_to_maxindex[height];
331}
332
333/*
334 * Extend a radix tree so it can store key @index.
335 */
336static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
337{
338 struct radix_tree_node *node;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800339 struct radix_tree_node *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 unsigned int height;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 int tag;
342
343 /* Figure out what the height should be. */
344 height = root->height + 1;
345 while (index > radix_tree_maxindex(height))
346 height++;
347
348 if (root->rnode == NULL) {
349 root->height = height;
350 goto out;
351 }
352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 do {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800354 unsigned int newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 if (!(node = radix_tree_node_alloc(root)))
356 return -ENOMEM;
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 /* Propagate the aggregated tag info into the new root */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800359 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -0700360 if (root_tag_get(root, tag))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 tag_set(node, tag, 0);
362 }
363
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800364 /* Increase the height. */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800365 newheight = root->height+1;
Johannes Weiner449dd692014-04-03 14:47:56 -0700366 BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK);
367 node->path = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 node->count = 1;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800369 node->parent = NULL;
370 slot = root->rnode;
Matthew Wilcox339e6352016-03-17 14:21:48 -0700371 if (radix_tree_is_indirect_ptr(slot) && newheight > 1) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800372 slot = indirect_to_ptr(slot);
373 slot->parent = node;
Matthew Wilcox339e6352016-03-17 14:21:48 -0700374 slot = ptr_to_indirect(slot);
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800375 }
376 node->slots[0] = slot;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800377 node = ptr_to_indirect(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800378 rcu_assign_pointer(root->rnode, node);
379 root->height = newheight;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 } while (height > root->height);
381out:
382 return 0;
383}
384
385/**
Johannes Weiner139e5612014-04-03 14:47:54 -0700386 * __radix_tree_create - create a slot in a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 * @root: radix tree root
388 * @index: index key
Johannes Weiner139e5612014-04-03 14:47:54 -0700389 * @nodep: returns node
390 * @slotp: returns slot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 *
Johannes Weiner139e5612014-04-03 14:47:54 -0700392 * Create, if necessary, and return the node and slot for an item
393 * at position @index in the radix tree @root.
394 *
395 * Until there is more than one item in the tree, no nodes are
396 * allocated and @root->rnode is used as a direct slot instead of
397 * pointing to a node, in which case *@nodep will be NULL.
398 *
399 * Returns -ENOMEM, or 0 for success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700401int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
402 struct radix_tree_node **nodep, void ***slotp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
Christoph Lameter201b6262005-09-06 15:16:46 -0700404 struct radix_tree_node *node = NULL, *slot;
Johannes Weiner139e5612014-04-03 14:47:54 -0700405 unsigned int height, shift, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 int error;
407
408 /* Make sure the tree is high enough. */
Nick Piggin612d6c12006-06-23 02:03:22 -0700409 if (index > radix_tree_maxindex(root->height)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 error = radix_tree_extend(root, index);
411 if (error)
412 return error;
413 }
414
Nick Piggin27d20fd2010-11-11 14:05:19 -0800415 slot = indirect_to_ptr(root->rnode);
Nick Pigginc0bc9872007-10-16 01:24:42 -0700416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 height = root->height;
Matthew Wilcox0070e282016-03-17 14:21:51 -0700418 shift = height * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 offset = 0; /* uninitialised var warning */
Matthew Wilcox0070e282016-03-17 14:21:51 -0700421 while (shift > 0) {
Christoph Lameter201b6262005-09-06 15:16:46 -0700422 if (slot == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 /* Have to add a child node. */
Christoph Lameter201b6262005-09-06 15:16:46 -0700424 if (!(slot = radix_tree_node_alloc(root)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 return -ENOMEM;
Johannes Weiner449dd692014-04-03 14:47:56 -0700426 slot->path = height;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800427 slot->parent = node;
Christoph Lameter201b6262005-09-06 15:16:46 -0700428 if (node) {
Matthew Wilcox339e6352016-03-17 14:21:48 -0700429 rcu_assign_pointer(node->slots[offset],
430 ptr_to_indirect(slot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 node->count++;
Johannes Weiner449dd692014-04-03 14:47:56 -0700432 slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT;
Christoph Lameter201b6262005-09-06 15:16:46 -0700433 } else
Matthew Wilcox339e6352016-03-17 14:21:48 -0700434 rcu_assign_pointer(root->rnode,
435 ptr_to_indirect(slot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 }
437
438 /* Go a level down */
Matthew Wilcox0070e282016-03-17 14:21:51 -0700439 shift -= RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Christoph Lameter201b6262005-09-06 15:16:46 -0700441 node = slot;
442 slot = node->slots[offset];
Matthew Wilcox339e6352016-03-17 14:21:48 -0700443 slot = indirect_to_ptr(slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 height--;
Nick Piggin612d6c12006-06-23 02:03:22 -0700445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Johannes Weiner139e5612014-04-03 14:47:54 -0700447 if (nodep)
448 *nodep = node;
449 if (slotp)
450 *slotp = node ? node->slots + offset : (void **)&root->rnode;
451 return 0;
452}
453
454/**
455 * radix_tree_insert - insert into a radix tree
456 * @root: radix tree root
457 * @index: index key
458 * @item: item to insert
459 *
460 * Insert an item into the radix tree at position @index.
461 */
462int radix_tree_insert(struct radix_tree_root *root,
463 unsigned long index, void *item)
464{
465 struct radix_tree_node *node;
466 void **slot;
467 int error;
468
469 BUG_ON(radix_tree_is_indirect_ptr(item));
470
471 error = __radix_tree_create(root, index, &node, &slot);
472 if (error)
473 return error;
474 if (*slot != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 return -EEXIST;
Johannes Weiner139e5612014-04-03 14:47:54 -0700476 rcu_assign_pointer(*slot, item);
Christoph Lameter201b6262005-09-06 15:16:46 -0700477
Nick Piggin612d6c12006-06-23 02:03:22 -0700478 if (node) {
479 node->count++;
Johannes Weiner139e5612014-04-03 14:47:54 -0700480 BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK));
481 BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK));
Nick Piggin612d6c12006-06-23 02:03:22 -0700482 } else {
Nick Piggin612d6c12006-06-23 02:03:22 -0700483 BUG_ON(root_tag_get(root, 0));
484 BUG_ON(root_tag_get(root, 1));
485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return 0;
488}
489EXPORT_SYMBOL(radix_tree_insert);
490
Johannes Weiner139e5612014-04-03 14:47:54 -0700491/**
492 * __radix_tree_lookup - lookup an item in a radix tree
493 * @root: radix tree root
494 * @index: index key
495 * @nodep: returns node
496 * @slotp: returns slot
497 *
498 * Lookup and return the item at position @index in the radix
499 * tree @root.
500 *
501 * Until there is more than one item in the tree, no nodes are
502 * allocated and @root->rnode is used as a direct slot instead of
503 * pointing to a node, in which case *@nodep will be NULL.
Hans Reisera4331362005-11-07 00:59:29 -0800504 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700505void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
506 struct radix_tree_node **nodep, void ***slotp)
Hans Reisera4331362005-11-07 00:59:29 -0800507{
Johannes Weiner139e5612014-04-03 14:47:54 -0700508 struct radix_tree_node *node, *parent;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800509 unsigned int height, shift;
Johannes Weiner139e5612014-04-03 14:47:54 -0700510 void **slot;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800511
Paul E. McKenney2676a582010-02-22 17:04:54 -0800512 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800513 if (node == NULL)
514 return NULL;
515
Nick Pigginc0bc9872007-10-16 01:24:42 -0700516 if (!radix_tree_is_indirect_ptr(node)) {
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800517 if (index > 0)
518 return NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -0700519
520 if (nodep)
521 *nodep = NULL;
522 if (slotp)
523 *slotp = (void **)&root->rnode;
524 return node;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800525 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800526 node = indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800527
Johannes Weiner449dd692014-04-03 14:47:56 -0700528 height = node->path & RADIX_TREE_HEIGHT_MASK;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800529 if (index > radix_tree_maxindex(height))
530 return NULL;
531
532 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
533
534 do {
Johannes Weiner139e5612014-04-03 14:47:54 -0700535 parent = node;
536 slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK);
Paul E. McKenney2676a582010-02-22 17:04:54 -0800537 node = rcu_dereference_raw(*slot);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800538 if (node == NULL)
539 return NULL;
Matthew Wilcox339e6352016-03-17 14:21:48 -0700540 node = indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800541
542 shift -= RADIX_TREE_MAP_SHIFT;
543 height--;
544 } while (height > 0);
545
Johannes Weiner139e5612014-04-03 14:47:54 -0700546 if (nodep)
547 *nodep = parent;
548 if (slotp)
549 *slotp = slot;
550 return node;
Huang Shijieb72b71c2009-06-16 15:33:42 -0700551}
552
553/**
554 * radix_tree_lookup_slot - lookup a slot in a radix tree
555 * @root: radix tree root
556 * @index: index key
557 *
558 * Returns: the slot corresponding to the position @index in the
559 * radix tree @root. This is useful for update-if-exists operations.
560 *
561 * This function can be called under rcu_read_lock iff the slot is not
562 * modified by radix_tree_replace_slot, otherwise it must be called
563 * exclusive from other writers. Any dereference of the slot must be done
564 * using radix_tree_deref_slot.
565 */
566void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
567{
Johannes Weiner139e5612014-04-03 14:47:54 -0700568 void **slot;
569
570 if (!__radix_tree_lookup(root, index, NULL, &slot))
571 return NULL;
572 return slot;
Hans Reisera4331362005-11-07 00:59:29 -0800573}
574EXPORT_SYMBOL(radix_tree_lookup_slot);
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576/**
577 * radix_tree_lookup - perform lookup operation on a radix tree
578 * @root: radix tree root
579 * @index: index key
580 *
581 * Lookup the item at the position @index in the radix tree @root.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800582 *
583 * This function can be called under rcu_read_lock, however the caller
584 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
585 * them safely). No RCU barriers are required to access or modify the
586 * returned item, however.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
588void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
589{
Johannes Weiner139e5612014-04-03 14:47:54 -0700590 return __radix_tree_lookup(root, index, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591}
592EXPORT_SYMBOL(radix_tree_lookup);
593
594/**
595 * radix_tree_tag_set - set a tag on a radix tree node
596 * @root: radix tree root
597 * @index: index key
598 * @tag: tag index
599 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800600 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
601 * corresponding to @index in the radix tree. From
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 * the root all the way down to the leaf node.
603 *
604 * Returns the address of the tagged item. Setting a tag on a not-present
605 * item is a bug.
606 */
607void *radix_tree_tag_set(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800608 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
610 unsigned int height, shift;
Christoph Lameter201b6262005-09-06 15:16:46 -0700611 struct radix_tree_node *slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
613 height = root->height;
Peter Zijlstra4c91c362006-06-23 02:03:25 -0700614 BUG_ON(index > radix_tree_maxindex(height));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Nick Piggin27d20fd2010-11-11 14:05:19 -0800616 slot = indirect_to_ptr(root->rnode);
Nick Piggin612d6c12006-06-23 02:03:22 -0700617 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 while (height > 0) {
620 int offset;
621
622 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Nick Piggind5274262006-01-08 01:01:41 -0800623 if (!tag_get(slot, tag, offset))
624 tag_set(slot, tag, offset);
Christoph Lameter201b6262005-09-06 15:16:46 -0700625 slot = slot->slots[offset];
626 BUG_ON(slot == NULL);
Matthew Wilcox339e6352016-03-17 14:21:48 -0700627 slot = indirect_to_ptr(slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 shift -= RADIX_TREE_MAP_SHIFT;
629 height--;
630 }
631
Nick Piggin612d6c12006-06-23 02:03:22 -0700632 /* set the root's tag bit */
633 if (slot && !root_tag_get(root, tag))
634 root_tag_set(root, tag);
635
Christoph Lameter201b6262005-09-06 15:16:46 -0700636 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638EXPORT_SYMBOL(radix_tree_tag_set);
639
640/**
641 * radix_tree_tag_clear - clear a tag on a radix tree node
642 * @root: radix tree root
643 * @index: index key
644 * @tag: tag index
645 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800646 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
647 * corresponding to @index in the radix tree. If
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 * this causes the leaf node to have no tags set then clear the tag in the
649 * next-to-leaf node, etc.
650 *
651 * Returns the address of the tagged item on success, else NULL. ie:
652 * has the same return value and semantics as radix_tree_lookup().
653 */
654void *radix_tree_tag_clear(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800655 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800657 struct radix_tree_node *node = NULL;
Nick Piggin612d6c12006-06-23 02:03:22 -0700658 struct radix_tree_node *slot = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 unsigned int height, shift;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800660 int uninitialized_var(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
662 height = root->height;
663 if (index > radix_tree_maxindex(height))
664 goto out;
665
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800666 shift = height * RADIX_TREE_MAP_SHIFT;
Matthew Wilcox339e6352016-03-17 14:21:48 -0700667 slot = root->rnode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800669 while (shift) {
Christoph Lameter201b6262005-09-06 15:16:46 -0700670 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 goto out;
Matthew Wilcox339e6352016-03-17 14:21:48 -0700672 slot = indirect_to_ptr(slot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 shift -= RADIX_TREE_MAP_SHIFT;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800675 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
676 node = slot;
677 slot = slot->slots[offset];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679
Nick Piggin612d6c12006-06-23 02:03:22 -0700680 if (slot == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 goto out;
682
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800683 while (node) {
684 if (!tag_get(node, tag, offset))
Nick Piggind5274262006-01-08 01:01:41 -0800685 goto out;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800686 tag_clear(node, tag, offset);
687 if (any_tag_set(node, tag))
Nick Piggin6e954b92006-01-08 01:01:40 -0800688 goto out;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800689
690 index >>= RADIX_TREE_MAP_SHIFT;
691 offset = index & RADIX_TREE_MAP_MASK;
692 node = node->parent;
Nick Piggin612d6c12006-06-23 02:03:22 -0700693 }
694
695 /* clear the root's tag bit */
696 if (root_tag_get(root, tag))
697 root_tag_clear(root, tag);
698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699out:
Nick Piggin612d6c12006-06-23 02:03:22 -0700700 return slot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702EXPORT_SYMBOL(radix_tree_tag_clear);
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704/**
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700705 * radix_tree_tag_get - get a tag on a radix tree node
706 * @root: radix tree root
707 * @index: index key
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800708 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 *
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700710 * Return values:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 *
Nick Piggin612d6c12006-06-23 02:03:22 -0700712 * 0: tag not present or not set
713 * 1: tag set
David Howellsce826532010-04-06 22:36:20 +0100714 *
715 * Note that the return value of this function may not be relied on, even if
716 * the RCU lock is held, unless tag modification and node deletion are excluded
717 * from concurrency.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 */
719int radix_tree_tag_get(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800720 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
722 unsigned int height, shift;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800723 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
Nick Piggin612d6c12006-06-23 02:03:22 -0700725 /* check the root's tag bit */
726 if (!root_tag_get(root, tag))
727 return 0;
728
Paul E. McKenney2676a582010-02-22 17:04:54 -0800729 node = rcu_dereference_raw(root->rnode);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800730 if (node == NULL)
731 return 0;
732
Nick Pigginc0bc9872007-10-16 01:24:42 -0700733 if (!radix_tree_is_indirect_ptr(node))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800734 return (index == 0);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800735 node = indirect_to_ptr(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800736
Johannes Weiner449dd692014-04-03 14:47:56 -0700737 height = node->path & RADIX_TREE_HEIGHT_MASK;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800738 if (index > radix_tree_maxindex(height))
739 return 0;
Nick Piggin612d6c12006-06-23 02:03:22 -0700740
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
743 for ( ; ; ) {
744 int offset;
745
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800746 if (node == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 return 0;
Matthew Wilcox339e6352016-03-17 14:21:48 -0700748 node = indirect_to_ptr(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800751 if (!tag_get(node, tag, offset))
Hugh Dickins3fa36ac2011-10-31 17:07:02 -0700752 return 0;
David Howellsce826532010-04-06 22:36:20 +0100753 if (height == 1)
Hugh Dickins3fa36ac2011-10-31 17:07:02 -0700754 return 1;
Paul E. McKenney2676a582010-02-22 17:04:54 -0800755 node = rcu_dereference_raw(node->slots[offset]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 shift -= RADIX_TREE_MAP_SHIFT;
757 height--;
758 }
759}
760EXPORT_SYMBOL(radix_tree_tag_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700762/**
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700763 * radix_tree_next_chunk - find next chunk of slots for iteration
764 *
765 * @root: radix tree root
766 * @iter: iterator state
767 * @flags: RADIX_TREE_ITER_* flags and tag index
768 * Returns: pointer to chunk first slot, or NULL if iteration is over
769 */
770void **radix_tree_next_chunk(struct radix_tree_root *root,
771 struct radix_tree_iter *iter, unsigned flags)
772{
773 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
774 struct radix_tree_node *rnode, *node;
Johannes Weiner449dd692014-04-03 14:47:56 -0700775 unsigned long index, offset, height;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700776
777 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
778 return NULL;
779
780 /*
781 * Catch next_index overflow after ~0UL. iter->index never overflows
782 * during iterating; it can be zero only at the beginning.
783 * And we cannot overflow iter->next_index in a single step,
784 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
Konstantin Khlebnikovfffaee32012-06-05 21:36:33 +0400785 *
786 * This condition also used by radix_tree_next_slot() to stop
787 * contiguous iterating, and forbid swithing to the next chunk.
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700788 */
789 index = iter->next_index;
790 if (!index && iter->index)
791 return NULL;
792
793 rnode = rcu_dereference_raw(root->rnode);
794 if (radix_tree_is_indirect_ptr(rnode)) {
795 rnode = indirect_to_ptr(rnode);
796 } else if (rnode && !index) {
797 /* Single-slot tree */
798 iter->index = 0;
799 iter->next_index = 1;
800 iter->tags = 1;
801 return (void **)&root->rnode;
802 } else
803 return NULL;
804
805restart:
Johannes Weiner449dd692014-04-03 14:47:56 -0700806 height = rnode->path & RADIX_TREE_HEIGHT_MASK;
807 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700808 offset = index >> shift;
809
810 /* Index outside of the tree */
811 if (offset >= RADIX_TREE_MAP_SIZE)
812 return NULL;
813
814 node = rnode;
815 while (1) {
816 if ((flags & RADIX_TREE_ITER_TAGGED) ?
817 !test_bit(offset, node->tags[tag]) :
818 !node->slots[offset]) {
819 /* Hole detected */
820 if (flags & RADIX_TREE_ITER_CONTIG)
821 return NULL;
822
823 if (flags & RADIX_TREE_ITER_TAGGED)
824 offset = radix_tree_find_next_bit(
825 node->tags[tag],
826 RADIX_TREE_MAP_SIZE,
827 offset + 1);
828 else
829 while (++offset < RADIX_TREE_MAP_SIZE) {
830 if (node->slots[offset])
831 break;
832 }
833 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
834 index += offset << shift;
835 /* Overflow after ~0UL */
836 if (!index)
837 return NULL;
838 if (offset == RADIX_TREE_MAP_SIZE)
839 goto restart;
840 }
841
842 /* This is leaf-node */
843 if (!shift)
844 break;
845
846 node = rcu_dereference_raw(node->slots[offset]);
847 if (node == NULL)
848 goto restart;
Matthew Wilcox339e6352016-03-17 14:21:48 -0700849 node = indirect_to_ptr(node);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700850 shift -= RADIX_TREE_MAP_SHIFT;
851 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
852 }
853
854 /* Update the iterator state */
855 iter->index = index;
856 iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1;
857
858 /* Construct iter->tags bit-mask from node->tags[tag] array */
859 if (flags & RADIX_TREE_ITER_TAGGED) {
860 unsigned tag_long, tag_bit;
861
862 tag_long = offset / BITS_PER_LONG;
863 tag_bit = offset % BITS_PER_LONG;
864 iter->tags = node->tags[tag][tag_long] >> tag_bit;
865 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
866 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
867 /* Pick tags from next element */
868 if (tag_bit)
869 iter->tags |= node->tags[tag][tag_long + 1] <<
870 (BITS_PER_LONG - tag_bit);
871 /* Clip chunk size, here only BITS_PER_LONG tags */
872 iter->next_index = index + BITS_PER_LONG;
873 }
874 }
875
876 return node->slots + offset;
877}
878EXPORT_SYMBOL(radix_tree_next_chunk);
879
880/**
Jan Karaebf8aa42010-08-09 17:19:11 -0700881 * radix_tree_range_tag_if_tagged - for each item in given range set given
882 * tag if item has another tag set
883 * @root: radix tree root
884 * @first_indexp: pointer to a starting index of a range to scan
885 * @last_index: last index of a range to scan
886 * @nr_to_tag: maximum number items to tag
887 * @iftag: tag index to test
888 * @settag: tag index to set if tested tag is set
889 *
890 * This function scans range of radix tree from first_index to last_index
891 * (inclusive). For each item in the range if iftag is set, the function sets
892 * also settag. The function stops either after tagging nr_to_tag items or
893 * after reaching last_index.
894 *
Dave Chinner144dcfc2010-08-23 10:33:53 +1000895 * The tags must be set from the leaf level only and propagated back up the
896 * path to the root. We must do this so that we resolve the full path before
897 * setting any tags on intermediate nodes. If we set tags as we descend, then
898 * we can get to the leaf node and find that the index that has the iftag
899 * set is outside the range we are scanning. This reults in dangling tags and
900 * can lead to problems with later tag operations (e.g. livelocks on lookups).
901 *
Jan Karaebf8aa42010-08-09 17:19:11 -0700902 * The function returns number of leaves where the tag was set and sets
903 * *first_indexp to the first unscanned index.
Jan Karad5ed3a42010-08-19 14:13:33 -0700904 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
905 * be prepared to handle that.
Jan Karaebf8aa42010-08-09 17:19:11 -0700906 */
907unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
908 unsigned long *first_indexp, unsigned long last_index,
909 unsigned long nr_to_tag,
910 unsigned int iftag, unsigned int settag)
911{
Dave Chinner144dcfc2010-08-23 10:33:53 +1000912 unsigned int height = root->height;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800913 struct radix_tree_node *node = NULL;
Dave Chinner144dcfc2010-08-23 10:33:53 +1000914 struct radix_tree_node *slot;
915 unsigned int shift;
916 unsigned long tagged = 0;
917 unsigned long index = *first_indexp;
Jan Karaebf8aa42010-08-09 17:19:11 -0700918
919 last_index = min(last_index, radix_tree_maxindex(height));
920 if (index > last_index)
921 return 0;
922 if (!nr_to_tag)
923 return 0;
924 if (!root_tag_get(root, iftag)) {
925 *first_indexp = last_index + 1;
926 return 0;
927 }
928 if (height == 0) {
929 *first_indexp = last_index + 1;
930 root_tag_set(root, settag);
931 return 1;
932 }
933
934 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800935 slot = indirect_to_ptr(root->rnode);
Jan Karaebf8aa42010-08-09 17:19:11 -0700936
937 for (;;) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800938 unsigned long upindex;
Jan Karaebf8aa42010-08-09 17:19:11 -0700939 int offset;
940
941 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
942 if (!slot->slots[offset])
943 goto next;
944 if (!tag_get(slot, iftag, offset))
945 goto next;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800946 if (shift) {
Dave Chinner144dcfc2010-08-23 10:33:53 +1000947 /* Go down one level */
Dave Chinner144dcfc2010-08-23 10:33:53 +1000948 shift -= RADIX_TREE_MAP_SHIFT;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800949 node = slot;
Dave Chinner144dcfc2010-08-23 10:33:53 +1000950 slot = slot->slots[offset];
Matthew Wilcox339e6352016-03-17 14:21:48 -0700951 slot = indirect_to_ptr(slot);
Dave Chinner144dcfc2010-08-23 10:33:53 +1000952 continue;
Jan Karaebf8aa42010-08-09 17:19:11 -0700953 }
Dave Chinner144dcfc2010-08-23 10:33:53 +1000954
955 /* tag the leaf */
956 tagged++;
957 tag_set(slot, settag, offset);
958
959 /* walk back up the path tagging interior nodes */
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800960 upindex = index;
961 while (node) {
962 upindex >>= RADIX_TREE_MAP_SHIFT;
963 offset = upindex & RADIX_TREE_MAP_MASK;
964
Dave Chinner144dcfc2010-08-23 10:33:53 +1000965 /* stop if we find a node with the tag already set */
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800966 if (tag_get(node, settag, offset))
Dave Chinner144dcfc2010-08-23 10:33:53 +1000967 break;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800968 tag_set(node, settag, offset);
969 node = node->parent;
Dave Chinner144dcfc2010-08-23 10:33:53 +1000970 }
971
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800972 /*
973 * Small optimization: now clear that node pointer.
974 * Since all of this slot's ancestors now have the tag set
975 * from setting it above, we have no further need to walk
976 * back up the tree setting tags, until we update slot to
977 * point to another radix_tree_node.
978 */
979 node = NULL;
980
Jan Karaebf8aa42010-08-09 17:19:11 -0700981next:
982 /* Go to next item at level determined by 'shift' */
983 index = ((index >> shift) + 1) << shift;
Jan Karad5ed3a42010-08-19 14:13:33 -0700984 /* Overflow can happen when last_index is ~0UL... */
985 if (index > last_index || !index)
Jan Karaebf8aa42010-08-09 17:19:11 -0700986 break;
987 if (tagged >= nr_to_tag)
988 break;
989 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
990 /*
991 * We've fully scanned this node. Go up. Because
992 * last_index is guaranteed to be in the tree, what
993 * we do below cannot wander astray.
994 */
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800995 slot = slot->parent;
Jan Karaebf8aa42010-08-09 17:19:11 -0700996 shift += RADIX_TREE_MAP_SHIFT;
997 }
998 }
999 /*
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -08001000 * We need not to tag the root tag if there is no tag which is set with
1001 * settag within the range from *first_indexp to last_index.
Jan Karaebf8aa42010-08-09 17:19:11 -07001002 */
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -08001003 if (tagged > 0)
1004 root_tag_set(root, settag);
Jan Karaebf8aa42010-08-09 17:19:11 -07001005 *first_indexp = index;
1006
1007 return tagged;
1008}
1009EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011/**
1012 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1013 * @root: radix tree root
1014 * @results: where the results of the lookup are placed
1015 * @first_index: start the lookup from this key
1016 * @max_items: place up to this many items at *results
1017 *
1018 * Performs an index-ascending scan of the tree for present items. Places
1019 * them at *@results and returns the number of items which were placed at
1020 * *@results.
1021 *
1022 * The implementation is naive.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001023 *
1024 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1025 * rcu_read_lock. In this case, rather than the returned results being
1026 * an atomic snapshot of the tree at a single point in time, the semantics
1027 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
1028 * have been issued in individual locks, and results stored in 'results'.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 */
1030unsigned int
1031radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1032 unsigned long first_index, unsigned int max_items)
1033{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001034 struct radix_tree_iter iter;
1035 void **slot;
1036 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001038 if (unlikely(!max_items))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001039 return 0;
1040
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001041 radix_tree_for_each_slot(slot, root, &iter, first_index) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001042 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001043 if (!results[ret])
1044 continue;
Matthew Wilcox46437f92016-02-02 16:57:52 -08001045 if (radix_tree_is_indirect_ptr(results[ret])) {
1046 slot = radix_tree_iter_retry(&iter);
1047 continue;
1048 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001049 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 return ret;
1054}
1055EXPORT_SYMBOL(radix_tree_gang_lookup);
1056
Nick Piggin47feff22008-07-25 19:45:29 -07001057/**
1058 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1059 * @root: radix tree root
1060 * @results: where the results of the lookup are placed
Hugh Dickins63286502011-08-03 16:21:18 -07001061 * @indices: where their indices should be placed (but usually NULL)
Nick Piggin47feff22008-07-25 19:45:29 -07001062 * @first_index: start the lookup from this key
1063 * @max_items: place up to this many items at *results
1064 *
1065 * Performs an index-ascending scan of the tree for present items. Places
1066 * their slots at *@results and returns the number of items which were
1067 * placed at *@results.
1068 *
1069 * The implementation is naive.
1070 *
1071 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1072 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1073 * protection, radix_tree_deref_slot may fail requiring a retry.
1074 */
1075unsigned int
Hugh Dickins63286502011-08-03 16:21:18 -07001076radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1077 void ***results, unsigned long *indices,
Nick Piggin47feff22008-07-25 19:45:29 -07001078 unsigned long first_index, unsigned int max_items)
1079{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001080 struct radix_tree_iter iter;
1081 void **slot;
1082 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001083
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001084 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001085 return 0;
1086
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001087 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1088 results[ret] = slot;
Hugh Dickins63286502011-08-03 16:21:18 -07001089 if (indices)
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001090 indices[ret] = iter.index;
1091 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001092 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001093 }
1094
1095 return ret;
1096}
1097EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1098
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099/**
1100 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1101 * based on a tag
1102 * @root: radix tree root
1103 * @results: where the results of the lookup are placed
1104 * @first_index: start the lookup from this key
1105 * @max_items: place up to this many items at *results
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001106 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 *
1108 * Performs an index-ascending scan of the tree for present items which
1109 * have the tag indexed by @tag set. Places the items at *@results and
1110 * returns the number of items which were placed at *@results.
1111 */
1112unsigned int
1113radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001114 unsigned long first_index, unsigned int max_items,
1115 unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001117 struct radix_tree_iter iter;
1118 void **slot;
1119 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001121 if (unlikely(!max_items))
Nick Piggin612d6c12006-06-23 02:03:22 -07001122 return 0;
1123
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001124 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001125 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001126 if (!results[ret])
1127 continue;
Matthew Wilcox46437f92016-02-02 16:57:52 -08001128 if (radix_tree_is_indirect_ptr(results[ret])) {
1129 slot = radix_tree_iter_retry(&iter);
1130 continue;
1131 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001132 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 return ret;
1137}
1138EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1139
1140/**
Nick Piggin47feff22008-07-25 19:45:29 -07001141 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1142 * radix tree based on a tag
1143 * @root: radix tree root
1144 * @results: where the results of the lookup are placed
1145 * @first_index: start the lookup from this key
1146 * @max_items: place up to this many items at *results
1147 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1148 *
1149 * Performs an index-ascending scan of the tree for present items which
1150 * have the tag indexed by @tag set. Places the slots at *@results and
1151 * returns the number of slots which were placed at *@results.
1152 */
1153unsigned int
1154radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1155 unsigned long first_index, unsigned int max_items,
1156 unsigned int tag)
1157{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001158 struct radix_tree_iter iter;
1159 void **slot;
1160 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001161
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001162 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001163 return 0;
1164
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001165 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1166 results[ret] = slot;
1167 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001168 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001169 }
1170
1171 return ret;
1172}
1173EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1174
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001175#if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1176#include <linux/sched.h> /* for cond_resched() */
1177
1178/*
1179 * This linear search is at present only useful to shmem_unuse_inode().
1180 */
1181static unsigned long __locate(struct radix_tree_node *slot, void *item,
1182 unsigned long index, unsigned long *found_index)
1183{
1184 unsigned int shift, height;
1185 unsigned long i;
1186
Johannes Weiner449dd692014-04-03 14:47:56 -07001187 height = slot->path & RADIX_TREE_HEIGHT_MASK;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001188 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1189
1190 for ( ; height > 1; height--) {
1191 i = (index >> shift) & RADIX_TREE_MAP_MASK;
1192 for (;;) {
1193 if (slot->slots[i] != NULL)
1194 break;
1195 index &= ~((1UL << shift) - 1);
1196 index += 1UL << shift;
1197 if (index == 0)
1198 goto out; /* 32-bit wraparound */
1199 i++;
1200 if (i == RADIX_TREE_MAP_SIZE)
1201 goto out;
1202 }
1203
1204 shift -= RADIX_TREE_MAP_SHIFT;
1205 slot = rcu_dereference_raw(slot->slots[i]);
1206 if (slot == NULL)
1207 goto out;
Matthew Wilcox339e6352016-03-17 14:21:48 -07001208 slot = indirect_to_ptr(slot);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001209 }
1210
1211 /* Bottom level: check items */
1212 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
1213 if (slot->slots[i] == item) {
1214 *found_index = index + i;
1215 index = 0;
1216 goto out;
1217 }
1218 }
1219 index += RADIX_TREE_MAP_SIZE;
1220out:
1221 return index;
1222}
1223
1224/**
1225 * radix_tree_locate_item - search through radix tree for item
1226 * @root: radix tree root
1227 * @item: item to be found
1228 *
1229 * Returns index where item was found, or -1 if not found.
1230 * Caller must hold no lock (since this time-consuming function needs
1231 * to be preemptible), and must check afterwards if item is still there.
1232 */
1233unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1234{
1235 struct radix_tree_node *node;
1236 unsigned long max_index;
1237 unsigned long cur_index = 0;
1238 unsigned long found_index = -1;
1239
1240 do {
1241 rcu_read_lock();
1242 node = rcu_dereference_raw(root->rnode);
1243 if (!radix_tree_is_indirect_ptr(node)) {
1244 rcu_read_unlock();
1245 if (node == item)
1246 found_index = 0;
1247 break;
1248 }
1249
1250 node = indirect_to_ptr(node);
Johannes Weiner449dd692014-04-03 14:47:56 -07001251 max_index = radix_tree_maxindex(node->path &
1252 RADIX_TREE_HEIGHT_MASK);
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001253 if (cur_index > max_index) {
1254 rcu_read_unlock();
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001255 break;
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001256 }
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001257
1258 cur_index = __locate(node, item, cur_index, &found_index);
1259 rcu_read_unlock();
1260 cond_resched();
1261 } while (cur_index != 0 && cur_index <= max_index);
1262
1263 return found_index;
1264}
1265#else
1266unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1267{
1268 return -1;
1269}
1270#endif /* CONFIG_SHMEM && CONFIG_SWAP */
Nick Piggin47feff22008-07-25 19:45:29 -07001271
1272/**
Nick Piggina5f51c92006-01-08 01:01:41 -08001273 * radix_tree_shrink - shrink height of a radix tree to minimal
1274 * @root radix tree root
1275 */
1276static inline void radix_tree_shrink(struct radix_tree_root *root)
1277{
1278 /* try to shrink tree height */
Nick Pigginc0bc9872007-10-16 01:24:42 -07001279 while (root->height > 0) {
Nick Piggina5f51c92006-01-08 01:01:41 -08001280 struct radix_tree_node *to_free = root->rnode;
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001281 struct radix_tree_node *slot;
Nick Piggina5f51c92006-01-08 01:01:41 -08001282
Nick Pigginc0bc9872007-10-16 01:24:42 -07001283 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
Nick Piggin27d20fd2010-11-11 14:05:19 -08001284 to_free = indirect_to_ptr(to_free);
Nick Pigginc0bc9872007-10-16 01:24:42 -07001285
1286 /*
1287 * The candidate node has more than one child, or its child
1288 * is not at the leftmost slot, we cannot shrink.
1289 */
1290 if (to_free->count != 1)
1291 break;
Matthew Wilcox339e6352016-03-17 14:21:48 -07001292 slot = to_free->slots[0];
1293 if (!slot)
Nick Pigginc0bc9872007-10-16 01:24:42 -07001294 break;
1295
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001296 /*
1297 * We don't need rcu_assign_pointer(), since we are simply
Nick Piggin27d20fd2010-11-11 14:05:19 -08001298 * moving the node from one part of the tree to another: if it
1299 * was safe to dereference the old pointer to it
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001300 * (to_free->slots[0]), it will be safe to dereference the new
Nick Piggin27d20fd2010-11-11 14:05:19 -08001301 * one (root->rnode) as far as dependent read barriers go.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001302 */
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001303 if (root->height > 1) {
Matthew Wilcox339e6352016-03-17 14:21:48 -07001304 slot = indirect_to_ptr(slot);
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001305 slot->parent = NULL;
1306 slot = ptr_to_indirect(slot);
1307 }
1308 root->rnode = slot;
Nick Piggina5f51c92006-01-08 01:01:41 -08001309 root->height--;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001310
1311 /*
1312 * We have a dilemma here. The node's slot[0] must not be
1313 * NULLed in case there are concurrent lookups expecting to
1314 * find the item. However if this was a bottom-level node,
1315 * then it may be subject to the slot pointer being visible
1316 * to callers dereferencing it. If item corresponding to
1317 * slot[0] is subsequently deleted, these callers would expect
1318 * their slot to become empty sooner or later.
1319 *
1320 * For example, lockless pagecache will look up a slot, deref
1321 * the page pointer, and if the page is 0 refcount it means it
1322 * was concurrently deleted from pagecache so try the deref
1323 * again. Fortunately there is already a requirement for logic
1324 * to retry the entire slot lookup -- the indirect pointer
1325 * problem (replacing direct root node with an indirect pointer
1326 * also results in a stale slot). So tag the slot as indirect
1327 * to force callers to retry.
1328 */
1329 if (root->height == 0)
1330 *((unsigned long *)&to_free->slots[0]) |=
1331 RADIX_TREE_INDIRECT_PTR;
1332
Nick Piggina5f51c92006-01-08 01:01:41 -08001333 radix_tree_node_free(to_free);
1334 }
1335}
1336
1337/**
Johannes Weiner139e5612014-04-03 14:47:54 -07001338 * __radix_tree_delete_node - try to free node after clearing a slot
1339 * @root: radix tree root
Johannes Weiner139e5612014-04-03 14:47:54 -07001340 * @node: node containing @index
1341 *
1342 * After clearing the slot at @index in @node from radix tree
1343 * rooted at @root, call this function to attempt freeing the
1344 * node and shrinking the tree.
1345 *
1346 * Returns %true if @node was freed, %false otherwise.
1347 */
Johannes Weiner449dd692014-04-03 14:47:56 -07001348bool __radix_tree_delete_node(struct radix_tree_root *root,
Johannes Weiner139e5612014-04-03 14:47:54 -07001349 struct radix_tree_node *node)
1350{
1351 bool deleted = false;
1352
1353 do {
1354 struct radix_tree_node *parent;
1355
1356 if (node->count) {
1357 if (node == indirect_to_ptr(root->rnode)) {
1358 radix_tree_shrink(root);
1359 if (root->height == 0)
1360 deleted = true;
1361 }
1362 return deleted;
1363 }
1364
1365 parent = node->parent;
1366 if (parent) {
Johannes Weiner449dd692014-04-03 14:47:56 -07001367 unsigned int offset;
Johannes Weiner139e5612014-04-03 14:47:54 -07001368
Johannes Weiner449dd692014-04-03 14:47:56 -07001369 offset = node->path >> RADIX_TREE_HEIGHT_SHIFT;
1370 parent->slots[offset] = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001371 parent->count--;
1372 } else {
1373 root_tag_clear_all(root);
1374 root->height = 0;
1375 root->rnode = NULL;
1376 }
1377
1378 radix_tree_node_free(node);
1379 deleted = true;
1380
1381 node = parent;
1382 } while (node);
1383
1384 return deleted;
1385}
1386
1387/**
Johannes Weiner53c59f22014-04-03 14:47:39 -07001388 * radix_tree_delete_item - delete an item from a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 * @root: radix tree root
1390 * @index: index key
Johannes Weiner53c59f22014-04-03 14:47:39 -07001391 * @item: expected item
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001393 * Remove @item at @index from the radix tree rooted at @root.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001395 * Returns the address of the deleted item, or NULL if it was not present
1396 * or the entry at the given @index was not @item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 */
Johannes Weiner53c59f22014-04-03 14:47:39 -07001398void *radix_tree_delete_item(struct radix_tree_root *root,
1399 unsigned long index, void *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Johannes Weiner139e5612014-04-03 14:47:54 -07001401 struct radix_tree_node *node;
1402 unsigned int offset;
1403 void **slot;
1404 void *entry;
Nick Piggind5274262006-01-08 01:01:41 -08001405 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
Johannes Weiner139e5612014-04-03 14:47:54 -07001407 entry = __radix_tree_lookup(root, index, &node, &slot);
1408 if (!entry)
1409 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Johannes Weiner139e5612014-04-03 14:47:54 -07001411 if (item && entry != item)
1412 return NULL;
1413
1414 if (!node) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001415 root_tag_clear_all(root);
1416 root->rnode = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001417 return entry;
Nick Piggin612d6c12006-06-23 02:03:22 -07001418 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Johannes Weiner139e5612014-04-03 14:47:54 -07001420 offset = index & RADIX_TREE_MAP_MASK;
Johannes Weiner53c59f22014-04-03 14:47:39 -07001421
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 /*
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001423 * Clear all tags associated with the item to be deleted.
1424 * This way of doing it would be inefficient, but seldom is any set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001426 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001427 if (tag_get(node, tag, offset))
Nick Piggin612d6c12006-06-23 02:03:22 -07001428 radix_tree_tag_clear(root, index, tag);
Nick Piggind5274262006-01-08 01:01:41 -08001429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
Johannes Weiner139e5612014-04-03 14:47:54 -07001431 node->slots[offset] = NULL;
1432 node->count--;
Nick Piggina5f51c92006-01-08 01:01:41 -08001433
Johannes Weiner449dd692014-04-03 14:47:56 -07001434 __radix_tree_delete_node(root, node);
Christoph Lameter201b6262005-09-06 15:16:46 -07001435
Johannes Weiner139e5612014-04-03 14:47:54 -07001436 return entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437}
Johannes Weiner53c59f22014-04-03 14:47:39 -07001438EXPORT_SYMBOL(radix_tree_delete_item);
1439
1440/**
1441 * radix_tree_delete - delete an item from a radix tree
1442 * @root: radix tree root
1443 * @index: index key
1444 *
1445 * Remove the item at @index from the radix tree rooted at @root.
1446 *
1447 * Returns the address of the deleted item, or NULL if it was not present.
1448 */
1449void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1450{
1451 return radix_tree_delete_item(root, index, NULL);
1452}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453EXPORT_SYMBOL(radix_tree_delete);
1454
1455/**
1456 * radix_tree_tagged - test whether any items in the tree are tagged
1457 * @root: radix tree root
1458 * @tag: tag to test
1459 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001460int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
Nick Piggin612d6c12006-06-23 02:03:22 -07001462 return root_tag_get(root, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464EXPORT_SYMBOL(radix_tree_tagged);
1465
1466static void
Johannes Weiner449dd692014-04-03 14:47:56 -07001467radix_tree_node_ctor(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
Johannes Weiner449dd692014-04-03 14:47:56 -07001469 struct radix_tree_node *node = arg;
1470
1471 memset(node, 0, sizeof(*node));
1472 INIT_LIST_HEAD(&node->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
1474
1475static __init unsigned long __maxindex(unsigned int height)
1476{
Peter Lund430d2752007-10-16 23:29:35 -07001477 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1478 int shift = RADIX_TREE_INDEX_BITS - width;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Peter Lund430d2752007-10-16 23:29:35 -07001480 if (shift < 0)
1481 return ~0UL;
1482 if (shift >= BITS_PER_LONG)
1483 return 0UL;
1484 return ~0UL >> shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485}
1486
1487static __init void radix_tree_init_maxindex(void)
1488{
1489 unsigned int i;
1490
1491 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1492 height_to_maxindex[i] = __maxindex(i);
1493}
1494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495static int radix_tree_callback(struct notifier_block *nfb,
1496 unsigned long action,
1497 void *hcpu)
1498{
1499 int cpu = (long)hcpu;
1500 struct radix_tree_preload *rtp;
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -07001501 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
1503 /* Free per-cpu pool of perloaded nodes */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07001504 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 rtp = &per_cpu(radix_tree_preloads, cpu);
1506 while (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -07001507 node = rtp->nodes;
1508 rtp->nodes = node->private_data;
1509 kmem_cache_free(radix_tree_node_cachep, node);
1510 rtp->nr--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 }
1512 }
1513 return NOTIFY_OK;
1514}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516void __init radix_tree_init(void)
1517{
1518 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1519 sizeof(struct radix_tree_node), 0,
Christoph Lameter488514d2008-04-28 02:12:05 -07001520 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1521 radix_tree_node_ctor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 radix_tree_init_maxindex();
1523 hotcpu_notifier(radix_tree_callback, 0);
1524}