blob: b65c83036ca49a290db94e8215806913e4b24720 [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
Matthew Wilcox6b053b82016-05-20 17:02:58 -07007 * Copyright (C) 2016 Intel, Matthew Wilcox
8 * Copyright (C) 2016 Intel, Ross Zwisler
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License as
12 * published by the Free Software Foundation; either version 2, or (at
13 * your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 */
24
25#include <linux/errno.h>
26#include <linux/init.h>
27#include <linux/kernel.h>
Paul Gortmaker8bc3bcc2011-11-16 21:29:17 -050028#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/radix-tree.h>
30#include <linux/percpu.h>
31#include <linux/slab.h>
Catalin Marinasce80b062014-06-06 14:38:18 -070032#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/notifier.h>
34#include <linux/cpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/string.h>
36#include <linux/bitops.h>
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080037#include <linux/rcupdate.h>
Frederic Weisbecker92cf2112015-05-12 16:41:46 +020038#include <linux/preempt.h> /* in_interrupt() */
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40
Jeff Moyer26fb1582007-10-16 01:24:49 -070041/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 * Radix tree node cache.
43 */
Christoph Lametere18b8902006-12-06 20:33:20 -080044static struct kmem_cache *radix_tree_node_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
Nick Piggin55368052012-05-29 15:07:34 -070047 * The radix tree is variable-height, so an insert operation not only has
48 * to build the branch to its corresponding item, it also has to build the
49 * branch to existing items if the size has to be increased (by
50 * radix_tree_extend).
51 *
52 * The worst case is a zero height tree with just a single item at index 0,
53 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
54 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
55 * Hence:
56 */
57#define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
58
59/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * Per-cpu pool of preloaded nodes
61 */
62struct radix_tree_preload {
Matthew Wilcox2fcd9002016-05-20 17:03:04 -070063 unsigned nr;
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -070064 /* nodes->private_data points to next preallocated node */
65 struct radix_tree_node *nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066};
Harvey Harrison8cef7d52009-01-06 14:40:50 -080067static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -070069static inline void *node_to_entry(void *ptr)
Nick Piggin27d20fd2010-11-11 14:05:19 -080070{
Matthew Wilcox30ff46cc2016-05-20 17:03:22 -070071 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
Nick Piggin27d20fd2010-11-11 14:05:19 -080072}
73
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -070074#define RADIX_TREE_RETRY node_to_entry(NULL)
Matthew Wilcoxafe0e392016-05-20 17:02:17 -070075
Matthew Wilcoxdb050f22016-05-20 17:01:57 -070076#ifdef CONFIG_RADIX_TREE_MULTIORDER
77/* Sibling slots point directly to another slot in the same node */
78static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
79{
80 void **ptr = node;
81 return (parent->slots <= ptr) &&
82 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
83}
84#else
85static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
86{
87 return false;
88}
89#endif
90
91static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
92 void **slot)
93{
94 return slot - parent->slots;
95}
96
97static unsigned radix_tree_descend(struct radix_tree_node *parent,
98 struct radix_tree_node **nodep, unsigned offset)
99{
100 void **entry = rcu_dereference_raw(parent->slots[offset]);
101
102#ifdef CONFIG_RADIX_TREE_MULTIORDER
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700103 if (radix_tree_is_internal_node(entry)) {
Matthew Wilcoxdb050f22016-05-20 17:01:57 -0700104 unsigned long siboff = get_slot_offset(parent, entry);
105 if (siboff < RADIX_TREE_MAP_SIZE) {
106 offset = siboff;
107 entry = rcu_dereference_raw(parent->slots[offset]);
108 }
109 }
110#endif
111
112 *nodep = (void *)entry;
113 return offset;
114}
115
Nick Piggin612d6c12006-06-23 02:03:22 -0700116static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
117{
118 return root->gfp_mask & __GFP_BITS_MASK;
119}
120
Nick Piggin643b52b2008-06-12 15:21:52 -0700121static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
122 int offset)
123{
124 __set_bit(offset, node->tags[tag]);
125}
126
127static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
128 int offset)
129{
130 __clear_bit(offset, node->tags[tag]);
131}
132
133static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
134 int offset)
135{
136 return test_bit(offset, node->tags[tag]);
137}
138
139static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
140{
141 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
142}
143
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700144static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
Nick Piggin643b52b2008-06-12 15:21:52 -0700145{
146 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
147}
148
149static inline void root_tag_clear_all(struct radix_tree_root *root)
150{
151 root->gfp_mask &= __GFP_BITS_MASK;
152}
153
154static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
155{
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700156 return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
Nick Piggin643b52b2008-06-12 15:21:52 -0700157}
158
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700159static inline unsigned root_tags_get(struct radix_tree_root *root)
160{
161 return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT;
162}
163
Nick Piggin643b52b2008-06-12 15:21:52 -0700164/*
165 * Returns 1 if any slot in the node has this tag set.
166 * Otherwise returns 0.
167 */
168static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
169{
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700170 unsigned idx;
Nick Piggin643b52b2008-06-12 15:21:52 -0700171 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
172 if (node->tags[tag][idx])
173 return 1;
174 }
175 return 0;
176}
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700177
178/**
179 * radix_tree_find_next_bit - find the next set bit in a memory region
180 *
181 * @addr: The address to base the search on
182 * @size: The bitmap size in bits
183 * @offset: The bitnumber to start searching at
184 *
185 * Unrollable variant of find_next_bit() for constant size arrays.
186 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
187 * Returns next bit offset, or size if nothing found.
188 */
189static __always_inline unsigned long
190radix_tree_find_next_bit(const unsigned long *addr,
191 unsigned long size, unsigned long offset)
192{
193 if (!__builtin_constant_p(size))
194 return find_next_bit(addr, size, offset);
195
196 if (offset < size) {
197 unsigned long tmp;
198
199 addr += offset / BITS_PER_LONG;
200 tmp = *addr >> (offset % BITS_PER_LONG);
201 if (tmp)
202 return __ffs(tmp) + offset;
203 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
204 while (offset < size) {
205 tmp = *++addr;
206 if (tmp)
207 return __ffs(tmp) + offset;
208 offset += BITS_PER_LONG;
209 }
210 }
211 return size;
212}
213
Ross Zwisler0796c582016-05-20 17:02:55 -0700214#ifndef __KERNEL__
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700215static void dump_node(struct radix_tree_node *node, unsigned long index)
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700216{
Ross Zwisler0796c582016-05-20 17:02:55 -0700217 unsigned long i;
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700218
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700219 pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d parent %p\n",
Matthew Wilcox0c7fa0a2016-05-20 17:03:07 -0700220 node, node->offset,
Ross Zwisler0796c582016-05-20 17:02:55 -0700221 node->tags[0][0], node->tags[1][0], node->tags[2][0],
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700222 node->shift, node->count, node->parent);
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700223
Ross Zwisler0796c582016-05-20 17:02:55 -0700224 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700225 unsigned long first = index | (i << node->shift);
226 unsigned long last = first | ((1UL << node->shift) - 1);
Ross Zwisler0796c582016-05-20 17:02:55 -0700227 void *entry = node->slots[i];
228 if (!entry)
229 continue;
230 if (is_sibling_entry(node, entry)) {
231 pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n",
232 entry, i,
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700233 *(void **)entry_to_node(entry),
Ross Zwisler0796c582016-05-20 17:02:55 -0700234 first, last);
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700235 } else if (!radix_tree_is_internal_node(entry)) {
Ross Zwisler0796c582016-05-20 17:02:55 -0700236 pr_debug("radix entry %p offset %ld indices %ld-%ld\n",
237 entry, i, first, last);
238 } else {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700239 dump_node(entry_to_node(entry), first);
Ross Zwisler0796c582016-05-20 17:02:55 -0700240 }
241 }
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700242}
243
244/* For debug */
245static void radix_tree_dump(struct radix_tree_root *root)
246{
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700247 pr_debug("radix root: %p rnode %p tags %x\n",
248 root, root->rnode,
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700249 root->gfp_mask >> __GFP_BITS_SHIFT);
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700250 if (!radix_tree_is_internal_node(root->rnode))
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700251 return;
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700252 dump_node(entry_to_node(root->rnode), 0);
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700253}
254#endif
255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256/*
257 * This assumes that the caller has performed appropriate preallocation, and
258 * that the caller has pinned this thread of control to the current CPU.
259 */
260static struct radix_tree_node *
261radix_tree_node_alloc(struct radix_tree_root *root)
262{
Nick Piggine2848a02008-02-04 22:29:10 -0800263 struct radix_tree_node *ret = NULL;
Nick Piggin612d6c12006-06-23 02:03:22 -0700264 gfp_t gfp_mask = root_gfp_mask(root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Jan Kara5e4c0d972013-09-11 14:26:05 -0700266 /*
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700267 * Preload code isn't irq safe and it doesn't make sense to use
268 * preloading during an interrupt anyway as all the allocations have
269 * to be atomic. So just do normal allocation when in interrupt.
Jan Kara5e4c0d972013-09-11 14:26:05 -0700270 */
Mel Gormand0164ad2015-11-06 16:28:21 -0800271 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 struct radix_tree_preload *rtp;
273
Nick Piggine2848a02008-02-04 22:29:10 -0800274 /*
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700275 * Even if the caller has preloaded, try to allocate from the
276 * cache first for the new node to get accounted.
277 */
278 ret = kmem_cache_alloc(radix_tree_node_cachep,
279 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
280 if (ret)
281 goto out;
282
283 /*
Nick Piggine2848a02008-02-04 22:29:10 -0800284 * Provided the caller has preloaded here, we will always
285 * succeed in getting a node here (and never reach
286 * kmem_cache_alloc)
287 */
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700288 rtp = this_cpu_ptr(&radix_tree_preloads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 if (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700290 ret = rtp->nodes;
291 rtp->nodes = ret->private_data;
292 ret->private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 rtp->nr--;
294 }
Catalin Marinasce80b062014-06-06 14:38:18 -0700295 /*
296 * Update the allocation stack trace as this is more useful
297 * for debugging.
298 */
299 kmemleak_update_trace(ret);
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700300 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 }
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700302 ret = kmem_cache_alloc(radix_tree_node_cachep,
303 gfp_mask | __GFP_ACCOUNT);
304out:
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700305 BUG_ON(radix_tree_is_internal_node(ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return ret;
307}
308
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800309static void radix_tree_node_rcu_free(struct rcu_head *head)
310{
311 struct radix_tree_node *node =
312 container_of(head, struct radix_tree_node, rcu_head);
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000313 int i;
Nick Piggin643b52b2008-06-12 15:21:52 -0700314
315 /*
316 * must only free zeroed nodes into the slab. radix_tree_shrink
317 * can leave us with a non-NULL entry in the first slot, so clear
318 * that here to make sure.
319 */
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000320 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
321 tag_clear(node, i, 0);
322
Nick Piggin643b52b2008-06-12 15:21:52 -0700323 node->slots[0] = NULL;
324 node->count = 0;
325
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800326 kmem_cache_free(radix_tree_node_cachep, node);
327}
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329static inline void
330radix_tree_node_free(struct radix_tree_node *node)
331{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800332 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334
335/*
336 * Load up this CPU's radix_tree_node buffer with sufficient objects to
337 * ensure that the addition of a single element in the tree cannot fail. On
338 * success, return zero, with preemption disabled. On error, return -ENOMEM
339 * with preemption not disabled.
David Howellsb34df792009-11-19 18:11:14 +0000340 *
341 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800342 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 */
Jan Kara5e4c0d972013-09-11 14:26:05 -0700344static int __radix_tree_preload(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
346 struct radix_tree_preload *rtp;
347 struct radix_tree_node *node;
348 int ret = -ENOMEM;
349
350 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700351 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700352 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 preempt_enable();
Christoph Lameter488514d2008-04-28 02:12:05 -0700354 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 if (node == NULL)
356 goto out;
357 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700358 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700359 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
360 node->private_data = rtp->nodes;
361 rtp->nodes = node;
362 rtp->nr++;
363 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 kmem_cache_free(radix_tree_node_cachep, node);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 }
367 ret = 0;
368out:
369 return ret;
370}
Jan Kara5e4c0d972013-09-11 14:26:05 -0700371
372/*
373 * Load up this CPU's radix_tree_node buffer with sufficient objects to
374 * ensure that the addition of a single element in the tree cannot fail. On
375 * success, return zero, with preemption disabled. On error, return -ENOMEM
376 * with preemption not disabled.
377 *
378 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800379 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Jan Kara5e4c0d972013-09-11 14:26:05 -0700380 */
381int radix_tree_preload(gfp_t gfp_mask)
382{
383 /* Warn on non-sensical use... */
Mel Gormand0164ad2015-11-06 16:28:21 -0800384 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
Jan Kara5e4c0d972013-09-11 14:26:05 -0700385 return __radix_tree_preload(gfp_mask);
386}
David Chinnerd7f09232007-07-14 16:05:04 +1000387EXPORT_SYMBOL(radix_tree_preload);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Nick Piggin6e954b92006-01-08 01:01:40 -0800389/*
Jan Kara5e4c0d972013-09-11 14:26:05 -0700390 * The same as above function, except we don't guarantee preloading happens.
391 * We do it, if we decide it helps. On success, return zero with preemption
392 * disabled. On error, return -ENOMEM with preemption not disabled.
393 */
394int radix_tree_maybe_preload(gfp_t gfp_mask)
395{
Mel Gormand0164ad2015-11-06 16:28:21 -0800396 if (gfpflags_allow_blocking(gfp_mask))
Jan Kara5e4c0d972013-09-11 14:26:05 -0700397 return __radix_tree_preload(gfp_mask);
398 /* Preloading doesn't help anything with this gfp mask, skip it */
399 preempt_disable();
400 return 0;
401}
402EXPORT_SYMBOL(radix_tree_maybe_preload);
403
404/*
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700405 * The maximum index which can be stored in a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 */
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700407static inline unsigned long shift_maxindex(unsigned int shift)
408{
409 return (RADIX_TREE_MAP_SIZE << shift) - 1;
410}
411
Matthew Wilcox1456a432016-05-20 17:02:08 -0700412static inline unsigned long node_maxindex(struct radix_tree_node *node)
413{
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700414 return shift_maxindex(node->shift);
Matthew Wilcox1456a432016-05-20 17:02:08 -0700415}
416
417static unsigned radix_tree_load_root(struct radix_tree_root *root,
418 struct radix_tree_node **nodep, unsigned long *maxindex)
419{
420 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
421
422 *nodep = node;
423
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700424 if (likely(radix_tree_is_internal_node(node))) {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700425 node = entry_to_node(node);
Matthew Wilcox1456a432016-05-20 17:02:08 -0700426 *maxindex = node_maxindex(node);
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700427 return node->shift + RADIX_TREE_MAP_SHIFT;
Matthew Wilcox1456a432016-05-20 17:02:08 -0700428 }
429
430 *maxindex = 0;
431 return 0;
432}
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434/*
435 * Extend a radix tree so it can store key @index.
436 */
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700437static int radix_tree_extend(struct radix_tree_root *root,
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700438 unsigned long index, unsigned int shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800440 struct radix_tree_node *slot;
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700441 unsigned int maxshift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 int tag;
443
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700444 /* Figure out what the shift should be. */
445 maxshift = shift;
446 while (index > shift_maxindex(maxshift))
447 maxshift += RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700449 slot = root->rnode;
450 if (!slot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 do {
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700454 struct radix_tree_node *node = radix_tree_node_alloc(root);
455
456 if (!node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 return -ENOMEM;
458
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 /* Propagate the aggregated tag info into the new root */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800460 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -0700461 if (root_tag_get(root, tag))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 tag_set(node, tag, 0);
463 }
464
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700465 BUG_ON(shift > BITS_PER_LONG);
466 node->shift = shift;
Matthew Wilcox0c7fa0a2016-05-20 17:03:07 -0700467 node->offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 node->count = 1;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800469 node->parent = NULL;
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700470 if (radix_tree_is_internal_node(slot))
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700471 entry_to_node(slot)->parent = node;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800472 node->slots[0] = slot;
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -0700473 slot = node_to_entry(node);
474 rcu_assign_pointer(root->rnode, slot);
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700475 shift += RADIX_TREE_MAP_SHIFT;
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700476 } while (shift <= maxshift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477out:
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700478 return maxshift + RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
481/**
Johannes Weiner139e5612014-04-03 14:47:54 -0700482 * __radix_tree_create - create a slot in a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 * @root: radix tree root
484 * @index: index key
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700485 * @order: index occupies 2^order aligned slots
Johannes Weiner139e5612014-04-03 14:47:54 -0700486 * @nodep: returns node
487 * @slotp: returns slot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 *
Johannes Weiner139e5612014-04-03 14:47:54 -0700489 * Create, if necessary, and return the node and slot for an item
490 * at position @index in the radix tree @root.
491 *
492 * Until there is more than one item in the tree, no nodes are
493 * allocated and @root->rnode is used as a direct slot instead of
494 * pointing to a node, in which case *@nodep will be NULL.
495 *
496 * Returns -ENOMEM, or 0 for success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700498int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700499 unsigned order, struct radix_tree_node **nodep,
500 void ***slotp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
Christoph Lameter201b6262005-09-06 15:16:46 -0700502 struct radix_tree_node *node = NULL, *slot;
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700503 unsigned long maxindex;
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700504 unsigned int shift, offset;
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700505 unsigned long max = index | ((1UL << order) - 1);
506
507 shift = radix_tree_load_root(root, &slot, &maxindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509 /* Make sure the tree is high enough. */
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700510 if (max > maxindex) {
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700511 int error = radix_tree_extend(root, max, shift);
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700512 if (error < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 return error;
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700514 shift = error;
515 slot = root->rnode;
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700516 if (order == shift)
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700517 shift += RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 }
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 offset = 0; /* uninitialised var warning */
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700521 while (shift > order) {
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700522 shift -= RADIX_TREE_MAP_SHIFT;
Christoph Lameter201b6262005-09-06 15:16:46 -0700523 if (slot == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 /* Have to add a child node. */
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700525 slot = radix_tree_node_alloc(root);
526 if (!slot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 return -ENOMEM;
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700528 slot->shift = shift;
Matthew Wilcox0c7fa0a2016-05-20 17:03:07 -0700529 slot->offset = offset;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800530 slot->parent = node;
Christoph Lameter201b6262005-09-06 15:16:46 -0700531 if (node) {
Matthew Wilcox339e6352016-03-17 14:21:48 -0700532 rcu_assign_pointer(node->slots[offset],
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -0700533 node_to_entry(slot));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 node->count++;
Christoph Lameter201b6262005-09-06 15:16:46 -0700535 } else
Matthew Wilcox339e6352016-03-17 14:21:48 -0700536 rcu_assign_pointer(root->rnode,
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -0700537 node_to_entry(slot));
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700538 } else if (!radix_tree_is_internal_node(slot))
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700539 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
541 /* Go a level down */
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700542 node = entry_to_node(slot);
Matthew Wilcox8a14f4d2016-05-20 17:02:44 -0700543 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
544 offset = radix_tree_descend(node, &slot, offset);
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700545 }
546
Matthew Wilcox57578c22016-05-20 17:01:54 -0700547#ifdef CONFIG_RADIX_TREE_MULTIORDER
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700548 /* Insert pointers to the canonical entry */
Matthew Wilcox3b8c00f2016-05-20 17:01:59 -0700549 if (order > shift) {
550 int i, n = 1 << (order - shift);
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700551 offset = offset & ~(n - 1);
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -0700552 slot = node_to_entry(&node->slots[offset]);
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700553 for (i = 0; i < n; i++) {
554 if (node->slots[offset + i])
555 return -EEXIST;
556 }
557
558 for (i = 1; i < n; i++) {
559 rcu_assign_pointer(node->slots[offset + i], slot);
560 node->count++;
561 }
Nick Piggin612d6c12006-06-23 02:03:22 -0700562 }
Matthew Wilcox57578c22016-05-20 17:01:54 -0700563#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Johannes Weiner139e5612014-04-03 14:47:54 -0700565 if (nodep)
566 *nodep = node;
567 if (slotp)
568 *slotp = node ? node->slots + offset : (void **)&root->rnode;
569 return 0;
570}
571
572/**
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700573 * __radix_tree_insert - insert into a radix tree
Johannes Weiner139e5612014-04-03 14:47:54 -0700574 * @root: radix tree root
575 * @index: index key
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700576 * @order: key covers the 2^order indices around index
Johannes Weiner139e5612014-04-03 14:47:54 -0700577 * @item: item to insert
578 *
579 * Insert an item into the radix tree at position @index.
580 */
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700581int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
582 unsigned order, void *item)
Johannes Weiner139e5612014-04-03 14:47:54 -0700583{
584 struct radix_tree_node *node;
585 void **slot;
586 int error;
587
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700588 BUG_ON(radix_tree_is_internal_node(item));
Johannes Weiner139e5612014-04-03 14:47:54 -0700589
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700590 error = __radix_tree_create(root, index, order, &node, &slot);
Johannes Weiner139e5612014-04-03 14:47:54 -0700591 if (error)
592 return error;
593 if (*slot != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 return -EEXIST;
Johannes Weiner139e5612014-04-03 14:47:54 -0700595 rcu_assign_pointer(*slot, item);
Christoph Lameter201b6262005-09-06 15:16:46 -0700596
Nick Piggin612d6c12006-06-23 02:03:22 -0700597 if (node) {
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700598 unsigned offset = get_slot_offset(node, slot);
Nick Piggin612d6c12006-06-23 02:03:22 -0700599 node->count++;
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700600 BUG_ON(tag_get(node, 0, offset));
601 BUG_ON(tag_get(node, 1, offset));
602 BUG_ON(tag_get(node, 2, offset));
Nick Piggin612d6c12006-06-23 02:03:22 -0700603 } else {
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700604 BUG_ON(root_tags_get(root));
Nick Piggin612d6c12006-06-23 02:03:22 -0700605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 return 0;
608}
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700609EXPORT_SYMBOL(__radix_tree_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
Johannes Weiner139e5612014-04-03 14:47:54 -0700611/**
612 * __radix_tree_lookup - lookup an item in a radix tree
613 * @root: radix tree root
614 * @index: index key
615 * @nodep: returns node
616 * @slotp: returns slot
617 *
618 * Lookup and return the item at position @index in the radix
619 * tree @root.
620 *
621 * Until there is more than one item in the tree, no nodes are
622 * allocated and @root->rnode is used as a direct slot instead of
623 * pointing to a node, in which case *@nodep will be NULL.
Hans Reisera4331362005-11-07 00:59:29 -0800624 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700625void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
626 struct radix_tree_node **nodep, void ***slotp)
Hans Reisera4331362005-11-07 00:59:29 -0800627{
Johannes Weiner139e5612014-04-03 14:47:54 -0700628 struct radix_tree_node *node, *parent;
Matthew Wilcox85829952016-05-20 17:02:20 -0700629 unsigned long maxindex;
630 unsigned int shift;
Johannes Weiner139e5612014-04-03 14:47:54 -0700631 void **slot;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800632
Matthew Wilcox85829952016-05-20 17:02:20 -0700633 restart:
634 parent = NULL;
635 slot = (void **)&root->rnode;
636 shift = radix_tree_load_root(root, &node, &maxindex);
637 if (index > maxindex)
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800638 return NULL;
639
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700640 while (radix_tree_is_internal_node(node)) {
Matthew Wilcox85829952016-05-20 17:02:20 -0700641 unsigned offset;
Johannes Weiner139e5612014-04-03 14:47:54 -0700642
Matthew Wilcox85829952016-05-20 17:02:20 -0700643 if (node == RADIX_TREE_RETRY)
644 goto restart;
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700645 parent = entry_to_node(node);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800646 shift -= RADIX_TREE_MAP_SHIFT;
Matthew Wilcox85829952016-05-20 17:02:20 -0700647 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
648 offset = radix_tree_descend(parent, &node, offset);
649 slot = parent->slots + offset;
650 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800651
Johannes Weiner139e5612014-04-03 14:47:54 -0700652 if (nodep)
653 *nodep = parent;
654 if (slotp)
655 *slotp = slot;
656 return node;
Huang Shijieb72b71c2009-06-16 15:33:42 -0700657}
658
659/**
660 * radix_tree_lookup_slot - lookup a slot in a radix tree
661 * @root: radix tree root
662 * @index: index key
663 *
664 * Returns: the slot corresponding to the position @index in the
665 * radix tree @root. This is useful for update-if-exists operations.
666 *
667 * This function can be called under rcu_read_lock iff the slot is not
668 * modified by radix_tree_replace_slot, otherwise it must be called
669 * exclusive from other writers. Any dereference of the slot must be done
670 * using radix_tree_deref_slot.
671 */
672void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
673{
Johannes Weiner139e5612014-04-03 14:47:54 -0700674 void **slot;
675
676 if (!__radix_tree_lookup(root, index, NULL, &slot))
677 return NULL;
678 return slot;
Hans Reisera4331362005-11-07 00:59:29 -0800679}
680EXPORT_SYMBOL(radix_tree_lookup_slot);
681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682/**
683 * radix_tree_lookup - perform lookup operation on a radix tree
684 * @root: radix tree root
685 * @index: index key
686 *
687 * Lookup the item at the position @index in the radix tree @root.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800688 *
689 * This function can be called under rcu_read_lock, however the caller
690 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
691 * them safely). No RCU barriers are required to access or modify the
692 * returned item, however.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 */
694void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
695{
Johannes Weiner139e5612014-04-03 14:47:54 -0700696 return __radix_tree_lookup(root, index, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697}
698EXPORT_SYMBOL(radix_tree_lookup);
699
700/**
701 * radix_tree_tag_set - set a tag on a radix tree node
702 * @root: radix tree root
703 * @index: index key
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700704 * @tag: tag index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800706 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
707 * corresponding to @index in the radix tree. From
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 * the root all the way down to the leaf node.
709 *
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700710 * Returns the address of the tagged item. Setting a tag on a not-present
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 * item is a bug.
712 */
713void *radix_tree_tag_set(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800714 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
Ross Zwislerfb969902016-05-20 17:02:32 -0700716 struct radix_tree_node *node, *parent;
717 unsigned long maxindex;
718 unsigned int shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Ross Zwislerfb969902016-05-20 17:02:32 -0700720 shift = radix_tree_load_root(root, &node, &maxindex);
721 BUG_ON(index > maxindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700723 while (radix_tree_is_internal_node(node)) {
Ross Zwislerfb969902016-05-20 17:02:32 -0700724 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 shift -= RADIX_TREE_MAP_SHIFT;
Ross Zwislerfb969902016-05-20 17:02:32 -0700727 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
728
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700729 parent = entry_to_node(node);
Ross Zwislerfb969902016-05-20 17:02:32 -0700730 offset = radix_tree_descend(parent, &node, offset);
731 BUG_ON(!node);
732
733 if (!tag_get(parent, tag, offset))
734 tag_set(parent, tag, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
736
Nick Piggin612d6c12006-06-23 02:03:22 -0700737 /* set the root's tag bit */
Ross Zwislerfb969902016-05-20 17:02:32 -0700738 if (!root_tag_get(root, tag))
Nick Piggin612d6c12006-06-23 02:03:22 -0700739 root_tag_set(root, tag);
740
Ross Zwislerfb969902016-05-20 17:02:32 -0700741 return node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743EXPORT_SYMBOL(radix_tree_tag_set);
744
745/**
746 * radix_tree_tag_clear - clear a tag on a radix tree node
747 * @root: radix tree root
748 * @index: index key
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700749 * @tag: tag index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800751 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700752 * corresponding to @index in the radix tree. If this causes
753 * the leaf node to have no tags set then clear the tag in the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 * next-to-leaf node, etc.
755 *
756 * Returns the address of the tagged item on success, else NULL. ie:
757 * has the same return value and semantics as radix_tree_lookup().
758 */
759void *radix_tree_tag_clear(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800760 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761{
Ross Zwisler00f47b52016-05-20 17:02:35 -0700762 struct radix_tree_node *node, *parent;
763 unsigned long maxindex;
764 unsigned int shift;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800765 int uninitialized_var(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
Ross Zwisler00f47b52016-05-20 17:02:35 -0700767 shift = radix_tree_load_root(root, &node, &maxindex);
768 if (index > maxindex)
769 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Ross Zwisler00f47b52016-05-20 17:02:35 -0700771 parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700773 while (radix_tree_is_internal_node(node)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 shift -= RADIX_TREE_MAP_SHIFT;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800775 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Ross Zwisler00f47b52016-05-20 17:02:35 -0700776
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700777 parent = entry_to_node(node);
Ross Zwisler00f47b52016-05-20 17:02:35 -0700778 offset = radix_tree_descend(parent, &node, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 }
780
Ross Zwisler00f47b52016-05-20 17:02:35 -0700781 if (node == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 goto out;
783
Ross Zwisler00f47b52016-05-20 17:02:35 -0700784 index >>= shift;
785
786 while (parent) {
787 if (!tag_get(parent, tag, offset))
Nick Piggind5274262006-01-08 01:01:41 -0800788 goto out;
Ross Zwisler00f47b52016-05-20 17:02:35 -0700789 tag_clear(parent, tag, offset);
790 if (any_tag_set(parent, tag))
Nick Piggin6e954b92006-01-08 01:01:40 -0800791 goto out;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800792
793 index >>= RADIX_TREE_MAP_SHIFT;
794 offset = index & RADIX_TREE_MAP_MASK;
Ross Zwisler00f47b52016-05-20 17:02:35 -0700795 parent = parent->parent;
Nick Piggin612d6c12006-06-23 02:03:22 -0700796 }
797
798 /* clear the root's tag bit */
799 if (root_tag_get(root, tag))
800 root_tag_clear(root, tag);
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802out:
Ross Zwisler00f47b52016-05-20 17:02:35 -0700803 return node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805EXPORT_SYMBOL(radix_tree_tag_clear);
806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807/**
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700808 * radix_tree_tag_get - get a tag on a radix tree node
809 * @root: radix tree root
810 * @index: index key
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700811 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 *
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700813 * Return values:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 *
Nick Piggin612d6c12006-06-23 02:03:22 -0700815 * 0: tag not present or not set
816 * 1: tag set
David Howellsce826532010-04-06 22:36:20 +0100817 *
818 * Note that the return value of this function may not be relied on, even if
819 * the RCU lock is held, unless tag modification and node deletion are excluded
820 * from concurrency.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 */
822int radix_tree_tag_get(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800823 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824{
Ross Zwisler4589ba62016-05-20 17:02:38 -0700825 struct radix_tree_node *node, *parent;
826 unsigned long maxindex;
827 unsigned int shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Nick Piggin612d6c12006-06-23 02:03:22 -0700829 if (!root_tag_get(root, tag))
830 return 0;
831
Ross Zwisler4589ba62016-05-20 17:02:38 -0700832 shift = radix_tree_load_root(root, &node, &maxindex);
833 if (index > maxindex)
834 return 0;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800835 if (node == NULL)
836 return 0;
837
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700838 while (radix_tree_is_internal_node(node)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 int offset;
840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 shift -= RADIX_TREE_MAP_SHIFT;
Ross Zwisler4589ba62016-05-20 17:02:38 -0700842 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
843
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700844 parent = entry_to_node(node);
Ross Zwisler4589ba62016-05-20 17:02:38 -0700845 offset = radix_tree_descend(parent, &node, offset);
846
847 if (!node)
848 return 0;
849 if (!tag_get(parent, tag, offset))
850 return 0;
851 if (node == RADIX_TREE_RETRY)
852 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 }
Ross Zwisler4589ba62016-05-20 17:02:38 -0700854
855 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857EXPORT_SYMBOL(radix_tree_tag_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Ross Zwisler21ef5332016-05-20 17:02:26 -0700859static inline void __set_iter_shift(struct radix_tree_iter *iter,
860 unsigned int shift)
861{
862#ifdef CONFIG_RADIX_TREE_MULTIORDER
863 iter->shift = shift;
864#endif
865}
866
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700867/**
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700868 * radix_tree_next_chunk - find next chunk of slots for iteration
869 *
870 * @root: radix tree root
871 * @iter: iterator state
872 * @flags: RADIX_TREE_ITER_* flags and tag index
873 * Returns: pointer to chunk first slot, or NULL if iteration is over
874 */
875void **radix_tree_next_chunk(struct radix_tree_root *root,
876 struct radix_tree_iter *iter, unsigned flags)
877{
878 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
879 struct radix_tree_node *rnode, *node;
Ross Zwisler21ef5332016-05-20 17:02:26 -0700880 unsigned long index, offset, maxindex;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700881
882 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
883 return NULL;
884
885 /*
886 * Catch next_index overflow after ~0UL. iter->index never overflows
887 * during iterating; it can be zero only at the beginning.
888 * And we cannot overflow iter->next_index in a single step,
889 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
Konstantin Khlebnikovfffaee32012-06-05 21:36:33 +0400890 *
891 * This condition also used by radix_tree_next_slot() to stop
892 * contiguous iterating, and forbid swithing to the next chunk.
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700893 */
894 index = iter->next_index;
895 if (!index && iter->index)
896 return NULL;
897
Ross Zwisler21ef5332016-05-20 17:02:26 -0700898 restart:
899 shift = radix_tree_load_root(root, &rnode, &maxindex);
900 if (index > maxindex)
901 return NULL;
902
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700903 if (radix_tree_is_internal_node(rnode)) {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700904 rnode = entry_to_node(rnode);
Ross Zwisler21ef5332016-05-20 17:02:26 -0700905 } else if (rnode) {
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700906 /* Single-slot tree */
Ross Zwisler21ef5332016-05-20 17:02:26 -0700907 iter->index = index;
908 iter->next_index = maxindex + 1;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700909 iter->tags = 1;
Ross Zwisler21ef5332016-05-20 17:02:26 -0700910 __set_iter_shift(iter, shift);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700911 return (void **)&root->rnode;
912 } else
913 return NULL;
914
Ross Zwisler21ef5332016-05-20 17:02:26 -0700915 shift -= RADIX_TREE_MAP_SHIFT;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700916 offset = index >> shift;
917
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700918 node = rnode;
919 while (1) {
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700920 struct radix_tree_node *slot;
Ross Zwisler21ef5332016-05-20 17:02:26 -0700921 unsigned new_off = radix_tree_descend(node, &slot, offset);
922
923 if (new_off < offset) {
924 offset = new_off;
925 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
926 index |= offset << shift;
927 }
928
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700929 if ((flags & RADIX_TREE_ITER_TAGGED) ?
Ross Zwisler21ef5332016-05-20 17:02:26 -0700930 !tag_get(node, tag, offset) : !slot) {
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700931 /* Hole detected */
932 if (flags & RADIX_TREE_ITER_CONTIG)
933 return NULL;
934
935 if (flags & RADIX_TREE_ITER_TAGGED)
936 offset = radix_tree_find_next_bit(
937 node->tags[tag],
938 RADIX_TREE_MAP_SIZE,
939 offset + 1);
940 else
941 while (++offset < RADIX_TREE_MAP_SIZE) {
Ross Zwisler21ef5332016-05-20 17:02:26 -0700942 void *slot = node->slots[offset];
943 if (is_sibling_entry(node, slot))
944 continue;
945 if (slot)
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700946 break;
947 }
948 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
949 index += offset << shift;
950 /* Overflow after ~0UL */
951 if (!index)
952 return NULL;
953 if (offset == RADIX_TREE_MAP_SIZE)
954 goto restart;
Ross Zwisler21ef5332016-05-20 17:02:26 -0700955 slot = rcu_dereference_raw(node->slots[offset]);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700956 }
957
Ross Zwisler21ef5332016-05-20 17:02:26 -0700958 if ((slot == NULL) || (slot == RADIX_TREE_RETRY))
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700959 goto restart;
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700960 if (!radix_tree_is_internal_node(slot))
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700961 break;
Ross Zwisler21ef5332016-05-20 17:02:26 -0700962
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700963 node = entry_to_node(slot);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700964 shift -= RADIX_TREE_MAP_SHIFT;
965 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
966 }
967
968 /* Update the iterator state */
Ross Zwisler21ef5332016-05-20 17:02:26 -0700969 iter->index = index & ~((1 << shift) - 1);
970 iter->next_index = (index | ((RADIX_TREE_MAP_SIZE << shift) - 1)) + 1;
971 __set_iter_shift(iter, shift);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700972
973 /* Construct iter->tags bit-mask from node->tags[tag] array */
974 if (flags & RADIX_TREE_ITER_TAGGED) {
975 unsigned tag_long, tag_bit;
976
977 tag_long = offset / BITS_PER_LONG;
978 tag_bit = offset % BITS_PER_LONG;
979 iter->tags = node->tags[tag][tag_long] >> tag_bit;
980 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
981 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
982 /* Pick tags from next element */
983 if (tag_bit)
984 iter->tags |= node->tags[tag][tag_long + 1] <<
985 (BITS_PER_LONG - tag_bit);
986 /* Clip chunk size, here only BITS_PER_LONG tags */
987 iter->next_index = index + BITS_PER_LONG;
988 }
989 }
990
991 return node->slots + offset;
992}
993EXPORT_SYMBOL(radix_tree_next_chunk);
994
995/**
Jan Karaebf8aa42010-08-09 17:19:11 -0700996 * radix_tree_range_tag_if_tagged - for each item in given range set given
997 * tag if item has another tag set
998 * @root: radix tree root
999 * @first_indexp: pointer to a starting index of a range to scan
1000 * @last_index: last index of a range to scan
1001 * @nr_to_tag: maximum number items to tag
1002 * @iftag: tag index to test
1003 * @settag: tag index to set if tested tag is set
1004 *
1005 * This function scans range of radix tree from first_index to last_index
1006 * (inclusive). For each item in the range if iftag is set, the function sets
1007 * also settag. The function stops either after tagging nr_to_tag items or
1008 * after reaching last_index.
1009 *
Dave Chinner144dcfc2010-08-23 10:33:53 +10001010 * The tags must be set from the leaf level only and propagated back up the
1011 * path to the root. We must do this so that we resolve the full path before
1012 * setting any tags on intermediate nodes. If we set tags as we descend, then
1013 * we can get to the leaf node and find that the index that has the iftag
1014 * set is outside the range we are scanning. This reults in dangling tags and
1015 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1016 *
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001017 * The function returns the number of leaves where the tag was set and sets
Jan Karaebf8aa42010-08-09 17:19:11 -07001018 * *first_indexp to the first unscanned index.
Jan Karad5ed3a42010-08-19 14:13:33 -07001019 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1020 * be prepared to handle that.
Jan Karaebf8aa42010-08-09 17:19:11 -07001021 */
1022unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1023 unsigned long *first_indexp, unsigned long last_index,
1024 unsigned long nr_to_tag,
1025 unsigned int iftag, unsigned int settag)
1026{
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001027 struct radix_tree_node *slot, *node = NULL;
1028 unsigned long maxindex;
1029 unsigned int shift = radix_tree_load_root(root, &slot, &maxindex);
Dave Chinner144dcfc2010-08-23 10:33:53 +10001030 unsigned long tagged = 0;
1031 unsigned long index = *first_indexp;
Jan Karaebf8aa42010-08-09 17:19:11 -07001032
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001033 last_index = min(last_index, maxindex);
Jan Karaebf8aa42010-08-09 17:19:11 -07001034 if (index > last_index)
1035 return 0;
1036 if (!nr_to_tag)
1037 return 0;
1038 if (!root_tag_get(root, iftag)) {
1039 *first_indexp = last_index + 1;
1040 return 0;
1041 }
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001042 if (!radix_tree_is_internal_node(slot)) {
Jan Karaebf8aa42010-08-09 17:19:11 -07001043 *first_indexp = last_index + 1;
1044 root_tag_set(root, settag);
1045 return 1;
1046 }
1047
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001048 node = entry_to_node(slot);
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001049 shift -= RADIX_TREE_MAP_SHIFT;
Jan Karaebf8aa42010-08-09 17:19:11 -07001050
1051 for (;;) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001052 unsigned long upindex;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001053 unsigned offset;
Jan Karaebf8aa42010-08-09 17:19:11 -07001054
1055 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001056 offset = radix_tree_descend(node, &slot, offset);
1057 if (!slot)
Jan Karaebf8aa42010-08-09 17:19:11 -07001058 goto next;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001059 if (!tag_get(node, iftag, offset))
Jan Karaebf8aa42010-08-09 17:19:11 -07001060 goto next;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001061 /* Sibling slots never have tags set on them */
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001062 if (radix_tree_is_internal_node(slot)) {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001063 node = entry_to_node(slot);
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001064 shift -= RADIX_TREE_MAP_SHIFT;
1065 continue;
Jan Karaebf8aa42010-08-09 17:19:11 -07001066 }
Dave Chinner144dcfc2010-08-23 10:33:53 +10001067
1068 /* tag the leaf */
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001069 tagged++;
1070 tag_set(node, settag, offset);
Dave Chinner144dcfc2010-08-23 10:33:53 +10001071
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001072 slot = node->parent;
Dave Chinner144dcfc2010-08-23 10:33:53 +10001073 /* walk back up the path tagging interior nodes */
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001074 upindex = index >> shift;
1075 while (slot) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001076 upindex >>= RADIX_TREE_MAP_SHIFT;
1077 offset = upindex & RADIX_TREE_MAP_MASK;
1078
Dave Chinner144dcfc2010-08-23 10:33:53 +10001079 /* stop if we find a node with the tag already set */
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001080 if (tag_get(slot, settag, offset))
Dave Chinner144dcfc2010-08-23 10:33:53 +10001081 break;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001082 tag_set(slot, settag, offset);
1083 slot = slot->parent;
Dave Chinner144dcfc2010-08-23 10:33:53 +10001084 }
1085
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001086 next:
Jan Karaebf8aa42010-08-09 17:19:11 -07001087 /* Go to next item at level determined by 'shift' */
1088 index = ((index >> shift) + 1) << shift;
Jan Karad5ed3a42010-08-19 14:13:33 -07001089 /* Overflow can happen when last_index is ~0UL... */
1090 if (index > last_index || !index)
Jan Karaebf8aa42010-08-09 17:19:11 -07001091 break;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001092 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1093 while (offset == 0) {
Jan Karaebf8aa42010-08-09 17:19:11 -07001094 /*
1095 * We've fully scanned this node. Go up. Because
1096 * last_index is guaranteed to be in the tree, what
1097 * we do below cannot wander astray.
1098 */
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001099 node = node->parent;
Jan Karaebf8aa42010-08-09 17:19:11 -07001100 shift += RADIX_TREE_MAP_SHIFT;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001101 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
Jan Karaebf8aa42010-08-09 17:19:11 -07001102 }
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001103 if (is_sibling_entry(node, node->slots[offset]))
1104 goto next;
1105 if (tagged >= nr_to_tag)
1106 break;
Jan Karaebf8aa42010-08-09 17:19:11 -07001107 }
1108 /*
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -08001109 * We need not to tag the root tag if there is no tag which is set with
1110 * settag within the range from *first_indexp to last_index.
Jan Karaebf8aa42010-08-09 17:19:11 -07001111 */
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -08001112 if (tagged > 0)
1113 root_tag_set(root, settag);
Jan Karaebf8aa42010-08-09 17:19:11 -07001114 *first_indexp = index;
1115
1116 return tagged;
1117}
1118EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120/**
1121 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1122 * @root: radix tree root
1123 * @results: where the results of the lookup are placed
1124 * @first_index: start the lookup from this key
1125 * @max_items: place up to this many items at *results
1126 *
1127 * Performs an index-ascending scan of the tree for present items. Places
1128 * them at *@results and returns the number of items which were placed at
1129 * *@results.
1130 *
1131 * The implementation is naive.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001132 *
1133 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1134 * rcu_read_lock. In this case, rather than the returned results being
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001135 * an atomic snapshot of the tree at a single point in time, the
1136 * semantics of an RCU protected gang lookup are as though multiple
1137 * radix_tree_lookups have been issued in individual locks, and results
1138 * stored in 'results'.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 */
1140unsigned int
1141radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1142 unsigned long first_index, unsigned int max_items)
1143{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001144 struct radix_tree_iter iter;
1145 void **slot;
1146 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001148 if (unlikely(!max_items))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001149 return 0;
1150
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001151 radix_tree_for_each_slot(slot, root, &iter, first_index) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001152 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001153 if (!results[ret])
1154 continue;
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001155 if (radix_tree_is_internal_node(results[ret])) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001156 slot = radix_tree_iter_retry(&iter);
1157 continue;
1158 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001159 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 return ret;
1164}
1165EXPORT_SYMBOL(radix_tree_gang_lookup);
1166
Nick Piggin47feff22008-07-25 19:45:29 -07001167/**
1168 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1169 * @root: radix tree root
1170 * @results: where the results of the lookup are placed
Hugh Dickins63286502011-08-03 16:21:18 -07001171 * @indices: where their indices should be placed (but usually NULL)
Nick Piggin47feff22008-07-25 19:45:29 -07001172 * @first_index: start the lookup from this key
1173 * @max_items: place up to this many items at *results
1174 *
1175 * Performs an index-ascending scan of the tree for present items. Places
1176 * their slots at *@results and returns the number of items which were
1177 * placed at *@results.
1178 *
1179 * The implementation is naive.
1180 *
1181 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1182 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1183 * protection, radix_tree_deref_slot may fail requiring a retry.
1184 */
1185unsigned int
Hugh Dickins63286502011-08-03 16:21:18 -07001186radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1187 void ***results, unsigned long *indices,
Nick Piggin47feff22008-07-25 19:45:29 -07001188 unsigned long first_index, unsigned int max_items)
1189{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001190 struct radix_tree_iter iter;
1191 void **slot;
1192 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001193
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001194 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001195 return 0;
1196
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001197 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1198 results[ret] = slot;
Hugh Dickins63286502011-08-03 16:21:18 -07001199 if (indices)
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001200 indices[ret] = iter.index;
1201 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001202 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001203 }
1204
1205 return ret;
1206}
1207EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1208
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209/**
1210 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1211 * based on a tag
1212 * @root: radix tree root
1213 * @results: where the results of the lookup are placed
1214 * @first_index: start the lookup from this key
1215 * @max_items: place up to this many items at *results
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001216 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 *
1218 * Performs an index-ascending scan of the tree for present items which
1219 * have the tag indexed by @tag set. Places the items at *@results and
1220 * returns the number of items which were placed at *@results.
1221 */
1222unsigned int
1223radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001224 unsigned long first_index, unsigned int max_items,
1225 unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001227 struct radix_tree_iter iter;
1228 void **slot;
1229 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001231 if (unlikely(!max_items))
Nick Piggin612d6c12006-06-23 02:03:22 -07001232 return 0;
1233
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001234 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001235 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001236 if (!results[ret])
1237 continue;
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001238 if (radix_tree_is_internal_node(results[ret])) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001239 slot = radix_tree_iter_retry(&iter);
1240 continue;
1241 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001242 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001245
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 return ret;
1247}
1248EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1249
1250/**
Nick Piggin47feff22008-07-25 19:45:29 -07001251 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1252 * radix tree based on a tag
1253 * @root: radix tree root
1254 * @results: where the results of the lookup are placed
1255 * @first_index: start the lookup from this key
1256 * @max_items: place up to this many items at *results
1257 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1258 *
1259 * Performs an index-ascending scan of the tree for present items which
1260 * have the tag indexed by @tag set. Places the slots at *@results and
1261 * returns the number of slots which were placed at *@results.
1262 */
1263unsigned int
1264radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1265 unsigned long first_index, unsigned int max_items,
1266 unsigned int tag)
1267{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001268 struct radix_tree_iter iter;
1269 void **slot;
1270 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001271
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001272 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001273 return 0;
1274
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001275 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1276 results[ret] = slot;
1277 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001278 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001279 }
1280
1281 return ret;
1282}
1283EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1284
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001285#if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1286#include <linux/sched.h> /* for cond_resched() */
1287
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001288struct locate_info {
1289 unsigned long found_index;
1290 bool stop;
1291};
1292
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001293/*
1294 * This linear search is at present only useful to shmem_unuse_inode().
1295 */
1296static unsigned long __locate(struct radix_tree_node *slot, void *item,
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001297 unsigned long index, struct locate_info *info)
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001298{
Matthew Wilcox0c7fa0a2016-05-20 17:03:07 -07001299 unsigned int shift;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001300 unsigned long i;
1301
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -07001302 shift = slot->shift + RADIX_TREE_MAP_SHIFT;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001303
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001304 do {
Matthew Wilcoxe6145232016-03-17 14:21:54 -07001305 shift -= RADIX_TREE_MAP_SHIFT;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001306
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001307 for (i = (index >> shift) & RADIX_TREE_MAP_MASK;
1308 i < RADIX_TREE_MAP_SIZE;
1309 i++, index += (1UL << shift)) {
1310 struct radix_tree_node *node =
1311 rcu_dereference_raw(slot->slots[i]);
1312 if (node == RADIX_TREE_RETRY)
1313 goto out;
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001314 if (!radix_tree_is_internal_node(node)) {
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001315 if (node == item) {
1316 info->found_index = index;
1317 info->stop = true;
1318 goto out;
1319 }
1320 continue;
1321 }
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001322 node = entry_to_node(node);
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001323 if (is_sibling_entry(slot, node))
1324 continue;
1325 slot = node;
1326 break;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001327 }
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001328 if (i == RADIX_TREE_MAP_SIZE)
1329 break;
1330 } while (shift);
1331
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001332out:
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001333 if ((index == 0) && (i == RADIX_TREE_MAP_SIZE))
1334 info->stop = true;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001335 return index;
1336}
1337
1338/**
1339 * radix_tree_locate_item - search through radix tree for item
1340 * @root: radix tree root
1341 * @item: item to be found
1342 *
1343 * Returns index where item was found, or -1 if not found.
1344 * Caller must hold no lock (since this time-consuming function needs
1345 * to be preemptible), and must check afterwards if item is still there.
1346 */
1347unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1348{
1349 struct radix_tree_node *node;
1350 unsigned long max_index;
1351 unsigned long cur_index = 0;
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001352 struct locate_info info = {
1353 .found_index = -1,
1354 .stop = false,
1355 };
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001356
1357 do {
1358 rcu_read_lock();
1359 node = rcu_dereference_raw(root->rnode);
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001360 if (!radix_tree_is_internal_node(node)) {
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001361 rcu_read_unlock();
1362 if (node == item)
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001363 info.found_index = 0;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001364 break;
1365 }
1366
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001367 node = entry_to_node(node);
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001368
1369 max_index = node_maxindex(node);
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001370 if (cur_index > max_index) {
1371 rcu_read_unlock();
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001372 break;
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001373 }
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001374
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001375 cur_index = __locate(node, item, cur_index, &info);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001376 rcu_read_unlock();
1377 cond_resched();
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001378 } while (!info.stop && cur_index <= max_index);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001379
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001380 return info.found_index;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001381}
1382#else
1383unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1384{
1385 return -1;
1386}
1387#endif /* CONFIG_SHMEM && CONFIG_SWAP */
Nick Piggin47feff22008-07-25 19:45:29 -07001388
1389/**
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001390 * radix_tree_shrink - shrink radix tree to minimum height
Nick Piggina5f51c92006-01-08 01:01:41 -08001391 * @root radix tree root
1392 */
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001393static inline bool radix_tree_shrink(struct radix_tree_root *root)
Nick Piggina5f51c92006-01-08 01:01:41 -08001394{
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001395 bool shrunk = false;
1396
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001397 for (;;) {
Nick Piggina5f51c92006-01-08 01:01:41 -08001398 struct radix_tree_node *to_free = root->rnode;
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001399 struct radix_tree_node *slot;
Nick Piggina5f51c92006-01-08 01:01:41 -08001400
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001401 if (!radix_tree_is_internal_node(to_free))
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001402 break;
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001403 to_free = entry_to_node(to_free);
Nick Pigginc0bc9872007-10-16 01:24:42 -07001404
1405 /*
1406 * The candidate node has more than one child, or its child
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001407 * is not at the leftmost slot, or the child is a multiorder
1408 * entry, we cannot shrink.
Nick Pigginc0bc9872007-10-16 01:24:42 -07001409 */
1410 if (to_free->count != 1)
1411 break;
Matthew Wilcox339e6352016-03-17 14:21:48 -07001412 slot = to_free->slots[0];
1413 if (!slot)
Nick Pigginc0bc9872007-10-16 01:24:42 -07001414 break;
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001415 if (!radix_tree_is_internal_node(slot) && to_free->shift)
Matthew Wilcoxafe0e392016-05-20 17:02:17 -07001416 break;
1417
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001418 if (radix_tree_is_internal_node(slot))
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001419 entry_to_node(slot)->parent = NULL;
Nick Pigginc0bc9872007-10-16 01:24:42 -07001420
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001421 /*
1422 * We don't need rcu_assign_pointer(), since we are simply
Nick Piggin27d20fd2010-11-11 14:05:19 -08001423 * moving the node from one part of the tree to another: if it
1424 * was safe to dereference the old pointer to it
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001425 * (to_free->slots[0]), it will be safe to dereference the new
Nick Piggin27d20fd2010-11-11 14:05:19 -08001426 * one (root->rnode) as far as dependent read barriers go.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001427 */
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001428 root->rnode = slot;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001429
1430 /*
1431 * We have a dilemma here. The node's slot[0] must not be
1432 * NULLed in case there are concurrent lookups expecting to
1433 * find the item. However if this was a bottom-level node,
1434 * then it may be subject to the slot pointer being visible
1435 * to callers dereferencing it. If item corresponding to
1436 * slot[0] is subsequently deleted, these callers would expect
1437 * their slot to become empty sooner or later.
1438 *
1439 * For example, lockless pagecache will look up a slot, deref
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001440 * the page pointer, and if the page has 0 refcount it means it
Nick Piggin27d20fd2010-11-11 14:05:19 -08001441 * was concurrently deleted from pagecache so try the deref
1442 * again. Fortunately there is already a requirement for logic
1443 * to retry the entire slot lookup -- the indirect pointer
1444 * problem (replacing direct root node with an indirect pointer
1445 * also results in a stale slot). So tag the slot as indirect
1446 * to force callers to retry.
1447 */
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001448 if (!radix_tree_is_internal_node(slot))
Matthew Wilcoxafe0e392016-05-20 17:02:17 -07001449 to_free->slots[0] = RADIX_TREE_RETRY;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001450
Nick Piggina5f51c92006-01-08 01:01:41 -08001451 radix_tree_node_free(to_free);
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001452 shrunk = true;
Nick Piggina5f51c92006-01-08 01:01:41 -08001453 }
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001454
1455 return shrunk;
Nick Piggina5f51c92006-01-08 01:01:41 -08001456}
1457
1458/**
Johannes Weiner139e5612014-04-03 14:47:54 -07001459 * __radix_tree_delete_node - try to free node after clearing a slot
1460 * @root: radix tree root
Johannes Weiner139e5612014-04-03 14:47:54 -07001461 * @node: node containing @index
1462 *
1463 * After clearing the slot at @index in @node from radix tree
1464 * rooted at @root, call this function to attempt freeing the
1465 * node and shrinking the tree.
1466 *
1467 * Returns %true if @node was freed, %false otherwise.
1468 */
Johannes Weiner449dd692014-04-03 14:47:56 -07001469bool __radix_tree_delete_node(struct radix_tree_root *root,
Johannes Weiner139e5612014-04-03 14:47:54 -07001470 struct radix_tree_node *node)
1471{
1472 bool deleted = false;
1473
1474 do {
1475 struct radix_tree_node *parent;
1476
1477 if (node->count) {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001478 if (node == entry_to_node(root->rnode))
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001479 deleted |= radix_tree_shrink(root);
Johannes Weiner139e5612014-04-03 14:47:54 -07001480 return deleted;
1481 }
1482
1483 parent = node->parent;
1484 if (parent) {
Matthew Wilcox0c7fa0a2016-05-20 17:03:07 -07001485 parent->slots[node->offset] = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001486 parent->count--;
1487 } else {
1488 root_tag_clear_all(root);
Johannes Weiner139e5612014-04-03 14:47:54 -07001489 root->rnode = NULL;
1490 }
1491
1492 radix_tree_node_free(node);
1493 deleted = true;
1494
1495 node = parent;
1496 } while (node);
1497
1498 return deleted;
1499}
1500
Matthew Wilcox57578c22016-05-20 17:01:54 -07001501static inline void delete_sibling_entries(struct radix_tree_node *node,
1502 void *ptr, unsigned offset)
1503{
1504#ifdef CONFIG_RADIX_TREE_MULTIORDER
1505 int i;
1506 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1507 if (node->slots[offset + i] != ptr)
1508 break;
1509 node->slots[offset + i] = NULL;
1510 node->count--;
1511 }
1512#endif
1513}
1514
Johannes Weiner139e5612014-04-03 14:47:54 -07001515/**
Johannes Weiner53c59f22014-04-03 14:47:39 -07001516 * radix_tree_delete_item - delete an item from a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 * @root: radix tree root
1518 * @index: index key
Johannes Weiner53c59f22014-04-03 14:47:39 -07001519 * @item: expected item
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001521 * Remove @item at @index from the radix tree rooted at @root.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001523 * Returns the address of the deleted item, or NULL if it was not present
1524 * or the entry at the given @index was not @item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 */
Johannes Weiner53c59f22014-04-03 14:47:39 -07001526void *radix_tree_delete_item(struct radix_tree_root *root,
1527 unsigned long index, void *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528{
Johannes Weiner139e5612014-04-03 14:47:54 -07001529 struct radix_tree_node *node;
Matthew Wilcox57578c22016-05-20 17:01:54 -07001530 unsigned int offset;
Johannes Weiner139e5612014-04-03 14:47:54 -07001531 void **slot;
1532 void *entry;
Nick Piggind5274262006-01-08 01:01:41 -08001533 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
Johannes Weiner139e5612014-04-03 14:47:54 -07001535 entry = __radix_tree_lookup(root, index, &node, &slot);
1536 if (!entry)
1537 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Johannes Weiner139e5612014-04-03 14:47:54 -07001539 if (item && entry != item)
1540 return NULL;
1541
1542 if (!node) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001543 root_tag_clear_all(root);
1544 root->rnode = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001545 return entry;
Nick Piggin612d6c12006-06-23 02:03:22 -07001546 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
Matthew Wilcox29e09672016-05-20 17:02:02 -07001548 offset = get_slot_offset(node, slot);
Johannes Weiner53c59f22014-04-03 14:47:39 -07001549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 /*
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001551 * Clear all tags associated with the item to be deleted.
1552 * This way of doing it would be inefficient, but seldom is any set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001554 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Hugh Dickinse2bdb932012-01-12 17:20:41 -08001555 if (tag_get(node, tag, offset))
Nick Piggin612d6c12006-06-23 02:03:22 -07001556 radix_tree_tag_clear(root, index, tag);
Nick Piggind5274262006-01-08 01:01:41 -08001557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -07001559 delete_sibling_entries(node, node_to_entry(slot), offset);
Johannes Weiner139e5612014-04-03 14:47:54 -07001560 node->slots[offset] = NULL;
1561 node->count--;
Nick Piggina5f51c92006-01-08 01:01:41 -08001562
Johannes Weiner449dd692014-04-03 14:47:56 -07001563 __radix_tree_delete_node(root, node);
Christoph Lameter201b6262005-09-06 15:16:46 -07001564
Johannes Weiner139e5612014-04-03 14:47:54 -07001565 return entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566}
Johannes Weiner53c59f22014-04-03 14:47:39 -07001567EXPORT_SYMBOL(radix_tree_delete_item);
1568
1569/**
1570 * radix_tree_delete - delete an item from a radix tree
1571 * @root: radix tree root
1572 * @index: index key
1573 *
1574 * Remove the item at @index from the radix tree rooted at @root.
1575 *
1576 * Returns the address of the deleted item, or NULL if it was not present.
1577 */
1578void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1579{
1580 return radix_tree_delete_item(root, index, NULL);
1581}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582EXPORT_SYMBOL(radix_tree_delete);
1583
1584/**
1585 * radix_tree_tagged - test whether any items in the tree are tagged
1586 * @root: radix tree root
1587 * @tag: tag to test
1588 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001589int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590{
Nick Piggin612d6c12006-06-23 02:03:22 -07001591 return root_tag_get(root, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592}
1593EXPORT_SYMBOL(radix_tree_tagged);
1594
1595static void
Johannes Weiner449dd692014-04-03 14:47:56 -07001596radix_tree_node_ctor(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
Johannes Weiner449dd692014-04-03 14:47:56 -07001598 struct radix_tree_node *node = arg;
1599
1600 memset(node, 0, sizeof(*node));
1601 INIT_LIST_HEAD(&node->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602}
1603
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604static int radix_tree_callback(struct notifier_block *nfb,
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001605 unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606{
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001607 int cpu = (long)hcpu;
1608 struct radix_tree_preload *rtp;
1609 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001611 /* Free per-cpu pool of preloaded nodes */
1612 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1613 rtp = &per_cpu(radix_tree_preloads, cpu);
1614 while (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -07001615 node = rtp->nodes;
1616 rtp->nodes = node->private_data;
1617 kmem_cache_free(radix_tree_node_cachep, node);
1618 rtp->nr--;
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001619 }
1620 }
1621 return NOTIFY_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624void __init radix_tree_init(void)
1625{
1626 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1627 sizeof(struct radix_tree_node), 0,
Christoph Lameter488514d2008-04-28 02:12:05 -07001628 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1629 radix_tree_node_ctor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 hotcpu_notifier(radix_tree_callback, 0);
1631}