blob: 8e6d552c40ddfc7fa745a8a8acf34ee17f02bc3d [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 Torvalds2b412262016-12-09 10:41:42 -080033#include <linux/notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#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
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -070041/* Number of nodes in fully populated tree of given height */
42static unsigned long height_to_maxnodes[RADIX_TREE_MAX_PATH + 1] __read_mostly;
43
Jeff Moyer26fb1582007-10-16 01:24:49 -070044/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 * Radix tree node cache.
46 */
Christoph Lametere18b8902006-12-06 20:33:20 -080047static struct kmem_cache *radix_tree_node_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49/*
Nick Piggin55368052012-05-29 15:07:34 -070050 * The radix tree is variable-height, so an insert operation not only has
51 * to build the branch to its corresponding item, it also has to build the
52 * branch to existing items if the size has to be increased (by
53 * radix_tree_extend).
54 *
55 * The worst case is a zero height tree with just a single item at index 0,
56 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
57 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
58 * Hence:
59 */
60#define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
61
62/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 * Per-cpu pool of preloaded nodes
64 */
65struct radix_tree_preload {
Matthew Wilcox2fcd9002016-05-20 17:03:04 -070066 unsigned nr;
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -070067 /* nodes->private_data points to next preallocated node */
68 struct radix_tree_node *nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069};
Harvey Harrison8cef7d52009-01-06 14:40:50 -080070static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -070072static inline void *node_to_entry(void *ptr)
Nick Piggin27d20fd2010-11-11 14:05:19 -080073{
Matthew Wilcox30ff46cc2016-05-20 17:03:22 -070074 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
Nick Piggin27d20fd2010-11-11 14:05:19 -080075}
76
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -070077#define RADIX_TREE_RETRY node_to_entry(NULL)
Matthew Wilcoxafe0e392016-05-20 17:02:17 -070078
Matthew Wilcoxdb050f22016-05-20 17:01:57 -070079#ifdef CONFIG_RADIX_TREE_MULTIORDER
80/* Sibling slots point directly to another slot in the same node */
81static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
82{
83 void **ptr = node;
84 return (parent->slots <= ptr) &&
85 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
86}
87#else
88static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
89{
90 return false;
91}
92#endif
93
94static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
95 void **slot)
96{
97 return slot - parent->slots;
98}
99
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700100static unsigned int radix_tree_descend(struct radix_tree_node *parent,
101 struct radix_tree_node **nodep, unsigned long index)
Matthew Wilcoxdb050f22016-05-20 17:01:57 -0700102{
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700103 unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK;
Matthew Wilcoxdb050f22016-05-20 17:01:57 -0700104 void **entry = rcu_dereference_raw(parent->slots[offset]);
105
106#ifdef CONFIG_RADIX_TREE_MULTIORDER
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700107 if (radix_tree_is_internal_node(entry)) {
Linus Torvalds8d2c0d32016-09-25 13:32:46 -0700108 if (is_sibling_entry(parent, entry)) {
109 void **sibentry = (void **) entry_to_node(entry);
110 offset = get_slot_offset(parent, sibentry);
111 entry = rcu_dereference_raw(*sibentry);
Matthew Wilcoxdb050f22016-05-20 17:01:57 -0700112 }
113 }
114#endif
115
116 *nodep = (void *)entry;
117 return offset;
118}
119
Nick Piggin612d6c12006-06-23 02:03:22 -0700120static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
121{
122 return root->gfp_mask & __GFP_BITS_MASK;
123}
124
Nick Piggin643b52b2008-06-12 15:21:52 -0700125static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
126 int offset)
127{
128 __set_bit(offset, node->tags[tag]);
129}
130
131static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
132 int offset)
133{
134 __clear_bit(offset, node->tags[tag]);
135}
136
137static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
138 int offset)
139{
140 return test_bit(offset, node->tags[tag]);
141}
142
143static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
144{
145 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
146}
147
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700148static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
Nick Piggin643b52b2008-06-12 15:21:52 -0700149{
150 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
151}
152
153static inline void root_tag_clear_all(struct radix_tree_root *root)
154{
155 root->gfp_mask &= __GFP_BITS_MASK;
156}
157
158static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
159{
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700160 return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
Nick Piggin643b52b2008-06-12 15:21:52 -0700161}
162
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700163static inline unsigned root_tags_get(struct radix_tree_root *root)
164{
165 return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT;
166}
167
Nick Piggin643b52b2008-06-12 15:21:52 -0700168/*
169 * Returns 1 if any slot in the node has this tag set.
170 * Otherwise returns 0.
171 */
172static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
173{
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700174 unsigned idx;
Nick Piggin643b52b2008-06-12 15:21:52 -0700175 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
176 if (node->tags[tag][idx])
177 return 1;
178 }
179 return 0;
180}
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700181
182/**
183 * radix_tree_find_next_bit - find the next set bit in a memory region
184 *
185 * @addr: The address to base the search on
186 * @size: The bitmap size in bits
187 * @offset: The bitnumber to start searching at
188 *
189 * Unrollable variant of find_next_bit() for constant size arrays.
190 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
191 * Returns next bit offset, or size if nothing found.
192 */
193static __always_inline unsigned long
194radix_tree_find_next_bit(const unsigned long *addr,
195 unsigned long size, unsigned long offset)
196{
197 if (!__builtin_constant_p(size))
198 return find_next_bit(addr, size, offset);
199
200 if (offset < size) {
201 unsigned long tmp;
202
203 addr += offset / BITS_PER_LONG;
204 tmp = *addr >> (offset % BITS_PER_LONG);
205 if (tmp)
206 return __ffs(tmp) + offset;
207 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
208 while (offset < size) {
209 tmp = *++addr;
210 if (tmp)
211 return __ffs(tmp) + offset;
212 offset += BITS_PER_LONG;
213 }
214 }
215 return size;
216}
217
Ross Zwisler0796c582016-05-20 17:02:55 -0700218#ifndef __KERNEL__
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700219static void dump_node(struct radix_tree_node *node, unsigned long index)
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700220{
Ross Zwisler0796c582016-05-20 17:02:55 -0700221 unsigned long i;
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700222
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700223 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 -0700224 node, node->offset,
Ross Zwisler0796c582016-05-20 17:02:55 -0700225 node->tags[0][0], node->tags[1][0], node->tags[2][0],
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700226 node->shift, node->count, node->parent);
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700227
Ross Zwisler0796c582016-05-20 17:02:55 -0700228 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700229 unsigned long first = index | (i << node->shift);
230 unsigned long last = first | ((1UL << node->shift) - 1);
Ross Zwisler0796c582016-05-20 17:02:55 -0700231 void *entry = node->slots[i];
232 if (!entry)
233 continue;
234 if (is_sibling_entry(node, entry)) {
235 pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n",
236 entry, i,
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700237 *(void **)entry_to_node(entry),
Ross Zwisler0796c582016-05-20 17:02:55 -0700238 first, last);
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700239 } else if (!radix_tree_is_internal_node(entry)) {
Ross Zwisler0796c582016-05-20 17:02:55 -0700240 pr_debug("radix entry %p offset %ld indices %ld-%ld\n",
241 entry, i, first, last);
242 } else {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700243 dump_node(entry_to_node(entry), first);
Ross Zwisler0796c582016-05-20 17:02:55 -0700244 }
245 }
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700246}
247
248/* For debug */
249static void radix_tree_dump(struct radix_tree_root *root)
250{
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700251 pr_debug("radix root: %p rnode %p tags %x\n",
252 root, root->rnode,
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700253 root->gfp_mask >> __GFP_BITS_SHIFT);
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700254 if (!radix_tree_is_internal_node(root->rnode))
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700255 return;
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700256 dump_node(entry_to_node(root->rnode), 0);
Matthew Wilcox7cf19af2016-03-17 14:21:57 -0700257}
258#endif
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260/*
261 * This assumes that the caller has performed appropriate preallocation, and
262 * that the caller has pinned this thread of control to the current CPU.
263 */
264static struct radix_tree_node *
265radix_tree_node_alloc(struct radix_tree_root *root)
266{
Nick Piggine2848a02008-02-04 22:29:10 -0800267 struct radix_tree_node *ret = NULL;
Nick Piggin612d6c12006-06-23 02:03:22 -0700268 gfp_t gfp_mask = root_gfp_mask(root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Jan Kara5e4c0d972013-09-11 14:26:05 -0700270 /*
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700271 * Preload code isn't irq safe and it doesn't make sense to use
272 * preloading during an interrupt anyway as all the allocations have
273 * to be atomic. So just do normal allocation when in interrupt.
Jan Kara5e4c0d972013-09-11 14:26:05 -0700274 */
Mel Gormand0164ad2015-11-06 16:28:21 -0800275 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 struct radix_tree_preload *rtp;
277
Nick Piggine2848a02008-02-04 22:29:10 -0800278 /*
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700279 * Even if the caller has preloaded, try to allocate from the
Vladimir Davydov05eb6e72016-08-02 14:03:01 -0700280 * cache first for the new node to get accounted to the memory
281 * cgroup.
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700282 */
283 ret = kmem_cache_alloc(radix_tree_node_cachep,
Vladimir Davydov05eb6e72016-08-02 14:03:01 -0700284 gfp_mask | __GFP_NOWARN);
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700285 if (ret)
286 goto out;
287
288 /*
Nick Piggine2848a02008-02-04 22:29:10 -0800289 * Provided the caller has preloaded here, we will always
290 * succeed in getting a node here (and never reach
291 * kmem_cache_alloc)
292 */
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700293 rtp = this_cpu_ptr(&radix_tree_preloads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 if (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700295 ret = rtp->nodes;
296 rtp->nodes = ret->private_data;
297 ret->private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 rtp->nr--;
299 }
Catalin Marinasce80b062014-06-06 14:38:18 -0700300 /*
301 * Update the allocation stack trace as this is more useful
302 * for debugging.
303 */
304 kmemleak_update_trace(ret);
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700305 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 }
Vladimir Davydov05eb6e72016-08-02 14:03:01 -0700307 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Vladimir Davydov58e698a2016-03-17 14:18:36 -0700308out:
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700309 BUG_ON(radix_tree_is_internal_node(ret));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 return ret;
311}
312
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800313static void radix_tree_node_rcu_free(struct rcu_head *head)
314{
315 struct radix_tree_node *node =
316 container_of(head, struct radix_tree_node, rcu_head);
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000317 int i;
Nick Piggin643b52b2008-06-12 15:21:52 -0700318
319 /*
320 * must only free zeroed nodes into the slab. radix_tree_shrink
321 * can leave us with a non-NULL entry in the first slot, so clear
322 * that here to make sure.
323 */
Dave Chinnerb6dd0862010-08-23 10:33:19 +1000324 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
325 tag_clear(node, i, 0);
326
Nick Piggin643b52b2008-06-12 15:21:52 -0700327 node->slots[0] = NULL;
328 node->count = 0;
329
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800330 kmem_cache_free(radix_tree_node_cachep, node);
331}
332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333static inline void
334radix_tree_node_free(struct radix_tree_node *node)
335{
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800336 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337}
338
339/*
340 * Load up this CPU's radix_tree_node buffer with sufficient objects to
341 * ensure that the addition of a single element in the tree cannot fail. On
342 * success, return zero, with preemption disabled. On error, return -ENOMEM
343 * with preemption not disabled.
David Howellsb34df792009-11-19 18:11:14 +0000344 *
345 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800346 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 */
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -0700348static int __radix_tree_preload(gfp_t gfp_mask, int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349{
350 struct radix_tree_preload *rtp;
351 struct radix_tree_node *node;
352 int ret = -ENOMEM;
353
Vladimir Davydov05eb6e72016-08-02 14:03:01 -0700354 /*
355 * Nodes preloaded by one cgroup can be be used by another cgroup, so
356 * they should never be accounted to any particular memory cgroup.
357 */
358 gfp_mask &= ~__GFP_ACCOUNT;
359
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700361 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -0700362 while (rtp->nr < nr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 preempt_enable();
Christoph Lameter488514d2008-04-28 02:12:05 -0700364 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (node == NULL)
366 goto out;
367 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -0700368 rtp = this_cpu_ptr(&radix_tree_preloads);
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -0700369 if (rtp->nr < nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700370 node->private_data = rtp->nodes;
371 rtp->nodes = node;
372 rtp->nr++;
373 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 kmem_cache_free(radix_tree_node_cachep, node);
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -0700375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 }
377 ret = 0;
378out:
379 return ret;
380}
Jan Kara5e4c0d972013-09-11 14:26:05 -0700381
382/*
383 * Load up this CPU's radix_tree_node buffer with sufficient objects to
384 * ensure that the addition of a single element in the tree cannot fail. On
385 * success, return zero, with preemption disabled. On error, return -ENOMEM
386 * with preemption not disabled.
387 *
388 * To make use of this facility, the radix tree must be initialised without
Mel Gormand0164ad2015-11-06 16:28:21 -0800389 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
Jan Kara5e4c0d972013-09-11 14:26:05 -0700390 */
391int radix_tree_preload(gfp_t gfp_mask)
392{
393 /* Warn on non-sensical use... */
Mel Gormand0164ad2015-11-06 16:28:21 -0800394 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -0700395 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
Jan Kara5e4c0d972013-09-11 14:26:05 -0700396}
David Chinnerd7f09232007-07-14 16:05:04 +1000397EXPORT_SYMBOL(radix_tree_preload);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Nick Piggin6e954b92006-01-08 01:01:40 -0800399/*
Jan Kara5e4c0d972013-09-11 14:26:05 -0700400 * The same as above function, except we don't guarantee preloading happens.
401 * We do it, if we decide it helps. On success, return zero with preemption
402 * disabled. On error, return -ENOMEM with preemption not disabled.
403 */
404int radix_tree_maybe_preload(gfp_t gfp_mask)
405{
Mel Gormand0164ad2015-11-06 16:28:21 -0800406 if (gfpflags_allow_blocking(gfp_mask))
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -0700407 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
Jan Kara5e4c0d972013-09-11 14:26:05 -0700408 /* Preloading doesn't help anything with this gfp mask, skip it */
409 preempt_disable();
410 return 0;
411}
412EXPORT_SYMBOL(radix_tree_maybe_preload);
413
414/*
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -0700415 * The same as function above, but preload number of nodes required to insert
416 * (1 << order) continuous naturally-aligned elements.
417 */
418int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order)
419{
420 unsigned long nr_subtrees;
421 int nr_nodes, subtree_height;
422
423 /* Preloading doesn't help anything with this gfp mask, skip it */
424 if (!gfpflags_allow_blocking(gfp_mask)) {
425 preempt_disable();
426 return 0;
427 }
428
429 /*
430 * Calculate number and height of fully populated subtrees it takes to
431 * store (1 << order) elements.
432 */
433 nr_subtrees = 1 << order;
434 for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE;
435 subtree_height++)
436 nr_subtrees >>= RADIX_TREE_MAP_SHIFT;
437
438 /*
439 * The worst case is zero height tree with a single item at index 0 and
440 * then inserting items starting at ULONG_MAX - (1 << order).
441 *
442 * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to
443 * 0-index item.
444 */
445 nr_nodes = RADIX_TREE_MAX_PATH;
446
447 /* Plus branch to fully populated subtrees. */
448 nr_nodes += RADIX_TREE_MAX_PATH - subtree_height;
449
450 /* Root node is shared. */
451 nr_nodes--;
452
453 /* Plus nodes required to build subtrees. */
454 nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height];
455
456 return __radix_tree_preload(gfp_mask, nr_nodes);
457}
458
459/*
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700460 * The maximum index which can be stored in a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 */
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700462static inline unsigned long shift_maxindex(unsigned int shift)
463{
464 return (RADIX_TREE_MAP_SIZE << shift) - 1;
465}
466
Matthew Wilcox1456a432016-05-20 17:02:08 -0700467static inline unsigned long node_maxindex(struct radix_tree_node *node)
468{
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700469 return shift_maxindex(node->shift);
Matthew Wilcox1456a432016-05-20 17:02:08 -0700470}
471
472static unsigned radix_tree_load_root(struct radix_tree_root *root,
473 struct radix_tree_node **nodep, unsigned long *maxindex)
474{
475 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
476
477 *nodep = node;
478
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700479 if (likely(radix_tree_is_internal_node(node))) {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700480 node = entry_to_node(node);
Matthew Wilcox1456a432016-05-20 17:02:08 -0700481 *maxindex = node_maxindex(node);
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700482 return node->shift + RADIX_TREE_MAP_SHIFT;
Matthew Wilcox1456a432016-05-20 17:02:08 -0700483 }
484
485 *maxindex = 0;
486 return 0;
487}
488
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489/*
490 * Extend a radix tree so it can store key @index.
491 */
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700492static int radix_tree_extend(struct radix_tree_root *root,
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700493 unsigned long index, unsigned int shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800495 struct radix_tree_node *slot;
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700496 unsigned int maxshift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 int tag;
498
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700499 /* Figure out what the shift should be. */
500 maxshift = shift;
501 while (index > shift_maxindex(maxshift))
502 maxshift += RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700504 slot = root->rnode;
505 if (!slot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 do {
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700509 struct radix_tree_node *node = radix_tree_node_alloc(root);
510
511 if (!node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return -ENOMEM;
513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 /* Propagate the aggregated tag info into the new root */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800515 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
Nick Piggin612d6c12006-06-23 02:03:22 -0700516 if (root_tag_get(root, tag))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 tag_set(node, tag, 0);
518 }
519
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700520 BUG_ON(shift > BITS_PER_LONG);
521 node->shift = shift;
Matthew Wilcox0c7fa0a2016-05-20 17:03:07 -0700522 node->offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 node->count = 1;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800524 node->parent = NULL;
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700525 if (radix_tree_is_internal_node(slot))
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700526 entry_to_node(slot)->parent = node;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800527 node->slots[0] = slot;
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -0700528 slot = node_to_entry(node);
529 rcu_assign_pointer(root->rnode, slot);
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700530 shift += RADIX_TREE_MAP_SHIFT;
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700531 } while (shift <= maxshift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532out:
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700533 return maxshift + RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534}
535
536/**
Johannes Weiner139e5612014-04-03 14:47:54 -0700537 * __radix_tree_create - create a slot in a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 * @root: radix tree root
539 * @index: index key
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700540 * @order: index occupies 2^order aligned slots
Johannes Weiner139e5612014-04-03 14:47:54 -0700541 * @nodep: returns node
542 * @slotp: returns slot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 *
Johannes Weiner139e5612014-04-03 14:47:54 -0700544 * Create, if necessary, and return the node and slot for an item
545 * at position @index in the radix tree @root.
546 *
547 * Until there is more than one item in the tree, no nodes are
548 * allocated and @root->rnode is used as a direct slot instead of
549 * pointing to a node, in which case *@nodep will be NULL.
550 *
551 * Returns -ENOMEM, or 0 for success.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700553int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700554 unsigned order, struct radix_tree_node **nodep,
555 void ***slotp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700557 struct radix_tree_node *node = NULL, *child;
558 void **slot = (void **)&root->rnode;
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700559 unsigned long maxindex;
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700560 unsigned int shift, offset = 0;
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700561 unsigned long max = index | ((1UL << order) - 1);
562
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700563 shift = radix_tree_load_root(root, &child, &maxindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
565 /* Make sure the tree is high enough. */
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700566 if (max > maxindex) {
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700567 int error = radix_tree_extend(root, max, shift);
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700568 if (error < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 return error;
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700570 shift = error;
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700571 child = root->rnode;
Matthew Wilcoxd0891262016-05-20 17:03:19 -0700572 if (order == shift)
Matthew Wilcox49ea6eb2016-05-20 17:02:11 -0700573 shift += RADIX_TREE_MAP_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 }
575
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700576 while (shift > order) {
Matthew Wilcoxc12e51b2016-05-20 17:03:10 -0700577 shift -= RADIX_TREE_MAP_SHIFT;
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700578 if (child == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 /* Have to add a child node. */
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700580 child = radix_tree_node_alloc(root);
581 if (!child)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 return -ENOMEM;
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700583 child->shift = shift;
584 child->offset = offset;
585 child->parent = node;
586 rcu_assign_pointer(*slot, node_to_entry(child));
587 if (node)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 node->count++;
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700589 } else if (!radix_tree_is_internal_node(child))
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700590 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 /* Go a level down */
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700593 node = entry_to_node(child);
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700594 offset = radix_tree_descend(node, &child, index);
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700595 slot = &node->slots[offset];
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700596 }
597
Matthew Wilcox57578c22016-05-20 17:01:54 -0700598#ifdef CONFIG_RADIX_TREE_MULTIORDER
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700599 /* Insert pointers to the canonical entry */
Matthew Wilcox3b8c00f2016-05-20 17:01:59 -0700600 if (order > shift) {
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700601 unsigned i, n = 1 << (order - shift);
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700602 offset = offset & ~(n - 1);
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700603 slot = &node->slots[offset];
604 child = node_to_entry(slot);
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700605 for (i = 0; i < n; i++) {
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700606 if (slot[i])
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700607 return -EEXIST;
608 }
609
610 for (i = 1; i < n; i++) {
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700611 rcu_assign_pointer(slot[i], child);
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700612 node->count++;
613 }
Nick Piggin612d6c12006-06-23 02:03:22 -0700614 }
Matthew Wilcox57578c22016-05-20 17:01:54 -0700615#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Johannes Weiner139e5612014-04-03 14:47:54 -0700617 if (nodep)
618 *nodep = node;
619 if (slotp)
Matthew Wilcox89148aa2016-05-20 17:03:42 -0700620 *slotp = slot;
Johannes Weiner139e5612014-04-03 14:47:54 -0700621 return 0;
622}
623
624/**
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700625 * __radix_tree_insert - insert into a radix tree
Johannes Weiner139e5612014-04-03 14:47:54 -0700626 * @root: radix tree root
627 * @index: index key
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700628 * @order: key covers the 2^order indices around index
Johannes Weiner139e5612014-04-03 14:47:54 -0700629 * @item: item to insert
630 *
631 * Insert an item into the radix tree at position @index.
632 */
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700633int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
634 unsigned order, void *item)
Johannes Weiner139e5612014-04-03 14:47:54 -0700635{
636 struct radix_tree_node *node;
637 void **slot;
638 int error;
639
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700640 BUG_ON(radix_tree_is_internal_node(item));
Johannes Weiner139e5612014-04-03 14:47:54 -0700641
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700642 error = __radix_tree_create(root, index, order, &node, &slot);
Johannes Weiner139e5612014-04-03 14:47:54 -0700643 if (error)
644 return error;
645 if (*slot != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 return -EEXIST;
Johannes Weiner139e5612014-04-03 14:47:54 -0700647 rcu_assign_pointer(*slot, item);
Christoph Lameter201b6262005-09-06 15:16:46 -0700648
Nick Piggin612d6c12006-06-23 02:03:22 -0700649 if (node) {
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700650 unsigned offset = get_slot_offset(node, slot);
Nick Piggin612d6c12006-06-23 02:03:22 -0700651 node->count++;
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700652 BUG_ON(tag_get(node, 0, offset));
653 BUG_ON(tag_get(node, 1, offset));
654 BUG_ON(tag_get(node, 2, offset));
Nick Piggin612d6c12006-06-23 02:03:22 -0700655 } else {
Matthew Wilcox7b60e9a2016-05-20 17:02:23 -0700656 BUG_ON(root_tags_get(root));
Nick Piggin612d6c12006-06-23 02:03:22 -0700657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 return 0;
660}
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700661EXPORT_SYMBOL(__radix_tree_insert);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Johannes Weiner139e5612014-04-03 14:47:54 -0700663/**
664 * __radix_tree_lookup - lookup an item in a radix tree
665 * @root: radix tree root
666 * @index: index key
667 * @nodep: returns node
668 * @slotp: returns slot
669 *
670 * Lookup and return the item at position @index in the radix
671 * tree @root.
672 *
673 * Until there is more than one item in the tree, no nodes are
674 * allocated and @root->rnode is used as a direct slot instead of
675 * pointing to a node, in which case *@nodep will be NULL.
Hans Reisera4331362005-11-07 00:59:29 -0800676 */
Johannes Weiner139e5612014-04-03 14:47:54 -0700677void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
678 struct radix_tree_node **nodep, void ***slotp)
Hans Reisera4331362005-11-07 00:59:29 -0800679{
Johannes Weiner139e5612014-04-03 14:47:54 -0700680 struct radix_tree_node *node, *parent;
Matthew Wilcox85829952016-05-20 17:02:20 -0700681 unsigned long maxindex;
Johannes Weiner139e5612014-04-03 14:47:54 -0700682 void **slot;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800683
Matthew Wilcox85829952016-05-20 17:02:20 -0700684 restart:
685 parent = NULL;
686 slot = (void **)&root->rnode;
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700687 radix_tree_load_root(root, &node, &maxindex);
Matthew Wilcox85829952016-05-20 17:02:20 -0700688 if (index > maxindex)
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800689 return NULL;
690
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700691 while (radix_tree_is_internal_node(node)) {
Matthew Wilcox85829952016-05-20 17:02:20 -0700692 unsigned offset;
Johannes Weiner139e5612014-04-03 14:47:54 -0700693
Matthew Wilcox85829952016-05-20 17:02:20 -0700694 if (node == RADIX_TREE_RETRY)
695 goto restart;
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700696 parent = entry_to_node(node);
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700697 offset = radix_tree_descend(parent, &node, index);
Matthew Wilcox85829952016-05-20 17:02:20 -0700698 slot = parent->slots + offset;
699 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800700
Johannes Weiner139e5612014-04-03 14:47:54 -0700701 if (nodep)
702 *nodep = parent;
703 if (slotp)
704 *slotp = slot;
705 return node;
Huang Shijieb72b71c2009-06-16 15:33:42 -0700706}
707
708/**
709 * radix_tree_lookup_slot - lookup a slot in a radix tree
710 * @root: radix tree root
711 * @index: index key
712 *
713 * Returns: the slot corresponding to the position @index in the
714 * radix tree @root. This is useful for update-if-exists operations.
715 *
716 * This function can be called under rcu_read_lock iff the slot is not
717 * modified by radix_tree_replace_slot, otherwise it must be called
718 * exclusive from other writers. Any dereference of the slot must be done
719 * using radix_tree_deref_slot.
720 */
721void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
722{
Johannes Weiner139e5612014-04-03 14:47:54 -0700723 void **slot;
724
725 if (!__radix_tree_lookup(root, index, NULL, &slot))
726 return NULL;
727 return slot;
Hans Reisera4331362005-11-07 00:59:29 -0800728}
729EXPORT_SYMBOL(radix_tree_lookup_slot);
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731/**
732 * radix_tree_lookup - perform lookup operation on a radix tree
733 * @root: radix tree root
734 * @index: index key
735 *
736 * Lookup the item at the position @index in the radix tree @root.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800737 *
738 * This function can be called under rcu_read_lock, however the caller
739 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
740 * them safely). No RCU barriers are required to access or modify the
741 * returned item, however.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 */
743void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
744{
Johannes Weiner139e5612014-04-03 14:47:54 -0700745 return __radix_tree_lookup(root, index, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746}
747EXPORT_SYMBOL(radix_tree_lookup);
748
749/**
750 * radix_tree_tag_set - set a tag on a radix tree node
751 * @root: radix tree root
752 * @index: index key
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700753 * @tag: tag index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800755 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
756 * corresponding to @index in the radix tree. From
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 * the root all the way down to the leaf node.
758 *
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700759 * Returns the address of the tagged item. Setting a tag on a not-present
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 * item is a bug.
761 */
762void *radix_tree_tag_set(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800763 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764{
Ross Zwislerfb969902016-05-20 17:02:32 -0700765 struct radix_tree_node *node, *parent;
766 unsigned long maxindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700768 radix_tree_load_root(root, &node, &maxindex);
Ross Zwislerfb969902016-05-20 17:02:32 -0700769 BUG_ON(index > maxindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700771 while (radix_tree_is_internal_node(node)) {
Ross Zwislerfb969902016-05-20 17:02:32 -0700772 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700774 parent = entry_to_node(node);
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700775 offset = radix_tree_descend(parent, &node, index);
Ross Zwislerfb969902016-05-20 17:02:32 -0700776 BUG_ON(!node);
777
778 if (!tag_get(parent, tag, offset))
779 tag_set(parent, tag, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 }
781
Nick Piggin612d6c12006-06-23 02:03:22 -0700782 /* set the root's tag bit */
Ross Zwislerfb969902016-05-20 17:02:32 -0700783 if (!root_tag_get(root, tag))
Nick Piggin612d6c12006-06-23 02:03:22 -0700784 root_tag_set(root, tag);
785
Ross Zwislerfb969902016-05-20 17:02:32 -0700786 return node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787}
788EXPORT_SYMBOL(radix_tree_tag_set);
789
Matthew Wilcoxd604c322016-05-20 17:03:45 -0700790static void node_tag_clear(struct radix_tree_root *root,
791 struct radix_tree_node *node,
792 unsigned int tag, unsigned int offset)
793{
794 while (node) {
795 if (!tag_get(node, tag, offset))
796 return;
797 tag_clear(node, tag, offset);
798 if (any_tag_set(node, tag))
799 return;
800
801 offset = node->offset;
802 node = node->parent;
803 }
804
805 /* clear the root's tag bit */
806 if (root_tag_get(root, tag))
807 root_tag_clear(root, tag);
808}
809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810/**
811 * radix_tree_tag_clear - clear a tag on a radix tree node
812 * @root: radix tree root
813 * @index: index key
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700814 * @tag: tag index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 *
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800816 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700817 * corresponding to @index in the radix tree. If this causes
818 * the leaf node to have no tags set then clear the tag in the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 * next-to-leaf node, etc.
820 *
821 * Returns the address of the tagged item on success, else NULL. ie:
822 * has the same return value and semantics as radix_tree_lookup().
823 */
824void *radix_tree_tag_clear(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800825 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826{
Ross Zwisler00f47b52016-05-20 17:02:35 -0700827 struct radix_tree_node *node, *parent;
828 unsigned long maxindex;
Hugh Dickinse2bdb932012-01-12 17:20:41 -0800829 int uninitialized_var(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700831 radix_tree_load_root(root, &node, &maxindex);
Ross Zwisler00f47b52016-05-20 17:02:35 -0700832 if (index > maxindex)
833 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Ross Zwisler00f47b52016-05-20 17:02:35 -0700835 parent = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700837 while (radix_tree_is_internal_node(node)) {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700838 parent = entry_to_node(node);
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700839 offset = radix_tree_descend(parent, &node, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
841
Matthew Wilcoxd604c322016-05-20 17:03:45 -0700842 if (node)
843 node_tag_clear(root, parent, tag, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Ross Zwisler00f47b52016-05-20 17:02:35 -0700845 return node;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846}
847EXPORT_SYMBOL(radix_tree_tag_clear);
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849/**
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700850 * radix_tree_tag_get - get a tag on a radix tree node
851 * @root: radix tree root
852 * @index: index key
Matthew Wilcox2fcd9002016-05-20 17:03:04 -0700853 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 *
Marcelo Tosatti32605a12005-09-06 15:16:48 -0700855 * Return values:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 *
Nick Piggin612d6c12006-06-23 02:03:22 -0700857 * 0: tag not present or not set
858 * 1: tag set
David Howellsce826532010-04-06 22:36:20 +0100859 *
860 * Note that the return value of this function may not be relied on, even if
861 * the RCU lock is held, unless tag modification and node deletion are excluded
862 * from concurrency.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 */
864int radix_tree_tag_get(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800865 unsigned long index, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
Ross Zwisler4589ba62016-05-20 17:02:38 -0700867 struct radix_tree_node *node, *parent;
868 unsigned long maxindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
Nick Piggin612d6c12006-06-23 02:03:22 -0700870 if (!root_tag_get(root, tag))
871 return 0;
872
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700873 radix_tree_load_root(root, &node, &maxindex);
Ross Zwisler4589ba62016-05-20 17:02:38 -0700874 if (index > maxindex)
875 return 0;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800876 if (node == NULL)
877 return 0;
878
Matthew Wilcoxb194d162016-05-20 17:03:30 -0700879 while (radix_tree_is_internal_node(node)) {
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700880 unsigned offset;
Ross Zwisler4589ba62016-05-20 17:02:38 -0700881
Matthew Wilcox4dd6c092016-05-20 17:03:27 -0700882 parent = entry_to_node(node);
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700883 offset = radix_tree_descend(parent, &node, index);
Ross Zwisler4589ba62016-05-20 17:02:38 -0700884
885 if (!node)
886 return 0;
887 if (!tag_get(parent, tag, offset))
888 return 0;
889 if (node == RADIX_TREE_RETRY)
890 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 }
Ross Zwisler4589ba62016-05-20 17:02:38 -0700892
893 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
895EXPORT_SYMBOL(radix_tree_tag_get);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
Ross Zwisler21ef5332016-05-20 17:02:26 -0700897static inline void __set_iter_shift(struct radix_tree_iter *iter,
898 unsigned int shift)
899{
900#ifdef CONFIG_RADIX_TREE_MULTIORDER
901 iter->shift = shift;
902#endif
903}
904
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700905/**
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700906 * radix_tree_next_chunk - find next chunk of slots for iteration
907 *
908 * @root: radix tree root
909 * @iter: iterator state
910 * @flags: RADIX_TREE_ITER_* flags and tag index
911 * Returns: pointer to chunk first slot, or NULL if iteration is over
912 */
913void **radix_tree_next_chunk(struct radix_tree_root *root,
914 struct radix_tree_iter *iter, unsigned flags)
915{
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700916 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700917 struct radix_tree_node *node, *child;
Ross Zwisler21ef5332016-05-20 17:02:26 -0700918 unsigned long index, offset, maxindex;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700919
920 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
921 return NULL;
922
923 /*
924 * Catch next_index overflow after ~0UL. iter->index never overflows
925 * during iterating; it can be zero only at the beginning.
926 * And we cannot overflow iter->next_index in a single step,
927 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
Konstantin Khlebnikovfffaee32012-06-05 21:36:33 +0400928 *
929 * This condition also used by radix_tree_next_slot() to stop
930 * contiguous iterating, and forbid swithing to the next chunk.
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700931 */
932 index = iter->next_index;
933 if (!index && iter->index)
934 return NULL;
935
Ross Zwisler21ef5332016-05-20 17:02:26 -0700936 restart:
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700937 radix_tree_load_root(root, &child, &maxindex);
Ross Zwisler21ef5332016-05-20 17:02:26 -0700938 if (index > maxindex)
939 return NULL;
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700940 if (!child)
941 return NULL;
Ross Zwisler21ef5332016-05-20 17:02:26 -0700942
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700943 if (!radix_tree_is_internal_node(child)) {
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700944 /* Single-slot tree */
Ross Zwisler21ef5332016-05-20 17:02:26 -0700945 iter->index = index;
946 iter->next_index = maxindex + 1;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700947 iter->tags = 1;
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700948 __set_iter_shift(iter, 0);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700949 return (void **)&root->rnode;
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700950 }
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700951
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700952 do {
953 node = entry_to_node(child);
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700954 offset = radix_tree_descend(node, &child, index);
Ross Zwisler21ef5332016-05-20 17:02:26 -0700955
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700956 if ((flags & RADIX_TREE_ITER_TAGGED) ?
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700957 !tag_get(node, tag, offset) : !child) {
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700958 /* Hole detected */
959 if (flags & RADIX_TREE_ITER_CONTIG)
960 return NULL;
961
962 if (flags & RADIX_TREE_ITER_TAGGED)
963 offset = radix_tree_find_next_bit(
964 node->tags[tag],
965 RADIX_TREE_MAP_SIZE,
966 offset + 1);
967 else
968 while (++offset < RADIX_TREE_MAP_SIZE) {
Ross Zwisler21ef5332016-05-20 17:02:26 -0700969 void *slot = node->slots[offset];
970 if (is_sibling_entry(node, slot))
971 continue;
972 if (slot)
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700973 break;
974 }
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700975 index &= ~node_maxindex(node);
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700976 index += offset << node->shift;
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700977 /* Overflow after ~0UL */
978 if (!index)
979 return NULL;
980 if (offset == RADIX_TREE_MAP_SIZE)
981 goto restart;
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700982 child = rcu_dereference_raw(node->slots[offset]);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700983 }
984
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700985 if ((child == NULL) || (child == RADIX_TREE_RETRY))
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700986 goto restart;
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700987 } while (radix_tree_is_internal_node(child));
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700988
989 /* Update the iterator state */
Matthew Wilcox8c1244d2016-05-20 17:03:36 -0700990 iter->index = (index &~ node_maxindex(node)) | (offset << node->shift);
991 iter->next_index = (index | node_maxindex(node)) + 1;
Matthew Wilcox9e85d812016-05-20 17:03:48 -0700992 __set_iter_shift(iter, node->shift);
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700993
994 /* Construct iter->tags bit-mask from node->tags[tag] array */
995 if (flags & RADIX_TREE_ITER_TAGGED) {
996 unsigned tag_long, tag_bit;
997
998 tag_long = offset / BITS_PER_LONG;
999 tag_bit = offset % BITS_PER_LONG;
1000 iter->tags = node->tags[tag][tag_long] >> tag_bit;
1001 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1002 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
1003 /* Pick tags from next element */
1004 if (tag_bit)
1005 iter->tags |= node->tags[tag][tag_long + 1] <<
1006 (BITS_PER_LONG - tag_bit);
1007 /* Clip chunk size, here only BITS_PER_LONG tags */
1008 iter->next_index = index + BITS_PER_LONG;
1009 }
1010 }
1011
1012 return node->slots + offset;
1013}
1014EXPORT_SYMBOL(radix_tree_next_chunk);
1015
1016/**
Jan Karaebf8aa42010-08-09 17:19:11 -07001017 * radix_tree_range_tag_if_tagged - for each item in given range set given
1018 * tag if item has another tag set
1019 * @root: radix tree root
1020 * @first_indexp: pointer to a starting index of a range to scan
1021 * @last_index: last index of a range to scan
1022 * @nr_to_tag: maximum number items to tag
1023 * @iftag: tag index to test
1024 * @settag: tag index to set if tested tag is set
1025 *
1026 * This function scans range of radix tree from first_index to last_index
1027 * (inclusive). For each item in the range if iftag is set, the function sets
1028 * also settag. The function stops either after tagging nr_to_tag items or
1029 * after reaching last_index.
1030 *
Dave Chinner144dcfc2010-08-23 10:33:53 +10001031 * The tags must be set from the leaf level only and propagated back up the
1032 * path to the root. We must do this so that we resolve the full path before
1033 * setting any tags on intermediate nodes. If we set tags as we descend, then
1034 * we can get to the leaf node and find that the index that has the iftag
1035 * set is outside the range we are scanning. This reults in dangling tags and
1036 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1037 *
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001038 * The function returns the number of leaves where the tag was set and sets
Jan Karaebf8aa42010-08-09 17:19:11 -07001039 * *first_indexp to the first unscanned index.
Jan Karad5ed3a42010-08-19 14:13:33 -07001040 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1041 * be prepared to handle that.
Jan Karaebf8aa42010-08-09 17:19:11 -07001042 */
1043unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1044 unsigned long *first_indexp, unsigned long last_index,
1045 unsigned long nr_to_tag,
1046 unsigned int iftag, unsigned int settag)
1047{
Matthew Wilcoxa8e4da22016-05-20 17:03:39 -07001048 struct radix_tree_node *parent, *node, *child;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001049 unsigned long maxindex;
Dave Chinner144dcfc2010-08-23 10:33:53 +10001050 unsigned long tagged = 0;
1051 unsigned long index = *first_indexp;
Jan Karaebf8aa42010-08-09 17:19:11 -07001052
Matthew Wilcox9e85d812016-05-20 17:03:48 -07001053 radix_tree_load_root(root, &child, &maxindex);
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001054 last_index = min(last_index, maxindex);
Jan Karaebf8aa42010-08-09 17:19:11 -07001055 if (index > last_index)
1056 return 0;
1057 if (!nr_to_tag)
1058 return 0;
1059 if (!root_tag_get(root, iftag)) {
1060 *first_indexp = last_index + 1;
1061 return 0;
1062 }
Matthew Wilcoxa8e4da22016-05-20 17:03:39 -07001063 if (!radix_tree_is_internal_node(child)) {
Jan Karaebf8aa42010-08-09 17:19:11 -07001064 *first_indexp = last_index + 1;
1065 root_tag_set(root, settag);
1066 return 1;
1067 }
1068
Matthew Wilcoxa8e4da22016-05-20 17:03:39 -07001069 node = entry_to_node(child);
Jan Karaebf8aa42010-08-09 17:19:11 -07001070
1071 for (;;) {
Matthew Wilcox9e85d812016-05-20 17:03:48 -07001072 unsigned offset = radix_tree_descend(node, &child, index);
Matthew Wilcoxa8e4da22016-05-20 17:03:39 -07001073 if (!child)
Jan Karaebf8aa42010-08-09 17:19:11 -07001074 goto next;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001075 if (!tag_get(node, iftag, offset))
Jan Karaebf8aa42010-08-09 17:19:11 -07001076 goto next;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001077 /* Sibling slots never have tags set on them */
Matthew Wilcoxa8e4da22016-05-20 17:03:39 -07001078 if (radix_tree_is_internal_node(child)) {
1079 node = entry_to_node(child);
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001080 continue;
Jan Karaebf8aa42010-08-09 17:19:11 -07001081 }
Dave Chinner144dcfc2010-08-23 10:33:53 +10001082
1083 /* tag the leaf */
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001084 tagged++;
1085 tag_set(node, settag, offset);
Dave Chinner144dcfc2010-08-23 10:33:53 +10001086
1087 /* walk back up the path tagging interior nodes */
Matthew Wilcoxa8e4da22016-05-20 17:03:39 -07001088 parent = node;
1089 for (;;) {
1090 offset = parent->offset;
1091 parent = parent->parent;
1092 if (!parent)
Dave Chinner144dcfc2010-08-23 10:33:53 +10001093 break;
Matthew Wilcoxa8e4da22016-05-20 17:03:39 -07001094 /* stop if we find a node with the tag already set */
1095 if (tag_get(parent, settag, offset))
1096 break;
1097 tag_set(parent, settag, offset);
Dave Chinner144dcfc2010-08-23 10:33:53 +10001098 }
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001099 next:
Matthew Wilcox9e85d812016-05-20 17:03:48 -07001100 /* Go to next entry in node */
1101 index = ((index >> node->shift) + 1) << node->shift;
Jan Karad5ed3a42010-08-19 14:13:33 -07001102 /* Overflow can happen when last_index is ~0UL... */
1103 if (index > last_index || !index)
Jan Karaebf8aa42010-08-09 17:19:11 -07001104 break;
Matthew Wilcox9e85d812016-05-20 17:03:48 -07001105 offset = (index >> node->shift) & RADIX_TREE_MAP_MASK;
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001106 while (offset == 0) {
Jan Karaebf8aa42010-08-09 17:19:11 -07001107 /*
1108 * We've fully scanned this node. Go up. Because
1109 * last_index is guaranteed to be in the tree, what
1110 * we do below cannot wander astray.
1111 */
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001112 node = node->parent;
Matthew Wilcox9e85d812016-05-20 17:03:48 -07001113 offset = (index >> node->shift) & RADIX_TREE_MAP_MASK;
Jan Karaebf8aa42010-08-09 17:19:11 -07001114 }
Matthew Wilcox070c5ac2016-05-20 17:02:52 -07001115 if (is_sibling_entry(node, node->slots[offset]))
1116 goto next;
1117 if (tagged >= nr_to_tag)
1118 break;
Jan Karaebf8aa42010-08-09 17:19:11 -07001119 }
1120 /*
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -08001121 * We need not to tag the root tag if there is no tag which is set with
1122 * settag within the range from *first_indexp to last_index.
Jan Karaebf8aa42010-08-09 17:19:11 -07001123 */
Toshiyuki Okajimaac15ee62011-01-25 15:07:32 -08001124 if (tagged > 0)
1125 root_tag_set(root, settag);
Jan Karaebf8aa42010-08-09 17:19:11 -07001126 *first_indexp = index;
1127
1128 return tagged;
1129}
1130EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132/**
1133 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1134 * @root: radix tree root
1135 * @results: where the results of the lookup are placed
1136 * @first_index: start the lookup from this key
1137 * @max_items: place up to this many items at *results
1138 *
1139 * Performs an index-ascending scan of the tree for present items. Places
1140 * them at *@results and returns the number of items which were placed at
1141 * *@results.
1142 *
1143 * The implementation is naive.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001144 *
1145 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1146 * rcu_read_lock. In this case, rather than the returned results being
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001147 * an atomic snapshot of the tree at a single point in time, the
1148 * semantics of an RCU protected gang lookup are as though multiple
1149 * radix_tree_lookups have been issued in individual locks, and results
1150 * stored in 'results'.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 */
1152unsigned int
1153radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1154 unsigned long first_index, unsigned int max_items)
1155{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001156 struct radix_tree_iter iter;
1157 void **slot;
1158 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001160 if (unlikely(!max_items))
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001161 return 0;
1162
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001163 radix_tree_for_each_slot(slot, root, &iter, first_index) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001164 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001165 if (!results[ret])
1166 continue;
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001167 if (radix_tree_is_internal_node(results[ret])) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001168 slot = radix_tree_iter_retry(&iter);
1169 continue;
1170 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001171 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 return ret;
1176}
1177EXPORT_SYMBOL(radix_tree_gang_lookup);
1178
Nick Piggin47feff22008-07-25 19:45:29 -07001179/**
1180 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1181 * @root: radix tree root
1182 * @results: where the results of the lookup are placed
Hugh Dickins63286502011-08-03 16:21:18 -07001183 * @indices: where their indices should be placed (but usually NULL)
Nick Piggin47feff22008-07-25 19:45:29 -07001184 * @first_index: start the lookup from this key
1185 * @max_items: place up to this many items at *results
1186 *
1187 * Performs an index-ascending scan of the tree for present items. Places
1188 * their slots at *@results and returns the number of items which were
1189 * placed at *@results.
1190 *
1191 * The implementation is naive.
1192 *
1193 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1194 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1195 * protection, radix_tree_deref_slot may fail requiring a retry.
1196 */
1197unsigned int
Hugh Dickins63286502011-08-03 16:21:18 -07001198radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1199 void ***results, unsigned long *indices,
Nick Piggin47feff22008-07-25 19:45:29 -07001200 unsigned long first_index, unsigned int max_items)
1201{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001202 struct radix_tree_iter iter;
1203 void **slot;
1204 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001205
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001206 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001207 return 0;
1208
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001209 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1210 results[ret] = slot;
Hugh Dickins63286502011-08-03 16:21:18 -07001211 if (indices)
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001212 indices[ret] = iter.index;
1213 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001214 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001215 }
1216
1217 return ret;
1218}
1219EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221/**
1222 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1223 * based on a tag
1224 * @root: radix tree root
1225 * @results: where the results of the lookup are placed
1226 * @first_index: start the lookup from this key
1227 * @max_items: place up to this many items at *results
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001228 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 *
1230 * Performs an index-ascending scan of the tree for present items which
1231 * have the tag indexed by @tag set. Places the items at *@results and
1232 * returns the number of items which were placed at *@results.
1233 */
1234unsigned int
1235radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001236 unsigned long first_index, unsigned int max_items,
1237 unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001239 struct radix_tree_iter iter;
1240 void **slot;
1241 unsigned int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001243 if (unlikely(!max_items))
Nick Piggin612d6c12006-06-23 02:03:22 -07001244 return 0;
1245
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001246 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001247 results[ret] = rcu_dereference_raw(*slot);
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001248 if (!results[ret])
1249 continue;
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001250 if (radix_tree_is_internal_node(results[ret])) {
Matthew Wilcox46437f92016-02-02 16:57:52 -08001251 slot = radix_tree_iter_retry(&iter);
1252 continue;
1253 }
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001254 if (++ret == max_items)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 }
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 return ret;
1259}
1260EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1261
1262/**
Nick Piggin47feff22008-07-25 19:45:29 -07001263 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1264 * radix tree based on a tag
1265 * @root: radix tree root
1266 * @results: where the results of the lookup are placed
1267 * @first_index: start the lookup from this key
1268 * @max_items: place up to this many items at *results
1269 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1270 *
1271 * Performs an index-ascending scan of the tree for present items which
1272 * have the tag indexed by @tag set. Places the slots at *@results and
1273 * returns the number of slots which were placed at *@results.
1274 */
1275unsigned int
1276radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1277 unsigned long first_index, unsigned int max_items,
1278 unsigned int tag)
1279{
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001280 struct radix_tree_iter iter;
1281 void **slot;
1282 unsigned int ret = 0;
Nick Piggin47feff22008-07-25 19:45:29 -07001283
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001284 if (unlikely(!max_items))
Nick Piggin47feff22008-07-25 19:45:29 -07001285 return 0;
1286
Konstantin Khlebnikovcebbd292012-03-28 14:42:53 -07001287 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1288 results[ret] = slot;
1289 if (++ret == max_items)
Nick Piggin47feff22008-07-25 19:45:29 -07001290 break;
Nick Piggin47feff22008-07-25 19:45:29 -07001291 }
1292
1293 return ret;
1294}
1295EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1296
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001297#if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1298#include <linux/sched.h> /* for cond_resched() */
1299
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001300struct locate_info {
1301 unsigned long found_index;
1302 bool stop;
1303};
1304
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001305/*
1306 * This linear search is at present only useful to shmem_unuse_inode().
1307 */
1308static unsigned long __locate(struct radix_tree_node *slot, void *item,
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001309 unsigned long index, struct locate_info *info)
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001310{
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001311 unsigned long i;
1312
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001313 do {
Matthew Wilcox9e85d812016-05-20 17:03:48 -07001314 unsigned int shift = slot->shift;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001315
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001316 for (i = (index >> shift) & RADIX_TREE_MAP_MASK;
1317 i < RADIX_TREE_MAP_SIZE;
1318 i++, index += (1UL << shift)) {
1319 struct radix_tree_node *node =
1320 rcu_dereference_raw(slot->slots[i]);
1321 if (node == RADIX_TREE_RETRY)
1322 goto out;
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001323 if (!radix_tree_is_internal_node(node)) {
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001324 if (node == item) {
1325 info->found_index = index;
1326 info->stop = true;
1327 goto out;
1328 }
1329 continue;
1330 }
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001331 node = entry_to_node(node);
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001332 if (is_sibling_entry(slot, node))
1333 continue;
1334 slot = node;
1335 break;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001336 }
Matthew Wilcox9e85d812016-05-20 17:03:48 -07001337 } while (i < RADIX_TREE_MAP_SIZE);
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001338
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001339out:
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001340 if ((index == 0) && (i == RADIX_TREE_MAP_SIZE))
1341 info->stop = true;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001342 return index;
1343}
1344
1345/**
1346 * radix_tree_locate_item - search through radix tree for item
1347 * @root: radix tree root
1348 * @item: item to be found
1349 *
1350 * Returns index where item was found, or -1 if not found.
1351 * Caller must hold no lock (since this time-consuming function needs
1352 * to be preemptible), and must check afterwards if item is still there.
1353 */
1354unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1355{
1356 struct radix_tree_node *node;
1357 unsigned long max_index;
1358 unsigned long cur_index = 0;
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001359 struct locate_info info = {
1360 .found_index = -1,
1361 .stop = false,
1362 };
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001363
1364 do {
1365 rcu_read_lock();
1366 node = rcu_dereference_raw(root->rnode);
Matthew Wilcoxb194d162016-05-20 17:03:30 -07001367 if (!radix_tree_is_internal_node(node)) {
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001368 rcu_read_unlock();
1369 if (node == item)
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001370 info.found_index = 0;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001371 break;
1372 }
1373
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001374 node = entry_to_node(node);
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001375
1376 max_index = node_maxindex(node);
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001377 if (cur_index > max_index) {
1378 rcu_read_unlock();
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001379 break;
Hugh Dickins5f30fc92014-03-03 15:38:23 -08001380 }
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001381
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001382 cur_index = __locate(node, item, cur_index, &info);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001383 rcu_read_unlock();
1384 cond_resched();
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001385 } while (!info.stop && cur_index <= max_index);
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001386
Matthew Wilcox0a2efc62016-05-20 17:02:46 -07001387 return info.found_index;
Hugh Dickinse504f3f2011-08-03 16:21:27 -07001388}
1389#else
1390unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1391{
1392 return -1;
1393}
1394#endif /* CONFIG_SHMEM && CONFIG_SWAP */
Nick Piggin47feff22008-07-25 19:45:29 -07001395
1396/**
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001397 * radix_tree_shrink - shrink radix tree to minimum height
Nick Piggina5f51c92006-01-08 01:01:41 -08001398 * @root radix tree root
1399 */
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001400static inline bool radix_tree_shrink(struct radix_tree_root *root)
Nick Piggina5f51c92006-01-08 01:01:41 -08001401{
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001402 bool shrunk = false;
1403
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001404 for (;;) {
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001405 struct radix_tree_node *node = root->rnode;
1406 struct radix_tree_node *child;
Nick Piggina5f51c92006-01-08 01:01:41 -08001407
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001408 if (!radix_tree_is_internal_node(node))
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001409 break;
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001410 node = entry_to_node(node);
Nick Pigginc0bc9872007-10-16 01:24:42 -07001411
1412 /*
1413 * The candidate node has more than one child, or its child
Matthew Wilcoxd0891262016-05-20 17:03:19 -07001414 * is not at the leftmost slot, or the child is a multiorder
1415 * entry, we cannot shrink.
Nick Pigginc0bc9872007-10-16 01:24:42 -07001416 */
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001417 if (node->count != 1)
Nick Pigginc0bc9872007-10-16 01:24:42 -07001418 break;
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001419 child = node->slots[0];
1420 if (!child)
Nick Pigginc0bc9872007-10-16 01:24:42 -07001421 break;
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001422 if (!radix_tree_is_internal_node(child) && node->shift)
Matthew Wilcoxafe0e392016-05-20 17:02:17 -07001423 break;
1424
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001425 if (radix_tree_is_internal_node(child))
1426 entry_to_node(child)->parent = NULL;
Nick Pigginc0bc9872007-10-16 01:24:42 -07001427
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001428 /*
1429 * We don't need rcu_assign_pointer(), since we are simply
Nick Piggin27d20fd2010-11-11 14:05:19 -08001430 * moving the node from one part of the tree to another: if it
1431 * was safe to dereference the old pointer to it
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001432 * (node->slots[0]), it will be safe to dereference the new
Nick Piggin27d20fd2010-11-11 14:05:19 -08001433 * one (root->rnode) as far as dependent read barriers go.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08001434 */
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001435 root->rnode = child;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001436
1437 /*
1438 * We have a dilemma here. The node's slot[0] must not be
1439 * NULLed in case there are concurrent lookups expecting to
1440 * find the item. However if this was a bottom-level node,
1441 * then it may be subject to the slot pointer being visible
1442 * to callers dereferencing it. If item corresponding to
1443 * slot[0] is subsequently deleted, these callers would expect
1444 * their slot to become empty sooner or later.
1445 *
1446 * For example, lockless pagecache will look up a slot, deref
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001447 * the page pointer, and if the page has 0 refcount it means it
Nick Piggin27d20fd2010-11-11 14:05:19 -08001448 * was concurrently deleted from pagecache so try the deref
1449 * again. Fortunately there is already a requirement for logic
1450 * to retry the entire slot lookup -- the indirect pointer
1451 * problem (replacing direct root node with an indirect pointer
1452 * also results in a stale slot). So tag the slot as indirect
1453 * to force callers to retry.
1454 */
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001455 if (!radix_tree_is_internal_node(child))
1456 node->slots[0] = RADIX_TREE_RETRY;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001457
Matthew Wilcoxaf49a632016-05-20 17:03:33 -07001458 radix_tree_node_free(node);
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001459 shrunk = true;
Nick Piggina5f51c92006-01-08 01:01:41 -08001460 }
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001461
1462 return shrunk;
Nick Piggina5f51c92006-01-08 01:01:41 -08001463}
1464
1465/**
Johannes Weiner139e5612014-04-03 14:47:54 -07001466 * __radix_tree_delete_node - try to free node after clearing a slot
1467 * @root: radix tree root
Johannes Weiner139e5612014-04-03 14:47:54 -07001468 * @node: node containing @index
1469 *
1470 * After clearing the slot at @index in @node from radix tree
1471 * rooted at @root, call this function to attempt freeing the
1472 * node and shrinking the tree.
1473 *
1474 * Returns %true if @node was freed, %false otherwise.
1475 */
Johannes Weiner449dd692014-04-03 14:47:56 -07001476bool __radix_tree_delete_node(struct radix_tree_root *root,
Johannes Weiner139e5612014-04-03 14:47:54 -07001477 struct radix_tree_node *node)
1478{
1479 bool deleted = false;
1480
1481 do {
1482 struct radix_tree_node *parent;
1483
1484 if (node->count) {
Matthew Wilcox4dd6c092016-05-20 17:03:27 -07001485 if (node == entry_to_node(root->rnode))
Matthew Wilcoxfb209012016-05-20 17:03:13 -07001486 deleted |= radix_tree_shrink(root);
Johannes Weiner139e5612014-04-03 14:47:54 -07001487 return deleted;
1488 }
1489
1490 parent = node->parent;
1491 if (parent) {
Matthew Wilcox0c7fa0a2016-05-20 17:03:07 -07001492 parent->slots[node->offset] = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001493 parent->count--;
1494 } else {
1495 root_tag_clear_all(root);
Johannes Weiner139e5612014-04-03 14:47:54 -07001496 root->rnode = NULL;
1497 }
1498
1499 radix_tree_node_free(node);
1500 deleted = true;
1501
1502 node = parent;
1503 } while (node);
1504
1505 return deleted;
1506}
1507
Matthew Wilcox57578c22016-05-20 17:01:54 -07001508static inline void delete_sibling_entries(struct radix_tree_node *node,
1509 void *ptr, unsigned offset)
1510{
1511#ifdef CONFIG_RADIX_TREE_MULTIORDER
1512 int i;
1513 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1514 if (node->slots[offset + i] != ptr)
1515 break;
1516 node->slots[offset + i] = NULL;
1517 node->count--;
1518 }
1519#endif
1520}
1521
Johannes Weiner139e5612014-04-03 14:47:54 -07001522/**
Johannes Weiner53c59f22014-04-03 14:47:39 -07001523 * radix_tree_delete_item - delete an item from a radix tree
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 * @root: radix tree root
1525 * @index: index key
Johannes Weiner53c59f22014-04-03 14:47:39 -07001526 * @item: expected item
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001528 * Remove @item at @index from the radix tree rooted at @root.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 *
Johannes Weiner53c59f22014-04-03 14:47:39 -07001530 * Returns the address of the deleted item, or NULL if it was not present
1531 * or the entry at the given @index was not @item.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 */
Johannes Weiner53c59f22014-04-03 14:47:39 -07001533void *radix_tree_delete_item(struct radix_tree_root *root,
1534 unsigned long index, void *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535{
Johannes Weiner139e5612014-04-03 14:47:54 -07001536 struct radix_tree_node *node;
Matthew Wilcox57578c22016-05-20 17:01:54 -07001537 unsigned int offset;
Johannes Weiner139e5612014-04-03 14:47:54 -07001538 void **slot;
1539 void *entry;
Nick Piggind5274262006-01-08 01:01:41 -08001540 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
Johannes Weiner139e5612014-04-03 14:47:54 -07001542 entry = __radix_tree_lookup(root, index, &node, &slot);
1543 if (!entry)
1544 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Johannes Weiner139e5612014-04-03 14:47:54 -07001546 if (item && entry != item)
1547 return NULL;
1548
1549 if (!node) {
Nick Piggin612d6c12006-06-23 02:03:22 -07001550 root_tag_clear_all(root);
1551 root->rnode = NULL;
Johannes Weiner139e5612014-04-03 14:47:54 -07001552 return entry;
Nick Piggin612d6c12006-06-23 02:03:22 -07001553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Matthew Wilcox29e09672016-05-20 17:02:02 -07001555 offset = get_slot_offset(node, slot);
Johannes Weiner53c59f22014-04-03 14:47:39 -07001556
Matthew Wilcoxd604c322016-05-20 17:03:45 -07001557 /* Clear all tags associated with the item to be deleted. */
1558 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1559 node_tag_clear(root, node, tag, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
Matthew Wilcoxa4db4dc2016-05-20 17:03:24 -07001561 delete_sibling_entries(node, node_to_entry(slot), offset);
Johannes Weiner139e5612014-04-03 14:47:54 -07001562 node->slots[offset] = NULL;
1563 node->count--;
Nick Piggina5f51c92006-01-08 01:01:41 -08001564
Johannes Weiner449dd692014-04-03 14:47:56 -07001565 __radix_tree_delete_node(root, node);
Christoph Lameter201b6262005-09-06 15:16:46 -07001566
Johannes Weiner139e5612014-04-03 14:47:54 -07001567 return entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568}
Johannes Weiner53c59f22014-04-03 14:47:39 -07001569EXPORT_SYMBOL(radix_tree_delete_item);
1570
1571/**
1572 * radix_tree_delete - delete an item from a radix tree
1573 * @root: radix tree root
1574 * @index: index key
1575 *
1576 * Remove the item at @index from the radix tree rooted at @root.
1577 *
1578 * Returns the address of the deleted item, or NULL if it was not present.
1579 */
1580void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1581{
1582 return radix_tree_delete_item(root, index, NULL);
1583}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584EXPORT_SYMBOL(radix_tree_delete);
1585
Johannes Weinerd3798ae2016-10-04 22:02:08 +02001586void radix_tree_clear_tags(struct radix_tree_root *root,
1587 struct radix_tree_node *node,
1588 void **slot)
Matthew Wilcoxd604c322016-05-20 17:03:45 -07001589{
Matthew Wilcoxd604c322016-05-20 17:03:45 -07001590 if (node) {
1591 unsigned int tag, offset = get_slot_offset(node, slot);
1592 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1593 node_tag_clear(root, node, tag, offset);
1594 } else {
1595 /* Clear root node tags */
1596 root->gfp_mask &= __GFP_BITS_MASK;
1597 }
Matthew Wilcoxd604c322016-05-20 17:03:45 -07001598}
1599
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600/**
1601 * radix_tree_tagged - test whether any items in the tree are tagged
1602 * @root: radix tree root
1603 * @tag: tag to test
1604 */
Jonathan Corbetdaff89f2006-03-25 03:08:05 -08001605int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606{
Nick Piggin612d6c12006-06-23 02:03:22 -07001607 return root_tag_get(root, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608}
1609EXPORT_SYMBOL(radix_tree_tagged);
1610
1611static void
Johannes Weiner449dd692014-04-03 14:47:56 -07001612radix_tree_node_ctor(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
Johannes Weiner449dd692014-04-03 14:47:56 -07001614 struct radix_tree_node *node = arg;
1615
1616 memset(node, 0, sizeof(*node));
1617 INIT_LIST_HEAD(&node->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618}
1619
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -07001620static __init unsigned long __maxindex(unsigned int height)
1621{
1622 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1623 int shift = RADIX_TREE_INDEX_BITS - width;
1624
1625 if (shift < 0)
1626 return ~0UL;
1627 if (shift >= BITS_PER_LONG)
1628 return 0UL;
1629 return ~0UL >> shift;
1630}
1631
1632static __init void radix_tree_init_maxnodes(void)
1633{
1634 unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1];
1635 unsigned int i, j;
1636
1637 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1638 height_to_maxindex[i] = __maxindex(i);
1639 for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) {
1640 for (j = i; j > 0; j--)
1641 height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1;
1642 }
1643}
1644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645static int radix_tree_callback(struct notifier_block *nfb,
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001646 unsigned long action, void *hcpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001648 int cpu = (long)hcpu;
1649 struct radix_tree_preload *rtp;
1650 struct radix_tree_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001652 /* Free per-cpu pool of preloaded nodes */
1653 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1654 rtp = &per_cpu(radix_tree_preloads, cpu);
1655 while (rtp->nr) {
Kirill A. Shutemov9d2a8da2015-06-25 15:02:19 -07001656 node = rtp->nodes;
1657 rtp->nodes = node->private_data;
1658 kmem_cache_free(radix_tree_node_cachep, node);
1659 rtp->nr--;
Matthew Wilcox2fcd9002016-05-20 17:03:04 -07001660 }
1661 }
1662 return NOTIFY_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665void __init radix_tree_init(void)
1666{
1667 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1668 sizeof(struct radix_tree_node), 0,
Christoph Lameter488514d2008-04-28 02:12:05 -07001669 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1670 radix_tree_node_ctor);
Kirill A. Shutemovc78c66d2016-07-26 15:26:02 -07001671 radix_tree_init_maxnodes();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 hotcpu_notifier(radix_tree_callback, 0);
1673}