proc: faster /proc/cmdline
[linux-2.6-block.git] / lib / radix-tree.c
CommitLineData
1da177e4
LT
1/*
2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
cde53535 4 * Copyright (C) 2005 SGI, Christoph Lameter
7cf9c2c7 5 * Copyright (C) 2006 Nick Piggin
78c1d784 6 * Copyright (C) 2012 Konstantin Khlebnikov
6b053b8e
MW
7 * Copyright (C) 2016 Intel, Matthew Wilcox
8 * Copyright (C) 2016 Intel, Ross Zwisler
1da177e4
LT
9 *
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
0a835c4f
MW
25#include <linux/bitmap.h>
26#include <linux/bitops.h>
460488c5 27#include <linux/bug.h>
e157b555 28#include <linux/cpu.h>
1da177e4 29#include <linux/errno.h>
0a835c4f
MW
30#include <linux/export.h>
31#include <linux/idr.h>
1da177e4
LT
32#include <linux/init.h>
33#include <linux/kernel.h>
0a835c4f 34#include <linux/kmemleak.h>
1da177e4 35#include <linux/percpu.h>
0a835c4f
MW
36#include <linux/preempt.h> /* in_interrupt() */
37#include <linux/radix-tree.h>
38#include <linux/rcupdate.h>
1da177e4 39#include <linux/slab.h>
1da177e4 40#include <linux/string.h>
1da177e4
LT
41
42
c78c66d1
KS
43/* Number of nodes in fully populated tree of given height */
44static unsigned long height_to_maxnodes[RADIX_TREE_MAX_PATH + 1] __read_mostly;
45
1da177e4
LT
46/*
47 * Radix tree node cache.
48 */
e18b890b 49static struct kmem_cache *radix_tree_node_cachep;
1da177e4 50
55368052
NP
51/*
52 * The radix tree is variable-height, so an insert operation not only has
53 * to build the branch to its corresponding item, it also has to build the
54 * branch to existing items if the size has to be increased (by
55 * radix_tree_extend).
56 *
57 * The worst case is a zero height tree with just a single item at index 0,
58 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
59 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
60 * Hence:
61 */
62#define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
63
0a835c4f
MW
64/*
65 * The IDR does not have to be as high as the radix tree since it uses
66 * signed integers, not unsigned longs.
67 */
68#define IDR_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(int) - 1)
69#define IDR_MAX_PATH (DIV_ROUND_UP(IDR_INDEX_BITS, \
70 RADIX_TREE_MAP_SHIFT))
71#define IDR_PRELOAD_SIZE (IDR_MAX_PATH * 2 - 1)
72
7ad3d4d8
MW
73/*
74 * The IDA is even shorter since it uses a bitmap at the last level.
75 */
76#define IDA_INDEX_BITS (8 * sizeof(int) - 1 - ilog2(IDA_BITMAP_BITS))
77#define IDA_MAX_PATH (DIV_ROUND_UP(IDA_INDEX_BITS, \
78 RADIX_TREE_MAP_SHIFT))
79#define IDA_PRELOAD_SIZE (IDA_MAX_PATH * 2 - 1)
80
1da177e4
LT
81/*
82 * Per-cpu pool of preloaded nodes
83 */
84struct radix_tree_preload {
2fcd9005 85 unsigned nr;
1293d5c5 86 /* nodes->parent points to next preallocated node */
9d2a8da0 87 struct radix_tree_node *nodes;
1da177e4 88};
8cef7d57 89static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
1da177e4 90
148deab2
MW
91static inline struct radix_tree_node *entry_to_node(void *ptr)
92{
93 return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE);
94}
95
a4db4dce 96static inline void *node_to_entry(void *ptr)
27d20fdd 97{
30ff46cc 98 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
27d20fdd
NP
99}
100
a4db4dce 101#define RADIX_TREE_RETRY node_to_entry(NULL)
afe0e395 102
db050f29
MW
103#ifdef CONFIG_RADIX_TREE_MULTIORDER
104/* Sibling slots point directly to another slot in the same node */
35534c86
MW
105static inline
106bool is_sibling_entry(const struct radix_tree_node *parent, void *node)
db050f29 107{
d7b62727 108 void __rcu **ptr = node;
db050f29
MW
109 return (parent->slots <= ptr) &&
110 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
111}
112#else
35534c86
MW
113static inline
114bool is_sibling_entry(const struct radix_tree_node *parent, void *node)
db050f29
MW
115{
116 return false;
117}
118#endif
119
d7b62727
MW
120static inline unsigned long
121get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot)
db050f29
MW
122{
123 return slot - parent->slots;
124}
125
35534c86 126static unsigned int radix_tree_descend(const struct radix_tree_node *parent,
9e85d811 127 struct radix_tree_node **nodep, unsigned long index)
db050f29 128{
9e85d811 129 unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK;
d7b62727 130 void __rcu **entry = rcu_dereference_raw(parent->slots[offset]);
db050f29
MW
131
132#ifdef CONFIG_RADIX_TREE_MULTIORDER
b194d16c 133 if (radix_tree_is_internal_node(entry)) {
8d2c0d36 134 if (is_sibling_entry(parent, entry)) {
d7b62727
MW
135 void __rcu **sibentry;
136 sibentry = (void __rcu **) entry_to_node(entry);
8d2c0d36
LT
137 offset = get_slot_offset(parent, sibentry);
138 entry = rcu_dereference_raw(*sibentry);
db050f29
MW
139 }
140 }
141#endif
142
143 *nodep = (void *)entry;
144 return offset;
145}
146
35534c86 147static inline gfp_t root_gfp_mask(const struct radix_tree_root *root)
612d6c19
NP
148{
149 return root->gfp_mask & __GFP_BITS_MASK;
150}
151
643b52b9
NP
152static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
153 int offset)
154{
155 __set_bit(offset, node->tags[tag]);
156}
157
158static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
159 int offset)
160{
161 __clear_bit(offset, node->tags[tag]);
162}
163
35534c86 164static inline int tag_get(const struct radix_tree_node *node, unsigned int tag,
643b52b9
NP
165 int offset)
166{
167 return test_bit(offset, node->tags[tag]);
168}
169
35534c86 170static inline void root_tag_set(struct radix_tree_root *root, unsigned tag)
643b52b9 171{
0a835c4f 172 root->gfp_mask |= (__force gfp_t)(1 << (tag + ROOT_TAG_SHIFT));
643b52b9
NP
173}
174
2fcd9005 175static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
643b52b9 176{
0a835c4f 177 root->gfp_mask &= (__force gfp_t)~(1 << (tag + ROOT_TAG_SHIFT));
643b52b9
NP
178}
179
180static inline void root_tag_clear_all(struct radix_tree_root *root)
181{
0a835c4f 182 root->gfp_mask &= (1 << ROOT_TAG_SHIFT) - 1;
643b52b9
NP
183}
184
35534c86 185static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag)
643b52b9 186{
0a835c4f 187 return (__force int)root->gfp_mask & (1 << (tag + ROOT_TAG_SHIFT));
643b52b9
NP
188}
189
35534c86 190static inline unsigned root_tags_get(const struct radix_tree_root *root)
643b52b9 191{
0a835c4f 192 return (__force unsigned)root->gfp_mask >> ROOT_TAG_SHIFT;
643b52b9
NP
193}
194
0a835c4f 195static inline bool is_idr(const struct radix_tree_root *root)
7b60e9ad 196{
0a835c4f 197 return !!(root->gfp_mask & ROOT_IS_IDR);
7b60e9ad
MW
198}
199
643b52b9
NP
200/*
201 * Returns 1 if any slot in the node has this tag set.
202 * Otherwise returns 0.
203 */
35534c86
MW
204static inline int any_tag_set(const struct radix_tree_node *node,
205 unsigned int tag)
643b52b9 206{
2fcd9005 207 unsigned idx;
643b52b9
NP
208 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
209 if (node->tags[tag][idx])
210 return 1;
211 }
212 return 0;
213}
78c1d784 214
0a835c4f
MW
215static inline void all_tag_set(struct radix_tree_node *node, unsigned int tag)
216{
217 bitmap_fill(node->tags[tag], RADIX_TREE_MAP_SIZE);
218}
219
78c1d784
KK
220/**
221 * radix_tree_find_next_bit - find the next set bit in a memory region
222 *
223 * @addr: The address to base the search on
224 * @size: The bitmap size in bits
225 * @offset: The bitnumber to start searching at
226 *
227 * Unrollable variant of find_next_bit() for constant size arrays.
228 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
229 * Returns next bit offset, or size if nothing found.
230 */
231static __always_inline unsigned long
bc412fca
MW
232radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag,
233 unsigned long offset)
78c1d784 234{
bc412fca 235 const unsigned long *addr = node->tags[tag];
78c1d784 236
bc412fca 237 if (offset < RADIX_TREE_MAP_SIZE) {
78c1d784
KK
238 unsigned long tmp;
239
240 addr += offset / BITS_PER_LONG;
241 tmp = *addr >> (offset % BITS_PER_LONG);
242 if (tmp)
243 return __ffs(tmp) + offset;
244 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
bc412fca 245 while (offset < RADIX_TREE_MAP_SIZE) {
78c1d784
KK
246 tmp = *++addr;
247 if (tmp)
248 return __ffs(tmp) + offset;
249 offset += BITS_PER_LONG;
250 }
251 }
bc412fca 252 return RADIX_TREE_MAP_SIZE;
78c1d784
KK
253}
254
268f42de
MW
255static unsigned int iter_offset(const struct radix_tree_iter *iter)
256{
257 return (iter->index >> iter_shift(iter)) & RADIX_TREE_MAP_MASK;
258}
259
218ed750
MW
260/*
261 * The maximum index which can be stored in a radix tree
262 */
263static inline unsigned long shift_maxindex(unsigned int shift)
264{
265 return (RADIX_TREE_MAP_SIZE << shift) - 1;
266}
267
35534c86 268static inline unsigned long node_maxindex(const struct radix_tree_node *node)
218ed750
MW
269{
270 return shift_maxindex(node->shift);
271}
272
0a835c4f
MW
273static unsigned long next_index(unsigned long index,
274 const struct radix_tree_node *node,
275 unsigned long offset)
276{
277 return (index & ~node_maxindex(node)) + (offset << node->shift);
278}
279
0796c583 280#ifndef __KERNEL__
d0891265 281static void dump_node(struct radix_tree_node *node, unsigned long index)
7cf19af4 282{
0796c583 283 unsigned long i;
7cf19af4 284
218ed750
MW
285 pr_debug("radix node: %p offset %d indices %lu-%lu parent %p tags %lx %lx %lx shift %d count %d exceptional %d\n",
286 node, node->offset, index, index | node_maxindex(node),
287 node->parent,
0796c583 288 node->tags[0][0], node->tags[1][0], node->tags[2][0],
218ed750 289 node->shift, node->count, node->exceptional);
0796c583
RZ
290
291 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
d0891265
MW
292 unsigned long first = index | (i << node->shift);
293 unsigned long last = first | ((1UL << node->shift) - 1);
0796c583
RZ
294 void *entry = node->slots[i];
295 if (!entry)
296 continue;
218ed750
MW
297 if (entry == RADIX_TREE_RETRY) {
298 pr_debug("radix retry offset %ld indices %lu-%lu parent %p\n",
299 i, first, last, node);
b194d16c 300 } else if (!radix_tree_is_internal_node(entry)) {
218ed750
MW
301 pr_debug("radix entry %p offset %ld indices %lu-%lu parent %p\n",
302 entry, i, first, last, node);
303 } else if (is_sibling_entry(node, entry)) {
304 pr_debug("radix sblng %p offset %ld indices %lu-%lu parent %p val %p\n",
305 entry, i, first, last, node,
306 *(void **)entry_to_node(entry));
0796c583 307 } else {
4dd6c098 308 dump_node(entry_to_node(entry), first);
0796c583
RZ
309 }
310 }
7cf19af4
MW
311}
312
313/* For debug */
314static void radix_tree_dump(struct radix_tree_root *root)
315{
d0891265
MW
316 pr_debug("radix root: %p rnode %p tags %x\n",
317 root, root->rnode,
0a835c4f 318 root->gfp_mask >> ROOT_TAG_SHIFT);
b194d16c 319 if (!radix_tree_is_internal_node(root->rnode))
7cf19af4 320 return;
4dd6c098 321 dump_node(entry_to_node(root->rnode), 0);
7cf19af4 322}
0a835c4f
MW
323
324static void dump_ida_node(void *entry, unsigned long index)
325{
326 unsigned long i;
327
328 if (!entry)
329 return;
330
331 if (radix_tree_is_internal_node(entry)) {
332 struct radix_tree_node *node = entry_to_node(entry);
333
334 pr_debug("ida node: %p offset %d indices %lu-%lu parent %p free %lx shift %d count %d\n",
335 node, node->offset, index * IDA_BITMAP_BITS,
336 ((index | node_maxindex(node)) + 1) *
337 IDA_BITMAP_BITS - 1,
338 node->parent, node->tags[0][0], node->shift,
339 node->count);
340 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++)
341 dump_ida_node(node->slots[i],
342 index | (i << node->shift));
d37cacc5
MW
343 } else if (radix_tree_exceptional_entry(entry)) {
344 pr_debug("ida excp: %p offset %d indices %lu-%lu data %lx\n",
345 entry, (int)(index & RADIX_TREE_MAP_MASK),
346 index * IDA_BITMAP_BITS,
347 index * IDA_BITMAP_BITS + BITS_PER_LONG -
348 RADIX_TREE_EXCEPTIONAL_SHIFT,
349 (unsigned long)entry >>
350 RADIX_TREE_EXCEPTIONAL_SHIFT);
0a835c4f
MW
351 } else {
352 struct ida_bitmap *bitmap = entry;
353
354 pr_debug("ida btmp: %p offset %d indices %lu-%lu data", bitmap,
355 (int)(index & RADIX_TREE_MAP_MASK),
356 index * IDA_BITMAP_BITS,
357 (index + 1) * IDA_BITMAP_BITS - 1);
358 for (i = 0; i < IDA_BITMAP_LONGS; i++)
359 pr_cont(" %lx", bitmap->bitmap[i]);
360 pr_cont("\n");
361 }
362}
363
364static void ida_dump(struct ida *ida)
365{
366 struct radix_tree_root *root = &ida->ida_rt;
7ad3d4d8
MW
367 pr_debug("ida: %p node %p free %d\n", ida, root->rnode,
368 root->gfp_mask >> ROOT_TAG_SHIFT);
0a835c4f
MW
369 dump_ida_node(root->rnode, 0);
370}
7cf19af4
MW
371#endif
372
1da177e4
LT
373/*
374 * This assumes that the caller has performed appropriate preallocation, and
375 * that the caller has pinned this thread of control to the current CPU.
376 */
377static struct radix_tree_node *
0a835c4f 378radix_tree_node_alloc(gfp_t gfp_mask, struct radix_tree_node *parent,
d58275bc 379 struct radix_tree_root *root,
e8de4340
MW
380 unsigned int shift, unsigned int offset,
381 unsigned int count, unsigned int exceptional)
1da177e4 382{
e2848a0e 383 struct radix_tree_node *ret = NULL;
1da177e4 384
5e4c0d97 385 /*
2fcd9005
MW
386 * Preload code isn't irq safe and it doesn't make sense to use
387 * preloading during an interrupt anyway as all the allocations have
388 * to be atomic. So just do normal allocation when in interrupt.
5e4c0d97 389 */
d0164adc 390 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
1da177e4
LT
391 struct radix_tree_preload *rtp;
392
58e698af
VD
393 /*
394 * Even if the caller has preloaded, try to allocate from the
05eb6e72
VD
395 * cache first for the new node to get accounted to the memory
396 * cgroup.
58e698af
VD
397 */
398 ret = kmem_cache_alloc(radix_tree_node_cachep,
05eb6e72 399 gfp_mask | __GFP_NOWARN);
58e698af
VD
400 if (ret)
401 goto out;
402
e2848a0e
NP
403 /*
404 * Provided the caller has preloaded here, we will always
405 * succeed in getting a node here (and never reach
406 * kmem_cache_alloc)
407 */
7c8e0181 408 rtp = this_cpu_ptr(&radix_tree_preloads);
1da177e4 409 if (rtp->nr) {
9d2a8da0 410 ret = rtp->nodes;
1293d5c5 411 rtp->nodes = ret->parent;
1da177e4
LT
412 rtp->nr--;
413 }
ce80b067
CM
414 /*
415 * Update the allocation stack trace as this is more useful
416 * for debugging.
417 */
418 kmemleak_update_trace(ret);
58e698af 419 goto out;
1da177e4 420 }
05eb6e72 421 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
58e698af 422out:
b194d16c 423 BUG_ON(radix_tree_is_internal_node(ret));
e8de4340 424 if (ret) {
e8de4340
MW
425 ret->shift = shift;
426 ret->offset = offset;
427 ret->count = count;
428 ret->exceptional = exceptional;
d58275bc
MW
429 ret->parent = parent;
430 ret->root = root;
e8de4340 431 }
1da177e4
LT
432 return ret;
433}
434
7cf9c2c7
NP
435static void radix_tree_node_rcu_free(struct rcu_head *head)
436{
437 struct radix_tree_node *node =
438 container_of(head, struct radix_tree_node, rcu_head);
643b52b9
NP
439
440 /*
175542f5
MW
441 * Must only free zeroed nodes into the slab. We can be left with
442 * non-NULL entries by radix_tree_free_nodes, so clear the entries
443 * and tags here.
643b52b9 444 */
175542f5
MW
445 memset(node->slots, 0, sizeof(node->slots));
446 memset(node->tags, 0, sizeof(node->tags));
91d9c05a 447 INIT_LIST_HEAD(&node->private_list);
643b52b9 448
7cf9c2c7
NP
449 kmem_cache_free(radix_tree_node_cachep, node);
450}
451
1da177e4
LT
452static inline void
453radix_tree_node_free(struct radix_tree_node *node)
454{
7cf9c2c7 455 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
1da177e4
LT
456}
457
458/*
459 * Load up this CPU's radix_tree_node buffer with sufficient objects to
460 * ensure that the addition of a single element in the tree cannot fail. On
461 * success, return zero, with preemption disabled. On error, return -ENOMEM
462 * with preemption not disabled.
b34df792
DH
463 *
464 * To make use of this facility, the radix tree must be initialised without
d0164adc 465 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
1da177e4 466 */
bc9ae224 467static __must_check int __radix_tree_preload(gfp_t gfp_mask, unsigned nr)
1da177e4
LT
468{
469 struct radix_tree_preload *rtp;
470 struct radix_tree_node *node;
471 int ret = -ENOMEM;
472
05eb6e72
VD
473 /*
474 * Nodes preloaded by one cgroup can be be used by another cgroup, so
475 * they should never be accounted to any particular memory cgroup.
476 */
477 gfp_mask &= ~__GFP_ACCOUNT;
478
1da177e4 479 preempt_disable();
7c8e0181 480 rtp = this_cpu_ptr(&radix_tree_preloads);
c78c66d1 481 while (rtp->nr < nr) {
1da177e4 482 preempt_enable();
488514d1 483 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
1da177e4
LT
484 if (node == NULL)
485 goto out;
486 preempt_disable();
7c8e0181 487 rtp = this_cpu_ptr(&radix_tree_preloads);
c78c66d1 488 if (rtp->nr < nr) {
1293d5c5 489 node->parent = rtp->nodes;
9d2a8da0
KS
490 rtp->nodes = node;
491 rtp->nr++;
492 } else {
1da177e4 493 kmem_cache_free(radix_tree_node_cachep, node);
9d2a8da0 494 }
1da177e4
LT
495 }
496 ret = 0;
497out:
498 return ret;
499}
5e4c0d97
JK
500
501/*
502 * Load up this CPU's radix_tree_node buffer with sufficient objects to
503 * ensure that the addition of a single element in the tree cannot fail. On
504 * success, return zero, with preemption disabled. On error, return -ENOMEM
505 * with preemption not disabled.
506 *
507 * To make use of this facility, the radix tree must be initialised without
d0164adc 508 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
5e4c0d97
JK
509 */
510int radix_tree_preload(gfp_t gfp_mask)
511{
512 /* Warn on non-sensical use... */
d0164adc 513 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
c78c66d1 514 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
5e4c0d97 515}
d7f0923d 516EXPORT_SYMBOL(radix_tree_preload);
1da177e4 517
5e4c0d97
JK
518/*
519 * The same as above function, except we don't guarantee preloading happens.
520 * We do it, if we decide it helps. On success, return zero with preemption
521 * disabled. On error, return -ENOMEM with preemption not disabled.
522 */
523int radix_tree_maybe_preload(gfp_t gfp_mask)
524{
d0164adc 525 if (gfpflags_allow_blocking(gfp_mask))
c78c66d1 526 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
5e4c0d97
JK
527 /* Preloading doesn't help anything with this gfp mask, skip it */
528 preempt_disable();
529 return 0;
530}
531EXPORT_SYMBOL(radix_tree_maybe_preload);
532
2791653a
MW
533#ifdef CONFIG_RADIX_TREE_MULTIORDER
534/*
535 * Preload with enough objects to ensure that we can split a single entry
536 * of order @old_order into many entries of size @new_order
537 */
538int radix_tree_split_preload(unsigned int old_order, unsigned int new_order,
539 gfp_t gfp_mask)
540{
541 unsigned top = 1 << (old_order % RADIX_TREE_MAP_SHIFT);
542 unsigned layers = (old_order / RADIX_TREE_MAP_SHIFT) -
543 (new_order / RADIX_TREE_MAP_SHIFT);
544 unsigned nr = 0;
545
546 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
547 BUG_ON(new_order >= old_order);
548
549 while (layers--)
550 nr = nr * RADIX_TREE_MAP_SIZE + 1;
551 return __radix_tree_preload(gfp_mask, top * nr);
552}
553#endif
554
c78c66d1
KS
555/*
556 * The same as function above, but preload number of nodes required to insert
557 * (1 << order) continuous naturally-aligned elements.
558 */
559int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order)
560{
561 unsigned long nr_subtrees;
562 int nr_nodes, subtree_height;
563
564 /* Preloading doesn't help anything with this gfp mask, skip it */
565 if (!gfpflags_allow_blocking(gfp_mask)) {
566 preempt_disable();
567 return 0;
568 }
569
570 /*
571 * Calculate number and height of fully populated subtrees it takes to
572 * store (1 << order) elements.
573 */
574 nr_subtrees = 1 << order;
575 for (subtree_height = 0; nr_subtrees > RADIX_TREE_MAP_SIZE;
576 subtree_height++)
577 nr_subtrees >>= RADIX_TREE_MAP_SHIFT;
578
579 /*
580 * The worst case is zero height tree with a single item at index 0 and
581 * then inserting items starting at ULONG_MAX - (1 << order).
582 *
583 * This requires RADIX_TREE_MAX_PATH nodes to build branch from root to
584 * 0-index item.
585 */
586 nr_nodes = RADIX_TREE_MAX_PATH;
587
588 /* Plus branch to fully populated subtrees. */
589 nr_nodes += RADIX_TREE_MAX_PATH - subtree_height;
590
591 /* Root node is shared. */
592 nr_nodes--;
593
594 /* Plus nodes required to build subtrees. */
595 nr_nodes += nr_subtrees * height_to_maxnodes[subtree_height];
596
597 return __radix_tree_preload(gfp_mask, nr_nodes);
598}
599
35534c86 600static unsigned radix_tree_load_root(const struct radix_tree_root *root,
1456a439
MW
601 struct radix_tree_node **nodep, unsigned long *maxindex)
602{
603 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
604
605 *nodep = node;
606
b194d16c 607 if (likely(radix_tree_is_internal_node(node))) {
4dd6c098 608 node = entry_to_node(node);
1456a439 609 *maxindex = node_maxindex(node);
c12e51b0 610 return node->shift + RADIX_TREE_MAP_SHIFT;
1456a439
MW
611 }
612
613 *maxindex = 0;
614 return 0;
615}
616
1da177e4
LT
617/*
618 * Extend a radix tree so it can store key @index.
619 */
0a835c4f 620static int radix_tree_extend(struct radix_tree_root *root, gfp_t gfp,
d0891265 621 unsigned long index, unsigned int shift)
1da177e4 622{
d7b62727 623 void *entry;
d0891265 624 unsigned int maxshift;
1da177e4
LT
625 int tag;
626
d0891265
MW
627 /* Figure out what the shift should be. */
628 maxshift = shift;
629 while (index > shift_maxindex(maxshift))
630 maxshift += RADIX_TREE_MAP_SHIFT;
1da177e4 631
d7b62727
MW
632 entry = rcu_dereference_raw(root->rnode);
633 if (!entry && (!is_idr(root) || root_tag_get(root, IDR_FREE)))
1da177e4 634 goto out;
1da177e4 635
1da177e4 636 do {
0a835c4f 637 struct radix_tree_node *node = radix_tree_node_alloc(gfp, NULL,
d58275bc 638 root, shift, 0, 1, 0);
2fcd9005 639 if (!node)
1da177e4
LT
640 return -ENOMEM;
641
0a835c4f
MW
642 if (is_idr(root)) {
643 all_tag_set(node, IDR_FREE);
644 if (!root_tag_get(root, IDR_FREE)) {
645 tag_clear(node, IDR_FREE, 0);
646 root_tag_set(root, IDR_FREE);
647 }
648 } else {
649 /* Propagate the aggregated tag info to the new child */
650 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
651 if (root_tag_get(root, tag))
652 tag_set(node, tag, 0);
653 }
1da177e4
LT
654 }
655
d0891265 656 BUG_ON(shift > BITS_PER_LONG);
d7b62727
MW
657 if (radix_tree_is_internal_node(entry)) {
658 entry_to_node(entry)->parent = node;
659 } else if (radix_tree_exceptional_entry(entry)) {
f7942430 660 /* Moving an exceptional root->rnode to a node */
e8de4340 661 node->exceptional = 1;
f7942430 662 }
d7b62727
MW
663 /*
664 * entry was already in the radix tree, so we do not need
665 * rcu_assign_pointer here
666 */
667 node->slots[0] = (void __rcu *)entry;
668 entry = node_to_entry(node);
669 rcu_assign_pointer(root->rnode, entry);
d0891265 670 shift += RADIX_TREE_MAP_SHIFT;
d0891265 671 } while (shift <= maxshift);
1da177e4 672out:
d0891265 673 return maxshift + RADIX_TREE_MAP_SHIFT;
1da177e4
LT
674}
675
f4b109c6
JW
676/**
677 * radix_tree_shrink - shrink radix tree to minimum height
678 * @root radix tree root
679 */
0ac398ef 680static inline bool radix_tree_shrink(struct radix_tree_root *root,
c7df8ad2 681 radix_tree_update_node_t update_node)
f4b109c6 682{
0ac398ef
MW
683 bool shrunk = false;
684
f4b109c6 685 for (;;) {
12320d0f 686 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
f4b109c6
JW
687 struct radix_tree_node *child;
688
689 if (!radix_tree_is_internal_node(node))
690 break;
691 node = entry_to_node(node);
692
693 /*
694 * The candidate node has more than one child, or its child
695 * is not at the leftmost slot, or the child is a multiorder
696 * entry, we cannot shrink.
697 */
698 if (node->count != 1)
699 break;
12320d0f 700 child = rcu_dereference_raw(node->slots[0]);
f4b109c6
JW
701 if (!child)
702 break;
703 if (!radix_tree_is_internal_node(child) && node->shift)
704 break;
705
706 if (radix_tree_is_internal_node(child))
707 entry_to_node(child)->parent = NULL;
708
709 /*
710 * We don't need rcu_assign_pointer(), since we are simply
711 * moving the node from one part of the tree to another: if it
712 * was safe to dereference the old pointer to it
713 * (node->slots[0]), it will be safe to dereference the new
714 * one (root->rnode) as far as dependent read barriers go.
715 */
d7b62727 716 root->rnode = (void __rcu *)child;
0a835c4f
MW
717 if (is_idr(root) && !tag_get(node, IDR_FREE, 0))
718 root_tag_clear(root, IDR_FREE);
f4b109c6
JW
719
720 /*
721 * We have a dilemma here. The node's slot[0] must not be
722 * NULLed in case there are concurrent lookups expecting to
723 * find the item. However if this was a bottom-level node,
724 * then it may be subject to the slot pointer being visible
725 * to callers dereferencing it. If item corresponding to
726 * slot[0] is subsequently deleted, these callers would expect
727 * their slot to become empty sooner or later.
728 *
729 * For example, lockless pagecache will look up a slot, deref
730 * the page pointer, and if the page has 0 refcount it means it
731 * was concurrently deleted from pagecache so try the deref
732 * again. Fortunately there is already a requirement for logic
733 * to retry the entire slot lookup -- the indirect pointer
734 * problem (replacing direct root node with an indirect pointer
735 * also results in a stale slot). So tag the slot as indirect
736 * to force callers to retry.
737 */
4d693d08
JW
738 node->count = 0;
739 if (!radix_tree_is_internal_node(child)) {
d7b62727 740 node->slots[0] = (void __rcu *)RADIX_TREE_RETRY;
4d693d08 741 if (update_node)
c7df8ad2 742 update_node(node);
4d693d08 743 }
f4b109c6 744
ea07b862 745 WARN_ON_ONCE(!list_empty(&node->private_list));
f4b109c6 746 radix_tree_node_free(node);
0ac398ef 747 shrunk = true;
f4b109c6 748 }
0ac398ef
MW
749
750 return shrunk;
f4b109c6
JW
751}
752
0ac398ef 753static bool delete_node(struct radix_tree_root *root,
4d693d08 754 struct radix_tree_node *node,
c7df8ad2 755 radix_tree_update_node_t update_node)
f4b109c6 756{
0ac398ef
MW
757 bool deleted = false;
758
f4b109c6
JW
759 do {
760 struct radix_tree_node *parent;
761
762 if (node->count) {
12320d0f
MW
763 if (node_to_entry(node) ==
764 rcu_dereference_raw(root->rnode))
c7df8ad2
MG
765 deleted |= radix_tree_shrink(root,
766 update_node);
0ac398ef 767 return deleted;
f4b109c6
JW
768 }
769
770 parent = node->parent;
771 if (parent) {
772 parent->slots[node->offset] = NULL;
773 parent->count--;
774 } else {
0a835c4f
MW
775 /*
776 * Shouldn't the tags already have all been cleared
777 * by the caller?
778 */
779 if (!is_idr(root))
780 root_tag_clear_all(root);
f4b109c6
JW
781 root->rnode = NULL;
782 }
783
ea07b862 784 WARN_ON_ONCE(!list_empty(&node->private_list));
f4b109c6 785 radix_tree_node_free(node);
0ac398ef 786 deleted = true;
f4b109c6
JW
787
788 node = parent;
789 } while (node);
0ac398ef
MW
790
791 return deleted;
f4b109c6
JW
792}
793
1da177e4 794/**
139e5616 795 * __radix_tree_create - create a slot in a radix tree
1da177e4
LT
796 * @root: radix tree root
797 * @index: index key
e6145236 798 * @order: index occupies 2^order aligned slots
139e5616
JW
799 * @nodep: returns node
800 * @slotp: returns slot
1da177e4 801 *
139e5616
JW
802 * Create, if necessary, and return the node and slot for an item
803 * at position @index in the radix tree @root.
804 *
805 * Until there is more than one item in the tree, no nodes are
806 * allocated and @root->rnode is used as a direct slot instead of
807 * pointing to a node, in which case *@nodep will be NULL.
808 *
809 * Returns -ENOMEM, or 0 for success.
1da177e4 810 */
139e5616 811int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
e6145236 812 unsigned order, struct radix_tree_node **nodep,
d7b62727 813 void __rcu ***slotp)
1da177e4 814{
89148aa4 815 struct radix_tree_node *node = NULL, *child;
d7b62727 816 void __rcu **slot = (void __rcu **)&root->rnode;
49ea6ebc 817 unsigned long maxindex;
89148aa4 818 unsigned int shift, offset = 0;
49ea6ebc 819 unsigned long max = index | ((1UL << order) - 1);
0a835c4f 820 gfp_t gfp = root_gfp_mask(root);
49ea6ebc 821
89148aa4 822 shift = radix_tree_load_root(root, &child, &maxindex);
1da177e4
LT
823
824 /* Make sure the tree is high enough. */
175542f5
MW
825 if (order > 0 && max == ((1UL << order) - 1))
826 max++;
49ea6ebc 827 if (max > maxindex) {
0a835c4f 828 int error = radix_tree_extend(root, gfp, max, shift);
49ea6ebc 829 if (error < 0)
1da177e4 830 return error;
49ea6ebc 831 shift = error;
12320d0f 832 child = rcu_dereference_raw(root->rnode);
1da177e4
LT
833 }
834
e6145236 835 while (shift > order) {
c12e51b0 836 shift -= RADIX_TREE_MAP_SHIFT;
89148aa4 837 if (child == NULL) {
1da177e4 838 /* Have to add a child node. */
d58275bc 839 child = radix_tree_node_alloc(gfp, node, root, shift,
e8de4340 840 offset, 0, 0);
89148aa4 841 if (!child)
1da177e4 842 return -ENOMEM;
89148aa4
MW
843 rcu_assign_pointer(*slot, node_to_entry(child));
844 if (node)
1da177e4 845 node->count++;
89148aa4 846 } else if (!radix_tree_is_internal_node(child))
e6145236 847 break;
1da177e4
LT
848
849 /* Go a level down */
89148aa4 850 node = entry_to_node(child);
9e85d811 851 offset = radix_tree_descend(node, &child, index);
89148aa4 852 slot = &node->slots[offset];
e6145236
MW
853 }
854
175542f5
MW
855 if (nodep)
856 *nodep = node;
857 if (slotp)
858 *slotp = slot;
859 return 0;
860}
861
175542f5
MW
862/*
863 * Free any nodes below this node. The tree is presumed to not need
864 * shrinking, and any user data in the tree is presumed to not need a
865 * destructor called on it. If we need to add a destructor, we can
866 * add that functionality later. Note that we may not clear tags or
867 * slots from the tree as an RCU walker may still have a pointer into
868 * this subtree. We could replace the entries with RADIX_TREE_RETRY,
869 * but we'll still have to clear those in rcu_free.
870 */
871static void radix_tree_free_nodes(struct radix_tree_node *node)
872{
873 unsigned offset = 0;
874 struct radix_tree_node *child = entry_to_node(node);
875
876 for (;;) {
12320d0f 877 void *entry = rcu_dereference_raw(child->slots[offset]);
175542f5
MW
878 if (radix_tree_is_internal_node(entry) &&
879 !is_sibling_entry(child, entry)) {
880 child = entry_to_node(entry);
881 offset = 0;
882 continue;
883 }
884 offset++;
885 while (offset == RADIX_TREE_MAP_SIZE) {
886 struct radix_tree_node *old = child;
887 offset = child->offset + 1;
888 child = child->parent;
dd040b6f 889 WARN_ON_ONCE(!list_empty(&old->private_list));
175542f5
MW
890 radix_tree_node_free(old);
891 if (old == entry_to_node(node))
892 return;
893 }
894 }
895}
896
0a835c4f 897#ifdef CONFIG_RADIX_TREE_MULTIORDER
d7b62727
MW
898static inline int insert_entries(struct radix_tree_node *node,
899 void __rcu **slot, void *item, unsigned order, bool replace)
175542f5
MW
900{
901 struct radix_tree_node *child;
902 unsigned i, n, tag, offset, tags = 0;
903
904 if (node) {
e157b555
MW
905 if (order > node->shift)
906 n = 1 << (order - node->shift);
907 else
908 n = 1;
175542f5
MW
909 offset = get_slot_offset(node, slot);
910 } else {
911 n = 1;
912 offset = 0;
913 }
914
915 if (n > 1) {
e6145236 916 offset = offset & ~(n - 1);
89148aa4 917 slot = &node->slots[offset];
175542f5
MW
918 }
919 child = node_to_entry(slot);
920
921 for (i = 0; i < n; i++) {
922 if (slot[i]) {
923 if (replace) {
924 node->count--;
925 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
926 if (tag_get(node, tag, offset + i))
927 tags |= 1 << tag;
928 } else
e6145236
MW
929 return -EEXIST;
930 }
175542f5 931 }
e6145236 932
175542f5 933 for (i = 0; i < n; i++) {
12320d0f 934 struct radix_tree_node *old = rcu_dereference_raw(slot[i]);
175542f5 935 if (i) {
89148aa4 936 rcu_assign_pointer(slot[i], child);
175542f5
MW
937 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
938 if (tags & (1 << tag))
939 tag_clear(node, tag, offset + i);
940 } else {
941 rcu_assign_pointer(slot[i], item);
942 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
943 if (tags & (1 << tag))
944 tag_set(node, tag, offset);
e6145236 945 }
175542f5 946 if (radix_tree_is_internal_node(old) &&
e157b555
MW
947 !is_sibling_entry(node, old) &&
948 (old != RADIX_TREE_RETRY))
175542f5
MW
949 radix_tree_free_nodes(old);
950 if (radix_tree_exceptional_entry(old))
951 node->exceptional--;
612d6c19 952 }
175542f5
MW
953 if (node) {
954 node->count += n;
955 if (radix_tree_exceptional_entry(item))
956 node->exceptional += n;
957 }
958 return n;
139e5616 959}
175542f5 960#else
d7b62727
MW
961static inline int insert_entries(struct radix_tree_node *node,
962 void __rcu **slot, void *item, unsigned order, bool replace)
175542f5
MW
963{
964 if (*slot)
965 return -EEXIST;
966 rcu_assign_pointer(*slot, item);
967 if (node) {
968 node->count++;
969 if (radix_tree_exceptional_entry(item))
970 node->exceptional++;
971 }
972 return 1;
973}
974#endif
139e5616
JW
975
976/**
e6145236 977 * __radix_tree_insert - insert into a radix tree
139e5616
JW
978 * @root: radix tree root
979 * @index: index key
e6145236 980 * @order: key covers the 2^order indices around index
139e5616
JW
981 * @item: item to insert
982 *
983 * Insert an item into the radix tree at position @index.
984 */
e6145236
MW
985int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
986 unsigned order, void *item)
139e5616
JW
987{
988 struct radix_tree_node *node;
d7b62727 989 void __rcu **slot;
139e5616
JW
990 int error;
991
b194d16c 992 BUG_ON(radix_tree_is_internal_node(item));
139e5616 993
e6145236 994 error = __radix_tree_create(root, index, order, &node, &slot);
139e5616
JW
995 if (error)
996 return error;
175542f5
MW
997
998 error = insert_entries(node, slot, item, order, false);
999 if (error < 0)
1000 return error;
201b6264 1001
612d6c19 1002 if (node) {
7b60e9ad 1003 unsigned offset = get_slot_offset(node, slot);
7b60e9ad
MW
1004 BUG_ON(tag_get(node, 0, offset));
1005 BUG_ON(tag_get(node, 1, offset));
1006 BUG_ON(tag_get(node, 2, offset));
612d6c19 1007 } else {
7b60e9ad 1008 BUG_ON(root_tags_get(root));
612d6c19 1009 }
1da177e4 1010
1da177e4
LT
1011 return 0;
1012}
e6145236 1013EXPORT_SYMBOL(__radix_tree_insert);
1da177e4 1014
139e5616
JW
1015/**
1016 * __radix_tree_lookup - lookup an item in a radix tree
1017 * @root: radix tree root
1018 * @index: index key
1019 * @nodep: returns node
1020 * @slotp: returns slot
1021 *
1022 * Lookup and return the item at position @index in the radix
1023 * tree @root.
1024 *
1025 * Until there is more than one item in the tree, no nodes are
1026 * allocated and @root->rnode is used as a direct slot instead of
1027 * pointing to a node, in which case *@nodep will be NULL.
7cf9c2c7 1028 */
35534c86
MW
1029void *__radix_tree_lookup(const struct radix_tree_root *root,
1030 unsigned long index, struct radix_tree_node **nodep,
d7b62727 1031 void __rcu ***slotp)
1da177e4 1032{
139e5616 1033 struct radix_tree_node *node, *parent;
85829954 1034 unsigned long maxindex;
d7b62727 1035 void __rcu **slot;
612d6c19 1036
85829954
MW
1037 restart:
1038 parent = NULL;
d7b62727 1039 slot = (void __rcu **)&root->rnode;
9e85d811 1040 radix_tree_load_root(root, &node, &maxindex);
85829954 1041 if (index > maxindex)
1da177e4
LT
1042 return NULL;
1043
b194d16c 1044 while (radix_tree_is_internal_node(node)) {
85829954 1045 unsigned offset;
1da177e4 1046
85829954
MW
1047 if (node == RADIX_TREE_RETRY)
1048 goto restart;
4dd6c098 1049 parent = entry_to_node(node);
9e85d811 1050 offset = radix_tree_descend(parent, &node, index);
85829954
MW
1051 slot = parent->slots + offset;
1052 }
1da177e4 1053
139e5616
JW
1054 if (nodep)
1055 *nodep = parent;
1056 if (slotp)
1057 *slotp = slot;
1058 return node;
b72b71c6
HS
1059}
1060
1061/**
1062 * radix_tree_lookup_slot - lookup a slot in a radix tree
1063 * @root: radix tree root
1064 * @index: index key
1065 *
1066 * Returns: the slot corresponding to the position @index in the
1067 * radix tree @root. This is useful for update-if-exists operations.
1068 *
1069 * This function can be called under rcu_read_lock iff the slot is not
1070 * modified by radix_tree_replace_slot, otherwise it must be called
1071 * exclusive from other writers. Any dereference of the slot must be done
1072 * using radix_tree_deref_slot.
1073 */
d7b62727 1074void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *root,
35534c86 1075 unsigned long index)
b72b71c6 1076{
d7b62727 1077 void __rcu **slot;
139e5616
JW
1078
1079 if (!__radix_tree_lookup(root, index, NULL, &slot))
1080 return NULL;
1081 return slot;
a4331366 1082}
a4331366
HR
1083EXPORT_SYMBOL(radix_tree_lookup_slot);
1084
1085/**
1086 * radix_tree_lookup - perform lookup operation on a radix tree
1087 * @root: radix tree root
1088 * @index: index key
1089 *
1090 * Lookup the item at the position @index in the radix tree @root.
7cf9c2c7
NP
1091 *
1092 * This function can be called under rcu_read_lock, however the caller
1093 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
1094 * them safely). No RCU barriers are required to access or modify the
1095 * returned item, however.
a4331366 1096 */
35534c86 1097void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index)
a4331366 1098{
139e5616 1099 return __radix_tree_lookup(root, index, NULL, NULL);
1da177e4
LT
1100}
1101EXPORT_SYMBOL(radix_tree_lookup);
1102
0a835c4f 1103static inline void replace_sibling_entries(struct radix_tree_node *node,
d7b62727 1104 void __rcu **slot, int count, int exceptional)
a90eb3a2 1105{
a90eb3a2
MW
1106#ifdef CONFIG_RADIX_TREE_MULTIORDER
1107 void *ptr = node_to_entry(slot);
0a835c4f 1108 unsigned offset = get_slot_offset(node, slot) + 1;
a90eb3a2 1109
0a835c4f 1110 while (offset < RADIX_TREE_MAP_SIZE) {
12320d0f 1111 if (rcu_dereference_raw(node->slots[offset]) != ptr)
a90eb3a2 1112 break;
0a835c4f
MW
1113 if (count < 0) {
1114 node->slots[offset] = NULL;
1115 node->count--;
1116 }
1117 node->exceptional += exceptional;
1118 offset++;
a90eb3a2
MW
1119 }
1120#endif
a90eb3a2
MW
1121}
1122
d7b62727
MW
1123static void replace_slot(void __rcu **slot, void *item,
1124 struct radix_tree_node *node, int count, int exceptional)
f7942430 1125{
0a835c4f
MW
1126 if (WARN_ON_ONCE(radix_tree_is_internal_node(item)))
1127 return;
f7942430 1128
0a835c4f 1129 if (node && (count || exceptional)) {
f4b109c6 1130 node->count += count;
0a835c4f
MW
1131 node->exceptional += exceptional;
1132 replace_sibling_entries(node, slot, count, exceptional);
f4b109c6 1133 }
f7942430
JW
1134
1135 rcu_assign_pointer(*slot, item);
1136}
1137
0a835c4f
MW
1138static bool node_tag_get(const struct radix_tree_root *root,
1139 const struct radix_tree_node *node,
1140 unsigned int tag, unsigned int offset)
a90eb3a2 1141{
0a835c4f
MW
1142 if (node)
1143 return tag_get(node, tag, offset);
1144 return root_tag_get(root, tag);
1145}
a90eb3a2 1146
0a835c4f
MW
1147/*
1148 * IDR users want to be able to store NULL in the tree, so if the slot isn't
1149 * free, don't adjust the count, even if it's transitioning between NULL and
1150 * non-NULL. For the IDA, we mark slots as being IDR_FREE while they still
1151 * have empty bits, but it only stores NULL in slots when they're being
1152 * deleted.
1153 */
1154static int calculate_count(struct radix_tree_root *root,
d7b62727 1155 struct radix_tree_node *node, void __rcu **slot,
0a835c4f
MW
1156 void *item, void *old)
1157{
1158 if (is_idr(root)) {
1159 unsigned offset = get_slot_offset(node, slot);
1160 bool free = node_tag_get(root, node, IDR_FREE, offset);
1161 if (!free)
1162 return 0;
1163 if (!old)
1164 return 1;
a90eb3a2 1165 }
0a835c4f 1166 return !!item - !!old;
a90eb3a2
MW
1167}
1168
6d75f366
JW
1169/**
1170 * __radix_tree_replace - replace item in a slot
4d693d08
JW
1171 * @root: radix tree root
1172 * @node: pointer to tree node
1173 * @slot: pointer to slot in @node
1174 * @item: new item to store in the slot.
1175 * @update_node: callback for changing leaf nodes
6d75f366
JW
1176 *
1177 * For use with __radix_tree_lookup(). Caller must hold tree write locked
1178 * across slot lookup and replacement.
1179 */
1180void __radix_tree_replace(struct radix_tree_root *root,
1181 struct radix_tree_node *node,
d7b62727 1182 void __rcu **slot, void *item,
c7df8ad2 1183 radix_tree_update_node_t update_node)
6d75f366 1184{
0a835c4f
MW
1185 void *old = rcu_dereference_raw(*slot);
1186 int exceptional = !!radix_tree_exceptional_entry(item) -
1187 !!radix_tree_exceptional_entry(old);
1188 int count = calculate_count(root, node, slot, item, old);
1189
6d75f366 1190 /*
f4b109c6
JW
1191 * This function supports replacing exceptional entries and
1192 * deleting entries, but that needs accounting against the
1193 * node unless the slot is root->rnode.
6d75f366 1194 */
d7b62727 1195 WARN_ON_ONCE(!node && (slot != (void __rcu **)&root->rnode) &&
0a835c4f
MW
1196 (count || exceptional));
1197 replace_slot(slot, item, node, count, exceptional);
f4b109c6 1198
4d693d08
JW
1199 if (!node)
1200 return;
1201
1202 if (update_node)
c7df8ad2 1203 update_node(node);
4d693d08 1204
c7df8ad2 1205 delete_node(root, node, update_node);
6d75f366
JW
1206}
1207
1208/**
1209 * radix_tree_replace_slot - replace item in a slot
1210 * @root: radix tree root
1211 * @slot: pointer to slot
1212 * @item: new item to store in the slot.
1213 *
1214 * For use with radix_tree_lookup_slot(), radix_tree_gang_lookup_slot(),
1215 * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked
1216 * across slot lookup and replacement.
1217 *
1218 * NOTE: This cannot be used to switch between non-entries (empty slots),
1219 * regular entries, and exceptional entries, as that requires accounting
f4b109c6 1220 * inside the radix tree node. When switching from one type of entry or
e157b555
MW
1221 * deleting, use __radix_tree_lookup() and __radix_tree_replace() or
1222 * radix_tree_iter_replace().
6d75f366
JW
1223 */
1224void radix_tree_replace_slot(struct radix_tree_root *root,
d7b62727 1225 void __rcu **slot, void *item)
6d75f366 1226{
c7df8ad2 1227 __radix_tree_replace(root, NULL, slot, item, NULL);
6d75f366 1228}
10257d71 1229EXPORT_SYMBOL(radix_tree_replace_slot);
6d75f366 1230
e157b555
MW
1231/**
1232 * radix_tree_iter_replace - replace item in a slot
1233 * @root: radix tree root
1234 * @slot: pointer to slot
1235 * @item: new item to store in the slot.
1236 *
1237 * For use with radix_tree_split() and radix_tree_for_each_slot().
1238 * Caller must hold tree write locked across split and replacement.
1239 */
1240void radix_tree_iter_replace(struct radix_tree_root *root,
d7b62727
MW
1241 const struct radix_tree_iter *iter,
1242 void __rcu **slot, void *item)
e157b555 1243{
c7df8ad2 1244 __radix_tree_replace(root, iter->node, slot, item, NULL);
e157b555
MW
1245}
1246
175542f5
MW
1247#ifdef CONFIG_RADIX_TREE_MULTIORDER
1248/**
1249 * radix_tree_join - replace multiple entries with one multiorder entry
1250 * @root: radix tree root
1251 * @index: an index inside the new entry
1252 * @order: order of the new entry
1253 * @item: new entry
1254 *
1255 * Call this function to replace several entries with one larger entry.
1256 * The existing entries are presumed to not need freeing as a result of
1257 * this call.
1258 *
1259 * The replacement entry will have all the tags set on it that were set
1260 * on any of the entries it is replacing.
1261 */
1262int radix_tree_join(struct radix_tree_root *root, unsigned long index,
1263 unsigned order, void *item)
1264{
1265 struct radix_tree_node *node;
d7b62727 1266 void __rcu **slot;
175542f5
MW
1267 int error;
1268
1269 BUG_ON(radix_tree_is_internal_node(item));
1270
1271 error = __radix_tree_create(root, index, order, &node, &slot);
1272 if (!error)
1273 error = insert_entries(node, slot, item, order, true);
1274 if (error > 0)
1275 error = 0;
1276
1277 return error;
1278}
e157b555
MW
1279
1280/**
1281 * radix_tree_split - Split an entry into smaller entries
1282 * @root: radix tree root
1283 * @index: An index within the large entry
1284 * @order: Order of new entries
1285 *
1286 * Call this function as the first step in replacing a multiorder entry
1287 * with several entries of lower order. After this function returns,
1288 * loop over the relevant portion of the tree using radix_tree_for_each_slot()
1289 * and call radix_tree_iter_replace() to set up each new entry.
1290 *
1291 * The tags from this entry are replicated to all the new entries.
1292 *
1293 * The radix tree should be locked against modification during the entire
1294 * replacement operation. Lock-free lookups will see RADIX_TREE_RETRY which
1295 * should prompt RCU walkers to restart the lookup from the root.
1296 */
1297int radix_tree_split(struct radix_tree_root *root, unsigned long index,
1298 unsigned order)
1299{
1300 struct radix_tree_node *parent, *node, *child;
d7b62727 1301 void __rcu **slot;
e157b555
MW
1302 unsigned int offset, end;
1303 unsigned n, tag, tags = 0;
0a835c4f 1304 gfp_t gfp = root_gfp_mask(root);
e157b555
MW
1305
1306 if (!__radix_tree_lookup(root, index, &parent, &slot))
1307 return -ENOENT;
1308 if (!parent)
1309 return -ENOENT;
1310
1311 offset = get_slot_offset(parent, slot);
1312
1313 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1314 if (tag_get(parent, tag, offset))
1315 tags |= 1 << tag;
1316
1317 for (end = offset + 1; end < RADIX_TREE_MAP_SIZE; end++) {
12320d0f
MW
1318 if (!is_sibling_entry(parent,
1319 rcu_dereference_raw(parent->slots[end])))
e157b555
MW
1320 break;
1321 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1322 if (tags & (1 << tag))
1323 tag_set(parent, tag, end);
1324 /* rcu_assign_pointer ensures tags are set before RETRY */
1325 rcu_assign_pointer(parent->slots[end], RADIX_TREE_RETRY);
1326 }
1327 rcu_assign_pointer(parent->slots[offset], RADIX_TREE_RETRY);
1328 parent->exceptional -= (end - offset);
1329
1330 if (order == parent->shift)
1331 return 0;
1332 if (order > parent->shift) {
1333 while (offset < end)
1334 offset += insert_entries(parent, &parent->slots[offset],
1335 RADIX_TREE_RETRY, order, true);
1336 return 0;
1337 }
1338
1339 node = parent;
1340
1341 for (;;) {
1342 if (node->shift > order) {
d58275bc 1343 child = radix_tree_node_alloc(gfp, node, root,
e8de4340
MW
1344 node->shift - RADIX_TREE_MAP_SHIFT,
1345 offset, 0, 0);
e157b555
MW
1346 if (!child)
1347 goto nomem;
e157b555
MW
1348 if (node != parent) {
1349 node->count++;
12320d0f
MW
1350 rcu_assign_pointer(node->slots[offset],
1351 node_to_entry(child));
e157b555
MW
1352 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1353 if (tags & (1 << tag))
1354 tag_set(node, tag, offset);
1355 }
1356
1357 node = child;
1358 offset = 0;
1359 continue;
1360 }
1361
1362 n = insert_entries(node, &node->slots[offset],
1363 RADIX_TREE_RETRY, order, false);
1364 BUG_ON(n > RADIX_TREE_MAP_SIZE);
1365
1366 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1367 if (tags & (1 << tag))
1368 tag_set(node, tag, offset);
1369 offset += n;
1370
1371 while (offset == RADIX_TREE_MAP_SIZE) {
1372 if (node == parent)
1373 break;
1374 offset = node->offset;
1375 child = node;
1376 node = node->parent;
1377 rcu_assign_pointer(node->slots[offset],
1378 node_to_entry(child));
1379 offset++;
1380 }
1381 if ((node == parent) && (offset == end))
1382 return 0;
1383 }
1384
1385 nomem:
1386 /* Shouldn't happen; did user forget to preload? */
1387 /* TODO: free all the allocated nodes */
1388 WARN_ON(1);
1389 return -ENOMEM;
1390}
175542f5
MW
1391#endif
1392
30b888ba
MW
1393static void node_tag_set(struct radix_tree_root *root,
1394 struct radix_tree_node *node,
1395 unsigned int tag, unsigned int offset)
1396{
1397 while (node) {
1398 if (tag_get(node, tag, offset))
1399 return;
1400 tag_set(node, tag, offset);
1401 offset = node->offset;
1402 node = node->parent;
1403 }
1404
1405 if (!root_tag_get(root, tag))
1406 root_tag_set(root, tag);
1407}
1408
1da177e4
LT
1409/**
1410 * radix_tree_tag_set - set a tag on a radix tree node
1411 * @root: radix tree root
1412 * @index: index key
2fcd9005 1413 * @tag: tag index
1da177e4 1414 *
daff89f3
JC
1415 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
1416 * corresponding to @index in the radix tree. From
1da177e4
LT
1417 * the root all the way down to the leaf node.
1418 *
2fcd9005 1419 * Returns the address of the tagged item. Setting a tag on a not-present
1da177e4
LT
1420 * item is a bug.
1421 */
1422void *radix_tree_tag_set(struct radix_tree_root *root,
daff89f3 1423 unsigned long index, unsigned int tag)
1da177e4 1424{
fb969909
RZ
1425 struct radix_tree_node *node, *parent;
1426 unsigned long maxindex;
1da177e4 1427
9e85d811 1428 radix_tree_load_root(root, &node, &maxindex);
fb969909 1429 BUG_ON(index > maxindex);
1da177e4 1430
b194d16c 1431 while (radix_tree_is_internal_node(node)) {
fb969909 1432 unsigned offset;
1da177e4 1433
4dd6c098 1434 parent = entry_to_node(node);
9e85d811 1435 offset = radix_tree_descend(parent, &node, index);
fb969909
RZ
1436 BUG_ON(!node);
1437
1438 if (!tag_get(parent, tag, offset))
1439 tag_set(parent, tag, offset);
1da177e4
LT
1440 }
1441
612d6c19 1442 /* set the root's tag bit */
fb969909 1443 if (!root_tag_get(root, tag))
612d6c19
NP
1444 root_tag_set(root, tag);
1445
fb969909 1446 return node;
1da177e4
LT
1447}
1448EXPORT_SYMBOL(radix_tree_tag_set);
1449
30b888ba
MW
1450/**
1451 * radix_tree_iter_tag_set - set a tag on the current iterator entry
1452 * @root: radix tree root
1453 * @iter: iterator state
1454 * @tag: tag to set
1455 */
1456void radix_tree_iter_tag_set(struct radix_tree_root *root,
1457 const struct radix_tree_iter *iter, unsigned int tag)
1458{
1459 node_tag_set(root, iter->node, tag, iter_offset(iter));
1460}
1461
d604c324
MW
1462static void node_tag_clear(struct radix_tree_root *root,
1463 struct radix_tree_node *node,
1464 unsigned int tag, unsigned int offset)
1465{
1466 while (node) {
1467 if (!tag_get(node, tag, offset))
1468 return;
1469 tag_clear(node, tag, offset);
1470 if (any_tag_set(node, tag))
1471 return;
1472
1473 offset = node->offset;
1474 node = node->parent;
1475 }
1476
1477 /* clear the root's tag bit */
1478 if (root_tag_get(root, tag))
1479 root_tag_clear(root, tag);
1480}
1481
1da177e4
LT
1482/**
1483 * radix_tree_tag_clear - clear a tag on a radix tree node
1484 * @root: radix tree root
1485 * @index: index key
2fcd9005 1486 * @tag: tag index
1da177e4 1487 *
daff89f3 1488 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
2fcd9005
MW
1489 * corresponding to @index in the radix tree. If this causes
1490 * the leaf node to have no tags set then clear the tag in the
1da177e4
LT
1491 * next-to-leaf node, etc.
1492 *
1493 * Returns the address of the tagged item on success, else NULL. ie:
1494 * has the same return value and semantics as radix_tree_lookup().
1495 */
1496void *radix_tree_tag_clear(struct radix_tree_root *root,
daff89f3 1497 unsigned long index, unsigned int tag)
1da177e4 1498{
00f47b58
RZ
1499 struct radix_tree_node *node, *parent;
1500 unsigned long maxindex;
e2bdb933 1501 int uninitialized_var(offset);
1da177e4 1502
9e85d811 1503 radix_tree_load_root(root, &node, &maxindex);
00f47b58
RZ
1504 if (index > maxindex)
1505 return NULL;
1da177e4 1506
00f47b58 1507 parent = NULL;
1da177e4 1508
b194d16c 1509 while (radix_tree_is_internal_node(node)) {
4dd6c098 1510 parent = entry_to_node(node);
9e85d811 1511 offset = radix_tree_descend(parent, &node, index);
1da177e4
LT
1512 }
1513
d604c324
MW
1514 if (node)
1515 node_tag_clear(root, parent, tag, offset);
1da177e4 1516
00f47b58 1517 return node;
1da177e4
LT
1518}
1519EXPORT_SYMBOL(radix_tree_tag_clear);
1520
30b888ba
MW
1521/**
1522 * radix_tree_iter_tag_clear - clear a tag on the current iterator entry
1523 * @root: radix tree root
1524 * @iter: iterator state
1525 * @tag: tag to clear
1526 */
1527void radix_tree_iter_tag_clear(struct radix_tree_root *root,
1528 const struct radix_tree_iter *iter, unsigned int tag)
1529{
1530 node_tag_clear(root, iter->node, tag, iter_offset(iter));
1531}
1532
1da177e4 1533/**
32605a18
MT
1534 * radix_tree_tag_get - get a tag on a radix tree node
1535 * @root: radix tree root
1536 * @index: index key
2fcd9005 1537 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
1da177e4 1538 *
32605a18 1539 * Return values:
1da177e4 1540 *
612d6c19
NP
1541 * 0: tag not present or not set
1542 * 1: tag set
ce82653d
DH
1543 *
1544 * Note that the return value of this function may not be relied on, even if
1545 * the RCU lock is held, unless tag modification and node deletion are excluded
1546 * from concurrency.
1da177e4 1547 */
35534c86 1548int radix_tree_tag_get(const struct radix_tree_root *root,
daff89f3 1549 unsigned long index, unsigned int tag)
1da177e4 1550{
4589ba6d
RZ
1551 struct radix_tree_node *node, *parent;
1552 unsigned long maxindex;
1da177e4 1553
612d6c19
NP
1554 if (!root_tag_get(root, tag))
1555 return 0;
1556
9e85d811 1557 radix_tree_load_root(root, &node, &maxindex);
4589ba6d
RZ
1558 if (index > maxindex)
1559 return 0;
7cf9c2c7 1560
b194d16c 1561 while (radix_tree_is_internal_node(node)) {
9e85d811 1562 unsigned offset;
1da177e4 1563
4dd6c098 1564 parent = entry_to_node(node);
9e85d811 1565 offset = radix_tree_descend(parent, &node, index);
1da177e4 1566
4589ba6d 1567 if (!tag_get(parent, tag, offset))
3fa36acb 1568 return 0;
4589ba6d
RZ
1569 if (node == RADIX_TREE_RETRY)
1570 break;
1da177e4 1571 }
4589ba6d
RZ
1572
1573 return 1;
1da177e4
LT
1574}
1575EXPORT_SYMBOL(radix_tree_tag_get);
1da177e4 1576
21ef5339
RZ
1577static inline void __set_iter_shift(struct radix_tree_iter *iter,
1578 unsigned int shift)
1579{
1580#ifdef CONFIG_RADIX_TREE_MULTIORDER
1581 iter->shift = shift;
1582#endif
1583}
1584
148deab2
MW
1585/* Construct iter->tags bit-mask from node->tags[tag] array */
1586static void set_iter_tags(struct radix_tree_iter *iter,
1587 struct radix_tree_node *node, unsigned offset,
1588 unsigned tag)
1589{
1590 unsigned tag_long = offset / BITS_PER_LONG;
1591 unsigned tag_bit = offset % BITS_PER_LONG;
1592
0a835c4f
MW
1593 if (!node) {
1594 iter->tags = 1;
1595 return;
1596 }
1597
148deab2
MW
1598 iter->tags = node->tags[tag][tag_long] >> tag_bit;
1599
1600 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1601 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
1602 /* Pick tags from next element */
1603 if (tag_bit)
1604 iter->tags |= node->tags[tag][tag_long + 1] <<
1605 (BITS_PER_LONG - tag_bit);
1606 /* Clip chunk size, here only BITS_PER_LONG tags */
1607 iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG);
1608 }
1609}
1610
1611#ifdef CONFIG_RADIX_TREE_MULTIORDER
d7b62727
MW
1612static void __rcu **skip_siblings(struct radix_tree_node **nodep,
1613 void __rcu **slot, struct radix_tree_iter *iter)
148deab2
MW
1614{
1615 void *sib = node_to_entry(slot - 1);
1616
1617 while (iter->index < iter->next_index) {
1618 *nodep = rcu_dereference_raw(*slot);
1619 if (*nodep && *nodep != sib)
1620 return slot;
1621 slot++;
1622 iter->index = __radix_tree_iter_add(iter, 1);
1623 iter->tags >>= 1;
1624 }
1625
1626 *nodep = NULL;
1627 return NULL;
1628}
1629
d7b62727
MW
1630void __rcu **__radix_tree_next_slot(void __rcu **slot,
1631 struct radix_tree_iter *iter, unsigned flags)
148deab2
MW
1632{
1633 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
1634 struct radix_tree_node *node = rcu_dereference_raw(*slot);
1635
1636 slot = skip_siblings(&node, slot, iter);
1637
1638 while (radix_tree_is_internal_node(node)) {
1639 unsigned offset;
1640 unsigned long next_index;
1641
1642 if (node == RADIX_TREE_RETRY)
1643 return slot;
1644 node = entry_to_node(node);
268f42de 1645 iter->node = node;
148deab2
MW
1646 iter->shift = node->shift;
1647
1648 if (flags & RADIX_TREE_ITER_TAGGED) {
1649 offset = radix_tree_find_next_bit(node, tag, 0);
1650 if (offset == RADIX_TREE_MAP_SIZE)
1651 return NULL;
1652 slot = &node->slots[offset];
1653 iter->index = __radix_tree_iter_add(iter, offset);
1654 set_iter_tags(iter, node, offset, tag);
1655 node = rcu_dereference_raw(*slot);
1656 } else {
1657 offset = 0;
1658 slot = &node->slots[0];
1659 for (;;) {
1660 node = rcu_dereference_raw(*slot);
1661 if (node)
1662 break;
1663 slot++;
1664 offset++;
1665 if (offset == RADIX_TREE_MAP_SIZE)
1666 return NULL;
1667 }
1668 iter->index = __radix_tree_iter_add(iter, offset);
1669 }
1670 if ((flags & RADIX_TREE_ITER_CONTIG) && (offset > 0))
1671 goto none;
1672 next_index = (iter->index | shift_maxindex(iter->shift)) + 1;
1673 if (next_index < iter->next_index)
1674 iter->next_index = next_index;
1675 }
1676
1677 return slot;
1678 none:
1679 iter->next_index = 0;
1680 return NULL;
1681}
1682EXPORT_SYMBOL(__radix_tree_next_slot);
1683#else
d7b62727
MW
1684static void __rcu **skip_siblings(struct radix_tree_node **nodep,
1685 void __rcu **slot, struct radix_tree_iter *iter)
148deab2
MW
1686{
1687 return slot;
1688}
1689#endif
1690
d7b62727
MW
1691void __rcu **radix_tree_iter_resume(void __rcu **slot,
1692 struct radix_tree_iter *iter)
148deab2
MW
1693{
1694 struct radix_tree_node *node;
1695
1696 slot++;
1697 iter->index = __radix_tree_iter_add(iter, 1);
148deab2
MW
1698 skip_siblings(&node, slot, iter);
1699 iter->next_index = iter->index;
1700 iter->tags = 0;
1701 return NULL;
1702}
1703EXPORT_SYMBOL(radix_tree_iter_resume);
1704
78c1d784
KK
1705/**
1706 * radix_tree_next_chunk - find next chunk of slots for iteration
1707 *
1708 * @root: radix tree root
1709 * @iter: iterator state
1710 * @flags: RADIX_TREE_ITER_* flags and tag index
1711 * Returns: pointer to chunk first slot, or NULL if iteration is over
1712 */
d7b62727 1713void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root,
78c1d784
KK
1714 struct radix_tree_iter *iter, unsigned flags)
1715{
9e85d811 1716 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
8c1244de 1717 struct radix_tree_node *node, *child;
21ef5339 1718 unsigned long index, offset, maxindex;
78c1d784
KK
1719
1720 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
1721 return NULL;
1722
1723 /*
1724 * Catch next_index overflow after ~0UL. iter->index never overflows
1725 * during iterating; it can be zero only at the beginning.
1726 * And we cannot overflow iter->next_index in a single step,
1727 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
fffaee36
KK
1728 *
1729 * This condition also used by radix_tree_next_slot() to stop
91b9677c 1730 * contiguous iterating, and forbid switching to the next chunk.
78c1d784
KK
1731 */
1732 index = iter->next_index;
1733 if (!index && iter->index)
1734 return NULL;
1735
21ef5339 1736 restart:
9e85d811 1737 radix_tree_load_root(root, &child, &maxindex);
21ef5339
RZ
1738 if (index > maxindex)
1739 return NULL;
8c1244de
MW
1740 if (!child)
1741 return NULL;
21ef5339 1742
8c1244de 1743 if (!radix_tree_is_internal_node(child)) {
78c1d784 1744 /* Single-slot tree */
21ef5339
RZ
1745 iter->index = index;
1746 iter->next_index = maxindex + 1;
78c1d784 1747 iter->tags = 1;
268f42de 1748 iter->node = NULL;
8c1244de 1749 __set_iter_shift(iter, 0);
d7b62727 1750 return (void __rcu **)&root->rnode;
8c1244de 1751 }
21ef5339 1752
8c1244de
MW
1753 do {
1754 node = entry_to_node(child);
9e85d811 1755 offset = radix_tree_descend(node, &child, index);
21ef5339 1756
78c1d784 1757 if ((flags & RADIX_TREE_ITER_TAGGED) ?
8c1244de 1758 !tag_get(node, tag, offset) : !child) {
78c1d784
KK
1759 /* Hole detected */
1760 if (flags & RADIX_TREE_ITER_CONTIG)
1761 return NULL;
1762
1763 if (flags & RADIX_TREE_ITER_TAGGED)
bc412fca 1764 offset = radix_tree_find_next_bit(node, tag,
78c1d784
KK
1765 offset + 1);
1766 else
1767 while (++offset < RADIX_TREE_MAP_SIZE) {
12320d0f
MW
1768 void *slot = rcu_dereference_raw(
1769 node->slots[offset]);
21ef5339
RZ
1770 if (is_sibling_entry(node, slot))
1771 continue;
1772 if (slot)
78c1d784
KK
1773 break;
1774 }
8c1244de 1775 index &= ~node_maxindex(node);
9e85d811 1776 index += offset << node->shift;
78c1d784
KK
1777 /* Overflow after ~0UL */
1778 if (!index)
1779 return NULL;
1780 if (offset == RADIX_TREE_MAP_SIZE)
1781 goto restart;
8c1244de 1782 child = rcu_dereference_raw(node->slots[offset]);
78c1d784
KK
1783 }
1784
e157b555 1785 if (!child)
78c1d784 1786 goto restart;
e157b555
MW
1787 if (child == RADIX_TREE_RETRY)
1788 break;
8c1244de 1789 } while (radix_tree_is_internal_node(child));
78c1d784
KK
1790
1791 /* Update the iterator state */
8c1244de
MW
1792 iter->index = (index &~ node_maxindex(node)) | (offset << node->shift);
1793 iter->next_index = (index | node_maxindex(node)) + 1;
268f42de 1794 iter->node = node;
9e85d811 1795 __set_iter_shift(iter, node->shift);
78c1d784 1796
148deab2
MW
1797 if (flags & RADIX_TREE_ITER_TAGGED)
1798 set_iter_tags(iter, node, offset, tag);
78c1d784
KK
1799
1800 return node->slots + offset;
1801}
1802EXPORT_SYMBOL(radix_tree_next_chunk);
1803
1da177e4
LT
1804/**
1805 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1806 * @root: radix tree root
1807 * @results: where the results of the lookup are placed
1808 * @first_index: start the lookup from this key
1809 * @max_items: place up to this many items at *results
1810 *
1811 * Performs an index-ascending scan of the tree for present items. Places
1812 * them at *@results and returns the number of items which were placed at
1813 * *@results.
1814 *
1815 * The implementation is naive.
7cf9c2c7
NP
1816 *
1817 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1818 * rcu_read_lock. In this case, rather than the returned results being
2fcd9005
MW
1819 * an atomic snapshot of the tree at a single point in time, the
1820 * semantics of an RCU protected gang lookup are as though multiple
1821 * radix_tree_lookups have been issued in individual locks, and results
1822 * stored in 'results'.
1da177e4
LT
1823 */
1824unsigned int
35534c86 1825radix_tree_gang_lookup(const struct radix_tree_root *root, void **results,
1da177e4
LT
1826 unsigned long first_index, unsigned int max_items)
1827{
cebbd29e 1828 struct radix_tree_iter iter;
d7b62727 1829 void __rcu **slot;
cebbd29e 1830 unsigned int ret = 0;
7cf9c2c7 1831
cebbd29e 1832 if (unlikely(!max_items))
7cf9c2c7 1833 return 0;
1da177e4 1834
cebbd29e 1835 radix_tree_for_each_slot(slot, root, &iter, first_index) {
46437f9a 1836 results[ret] = rcu_dereference_raw(*slot);
cebbd29e
KK
1837 if (!results[ret])
1838 continue;
b194d16c 1839 if (radix_tree_is_internal_node(results[ret])) {
46437f9a
MW
1840 slot = radix_tree_iter_retry(&iter);
1841 continue;
1842 }
cebbd29e 1843 if (++ret == max_items)
1da177e4 1844 break;
1da177e4 1845 }
7cf9c2c7 1846
1da177e4
LT
1847 return ret;
1848}
1849EXPORT_SYMBOL(radix_tree_gang_lookup);
1850
47feff2c
NP
1851/**
1852 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1853 * @root: radix tree root
1854 * @results: where the results of the lookup are placed
6328650b 1855 * @indices: where their indices should be placed (but usually NULL)
47feff2c
NP
1856 * @first_index: start the lookup from this key
1857 * @max_items: place up to this many items at *results
1858 *
1859 * Performs an index-ascending scan of the tree for present items. Places
1860 * their slots at *@results and returns the number of items which were
1861 * placed at *@results.
1862 *
1863 * The implementation is naive.
1864 *
1865 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1866 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1867 * protection, radix_tree_deref_slot may fail requiring a retry.
1868 */
1869unsigned int
35534c86 1870radix_tree_gang_lookup_slot(const struct radix_tree_root *root,
d7b62727 1871 void __rcu ***results, unsigned long *indices,
47feff2c
NP
1872 unsigned long first_index, unsigned int max_items)
1873{
cebbd29e 1874 struct radix_tree_iter iter;
d7b62727 1875 void __rcu **slot;
cebbd29e 1876 unsigned int ret = 0;
47feff2c 1877
cebbd29e 1878 if (unlikely(!max_items))
47feff2c
NP
1879 return 0;
1880
cebbd29e
KK
1881 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1882 results[ret] = slot;
6328650b 1883 if (indices)
cebbd29e
KK
1884 indices[ret] = iter.index;
1885 if (++ret == max_items)
47feff2c 1886 break;
47feff2c
NP
1887 }
1888
1889 return ret;
1890}
1891EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1892
1da177e4
LT
1893/**
1894 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1895 * based on a tag
1896 * @root: radix tree root
1897 * @results: where the results of the lookup are placed
1898 * @first_index: start the lookup from this key
1899 * @max_items: place up to this many items at *results
daff89f3 1900 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1da177e4
LT
1901 *
1902 * Performs an index-ascending scan of the tree for present items which
1903 * have the tag indexed by @tag set. Places the items at *@results and
1904 * returns the number of items which were placed at *@results.
1905 */
1906unsigned int
35534c86 1907radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results,
daff89f3
JC
1908 unsigned long first_index, unsigned int max_items,
1909 unsigned int tag)
1da177e4 1910{
cebbd29e 1911 struct radix_tree_iter iter;
d7b62727 1912 void __rcu **slot;
cebbd29e 1913 unsigned int ret = 0;
612d6c19 1914
cebbd29e 1915 if (unlikely(!max_items))
7cf9c2c7
NP
1916 return 0;
1917
cebbd29e 1918 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
46437f9a 1919 results[ret] = rcu_dereference_raw(*slot);
cebbd29e
KK
1920 if (!results[ret])
1921 continue;
b194d16c 1922 if (radix_tree_is_internal_node(results[ret])) {
46437f9a
MW
1923 slot = radix_tree_iter_retry(&iter);
1924 continue;
1925 }
cebbd29e 1926 if (++ret == max_items)
1da177e4 1927 break;
1da177e4 1928 }
7cf9c2c7 1929
1da177e4
LT
1930 return ret;
1931}
1932EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1933
47feff2c
NP
1934/**
1935 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1936 * radix tree based on a tag
1937 * @root: radix tree root
1938 * @results: where the results of the lookup are placed
1939 * @first_index: start the lookup from this key
1940 * @max_items: place up to this many items at *results
1941 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1942 *
1943 * Performs an index-ascending scan of the tree for present items which
1944 * have the tag indexed by @tag set. Places the slots at *@results and
1945 * returns the number of slots which were placed at *@results.
1946 */
1947unsigned int
35534c86 1948radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root,
d7b62727 1949 void __rcu ***results, unsigned long first_index,
35534c86 1950 unsigned int max_items, unsigned int tag)
47feff2c 1951{
cebbd29e 1952 struct radix_tree_iter iter;
d7b62727 1953 void __rcu **slot;
cebbd29e 1954 unsigned int ret = 0;
47feff2c 1955
cebbd29e 1956 if (unlikely(!max_items))
47feff2c
NP
1957 return 0;
1958
cebbd29e
KK
1959 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1960 results[ret] = slot;
1961 if (++ret == max_items)
47feff2c 1962 break;
47feff2c
NP
1963 }
1964
1965 return ret;
1966}
1967EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1968
139e5616
JW
1969/**
1970 * __radix_tree_delete_node - try to free node after clearing a slot
1971 * @root: radix tree root
139e5616 1972 * @node: node containing @index
ea07b862 1973 * @update_node: callback for changing leaf nodes
139e5616
JW
1974 *
1975 * After clearing the slot at @index in @node from radix tree
1976 * rooted at @root, call this function to attempt freeing the
1977 * node and shrinking the tree.
139e5616 1978 */
14b46879 1979void __radix_tree_delete_node(struct radix_tree_root *root,
ea07b862 1980 struct radix_tree_node *node,
c7df8ad2 1981 radix_tree_update_node_t update_node)
139e5616 1982{
c7df8ad2 1983 delete_node(root, node, update_node);
139e5616
JW
1984}
1985
0ac398ef 1986static bool __radix_tree_delete(struct radix_tree_root *root,
d7b62727 1987 struct radix_tree_node *node, void __rcu **slot)
0ac398ef 1988{
0a835c4f
MW
1989 void *old = rcu_dereference_raw(*slot);
1990 int exceptional = radix_tree_exceptional_entry(old) ? -1 : 0;
0ac398ef
MW
1991 unsigned offset = get_slot_offset(node, slot);
1992 int tag;
1993
0a835c4f
MW
1994 if (is_idr(root))
1995 node_tag_set(root, node, IDR_FREE, offset);
1996 else
1997 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1998 node_tag_clear(root, node, tag, offset);
0ac398ef 1999
0a835c4f 2000 replace_slot(slot, NULL, node, -1, exceptional);
c7df8ad2 2001 return node && delete_node(root, node, NULL);
0ac398ef
MW
2002}
2003
1da177e4 2004/**
0ac398ef
MW
2005 * radix_tree_iter_delete - delete the entry at this iterator position
2006 * @root: radix tree root
2007 * @iter: iterator state
2008 * @slot: pointer to slot
1da177e4 2009 *
0ac398ef
MW
2010 * Delete the entry at the position currently pointed to by the iterator.
2011 * This may result in the current node being freed; if it is, the iterator
2012 * is advanced so that it will not reference the freed memory. This
2013 * function may be called without any locking if there are no other threads
2014 * which can access this tree.
2015 */
2016void radix_tree_iter_delete(struct radix_tree_root *root,
d7b62727 2017 struct radix_tree_iter *iter, void __rcu **slot)
0ac398ef
MW
2018{
2019 if (__radix_tree_delete(root, iter->node, slot))
2020 iter->index = iter->next_index;
2021}
d1b48c1e 2022EXPORT_SYMBOL(radix_tree_iter_delete);
0ac398ef
MW
2023
2024/**
2025 * radix_tree_delete_item - delete an item from a radix tree
2026 * @root: radix tree root
2027 * @index: index key
2028 * @item: expected item
1da177e4 2029 *
0ac398ef 2030 * Remove @item at @index from the radix tree rooted at @root.
1da177e4 2031 *
0ac398ef
MW
2032 * Return: the deleted entry, or %NULL if it was not present
2033 * or the entry at the given @index was not @item.
1da177e4 2034 */
53c59f26
JW
2035void *radix_tree_delete_item(struct radix_tree_root *root,
2036 unsigned long index, void *item)
1da177e4 2037{
0a835c4f 2038 struct radix_tree_node *node = NULL;
d7b62727 2039 void __rcu **slot;
139e5616 2040 void *entry;
1da177e4 2041
139e5616 2042 entry = __radix_tree_lookup(root, index, &node, &slot);
0a835c4f
MW
2043 if (!entry && (!is_idr(root) || node_tag_get(root, node, IDR_FREE,
2044 get_slot_offset(node, slot))))
139e5616 2045 return NULL;
1da177e4 2046
139e5616
JW
2047 if (item && entry != item)
2048 return NULL;
2049
0ac398ef 2050 __radix_tree_delete(root, node, slot);
612d6c19 2051
139e5616 2052 return entry;
1da177e4 2053}
53c59f26
JW
2054EXPORT_SYMBOL(radix_tree_delete_item);
2055
2056/**
0ac398ef
MW
2057 * radix_tree_delete - delete an entry from a radix tree
2058 * @root: radix tree root
2059 * @index: index key
53c59f26 2060 *
0ac398ef 2061 * Remove the entry at @index from the radix tree rooted at @root.
53c59f26 2062 *
0ac398ef 2063 * Return: The deleted entry, or %NULL if it was not present.
53c59f26
JW
2064 */
2065void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
2066{
2067 return radix_tree_delete_item(root, index, NULL);
2068}
1da177e4
LT
2069EXPORT_SYMBOL(radix_tree_delete);
2070
d3798ae8
JW
2071void radix_tree_clear_tags(struct radix_tree_root *root,
2072 struct radix_tree_node *node,
d7b62727 2073 void __rcu **slot)
d604c324 2074{
d604c324
MW
2075 if (node) {
2076 unsigned int tag, offset = get_slot_offset(node, slot);
2077 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
2078 node_tag_clear(root, node, tag, offset);
2079 } else {
0a835c4f 2080 root_tag_clear_all(root);
d604c324 2081 }
d604c324
MW
2082}
2083
1da177e4
LT
2084/**
2085 * radix_tree_tagged - test whether any items in the tree are tagged
2086 * @root: radix tree root
2087 * @tag: tag to test
2088 */
35534c86 2089int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag)
1da177e4 2090{
612d6c19 2091 return root_tag_get(root, tag);
1da177e4
LT
2092}
2093EXPORT_SYMBOL(radix_tree_tagged);
2094
0a835c4f
MW
2095/**
2096 * idr_preload - preload for idr_alloc()
2097 * @gfp_mask: allocation mask to use for preloading
2098 *
2099 * Preallocate memory to use for the next call to idr_alloc(). This function
2100 * returns with preemption disabled. It will be enabled by idr_preload_end().
2101 */
2102void idr_preload(gfp_t gfp_mask)
2103{
bc9ae224
ED
2104 if (__radix_tree_preload(gfp_mask, IDR_PRELOAD_SIZE))
2105 preempt_disable();
0a835c4f
MW
2106}
2107EXPORT_SYMBOL(idr_preload);
2108
7ad3d4d8
MW
2109/**
2110 * ida_pre_get - reserve resources for ida allocation
2111 * @ida: ida handle
2112 * @gfp: memory allocation flags
2113 *
2114 * This function should be called before calling ida_get_new_above(). If it
2115 * is unable to allocate memory, it will return %0. On success, it returns %1.
2116 */
2117int ida_pre_get(struct ida *ida, gfp_t gfp)
2118{
7ad3d4d8
MW
2119 /*
2120 * The IDA API has no preload_end() equivalent. Instead,
2121 * ida_get_new() can return -EAGAIN, prompting the caller
2122 * to return to the ida_pre_get() step.
2123 */
bc9ae224
ED
2124 if (!__radix_tree_preload(gfp, IDA_PRELOAD_SIZE))
2125 preempt_enable();
7ad3d4d8
MW
2126
2127 if (!this_cpu_read(ida_bitmap)) {
b1a8a7a7 2128 struct ida_bitmap *bitmap = kzalloc(sizeof(*bitmap), gfp);
7ad3d4d8
MW
2129 if (!bitmap)
2130 return 0;
4ecd9542
MW
2131 if (this_cpu_cmpxchg(ida_bitmap, NULL, bitmap))
2132 kfree(bitmap);
7ad3d4d8
MW
2133 }
2134
2135 return 1;
2136}
2137EXPORT_SYMBOL(ida_pre_get);
2138
460488c5 2139void __rcu **idr_get_free(struct radix_tree_root *root,
388f79fd
CM
2140 struct radix_tree_iter *iter, gfp_t gfp,
2141 unsigned long max)
0a835c4f
MW
2142{
2143 struct radix_tree_node *node = NULL, *child;
d7b62727 2144 void __rcu **slot = (void __rcu **)&root->rnode;
0a835c4f 2145 unsigned long maxindex, start = iter->next_index;
0a835c4f
MW
2146 unsigned int shift, offset = 0;
2147
2148 grow:
2149 shift = radix_tree_load_root(root, &child, &maxindex);
2150 if (!radix_tree_tagged(root, IDR_FREE))
2151 start = max(start, maxindex + 1);
2152 if (start > max)
2153 return ERR_PTR(-ENOSPC);
2154
2155 if (start > maxindex) {
2156 int error = radix_tree_extend(root, gfp, start, shift);
2157 if (error < 0)
2158 return ERR_PTR(error);
2159 shift = error;
2160 child = rcu_dereference_raw(root->rnode);
2161 }
2162
2163 while (shift) {
2164 shift -= RADIX_TREE_MAP_SHIFT;
2165 if (child == NULL) {
2166 /* Have to add a child node. */
d58275bc
MW
2167 child = radix_tree_node_alloc(gfp, node, root, shift,
2168 offset, 0, 0);
0a835c4f
MW
2169 if (!child)
2170 return ERR_PTR(-ENOMEM);
2171 all_tag_set(child, IDR_FREE);
2172 rcu_assign_pointer(*slot, node_to_entry(child));
2173 if (node)
2174 node->count++;
2175 } else if (!radix_tree_is_internal_node(child))
2176 break;
2177
2178 node = entry_to_node(child);
2179 offset = radix_tree_descend(node, &child, start);
2180 if (!tag_get(node, IDR_FREE, offset)) {
2181 offset = radix_tree_find_next_bit(node, IDR_FREE,
2182 offset + 1);
2183 start = next_index(start, node, offset);
2184 if (start > max)
2185 return ERR_PTR(-ENOSPC);
2186 while (offset == RADIX_TREE_MAP_SIZE) {
2187 offset = node->offset + 1;
2188 node = node->parent;
2189 if (!node)
2190 goto grow;
2191 shift = node->shift;
2192 }
2193 child = rcu_dereference_raw(node->slots[offset]);
2194 }
2195 slot = &node->slots[offset];
2196 }
2197
2198 iter->index = start;
2199 if (node)
2200 iter->next_index = 1 + min(max, (start | node_maxindex(node)));
2201 else
2202 iter->next_index = 1;
2203 iter->node = node;
2204 __set_iter_shift(iter, shift);
2205 set_iter_tags(iter, node, offset, IDR_FREE);
2206
2207 return slot;
2208}
2209
2210/**
2211 * idr_destroy - release all internal memory from an IDR
2212 * @idr: idr handle
2213 *
2214 * After this function is called, the IDR is empty, and may be reused or
2215 * the data structure containing it may be freed.
2216 *
2217 * A typical clean-up sequence for objects stored in an idr tree will use
2218 * idr_for_each() to free all objects, if necessary, then idr_destroy() to
2219 * free the memory used to keep track of those objects.
2220 */
2221void idr_destroy(struct idr *idr)
2222{
2223 struct radix_tree_node *node = rcu_dereference_raw(idr->idr_rt.rnode);
2224 if (radix_tree_is_internal_node(node))
2225 radix_tree_free_nodes(node);
2226 idr->idr_rt.rnode = NULL;
2227 root_tag_set(&idr->idr_rt, IDR_FREE);
2228}
2229EXPORT_SYMBOL(idr_destroy);
2230
1da177e4 2231static void
449dd698 2232radix_tree_node_ctor(void *arg)
1da177e4 2233{
449dd698
JW
2234 struct radix_tree_node *node = arg;
2235
2236 memset(node, 0, sizeof(*node));
2237 INIT_LIST_HEAD(&node->private_list);
1da177e4
LT
2238}
2239
c78c66d1
KS
2240static __init unsigned long __maxindex(unsigned int height)
2241{
2242 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
2243 int shift = RADIX_TREE_INDEX_BITS - width;
2244
2245 if (shift < 0)
2246 return ~0UL;
2247 if (shift >= BITS_PER_LONG)
2248 return 0UL;
2249 return ~0UL >> shift;
2250}
2251
2252static __init void radix_tree_init_maxnodes(void)
2253{
2254 unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1];
2255 unsigned int i, j;
2256
2257 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
2258 height_to_maxindex[i] = __maxindex(i);
2259 for (i = 0; i < ARRAY_SIZE(height_to_maxnodes); i++) {
2260 for (j = i; j > 0; j--)
2261 height_to_maxnodes[i] += height_to_maxindex[j - 1] + 1;
2262 }
2263}
2264
d544abd5 2265static int radix_tree_cpu_dead(unsigned int cpu)
1da177e4 2266{
2fcd9005
MW
2267 struct radix_tree_preload *rtp;
2268 struct radix_tree_node *node;
2269
2270 /* Free per-cpu pool of preloaded nodes */
d544abd5
SAS
2271 rtp = &per_cpu(radix_tree_preloads, cpu);
2272 while (rtp->nr) {
2273 node = rtp->nodes;
1293d5c5 2274 rtp->nodes = node->parent;
d544abd5
SAS
2275 kmem_cache_free(radix_tree_node_cachep, node);
2276 rtp->nr--;
2fcd9005 2277 }
7ad3d4d8
MW
2278 kfree(per_cpu(ida_bitmap, cpu));
2279 per_cpu(ida_bitmap, cpu) = NULL;
d544abd5 2280 return 0;
1da177e4 2281}
1da177e4
LT
2282
2283void __init radix_tree_init(void)
2284{
d544abd5 2285 int ret;
7e784422
MH
2286
2287 BUILD_BUG_ON(RADIX_TREE_MAX_TAGS + __GFP_BITS_SHIFT > 32);
1da177e4
LT
2288 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
2289 sizeof(struct radix_tree_node), 0,
488514d1
CL
2290 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
2291 radix_tree_node_ctor);
c78c66d1 2292 radix_tree_init_maxnodes();
d544abd5
SAS
2293 ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead",
2294 NULL, radix_tree_cpu_dead);
2295 WARN_ON(ret < 0);
1da177e4 2296}