1/*
2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
4 * Copyright (C) 2006 Nick Piggin
5 * Copyright (C) 2012 Konstantin Khlebnikov
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License as
9 * published by the Free Software Foundation; either version 2, or (at
10 * your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 */
21#ifndef _LINUX_RADIX_TREE_H
22#define _LINUX_RADIX_TREE_H
23
24#include <linux/bitops.h>
25#include <linux/kernel.h>
26#include <linux/list.h>
27#include <linux/preempt.h>
28#include <linux/rcupdate.h>
29#include <linux/spinlock.h>
30#include <linux/types.h>
31#include <linux/xarray.h>
32
33/* Keep unconverted code working */
34#define radix_tree_root xarray
35#define radix_tree_node xa_node
36
37/*
38 * The bottom two bits of the slot determine how the remaining bits in the
39 * slot are interpreted:
40 *
41 * 00 - data pointer
42 * 10 - internal entry
43 * x1 - value entry
44 *
45 * The internal entry may be a pointer to the next level in the tree, a
46 * sibling entry, or an indicator that the entry in this slot has been moved
47 * to another location in the tree and the lookup should be restarted. While
48 * NULL fits the 'data pointer' pattern, it means that there is no entry in
49 * the tree for this index (no matter what level of the tree it is found at).
50 * This means that storing a NULL entry in the tree is the same as deleting
51 * the entry from the tree.
52 */
53#define RADIX_TREE_ENTRY_MASK 3UL
54#define RADIX_TREE_INTERNAL_NODE 2UL
55
56static inline bool radix_tree_is_internal_node(void *ptr)
57{
58 return ((unsigned long)ptr & RADIX_TREE_ENTRY_MASK) ==
59 RADIX_TREE_INTERNAL_NODE;
60}
61
62/*** radix-tree API starts here ***/
63
64#define RADIX_TREE_MAP_SHIFT XA_CHUNK_SHIFT
65#define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
66#define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
67
68#define RADIX_TREE_MAX_TAGS XA_MAX_MARKS
69#define RADIX_TREE_TAG_LONGS XA_MARK_LONGS
70
71#define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
72#define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
73 RADIX_TREE_MAP_SHIFT))
74
75/* The IDR tag is stored in the low bits of xa_flags */
76#define ROOT_IS_IDR ((__force gfp_t)4)
77/* The top bits of xa_flags are used to store the root tags */
78#define ROOT_TAG_SHIFT (__GFP_BITS_SHIFT)
79
80#define RADIX_TREE_INIT(name, mask) XARRAY_INIT(name, mask)
81
82#define RADIX_TREE(name, mask) \
83 struct radix_tree_root name = RADIX_TREE_INIT(name, mask)
84
85#define INIT_RADIX_TREE(root, mask) xa_init_flags(root, mask)
86
87static inline bool radix_tree_empty(const struct radix_tree_root *root)
88{
89 return root->xa_head == NULL;
90}
91
92/**
93 * struct radix_tree_iter - radix tree iterator state
94 *
95 * @index: index of current slot
96 * @next_index: one beyond the last index for this chunk
97 * @tags: bit-mask for tag-iterating
98 * @node: node that contains current slot
99 *
100 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a
101 * subinterval of slots contained within one radix tree leaf node. It is
102 * described by a pointer to its first slot and a struct radix_tree_iter
103 * which holds the chunk's position in the tree and its size. For tagged
104 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
105 * radix tree tag.
106 */
107struct radix_tree_iter {
108 unsigned long index;
109 unsigned long next_index;
110 unsigned long tags;
111 struct radix_tree_node *node;
112};
113
114/**
115 * Radix-tree synchronization
116 *
117 * The radix-tree API requires that users provide all synchronisation (with
118 * specific exceptions, noted below).
119 *
120 * Synchronization of access to the data items being stored in the tree, and
121 * management of their lifetimes must be completely managed by API users.
122 *
123 * For API usage, in general,
124 * - any function _modifying_ the tree or tags (inserting or deleting
125 * items, setting or clearing tags) must exclude other modifications, and
126 * exclude any functions reading the tree.
127 * - any function _reading_ the tree or tags (looking up items or tags,
128 * gang lookups) must exclude modifications to the tree, but may occur
129 * concurrently with other readers.
130 *
131 * The notable exceptions to this rule are the following functions:
132 * __radix_tree_lookup
133 * radix_tree_lookup
134 * radix_tree_lookup_slot
135 * radix_tree_tag_get
136 * radix_tree_gang_lookup
137 * radix_tree_gang_lookup_tag
138 * radix_tree_gang_lookup_tag_slot
139 * radix_tree_tagged
140 *
141 * The first 7 functions are able to be called locklessly, using RCU. The
142 * caller must ensure calls to these functions are made within rcu_read_lock()
143 * regions. Other readers (lock-free or otherwise) and modifications may be
144 * running concurrently.
145 *
146 * It is still required that the caller manage the synchronization and lifetimes
147 * of the items. So if RCU lock-free lookups are used, typically this would mean
148 * that the items have their own locks, or are amenable to lock-free access; and
149 * that the items are freed by RCU (or only freed after having been deleted from
150 * the radix tree *and* a synchronize_rcu() grace period).
151 *
152 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
153 * access to data items when inserting into or looking up from the radix tree)
154 *
155 * Note that the value returned by radix_tree_tag_get() may not be relied upon
156 * if only the RCU read lock is held. Functions to set/clear tags and to
157 * delete nodes running concurrently with it may affect its result such that
158 * two consecutive reads in the same locked section may return different
159 * values. If reliability is required, modification functions must also be
160 * excluded from concurrency.
161 *
162 * radix_tree_tagged is able to be called without locking or RCU.
163 */
164
165/**
166 * radix_tree_deref_slot - dereference a slot
167 * @slot: slot pointer, returned by radix_tree_lookup_slot
168 *
169 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read
170 * locked across slot lookup and dereference. Not required if write lock is
171 * held (ie. items cannot be concurrently inserted).
172 *
173 * radix_tree_deref_retry must be used to confirm validity of the pointer if
174 * only the read lock is held.
175 *
176 * Return: entry stored in that slot.
177 */
178static inline void *radix_tree_deref_slot(void __rcu **slot)
179{
180 return rcu_dereference(*slot);
181}
182
183/**
184 * radix_tree_deref_slot_protected - dereference a slot with tree lock held
185 * @slot: slot pointer, returned by radix_tree_lookup_slot
186 *
187 * Similar to radix_tree_deref_slot. The caller does not hold the RCU read
188 * lock but it must hold the tree lock to prevent parallel updates.
189 *
190 * Return: entry stored in that slot.
191 */
192static inline void *radix_tree_deref_slot_protected(void __rcu **slot,
193 spinlock_t *treelock)
194{
195 return rcu_dereference_protected(*slot, lockdep_is_held(treelock));
196}
197
198/**
199 * radix_tree_deref_retry - check radix_tree_deref_slot
200 * @arg: pointer returned by radix_tree_deref_slot
201 * Returns: 0 if retry is not required, otherwise retry is required
202 *
203 * radix_tree_deref_retry must be used with radix_tree_deref_slot.
204 */
205static inline int radix_tree_deref_retry(void *arg)
206{
207 return unlikely(radix_tree_is_internal_node(arg));
208}
209
210/**
211 * radix_tree_exception - radix_tree_deref_slot returned either exception?
212 * @arg: value returned by radix_tree_deref_slot
213 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception.
214 */
215static inline int radix_tree_exception(void *arg)
216{
217 return unlikely((unsigned long)arg & RADIX_TREE_ENTRY_MASK);
218}
219
220int radix_tree_insert(struct radix_tree_root *, unsigned long index,
221 void *);
222void *__radix_tree_lookup(const struct radix_tree_root *, unsigned long index,
223 struct radix_tree_node **nodep, void __rcu ***slotp);
224void *radix_tree_lookup(const struct radix_tree_root *, unsigned long);
225void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *,
226 unsigned long index);
227void __radix_tree_replace(struct radix_tree_root *, struct radix_tree_node *,
228 void __rcu **slot, void *entry);
229void radix_tree_iter_replace(struct radix_tree_root *,
230 const struct radix_tree_iter *, void __rcu **slot, void *entry);
231void radix_tree_replace_slot(struct radix_tree_root *,
232 void __rcu **slot, void *entry);
233void radix_tree_iter_delete(struct radix_tree_root *,
234 struct radix_tree_iter *iter, void __rcu **slot);
235void *radix_tree_delete_item(struct radix_tree_root *, unsigned long, void *);
236void *radix_tree_delete(struct radix_tree_root *, unsigned long);
237unsigned int radix_tree_gang_lookup(const struct radix_tree_root *,
238 void **results, unsigned long first_index,
239 unsigned int max_items);
240int radix_tree_preload(gfp_t gfp_mask);
241int radix_tree_maybe_preload(gfp_t gfp_mask);
242void radix_tree_init(void);
243void *radix_tree_tag_set(struct radix_tree_root *,
244 unsigned long index, unsigned int tag);
245void *radix_tree_tag_clear(struct radix_tree_root *,
246 unsigned long index, unsigned int tag);
247int radix_tree_tag_get(const struct radix_tree_root *,
248 unsigned long index, unsigned int tag);
249void radix_tree_iter_tag_clear(struct radix_tree_root *,
250 const struct radix_tree_iter *iter, unsigned int tag);
251unsigned int radix_tree_gang_lookup_tag(const struct radix_tree_root *,
252 void **results, unsigned long first_index,
253 unsigned int max_items, unsigned int tag);
254unsigned int radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *,
255 void __rcu ***results, unsigned long first_index,
256 unsigned int max_items, unsigned int tag);
257int radix_tree_tagged(const struct radix_tree_root *, unsigned int tag);
258
259static inline void radix_tree_preload_end(void)
260{
261 preempt_enable();
262}
263
264void __rcu **idr_get_free(struct radix_tree_root *root,
265 struct radix_tree_iter *iter, gfp_t gfp,
266 unsigned long max);
267
268enum {
269 RADIX_TREE_ITER_TAG_MASK = 0x0f, /* tag index in lower nybble */
270 RADIX_TREE_ITER_TAGGED = 0x10, /* lookup tagged slots */
271 RADIX_TREE_ITER_CONTIG = 0x20, /* stop at first hole */
272};
273
274/**
275 * radix_tree_iter_init - initialize radix tree iterator
276 *
277 * @iter: pointer to iterator state
278 * @start: iteration starting index
279 * Returns: NULL
280 */
281static __always_inline void __rcu **
282radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start)
283{
284 /*
285 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
286 * in the case of a successful tagged chunk lookup. If the lookup was
287 * unsuccessful or non-tagged then nobody cares about ->tags.
288 *
289 * Set index to zero to bypass next_index overflow protection.
290 * See the comment in radix_tree_next_chunk() for details.
291 */
292 iter->index = 0;
293 iter->next_index = start;
294 return NULL;
295}
296
297/**
298 * radix_tree_next_chunk - find next chunk of slots for iteration
299 *
300 * @root: radix tree root
301 * @iter: iterator state
302 * @flags: RADIX_TREE_ITER_* flags and tag index
303 * Returns: pointer to chunk first slot, or NULL if there no more left
304 *
305 * This function looks up the next chunk in the radix tree starting from
306 * @iter->next_index. It returns a pointer to the chunk's first slot.
307 * Also it fills @iter with data about chunk: position in the tree (index),
308 * its end (next_index), and constructs a bit mask for tagged iterating (tags).
309 */
310void __rcu **radix_tree_next_chunk(const struct radix_tree_root *,
311 struct radix_tree_iter *iter, unsigned flags);
312
313/**
314 * radix_tree_iter_lookup - look up an index in the radix tree
315 * @root: radix tree root
316 * @iter: iterator state
317 * @index: key to look up
318 *
319 * If @index is present in the radix tree, this function returns the slot
320 * containing it and updates @iter to describe the entry. If @index is not
321 * present, it returns NULL.
322 */
323static inline void __rcu **
324radix_tree_iter_lookup(const struct radix_tree_root *root,
325 struct radix_tree_iter *iter, unsigned long index)
326{
327 radix_tree_iter_init(iter, index);
328 return radix_tree_next_chunk(root, iter, RADIX_TREE_ITER_CONTIG);
329}
330
331/**
332 * radix_tree_iter_find - find a present entry
333 * @root: radix tree root
334 * @iter: iterator state
335 * @index: start location
336 *
337 * This function returns the slot containing the entry with the lowest index
338 * which is at least @index. If @index is larger than any present entry, this
339 * function returns NULL. The @iter is updated to describe the entry found.
340 */
341static inline void __rcu **
342radix_tree_iter_find(const struct radix_tree_root *root,
343 struct radix_tree_iter *iter, unsigned long index)
344{
345 radix_tree_iter_init(iter, index);
346 return radix_tree_next_chunk(root, iter, 0);
347}
348
349/**
350 * radix_tree_iter_retry - retry this chunk of the iteration
351 * @iter: iterator state
352 *
353 * If we iterate over a tree protected only by the RCU lock, a race
354 * against deletion or creation may result in seeing a slot for which
355 * radix_tree_deref_retry() returns true. If so, call this function
356 * and continue the iteration.
357 */
358static inline __must_check
359void __rcu **radix_tree_iter_retry(struct radix_tree_iter *iter)
360{
361 iter->next_index = iter->index;
362 iter->tags = 0;
363 return NULL;
364}
365
366static inline unsigned long
367__radix_tree_iter_add(struct radix_tree_iter *iter, unsigned long slots)
368{
369 return iter->index + slots;
370}
371
372/**
373 * radix_tree_iter_resume - resume iterating when the chunk may be invalid
374 * @slot: pointer to current slot
375 * @iter: iterator state
376 * Returns: New slot pointer
377 *
378 * If the iterator needs to release then reacquire a lock, the chunk may
379 * have been invalidated by an insertion or deletion. Call this function
380 * before releasing the lock to continue the iteration from the next index.
381 */
382void __rcu **__must_check radix_tree_iter_resume(void __rcu **slot,
383 struct radix_tree_iter *iter);
384
385/**
386 * radix_tree_chunk_size - get current chunk size
387 *
388 * @iter: pointer to radix tree iterator
389 * Returns: current chunk size
390 */
391static __always_inline long
392radix_tree_chunk_size(struct radix_tree_iter *iter)
393{
394 return iter->next_index - iter->index;
395}
396
397/**
398 * radix_tree_next_slot - find next slot in chunk
399 *
400 * @slot: pointer to current slot
401 * @iter: pointer to interator state
402 * @flags: RADIX_TREE_ITER_*, should be constant
403 * Returns: pointer to next slot, or NULL if there no more left
404 *
405 * This function updates @iter->index in the case of a successful lookup.
406 * For tagged lookup it also eats @iter->tags.
407 *
408 * There are several cases where 'slot' can be passed in as NULL to this
409 * function. These cases result from the use of radix_tree_iter_resume() or
410 * radix_tree_iter_retry(). In these cases we don't end up dereferencing
411 * 'slot' because either:
412 * a) we are doing tagged iteration and iter->tags has been set to 0, or
413 * b) we are doing non-tagged iteration, and iter->index and iter->next_index
414 * have been set up so that radix_tree_chunk_size() returns 1 or 0.
415 */
416static __always_inline void __rcu **radix_tree_next_slot(void __rcu **slot,
417 struct radix_tree_iter *iter, unsigned flags)
418{
419 if (flags & RADIX_TREE_ITER_TAGGED) {
420 iter->tags >>= 1;
421 if (unlikely(!iter->tags))
422 return NULL;
423 if (likely(iter->tags & 1ul)) {
424 iter->index = __radix_tree_iter_add(iter, 1);
425 slot++;
426 goto found;
427 }
428 if (!(flags & RADIX_TREE_ITER_CONTIG)) {
429 unsigned offset = __ffs(iter->tags);
430
431 iter->tags >>= offset++;
432 iter->index = __radix_tree_iter_add(iter, offset);
433 slot += offset;
434 goto found;
435 }
436 } else {
437 long count = radix_tree_chunk_size(iter);
438
439 while (--count > 0) {
440 slot++;
441 iter->index = __radix_tree_iter_add(iter, 1);
442
443 if (likely(*slot))
444 goto found;
445 if (flags & RADIX_TREE_ITER_CONTIG) {
446 /* forbid switching to the next chunk */
447 iter->next_index = 0;
448 break;
449 }
450 }
451 }
452 return NULL;
453
454 found:
455 return slot;
456}
457
458/**
459 * radix_tree_for_each_slot - iterate over non-empty slots
460 *
461 * @slot: the void** variable for pointer to slot
462 * @root: the struct radix_tree_root pointer
463 * @iter: the struct radix_tree_iter pointer
464 * @start: iteration starting index
465 *
466 * @slot points to radix tree slot, @iter->index contains its index.
467 */
468#define radix_tree_for_each_slot(slot, root, iter, start) \
469 for (slot = radix_tree_iter_init(iter, start) ; \
470 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \
471 slot = radix_tree_next_slot(slot, iter, 0))
472
473/**
474 * radix_tree_for_each_tagged - iterate over tagged slots
475 *
476 * @slot: the void** variable for pointer to slot
477 * @root: the struct radix_tree_root pointer
478 * @iter: the struct radix_tree_iter pointer
479 * @start: iteration starting index
480 * @tag: tag index
481 *
482 * @slot points to radix tree slot, @iter->index contains its index.
483 */
484#define radix_tree_for_each_tagged(slot, root, iter, start, tag) \
485 for (slot = radix_tree_iter_init(iter, start) ; \
486 slot || (slot = radix_tree_next_chunk(root, iter, \
487 RADIX_TREE_ITER_TAGGED | tag)) ; \
488 slot = radix_tree_next_slot(slot, iter, \
489 RADIX_TREE_ITER_TAGGED | tag))
490
491#endif /* _LINUX_RADIX_TREE_H */
492