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
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.
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.
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.
21 #ifndef _LINUX_RADIX_TREE_H
22 #define _LINUX_RADIX_TREE_H
24 #include <linux/bitops.h>
25 #include <linux/bug.h>
26 #include <linux/kernel.h>
27 #include <linux/list.h>
28 #include <linux/preempt.h>
29 #include <linux/rcupdate.h>
30 #include <linux/spinlock.h>
31 #include <linux/types.h>
34 * The bottom two bits of the slot determine how the remaining bits in the
35 * slot are interpreted:
39 * 10 - exceptional entry
40 * 11 - this bit combination is currently unused/reserved
42 * The internal entry may be a pointer to the next level in the tree, a
43 * sibling entry, or an indicator that the entry in this slot has been moved
44 * to another location in the tree and the lookup should be restarted. While
45 * NULL fits the 'data pointer' pattern, it means that there is no entry in
46 * the tree for this index (no matter what level of the tree it is found at).
47 * This means that you cannot store NULL in the tree as a value for the index.
49 #define RADIX_TREE_ENTRY_MASK 3UL
50 #define RADIX_TREE_INTERNAL_NODE 1UL
53 * Most users of the radix tree store pointers but shmem/tmpfs stores swap
54 * entries in the same tree. They are marked as exceptional entries to
55 * distinguish them from pointers to struct page.
56 * EXCEPTIONAL_ENTRY tests the bit, EXCEPTIONAL_SHIFT shifts content past it.
58 #define RADIX_TREE_EXCEPTIONAL_ENTRY 2
59 #define RADIX_TREE_EXCEPTIONAL_SHIFT 2
61 static inline bool radix_tree_is_internal_node(void *ptr)
63 return ((unsigned long)ptr & RADIX_TREE_ENTRY_MASK) ==
64 RADIX_TREE_INTERNAL_NODE;
67 /*** radix-tree API starts here ***/
69 #define RADIX_TREE_MAX_TAGS 3
71 #ifndef RADIX_TREE_MAP_SHIFT
72 #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
75 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
76 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
78 #define RADIX_TREE_TAG_LONGS \
79 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
81 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
82 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
83 RADIX_TREE_MAP_SHIFT))
86 * @count is the count of every non-NULL element in the ->slots array
87 * whether that is an exceptional entry, a retry entry, a user pointer,
88 * a sibling entry or a pointer to the next level of the tree.
89 * @exceptional is the count of every element in ->slots which is
90 * either radix_tree_exceptional_entry() or is a sibling entry for an
93 struct radix_tree_node {
94 unsigned char shift; /* Bits remaining in each slot */
95 unsigned char offset; /* Slot offset in parent */
96 unsigned char count; /* Total entry count */
97 unsigned char exceptional; /* Exceptional entry count */
98 struct radix_tree_node *parent; /* Used when ascending tree */
99 struct radix_tree_root *root; /* The tree we belong to */
101 struct list_head private_list; /* For tree user */
102 struct rcu_head rcu_head; /* Used when freeing node */
104 void __rcu *slots[RADIX_TREE_MAP_SIZE];
105 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
108 /* The top bits of gfp_mask are used to store the root tags and the IDR flag */
109 #define ROOT_IS_IDR ((__force gfp_t)(1 << __GFP_BITS_SHIFT))
110 #define ROOT_TAG_SHIFT (__GFP_BITS_SHIFT + 1)
112 struct radix_tree_root {
114 struct radix_tree_node __rcu *rnode;
117 #define RADIX_TREE_INIT(mask) { \
118 .gfp_mask = (mask), \
122 #define RADIX_TREE(name, mask) \
123 struct radix_tree_root name = RADIX_TREE_INIT(mask)
125 #define INIT_RADIX_TREE(root, mask) \
127 (root)->gfp_mask = (mask); \
128 (root)->rnode = NULL; \
131 static inline bool radix_tree_empty(const struct radix_tree_root *root)
133 return root->rnode == NULL;
137 * struct radix_tree_iter - radix tree iterator state
139 * @index: index of current slot
140 * @next_index: one beyond the last index for this chunk
141 * @tags: bit-mask for tag-iterating
142 * @node: node that contains current slot
143 * @shift: shift for the node that holds our slots
145 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a
146 * subinterval of slots contained within one radix tree leaf node. It is
147 * described by a pointer to its first slot and a struct radix_tree_iter
148 * which holds the chunk's position in the tree and its size. For tagged
149 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
152 struct radix_tree_iter {
154 unsigned long next_index;
156 struct radix_tree_node *node;
157 #ifdef CONFIG_RADIX_TREE_MULTIORDER
162 static inline unsigned int iter_shift(const struct radix_tree_iter *iter)
164 #ifdef CONFIG_RADIX_TREE_MULTIORDER
172 * Radix-tree synchronization
174 * The radix-tree API requires that users provide all synchronisation (with
175 * specific exceptions, noted below).
177 * Synchronization of access to the data items being stored in the tree, and
178 * management of their lifetimes must be completely managed by API users.
180 * For API usage, in general,
181 * - any function _modifying_ the tree or tags (inserting or deleting
182 * items, setting or clearing tags) must exclude other modifications, and
183 * exclude any functions reading the tree.
184 * - any function _reading_ the tree or tags (looking up items or tags,
185 * gang lookups) must exclude modifications to the tree, but may occur
186 * concurrently with other readers.
188 * The notable exceptions to this rule are the following functions:
189 * __radix_tree_lookup
191 * radix_tree_lookup_slot
193 * radix_tree_gang_lookup
194 * radix_tree_gang_lookup_slot
195 * radix_tree_gang_lookup_tag
196 * radix_tree_gang_lookup_tag_slot
199 * The first 8 functions are able to be called locklessly, using RCU. The
200 * caller must ensure calls to these functions are made within rcu_read_lock()
201 * regions. Other readers (lock-free or otherwise) and modifications may be
202 * running concurrently.
204 * It is still required that the caller manage the synchronization and lifetimes
205 * of the items. So if RCU lock-free lookups are used, typically this would mean
206 * that the items have their own locks, or are amenable to lock-free access; and
207 * that the items are freed by RCU (or only freed after having been deleted from
208 * the radix tree *and* a synchronize_rcu() grace period).
210 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
211 * access to data items when inserting into or looking up from the radix tree)
213 * Note that the value returned by radix_tree_tag_get() may not be relied upon
214 * if only the RCU read lock is held. Functions to set/clear tags and to
215 * delete nodes running concurrently with it may affect its result such that
216 * two consecutive reads in the same locked section may return different
217 * values. If reliability is required, modification functions must also be
218 * excluded from concurrency.
220 * radix_tree_tagged is able to be called without locking or RCU.
224 * radix_tree_deref_slot - dereference a slot
225 * @slot: slot pointer, returned by radix_tree_lookup_slot
227 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read
228 * locked across slot lookup and dereference. Not required if write lock is
229 * held (ie. items cannot be concurrently inserted).
231 * radix_tree_deref_retry must be used to confirm validity of the pointer if
232 * only the read lock is held.
234 * Return: entry stored in that slot.
236 static inline void *radix_tree_deref_slot(void __rcu **slot)
238 return rcu_dereference(*slot);
242 * radix_tree_deref_slot_protected - dereference a slot with tree lock held
243 * @slot: slot pointer, returned by radix_tree_lookup_slot
245 * Similar to radix_tree_deref_slot. The caller does not hold the RCU read
246 * lock but it must hold the tree lock to prevent parallel updates.
248 * Return: entry stored in that slot.
250 static inline void *radix_tree_deref_slot_protected(void __rcu **slot,
251 spinlock_t *treelock)
253 return rcu_dereference_protected(*slot, lockdep_is_held(treelock));
257 * radix_tree_deref_retry - check radix_tree_deref_slot
258 * @arg: pointer returned by radix_tree_deref_slot
259 * Returns: 0 if retry is not required, otherwise retry is required
261 * radix_tree_deref_retry must be used with radix_tree_deref_slot.
263 static inline int radix_tree_deref_retry(void *arg)
265 return unlikely(radix_tree_is_internal_node(arg));
269 * radix_tree_exceptional_entry - radix_tree_deref_slot gave exceptional entry?
270 * @arg: value returned by radix_tree_deref_slot
271 * Returns: 0 if well-aligned pointer, non-0 if exceptional entry.
273 static inline int radix_tree_exceptional_entry(void *arg)
275 /* Not unlikely because radix_tree_exception often tested first */
276 return (unsigned long)arg & RADIX_TREE_EXCEPTIONAL_ENTRY;
280 * radix_tree_exception - radix_tree_deref_slot returned either exception?
281 * @arg: value returned by radix_tree_deref_slot
282 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception.
284 static inline int radix_tree_exception(void *arg)
286 return unlikely((unsigned long)arg & RADIX_TREE_ENTRY_MASK);
289 int __radix_tree_create(struct radix_tree_root *, unsigned long index,
290 unsigned order, struct radix_tree_node **nodep,
291 void __rcu ***slotp);
292 int __radix_tree_insert(struct radix_tree_root *, unsigned long index,
293 unsigned order, void *);
294 static inline int radix_tree_insert(struct radix_tree_root *root,
295 unsigned long index, void *entry)
297 return __radix_tree_insert(root, index, 0, entry);
299 void *__radix_tree_lookup(const struct radix_tree_root *, unsigned long index,
300 struct radix_tree_node **nodep, void __rcu ***slotp);
301 void *radix_tree_lookup(const struct radix_tree_root *, unsigned long);
302 void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *,
303 unsigned long index);
304 typedef void (*radix_tree_update_node_t)(struct radix_tree_node *, void *);
305 void __radix_tree_replace(struct radix_tree_root *, struct radix_tree_node *,
306 void __rcu **slot, void *entry,
307 radix_tree_update_node_t update_node, void *private);
308 void radix_tree_iter_replace(struct radix_tree_root *,
309 const struct radix_tree_iter *, void __rcu **slot, void *entry);
310 void radix_tree_replace_slot(struct radix_tree_root *,
311 void __rcu **slot, void *entry);
312 void __radix_tree_delete_node(struct radix_tree_root *,
313 struct radix_tree_node *,
314 radix_tree_update_node_t update_node,
316 void radix_tree_iter_delete(struct radix_tree_root *,
317 struct radix_tree_iter *iter, void __rcu **slot);
318 void *radix_tree_delete_item(struct radix_tree_root *, unsigned long, void *);
319 void *radix_tree_delete(struct radix_tree_root *, unsigned long);
320 void radix_tree_clear_tags(struct radix_tree_root *, struct radix_tree_node *,
322 unsigned int radix_tree_gang_lookup(const struct radix_tree_root *,
323 void **results, unsigned long first_index,
324 unsigned int max_items);
325 unsigned int radix_tree_gang_lookup_slot(const struct radix_tree_root *,
326 void __rcu ***results, unsigned long *indices,
327 unsigned long first_index, unsigned int max_items);
328 int radix_tree_preload(gfp_t gfp_mask);
329 int radix_tree_maybe_preload(gfp_t gfp_mask);
330 int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order);
331 void radix_tree_init(void);
332 void *radix_tree_tag_set(struct radix_tree_root *,
333 unsigned long index, unsigned int tag);
334 void *radix_tree_tag_clear(struct radix_tree_root *,
335 unsigned long index, unsigned int tag);
336 int radix_tree_tag_get(const struct radix_tree_root *,
337 unsigned long index, unsigned int tag);
338 void radix_tree_iter_tag_set(struct radix_tree_root *,
339 const struct radix_tree_iter *iter, unsigned int tag);
340 void radix_tree_iter_tag_clear(struct radix_tree_root *,
341 const struct radix_tree_iter *iter, unsigned int tag);
342 unsigned int radix_tree_gang_lookup_tag(const struct radix_tree_root *,
343 void **results, unsigned long first_index,
344 unsigned int max_items, unsigned int tag);
345 unsigned int radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *,
346 void __rcu ***results, unsigned long first_index,
347 unsigned int max_items, unsigned int tag);
348 int radix_tree_tagged(const struct radix_tree_root *, unsigned int tag);
350 static inline void radix_tree_preload_end(void)
355 int radix_tree_split_preload(unsigned old_order, unsigned new_order, gfp_t);
356 int radix_tree_split(struct radix_tree_root *, unsigned long index,
358 int radix_tree_join(struct radix_tree_root *, unsigned long index,
359 unsigned new_order, void *);
360 void __rcu **idr_get_free(struct radix_tree_root *, struct radix_tree_iter *,
364 RADIX_TREE_ITER_TAG_MASK = 0x0f, /* tag index in lower nybble */
365 RADIX_TREE_ITER_TAGGED = 0x10, /* lookup tagged slots */
366 RADIX_TREE_ITER_CONTIG = 0x20, /* stop at first hole */
370 * radix_tree_iter_init - initialize radix tree iterator
372 * @iter: pointer to iterator state
373 * @start: iteration starting index
376 static __always_inline void __rcu **
377 radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start)
380 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
381 * in the case of a successful tagged chunk lookup. If the lookup was
382 * unsuccessful or non-tagged then nobody cares about ->tags.
384 * Set index to zero to bypass next_index overflow protection.
385 * See the comment in radix_tree_next_chunk() for details.
388 iter->next_index = start;
393 * radix_tree_next_chunk - find next chunk of slots for iteration
395 * @root: radix tree root
396 * @iter: iterator state
397 * @flags: RADIX_TREE_ITER_* flags and tag index
398 * Returns: pointer to chunk first slot, or NULL if there no more left
400 * This function looks up the next chunk in the radix tree starting from
401 * @iter->next_index. It returns a pointer to the chunk's first slot.
402 * Also it fills @iter with data about chunk: position in the tree (index),
403 * its end (next_index), and constructs a bit mask for tagged iterating (tags).
405 void __rcu **radix_tree_next_chunk(const struct radix_tree_root *,
406 struct radix_tree_iter *iter, unsigned flags);
409 * radix_tree_iter_lookup - look up an index in the radix tree
410 * @root: radix tree root
411 * @iter: iterator state
412 * @index: key to look up
414 * If @index is present in the radix tree, this function returns the slot
415 * containing it and updates @iter to describe the entry. If @index is not
416 * present, it returns NULL.
418 static inline void __rcu **
419 radix_tree_iter_lookup(const struct radix_tree_root *root,
420 struct radix_tree_iter *iter, unsigned long index)
422 radix_tree_iter_init(iter, index);
423 return radix_tree_next_chunk(root, iter, RADIX_TREE_ITER_CONTIG);
427 * radix_tree_iter_find - find a present entry
428 * @root: radix tree root
429 * @iter: iterator state
430 * @index: start location
432 * This function returns the slot containing the entry with the lowest index
433 * which is at least @index. If @index is larger than any present entry, this
434 * function returns NULL. The @iter is updated to describe the entry found.
436 static inline void __rcu **
437 radix_tree_iter_find(const struct radix_tree_root *root,
438 struct radix_tree_iter *iter, unsigned long index)
440 radix_tree_iter_init(iter, index);
441 return radix_tree_next_chunk(root, iter, 0);
445 * radix_tree_iter_retry - retry this chunk of the iteration
446 * @iter: iterator state
448 * If we iterate over a tree protected only by the RCU lock, a race
449 * against deletion or creation may result in seeing a slot for which
450 * radix_tree_deref_retry() returns true. If so, call this function
451 * and continue the iteration.
453 static inline __must_check
454 void __rcu **radix_tree_iter_retry(struct radix_tree_iter *iter)
456 iter->next_index = iter->index;
461 static inline unsigned long
462 __radix_tree_iter_add(struct radix_tree_iter *iter, unsigned long slots)
464 return iter->index + (slots << iter_shift(iter));
468 * radix_tree_iter_resume - resume iterating when the chunk may be invalid
469 * @slot: pointer to current slot
470 * @iter: iterator state
471 * Returns: New slot pointer
473 * If the iterator needs to release then reacquire a lock, the chunk may
474 * have been invalidated by an insertion or deletion. Call this function
475 * before releasing the lock to continue the iteration from the next index.
477 void __rcu **__must_check radix_tree_iter_resume(void __rcu **slot,
478 struct radix_tree_iter *iter);
481 * radix_tree_chunk_size - get current chunk size
483 * @iter: pointer to radix tree iterator
484 * Returns: current chunk size
486 static __always_inline long
487 radix_tree_chunk_size(struct radix_tree_iter *iter)
489 return (iter->next_index - iter->index) >> iter_shift(iter);
492 #ifdef CONFIG_RADIX_TREE_MULTIORDER
493 void __rcu **__radix_tree_next_slot(void __rcu **slot,
494 struct radix_tree_iter *iter, unsigned flags);
496 /* Can't happen without sibling entries, but the compiler can't tell that */
497 static inline void __rcu **__radix_tree_next_slot(void __rcu **slot,
498 struct radix_tree_iter *iter, unsigned flags)
505 * radix_tree_next_slot - find next slot in chunk
507 * @slot: pointer to current slot
508 * @iter: pointer to interator state
509 * @flags: RADIX_TREE_ITER_*, should be constant
510 * Returns: pointer to next slot, or NULL if there no more left
512 * This function updates @iter->index in the case of a successful lookup.
513 * For tagged lookup it also eats @iter->tags.
515 * There are several cases where 'slot' can be passed in as NULL to this
516 * function. These cases result from the use of radix_tree_iter_resume() or
517 * radix_tree_iter_retry(). In these cases we don't end up dereferencing
518 * 'slot' because either:
519 * a) we are doing tagged iteration and iter->tags has been set to 0, or
520 * b) we are doing non-tagged iteration, and iter->index and iter->next_index
521 * have been set up so that radix_tree_chunk_size() returns 1 or 0.
523 static __always_inline void __rcu **radix_tree_next_slot(void __rcu **slot,
524 struct radix_tree_iter *iter, unsigned flags)
526 if (flags & RADIX_TREE_ITER_TAGGED) {
528 if (unlikely(!iter->tags))
530 if (likely(iter->tags & 1ul)) {
531 iter->index = __radix_tree_iter_add(iter, 1);
535 if (!(flags & RADIX_TREE_ITER_CONTIG)) {
536 unsigned offset = __ffs(iter->tags);
538 iter->tags >>= offset++;
539 iter->index = __radix_tree_iter_add(iter, offset);
544 long count = radix_tree_chunk_size(iter);
546 while (--count > 0) {
548 iter->index = __radix_tree_iter_add(iter, 1);
552 if (flags & RADIX_TREE_ITER_CONTIG) {
553 /* forbid switching to the next chunk */
554 iter->next_index = 0;
562 if (unlikely(radix_tree_is_internal_node(rcu_dereference_raw(*slot))))
563 return __radix_tree_next_slot(slot, iter, flags);
568 * radix_tree_for_each_slot - iterate over non-empty slots
570 * @slot: the void** variable for pointer to slot
571 * @root: the struct radix_tree_root pointer
572 * @iter: the struct radix_tree_iter pointer
573 * @start: iteration starting index
575 * @slot points to radix tree slot, @iter->index contains its index.
577 #define radix_tree_for_each_slot(slot, root, iter, start) \
578 for (slot = radix_tree_iter_init(iter, start) ; \
579 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \
580 slot = radix_tree_next_slot(slot, iter, 0))
583 * radix_tree_for_each_contig - iterate over contiguous slots
585 * @slot: the void** variable for pointer to slot
586 * @root: the struct radix_tree_root pointer
587 * @iter: the struct radix_tree_iter pointer
588 * @start: iteration starting index
590 * @slot points to radix tree slot, @iter->index contains its index.
592 #define radix_tree_for_each_contig(slot, root, iter, start) \
593 for (slot = radix_tree_iter_init(iter, start) ; \
594 slot || (slot = radix_tree_next_chunk(root, iter, \
595 RADIX_TREE_ITER_CONTIG)) ; \
596 slot = radix_tree_next_slot(slot, iter, \
597 RADIX_TREE_ITER_CONTIG))
600 * radix_tree_for_each_tagged - iterate over tagged slots
602 * @slot: the void** variable for pointer to slot
603 * @root: the struct radix_tree_root pointer
604 * @iter: the struct radix_tree_iter pointer
605 * @start: iteration starting index
608 * @slot points to radix tree slot, @iter->index contains its index.
610 #define radix_tree_for_each_tagged(slot, root, iter, start, tag) \
611 for (slot = radix_tree_iter_init(iter, start) ; \
612 slot || (slot = radix_tree_next_chunk(root, iter, \
613 RADIX_TREE_ITER_TAGGED | tag)) ; \
614 slot = radix_tree_next_slot(slot, iter, \
615 RADIX_TREE_ITER_TAGGED | tag))
617 #endif /* _LINUX_RADIX_TREE_H */