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/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>
33 * The bottom two bits of the slot determine how the remaining bits in the
34 * slot are interpreted:
38 * 10 - exceptional entry
39 * 11 - this bit combination is currently unused/reserved
41 * The internal entry may be a pointer to the next level in the tree, a
42 * sibling entry, or an indicator that the entry in this slot has been moved
43 * to another location in the tree and the lookup should be restarted. While
44 * NULL fits the 'data pointer' pattern, it means that there is no entry in
45 * the tree for this index (no matter what level of the tree it is found at).
46 * This means that you cannot store NULL in the tree as a value for the index.
48 #define RADIX_TREE_ENTRY_MASK 3UL
49 #define RADIX_TREE_INTERNAL_NODE 1UL
52 * Most users of the radix tree store pointers but shmem/tmpfs stores swap
53 * entries in the same tree. They are marked as exceptional entries to
54 * distinguish them from pointers to struct page.
55 * EXCEPTIONAL_ENTRY tests the bit, EXCEPTIONAL_SHIFT shifts content past it.
57 #define RADIX_TREE_EXCEPTIONAL_ENTRY 2
58 #define RADIX_TREE_EXCEPTIONAL_SHIFT 2
60 static inline bool radix_tree_is_internal_node(void *ptr)
62 return ((unsigned long)ptr & RADIX_TREE_ENTRY_MASK) ==
63 RADIX_TREE_INTERNAL_NODE;
66 /*** radix-tree API starts here ***/
68 #define RADIX_TREE_MAX_TAGS 3
70 #ifndef RADIX_TREE_MAP_SHIFT
71 #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
74 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
75 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
77 #define RADIX_TREE_TAG_LONGS \
78 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
80 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
81 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
82 RADIX_TREE_MAP_SHIFT))
85 * @count is the count of every non-NULL element in the ->slots array
86 * whether that is an exceptional entry, a retry entry, a user pointer,
87 * a sibling entry or a pointer to the next level of the tree.
88 * @exceptional is the count of every element in ->slots which is
89 * either radix_tree_exceptional_entry() or is a sibling entry for an
92 struct radix_tree_node {
93 unsigned char shift; /* Bits remaining in each slot */
94 unsigned char offset; /* Slot offset in parent */
95 unsigned char count; /* Total entry count */
96 unsigned char exceptional; /* Exceptional entry count */
97 struct radix_tree_node *parent; /* Used when ascending tree */
98 struct radix_tree_root *root; /* The tree we belong to */
100 struct list_head private_list; /* For tree user */
101 struct rcu_head rcu_head; /* Used when freeing node */
103 void __rcu *slots[RADIX_TREE_MAP_SIZE];
104 unsigned long tags[RADIX_TREE_MAX_TAGS][RADIX_TREE_TAG_LONGS];
107 /* The IDR tag is stored in the low bits of the GFP flags */
108 #define ROOT_IS_IDR ((__force gfp_t)4)
109 /* The top bits of gfp_mask are used to store the root tags */
110 #define ROOT_TAG_SHIFT (__GFP_BITS_SHIFT)
112 struct radix_tree_root {
115 struct radix_tree_node __rcu *rnode;
118 #define RADIX_TREE_INIT(name, mask) { \
119 .xa_lock = __SPIN_LOCK_UNLOCKED(name.xa_lock), \
120 .gfp_mask = (mask), \
124 #define RADIX_TREE(name, mask) \
125 struct radix_tree_root name = RADIX_TREE_INIT(name, mask)
127 #define INIT_RADIX_TREE(root, mask) \
129 spin_lock_init(&(root)->xa_lock); \
130 (root)->gfp_mask = (mask); \
131 (root)->rnode = NULL; \
134 static inline bool radix_tree_empty(const struct radix_tree_root *root)
136 return root->rnode == NULL;
140 * struct radix_tree_iter - radix tree iterator state
142 * @index: index of current slot
143 * @next_index: one beyond the last index for this chunk
144 * @tags: bit-mask for tag-iterating
145 * @node: node that contains current slot
146 * @shift: shift for the node that holds our slots
148 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a
149 * subinterval of slots contained within one radix tree leaf node. It is
150 * described by a pointer to its first slot and a struct radix_tree_iter
151 * which holds the chunk's position in the tree and its size. For tagged
152 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
155 struct radix_tree_iter {
157 unsigned long next_index;
159 struct radix_tree_node *node;
160 #ifdef CONFIG_RADIX_TREE_MULTIORDER
165 static inline unsigned int iter_shift(const struct radix_tree_iter *iter)
167 #ifdef CONFIG_RADIX_TREE_MULTIORDER
175 * Radix-tree synchronization
177 * The radix-tree API requires that users provide all synchronisation (with
178 * specific exceptions, noted below).
180 * Synchronization of access to the data items being stored in the tree, and
181 * management of their lifetimes must be completely managed by API users.
183 * For API usage, in general,
184 * - any function _modifying_ the tree or tags (inserting or deleting
185 * items, setting or clearing tags) must exclude other modifications, and
186 * exclude any functions reading the tree.
187 * - any function _reading_ the tree or tags (looking up items or tags,
188 * gang lookups) must exclude modifications to the tree, but may occur
189 * concurrently with other readers.
191 * The notable exceptions to this rule are the following functions:
192 * __radix_tree_lookup
194 * radix_tree_lookup_slot
196 * radix_tree_gang_lookup
197 * radix_tree_gang_lookup_slot
198 * radix_tree_gang_lookup_tag
199 * radix_tree_gang_lookup_tag_slot
202 * The first 8 functions are able to be called locklessly, using RCU. The
203 * caller must ensure calls to these functions are made within rcu_read_lock()
204 * regions. Other readers (lock-free or otherwise) and modifications may be
205 * running concurrently.
207 * It is still required that the caller manage the synchronization and lifetimes
208 * of the items. So if RCU lock-free lookups are used, typically this would mean
209 * that the items have their own locks, or are amenable to lock-free access; and
210 * that the items are freed by RCU (or only freed after having been deleted from
211 * the radix tree *and* a synchronize_rcu() grace period).
213 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
214 * access to data items when inserting into or looking up from the radix tree)
216 * Note that the value returned by radix_tree_tag_get() may not be relied upon
217 * if only the RCU read lock is held. Functions to set/clear tags and to
218 * delete nodes running concurrently with it may affect its result such that
219 * two consecutive reads in the same locked section may return different
220 * values. If reliability is required, modification functions must also be
221 * excluded from concurrency.
223 * radix_tree_tagged is able to be called without locking or RCU.
227 * radix_tree_deref_slot - dereference a slot
228 * @slot: slot pointer, returned by radix_tree_lookup_slot
230 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read
231 * locked across slot lookup and dereference. Not required if write lock is
232 * held (ie. items cannot be concurrently inserted).
234 * radix_tree_deref_retry must be used to confirm validity of the pointer if
235 * only the read lock is held.
237 * Return: entry stored in that slot.
239 static inline void *radix_tree_deref_slot(void __rcu **slot)
241 return rcu_dereference(*slot);
245 * radix_tree_deref_slot_protected - dereference a slot with tree lock held
246 * @slot: slot pointer, returned by radix_tree_lookup_slot
248 * Similar to radix_tree_deref_slot. The caller does not hold the RCU read
249 * lock but it must hold the tree lock to prevent parallel updates.
251 * Return: entry stored in that slot.
253 static inline void *radix_tree_deref_slot_protected(void __rcu **slot,
254 spinlock_t *treelock)
256 return rcu_dereference_protected(*slot, lockdep_is_held(treelock));
260 * radix_tree_deref_retry - check radix_tree_deref_slot
261 * @arg: pointer returned by radix_tree_deref_slot
262 * Returns: 0 if retry is not required, otherwise retry is required
264 * radix_tree_deref_retry must be used with radix_tree_deref_slot.
266 static inline int radix_tree_deref_retry(void *arg)
268 return unlikely(radix_tree_is_internal_node(arg));
272 * radix_tree_exceptional_entry - radix_tree_deref_slot gave exceptional entry?
273 * @arg: value returned by radix_tree_deref_slot
274 * Returns: 0 if well-aligned pointer, non-0 if exceptional entry.
276 static inline int radix_tree_exceptional_entry(void *arg)
278 /* Not unlikely because radix_tree_exception often tested first */
279 return (unsigned long)arg & RADIX_TREE_EXCEPTIONAL_ENTRY;
283 * radix_tree_exception - radix_tree_deref_slot returned either exception?
284 * @arg: value returned by radix_tree_deref_slot
285 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception.
287 static inline int radix_tree_exception(void *arg)
289 return unlikely((unsigned long)arg & RADIX_TREE_ENTRY_MASK);
292 int __radix_tree_create(struct radix_tree_root *, unsigned long index,
293 unsigned order, struct radix_tree_node **nodep,
294 void __rcu ***slotp);
295 int __radix_tree_insert(struct radix_tree_root *, unsigned long index,
296 unsigned order, void *);
297 static inline int radix_tree_insert(struct radix_tree_root *root,
298 unsigned long index, void *entry)
300 return __radix_tree_insert(root, index, 0, entry);
302 void *__radix_tree_lookup(const struct radix_tree_root *, unsigned long index,
303 struct radix_tree_node **nodep, void __rcu ***slotp);
304 void *radix_tree_lookup(const struct radix_tree_root *, unsigned long);
305 void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *,
306 unsigned long index);
307 typedef void (*radix_tree_update_node_t)(struct radix_tree_node *);
308 void __radix_tree_replace(struct radix_tree_root *, struct radix_tree_node *,
309 void __rcu **slot, void *entry,
310 radix_tree_update_node_t update_node);
311 void radix_tree_iter_replace(struct radix_tree_root *,
312 const struct radix_tree_iter *, void __rcu **slot, void *entry);
313 void radix_tree_replace_slot(struct radix_tree_root *,
314 void __rcu **slot, void *entry);
315 void __radix_tree_delete_node(struct radix_tree_root *,
316 struct radix_tree_node *,
317 radix_tree_update_node_t update_node);
318 void radix_tree_iter_delete(struct radix_tree_root *,
319 struct radix_tree_iter *iter, void __rcu **slot);
320 void *radix_tree_delete_item(struct radix_tree_root *, unsigned long, void *);
321 void *radix_tree_delete(struct radix_tree_root *, unsigned long);
322 void radix_tree_clear_tags(struct radix_tree_root *, struct radix_tree_node *,
324 unsigned int radix_tree_gang_lookup(const struct radix_tree_root *,
325 void **results, unsigned long first_index,
326 unsigned int max_items);
327 unsigned int radix_tree_gang_lookup_slot(const struct radix_tree_root *,
328 void __rcu ***results, unsigned long *indices,
329 unsigned long first_index, unsigned int max_items);
330 int radix_tree_preload(gfp_t gfp_mask);
331 int radix_tree_maybe_preload(gfp_t gfp_mask);
332 int radix_tree_maybe_preload_order(gfp_t gfp_mask, int order);
333 void radix_tree_init(void);
334 void *radix_tree_tag_set(struct radix_tree_root *,
335 unsigned long index, unsigned int tag);
336 void *radix_tree_tag_clear(struct radix_tree_root *,
337 unsigned long index, unsigned int tag);
338 int radix_tree_tag_get(const struct radix_tree_root *,
339 unsigned long index, unsigned int tag);
340 void radix_tree_iter_tag_set(struct radix_tree_root *,
341 const struct radix_tree_iter *iter, unsigned int tag);
342 void radix_tree_iter_tag_clear(struct radix_tree_root *,
343 const struct radix_tree_iter *iter, unsigned int tag);
344 unsigned int radix_tree_gang_lookup_tag(const struct radix_tree_root *,
345 void **results, unsigned long first_index,
346 unsigned int max_items, unsigned int tag);
347 unsigned int radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *,
348 void __rcu ***results, unsigned long first_index,
349 unsigned int max_items, unsigned int tag);
350 int radix_tree_tagged(const struct radix_tree_root *, unsigned int tag);
352 static inline void radix_tree_preload_end(void)
357 int radix_tree_split_preload(unsigned old_order, unsigned new_order, gfp_t);
358 int radix_tree_split(struct radix_tree_root *, unsigned long index,
360 int radix_tree_join(struct radix_tree_root *, unsigned long index,
361 unsigned new_order, void *);
363 void __rcu **idr_get_free(struct radix_tree_root *root,
364 struct radix_tree_iter *iter, gfp_t gfp,
368 RADIX_TREE_ITER_TAG_MASK = 0x0f, /* tag index in lower nybble */
369 RADIX_TREE_ITER_TAGGED = 0x10, /* lookup tagged slots */
370 RADIX_TREE_ITER_CONTIG = 0x20, /* stop at first hole */
374 * radix_tree_iter_init - initialize radix tree iterator
376 * @iter: pointer to iterator state
377 * @start: iteration starting index
380 static __always_inline void __rcu **
381 radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start)
384 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
385 * in the case of a successful tagged chunk lookup. If the lookup was
386 * unsuccessful or non-tagged then nobody cares about ->tags.
388 * Set index to zero to bypass next_index overflow protection.
389 * See the comment in radix_tree_next_chunk() for details.
392 iter->next_index = start;
397 * radix_tree_next_chunk - find next chunk of slots for iteration
399 * @root: radix tree root
400 * @iter: iterator state
401 * @flags: RADIX_TREE_ITER_* flags and tag index
402 * Returns: pointer to chunk first slot, or NULL if there no more left
404 * This function looks up the next chunk in the radix tree starting from
405 * @iter->next_index. It returns a pointer to the chunk's first slot.
406 * Also it fills @iter with data about chunk: position in the tree (index),
407 * its end (next_index), and constructs a bit mask for tagged iterating (tags).
409 void __rcu **radix_tree_next_chunk(const struct radix_tree_root *,
410 struct radix_tree_iter *iter, unsigned flags);
413 * radix_tree_iter_lookup - look up an index in the radix tree
414 * @root: radix tree root
415 * @iter: iterator state
416 * @index: key to look up
418 * If @index is present in the radix tree, this function returns the slot
419 * containing it and updates @iter to describe the entry. If @index is not
420 * present, it returns NULL.
422 static inline void __rcu **
423 radix_tree_iter_lookup(const struct radix_tree_root *root,
424 struct radix_tree_iter *iter, unsigned long index)
426 radix_tree_iter_init(iter, index);
427 return radix_tree_next_chunk(root, iter, RADIX_TREE_ITER_CONTIG);
431 * radix_tree_iter_find - find a present entry
432 * @root: radix tree root
433 * @iter: iterator state
434 * @index: start location
436 * This function returns the slot containing the entry with the lowest index
437 * which is at least @index. If @index is larger than any present entry, this
438 * function returns NULL. The @iter is updated to describe the entry found.
440 static inline void __rcu **
441 radix_tree_iter_find(const struct radix_tree_root *root,
442 struct radix_tree_iter *iter, unsigned long index)
444 radix_tree_iter_init(iter, index);
445 return radix_tree_next_chunk(root, iter, 0);
449 * radix_tree_iter_retry - retry this chunk of the iteration
450 * @iter: iterator state
452 * If we iterate over a tree protected only by the RCU lock, a race
453 * against deletion or creation may result in seeing a slot for which
454 * radix_tree_deref_retry() returns true. If so, call this function
455 * and continue the iteration.
457 static inline __must_check
458 void __rcu **radix_tree_iter_retry(struct radix_tree_iter *iter)
460 iter->next_index = iter->index;
465 static inline unsigned long
466 __radix_tree_iter_add(struct radix_tree_iter *iter, unsigned long slots)
468 return iter->index + (slots << iter_shift(iter));
472 * radix_tree_iter_resume - resume iterating when the chunk may be invalid
473 * @slot: pointer to current slot
474 * @iter: iterator state
475 * Returns: New slot pointer
477 * If the iterator needs to release then reacquire a lock, the chunk may
478 * have been invalidated by an insertion or deletion. Call this function
479 * before releasing the lock to continue the iteration from the next index.
481 void __rcu **__must_check radix_tree_iter_resume(void __rcu **slot,
482 struct radix_tree_iter *iter);
485 * radix_tree_chunk_size - get current chunk size
487 * @iter: pointer to radix tree iterator
488 * Returns: current chunk size
490 static __always_inline long
491 radix_tree_chunk_size(struct radix_tree_iter *iter)
493 return (iter->next_index - iter->index) >> iter_shift(iter);
496 #ifdef CONFIG_RADIX_TREE_MULTIORDER
497 void __rcu **__radix_tree_next_slot(void __rcu **slot,
498 struct radix_tree_iter *iter, unsigned flags);
500 /* Can't happen without sibling entries, but the compiler can't tell that */
501 static inline void __rcu **__radix_tree_next_slot(void __rcu **slot,
502 struct radix_tree_iter *iter, unsigned flags)
509 * radix_tree_next_slot - find next slot in chunk
511 * @slot: pointer to current slot
512 * @iter: pointer to interator state
513 * @flags: RADIX_TREE_ITER_*, should be constant
514 * Returns: pointer to next slot, or NULL if there no more left
516 * This function updates @iter->index in the case of a successful lookup.
517 * For tagged lookup it also eats @iter->tags.
519 * There are several cases where 'slot' can be passed in as NULL to this
520 * function. These cases result from the use of radix_tree_iter_resume() or
521 * radix_tree_iter_retry(). In these cases we don't end up dereferencing
522 * 'slot' because either:
523 * a) we are doing tagged iteration and iter->tags has been set to 0, or
524 * b) we are doing non-tagged iteration, and iter->index and iter->next_index
525 * have been set up so that radix_tree_chunk_size() returns 1 or 0.
527 static __always_inline void __rcu **radix_tree_next_slot(void __rcu **slot,
528 struct radix_tree_iter *iter, unsigned flags)
530 if (flags & RADIX_TREE_ITER_TAGGED) {
532 if (unlikely(!iter->tags))
534 if (likely(iter->tags & 1ul)) {
535 iter->index = __radix_tree_iter_add(iter, 1);
539 if (!(flags & RADIX_TREE_ITER_CONTIG)) {
540 unsigned offset = __ffs(iter->tags);
542 iter->tags >>= offset++;
543 iter->index = __radix_tree_iter_add(iter, offset);
548 long count = radix_tree_chunk_size(iter);
550 while (--count > 0) {
552 iter->index = __radix_tree_iter_add(iter, 1);
556 if (flags & RADIX_TREE_ITER_CONTIG) {
557 /* forbid switching to the next chunk */
558 iter->next_index = 0;
566 if (unlikely(radix_tree_is_internal_node(rcu_dereference_raw(*slot))))
567 return __radix_tree_next_slot(slot, iter, flags);
572 * radix_tree_for_each_slot - iterate over non-empty slots
574 * @slot: the void** variable for pointer to slot
575 * @root: the struct radix_tree_root pointer
576 * @iter: the struct radix_tree_iter pointer
577 * @start: iteration starting index
579 * @slot points to radix tree slot, @iter->index contains its index.
581 #define radix_tree_for_each_slot(slot, root, iter, start) \
582 for (slot = radix_tree_iter_init(iter, start) ; \
583 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \
584 slot = radix_tree_next_slot(slot, iter, 0))
587 * radix_tree_for_each_contig - iterate over contiguous slots
589 * @slot: the void** variable for pointer to slot
590 * @root: the struct radix_tree_root pointer
591 * @iter: the struct radix_tree_iter pointer
592 * @start: iteration starting index
594 * @slot points to radix tree slot, @iter->index contains its index.
596 #define radix_tree_for_each_contig(slot, root, iter, start) \
597 for (slot = radix_tree_iter_init(iter, start) ; \
598 slot || (slot = radix_tree_next_chunk(root, iter, \
599 RADIX_TREE_ITER_CONTIG)) ; \
600 slot = radix_tree_next_slot(slot, iter, \
601 RADIX_TREE_ITER_CONTIG))
604 * radix_tree_for_each_tagged - iterate over tagged slots
606 * @slot: the void** variable for pointer to slot
607 * @root: the struct radix_tree_root pointer
608 * @iter: the struct radix_tree_iter pointer
609 * @start: iteration starting index
612 * @slot points to radix tree slot, @iter->index contains its index.
614 #define radix_tree_for_each_tagged(slot, root, iter, start, tag) \
615 for (slot = radix_tree_iter_init(iter, start) ; \
616 slot || (slot = radix_tree_next_chunk(root, iter, \
617 RADIX_TREE_ITER_TAGGED | tag)) ; \
618 slot = radix_tree_next_slot(slot, iter, \
619 RADIX_TREE_ITER_TAGGED | tag))
621 #endif /* _LINUX_RADIX_TREE_H */