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 void *private_data
; /* For tree user */
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 /* root tags are stored in gfp_mask, shifted by __GFP_BITS_SHIFT */
109 struct radix_tree_root
{
111 struct radix_tree_node __rcu
*rnode
;
114 #define RADIX_TREE_INIT(mask) { \
115 .gfp_mask = (mask), \
119 #define RADIX_TREE(name, mask) \
120 struct radix_tree_root name = RADIX_TREE_INIT(mask)
122 #define INIT_RADIX_TREE(root, mask) \
124 (root)->gfp_mask = (mask); \
125 (root)->rnode = NULL; \
128 static inline bool radix_tree_empty(const struct radix_tree_root
*root
)
130 return root
->rnode
== NULL
;
134 * struct radix_tree_iter - radix tree iterator state
136 * @index: index of current slot
137 * @next_index: one beyond the last index for this chunk
138 * @tags: bit-mask for tag-iterating
139 * @node: node that contains current slot
140 * @shift: shift for the node that holds our slots
142 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a
143 * subinterval of slots contained within one radix tree leaf node. It is
144 * described by a pointer to its first slot and a struct radix_tree_iter
145 * which holds the chunk's position in the tree and its size. For tagged
146 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
149 struct radix_tree_iter
{
151 unsigned long next_index
;
153 struct radix_tree_node
*node
;
154 #ifdef CONFIG_RADIX_TREE_MULTIORDER
159 static inline unsigned int iter_shift(const struct radix_tree_iter
*iter
)
161 #ifdef CONFIG_RADIX_TREE_MULTIORDER
169 * Radix-tree synchronization
171 * The radix-tree API requires that users provide all synchronisation (with
172 * specific exceptions, noted below).
174 * Synchronization of access to the data items being stored in the tree, and
175 * management of their lifetimes must be completely managed by API users.
177 * For API usage, in general,
178 * - any function _modifying_ the tree or tags (inserting or deleting
179 * items, setting or clearing tags) must exclude other modifications, and
180 * exclude any functions reading the tree.
181 * - any function _reading_ the tree or tags (looking up items or tags,
182 * gang lookups) must exclude modifications to the tree, but may occur
183 * concurrently with other readers.
185 * The notable exceptions to this rule are the following functions:
186 * __radix_tree_lookup
188 * radix_tree_lookup_slot
190 * radix_tree_gang_lookup
191 * radix_tree_gang_lookup_slot
192 * radix_tree_gang_lookup_tag
193 * radix_tree_gang_lookup_tag_slot
196 * The first 8 functions are able to be called locklessly, using RCU. The
197 * caller must ensure calls to these functions are made within rcu_read_lock()
198 * regions. Other readers (lock-free or otherwise) and modifications may be
199 * running concurrently.
201 * It is still required that the caller manage the synchronization and lifetimes
202 * of the items. So if RCU lock-free lookups are used, typically this would mean
203 * that the items have their own locks, or are amenable to lock-free access; and
204 * that the items are freed by RCU (or only freed after having been deleted from
205 * the radix tree *and* a synchronize_rcu() grace period).
207 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
208 * access to data items when inserting into or looking up from the radix tree)
210 * Note that the value returned by radix_tree_tag_get() may not be relied upon
211 * if only the RCU read lock is held. Functions to set/clear tags and to
212 * delete nodes running concurrently with it may affect its result such that
213 * two consecutive reads in the same locked section may return different
214 * values. If reliability is required, modification functions must also be
215 * excluded from concurrency.
217 * radix_tree_tagged is able to be called without locking or RCU.
221 * radix_tree_deref_slot - dereference a slot
222 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
223 * Returns: item that was stored in that slot with any direct pointer flag
226 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read
227 * locked across slot lookup and dereference. Not required if write lock is
228 * held (ie. items cannot be concurrently inserted).
230 * radix_tree_deref_retry must be used to confirm validity of the pointer if
231 * only the read lock is held.
233 static inline void *radix_tree_deref_slot(void **pslot
)
235 return rcu_dereference(*pslot
);
239 * radix_tree_deref_slot_protected - dereference a slot without RCU lock but with tree lock held
240 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
241 * Returns: item that was stored in that slot with any direct pointer flag
244 * Similar to radix_tree_deref_slot but only used during migration when a pages
245 * mapping is being moved. The caller does not hold the RCU read lock but it
246 * must hold the tree lock to prevent parallel updates.
248 static inline void *radix_tree_deref_slot_protected(void **pslot
,
249 spinlock_t
*treelock
)
251 return rcu_dereference_protected(*pslot
, lockdep_is_held(treelock
));
255 * radix_tree_deref_retry - check radix_tree_deref_slot
256 * @arg: pointer returned by radix_tree_deref_slot
257 * Returns: 0 if retry is not required, otherwise retry is required
259 * radix_tree_deref_retry must be used with radix_tree_deref_slot.
261 static inline int radix_tree_deref_retry(void *arg
)
263 return unlikely(radix_tree_is_internal_node(arg
));
267 * radix_tree_exceptional_entry - radix_tree_deref_slot gave exceptional entry?
268 * @arg: value returned by radix_tree_deref_slot
269 * Returns: 0 if well-aligned pointer, non-0 if exceptional entry.
271 static inline int radix_tree_exceptional_entry(void *arg
)
273 /* Not unlikely because radix_tree_exception often tested first */
274 return (unsigned long)arg
& RADIX_TREE_EXCEPTIONAL_ENTRY
;
278 * radix_tree_exception - radix_tree_deref_slot returned either exception?
279 * @arg: value returned by radix_tree_deref_slot
280 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception.
282 static inline int radix_tree_exception(void *arg
)
284 return unlikely((unsigned long)arg
& RADIX_TREE_ENTRY_MASK
);
287 int __radix_tree_create(struct radix_tree_root
*root
, unsigned long index
,
288 unsigned order
, struct radix_tree_node
**nodep
,
290 int __radix_tree_insert(struct radix_tree_root
*, unsigned long index
,
291 unsigned order
, void *);
292 static inline int radix_tree_insert(struct radix_tree_root
*root
,
293 unsigned long index
, void *entry
)
295 return __radix_tree_insert(root
, index
, 0, entry
);
297 void *__radix_tree_lookup(const struct radix_tree_root
*, unsigned long index
,
298 struct radix_tree_node
**nodep
, void ***slotp
);
299 void *radix_tree_lookup(const struct radix_tree_root
*, unsigned long);
300 void **radix_tree_lookup_slot(const struct radix_tree_root
*, unsigned long);
301 typedef void (*radix_tree_update_node_t
)(struct radix_tree_node
*, void *);
302 void __radix_tree_replace(struct radix_tree_root
*root
,
303 struct radix_tree_node
*node
,
304 void **slot
, void *item
,
305 radix_tree_update_node_t update_node
, void *private);
306 void radix_tree_iter_replace(struct radix_tree_root
*,
307 const struct radix_tree_iter
*, void **slot
, void *item
);
308 void radix_tree_replace_slot(struct radix_tree_root
*root
,
309 void **slot
, void *item
);
310 void __radix_tree_delete_node(struct radix_tree_root
*root
,
311 struct radix_tree_node
*node
,
312 radix_tree_update_node_t update_node
,
314 void *radix_tree_delete_item(struct radix_tree_root
*, unsigned long, void *);
315 void *radix_tree_delete(struct radix_tree_root
*, unsigned long);
316 void radix_tree_clear_tags(struct radix_tree_root
*root
,
317 struct radix_tree_node
*node
,
319 unsigned int radix_tree_gang_lookup(const struct radix_tree_root
*,
320 void **results
, unsigned long first_index
,
321 unsigned int max_items
);
322 unsigned int radix_tree_gang_lookup_slot(const struct radix_tree_root
*,
323 void ***results
, unsigned long *indices
,
324 unsigned long first_index
, unsigned int max_items
);
325 int radix_tree_preload(gfp_t gfp_mask
);
326 int radix_tree_maybe_preload(gfp_t gfp_mask
);
327 int radix_tree_maybe_preload_order(gfp_t gfp_mask
, int order
);
328 void radix_tree_init(void);
329 void *radix_tree_tag_set(struct radix_tree_root
*root
,
330 unsigned long index
, unsigned int tag
);
331 void *radix_tree_tag_clear(struct radix_tree_root
*root
,
332 unsigned long index
, unsigned int tag
);
333 int radix_tree_tag_get(const struct radix_tree_root
*,
334 unsigned long index
, unsigned int tag
);
335 void radix_tree_iter_tag_set(struct radix_tree_root
*root
,
336 const struct radix_tree_iter
*iter
, unsigned int tag
);
338 radix_tree_gang_lookup_tag(const struct radix_tree_root
*, void **results
,
339 unsigned long first_index
, unsigned int max_items
,
342 radix_tree_gang_lookup_tag_slot(const struct radix_tree_root
*, void ***results
,
343 unsigned long first_index
, unsigned int max_items
,
345 int radix_tree_tagged(const struct radix_tree_root
*root
, unsigned int tag
);
347 static inline void radix_tree_preload_end(void)
352 int radix_tree_split_preload(unsigned old_order
, unsigned new_order
, gfp_t
);
353 int radix_tree_split(struct radix_tree_root
*, unsigned long index
,
355 int radix_tree_join(struct radix_tree_root
*, unsigned long index
,
356 unsigned new_order
, void *);
358 #define RADIX_TREE_ITER_TAG_MASK 0x00FF /* tag index in lower byte */
359 #define RADIX_TREE_ITER_TAGGED 0x0100 /* lookup tagged slots */
360 #define RADIX_TREE_ITER_CONTIG 0x0200 /* stop at first hole */
363 * radix_tree_iter_init - initialize radix tree iterator
365 * @iter: pointer to iterator state
366 * @start: iteration starting index
369 static __always_inline
void **
370 radix_tree_iter_init(struct radix_tree_iter
*iter
, unsigned long start
)
373 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
374 * in the case of a successful tagged chunk lookup. If the lookup was
375 * unsuccessful or non-tagged then nobody cares about ->tags.
377 * Set index to zero to bypass next_index overflow protection.
378 * See the comment in radix_tree_next_chunk() for details.
381 iter
->next_index
= start
;
386 * radix_tree_next_chunk - find next chunk of slots for iteration
388 * @root: radix tree root
389 * @iter: iterator state
390 * @flags: RADIX_TREE_ITER_* flags and tag index
391 * Returns: pointer to chunk first slot, or NULL if there no more left
393 * This function looks up the next chunk in the radix tree starting from
394 * @iter->next_index. It returns a pointer to the chunk's first slot.
395 * Also it fills @iter with data about chunk: position in the tree (index),
396 * its end (next_index), and constructs a bit mask for tagged iterating (tags).
398 void **radix_tree_next_chunk(const struct radix_tree_root
*,
399 struct radix_tree_iter
*iter
, unsigned flags
);
402 * radix_tree_iter_retry - retry this chunk of the iteration
403 * @iter: iterator state
405 * If we iterate over a tree protected only by the RCU lock, a race
406 * against deletion or creation may result in seeing a slot for which
407 * radix_tree_deref_retry() returns true. If so, call this function
408 * and continue the iteration.
410 static inline __must_check
411 void **radix_tree_iter_retry(struct radix_tree_iter
*iter
)
413 iter
->next_index
= iter
->index
;
418 static inline unsigned long
419 __radix_tree_iter_add(struct radix_tree_iter
*iter
, unsigned long slots
)
421 return iter
->index
+ (slots
<< iter_shift(iter
));
425 * radix_tree_iter_resume - resume iterating when the chunk may be invalid
426 * @slot: pointer to current slot
427 * @iter: iterator state
428 * Returns: New slot pointer
430 * If the iterator needs to release then reacquire a lock, the chunk may
431 * have been invalidated by an insertion or deletion. Call this function
432 * before releasing the lock to continue the iteration from the next index.
434 void **__must_check
radix_tree_iter_resume(void **slot
,
435 struct radix_tree_iter
*iter
);
438 * radix_tree_chunk_size - get current chunk size
440 * @iter: pointer to radix tree iterator
441 * Returns: current chunk size
443 static __always_inline
long
444 radix_tree_chunk_size(struct radix_tree_iter
*iter
)
446 return (iter
->next_index
- iter
->index
) >> iter_shift(iter
);
449 #ifdef CONFIG_RADIX_TREE_MULTIORDER
450 void ** __radix_tree_next_slot(void **slot
, struct radix_tree_iter
*iter
,
453 /* Can't happen without sibling entries, but the compiler can't tell that */
454 static inline void ** __radix_tree_next_slot(void **slot
,
455 struct radix_tree_iter
*iter
, unsigned flags
)
462 * radix_tree_next_slot - find next slot in chunk
464 * @slot: pointer to current slot
465 * @iter: pointer to interator state
466 * @flags: RADIX_TREE_ITER_*, should be constant
467 * Returns: pointer to next slot, or NULL if there no more left
469 * This function updates @iter->index in the case of a successful lookup.
470 * For tagged lookup it also eats @iter->tags.
472 * There are several cases where 'slot' can be passed in as NULL to this
473 * function. These cases result from the use of radix_tree_iter_resume() or
474 * radix_tree_iter_retry(). In these cases we don't end up dereferencing
475 * 'slot' because either:
476 * a) we are doing tagged iteration and iter->tags has been set to 0, or
477 * b) we are doing non-tagged iteration, and iter->index and iter->next_index
478 * have been set up so that radix_tree_chunk_size() returns 1 or 0.
480 static __always_inline
void **
481 radix_tree_next_slot(void **slot
, struct radix_tree_iter
*iter
, unsigned flags
)
483 if (flags
& RADIX_TREE_ITER_TAGGED
) {
485 if (unlikely(!iter
->tags
))
487 if (likely(iter
->tags
& 1ul)) {
488 iter
->index
= __radix_tree_iter_add(iter
, 1);
492 if (!(flags
& RADIX_TREE_ITER_CONTIG
)) {
493 unsigned offset
= __ffs(iter
->tags
);
495 iter
->tags
>>= offset
++;
496 iter
->index
= __radix_tree_iter_add(iter
, offset
);
501 long count
= radix_tree_chunk_size(iter
);
503 while (--count
> 0) {
505 iter
->index
= __radix_tree_iter_add(iter
, 1);
509 if (flags
& RADIX_TREE_ITER_CONTIG
) {
510 /* forbid switching to the next chunk */
511 iter
->next_index
= 0;
519 if (unlikely(radix_tree_is_internal_node(*slot
)))
520 return __radix_tree_next_slot(slot
, iter
, flags
);
525 * radix_tree_for_each_slot - iterate over non-empty slots
527 * @slot: the void** variable for pointer to slot
528 * @root: the struct radix_tree_root pointer
529 * @iter: the struct radix_tree_iter pointer
530 * @start: iteration starting index
532 * @slot points to radix tree slot, @iter->index contains its index.
534 #define radix_tree_for_each_slot(slot, root, iter, start) \
535 for (slot = radix_tree_iter_init(iter, start) ; \
536 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \
537 slot = radix_tree_next_slot(slot, iter, 0))
540 * radix_tree_for_each_contig - iterate over contiguous slots
542 * @slot: the void** variable for pointer to slot
543 * @root: the struct radix_tree_root pointer
544 * @iter: the struct radix_tree_iter pointer
545 * @start: iteration starting index
547 * @slot points to radix tree slot, @iter->index contains its index.
549 #define radix_tree_for_each_contig(slot, root, iter, start) \
550 for (slot = radix_tree_iter_init(iter, start) ; \
551 slot || (slot = radix_tree_next_chunk(root, iter, \
552 RADIX_TREE_ITER_CONTIG)) ; \
553 slot = radix_tree_next_slot(slot, iter, \
554 RADIX_TREE_ITER_CONTIG))
557 * radix_tree_for_each_tagged - iterate over tagged slots
559 * @slot: the void** variable for pointer to slot
560 * @root: the struct radix_tree_root pointer
561 * @iter: the struct radix_tree_iter pointer
562 * @start: iteration starting index
565 * @slot points to radix tree slot, @iter->index contains its index.
567 #define radix_tree_for_each_tagged(slot, root, iter, start, tag) \
568 for (slot = radix_tree_iter_init(iter, start) ; \
569 slot || (slot = radix_tree_next_chunk(root, iter, \
570 RADIX_TREE_ITER_TAGGED | tag)) ; \
571 slot = radix_tree_next_slot(slot, iter, \
572 RADIX_TREE_ITER_TAGGED | tag))
574 #endif /* _LINUX_RADIX_TREE_H */