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/preempt.h>
26 #include <linux/types.h>
27 #include <linux/bug.h>
28 #include <linux/kernel.h>
29 #include <linux/rcupdate.h>
32 * An indirect pointer (root->rnode pointing to a radix_tree_node, rather
33 * than a data item) is signalled by the low bit set in the root->rnode
36 * In this case root->height is > 0, but the indirect pointer tests are
37 * needed for RCU lookups (because root->height is unreliable). The only
38 * time callers need worry about this is when doing a lookup_slot under
41 * Indirect pointer in fact is also used to tag the last pointer of a node
42 * when it is shrunk, before we rcu free the node. See shrink code for
45 #define RADIX_TREE_INDIRECT_PTR 1
47 * A common use of the radix tree is to store pointers to struct pages;
48 * but shmem/tmpfs needs also to store swap entries in the same tree:
49 * those are marked as exceptional entries to distinguish them.
50 * EXCEPTIONAL_ENTRY tests the bit, EXCEPTIONAL_SHIFT shifts content past it.
52 #define RADIX_TREE_EXCEPTIONAL_ENTRY 2
53 #define RADIX_TREE_EXCEPTIONAL_SHIFT 2
55 #define RADIX_DAX_MASK 0xf
56 #define RADIX_DAX_SHIFT 4
57 #define RADIX_DAX_PTE (0x4 | RADIX_TREE_EXCEPTIONAL_ENTRY)
58 #define RADIX_DAX_PMD (0x8 | RADIX_TREE_EXCEPTIONAL_ENTRY)
59 #define RADIX_DAX_TYPE(entry) ((unsigned long)entry & RADIX_DAX_MASK)
60 #define RADIX_DAX_SECTOR(entry) (((unsigned long)entry >> RADIX_DAX_SHIFT))
61 #define RADIX_DAX_ENTRY(sector, pmd) ((void *)((unsigned long)sector << \
62 RADIX_DAX_SHIFT | (pmd ? RADIX_DAX_PMD : RADIX_DAX_PTE)))
64 static inline int radix_tree_is_indirect_ptr(void *ptr
)
66 return (int)((unsigned long)ptr
& RADIX_TREE_INDIRECT_PTR
);
69 /*** radix-tree API starts here ***/
71 #define RADIX_TREE_MAX_TAGS 3
73 #ifndef RADIX_TREE_MAP_SHIFT
74 #define RADIX_TREE_MAP_SHIFT (CONFIG_BASE_SMALL ? 4 : 6)
77 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
78 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
80 #define RADIX_TREE_TAG_LONGS \
81 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
83 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
84 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
85 RADIX_TREE_MAP_SHIFT))
87 /* Height component in node->path */
88 #define RADIX_TREE_HEIGHT_SHIFT (RADIX_TREE_MAX_PATH + 1)
89 #define RADIX_TREE_HEIGHT_MASK ((1UL << RADIX_TREE_HEIGHT_SHIFT) - 1)
91 /* Internally used bits of node->count */
92 #define RADIX_TREE_COUNT_SHIFT (RADIX_TREE_MAP_SHIFT + 1)
93 #define RADIX_TREE_COUNT_MASK ((1UL << RADIX_TREE_COUNT_SHIFT) - 1)
95 struct radix_tree_node
{
96 unsigned int path
; /* Offset in parent & height from the bottom */
100 /* Used when ascending tree */
101 struct radix_tree_node
*parent
;
105 /* Used when freeing node */
106 struct rcu_head rcu_head
;
109 struct list_head private_list
;
110 void __rcu
*slots
[RADIX_TREE_MAP_SIZE
];
111 unsigned long tags
[RADIX_TREE_MAX_TAGS
][RADIX_TREE_TAG_LONGS
];
114 /* root tags are stored in gfp_mask, shifted by __GFP_BITS_SHIFT */
115 struct radix_tree_root
{
118 struct radix_tree_node __rcu
*rnode
;
121 #define RADIX_TREE_INIT(mask) { \
123 .gfp_mask = (mask), \
127 #define RADIX_TREE(name, mask) \
128 struct radix_tree_root name = RADIX_TREE_INIT(mask)
130 #define INIT_RADIX_TREE(root, mask) \
132 (root)->height = 0; \
133 (root)->gfp_mask = (mask); \
134 (root)->rnode = NULL; \
137 static inline bool radix_tree_empty(struct radix_tree_root
*root
)
139 return root
->rnode
== NULL
;
143 * Radix-tree synchronization
145 * The radix-tree API requires that users provide all synchronisation (with
146 * specific exceptions, noted below).
148 * Synchronization of access to the data items being stored in the tree, and
149 * management of their lifetimes must be completely managed by API users.
151 * For API usage, in general,
152 * - any function _modifying_ the tree or tags (inserting or deleting
153 * items, setting or clearing tags) must exclude other modifications, and
154 * exclude any functions reading the tree.
155 * - any function _reading_ the tree or tags (looking up items or tags,
156 * gang lookups) must exclude modifications to the tree, but may occur
157 * concurrently with other readers.
159 * The notable exceptions to this rule are the following functions:
160 * __radix_tree_lookup
162 * radix_tree_lookup_slot
164 * radix_tree_gang_lookup
165 * radix_tree_gang_lookup_slot
166 * radix_tree_gang_lookup_tag
167 * radix_tree_gang_lookup_tag_slot
170 * The first 8 functions are able to be called locklessly, using RCU. The
171 * caller must ensure calls to these functions are made within rcu_read_lock()
172 * regions. Other readers (lock-free or otherwise) and modifications may be
173 * running concurrently.
175 * It is still required that the caller manage the synchronization and lifetimes
176 * of the items. So if RCU lock-free lookups are used, typically this would mean
177 * that the items have their own locks, or are amenable to lock-free access; and
178 * that the items are freed by RCU (or only freed after having been deleted from
179 * the radix tree *and* a synchronize_rcu() grace period).
181 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
182 * access to data items when inserting into or looking up from the radix tree)
184 * Note that the value returned by radix_tree_tag_get() may not be relied upon
185 * if only the RCU read lock is held. Functions to set/clear tags and to
186 * delete nodes running concurrently with it may affect its result such that
187 * two consecutive reads in the same locked section may return different
188 * values. If reliability is required, modification functions must also be
189 * excluded from concurrency.
191 * radix_tree_tagged is able to be called without locking or RCU.
195 * radix_tree_deref_slot - dereference a slot
196 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
197 * Returns: item that was stored in that slot with any direct pointer flag
200 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read
201 * locked across slot lookup and dereference. Not required if write lock is
202 * held (ie. items cannot be concurrently inserted).
204 * radix_tree_deref_retry must be used to confirm validity of the pointer if
205 * only the read lock is held.
207 static inline void *radix_tree_deref_slot(void **pslot
)
209 return rcu_dereference(*pslot
);
213 * radix_tree_deref_slot_protected - dereference a slot without RCU lock but with tree lock held
214 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
215 * Returns: item that was stored in that slot with any direct pointer flag
218 * Similar to radix_tree_deref_slot but only used during migration when a pages
219 * mapping is being moved. The caller does not hold the RCU read lock but it
220 * must hold the tree lock to prevent parallel updates.
222 static inline void *radix_tree_deref_slot_protected(void **pslot
,
223 spinlock_t
*treelock
)
225 return rcu_dereference_protected(*pslot
, lockdep_is_held(treelock
));
229 * radix_tree_deref_retry - check radix_tree_deref_slot
230 * @arg: pointer returned by radix_tree_deref_slot
231 * Returns: 0 if retry is not required, otherwise retry is required
233 * radix_tree_deref_retry must be used with radix_tree_deref_slot.
235 static inline int radix_tree_deref_retry(void *arg
)
237 return unlikely((unsigned long)arg
& RADIX_TREE_INDIRECT_PTR
);
241 * radix_tree_exceptional_entry - radix_tree_deref_slot gave exceptional entry?
242 * @arg: value returned by radix_tree_deref_slot
243 * Returns: 0 if well-aligned pointer, non-0 if exceptional entry.
245 static inline int radix_tree_exceptional_entry(void *arg
)
247 /* Not unlikely because radix_tree_exception often tested first */
248 return (unsigned long)arg
& RADIX_TREE_EXCEPTIONAL_ENTRY
;
252 * radix_tree_exception - radix_tree_deref_slot returned either exception?
253 * @arg: value returned by radix_tree_deref_slot
254 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception.
256 static inline int radix_tree_exception(void *arg
)
258 return unlikely((unsigned long)arg
&
259 (RADIX_TREE_INDIRECT_PTR
| RADIX_TREE_EXCEPTIONAL_ENTRY
));
263 * radix_tree_replace_slot - replace item in a slot
264 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
265 * @item: new item to store in the slot.
267 * For use with radix_tree_lookup_slot(). Caller must hold tree write locked
268 * across slot lookup and replacement.
270 static inline void radix_tree_replace_slot(void **pslot
, void *item
)
272 BUG_ON(radix_tree_is_indirect_ptr(item
));
273 rcu_assign_pointer(*pslot
, item
);
276 int __radix_tree_create(struct radix_tree_root
*root
, unsigned long index
,
277 unsigned order
, struct radix_tree_node
**nodep
,
279 int __radix_tree_insert(struct radix_tree_root
*, unsigned long index
,
280 unsigned order
, void *);
281 static inline int radix_tree_insert(struct radix_tree_root
*root
,
282 unsigned long index
, void *entry
)
284 return __radix_tree_insert(root
, index
, 0, entry
);
286 void *__radix_tree_lookup(struct radix_tree_root
*root
, unsigned long index
,
287 struct radix_tree_node
**nodep
, void ***slotp
);
288 void *radix_tree_lookup(struct radix_tree_root
*, unsigned long);
289 void **radix_tree_lookup_slot(struct radix_tree_root
*, unsigned long);
290 bool __radix_tree_delete_node(struct radix_tree_root
*root
,
291 struct radix_tree_node
*node
);
292 void *radix_tree_delete_item(struct radix_tree_root
*, unsigned long, void *);
293 void *radix_tree_delete(struct radix_tree_root
*, unsigned long);
295 radix_tree_gang_lookup(struct radix_tree_root
*root
, void **results
,
296 unsigned long first_index
, unsigned int max_items
);
297 unsigned int radix_tree_gang_lookup_slot(struct radix_tree_root
*root
,
298 void ***results
, unsigned long *indices
,
299 unsigned long first_index
, unsigned int max_items
);
300 int radix_tree_preload(gfp_t gfp_mask
);
301 int radix_tree_maybe_preload(gfp_t gfp_mask
);
302 void radix_tree_init(void);
303 void *radix_tree_tag_set(struct radix_tree_root
*root
,
304 unsigned long index
, unsigned int tag
);
305 void *radix_tree_tag_clear(struct radix_tree_root
*root
,
306 unsigned long index
, unsigned int tag
);
307 int radix_tree_tag_get(struct radix_tree_root
*root
,
308 unsigned long index
, unsigned int tag
);
310 radix_tree_gang_lookup_tag(struct radix_tree_root
*root
, void **results
,
311 unsigned long first_index
, unsigned int max_items
,
314 radix_tree_gang_lookup_tag_slot(struct radix_tree_root
*root
, void ***results
,
315 unsigned long first_index
, unsigned int max_items
,
317 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root
*root
,
318 unsigned long *first_indexp
, unsigned long last_index
,
319 unsigned long nr_to_tag
,
320 unsigned int fromtag
, unsigned int totag
);
321 int radix_tree_tagged(struct radix_tree_root
*root
, unsigned int tag
);
322 unsigned long radix_tree_locate_item(struct radix_tree_root
*root
, void *item
);
324 static inline void radix_tree_preload_end(void)
330 * struct radix_tree_iter - radix tree iterator state
332 * @index: index of current slot
333 * @next_index: next-to-last index for this chunk
334 * @tags: bit-mask for tag-iterating
336 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a
337 * subinterval of slots contained within one radix tree leaf node. It is
338 * described by a pointer to its first slot and a struct radix_tree_iter
339 * which holds the chunk's position in the tree and its size. For tagged
340 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
343 struct radix_tree_iter
{
345 unsigned long next_index
;
349 #define RADIX_TREE_ITER_TAG_MASK 0x00FF /* tag index in lower byte */
350 #define RADIX_TREE_ITER_TAGGED 0x0100 /* lookup tagged slots */
351 #define RADIX_TREE_ITER_CONTIG 0x0200 /* stop at first hole */
354 * radix_tree_iter_init - initialize radix tree iterator
356 * @iter: pointer to iterator state
357 * @start: iteration starting index
360 static __always_inline
void **
361 radix_tree_iter_init(struct radix_tree_iter
*iter
, unsigned long start
)
364 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
365 * in the case of a successful tagged chunk lookup. If the lookup was
366 * unsuccessful or non-tagged then nobody cares about ->tags.
368 * Set index to zero to bypass next_index overflow protection.
369 * See the comment in radix_tree_next_chunk() for details.
372 iter
->next_index
= start
;
377 * radix_tree_next_chunk - find next chunk of slots for iteration
379 * @root: radix tree root
380 * @iter: iterator state
381 * @flags: RADIX_TREE_ITER_* flags and tag index
382 * Returns: pointer to chunk first slot, or NULL if there no more left
384 * This function looks up the next chunk in the radix tree starting from
385 * @iter->next_index. It returns a pointer to the chunk's first slot.
386 * Also it fills @iter with data about chunk: position in the tree (index),
387 * its end (next_index), and constructs a bit mask for tagged iterating (tags).
389 void **radix_tree_next_chunk(struct radix_tree_root
*root
,
390 struct radix_tree_iter
*iter
, unsigned flags
);
393 * radix_tree_iter_retry - retry this chunk of the iteration
394 * @iter: iterator state
396 * If we iterate over a tree protected only by the RCU lock, a race
397 * against deletion or creation may result in seeing a slot for which
398 * radix_tree_deref_retry() returns true. If so, call this function
399 * and continue the iteration.
401 static inline __must_check
402 void **radix_tree_iter_retry(struct radix_tree_iter
*iter
)
404 iter
->next_index
= iter
->index
;
409 * radix_tree_iter_next - resume iterating when the chunk may be invalid
410 * @iter: iterator state
412 * If the iterator needs to release then reacquire a lock, the chunk may
413 * have been invalidated by an insertion or deletion. Call this function
414 * to continue the iteration from the next index.
416 static inline __must_check
417 void **radix_tree_iter_next(struct radix_tree_iter
*iter
)
419 iter
->next_index
= iter
->index
+ 1;
425 * radix_tree_chunk_size - get current chunk size
427 * @iter: pointer to radix tree iterator
428 * Returns: current chunk size
430 static __always_inline
long
431 radix_tree_chunk_size(struct radix_tree_iter
*iter
)
433 return iter
->next_index
- iter
->index
;
437 * radix_tree_next_slot - find next slot in chunk
439 * @slot: pointer to current slot
440 * @iter: pointer to interator state
441 * @flags: RADIX_TREE_ITER_*, should be constant
442 * Returns: pointer to next slot, or NULL if there no more left
444 * This function updates @iter->index in the case of a successful lookup.
445 * For tagged lookup it also eats @iter->tags.
447 static __always_inline
void **
448 radix_tree_next_slot(void **slot
, struct radix_tree_iter
*iter
, unsigned flags
)
450 if (flags
& RADIX_TREE_ITER_TAGGED
) {
452 if (likely(iter
->tags
& 1ul)) {
456 if (!(flags
& RADIX_TREE_ITER_CONTIG
) && likely(iter
->tags
)) {
457 unsigned offset
= __ffs(iter
->tags
);
459 iter
->tags
>>= offset
;
460 iter
->index
+= offset
+ 1;
461 return slot
+ offset
+ 1;
464 long size
= radix_tree_chunk_size(iter
);
471 if (flags
& RADIX_TREE_ITER_CONTIG
) {
472 /* forbid switching to the next chunk */
473 iter
->next_index
= 0;
482 * radix_tree_for_each_chunk - iterate over chunks
484 * @slot: the void** variable for pointer to chunk first slot
485 * @root: the struct radix_tree_root pointer
486 * @iter: the struct radix_tree_iter pointer
487 * @start: iteration starting index
488 * @flags: RADIX_TREE_ITER_* and tag index
490 * Locks can be released and reacquired between iterations.
492 #define radix_tree_for_each_chunk(slot, root, iter, start, flags) \
493 for (slot = radix_tree_iter_init(iter, start) ; \
494 (slot = radix_tree_next_chunk(root, iter, flags)) ;)
497 * radix_tree_for_each_chunk_slot - iterate over slots in one chunk
499 * @slot: the void** variable, at the beginning points to chunk first slot
500 * @iter: the struct radix_tree_iter pointer
501 * @flags: RADIX_TREE_ITER_*, should be constant
503 * This macro is designed to be nested inside radix_tree_for_each_chunk().
504 * @slot points to the radix tree slot, @iter->index contains its index.
506 #define radix_tree_for_each_chunk_slot(slot, iter, flags) \
507 for (; slot ; slot = radix_tree_next_slot(slot, iter, flags))
510 * radix_tree_for_each_slot - iterate over non-empty slots
512 * @slot: the void** variable for pointer to slot
513 * @root: the struct radix_tree_root pointer
514 * @iter: the struct radix_tree_iter pointer
515 * @start: iteration starting index
517 * @slot points to radix tree slot, @iter->index contains its index.
519 #define radix_tree_for_each_slot(slot, root, iter, start) \
520 for (slot = radix_tree_iter_init(iter, start) ; \
521 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \
522 slot = radix_tree_next_slot(slot, iter, 0))
525 * radix_tree_for_each_contig - iterate over contiguous 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_contig(slot, root, iter, start) \
535 for (slot = radix_tree_iter_init(iter, start) ; \
536 slot || (slot = radix_tree_next_chunk(root, iter, \
537 RADIX_TREE_ITER_CONTIG)) ; \
538 slot = radix_tree_next_slot(slot, iter, \
539 RADIX_TREE_ITER_CONTIG))
542 * radix_tree_for_each_tagged - iterate over tagged slots
544 * @slot: the void** variable for pointer to slot
545 * @root: the struct radix_tree_root pointer
546 * @iter: the struct radix_tree_iter pointer
547 * @start: iteration starting index
550 * @slot points to radix tree slot, @iter->index contains its index.
552 #define radix_tree_for_each_tagged(slot, root, iter, start, tag) \
553 for (slot = radix_tree_iter_init(iter, start) ; \
554 slot || (slot = radix_tree_next_chunk(root, iter, \
555 RADIX_TREE_ITER_TAGGED | tag)) ; \
556 slot = radix_tree_next_slot(slot, iter, \
557 RADIX_TREE_ITER_TAGGED))
559 #endif /* _LINUX_RADIX_TREE_H */