blob: 0d04cd69ab9b8c483895fbf7bab7cd691508e77a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
Nick Piggin7cf9c2c2006-12-06 20:33:44 -08004 * Copyright (C) 2006 Nick Piggin
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -07005 * Copyright (C) 2012 Konstantin Khlebnikov
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
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/preempt.h>
25#include <linux/types.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050026#include <linux/bug.h>
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080027#include <linux/kernel.h>
28#include <linux/rcupdate.h>
29
30/*
Nick Pigginc0bc9872007-10-16 01:24:42 -070031 * An indirect pointer (root->rnode pointing to a radix_tree_node, rather
32 * than a data item) is signalled by the low bit set in the root->rnode
33 * pointer.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080034 *
Nick Pigginc0bc9872007-10-16 01:24:42 -070035 * In this case root->height is > 0, but the indirect pointer tests are
36 * needed for RCU lookups (because root->height is unreliable). The only
37 * time callers need worry about this is when doing a lookup_slot under
38 * RCU.
Nick Piggin27d20fd2010-11-11 14:05:19 -080039 *
40 * Indirect pointer in fact is also used to tag the last pointer of a node
41 * when it is shrunk, before we rcu free the node. See shrink code for
42 * details.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080043 */
Hugh Dickins63286502011-08-03 16:21:18 -070044#define RADIX_TREE_INDIRECT_PTR 1
45/*
46 * A common use of the radix tree is to store pointers to struct pages;
47 * but shmem/tmpfs needs also to store swap entries in the same tree:
48 * those are marked as exceptional entries to distinguish them.
49 * EXCEPTIONAL_ENTRY tests the bit, EXCEPTIONAL_SHIFT shifts content past it.
50 */
51#define RADIX_TREE_EXCEPTIONAL_ENTRY 2
52#define RADIX_TREE_EXCEPTIONAL_SHIFT 2
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080053
Nick Pigginc0bc9872007-10-16 01:24:42 -070054static inline int radix_tree_is_indirect_ptr(void *ptr)
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080055{
Nick Pigginc0bc9872007-10-16 01:24:42 -070056 return (int)((unsigned long)ptr & RADIX_TREE_INDIRECT_PTR);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080057}
58
59/*** radix-tree API starts here ***/
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Jan Karaf446daae2010-08-09 17:19:12 -070061#define RADIX_TREE_MAX_TAGS 3
Nick Piggin612d6c12006-06-23 02:03:22 -070062
63/* root tags are stored in gfp_mask, shifted by __GFP_BITS_SHIFT */
Linus Torvalds1da177e2005-04-16 15:20:36 -070064struct radix_tree_root {
65 unsigned int height;
Al Virofd4f2df2005-10-21 03:18:50 -040066 gfp_t gfp_mask;
Arnd Bergmanna1115572010-02-25 23:43:52 +010067 struct radix_tree_node __rcu *rnode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068};
69
70#define RADIX_TREE_INIT(mask) { \
71 .height = 0, \
72 .gfp_mask = (mask), \
73 .rnode = NULL, \
74}
75
76#define RADIX_TREE(name, mask) \
77 struct radix_tree_root name = RADIX_TREE_INIT(mask)
78
79#define INIT_RADIX_TREE(root, mask) \
80do { \
81 (root)->height = 0; \
82 (root)->gfp_mask = (mask); \
83 (root)->rnode = NULL; \
84} while (0)
85
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080086/**
87 * Radix-tree synchronization
88 *
89 * The radix-tree API requires that users provide all synchronisation (with
90 * specific exceptions, noted below).
91 *
92 * Synchronization of access to the data items being stored in the tree, and
93 * management of their lifetimes must be completely managed by API users.
94 *
95 * For API usage, in general,
Michael Opdenacker59c51592007-05-09 08:57:56 +020096 * - any function _modifying_ the tree or tags (inserting or deleting
Tim Peppereb8dc5e2008-02-03 16:12:47 +020097 * items, setting or clearing tags) must exclude other modifications, and
Nick Piggin7cf9c2c2006-12-06 20:33:44 -080098 * exclude any functions reading the tree.
Michael Opdenacker59c51592007-05-09 08:57:56 +020099 * - any function _reading_ the tree or tags (looking up items or tags,
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800100 * gang lookups) must exclude modifications to the tree, but may occur
101 * concurrently with other readers.
102 *
103 * The notable exceptions to this rule are the following functions:
104 * radix_tree_lookup
Nick Piggin47feff22008-07-25 19:45:29 -0700105 * radix_tree_lookup_slot
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800106 * radix_tree_tag_get
107 * radix_tree_gang_lookup
Nick Piggin47feff22008-07-25 19:45:29 -0700108 * radix_tree_gang_lookup_slot
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800109 * radix_tree_gang_lookup_tag
Nick Piggin47feff22008-07-25 19:45:29 -0700110 * radix_tree_gang_lookup_tag_slot
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800111 * radix_tree_tagged
112 *
Nick Piggin47feff22008-07-25 19:45:29 -0700113 * The first 7 functions are able to be called locklessly, using RCU. The
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800114 * caller must ensure calls to these functions are made within rcu_read_lock()
115 * regions. Other readers (lock-free or otherwise) and modifications may be
116 * running concurrently.
117 *
118 * It is still required that the caller manage the synchronization and lifetimes
119 * of the items. So if RCU lock-free lookups are used, typically this would mean
120 * that the items have their own locks, or are amenable to lock-free access; and
121 * that the items are freed by RCU (or only freed after having been deleted from
122 * the radix tree *and* a synchronize_rcu() grace period).
123 *
124 * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
125 * access to data items when inserting into or looking up from the radix tree)
126 *
David Howellsce826532010-04-06 22:36:20 +0100127 * Note that the value returned by radix_tree_tag_get() may not be relied upon
128 * if only the RCU read lock is held. Functions to set/clear tags and to
129 * delete nodes running concurrently with it may affect its result such that
130 * two consecutive reads in the same locked section may return different
131 * values. If reliability is required, modification functions must also be
132 * excluded from concurrency.
133 *
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800134 * radix_tree_tagged is able to be called without locking or RCU.
135 */
136
137/**
138 * radix_tree_deref_slot - dereference a slot
139 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
140 * Returns: item that was stored in that slot with any direct pointer flag
141 * removed.
142 *
143 * For use with radix_tree_lookup_slot(). Caller must hold tree at least read
Nick Piggin27d20fd2010-11-11 14:05:19 -0800144 * locked across slot lookup and dereference. Not required if write lock is
145 * held (ie. items cannot be concurrently inserted).
146 *
147 * radix_tree_deref_retry must be used to confirm validity of the pointer if
148 * only the read lock is held.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800149 */
150static inline void *radix_tree_deref_slot(void **pslot)
151{
Nick Piggin27d20fd2010-11-11 14:05:19 -0800152 return rcu_dereference(*pslot);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800153}
Nick Piggin27d20fd2010-11-11 14:05:19 -0800154
155/**
Mel Gorman29c1f672011-01-13 15:47:21 -0800156 * radix_tree_deref_slot_protected - dereference a slot without RCU lock but with tree lock held
157 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
158 * Returns: item that was stored in that slot with any direct pointer flag
159 * removed.
160 *
161 * Similar to radix_tree_deref_slot but only used during migration when a pages
162 * mapping is being moved. The caller does not hold the RCU read lock but it
163 * must hold the tree lock to prevent parallel updates.
164 */
165static inline void *radix_tree_deref_slot_protected(void **pslot,
166 spinlock_t *treelock)
167{
168 return rcu_dereference_protected(*pslot, lockdep_is_held(treelock));
169}
170
171/**
Nick Piggin27d20fd2010-11-11 14:05:19 -0800172 * radix_tree_deref_retry - check radix_tree_deref_slot
173 * @arg: pointer returned by radix_tree_deref_slot
174 * Returns: 0 if retry is not required, otherwise retry is required
175 *
176 * radix_tree_deref_retry must be used with radix_tree_deref_slot.
177 */
178static inline int radix_tree_deref_retry(void *arg)
179{
180 return unlikely((unsigned long)arg & RADIX_TREE_INDIRECT_PTR);
181}
182
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800183/**
Hugh Dickins63286502011-08-03 16:21:18 -0700184 * radix_tree_exceptional_entry - radix_tree_deref_slot gave exceptional entry?
185 * @arg: value returned by radix_tree_deref_slot
186 * Returns: 0 if well-aligned pointer, non-0 if exceptional entry.
187 */
188static inline int radix_tree_exceptional_entry(void *arg)
189{
190 /* Not unlikely because radix_tree_exception often tested first */
191 return (unsigned long)arg & RADIX_TREE_EXCEPTIONAL_ENTRY;
192}
193
194/**
195 * radix_tree_exception - radix_tree_deref_slot returned either exception?
196 * @arg: value returned by radix_tree_deref_slot
197 * Returns: 0 if well-aligned pointer, non-0 if either kind of exception.
198 */
199static inline int radix_tree_exception(void *arg)
200{
201 return unlikely((unsigned long)arg &
202 (RADIX_TREE_INDIRECT_PTR | RADIX_TREE_EXCEPTIONAL_ENTRY));
203}
204
205/**
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800206 * radix_tree_replace_slot - replace item in a slot
207 * @pslot: pointer to slot, returned by radix_tree_lookup_slot
208 * @item: new item to store in the slot.
209 *
210 * For use with radix_tree_lookup_slot(). Caller must hold tree write locked
211 * across slot lookup and replacement.
212 */
213static inline void radix_tree_replace_slot(void **pslot, void *item)
214{
Nick Pigginc0bc9872007-10-16 01:24:42 -0700215 BUG_ON(radix_tree_is_indirect_ptr(item));
216 rcu_assign_pointer(*pslot, item);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800217}
218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219int radix_tree_insert(struct radix_tree_root *, unsigned long, void *);
220void *radix_tree_lookup(struct radix_tree_root *, unsigned long);
Hans Reisera4331362005-11-07 00:59:29 -0800221void **radix_tree_lookup_slot(struct radix_tree_root *, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222void *radix_tree_delete(struct radix_tree_root *, unsigned long);
223unsigned int
224radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
225 unsigned long first_index, unsigned int max_items);
Hugh Dickins63286502011-08-03 16:21:18 -0700226unsigned int radix_tree_gang_lookup_slot(struct radix_tree_root *root,
227 void ***results, unsigned long *indices,
Nick Piggin47feff22008-07-25 19:45:29 -0700228 unsigned long first_index, unsigned int max_items);
Fengguang Wu6df8ba42007-10-16 01:24:33 -0700229unsigned long radix_tree_next_hole(struct radix_tree_root *root,
230 unsigned long index, unsigned long max_scan);
Wu Fengguangdc566122009-06-16 15:31:32 -0700231unsigned long radix_tree_prev_hole(struct radix_tree_root *root,
232 unsigned long index, unsigned long max_scan);
Al Virodd0fc662005-10-07 07:46:04 +0100233int radix_tree_preload(gfp_t gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234void radix_tree_init(void);
235void *radix_tree_tag_set(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800236 unsigned long index, unsigned int tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237void *radix_tree_tag_clear(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800238 unsigned long index, unsigned int tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239int radix_tree_tag_get(struct radix_tree_root *root,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800240 unsigned long index, unsigned int tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241unsigned int
242radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800243 unsigned long first_index, unsigned int max_items,
244 unsigned int tag);
Nick Piggin47feff22008-07-25 19:45:29 -0700245unsigned int
246radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
247 unsigned long first_index, unsigned int max_items,
248 unsigned int tag);
Jan Karaebf8aa42010-08-09 17:19:11 -0700249unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
250 unsigned long *first_indexp, unsigned long last_index,
251 unsigned long nr_to_tag,
252 unsigned int fromtag, unsigned int totag);
Jonathan Corbetdaff89f2006-03-25 03:08:05 -0800253int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700254unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256static inline void radix_tree_preload_end(void)
257{
258 preempt_enable();
259}
260
Konstantin Khlebnikov78c1d782012-03-28 14:42:53 -0700261/**
262 * struct radix_tree_iter - radix tree iterator state
263 *
264 * @index: index of current slot
265 * @next_index: next-to-last index for this chunk
266 * @tags: bit-mask for tag-iterating
267 *
268 * This radix tree iterator works in terms of "chunks" of slots. A chunk is a
269 * subinterval of slots contained within one radix tree leaf node. It is
270 * described by a pointer to its first slot and a struct radix_tree_iter
271 * which holds the chunk's position in the tree and its size. For tagged
272 * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
273 * radix tree tag.
274 */
275struct radix_tree_iter {
276 unsigned long index;
277 unsigned long next_index;
278 unsigned long tags;
279};
280
281#define RADIX_TREE_ITER_TAG_MASK 0x00FF /* tag index in lower byte */
282#define RADIX_TREE_ITER_TAGGED 0x0100 /* lookup tagged slots */
283#define RADIX_TREE_ITER_CONTIG 0x0200 /* stop at first hole */
284
285/**
286 * radix_tree_iter_init - initialize radix tree iterator
287 *
288 * @iter: pointer to iterator state
289 * @start: iteration starting index
290 * Returns: NULL
291 */
292static __always_inline void **
293radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start)
294{
295 /*
296 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
297 * in the case of a successful tagged chunk lookup. If the lookup was
298 * unsuccessful or non-tagged then nobody cares about ->tags.
299 *
300 * Set index to zero to bypass next_index overflow protection.
301 * See the comment in radix_tree_next_chunk() for details.
302 */
303 iter->index = 0;
304 iter->next_index = start;
305 return NULL;
306}
307
308/**
309 * radix_tree_next_chunk - find next chunk of slots for iteration
310 *
311 * @root: radix tree root
312 * @iter: iterator state
313 * @flags: RADIX_TREE_ITER_* flags and tag index
314 * Returns: pointer to chunk first slot, or NULL if there no more left
315 *
316 * This function looks up the next chunk in the radix tree starting from
317 * @iter->next_index. It returns a pointer to the chunk's first slot.
318 * Also it fills @iter with data about chunk: position in the tree (index),
319 * its end (next_index), and constructs a bit mask for tagged iterating (tags).
320 */
321void **radix_tree_next_chunk(struct radix_tree_root *root,
322 struct radix_tree_iter *iter, unsigned flags);
323
324/**
325 * radix_tree_chunk_size - get current chunk size
326 *
327 * @iter: pointer to radix tree iterator
328 * Returns: current chunk size
329 */
330static __always_inline unsigned
331radix_tree_chunk_size(struct radix_tree_iter *iter)
332{
333 return iter->next_index - iter->index;
334}
335
336/**
337 * radix_tree_next_slot - find next slot in chunk
338 *
339 * @slot: pointer to current slot
340 * @iter: pointer to interator state
341 * @flags: RADIX_TREE_ITER_*, should be constant
342 * Returns: pointer to next slot, or NULL if there no more left
343 *
344 * This function updates @iter->index in the case of a successful lookup.
345 * For tagged lookup it also eats @iter->tags.
346 */
347static __always_inline void **
348radix_tree_next_slot(void **slot, struct radix_tree_iter *iter, unsigned flags)
349{
350 if (flags & RADIX_TREE_ITER_TAGGED) {
351 iter->tags >>= 1;
352 if (likely(iter->tags & 1ul)) {
353 iter->index++;
354 return slot + 1;
355 }
356 if (!(flags & RADIX_TREE_ITER_CONTIG) && likely(iter->tags)) {
357 unsigned offset = __ffs(iter->tags);
358
359 iter->tags >>= offset;
360 iter->index += offset + 1;
361 return slot + offset + 1;
362 }
363 } else {
364 unsigned size = radix_tree_chunk_size(iter) - 1;
365
366 while (size--) {
367 slot++;
368 iter->index++;
369 if (likely(*slot))
370 return slot;
371 if (flags & RADIX_TREE_ITER_CONTIG)
372 break;
373 }
374 }
375 return NULL;
376}
377
378/**
379 * radix_tree_for_each_chunk - iterate over chunks
380 *
381 * @slot: the void** variable for pointer to chunk first slot
382 * @root: the struct radix_tree_root pointer
383 * @iter: the struct radix_tree_iter pointer
384 * @start: iteration starting index
385 * @flags: RADIX_TREE_ITER_* and tag index
386 *
387 * Locks can be released and reacquired between iterations.
388 */
389#define radix_tree_for_each_chunk(slot, root, iter, start, flags) \
390 for (slot = radix_tree_iter_init(iter, start) ; \
391 (slot = radix_tree_next_chunk(root, iter, flags)) ;)
392
393/**
394 * radix_tree_for_each_chunk_slot - iterate over slots in one chunk
395 *
396 * @slot: the void** variable, at the beginning points to chunk first slot
397 * @iter: the struct radix_tree_iter pointer
398 * @flags: RADIX_TREE_ITER_*, should be constant
399 *
400 * This macro is designed to be nested inside radix_tree_for_each_chunk().
401 * @slot points to the radix tree slot, @iter->index contains its index.
402 */
403#define radix_tree_for_each_chunk_slot(slot, iter, flags) \
404 for (; slot ; slot = radix_tree_next_slot(slot, iter, flags))
405
406/**
407 * radix_tree_for_each_slot - iterate over non-empty slots
408 *
409 * @slot: the void** variable for pointer to slot
410 * @root: the struct radix_tree_root pointer
411 * @iter: the struct radix_tree_iter pointer
412 * @start: iteration starting index
413 *
414 * @slot points to radix tree slot, @iter->index contains its index.
415 */
416#define radix_tree_for_each_slot(slot, root, iter, start) \
417 for (slot = radix_tree_iter_init(iter, start) ; \
418 slot || (slot = radix_tree_next_chunk(root, iter, 0)) ; \
419 slot = radix_tree_next_slot(slot, iter, 0))
420
421/**
422 * radix_tree_for_each_contig - iterate over contiguous slots
423 *
424 * @slot: the void** variable for pointer to slot
425 * @root: the struct radix_tree_root pointer
426 * @iter: the struct radix_tree_iter pointer
427 * @start: iteration starting index
428 *
429 * @slot points to radix tree slot, @iter->index contains its index.
430 */
431#define radix_tree_for_each_contig(slot, root, iter, start) \
432 for (slot = radix_tree_iter_init(iter, start) ; \
433 slot || (slot = radix_tree_next_chunk(root, iter, \
434 RADIX_TREE_ITER_CONTIG)) ; \
435 slot = radix_tree_next_slot(slot, iter, \
436 RADIX_TREE_ITER_CONTIG))
437
438/**
439 * radix_tree_for_each_tagged - iterate over tagged slots
440 *
441 * @slot: the void** variable for pointer to slot
442 * @root: the struct radix_tree_root pointer
443 * @iter: the struct radix_tree_iter pointer
444 * @start: iteration starting index
445 * @tag: tag index
446 *
447 * @slot points to radix tree slot, @iter->index contains its index.
448 */
449#define radix_tree_for_each_tagged(slot, root, iter, start, tag) \
450 for (slot = radix_tree_iter_init(iter, start) ; \
451 slot || (slot = radix_tree_next_chunk(root, iter, \
452 RADIX_TREE_ITER_TAGGED | tag)) ; \
453 slot = radix_tree_next_slot(slot, iter, \
454 RADIX_TREE_ITER_TAGGED))
455
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456#endif /* _LINUX_RADIX_TREE_H */