1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * Copyright (C) 2001 Momchil Velikov
4  * Portions Copyright (C) 2001 Christoph Hellwig
5  * Copyright (C) 2006 Nick Piggin
6  * Copyright (C) 2012 Konstantin Khlebnikov
7  */
8 #ifndef _LINUX_RADIX_TREE_H
9 #define _LINUX_RADIX_TREE_H
10 
11 #include <linux/bitops.h>
12 #include <linux/gfp_types.h>
13 #include <linux/list.h>
14 #include <linux/lockdep.h>
15 #include <linux/math.h>
16 #include <linux/percpu.h>
17 #include <linux/preempt.h>
18 #include <linux/rcupdate.h>
19 #include <linux/spinlock.h>
20 #include <linux/types.h>
21 #include <linux/xarray.h>
22 #include <linux/local_lock.h>
23 
24 /* Keep unconverted code working */
25 #define radix_tree_root		xarray
26 #define radix_tree_node		xa_node
27 
28 struct radix_tree_preload {
29 	local_lock_t lock;
30 	unsigned nr;
31 	/* nodes->parent points to next preallocated node */
32 	struct radix_tree_node *nodes;
33 };
34 DECLARE_PER_CPU(struct radix_tree_preload, radix_tree_preloads);
35 
36 /*
37  * The bottom two bits of the slot determine how the remaining bits in the
38  * slot are interpreted:
39  *
40  * 00 - data pointer
41  * 10 - internal entry
42  * x1 - value entry
43  *
44  * The internal entry may be a pointer to the next level in the tree, a
45  * sibling entry, or an indicator that the entry in this slot has been moved
46  * to another location in the tree and the lookup should be restarted.  While
47  * NULL fits the 'data pointer' pattern, it means that there is no entry in
48  * the tree for this index (no matter what level of the tree it is found at).
49  * This means that storing a NULL entry in the tree is the same as deleting
50  * the entry from the tree.
51  */
52 #define RADIX_TREE_ENTRY_MASK		3UL
53 #define RADIX_TREE_INTERNAL_NODE	2UL
54 
radix_tree_is_internal_node(void * ptr)55 static inline bool radix_tree_is_internal_node(void *ptr)
56 {
57 	return ((unsigned long)ptr & RADIX_TREE_ENTRY_MASK) ==
58 				RADIX_TREE_INTERNAL_NODE;
59 }
60 
61 /*** radix-tree API starts here ***/
62 
63 #define RADIX_TREE_MAP_SHIFT	XA_CHUNK_SHIFT
64 #define RADIX_TREE_MAP_SIZE	(1UL << RADIX_TREE_MAP_SHIFT)
65 #define RADIX_TREE_MAP_MASK	(RADIX_TREE_MAP_SIZE-1)
66 
67 #define RADIX_TREE_MAX_TAGS	XA_MAX_MARKS
68 #define RADIX_TREE_TAG_LONGS	XA_MARK_LONGS
69 
70 #define RADIX_TREE_INDEX_BITS  (8 /* CHAR_BIT */ * sizeof(unsigned long))
71 #define RADIX_TREE_MAX_PATH (DIV_ROUND_UP(RADIX_TREE_INDEX_BITS, \
72 					  RADIX_TREE_MAP_SHIFT))
73 
74 /* The IDR tag is stored in the low bits of xa_flags */
75 #define ROOT_IS_IDR	((__force gfp_t)4)
76 /* The top bits of xa_flags are used to store the root tags */
77 #define ROOT_TAG_SHIFT	(__GFP_BITS_SHIFT)
78 
79 #define RADIX_TREE_INIT(name, mask)	XARRAY_INIT(name, mask)
80 
81 #define RADIX_TREE(name, mask) \
82 	struct radix_tree_root name = RADIX_TREE_INIT(name, mask)
83 
84 #define INIT_RADIX_TREE(root, mask) xa_init_flags(root, mask)
85 
radix_tree_empty(const struct radix_tree_root * root)86 static inline bool radix_tree_empty(const struct radix_tree_root *root)
87 {
88 	return root->xa_head == NULL;
89 }
90 
91 /**
92  * struct radix_tree_iter - radix tree iterator state
93  *
94  * @index:	index of current slot
95  * @next_index:	one beyond the last index for this chunk
96  * @tags:	bit-mask for tag-iterating
97  * @node:	node that contains current slot
98  *
99  * This radix tree iterator works in terms of "chunks" of slots.  A chunk is a
100  * subinterval of slots contained within one radix tree leaf node.  It is
101  * described by a pointer to its first slot and a struct radix_tree_iter
102  * which holds the chunk's position in the tree and its size.  For tagged
103  * iteration radix_tree_iter also holds the slots' bit-mask for one chosen
104  * radix tree tag.
105  */
106 struct radix_tree_iter {
107 	unsigned long	index;
108 	unsigned long	next_index;
109 	unsigned long	tags;
110 	struct radix_tree_node *node;
111 };
112 
113 /**
114  * Radix-tree synchronization
115  *
116  * The radix-tree API requires that users provide all synchronisation (with
117  * specific exceptions, noted below).
118  *
119  * Synchronization of access to the data items being stored in the tree, and
120  * management of their lifetimes must be completely managed by API users.
121  *
122  * For API usage, in general,
123  * - any function _modifying_ the tree or tags (inserting or deleting
124  *   items, setting or clearing tags) must exclude other modifications, and
125  *   exclude any functions reading the tree.
126  * - any function _reading_ the tree or tags (looking up items or tags,
127  *   gang lookups) must exclude modifications to the tree, but may occur
128  *   concurrently with other readers.
129  *
130  * The notable exceptions to this rule are the following functions:
131  * __radix_tree_lookup
132  * radix_tree_lookup
133  * radix_tree_lookup_slot
134  * radix_tree_tag_get
135  * radix_tree_gang_lookup
136  * radix_tree_gang_lookup_tag
137  * radix_tree_gang_lookup_tag_slot
138  * radix_tree_tagged
139  *
140  * The first 7 functions are able to be called locklessly, using RCU. The
141  * caller must ensure calls to these functions are made within rcu_read_lock()
142  * regions. Other readers (lock-free or otherwise) and modifications may be
143  * running concurrently.
144  *
145  * It is still required that the caller manage the synchronization and lifetimes
146  * of the items. So if RCU lock-free lookups are used, typically this would mean
147  * that the items have their own locks, or are amenable to lock-free access; and
148  * that the items are freed by RCU (or only freed after having been deleted from
149  * the radix tree *and* a synchronize_rcu() grace period).
150  *
151  * (Note, rcu_assign_pointer and rcu_dereference are not needed to control
152  * access to data items when inserting into or looking up from the radix tree)
153  *
154  * Note that the value returned by radix_tree_tag_get() may not be relied upon
155  * if only the RCU read lock is held.  Functions to set/clear tags and to
156  * delete nodes running concurrently with it may affect its result such that
157  * two consecutive reads in the same locked section may return different
158  * values.  If reliability is required, modification functions must also be
159  * excluded from concurrency.
160  *
161  * radix_tree_tagged is able to be called without locking or RCU.
162  */
163 
164 /**
165  * radix_tree_deref_slot - dereference a slot
166  * @slot: slot pointer, returned by radix_tree_lookup_slot
167  *
168  * For use with radix_tree_lookup_slot().  Caller must hold tree at least read
169  * locked across slot lookup and dereference. Not required if write lock is
170  * held (ie. items cannot be concurrently inserted).
171  *
172  * radix_tree_deref_retry must be used to confirm validity of the pointer if
173  * only the read lock is held.
174  *
175  * Return: entry stored in that slot.
176  */
radix_tree_deref_slot(void __rcu ** slot)177 static inline void *radix_tree_deref_slot(void __rcu **slot)
178 {
179 	return rcu_dereference(*slot);
180 }
181 
182 /**
183  * radix_tree_deref_slot_protected - dereference a slot with tree lock held
184  * @slot: slot pointer, returned by radix_tree_lookup_slot
185  *
186  * Similar to radix_tree_deref_slot.  The caller does not hold the RCU read
187  * lock but it must hold the tree lock to prevent parallel updates.
188  *
189  * Return: entry stored in that slot.
190  */
radix_tree_deref_slot_protected(void __rcu ** slot,spinlock_t * treelock)191 static inline void *radix_tree_deref_slot_protected(void __rcu **slot,
192 							spinlock_t *treelock)
193 {
194 	return rcu_dereference_protected(*slot, lockdep_is_held(treelock));
195 }
196 
197 /**
198  * radix_tree_deref_retry	- check radix_tree_deref_slot
199  * @arg:	pointer returned by radix_tree_deref_slot
200  * Returns:	0 if retry is not required, otherwise retry is required
201  *
202  * radix_tree_deref_retry must be used with radix_tree_deref_slot.
203  */
radix_tree_deref_retry(void * arg)204 static inline int radix_tree_deref_retry(void *arg)
205 {
206 	return unlikely(radix_tree_is_internal_node(arg));
207 }
208 
209 /**
210  * radix_tree_exception	- radix_tree_deref_slot returned either exception?
211  * @arg:	value returned by radix_tree_deref_slot
212  * Returns:	0 if well-aligned pointer, non-0 if either kind of exception.
213  */
radix_tree_exception(void * arg)214 static inline int radix_tree_exception(void *arg)
215 {
216 	return unlikely((unsigned long)arg & RADIX_TREE_ENTRY_MASK);
217 }
218 
219 int radix_tree_insert(struct radix_tree_root *, unsigned long index,
220 			void *);
221 void *__radix_tree_lookup(const struct radix_tree_root *, unsigned long index,
222 			  struct radix_tree_node **nodep, void __rcu ***slotp);
223 void *radix_tree_lookup(const struct radix_tree_root *, unsigned long);
224 void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *,
225 					unsigned long index);
226 void __radix_tree_replace(struct radix_tree_root *, struct radix_tree_node *,
227 			  void __rcu **slot, void *entry);
228 void radix_tree_iter_replace(struct radix_tree_root *,
229 		const struct radix_tree_iter *, void __rcu **slot, void *entry);
230 void radix_tree_replace_slot(struct radix_tree_root *,
231 			     void __rcu **slot, void *entry);
232 void radix_tree_iter_delete(struct radix_tree_root *,
233 			struct radix_tree_iter *iter, void __rcu **slot);
234 void *radix_tree_delete_item(struct radix_tree_root *, unsigned long, void *);
235 void *radix_tree_delete(struct radix_tree_root *, unsigned long);
236 unsigned int radix_tree_gang_lookup(const struct radix_tree_root *,
237 			void **results, unsigned long first_index,
238 			unsigned int max_items);
239 int radix_tree_preload(gfp_t gfp_mask);
240 int radix_tree_maybe_preload(gfp_t gfp_mask);
241 void radix_tree_init(void);
242 void *radix_tree_tag_set(struct radix_tree_root *,
243 			unsigned long index, unsigned int tag);
244 void *radix_tree_tag_clear(struct radix_tree_root *,
245 			unsigned long index, unsigned int tag);
246 int radix_tree_tag_get(const struct radix_tree_root *,
247 			unsigned long index, unsigned int tag);
248 void radix_tree_iter_tag_clear(struct radix_tree_root *,
249 		const struct radix_tree_iter *iter, unsigned int tag);
250 unsigned int radix_tree_gang_lookup_tag(const struct radix_tree_root *,
251 		void **results, unsigned long first_index,
252 		unsigned int max_items, unsigned int tag);
253 unsigned int radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *,
254 		void __rcu ***results, unsigned long first_index,
255 		unsigned int max_items, unsigned int tag);
256 int radix_tree_tagged(const struct radix_tree_root *, unsigned int tag);
257 
radix_tree_preload_end(void)258 static inline void radix_tree_preload_end(void)
259 {
260 	local_unlock(&radix_tree_preloads.lock);
261 }
262 
263 void __rcu **idr_get_free(struct radix_tree_root *root,
264 			      struct radix_tree_iter *iter, gfp_t gfp,
265 			      unsigned long max);
266 
267 enum {
268 	RADIX_TREE_ITER_TAG_MASK = 0x0f,	/* tag index in lower nybble */
269 	RADIX_TREE_ITER_TAGGED   = 0x10,	/* lookup tagged slots */
270 	RADIX_TREE_ITER_CONTIG   = 0x20,	/* stop at first hole */
271 };
272 
273 /**
274  * radix_tree_iter_init - initialize radix tree iterator
275  *
276  * @iter:	pointer to iterator state
277  * @start:	iteration starting index
278  * Returns:	NULL
279  */
280 static __always_inline void __rcu **
radix_tree_iter_init(struct radix_tree_iter * iter,unsigned long start)281 radix_tree_iter_init(struct radix_tree_iter *iter, unsigned long start)
282 {
283 	/*
284 	 * Leave iter->tags uninitialized. radix_tree_next_chunk() will fill it
285 	 * in the case of a successful tagged chunk lookup.  If the lookup was
286 	 * unsuccessful or non-tagged then nobody cares about ->tags.
287 	 *
288 	 * Set index to zero to bypass next_index overflow protection.
289 	 * See the comment in radix_tree_next_chunk() for details.
290 	 */
291 	iter->index = 0;
292 	iter->next_index = start;
293 	return NULL;
294 }
295 
296 /**
297  * radix_tree_next_chunk - find next chunk of slots for iteration
298  *
299  * @root:	radix tree root
300  * @iter:	iterator state
301  * @flags:	RADIX_TREE_ITER_* flags and tag index
302  * Returns:	pointer to chunk first slot, or NULL if there no more left
303  *
304  * This function looks up the next chunk in the radix tree starting from
305  * @iter->next_index.  It returns a pointer to the chunk's first slot.
306  * Also it fills @iter with data about chunk: position in the tree (index),
307  * its end (next_index), and constructs a bit mask for tagged iterating (tags).
308  */
309 void __rcu **radix_tree_next_chunk(const struct radix_tree_root *,
310 			     struct radix_tree_iter *iter, unsigned flags);
311 
312 /**
313  * radix_tree_iter_lookup - look up an index in the radix tree
314  * @root: radix tree root
315  * @iter: iterator state
316  * @index: key to look up
317  *
318  * If @index is present in the radix tree, this function returns the slot
319  * containing it and updates @iter to describe the entry.  If @index is not
320  * present, it returns NULL.
321  */
322 static inline void __rcu **
radix_tree_iter_lookup(const struct radix_tree_root * root,struct radix_tree_iter * iter,unsigned long index)323 radix_tree_iter_lookup(const struct radix_tree_root *root,
324 			struct radix_tree_iter *iter, unsigned long index)
325 {
326 	radix_tree_iter_init(iter, index);
327 	return radix_tree_next_chunk(root, iter, RADIX_TREE_ITER_CONTIG);
328 }
329 
330 /**
331  * radix_tree_iter_retry - retry this chunk of the iteration
332  * @iter:	iterator state
333  *
334  * If we iterate over a tree protected only by the RCU lock, a race
335  * against deletion or creation may result in seeing a slot for which
336  * radix_tree_deref_retry() returns true.  If so, call this function
337  * and continue the iteration.
338  */
339 static inline __must_check
radix_tree_iter_retry(struct radix_tree_iter * iter)340 void __rcu **radix_tree_iter_retry(struct radix_tree_iter *iter)
341 {
342 	iter->next_index = iter->index;
343 	iter->tags = 0;
344 	return NULL;
345 }
346 
347 static inline unsigned long
__radix_tree_iter_add(struct radix_tree_iter * iter,unsigned long slots)348 __radix_tree_iter_add(struct radix_tree_iter *iter, unsigned long slots)
349 {
350 	return iter->index + slots;
351 }
352 
353 /**
354  * radix_tree_iter_resume - resume iterating when the chunk may be invalid
355  * @slot: pointer to current slot
356  * @iter: iterator state
357  * Returns: New slot pointer
358  *
359  * If the iterator needs to release then reacquire a lock, the chunk may
360  * have been invalidated by an insertion or deletion.  Call this function
361  * before releasing the lock to continue the iteration from the next index.
362  */
363 void __rcu **__must_check radix_tree_iter_resume(void __rcu **slot,
364 					struct radix_tree_iter *iter);
365 
366 /**
367  * radix_tree_chunk_size - get current chunk size
368  *
369  * @iter:	pointer to radix tree iterator
370  * Returns:	current chunk size
371  */
372 static __always_inline long
radix_tree_chunk_size(struct radix_tree_iter * iter)373 radix_tree_chunk_size(struct radix_tree_iter *iter)
374 {
375 	return iter->next_index - iter->index;
376 }
377 
378 /**
379  * radix_tree_next_slot - find next slot in chunk
380  *
381  * @slot:	pointer to current slot
382  * @iter:	pointer to iterator state
383  * @flags:	RADIX_TREE_ITER_*, should be constant
384  * Returns:	pointer to next slot, or NULL if there no more left
385  *
386  * This function updates @iter->index in the case of a successful lookup.
387  * For tagged lookup it also eats @iter->tags.
388  *
389  * There are several cases where 'slot' can be passed in as NULL to this
390  * function.  These cases result from the use of radix_tree_iter_resume() or
391  * radix_tree_iter_retry().  In these cases we don't end up dereferencing
392  * 'slot' because either:
393  * a) we are doing tagged iteration and iter->tags has been set to 0, or
394  * b) we are doing non-tagged iteration, and iter->index and iter->next_index
395  *    have been set up so that radix_tree_chunk_size() returns 1 or 0.
396  */
radix_tree_next_slot(void __rcu ** slot,struct radix_tree_iter * iter,unsigned flags)397 static __always_inline void __rcu **radix_tree_next_slot(void __rcu **slot,
398 				struct radix_tree_iter *iter, unsigned flags)
399 {
400 	if (flags & RADIX_TREE_ITER_TAGGED) {
401 		iter->tags >>= 1;
402 		if (unlikely(!iter->tags))
403 			return NULL;
404 		if (likely(iter->tags & 1ul)) {
405 			iter->index = __radix_tree_iter_add(iter, 1);
406 			slot++;
407 			goto found;
408 		}
409 		if (!(flags & RADIX_TREE_ITER_CONTIG)) {
410 			unsigned offset = __ffs(iter->tags);
411 
412 			iter->tags >>= offset++;
413 			iter->index = __radix_tree_iter_add(iter, offset);
414 			slot += offset;
415 			goto found;
416 		}
417 	} else {
418 		long count = radix_tree_chunk_size(iter);
419 
420 		while (--count > 0) {
421 			slot++;
422 			iter->index = __radix_tree_iter_add(iter, 1);
423 
424 			if (likely(*slot))
425 				goto found;
426 			if (flags & RADIX_TREE_ITER_CONTIG) {
427 				/* forbid switching to the next chunk */
428 				iter->next_index = 0;
429 				break;
430 			}
431 		}
432 	}
433 	return NULL;
434 
435  found:
436 	return slot;
437 }
438 
439 /**
440  * radix_tree_for_each_slot - iterate over non-empty slots
441  *
442  * @slot:	the void** variable for pointer to slot
443  * @root:	the struct radix_tree_root pointer
444  * @iter:	the struct radix_tree_iter pointer
445  * @start:	iteration starting index
446  *
447  * @slot points to radix tree slot, @iter->index contains its index.
448  */
449 #define radix_tree_for_each_slot(slot, root, iter, start)		\
450 	for (slot = radix_tree_iter_init(iter, start) ;			\
451 	     slot || (slot = radix_tree_next_chunk(root, iter, 0)) ;	\
452 	     slot = radix_tree_next_slot(slot, iter, 0))
453 
454 /**
455  * radix_tree_for_each_tagged - iterate over tagged slots
456  *
457  * @slot:	the void** variable for pointer to slot
458  * @root:	the struct radix_tree_root pointer
459  * @iter:	the struct radix_tree_iter pointer
460  * @start:	iteration starting index
461  * @tag:	tag index
462  *
463  * @slot points to radix tree slot, @iter->index contains its index.
464  */
465 #define radix_tree_for_each_tagged(slot, root, iter, start, tag)	\
466 	for (slot = radix_tree_iter_init(iter, start) ;			\
467 	     slot || (slot = radix_tree_next_chunk(root, iter,		\
468 			      RADIX_TREE_ITER_TAGGED | tag)) ;		\
469 	     slot = radix_tree_next_slot(slot, iter,			\
470 				RADIX_TREE_ITER_TAGGED | tag))
471 
472 #endif /* _LINUX_RADIX_TREE_H */
473