Loading...
1/*
2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
4 * Copyright (C) 2005 SGI, Christoph Lameter
5 * Copyright (C) 2006 Nick Piggin
6 * Copyright (C) 2012 Konstantin Khlebnikov
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as
10 * published by the Free Software Foundation; either version 2, or (at
11 * your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 */
22
23#include <linux/errno.h>
24#include <linux/init.h>
25#include <linux/kernel.h>
26#include <linux/export.h>
27#include <linux/radix-tree.h>
28#include <linux/percpu.h>
29#include <linux/slab.h>
30#include <linux/notifier.h>
31#include <linux/cpu.h>
32#include <linux/string.h>
33#include <linux/bitops.h>
34#include <linux/rcupdate.h>
35#include <linux/hardirq.h> /* in_interrupt() */
36
37
38/*
39 * The height_to_maxindex array needs to be one deeper than the maximum
40 * path as height 0 holds only 1 entry.
41 */
42static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH + 1] __read_mostly;
43
44/*
45 * Radix tree node cache.
46 */
47static struct kmem_cache *radix_tree_node_cachep;
48
49/*
50 * The radix tree is variable-height, so an insert operation not only has
51 * to build the branch to its corresponding item, it also has to build the
52 * branch to existing items if the size has to be increased (by
53 * radix_tree_extend).
54 *
55 * The worst case is a zero height tree with just a single item at index 0,
56 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
57 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
58 * Hence:
59 */
60#define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
61
62/*
63 * Per-cpu pool of preloaded nodes
64 */
65struct radix_tree_preload {
66 int nr;
67 struct radix_tree_node *nodes[RADIX_TREE_PRELOAD_SIZE];
68};
69static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
70
71static inline void *ptr_to_indirect(void *ptr)
72{
73 return (void *)((unsigned long)ptr | RADIX_TREE_INDIRECT_PTR);
74}
75
76static inline void *indirect_to_ptr(void *ptr)
77{
78 return (void *)((unsigned long)ptr & ~RADIX_TREE_INDIRECT_PTR);
79}
80
81static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
82{
83 return root->gfp_mask & __GFP_BITS_MASK;
84}
85
86static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
87 int offset)
88{
89 __set_bit(offset, node->tags[tag]);
90}
91
92static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
93 int offset)
94{
95 __clear_bit(offset, node->tags[tag]);
96}
97
98static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
99 int offset)
100{
101 return test_bit(offset, node->tags[tag]);
102}
103
104static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
105{
106 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
107}
108
109static inline void root_tag_clear(struct radix_tree_root *root, unsigned int tag)
110{
111 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
112}
113
114static inline void root_tag_clear_all(struct radix_tree_root *root)
115{
116 root->gfp_mask &= __GFP_BITS_MASK;
117}
118
119static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
120{
121 return (__force unsigned)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
122}
123
124/*
125 * Returns 1 if any slot in the node has this tag set.
126 * Otherwise returns 0.
127 */
128static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
129{
130 int idx;
131 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
132 if (node->tags[tag][idx])
133 return 1;
134 }
135 return 0;
136}
137
138/**
139 * radix_tree_find_next_bit - find the next set bit in a memory region
140 *
141 * @addr: The address to base the search on
142 * @size: The bitmap size in bits
143 * @offset: The bitnumber to start searching at
144 *
145 * Unrollable variant of find_next_bit() for constant size arrays.
146 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
147 * Returns next bit offset, or size if nothing found.
148 */
149static __always_inline unsigned long
150radix_tree_find_next_bit(const unsigned long *addr,
151 unsigned long size, unsigned long offset)
152{
153 if (!__builtin_constant_p(size))
154 return find_next_bit(addr, size, offset);
155
156 if (offset < size) {
157 unsigned long tmp;
158
159 addr += offset / BITS_PER_LONG;
160 tmp = *addr >> (offset % BITS_PER_LONG);
161 if (tmp)
162 return __ffs(tmp) + offset;
163 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
164 while (offset < size) {
165 tmp = *++addr;
166 if (tmp)
167 return __ffs(tmp) + offset;
168 offset += BITS_PER_LONG;
169 }
170 }
171 return size;
172}
173
174/*
175 * This assumes that the caller has performed appropriate preallocation, and
176 * that the caller has pinned this thread of control to the current CPU.
177 */
178static struct radix_tree_node *
179radix_tree_node_alloc(struct radix_tree_root *root)
180{
181 struct radix_tree_node *ret = NULL;
182 gfp_t gfp_mask = root_gfp_mask(root);
183
184 /*
185 * Preload code isn't irq safe and it doesn't make sence to use
186 * preloading in the interrupt anyway as all the allocations have to
187 * be atomic. So just do normal allocation when in interrupt.
188 */
189 if (!(gfp_mask & __GFP_WAIT) && !in_interrupt()) {
190 struct radix_tree_preload *rtp;
191
192 /*
193 * Provided the caller has preloaded here, we will always
194 * succeed in getting a node here (and never reach
195 * kmem_cache_alloc)
196 */
197 rtp = &__get_cpu_var(radix_tree_preloads);
198 if (rtp->nr) {
199 ret = rtp->nodes[rtp->nr - 1];
200 rtp->nodes[rtp->nr - 1] = NULL;
201 rtp->nr--;
202 }
203 }
204 if (ret == NULL)
205 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
206
207 BUG_ON(radix_tree_is_indirect_ptr(ret));
208 return ret;
209}
210
211static void radix_tree_node_rcu_free(struct rcu_head *head)
212{
213 struct radix_tree_node *node =
214 container_of(head, struct radix_tree_node, rcu_head);
215 int i;
216
217 /*
218 * must only free zeroed nodes into the slab. radix_tree_shrink
219 * can leave us with a non-NULL entry in the first slot, so clear
220 * that here to make sure.
221 */
222 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
223 tag_clear(node, i, 0);
224
225 node->slots[0] = NULL;
226 node->count = 0;
227
228 kmem_cache_free(radix_tree_node_cachep, node);
229}
230
231static inline void
232radix_tree_node_free(struct radix_tree_node *node)
233{
234 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
235}
236
237/*
238 * Load up this CPU's radix_tree_node buffer with sufficient objects to
239 * ensure that the addition of a single element in the tree cannot fail. On
240 * success, return zero, with preemption disabled. On error, return -ENOMEM
241 * with preemption not disabled.
242 *
243 * To make use of this facility, the radix tree must be initialised without
244 * __GFP_WAIT being passed to INIT_RADIX_TREE().
245 */
246static int __radix_tree_preload(gfp_t gfp_mask)
247{
248 struct radix_tree_preload *rtp;
249 struct radix_tree_node *node;
250 int ret = -ENOMEM;
251
252 preempt_disable();
253 rtp = &__get_cpu_var(radix_tree_preloads);
254 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
255 preempt_enable();
256 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
257 if (node == NULL)
258 goto out;
259 preempt_disable();
260 rtp = &__get_cpu_var(radix_tree_preloads);
261 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
262 rtp->nodes[rtp->nr++] = node;
263 else
264 kmem_cache_free(radix_tree_node_cachep, node);
265 }
266 ret = 0;
267out:
268 return ret;
269}
270
271/*
272 * Load up this CPU's radix_tree_node buffer with sufficient objects to
273 * ensure that the addition of a single element in the tree cannot fail. On
274 * success, return zero, with preemption disabled. On error, return -ENOMEM
275 * with preemption not disabled.
276 *
277 * To make use of this facility, the radix tree must be initialised without
278 * __GFP_WAIT being passed to INIT_RADIX_TREE().
279 */
280int radix_tree_preload(gfp_t gfp_mask)
281{
282 /* Warn on non-sensical use... */
283 WARN_ON_ONCE(!(gfp_mask & __GFP_WAIT));
284 return __radix_tree_preload(gfp_mask);
285}
286EXPORT_SYMBOL(radix_tree_preload);
287
288/*
289 * The same as above function, except we don't guarantee preloading happens.
290 * We do it, if we decide it helps. On success, return zero with preemption
291 * disabled. On error, return -ENOMEM with preemption not disabled.
292 */
293int radix_tree_maybe_preload(gfp_t gfp_mask)
294{
295 if (gfp_mask & __GFP_WAIT)
296 return __radix_tree_preload(gfp_mask);
297 /* Preloading doesn't help anything with this gfp mask, skip it */
298 preempt_disable();
299 return 0;
300}
301EXPORT_SYMBOL(radix_tree_maybe_preload);
302
303/*
304 * Return the maximum key which can be store into a
305 * radix tree with height HEIGHT.
306 */
307static inline unsigned long radix_tree_maxindex(unsigned int height)
308{
309 return height_to_maxindex[height];
310}
311
312/*
313 * Extend a radix tree so it can store key @index.
314 */
315static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
316{
317 struct radix_tree_node *node;
318 struct radix_tree_node *slot;
319 unsigned int height;
320 int tag;
321
322 /* Figure out what the height should be. */
323 height = root->height + 1;
324 while (index > radix_tree_maxindex(height))
325 height++;
326
327 if (root->rnode == NULL) {
328 root->height = height;
329 goto out;
330 }
331
332 do {
333 unsigned int newheight;
334 if (!(node = radix_tree_node_alloc(root)))
335 return -ENOMEM;
336
337 /* Propagate the aggregated tag info into the new root */
338 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
339 if (root_tag_get(root, tag))
340 tag_set(node, tag, 0);
341 }
342
343 /* Increase the height. */
344 newheight = root->height+1;
345 BUG_ON(newheight & ~RADIX_TREE_HEIGHT_MASK);
346 node->path = newheight;
347 node->count = 1;
348 node->parent = NULL;
349 slot = root->rnode;
350 if (newheight > 1) {
351 slot = indirect_to_ptr(slot);
352 slot->parent = node;
353 }
354 node->slots[0] = slot;
355 node = ptr_to_indirect(node);
356 rcu_assign_pointer(root->rnode, node);
357 root->height = newheight;
358 } while (height > root->height);
359out:
360 return 0;
361}
362
363/**
364 * __radix_tree_create - create a slot in a radix tree
365 * @root: radix tree root
366 * @index: index key
367 * @nodep: returns node
368 * @slotp: returns slot
369 *
370 * Create, if necessary, and return the node and slot for an item
371 * at position @index in the radix tree @root.
372 *
373 * Until there is more than one item in the tree, no nodes are
374 * allocated and @root->rnode is used as a direct slot instead of
375 * pointing to a node, in which case *@nodep will be NULL.
376 *
377 * Returns -ENOMEM, or 0 for success.
378 */
379int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
380 struct radix_tree_node **nodep, void ***slotp)
381{
382 struct radix_tree_node *node = NULL, *slot;
383 unsigned int height, shift, offset;
384 int error;
385
386 /* Make sure the tree is high enough. */
387 if (index > radix_tree_maxindex(root->height)) {
388 error = radix_tree_extend(root, index);
389 if (error)
390 return error;
391 }
392
393 slot = indirect_to_ptr(root->rnode);
394
395 height = root->height;
396 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
397
398 offset = 0; /* uninitialised var warning */
399 while (height > 0) {
400 if (slot == NULL) {
401 /* Have to add a child node. */
402 if (!(slot = radix_tree_node_alloc(root)))
403 return -ENOMEM;
404 slot->path = height;
405 slot->parent = node;
406 if (node) {
407 rcu_assign_pointer(node->slots[offset], slot);
408 node->count++;
409 slot->path |= offset << RADIX_TREE_HEIGHT_SHIFT;
410 } else
411 rcu_assign_pointer(root->rnode, ptr_to_indirect(slot));
412 }
413
414 /* Go a level down */
415 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
416 node = slot;
417 slot = node->slots[offset];
418 shift -= RADIX_TREE_MAP_SHIFT;
419 height--;
420 }
421
422 if (nodep)
423 *nodep = node;
424 if (slotp)
425 *slotp = node ? node->slots + offset : (void **)&root->rnode;
426 return 0;
427}
428
429/**
430 * radix_tree_insert - insert into a radix tree
431 * @root: radix tree root
432 * @index: index key
433 * @item: item to insert
434 *
435 * Insert an item into the radix tree at position @index.
436 */
437int radix_tree_insert(struct radix_tree_root *root,
438 unsigned long index, void *item)
439{
440 struct radix_tree_node *node;
441 void **slot;
442 int error;
443
444 BUG_ON(radix_tree_is_indirect_ptr(item));
445
446 error = __radix_tree_create(root, index, &node, &slot);
447 if (error)
448 return error;
449 if (*slot != NULL)
450 return -EEXIST;
451 rcu_assign_pointer(*slot, item);
452
453 if (node) {
454 node->count++;
455 BUG_ON(tag_get(node, 0, index & RADIX_TREE_MAP_MASK));
456 BUG_ON(tag_get(node, 1, index & RADIX_TREE_MAP_MASK));
457 } else {
458 BUG_ON(root_tag_get(root, 0));
459 BUG_ON(root_tag_get(root, 1));
460 }
461
462 return 0;
463}
464EXPORT_SYMBOL(radix_tree_insert);
465
466/**
467 * __radix_tree_lookup - lookup an item in a radix tree
468 * @root: radix tree root
469 * @index: index key
470 * @nodep: returns node
471 * @slotp: returns slot
472 *
473 * Lookup and return the item at position @index in the radix
474 * tree @root.
475 *
476 * Until there is more than one item in the tree, no nodes are
477 * allocated and @root->rnode is used as a direct slot instead of
478 * pointing to a node, in which case *@nodep will be NULL.
479 */
480void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
481 struct radix_tree_node **nodep, void ***slotp)
482{
483 struct radix_tree_node *node, *parent;
484 unsigned int height, shift;
485 void **slot;
486
487 node = rcu_dereference_raw(root->rnode);
488 if (node == NULL)
489 return NULL;
490
491 if (!radix_tree_is_indirect_ptr(node)) {
492 if (index > 0)
493 return NULL;
494
495 if (nodep)
496 *nodep = NULL;
497 if (slotp)
498 *slotp = (void **)&root->rnode;
499 return node;
500 }
501 node = indirect_to_ptr(node);
502
503 height = node->path & RADIX_TREE_HEIGHT_MASK;
504 if (index > radix_tree_maxindex(height))
505 return NULL;
506
507 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
508
509 do {
510 parent = node;
511 slot = node->slots + ((index >> shift) & RADIX_TREE_MAP_MASK);
512 node = rcu_dereference_raw(*slot);
513 if (node == NULL)
514 return NULL;
515
516 shift -= RADIX_TREE_MAP_SHIFT;
517 height--;
518 } while (height > 0);
519
520 if (nodep)
521 *nodep = parent;
522 if (slotp)
523 *slotp = slot;
524 return node;
525}
526
527/**
528 * radix_tree_lookup_slot - lookup a slot in a radix tree
529 * @root: radix tree root
530 * @index: index key
531 *
532 * Returns: the slot corresponding to the position @index in the
533 * radix tree @root. This is useful for update-if-exists operations.
534 *
535 * This function can be called under rcu_read_lock iff the slot is not
536 * modified by radix_tree_replace_slot, otherwise it must be called
537 * exclusive from other writers. Any dereference of the slot must be done
538 * using radix_tree_deref_slot.
539 */
540void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
541{
542 void **slot;
543
544 if (!__radix_tree_lookup(root, index, NULL, &slot))
545 return NULL;
546 return slot;
547}
548EXPORT_SYMBOL(radix_tree_lookup_slot);
549
550/**
551 * radix_tree_lookup - perform lookup operation on a radix tree
552 * @root: radix tree root
553 * @index: index key
554 *
555 * Lookup the item at the position @index in the radix tree @root.
556 *
557 * This function can be called under rcu_read_lock, however the caller
558 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
559 * them safely). No RCU barriers are required to access or modify the
560 * returned item, however.
561 */
562void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
563{
564 return __radix_tree_lookup(root, index, NULL, NULL);
565}
566EXPORT_SYMBOL(radix_tree_lookup);
567
568/**
569 * radix_tree_tag_set - set a tag on a radix tree node
570 * @root: radix tree root
571 * @index: index key
572 * @tag: tag index
573 *
574 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
575 * corresponding to @index in the radix tree. From
576 * the root all the way down to the leaf node.
577 *
578 * Returns the address of the tagged item. Setting a tag on a not-present
579 * item is a bug.
580 */
581void *radix_tree_tag_set(struct radix_tree_root *root,
582 unsigned long index, unsigned int tag)
583{
584 unsigned int height, shift;
585 struct radix_tree_node *slot;
586
587 height = root->height;
588 BUG_ON(index > radix_tree_maxindex(height));
589
590 slot = indirect_to_ptr(root->rnode);
591 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
592
593 while (height > 0) {
594 int offset;
595
596 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
597 if (!tag_get(slot, tag, offset))
598 tag_set(slot, tag, offset);
599 slot = slot->slots[offset];
600 BUG_ON(slot == NULL);
601 shift -= RADIX_TREE_MAP_SHIFT;
602 height--;
603 }
604
605 /* set the root's tag bit */
606 if (slot && !root_tag_get(root, tag))
607 root_tag_set(root, tag);
608
609 return slot;
610}
611EXPORT_SYMBOL(radix_tree_tag_set);
612
613/**
614 * radix_tree_tag_clear - clear a tag on a radix tree node
615 * @root: radix tree root
616 * @index: index key
617 * @tag: tag index
618 *
619 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
620 * corresponding to @index in the radix tree. If
621 * this causes the leaf node to have no tags set then clear the tag in the
622 * next-to-leaf node, etc.
623 *
624 * Returns the address of the tagged item on success, else NULL. ie:
625 * has the same return value and semantics as radix_tree_lookup().
626 */
627void *radix_tree_tag_clear(struct radix_tree_root *root,
628 unsigned long index, unsigned int tag)
629{
630 struct radix_tree_node *node = NULL;
631 struct radix_tree_node *slot = NULL;
632 unsigned int height, shift;
633 int uninitialized_var(offset);
634
635 height = root->height;
636 if (index > radix_tree_maxindex(height))
637 goto out;
638
639 shift = height * RADIX_TREE_MAP_SHIFT;
640 slot = indirect_to_ptr(root->rnode);
641
642 while (shift) {
643 if (slot == NULL)
644 goto out;
645
646 shift -= RADIX_TREE_MAP_SHIFT;
647 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
648 node = slot;
649 slot = slot->slots[offset];
650 }
651
652 if (slot == NULL)
653 goto out;
654
655 while (node) {
656 if (!tag_get(node, tag, offset))
657 goto out;
658 tag_clear(node, tag, offset);
659 if (any_tag_set(node, tag))
660 goto out;
661
662 index >>= RADIX_TREE_MAP_SHIFT;
663 offset = index & RADIX_TREE_MAP_MASK;
664 node = node->parent;
665 }
666
667 /* clear the root's tag bit */
668 if (root_tag_get(root, tag))
669 root_tag_clear(root, tag);
670
671out:
672 return slot;
673}
674EXPORT_SYMBOL(radix_tree_tag_clear);
675
676/**
677 * radix_tree_tag_get - get a tag on a radix tree node
678 * @root: radix tree root
679 * @index: index key
680 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
681 *
682 * Return values:
683 *
684 * 0: tag not present or not set
685 * 1: tag set
686 *
687 * Note that the return value of this function may not be relied on, even if
688 * the RCU lock is held, unless tag modification and node deletion are excluded
689 * from concurrency.
690 */
691int radix_tree_tag_get(struct radix_tree_root *root,
692 unsigned long index, unsigned int tag)
693{
694 unsigned int height, shift;
695 struct radix_tree_node *node;
696
697 /* check the root's tag bit */
698 if (!root_tag_get(root, tag))
699 return 0;
700
701 node = rcu_dereference_raw(root->rnode);
702 if (node == NULL)
703 return 0;
704
705 if (!radix_tree_is_indirect_ptr(node))
706 return (index == 0);
707 node = indirect_to_ptr(node);
708
709 height = node->path & RADIX_TREE_HEIGHT_MASK;
710 if (index > radix_tree_maxindex(height))
711 return 0;
712
713 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
714
715 for ( ; ; ) {
716 int offset;
717
718 if (node == NULL)
719 return 0;
720
721 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
722 if (!tag_get(node, tag, offset))
723 return 0;
724 if (height == 1)
725 return 1;
726 node = rcu_dereference_raw(node->slots[offset]);
727 shift -= RADIX_TREE_MAP_SHIFT;
728 height--;
729 }
730}
731EXPORT_SYMBOL(radix_tree_tag_get);
732
733/**
734 * radix_tree_next_chunk - find next chunk of slots for iteration
735 *
736 * @root: radix tree root
737 * @iter: iterator state
738 * @flags: RADIX_TREE_ITER_* flags and tag index
739 * Returns: pointer to chunk first slot, or NULL if iteration is over
740 */
741void **radix_tree_next_chunk(struct radix_tree_root *root,
742 struct radix_tree_iter *iter, unsigned flags)
743{
744 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
745 struct radix_tree_node *rnode, *node;
746 unsigned long index, offset, height;
747
748 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
749 return NULL;
750
751 /*
752 * Catch next_index overflow after ~0UL. iter->index never overflows
753 * during iterating; it can be zero only at the beginning.
754 * And we cannot overflow iter->next_index in a single step,
755 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
756 *
757 * This condition also used by radix_tree_next_slot() to stop
758 * contiguous iterating, and forbid swithing to the next chunk.
759 */
760 index = iter->next_index;
761 if (!index && iter->index)
762 return NULL;
763
764 rnode = rcu_dereference_raw(root->rnode);
765 if (radix_tree_is_indirect_ptr(rnode)) {
766 rnode = indirect_to_ptr(rnode);
767 } else if (rnode && !index) {
768 /* Single-slot tree */
769 iter->index = 0;
770 iter->next_index = 1;
771 iter->tags = 1;
772 return (void **)&root->rnode;
773 } else
774 return NULL;
775
776restart:
777 height = rnode->path & RADIX_TREE_HEIGHT_MASK;
778 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
779 offset = index >> shift;
780
781 /* Index outside of the tree */
782 if (offset >= RADIX_TREE_MAP_SIZE)
783 return NULL;
784
785 node = rnode;
786 while (1) {
787 if ((flags & RADIX_TREE_ITER_TAGGED) ?
788 !test_bit(offset, node->tags[tag]) :
789 !node->slots[offset]) {
790 /* Hole detected */
791 if (flags & RADIX_TREE_ITER_CONTIG)
792 return NULL;
793
794 if (flags & RADIX_TREE_ITER_TAGGED)
795 offset = radix_tree_find_next_bit(
796 node->tags[tag],
797 RADIX_TREE_MAP_SIZE,
798 offset + 1);
799 else
800 while (++offset < RADIX_TREE_MAP_SIZE) {
801 if (node->slots[offset])
802 break;
803 }
804 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
805 index += offset << shift;
806 /* Overflow after ~0UL */
807 if (!index)
808 return NULL;
809 if (offset == RADIX_TREE_MAP_SIZE)
810 goto restart;
811 }
812
813 /* This is leaf-node */
814 if (!shift)
815 break;
816
817 node = rcu_dereference_raw(node->slots[offset]);
818 if (node == NULL)
819 goto restart;
820 shift -= RADIX_TREE_MAP_SHIFT;
821 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
822 }
823
824 /* Update the iterator state */
825 iter->index = index;
826 iter->next_index = (index | RADIX_TREE_MAP_MASK) + 1;
827
828 /* Construct iter->tags bit-mask from node->tags[tag] array */
829 if (flags & RADIX_TREE_ITER_TAGGED) {
830 unsigned tag_long, tag_bit;
831
832 tag_long = offset / BITS_PER_LONG;
833 tag_bit = offset % BITS_PER_LONG;
834 iter->tags = node->tags[tag][tag_long] >> tag_bit;
835 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
836 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
837 /* Pick tags from next element */
838 if (tag_bit)
839 iter->tags |= node->tags[tag][tag_long + 1] <<
840 (BITS_PER_LONG - tag_bit);
841 /* Clip chunk size, here only BITS_PER_LONG tags */
842 iter->next_index = index + BITS_PER_LONG;
843 }
844 }
845
846 return node->slots + offset;
847}
848EXPORT_SYMBOL(radix_tree_next_chunk);
849
850/**
851 * radix_tree_range_tag_if_tagged - for each item in given range set given
852 * tag if item has another tag set
853 * @root: radix tree root
854 * @first_indexp: pointer to a starting index of a range to scan
855 * @last_index: last index of a range to scan
856 * @nr_to_tag: maximum number items to tag
857 * @iftag: tag index to test
858 * @settag: tag index to set if tested tag is set
859 *
860 * This function scans range of radix tree from first_index to last_index
861 * (inclusive). For each item in the range if iftag is set, the function sets
862 * also settag. The function stops either after tagging nr_to_tag items or
863 * after reaching last_index.
864 *
865 * The tags must be set from the leaf level only and propagated back up the
866 * path to the root. We must do this so that we resolve the full path before
867 * setting any tags on intermediate nodes. If we set tags as we descend, then
868 * we can get to the leaf node and find that the index that has the iftag
869 * set is outside the range we are scanning. This reults in dangling tags and
870 * can lead to problems with later tag operations (e.g. livelocks on lookups).
871 *
872 * The function returns number of leaves where the tag was set and sets
873 * *first_indexp to the first unscanned index.
874 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
875 * be prepared to handle that.
876 */
877unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
878 unsigned long *first_indexp, unsigned long last_index,
879 unsigned long nr_to_tag,
880 unsigned int iftag, unsigned int settag)
881{
882 unsigned int height = root->height;
883 struct radix_tree_node *node = NULL;
884 struct radix_tree_node *slot;
885 unsigned int shift;
886 unsigned long tagged = 0;
887 unsigned long index = *first_indexp;
888
889 last_index = min(last_index, radix_tree_maxindex(height));
890 if (index > last_index)
891 return 0;
892 if (!nr_to_tag)
893 return 0;
894 if (!root_tag_get(root, iftag)) {
895 *first_indexp = last_index + 1;
896 return 0;
897 }
898 if (height == 0) {
899 *first_indexp = last_index + 1;
900 root_tag_set(root, settag);
901 return 1;
902 }
903
904 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
905 slot = indirect_to_ptr(root->rnode);
906
907 for (;;) {
908 unsigned long upindex;
909 int offset;
910
911 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
912 if (!slot->slots[offset])
913 goto next;
914 if (!tag_get(slot, iftag, offset))
915 goto next;
916 if (shift) {
917 /* Go down one level */
918 shift -= RADIX_TREE_MAP_SHIFT;
919 node = slot;
920 slot = slot->slots[offset];
921 continue;
922 }
923
924 /* tag the leaf */
925 tagged++;
926 tag_set(slot, settag, offset);
927
928 /* walk back up the path tagging interior nodes */
929 upindex = index;
930 while (node) {
931 upindex >>= RADIX_TREE_MAP_SHIFT;
932 offset = upindex & RADIX_TREE_MAP_MASK;
933
934 /* stop if we find a node with the tag already set */
935 if (tag_get(node, settag, offset))
936 break;
937 tag_set(node, settag, offset);
938 node = node->parent;
939 }
940
941 /*
942 * Small optimization: now clear that node pointer.
943 * Since all of this slot's ancestors now have the tag set
944 * from setting it above, we have no further need to walk
945 * back up the tree setting tags, until we update slot to
946 * point to another radix_tree_node.
947 */
948 node = NULL;
949
950next:
951 /* Go to next item at level determined by 'shift' */
952 index = ((index >> shift) + 1) << shift;
953 /* Overflow can happen when last_index is ~0UL... */
954 if (index > last_index || !index)
955 break;
956 if (tagged >= nr_to_tag)
957 break;
958 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
959 /*
960 * We've fully scanned this node. Go up. Because
961 * last_index is guaranteed to be in the tree, what
962 * we do below cannot wander astray.
963 */
964 slot = slot->parent;
965 shift += RADIX_TREE_MAP_SHIFT;
966 }
967 }
968 /*
969 * We need not to tag the root tag if there is no tag which is set with
970 * settag within the range from *first_indexp to last_index.
971 */
972 if (tagged > 0)
973 root_tag_set(root, settag);
974 *first_indexp = index;
975
976 return tagged;
977}
978EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
979
980/**
981 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
982 * @root: radix tree root
983 * @results: where the results of the lookup are placed
984 * @first_index: start the lookup from this key
985 * @max_items: place up to this many items at *results
986 *
987 * Performs an index-ascending scan of the tree for present items. Places
988 * them at *@results and returns the number of items which were placed at
989 * *@results.
990 *
991 * The implementation is naive.
992 *
993 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
994 * rcu_read_lock. In this case, rather than the returned results being
995 * an atomic snapshot of the tree at a single point in time, the semantics
996 * of an RCU protected gang lookup are as though multiple radix_tree_lookups
997 * have been issued in individual locks, and results stored in 'results'.
998 */
999unsigned int
1000radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1001 unsigned long first_index, unsigned int max_items)
1002{
1003 struct radix_tree_iter iter;
1004 void **slot;
1005 unsigned int ret = 0;
1006
1007 if (unlikely(!max_items))
1008 return 0;
1009
1010 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1011 results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot));
1012 if (!results[ret])
1013 continue;
1014 if (++ret == max_items)
1015 break;
1016 }
1017
1018 return ret;
1019}
1020EXPORT_SYMBOL(radix_tree_gang_lookup);
1021
1022/**
1023 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1024 * @root: radix tree root
1025 * @results: where the results of the lookup are placed
1026 * @indices: where their indices should be placed (but usually NULL)
1027 * @first_index: start the lookup from this key
1028 * @max_items: place up to this many items at *results
1029 *
1030 * Performs an index-ascending scan of the tree for present items. Places
1031 * their slots at *@results and returns the number of items which were
1032 * placed at *@results.
1033 *
1034 * The implementation is naive.
1035 *
1036 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1037 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1038 * protection, radix_tree_deref_slot may fail requiring a retry.
1039 */
1040unsigned int
1041radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1042 void ***results, unsigned long *indices,
1043 unsigned long first_index, unsigned int max_items)
1044{
1045 struct radix_tree_iter iter;
1046 void **slot;
1047 unsigned int ret = 0;
1048
1049 if (unlikely(!max_items))
1050 return 0;
1051
1052 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1053 results[ret] = slot;
1054 if (indices)
1055 indices[ret] = iter.index;
1056 if (++ret == max_items)
1057 break;
1058 }
1059
1060 return ret;
1061}
1062EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1063
1064/**
1065 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1066 * based on a tag
1067 * @root: radix tree root
1068 * @results: where the results of the lookup are placed
1069 * @first_index: start the lookup from this key
1070 * @max_items: place up to this many items at *results
1071 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1072 *
1073 * Performs an index-ascending scan of the tree for present items which
1074 * have the tag indexed by @tag set. Places the items at *@results and
1075 * returns the number of items which were placed at *@results.
1076 */
1077unsigned int
1078radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1079 unsigned long first_index, unsigned int max_items,
1080 unsigned int tag)
1081{
1082 struct radix_tree_iter iter;
1083 void **slot;
1084 unsigned int ret = 0;
1085
1086 if (unlikely(!max_items))
1087 return 0;
1088
1089 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1090 results[ret] = indirect_to_ptr(rcu_dereference_raw(*slot));
1091 if (!results[ret])
1092 continue;
1093 if (++ret == max_items)
1094 break;
1095 }
1096
1097 return ret;
1098}
1099EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1100
1101/**
1102 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1103 * radix tree based on a tag
1104 * @root: radix tree root
1105 * @results: where the results of the lookup are placed
1106 * @first_index: start the lookup from this key
1107 * @max_items: place up to this many items at *results
1108 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1109 *
1110 * Performs an index-ascending scan of the tree for present items which
1111 * have the tag indexed by @tag set. Places the slots at *@results and
1112 * returns the number of slots which were placed at *@results.
1113 */
1114unsigned int
1115radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1116 unsigned long first_index, unsigned int max_items,
1117 unsigned int tag)
1118{
1119 struct radix_tree_iter iter;
1120 void **slot;
1121 unsigned int ret = 0;
1122
1123 if (unlikely(!max_items))
1124 return 0;
1125
1126 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1127 results[ret] = slot;
1128 if (++ret == max_items)
1129 break;
1130 }
1131
1132 return ret;
1133}
1134EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1135
1136#if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1137#include <linux/sched.h> /* for cond_resched() */
1138
1139/*
1140 * This linear search is at present only useful to shmem_unuse_inode().
1141 */
1142static unsigned long __locate(struct radix_tree_node *slot, void *item,
1143 unsigned long index, unsigned long *found_index)
1144{
1145 unsigned int shift, height;
1146 unsigned long i;
1147
1148 height = slot->path & RADIX_TREE_HEIGHT_MASK;
1149 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
1150
1151 for ( ; height > 1; height--) {
1152 i = (index >> shift) & RADIX_TREE_MAP_MASK;
1153 for (;;) {
1154 if (slot->slots[i] != NULL)
1155 break;
1156 index &= ~((1UL << shift) - 1);
1157 index += 1UL << shift;
1158 if (index == 0)
1159 goto out; /* 32-bit wraparound */
1160 i++;
1161 if (i == RADIX_TREE_MAP_SIZE)
1162 goto out;
1163 }
1164
1165 shift -= RADIX_TREE_MAP_SHIFT;
1166 slot = rcu_dereference_raw(slot->slots[i]);
1167 if (slot == NULL)
1168 goto out;
1169 }
1170
1171 /* Bottom level: check items */
1172 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
1173 if (slot->slots[i] == item) {
1174 *found_index = index + i;
1175 index = 0;
1176 goto out;
1177 }
1178 }
1179 index += RADIX_TREE_MAP_SIZE;
1180out:
1181 return index;
1182}
1183
1184/**
1185 * radix_tree_locate_item - search through radix tree for item
1186 * @root: radix tree root
1187 * @item: item to be found
1188 *
1189 * Returns index where item was found, or -1 if not found.
1190 * Caller must hold no lock (since this time-consuming function needs
1191 * to be preemptible), and must check afterwards if item is still there.
1192 */
1193unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1194{
1195 struct radix_tree_node *node;
1196 unsigned long max_index;
1197 unsigned long cur_index = 0;
1198 unsigned long found_index = -1;
1199
1200 do {
1201 rcu_read_lock();
1202 node = rcu_dereference_raw(root->rnode);
1203 if (!radix_tree_is_indirect_ptr(node)) {
1204 rcu_read_unlock();
1205 if (node == item)
1206 found_index = 0;
1207 break;
1208 }
1209
1210 node = indirect_to_ptr(node);
1211 max_index = radix_tree_maxindex(node->path &
1212 RADIX_TREE_HEIGHT_MASK);
1213 if (cur_index > max_index) {
1214 rcu_read_unlock();
1215 break;
1216 }
1217
1218 cur_index = __locate(node, item, cur_index, &found_index);
1219 rcu_read_unlock();
1220 cond_resched();
1221 } while (cur_index != 0 && cur_index <= max_index);
1222
1223 return found_index;
1224}
1225#else
1226unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1227{
1228 return -1;
1229}
1230#endif /* CONFIG_SHMEM && CONFIG_SWAP */
1231
1232/**
1233 * radix_tree_shrink - shrink height of a radix tree to minimal
1234 * @root radix tree root
1235 */
1236static inline void radix_tree_shrink(struct radix_tree_root *root)
1237{
1238 /* try to shrink tree height */
1239 while (root->height > 0) {
1240 struct radix_tree_node *to_free = root->rnode;
1241 struct radix_tree_node *slot;
1242
1243 BUG_ON(!radix_tree_is_indirect_ptr(to_free));
1244 to_free = indirect_to_ptr(to_free);
1245
1246 /*
1247 * The candidate node has more than one child, or its child
1248 * is not at the leftmost slot, we cannot shrink.
1249 */
1250 if (to_free->count != 1)
1251 break;
1252 if (!to_free->slots[0])
1253 break;
1254
1255 /*
1256 * We don't need rcu_assign_pointer(), since we are simply
1257 * moving the node from one part of the tree to another: if it
1258 * was safe to dereference the old pointer to it
1259 * (to_free->slots[0]), it will be safe to dereference the new
1260 * one (root->rnode) as far as dependent read barriers go.
1261 */
1262 slot = to_free->slots[0];
1263 if (root->height > 1) {
1264 slot->parent = NULL;
1265 slot = ptr_to_indirect(slot);
1266 }
1267 root->rnode = slot;
1268 root->height--;
1269
1270 /*
1271 * We have a dilemma here. The node's slot[0] must not be
1272 * NULLed in case there are concurrent lookups expecting to
1273 * find the item. However if this was a bottom-level node,
1274 * then it may be subject to the slot pointer being visible
1275 * to callers dereferencing it. If item corresponding to
1276 * slot[0] is subsequently deleted, these callers would expect
1277 * their slot to become empty sooner or later.
1278 *
1279 * For example, lockless pagecache will look up a slot, deref
1280 * the page pointer, and if the page is 0 refcount it means it
1281 * was concurrently deleted from pagecache so try the deref
1282 * again. Fortunately there is already a requirement for logic
1283 * to retry the entire slot lookup -- the indirect pointer
1284 * problem (replacing direct root node with an indirect pointer
1285 * also results in a stale slot). So tag the slot as indirect
1286 * to force callers to retry.
1287 */
1288 if (root->height == 0)
1289 *((unsigned long *)&to_free->slots[0]) |=
1290 RADIX_TREE_INDIRECT_PTR;
1291
1292 radix_tree_node_free(to_free);
1293 }
1294}
1295
1296/**
1297 * __radix_tree_delete_node - try to free node after clearing a slot
1298 * @root: radix tree root
1299 * @index: index key
1300 * @node: node containing @index
1301 *
1302 * After clearing the slot at @index in @node from radix tree
1303 * rooted at @root, call this function to attempt freeing the
1304 * node and shrinking the tree.
1305 *
1306 * Returns %true if @node was freed, %false otherwise.
1307 */
1308bool __radix_tree_delete_node(struct radix_tree_root *root,
1309 struct radix_tree_node *node)
1310{
1311 bool deleted = false;
1312
1313 do {
1314 struct radix_tree_node *parent;
1315
1316 if (node->count) {
1317 if (node == indirect_to_ptr(root->rnode)) {
1318 radix_tree_shrink(root);
1319 if (root->height == 0)
1320 deleted = true;
1321 }
1322 return deleted;
1323 }
1324
1325 parent = node->parent;
1326 if (parent) {
1327 unsigned int offset;
1328
1329 offset = node->path >> RADIX_TREE_HEIGHT_SHIFT;
1330 parent->slots[offset] = NULL;
1331 parent->count--;
1332 } else {
1333 root_tag_clear_all(root);
1334 root->height = 0;
1335 root->rnode = NULL;
1336 }
1337
1338 radix_tree_node_free(node);
1339 deleted = true;
1340
1341 node = parent;
1342 } while (node);
1343
1344 return deleted;
1345}
1346
1347/**
1348 * radix_tree_delete_item - delete an item from a radix tree
1349 * @root: radix tree root
1350 * @index: index key
1351 * @item: expected item
1352 *
1353 * Remove @item at @index from the radix tree rooted at @root.
1354 *
1355 * Returns the address of the deleted item, or NULL if it was not present
1356 * or the entry at the given @index was not @item.
1357 */
1358void *radix_tree_delete_item(struct radix_tree_root *root,
1359 unsigned long index, void *item)
1360{
1361 struct radix_tree_node *node;
1362 unsigned int offset;
1363 void **slot;
1364 void *entry;
1365 int tag;
1366
1367 entry = __radix_tree_lookup(root, index, &node, &slot);
1368 if (!entry)
1369 return NULL;
1370
1371 if (item && entry != item)
1372 return NULL;
1373
1374 if (!node) {
1375 root_tag_clear_all(root);
1376 root->rnode = NULL;
1377 return entry;
1378 }
1379
1380 offset = index & RADIX_TREE_MAP_MASK;
1381
1382 /*
1383 * Clear all tags associated with the item to be deleted.
1384 * This way of doing it would be inefficient, but seldom is any set.
1385 */
1386 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1387 if (tag_get(node, tag, offset))
1388 radix_tree_tag_clear(root, index, tag);
1389 }
1390
1391 node->slots[offset] = NULL;
1392 node->count--;
1393
1394 __radix_tree_delete_node(root, node);
1395
1396 return entry;
1397}
1398EXPORT_SYMBOL(radix_tree_delete_item);
1399
1400/**
1401 * radix_tree_delete - delete an item from a radix tree
1402 * @root: radix tree root
1403 * @index: index key
1404 *
1405 * Remove the item at @index from the radix tree rooted at @root.
1406 *
1407 * Returns the address of the deleted item, or NULL if it was not present.
1408 */
1409void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1410{
1411 return radix_tree_delete_item(root, index, NULL);
1412}
1413EXPORT_SYMBOL(radix_tree_delete);
1414
1415/**
1416 * radix_tree_tagged - test whether any items in the tree are tagged
1417 * @root: radix tree root
1418 * @tag: tag to test
1419 */
1420int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1421{
1422 return root_tag_get(root, tag);
1423}
1424EXPORT_SYMBOL(radix_tree_tagged);
1425
1426static void
1427radix_tree_node_ctor(void *arg)
1428{
1429 struct radix_tree_node *node = arg;
1430
1431 memset(node, 0, sizeof(*node));
1432 INIT_LIST_HEAD(&node->private_list);
1433}
1434
1435static __init unsigned long __maxindex(unsigned int height)
1436{
1437 unsigned int width = height * RADIX_TREE_MAP_SHIFT;
1438 int shift = RADIX_TREE_INDEX_BITS - width;
1439
1440 if (shift < 0)
1441 return ~0UL;
1442 if (shift >= BITS_PER_LONG)
1443 return 0UL;
1444 return ~0UL >> shift;
1445}
1446
1447static __init void radix_tree_init_maxindex(void)
1448{
1449 unsigned int i;
1450
1451 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
1452 height_to_maxindex[i] = __maxindex(i);
1453}
1454
1455static int radix_tree_callback(struct notifier_block *nfb,
1456 unsigned long action,
1457 void *hcpu)
1458{
1459 int cpu = (long)hcpu;
1460 struct radix_tree_preload *rtp;
1461
1462 /* Free per-cpu pool of perloaded nodes */
1463 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1464 rtp = &per_cpu(radix_tree_preloads, cpu);
1465 while (rtp->nr) {
1466 kmem_cache_free(radix_tree_node_cachep,
1467 rtp->nodes[rtp->nr-1]);
1468 rtp->nodes[rtp->nr-1] = NULL;
1469 rtp->nr--;
1470 }
1471 }
1472 return NOTIFY_OK;
1473}
1474
1475void __init radix_tree_init(void)
1476{
1477 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1478 sizeof(struct radix_tree_node), 0,
1479 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1480 radix_tree_node_ctor);
1481 radix_tree_init_maxindex();
1482 hotcpu_notifier(radix_tree_callback, 0);
1483}
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) 2005 SGI, Christoph Lameter
6 * Copyright (C) 2006 Nick Piggin
7 * Copyright (C) 2012 Konstantin Khlebnikov
8 * Copyright (C) 2016 Intel, Matthew Wilcox
9 * Copyright (C) 2016 Intel, Ross Zwisler
10 */
11
12#include <linux/bitmap.h>
13#include <linux/bitops.h>
14#include <linux/bug.h>
15#include <linux/cpu.h>
16#include <linux/errno.h>
17#include <linux/export.h>
18#include <linux/idr.h>
19#include <linux/init.h>
20#include <linux/kernel.h>
21#include <linux/kmemleak.h>
22#include <linux/percpu.h>
23#include <linux/preempt.h> /* in_interrupt() */
24#include <linux/radix-tree.h>
25#include <linux/rcupdate.h>
26#include <linux/slab.h>
27#include <linux/string.h>
28#include <linux/xarray.h>
29
30#include "radix-tree.h"
31
32/*
33 * Radix tree node cache.
34 */
35struct kmem_cache *radix_tree_node_cachep;
36
37/*
38 * The radix tree is variable-height, so an insert operation not only has
39 * to build the branch to its corresponding item, it also has to build the
40 * branch to existing items if the size has to be increased (by
41 * radix_tree_extend).
42 *
43 * The worst case is a zero height tree with just a single item at index 0,
44 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
45 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
46 * Hence:
47 */
48#define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
49
50/*
51 * The IDR does not have to be as high as the radix tree since it uses
52 * signed integers, not unsigned longs.
53 */
54#define IDR_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(int) - 1)
55#define IDR_MAX_PATH (DIV_ROUND_UP(IDR_INDEX_BITS, \
56 RADIX_TREE_MAP_SHIFT))
57#define IDR_PRELOAD_SIZE (IDR_MAX_PATH * 2 - 1)
58
59/*
60 * Per-cpu pool of preloaded nodes
61 */
62DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = {
63 .lock = INIT_LOCAL_LOCK(lock),
64};
65EXPORT_PER_CPU_SYMBOL_GPL(radix_tree_preloads);
66
67static inline struct radix_tree_node *entry_to_node(void *ptr)
68{
69 return (void *)((unsigned long)ptr & ~RADIX_TREE_INTERNAL_NODE);
70}
71
72static inline void *node_to_entry(void *ptr)
73{
74 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
75}
76
77#define RADIX_TREE_RETRY XA_RETRY_ENTRY
78
79static inline unsigned long
80get_slot_offset(const struct radix_tree_node *parent, void __rcu **slot)
81{
82 return parent ? slot - parent->slots : 0;
83}
84
85static unsigned int radix_tree_descend(const struct radix_tree_node *parent,
86 struct radix_tree_node **nodep, unsigned long index)
87{
88 unsigned int offset = (index >> parent->shift) & RADIX_TREE_MAP_MASK;
89 void __rcu **entry = rcu_dereference_raw(parent->slots[offset]);
90
91 *nodep = (void *)entry;
92 return offset;
93}
94
95static inline gfp_t root_gfp_mask(const struct radix_tree_root *root)
96{
97 return root->xa_flags & (__GFP_BITS_MASK & ~GFP_ZONEMASK);
98}
99
100static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
101 int offset)
102{
103 __set_bit(offset, node->tags[tag]);
104}
105
106static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
107 int offset)
108{
109 __clear_bit(offset, node->tags[tag]);
110}
111
112static inline int tag_get(const struct radix_tree_node *node, unsigned int tag,
113 int offset)
114{
115 return test_bit(offset, node->tags[tag]);
116}
117
118static inline void root_tag_set(struct radix_tree_root *root, unsigned tag)
119{
120 root->xa_flags |= (__force gfp_t)(1 << (tag + ROOT_TAG_SHIFT));
121}
122
123static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
124{
125 root->xa_flags &= (__force gfp_t)~(1 << (tag + ROOT_TAG_SHIFT));
126}
127
128static inline void root_tag_clear_all(struct radix_tree_root *root)
129{
130 root->xa_flags &= (__force gfp_t)((1 << ROOT_TAG_SHIFT) - 1);
131}
132
133static inline int root_tag_get(const struct radix_tree_root *root, unsigned tag)
134{
135 return (__force int)root->xa_flags & (1 << (tag + ROOT_TAG_SHIFT));
136}
137
138static inline unsigned root_tags_get(const struct radix_tree_root *root)
139{
140 return (__force unsigned)root->xa_flags >> ROOT_TAG_SHIFT;
141}
142
143static inline bool is_idr(const struct radix_tree_root *root)
144{
145 return !!(root->xa_flags & ROOT_IS_IDR);
146}
147
148/*
149 * Returns 1 if any slot in the node has this tag set.
150 * Otherwise returns 0.
151 */
152static inline int any_tag_set(const struct radix_tree_node *node,
153 unsigned int tag)
154{
155 unsigned idx;
156 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
157 if (node->tags[tag][idx])
158 return 1;
159 }
160 return 0;
161}
162
163static inline void all_tag_set(struct radix_tree_node *node, unsigned int tag)
164{
165 bitmap_fill(node->tags[tag], RADIX_TREE_MAP_SIZE);
166}
167
168/**
169 * radix_tree_find_next_bit - find the next set bit in a memory region
170 *
171 * @node: where to begin the search
172 * @tag: the tag index
173 * @offset: the bitnumber to start searching at
174 *
175 * Unrollable variant of find_next_bit() for constant size arrays.
176 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
177 * Returns next bit offset, or size if nothing found.
178 */
179static __always_inline unsigned long
180radix_tree_find_next_bit(struct radix_tree_node *node, unsigned int tag,
181 unsigned long offset)
182{
183 const unsigned long *addr = node->tags[tag];
184
185 if (offset < RADIX_TREE_MAP_SIZE) {
186 unsigned long tmp;
187
188 addr += offset / BITS_PER_LONG;
189 tmp = *addr >> (offset % BITS_PER_LONG);
190 if (tmp)
191 return __ffs(tmp) + offset;
192 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
193 while (offset < RADIX_TREE_MAP_SIZE) {
194 tmp = *++addr;
195 if (tmp)
196 return __ffs(tmp) + offset;
197 offset += BITS_PER_LONG;
198 }
199 }
200 return RADIX_TREE_MAP_SIZE;
201}
202
203static unsigned int iter_offset(const struct radix_tree_iter *iter)
204{
205 return iter->index & RADIX_TREE_MAP_MASK;
206}
207
208/*
209 * The maximum index which can be stored in a radix tree
210 */
211static inline unsigned long shift_maxindex(unsigned int shift)
212{
213 return (RADIX_TREE_MAP_SIZE << shift) - 1;
214}
215
216static inline unsigned long node_maxindex(const struct radix_tree_node *node)
217{
218 return shift_maxindex(node->shift);
219}
220
221static unsigned long next_index(unsigned long index,
222 const struct radix_tree_node *node,
223 unsigned long offset)
224{
225 return (index & ~node_maxindex(node)) + (offset << node->shift);
226}
227
228/*
229 * This assumes that the caller has performed appropriate preallocation, and
230 * that the caller has pinned this thread of control to the current CPU.
231 */
232static struct radix_tree_node *
233radix_tree_node_alloc(gfp_t gfp_mask, struct radix_tree_node *parent,
234 struct radix_tree_root *root,
235 unsigned int shift, unsigned int offset,
236 unsigned int count, unsigned int nr_values)
237{
238 struct radix_tree_node *ret = NULL;
239
240 /*
241 * Preload code isn't irq safe and it doesn't make sense to use
242 * preloading during an interrupt anyway as all the allocations have
243 * to be atomic. So just do normal allocation when in interrupt.
244 */
245 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
246 struct radix_tree_preload *rtp;
247
248 /*
249 * Even if the caller has preloaded, try to allocate from the
250 * cache first for the new node to get accounted to the memory
251 * cgroup.
252 */
253 ret = kmem_cache_alloc(radix_tree_node_cachep,
254 gfp_mask | __GFP_NOWARN);
255 if (ret)
256 goto out;
257
258 /*
259 * Provided the caller has preloaded here, we will always
260 * succeed in getting a node here (and never reach
261 * kmem_cache_alloc)
262 */
263 rtp = this_cpu_ptr(&radix_tree_preloads);
264 if (rtp->nr) {
265 ret = rtp->nodes;
266 rtp->nodes = ret->parent;
267 rtp->nr--;
268 }
269 /*
270 * Update the allocation stack trace as this is more useful
271 * for debugging.
272 */
273 kmemleak_update_trace(ret);
274 goto out;
275 }
276 ret = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
277out:
278 BUG_ON(radix_tree_is_internal_node(ret));
279 if (ret) {
280 ret->shift = shift;
281 ret->offset = offset;
282 ret->count = count;
283 ret->nr_values = nr_values;
284 ret->parent = parent;
285 ret->array = root;
286 }
287 return ret;
288}
289
290void radix_tree_node_rcu_free(struct rcu_head *head)
291{
292 struct radix_tree_node *node =
293 container_of(head, struct radix_tree_node, rcu_head);
294
295 /*
296 * Must only free zeroed nodes into the slab. We can be left with
297 * non-NULL entries by radix_tree_free_nodes, so clear the entries
298 * and tags here.
299 */
300 memset(node->slots, 0, sizeof(node->slots));
301 memset(node->tags, 0, sizeof(node->tags));
302 INIT_LIST_HEAD(&node->private_list);
303
304 kmem_cache_free(radix_tree_node_cachep, node);
305}
306
307static inline void
308radix_tree_node_free(struct radix_tree_node *node)
309{
310 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
311}
312
313/*
314 * Load up this CPU's radix_tree_node buffer with sufficient objects to
315 * ensure that the addition of a single element in the tree cannot fail. On
316 * success, return zero, with preemption disabled. On error, return -ENOMEM
317 * with preemption not disabled.
318 *
319 * To make use of this facility, the radix tree must be initialised without
320 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
321 */
322static __must_check int __radix_tree_preload(gfp_t gfp_mask, unsigned nr)
323{
324 struct radix_tree_preload *rtp;
325 struct radix_tree_node *node;
326 int ret = -ENOMEM;
327
328 /*
329 * Nodes preloaded by one cgroup can be used by another cgroup, so
330 * they should never be accounted to any particular memory cgroup.
331 */
332 gfp_mask &= ~__GFP_ACCOUNT;
333
334 local_lock(&radix_tree_preloads.lock);
335 rtp = this_cpu_ptr(&radix_tree_preloads);
336 while (rtp->nr < nr) {
337 local_unlock(&radix_tree_preloads.lock);
338 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
339 if (node == NULL)
340 goto out;
341 local_lock(&radix_tree_preloads.lock);
342 rtp = this_cpu_ptr(&radix_tree_preloads);
343 if (rtp->nr < nr) {
344 node->parent = rtp->nodes;
345 rtp->nodes = node;
346 rtp->nr++;
347 } else {
348 kmem_cache_free(radix_tree_node_cachep, node);
349 }
350 }
351 ret = 0;
352out:
353 return ret;
354}
355
356/*
357 * Load up this CPU's radix_tree_node buffer with sufficient objects to
358 * ensure that the addition of a single element in the tree cannot fail. On
359 * success, return zero, with preemption disabled. On error, return -ENOMEM
360 * with preemption not disabled.
361 *
362 * To make use of this facility, the radix tree must be initialised without
363 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
364 */
365int radix_tree_preload(gfp_t gfp_mask)
366{
367 /* Warn on non-sensical use... */
368 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
369 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
370}
371EXPORT_SYMBOL(radix_tree_preload);
372
373/*
374 * The same as above function, except we don't guarantee preloading happens.
375 * We do it, if we decide it helps. On success, return zero with preemption
376 * disabled. On error, return -ENOMEM with preemption not disabled.
377 */
378int radix_tree_maybe_preload(gfp_t gfp_mask)
379{
380 if (gfpflags_allow_blocking(gfp_mask))
381 return __radix_tree_preload(gfp_mask, RADIX_TREE_PRELOAD_SIZE);
382 /* Preloading doesn't help anything with this gfp mask, skip it */
383 local_lock(&radix_tree_preloads.lock);
384 return 0;
385}
386EXPORT_SYMBOL(radix_tree_maybe_preload);
387
388static unsigned radix_tree_load_root(const struct radix_tree_root *root,
389 struct radix_tree_node **nodep, unsigned long *maxindex)
390{
391 struct radix_tree_node *node = rcu_dereference_raw(root->xa_head);
392
393 *nodep = node;
394
395 if (likely(radix_tree_is_internal_node(node))) {
396 node = entry_to_node(node);
397 *maxindex = node_maxindex(node);
398 return node->shift + RADIX_TREE_MAP_SHIFT;
399 }
400
401 *maxindex = 0;
402 return 0;
403}
404
405/*
406 * Extend a radix tree so it can store key @index.
407 */
408static int radix_tree_extend(struct radix_tree_root *root, gfp_t gfp,
409 unsigned long index, unsigned int shift)
410{
411 void *entry;
412 unsigned int maxshift;
413 int tag;
414
415 /* Figure out what the shift should be. */
416 maxshift = shift;
417 while (index > shift_maxindex(maxshift))
418 maxshift += RADIX_TREE_MAP_SHIFT;
419
420 entry = rcu_dereference_raw(root->xa_head);
421 if (!entry && (!is_idr(root) || root_tag_get(root, IDR_FREE)))
422 goto out;
423
424 do {
425 struct radix_tree_node *node = radix_tree_node_alloc(gfp, NULL,
426 root, shift, 0, 1, 0);
427 if (!node)
428 return -ENOMEM;
429
430 if (is_idr(root)) {
431 all_tag_set(node, IDR_FREE);
432 if (!root_tag_get(root, IDR_FREE)) {
433 tag_clear(node, IDR_FREE, 0);
434 root_tag_set(root, IDR_FREE);
435 }
436 } else {
437 /* Propagate the aggregated tag info to the new child */
438 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
439 if (root_tag_get(root, tag))
440 tag_set(node, tag, 0);
441 }
442 }
443
444 BUG_ON(shift > BITS_PER_LONG);
445 if (radix_tree_is_internal_node(entry)) {
446 entry_to_node(entry)->parent = node;
447 } else if (xa_is_value(entry)) {
448 /* Moving a value entry root->xa_head to a node */
449 node->nr_values = 1;
450 }
451 /*
452 * entry was already in the radix tree, so we do not need
453 * rcu_assign_pointer here
454 */
455 node->slots[0] = (void __rcu *)entry;
456 entry = node_to_entry(node);
457 rcu_assign_pointer(root->xa_head, entry);
458 shift += RADIX_TREE_MAP_SHIFT;
459 } while (shift <= maxshift);
460out:
461 return maxshift + RADIX_TREE_MAP_SHIFT;
462}
463
464/**
465 * radix_tree_shrink - shrink radix tree to minimum height
466 * @root: radix tree root
467 */
468static inline bool radix_tree_shrink(struct radix_tree_root *root)
469{
470 bool shrunk = false;
471
472 for (;;) {
473 struct radix_tree_node *node = rcu_dereference_raw(root->xa_head);
474 struct radix_tree_node *child;
475
476 if (!radix_tree_is_internal_node(node))
477 break;
478 node = entry_to_node(node);
479
480 /*
481 * The candidate node has more than one child, or its child
482 * is not at the leftmost slot, we cannot shrink.
483 */
484 if (node->count != 1)
485 break;
486 child = rcu_dereference_raw(node->slots[0]);
487 if (!child)
488 break;
489
490 /*
491 * For an IDR, we must not shrink entry 0 into the root in
492 * case somebody calls idr_replace() with a pointer that
493 * appears to be an internal entry
494 */
495 if (!node->shift && is_idr(root))
496 break;
497
498 if (radix_tree_is_internal_node(child))
499 entry_to_node(child)->parent = NULL;
500
501 /*
502 * We don't need rcu_assign_pointer(), since we are simply
503 * moving the node from one part of the tree to another: if it
504 * was safe to dereference the old pointer to it
505 * (node->slots[0]), it will be safe to dereference the new
506 * one (root->xa_head) as far as dependent read barriers go.
507 */
508 root->xa_head = (void __rcu *)child;
509 if (is_idr(root) && !tag_get(node, IDR_FREE, 0))
510 root_tag_clear(root, IDR_FREE);
511
512 /*
513 * We have a dilemma here. The node's slot[0] must not be
514 * NULLed in case there are concurrent lookups expecting to
515 * find the item. However if this was a bottom-level node,
516 * then it may be subject to the slot pointer being visible
517 * to callers dereferencing it. If item corresponding to
518 * slot[0] is subsequently deleted, these callers would expect
519 * their slot to become empty sooner or later.
520 *
521 * For example, lockless pagecache will look up a slot, deref
522 * the page pointer, and if the page has 0 refcount it means it
523 * was concurrently deleted from pagecache so try the deref
524 * again. Fortunately there is already a requirement for logic
525 * to retry the entire slot lookup -- the indirect pointer
526 * problem (replacing direct root node with an indirect pointer
527 * also results in a stale slot). So tag the slot as indirect
528 * to force callers to retry.
529 */
530 node->count = 0;
531 if (!radix_tree_is_internal_node(child)) {
532 node->slots[0] = (void __rcu *)RADIX_TREE_RETRY;
533 }
534
535 WARN_ON_ONCE(!list_empty(&node->private_list));
536 radix_tree_node_free(node);
537 shrunk = true;
538 }
539
540 return shrunk;
541}
542
543static bool delete_node(struct radix_tree_root *root,
544 struct radix_tree_node *node)
545{
546 bool deleted = false;
547
548 do {
549 struct radix_tree_node *parent;
550
551 if (node->count) {
552 if (node_to_entry(node) ==
553 rcu_dereference_raw(root->xa_head))
554 deleted |= radix_tree_shrink(root);
555 return deleted;
556 }
557
558 parent = node->parent;
559 if (parent) {
560 parent->slots[node->offset] = NULL;
561 parent->count--;
562 } else {
563 /*
564 * Shouldn't the tags already have all been cleared
565 * by the caller?
566 */
567 if (!is_idr(root))
568 root_tag_clear_all(root);
569 root->xa_head = NULL;
570 }
571
572 WARN_ON_ONCE(!list_empty(&node->private_list));
573 radix_tree_node_free(node);
574 deleted = true;
575
576 node = parent;
577 } while (node);
578
579 return deleted;
580}
581
582/**
583 * __radix_tree_create - create a slot in a radix tree
584 * @root: radix tree root
585 * @index: index key
586 * @nodep: returns node
587 * @slotp: returns slot
588 *
589 * Create, if necessary, and return the node and slot for an item
590 * at position @index in the radix tree @root.
591 *
592 * Until there is more than one item in the tree, no nodes are
593 * allocated and @root->xa_head is used as a direct slot instead of
594 * pointing to a node, in which case *@nodep will be NULL.
595 *
596 * Returns -ENOMEM, or 0 for success.
597 */
598static int __radix_tree_create(struct radix_tree_root *root,
599 unsigned long index, struct radix_tree_node **nodep,
600 void __rcu ***slotp)
601{
602 struct radix_tree_node *node = NULL, *child;
603 void __rcu **slot = (void __rcu **)&root->xa_head;
604 unsigned long maxindex;
605 unsigned int shift, offset = 0;
606 unsigned long max = index;
607 gfp_t gfp = root_gfp_mask(root);
608
609 shift = radix_tree_load_root(root, &child, &maxindex);
610
611 /* Make sure the tree is high enough. */
612 if (max > maxindex) {
613 int error = radix_tree_extend(root, gfp, max, shift);
614 if (error < 0)
615 return error;
616 shift = error;
617 child = rcu_dereference_raw(root->xa_head);
618 }
619
620 while (shift > 0) {
621 shift -= RADIX_TREE_MAP_SHIFT;
622 if (child == NULL) {
623 /* Have to add a child node. */
624 child = radix_tree_node_alloc(gfp, node, root, shift,
625 offset, 0, 0);
626 if (!child)
627 return -ENOMEM;
628 rcu_assign_pointer(*slot, node_to_entry(child));
629 if (node)
630 node->count++;
631 } else if (!radix_tree_is_internal_node(child))
632 break;
633
634 /* Go a level down */
635 node = entry_to_node(child);
636 offset = radix_tree_descend(node, &child, index);
637 slot = &node->slots[offset];
638 }
639
640 if (nodep)
641 *nodep = node;
642 if (slotp)
643 *slotp = slot;
644 return 0;
645}
646
647/*
648 * Free any nodes below this node. The tree is presumed to not need
649 * shrinking, and any user data in the tree is presumed to not need a
650 * destructor called on it. If we need to add a destructor, we can
651 * add that functionality later. Note that we may not clear tags or
652 * slots from the tree as an RCU walker may still have a pointer into
653 * this subtree. We could replace the entries with RADIX_TREE_RETRY,
654 * but we'll still have to clear those in rcu_free.
655 */
656static void radix_tree_free_nodes(struct radix_tree_node *node)
657{
658 unsigned offset = 0;
659 struct radix_tree_node *child = entry_to_node(node);
660
661 for (;;) {
662 void *entry = rcu_dereference_raw(child->slots[offset]);
663 if (xa_is_node(entry) && child->shift) {
664 child = entry_to_node(entry);
665 offset = 0;
666 continue;
667 }
668 offset++;
669 while (offset == RADIX_TREE_MAP_SIZE) {
670 struct radix_tree_node *old = child;
671 offset = child->offset + 1;
672 child = child->parent;
673 WARN_ON_ONCE(!list_empty(&old->private_list));
674 radix_tree_node_free(old);
675 if (old == entry_to_node(node))
676 return;
677 }
678 }
679}
680
681static inline int insert_entries(struct radix_tree_node *node,
682 void __rcu **slot, void *item)
683{
684 if (*slot)
685 return -EEXIST;
686 rcu_assign_pointer(*slot, item);
687 if (node) {
688 node->count++;
689 if (xa_is_value(item))
690 node->nr_values++;
691 }
692 return 1;
693}
694
695/**
696 * radix_tree_insert - insert into a radix tree
697 * @root: radix tree root
698 * @index: index key
699 * @item: item to insert
700 *
701 * Insert an item into the radix tree at position @index.
702 */
703int radix_tree_insert(struct radix_tree_root *root, unsigned long index,
704 void *item)
705{
706 struct radix_tree_node *node;
707 void __rcu **slot;
708 int error;
709
710 BUG_ON(radix_tree_is_internal_node(item));
711
712 error = __radix_tree_create(root, index, &node, &slot);
713 if (error)
714 return error;
715
716 error = insert_entries(node, slot, item);
717 if (error < 0)
718 return error;
719
720 if (node) {
721 unsigned offset = get_slot_offset(node, slot);
722 BUG_ON(tag_get(node, 0, offset));
723 BUG_ON(tag_get(node, 1, offset));
724 BUG_ON(tag_get(node, 2, offset));
725 } else {
726 BUG_ON(root_tags_get(root));
727 }
728
729 return 0;
730}
731EXPORT_SYMBOL(radix_tree_insert);
732
733/**
734 * __radix_tree_lookup - lookup an item in a radix tree
735 * @root: radix tree root
736 * @index: index key
737 * @nodep: returns node
738 * @slotp: returns slot
739 *
740 * Lookup and return the item at position @index in the radix
741 * tree @root.
742 *
743 * Until there is more than one item in the tree, no nodes are
744 * allocated and @root->xa_head is used as a direct slot instead of
745 * pointing to a node, in which case *@nodep will be NULL.
746 */
747void *__radix_tree_lookup(const struct radix_tree_root *root,
748 unsigned long index, struct radix_tree_node **nodep,
749 void __rcu ***slotp)
750{
751 struct radix_tree_node *node, *parent;
752 unsigned long maxindex;
753 void __rcu **slot;
754
755 restart:
756 parent = NULL;
757 slot = (void __rcu **)&root->xa_head;
758 radix_tree_load_root(root, &node, &maxindex);
759 if (index > maxindex)
760 return NULL;
761
762 while (radix_tree_is_internal_node(node)) {
763 unsigned offset;
764
765 parent = entry_to_node(node);
766 offset = radix_tree_descend(parent, &node, index);
767 slot = parent->slots + offset;
768 if (node == RADIX_TREE_RETRY)
769 goto restart;
770 if (parent->shift == 0)
771 break;
772 }
773
774 if (nodep)
775 *nodep = parent;
776 if (slotp)
777 *slotp = slot;
778 return node;
779}
780
781/**
782 * radix_tree_lookup_slot - lookup a slot in a radix tree
783 * @root: radix tree root
784 * @index: index key
785 *
786 * Returns: the slot corresponding to the position @index in the
787 * radix tree @root. This is useful for update-if-exists operations.
788 *
789 * This function can be called under rcu_read_lock iff the slot is not
790 * modified by radix_tree_replace_slot, otherwise it must be called
791 * exclusive from other writers. Any dereference of the slot must be done
792 * using radix_tree_deref_slot.
793 */
794void __rcu **radix_tree_lookup_slot(const struct radix_tree_root *root,
795 unsigned long index)
796{
797 void __rcu **slot;
798
799 if (!__radix_tree_lookup(root, index, NULL, &slot))
800 return NULL;
801 return slot;
802}
803EXPORT_SYMBOL(radix_tree_lookup_slot);
804
805/**
806 * radix_tree_lookup - perform lookup operation on a radix tree
807 * @root: radix tree root
808 * @index: index key
809 *
810 * Lookup the item at the position @index in the radix tree @root.
811 *
812 * This function can be called under rcu_read_lock, however the caller
813 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
814 * them safely). No RCU barriers are required to access or modify the
815 * returned item, however.
816 */
817void *radix_tree_lookup(const struct radix_tree_root *root, unsigned long index)
818{
819 return __radix_tree_lookup(root, index, NULL, NULL);
820}
821EXPORT_SYMBOL(radix_tree_lookup);
822
823static void replace_slot(void __rcu **slot, void *item,
824 struct radix_tree_node *node, int count, int values)
825{
826 if (node && (count || values)) {
827 node->count += count;
828 node->nr_values += values;
829 }
830
831 rcu_assign_pointer(*slot, item);
832}
833
834static bool node_tag_get(const struct radix_tree_root *root,
835 const struct radix_tree_node *node,
836 unsigned int tag, unsigned int offset)
837{
838 if (node)
839 return tag_get(node, tag, offset);
840 return root_tag_get(root, tag);
841}
842
843/*
844 * IDR users want to be able to store NULL in the tree, so if the slot isn't
845 * free, don't adjust the count, even if it's transitioning between NULL and
846 * non-NULL. For the IDA, we mark slots as being IDR_FREE while they still
847 * have empty bits, but it only stores NULL in slots when they're being
848 * deleted.
849 */
850static int calculate_count(struct radix_tree_root *root,
851 struct radix_tree_node *node, void __rcu **slot,
852 void *item, void *old)
853{
854 if (is_idr(root)) {
855 unsigned offset = get_slot_offset(node, slot);
856 bool free = node_tag_get(root, node, IDR_FREE, offset);
857 if (!free)
858 return 0;
859 if (!old)
860 return 1;
861 }
862 return !!item - !!old;
863}
864
865/**
866 * __radix_tree_replace - replace item in a slot
867 * @root: radix tree root
868 * @node: pointer to tree node
869 * @slot: pointer to slot in @node
870 * @item: new item to store in the slot.
871 *
872 * For use with __radix_tree_lookup(). Caller must hold tree write locked
873 * across slot lookup and replacement.
874 */
875void __radix_tree_replace(struct radix_tree_root *root,
876 struct radix_tree_node *node,
877 void __rcu **slot, void *item)
878{
879 void *old = rcu_dereference_raw(*slot);
880 int values = !!xa_is_value(item) - !!xa_is_value(old);
881 int count = calculate_count(root, node, slot, item, old);
882
883 /*
884 * This function supports replacing value entries and
885 * deleting entries, but that needs accounting against the
886 * node unless the slot is root->xa_head.
887 */
888 WARN_ON_ONCE(!node && (slot != (void __rcu **)&root->xa_head) &&
889 (count || values));
890 replace_slot(slot, item, node, count, values);
891
892 if (!node)
893 return;
894
895 delete_node(root, node);
896}
897
898/**
899 * radix_tree_replace_slot - replace item in a slot
900 * @root: radix tree root
901 * @slot: pointer to slot
902 * @item: new item to store in the slot.
903 *
904 * For use with radix_tree_lookup_slot() and
905 * radix_tree_gang_lookup_tag_slot(). Caller must hold tree write locked
906 * across slot lookup and replacement.
907 *
908 * NOTE: This cannot be used to switch between non-entries (empty slots),
909 * regular entries, and value entries, as that requires accounting
910 * inside the radix tree node. When switching from one type of entry or
911 * deleting, use __radix_tree_lookup() and __radix_tree_replace() or
912 * radix_tree_iter_replace().
913 */
914void radix_tree_replace_slot(struct radix_tree_root *root,
915 void __rcu **slot, void *item)
916{
917 __radix_tree_replace(root, NULL, slot, item);
918}
919EXPORT_SYMBOL(radix_tree_replace_slot);
920
921/**
922 * radix_tree_iter_replace - replace item in a slot
923 * @root: radix tree root
924 * @iter: iterator state
925 * @slot: pointer to slot
926 * @item: new item to store in the slot.
927 *
928 * For use with radix_tree_for_each_slot().
929 * Caller must hold tree write locked.
930 */
931void radix_tree_iter_replace(struct radix_tree_root *root,
932 const struct radix_tree_iter *iter,
933 void __rcu **slot, void *item)
934{
935 __radix_tree_replace(root, iter->node, slot, item);
936}
937
938static void node_tag_set(struct radix_tree_root *root,
939 struct radix_tree_node *node,
940 unsigned int tag, unsigned int offset)
941{
942 while (node) {
943 if (tag_get(node, tag, offset))
944 return;
945 tag_set(node, tag, offset);
946 offset = node->offset;
947 node = node->parent;
948 }
949
950 if (!root_tag_get(root, tag))
951 root_tag_set(root, tag);
952}
953
954/**
955 * radix_tree_tag_set - set a tag on a radix tree node
956 * @root: radix tree root
957 * @index: index key
958 * @tag: tag index
959 *
960 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
961 * corresponding to @index in the radix tree. From
962 * the root all the way down to the leaf node.
963 *
964 * Returns the address of the tagged item. Setting a tag on a not-present
965 * item is a bug.
966 */
967void *radix_tree_tag_set(struct radix_tree_root *root,
968 unsigned long index, unsigned int tag)
969{
970 struct radix_tree_node *node, *parent;
971 unsigned long maxindex;
972
973 radix_tree_load_root(root, &node, &maxindex);
974 BUG_ON(index > maxindex);
975
976 while (radix_tree_is_internal_node(node)) {
977 unsigned offset;
978
979 parent = entry_to_node(node);
980 offset = radix_tree_descend(parent, &node, index);
981 BUG_ON(!node);
982
983 if (!tag_get(parent, tag, offset))
984 tag_set(parent, tag, offset);
985 }
986
987 /* set the root's tag bit */
988 if (!root_tag_get(root, tag))
989 root_tag_set(root, tag);
990
991 return node;
992}
993EXPORT_SYMBOL(radix_tree_tag_set);
994
995static void node_tag_clear(struct radix_tree_root *root,
996 struct radix_tree_node *node,
997 unsigned int tag, unsigned int offset)
998{
999 while (node) {
1000 if (!tag_get(node, tag, offset))
1001 return;
1002 tag_clear(node, tag, offset);
1003 if (any_tag_set(node, tag))
1004 return;
1005
1006 offset = node->offset;
1007 node = node->parent;
1008 }
1009
1010 /* clear the root's tag bit */
1011 if (root_tag_get(root, tag))
1012 root_tag_clear(root, tag);
1013}
1014
1015/**
1016 * radix_tree_tag_clear - clear a tag on a radix tree node
1017 * @root: radix tree root
1018 * @index: index key
1019 * @tag: tag index
1020 *
1021 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
1022 * corresponding to @index in the radix tree. If this causes
1023 * the leaf node to have no tags set then clear the tag in the
1024 * next-to-leaf node, etc.
1025 *
1026 * Returns the address of the tagged item on success, else NULL. ie:
1027 * has the same return value and semantics as radix_tree_lookup().
1028 */
1029void *radix_tree_tag_clear(struct radix_tree_root *root,
1030 unsigned long index, unsigned int tag)
1031{
1032 struct radix_tree_node *node, *parent;
1033 unsigned long maxindex;
1034 int offset = 0;
1035
1036 radix_tree_load_root(root, &node, &maxindex);
1037 if (index > maxindex)
1038 return NULL;
1039
1040 parent = NULL;
1041
1042 while (radix_tree_is_internal_node(node)) {
1043 parent = entry_to_node(node);
1044 offset = radix_tree_descend(parent, &node, index);
1045 }
1046
1047 if (node)
1048 node_tag_clear(root, parent, tag, offset);
1049
1050 return node;
1051}
1052EXPORT_SYMBOL(radix_tree_tag_clear);
1053
1054/**
1055 * radix_tree_iter_tag_clear - clear a tag on the current iterator entry
1056 * @root: radix tree root
1057 * @iter: iterator state
1058 * @tag: tag to clear
1059 */
1060void radix_tree_iter_tag_clear(struct radix_tree_root *root,
1061 const struct radix_tree_iter *iter, unsigned int tag)
1062{
1063 node_tag_clear(root, iter->node, tag, iter_offset(iter));
1064}
1065
1066/**
1067 * radix_tree_tag_get - get a tag on a radix tree node
1068 * @root: radix tree root
1069 * @index: index key
1070 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
1071 *
1072 * Return values:
1073 *
1074 * 0: tag not present or not set
1075 * 1: tag set
1076 *
1077 * Note that the return value of this function may not be relied on, even if
1078 * the RCU lock is held, unless tag modification and node deletion are excluded
1079 * from concurrency.
1080 */
1081int radix_tree_tag_get(const struct radix_tree_root *root,
1082 unsigned long index, unsigned int tag)
1083{
1084 struct radix_tree_node *node, *parent;
1085 unsigned long maxindex;
1086
1087 if (!root_tag_get(root, tag))
1088 return 0;
1089
1090 radix_tree_load_root(root, &node, &maxindex);
1091 if (index > maxindex)
1092 return 0;
1093
1094 while (radix_tree_is_internal_node(node)) {
1095 unsigned offset;
1096
1097 parent = entry_to_node(node);
1098 offset = radix_tree_descend(parent, &node, index);
1099
1100 if (!tag_get(parent, tag, offset))
1101 return 0;
1102 if (node == RADIX_TREE_RETRY)
1103 break;
1104 }
1105
1106 return 1;
1107}
1108EXPORT_SYMBOL(radix_tree_tag_get);
1109
1110/* Construct iter->tags bit-mask from node->tags[tag] array */
1111static void set_iter_tags(struct radix_tree_iter *iter,
1112 struct radix_tree_node *node, unsigned offset,
1113 unsigned tag)
1114{
1115 unsigned tag_long = offset / BITS_PER_LONG;
1116 unsigned tag_bit = offset % BITS_PER_LONG;
1117
1118 if (!node) {
1119 iter->tags = 1;
1120 return;
1121 }
1122
1123 iter->tags = node->tags[tag][tag_long] >> tag_bit;
1124
1125 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
1126 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
1127 /* Pick tags from next element */
1128 if (tag_bit)
1129 iter->tags |= node->tags[tag][tag_long + 1] <<
1130 (BITS_PER_LONG - tag_bit);
1131 /* Clip chunk size, here only BITS_PER_LONG tags */
1132 iter->next_index = __radix_tree_iter_add(iter, BITS_PER_LONG);
1133 }
1134}
1135
1136void __rcu **radix_tree_iter_resume(void __rcu **slot,
1137 struct radix_tree_iter *iter)
1138{
1139 iter->index = __radix_tree_iter_add(iter, 1);
1140 iter->next_index = iter->index;
1141 iter->tags = 0;
1142 return NULL;
1143}
1144EXPORT_SYMBOL(radix_tree_iter_resume);
1145
1146/**
1147 * radix_tree_next_chunk - find next chunk of slots for iteration
1148 *
1149 * @root: radix tree root
1150 * @iter: iterator state
1151 * @flags: RADIX_TREE_ITER_* flags and tag index
1152 * Returns: pointer to chunk first slot, or NULL if iteration is over
1153 */
1154void __rcu **radix_tree_next_chunk(const struct radix_tree_root *root,
1155 struct radix_tree_iter *iter, unsigned flags)
1156{
1157 unsigned tag = flags & RADIX_TREE_ITER_TAG_MASK;
1158 struct radix_tree_node *node, *child;
1159 unsigned long index, offset, maxindex;
1160
1161 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
1162 return NULL;
1163
1164 /*
1165 * Catch next_index overflow after ~0UL. iter->index never overflows
1166 * during iterating; it can be zero only at the beginning.
1167 * And we cannot overflow iter->next_index in a single step,
1168 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
1169 *
1170 * This condition also used by radix_tree_next_slot() to stop
1171 * contiguous iterating, and forbid switching to the next chunk.
1172 */
1173 index = iter->next_index;
1174 if (!index && iter->index)
1175 return NULL;
1176
1177 restart:
1178 radix_tree_load_root(root, &child, &maxindex);
1179 if (index > maxindex)
1180 return NULL;
1181 if (!child)
1182 return NULL;
1183
1184 if (!radix_tree_is_internal_node(child)) {
1185 /* Single-slot tree */
1186 iter->index = index;
1187 iter->next_index = maxindex + 1;
1188 iter->tags = 1;
1189 iter->node = NULL;
1190 return (void __rcu **)&root->xa_head;
1191 }
1192
1193 do {
1194 node = entry_to_node(child);
1195 offset = radix_tree_descend(node, &child, index);
1196
1197 if ((flags & RADIX_TREE_ITER_TAGGED) ?
1198 !tag_get(node, tag, offset) : !child) {
1199 /* Hole detected */
1200 if (flags & RADIX_TREE_ITER_CONTIG)
1201 return NULL;
1202
1203 if (flags & RADIX_TREE_ITER_TAGGED)
1204 offset = radix_tree_find_next_bit(node, tag,
1205 offset + 1);
1206 else
1207 while (++offset < RADIX_TREE_MAP_SIZE) {
1208 void *slot = rcu_dereference_raw(
1209 node->slots[offset]);
1210 if (slot)
1211 break;
1212 }
1213 index &= ~node_maxindex(node);
1214 index += offset << node->shift;
1215 /* Overflow after ~0UL */
1216 if (!index)
1217 return NULL;
1218 if (offset == RADIX_TREE_MAP_SIZE)
1219 goto restart;
1220 child = rcu_dereference_raw(node->slots[offset]);
1221 }
1222
1223 if (!child)
1224 goto restart;
1225 if (child == RADIX_TREE_RETRY)
1226 break;
1227 } while (node->shift && radix_tree_is_internal_node(child));
1228
1229 /* Update the iterator state */
1230 iter->index = (index &~ node_maxindex(node)) | offset;
1231 iter->next_index = (index | node_maxindex(node)) + 1;
1232 iter->node = node;
1233
1234 if (flags & RADIX_TREE_ITER_TAGGED)
1235 set_iter_tags(iter, node, offset, tag);
1236
1237 return node->slots + offset;
1238}
1239EXPORT_SYMBOL(radix_tree_next_chunk);
1240
1241/**
1242 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1243 * @root: radix tree root
1244 * @results: where the results of the lookup are placed
1245 * @first_index: start the lookup from this key
1246 * @max_items: place up to this many items at *results
1247 *
1248 * Performs an index-ascending scan of the tree for present items. Places
1249 * them at *@results and returns the number of items which were placed at
1250 * *@results.
1251 *
1252 * The implementation is naive.
1253 *
1254 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1255 * rcu_read_lock. In this case, rather than the returned results being
1256 * an atomic snapshot of the tree at a single point in time, the
1257 * semantics of an RCU protected gang lookup are as though multiple
1258 * radix_tree_lookups have been issued in individual locks, and results
1259 * stored in 'results'.
1260 */
1261unsigned int
1262radix_tree_gang_lookup(const struct radix_tree_root *root, void **results,
1263 unsigned long first_index, unsigned int max_items)
1264{
1265 struct radix_tree_iter iter;
1266 void __rcu **slot;
1267 unsigned int ret = 0;
1268
1269 if (unlikely(!max_items))
1270 return 0;
1271
1272 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1273 results[ret] = rcu_dereference_raw(*slot);
1274 if (!results[ret])
1275 continue;
1276 if (radix_tree_is_internal_node(results[ret])) {
1277 slot = radix_tree_iter_retry(&iter);
1278 continue;
1279 }
1280 if (++ret == max_items)
1281 break;
1282 }
1283
1284 return ret;
1285}
1286EXPORT_SYMBOL(radix_tree_gang_lookup);
1287
1288/**
1289 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1290 * based on a tag
1291 * @root: radix tree root
1292 * @results: where the results of the lookup are placed
1293 * @first_index: start the lookup from this key
1294 * @max_items: place up to this many items at *results
1295 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1296 *
1297 * Performs an index-ascending scan of the tree for present items which
1298 * have the tag indexed by @tag set. Places the items at *@results and
1299 * returns the number of items which were placed at *@results.
1300 */
1301unsigned int
1302radix_tree_gang_lookup_tag(const struct radix_tree_root *root, void **results,
1303 unsigned long first_index, unsigned int max_items,
1304 unsigned int tag)
1305{
1306 struct radix_tree_iter iter;
1307 void __rcu **slot;
1308 unsigned int ret = 0;
1309
1310 if (unlikely(!max_items))
1311 return 0;
1312
1313 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1314 results[ret] = rcu_dereference_raw(*slot);
1315 if (!results[ret])
1316 continue;
1317 if (radix_tree_is_internal_node(results[ret])) {
1318 slot = radix_tree_iter_retry(&iter);
1319 continue;
1320 }
1321 if (++ret == max_items)
1322 break;
1323 }
1324
1325 return ret;
1326}
1327EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1328
1329/**
1330 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1331 * radix tree based on a tag
1332 * @root: radix tree root
1333 * @results: where the results of the lookup are placed
1334 * @first_index: start the lookup from this key
1335 * @max_items: place up to this many items at *results
1336 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1337 *
1338 * Performs an index-ascending scan of the tree for present items which
1339 * have the tag indexed by @tag set. Places the slots at *@results and
1340 * returns the number of slots which were placed at *@results.
1341 */
1342unsigned int
1343radix_tree_gang_lookup_tag_slot(const struct radix_tree_root *root,
1344 void __rcu ***results, unsigned long first_index,
1345 unsigned int max_items, unsigned int tag)
1346{
1347 struct radix_tree_iter iter;
1348 void __rcu **slot;
1349 unsigned int ret = 0;
1350
1351 if (unlikely(!max_items))
1352 return 0;
1353
1354 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1355 results[ret] = slot;
1356 if (++ret == max_items)
1357 break;
1358 }
1359
1360 return ret;
1361}
1362EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1363
1364static bool __radix_tree_delete(struct radix_tree_root *root,
1365 struct radix_tree_node *node, void __rcu **slot)
1366{
1367 void *old = rcu_dereference_raw(*slot);
1368 int values = xa_is_value(old) ? -1 : 0;
1369 unsigned offset = get_slot_offset(node, slot);
1370 int tag;
1371
1372 if (is_idr(root))
1373 node_tag_set(root, node, IDR_FREE, offset);
1374 else
1375 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++)
1376 node_tag_clear(root, node, tag, offset);
1377
1378 replace_slot(slot, NULL, node, -1, values);
1379 return node && delete_node(root, node);
1380}
1381
1382/**
1383 * radix_tree_iter_delete - delete the entry at this iterator position
1384 * @root: radix tree root
1385 * @iter: iterator state
1386 * @slot: pointer to slot
1387 *
1388 * Delete the entry at the position currently pointed to by the iterator.
1389 * This may result in the current node being freed; if it is, the iterator
1390 * is advanced so that it will not reference the freed memory. This
1391 * function may be called without any locking if there are no other threads
1392 * which can access this tree.
1393 */
1394void radix_tree_iter_delete(struct radix_tree_root *root,
1395 struct radix_tree_iter *iter, void __rcu **slot)
1396{
1397 if (__radix_tree_delete(root, iter->node, slot))
1398 iter->index = iter->next_index;
1399}
1400EXPORT_SYMBOL(radix_tree_iter_delete);
1401
1402/**
1403 * radix_tree_delete_item - delete an item from a radix tree
1404 * @root: radix tree root
1405 * @index: index key
1406 * @item: expected item
1407 *
1408 * Remove @item at @index from the radix tree rooted at @root.
1409 *
1410 * Return: the deleted entry, or %NULL if it was not present
1411 * or the entry at the given @index was not @item.
1412 */
1413void *radix_tree_delete_item(struct radix_tree_root *root,
1414 unsigned long index, void *item)
1415{
1416 struct radix_tree_node *node = NULL;
1417 void __rcu **slot = NULL;
1418 void *entry;
1419
1420 entry = __radix_tree_lookup(root, index, &node, &slot);
1421 if (!slot)
1422 return NULL;
1423 if (!entry && (!is_idr(root) || node_tag_get(root, node, IDR_FREE,
1424 get_slot_offset(node, slot))))
1425 return NULL;
1426
1427 if (item && entry != item)
1428 return NULL;
1429
1430 __radix_tree_delete(root, node, slot);
1431
1432 return entry;
1433}
1434EXPORT_SYMBOL(radix_tree_delete_item);
1435
1436/**
1437 * radix_tree_delete - delete an entry from a radix tree
1438 * @root: radix tree root
1439 * @index: index key
1440 *
1441 * Remove the entry at @index from the radix tree rooted at @root.
1442 *
1443 * Return: The deleted entry, or %NULL if it was not present.
1444 */
1445void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1446{
1447 return radix_tree_delete_item(root, index, NULL);
1448}
1449EXPORT_SYMBOL(radix_tree_delete);
1450
1451/**
1452 * radix_tree_tagged - test whether any items in the tree are tagged
1453 * @root: radix tree root
1454 * @tag: tag to test
1455 */
1456int radix_tree_tagged(const struct radix_tree_root *root, unsigned int tag)
1457{
1458 return root_tag_get(root, tag);
1459}
1460EXPORT_SYMBOL(radix_tree_tagged);
1461
1462/**
1463 * idr_preload - preload for idr_alloc()
1464 * @gfp_mask: allocation mask to use for preloading
1465 *
1466 * Preallocate memory to use for the next call to idr_alloc(). This function
1467 * returns with preemption disabled. It will be enabled by idr_preload_end().
1468 */
1469void idr_preload(gfp_t gfp_mask)
1470{
1471 if (__radix_tree_preload(gfp_mask, IDR_PRELOAD_SIZE))
1472 local_lock(&radix_tree_preloads.lock);
1473}
1474EXPORT_SYMBOL(idr_preload);
1475
1476void __rcu **idr_get_free(struct radix_tree_root *root,
1477 struct radix_tree_iter *iter, gfp_t gfp,
1478 unsigned long max)
1479{
1480 struct radix_tree_node *node = NULL, *child;
1481 void __rcu **slot = (void __rcu **)&root->xa_head;
1482 unsigned long maxindex, start = iter->next_index;
1483 unsigned int shift, offset = 0;
1484
1485 grow:
1486 shift = radix_tree_load_root(root, &child, &maxindex);
1487 if (!radix_tree_tagged(root, IDR_FREE))
1488 start = max(start, maxindex + 1);
1489 if (start > max)
1490 return ERR_PTR(-ENOSPC);
1491
1492 if (start > maxindex) {
1493 int error = radix_tree_extend(root, gfp, start, shift);
1494 if (error < 0)
1495 return ERR_PTR(error);
1496 shift = error;
1497 child = rcu_dereference_raw(root->xa_head);
1498 }
1499 if (start == 0 && shift == 0)
1500 shift = RADIX_TREE_MAP_SHIFT;
1501
1502 while (shift) {
1503 shift -= RADIX_TREE_MAP_SHIFT;
1504 if (child == NULL) {
1505 /* Have to add a child node. */
1506 child = radix_tree_node_alloc(gfp, node, root, shift,
1507 offset, 0, 0);
1508 if (!child)
1509 return ERR_PTR(-ENOMEM);
1510 all_tag_set(child, IDR_FREE);
1511 rcu_assign_pointer(*slot, node_to_entry(child));
1512 if (node)
1513 node->count++;
1514 } else if (!radix_tree_is_internal_node(child))
1515 break;
1516
1517 node = entry_to_node(child);
1518 offset = radix_tree_descend(node, &child, start);
1519 if (!tag_get(node, IDR_FREE, offset)) {
1520 offset = radix_tree_find_next_bit(node, IDR_FREE,
1521 offset + 1);
1522 start = next_index(start, node, offset);
1523 if (start > max || start == 0)
1524 return ERR_PTR(-ENOSPC);
1525 while (offset == RADIX_TREE_MAP_SIZE) {
1526 offset = node->offset + 1;
1527 node = node->parent;
1528 if (!node)
1529 goto grow;
1530 shift = node->shift;
1531 }
1532 child = rcu_dereference_raw(node->slots[offset]);
1533 }
1534 slot = &node->slots[offset];
1535 }
1536
1537 iter->index = start;
1538 if (node)
1539 iter->next_index = 1 + min(max, (start | node_maxindex(node)));
1540 else
1541 iter->next_index = 1;
1542 iter->node = node;
1543 set_iter_tags(iter, node, offset, IDR_FREE);
1544
1545 return slot;
1546}
1547
1548/**
1549 * idr_destroy - release all internal memory from an IDR
1550 * @idr: idr handle
1551 *
1552 * After this function is called, the IDR is empty, and may be reused or
1553 * the data structure containing it may be freed.
1554 *
1555 * A typical clean-up sequence for objects stored in an idr tree will use
1556 * idr_for_each() to free all objects, if necessary, then idr_destroy() to
1557 * free the memory used to keep track of those objects.
1558 */
1559void idr_destroy(struct idr *idr)
1560{
1561 struct radix_tree_node *node = rcu_dereference_raw(idr->idr_rt.xa_head);
1562 if (radix_tree_is_internal_node(node))
1563 radix_tree_free_nodes(node);
1564 idr->idr_rt.xa_head = NULL;
1565 root_tag_set(&idr->idr_rt, IDR_FREE);
1566}
1567EXPORT_SYMBOL(idr_destroy);
1568
1569static void
1570radix_tree_node_ctor(void *arg)
1571{
1572 struct radix_tree_node *node = arg;
1573
1574 memset(node, 0, sizeof(*node));
1575 INIT_LIST_HEAD(&node->private_list);
1576}
1577
1578static int radix_tree_cpu_dead(unsigned int cpu)
1579{
1580 struct radix_tree_preload *rtp;
1581 struct radix_tree_node *node;
1582
1583 /* Free per-cpu pool of preloaded nodes */
1584 rtp = &per_cpu(radix_tree_preloads, cpu);
1585 while (rtp->nr) {
1586 node = rtp->nodes;
1587 rtp->nodes = node->parent;
1588 kmem_cache_free(radix_tree_node_cachep, node);
1589 rtp->nr--;
1590 }
1591 return 0;
1592}
1593
1594void __init radix_tree_init(void)
1595{
1596 int ret;
1597
1598 BUILD_BUG_ON(RADIX_TREE_MAX_TAGS + __GFP_BITS_SHIFT > 32);
1599 BUILD_BUG_ON(ROOT_IS_IDR & ~GFP_ZONEMASK);
1600 BUILD_BUG_ON(XA_CHUNK_SIZE > 255);
1601 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1602 sizeof(struct radix_tree_node), 0,
1603 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1604 radix_tree_node_ctor);
1605 ret = cpuhp_setup_state_nocalls(CPUHP_RADIX_DEAD, "lib/radix:dead",
1606 NULL, radix_tree_cpu_dead);
1607 WARN_ON(ret < 0);
1608}