radix-tree: rename INDIRECT_PTR to INTERNAL_NODE
[deliverable/linux.git] / lib / radix-tree.c
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 * Copyright (C) 2016 Intel, Matthew Wilcox
8 * Copyright (C) 2016 Intel, Ross Zwisler
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License as
12 * published by the Free Software Foundation; either version 2, or (at
13 * your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 */
24
25 #include <linux/errno.h>
26 #include <linux/init.h>
27 #include <linux/kernel.h>
28 #include <linux/export.h>
29 #include <linux/radix-tree.h>
30 #include <linux/percpu.h>
31 #include <linux/slab.h>
32 #include <linux/kmemleak.h>
33 #include <linux/notifier.h>
34 #include <linux/cpu.h>
35 #include <linux/string.h>
36 #include <linux/bitops.h>
37 #include <linux/rcupdate.h>
38 #include <linux/preempt.h> /* in_interrupt() */
39
40
41 /*
42 * Radix tree node cache.
43 */
44 static struct kmem_cache *radix_tree_node_cachep;
45
46 /*
47 * The radix tree is variable-height, so an insert operation not only has
48 * to build the branch to its corresponding item, it also has to build the
49 * branch to existing items if the size has to be increased (by
50 * radix_tree_extend).
51 *
52 * The worst case is a zero height tree with just a single item at index 0,
53 * and then inserting an item at index ULONG_MAX. This requires 2 new branches
54 * of RADIX_TREE_MAX_PATH size to be created, with only the root node shared.
55 * Hence:
56 */
57 #define RADIX_TREE_PRELOAD_SIZE (RADIX_TREE_MAX_PATH * 2 - 1)
58
59 /*
60 * Per-cpu pool of preloaded nodes
61 */
62 struct radix_tree_preload {
63 unsigned nr;
64 /* nodes->private_data points to next preallocated node */
65 struct radix_tree_node *nodes;
66 };
67 static DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
68
69 static inline void *ptr_to_indirect(void *ptr)
70 {
71 return (void *)((unsigned long)ptr | RADIX_TREE_INTERNAL_NODE);
72 }
73
74 #define RADIX_TREE_RETRY ptr_to_indirect(NULL)
75
76 #ifdef CONFIG_RADIX_TREE_MULTIORDER
77 /* Sibling slots point directly to another slot in the same node */
78 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
79 {
80 void **ptr = node;
81 return (parent->slots <= ptr) &&
82 (ptr < parent->slots + RADIX_TREE_MAP_SIZE);
83 }
84 #else
85 static inline bool is_sibling_entry(struct radix_tree_node *parent, void *node)
86 {
87 return false;
88 }
89 #endif
90
91 static inline unsigned long get_slot_offset(struct radix_tree_node *parent,
92 void **slot)
93 {
94 return slot - parent->slots;
95 }
96
97 static unsigned radix_tree_descend(struct radix_tree_node *parent,
98 struct radix_tree_node **nodep, unsigned offset)
99 {
100 void **entry = rcu_dereference_raw(parent->slots[offset]);
101
102 #ifdef CONFIG_RADIX_TREE_MULTIORDER
103 if (radix_tree_is_indirect_ptr(entry)) {
104 unsigned long siboff = get_slot_offset(parent, entry);
105 if (siboff < RADIX_TREE_MAP_SIZE) {
106 offset = siboff;
107 entry = rcu_dereference_raw(parent->slots[offset]);
108 }
109 }
110 #endif
111
112 *nodep = (void *)entry;
113 return offset;
114 }
115
116 static inline gfp_t root_gfp_mask(struct radix_tree_root *root)
117 {
118 return root->gfp_mask & __GFP_BITS_MASK;
119 }
120
121 static inline void tag_set(struct radix_tree_node *node, unsigned int tag,
122 int offset)
123 {
124 __set_bit(offset, node->tags[tag]);
125 }
126
127 static inline void tag_clear(struct radix_tree_node *node, unsigned int tag,
128 int offset)
129 {
130 __clear_bit(offset, node->tags[tag]);
131 }
132
133 static inline int tag_get(struct radix_tree_node *node, unsigned int tag,
134 int offset)
135 {
136 return test_bit(offset, node->tags[tag]);
137 }
138
139 static inline void root_tag_set(struct radix_tree_root *root, unsigned int tag)
140 {
141 root->gfp_mask |= (__force gfp_t)(1 << (tag + __GFP_BITS_SHIFT));
142 }
143
144 static inline void root_tag_clear(struct radix_tree_root *root, unsigned tag)
145 {
146 root->gfp_mask &= (__force gfp_t)~(1 << (tag + __GFP_BITS_SHIFT));
147 }
148
149 static inline void root_tag_clear_all(struct radix_tree_root *root)
150 {
151 root->gfp_mask &= __GFP_BITS_MASK;
152 }
153
154 static inline int root_tag_get(struct radix_tree_root *root, unsigned int tag)
155 {
156 return (__force int)root->gfp_mask & (1 << (tag + __GFP_BITS_SHIFT));
157 }
158
159 static inline unsigned root_tags_get(struct radix_tree_root *root)
160 {
161 return (__force unsigned)root->gfp_mask >> __GFP_BITS_SHIFT;
162 }
163
164 /*
165 * Returns 1 if any slot in the node has this tag set.
166 * Otherwise returns 0.
167 */
168 static inline int any_tag_set(struct radix_tree_node *node, unsigned int tag)
169 {
170 unsigned idx;
171 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
172 if (node->tags[tag][idx])
173 return 1;
174 }
175 return 0;
176 }
177
178 /**
179 * radix_tree_find_next_bit - find the next set bit in a memory region
180 *
181 * @addr: The address to base the search on
182 * @size: The bitmap size in bits
183 * @offset: The bitnumber to start searching at
184 *
185 * Unrollable variant of find_next_bit() for constant size arrays.
186 * Tail bits starting from size to roundup(size, BITS_PER_LONG) must be zero.
187 * Returns next bit offset, or size if nothing found.
188 */
189 static __always_inline unsigned long
190 radix_tree_find_next_bit(const unsigned long *addr,
191 unsigned long size, unsigned long offset)
192 {
193 if (!__builtin_constant_p(size))
194 return find_next_bit(addr, size, offset);
195
196 if (offset < size) {
197 unsigned long tmp;
198
199 addr += offset / BITS_PER_LONG;
200 tmp = *addr >> (offset % BITS_PER_LONG);
201 if (tmp)
202 return __ffs(tmp) + offset;
203 offset = (offset + BITS_PER_LONG) & ~(BITS_PER_LONG - 1);
204 while (offset < size) {
205 tmp = *++addr;
206 if (tmp)
207 return __ffs(tmp) + offset;
208 offset += BITS_PER_LONG;
209 }
210 }
211 return size;
212 }
213
214 #ifndef __KERNEL__
215 static void dump_node(struct radix_tree_node *node, unsigned long index)
216 {
217 unsigned long i;
218
219 pr_debug("radix node: %p offset %d tags %lx %lx %lx shift %d count %d parent %p\n",
220 node, node->offset,
221 node->tags[0][0], node->tags[1][0], node->tags[2][0],
222 node->shift, node->count, node->parent);
223
224 for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
225 unsigned long first = index | (i << node->shift);
226 unsigned long last = first | ((1UL << node->shift) - 1);
227 void *entry = node->slots[i];
228 if (!entry)
229 continue;
230 if (is_sibling_entry(node, entry)) {
231 pr_debug("radix sblng %p offset %ld val %p indices %ld-%ld\n",
232 entry, i,
233 *(void **)indirect_to_ptr(entry),
234 first, last);
235 } else if (!radix_tree_is_indirect_ptr(entry)) {
236 pr_debug("radix entry %p offset %ld indices %ld-%ld\n",
237 entry, i, first, last);
238 } else {
239 dump_node(indirect_to_ptr(entry), first);
240 }
241 }
242 }
243
244 /* For debug */
245 static void radix_tree_dump(struct radix_tree_root *root)
246 {
247 pr_debug("radix root: %p rnode %p tags %x\n",
248 root, root->rnode,
249 root->gfp_mask >> __GFP_BITS_SHIFT);
250 if (!radix_tree_is_indirect_ptr(root->rnode))
251 return;
252 dump_node(indirect_to_ptr(root->rnode), 0);
253 }
254 #endif
255
256 /*
257 * This assumes that the caller has performed appropriate preallocation, and
258 * that the caller has pinned this thread of control to the current CPU.
259 */
260 static struct radix_tree_node *
261 radix_tree_node_alloc(struct radix_tree_root *root)
262 {
263 struct radix_tree_node *ret = NULL;
264 gfp_t gfp_mask = root_gfp_mask(root);
265
266 /*
267 * Preload code isn't irq safe and it doesn't make sense to use
268 * preloading during an interrupt anyway as all the allocations have
269 * to be atomic. So just do normal allocation when in interrupt.
270 */
271 if (!gfpflags_allow_blocking(gfp_mask) && !in_interrupt()) {
272 struct radix_tree_preload *rtp;
273
274 /*
275 * Even if the caller has preloaded, try to allocate from the
276 * cache first for the new node to get accounted.
277 */
278 ret = kmem_cache_alloc(radix_tree_node_cachep,
279 gfp_mask | __GFP_ACCOUNT | __GFP_NOWARN);
280 if (ret)
281 goto out;
282
283 /*
284 * Provided the caller has preloaded here, we will always
285 * succeed in getting a node here (and never reach
286 * kmem_cache_alloc)
287 */
288 rtp = this_cpu_ptr(&radix_tree_preloads);
289 if (rtp->nr) {
290 ret = rtp->nodes;
291 rtp->nodes = ret->private_data;
292 ret->private_data = NULL;
293 rtp->nr--;
294 }
295 /*
296 * Update the allocation stack trace as this is more useful
297 * for debugging.
298 */
299 kmemleak_update_trace(ret);
300 goto out;
301 }
302 ret = kmem_cache_alloc(radix_tree_node_cachep,
303 gfp_mask | __GFP_ACCOUNT);
304 out:
305 BUG_ON(radix_tree_is_indirect_ptr(ret));
306 return ret;
307 }
308
309 static void radix_tree_node_rcu_free(struct rcu_head *head)
310 {
311 struct radix_tree_node *node =
312 container_of(head, struct radix_tree_node, rcu_head);
313 int i;
314
315 /*
316 * must only free zeroed nodes into the slab. radix_tree_shrink
317 * can leave us with a non-NULL entry in the first slot, so clear
318 * that here to make sure.
319 */
320 for (i = 0; i < RADIX_TREE_MAX_TAGS; i++)
321 tag_clear(node, i, 0);
322
323 node->slots[0] = NULL;
324 node->count = 0;
325
326 kmem_cache_free(radix_tree_node_cachep, node);
327 }
328
329 static inline void
330 radix_tree_node_free(struct radix_tree_node *node)
331 {
332 call_rcu(&node->rcu_head, radix_tree_node_rcu_free);
333 }
334
335 /*
336 * Load up this CPU's radix_tree_node buffer with sufficient objects to
337 * ensure that the addition of a single element in the tree cannot fail. On
338 * success, return zero, with preemption disabled. On error, return -ENOMEM
339 * with preemption not disabled.
340 *
341 * To make use of this facility, the radix tree must be initialised without
342 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
343 */
344 static int __radix_tree_preload(gfp_t gfp_mask)
345 {
346 struct radix_tree_preload *rtp;
347 struct radix_tree_node *node;
348 int ret = -ENOMEM;
349
350 preempt_disable();
351 rtp = this_cpu_ptr(&radix_tree_preloads);
352 while (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
353 preempt_enable();
354 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
355 if (node == NULL)
356 goto out;
357 preempt_disable();
358 rtp = this_cpu_ptr(&radix_tree_preloads);
359 if (rtp->nr < RADIX_TREE_PRELOAD_SIZE) {
360 node->private_data = rtp->nodes;
361 rtp->nodes = node;
362 rtp->nr++;
363 } else {
364 kmem_cache_free(radix_tree_node_cachep, node);
365 }
366 }
367 ret = 0;
368 out:
369 return ret;
370 }
371
372 /*
373 * Load up this CPU's radix_tree_node buffer with sufficient objects to
374 * ensure that the addition of a single element in the tree cannot fail. On
375 * success, return zero, with preemption disabled. On error, return -ENOMEM
376 * with preemption not disabled.
377 *
378 * To make use of this facility, the radix tree must be initialised without
379 * __GFP_DIRECT_RECLAIM being passed to INIT_RADIX_TREE().
380 */
381 int radix_tree_preload(gfp_t gfp_mask)
382 {
383 /* Warn on non-sensical use... */
384 WARN_ON_ONCE(!gfpflags_allow_blocking(gfp_mask));
385 return __radix_tree_preload(gfp_mask);
386 }
387 EXPORT_SYMBOL(radix_tree_preload);
388
389 /*
390 * The same as above function, except we don't guarantee preloading happens.
391 * We do it, if we decide it helps. On success, return zero with preemption
392 * disabled. On error, return -ENOMEM with preemption not disabled.
393 */
394 int radix_tree_maybe_preload(gfp_t gfp_mask)
395 {
396 if (gfpflags_allow_blocking(gfp_mask))
397 return __radix_tree_preload(gfp_mask);
398 /* Preloading doesn't help anything with this gfp mask, skip it */
399 preempt_disable();
400 return 0;
401 }
402 EXPORT_SYMBOL(radix_tree_maybe_preload);
403
404 /*
405 * The maximum index which can be stored in a radix tree
406 */
407 static inline unsigned long shift_maxindex(unsigned int shift)
408 {
409 return (RADIX_TREE_MAP_SIZE << shift) - 1;
410 }
411
412 static inline unsigned long node_maxindex(struct radix_tree_node *node)
413 {
414 return shift_maxindex(node->shift);
415 }
416
417 static unsigned radix_tree_load_root(struct radix_tree_root *root,
418 struct radix_tree_node **nodep, unsigned long *maxindex)
419 {
420 struct radix_tree_node *node = rcu_dereference_raw(root->rnode);
421
422 *nodep = node;
423
424 if (likely(radix_tree_is_indirect_ptr(node))) {
425 node = indirect_to_ptr(node);
426 *maxindex = node_maxindex(node);
427 return node->shift + RADIX_TREE_MAP_SHIFT;
428 }
429
430 *maxindex = 0;
431 return 0;
432 }
433
434 /*
435 * Extend a radix tree so it can store key @index.
436 */
437 static int radix_tree_extend(struct radix_tree_root *root,
438 unsigned long index, unsigned int shift)
439 {
440 struct radix_tree_node *slot;
441 unsigned int maxshift;
442 int tag;
443
444 /* Figure out what the shift should be. */
445 maxshift = shift;
446 while (index > shift_maxindex(maxshift))
447 maxshift += RADIX_TREE_MAP_SHIFT;
448
449 slot = root->rnode;
450 if (!slot)
451 goto out;
452
453 do {
454 struct radix_tree_node *node = radix_tree_node_alloc(root);
455
456 if (!node)
457 return -ENOMEM;
458
459 /* Propagate the aggregated tag info into the new root */
460 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
461 if (root_tag_get(root, tag))
462 tag_set(node, tag, 0);
463 }
464
465 BUG_ON(shift > BITS_PER_LONG);
466 node->shift = shift;
467 node->offset = 0;
468 node->count = 1;
469 node->parent = NULL;
470 if (radix_tree_is_indirect_ptr(slot)) {
471 slot = indirect_to_ptr(slot);
472 slot->parent = node;
473 slot = ptr_to_indirect(slot);
474 }
475 node->slots[0] = slot;
476 node = ptr_to_indirect(node);
477 rcu_assign_pointer(root->rnode, node);
478 shift += RADIX_TREE_MAP_SHIFT;
479 slot = node;
480 } while (shift <= maxshift);
481 out:
482 return maxshift + RADIX_TREE_MAP_SHIFT;
483 }
484
485 /**
486 * __radix_tree_create - create a slot in a radix tree
487 * @root: radix tree root
488 * @index: index key
489 * @order: index occupies 2^order aligned slots
490 * @nodep: returns node
491 * @slotp: returns slot
492 *
493 * Create, if necessary, and return the node and slot for an item
494 * at position @index in the radix tree @root.
495 *
496 * Until there is more than one item in the tree, no nodes are
497 * allocated and @root->rnode is used as a direct slot instead of
498 * pointing to a node, in which case *@nodep will be NULL.
499 *
500 * Returns -ENOMEM, or 0 for success.
501 */
502 int __radix_tree_create(struct radix_tree_root *root, unsigned long index,
503 unsigned order, struct radix_tree_node **nodep,
504 void ***slotp)
505 {
506 struct radix_tree_node *node = NULL, *slot;
507 unsigned long maxindex;
508 unsigned int shift, offset;
509 unsigned long max = index | ((1UL << order) - 1);
510
511 shift = radix_tree_load_root(root, &slot, &maxindex);
512
513 /* Make sure the tree is high enough. */
514 if (max > maxindex) {
515 int error = radix_tree_extend(root, max, shift);
516 if (error < 0)
517 return error;
518 shift = error;
519 slot = root->rnode;
520 if (order == shift)
521 shift += RADIX_TREE_MAP_SHIFT;
522 }
523
524 offset = 0; /* uninitialised var warning */
525 while (shift > order) {
526 shift -= RADIX_TREE_MAP_SHIFT;
527 if (slot == NULL) {
528 /* Have to add a child node. */
529 slot = radix_tree_node_alloc(root);
530 if (!slot)
531 return -ENOMEM;
532 slot->shift = shift;
533 slot->offset = offset;
534 slot->parent = node;
535 if (node) {
536 rcu_assign_pointer(node->slots[offset],
537 ptr_to_indirect(slot));
538 node->count++;
539 } else
540 rcu_assign_pointer(root->rnode,
541 ptr_to_indirect(slot));
542 } else if (!radix_tree_is_indirect_ptr(slot))
543 break;
544
545 /* Go a level down */
546 node = indirect_to_ptr(slot);
547 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
548 offset = radix_tree_descend(node, &slot, offset);
549 }
550
551 #ifdef CONFIG_RADIX_TREE_MULTIORDER
552 /* Insert pointers to the canonical entry */
553 if (order > shift) {
554 int i, n = 1 << (order - shift);
555 offset = offset & ~(n - 1);
556 slot = ptr_to_indirect(&node->slots[offset]);
557 for (i = 0; i < n; i++) {
558 if (node->slots[offset + i])
559 return -EEXIST;
560 }
561
562 for (i = 1; i < n; i++) {
563 rcu_assign_pointer(node->slots[offset + i], slot);
564 node->count++;
565 }
566 }
567 #endif
568
569 if (nodep)
570 *nodep = node;
571 if (slotp)
572 *slotp = node ? node->slots + offset : (void **)&root->rnode;
573 return 0;
574 }
575
576 /**
577 * __radix_tree_insert - insert into a radix tree
578 * @root: radix tree root
579 * @index: index key
580 * @order: key covers the 2^order indices around index
581 * @item: item to insert
582 *
583 * Insert an item into the radix tree at position @index.
584 */
585 int __radix_tree_insert(struct radix_tree_root *root, unsigned long index,
586 unsigned order, void *item)
587 {
588 struct radix_tree_node *node;
589 void **slot;
590 int error;
591
592 BUG_ON(radix_tree_is_indirect_ptr(item));
593
594 error = __radix_tree_create(root, index, order, &node, &slot);
595 if (error)
596 return error;
597 if (*slot != NULL)
598 return -EEXIST;
599 rcu_assign_pointer(*slot, item);
600
601 if (node) {
602 unsigned offset = get_slot_offset(node, slot);
603 node->count++;
604 BUG_ON(tag_get(node, 0, offset));
605 BUG_ON(tag_get(node, 1, offset));
606 BUG_ON(tag_get(node, 2, offset));
607 } else {
608 BUG_ON(root_tags_get(root));
609 }
610
611 return 0;
612 }
613 EXPORT_SYMBOL(__radix_tree_insert);
614
615 /**
616 * __radix_tree_lookup - lookup an item in a radix tree
617 * @root: radix tree root
618 * @index: index key
619 * @nodep: returns node
620 * @slotp: returns slot
621 *
622 * Lookup and return the item at position @index in the radix
623 * tree @root.
624 *
625 * Until there is more than one item in the tree, no nodes are
626 * allocated and @root->rnode is used as a direct slot instead of
627 * pointing to a node, in which case *@nodep will be NULL.
628 */
629 void *__radix_tree_lookup(struct radix_tree_root *root, unsigned long index,
630 struct radix_tree_node **nodep, void ***slotp)
631 {
632 struct radix_tree_node *node, *parent;
633 unsigned long maxindex;
634 unsigned int shift;
635 void **slot;
636
637 restart:
638 parent = NULL;
639 slot = (void **)&root->rnode;
640 shift = radix_tree_load_root(root, &node, &maxindex);
641 if (index > maxindex)
642 return NULL;
643
644 while (radix_tree_is_indirect_ptr(node)) {
645 unsigned offset;
646
647 if (node == RADIX_TREE_RETRY)
648 goto restart;
649 parent = indirect_to_ptr(node);
650 shift -= RADIX_TREE_MAP_SHIFT;
651 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
652 offset = radix_tree_descend(parent, &node, offset);
653 slot = parent->slots + offset;
654 }
655
656 if (nodep)
657 *nodep = parent;
658 if (slotp)
659 *slotp = slot;
660 return node;
661 }
662
663 /**
664 * radix_tree_lookup_slot - lookup a slot in a radix tree
665 * @root: radix tree root
666 * @index: index key
667 *
668 * Returns: the slot corresponding to the position @index in the
669 * radix tree @root. This is useful for update-if-exists operations.
670 *
671 * This function can be called under rcu_read_lock iff the slot is not
672 * modified by radix_tree_replace_slot, otherwise it must be called
673 * exclusive from other writers. Any dereference of the slot must be done
674 * using radix_tree_deref_slot.
675 */
676 void **radix_tree_lookup_slot(struct radix_tree_root *root, unsigned long index)
677 {
678 void **slot;
679
680 if (!__radix_tree_lookup(root, index, NULL, &slot))
681 return NULL;
682 return slot;
683 }
684 EXPORT_SYMBOL(radix_tree_lookup_slot);
685
686 /**
687 * radix_tree_lookup - perform lookup operation on a radix tree
688 * @root: radix tree root
689 * @index: index key
690 *
691 * Lookup the item at the position @index in the radix tree @root.
692 *
693 * This function can be called under rcu_read_lock, however the caller
694 * must manage lifetimes of leaf nodes (eg. RCU may also be used to free
695 * them safely). No RCU barriers are required to access or modify the
696 * returned item, however.
697 */
698 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
699 {
700 return __radix_tree_lookup(root, index, NULL, NULL);
701 }
702 EXPORT_SYMBOL(radix_tree_lookup);
703
704 /**
705 * radix_tree_tag_set - set a tag on a radix tree node
706 * @root: radix tree root
707 * @index: index key
708 * @tag: tag index
709 *
710 * Set the search tag (which must be < RADIX_TREE_MAX_TAGS)
711 * corresponding to @index in the radix tree. From
712 * the root all the way down to the leaf node.
713 *
714 * Returns the address of the tagged item. Setting a tag on a not-present
715 * item is a bug.
716 */
717 void *radix_tree_tag_set(struct radix_tree_root *root,
718 unsigned long index, unsigned int tag)
719 {
720 struct radix_tree_node *node, *parent;
721 unsigned long maxindex;
722 unsigned int shift;
723
724 shift = radix_tree_load_root(root, &node, &maxindex);
725 BUG_ON(index > maxindex);
726
727 while (radix_tree_is_indirect_ptr(node)) {
728 unsigned offset;
729
730 shift -= RADIX_TREE_MAP_SHIFT;
731 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
732
733 parent = indirect_to_ptr(node);
734 offset = radix_tree_descend(parent, &node, offset);
735 BUG_ON(!node);
736
737 if (!tag_get(parent, tag, offset))
738 tag_set(parent, tag, offset);
739 }
740
741 /* set the root's tag bit */
742 if (!root_tag_get(root, tag))
743 root_tag_set(root, tag);
744
745 return node;
746 }
747 EXPORT_SYMBOL(radix_tree_tag_set);
748
749 /**
750 * radix_tree_tag_clear - clear a tag on a radix tree node
751 * @root: radix tree root
752 * @index: index key
753 * @tag: tag index
754 *
755 * Clear the search tag (which must be < RADIX_TREE_MAX_TAGS)
756 * corresponding to @index in the radix tree. If this causes
757 * the leaf node to have no tags set then clear the tag in the
758 * next-to-leaf node, etc.
759 *
760 * Returns the address of the tagged item on success, else NULL. ie:
761 * has the same return value and semantics as radix_tree_lookup().
762 */
763 void *radix_tree_tag_clear(struct radix_tree_root *root,
764 unsigned long index, unsigned int tag)
765 {
766 struct radix_tree_node *node, *parent;
767 unsigned long maxindex;
768 unsigned int shift;
769 int uninitialized_var(offset);
770
771 shift = radix_tree_load_root(root, &node, &maxindex);
772 if (index > maxindex)
773 return NULL;
774
775 parent = NULL;
776
777 while (radix_tree_is_indirect_ptr(node)) {
778 shift -= RADIX_TREE_MAP_SHIFT;
779 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
780
781 parent = indirect_to_ptr(node);
782 offset = radix_tree_descend(parent, &node, offset);
783 }
784
785 if (node == NULL)
786 goto out;
787
788 index >>= shift;
789
790 while (parent) {
791 if (!tag_get(parent, tag, offset))
792 goto out;
793 tag_clear(parent, tag, offset);
794 if (any_tag_set(parent, tag))
795 goto out;
796
797 index >>= RADIX_TREE_MAP_SHIFT;
798 offset = index & RADIX_TREE_MAP_MASK;
799 parent = parent->parent;
800 }
801
802 /* clear the root's tag bit */
803 if (root_tag_get(root, tag))
804 root_tag_clear(root, tag);
805
806 out:
807 return node;
808 }
809 EXPORT_SYMBOL(radix_tree_tag_clear);
810
811 /**
812 * radix_tree_tag_get - get a tag on a radix tree node
813 * @root: radix tree root
814 * @index: index key
815 * @tag: tag index (< RADIX_TREE_MAX_TAGS)
816 *
817 * Return values:
818 *
819 * 0: tag not present or not set
820 * 1: tag set
821 *
822 * Note that the return value of this function may not be relied on, even if
823 * the RCU lock is held, unless tag modification and node deletion are excluded
824 * from concurrency.
825 */
826 int radix_tree_tag_get(struct radix_tree_root *root,
827 unsigned long index, unsigned int tag)
828 {
829 struct radix_tree_node *node, *parent;
830 unsigned long maxindex;
831 unsigned int shift;
832
833 if (!root_tag_get(root, tag))
834 return 0;
835
836 shift = radix_tree_load_root(root, &node, &maxindex);
837 if (index > maxindex)
838 return 0;
839 if (node == NULL)
840 return 0;
841
842 while (radix_tree_is_indirect_ptr(node)) {
843 int offset;
844
845 shift -= RADIX_TREE_MAP_SHIFT;
846 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
847
848 parent = indirect_to_ptr(node);
849 offset = radix_tree_descend(parent, &node, offset);
850
851 if (!node)
852 return 0;
853 if (!tag_get(parent, tag, offset))
854 return 0;
855 if (node == RADIX_TREE_RETRY)
856 break;
857 }
858
859 return 1;
860 }
861 EXPORT_SYMBOL(radix_tree_tag_get);
862
863 static inline void __set_iter_shift(struct radix_tree_iter *iter,
864 unsigned int shift)
865 {
866 #ifdef CONFIG_RADIX_TREE_MULTIORDER
867 iter->shift = shift;
868 #endif
869 }
870
871 /**
872 * radix_tree_next_chunk - find next chunk of slots for iteration
873 *
874 * @root: radix tree root
875 * @iter: iterator state
876 * @flags: RADIX_TREE_ITER_* flags and tag index
877 * Returns: pointer to chunk first slot, or NULL if iteration is over
878 */
879 void **radix_tree_next_chunk(struct radix_tree_root *root,
880 struct radix_tree_iter *iter, unsigned flags)
881 {
882 unsigned shift, tag = flags & RADIX_TREE_ITER_TAG_MASK;
883 struct radix_tree_node *rnode, *node;
884 unsigned long index, offset, maxindex;
885
886 if ((flags & RADIX_TREE_ITER_TAGGED) && !root_tag_get(root, tag))
887 return NULL;
888
889 /*
890 * Catch next_index overflow after ~0UL. iter->index never overflows
891 * during iterating; it can be zero only at the beginning.
892 * And we cannot overflow iter->next_index in a single step,
893 * because RADIX_TREE_MAP_SHIFT < BITS_PER_LONG.
894 *
895 * This condition also used by radix_tree_next_slot() to stop
896 * contiguous iterating, and forbid swithing to the next chunk.
897 */
898 index = iter->next_index;
899 if (!index && iter->index)
900 return NULL;
901
902 restart:
903 shift = radix_tree_load_root(root, &rnode, &maxindex);
904 if (index > maxindex)
905 return NULL;
906
907 if (radix_tree_is_indirect_ptr(rnode)) {
908 rnode = indirect_to_ptr(rnode);
909 } else if (rnode) {
910 /* Single-slot tree */
911 iter->index = index;
912 iter->next_index = maxindex + 1;
913 iter->tags = 1;
914 __set_iter_shift(iter, shift);
915 return (void **)&root->rnode;
916 } else
917 return NULL;
918
919 shift -= RADIX_TREE_MAP_SHIFT;
920 offset = index >> shift;
921
922 node = rnode;
923 while (1) {
924 struct radix_tree_node *slot;
925 unsigned new_off = radix_tree_descend(node, &slot, offset);
926
927 if (new_off < offset) {
928 offset = new_off;
929 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
930 index |= offset << shift;
931 }
932
933 if ((flags & RADIX_TREE_ITER_TAGGED) ?
934 !tag_get(node, tag, offset) : !slot) {
935 /* Hole detected */
936 if (flags & RADIX_TREE_ITER_CONTIG)
937 return NULL;
938
939 if (flags & RADIX_TREE_ITER_TAGGED)
940 offset = radix_tree_find_next_bit(
941 node->tags[tag],
942 RADIX_TREE_MAP_SIZE,
943 offset + 1);
944 else
945 while (++offset < RADIX_TREE_MAP_SIZE) {
946 void *slot = node->slots[offset];
947 if (is_sibling_entry(node, slot))
948 continue;
949 if (slot)
950 break;
951 }
952 index &= ~((RADIX_TREE_MAP_SIZE << shift) - 1);
953 index += offset << shift;
954 /* Overflow after ~0UL */
955 if (!index)
956 return NULL;
957 if (offset == RADIX_TREE_MAP_SIZE)
958 goto restart;
959 slot = rcu_dereference_raw(node->slots[offset]);
960 }
961
962 if ((slot == NULL) || (slot == RADIX_TREE_RETRY))
963 goto restart;
964 if (!radix_tree_is_indirect_ptr(slot))
965 break;
966
967 node = indirect_to_ptr(slot);
968 shift -= RADIX_TREE_MAP_SHIFT;
969 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
970 }
971
972 /* Update the iterator state */
973 iter->index = index & ~((1 << shift) - 1);
974 iter->next_index = (index | ((RADIX_TREE_MAP_SIZE << shift) - 1)) + 1;
975 __set_iter_shift(iter, shift);
976
977 /* Construct iter->tags bit-mask from node->tags[tag] array */
978 if (flags & RADIX_TREE_ITER_TAGGED) {
979 unsigned tag_long, tag_bit;
980
981 tag_long = offset / BITS_PER_LONG;
982 tag_bit = offset % BITS_PER_LONG;
983 iter->tags = node->tags[tag][tag_long] >> tag_bit;
984 /* This never happens if RADIX_TREE_TAG_LONGS == 1 */
985 if (tag_long < RADIX_TREE_TAG_LONGS - 1) {
986 /* Pick tags from next element */
987 if (tag_bit)
988 iter->tags |= node->tags[tag][tag_long + 1] <<
989 (BITS_PER_LONG - tag_bit);
990 /* Clip chunk size, here only BITS_PER_LONG tags */
991 iter->next_index = index + BITS_PER_LONG;
992 }
993 }
994
995 return node->slots + offset;
996 }
997 EXPORT_SYMBOL(radix_tree_next_chunk);
998
999 /**
1000 * radix_tree_range_tag_if_tagged - for each item in given range set given
1001 * tag if item has another tag set
1002 * @root: radix tree root
1003 * @first_indexp: pointer to a starting index of a range to scan
1004 * @last_index: last index of a range to scan
1005 * @nr_to_tag: maximum number items to tag
1006 * @iftag: tag index to test
1007 * @settag: tag index to set if tested tag is set
1008 *
1009 * This function scans range of radix tree from first_index to last_index
1010 * (inclusive). For each item in the range if iftag is set, the function sets
1011 * also settag. The function stops either after tagging nr_to_tag items or
1012 * after reaching last_index.
1013 *
1014 * The tags must be set from the leaf level only and propagated back up the
1015 * path to the root. We must do this so that we resolve the full path before
1016 * setting any tags on intermediate nodes. If we set tags as we descend, then
1017 * we can get to the leaf node and find that the index that has the iftag
1018 * set is outside the range we are scanning. This reults in dangling tags and
1019 * can lead to problems with later tag operations (e.g. livelocks on lookups).
1020 *
1021 * The function returns the number of leaves where the tag was set and sets
1022 * *first_indexp to the first unscanned index.
1023 * WARNING! *first_indexp can wrap if last_index is ULONG_MAX. Caller must
1024 * be prepared to handle that.
1025 */
1026 unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
1027 unsigned long *first_indexp, unsigned long last_index,
1028 unsigned long nr_to_tag,
1029 unsigned int iftag, unsigned int settag)
1030 {
1031 struct radix_tree_node *slot, *node = NULL;
1032 unsigned long maxindex;
1033 unsigned int shift = radix_tree_load_root(root, &slot, &maxindex);
1034 unsigned long tagged = 0;
1035 unsigned long index = *first_indexp;
1036
1037 last_index = min(last_index, maxindex);
1038 if (index > last_index)
1039 return 0;
1040 if (!nr_to_tag)
1041 return 0;
1042 if (!root_tag_get(root, iftag)) {
1043 *first_indexp = last_index + 1;
1044 return 0;
1045 }
1046 if (!radix_tree_is_indirect_ptr(slot)) {
1047 *first_indexp = last_index + 1;
1048 root_tag_set(root, settag);
1049 return 1;
1050 }
1051
1052 node = indirect_to_ptr(slot);
1053 shift -= RADIX_TREE_MAP_SHIFT;
1054
1055 for (;;) {
1056 unsigned long upindex;
1057 unsigned offset;
1058
1059 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1060 offset = radix_tree_descend(node, &slot, offset);
1061 if (!slot)
1062 goto next;
1063 if (!tag_get(node, iftag, offset))
1064 goto next;
1065 /* Sibling slots never have tags set on them */
1066 if (radix_tree_is_indirect_ptr(slot)) {
1067 node = indirect_to_ptr(slot);
1068 shift -= RADIX_TREE_MAP_SHIFT;
1069 continue;
1070 }
1071
1072 /* tag the leaf */
1073 tagged++;
1074 tag_set(node, settag, offset);
1075
1076 slot = node->parent;
1077 /* walk back up the path tagging interior nodes */
1078 upindex = index >> shift;
1079 while (slot) {
1080 upindex >>= RADIX_TREE_MAP_SHIFT;
1081 offset = upindex & RADIX_TREE_MAP_MASK;
1082
1083 /* stop if we find a node with the tag already set */
1084 if (tag_get(slot, settag, offset))
1085 break;
1086 tag_set(slot, settag, offset);
1087 slot = slot->parent;
1088 }
1089
1090 next:
1091 /* Go to next item at level determined by 'shift' */
1092 index = ((index >> shift) + 1) << shift;
1093 /* Overflow can happen when last_index is ~0UL... */
1094 if (index > last_index || !index)
1095 break;
1096 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1097 while (offset == 0) {
1098 /*
1099 * We've fully scanned this node. Go up. Because
1100 * last_index is guaranteed to be in the tree, what
1101 * we do below cannot wander astray.
1102 */
1103 node = node->parent;
1104 shift += RADIX_TREE_MAP_SHIFT;
1105 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
1106 }
1107 if (is_sibling_entry(node, node->slots[offset]))
1108 goto next;
1109 if (tagged >= nr_to_tag)
1110 break;
1111 }
1112 /*
1113 * We need not to tag the root tag if there is no tag which is set with
1114 * settag within the range from *first_indexp to last_index.
1115 */
1116 if (tagged > 0)
1117 root_tag_set(root, settag);
1118 *first_indexp = index;
1119
1120 return tagged;
1121 }
1122 EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
1123
1124 /**
1125 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
1126 * @root: radix tree root
1127 * @results: where the results of the lookup are placed
1128 * @first_index: start the lookup from this key
1129 * @max_items: place up to this many items at *results
1130 *
1131 * Performs an index-ascending scan of the tree for present items. Places
1132 * them at *@results and returns the number of items which were placed at
1133 * *@results.
1134 *
1135 * The implementation is naive.
1136 *
1137 * Like radix_tree_lookup, radix_tree_gang_lookup may be called under
1138 * rcu_read_lock. In this case, rather than the returned results being
1139 * an atomic snapshot of the tree at a single point in time, the
1140 * semantics of an RCU protected gang lookup are as though multiple
1141 * radix_tree_lookups have been issued in individual locks, and results
1142 * stored in 'results'.
1143 */
1144 unsigned int
1145 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
1146 unsigned long first_index, unsigned int max_items)
1147 {
1148 struct radix_tree_iter iter;
1149 void **slot;
1150 unsigned int ret = 0;
1151
1152 if (unlikely(!max_items))
1153 return 0;
1154
1155 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1156 results[ret] = rcu_dereference_raw(*slot);
1157 if (!results[ret])
1158 continue;
1159 if (radix_tree_is_indirect_ptr(results[ret])) {
1160 slot = radix_tree_iter_retry(&iter);
1161 continue;
1162 }
1163 if (++ret == max_items)
1164 break;
1165 }
1166
1167 return ret;
1168 }
1169 EXPORT_SYMBOL(radix_tree_gang_lookup);
1170
1171 /**
1172 * radix_tree_gang_lookup_slot - perform multiple slot lookup on radix tree
1173 * @root: radix tree root
1174 * @results: where the results of the lookup are placed
1175 * @indices: where their indices should be placed (but usually NULL)
1176 * @first_index: start the lookup from this key
1177 * @max_items: place up to this many items at *results
1178 *
1179 * Performs an index-ascending scan of the tree for present items. Places
1180 * their slots at *@results and returns the number of items which were
1181 * placed at *@results.
1182 *
1183 * The implementation is naive.
1184 *
1185 * Like radix_tree_gang_lookup as far as RCU and locking goes. Slots must
1186 * be dereferenced with radix_tree_deref_slot, and if using only RCU
1187 * protection, radix_tree_deref_slot may fail requiring a retry.
1188 */
1189 unsigned int
1190 radix_tree_gang_lookup_slot(struct radix_tree_root *root,
1191 void ***results, unsigned long *indices,
1192 unsigned long first_index, unsigned int max_items)
1193 {
1194 struct radix_tree_iter iter;
1195 void **slot;
1196 unsigned int ret = 0;
1197
1198 if (unlikely(!max_items))
1199 return 0;
1200
1201 radix_tree_for_each_slot(slot, root, &iter, first_index) {
1202 results[ret] = slot;
1203 if (indices)
1204 indices[ret] = iter.index;
1205 if (++ret == max_items)
1206 break;
1207 }
1208
1209 return ret;
1210 }
1211 EXPORT_SYMBOL(radix_tree_gang_lookup_slot);
1212
1213 /**
1214 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
1215 * based on a tag
1216 * @root: radix tree root
1217 * @results: where the results of the lookup are placed
1218 * @first_index: start the lookup from this key
1219 * @max_items: place up to this many items at *results
1220 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1221 *
1222 * Performs an index-ascending scan of the tree for present items which
1223 * have the tag indexed by @tag set. Places the items at *@results and
1224 * returns the number of items which were placed at *@results.
1225 */
1226 unsigned int
1227 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
1228 unsigned long first_index, unsigned int max_items,
1229 unsigned int tag)
1230 {
1231 struct radix_tree_iter iter;
1232 void **slot;
1233 unsigned int ret = 0;
1234
1235 if (unlikely(!max_items))
1236 return 0;
1237
1238 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1239 results[ret] = rcu_dereference_raw(*slot);
1240 if (!results[ret])
1241 continue;
1242 if (radix_tree_is_indirect_ptr(results[ret])) {
1243 slot = radix_tree_iter_retry(&iter);
1244 continue;
1245 }
1246 if (++ret == max_items)
1247 break;
1248 }
1249
1250 return ret;
1251 }
1252 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
1253
1254 /**
1255 * radix_tree_gang_lookup_tag_slot - perform multiple slot lookup on a
1256 * radix tree based on a tag
1257 * @root: radix tree root
1258 * @results: where the results of the lookup are placed
1259 * @first_index: start the lookup from this key
1260 * @max_items: place up to this many items at *results
1261 * @tag: the tag index (< RADIX_TREE_MAX_TAGS)
1262 *
1263 * Performs an index-ascending scan of the tree for present items which
1264 * have the tag indexed by @tag set. Places the slots at *@results and
1265 * returns the number of slots which were placed at *@results.
1266 */
1267 unsigned int
1268 radix_tree_gang_lookup_tag_slot(struct radix_tree_root *root, void ***results,
1269 unsigned long first_index, unsigned int max_items,
1270 unsigned int tag)
1271 {
1272 struct radix_tree_iter iter;
1273 void **slot;
1274 unsigned int ret = 0;
1275
1276 if (unlikely(!max_items))
1277 return 0;
1278
1279 radix_tree_for_each_tagged(slot, root, &iter, first_index, tag) {
1280 results[ret] = slot;
1281 if (++ret == max_items)
1282 break;
1283 }
1284
1285 return ret;
1286 }
1287 EXPORT_SYMBOL(radix_tree_gang_lookup_tag_slot);
1288
1289 #if defined(CONFIG_SHMEM) && defined(CONFIG_SWAP)
1290 #include <linux/sched.h> /* for cond_resched() */
1291
1292 struct locate_info {
1293 unsigned long found_index;
1294 bool stop;
1295 };
1296
1297 /*
1298 * This linear search is at present only useful to shmem_unuse_inode().
1299 */
1300 static unsigned long __locate(struct radix_tree_node *slot, void *item,
1301 unsigned long index, struct locate_info *info)
1302 {
1303 unsigned int shift;
1304 unsigned long i;
1305
1306 shift = slot->shift + RADIX_TREE_MAP_SHIFT;
1307
1308 do {
1309 shift -= RADIX_TREE_MAP_SHIFT;
1310
1311 for (i = (index >> shift) & RADIX_TREE_MAP_MASK;
1312 i < RADIX_TREE_MAP_SIZE;
1313 i++, index += (1UL << shift)) {
1314 struct radix_tree_node *node =
1315 rcu_dereference_raw(slot->slots[i]);
1316 if (node == RADIX_TREE_RETRY)
1317 goto out;
1318 if (!radix_tree_is_indirect_ptr(node)) {
1319 if (node == item) {
1320 info->found_index = index;
1321 info->stop = true;
1322 goto out;
1323 }
1324 continue;
1325 }
1326 node = indirect_to_ptr(node);
1327 if (is_sibling_entry(slot, node))
1328 continue;
1329 slot = node;
1330 break;
1331 }
1332 if (i == RADIX_TREE_MAP_SIZE)
1333 break;
1334 } while (shift);
1335
1336 out:
1337 if ((index == 0) && (i == RADIX_TREE_MAP_SIZE))
1338 info->stop = true;
1339 return index;
1340 }
1341
1342 /**
1343 * radix_tree_locate_item - search through radix tree for item
1344 * @root: radix tree root
1345 * @item: item to be found
1346 *
1347 * Returns index where item was found, or -1 if not found.
1348 * Caller must hold no lock (since this time-consuming function needs
1349 * to be preemptible), and must check afterwards if item is still there.
1350 */
1351 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1352 {
1353 struct radix_tree_node *node;
1354 unsigned long max_index;
1355 unsigned long cur_index = 0;
1356 struct locate_info info = {
1357 .found_index = -1,
1358 .stop = false,
1359 };
1360
1361 do {
1362 rcu_read_lock();
1363 node = rcu_dereference_raw(root->rnode);
1364 if (!radix_tree_is_indirect_ptr(node)) {
1365 rcu_read_unlock();
1366 if (node == item)
1367 info.found_index = 0;
1368 break;
1369 }
1370
1371 node = indirect_to_ptr(node);
1372
1373 max_index = node_maxindex(node);
1374 if (cur_index > max_index) {
1375 rcu_read_unlock();
1376 break;
1377 }
1378
1379 cur_index = __locate(node, item, cur_index, &info);
1380 rcu_read_unlock();
1381 cond_resched();
1382 } while (!info.stop && cur_index <= max_index);
1383
1384 return info.found_index;
1385 }
1386 #else
1387 unsigned long radix_tree_locate_item(struct radix_tree_root *root, void *item)
1388 {
1389 return -1;
1390 }
1391 #endif /* CONFIG_SHMEM && CONFIG_SWAP */
1392
1393 /**
1394 * radix_tree_shrink - shrink radix tree to minimum height
1395 * @root radix tree root
1396 */
1397 static inline bool radix_tree_shrink(struct radix_tree_root *root)
1398 {
1399 bool shrunk = false;
1400
1401 for (;;) {
1402 struct radix_tree_node *to_free = root->rnode;
1403 struct radix_tree_node *slot;
1404
1405 if (!radix_tree_is_indirect_ptr(to_free))
1406 break;
1407 to_free = indirect_to_ptr(to_free);
1408
1409 /*
1410 * The candidate node has more than one child, or its child
1411 * is not at the leftmost slot, or the child is a multiorder
1412 * entry, we cannot shrink.
1413 */
1414 if (to_free->count != 1)
1415 break;
1416 slot = to_free->slots[0];
1417 if (!slot)
1418 break;
1419 if (!radix_tree_is_indirect_ptr(slot) && to_free->shift)
1420 break;
1421
1422 if (radix_tree_is_indirect_ptr(slot)) {
1423 slot = indirect_to_ptr(slot);
1424 slot->parent = NULL;
1425 slot = ptr_to_indirect(slot);
1426 }
1427
1428 /*
1429 * We don't need rcu_assign_pointer(), since we are simply
1430 * moving the node from one part of the tree to another: if it
1431 * was safe to dereference the old pointer to it
1432 * (to_free->slots[0]), it will be safe to dereference the new
1433 * one (root->rnode) as far as dependent read barriers go.
1434 */
1435 root->rnode = slot;
1436
1437 /*
1438 * We have a dilemma here. The node's slot[0] must not be
1439 * NULLed in case there are concurrent lookups expecting to
1440 * find the item. However if this was a bottom-level node,
1441 * then it may be subject to the slot pointer being visible
1442 * to callers dereferencing it. If item corresponding to
1443 * slot[0] is subsequently deleted, these callers would expect
1444 * their slot to become empty sooner or later.
1445 *
1446 * For example, lockless pagecache will look up a slot, deref
1447 * the page pointer, and if the page has 0 refcount it means it
1448 * was concurrently deleted from pagecache so try the deref
1449 * again. Fortunately there is already a requirement for logic
1450 * to retry the entire slot lookup -- the indirect pointer
1451 * problem (replacing direct root node with an indirect pointer
1452 * also results in a stale slot). So tag the slot as indirect
1453 * to force callers to retry.
1454 */
1455 if (!radix_tree_is_indirect_ptr(slot))
1456 to_free->slots[0] = RADIX_TREE_RETRY;
1457
1458 radix_tree_node_free(to_free);
1459 shrunk = true;
1460 }
1461
1462 return shrunk;
1463 }
1464
1465 /**
1466 * __radix_tree_delete_node - try to free node after clearing a slot
1467 * @root: radix tree root
1468 * @node: node containing @index
1469 *
1470 * After clearing the slot at @index in @node from radix tree
1471 * rooted at @root, call this function to attempt freeing the
1472 * node and shrinking the tree.
1473 *
1474 * Returns %true if @node was freed, %false otherwise.
1475 */
1476 bool __radix_tree_delete_node(struct radix_tree_root *root,
1477 struct radix_tree_node *node)
1478 {
1479 bool deleted = false;
1480
1481 do {
1482 struct radix_tree_node *parent;
1483
1484 if (node->count) {
1485 if (node == indirect_to_ptr(root->rnode))
1486 deleted |= radix_tree_shrink(root);
1487 return deleted;
1488 }
1489
1490 parent = node->parent;
1491 if (parent) {
1492 parent->slots[node->offset] = NULL;
1493 parent->count--;
1494 } else {
1495 root_tag_clear_all(root);
1496 root->rnode = NULL;
1497 }
1498
1499 radix_tree_node_free(node);
1500 deleted = true;
1501
1502 node = parent;
1503 } while (node);
1504
1505 return deleted;
1506 }
1507
1508 static inline void delete_sibling_entries(struct radix_tree_node *node,
1509 void *ptr, unsigned offset)
1510 {
1511 #ifdef CONFIG_RADIX_TREE_MULTIORDER
1512 int i;
1513 for (i = 1; offset + i < RADIX_TREE_MAP_SIZE; i++) {
1514 if (node->slots[offset + i] != ptr)
1515 break;
1516 node->slots[offset + i] = NULL;
1517 node->count--;
1518 }
1519 #endif
1520 }
1521
1522 /**
1523 * radix_tree_delete_item - delete an item from a radix tree
1524 * @root: radix tree root
1525 * @index: index key
1526 * @item: expected item
1527 *
1528 * Remove @item at @index from the radix tree rooted at @root.
1529 *
1530 * Returns the address of the deleted item, or NULL if it was not present
1531 * or the entry at the given @index was not @item.
1532 */
1533 void *radix_tree_delete_item(struct radix_tree_root *root,
1534 unsigned long index, void *item)
1535 {
1536 struct radix_tree_node *node;
1537 unsigned int offset;
1538 void **slot;
1539 void *entry;
1540 int tag;
1541
1542 entry = __radix_tree_lookup(root, index, &node, &slot);
1543 if (!entry)
1544 return NULL;
1545
1546 if (item && entry != item)
1547 return NULL;
1548
1549 if (!node) {
1550 root_tag_clear_all(root);
1551 root->rnode = NULL;
1552 return entry;
1553 }
1554
1555 offset = get_slot_offset(node, slot);
1556
1557 /*
1558 * Clear all tags associated with the item to be deleted.
1559 * This way of doing it would be inefficient, but seldom is any set.
1560 */
1561 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
1562 if (tag_get(node, tag, offset))
1563 radix_tree_tag_clear(root, index, tag);
1564 }
1565
1566 delete_sibling_entries(node, ptr_to_indirect(slot), offset);
1567 node->slots[offset] = NULL;
1568 node->count--;
1569
1570 __radix_tree_delete_node(root, node);
1571
1572 return entry;
1573 }
1574 EXPORT_SYMBOL(radix_tree_delete_item);
1575
1576 /**
1577 * radix_tree_delete - delete an item from a radix tree
1578 * @root: radix tree root
1579 * @index: index key
1580 *
1581 * Remove the item at @index from the radix tree rooted at @root.
1582 *
1583 * Returns the address of the deleted item, or NULL if it was not present.
1584 */
1585 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
1586 {
1587 return radix_tree_delete_item(root, index, NULL);
1588 }
1589 EXPORT_SYMBOL(radix_tree_delete);
1590
1591 /**
1592 * radix_tree_tagged - test whether any items in the tree are tagged
1593 * @root: radix tree root
1594 * @tag: tag to test
1595 */
1596 int radix_tree_tagged(struct radix_tree_root *root, unsigned int tag)
1597 {
1598 return root_tag_get(root, tag);
1599 }
1600 EXPORT_SYMBOL(radix_tree_tagged);
1601
1602 static void
1603 radix_tree_node_ctor(void *arg)
1604 {
1605 struct radix_tree_node *node = arg;
1606
1607 memset(node, 0, sizeof(*node));
1608 INIT_LIST_HEAD(&node->private_list);
1609 }
1610
1611 static int radix_tree_callback(struct notifier_block *nfb,
1612 unsigned long action, void *hcpu)
1613 {
1614 int cpu = (long)hcpu;
1615 struct radix_tree_preload *rtp;
1616 struct radix_tree_node *node;
1617
1618 /* Free per-cpu pool of preloaded nodes */
1619 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
1620 rtp = &per_cpu(radix_tree_preloads, cpu);
1621 while (rtp->nr) {
1622 node = rtp->nodes;
1623 rtp->nodes = node->private_data;
1624 kmem_cache_free(radix_tree_node_cachep, node);
1625 rtp->nr--;
1626 }
1627 }
1628 return NOTIFY_OK;
1629 }
1630
1631 void __init radix_tree_init(void)
1632 {
1633 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
1634 sizeof(struct radix_tree_node), 0,
1635 SLAB_PANIC | SLAB_RECLAIM_ACCOUNT,
1636 radix_tree_node_ctor);
1637 hotcpu_notifier(radix_tree_callback, 0);
1638 }
This page took 0.069485 seconds and 6 git commands to generate.