[PATCH] mm: dup_mmap use oldmm more
[linux-2.6.22.y-op.git] / lib / radix-tree.c
blobd1c057e71b683db7328f1c8ef619e93866e486cf
1 /*
2 * Copyright (C) 2001 Momchil Velikov
3 * Portions Copyright (C) 2001 Christoph Hellwig
4 * Copyright (C) 2005 SGI, Christoph Lameter <clameter@sgi.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation; either version 2, or (at
9 * your option) any later version.
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include <linux/errno.h>
22 #include <linux/init.h>
23 #include <linux/kernel.h>
24 #include <linux/module.h>
25 #include <linux/radix-tree.h>
26 #include <linux/percpu.h>
27 #include <linux/slab.h>
28 #include <linux/notifier.h>
29 #include <linux/cpu.h>
30 #include <linux/gfp.h>
31 #include <linux/string.h>
32 #include <linux/bitops.h>
35 #ifdef __KERNEL__
36 #define RADIX_TREE_MAP_SHIFT 6
37 #else
38 #define RADIX_TREE_MAP_SHIFT 3 /* For more stressful testing */
39 #endif
40 #define RADIX_TREE_TAGS 2
42 #define RADIX_TREE_MAP_SIZE (1UL << RADIX_TREE_MAP_SHIFT)
43 #define RADIX_TREE_MAP_MASK (RADIX_TREE_MAP_SIZE-1)
45 #define RADIX_TREE_TAG_LONGS \
46 ((RADIX_TREE_MAP_SIZE + BITS_PER_LONG - 1) / BITS_PER_LONG)
48 struct radix_tree_node {
49 unsigned int count;
50 void *slots[RADIX_TREE_MAP_SIZE];
51 unsigned long tags[RADIX_TREE_TAGS][RADIX_TREE_TAG_LONGS];
54 struct radix_tree_path {
55 struct radix_tree_node *node;
56 int offset;
59 #define RADIX_TREE_INDEX_BITS (8 /* CHAR_BIT */ * sizeof(unsigned long))
60 #define RADIX_TREE_MAX_PATH (RADIX_TREE_INDEX_BITS/RADIX_TREE_MAP_SHIFT + 2)
62 static unsigned long height_to_maxindex[RADIX_TREE_MAX_PATH] __read_mostly;
65 * Radix tree node cache.
67 static kmem_cache_t *radix_tree_node_cachep;
70 * Per-cpu pool of preloaded nodes
72 struct radix_tree_preload {
73 int nr;
74 struct radix_tree_node *nodes[RADIX_TREE_MAX_PATH];
76 DEFINE_PER_CPU(struct radix_tree_preload, radix_tree_preloads) = { 0, };
79 * This assumes that the caller has performed appropriate preallocation, and
80 * that the caller has pinned this thread of control to the current CPU.
82 static struct radix_tree_node *
83 radix_tree_node_alloc(struct radix_tree_root *root)
85 struct radix_tree_node *ret;
87 ret = kmem_cache_alloc(radix_tree_node_cachep, root->gfp_mask);
88 if (ret == NULL && !(root->gfp_mask & __GFP_WAIT)) {
89 struct radix_tree_preload *rtp;
91 rtp = &__get_cpu_var(radix_tree_preloads);
92 if (rtp->nr) {
93 ret = rtp->nodes[rtp->nr - 1];
94 rtp->nodes[rtp->nr - 1] = NULL;
95 rtp->nr--;
98 return ret;
101 static inline void
102 radix_tree_node_free(struct radix_tree_node *node)
104 kmem_cache_free(radix_tree_node_cachep, node);
108 * Load up this CPU's radix_tree_node buffer with sufficient objects to
109 * ensure that the addition of a single element in the tree cannot fail. On
110 * success, return zero, with preemption disabled. On error, return -ENOMEM
111 * with preemption not disabled.
113 int radix_tree_preload(gfp_t gfp_mask)
115 struct radix_tree_preload *rtp;
116 struct radix_tree_node *node;
117 int ret = -ENOMEM;
119 preempt_disable();
120 rtp = &__get_cpu_var(radix_tree_preloads);
121 while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
122 preempt_enable();
123 node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
124 if (node == NULL)
125 goto out;
126 preempt_disable();
127 rtp = &__get_cpu_var(radix_tree_preloads);
128 if (rtp->nr < ARRAY_SIZE(rtp->nodes))
129 rtp->nodes[rtp->nr++] = node;
130 else
131 kmem_cache_free(radix_tree_node_cachep, node);
133 ret = 0;
134 out:
135 return ret;
138 static inline void tag_set(struct radix_tree_node *node, int tag, int offset)
140 if (!test_bit(offset, &node->tags[tag][0]))
141 __set_bit(offset, &node->tags[tag][0]);
144 static inline void tag_clear(struct radix_tree_node *node, int tag, int offset)
146 __clear_bit(offset, &node->tags[tag][0]);
149 static inline int tag_get(struct radix_tree_node *node, int tag, int offset)
151 return test_bit(offset, &node->tags[tag][0]);
155 * Return the maximum key which can be store into a
156 * radix tree with height HEIGHT.
158 static inline unsigned long radix_tree_maxindex(unsigned int height)
160 return height_to_maxindex[height];
164 * Extend a radix tree so it can store key @index.
166 static int radix_tree_extend(struct radix_tree_root *root, unsigned long index)
168 struct radix_tree_node *node;
169 unsigned int height;
170 char tags[RADIX_TREE_TAGS];
171 int tag;
173 /* Figure out what the height should be. */
174 height = root->height + 1;
175 while (index > radix_tree_maxindex(height))
176 height++;
178 if (root->rnode == NULL) {
179 root->height = height;
180 goto out;
184 * Prepare the tag status of the top-level node for propagation
185 * into the newly-pushed top-level node(s)
187 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
188 int idx;
190 tags[tag] = 0;
191 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
192 if (root->rnode->tags[tag][idx]) {
193 tags[tag] = 1;
194 break;
199 do {
200 if (!(node = radix_tree_node_alloc(root)))
201 return -ENOMEM;
203 /* Increase the height. */
204 node->slots[0] = root->rnode;
206 /* Propagate the aggregated tag info into the new root */
207 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
208 if (tags[tag])
209 tag_set(node, tag, 0);
212 node->count = 1;
213 root->rnode = node;
214 root->height++;
215 } while (height > root->height);
216 out:
217 return 0;
221 * radix_tree_insert - insert into a radix tree
222 * @root: radix tree root
223 * @index: index key
224 * @item: item to insert
226 * Insert an item into the radix tree at position @index.
228 int radix_tree_insert(struct radix_tree_root *root,
229 unsigned long index, void *item)
231 struct radix_tree_node *node = NULL, *slot;
232 unsigned int height, shift;
233 int offset;
234 int error;
236 /* Make sure the tree is high enough. */
237 if ((!index && !root->rnode) ||
238 index > radix_tree_maxindex(root->height)) {
239 error = radix_tree_extend(root, index);
240 if (error)
241 return error;
244 slot = root->rnode;
245 height = root->height;
246 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
248 offset = 0; /* uninitialised var warning */
249 while (height > 0) {
250 if (slot == NULL) {
251 /* Have to add a child node. */
252 if (!(slot = radix_tree_node_alloc(root)))
253 return -ENOMEM;
254 if (node) {
255 node->slots[offset] = slot;
256 node->count++;
257 } else
258 root->rnode = slot;
261 /* Go a level down */
262 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
263 node = slot;
264 slot = node->slots[offset];
265 shift -= RADIX_TREE_MAP_SHIFT;
266 height--;
269 if (slot != NULL)
270 return -EEXIST;
272 if (node) {
273 node->count++;
274 node->slots[offset] = item;
275 BUG_ON(tag_get(node, 0, offset));
276 BUG_ON(tag_get(node, 1, offset));
277 } else
278 root->rnode = item;
280 return 0;
282 EXPORT_SYMBOL(radix_tree_insert);
285 * radix_tree_lookup - perform lookup operation on a radix tree
286 * @root: radix tree root
287 * @index: index key
289 * Lookup the item at the position @index in the radix tree @root.
291 void *radix_tree_lookup(struct radix_tree_root *root, unsigned long index)
293 unsigned int height, shift;
294 struct radix_tree_node *slot;
296 height = root->height;
297 if (index > radix_tree_maxindex(height))
298 return NULL;
300 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
301 slot = root->rnode;
303 while (height > 0) {
304 if (slot == NULL)
305 return NULL;
307 slot = slot->slots[(index >> shift) & RADIX_TREE_MAP_MASK];
308 shift -= RADIX_TREE_MAP_SHIFT;
309 height--;
312 return slot;
314 EXPORT_SYMBOL(radix_tree_lookup);
317 * radix_tree_tag_set - set a tag on a radix tree node
318 * @root: radix tree root
319 * @index: index key
320 * @tag: tag index
322 * Set the search tag corresponging to @index in the radix tree. From
323 * the root all the way down to the leaf node.
325 * Returns the address of the tagged item. Setting a tag on a not-present
326 * item is a bug.
328 void *radix_tree_tag_set(struct radix_tree_root *root,
329 unsigned long index, int tag)
331 unsigned int height, shift;
332 struct radix_tree_node *slot;
334 height = root->height;
335 if (index > radix_tree_maxindex(height))
336 return NULL;
338 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
339 slot = root->rnode;
341 while (height > 0) {
342 int offset;
344 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
345 tag_set(slot, tag, offset);
346 slot = slot->slots[offset];
347 BUG_ON(slot == NULL);
348 shift -= RADIX_TREE_MAP_SHIFT;
349 height--;
352 return slot;
354 EXPORT_SYMBOL(radix_tree_tag_set);
357 * radix_tree_tag_clear - clear a tag on a radix tree node
358 * @root: radix tree root
359 * @index: index key
360 * @tag: tag index
362 * Clear the search tag corresponging to @index in the radix tree. If
363 * this causes the leaf node to have no tags set then clear the tag in the
364 * next-to-leaf node, etc.
366 * Returns the address of the tagged item on success, else NULL. ie:
367 * has the same return value and semantics as radix_tree_lookup().
369 void *radix_tree_tag_clear(struct radix_tree_root *root,
370 unsigned long index, int tag)
372 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
373 struct radix_tree_node *slot;
374 unsigned int height, shift;
375 void *ret = NULL;
377 height = root->height;
378 if (index > radix_tree_maxindex(height))
379 goto out;
381 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
382 pathp->node = NULL;
383 slot = root->rnode;
385 while (height > 0) {
386 int offset;
388 if (slot == NULL)
389 goto out;
391 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
392 pathp[1].offset = offset;
393 pathp[1].node = slot;
394 slot = slot->slots[offset];
395 pathp++;
396 shift -= RADIX_TREE_MAP_SHIFT;
397 height--;
400 ret = slot;
401 if (ret == NULL)
402 goto out;
404 do {
405 int idx;
407 tag_clear(pathp->node, tag, pathp->offset);
408 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
409 if (pathp->node->tags[tag][idx])
410 goto out;
412 pathp--;
413 } while (pathp->node);
414 out:
415 return ret;
417 EXPORT_SYMBOL(radix_tree_tag_clear);
419 #ifndef __KERNEL__ /* Only the test harness uses this at present */
421 * radix_tree_tag_get - get a tag on a radix tree node
422 * @root: radix tree root
423 * @index: index key
424 * @tag: tag index
426 * Return values:
428 * 0: tag not present
429 * 1: tag present, set
430 * -1: tag present, unset
432 int radix_tree_tag_get(struct radix_tree_root *root,
433 unsigned long index, int tag)
435 unsigned int height, shift;
436 struct radix_tree_node *slot;
437 int saw_unset_tag = 0;
439 height = root->height;
440 if (index > radix_tree_maxindex(height))
441 return 0;
443 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
444 slot = root->rnode;
446 for ( ; ; ) {
447 int offset;
449 if (slot == NULL)
450 return 0;
452 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
455 * This is just a debug check. Later, we can bale as soon as
456 * we see an unset tag.
458 if (!tag_get(slot, tag, offset))
459 saw_unset_tag = 1;
460 if (height == 1) {
461 int ret = tag_get(slot, tag, offset);
463 BUG_ON(ret && saw_unset_tag);
464 return ret ? 1 : -1;
466 slot = slot->slots[offset];
467 shift -= RADIX_TREE_MAP_SHIFT;
468 height--;
471 EXPORT_SYMBOL(radix_tree_tag_get);
472 #endif
474 static unsigned int
475 __lookup(struct radix_tree_root *root, void **results, unsigned long index,
476 unsigned int max_items, unsigned long *next_index)
478 unsigned int nr_found = 0;
479 unsigned int shift, height;
480 struct radix_tree_node *slot;
481 unsigned long i;
483 height = root->height;
484 if (height == 0)
485 goto out;
487 shift = (height-1) * RADIX_TREE_MAP_SHIFT;
488 slot = root->rnode;
490 for ( ; height > 1; height--) {
492 for (i = (index >> shift) & RADIX_TREE_MAP_MASK ;
493 i < RADIX_TREE_MAP_SIZE; i++) {
494 if (slot->slots[i] != NULL)
495 break;
496 index &= ~((1UL << shift) - 1);
497 index += 1UL << shift;
498 if (index == 0)
499 goto out; /* 32-bit wraparound */
501 if (i == RADIX_TREE_MAP_SIZE)
502 goto out;
504 shift -= RADIX_TREE_MAP_SHIFT;
505 slot = slot->slots[i];
508 /* Bottom level: grab some items */
509 for (i = index & RADIX_TREE_MAP_MASK; i < RADIX_TREE_MAP_SIZE; i++) {
510 index++;
511 if (slot->slots[i]) {
512 results[nr_found++] = slot->slots[i];
513 if (nr_found == max_items)
514 goto out;
517 out:
518 *next_index = index;
519 return nr_found;
523 * radix_tree_gang_lookup - perform multiple lookup on a radix tree
524 * @root: radix tree root
525 * @results: where the results of the lookup are placed
526 * @first_index: start the lookup from this key
527 * @max_items: place up to this many items at *results
529 * Performs an index-ascending scan of the tree for present items. Places
530 * them at *@results and returns the number of items which were placed at
531 * *@results.
533 * The implementation is naive.
535 unsigned int
536 radix_tree_gang_lookup(struct radix_tree_root *root, void **results,
537 unsigned long first_index, unsigned int max_items)
539 const unsigned long max_index = radix_tree_maxindex(root->height);
540 unsigned long cur_index = first_index;
541 unsigned int ret = 0;
543 while (ret < max_items) {
544 unsigned int nr_found;
545 unsigned long next_index; /* Index of next search */
547 if (cur_index > max_index)
548 break;
549 nr_found = __lookup(root, results + ret, cur_index,
550 max_items - ret, &next_index);
551 ret += nr_found;
552 if (next_index == 0)
553 break;
554 cur_index = next_index;
556 return ret;
558 EXPORT_SYMBOL(radix_tree_gang_lookup);
561 * FIXME: the two tag_get()s here should use find_next_bit() instead of
562 * open-coding the search.
564 static unsigned int
565 __lookup_tag(struct radix_tree_root *root, void **results, unsigned long index,
566 unsigned int max_items, unsigned long *next_index, int tag)
568 unsigned int nr_found = 0;
569 unsigned int shift;
570 unsigned int height = root->height;
571 struct radix_tree_node *slot;
573 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
574 slot = root->rnode;
576 while (height > 0) {
577 unsigned long i = (index >> shift) & RADIX_TREE_MAP_MASK;
579 for ( ; i < RADIX_TREE_MAP_SIZE; i++) {
580 if (tag_get(slot, tag, i)) {
581 BUG_ON(slot->slots[i] == NULL);
582 break;
584 index &= ~((1UL << shift) - 1);
585 index += 1UL << shift;
586 if (index == 0)
587 goto out; /* 32-bit wraparound */
589 if (i == RADIX_TREE_MAP_SIZE)
590 goto out;
591 height--;
592 if (height == 0) { /* Bottom level: grab some items */
593 unsigned long j = index & RADIX_TREE_MAP_MASK;
595 for ( ; j < RADIX_TREE_MAP_SIZE; j++) {
596 index++;
597 if (tag_get(slot, tag, j)) {
598 BUG_ON(slot->slots[j] == NULL);
599 results[nr_found++] = slot->slots[j];
600 if (nr_found == max_items)
601 goto out;
605 shift -= RADIX_TREE_MAP_SHIFT;
606 slot = slot->slots[i];
608 out:
609 *next_index = index;
610 return nr_found;
614 * radix_tree_gang_lookup_tag - perform multiple lookup on a radix tree
615 * based on a tag
616 * @root: radix tree root
617 * @results: where the results of the lookup are placed
618 * @first_index: start the lookup from this key
619 * @max_items: place up to this many items at *results
620 * @tag: the tag index
622 * Performs an index-ascending scan of the tree for present items which
623 * have the tag indexed by @tag set. Places the items at *@results and
624 * returns the number of items which were placed at *@results.
626 unsigned int
627 radix_tree_gang_lookup_tag(struct radix_tree_root *root, void **results,
628 unsigned long first_index, unsigned int max_items, int tag)
630 const unsigned long max_index = radix_tree_maxindex(root->height);
631 unsigned long cur_index = first_index;
632 unsigned int ret = 0;
634 while (ret < max_items) {
635 unsigned int nr_found;
636 unsigned long next_index; /* Index of next search */
638 if (cur_index > max_index)
639 break;
640 nr_found = __lookup_tag(root, results + ret, cur_index,
641 max_items - ret, &next_index, tag);
642 ret += nr_found;
643 if (next_index == 0)
644 break;
645 cur_index = next_index;
647 return ret;
649 EXPORT_SYMBOL(radix_tree_gang_lookup_tag);
652 * radix_tree_delete - delete an item from a radix tree
653 * @root: radix tree root
654 * @index: index key
656 * Remove the item at @index from the radix tree rooted at @root.
658 * Returns the address of the deleted item, or NULL if it was not present.
660 void *radix_tree_delete(struct radix_tree_root *root, unsigned long index)
662 struct radix_tree_path path[RADIX_TREE_MAX_PATH], *pathp = path;
663 struct radix_tree_path *orig_pathp;
664 struct radix_tree_node *slot;
665 unsigned int height, shift;
666 void *ret = NULL;
667 char tags[RADIX_TREE_TAGS];
668 int nr_cleared_tags;
670 height = root->height;
671 if (index > radix_tree_maxindex(height))
672 goto out;
674 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
675 pathp->node = NULL;
676 slot = root->rnode;
678 for ( ; height > 0; height--) {
679 int offset;
681 if (slot == NULL)
682 goto out;
684 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
685 pathp[1].offset = offset;
686 pathp[1].node = slot;
687 slot = slot->slots[offset];
688 pathp++;
689 shift -= RADIX_TREE_MAP_SHIFT;
692 ret = slot;
693 if (ret == NULL)
694 goto out;
696 orig_pathp = pathp;
699 * Clear all tags associated with the just-deleted item
701 memset(tags, 0, sizeof(tags));
702 do {
703 int tag;
705 nr_cleared_tags = RADIX_TREE_TAGS;
706 for (tag = 0; tag < RADIX_TREE_TAGS; tag++) {
707 int idx;
709 if (tags[tag])
710 continue;
712 tag_clear(pathp->node, tag, pathp->offset);
714 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
715 if (pathp->node->tags[tag][idx]) {
716 tags[tag] = 1;
717 nr_cleared_tags--;
718 break;
722 pathp--;
723 } while (pathp->node && nr_cleared_tags);
725 /* Now free the nodes we do not need anymore */
726 for (pathp = orig_pathp; pathp->node; pathp--) {
727 pathp->node->slots[pathp->offset] = NULL;
728 if (--pathp->node->count)
729 goto out;
731 /* Node with zero slots in use so free it */
732 radix_tree_node_free(pathp->node);
734 root->rnode = NULL;
735 root->height = 0;
736 out:
737 return ret;
739 EXPORT_SYMBOL(radix_tree_delete);
742 * radix_tree_tagged - test whether any items in the tree are tagged
743 * @root: radix tree root
744 * @tag: tag to test
746 int radix_tree_tagged(struct radix_tree_root *root, int tag)
748 int idx;
750 if (!root->rnode)
751 return 0;
752 for (idx = 0; idx < RADIX_TREE_TAG_LONGS; idx++) {
753 if (root->rnode->tags[tag][idx])
754 return 1;
756 return 0;
758 EXPORT_SYMBOL(radix_tree_tagged);
760 static void
761 radix_tree_node_ctor(void *node, kmem_cache_t *cachep, unsigned long flags)
763 memset(node, 0, sizeof(struct radix_tree_node));
766 static __init unsigned long __maxindex(unsigned int height)
768 unsigned int tmp = height * RADIX_TREE_MAP_SHIFT;
769 unsigned long index = (~0UL >> (RADIX_TREE_INDEX_BITS - tmp - 1)) >> 1;
771 if (tmp >= RADIX_TREE_INDEX_BITS)
772 index = ~0UL;
773 return index;
776 static __init void radix_tree_init_maxindex(void)
778 unsigned int i;
780 for (i = 0; i < ARRAY_SIZE(height_to_maxindex); i++)
781 height_to_maxindex[i] = __maxindex(i);
784 #ifdef CONFIG_HOTPLUG_CPU
785 static int radix_tree_callback(struct notifier_block *nfb,
786 unsigned long action,
787 void *hcpu)
789 int cpu = (long)hcpu;
790 struct radix_tree_preload *rtp;
792 /* Free per-cpu pool of perloaded nodes */
793 if (action == CPU_DEAD) {
794 rtp = &per_cpu(radix_tree_preloads, cpu);
795 while (rtp->nr) {
796 kmem_cache_free(radix_tree_node_cachep,
797 rtp->nodes[rtp->nr-1]);
798 rtp->nodes[rtp->nr-1] = NULL;
799 rtp->nr--;
802 return NOTIFY_OK;
804 #endif /* CONFIG_HOTPLUG_CPU */
806 void __init radix_tree_init(void)
808 radix_tree_node_cachep = kmem_cache_create("radix_tree_node",
809 sizeof(struct radix_tree_node), 0,
810 SLAB_PANIC, radix_tree_node_ctor, NULL);
811 radix_tree_init_maxindex();
812 hotcpu_notifier(radix_tree_callback, 0);