xref: /freebsd/sys/contrib/openzfs/module/avl/avl.c (revision 22649d4dba730d46244fd2dff4fd174903c8379f)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * This file and its contents are supplied under the terms of the
4  * Common Development and Distribution License ("CDDL"), version 1.0.
5  * You may only use this file in accordance with the terms of version
6  * 1.0 of the CDDL.
7  *
8  * A full copy of the text of the CDDL should have accompanied this
9  * source.  A copy of the CDDL is also available via the Internet at
10  * https://opensource.org/license/CDDL-1.0.
11  */
12 /*
13  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
14  * Use is subject to license terms.
15  */
16 
17 /*
18  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
19  * Copyright (c) 2015 by Delphix. All rights reserved.
20  */
21 
22 /*
23  * AVL - generic AVL tree implementation for kernel use
24  *
25  * A complete description of AVL trees can be found in many CS textbooks.
26  *
27  * Here is a very brief overview. An AVL tree is a binary search tree that is
28  * almost perfectly balanced. By "almost" perfectly balanced, we mean that at
29  * any given node, the left and right subtrees are allowed to differ in height
30  * by at most 1 level.
31  *
32  * This relaxation from a perfectly balanced binary tree allows doing
33  * insertion and deletion relatively efficiently. Searching the tree is
34  * still a fast operation, roughly O(log(N)).
35  *
36  * The key to insertion and deletion is a set of tree manipulations called
37  * rotations, which bring unbalanced subtrees back into the semi-balanced state.
38  *
39  * This implementation of AVL trees has the following peculiarities:
40  *
41  *	- The AVL specific data structures are physically embedded as fields
42  *	  in the "using" data structures.  To maintain generality the code
43  *	  must constantly translate between "avl_node_t *" and containing
44  *	  data structure "void *"s by adding/subtracting the avl_offset.
45  *
46  *	- Since the AVL data is always embedded in other structures, there is
47  *	  no locking or memory allocation in the AVL routines. This must be
48  *	  provided for by the enclosing data structure's semantics. Typically,
49  *	  avl_insert()/_add()/_remove()/avl_insert_here() require some kind of
50  *	  exclusive write lock. Other operations require a read lock.
51  *
52  *      - The implementation uses iteration instead of explicit recursion,
53  *	  since it is intended to run on limited size kernel stacks. Since
54  *	  there is no recursion stack present to move "up" in the tree,
55  *	  there is an explicit "parent" link in the avl_node_t.
56  *
57  *      - The left/right children pointers of a node are in an array.
58  *	  In the code, variables (instead of constants) are used to represent
59  *	  left and right indices.  The implementation is written as if it only
60  *	  dealt with left handed manipulations.  By changing the value assigned
61  *	  to "left", the code also works for right handed trees.  The
62  *	  following variables/terms are frequently used:
63  *
64  *		int left;	// 0 when dealing with left children,
65  *				// 1 for dealing with right children
66  *
67  *		int left_heavy;	// -1 when left subtree is taller at some node,
68  *				// +1 when right subtree is taller
69  *
70  *		int right;	// will be the opposite of left (0 or 1)
71  *		int right_heavy;// will be the opposite of left_heavy (-1 or 1)
72  *
73  *		int direction;  // 0 for "<" (ie. left child); 1 for ">" (right)
74  *
75  *	  Though it is a little more confusing to read the code, the approach
76  *	  allows using half as much code (and hence cache footprint) for tree
77  *	  manipulations and eliminates many conditional branches.
78  *
79  *	- The avl_index_t is an opaque "cookie" used to find nodes at or
80  *	  adjacent to where a new value would be inserted in the tree. The value
81  *	  is a modified "avl_node_t *".  The bottom bit (normally 0 for a
82  *	  pointer) is set to indicate if that the new node has a value greater
83  *	  than the value of the indicated "avl_node_t *".
84  *
85  * Note - in addition to userland (e.g. libavl and libutil) and the kernel
86  * (e.g. genunix), avl.c is compiled into ld.so and kmdb's genunix module,
87  * which each have their own compilation environments and subsequent
88  * requirements. Each of these environments must be considered when adding
89  * dependencies from avl.c.
90  *
91  * Link to Illumos.org for more information on avl function:
92  * [1] https://illumos.org/man/9f/avl
93  */
94 
95 #include <sys/types.h>
96 #include <sys/param.h>
97 #include <sys/debug.h>
98 #include <sys/avl.h>
99 #include <sys/cmn_err.h>
100 #include <sys/mod.h>
101 
102 #ifndef _KERNEL
103 #include <string.h>
104 #endif
105 
106 /*
107  * Walk from one node to the previous valued node (ie. an infix walk
108  * towards the left). At any given node we do one of 2 things:
109  *
110  * - If there is a left child, go to it, then to it's rightmost descendant.
111  *
112  * - otherwise we return through parent nodes until we've come from a right
113  *   child.
114  *
115  * Return Value:
116  * NULL - if at the end of the nodes
117  * otherwise next node
118  */
119 void *
avl_walk(avl_tree_t * tree,void * oldnode,int left)120 avl_walk(avl_tree_t *tree, void	*oldnode, int left)
121 {
122 	size_t off = tree->avl_offset;
123 	avl_node_t *node = AVL_DATA2NODE(oldnode, off);
124 	int right = 1 - left;
125 	int was_child;
126 
127 
128 	/*
129 	 * nowhere to walk to if tree is empty
130 	 */
131 	if (node == NULL)
132 		return (NULL);
133 
134 	/*
135 	 * Visit the previous valued node. There are two possibilities:
136 	 *
137 	 * If this node has a left child, go down one left, then all
138 	 * the way right.
139 	 */
140 	if (node->avl_child[left] != NULL) {
141 		for (node = node->avl_child[left];
142 		    node->avl_child[right] != NULL;
143 		    node = node->avl_child[right])
144 			;
145 	/*
146 	 * Otherwise, return through left children as far as we can.
147 	 */
148 	} else {
149 		for (;;) {
150 			was_child = AVL_XCHILD(node);
151 			node = AVL_XPARENT(node);
152 			if (node == NULL)
153 				return (NULL);
154 			if (was_child == right)
155 				break;
156 		}
157 	}
158 
159 	return (AVL_NODE2DATA(node, off));
160 }
161 
162 /*
163  * Return the lowest valued node in a tree or NULL.
164  * (leftmost child from root of tree)
165  */
166 void *
avl_first(avl_tree_t * tree)167 avl_first(avl_tree_t *tree)
168 {
169 	avl_node_t *node;
170 	avl_node_t *prev = NULL;
171 	size_t off = tree->avl_offset;
172 
173 	for (node = tree->avl_root; node != NULL; node = node->avl_child[0])
174 		prev = node;
175 
176 	if (prev != NULL)
177 		return (AVL_NODE2DATA(prev, off));
178 	return (NULL);
179 }
180 
181 /*
182  * Return the highest valued node in a tree or NULL.
183  * (rightmost child from root of tree)
184  */
185 void *
avl_last(avl_tree_t * tree)186 avl_last(avl_tree_t *tree)
187 {
188 	avl_node_t *node;
189 	avl_node_t *prev = NULL;
190 	size_t off = tree->avl_offset;
191 
192 	for (node = tree->avl_root; node != NULL; node = node->avl_child[1])
193 		prev = node;
194 
195 	if (prev != NULL)
196 		return (AVL_NODE2DATA(prev, off));
197 	return (NULL);
198 }
199 
200 /*
201  * Access the node immediately before or after an insertion point.
202  *
203  * "avl_index_t" is a (avl_node_t *) with the bottom bit indicating a child
204  *
205  * Return value:
206  *	NULL: no node in the given direction
207  *	"void *"  of the found tree node
208  */
209 void *
avl_nearest(avl_tree_t * tree,avl_index_t where,int direction)210 avl_nearest(avl_tree_t *tree, avl_index_t where, int direction)
211 {
212 	int child = AVL_INDEX2CHILD(where);
213 	avl_node_t *node = AVL_INDEX2NODE(where);
214 	void *data;
215 	size_t off = tree->avl_offset;
216 
217 	if (node == NULL) {
218 		ASSERT0P(tree->avl_root);
219 		return (NULL);
220 	}
221 	data = AVL_NODE2DATA(node, off);
222 	if (child != direction)
223 		return (data);
224 
225 	return (avl_walk(tree, data, direction));
226 }
227 
228 
229 /*
230  * Search for the node which contains "value".  The algorithm is a
231  * simple binary tree search.
232  *
233  * return value:
234  *	NULL: the value is not in the AVL tree
235  *		*where (if not NULL)  is set to indicate the insertion point
236  *	"void *"  of the found tree node
237  */
238 void *
avl_find(avl_tree_t * tree,const void * value,avl_index_t * where)239 avl_find(avl_tree_t *tree, const void *value, avl_index_t *where)
240 {
241 	avl_node_t *node;
242 	avl_node_t *prev = NULL;
243 	int child = 0;
244 	int diff;
245 	size_t off = tree->avl_offset;
246 
247 	for (node = tree->avl_root; node != NULL;
248 	    node = node->avl_child[child]) {
249 
250 		prev = node;
251 
252 		diff = tree->avl_compar(value, AVL_NODE2DATA(node, off));
253 		ASSERT(-1 <= diff && diff <= 1);
254 		if (diff == 0) {
255 #ifdef ZFS_DEBUG
256 			if (where != NULL)
257 				*where = 0;
258 #endif
259 			return (AVL_NODE2DATA(node, off));
260 		}
261 		child = (diff > 0);
262 	}
263 
264 	if (where != NULL)
265 		*where = AVL_MKINDEX(prev, child);
266 
267 	return (NULL);
268 }
269 
270 
271 /*
272  * Perform a rotation to restore balance at the subtree given by depth.
273  *
274  * This routine is used by both insertion and deletion. The return value
275  * indicates:
276  *	 0 : subtree did not change height
277  *	!0 : subtree was reduced in height
278  *
279  * The code is written as if handling left rotations, right rotations are
280  * symmetric and handled by swapping values of variables right/left[_heavy]
281  *
282  * On input balance is the "new" balance at "node". This value is either
283  * -2 or +2.
284  */
285 static int
avl_rotation(avl_tree_t * tree,avl_node_t * node,int balance)286 avl_rotation(avl_tree_t *tree, avl_node_t *node, int balance)
287 {
288 	int left = !(balance < 0);	/* when balance = -2, left will be 0 */
289 	int right = 1 - left;
290 	int left_heavy = balance >> 1;
291 	int right_heavy = -left_heavy;
292 	avl_node_t *parent = AVL_XPARENT(node);
293 	avl_node_t *child = node->avl_child[left];
294 	avl_node_t *cright;
295 	avl_node_t *gchild;
296 	avl_node_t *gright;
297 	avl_node_t *gleft;
298 	int which_child = AVL_XCHILD(node);
299 	int child_bal = AVL_XBALANCE(child);
300 
301 	/*
302 	 * case 1 : node is overly left heavy, the left child is balanced or
303 	 * also left heavy. This requires the following rotation.
304 	 *
305 	 *                   (node bal:-2)
306 	 *                    /           \
307 	 *                   /             \
308 	 *              (child bal:0 or -1)
309 	 *              /    \
310 	 *             /      \
311 	 *                     cright
312 	 *
313 	 * becomes:
314 	 *
315 	 *              (child bal:1 or 0)
316 	 *              /        \
317 	 *             /          \
318 	 *                        (node bal:-1 or 0)
319 	 *                         /     \
320 	 *                        /       \
321 	 *                     cright
322 	 *
323 	 * we detect this situation by noting that child's balance is not
324 	 * right_heavy.
325 	 */
326 	if (child_bal != right_heavy) {
327 
328 		/*
329 		 * compute new balance of nodes
330 		 *
331 		 * If child used to be left heavy (now balanced) we reduced
332 		 * the height of this sub-tree -- used in "return...;" below
333 		 */
334 		child_bal += right_heavy; /* adjust towards right */
335 
336 		/*
337 		 * move "cright" to be node's left child
338 		 */
339 		cright = child->avl_child[right];
340 		node->avl_child[left] = cright;
341 		if (cright != NULL) {
342 			AVL_SETPARENT(cright, node);
343 			AVL_SETCHILD(cright, left);
344 		}
345 
346 		/*
347 		 * move node to be child's right child
348 		 */
349 		child->avl_child[right] = node;
350 		AVL_SETBALANCE(node, -child_bal);
351 		AVL_SETCHILD(node, right);
352 		AVL_SETPARENT(node, child);
353 
354 		/*
355 		 * update the pointer into this subtree
356 		 */
357 		AVL_SETBALANCE(child, child_bal);
358 		AVL_SETCHILD(child, which_child);
359 		AVL_SETPARENT(child, parent);
360 		if (parent != NULL)
361 			parent->avl_child[which_child] = child;
362 		else
363 			tree->avl_root = child;
364 
365 		return (child_bal == 0);
366 	}
367 
368 	/*
369 	 * case 2 : When node is left heavy, but child is right heavy we use
370 	 * a different rotation.
371 	 *
372 	 *                   (node b:-2)
373 	 *                    /   \
374 	 *                   /     \
375 	 *                  /       \
376 	 *             (child b:+1)
377 	 *              /     \
378 	 *             /       \
379 	 *                   (gchild b: != 0)
380 	 *                     /  \
381 	 *                    /    \
382 	 *                 gleft   gright
383 	 *
384 	 * becomes:
385 	 *
386 	 *              (gchild b:0)
387 	 *              /       \
388 	 *             /         \
389 	 *            /           \
390 	 *        (child b:?)   (node b:?)
391 	 *         /  \          /   \
392 	 *        /    \        /     \
393 	 *            gleft   gright
394 	 *
395 	 * computing the new balances is more complicated. As an example:
396 	 *	 if gchild was right_heavy, then child is now left heavy
397 	 *		else it is balanced
398 	 */
399 	gchild = child->avl_child[right];
400 	gleft = gchild->avl_child[left];
401 	gright = gchild->avl_child[right];
402 
403 	/*
404 	 * move gright to left child of node and
405 	 *
406 	 * move gleft to right child of node
407 	 */
408 	node->avl_child[left] = gright;
409 	if (gright != NULL) {
410 		AVL_SETPARENT(gright, node);
411 		AVL_SETCHILD(gright, left);
412 	}
413 
414 	child->avl_child[right] = gleft;
415 	if (gleft != NULL) {
416 		AVL_SETPARENT(gleft, child);
417 		AVL_SETCHILD(gleft, right);
418 	}
419 
420 	/*
421 	 * move child to left child of gchild and
422 	 *
423 	 * move node to right child of gchild and
424 	 *
425 	 * fixup parent of all this to point to gchild
426 	 */
427 	balance = AVL_XBALANCE(gchild);
428 	gchild->avl_child[left] = child;
429 	AVL_SETBALANCE(child, (balance == right_heavy ? left_heavy : 0));
430 	AVL_SETPARENT(child, gchild);
431 	AVL_SETCHILD(child, left);
432 
433 	gchild->avl_child[right] = node;
434 	AVL_SETBALANCE(node, (balance == left_heavy ? right_heavy : 0));
435 	AVL_SETPARENT(node, gchild);
436 	AVL_SETCHILD(node, right);
437 
438 	AVL_SETBALANCE(gchild, 0);
439 	AVL_SETPARENT(gchild, parent);
440 	AVL_SETCHILD(gchild, which_child);
441 	if (parent != NULL)
442 		parent->avl_child[which_child] = gchild;
443 	else
444 		tree->avl_root = gchild;
445 
446 	return (1);	/* the new tree is always shorter */
447 }
448 
449 
450 /*
451  * Insert a new node into an AVL tree at the specified (from avl_find()) place.
452  *
453  * Newly inserted nodes are always leaf nodes in the tree, since avl_find()
454  * searches out to the leaf positions.  The avl_index_t indicates the node
455  * which will be the parent of the new node.
456  *
457  * After the node is inserted, a single rotation further up the tree may
458  * be necessary to maintain an acceptable AVL balance.
459  */
460 void
avl_insert(avl_tree_t * tree,void * new_data,avl_index_t where)461 avl_insert(avl_tree_t *tree, void *new_data, avl_index_t where)
462 {
463 	avl_node_t *node;
464 	avl_node_t *parent = AVL_INDEX2NODE(where);
465 	int old_balance;
466 	int new_balance;
467 	int which_child = AVL_INDEX2CHILD(where);
468 	size_t off = tree->avl_offset;
469 
470 #ifdef _LP64
471 	ASSERT0(((uintptr_t)new_data & 0x7));
472 #endif
473 
474 	node = AVL_DATA2NODE(new_data, off);
475 
476 	/*
477 	 * First, add the node to the tree at the indicated position.
478 	 */
479 	++tree->avl_numnodes;
480 
481 	node->avl_child[0] = NULL;
482 	node->avl_child[1] = NULL;
483 
484 	AVL_SETCHILD(node, which_child);
485 	AVL_SETBALANCE(node, 0);
486 	AVL_SETPARENT(node, parent);
487 	if (parent != NULL) {
488 		ASSERT0P(parent->avl_child[which_child]);
489 		parent->avl_child[which_child] = node;
490 	} else {
491 		ASSERT0P(tree->avl_root);
492 		tree->avl_root = node;
493 	}
494 	/*
495 	 * Now, back up the tree modifying the balance of all nodes above the
496 	 * insertion point. If we get to a highly unbalanced ancestor, we
497 	 * need to do a rotation.  If we back out of the tree we are done.
498 	 * If we brought any subtree into perfect balance (0), we are also done.
499 	 */
500 	for (;;) {
501 		node = parent;
502 		if (node == NULL)
503 			return;
504 
505 		/*
506 		 * Compute the new balance
507 		 */
508 		old_balance = AVL_XBALANCE(node);
509 		new_balance = old_balance + (which_child ? 1 : -1);
510 
511 		/*
512 		 * If we introduced equal balance, then we are done immediately
513 		 */
514 		if (new_balance == 0) {
515 			AVL_SETBALANCE(node, 0);
516 			return;
517 		}
518 
519 		/*
520 		 * If both old and new are not zero we went
521 		 * from -1 to -2 balance, do a rotation.
522 		 */
523 		if (old_balance != 0)
524 			break;
525 
526 		AVL_SETBALANCE(node, new_balance);
527 		parent = AVL_XPARENT(node);
528 		which_child = AVL_XCHILD(node);
529 	}
530 
531 	/*
532 	 * perform a rotation to fix the tree and return
533 	 */
534 	(void) avl_rotation(tree, node, new_balance);
535 }
536 
537 /*
538  * Insert "new_data" in "tree" in the given "direction" either after or
539  * before (AVL_AFTER, AVL_BEFORE) the data "here".
540  *
541  * Insertions can only be done at empty leaf points in the tree, therefore
542  * if the given child of the node is already present we move to either
543  * the AVL_PREV or AVL_NEXT and reverse the insertion direction. Since
544  * every other node in the tree is a leaf, this always works.
545  *
546  * To help developers using this interface, we assert that the new node
547  * is correctly ordered at every step of the way in DEBUG kernels.
548  */
549 void
avl_insert_here(avl_tree_t * tree,void * new_data,void * here,int direction)550 avl_insert_here(
551 	avl_tree_t *tree,
552 	void *new_data,
553 	void *here,
554 	int direction)
555 {
556 	avl_node_t *node;
557 	int child = direction;	/* rely on AVL_BEFORE == 0, AVL_AFTER == 1 */
558 #ifdef ZFS_DEBUG
559 	int diff;
560 #endif
561 
562 	ASSERT(tree != NULL);
563 	ASSERT(new_data != NULL);
564 	ASSERT(here != NULL);
565 	ASSERT(direction == AVL_BEFORE || direction == AVL_AFTER);
566 
567 	/*
568 	 * If corresponding child of node is not NULL, go to the neighboring
569 	 * node and reverse the insertion direction.
570 	 */
571 	node = AVL_DATA2NODE(here, tree->avl_offset);
572 
573 #ifdef ZFS_DEBUG
574 	diff = tree->avl_compar(new_data, here);
575 	ASSERT(-1 <= diff && diff <= 1);
576 	ASSERT(diff != 0);
577 	ASSERT(diff > 0 ? child == 1 : child == 0);
578 #endif
579 
580 	if (node->avl_child[child] != NULL) {
581 		node = node->avl_child[child];
582 		child = 1 - child;
583 		while (node->avl_child[child] != NULL) {
584 #ifdef ZFS_DEBUG
585 			diff = tree->avl_compar(new_data,
586 			    AVL_NODE2DATA(node, tree->avl_offset));
587 			ASSERT(-1 <= diff && diff <= 1);
588 			ASSERT(diff != 0);
589 			ASSERT(diff > 0 ? child == 1 : child == 0);
590 #endif
591 			node = node->avl_child[child];
592 		}
593 #ifdef ZFS_DEBUG
594 		diff = tree->avl_compar(new_data,
595 		    AVL_NODE2DATA(node, tree->avl_offset));
596 		ASSERT(-1 <= diff && diff <= 1);
597 		ASSERT(diff != 0);
598 		ASSERT(diff > 0 ? child == 1 : child == 0);
599 #endif
600 	}
601 	ASSERT0P(node->avl_child[child]);
602 
603 	avl_insert(tree, new_data, AVL_MKINDEX(node, child));
604 }
605 
606 /*
607  * Add a new node to an AVL tree.  Strictly enforce that no duplicates can
608  * be added to the tree with a VERIFY which is enabled for non-DEBUG builds.
609  */
610 void
avl_add(avl_tree_t * tree,void * new_node)611 avl_add(avl_tree_t *tree, void *new_node)
612 {
613 	avl_index_t where = 0;
614 
615 	VERIFY(avl_find(tree, new_node, &where) == NULL);
616 
617 	avl_insert(tree, new_node, where);
618 }
619 
620 /*
621  * Delete a node from the AVL tree.  Deletion is similar to insertion, but
622  * with 2 complications.
623  *
624  * First, we may be deleting an interior node. Consider the following subtree:
625  *
626  *     d           c            c
627  *    / \         / \          / \
628  *   b   e       b   e        b   e
629  *  / \	        / \          /
630  * a   c       a            a
631  *
632  * When we are deleting node (d), we find and bring up an adjacent valued leaf
633  * node, say (c), to take the interior node's place. In the code this is
634  * handled by temporarily swapping (d) and (c) in the tree and then using
635  * common code to delete (d) from the leaf position.
636  *
637  * Secondly, an interior deletion from a deep tree may require more than one
638  * rotation to fix the balance. This is handled by moving up the tree through
639  * parents and applying rotations as needed. The return value from
640  * avl_rotation() is used to detect when a subtree did not change overall
641  * height due to a rotation.
642  */
643 void
avl_remove(avl_tree_t * tree,void * data)644 avl_remove(avl_tree_t *tree, void *data)
645 {
646 	avl_node_t *delete;
647 	avl_node_t *parent;
648 	avl_node_t *node;
649 	avl_node_t tmp;
650 	int old_balance;
651 	int new_balance;
652 	int left;
653 	int right;
654 	int which_child;
655 	size_t off = tree->avl_offset;
656 
657 	delete = AVL_DATA2NODE(data, off);
658 
659 	/*
660 	 * Deletion is easiest with a node that has at most 1 child.
661 	 * We swap a node with 2 children with a sequentially valued
662 	 * neighbor node. That node will have at most 1 child. Note this
663 	 * has no effect on the ordering of the remaining nodes.
664 	 *
665 	 * As an optimization, we choose the greater neighbor if the tree
666 	 * is right heavy, otherwise the left neighbor. This reduces the
667 	 * number of rotations needed.
668 	 */
669 	if (delete->avl_child[0] != NULL && delete->avl_child[1] != NULL) {
670 
671 		/*
672 		 * choose node to swap from whichever side is taller
673 		 */
674 		old_balance = AVL_XBALANCE(delete);
675 		left = (old_balance > 0);
676 		right = 1 - left;
677 
678 		/*
679 		 * get to the previous value'd node
680 		 * (down 1 left, as far as possible right)
681 		 */
682 		for (node = delete->avl_child[left];
683 		    node->avl_child[right] != NULL;
684 		    node = node->avl_child[right])
685 			;
686 
687 		/*
688 		 * create a temp placeholder for 'node'
689 		 * move 'node' to delete's spot in the tree
690 		 */
691 		tmp = *node;
692 
693 		memcpy(node, delete, sizeof (*node));
694 		if (node->avl_child[left] == node)
695 			node->avl_child[left] = &tmp;
696 
697 		parent = AVL_XPARENT(node);
698 		if (parent != NULL)
699 			parent->avl_child[AVL_XCHILD(node)] = node;
700 		else
701 			tree->avl_root = node;
702 		AVL_SETPARENT(node->avl_child[left], node);
703 		AVL_SETPARENT(node->avl_child[right], node);
704 
705 		/*
706 		 * Put tmp where node used to be (just temporary).
707 		 * It always has a parent and at most 1 child.
708 		 */
709 		delete = &tmp;
710 		parent = AVL_XPARENT(delete);
711 		parent->avl_child[AVL_XCHILD(delete)] = delete;
712 		which_child = (delete->avl_child[1] != 0);
713 		if (delete->avl_child[which_child] != NULL)
714 			AVL_SETPARENT(delete->avl_child[which_child], delete);
715 	}
716 
717 
718 	/*
719 	 * Here we know "delete" is at least partially a leaf node. It can
720 	 * be easily removed from the tree.
721 	 */
722 	ASSERT(tree->avl_numnodes > 0);
723 	--tree->avl_numnodes;
724 	parent = AVL_XPARENT(delete);
725 	which_child = AVL_XCHILD(delete);
726 	if (delete->avl_child[0] != NULL)
727 		node = delete->avl_child[0];
728 	else
729 		node = delete->avl_child[1];
730 
731 	/*
732 	 * Connect parent directly to node (leaving out delete).
733 	 */
734 	if (node != NULL) {
735 		AVL_SETPARENT(node, parent);
736 		AVL_SETCHILD(node, which_child);
737 	}
738 	if (parent == NULL) {
739 		tree->avl_root = node;
740 		return;
741 	}
742 	parent->avl_child[which_child] = node;
743 
744 
745 	/*
746 	 * Since the subtree is now shorter, begin adjusting parent balances
747 	 * and performing any needed rotations.
748 	 */
749 	do {
750 
751 		/*
752 		 * Move up the tree and adjust the balance
753 		 *
754 		 * Capture the parent and which_child values for the next
755 		 * iteration before any rotations occur.
756 		 */
757 		node = parent;
758 		old_balance = AVL_XBALANCE(node);
759 		new_balance = old_balance - (which_child ? 1 : -1);
760 		parent = AVL_XPARENT(node);
761 		which_child = AVL_XCHILD(node);
762 
763 		/*
764 		 * If a node was in perfect balance but isn't anymore then
765 		 * we can stop, since the height didn't change above this point
766 		 * due to a deletion.
767 		 */
768 		if (old_balance == 0) {
769 			AVL_SETBALANCE(node, new_balance);
770 			break;
771 		}
772 
773 		/*
774 		 * If the new balance is zero, we don't need to rotate
775 		 * else
776 		 * need a rotation to fix the balance.
777 		 * If the rotation doesn't change the height
778 		 * of the sub-tree we have finished adjusting.
779 		 */
780 		if (new_balance == 0)
781 			AVL_SETBALANCE(node, new_balance);
782 		else if (!avl_rotation(tree, node, new_balance))
783 			break;
784 	} while (parent != NULL);
785 }
786 
787 #define	AVL_REINSERT(tree, obj)		\
788 	avl_remove((tree), (obj));	\
789 	avl_add((tree), (obj))
790 
791 boolean_t
avl_update_lt(avl_tree_t * t,void * obj)792 avl_update_lt(avl_tree_t *t, void *obj)
793 {
794 	void *neighbor;
795 
796 	ASSERT(((neighbor = AVL_NEXT(t, obj)) == NULL) ||
797 	    (t->avl_compar(obj, neighbor) <= 0));
798 
799 	neighbor = AVL_PREV(t, obj);
800 	if ((neighbor != NULL) && (t->avl_compar(obj, neighbor) < 0)) {
801 		AVL_REINSERT(t, obj);
802 		return (B_TRUE);
803 	}
804 
805 	return (B_FALSE);
806 }
807 
808 boolean_t
avl_update_gt(avl_tree_t * t,void * obj)809 avl_update_gt(avl_tree_t *t, void *obj)
810 {
811 	void *neighbor;
812 
813 	ASSERT(((neighbor = AVL_PREV(t, obj)) == NULL) ||
814 	    (t->avl_compar(obj, neighbor) >= 0));
815 
816 	neighbor = AVL_NEXT(t, obj);
817 	if ((neighbor != NULL) && (t->avl_compar(obj, neighbor) > 0)) {
818 		AVL_REINSERT(t, obj);
819 		return (B_TRUE);
820 	}
821 
822 	return (B_FALSE);
823 }
824 
825 boolean_t
avl_update(avl_tree_t * t,void * obj)826 avl_update(avl_tree_t *t, void *obj)
827 {
828 	void *neighbor;
829 
830 	neighbor = AVL_PREV(t, obj);
831 	if ((neighbor != NULL) && (t->avl_compar(obj, neighbor) < 0)) {
832 		AVL_REINSERT(t, obj);
833 		return (B_TRUE);
834 	}
835 
836 	neighbor = AVL_NEXT(t, obj);
837 	if ((neighbor != NULL) && (t->avl_compar(obj, neighbor) > 0)) {
838 		AVL_REINSERT(t, obj);
839 		return (B_TRUE);
840 	}
841 
842 	return (B_FALSE);
843 }
844 
845 void
avl_swap(avl_tree_t * tree1,avl_tree_t * tree2)846 avl_swap(avl_tree_t *tree1, avl_tree_t *tree2)
847 {
848 	avl_node_t *temp_node;
849 	ulong_t temp_numnodes;
850 
851 	ASSERT3P(tree1->avl_compar, ==, tree2->avl_compar);
852 	ASSERT3U(tree1->avl_offset, ==, tree2->avl_offset);
853 
854 	temp_node = tree1->avl_root;
855 	temp_numnodes = tree1->avl_numnodes;
856 	tree1->avl_root = tree2->avl_root;
857 	tree1->avl_numnodes = tree2->avl_numnodes;
858 	tree2->avl_root = temp_node;
859 	tree2->avl_numnodes = temp_numnodes;
860 }
861 
862 /*
863  * initialize a new AVL tree
864  */
865 void
avl_create(avl_tree_t * tree,int (* compar)(const void *,const void *),size_t size,size_t offset)866 avl_create(avl_tree_t *tree, int (*compar) (const void *, const void *),
867     size_t size, size_t offset)
868 {
869 	ASSERT(tree);
870 	ASSERT(compar);
871 	ASSERT(size > 0);
872 	ASSERT(size >= offset + sizeof (avl_node_t));
873 #ifdef _LP64
874 	ASSERT0((offset & 0x7));
875 #endif
876 
877 	tree->avl_compar = compar;
878 	tree->avl_root = NULL;
879 	tree->avl_numnodes = 0;
880 	tree->avl_offset = offset;
881 }
882 
883 /*
884  * Delete a tree.
885  */
886 void
avl_destroy(avl_tree_t * tree)887 avl_destroy(avl_tree_t *tree)
888 {
889 	ASSERT(tree);
890 	ASSERT0(tree->avl_numnodes);
891 	ASSERT0P(tree->avl_root);
892 }
893 
894 
895 /*
896  * Return the number of nodes in an AVL tree.
897  */
898 ulong_t
avl_numnodes(avl_tree_t * tree)899 avl_numnodes(avl_tree_t *tree)
900 {
901 	ASSERT(tree);
902 	return (tree->avl_numnodes);
903 }
904 
905 boolean_t
avl_is_empty(avl_tree_t * tree)906 avl_is_empty(avl_tree_t *tree)
907 {
908 	ASSERT(tree);
909 	return (tree->avl_numnodes == 0);
910 }
911 
912 #define	CHILDBIT	(1L)
913 
914 /*
915  * Post-order tree walk used to visit all tree nodes and destroy the tree
916  * in post order. This is used for removing all the nodes from a tree without
917  * paying any cost for rebalancing it.
918  *
919  * example:
920  *
921  *	void *cookie = NULL;
922  *	my_data_t *node;
923  *
924  *	while ((node = avl_destroy_nodes(tree, &cookie)) != NULL)
925  *		free(node);
926  *	avl_destroy(tree);
927  *
928  * The cookie is really an avl_node_t to the current node's parent and
929  * an indication of which child you looked at last.
930  *
931  * On input, a cookie value of CHILDBIT indicates the tree is done.
932  */
933 void *
avl_destroy_nodes(avl_tree_t * tree,void ** cookie)934 avl_destroy_nodes(avl_tree_t *tree, void **cookie)
935 {
936 	avl_node_t	*node;
937 	avl_node_t	*parent;
938 	int		child;
939 	void		*first;
940 	size_t		off = tree->avl_offset;
941 
942 	/*
943 	 * Initial calls go to the first node or it's right descendant.
944 	 */
945 	if (*cookie == NULL) {
946 		first = avl_first(tree);
947 
948 		/*
949 		 * deal with an empty tree
950 		 */
951 		if (first == NULL) {
952 			*cookie = (void *)CHILDBIT;
953 			return (NULL);
954 		}
955 
956 		node = AVL_DATA2NODE(first, off);
957 		parent = AVL_XPARENT(node);
958 		goto check_right_side;
959 	}
960 
961 	/*
962 	 * If there is no parent to return to we are done.
963 	 */
964 	parent = (avl_node_t *)((uintptr_t)(*cookie) & ~CHILDBIT);
965 	if (parent == NULL) {
966 		if (tree->avl_root != NULL) {
967 			ASSERT(tree->avl_numnodes == 1);
968 			tree->avl_root = NULL;
969 			tree->avl_numnodes = 0;
970 		}
971 		return (NULL);
972 	}
973 
974 	/*
975 	 * Remove the child pointer we just visited from the parent and tree.
976 	 */
977 	child = (uintptr_t)(*cookie) & CHILDBIT;
978 	parent->avl_child[child] = NULL;
979 	ASSERT(tree->avl_numnodes > 1);
980 	--tree->avl_numnodes;
981 
982 	/*
983 	 * If we just removed a right child or there isn't one, go up to parent.
984 	 */
985 	if (child == 1 || parent->avl_child[1] == NULL) {
986 		node = parent;
987 		parent = AVL_XPARENT(parent);
988 		goto done;
989 	}
990 
991 	/*
992 	 * Do parent's right child, then leftmost descendent.
993 	 */
994 	node = parent->avl_child[1];
995 	while (node->avl_child[0] != NULL) {
996 		parent = node;
997 		node = node->avl_child[0];
998 	}
999 
1000 	/*
1001 	 * If here, we moved to a left child. It may have one
1002 	 * child on the right (when balance == +1).
1003 	 */
1004 check_right_side:
1005 	if (node->avl_child[1] != NULL) {
1006 		ASSERT(AVL_XBALANCE(node) == 1);
1007 		parent = node;
1008 		node = node->avl_child[1];
1009 		ASSERT(node->avl_child[0] == NULL &&
1010 		    node->avl_child[1] == NULL);
1011 	} else {
1012 		ASSERT(AVL_XBALANCE(node) <= 0);
1013 	}
1014 
1015 done:
1016 	if (parent == NULL) {
1017 		*cookie = (void *)CHILDBIT;
1018 		ASSERT(node == tree->avl_root);
1019 	} else {
1020 		*cookie = (void *)((uintptr_t)parent | AVL_XCHILD(node));
1021 	}
1022 
1023 	return (AVL_NODE2DATA(node, off));
1024 }
1025 
1026 EXPORT_SYMBOL(avl_create);
1027 EXPORT_SYMBOL(avl_find);
1028 EXPORT_SYMBOL(avl_insert);
1029 EXPORT_SYMBOL(avl_insert_here);
1030 EXPORT_SYMBOL(avl_walk);
1031 EXPORT_SYMBOL(avl_first);
1032 EXPORT_SYMBOL(avl_last);
1033 EXPORT_SYMBOL(avl_nearest);
1034 EXPORT_SYMBOL(avl_add);
1035 EXPORT_SYMBOL(avl_swap);
1036 EXPORT_SYMBOL(avl_is_empty);
1037 EXPORT_SYMBOL(avl_remove);
1038 EXPORT_SYMBOL(avl_numnodes);
1039 EXPORT_SYMBOL(avl_destroy_nodes);
1040 EXPORT_SYMBOL(avl_destroy);
1041 EXPORT_SYMBOL(avl_update_lt);
1042 EXPORT_SYMBOL(avl_update_gt);
1043 EXPORT_SYMBOL(avl_update);
1044