xref: /freebsd/sys/contrib/openzfs/module/zfs/range_tree.c (revision a0b956f5ac5e0941f9e74e24c1c53e05ad061a38)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * Copyright (c) 2013, 2019 by Delphix. All rights reserved.
27  * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
28  */
29 
30 #include <sys/zfs_context.h>
31 #include <sys/spa.h>
32 #include <sys/dmu.h>
33 #include <sys/dnode.h>
34 #include <sys/zio.h>
35 #include <sys/range_tree.h>
36 
37 /*
38  * Range trees are tree-based data structures that can be used to
39  * track free space or generally any space allocation information.
40  * A range tree keeps track of individual segments and automatically
41  * provides facilities such as adjacent extent merging and extent
42  * splitting in response to range add/remove requests.
43  *
44  * A range tree starts out completely empty, with no segments in it.
45  * Adding an allocation via range_tree_add to the range tree can either:
46  * 1) create a new extent
47  * 2) extend an adjacent extent
48  * 3) merge two adjacent extents
49  * Conversely, removing an allocation via range_tree_remove can:
50  * 1) completely remove an extent
51  * 2) shorten an extent (if the allocation was near one of its ends)
52  * 3) split an extent into two extents, in effect punching a hole
53  *
54  * A range tree is also capable of 'bridging' gaps when adding
55  * allocations. This is useful for cases when close proximity of
56  * allocations is an important detail that needs to be represented
57  * in the range tree. See range_tree_set_gap(). The default behavior
58  * is not to bridge gaps (i.e. the maximum allowed gap size is 0).
59  *
60  * In order to traverse a range tree, use either the range_tree_walk()
61  * or range_tree_vacate() functions.
62  *
63  * To obtain more accurate information on individual segment
64  * operations that the range tree performs "under the hood", you can
65  * specify a set of callbacks by passing a range_tree_ops_t structure
66  * to the range_tree_create function. Any callbacks that are non-NULL
67  * are then called at the appropriate times.
68  *
69  * The range tree code also supports a special variant of range trees
70  * that can bridge small gaps between segments. This kind of tree is used
71  * by the dsl scanning code to group I/Os into mostly sequential chunks to
72  * optimize disk performance. The code here attempts to do this with as
73  * little memory and computational overhead as possible. One limitation of
74  * this implementation is that segments of range trees with gaps can only
75  * support removing complete segments.
76  */
77 
78 static inline void
79 rs_copy(range_seg_t *src, range_seg_t *dest, range_tree_t *rt)
80 {
81 	ASSERT3U(rt->rt_type, <, RANGE_SEG_NUM_TYPES);
82 	size_t size = 0;
83 	switch (rt->rt_type) {
84 	case RANGE_SEG32:
85 		size = sizeof (range_seg32_t);
86 		break;
87 	case RANGE_SEG64:
88 		size = sizeof (range_seg64_t);
89 		break;
90 	case RANGE_SEG_GAP:
91 		size = sizeof (range_seg_gap_t);
92 		break;
93 	default:
94 		__builtin_unreachable();
95 	}
96 	memcpy(dest, src, size);
97 }
98 
99 void
100 range_tree_stat_verify(range_tree_t *rt)
101 {
102 	range_seg_t *rs;
103 	zfs_btree_index_t where;
104 	uint64_t hist[RANGE_TREE_HISTOGRAM_SIZE] = { 0 };
105 	int i;
106 
107 	for (rs = zfs_btree_first(&rt->rt_root, &where); rs != NULL;
108 	    rs = zfs_btree_next(&rt->rt_root, &where, &where)) {
109 		uint64_t size = rs_get_end(rs, rt) - rs_get_start(rs, rt);
110 		int idx	= highbit64(size) - 1;
111 
112 		hist[idx]++;
113 		ASSERT3U(hist[idx], !=, 0);
114 	}
115 
116 	for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
117 		if (hist[i] != rt->rt_histogram[i]) {
118 			zfs_dbgmsg("i=%d, hist=%px, hist=%llu, rt_hist=%llu",
119 			    i, hist, (u_longlong_t)hist[i],
120 			    (u_longlong_t)rt->rt_histogram[i]);
121 		}
122 		VERIFY3U(hist[i], ==, rt->rt_histogram[i]);
123 	}
124 }
125 
126 static void
127 range_tree_stat_incr(range_tree_t *rt, range_seg_t *rs)
128 {
129 	uint64_t size = rs_get_end(rs, rt) - rs_get_start(rs, rt);
130 	int idx = highbit64(size) - 1;
131 
132 	ASSERT(size != 0);
133 	ASSERT3U(idx, <,
134 	    sizeof (rt->rt_histogram) / sizeof (*rt->rt_histogram));
135 
136 	rt->rt_histogram[idx]++;
137 	ASSERT3U(rt->rt_histogram[idx], !=, 0);
138 }
139 
140 static void
141 range_tree_stat_decr(range_tree_t *rt, range_seg_t *rs)
142 {
143 	uint64_t size = rs_get_end(rs, rt) - rs_get_start(rs, rt);
144 	int idx = highbit64(size) - 1;
145 
146 	ASSERT(size != 0);
147 	ASSERT3U(idx, <,
148 	    sizeof (rt->rt_histogram) / sizeof (*rt->rt_histogram));
149 
150 	ASSERT3U(rt->rt_histogram[idx], !=, 0);
151 	rt->rt_histogram[idx]--;
152 }
153 
154 static int
155 range_tree_seg32_compare(const void *x1, const void *x2)
156 {
157 	const range_seg32_t *r1 = x1;
158 	const range_seg32_t *r2 = x2;
159 
160 	ASSERT3U(r1->rs_start, <=, r1->rs_end);
161 	ASSERT3U(r2->rs_start, <=, r2->rs_end);
162 
163 	return ((r1->rs_start >= r2->rs_end) - (r1->rs_end <= r2->rs_start));
164 }
165 
166 static int
167 range_tree_seg64_compare(const void *x1, const void *x2)
168 {
169 	const range_seg64_t *r1 = x1;
170 	const range_seg64_t *r2 = x2;
171 
172 	ASSERT3U(r1->rs_start, <=, r1->rs_end);
173 	ASSERT3U(r2->rs_start, <=, r2->rs_end);
174 
175 	return ((r1->rs_start >= r2->rs_end) - (r1->rs_end <= r2->rs_start));
176 }
177 
178 static int
179 range_tree_seg_gap_compare(const void *x1, const void *x2)
180 {
181 	const range_seg_gap_t *r1 = x1;
182 	const range_seg_gap_t *r2 = x2;
183 
184 	ASSERT3U(r1->rs_start, <=, r1->rs_end);
185 	ASSERT3U(r2->rs_start, <=, r2->rs_end);
186 
187 	return ((r1->rs_start >= r2->rs_end) - (r1->rs_end <= r2->rs_start));
188 }
189 
190 range_tree_t *
191 range_tree_create_gap(const range_tree_ops_t *ops, range_seg_type_t type,
192     void *arg, uint64_t start, uint64_t shift, uint64_t gap)
193 {
194 	range_tree_t *rt = kmem_zalloc(sizeof (range_tree_t), KM_SLEEP);
195 
196 	ASSERT3U(shift, <, 64);
197 	ASSERT3U(type, <=, RANGE_SEG_NUM_TYPES);
198 	size_t size;
199 	int (*compare) (const void *, const void *);
200 	switch (type) {
201 	case RANGE_SEG32:
202 		size = sizeof (range_seg32_t);
203 		compare = range_tree_seg32_compare;
204 		break;
205 	case RANGE_SEG64:
206 		size = sizeof (range_seg64_t);
207 		compare = range_tree_seg64_compare;
208 		break;
209 	case RANGE_SEG_GAP:
210 		size = sizeof (range_seg_gap_t);
211 		compare = range_tree_seg_gap_compare;
212 		break;
213 	default:
214 		panic("Invalid range seg type %d", type);
215 	}
216 	zfs_btree_create(&rt->rt_root, compare, size);
217 
218 	rt->rt_ops = ops;
219 	rt->rt_gap = gap;
220 	rt->rt_arg = arg;
221 	rt->rt_type = type;
222 	rt->rt_start = start;
223 	rt->rt_shift = shift;
224 
225 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_create != NULL)
226 		rt->rt_ops->rtop_create(rt, rt->rt_arg);
227 
228 	return (rt);
229 }
230 
231 range_tree_t *
232 range_tree_create(const range_tree_ops_t *ops, range_seg_type_t type,
233     void *arg, uint64_t start, uint64_t shift)
234 {
235 	return (range_tree_create_gap(ops, type, arg, start, shift, 0));
236 }
237 
238 void
239 range_tree_destroy(range_tree_t *rt)
240 {
241 	VERIFY0(rt->rt_space);
242 
243 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_destroy != NULL)
244 		rt->rt_ops->rtop_destroy(rt, rt->rt_arg);
245 
246 	zfs_btree_destroy(&rt->rt_root);
247 	kmem_free(rt, sizeof (*rt));
248 }
249 
250 void
251 range_tree_adjust_fill(range_tree_t *rt, range_seg_t *rs, int64_t delta)
252 {
253 	if (delta < 0 && delta * -1 >= rs_get_fill(rs, rt)) {
254 		zfs_panic_recover("zfs: attempting to decrease fill to or "
255 		    "below 0; probable double remove in segment [%llx:%llx]",
256 		    (longlong_t)rs_get_start(rs, rt),
257 		    (longlong_t)rs_get_end(rs, rt));
258 	}
259 	if (rs_get_fill(rs, rt) + delta > rs_get_end(rs, rt) -
260 	    rs_get_start(rs, rt)) {
261 		zfs_panic_recover("zfs: attempting to increase fill beyond "
262 		    "max; probable double add in segment [%llx:%llx]",
263 		    (longlong_t)rs_get_start(rs, rt),
264 		    (longlong_t)rs_get_end(rs, rt));
265 	}
266 
267 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_remove != NULL)
268 		rt->rt_ops->rtop_remove(rt, rs, rt->rt_arg);
269 	rs_set_fill(rs, rt, rs_get_fill(rs, rt) + delta);
270 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_add != NULL)
271 		rt->rt_ops->rtop_add(rt, rs, rt->rt_arg);
272 }
273 
274 static void
275 range_tree_add_impl(void *arg, uint64_t start, uint64_t size, uint64_t fill)
276 {
277 	range_tree_t *rt = arg;
278 	zfs_btree_index_t where;
279 	range_seg_t *rs_before, *rs_after, *rs;
280 	range_seg_max_t tmp, rsearch;
281 	uint64_t end = start + size, gap = rt->rt_gap;
282 	uint64_t bridge_size = 0;
283 	boolean_t merge_before, merge_after;
284 
285 	ASSERT3U(size, !=, 0);
286 	ASSERT3U(fill, <=, size);
287 	ASSERT3U(start + size, >, start);
288 
289 	rs_set_start(&rsearch, rt, start);
290 	rs_set_end(&rsearch, rt, end);
291 	rs = zfs_btree_find(&rt->rt_root, &rsearch, &where);
292 
293 	/*
294 	 * If this is a gap-supporting range tree, it is possible that we
295 	 * are inserting into an existing segment. In this case simply
296 	 * bump the fill count and call the remove / add callbacks. If the
297 	 * new range will extend an existing segment, we remove the
298 	 * existing one, apply the new extent to it and re-insert it using
299 	 * the normal code paths.
300 	 */
301 	if (rs != NULL) {
302 		if (gap == 0) {
303 			zfs_panic_recover("zfs: adding existent segment to "
304 			    "range tree (offset=%llx size=%llx)",
305 			    (longlong_t)start, (longlong_t)size);
306 			return;
307 		}
308 		uint64_t rstart = rs_get_start(rs, rt);
309 		uint64_t rend = rs_get_end(rs, rt);
310 		if (rstart <= start && rend >= end) {
311 			range_tree_adjust_fill(rt, rs, fill);
312 			return;
313 		}
314 
315 		if (rt->rt_ops != NULL && rt->rt_ops->rtop_remove != NULL)
316 			rt->rt_ops->rtop_remove(rt, rs, rt->rt_arg);
317 
318 		range_tree_stat_decr(rt, rs);
319 		rt->rt_space -= rend - rstart;
320 
321 		fill += rs_get_fill(rs, rt);
322 		start = MIN(start, rstart);
323 		end = MAX(end, rend);
324 		size = end - start;
325 
326 		zfs_btree_remove(&rt->rt_root, rs);
327 		range_tree_add_impl(rt, start, size, fill);
328 		return;
329 	}
330 
331 	ASSERT3P(rs, ==, NULL);
332 
333 	/*
334 	 * Determine whether or not we will have to merge with our neighbors.
335 	 * If gap != 0, we might need to merge with our neighbors even if we
336 	 * aren't directly touching.
337 	 */
338 	zfs_btree_index_t where_before, where_after;
339 	rs_before = zfs_btree_prev(&rt->rt_root, &where, &where_before);
340 	rs_after = zfs_btree_next(&rt->rt_root, &where, &where_after);
341 
342 	merge_before = (rs_before != NULL && rs_get_end(rs_before, rt) >=
343 	    start - gap);
344 	merge_after = (rs_after != NULL && rs_get_start(rs_after, rt) <= end +
345 	    gap);
346 
347 	if (merge_before && gap != 0)
348 		bridge_size += start - rs_get_end(rs_before, rt);
349 	if (merge_after && gap != 0)
350 		bridge_size += rs_get_start(rs_after, rt) - end;
351 
352 	if (merge_before && merge_after) {
353 		if (rt->rt_ops != NULL && rt->rt_ops->rtop_remove != NULL) {
354 			rt->rt_ops->rtop_remove(rt, rs_before, rt->rt_arg);
355 			rt->rt_ops->rtop_remove(rt, rs_after, rt->rt_arg);
356 		}
357 
358 		range_tree_stat_decr(rt, rs_before);
359 		range_tree_stat_decr(rt, rs_after);
360 
361 		rs_copy(rs_after, &tmp, rt);
362 		uint64_t before_start = rs_get_start_raw(rs_before, rt);
363 		uint64_t before_fill = rs_get_fill(rs_before, rt);
364 		uint64_t after_fill = rs_get_fill(rs_after, rt);
365 		zfs_btree_remove_idx(&rt->rt_root, &where_before);
366 
367 		/*
368 		 * We have to re-find the node because our old reference is
369 		 * invalid as soon as we do any mutating btree operations.
370 		 */
371 		rs_after = zfs_btree_find(&rt->rt_root, &tmp, &where_after);
372 		rs_set_start_raw(rs_after, rt, before_start);
373 		rs_set_fill(rs_after, rt, after_fill + before_fill + fill);
374 		rs = rs_after;
375 	} else if (merge_before) {
376 		if (rt->rt_ops != NULL && rt->rt_ops->rtop_remove != NULL)
377 			rt->rt_ops->rtop_remove(rt, rs_before, rt->rt_arg);
378 
379 		range_tree_stat_decr(rt, rs_before);
380 
381 		uint64_t before_fill = rs_get_fill(rs_before, rt);
382 		rs_set_end(rs_before, rt, end);
383 		rs_set_fill(rs_before, rt, before_fill + fill);
384 		rs = rs_before;
385 	} else if (merge_after) {
386 		if (rt->rt_ops != NULL && rt->rt_ops->rtop_remove != NULL)
387 			rt->rt_ops->rtop_remove(rt, rs_after, rt->rt_arg);
388 
389 		range_tree_stat_decr(rt, rs_after);
390 
391 		uint64_t after_fill = rs_get_fill(rs_after, rt);
392 		rs_set_start(rs_after, rt, start);
393 		rs_set_fill(rs_after, rt, after_fill + fill);
394 		rs = rs_after;
395 	} else {
396 		rs = &tmp;
397 
398 		rs_set_start(rs, rt, start);
399 		rs_set_end(rs, rt, end);
400 		rs_set_fill(rs, rt, fill);
401 		zfs_btree_add_idx(&rt->rt_root, rs, &where);
402 	}
403 
404 	if (gap != 0) {
405 		ASSERT3U(rs_get_fill(rs, rt), <=, rs_get_end(rs, rt) -
406 		    rs_get_start(rs, rt));
407 	} else {
408 		ASSERT3U(rs_get_fill(rs, rt), ==, rs_get_end(rs, rt) -
409 		    rs_get_start(rs, rt));
410 	}
411 
412 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_add != NULL)
413 		rt->rt_ops->rtop_add(rt, rs, rt->rt_arg);
414 
415 	range_tree_stat_incr(rt, rs);
416 	rt->rt_space += size + bridge_size;
417 }
418 
419 void
420 range_tree_add(void *arg, uint64_t start, uint64_t size)
421 {
422 	range_tree_add_impl(arg, start, size, size);
423 }
424 
425 static void
426 range_tree_remove_impl(range_tree_t *rt, uint64_t start, uint64_t size,
427     boolean_t do_fill)
428 {
429 	zfs_btree_index_t where;
430 	range_seg_t *rs;
431 	range_seg_max_t rsearch, rs_tmp;
432 	uint64_t end = start + size;
433 	boolean_t left_over, right_over;
434 
435 	VERIFY3U(size, !=, 0);
436 	VERIFY3U(size, <=, rt->rt_space);
437 	if (rt->rt_type == RANGE_SEG64)
438 		ASSERT3U(start + size, >, start);
439 
440 	rs_set_start(&rsearch, rt, start);
441 	rs_set_end(&rsearch, rt, end);
442 	rs = zfs_btree_find(&rt->rt_root, &rsearch, &where);
443 
444 	/* Make sure we completely overlap with someone */
445 	if (rs == NULL) {
446 		zfs_panic_recover("zfs: removing nonexistent segment from "
447 		    "range tree (offset=%llx size=%llx)",
448 		    (longlong_t)start, (longlong_t)size);
449 		return;
450 	}
451 
452 	/*
453 	 * Range trees with gap support must only remove complete segments
454 	 * from the tree. This allows us to maintain accurate fill accounting
455 	 * and to ensure that bridged sections are not leaked. If we need to
456 	 * remove less than the full segment, we can only adjust the fill count.
457 	 */
458 	if (rt->rt_gap != 0) {
459 		if (do_fill) {
460 			if (rs_get_fill(rs, rt) == size) {
461 				start = rs_get_start(rs, rt);
462 				end = rs_get_end(rs, rt);
463 				size = end - start;
464 			} else {
465 				range_tree_adjust_fill(rt, rs, -size);
466 				return;
467 			}
468 		} else if (rs_get_start(rs, rt) != start ||
469 		    rs_get_end(rs, rt) != end) {
470 			zfs_panic_recover("zfs: freeing partial segment of "
471 			    "gap tree (offset=%llx size=%llx) of "
472 			    "(offset=%llx size=%llx)",
473 			    (longlong_t)start, (longlong_t)size,
474 			    (longlong_t)rs_get_start(rs, rt),
475 			    (longlong_t)rs_get_end(rs, rt) - rs_get_start(rs,
476 			    rt));
477 			return;
478 		}
479 	}
480 
481 	VERIFY3U(rs_get_start(rs, rt), <=, start);
482 	VERIFY3U(rs_get_end(rs, rt), >=, end);
483 
484 	left_over = (rs_get_start(rs, rt) != start);
485 	right_over = (rs_get_end(rs, rt) != end);
486 
487 	range_tree_stat_decr(rt, rs);
488 
489 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_remove != NULL)
490 		rt->rt_ops->rtop_remove(rt, rs, rt->rt_arg);
491 
492 	if (left_over && right_over) {
493 		range_seg_max_t newseg;
494 		rs_set_start(&newseg, rt, end);
495 		rs_set_end_raw(&newseg, rt, rs_get_end_raw(rs, rt));
496 		rs_set_fill(&newseg, rt, rs_get_end(rs, rt) - end);
497 		range_tree_stat_incr(rt, &newseg);
498 
499 		// This modifies the buffer already inside the range tree
500 		rs_set_end(rs, rt, start);
501 
502 		rs_copy(rs, &rs_tmp, rt);
503 		if (zfs_btree_next(&rt->rt_root, &where, &where) != NULL)
504 			zfs_btree_add_idx(&rt->rt_root, &newseg, &where);
505 		else
506 			zfs_btree_add(&rt->rt_root, &newseg);
507 
508 		if (rt->rt_ops != NULL && rt->rt_ops->rtop_add != NULL)
509 			rt->rt_ops->rtop_add(rt, &newseg, rt->rt_arg);
510 	} else if (left_over) {
511 		// This modifies the buffer already inside the range tree
512 		rs_set_end(rs, rt, start);
513 		rs_copy(rs, &rs_tmp, rt);
514 	} else if (right_over) {
515 		// This modifies the buffer already inside the range tree
516 		rs_set_start(rs, rt, end);
517 		rs_copy(rs, &rs_tmp, rt);
518 	} else {
519 		zfs_btree_remove_idx(&rt->rt_root, &where);
520 		rs = NULL;
521 	}
522 
523 	if (rs != NULL) {
524 		/*
525 		 * The fill of the leftover segment will always be equal to
526 		 * the size, since we do not support removing partial segments
527 		 * of range trees with gaps.
528 		 */
529 		rs_set_fill_raw(rs, rt, rs_get_end_raw(rs, rt) -
530 		    rs_get_start_raw(rs, rt));
531 		range_tree_stat_incr(rt, &rs_tmp);
532 
533 		if (rt->rt_ops != NULL && rt->rt_ops->rtop_add != NULL)
534 			rt->rt_ops->rtop_add(rt, &rs_tmp, rt->rt_arg);
535 	}
536 
537 	rt->rt_space -= size;
538 }
539 
540 void
541 range_tree_remove(void *arg, uint64_t start, uint64_t size)
542 {
543 	range_tree_remove_impl(arg, start, size, B_FALSE);
544 }
545 
546 void
547 range_tree_remove_fill(range_tree_t *rt, uint64_t start, uint64_t size)
548 {
549 	range_tree_remove_impl(rt, start, size, B_TRUE);
550 }
551 
552 void
553 range_tree_resize_segment(range_tree_t *rt, range_seg_t *rs,
554     uint64_t newstart, uint64_t newsize)
555 {
556 	int64_t delta = newsize - (rs_get_end(rs, rt) - rs_get_start(rs, rt));
557 
558 	range_tree_stat_decr(rt, rs);
559 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_remove != NULL)
560 		rt->rt_ops->rtop_remove(rt, rs, rt->rt_arg);
561 
562 	rs_set_start(rs, rt, newstart);
563 	rs_set_end(rs, rt, newstart + newsize);
564 
565 	range_tree_stat_incr(rt, rs);
566 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_add != NULL)
567 		rt->rt_ops->rtop_add(rt, rs, rt->rt_arg);
568 
569 	rt->rt_space += delta;
570 }
571 
572 static range_seg_t *
573 range_tree_find_impl(range_tree_t *rt, uint64_t start, uint64_t size)
574 {
575 	range_seg_max_t rsearch;
576 	uint64_t end = start + size;
577 
578 	VERIFY(size != 0);
579 
580 	rs_set_start(&rsearch, rt, start);
581 	rs_set_end(&rsearch, rt, end);
582 	return (zfs_btree_find(&rt->rt_root, &rsearch, NULL));
583 }
584 
585 range_seg_t *
586 range_tree_find(range_tree_t *rt, uint64_t start, uint64_t size)
587 {
588 	if (rt->rt_type == RANGE_SEG64)
589 		ASSERT3U(start + size, >, start);
590 
591 	range_seg_t *rs = range_tree_find_impl(rt, start, size);
592 	if (rs != NULL && rs_get_start(rs, rt) <= start &&
593 	    rs_get_end(rs, rt) >= start + size) {
594 		return (rs);
595 	}
596 	return (NULL);
597 }
598 
599 void
600 range_tree_verify_not_present(range_tree_t *rt, uint64_t off, uint64_t size)
601 {
602 	range_seg_t *rs = range_tree_find(rt, off, size);
603 	if (rs != NULL)
604 		panic("segment already in tree; rs=%p", (void *)rs);
605 }
606 
607 boolean_t
608 range_tree_contains(range_tree_t *rt, uint64_t start, uint64_t size)
609 {
610 	return (range_tree_find(rt, start, size) != NULL);
611 }
612 
613 /*
614  * Returns the first subset of the given range which overlaps with the range
615  * tree. Returns true if there is a segment in the range, and false if there
616  * isn't.
617  */
618 boolean_t
619 range_tree_find_in(range_tree_t *rt, uint64_t start, uint64_t size,
620     uint64_t *ostart, uint64_t *osize)
621 {
622 	if (rt->rt_type == RANGE_SEG64)
623 		ASSERT3U(start + size, >, start);
624 
625 	range_seg_max_t rsearch;
626 	rs_set_start(&rsearch, rt, start);
627 	rs_set_end_raw(&rsearch, rt, rs_get_start_raw(&rsearch, rt) + 1);
628 
629 	zfs_btree_index_t where;
630 	range_seg_t *rs = zfs_btree_find(&rt->rt_root, &rsearch, &where);
631 	if (rs != NULL) {
632 		*ostart = start;
633 		*osize = MIN(size, rs_get_end(rs, rt) - start);
634 		return (B_TRUE);
635 	}
636 
637 	rs = zfs_btree_next(&rt->rt_root, &where, &where);
638 	if (rs == NULL || rs_get_start(rs, rt) > start + size)
639 		return (B_FALSE);
640 
641 	*ostart = rs_get_start(rs, rt);
642 	*osize = MIN(start + size, rs_get_end(rs, rt)) -
643 	    rs_get_start(rs, rt);
644 	return (B_TRUE);
645 }
646 
647 /*
648  * Ensure that this range is not in the tree, regardless of whether
649  * it is currently in the tree.
650  */
651 void
652 range_tree_clear(range_tree_t *rt, uint64_t start, uint64_t size)
653 {
654 	range_seg_t *rs;
655 
656 	if (size == 0)
657 		return;
658 
659 	if (rt->rt_type == RANGE_SEG64)
660 		ASSERT3U(start + size, >, start);
661 
662 	while ((rs = range_tree_find_impl(rt, start, size)) != NULL) {
663 		uint64_t free_start = MAX(rs_get_start(rs, rt), start);
664 		uint64_t free_end = MIN(rs_get_end(rs, rt), start + size);
665 		range_tree_remove(rt, free_start, free_end - free_start);
666 	}
667 }
668 
669 void
670 range_tree_swap(range_tree_t **rtsrc, range_tree_t **rtdst)
671 {
672 	range_tree_t *rt;
673 
674 	ASSERT0(range_tree_space(*rtdst));
675 	ASSERT0(zfs_btree_numnodes(&(*rtdst)->rt_root));
676 
677 	rt = *rtsrc;
678 	*rtsrc = *rtdst;
679 	*rtdst = rt;
680 }
681 
682 void
683 range_tree_vacate(range_tree_t *rt, range_tree_func_t *func, void *arg)
684 {
685 	if (rt->rt_ops != NULL && rt->rt_ops->rtop_vacate != NULL)
686 		rt->rt_ops->rtop_vacate(rt, rt->rt_arg);
687 
688 	if (func != NULL) {
689 		range_seg_t *rs;
690 		zfs_btree_index_t *cookie = NULL;
691 
692 		while ((rs = zfs_btree_destroy_nodes(&rt->rt_root, &cookie)) !=
693 		    NULL) {
694 			func(arg, rs_get_start(rs, rt), rs_get_end(rs, rt) -
695 			    rs_get_start(rs, rt));
696 		}
697 	} else {
698 		zfs_btree_clear(&rt->rt_root);
699 	}
700 
701 	memset(rt->rt_histogram, 0, sizeof (rt->rt_histogram));
702 	rt->rt_space = 0;
703 }
704 
705 void
706 range_tree_walk(range_tree_t *rt, range_tree_func_t *func, void *arg)
707 {
708 	zfs_btree_index_t where;
709 	for (range_seg_t *rs = zfs_btree_first(&rt->rt_root, &where);
710 	    rs != NULL; rs = zfs_btree_next(&rt->rt_root, &where, &where)) {
711 		func(arg, rs_get_start(rs, rt), rs_get_end(rs, rt) -
712 		    rs_get_start(rs, rt));
713 	}
714 }
715 
716 range_seg_t *
717 range_tree_first(range_tree_t *rt)
718 {
719 	return (zfs_btree_first(&rt->rt_root, NULL));
720 }
721 
722 uint64_t
723 range_tree_space(range_tree_t *rt)
724 {
725 	return (rt->rt_space);
726 }
727 
728 uint64_t
729 range_tree_numsegs(range_tree_t *rt)
730 {
731 	return ((rt == NULL) ? 0 : zfs_btree_numnodes(&rt->rt_root));
732 }
733 
734 boolean_t
735 range_tree_is_empty(range_tree_t *rt)
736 {
737 	ASSERT(rt != NULL);
738 	return (range_tree_space(rt) == 0);
739 }
740 
741 /*
742  * Remove any overlapping ranges between the given segment [start, end)
743  * from removefrom. Add non-overlapping leftovers to addto.
744  */
745 void
746 range_tree_remove_xor_add_segment(uint64_t start, uint64_t end,
747     range_tree_t *removefrom, range_tree_t *addto)
748 {
749 	zfs_btree_index_t where;
750 	range_seg_max_t starting_rs;
751 	rs_set_start(&starting_rs, removefrom, start);
752 	rs_set_end_raw(&starting_rs, removefrom, rs_get_start_raw(&starting_rs,
753 	    removefrom) + 1);
754 
755 	range_seg_t *curr = zfs_btree_find(&removefrom->rt_root,
756 	    &starting_rs, &where);
757 
758 	if (curr == NULL)
759 		curr = zfs_btree_next(&removefrom->rt_root, &where, &where);
760 
761 	range_seg_t *next;
762 	for (; curr != NULL; curr = next) {
763 		if (start == end)
764 			return;
765 		VERIFY3U(start, <, end);
766 
767 		/* there is no overlap */
768 		if (end <= rs_get_start(curr, removefrom)) {
769 			range_tree_add(addto, start, end - start);
770 			return;
771 		}
772 
773 		uint64_t overlap_start = MAX(rs_get_start(curr, removefrom),
774 		    start);
775 		uint64_t overlap_end = MIN(rs_get_end(curr, removefrom),
776 		    end);
777 		uint64_t overlap_size = overlap_end - overlap_start;
778 		ASSERT3S(overlap_size, >, 0);
779 		range_seg_max_t rs;
780 		rs_copy(curr, &rs, removefrom);
781 
782 		range_tree_remove(removefrom, overlap_start, overlap_size);
783 
784 		if (start < overlap_start)
785 			range_tree_add(addto, start, overlap_start - start);
786 
787 		start = overlap_end;
788 		next = zfs_btree_find(&removefrom->rt_root, &rs, &where);
789 		/*
790 		 * If we find something here, we only removed part of the
791 		 * curr segment. Either there's some left at the end
792 		 * because we've reached the end of the range we're removing,
793 		 * or there's some left at the start because we started
794 		 * partway through the range.  Either way, we continue with
795 		 * the loop. If it's the former, we'll return at the start of
796 		 * the loop, and if it's the latter we'll see if there is more
797 		 * area to process.
798 		 */
799 		if (next != NULL) {
800 			ASSERT(start == end || start == rs_get_end(&rs,
801 			    removefrom));
802 		}
803 
804 		next = zfs_btree_next(&removefrom->rt_root, &where, &where);
805 	}
806 	VERIFY3P(curr, ==, NULL);
807 
808 	if (start != end) {
809 		VERIFY3U(start, <, end);
810 		range_tree_add(addto, start, end - start);
811 	} else {
812 		VERIFY3U(start, ==, end);
813 	}
814 }
815 
816 /*
817  * For each entry in rt, if it exists in removefrom, remove it
818  * from removefrom. Otherwise, add it to addto.
819  */
820 void
821 range_tree_remove_xor_add(range_tree_t *rt, range_tree_t *removefrom,
822     range_tree_t *addto)
823 {
824 	zfs_btree_index_t where;
825 	for (range_seg_t *rs = zfs_btree_first(&rt->rt_root, &where); rs;
826 	    rs = zfs_btree_next(&rt->rt_root, &where, &where)) {
827 		range_tree_remove_xor_add_segment(rs_get_start(rs, rt),
828 		    rs_get_end(rs, rt), removefrom, addto);
829 	}
830 }
831 
832 uint64_t
833 range_tree_min(range_tree_t *rt)
834 {
835 	range_seg_t *rs = zfs_btree_first(&rt->rt_root, NULL);
836 	return (rs != NULL ? rs_get_start(rs, rt) : 0);
837 }
838 
839 uint64_t
840 range_tree_max(range_tree_t *rt)
841 {
842 	range_seg_t *rs = zfs_btree_last(&rt->rt_root, NULL);
843 	return (rs != NULL ? rs_get_end(rs, rt) : 0);
844 }
845 
846 uint64_t
847 range_tree_span(range_tree_t *rt)
848 {
849 	return (range_tree_max(rt) - range_tree_min(rt));
850 }
851