xref: /linux/fs/btrfs/tests/extent-map-tests.c (revision b8fc42dc065742bc68df6a61a2aff8cbe364fa17)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2017 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/types.h>
7 #include "btrfs-tests.h"
8 #include "../ctree.h"
9 #include "../btrfs_inode.h"
10 #include "../volumes.h"
11 #include "../disk-io.h"
12 #include "../block-group.h"
13 
14 static int free_extent_map_tree(struct btrfs_inode *inode)
15 {
16 	struct extent_map_tree *em_tree = &inode->extent_tree;
17 	struct extent_map *em;
18 	struct rb_node *node;
19 	int ret = 0;
20 
21 	write_lock(&em_tree->lock);
22 	while (!RB_EMPTY_ROOT(&em_tree->root)) {
23 		node = rb_first(&em_tree->root);
24 		em = rb_entry(node, struct extent_map, rb_node);
25 		remove_extent_mapping(inode, em);
26 
27 #ifdef CONFIG_BTRFS_DEBUG
28 		if (refcount_read(&em->refs) != 1) {
29 			ret = -EINVAL;
30 			test_err(
31 "em leak: em (start %llu len %llu disk_bytenr %llu disk_num_bytes %llu offset %llu) refs %d",
32 				 em->start, em->len, em->disk_bytenr,
33 				 em->disk_num_bytes, em->offset,
34 				 refcount_read(&em->refs));
35 
36 			refcount_set(&em->refs, 1);
37 		}
38 #endif
39 		free_extent_map(em);
40 	}
41 	write_unlock(&em_tree->lock);
42 
43 	return ret;
44 }
45 
46 /*
47  * Test scenario:
48  *
49  * Suppose that no extent map has been loaded into memory yet, there is a file
50  * extent [0, 16K), followed by another file extent [16K, 20K), two dio reads
51  * are entering btrfs_get_extent() concurrently, t1 is reading [8K, 16K), t2 is
52  * reading [0, 8K)
53  *
54  *     t1                            t2
55  *  btrfs_get_extent()              btrfs_get_extent()
56  *    -> lookup_extent_mapping()      ->lookup_extent_mapping()
57  *    -> add_extent_mapping(0, 16K)
58  *    -> return em
59  *                                    ->add_extent_mapping(0, 16K)
60  *                                    -> #handle -EEXIST
61  */
62 static int test_case_1(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
63 {
64 	struct extent_map_tree *em_tree = &inode->extent_tree;
65 	struct extent_map *em;
66 	u64 start = 0;
67 	u64 len = SZ_8K;
68 	int ret;
69 	int ret2;
70 
71 	em = alloc_extent_map();
72 	if (!em) {
73 		test_std_err(TEST_ALLOC_EXTENT_MAP);
74 		return -ENOMEM;
75 	}
76 
77 	/* Add [0, 16K) */
78 	em->start = 0;
79 	em->len = SZ_16K;
80 	em->disk_bytenr = 0;
81 	em->disk_num_bytes = SZ_16K;
82 	em->ram_bytes = SZ_16K;
83 	write_lock(&em_tree->lock);
84 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
85 	write_unlock(&em_tree->lock);
86 	if (ret < 0) {
87 		test_err("cannot add extent range [0, 16K)");
88 		goto out;
89 	}
90 	free_extent_map(em);
91 
92 	/* Add [16K, 20K) following [0, 16K)  */
93 	em = alloc_extent_map();
94 	if (!em) {
95 		test_std_err(TEST_ALLOC_EXTENT_MAP);
96 		ret = -ENOMEM;
97 		goto out;
98 	}
99 
100 	em->start = SZ_16K;
101 	em->len = SZ_4K;
102 	em->disk_bytenr = SZ_32K; /* avoid merging */
103 	em->disk_num_bytes = SZ_4K;
104 	em->ram_bytes = SZ_4K;
105 	write_lock(&em_tree->lock);
106 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
107 	write_unlock(&em_tree->lock);
108 	if (ret < 0) {
109 		test_err("cannot add extent range [16K, 20K)");
110 		goto out;
111 	}
112 	free_extent_map(em);
113 
114 	em = alloc_extent_map();
115 	if (!em) {
116 		test_std_err(TEST_ALLOC_EXTENT_MAP);
117 		ret = -ENOMEM;
118 		goto out;
119 	}
120 
121 	/* Add [0, 8K), should return [0, 16K) instead. */
122 	em->start = start;
123 	em->len = len;
124 	em->disk_bytenr = start;
125 	em->disk_num_bytes = len;
126 	em->ram_bytes = len;
127 	write_lock(&em_tree->lock);
128 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
129 	write_unlock(&em_tree->lock);
130 	if (ret) {
131 		test_err("case1 [%llu %llu]: ret %d", start, start + len, ret);
132 		goto out;
133 	}
134 	if (!em) {
135 		test_err("case1 [%llu %llu]: no extent map returned",
136 			 start, start + len);
137 		ret = -ENOENT;
138 		goto out;
139 	}
140 	if (em->start != 0 || extent_map_end(em) != SZ_16K ||
141 	    em->disk_bytenr != 0 || em->disk_num_bytes != SZ_16K) {
142 		test_err(
143 "case1 [%llu %llu]: ret %d return a wrong em (start %llu len %llu disk_bytenr %llu disk_num_bytes %llu",
144 			 start, start + len, ret, em->start, em->len,
145 			 em->disk_bytenr, em->disk_num_bytes);
146 		ret = -EINVAL;
147 	}
148 	free_extent_map(em);
149 out:
150 	ret2 = free_extent_map_tree(inode);
151 	if (ret == 0)
152 		ret = ret2;
153 
154 	return ret;
155 }
156 
157 /*
158  * Test scenario:
159  *
160  * Reading the inline ending up with EEXIST, ie. read an inline
161  * extent and discard page cache and read it again.
162  */
163 static int test_case_2(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
164 {
165 	struct extent_map_tree *em_tree = &inode->extent_tree;
166 	struct extent_map *em;
167 	int ret;
168 	int ret2;
169 
170 	em = alloc_extent_map();
171 	if (!em) {
172 		test_std_err(TEST_ALLOC_EXTENT_MAP);
173 		return -ENOMEM;
174 	}
175 
176 	/* Add [0, 1K) */
177 	em->start = 0;
178 	em->len = SZ_1K;
179 	em->disk_bytenr = EXTENT_MAP_INLINE;
180 	em->disk_num_bytes = 0;
181 	em->ram_bytes = SZ_1K;
182 	write_lock(&em_tree->lock);
183 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
184 	write_unlock(&em_tree->lock);
185 	if (ret < 0) {
186 		test_err("cannot add extent range [0, 1K)");
187 		goto out;
188 	}
189 	free_extent_map(em);
190 
191 	/* Add [4K, 8K) following [0, 1K)  */
192 	em = alloc_extent_map();
193 	if (!em) {
194 		test_std_err(TEST_ALLOC_EXTENT_MAP);
195 		ret = -ENOMEM;
196 		goto out;
197 	}
198 
199 	em->start = SZ_4K;
200 	em->len = SZ_4K;
201 	em->disk_bytenr = SZ_4K;
202 	em->disk_num_bytes = SZ_4K;
203 	em->ram_bytes = SZ_4K;
204 	write_lock(&em_tree->lock);
205 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
206 	write_unlock(&em_tree->lock);
207 	if (ret < 0) {
208 		test_err("cannot add extent range [4K, 8K)");
209 		goto out;
210 	}
211 	free_extent_map(em);
212 
213 	em = alloc_extent_map();
214 	if (!em) {
215 		test_std_err(TEST_ALLOC_EXTENT_MAP);
216 		ret = -ENOMEM;
217 		goto out;
218 	}
219 
220 	/* Add [0, 1K) */
221 	em->start = 0;
222 	em->len = SZ_1K;
223 	em->disk_bytenr = EXTENT_MAP_INLINE;
224 	em->disk_num_bytes = 0;
225 	em->ram_bytes = SZ_1K;
226 	write_lock(&em_tree->lock);
227 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
228 	write_unlock(&em_tree->lock);
229 	if (ret) {
230 		test_err("case2 [0 1K]: ret %d", ret);
231 		goto out;
232 	}
233 	if (!em) {
234 		test_err("case2 [0 1K]: no extent map returned");
235 		ret = -ENOENT;
236 		goto out;
237 	}
238 	if (em->start != 0 || extent_map_end(em) != SZ_1K ||
239 	    em->disk_bytenr != EXTENT_MAP_INLINE) {
240 		test_err(
241 "case2 [0 1K]: ret %d return a wrong em (start %llu len %llu disk_bytenr %llu",
242 			 ret, em->start, em->len, em->disk_bytenr);
243 		ret = -EINVAL;
244 	}
245 	free_extent_map(em);
246 out:
247 	ret2 = free_extent_map_tree(inode);
248 	if (ret == 0)
249 		ret = ret2;
250 
251 	return ret;
252 }
253 
254 static int __test_case_3(struct btrfs_fs_info *fs_info,
255 			 struct btrfs_inode *inode, u64 start)
256 {
257 	struct extent_map_tree *em_tree = &inode->extent_tree;
258 	struct extent_map *em;
259 	u64 len = SZ_4K;
260 	int ret;
261 	int ret2;
262 
263 	em = alloc_extent_map();
264 	if (!em) {
265 		test_std_err(TEST_ALLOC_EXTENT_MAP);
266 		return -ENOMEM;
267 	}
268 
269 	/* Add [4K, 8K) */
270 	em->start = SZ_4K;
271 	em->len = SZ_4K;
272 	em->disk_bytenr = SZ_4K;
273 	em->disk_num_bytes = SZ_4K;
274 	em->ram_bytes = SZ_4K;
275 	write_lock(&em_tree->lock);
276 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
277 	write_unlock(&em_tree->lock);
278 	if (ret < 0) {
279 		test_err("cannot add extent range [4K, 8K)");
280 		goto out;
281 	}
282 	free_extent_map(em);
283 
284 	em = alloc_extent_map();
285 	if (!em) {
286 		test_std_err(TEST_ALLOC_EXTENT_MAP);
287 		ret = -ENOMEM;
288 		goto out;
289 	}
290 
291 	/* Add [0, 16K) */
292 	em->start = 0;
293 	em->len = SZ_16K;
294 	em->disk_bytenr = 0;
295 	em->disk_num_bytes = SZ_16K;
296 	em->ram_bytes = SZ_16K;
297 	write_lock(&em_tree->lock);
298 	ret = btrfs_add_extent_mapping(inode, &em, start, len);
299 	write_unlock(&em_tree->lock);
300 	if (ret) {
301 		test_err("case3 [%llu %llu): ret %d",
302 			 start, start + len, ret);
303 		goto out;
304 	}
305 	if (!em) {
306 		test_err("case3 [%llu %llu): no extent map returned",
307 			 start, start + len);
308 		ret = -ENOENT;
309 		goto out;
310 	}
311 	/*
312 	 * Since bytes within em are contiguous, em->block_start is identical to
313 	 * em->start.
314 	 */
315 	if (start < em->start || start + len > extent_map_end(em) ||
316 	    em->start != extent_map_block_start(em)) {
317 		test_err(
318 "case3 [%llu %llu): ret %d em (start %llu len %llu disk_bytenr %llu block_len %llu)",
319 			 start, start + len, ret, em->start, em->len,
320 			 em->disk_bytenr, em->disk_num_bytes);
321 		ret = -EINVAL;
322 	}
323 	free_extent_map(em);
324 out:
325 	ret2 = free_extent_map_tree(inode);
326 	if (ret == 0)
327 		ret = ret2;
328 
329 	return ret;
330 }
331 
332 /*
333  * Test scenario:
334  *
335  * Suppose that no extent map has been loaded into memory yet.
336  * There is a file extent [0, 16K), two jobs are running concurrently
337  * against it, t1 is buffered writing to [4K, 8K) and t2 is doing dio
338  * read from [0, 4K) or [8K, 12K) or [12K, 16K).
339  *
340  * t1 goes ahead of t2 and adds em [4K, 8K) into tree.
341  *
342  *         t1                       t2
343  *  cow_file_range()	     btrfs_get_extent()
344  *                            -> lookup_extent_mapping()
345  *   -> add_extent_mapping()
346  *                            -> add_extent_mapping()
347  */
348 static int test_case_3(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
349 {
350 	int ret;
351 
352 	ret = __test_case_3(fs_info, inode, 0);
353 	if (ret)
354 		return ret;
355 	ret = __test_case_3(fs_info, inode, SZ_8K);
356 	if (ret)
357 		return ret;
358 	ret = __test_case_3(fs_info, inode, (12 * SZ_1K));
359 
360 	return ret;
361 }
362 
363 static int __test_case_4(struct btrfs_fs_info *fs_info,
364 			 struct btrfs_inode *inode, u64 start)
365 {
366 	struct extent_map_tree *em_tree = &inode->extent_tree;
367 	struct extent_map *em;
368 	u64 len = SZ_4K;
369 	int ret;
370 	int ret2;
371 
372 	em = alloc_extent_map();
373 	if (!em) {
374 		test_std_err(TEST_ALLOC_EXTENT_MAP);
375 		return -ENOMEM;
376 	}
377 
378 	/* Add [0K, 8K) */
379 	em->start = 0;
380 	em->len = SZ_8K;
381 	em->disk_bytenr = 0;
382 	em->disk_num_bytes = SZ_8K;
383 	em->ram_bytes = SZ_8K;
384 	write_lock(&em_tree->lock);
385 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
386 	write_unlock(&em_tree->lock);
387 	if (ret < 0) {
388 		test_err("cannot add extent range [0, 8K)");
389 		goto out;
390 	}
391 	free_extent_map(em);
392 
393 	em = alloc_extent_map();
394 	if (!em) {
395 		test_std_err(TEST_ALLOC_EXTENT_MAP);
396 		ret = -ENOMEM;
397 		goto out;
398 	}
399 
400 	/* Add [8K, 32K) */
401 	em->start = SZ_8K;
402 	em->len = 24 * SZ_1K;
403 	em->disk_bytenr = SZ_16K; /* avoid merging */
404 	em->disk_num_bytes = 24 * SZ_1K;
405 	em->ram_bytes = 24 * SZ_1K;
406 	write_lock(&em_tree->lock);
407 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
408 	write_unlock(&em_tree->lock);
409 	if (ret < 0) {
410 		test_err("cannot add extent range [8K, 32K)");
411 		goto out;
412 	}
413 	free_extent_map(em);
414 
415 	em = alloc_extent_map();
416 	if (!em) {
417 		test_std_err(TEST_ALLOC_EXTENT_MAP);
418 		ret = -ENOMEM;
419 		goto out;
420 	}
421 	/* Add [0K, 32K) */
422 	em->start = 0;
423 	em->len = SZ_32K;
424 	em->disk_bytenr = 0;
425 	em->disk_num_bytes = SZ_32K;
426 	em->ram_bytes = SZ_32K;
427 	write_lock(&em_tree->lock);
428 	ret = btrfs_add_extent_mapping(inode, &em, start, len);
429 	write_unlock(&em_tree->lock);
430 	if (ret) {
431 		test_err("case4 [%llu %llu): ret %d",
432 			 start, start + len, ret);
433 		goto out;
434 	}
435 	if (!em) {
436 		test_err("case4 [%llu %llu): no extent map returned",
437 			 start, start + len);
438 		ret = -ENOENT;
439 		goto out;
440 	}
441 	if (start < em->start || start + len > extent_map_end(em)) {
442 		test_err(
443 "case4 [%llu %llu): ret %d, added wrong em (start %llu len %llu disk_bytenr %llu disk_num_bytes %llu)",
444 			 start, start + len, ret, em->start, em->len,
445 			 em->disk_bytenr, em->disk_num_bytes);
446 		ret = -EINVAL;
447 	}
448 	free_extent_map(em);
449 out:
450 	ret2 = free_extent_map_tree(inode);
451 	if (ret == 0)
452 		ret = ret2;
453 
454 	return ret;
455 }
456 
457 /*
458  * Test scenario:
459  *
460  * Suppose that no extent map has been loaded into memory yet.
461  * There is a file extent [0, 32K), two jobs are running concurrently
462  * against it, t1 is doing dio write to [8K, 32K) and t2 is doing dio
463  * read from [0, 4K) or [4K, 8K).
464  *
465  * t1 goes ahead of t2 and splits em [0, 32K) to em [0K, 8K) and [8K 32K).
466  *
467  *         t1                                t2
468  *  btrfs_get_blocks_direct()	       btrfs_get_blocks_direct()
469  *   -> btrfs_get_extent()              -> btrfs_get_extent()
470  *       -> lookup_extent_mapping()
471  *       -> add_extent_mapping()            -> lookup_extent_mapping()
472  *          # load [0, 32K)
473  *   -> btrfs_new_extent_direct()
474  *       -> btrfs_drop_extent_cache()
475  *          # split [0, 32K)
476  *       -> add_extent_mapping()
477  *          # add [8K, 32K)
478  *                                          -> add_extent_mapping()
479  *                                             # handle -EEXIST when adding
480  *                                             # [0, 32K)
481  */
482 static int test_case_4(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
483 {
484 	int ret;
485 
486 	ret = __test_case_4(fs_info, inode, 0);
487 	if (ret)
488 		return ret;
489 	ret = __test_case_4(fs_info, inode, SZ_4K);
490 
491 	return ret;
492 }
493 
494 static int add_compressed_extent(struct btrfs_inode *inode,
495 				 u64 start, u64 len, u64 block_start)
496 {
497 	struct extent_map_tree *em_tree = &inode->extent_tree;
498 	struct extent_map *em;
499 	int ret;
500 
501 	em = alloc_extent_map();
502 	if (!em) {
503 		test_std_err(TEST_ALLOC_EXTENT_MAP);
504 		return -ENOMEM;
505 	}
506 
507 	em->start = start;
508 	em->len = len;
509 	em->disk_bytenr = block_start;
510 	em->disk_num_bytes = SZ_4K;
511 	em->ram_bytes = len;
512 	em->flags |= EXTENT_FLAG_COMPRESS_ZLIB;
513 	write_lock(&em_tree->lock);
514 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
515 	write_unlock(&em_tree->lock);
516 	free_extent_map(em);
517 	if (ret < 0) {
518 		test_err("cannot add extent map [%llu, %llu)", start, start + len);
519 		return ret;
520 	}
521 
522 	return 0;
523 }
524 
525 struct extent_range {
526 	u64 start;
527 	u64 len;
528 };
529 
530 /* The valid states of the tree after every drop, as described below. */
531 struct extent_range valid_ranges[][7] = {
532 	{
533 	  { .start = 0,			.len = SZ_8K },		/* [0, 8K) */
534 	  { .start = SZ_4K * 3,		.len = SZ_4K * 3},	/* [12k, 24k) */
535 	  { .start = SZ_4K * 6,		.len = SZ_4K * 3},	/* [24k, 36k) */
536 	  { .start = SZ_32K + SZ_4K,	.len = SZ_4K},		/* [36k, 40k) */
537 	  { .start = SZ_4K * 10,	.len = SZ_4K * 6},	/* [40k, 64k) */
538 	},
539 	{
540 	  { .start = 0,			.len = SZ_8K },		/* [0, 8K) */
541 	  { .start = SZ_4K * 5,		.len = SZ_4K},		/* [20k, 24k) */
542 	  { .start = SZ_4K * 6,		.len = SZ_4K * 3},	/* [24k, 36k) */
543 	  { .start = SZ_32K + SZ_4K,	.len = SZ_4K},		/* [36k, 40k) */
544 	  { .start = SZ_4K * 10,	.len = SZ_4K * 6},	/* [40k, 64k) */
545 	},
546 	{
547 	  { .start = 0,			.len = SZ_8K },		/* [0, 8K) */
548 	  { .start = SZ_4K * 5,		.len = SZ_4K},		/* [20k, 24k) */
549 	  { .start = SZ_4K * 6,		.len = SZ_4K},		/* [24k, 28k) */
550 	  { .start = SZ_32K,		.len = SZ_4K},		/* [32k, 36k) */
551 	  { .start = SZ_32K + SZ_4K,	.len = SZ_4K},		/* [36k, 40k) */
552 	  { .start = SZ_4K * 10,	.len = SZ_4K * 6},	/* [40k, 64k) */
553 	},
554 	{
555 	  { .start = 0,			.len = SZ_8K},		/* [0, 8K) */
556 	  { .start = SZ_4K * 5,		.len = SZ_4K},		/* [20k, 24k) */
557 	  { .start = SZ_4K * 6,		.len = SZ_4K},		/* [24k, 28k) */
558 	}
559 };
560 
561 static int validate_range(struct extent_map_tree *em_tree, int index)
562 {
563 	struct rb_node *n;
564 	int i;
565 
566 	for (i = 0, n = rb_first(&em_tree->root);
567 	     valid_ranges[index][i].len && n;
568 	     i++, n = rb_next(n)) {
569 		struct extent_map *entry = rb_entry(n, struct extent_map, rb_node);
570 
571 		if (entry->start != valid_ranges[index][i].start) {
572 			test_err("mapping has start %llu expected %llu",
573 				 entry->start, valid_ranges[index][i].start);
574 			return -EINVAL;
575 		}
576 
577 		if (entry->len != valid_ranges[index][i].len) {
578 			test_err("mapping has len %llu expected %llu",
579 				 entry->len, valid_ranges[index][i].len);
580 			return -EINVAL;
581 		}
582 	}
583 
584 	/*
585 	 * We exited because we don't have any more entries in the extent_map
586 	 * but we still expect more valid entries.
587 	 */
588 	if (valid_ranges[index][i].len) {
589 		test_err("missing an entry");
590 		return -EINVAL;
591 	}
592 
593 	/* We exited the loop but still have entries in the extent map. */
594 	if (n) {
595 		test_err("we have a left over entry in the extent map we didn't expect");
596 		return -EINVAL;
597 	}
598 
599 	return 0;
600 }
601 
602 /*
603  * Test scenario:
604  *
605  * Test the various edge cases of btrfs_drop_extent_map_range, create the
606  * following ranges
607  *
608  * [0, 12k)[12k, 24k)[24k, 36k)[36k, 40k)[40k,64k)
609  *
610  * And then we'll drop:
611  *
612  * [8k, 12k) - test the single front split
613  * [12k, 20k) - test the single back split
614  * [28k, 32k) - test the double split
615  * [32k, 64k) - test whole em dropping
616  *
617  * They'll have the EXTENT_FLAG_COMPRESSED flag set to keep the em tree from
618  * merging the em's.
619  */
620 static int test_case_5(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
621 {
622 	u64 start, end;
623 	int ret;
624 	int ret2;
625 
626 	test_msg("Running btrfs_drop_extent_map_range tests");
627 
628 	/* [0, 12k) */
629 	ret = add_compressed_extent(inode, 0, SZ_4K * 3, 0);
630 	if (ret) {
631 		test_err("cannot add extent range [0, 12K)");
632 		goto out;
633 	}
634 
635 	/* [12k, 24k) */
636 	ret = add_compressed_extent(inode, SZ_4K * 3, SZ_4K * 3, SZ_4K);
637 	if (ret) {
638 		test_err("cannot add extent range [12k, 24k)");
639 		goto out;
640 	}
641 
642 	/* [24k, 36k) */
643 	ret = add_compressed_extent(inode, SZ_4K * 6, SZ_4K * 3, SZ_8K);
644 	if (ret) {
645 		test_err("cannot add extent range [12k, 24k)");
646 		goto out;
647 	}
648 
649 	/* [36k, 40k) */
650 	ret = add_compressed_extent(inode, SZ_32K + SZ_4K, SZ_4K, SZ_4K * 3);
651 	if (ret) {
652 		test_err("cannot add extent range [12k, 24k)");
653 		goto out;
654 	}
655 
656 	/* [40k, 64k) */
657 	ret = add_compressed_extent(inode, SZ_4K * 10, SZ_4K * 6, SZ_16K);
658 	if (ret) {
659 		test_err("cannot add extent range [12k, 24k)");
660 		goto out;
661 	}
662 
663 	/* Drop [8k, 12k) */
664 	start = SZ_8K;
665 	end = (3 * SZ_4K) - 1;
666 	btrfs_drop_extent_map_range(inode, start, end, false);
667 	ret = validate_range(&inode->extent_tree, 0);
668 	if (ret)
669 		goto out;
670 
671 	/* Drop [12k, 20k) */
672 	start = SZ_4K * 3;
673 	end = SZ_16K + SZ_4K - 1;
674 	btrfs_drop_extent_map_range(inode, start, end, false);
675 	ret = validate_range(&inode->extent_tree, 1);
676 	if (ret)
677 		goto out;
678 
679 	/* Drop [28k, 32k) */
680 	start = SZ_32K - SZ_4K;
681 	end = SZ_32K - 1;
682 	btrfs_drop_extent_map_range(inode, start, end, false);
683 	ret = validate_range(&inode->extent_tree, 2);
684 	if (ret)
685 		goto out;
686 
687 	/* Drop [32k, 64k) */
688 	start = SZ_32K;
689 	end = SZ_64K - 1;
690 	btrfs_drop_extent_map_range(inode, start, end, false);
691 	ret = validate_range(&inode->extent_tree, 3);
692 	if (ret)
693 		goto out;
694 out:
695 	ret2 = free_extent_map_tree(inode);
696 	if (ret == 0)
697 		ret = ret2;
698 
699 	return ret;
700 }
701 
702 /*
703  * Test the btrfs_add_extent_mapping helper which will attempt to create an em
704  * for areas between two existing ems.  Validate it doesn't do this when there
705  * are two unmerged em's side by side.
706  */
707 static int test_case_6(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
708 {
709 	struct extent_map_tree *em_tree = &inode->extent_tree;
710 	struct extent_map *em = NULL;
711 	int ret;
712 	int ret2;
713 
714 	ret = add_compressed_extent(inode, 0, SZ_4K, 0);
715 	if (ret)
716 		goto out;
717 
718 	ret = add_compressed_extent(inode, SZ_4K, SZ_4K, 0);
719 	if (ret)
720 		goto out;
721 
722 	em = alloc_extent_map();
723 	if (!em) {
724 		test_std_err(TEST_ALLOC_EXTENT_MAP);
725 		ret = -ENOMEM;
726 		goto out;
727 	}
728 
729 	em->start = SZ_4K;
730 	em->len = SZ_4K;
731 	em->disk_bytenr = SZ_16K;
732 	em->disk_num_bytes = SZ_16K;
733 	em->ram_bytes = SZ_16K;
734 	write_lock(&em_tree->lock);
735 	ret = btrfs_add_extent_mapping(inode, &em, 0, SZ_8K);
736 	write_unlock(&em_tree->lock);
737 
738 	if (ret != 0) {
739 		test_err("got an error when adding our em: %d", ret);
740 		goto out;
741 	}
742 
743 	ret = -EINVAL;
744 	if (em->start != 0) {
745 		test_err("unexpected em->start at %llu, wanted 0", em->start);
746 		goto out;
747 	}
748 	if (em->len != SZ_4K) {
749 		test_err("unexpected em->len %llu, expected 4K", em->len);
750 		goto out;
751 	}
752 	ret = 0;
753 out:
754 	free_extent_map(em);
755 	ret2 = free_extent_map_tree(inode);
756 	if (ret == 0)
757 		ret = ret2;
758 
759 	return ret;
760 }
761 
762 /*
763  * Regression test for btrfs_drop_extent_map_range.  Calling with skip_pinned ==
764  * true would mess up the start/end calculations and subsequent splits would be
765  * incorrect.
766  */
767 static int test_case_7(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
768 {
769 	struct extent_map_tree *em_tree = &inode->extent_tree;
770 	struct extent_map *em;
771 	int ret;
772 	int ret2;
773 
774 	test_msg("Running btrfs_drop_extent_cache with pinned");
775 
776 	em = alloc_extent_map();
777 	if (!em) {
778 		test_std_err(TEST_ALLOC_EXTENT_MAP);
779 		return -ENOMEM;
780 	}
781 
782 	/* [0, 16K), pinned */
783 	em->start = 0;
784 	em->len = SZ_16K;
785 	em->disk_bytenr = 0;
786 	em->disk_num_bytes = SZ_4K;
787 	em->ram_bytes = SZ_16K;
788 	em->flags |= (EXTENT_FLAG_PINNED | EXTENT_FLAG_COMPRESS_ZLIB);
789 	write_lock(&em_tree->lock);
790 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
791 	write_unlock(&em_tree->lock);
792 	if (ret < 0) {
793 		test_err("couldn't add extent map");
794 		goto out;
795 	}
796 	free_extent_map(em);
797 
798 	em = alloc_extent_map();
799 	if (!em) {
800 		test_std_err(TEST_ALLOC_EXTENT_MAP);
801 		ret = -ENOMEM;
802 		goto out;
803 	}
804 
805 	/* [32K, 48K), not pinned */
806 	em->start = SZ_32K;
807 	em->len = SZ_16K;
808 	em->disk_bytenr = SZ_32K;
809 	em->disk_num_bytes = SZ_16K;
810 	em->ram_bytes = SZ_16K;
811 	write_lock(&em_tree->lock);
812 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
813 	write_unlock(&em_tree->lock);
814 	if (ret < 0) {
815 		test_err("couldn't add extent map");
816 		goto out;
817 	}
818 	free_extent_map(em);
819 
820 	/*
821 	 * Drop [0, 36K) This should skip the [0, 4K) extent and then split the
822 	 * [32K, 48K) extent.
823 	 */
824 	btrfs_drop_extent_map_range(inode, 0, (36 * SZ_1K) - 1, true);
825 
826 	/* Make sure our extent maps look sane. */
827 	ret = -EINVAL;
828 
829 	em = lookup_extent_mapping(em_tree, 0, SZ_16K);
830 	if (!em) {
831 		test_err("didn't find an em at 0 as expected");
832 		goto out;
833 	}
834 
835 	if (em->start != 0) {
836 		test_err("em->start is %llu, expected 0", em->start);
837 		goto out;
838 	}
839 
840 	if (em->len != SZ_16K) {
841 		test_err("em->len is %llu, expected 16K", em->len);
842 		goto out;
843 	}
844 
845 	free_extent_map(em);
846 
847 	read_lock(&em_tree->lock);
848 	em = lookup_extent_mapping(em_tree, SZ_16K, SZ_16K);
849 	read_unlock(&em_tree->lock);
850 	if (em) {
851 		test_err("found an em when we weren't expecting one");
852 		goto out;
853 	}
854 
855 	read_lock(&em_tree->lock);
856 	em = lookup_extent_mapping(em_tree, SZ_32K, SZ_16K);
857 	read_unlock(&em_tree->lock);
858 	if (!em) {
859 		test_err("didn't find an em at 32K as expected");
860 		goto out;
861 	}
862 
863 	if (em->start != (36 * SZ_1K)) {
864 		test_err("em->start is %llu, expected 36K", em->start);
865 		goto out;
866 	}
867 
868 	if (em->len != (12 * SZ_1K)) {
869 		test_err("em->len is %llu, expected 12K", em->len);
870 		goto out;
871 	}
872 
873 	if (extent_map_block_start(em) != SZ_32K + SZ_4K) {
874 		test_err("em->block_start is %llu, expected 36K",
875 				extent_map_block_start(em));
876 		goto out;
877 	}
878 
879 	free_extent_map(em);
880 
881 	read_lock(&em_tree->lock);
882 	em = lookup_extent_mapping(em_tree, 48 * SZ_1K, (u64)-1);
883 	read_unlock(&em_tree->lock);
884 	if (em) {
885 		test_err("found an unexpected em above 48K");
886 		goto out;
887 	}
888 
889 	ret = 0;
890 out:
891 	free_extent_map(em);
892 	/* Unpin our extent to prevent warning when removing it below. */
893 	ret2 = unpin_extent_cache(inode, 0, SZ_16K, 0);
894 	if (ret == 0)
895 		ret = ret2;
896 	ret2 = free_extent_map_tree(inode);
897 	if (ret == 0)
898 		ret = ret2;
899 
900 	return ret;
901 }
902 
903 /*
904  * Test a regression for compressed extent map adjustment when we attempt to
905  * add an extent map that is partially overlapped by another existing extent
906  * map. The resulting extent map offset was left unchanged despite having
907  * incremented its start offset.
908  */
909 static int test_case_8(struct btrfs_fs_info *fs_info, struct btrfs_inode *inode)
910 {
911 	struct extent_map_tree *em_tree = &inode->extent_tree;
912 	struct extent_map *em;
913 	int ret;
914 	int ret2;
915 
916 	em = alloc_extent_map();
917 	if (!em) {
918 		test_std_err(TEST_ALLOC_EXTENT_MAP);
919 		return -ENOMEM;
920 	}
921 
922 	/* Compressed extent for the file range [120K, 128K). */
923 	em->start = SZ_1K * 120;
924 	em->len = SZ_8K;
925 	em->disk_num_bytes = SZ_4K;
926 	em->ram_bytes = SZ_8K;
927 	em->flags |= EXTENT_FLAG_COMPRESS_ZLIB;
928 	write_lock(&em_tree->lock);
929 	ret = btrfs_add_extent_mapping(inode, &em, em->start, em->len);
930 	write_unlock(&em_tree->lock);
931 	free_extent_map(em);
932 	if (ret < 0) {
933 		test_err("couldn't add extent map for range [120K, 128K)");
934 		goto out;
935 	}
936 
937 	em = alloc_extent_map();
938 	if (!em) {
939 		test_std_err(TEST_ALLOC_EXTENT_MAP);
940 		ret = -ENOMEM;
941 		goto out;
942 	}
943 
944 	/*
945 	 * Compressed extent for the file range [108K, 144K), which overlaps
946 	 * with the [120K, 128K) we previously inserted.
947 	 */
948 	em->start = SZ_1K * 108;
949 	em->len = SZ_1K * 36;
950 	em->disk_num_bytes = SZ_4K;
951 	em->ram_bytes = SZ_1K * 36;
952 	em->flags |= EXTENT_FLAG_COMPRESS_ZLIB;
953 
954 	/*
955 	 * Try to add the extent map but with a search range of [140K, 144K),
956 	 * this should succeed and adjust the extent map to the range
957 	 * [128K, 144K), with a length of 16K and an offset of 20K.
958 	 *
959 	 * This simulates a scenario where in the subvolume tree of an inode we
960 	 * have a compressed file extent item for the range [108K, 144K) and we
961 	 * have an overlapping compressed extent map for the range [120K, 128K),
962 	 * which was created by an encoded write, but its ordered extent was not
963 	 * yet completed, so the subvolume tree doesn't have yet the file extent
964 	 * item for that range - we only have the extent map in the inode's
965 	 * extent map tree.
966 	 */
967 	write_lock(&em_tree->lock);
968 	ret = btrfs_add_extent_mapping(inode, &em, SZ_1K * 140, SZ_4K);
969 	write_unlock(&em_tree->lock);
970 	free_extent_map(em);
971 	if (ret < 0) {
972 		test_err("couldn't add extent map for range [108K, 144K)");
973 		goto out;
974 	}
975 
976 	if (em->start != SZ_128K) {
977 		test_err("unexpected extent map start %llu (should be 128K)", em->start);
978 		ret = -EINVAL;
979 		goto out;
980 	}
981 	if (em->len != SZ_16K) {
982 		test_err("unexpected extent map length %llu (should be 16K)", em->len);
983 		ret = -EINVAL;
984 		goto out;
985 	}
986 	if (em->offset != SZ_1K * 20) {
987 		test_err("unexpected extent map offset %llu (should be 20K)", em->offset);
988 		ret = -EINVAL;
989 		goto out;
990 	}
991 out:
992 	ret2 = free_extent_map_tree(inode);
993 	if (ret == 0)
994 		ret = ret2;
995 
996 	return ret;
997 }
998 
999 struct rmap_test_vector {
1000 	u64 raid_type;
1001 	u64 physical_start;
1002 	u64 data_stripe_size;
1003 	u64 num_data_stripes;
1004 	u64 num_stripes;
1005 	/* Assume we won't have more than 5 physical stripes */
1006 	u64 data_stripe_phys_start[5];
1007 	bool expected_mapped_addr;
1008 	/* Physical to logical addresses */
1009 	u64 mapped_logical[5];
1010 };
1011 
1012 static int test_rmap_block(struct btrfs_fs_info *fs_info,
1013 			   struct rmap_test_vector *test)
1014 {
1015 	struct btrfs_chunk_map *map;
1016 	u64 *logical = NULL;
1017 	int i, out_ndaddrs, out_stripe_len;
1018 	int ret;
1019 
1020 	map = btrfs_alloc_chunk_map(test->num_stripes, GFP_KERNEL);
1021 	if (!map) {
1022 		test_std_err(TEST_ALLOC_CHUNK_MAP);
1023 		return -ENOMEM;
1024 	}
1025 
1026 	/* Start at 4GiB logical address */
1027 	map->start = SZ_4G;
1028 	map->chunk_len = test->data_stripe_size * test->num_data_stripes;
1029 	map->stripe_size = test->data_stripe_size;
1030 	map->num_stripes = test->num_stripes;
1031 	map->type = test->raid_type;
1032 
1033 	for (i = 0; i < map->num_stripes; i++) {
1034 		struct btrfs_device *dev = btrfs_alloc_dummy_device(fs_info);
1035 
1036 		if (IS_ERR(dev)) {
1037 			test_err("cannot allocate device");
1038 			ret = PTR_ERR(dev);
1039 			goto out;
1040 		}
1041 		map->stripes[i].dev = dev;
1042 		map->stripes[i].physical = test->data_stripe_phys_start[i];
1043 	}
1044 
1045 	ret = btrfs_add_chunk_map(fs_info, map);
1046 	if (ret) {
1047 		test_err("error adding chunk map to mapping tree");
1048 		goto out_free;
1049 	}
1050 
1051 	ret = btrfs_rmap_block(fs_info, map->start, btrfs_sb_offset(1),
1052 			       &logical, &out_ndaddrs, &out_stripe_len);
1053 	if (ret || (out_ndaddrs == 0 && test->expected_mapped_addr)) {
1054 		test_err("didn't rmap anything but expected %d",
1055 			 test->expected_mapped_addr);
1056 		goto out;
1057 	}
1058 
1059 	if (out_stripe_len != BTRFS_STRIPE_LEN) {
1060 		test_err("calculated stripe length doesn't match");
1061 		goto out;
1062 	}
1063 
1064 	if (out_ndaddrs != test->expected_mapped_addr) {
1065 		for (i = 0; i < out_ndaddrs; i++)
1066 			test_msg("mapped %llu", logical[i]);
1067 		test_err("unexpected number of mapped addresses: %d", out_ndaddrs);
1068 		goto out;
1069 	}
1070 
1071 	for (i = 0; i < out_ndaddrs; i++) {
1072 		if (logical[i] != test->mapped_logical[i]) {
1073 			test_err("unexpected logical address mapped");
1074 			goto out;
1075 		}
1076 	}
1077 
1078 	ret = 0;
1079 out:
1080 	btrfs_remove_chunk_map(fs_info, map);
1081 out_free:
1082 	kfree(logical);
1083 	return ret;
1084 }
1085 
1086 int btrfs_test_extent_map(void)
1087 {
1088 	struct btrfs_fs_info *fs_info = NULL;
1089 	struct inode *inode;
1090 	struct btrfs_root *root = NULL;
1091 	int ret = 0, i;
1092 	struct rmap_test_vector rmap_tests[] = {
1093 		{
1094 			/*
1095 			 * Test a chunk with 2 data stripes one of which
1096 			 * intersects the physical address of the super block
1097 			 * is correctly recognised.
1098 			 */
1099 			.raid_type = BTRFS_BLOCK_GROUP_RAID1,
1100 			.physical_start = SZ_64M - SZ_4M,
1101 			.data_stripe_size = SZ_256M,
1102 			.num_data_stripes = 2,
1103 			.num_stripes = 2,
1104 			.data_stripe_phys_start =
1105 				{SZ_64M - SZ_4M, SZ_64M - SZ_4M + SZ_256M},
1106 			.expected_mapped_addr = true,
1107 			.mapped_logical= {SZ_4G + SZ_4M}
1108 		},
1109 		{
1110 			/*
1111 			 * Test that out-of-range physical addresses are
1112 			 * ignored
1113 			 */
1114 
1115 			 /* SINGLE chunk type */
1116 			.raid_type = 0,
1117 			.physical_start = SZ_4G,
1118 			.data_stripe_size = SZ_256M,
1119 			.num_data_stripes = 1,
1120 			.num_stripes = 1,
1121 			.data_stripe_phys_start = {SZ_256M},
1122 			.expected_mapped_addr = false,
1123 			.mapped_logical = {0}
1124 		}
1125 	};
1126 
1127 	test_msg("running extent_map tests");
1128 
1129 	/*
1130 	 * Note: the fs_info is not set up completely, we only need
1131 	 * fs_info::fsid for the tracepoint.
1132 	 */
1133 	fs_info = btrfs_alloc_dummy_fs_info(PAGE_SIZE, PAGE_SIZE);
1134 	if (!fs_info) {
1135 		test_std_err(TEST_ALLOC_FS_INFO);
1136 		return -ENOMEM;
1137 	}
1138 
1139 	inode = btrfs_new_test_inode();
1140 	if (!inode) {
1141 		test_std_err(TEST_ALLOC_INODE);
1142 		ret = -ENOMEM;
1143 		goto out;
1144 	}
1145 
1146 	root = btrfs_alloc_dummy_root(fs_info);
1147 	if (IS_ERR(root)) {
1148 		test_std_err(TEST_ALLOC_ROOT);
1149 		ret = PTR_ERR(root);
1150 		root = NULL;
1151 		goto out;
1152 	}
1153 
1154 	BTRFS_I(inode)->root = root;
1155 
1156 	ret = test_case_1(fs_info, BTRFS_I(inode));
1157 	if (ret)
1158 		goto out;
1159 	ret = test_case_2(fs_info, BTRFS_I(inode));
1160 	if (ret)
1161 		goto out;
1162 	ret = test_case_3(fs_info, BTRFS_I(inode));
1163 	if (ret)
1164 		goto out;
1165 	ret = test_case_4(fs_info, BTRFS_I(inode));
1166 	if (ret)
1167 		goto out;
1168 	ret = test_case_5(fs_info, BTRFS_I(inode));
1169 	if (ret)
1170 		goto out;
1171 	ret = test_case_6(fs_info, BTRFS_I(inode));
1172 	if (ret)
1173 		goto out;
1174 	ret = test_case_7(fs_info, BTRFS_I(inode));
1175 	if (ret)
1176 		goto out;
1177 	ret = test_case_8(fs_info, BTRFS_I(inode));
1178 	if (ret)
1179 		goto out;
1180 
1181 	test_msg("running rmap tests");
1182 	for (i = 0; i < ARRAY_SIZE(rmap_tests); i++) {
1183 		ret = test_rmap_block(fs_info, &rmap_tests[i]);
1184 		if (ret)
1185 			goto out;
1186 	}
1187 
1188 out:
1189 	iput(inode);
1190 	btrfs_free_dummy_root(root);
1191 	btrfs_free_dummy_fs_info(fs_info);
1192 
1193 	return ret;
1194 }
1195