xref: /linux/fs/btrfs/zoned.c (revision bf80eef2212a1e8451df13b52533f4bc31bb4f8e)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
5 #include <linux/blkdev.h>
6 #include <linux/sched/mm.h>
7 #include <linux/atomic.h>
8 #include <linux/vmalloc.h>
9 #include "ctree.h"
10 #include "volumes.h"
11 #include "zoned.h"
12 #include "rcu-string.h"
13 #include "disk-io.h"
14 #include "block-group.h"
15 #include "transaction.h"
16 #include "dev-replace.h"
17 #include "space-info.h"
18 
19 /* Maximum number of zones to report per blkdev_report_zones() call */
20 #define BTRFS_REPORT_NR_ZONES   4096
21 /* Invalid allocation pointer value for missing devices */
22 #define WP_MISSING_DEV ((u64)-1)
23 /* Pseudo write pointer value for conventional zone */
24 #define WP_CONVENTIONAL ((u64)-2)
25 
26 /*
27  * Location of the first zone of superblock logging zone pairs.
28  *
29  * - primary superblock:    0B (zone 0)
30  * - first copy:          512G (zone starting at that offset)
31  * - second copy:           4T (zone starting at that offset)
32  */
33 #define BTRFS_SB_LOG_PRIMARY_OFFSET	(0ULL)
34 #define BTRFS_SB_LOG_FIRST_OFFSET	(512ULL * SZ_1G)
35 #define BTRFS_SB_LOG_SECOND_OFFSET	(4096ULL * SZ_1G)
36 
37 #define BTRFS_SB_LOG_FIRST_SHIFT	const_ilog2(BTRFS_SB_LOG_FIRST_OFFSET)
38 #define BTRFS_SB_LOG_SECOND_SHIFT	const_ilog2(BTRFS_SB_LOG_SECOND_OFFSET)
39 
40 /* Number of superblock log zones */
41 #define BTRFS_NR_SB_LOG_ZONES 2
42 
43 /*
44  * Minimum of active zones we need:
45  *
46  * - BTRFS_SUPER_MIRROR_MAX zones for superblock mirrors
47  * - 3 zones to ensure at least one zone per SYSTEM, META and DATA block group
48  * - 1 zone for tree-log dedicated block group
49  * - 1 zone for relocation
50  */
51 #define BTRFS_MIN_ACTIVE_ZONES		(BTRFS_SUPER_MIRROR_MAX + 5)
52 
53 /*
54  * Minimum / maximum supported zone size. Currently, SMR disks have a zone
55  * size of 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range.
56  * We do not expect the zone size to become larger than 8GiB or smaller than
57  * 4MiB in the near future.
58  */
59 #define BTRFS_MAX_ZONE_SIZE		SZ_8G
60 #define BTRFS_MIN_ZONE_SIZE		SZ_4M
61 
62 #define SUPER_INFO_SECTORS	((u64)BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT)
63 
64 static inline bool sb_zone_is_full(const struct blk_zone *zone)
65 {
66 	return (zone->cond == BLK_ZONE_COND_FULL) ||
67 		(zone->wp + SUPER_INFO_SECTORS > zone->start + zone->capacity);
68 }
69 
70 static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data)
71 {
72 	struct blk_zone *zones = data;
73 
74 	memcpy(&zones[idx], zone, sizeof(*zone));
75 
76 	return 0;
77 }
78 
79 static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones,
80 			    u64 *wp_ret)
81 {
82 	bool empty[BTRFS_NR_SB_LOG_ZONES];
83 	bool full[BTRFS_NR_SB_LOG_ZONES];
84 	sector_t sector;
85 	int i;
86 
87 	for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
88 		ASSERT(zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL);
89 		empty[i] = (zones[i].cond == BLK_ZONE_COND_EMPTY);
90 		full[i] = sb_zone_is_full(&zones[i]);
91 	}
92 
93 	/*
94 	 * Possible states of log buffer zones
95 	 *
96 	 *           Empty[0]  In use[0]  Full[0]
97 	 * Empty[1]         *          0        1
98 	 * In use[1]        x          x        1
99 	 * Full[1]          0          0        C
100 	 *
101 	 * Log position:
102 	 *   *: Special case, no superblock is written
103 	 *   0: Use write pointer of zones[0]
104 	 *   1: Use write pointer of zones[1]
105 	 *   C: Compare super blocks from zones[0] and zones[1], use the latest
106 	 *      one determined by generation
107 	 *   x: Invalid state
108 	 */
109 
110 	if (empty[0] && empty[1]) {
111 		/* Special case to distinguish no superblock to read */
112 		*wp_ret = zones[0].start << SECTOR_SHIFT;
113 		return -ENOENT;
114 	} else if (full[0] && full[1]) {
115 		/* Compare two super blocks */
116 		struct address_space *mapping = bdev->bd_inode->i_mapping;
117 		struct page *page[BTRFS_NR_SB_LOG_ZONES];
118 		struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES];
119 		int i;
120 
121 		for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
122 			u64 bytenr;
123 
124 			bytenr = ((zones[i].start + zones[i].len)
125 				   << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE;
126 
127 			page[i] = read_cache_page_gfp(mapping,
128 					bytenr >> PAGE_SHIFT, GFP_NOFS);
129 			if (IS_ERR(page[i])) {
130 				if (i == 1)
131 					btrfs_release_disk_super(super[0]);
132 				return PTR_ERR(page[i]);
133 			}
134 			super[i] = page_address(page[i]);
135 		}
136 
137 		if (super[0]->generation > super[1]->generation)
138 			sector = zones[1].start;
139 		else
140 			sector = zones[0].start;
141 
142 		for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++)
143 			btrfs_release_disk_super(super[i]);
144 	} else if (!full[0] && (empty[1] || full[1])) {
145 		sector = zones[0].wp;
146 	} else if (full[0]) {
147 		sector = zones[1].wp;
148 	} else {
149 		return -EUCLEAN;
150 	}
151 	*wp_ret = sector << SECTOR_SHIFT;
152 	return 0;
153 }
154 
155 /*
156  * Get the first zone number of the superblock mirror
157  */
158 static inline u32 sb_zone_number(int shift, int mirror)
159 {
160 	u64 zone;
161 
162 	ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX);
163 	switch (mirror) {
164 	case 0: zone = 0; break;
165 	case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break;
166 	case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break;
167 	}
168 
169 	ASSERT(zone <= U32_MAX);
170 
171 	return (u32)zone;
172 }
173 
174 static inline sector_t zone_start_sector(u32 zone_number,
175 					 struct block_device *bdev)
176 {
177 	return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev));
178 }
179 
180 static inline u64 zone_start_physical(u32 zone_number,
181 				      struct btrfs_zoned_device_info *zone_info)
182 {
183 	return (u64)zone_number << zone_info->zone_size_shift;
184 }
185 
186 /*
187  * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block
188  * device into static sized chunks and fake a conventional zone on each of
189  * them.
190  */
191 static int emulate_report_zones(struct btrfs_device *device, u64 pos,
192 				struct blk_zone *zones, unsigned int nr_zones)
193 {
194 	const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT;
195 	sector_t bdev_size = bdev_nr_sectors(device->bdev);
196 	unsigned int i;
197 
198 	pos >>= SECTOR_SHIFT;
199 	for (i = 0; i < nr_zones; i++) {
200 		zones[i].start = i * zone_sectors + pos;
201 		zones[i].len = zone_sectors;
202 		zones[i].capacity = zone_sectors;
203 		zones[i].wp = zones[i].start + zone_sectors;
204 		zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL;
205 		zones[i].cond = BLK_ZONE_COND_NOT_WP;
206 
207 		if (zones[i].wp >= bdev_size) {
208 			i++;
209 			break;
210 		}
211 	}
212 
213 	return i;
214 }
215 
216 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
217 			       struct blk_zone *zones, unsigned int *nr_zones)
218 {
219 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
220 	u32 zno;
221 	int ret;
222 
223 	if (!*nr_zones)
224 		return 0;
225 
226 	if (!bdev_is_zoned(device->bdev)) {
227 		ret = emulate_report_zones(device, pos, zones, *nr_zones);
228 		*nr_zones = ret;
229 		return 0;
230 	}
231 
232 	/* Check cache */
233 	if (zinfo->zone_cache) {
234 		unsigned int i;
235 
236 		ASSERT(IS_ALIGNED(pos, zinfo->zone_size));
237 		zno = pos >> zinfo->zone_size_shift;
238 		/*
239 		 * We cannot report zones beyond the zone end. So, it is OK to
240 		 * cap *nr_zones to at the end.
241 		 */
242 		*nr_zones = min_t(u32, *nr_zones, zinfo->nr_zones - zno);
243 
244 		for (i = 0; i < *nr_zones; i++) {
245 			struct blk_zone *zone_info;
246 
247 			zone_info = &zinfo->zone_cache[zno + i];
248 			if (!zone_info->len)
249 				break;
250 		}
251 
252 		if (i == *nr_zones) {
253 			/* Cache hit on all the zones */
254 			memcpy(zones, zinfo->zone_cache + zno,
255 			       sizeof(*zinfo->zone_cache) * *nr_zones);
256 			return 0;
257 		}
258 	}
259 
260 	ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
261 				  copy_zone_info_cb, zones);
262 	if (ret < 0) {
263 		btrfs_err_in_rcu(device->fs_info,
264 				 "zoned: failed to read zone %llu on %s (devid %llu)",
265 				 pos, rcu_str_deref(device->name),
266 				 device->devid);
267 		return ret;
268 	}
269 	*nr_zones = ret;
270 	if (!ret)
271 		return -EIO;
272 
273 	/* Populate cache */
274 	if (zinfo->zone_cache)
275 		memcpy(zinfo->zone_cache + zno, zones,
276 		       sizeof(*zinfo->zone_cache) * *nr_zones);
277 
278 	return 0;
279 }
280 
281 /* The emulated zone size is determined from the size of device extent */
282 static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info)
283 {
284 	struct btrfs_path *path;
285 	struct btrfs_root *root = fs_info->dev_root;
286 	struct btrfs_key key;
287 	struct extent_buffer *leaf;
288 	struct btrfs_dev_extent *dext;
289 	int ret = 0;
290 
291 	key.objectid = 1;
292 	key.type = BTRFS_DEV_EXTENT_KEY;
293 	key.offset = 0;
294 
295 	path = btrfs_alloc_path();
296 	if (!path)
297 		return -ENOMEM;
298 
299 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
300 	if (ret < 0)
301 		goto out;
302 
303 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
304 		ret = btrfs_next_leaf(root, path);
305 		if (ret < 0)
306 			goto out;
307 		/* No dev extents at all? Not good */
308 		if (ret > 0) {
309 			ret = -EUCLEAN;
310 			goto out;
311 		}
312 	}
313 
314 	leaf = path->nodes[0];
315 	dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
316 	fs_info->zone_size = btrfs_dev_extent_length(leaf, dext);
317 	ret = 0;
318 
319 out:
320 	btrfs_free_path(path);
321 
322 	return ret;
323 }
324 
325 int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
326 {
327 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
328 	struct btrfs_device *device;
329 	int ret = 0;
330 
331 	/* fs_info->zone_size might not set yet. Use the incomapt flag here. */
332 	if (!btrfs_fs_incompat(fs_info, ZONED))
333 		return 0;
334 
335 	mutex_lock(&fs_devices->device_list_mutex);
336 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
337 		/* We can skip reading of zone info for missing devices */
338 		if (!device->bdev)
339 			continue;
340 
341 		ret = btrfs_get_dev_zone_info(device, true);
342 		if (ret)
343 			break;
344 	}
345 	mutex_unlock(&fs_devices->device_list_mutex);
346 
347 	return ret;
348 }
349 
350 int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache)
351 {
352 	struct btrfs_fs_info *fs_info = device->fs_info;
353 	struct btrfs_zoned_device_info *zone_info = NULL;
354 	struct block_device *bdev = device->bdev;
355 	unsigned int max_active_zones;
356 	unsigned int nactive;
357 	sector_t nr_sectors;
358 	sector_t sector = 0;
359 	struct blk_zone *zones = NULL;
360 	unsigned int i, nreported = 0, nr_zones;
361 	sector_t zone_sectors;
362 	char *model, *emulated;
363 	int ret;
364 
365 	/*
366 	 * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not
367 	 * yet be set.
368 	 */
369 	if (!btrfs_fs_incompat(fs_info, ZONED))
370 		return 0;
371 
372 	if (device->zone_info)
373 		return 0;
374 
375 	zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL);
376 	if (!zone_info)
377 		return -ENOMEM;
378 
379 	device->zone_info = zone_info;
380 
381 	if (!bdev_is_zoned(bdev)) {
382 		if (!fs_info->zone_size) {
383 			ret = calculate_emulated_zone_size(fs_info);
384 			if (ret)
385 				goto out;
386 		}
387 
388 		ASSERT(fs_info->zone_size);
389 		zone_sectors = fs_info->zone_size >> SECTOR_SHIFT;
390 	} else {
391 		zone_sectors = bdev_zone_sectors(bdev);
392 	}
393 
394 	/* Check if it's power of 2 (see is_power_of_2) */
395 	ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0);
396 	zone_info->zone_size = zone_sectors << SECTOR_SHIFT;
397 
398 	/* We reject devices with a zone size larger than 8GB */
399 	if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) {
400 		btrfs_err_in_rcu(fs_info,
401 		"zoned: %s: zone size %llu larger than supported maximum %llu",
402 				 rcu_str_deref(device->name),
403 				 zone_info->zone_size, BTRFS_MAX_ZONE_SIZE);
404 		ret = -EINVAL;
405 		goto out;
406 	} else if (zone_info->zone_size < BTRFS_MIN_ZONE_SIZE) {
407 		btrfs_err_in_rcu(fs_info,
408 		"zoned: %s: zone size %llu smaller than supported minimum %u",
409 				 rcu_str_deref(device->name),
410 				 zone_info->zone_size, BTRFS_MIN_ZONE_SIZE);
411 		ret = -EINVAL;
412 		goto out;
413 	}
414 
415 	nr_sectors = bdev_nr_sectors(bdev);
416 	zone_info->zone_size_shift = ilog2(zone_info->zone_size);
417 	zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors);
418 	/*
419 	 * We limit max_zone_append_size also by max_segments *
420 	 * PAGE_SIZE. Technically, we can have multiple pages per segment. But,
421 	 * since btrfs adds the pages one by one to a bio, and btrfs cannot
422 	 * increase the metadata reservation even if it increases the number of
423 	 * extents, it is safe to stick with the limit.
424 	 *
425 	 * With the zoned emulation, we can have non-zoned device on the zoned
426 	 * mode. In this case, we don't have a valid max zone append size. So,
427 	 * use max_segments * PAGE_SIZE as the pseudo max_zone_append_size.
428 	 */
429 	if (bdev_is_zoned(bdev)) {
430 		zone_info->max_zone_append_size = min_t(u64,
431 			(u64)bdev_max_zone_append_sectors(bdev) << SECTOR_SHIFT,
432 			(u64)bdev_max_segments(bdev) << PAGE_SHIFT);
433 	} else {
434 		zone_info->max_zone_append_size =
435 			(u64)bdev_max_segments(bdev) << PAGE_SHIFT;
436 	}
437 	if (!IS_ALIGNED(nr_sectors, zone_sectors))
438 		zone_info->nr_zones++;
439 
440 	max_active_zones = bdev_max_active_zones(bdev);
441 	if (max_active_zones && max_active_zones < BTRFS_MIN_ACTIVE_ZONES) {
442 		btrfs_err_in_rcu(fs_info,
443 "zoned: %s: max active zones %u is too small, need at least %u active zones",
444 				 rcu_str_deref(device->name), max_active_zones,
445 				 BTRFS_MIN_ACTIVE_ZONES);
446 		ret = -EINVAL;
447 		goto out;
448 	}
449 	zone_info->max_active_zones = max_active_zones;
450 
451 	zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
452 	if (!zone_info->seq_zones) {
453 		ret = -ENOMEM;
454 		goto out;
455 	}
456 
457 	zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
458 	if (!zone_info->empty_zones) {
459 		ret = -ENOMEM;
460 		goto out;
461 	}
462 
463 	zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
464 	if (!zone_info->active_zones) {
465 		ret = -ENOMEM;
466 		goto out;
467 	}
468 
469 	zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
470 	if (!zones) {
471 		ret = -ENOMEM;
472 		goto out;
473 	}
474 
475 	/*
476 	 * Enable zone cache only for a zoned device. On a non-zoned device, we
477 	 * fill the zone info with emulated CONVENTIONAL zones, so no need to
478 	 * use the cache.
479 	 */
480 	if (populate_cache && bdev_is_zoned(device->bdev)) {
481 		zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) *
482 						zone_info->nr_zones);
483 		if (!zone_info->zone_cache) {
484 			btrfs_err_in_rcu(device->fs_info,
485 				"zoned: failed to allocate zone cache for %s",
486 				rcu_str_deref(device->name));
487 			ret = -ENOMEM;
488 			goto out;
489 		}
490 	}
491 
492 	/* Get zones type */
493 	nactive = 0;
494 	while (sector < nr_sectors) {
495 		nr_zones = BTRFS_REPORT_NR_ZONES;
496 		ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
497 					  &nr_zones);
498 		if (ret)
499 			goto out;
500 
501 		for (i = 0; i < nr_zones; i++) {
502 			if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
503 				__set_bit(nreported, zone_info->seq_zones);
504 			switch (zones[i].cond) {
505 			case BLK_ZONE_COND_EMPTY:
506 				__set_bit(nreported, zone_info->empty_zones);
507 				break;
508 			case BLK_ZONE_COND_IMP_OPEN:
509 			case BLK_ZONE_COND_EXP_OPEN:
510 			case BLK_ZONE_COND_CLOSED:
511 				__set_bit(nreported, zone_info->active_zones);
512 				nactive++;
513 				break;
514 			}
515 			nreported++;
516 		}
517 		sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
518 	}
519 
520 	if (nreported != zone_info->nr_zones) {
521 		btrfs_err_in_rcu(device->fs_info,
522 				 "inconsistent number of zones on %s (%u/%u)",
523 				 rcu_str_deref(device->name), nreported,
524 				 zone_info->nr_zones);
525 		ret = -EIO;
526 		goto out;
527 	}
528 
529 	if (max_active_zones) {
530 		if (nactive > max_active_zones) {
531 			btrfs_err_in_rcu(device->fs_info,
532 			"zoned: %u active zones on %s exceeds max_active_zones %u",
533 					 nactive, rcu_str_deref(device->name),
534 					 max_active_zones);
535 			ret = -EIO;
536 			goto out;
537 		}
538 		atomic_set(&zone_info->active_zones_left,
539 			   max_active_zones - nactive);
540 	}
541 
542 	/* Validate superblock log */
543 	nr_zones = BTRFS_NR_SB_LOG_ZONES;
544 	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
545 		u32 sb_zone;
546 		u64 sb_wp;
547 		int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
548 
549 		sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
550 		if (sb_zone + 1 >= zone_info->nr_zones)
551 			continue;
552 
553 		ret = btrfs_get_dev_zones(device,
554 					  zone_start_physical(sb_zone, zone_info),
555 					  &zone_info->sb_zones[sb_pos],
556 					  &nr_zones);
557 		if (ret)
558 			goto out;
559 
560 		if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
561 			btrfs_err_in_rcu(device->fs_info,
562 	"zoned: failed to read super block log zone info at devid %llu zone %u",
563 					 device->devid, sb_zone);
564 			ret = -EUCLEAN;
565 			goto out;
566 		}
567 
568 		/*
569 		 * If zones[0] is conventional, always use the beginning of the
570 		 * zone to record superblock. No need to validate in that case.
571 		 */
572 		if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
573 		    BLK_ZONE_TYPE_CONVENTIONAL)
574 			continue;
575 
576 		ret = sb_write_pointer(device->bdev,
577 				       &zone_info->sb_zones[sb_pos], &sb_wp);
578 		if (ret != -ENOENT && ret) {
579 			btrfs_err_in_rcu(device->fs_info,
580 			"zoned: super block log zone corrupted devid %llu zone %u",
581 					 device->devid, sb_zone);
582 			ret = -EUCLEAN;
583 			goto out;
584 		}
585 	}
586 
587 
588 	kfree(zones);
589 
590 	switch (bdev_zoned_model(bdev)) {
591 	case BLK_ZONED_HM:
592 		model = "host-managed zoned";
593 		emulated = "";
594 		break;
595 	case BLK_ZONED_HA:
596 		model = "host-aware zoned";
597 		emulated = "";
598 		break;
599 	case BLK_ZONED_NONE:
600 		model = "regular";
601 		emulated = "emulated ";
602 		break;
603 	default:
604 		/* Just in case */
605 		btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
606 				 bdev_zoned_model(bdev),
607 				 rcu_str_deref(device->name));
608 		ret = -EOPNOTSUPP;
609 		goto out_free_zone_info;
610 	}
611 
612 	btrfs_info_in_rcu(fs_info,
613 		"%s block device %s, %u %szones of %llu bytes",
614 		model, rcu_str_deref(device->name), zone_info->nr_zones,
615 		emulated, zone_info->zone_size);
616 
617 	return 0;
618 
619 out:
620 	kfree(zones);
621 out_free_zone_info:
622 	btrfs_destroy_dev_zone_info(device);
623 
624 	return ret;
625 }
626 
627 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
628 {
629 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
630 
631 	if (!zone_info)
632 		return;
633 
634 	bitmap_free(zone_info->active_zones);
635 	bitmap_free(zone_info->seq_zones);
636 	bitmap_free(zone_info->empty_zones);
637 	vfree(zone_info->zone_cache);
638 	kfree(zone_info);
639 	device->zone_info = NULL;
640 }
641 
642 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
643 		       struct blk_zone *zone)
644 {
645 	unsigned int nr_zones = 1;
646 	int ret;
647 
648 	ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
649 	if (ret != 0 || !nr_zones)
650 		return ret ? ret : -EIO;
651 
652 	return 0;
653 }
654 
655 static int btrfs_check_for_zoned_device(struct btrfs_fs_info *fs_info)
656 {
657 	struct btrfs_device *device;
658 
659 	list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
660 		if (device->bdev &&
661 		    bdev_zoned_model(device->bdev) == BLK_ZONED_HM) {
662 			btrfs_err(fs_info,
663 				"zoned: mode not enabled but zoned device found: %pg",
664 				device->bdev);
665 			return -EINVAL;
666 		}
667 	}
668 
669 	return 0;
670 }
671 
672 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
673 {
674 	struct btrfs_device *device;
675 	u64 zone_size = 0;
676 	u64 max_zone_append_size = 0;
677 	int ret;
678 
679 	/*
680 	 * Host-Managed devices can't be used without the ZONED flag.  With the
681 	 * ZONED all devices can be used, using zone emulation if required.
682 	 */
683 	if (!btrfs_fs_incompat(fs_info, ZONED))
684 		return btrfs_check_for_zoned_device(fs_info);
685 
686 	list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
687 		struct btrfs_zoned_device_info *zone_info = device->zone_info;
688 
689 		if (!device->bdev)
690 			continue;
691 
692 		if (!zone_size) {
693 			zone_size = zone_info->zone_size;
694 		} else if (zone_info->zone_size != zone_size) {
695 			btrfs_err(fs_info,
696 		"zoned: unequal block device zone sizes: have %llu found %llu",
697 				  zone_info->zone_size, zone_size);
698 			return -EINVAL;
699 		}
700 		if (!max_zone_append_size ||
701 		    (zone_info->max_zone_append_size &&
702 		     zone_info->max_zone_append_size < max_zone_append_size))
703 			max_zone_append_size = zone_info->max_zone_append_size;
704 	}
705 
706 	/*
707 	 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
708 	 * btrfs_create_chunk(). Since we want stripe_len == zone_size,
709 	 * check the alignment here.
710 	 */
711 	if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
712 		btrfs_err(fs_info,
713 			  "zoned: zone size %llu not aligned to stripe %u",
714 			  zone_size, BTRFS_STRIPE_LEN);
715 		return -EINVAL;
716 	}
717 
718 	if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
719 		btrfs_err(fs_info, "zoned: mixed block groups not supported");
720 		return -EINVAL;
721 	}
722 
723 	fs_info->zone_size = zone_size;
724 	fs_info->max_zone_append_size = ALIGN_DOWN(max_zone_append_size,
725 						   fs_info->sectorsize);
726 	fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
727 	if (fs_info->max_zone_append_size < fs_info->max_extent_size)
728 		fs_info->max_extent_size = fs_info->max_zone_append_size;
729 
730 	/*
731 	 * Check mount options here, because we might change fs_info->zoned
732 	 * from fs_info->zone_size.
733 	 */
734 	ret = btrfs_check_mountopts_zoned(fs_info);
735 	if (ret)
736 		return ret;
737 
738 	btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
739 	return 0;
740 }
741 
742 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
743 {
744 	if (!btrfs_is_zoned(info))
745 		return 0;
746 
747 	/*
748 	 * Space cache writing is not COWed. Disable that to avoid write errors
749 	 * in sequential zones.
750 	 */
751 	if (btrfs_test_opt(info, SPACE_CACHE)) {
752 		btrfs_err(info, "zoned: space cache v1 is not supported");
753 		return -EINVAL;
754 	}
755 
756 	if (btrfs_test_opt(info, NODATACOW)) {
757 		btrfs_err(info, "zoned: NODATACOW not supported");
758 		return -EINVAL;
759 	}
760 
761 	return 0;
762 }
763 
764 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
765 			   int rw, u64 *bytenr_ret)
766 {
767 	u64 wp;
768 	int ret;
769 
770 	if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
771 		*bytenr_ret = zones[0].start << SECTOR_SHIFT;
772 		return 0;
773 	}
774 
775 	ret = sb_write_pointer(bdev, zones, &wp);
776 	if (ret != -ENOENT && ret < 0)
777 		return ret;
778 
779 	if (rw == WRITE) {
780 		struct blk_zone *reset = NULL;
781 
782 		if (wp == zones[0].start << SECTOR_SHIFT)
783 			reset = &zones[0];
784 		else if (wp == zones[1].start << SECTOR_SHIFT)
785 			reset = &zones[1];
786 
787 		if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
788 			ASSERT(sb_zone_is_full(reset));
789 
790 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
791 					       reset->start, reset->len,
792 					       GFP_NOFS);
793 			if (ret)
794 				return ret;
795 
796 			reset->cond = BLK_ZONE_COND_EMPTY;
797 			reset->wp = reset->start;
798 		}
799 	} else if (ret != -ENOENT) {
800 		/*
801 		 * For READ, we want the previous one. Move write pointer to
802 		 * the end of a zone, if it is at the head of a zone.
803 		 */
804 		u64 zone_end = 0;
805 
806 		if (wp == zones[0].start << SECTOR_SHIFT)
807 			zone_end = zones[1].start + zones[1].capacity;
808 		else if (wp == zones[1].start << SECTOR_SHIFT)
809 			zone_end = zones[0].start + zones[0].capacity;
810 		if (zone_end)
811 			wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT,
812 					BTRFS_SUPER_INFO_SIZE);
813 
814 		wp -= BTRFS_SUPER_INFO_SIZE;
815 	}
816 
817 	*bytenr_ret = wp;
818 	return 0;
819 
820 }
821 
822 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
823 			       u64 *bytenr_ret)
824 {
825 	struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
826 	sector_t zone_sectors;
827 	u32 sb_zone;
828 	int ret;
829 	u8 zone_sectors_shift;
830 	sector_t nr_sectors;
831 	u32 nr_zones;
832 
833 	if (!bdev_is_zoned(bdev)) {
834 		*bytenr_ret = btrfs_sb_offset(mirror);
835 		return 0;
836 	}
837 
838 	ASSERT(rw == READ || rw == WRITE);
839 
840 	zone_sectors = bdev_zone_sectors(bdev);
841 	if (!is_power_of_2(zone_sectors))
842 		return -EINVAL;
843 	zone_sectors_shift = ilog2(zone_sectors);
844 	nr_sectors = bdev_nr_sectors(bdev);
845 	nr_zones = nr_sectors >> zone_sectors_shift;
846 
847 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
848 	if (sb_zone + 1 >= nr_zones)
849 		return -ENOENT;
850 
851 	ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
852 				  BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
853 				  zones);
854 	if (ret < 0)
855 		return ret;
856 	if (ret != BTRFS_NR_SB_LOG_ZONES)
857 		return -EIO;
858 
859 	return sb_log_location(bdev, zones, rw, bytenr_ret);
860 }
861 
862 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
863 			  u64 *bytenr_ret)
864 {
865 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
866 	u32 zone_num;
867 
868 	/*
869 	 * For a zoned filesystem on a non-zoned block device, use the same
870 	 * super block locations as regular filesystem. Doing so, the super
871 	 * block can always be retrieved and the zoned flag of the volume
872 	 * detected from the super block information.
873 	 */
874 	if (!bdev_is_zoned(device->bdev)) {
875 		*bytenr_ret = btrfs_sb_offset(mirror);
876 		return 0;
877 	}
878 
879 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
880 	if (zone_num + 1 >= zinfo->nr_zones)
881 		return -ENOENT;
882 
883 	return sb_log_location(device->bdev,
884 			       &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
885 			       rw, bytenr_ret);
886 }
887 
888 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
889 				  int mirror)
890 {
891 	u32 zone_num;
892 
893 	if (!zinfo)
894 		return false;
895 
896 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
897 	if (zone_num + 1 >= zinfo->nr_zones)
898 		return false;
899 
900 	if (!test_bit(zone_num, zinfo->seq_zones))
901 		return false;
902 
903 	return true;
904 }
905 
906 int btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
907 {
908 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
909 	struct blk_zone *zone;
910 	int i;
911 
912 	if (!is_sb_log_zone(zinfo, mirror))
913 		return 0;
914 
915 	zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
916 	for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
917 		/* Advance the next zone */
918 		if (zone->cond == BLK_ZONE_COND_FULL) {
919 			zone++;
920 			continue;
921 		}
922 
923 		if (zone->cond == BLK_ZONE_COND_EMPTY)
924 			zone->cond = BLK_ZONE_COND_IMP_OPEN;
925 
926 		zone->wp += SUPER_INFO_SECTORS;
927 
928 		if (sb_zone_is_full(zone)) {
929 			/*
930 			 * No room left to write new superblock. Since
931 			 * superblock is written with REQ_SYNC, it is safe to
932 			 * finish the zone now.
933 			 *
934 			 * If the write pointer is exactly at the capacity,
935 			 * explicit ZONE_FINISH is not necessary.
936 			 */
937 			if (zone->wp != zone->start + zone->capacity) {
938 				int ret;
939 
940 				ret = blkdev_zone_mgmt(device->bdev,
941 						REQ_OP_ZONE_FINISH, zone->start,
942 						zone->len, GFP_NOFS);
943 				if (ret)
944 					return ret;
945 			}
946 
947 			zone->wp = zone->start + zone->len;
948 			zone->cond = BLK_ZONE_COND_FULL;
949 		}
950 		return 0;
951 	}
952 
953 	/* All the zones are FULL. Should not reach here. */
954 	ASSERT(0);
955 	return -EIO;
956 }
957 
958 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
959 {
960 	sector_t zone_sectors;
961 	sector_t nr_sectors;
962 	u8 zone_sectors_shift;
963 	u32 sb_zone;
964 	u32 nr_zones;
965 
966 	zone_sectors = bdev_zone_sectors(bdev);
967 	zone_sectors_shift = ilog2(zone_sectors);
968 	nr_sectors = bdev_nr_sectors(bdev);
969 	nr_zones = nr_sectors >> zone_sectors_shift;
970 
971 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
972 	if (sb_zone + 1 >= nr_zones)
973 		return -ENOENT;
974 
975 	return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
976 				zone_start_sector(sb_zone, bdev),
977 				zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
978 }
979 
980 /**
981  * btrfs_find_allocatable_zones - find allocatable zones within a given region
982  *
983  * @device:	the device to allocate a region on
984  * @hole_start: the position of the hole to allocate the region
985  * @num_bytes:	size of wanted region
986  * @hole_end:	the end of the hole
987  * @return:	position of allocatable zones
988  *
989  * Allocatable region should not contain any superblock locations.
990  */
991 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
992 				 u64 hole_end, u64 num_bytes)
993 {
994 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
995 	const u8 shift = zinfo->zone_size_shift;
996 	u64 nzones = num_bytes >> shift;
997 	u64 pos = hole_start;
998 	u64 begin, end;
999 	bool have_sb;
1000 	int i;
1001 
1002 	ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
1003 	ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
1004 
1005 	while (pos < hole_end) {
1006 		begin = pos >> shift;
1007 		end = begin + nzones;
1008 
1009 		if (end > zinfo->nr_zones)
1010 			return hole_end;
1011 
1012 		/* Check if zones in the region are all empty */
1013 		if (btrfs_dev_is_sequential(device, pos) &&
1014 		    find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
1015 			pos += zinfo->zone_size;
1016 			continue;
1017 		}
1018 
1019 		have_sb = false;
1020 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1021 			u32 sb_zone;
1022 			u64 sb_pos;
1023 
1024 			sb_zone = sb_zone_number(shift, i);
1025 			if (!(end <= sb_zone ||
1026 			      sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
1027 				have_sb = true;
1028 				pos = zone_start_physical(
1029 					sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
1030 				break;
1031 			}
1032 
1033 			/* We also need to exclude regular superblock positions */
1034 			sb_pos = btrfs_sb_offset(i);
1035 			if (!(pos + num_bytes <= sb_pos ||
1036 			      sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
1037 				have_sb = true;
1038 				pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
1039 					    zinfo->zone_size);
1040 				break;
1041 			}
1042 		}
1043 		if (!have_sb)
1044 			break;
1045 	}
1046 
1047 	return pos;
1048 }
1049 
1050 static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos)
1051 {
1052 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
1053 	unsigned int zno = (pos >> zone_info->zone_size_shift);
1054 
1055 	/* We can use any number of zones */
1056 	if (zone_info->max_active_zones == 0)
1057 		return true;
1058 
1059 	if (!test_bit(zno, zone_info->active_zones)) {
1060 		/* Active zone left? */
1061 		if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0)
1062 			return false;
1063 		if (test_and_set_bit(zno, zone_info->active_zones)) {
1064 			/* Someone already set the bit */
1065 			atomic_inc(&zone_info->active_zones_left);
1066 		}
1067 	}
1068 
1069 	return true;
1070 }
1071 
1072 static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos)
1073 {
1074 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
1075 	unsigned int zno = (pos >> zone_info->zone_size_shift);
1076 
1077 	/* We can use any number of zones */
1078 	if (zone_info->max_active_zones == 0)
1079 		return;
1080 
1081 	if (test_and_clear_bit(zno, zone_info->active_zones))
1082 		atomic_inc(&zone_info->active_zones_left);
1083 }
1084 
1085 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
1086 			    u64 length, u64 *bytes)
1087 {
1088 	int ret;
1089 
1090 	*bytes = 0;
1091 	ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
1092 			       physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
1093 			       GFP_NOFS);
1094 	if (ret)
1095 		return ret;
1096 
1097 	*bytes = length;
1098 	while (length) {
1099 		btrfs_dev_set_zone_empty(device, physical);
1100 		btrfs_dev_clear_active_zone(device, physical);
1101 		physical += device->zone_info->zone_size;
1102 		length -= device->zone_info->zone_size;
1103 	}
1104 
1105 	return 0;
1106 }
1107 
1108 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
1109 {
1110 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
1111 	const u8 shift = zinfo->zone_size_shift;
1112 	unsigned long begin = start >> shift;
1113 	unsigned long end = (start + size) >> shift;
1114 	u64 pos;
1115 	int ret;
1116 
1117 	ASSERT(IS_ALIGNED(start, zinfo->zone_size));
1118 	ASSERT(IS_ALIGNED(size, zinfo->zone_size));
1119 
1120 	if (end > zinfo->nr_zones)
1121 		return -ERANGE;
1122 
1123 	/* All the zones are conventional */
1124 	if (find_next_bit(zinfo->seq_zones, begin, end) == end)
1125 		return 0;
1126 
1127 	/* All the zones are sequential and empty */
1128 	if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
1129 	    find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
1130 		return 0;
1131 
1132 	for (pos = start; pos < start + size; pos += zinfo->zone_size) {
1133 		u64 reset_bytes;
1134 
1135 		if (!btrfs_dev_is_sequential(device, pos) ||
1136 		    btrfs_dev_is_empty_zone(device, pos))
1137 			continue;
1138 
1139 		/* Free regions should be empty */
1140 		btrfs_warn_in_rcu(
1141 			device->fs_info,
1142 		"zoned: resetting device %s (devid %llu) zone %llu for allocation",
1143 			rcu_str_deref(device->name), device->devid, pos >> shift);
1144 		WARN_ON_ONCE(1);
1145 
1146 		ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1147 					      &reset_bytes);
1148 		if (ret)
1149 			return ret;
1150 	}
1151 
1152 	return 0;
1153 }
1154 
1155 /*
1156  * Calculate an allocation pointer from the extent allocation information
1157  * for a block group consist of conventional zones. It is pointed to the
1158  * end of the highest addressed extent in the block group as an allocation
1159  * offset.
1160  */
1161 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1162 				   u64 *offset_ret, bool new)
1163 {
1164 	struct btrfs_fs_info *fs_info = cache->fs_info;
1165 	struct btrfs_root *root;
1166 	struct btrfs_path *path;
1167 	struct btrfs_key key;
1168 	struct btrfs_key found_key;
1169 	int ret;
1170 	u64 length;
1171 
1172 	/*
1173 	 * Avoid  tree lookups for a new block group, there's no use for it.
1174 	 * It must always be 0.
1175 	 *
1176 	 * Also, we have a lock chain of extent buffer lock -> chunk mutex.
1177 	 * For new a block group, this function is called from
1178 	 * btrfs_make_block_group() which is already taking the chunk mutex.
1179 	 * Thus, we cannot call calculate_alloc_pointer() which takes extent
1180 	 * buffer locks to avoid deadlock.
1181 	 */
1182 	if (new) {
1183 		*offset_ret = 0;
1184 		return 0;
1185 	}
1186 
1187 	path = btrfs_alloc_path();
1188 	if (!path)
1189 		return -ENOMEM;
1190 
1191 	key.objectid = cache->start + cache->length;
1192 	key.type = 0;
1193 	key.offset = 0;
1194 
1195 	root = btrfs_extent_root(fs_info, key.objectid);
1196 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1197 	/* We should not find the exact match */
1198 	if (!ret)
1199 		ret = -EUCLEAN;
1200 	if (ret < 0)
1201 		goto out;
1202 
1203 	ret = btrfs_previous_extent_item(root, path, cache->start);
1204 	if (ret) {
1205 		if (ret == 1) {
1206 			ret = 0;
1207 			*offset_ret = 0;
1208 		}
1209 		goto out;
1210 	}
1211 
1212 	btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1213 
1214 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1215 		length = found_key.offset;
1216 	else
1217 		length = fs_info->nodesize;
1218 
1219 	if (!(found_key.objectid >= cache->start &&
1220 	       found_key.objectid + length <= cache->start + cache->length)) {
1221 		ret = -EUCLEAN;
1222 		goto out;
1223 	}
1224 	*offset_ret = found_key.objectid + length - cache->start;
1225 	ret = 0;
1226 
1227 out:
1228 	btrfs_free_path(path);
1229 	return ret;
1230 }
1231 
1232 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1233 {
1234 	struct btrfs_fs_info *fs_info = cache->fs_info;
1235 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1236 	struct extent_map *em;
1237 	struct map_lookup *map;
1238 	struct btrfs_device *device;
1239 	u64 logical = cache->start;
1240 	u64 length = cache->length;
1241 	int ret;
1242 	int i;
1243 	unsigned int nofs_flag;
1244 	u64 *alloc_offsets = NULL;
1245 	u64 *caps = NULL;
1246 	u64 *physical = NULL;
1247 	unsigned long *active = NULL;
1248 	u64 last_alloc = 0;
1249 	u32 num_sequential = 0, num_conventional = 0;
1250 
1251 	if (!btrfs_is_zoned(fs_info))
1252 		return 0;
1253 
1254 	/* Sanity check */
1255 	if (!IS_ALIGNED(length, fs_info->zone_size)) {
1256 		btrfs_err(fs_info,
1257 		"zoned: block group %llu len %llu unaligned to zone size %llu",
1258 			  logical, length, fs_info->zone_size);
1259 		return -EIO;
1260 	}
1261 
1262 	/* Get the chunk mapping */
1263 	read_lock(&em_tree->lock);
1264 	em = lookup_extent_mapping(em_tree, logical, length);
1265 	read_unlock(&em_tree->lock);
1266 
1267 	if (!em)
1268 		return -EINVAL;
1269 
1270 	map = em->map_lookup;
1271 
1272 	cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS);
1273 	if (!cache->physical_map) {
1274 		ret = -ENOMEM;
1275 		goto out;
1276 	}
1277 
1278 	alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1279 	if (!alloc_offsets) {
1280 		ret = -ENOMEM;
1281 		goto out;
1282 	}
1283 
1284 	caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS);
1285 	if (!caps) {
1286 		ret = -ENOMEM;
1287 		goto out;
1288 	}
1289 
1290 	physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS);
1291 	if (!physical) {
1292 		ret = -ENOMEM;
1293 		goto out;
1294 	}
1295 
1296 	active = bitmap_zalloc(map->num_stripes, GFP_NOFS);
1297 	if (!active) {
1298 		ret = -ENOMEM;
1299 		goto out;
1300 	}
1301 
1302 	for (i = 0; i < map->num_stripes; i++) {
1303 		bool is_sequential;
1304 		struct blk_zone zone;
1305 		struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1306 		int dev_replace_is_ongoing = 0;
1307 
1308 		device = map->stripes[i].dev;
1309 		physical[i] = map->stripes[i].physical;
1310 
1311 		if (device->bdev == NULL) {
1312 			alloc_offsets[i] = WP_MISSING_DEV;
1313 			continue;
1314 		}
1315 
1316 		is_sequential = btrfs_dev_is_sequential(device, physical[i]);
1317 		if (is_sequential)
1318 			num_sequential++;
1319 		else
1320 			num_conventional++;
1321 
1322 		/*
1323 		 * Consider a zone as active if we can allow any number of
1324 		 * active zones.
1325 		 */
1326 		if (!device->zone_info->max_active_zones)
1327 			__set_bit(i, active);
1328 
1329 		if (!is_sequential) {
1330 			alloc_offsets[i] = WP_CONVENTIONAL;
1331 			continue;
1332 		}
1333 
1334 		/*
1335 		 * This zone will be used for allocation, so mark this zone
1336 		 * non-empty.
1337 		 */
1338 		btrfs_dev_clear_zone_empty(device, physical[i]);
1339 
1340 		down_read(&dev_replace->rwsem);
1341 		dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1342 		if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1343 			btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]);
1344 		up_read(&dev_replace->rwsem);
1345 
1346 		/*
1347 		 * The group is mapped to a sequential zone. Get the zone write
1348 		 * pointer to determine the allocation offset within the zone.
1349 		 */
1350 		WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size));
1351 		nofs_flag = memalloc_nofs_save();
1352 		ret = btrfs_get_dev_zone(device, physical[i], &zone);
1353 		memalloc_nofs_restore(nofs_flag);
1354 		if (ret == -EIO || ret == -EOPNOTSUPP) {
1355 			ret = 0;
1356 			alloc_offsets[i] = WP_MISSING_DEV;
1357 			continue;
1358 		} else if (ret) {
1359 			goto out;
1360 		}
1361 
1362 		if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1363 			btrfs_err_in_rcu(fs_info,
1364 	"zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1365 				zone.start << SECTOR_SHIFT,
1366 				rcu_str_deref(device->name), device->devid);
1367 			ret = -EIO;
1368 			goto out;
1369 		}
1370 
1371 		caps[i] = (zone.capacity << SECTOR_SHIFT);
1372 
1373 		switch (zone.cond) {
1374 		case BLK_ZONE_COND_OFFLINE:
1375 		case BLK_ZONE_COND_READONLY:
1376 			btrfs_err(fs_info,
1377 		"zoned: offline/readonly zone %llu on device %s (devid %llu)",
1378 				  physical[i] >> device->zone_info->zone_size_shift,
1379 				  rcu_str_deref(device->name), device->devid);
1380 			alloc_offsets[i] = WP_MISSING_DEV;
1381 			break;
1382 		case BLK_ZONE_COND_EMPTY:
1383 			alloc_offsets[i] = 0;
1384 			break;
1385 		case BLK_ZONE_COND_FULL:
1386 			alloc_offsets[i] = caps[i];
1387 			break;
1388 		default:
1389 			/* Partially used zone */
1390 			alloc_offsets[i] =
1391 					((zone.wp - zone.start) << SECTOR_SHIFT);
1392 			__set_bit(i, active);
1393 			break;
1394 		}
1395 	}
1396 
1397 	if (num_sequential > 0)
1398 		cache->seq_zone = true;
1399 
1400 	if (num_conventional > 0) {
1401 		/* Zone capacity is always zone size in emulation */
1402 		cache->zone_capacity = cache->length;
1403 		ret = calculate_alloc_pointer(cache, &last_alloc, new);
1404 		if (ret) {
1405 			btrfs_err(fs_info,
1406 			"zoned: failed to determine allocation offset of bg %llu",
1407 				  cache->start);
1408 			goto out;
1409 		} else if (map->num_stripes == num_conventional) {
1410 			cache->alloc_offset = last_alloc;
1411 			set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1412 			goto out;
1413 		}
1414 	}
1415 
1416 	switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1417 	case 0: /* single */
1418 		if (alloc_offsets[0] == WP_MISSING_DEV) {
1419 			btrfs_err(fs_info,
1420 			"zoned: cannot recover write pointer for zone %llu",
1421 				physical[0]);
1422 			ret = -EIO;
1423 			goto out;
1424 		}
1425 		cache->alloc_offset = alloc_offsets[0];
1426 		cache->zone_capacity = caps[0];
1427 		if (test_bit(0, active))
1428 			set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1429 		break;
1430 	case BTRFS_BLOCK_GROUP_DUP:
1431 		if (map->type & BTRFS_BLOCK_GROUP_DATA) {
1432 			btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg");
1433 			ret = -EINVAL;
1434 			goto out;
1435 		}
1436 		if (alloc_offsets[0] == WP_MISSING_DEV) {
1437 			btrfs_err(fs_info,
1438 			"zoned: cannot recover write pointer for zone %llu",
1439 				physical[0]);
1440 			ret = -EIO;
1441 			goto out;
1442 		}
1443 		if (alloc_offsets[1] == WP_MISSING_DEV) {
1444 			btrfs_err(fs_info,
1445 			"zoned: cannot recover write pointer for zone %llu",
1446 				physical[1]);
1447 			ret = -EIO;
1448 			goto out;
1449 		}
1450 		if (alloc_offsets[0] != alloc_offsets[1]) {
1451 			btrfs_err(fs_info,
1452 			"zoned: write pointer offset mismatch of zones in DUP profile");
1453 			ret = -EIO;
1454 			goto out;
1455 		}
1456 		if (test_bit(0, active) != test_bit(1, active)) {
1457 			if (!btrfs_zone_activate(cache)) {
1458 				ret = -EIO;
1459 				goto out;
1460 			}
1461 		} else {
1462 			if (test_bit(0, active))
1463 				set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
1464 					&cache->runtime_flags);
1465 		}
1466 		cache->alloc_offset = alloc_offsets[0];
1467 		cache->zone_capacity = min(caps[0], caps[1]);
1468 		break;
1469 	case BTRFS_BLOCK_GROUP_RAID1:
1470 	case BTRFS_BLOCK_GROUP_RAID0:
1471 	case BTRFS_BLOCK_GROUP_RAID10:
1472 	case BTRFS_BLOCK_GROUP_RAID5:
1473 	case BTRFS_BLOCK_GROUP_RAID6:
1474 		/* non-single profiles are not supported yet */
1475 	default:
1476 		btrfs_err(fs_info, "zoned: profile %s not yet supported",
1477 			  btrfs_bg_type_to_raid_name(map->type));
1478 		ret = -EINVAL;
1479 		goto out;
1480 	}
1481 
1482 out:
1483 	if (cache->alloc_offset > fs_info->zone_size) {
1484 		btrfs_err(fs_info,
1485 			"zoned: invalid write pointer %llu in block group %llu",
1486 			cache->alloc_offset, cache->start);
1487 		ret = -EIO;
1488 	}
1489 
1490 	if (cache->alloc_offset > cache->zone_capacity) {
1491 		btrfs_err(fs_info,
1492 "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu",
1493 			  cache->alloc_offset, cache->zone_capacity,
1494 			  cache->start);
1495 		ret = -EIO;
1496 	}
1497 
1498 	/* An extent is allocated after the write pointer */
1499 	if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1500 		btrfs_err(fs_info,
1501 			  "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1502 			  logical, last_alloc, cache->alloc_offset);
1503 		ret = -EIO;
1504 	}
1505 
1506 	if (!ret) {
1507 		cache->meta_write_pointer = cache->alloc_offset + cache->start;
1508 		if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags)) {
1509 			btrfs_get_block_group(cache);
1510 			spin_lock(&fs_info->zone_active_bgs_lock);
1511 			list_add_tail(&cache->active_bg_list,
1512 				      &fs_info->zone_active_bgs);
1513 			spin_unlock(&fs_info->zone_active_bgs_lock);
1514 		}
1515 	} else {
1516 		kfree(cache->physical_map);
1517 		cache->physical_map = NULL;
1518 	}
1519 	bitmap_free(active);
1520 	kfree(physical);
1521 	kfree(caps);
1522 	kfree(alloc_offsets);
1523 	free_extent_map(em);
1524 
1525 	return ret;
1526 }
1527 
1528 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1529 {
1530 	u64 unusable, free;
1531 
1532 	if (!btrfs_is_zoned(cache->fs_info))
1533 		return;
1534 
1535 	WARN_ON(cache->bytes_super != 0);
1536 	unusable = (cache->alloc_offset - cache->used) +
1537 		   (cache->length - cache->zone_capacity);
1538 	free = cache->zone_capacity - cache->alloc_offset;
1539 
1540 	/* We only need ->free_space in ALLOC_SEQ block groups */
1541 	cache->cached = BTRFS_CACHE_FINISHED;
1542 	cache->free_space_ctl->free_space = free;
1543 	cache->zone_unusable = unusable;
1544 }
1545 
1546 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1547 			    struct extent_buffer *eb)
1548 {
1549 	struct btrfs_fs_info *fs_info = eb->fs_info;
1550 
1551 	if (!btrfs_is_zoned(fs_info) ||
1552 	    btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1553 	    !list_empty(&eb->release_list))
1554 		return;
1555 
1556 	set_extent_buffer_dirty(eb);
1557 	set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1558 			       eb->start + eb->len - 1, EXTENT_DIRTY);
1559 	memzero_extent_buffer(eb, 0, eb->len);
1560 	set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1561 
1562 	spin_lock(&trans->releasing_ebs_lock);
1563 	list_add_tail(&eb->release_list, &trans->releasing_ebs);
1564 	spin_unlock(&trans->releasing_ebs_lock);
1565 	atomic_inc(&eb->refs);
1566 }
1567 
1568 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1569 {
1570 	spin_lock(&trans->releasing_ebs_lock);
1571 	while (!list_empty(&trans->releasing_ebs)) {
1572 		struct extent_buffer *eb;
1573 
1574 		eb = list_first_entry(&trans->releasing_ebs,
1575 				      struct extent_buffer, release_list);
1576 		list_del_init(&eb->release_list);
1577 		free_extent_buffer(eb);
1578 	}
1579 	spin_unlock(&trans->releasing_ebs_lock);
1580 }
1581 
1582 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1583 {
1584 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1585 	struct btrfs_block_group *cache;
1586 	bool ret = false;
1587 
1588 	if (!btrfs_is_zoned(fs_info))
1589 		return false;
1590 
1591 	if (!is_data_inode(&inode->vfs_inode))
1592 		return false;
1593 
1594 	/*
1595 	 * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1596 	 * extent layout the relocation code has.
1597 	 * Furthermore we have set aside own block-group from which only the
1598 	 * relocation "process" can allocate and make sure only one process at a
1599 	 * time can add pages to an extent that gets relocated, so it's safe to
1600 	 * use regular REQ_OP_WRITE for this special case.
1601 	 */
1602 	if (btrfs_is_data_reloc_root(inode->root))
1603 		return false;
1604 
1605 	cache = btrfs_lookup_block_group(fs_info, start);
1606 	ASSERT(cache);
1607 	if (!cache)
1608 		return false;
1609 
1610 	ret = cache->seq_zone;
1611 	btrfs_put_block_group(cache);
1612 
1613 	return ret;
1614 }
1615 
1616 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1617 				 struct bio *bio)
1618 {
1619 	struct btrfs_ordered_extent *ordered;
1620 	const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1621 
1622 	if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1623 		return;
1624 
1625 	ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1626 	if (WARN_ON(!ordered))
1627 		return;
1628 
1629 	ordered->physical = physical;
1630 	ordered->bdev = bio->bi_bdev;
1631 
1632 	btrfs_put_ordered_extent(ordered);
1633 }
1634 
1635 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1636 {
1637 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1638 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1639 	struct extent_map_tree *em_tree;
1640 	struct extent_map *em;
1641 	struct btrfs_ordered_sum *sum;
1642 	u64 orig_logical = ordered->disk_bytenr;
1643 	u64 *logical = NULL;
1644 	int nr, stripe_len;
1645 
1646 	/* Zoned devices should not have partitions. So, we can assume it is 0 */
1647 	ASSERT(!bdev_is_partition(ordered->bdev));
1648 	if (WARN_ON(!ordered->bdev))
1649 		return;
1650 
1651 	if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev,
1652 				     ordered->physical, &logical, &nr,
1653 				     &stripe_len)))
1654 		goto out;
1655 
1656 	WARN_ON(nr != 1);
1657 
1658 	if (orig_logical == *logical)
1659 		goto out;
1660 
1661 	ordered->disk_bytenr = *logical;
1662 
1663 	em_tree = &inode->extent_tree;
1664 	write_lock(&em_tree->lock);
1665 	em = search_extent_mapping(em_tree, ordered->file_offset,
1666 				   ordered->num_bytes);
1667 	em->block_start = *logical;
1668 	free_extent_map(em);
1669 	write_unlock(&em_tree->lock);
1670 
1671 	list_for_each_entry(sum, &ordered->list, list) {
1672 		if (*logical < orig_logical)
1673 			sum->bytenr -= orig_logical - *logical;
1674 		else
1675 			sum->bytenr += *logical - orig_logical;
1676 	}
1677 
1678 out:
1679 	kfree(logical);
1680 }
1681 
1682 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1683 				    struct extent_buffer *eb,
1684 				    struct btrfs_block_group **cache_ret)
1685 {
1686 	struct btrfs_block_group *cache;
1687 	bool ret = true;
1688 
1689 	if (!btrfs_is_zoned(fs_info))
1690 		return true;
1691 
1692 	cache = btrfs_lookup_block_group(fs_info, eb->start);
1693 	if (!cache)
1694 		return true;
1695 
1696 	if (cache->meta_write_pointer != eb->start) {
1697 		btrfs_put_block_group(cache);
1698 		cache = NULL;
1699 		ret = false;
1700 	} else {
1701 		cache->meta_write_pointer = eb->start + eb->len;
1702 	}
1703 
1704 	*cache_ret = cache;
1705 
1706 	return ret;
1707 }
1708 
1709 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1710 				     struct extent_buffer *eb)
1711 {
1712 	if (!btrfs_is_zoned(eb->fs_info) || !cache)
1713 		return;
1714 
1715 	ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1716 	cache->meta_write_pointer = eb->start;
1717 }
1718 
1719 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1720 {
1721 	if (!btrfs_dev_is_sequential(device, physical))
1722 		return -EOPNOTSUPP;
1723 
1724 	return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1725 				    length >> SECTOR_SHIFT, GFP_NOFS, 0);
1726 }
1727 
1728 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1729 			  struct blk_zone *zone)
1730 {
1731 	struct btrfs_io_context *bioc = NULL;
1732 	u64 mapped_length = PAGE_SIZE;
1733 	unsigned int nofs_flag;
1734 	int nmirrors;
1735 	int i, ret;
1736 
1737 	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1738 			       &mapped_length, &bioc);
1739 	if (ret || !bioc || mapped_length < PAGE_SIZE) {
1740 		ret = -EIO;
1741 		goto out_put_bioc;
1742 	}
1743 
1744 	if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1745 		ret = -EINVAL;
1746 		goto out_put_bioc;
1747 	}
1748 
1749 	nofs_flag = memalloc_nofs_save();
1750 	nmirrors = (int)bioc->num_stripes;
1751 	for (i = 0; i < nmirrors; i++) {
1752 		u64 physical = bioc->stripes[i].physical;
1753 		struct btrfs_device *dev = bioc->stripes[i].dev;
1754 
1755 		/* Missing device */
1756 		if (!dev->bdev)
1757 			continue;
1758 
1759 		ret = btrfs_get_dev_zone(dev, physical, zone);
1760 		/* Failing device */
1761 		if (ret == -EIO || ret == -EOPNOTSUPP)
1762 			continue;
1763 		break;
1764 	}
1765 	memalloc_nofs_restore(nofs_flag);
1766 out_put_bioc:
1767 	btrfs_put_bioc(bioc);
1768 	return ret;
1769 }
1770 
1771 /*
1772  * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1773  * filling zeros between @physical_pos to a write pointer of dev-replace
1774  * source device.
1775  */
1776 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1777 				    u64 physical_start, u64 physical_pos)
1778 {
1779 	struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1780 	struct blk_zone zone;
1781 	u64 length;
1782 	u64 wp;
1783 	int ret;
1784 
1785 	if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1786 		return 0;
1787 
1788 	ret = read_zone_info(fs_info, logical, &zone);
1789 	if (ret)
1790 		return ret;
1791 
1792 	wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1793 
1794 	if (physical_pos == wp)
1795 		return 0;
1796 
1797 	if (physical_pos > wp)
1798 		return -EUCLEAN;
1799 
1800 	length = wp - physical_pos;
1801 	return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1802 }
1803 
1804 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1805 					    u64 logical, u64 length)
1806 {
1807 	struct btrfs_device *device;
1808 	struct extent_map *em;
1809 	struct map_lookup *map;
1810 
1811 	em = btrfs_get_chunk_map(fs_info, logical, length);
1812 	if (IS_ERR(em))
1813 		return ERR_CAST(em);
1814 
1815 	map = em->map_lookup;
1816 	/* We only support single profile for now */
1817 	device = map->stripes[0].dev;
1818 
1819 	free_extent_map(em);
1820 
1821 	return device;
1822 }
1823 
1824 /**
1825  * Activate block group and underlying device zones
1826  *
1827  * @block_group: the block group to activate
1828  *
1829  * Return: true on success, false otherwise
1830  */
1831 bool btrfs_zone_activate(struct btrfs_block_group *block_group)
1832 {
1833 	struct btrfs_fs_info *fs_info = block_group->fs_info;
1834 	struct btrfs_space_info *space_info = block_group->space_info;
1835 	struct map_lookup *map;
1836 	struct btrfs_device *device;
1837 	u64 physical;
1838 	bool ret;
1839 	int i;
1840 
1841 	if (!btrfs_is_zoned(block_group->fs_info))
1842 		return true;
1843 
1844 	map = block_group->physical_map;
1845 
1846 	spin_lock(&space_info->lock);
1847 	spin_lock(&block_group->lock);
1848 	if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1849 		ret = true;
1850 		goto out_unlock;
1851 	}
1852 
1853 	/* No space left */
1854 	if (btrfs_zoned_bg_is_full(block_group)) {
1855 		ret = false;
1856 		goto out_unlock;
1857 	}
1858 
1859 	for (i = 0; i < map->num_stripes; i++) {
1860 		device = map->stripes[i].dev;
1861 		physical = map->stripes[i].physical;
1862 
1863 		if (device->zone_info->max_active_zones == 0)
1864 			continue;
1865 
1866 		if (!btrfs_dev_set_active_zone(device, physical)) {
1867 			/* Cannot activate the zone */
1868 			ret = false;
1869 			goto out_unlock;
1870 		}
1871 	}
1872 
1873 	/* Successfully activated all the zones */
1874 	set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
1875 	space_info->active_total_bytes += block_group->length;
1876 	spin_unlock(&block_group->lock);
1877 	btrfs_try_granting_tickets(fs_info, space_info);
1878 	spin_unlock(&space_info->lock);
1879 
1880 	/* For the active block group list */
1881 	btrfs_get_block_group(block_group);
1882 
1883 	spin_lock(&fs_info->zone_active_bgs_lock);
1884 	list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs);
1885 	spin_unlock(&fs_info->zone_active_bgs_lock);
1886 
1887 	return true;
1888 
1889 out_unlock:
1890 	spin_unlock(&block_group->lock);
1891 	spin_unlock(&space_info->lock);
1892 	return ret;
1893 }
1894 
1895 static void wait_eb_writebacks(struct btrfs_block_group *block_group)
1896 {
1897 	struct btrfs_fs_info *fs_info = block_group->fs_info;
1898 	const u64 end = block_group->start + block_group->length;
1899 	struct radix_tree_iter iter;
1900 	struct extent_buffer *eb;
1901 	void __rcu **slot;
1902 
1903 	rcu_read_lock();
1904 	radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter,
1905 				 block_group->start >> fs_info->sectorsize_bits) {
1906 		eb = radix_tree_deref_slot(slot);
1907 		if (!eb)
1908 			continue;
1909 		if (radix_tree_deref_retry(eb)) {
1910 			slot = radix_tree_iter_retry(&iter);
1911 			continue;
1912 		}
1913 
1914 		if (eb->start < block_group->start)
1915 			continue;
1916 		if (eb->start >= end)
1917 			break;
1918 
1919 		slot = radix_tree_iter_resume(slot, &iter);
1920 		rcu_read_unlock();
1921 		wait_on_extent_buffer_writeback(eb);
1922 		rcu_read_lock();
1923 	}
1924 	rcu_read_unlock();
1925 }
1926 
1927 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
1928 {
1929 	struct btrfs_fs_info *fs_info = block_group->fs_info;
1930 	struct map_lookup *map;
1931 	const bool is_metadata = (block_group->flags &
1932 			(BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM));
1933 	int ret = 0;
1934 	int i;
1935 
1936 	spin_lock(&block_group->lock);
1937 	if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1938 		spin_unlock(&block_group->lock);
1939 		return 0;
1940 	}
1941 
1942 	/* Check if we have unwritten allocated space */
1943 	if (is_metadata &&
1944 	    block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
1945 		spin_unlock(&block_group->lock);
1946 		return -EAGAIN;
1947 	}
1948 
1949 	/*
1950 	 * If we are sure that the block group is full (= no more room left for
1951 	 * new allocation) and the IO for the last usable block is completed, we
1952 	 * don't need to wait for the other IOs. This holds because we ensure
1953 	 * the sequential IO submissions using the ZONE_APPEND command for data
1954 	 * and block_group->meta_write_pointer for metadata.
1955 	 */
1956 	if (!fully_written) {
1957 		spin_unlock(&block_group->lock);
1958 
1959 		ret = btrfs_inc_block_group_ro(block_group, false);
1960 		if (ret)
1961 			return ret;
1962 
1963 		/* Ensure all writes in this block group finish */
1964 		btrfs_wait_block_group_reservations(block_group);
1965 		/* No need to wait for NOCOW writers. Zoned mode does not allow that */
1966 		btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
1967 					 block_group->length);
1968 		/* Wait for extent buffers to be written. */
1969 		if (is_metadata)
1970 			wait_eb_writebacks(block_group);
1971 
1972 		spin_lock(&block_group->lock);
1973 
1974 		/*
1975 		 * Bail out if someone already deactivated the block group, or
1976 		 * allocated space is left in the block group.
1977 		 */
1978 		if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
1979 			      &block_group->runtime_flags)) {
1980 			spin_unlock(&block_group->lock);
1981 			btrfs_dec_block_group_ro(block_group);
1982 			return 0;
1983 		}
1984 
1985 		if (block_group->reserved) {
1986 			spin_unlock(&block_group->lock);
1987 			btrfs_dec_block_group_ro(block_group);
1988 			return -EAGAIN;
1989 		}
1990 	}
1991 
1992 	clear_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
1993 	block_group->alloc_offset = block_group->zone_capacity;
1994 	block_group->free_space_ctl->free_space = 0;
1995 	btrfs_clear_treelog_bg(block_group);
1996 	btrfs_clear_data_reloc_bg(block_group);
1997 	spin_unlock(&block_group->lock);
1998 
1999 	map = block_group->physical_map;
2000 	for (i = 0; i < map->num_stripes; i++) {
2001 		struct btrfs_device *device = map->stripes[i].dev;
2002 		const u64 physical = map->stripes[i].physical;
2003 
2004 		if (device->zone_info->max_active_zones == 0)
2005 			continue;
2006 
2007 		ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH,
2008 				       physical >> SECTOR_SHIFT,
2009 				       device->zone_info->zone_size >> SECTOR_SHIFT,
2010 				       GFP_NOFS);
2011 
2012 		if (ret)
2013 			return ret;
2014 
2015 		btrfs_dev_clear_active_zone(device, physical);
2016 	}
2017 
2018 	if (!fully_written)
2019 		btrfs_dec_block_group_ro(block_group);
2020 
2021 	spin_lock(&fs_info->zone_active_bgs_lock);
2022 	ASSERT(!list_empty(&block_group->active_bg_list));
2023 	list_del_init(&block_group->active_bg_list);
2024 	spin_unlock(&fs_info->zone_active_bgs_lock);
2025 
2026 	/* For active_bg_list */
2027 	btrfs_put_block_group(block_group);
2028 
2029 	clear_and_wake_up_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2030 
2031 	return 0;
2032 }
2033 
2034 int btrfs_zone_finish(struct btrfs_block_group *block_group)
2035 {
2036 	if (!btrfs_is_zoned(block_group->fs_info))
2037 		return 0;
2038 
2039 	return do_zone_finish(block_group, false);
2040 }
2041 
2042 bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags)
2043 {
2044 	struct btrfs_fs_info *fs_info = fs_devices->fs_info;
2045 	struct btrfs_device *device;
2046 	bool ret = false;
2047 
2048 	if (!btrfs_is_zoned(fs_info))
2049 		return true;
2050 
2051 	/* Check if there is a device with active zones left */
2052 	mutex_lock(&fs_info->chunk_mutex);
2053 	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
2054 		struct btrfs_zoned_device_info *zinfo = device->zone_info;
2055 
2056 		if (!device->bdev)
2057 			continue;
2058 
2059 		if (!zinfo->max_active_zones ||
2060 		    atomic_read(&zinfo->active_zones_left)) {
2061 			ret = true;
2062 			break;
2063 		}
2064 	}
2065 	mutex_unlock(&fs_info->chunk_mutex);
2066 
2067 	if (!ret)
2068 		set_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2069 
2070 	return ret;
2071 }
2072 
2073 void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length)
2074 {
2075 	struct btrfs_block_group *block_group;
2076 	u64 min_alloc_bytes;
2077 
2078 	if (!btrfs_is_zoned(fs_info))
2079 		return;
2080 
2081 	block_group = btrfs_lookup_block_group(fs_info, logical);
2082 	ASSERT(block_group);
2083 
2084 	/* No MIXED_BG on zoned btrfs. */
2085 	if (block_group->flags & BTRFS_BLOCK_GROUP_DATA)
2086 		min_alloc_bytes = fs_info->sectorsize;
2087 	else
2088 		min_alloc_bytes = fs_info->nodesize;
2089 
2090 	/* Bail out if we can allocate more data from this block group. */
2091 	if (logical + length + min_alloc_bytes <=
2092 	    block_group->start + block_group->zone_capacity)
2093 		goto out;
2094 
2095 	do_zone_finish(block_group, true);
2096 
2097 out:
2098 	btrfs_put_block_group(block_group);
2099 }
2100 
2101 static void btrfs_zone_finish_endio_workfn(struct work_struct *work)
2102 {
2103 	struct btrfs_block_group *bg =
2104 		container_of(work, struct btrfs_block_group, zone_finish_work);
2105 
2106 	wait_on_extent_buffer_writeback(bg->last_eb);
2107 	free_extent_buffer(bg->last_eb);
2108 	btrfs_zone_finish_endio(bg->fs_info, bg->start, bg->length);
2109 	btrfs_put_block_group(bg);
2110 }
2111 
2112 void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg,
2113 				   struct extent_buffer *eb)
2114 {
2115 	if (!bg->seq_zone || eb->start + eb->len * 2 <= bg->start + bg->zone_capacity)
2116 		return;
2117 
2118 	if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) {
2119 		btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing",
2120 			  bg->start);
2121 		return;
2122 	}
2123 
2124 	/* For the work */
2125 	btrfs_get_block_group(bg);
2126 	atomic_inc(&eb->refs);
2127 	bg->last_eb = eb;
2128 	INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn);
2129 	queue_work(system_unbound_wq, &bg->zone_finish_work);
2130 }
2131 
2132 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
2133 {
2134 	struct btrfs_fs_info *fs_info = bg->fs_info;
2135 
2136 	spin_lock(&fs_info->relocation_bg_lock);
2137 	if (fs_info->data_reloc_bg == bg->start)
2138 		fs_info->data_reloc_bg = 0;
2139 	spin_unlock(&fs_info->relocation_bg_lock);
2140 }
2141 
2142 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
2143 {
2144 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2145 	struct btrfs_device *device;
2146 
2147 	if (!btrfs_is_zoned(fs_info))
2148 		return;
2149 
2150 	mutex_lock(&fs_devices->device_list_mutex);
2151 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2152 		if (device->zone_info) {
2153 			vfree(device->zone_info->zone_cache);
2154 			device->zone_info->zone_cache = NULL;
2155 		}
2156 	}
2157 	mutex_unlock(&fs_devices->device_list_mutex);
2158 }
2159 
2160 bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info)
2161 {
2162 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2163 	struct btrfs_device *device;
2164 	u64 used = 0;
2165 	u64 total = 0;
2166 	u64 factor;
2167 
2168 	ASSERT(btrfs_is_zoned(fs_info));
2169 
2170 	if (fs_info->bg_reclaim_threshold == 0)
2171 		return false;
2172 
2173 	mutex_lock(&fs_devices->device_list_mutex);
2174 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2175 		if (!device->bdev)
2176 			continue;
2177 
2178 		total += device->disk_total_bytes;
2179 		used += device->bytes_used;
2180 	}
2181 	mutex_unlock(&fs_devices->device_list_mutex);
2182 
2183 	factor = div64_u64(used * 100, total);
2184 	return factor >= fs_info->bg_reclaim_threshold;
2185 }
2186 
2187 void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical,
2188 				       u64 length)
2189 {
2190 	struct btrfs_block_group *block_group;
2191 
2192 	if (!btrfs_is_zoned(fs_info))
2193 		return;
2194 
2195 	block_group = btrfs_lookup_block_group(fs_info, logical);
2196 	/* It should be called on a previous data relocation block group. */
2197 	ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA));
2198 
2199 	spin_lock(&block_group->lock);
2200 	if (!test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))
2201 		goto out;
2202 
2203 	/* All relocation extents are written. */
2204 	if (block_group->start + block_group->alloc_offset == logical + length) {
2205 		/* Now, release this block group for further allocations. */
2206 		clear_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC,
2207 			  &block_group->runtime_flags);
2208 	}
2209 
2210 out:
2211 	spin_unlock(&block_group->lock);
2212 	btrfs_put_block_group(block_group);
2213 }
2214 
2215 int btrfs_zone_finish_one_bg(struct btrfs_fs_info *fs_info)
2216 {
2217 	struct btrfs_block_group *block_group;
2218 	struct btrfs_block_group *min_bg = NULL;
2219 	u64 min_avail = U64_MAX;
2220 	int ret;
2221 
2222 	spin_lock(&fs_info->zone_active_bgs_lock);
2223 	list_for_each_entry(block_group, &fs_info->zone_active_bgs,
2224 			    active_bg_list) {
2225 		u64 avail;
2226 
2227 		spin_lock(&block_group->lock);
2228 		if (block_group->reserved ||
2229 		    (block_group->flags & BTRFS_BLOCK_GROUP_SYSTEM)) {
2230 			spin_unlock(&block_group->lock);
2231 			continue;
2232 		}
2233 
2234 		avail = block_group->zone_capacity - block_group->alloc_offset;
2235 		if (min_avail > avail) {
2236 			if (min_bg)
2237 				btrfs_put_block_group(min_bg);
2238 			min_bg = block_group;
2239 			min_avail = avail;
2240 			btrfs_get_block_group(min_bg);
2241 		}
2242 		spin_unlock(&block_group->lock);
2243 	}
2244 	spin_unlock(&fs_info->zone_active_bgs_lock);
2245 
2246 	if (!min_bg)
2247 		return 0;
2248 
2249 	ret = btrfs_zone_finish(min_bg);
2250 	btrfs_put_block_group(min_bg);
2251 
2252 	return ret < 0 ? ret : 1;
2253 }
2254 
2255 int btrfs_zoned_activate_one_bg(struct btrfs_fs_info *fs_info,
2256 				struct btrfs_space_info *space_info,
2257 				bool do_finish)
2258 {
2259 	struct btrfs_block_group *bg;
2260 	int index;
2261 
2262 	if (!btrfs_is_zoned(fs_info) || (space_info->flags & BTRFS_BLOCK_GROUP_DATA))
2263 		return 0;
2264 
2265 	/* No more block groups to activate */
2266 	if (space_info->active_total_bytes == space_info->total_bytes)
2267 		return 0;
2268 
2269 	for (;;) {
2270 		int ret;
2271 		bool need_finish = false;
2272 
2273 		down_read(&space_info->groups_sem);
2274 		for (index = 0; index < BTRFS_NR_RAID_TYPES; index++) {
2275 			list_for_each_entry(bg, &space_info->block_groups[index],
2276 					    list) {
2277 				if (!spin_trylock(&bg->lock))
2278 					continue;
2279 				if (btrfs_zoned_bg_is_full(bg) ||
2280 				    test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2281 					     &bg->runtime_flags)) {
2282 					spin_unlock(&bg->lock);
2283 					continue;
2284 				}
2285 				spin_unlock(&bg->lock);
2286 
2287 				if (btrfs_zone_activate(bg)) {
2288 					up_read(&space_info->groups_sem);
2289 					return 1;
2290 				}
2291 
2292 				need_finish = true;
2293 			}
2294 		}
2295 		up_read(&space_info->groups_sem);
2296 
2297 		if (!do_finish || !need_finish)
2298 			break;
2299 
2300 		ret = btrfs_zone_finish_one_bg(fs_info);
2301 		if (ret == 0)
2302 			break;
2303 		if (ret < 0)
2304 			return ret;
2305 	}
2306 
2307 	return 0;
2308 }
2309