xref: /linux/fs/btrfs/zoned.c (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
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 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
656 {
657 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
658 	struct btrfs_device *device;
659 	u64 zoned_devices = 0;
660 	u64 nr_devices = 0;
661 	u64 zone_size = 0;
662 	u64 max_zone_append_size = 0;
663 	const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED);
664 	int ret = 0;
665 
666 	/* Count zoned devices */
667 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
668 		enum blk_zoned_model model;
669 
670 		if (!device->bdev)
671 			continue;
672 
673 		model = bdev_zoned_model(device->bdev);
674 		/*
675 		 * A Host-Managed zoned device must be used as a zoned device.
676 		 * A Host-Aware zoned device and a non-zoned devices can be
677 		 * treated as a zoned device, if ZONED flag is enabled in the
678 		 * superblock.
679 		 */
680 		if (model == BLK_ZONED_HM ||
681 		    (model == BLK_ZONED_HA && incompat_zoned) ||
682 		    (model == BLK_ZONED_NONE && incompat_zoned)) {
683 			struct btrfs_zoned_device_info *zone_info;
684 
685 			zone_info = device->zone_info;
686 			zoned_devices++;
687 			if (!zone_size) {
688 				zone_size = zone_info->zone_size;
689 			} else if (zone_info->zone_size != zone_size) {
690 				btrfs_err(fs_info,
691 		"zoned: unequal block device zone sizes: have %llu found %llu",
692 					  device->zone_info->zone_size,
693 					  zone_size);
694 				ret = -EINVAL;
695 				goto out;
696 			}
697 			if (!max_zone_append_size ||
698 			    (zone_info->max_zone_append_size &&
699 			     zone_info->max_zone_append_size < max_zone_append_size))
700 				max_zone_append_size =
701 					zone_info->max_zone_append_size;
702 		}
703 		nr_devices++;
704 	}
705 
706 	if (!zoned_devices && !incompat_zoned)
707 		goto out;
708 
709 	if (!zoned_devices && incompat_zoned) {
710 		/* No zoned block device found on ZONED filesystem */
711 		btrfs_err(fs_info,
712 			  "zoned: no zoned devices found on a zoned filesystem");
713 		ret = -EINVAL;
714 		goto out;
715 	}
716 
717 	if (zoned_devices && !incompat_zoned) {
718 		btrfs_err(fs_info,
719 			  "zoned: mode not enabled but zoned device found");
720 		ret = -EINVAL;
721 		goto out;
722 	}
723 
724 	if (zoned_devices != nr_devices) {
725 		btrfs_err(fs_info,
726 			  "zoned: cannot mix zoned and regular devices");
727 		ret = -EINVAL;
728 		goto out;
729 	}
730 
731 	/*
732 	 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
733 	 * btrfs_create_chunk(). Since we want stripe_len == zone_size,
734 	 * check the alignment here.
735 	 */
736 	if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
737 		btrfs_err(fs_info,
738 			  "zoned: zone size %llu not aligned to stripe %u",
739 			  zone_size, BTRFS_STRIPE_LEN);
740 		ret = -EINVAL;
741 		goto out;
742 	}
743 
744 	if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
745 		btrfs_err(fs_info, "zoned: mixed block groups not supported");
746 		ret = -EINVAL;
747 		goto out;
748 	}
749 
750 	fs_info->zone_size = zone_size;
751 	fs_info->max_zone_append_size = ALIGN_DOWN(max_zone_append_size,
752 						   fs_info->sectorsize);
753 	fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
754 	if (fs_info->max_zone_append_size < fs_info->max_extent_size)
755 		fs_info->max_extent_size = fs_info->max_zone_append_size;
756 
757 	/*
758 	 * Check mount options here, because we might change fs_info->zoned
759 	 * from fs_info->zone_size.
760 	 */
761 	ret = btrfs_check_mountopts_zoned(fs_info);
762 	if (ret)
763 		goto out;
764 
765 	btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
766 out:
767 	return ret;
768 }
769 
770 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
771 {
772 	if (!btrfs_is_zoned(info))
773 		return 0;
774 
775 	/*
776 	 * Space cache writing is not COWed. Disable that to avoid write errors
777 	 * in sequential zones.
778 	 */
779 	if (btrfs_test_opt(info, SPACE_CACHE)) {
780 		btrfs_err(info, "zoned: space cache v1 is not supported");
781 		return -EINVAL;
782 	}
783 
784 	if (btrfs_test_opt(info, NODATACOW)) {
785 		btrfs_err(info, "zoned: NODATACOW not supported");
786 		return -EINVAL;
787 	}
788 
789 	return 0;
790 }
791 
792 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
793 			   int rw, u64 *bytenr_ret)
794 {
795 	u64 wp;
796 	int ret;
797 
798 	if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
799 		*bytenr_ret = zones[0].start << SECTOR_SHIFT;
800 		return 0;
801 	}
802 
803 	ret = sb_write_pointer(bdev, zones, &wp);
804 	if (ret != -ENOENT && ret < 0)
805 		return ret;
806 
807 	if (rw == WRITE) {
808 		struct blk_zone *reset = NULL;
809 
810 		if (wp == zones[0].start << SECTOR_SHIFT)
811 			reset = &zones[0];
812 		else if (wp == zones[1].start << SECTOR_SHIFT)
813 			reset = &zones[1];
814 
815 		if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
816 			ASSERT(sb_zone_is_full(reset));
817 
818 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
819 					       reset->start, reset->len,
820 					       GFP_NOFS);
821 			if (ret)
822 				return ret;
823 
824 			reset->cond = BLK_ZONE_COND_EMPTY;
825 			reset->wp = reset->start;
826 		}
827 	} else if (ret != -ENOENT) {
828 		/*
829 		 * For READ, we want the previous one. Move write pointer to
830 		 * the end of a zone, if it is at the head of a zone.
831 		 */
832 		u64 zone_end = 0;
833 
834 		if (wp == zones[0].start << SECTOR_SHIFT)
835 			zone_end = zones[1].start + zones[1].capacity;
836 		else if (wp == zones[1].start << SECTOR_SHIFT)
837 			zone_end = zones[0].start + zones[0].capacity;
838 		if (zone_end)
839 			wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT,
840 					BTRFS_SUPER_INFO_SIZE);
841 
842 		wp -= BTRFS_SUPER_INFO_SIZE;
843 	}
844 
845 	*bytenr_ret = wp;
846 	return 0;
847 
848 }
849 
850 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
851 			       u64 *bytenr_ret)
852 {
853 	struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
854 	sector_t zone_sectors;
855 	u32 sb_zone;
856 	int ret;
857 	u8 zone_sectors_shift;
858 	sector_t nr_sectors;
859 	u32 nr_zones;
860 
861 	if (!bdev_is_zoned(bdev)) {
862 		*bytenr_ret = btrfs_sb_offset(mirror);
863 		return 0;
864 	}
865 
866 	ASSERT(rw == READ || rw == WRITE);
867 
868 	zone_sectors = bdev_zone_sectors(bdev);
869 	if (!is_power_of_2(zone_sectors))
870 		return -EINVAL;
871 	zone_sectors_shift = ilog2(zone_sectors);
872 	nr_sectors = bdev_nr_sectors(bdev);
873 	nr_zones = nr_sectors >> zone_sectors_shift;
874 
875 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
876 	if (sb_zone + 1 >= nr_zones)
877 		return -ENOENT;
878 
879 	ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
880 				  BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
881 				  zones);
882 	if (ret < 0)
883 		return ret;
884 	if (ret != BTRFS_NR_SB_LOG_ZONES)
885 		return -EIO;
886 
887 	return sb_log_location(bdev, zones, rw, bytenr_ret);
888 }
889 
890 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
891 			  u64 *bytenr_ret)
892 {
893 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
894 	u32 zone_num;
895 
896 	/*
897 	 * For a zoned filesystem on a non-zoned block device, use the same
898 	 * super block locations as regular filesystem. Doing so, the super
899 	 * block can always be retrieved and the zoned flag of the volume
900 	 * detected from the super block information.
901 	 */
902 	if (!bdev_is_zoned(device->bdev)) {
903 		*bytenr_ret = btrfs_sb_offset(mirror);
904 		return 0;
905 	}
906 
907 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
908 	if (zone_num + 1 >= zinfo->nr_zones)
909 		return -ENOENT;
910 
911 	return sb_log_location(device->bdev,
912 			       &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
913 			       rw, bytenr_ret);
914 }
915 
916 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
917 				  int mirror)
918 {
919 	u32 zone_num;
920 
921 	if (!zinfo)
922 		return false;
923 
924 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
925 	if (zone_num + 1 >= zinfo->nr_zones)
926 		return false;
927 
928 	if (!test_bit(zone_num, zinfo->seq_zones))
929 		return false;
930 
931 	return true;
932 }
933 
934 int btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
935 {
936 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
937 	struct blk_zone *zone;
938 	int i;
939 
940 	if (!is_sb_log_zone(zinfo, mirror))
941 		return 0;
942 
943 	zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
944 	for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
945 		/* Advance the next zone */
946 		if (zone->cond == BLK_ZONE_COND_FULL) {
947 			zone++;
948 			continue;
949 		}
950 
951 		if (zone->cond == BLK_ZONE_COND_EMPTY)
952 			zone->cond = BLK_ZONE_COND_IMP_OPEN;
953 
954 		zone->wp += SUPER_INFO_SECTORS;
955 
956 		if (sb_zone_is_full(zone)) {
957 			/*
958 			 * No room left to write new superblock. Since
959 			 * superblock is written with REQ_SYNC, it is safe to
960 			 * finish the zone now.
961 			 *
962 			 * If the write pointer is exactly at the capacity,
963 			 * explicit ZONE_FINISH is not necessary.
964 			 */
965 			if (zone->wp != zone->start + zone->capacity) {
966 				int ret;
967 
968 				ret = blkdev_zone_mgmt(device->bdev,
969 						REQ_OP_ZONE_FINISH, zone->start,
970 						zone->len, GFP_NOFS);
971 				if (ret)
972 					return ret;
973 			}
974 
975 			zone->wp = zone->start + zone->len;
976 			zone->cond = BLK_ZONE_COND_FULL;
977 		}
978 		return 0;
979 	}
980 
981 	/* All the zones are FULL. Should not reach here. */
982 	ASSERT(0);
983 	return -EIO;
984 }
985 
986 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
987 {
988 	sector_t zone_sectors;
989 	sector_t nr_sectors;
990 	u8 zone_sectors_shift;
991 	u32 sb_zone;
992 	u32 nr_zones;
993 
994 	zone_sectors = bdev_zone_sectors(bdev);
995 	zone_sectors_shift = ilog2(zone_sectors);
996 	nr_sectors = bdev_nr_sectors(bdev);
997 	nr_zones = nr_sectors >> zone_sectors_shift;
998 
999 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
1000 	if (sb_zone + 1 >= nr_zones)
1001 		return -ENOENT;
1002 
1003 	return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1004 				zone_start_sector(sb_zone, bdev),
1005 				zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
1006 }
1007 
1008 /**
1009  * btrfs_find_allocatable_zones - find allocatable zones within a given region
1010  *
1011  * @device:	the device to allocate a region on
1012  * @hole_start: the position of the hole to allocate the region
1013  * @num_bytes:	size of wanted region
1014  * @hole_end:	the end of the hole
1015  * @return:	position of allocatable zones
1016  *
1017  * Allocatable region should not contain any superblock locations.
1018  */
1019 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
1020 				 u64 hole_end, u64 num_bytes)
1021 {
1022 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
1023 	const u8 shift = zinfo->zone_size_shift;
1024 	u64 nzones = num_bytes >> shift;
1025 	u64 pos = hole_start;
1026 	u64 begin, end;
1027 	bool have_sb;
1028 	int i;
1029 
1030 	ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
1031 	ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
1032 
1033 	while (pos < hole_end) {
1034 		begin = pos >> shift;
1035 		end = begin + nzones;
1036 
1037 		if (end > zinfo->nr_zones)
1038 			return hole_end;
1039 
1040 		/* Check if zones in the region are all empty */
1041 		if (btrfs_dev_is_sequential(device, pos) &&
1042 		    find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
1043 			pos += zinfo->zone_size;
1044 			continue;
1045 		}
1046 
1047 		have_sb = false;
1048 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1049 			u32 sb_zone;
1050 			u64 sb_pos;
1051 
1052 			sb_zone = sb_zone_number(shift, i);
1053 			if (!(end <= sb_zone ||
1054 			      sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
1055 				have_sb = true;
1056 				pos = zone_start_physical(
1057 					sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
1058 				break;
1059 			}
1060 
1061 			/* We also need to exclude regular superblock positions */
1062 			sb_pos = btrfs_sb_offset(i);
1063 			if (!(pos + num_bytes <= sb_pos ||
1064 			      sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
1065 				have_sb = true;
1066 				pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
1067 					    zinfo->zone_size);
1068 				break;
1069 			}
1070 		}
1071 		if (!have_sb)
1072 			break;
1073 	}
1074 
1075 	return pos;
1076 }
1077 
1078 static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos)
1079 {
1080 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
1081 	unsigned int zno = (pos >> zone_info->zone_size_shift);
1082 
1083 	/* We can use any number of zones */
1084 	if (zone_info->max_active_zones == 0)
1085 		return true;
1086 
1087 	if (!test_bit(zno, zone_info->active_zones)) {
1088 		/* Active zone left? */
1089 		if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0)
1090 			return false;
1091 		if (test_and_set_bit(zno, zone_info->active_zones)) {
1092 			/* Someone already set the bit */
1093 			atomic_inc(&zone_info->active_zones_left);
1094 		}
1095 	}
1096 
1097 	return true;
1098 }
1099 
1100 static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos)
1101 {
1102 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
1103 	unsigned int zno = (pos >> zone_info->zone_size_shift);
1104 
1105 	/* We can use any number of zones */
1106 	if (zone_info->max_active_zones == 0)
1107 		return;
1108 
1109 	if (test_and_clear_bit(zno, zone_info->active_zones))
1110 		atomic_inc(&zone_info->active_zones_left);
1111 }
1112 
1113 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
1114 			    u64 length, u64 *bytes)
1115 {
1116 	int ret;
1117 
1118 	*bytes = 0;
1119 	ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
1120 			       physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
1121 			       GFP_NOFS);
1122 	if (ret)
1123 		return ret;
1124 
1125 	*bytes = length;
1126 	while (length) {
1127 		btrfs_dev_set_zone_empty(device, physical);
1128 		btrfs_dev_clear_active_zone(device, physical);
1129 		physical += device->zone_info->zone_size;
1130 		length -= device->zone_info->zone_size;
1131 	}
1132 
1133 	return 0;
1134 }
1135 
1136 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
1137 {
1138 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
1139 	const u8 shift = zinfo->zone_size_shift;
1140 	unsigned long begin = start >> shift;
1141 	unsigned long end = (start + size) >> shift;
1142 	u64 pos;
1143 	int ret;
1144 
1145 	ASSERT(IS_ALIGNED(start, zinfo->zone_size));
1146 	ASSERT(IS_ALIGNED(size, zinfo->zone_size));
1147 
1148 	if (end > zinfo->nr_zones)
1149 		return -ERANGE;
1150 
1151 	/* All the zones are conventional */
1152 	if (find_next_bit(zinfo->seq_zones, begin, end) == end)
1153 		return 0;
1154 
1155 	/* All the zones are sequential and empty */
1156 	if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
1157 	    find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
1158 		return 0;
1159 
1160 	for (pos = start; pos < start + size; pos += zinfo->zone_size) {
1161 		u64 reset_bytes;
1162 
1163 		if (!btrfs_dev_is_sequential(device, pos) ||
1164 		    btrfs_dev_is_empty_zone(device, pos))
1165 			continue;
1166 
1167 		/* Free regions should be empty */
1168 		btrfs_warn_in_rcu(
1169 			device->fs_info,
1170 		"zoned: resetting device %s (devid %llu) zone %llu for allocation",
1171 			rcu_str_deref(device->name), device->devid, pos >> shift);
1172 		WARN_ON_ONCE(1);
1173 
1174 		ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1175 					      &reset_bytes);
1176 		if (ret)
1177 			return ret;
1178 	}
1179 
1180 	return 0;
1181 }
1182 
1183 /*
1184  * Calculate an allocation pointer from the extent allocation information
1185  * for a block group consist of conventional zones. It is pointed to the
1186  * end of the highest addressed extent in the block group as an allocation
1187  * offset.
1188  */
1189 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1190 				   u64 *offset_ret, bool new)
1191 {
1192 	struct btrfs_fs_info *fs_info = cache->fs_info;
1193 	struct btrfs_root *root;
1194 	struct btrfs_path *path;
1195 	struct btrfs_key key;
1196 	struct btrfs_key found_key;
1197 	int ret;
1198 	u64 length;
1199 
1200 	/*
1201 	 * Avoid  tree lookups for a new block group, there's no use for it.
1202 	 * It must always be 0.
1203 	 *
1204 	 * Also, we have a lock chain of extent buffer lock -> chunk mutex.
1205 	 * For new a block group, this function is called from
1206 	 * btrfs_make_block_group() which is already taking the chunk mutex.
1207 	 * Thus, we cannot call calculate_alloc_pointer() which takes extent
1208 	 * buffer locks to avoid deadlock.
1209 	 */
1210 	if (new) {
1211 		*offset_ret = 0;
1212 		return 0;
1213 	}
1214 
1215 	path = btrfs_alloc_path();
1216 	if (!path)
1217 		return -ENOMEM;
1218 
1219 	key.objectid = cache->start + cache->length;
1220 	key.type = 0;
1221 	key.offset = 0;
1222 
1223 	root = btrfs_extent_root(fs_info, key.objectid);
1224 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1225 	/* We should not find the exact match */
1226 	if (!ret)
1227 		ret = -EUCLEAN;
1228 	if (ret < 0)
1229 		goto out;
1230 
1231 	ret = btrfs_previous_extent_item(root, path, cache->start);
1232 	if (ret) {
1233 		if (ret == 1) {
1234 			ret = 0;
1235 			*offset_ret = 0;
1236 		}
1237 		goto out;
1238 	}
1239 
1240 	btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1241 
1242 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1243 		length = found_key.offset;
1244 	else
1245 		length = fs_info->nodesize;
1246 
1247 	if (!(found_key.objectid >= cache->start &&
1248 	       found_key.objectid + length <= cache->start + cache->length)) {
1249 		ret = -EUCLEAN;
1250 		goto out;
1251 	}
1252 	*offset_ret = found_key.objectid + length - cache->start;
1253 	ret = 0;
1254 
1255 out:
1256 	btrfs_free_path(path);
1257 	return ret;
1258 }
1259 
1260 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1261 {
1262 	struct btrfs_fs_info *fs_info = cache->fs_info;
1263 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1264 	struct extent_map *em;
1265 	struct map_lookup *map;
1266 	struct btrfs_device *device;
1267 	u64 logical = cache->start;
1268 	u64 length = cache->length;
1269 	int ret;
1270 	int i;
1271 	unsigned int nofs_flag;
1272 	u64 *alloc_offsets = NULL;
1273 	u64 *caps = NULL;
1274 	u64 *physical = NULL;
1275 	unsigned long *active = NULL;
1276 	u64 last_alloc = 0;
1277 	u32 num_sequential = 0, num_conventional = 0;
1278 
1279 	if (!btrfs_is_zoned(fs_info))
1280 		return 0;
1281 
1282 	/* Sanity check */
1283 	if (!IS_ALIGNED(length, fs_info->zone_size)) {
1284 		btrfs_err(fs_info,
1285 		"zoned: block group %llu len %llu unaligned to zone size %llu",
1286 			  logical, length, fs_info->zone_size);
1287 		return -EIO;
1288 	}
1289 
1290 	/* Get the chunk mapping */
1291 	read_lock(&em_tree->lock);
1292 	em = lookup_extent_mapping(em_tree, logical, length);
1293 	read_unlock(&em_tree->lock);
1294 
1295 	if (!em)
1296 		return -EINVAL;
1297 
1298 	map = em->map_lookup;
1299 
1300 	cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS);
1301 	if (!cache->physical_map) {
1302 		ret = -ENOMEM;
1303 		goto out;
1304 	}
1305 
1306 	alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1307 	if (!alloc_offsets) {
1308 		ret = -ENOMEM;
1309 		goto out;
1310 	}
1311 
1312 	caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS);
1313 	if (!caps) {
1314 		ret = -ENOMEM;
1315 		goto out;
1316 	}
1317 
1318 	physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS);
1319 	if (!physical) {
1320 		ret = -ENOMEM;
1321 		goto out;
1322 	}
1323 
1324 	active = bitmap_zalloc(map->num_stripes, GFP_NOFS);
1325 	if (!active) {
1326 		ret = -ENOMEM;
1327 		goto out;
1328 	}
1329 
1330 	for (i = 0; i < map->num_stripes; i++) {
1331 		bool is_sequential;
1332 		struct blk_zone zone;
1333 		struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1334 		int dev_replace_is_ongoing = 0;
1335 
1336 		device = map->stripes[i].dev;
1337 		physical[i] = map->stripes[i].physical;
1338 
1339 		if (device->bdev == NULL) {
1340 			alloc_offsets[i] = WP_MISSING_DEV;
1341 			continue;
1342 		}
1343 
1344 		is_sequential = btrfs_dev_is_sequential(device, physical[i]);
1345 		if (is_sequential)
1346 			num_sequential++;
1347 		else
1348 			num_conventional++;
1349 
1350 		/*
1351 		 * Consider a zone as active if we can allow any number of
1352 		 * active zones.
1353 		 */
1354 		if (!device->zone_info->max_active_zones)
1355 			__set_bit(i, active);
1356 
1357 		if (!is_sequential) {
1358 			alloc_offsets[i] = WP_CONVENTIONAL;
1359 			continue;
1360 		}
1361 
1362 		/*
1363 		 * This zone will be used for allocation, so mark this zone
1364 		 * non-empty.
1365 		 */
1366 		btrfs_dev_clear_zone_empty(device, physical[i]);
1367 
1368 		down_read(&dev_replace->rwsem);
1369 		dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1370 		if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1371 			btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]);
1372 		up_read(&dev_replace->rwsem);
1373 
1374 		/*
1375 		 * The group is mapped to a sequential zone. Get the zone write
1376 		 * pointer to determine the allocation offset within the zone.
1377 		 */
1378 		WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size));
1379 		nofs_flag = memalloc_nofs_save();
1380 		ret = btrfs_get_dev_zone(device, physical[i], &zone);
1381 		memalloc_nofs_restore(nofs_flag);
1382 		if (ret == -EIO || ret == -EOPNOTSUPP) {
1383 			ret = 0;
1384 			alloc_offsets[i] = WP_MISSING_DEV;
1385 			continue;
1386 		} else if (ret) {
1387 			goto out;
1388 		}
1389 
1390 		if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1391 			btrfs_err_in_rcu(fs_info,
1392 	"zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1393 				zone.start << SECTOR_SHIFT,
1394 				rcu_str_deref(device->name), device->devid);
1395 			ret = -EIO;
1396 			goto out;
1397 		}
1398 
1399 		caps[i] = (zone.capacity << SECTOR_SHIFT);
1400 
1401 		switch (zone.cond) {
1402 		case BLK_ZONE_COND_OFFLINE:
1403 		case BLK_ZONE_COND_READONLY:
1404 			btrfs_err(fs_info,
1405 		"zoned: offline/readonly zone %llu on device %s (devid %llu)",
1406 				  physical[i] >> device->zone_info->zone_size_shift,
1407 				  rcu_str_deref(device->name), device->devid);
1408 			alloc_offsets[i] = WP_MISSING_DEV;
1409 			break;
1410 		case BLK_ZONE_COND_EMPTY:
1411 			alloc_offsets[i] = 0;
1412 			break;
1413 		case BLK_ZONE_COND_FULL:
1414 			alloc_offsets[i] = caps[i];
1415 			break;
1416 		default:
1417 			/* Partially used zone */
1418 			alloc_offsets[i] =
1419 					((zone.wp - zone.start) << SECTOR_SHIFT);
1420 			__set_bit(i, active);
1421 			break;
1422 		}
1423 	}
1424 
1425 	if (num_sequential > 0)
1426 		cache->seq_zone = true;
1427 
1428 	if (num_conventional > 0) {
1429 		/* Zone capacity is always zone size in emulation */
1430 		cache->zone_capacity = cache->length;
1431 		ret = calculate_alloc_pointer(cache, &last_alloc, new);
1432 		if (ret) {
1433 			btrfs_err(fs_info,
1434 			"zoned: failed to determine allocation offset of bg %llu",
1435 				  cache->start);
1436 			goto out;
1437 		} else if (map->num_stripes == num_conventional) {
1438 			cache->alloc_offset = last_alloc;
1439 			cache->zone_is_active = 1;
1440 			goto out;
1441 		}
1442 	}
1443 
1444 	switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1445 	case 0: /* single */
1446 		if (alloc_offsets[0] == WP_MISSING_DEV) {
1447 			btrfs_err(fs_info,
1448 			"zoned: cannot recover write pointer for zone %llu",
1449 				physical[0]);
1450 			ret = -EIO;
1451 			goto out;
1452 		}
1453 		cache->alloc_offset = alloc_offsets[0];
1454 		cache->zone_capacity = caps[0];
1455 		cache->zone_is_active = test_bit(0, active);
1456 		break;
1457 	case BTRFS_BLOCK_GROUP_DUP:
1458 		if (map->type & BTRFS_BLOCK_GROUP_DATA) {
1459 			btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg");
1460 			ret = -EINVAL;
1461 			goto out;
1462 		}
1463 		if (alloc_offsets[0] == WP_MISSING_DEV) {
1464 			btrfs_err(fs_info,
1465 			"zoned: cannot recover write pointer for zone %llu",
1466 				physical[0]);
1467 			ret = -EIO;
1468 			goto out;
1469 		}
1470 		if (alloc_offsets[1] == WP_MISSING_DEV) {
1471 			btrfs_err(fs_info,
1472 			"zoned: cannot recover write pointer for zone %llu",
1473 				physical[1]);
1474 			ret = -EIO;
1475 			goto out;
1476 		}
1477 		if (alloc_offsets[0] != alloc_offsets[1]) {
1478 			btrfs_err(fs_info,
1479 			"zoned: write pointer offset mismatch of zones in DUP profile");
1480 			ret = -EIO;
1481 			goto out;
1482 		}
1483 		if (test_bit(0, active) != test_bit(1, active)) {
1484 			if (!btrfs_zone_activate(cache)) {
1485 				ret = -EIO;
1486 				goto out;
1487 			}
1488 		} else {
1489 			cache->zone_is_active = test_bit(0, active);
1490 		}
1491 		cache->alloc_offset = alloc_offsets[0];
1492 		cache->zone_capacity = min(caps[0], caps[1]);
1493 		break;
1494 	case BTRFS_BLOCK_GROUP_RAID1:
1495 	case BTRFS_BLOCK_GROUP_RAID0:
1496 	case BTRFS_BLOCK_GROUP_RAID10:
1497 	case BTRFS_BLOCK_GROUP_RAID5:
1498 	case BTRFS_BLOCK_GROUP_RAID6:
1499 		/* non-single profiles are not supported yet */
1500 	default:
1501 		btrfs_err(fs_info, "zoned: profile %s not yet supported",
1502 			  btrfs_bg_type_to_raid_name(map->type));
1503 		ret = -EINVAL;
1504 		goto out;
1505 	}
1506 
1507 out:
1508 	if (cache->alloc_offset > fs_info->zone_size) {
1509 		btrfs_err(fs_info,
1510 			"zoned: invalid write pointer %llu in block group %llu",
1511 			cache->alloc_offset, cache->start);
1512 		ret = -EIO;
1513 	}
1514 
1515 	if (cache->alloc_offset > cache->zone_capacity) {
1516 		btrfs_err(fs_info,
1517 "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu",
1518 			  cache->alloc_offset, cache->zone_capacity,
1519 			  cache->start);
1520 		ret = -EIO;
1521 	}
1522 
1523 	/* An extent is allocated after the write pointer */
1524 	if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1525 		btrfs_err(fs_info,
1526 			  "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1527 			  logical, last_alloc, cache->alloc_offset);
1528 		ret = -EIO;
1529 	}
1530 
1531 	if (!ret) {
1532 		cache->meta_write_pointer = cache->alloc_offset + cache->start;
1533 		if (cache->zone_is_active) {
1534 			btrfs_get_block_group(cache);
1535 			spin_lock(&fs_info->zone_active_bgs_lock);
1536 			list_add_tail(&cache->active_bg_list,
1537 				      &fs_info->zone_active_bgs);
1538 			spin_unlock(&fs_info->zone_active_bgs_lock);
1539 		}
1540 	} else {
1541 		kfree(cache->physical_map);
1542 		cache->physical_map = NULL;
1543 	}
1544 	bitmap_free(active);
1545 	kfree(physical);
1546 	kfree(caps);
1547 	kfree(alloc_offsets);
1548 	free_extent_map(em);
1549 
1550 	return ret;
1551 }
1552 
1553 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1554 {
1555 	u64 unusable, free;
1556 
1557 	if (!btrfs_is_zoned(cache->fs_info))
1558 		return;
1559 
1560 	WARN_ON(cache->bytes_super != 0);
1561 	unusable = (cache->alloc_offset - cache->used) +
1562 		   (cache->length - cache->zone_capacity);
1563 	free = cache->zone_capacity - cache->alloc_offset;
1564 
1565 	/* We only need ->free_space in ALLOC_SEQ block groups */
1566 	cache->last_byte_to_unpin = (u64)-1;
1567 	cache->cached = BTRFS_CACHE_FINISHED;
1568 	cache->free_space_ctl->free_space = free;
1569 	cache->zone_unusable = unusable;
1570 }
1571 
1572 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1573 			    struct extent_buffer *eb)
1574 {
1575 	struct btrfs_fs_info *fs_info = eb->fs_info;
1576 
1577 	if (!btrfs_is_zoned(fs_info) ||
1578 	    btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1579 	    !list_empty(&eb->release_list))
1580 		return;
1581 
1582 	set_extent_buffer_dirty(eb);
1583 	set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1584 			       eb->start + eb->len - 1, EXTENT_DIRTY);
1585 	memzero_extent_buffer(eb, 0, eb->len);
1586 	set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1587 
1588 	spin_lock(&trans->releasing_ebs_lock);
1589 	list_add_tail(&eb->release_list, &trans->releasing_ebs);
1590 	spin_unlock(&trans->releasing_ebs_lock);
1591 	atomic_inc(&eb->refs);
1592 }
1593 
1594 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1595 {
1596 	spin_lock(&trans->releasing_ebs_lock);
1597 	while (!list_empty(&trans->releasing_ebs)) {
1598 		struct extent_buffer *eb;
1599 
1600 		eb = list_first_entry(&trans->releasing_ebs,
1601 				      struct extent_buffer, release_list);
1602 		list_del_init(&eb->release_list);
1603 		free_extent_buffer(eb);
1604 	}
1605 	spin_unlock(&trans->releasing_ebs_lock);
1606 }
1607 
1608 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1609 {
1610 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1611 	struct btrfs_block_group *cache;
1612 	bool ret = false;
1613 
1614 	if (!btrfs_is_zoned(fs_info))
1615 		return false;
1616 
1617 	if (!is_data_inode(&inode->vfs_inode))
1618 		return false;
1619 
1620 	/*
1621 	 * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1622 	 * extent layout the relocation code has.
1623 	 * Furthermore we have set aside own block-group from which only the
1624 	 * relocation "process" can allocate and make sure only one process at a
1625 	 * time can add pages to an extent that gets relocated, so it's safe to
1626 	 * use regular REQ_OP_WRITE for this special case.
1627 	 */
1628 	if (btrfs_is_data_reloc_root(inode->root))
1629 		return false;
1630 
1631 	cache = btrfs_lookup_block_group(fs_info, start);
1632 	ASSERT(cache);
1633 	if (!cache)
1634 		return false;
1635 
1636 	ret = cache->seq_zone;
1637 	btrfs_put_block_group(cache);
1638 
1639 	return ret;
1640 }
1641 
1642 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1643 				 struct bio *bio)
1644 {
1645 	struct btrfs_ordered_extent *ordered;
1646 	const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1647 
1648 	if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1649 		return;
1650 
1651 	ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1652 	if (WARN_ON(!ordered))
1653 		return;
1654 
1655 	ordered->physical = physical;
1656 	ordered->bdev = bio->bi_bdev;
1657 
1658 	btrfs_put_ordered_extent(ordered);
1659 }
1660 
1661 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1662 {
1663 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1664 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1665 	struct extent_map_tree *em_tree;
1666 	struct extent_map *em;
1667 	struct btrfs_ordered_sum *sum;
1668 	u64 orig_logical = ordered->disk_bytenr;
1669 	u64 *logical = NULL;
1670 	int nr, stripe_len;
1671 
1672 	/* Zoned devices should not have partitions. So, we can assume it is 0 */
1673 	ASSERT(!bdev_is_partition(ordered->bdev));
1674 	if (WARN_ON(!ordered->bdev))
1675 		return;
1676 
1677 	if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev,
1678 				     ordered->physical, &logical, &nr,
1679 				     &stripe_len)))
1680 		goto out;
1681 
1682 	WARN_ON(nr != 1);
1683 
1684 	if (orig_logical == *logical)
1685 		goto out;
1686 
1687 	ordered->disk_bytenr = *logical;
1688 
1689 	em_tree = &inode->extent_tree;
1690 	write_lock(&em_tree->lock);
1691 	em = search_extent_mapping(em_tree, ordered->file_offset,
1692 				   ordered->num_bytes);
1693 	em->block_start = *logical;
1694 	free_extent_map(em);
1695 	write_unlock(&em_tree->lock);
1696 
1697 	list_for_each_entry(sum, &ordered->list, list) {
1698 		if (*logical < orig_logical)
1699 			sum->bytenr -= orig_logical - *logical;
1700 		else
1701 			sum->bytenr += *logical - orig_logical;
1702 	}
1703 
1704 out:
1705 	kfree(logical);
1706 }
1707 
1708 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1709 				    struct extent_buffer *eb,
1710 				    struct btrfs_block_group **cache_ret)
1711 {
1712 	struct btrfs_block_group *cache;
1713 	bool ret = true;
1714 
1715 	if (!btrfs_is_zoned(fs_info))
1716 		return true;
1717 
1718 	cache = btrfs_lookup_block_group(fs_info, eb->start);
1719 	if (!cache)
1720 		return true;
1721 
1722 	if (cache->meta_write_pointer != eb->start) {
1723 		btrfs_put_block_group(cache);
1724 		cache = NULL;
1725 		ret = false;
1726 	} else {
1727 		cache->meta_write_pointer = eb->start + eb->len;
1728 	}
1729 
1730 	*cache_ret = cache;
1731 
1732 	return ret;
1733 }
1734 
1735 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1736 				     struct extent_buffer *eb)
1737 {
1738 	if (!btrfs_is_zoned(eb->fs_info) || !cache)
1739 		return;
1740 
1741 	ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1742 	cache->meta_write_pointer = eb->start;
1743 }
1744 
1745 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1746 {
1747 	if (!btrfs_dev_is_sequential(device, physical))
1748 		return -EOPNOTSUPP;
1749 
1750 	return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1751 				    length >> SECTOR_SHIFT, GFP_NOFS, 0);
1752 }
1753 
1754 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1755 			  struct blk_zone *zone)
1756 {
1757 	struct btrfs_io_context *bioc = NULL;
1758 	u64 mapped_length = PAGE_SIZE;
1759 	unsigned int nofs_flag;
1760 	int nmirrors;
1761 	int i, ret;
1762 
1763 	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1764 			       &mapped_length, &bioc);
1765 	if (ret || !bioc || mapped_length < PAGE_SIZE) {
1766 		ret = -EIO;
1767 		goto out_put_bioc;
1768 	}
1769 
1770 	if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1771 		ret = -EINVAL;
1772 		goto out_put_bioc;
1773 	}
1774 
1775 	nofs_flag = memalloc_nofs_save();
1776 	nmirrors = (int)bioc->num_stripes;
1777 	for (i = 0; i < nmirrors; i++) {
1778 		u64 physical = bioc->stripes[i].physical;
1779 		struct btrfs_device *dev = bioc->stripes[i].dev;
1780 
1781 		/* Missing device */
1782 		if (!dev->bdev)
1783 			continue;
1784 
1785 		ret = btrfs_get_dev_zone(dev, physical, zone);
1786 		/* Failing device */
1787 		if (ret == -EIO || ret == -EOPNOTSUPP)
1788 			continue;
1789 		break;
1790 	}
1791 	memalloc_nofs_restore(nofs_flag);
1792 out_put_bioc:
1793 	btrfs_put_bioc(bioc);
1794 	return ret;
1795 }
1796 
1797 /*
1798  * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1799  * filling zeros between @physical_pos to a write pointer of dev-replace
1800  * source device.
1801  */
1802 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1803 				    u64 physical_start, u64 physical_pos)
1804 {
1805 	struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1806 	struct blk_zone zone;
1807 	u64 length;
1808 	u64 wp;
1809 	int ret;
1810 
1811 	if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1812 		return 0;
1813 
1814 	ret = read_zone_info(fs_info, logical, &zone);
1815 	if (ret)
1816 		return ret;
1817 
1818 	wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1819 
1820 	if (physical_pos == wp)
1821 		return 0;
1822 
1823 	if (physical_pos > wp)
1824 		return -EUCLEAN;
1825 
1826 	length = wp - physical_pos;
1827 	return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1828 }
1829 
1830 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1831 					    u64 logical, u64 length)
1832 {
1833 	struct btrfs_device *device;
1834 	struct extent_map *em;
1835 	struct map_lookup *map;
1836 
1837 	em = btrfs_get_chunk_map(fs_info, logical, length);
1838 	if (IS_ERR(em))
1839 		return ERR_CAST(em);
1840 
1841 	map = em->map_lookup;
1842 	/* We only support single profile for now */
1843 	device = map->stripes[0].dev;
1844 
1845 	free_extent_map(em);
1846 
1847 	return device;
1848 }
1849 
1850 /**
1851  * Activate block group and underlying device zones
1852  *
1853  * @block_group: the block group to activate
1854  *
1855  * Return: true on success, false otherwise
1856  */
1857 bool btrfs_zone_activate(struct btrfs_block_group *block_group)
1858 {
1859 	struct btrfs_fs_info *fs_info = block_group->fs_info;
1860 	struct btrfs_space_info *space_info = block_group->space_info;
1861 	struct map_lookup *map;
1862 	struct btrfs_device *device;
1863 	u64 physical;
1864 	bool ret;
1865 	int i;
1866 
1867 	if (!btrfs_is_zoned(block_group->fs_info))
1868 		return true;
1869 
1870 	map = block_group->physical_map;
1871 
1872 	spin_lock(&space_info->lock);
1873 	spin_lock(&block_group->lock);
1874 	if (block_group->zone_is_active) {
1875 		ret = true;
1876 		goto out_unlock;
1877 	}
1878 
1879 	/* No space left */
1880 	if (btrfs_zoned_bg_is_full(block_group)) {
1881 		ret = false;
1882 		goto out_unlock;
1883 	}
1884 
1885 	for (i = 0; i < map->num_stripes; i++) {
1886 		device = map->stripes[i].dev;
1887 		physical = map->stripes[i].physical;
1888 
1889 		if (device->zone_info->max_active_zones == 0)
1890 			continue;
1891 
1892 		if (!btrfs_dev_set_active_zone(device, physical)) {
1893 			/* Cannot activate the zone */
1894 			ret = false;
1895 			goto out_unlock;
1896 		}
1897 	}
1898 
1899 	/* Successfully activated all the zones */
1900 	block_group->zone_is_active = 1;
1901 	space_info->active_total_bytes += block_group->length;
1902 	spin_unlock(&block_group->lock);
1903 	btrfs_try_granting_tickets(fs_info, space_info);
1904 	spin_unlock(&space_info->lock);
1905 
1906 	/* For the active block group list */
1907 	btrfs_get_block_group(block_group);
1908 
1909 	spin_lock(&fs_info->zone_active_bgs_lock);
1910 	list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs);
1911 	spin_unlock(&fs_info->zone_active_bgs_lock);
1912 
1913 	return true;
1914 
1915 out_unlock:
1916 	spin_unlock(&block_group->lock);
1917 	spin_unlock(&space_info->lock);
1918 	return ret;
1919 }
1920 
1921 static void wait_eb_writebacks(struct btrfs_block_group *block_group)
1922 {
1923 	struct btrfs_fs_info *fs_info = block_group->fs_info;
1924 	const u64 end = block_group->start + block_group->length;
1925 	struct radix_tree_iter iter;
1926 	struct extent_buffer *eb;
1927 	void __rcu **slot;
1928 
1929 	rcu_read_lock();
1930 	radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter,
1931 				 block_group->start >> fs_info->sectorsize_bits) {
1932 		eb = radix_tree_deref_slot(slot);
1933 		if (!eb)
1934 			continue;
1935 		if (radix_tree_deref_retry(eb)) {
1936 			slot = radix_tree_iter_retry(&iter);
1937 			continue;
1938 		}
1939 
1940 		if (eb->start < block_group->start)
1941 			continue;
1942 		if (eb->start >= end)
1943 			break;
1944 
1945 		slot = radix_tree_iter_resume(slot, &iter);
1946 		rcu_read_unlock();
1947 		wait_on_extent_buffer_writeback(eb);
1948 		rcu_read_lock();
1949 	}
1950 	rcu_read_unlock();
1951 }
1952 
1953 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
1954 {
1955 	struct btrfs_fs_info *fs_info = block_group->fs_info;
1956 	struct map_lookup *map;
1957 	const bool is_metadata = (block_group->flags &
1958 			(BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM));
1959 	int ret = 0;
1960 	int i;
1961 
1962 	spin_lock(&block_group->lock);
1963 	if (!block_group->zone_is_active) {
1964 		spin_unlock(&block_group->lock);
1965 		return 0;
1966 	}
1967 
1968 	/* Check if we have unwritten allocated space */
1969 	if (is_metadata &&
1970 	    block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
1971 		spin_unlock(&block_group->lock);
1972 		return -EAGAIN;
1973 	}
1974 
1975 	/*
1976 	 * If we are sure that the block group is full (= no more room left for
1977 	 * new allocation) and the IO for the last usable block is completed, we
1978 	 * don't need to wait for the other IOs. This holds because we ensure
1979 	 * the sequential IO submissions using the ZONE_APPEND command for data
1980 	 * and block_group->meta_write_pointer for metadata.
1981 	 */
1982 	if (!fully_written) {
1983 		spin_unlock(&block_group->lock);
1984 
1985 		ret = btrfs_inc_block_group_ro(block_group, false);
1986 		if (ret)
1987 			return ret;
1988 
1989 		/* Ensure all writes in this block group finish */
1990 		btrfs_wait_block_group_reservations(block_group);
1991 		/* No need to wait for NOCOW writers. Zoned mode does not allow that */
1992 		btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
1993 					 block_group->length);
1994 		/* Wait for extent buffers to be written. */
1995 		if (is_metadata)
1996 			wait_eb_writebacks(block_group);
1997 
1998 		spin_lock(&block_group->lock);
1999 
2000 		/*
2001 		 * Bail out if someone already deactivated the block group, or
2002 		 * allocated space is left in the block group.
2003 		 */
2004 		if (!block_group->zone_is_active) {
2005 			spin_unlock(&block_group->lock);
2006 			btrfs_dec_block_group_ro(block_group);
2007 			return 0;
2008 		}
2009 
2010 		if (block_group->reserved) {
2011 			spin_unlock(&block_group->lock);
2012 			btrfs_dec_block_group_ro(block_group);
2013 			return -EAGAIN;
2014 		}
2015 	}
2016 
2017 	block_group->zone_is_active = 0;
2018 	block_group->alloc_offset = block_group->zone_capacity;
2019 	block_group->free_space_ctl->free_space = 0;
2020 	btrfs_clear_treelog_bg(block_group);
2021 	btrfs_clear_data_reloc_bg(block_group);
2022 	spin_unlock(&block_group->lock);
2023 
2024 	map = block_group->physical_map;
2025 	for (i = 0; i < map->num_stripes; i++) {
2026 		struct btrfs_device *device = map->stripes[i].dev;
2027 		const u64 physical = map->stripes[i].physical;
2028 
2029 		if (device->zone_info->max_active_zones == 0)
2030 			continue;
2031 
2032 		ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH,
2033 				       physical >> SECTOR_SHIFT,
2034 				       device->zone_info->zone_size >> SECTOR_SHIFT,
2035 				       GFP_NOFS);
2036 
2037 		if (ret)
2038 			return ret;
2039 
2040 		btrfs_dev_clear_active_zone(device, physical);
2041 	}
2042 
2043 	if (!fully_written)
2044 		btrfs_dec_block_group_ro(block_group);
2045 
2046 	spin_lock(&fs_info->zone_active_bgs_lock);
2047 	ASSERT(!list_empty(&block_group->active_bg_list));
2048 	list_del_init(&block_group->active_bg_list);
2049 	spin_unlock(&fs_info->zone_active_bgs_lock);
2050 
2051 	/* For active_bg_list */
2052 	btrfs_put_block_group(block_group);
2053 
2054 	clear_and_wake_up_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2055 
2056 	return 0;
2057 }
2058 
2059 int btrfs_zone_finish(struct btrfs_block_group *block_group)
2060 {
2061 	if (!btrfs_is_zoned(block_group->fs_info))
2062 		return 0;
2063 
2064 	return do_zone_finish(block_group, false);
2065 }
2066 
2067 bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags)
2068 {
2069 	struct btrfs_fs_info *fs_info = fs_devices->fs_info;
2070 	struct btrfs_device *device;
2071 	bool ret = false;
2072 
2073 	if (!btrfs_is_zoned(fs_info))
2074 		return true;
2075 
2076 	/* Check if there is a device with active zones left */
2077 	mutex_lock(&fs_info->chunk_mutex);
2078 	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
2079 		struct btrfs_zoned_device_info *zinfo = device->zone_info;
2080 
2081 		if (!device->bdev)
2082 			continue;
2083 
2084 		if (!zinfo->max_active_zones ||
2085 		    atomic_read(&zinfo->active_zones_left)) {
2086 			ret = true;
2087 			break;
2088 		}
2089 	}
2090 	mutex_unlock(&fs_info->chunk_mutex);
2091 
2092 	if (!ret)
2093 		set_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2094 
2095 	return ret;
2096 }
2097 
2098 void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length)
2099 {
2100 	struct btrfs_block_group *block_group;
2101 	u64 min_alloc_bytes;
2102 
2103 	if (!btrfs_is_zoned(fs_info))
2104 		return;
2105 
2106 	block_group = btrfs_lookup_block_group(fs_info, logical);
2107 	ASSERT(block_group);
2108 
2109 	/* No MIXED_BG on zoned btrfs. */
2110 	if (block_group->flags & BTRFS_BLOCK_GROUP_DATA)
2111 		min_alloc_bytes = fs_info->sectorsize;
2112 	else
2113 		min_alloc_bytes = fs_info->nodesize;
2114 
2115 	/* Bail out if we can allocate more data from this block group. */
2116 	if (logical + length + min_alloc_bytes <=
2117 	    block_group->start + block_group->zone_capacity)
2118 		goto out;
2119 
2120 	do_zone_finish(block_group, true);
2121 
2122 out:
2123 	btrfs_put_block_group(block_group);
2124 }
2125 
2126 static void btrfs_zone_finish_endio_workfn(struct work_struct *work)
2127 {
2128 	struct btrfs_block_group *bg =
2129 		container_of(work, struct btrfs_block_group, zone_finish_work);
2130 
2131 	wait_on_extent_buffer_writeback(bg->last_eb);
2132 	free_extent_buffer(bg->last_eb);
2133 	btrfs_zone_finish_endio(bg->fs_info, bg->start, bg->length);
2134 	btrfs_put_block_group(bg);
2135 }
2136 
2137 void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg,
2138 				   struct extent_buffer *eb)
2139 {
2140 	if (!bg->seq_zone || eb->start + eb->len * 2 <= bg->start + bg->zone_capacity)
2141 		return;
2142 
2143 	if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) {
2144 		btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing",
2145 			  bg->start);
2146 		return;
2147 	}
2148 
2149 	/* For the work */
2150 	btrfs_get_block_group(bg);
2151 	atomic_inc(&eb->refs);
2152 	bg->last_eb = eb;
2153 	INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn);
2154 	queue_work(system_unbound_wq, &bg->zone_finish_work);
2155 }
2156 
2157 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
2158 {
2159 	struct btrfs_fs_info *fs_info = bg->fs_info;
2160 
2161 	spin_lock(&fs_info->relocation_bg_lock);
2162 	if (fs_info->data_reloc_bg == bg->start)
2163 		fs_info->data_reloc_bg = 0;
2164 	spin_unlock(&fs_info->relocation_bg_lock);
2165 }
2166 
2167 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
2168 {
2169 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2170 	struct btrfs_device *device;
2171 
2172 	if (!btrfs_is_zoned(fs_info))
2173 		return;
2174 
2175 	mutex_lock(&fs_devices->device_list_mutex);
2176 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2177 		if (device->zone_info) {
2178 			vfree(device->zone_info->zone_cache);
2179 			device->zone_info->zone_cache = NULL;
2180 		}
2181 	}
2182 	mutex_unlock(&fs_devices->device_list_mutex);
2183 }
2184 
2185 bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info)
2186 {
2187 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2188 	struct btrfs_device *device;
2189 	u64 used = 0;
2190 	u64 total = 0;
2191 	u64 factor;
2192 
2193 	ASSERT(btrfs_is_zoned(fs_info));
2194 
2195 	if (fs_info->bg_reclaim_threshold == 0)
2196 		return false;
2197 
2198 	mutex_lock(&fs_devices->device_list_mutex);
2199 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2200 		if (!device->bdev)
2201 			continue;
2202 
2203 		total += device->disk_total_bytes;
2204 		used += device->bytes_used;
2205 	}
2206 	mutex_unlock(&fs_devices->device_list_mutex);
2207 
2208 	factor = div64_u64(used * 100, total);
2209 	return factor >= fs_info->bg_reclaim_threshold;
2210 }
2211 
2212 void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical,
2213 				       u64 length)
2214 {
2215 	struct btrfs_block_group *block_group;
2216 
2217 	if (!btrfs_is_zoned(fs_info))
2218 		return;
2219 
2220 	block_group = btrfs_lookup_block_group(fs_info, logical);
2221 	/* It should be called on a previous data relocation block group. */
2222 	ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA));
2223 
2224 	spin_lock(&block_group->lock);
2225 	if (!block_group->zoned_data_reloc_ongoing)
2226 		goto out;
2227 
2228 	/* All relocation extents are written. */
2229 	if (block_group->start + block_group->alloc_offset == logical + length) {
2230 		/* Now, release this block group for further allocations. */
2231 		block_group->zoned_data_reloc_ongoing = 0;
2232 	}
2233 
2234 out:
2235 	spin_unlock(&block_group->lock);
2236 	btrfs_put_block_group(block_group);
2237 }
2238 
2239 int btrfs_zone_finish_one_bg(struct btrfs_fs_info *fs_info)
2240 {
2241 	struct btrfs_block_group *block_group;
2242 	struct btrfs_block_group *min_bg = NULL;
2243 	u64 min_avail = U64_MAX;
2244 	int ret;
2245 
2246 	spin_lock(&fs_info->zone_active_bgs_lock);
2247 	list_for_each_entry(block_group, &fs_info->zone_active_bgs,
2248 			    active_bg_list) {
2249 		u64 avail;
2250 
2251 		spin_lock(&block_group->lock);
2252 		if (block_group->reserved ||
2253 		    (block_group->flags & BTRFS_BLOCK_GROUP_SYSTEM)) {
2254 			spin_unlock(&block_group->lock);
2255 			continue;
2256 		}
2257 
2258 		avail = block_group->zone_capacity - block_group->alloc_offset;
2259 		if (min_avail > avail) {
2260 			if (min_bg)
2261 				btrfs_put_block_group(min_bg);
2262 			min_bg = block_group;
2263 			min_avail = avail;
2264 			btrfs_get_block_group(min_bg);
2265 		}
2266 		spin_unlock(&block_group->lock);
2267 	}
2268 	spin_unlock(&fs_info->zone_active_bgs_lock);
2269 
2270 	if (!min_bg)
2271 		return 0;
2272 
2273 	ret = btrfs_zone_finish(min_bg);
2274 	btrfs_put_block_group(min_bg);
2275 
2276 	return ret < 0 ? ret : 1;
2277 }
2278 
2279 int btrfs_zoned_activate_one_bg(struct btrfs_fs_info *fs_info,
2280 				struct btrfs_space_info *space_info,
2281 				bool do_finish)
2282 {
2283 	struct btrfs_block_group *bg;
2284 	int index;
2285 
2286 	if (!btrfs_is_zoned(fs_info) || (space_info->flags & BTRFS_BLOCK_GROUP_DATA))
2287 		return 0;
2288 
2289 	/* No more block groups to activate */
2290 	if (space_info->active_total_bytes == space_info->total_bytes)
2291 		return 0;
2292 
2293 	for (;;) {
2294 		int ret;
2295 		bool need_finish = false;
2296 
2297 		down_read(&space_info->groups_sem);
2298 		for (index = 0; index < BTRFS_NR_RAID_TYPES; index++) {
2299 			list_for_each_entry(bg, &space_info->block_groups[index],
2300 					    list) {
2301 				if (!spin_trylock(&bg->lock))
2302 					continue;
2303 				if (btrfs_zoned_bg_is_full(bg) || bg->zone_is_active) {
2304 					spin_unlock(&bg->lock);
2305 					continue;
2306 				}
2307 				spin_unlock(&bg->lock);
2308 
2309 				if (btrfs_zone_activate(bg)) {
2310 					up_read(&space_info->groups_sem);
2311 					return 1;
2312 				}
2313 
2314 				need_finish = true;
2315 			}
2316 		}
2317 		up_read(&space_info->groups_sem);
2318 
2319 		if (!do_finish || !need_finish)
2320 			break;
2321 
2322 		ret = btrfs_zone_finish_one_bg(fs_info);
2323 		if (ret == 0)
2324 			break;
2325 		if (ret < 0)
2326 			return ret;
2327 	}
2328 
2329 	return 0;
2330 }
2331