xref: /linux/fs/btrfs/zoned.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
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 "ctree.h"
8 #include "volumes.h"
9 #include "zoned.h"
10 #include "rcu-string.h"
11 #include "disk-io.h"
12 #include "block-group.h"
13 #include "transaction.h"
14 #include "dev-replace.h"
15 #include "space-info.h"
16 
17 /* Maximum number of zones to report per blkdev_report_zones() call */
18 #define BTRFS_REPORT_NR_ZONES   4096
19 /* Invalid allocation pointer value for missing devices */
20 #define WP_MISSING_DEV ((u64)-1)
21 /* Pseudo write pointer value for conventional zone */
22 #define WP_CONVENTIONAL ((u64)-2)
23 
24 /*
25  * Location of the first zone of superblock logging zone pairs.
26  *
27  * - primary superblock:    0B (zone 0)
28  * - first copy:          512G (zone starting at that offset)
29  * - second copy:           4T (zone starting at that offset)
30  */
31 #define BTRFS_SB_LOG_PRIMARY_OFFSET	(0ULL)
32 #define BTRFS_SB_LOG_FIRST_OFFSET	(512ULL * SZ_1G)
33 #define BTRFS_SB_LOG_SECOND_OFFSET	(4096ULL * SZ_1G)
34 
35 #define BTRFS_SB_LOG_FIRST_SHIFT	const_ilog2(BTRFS_SB_LOG_FIRST_OFFSET)
36 #define BTRFS_SB_LOG_SECOND_SHIFT	const_ilog2(BTRFS_SB_LOG_SECOND_OFFSET)
37 
38 /* Number of superblock log zones */
39 #define BTRFS_NR_SB_LOG_ZONES 2
40 
41 /*
42  * Maximum supported zone size. Currently, SMR disks have a zone size of
43  * 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range. We do not
44  * expect the zone size to become larger than 8GiB in the near future.
45  */
46 #define BTRFS_MAX_ZONE_SIZE		SZ_8G
47 
48 static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data)
49 {
50 	struct blk_zone *zones = data;
51 
52 	memcpy(&zones[idx], zone, sizeof(*zone));
53 
54 	return 0;
55 }
56 
57 static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones,
58 			    u64 *wp_ret)
59 {
60 	bool empty[BTRFS_NR_SB_LOG_ZONES];
61 	bool full[BTRFS_NR_SB_LOG_ZONES];
62 	sector_t sector;
63 
64 	ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL &&
65 	       zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL);
66 
67 	empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY);
68 	empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY);
69 	full[0] = (zones[0].cond == BLK_ZONE_COND_FULL);
70 	full[1] = (zones[1].cond == BLK_ZONE_COND_FULL);
71 
72 	/*
73 	 * Possible states of log buffer zones
74 	 *
75 	 *           Empty[0]  In use[0]  Full[0]
76 	 * Empty[1]         *          x        0
77 	 * In use[1]        0          x        0
78 	 * Full[1]          1          1        C
79 	 *
80 	 * Log position:
81 	 *   *: Special case, no superblock is written
82 	 *   0: Use write pointer of zones[0]
83 	 *   1: Use write pointer of zones[1]
84 	 *   C: Compare super blcoks from zones[0] and zones[1], use the latest
85 	 *      one determined by generation
86 	 *   x: Invalid state
87 	 */
88 
89 	if (empty[0] && empty[1]) {
90 		/* Special case to distinguish no superblock to read */
91 		*wp_ret = zones[0].start << SECTOR_SHIFT;
92 		return -ENOENT;
93 	} else if (full[0] && full[1]) {
94 		/* Compare two super blocks */
95 		struct address_space *mapping = bdev->bd_inode->i_mapping;
96 		struct page *page[BTRFS_NR_SB_LOG_ZONES];
97 		struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES];
98 		int i;
99 
100 		for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
101 			u64 bytenr;
102 
103 			bytenr = ((zones[i].start + zones[i].len)
104 				   << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE;
105 
106 			page[i] = read_cache_page_gfp(mapping,
107 					bytenr >> PAGE_SHIFT, GFP_NOFS);
108 			if (IS_ERR(page[i])) {
109 				if (i == 1)
110 					btrfs_release_disk_super(super[0]);
111 				return PTR_ERR(page[i]);
112 			}
113 			super[i] = page_address(page[i]);
114 		}
115 
116 		if (super[0]->generation > super[1]->generation)
117 			sector = zones[1].start;
118 		else
119 			sector = zones[0].start;
120 
121 		for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++)
122 			btrfs_release_disk_super(super[i]);
123 	} else if (!full[0] && (empty[1] || full[1])) {
124 		sector = zones[0].wp;
125 	} else if (full[0]) {
126 		sector = zones[1].wp;
127 	} else {
128 		return -EUCLEAN;
129 	}
130 	*wp_ret = sector << SECTOR_SHIFT;
131 	return 0;
132 }
133 
134 /*
135  * Get the first zone number of the superblock mirror
136  */
137 static inline u32 sb_zone_number(int shift, int mirror)
138 {
139 	u64 zone;
140 
141 	ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX);
142 	switch (mirror) {
143 	case 0: zone = 0; break;
144 	case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break;
145 	case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break;
146 	}
147 
148 	ASSERT(zone <= U32_MAX);
149 
150 	return (u32)zone;
151 }
152 
153 /*
154  * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block
155  * device into static sized chunks and fake a conventional zone on each of
156  * them.
157  */
158 static int emulate_report_zones(struct btrfs_device *device, u64 pos,
159 				struct blk_zone *zones, unsigned int nr_zones)
160 {
161 	const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT;
162 	sector_t bdev_size = bdev_nr_sectors(device->bdev);
163 	unsigned int i;
164 
165 	pos >>= SECTOR_SHIFT;
166 	for (i = 0; i < nr_zones; i++) {
167 		zones[i].start = i * zone_sectors + pos;
168 		zones[i].len = zone_sectors;
169 		zones[i].capacity = zone_sectors;
170 		zones[i].wp = zones[i].start + zone_sectors;
171 		zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL;
172 		zones[i].cond = BLK_ZONE_COND_NOT_WP;
173 
174 		if (zones[i].wp >= bdev_size) {
175 			i++;
176 			break;
177 		}
178 	}
179 
180 	return i;
181 }
182 
183 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
184 			       struct blk_zone *zones, unsigned int *nr_zones)
185 {
186 	int ret;
187 
188 	if (!*nr_zones)
189 		return 0;
190 
191 	if (!bdev_is_zoned(device->bdev)) {
192 		ret = emulate_report_zones(device, pos, zones, *nr_zones);
193 		*nr_zones = ret;
194 		return 0;
195 	}
196 
197 	ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
198 				  copy_zone_info_cb, zones);
199 	if (ret < 0) {
200 		btrfs_err_in_rcu(device->fs_info,
201 				 "zoned: failed to read zone %llu on %s (devid %llu)",
202 				 pos, rcu_str_deref(device->name),
203 				 device->devid);
204 		return ret;
205 	}
206 	*nr_zones = ret;
207 	if (!ret)
208 		return -EIO;
209 
210 	return 0;
211 }
212 
213 /* The emulated zone size is determined from the size of device extent */
214 static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info)
215 {
216 	struct btrfs_path *path;
217 	struct btrfs_root *root = fs_info->dev_root;
218 	struct btrfs_key key;
219 	struct extent_buffer *leaf;
220 	struct btrfs_dev_extent *dext;
221 	int ret = 0;
222 
223 	key.objectid = 1;
224 	key.type = BTRFS_DEV_EXTENT_KEY;
225 	key.offset = 0;
226 
227 	path = btrfs_alloc_path();
228 	if (!path)
229 		return -ENOMEM;
230 
231 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
232 	if (ret < 0)
233 		goto out;
234 
235 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
236 		ret = btrfs_next_item(root, path);
237 		if (ret < 0)
238 			goto out;
239 		/* No dev extents at all? Not good */
240 		if (ret > 0) {
241 			ret = -EUCLEAN;
242 			goto out;
243 		}
244 	}
245 
246 	leaf = path->nodes[0];
247 	dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
248 	fs_info->zone_size = btrfs_dev_extent_length(leaf, dext);
249 	ret = 0;
250 
251 out:
252 	btrfs_free_path(path);
253 
254 	return ret;
255 }
256 
257 int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
258 {
259 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
260 	struct btrfs_device *device;
261 	int ret = 0;
262 
263 	/* fs_info->zone_size might not set yet. Use the incomapt flag here. */
264 	if (!btrfs_fs_incompat(fs_info, ZONED))
265 		return 0;
266 
267 	mutex_lock(&fs_devices->device_list_mutex);
268 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
269 		/* We can skip reading of zone info for missing devices */
270 		if (!device->bdev)
271 			continue;
272 
273 		ret = btrfs_get_dev_zone_info(device);
274 		if (ret)
275 			break;
276 	}
277 	mutex_unlock(&fs_devices->device_list_mutex);
278 
279 	return ret;
280 }
281 
282 int btrfs_get_dev_zone_info(struct btrfs_device *device)
283 {
284 	struct btrfs_fs_info *fs_info = device->fs_info;
285 	struct btrfs_zoned_device_info *zone_info = NULL;
286 	struct block_device *bdev = device->bdev;
287 	struct request_queue *queue = bdev_get_queue(bdev);
288 	sector_t nr_sectors;
289 	sector_t sector = 0;
290 	struct blk_zone *zones = NULL;
291 	unsigned int i, nreported = 0, nr_zones;
292 	sector_t zone_sectors;
293 	char *model, *emulated;
294 	int ret;
295 
296 	/*
297 	 * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not
298 	 * yet be set.
299 	 */
300 	if (!btrfs_fs_incompat(fs_info, ZONED))
301 		return 0;
302 
303 	if (device->zone_info)
304 		return 0;
305 
306 	zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL);
307 	if (!zone_info)
308 		return -ENOMEM;
309 
310 	if (!bdev_is_zoned(bdev)) {
311 		if (!fs_info->zone_size) {
312 			ret = calculate_emulated_zone_size(fs_info);
313 			if (ret)
314 				goto out;
315 		}
316 
317 		ASSERT(fs_info->zone_size);
318 		zone_sectors = fs_info->zone_size >> SECTOR_SHIFT;
319 	} else {
320 		zone_sectors = bdev_zone_sectors(bdev);
321 	}
322 
323 	/* Check if it's power of 2 (see is_power_of_2) */
324 	ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0);
325 	zone_info->zone_size = zone_sectors << SECTOR_SHIFT;
326 
327 	/* We reject devices with a zone size larger than 8GB */
328 	if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) {
329 		btrfs_err_in_rcu(fs_info,
330 		"zoned: %s: zone size %llu larger than supported maximum %llu",
331 				 rcu_str_deref(device->name),
332 				 zone_info->zone_size, BTRFS_MAX_ZONE_SIZE);
333 		ret = -EINVAL;
334 		goto out;
335 	}
336 
337 	nr_sectors = bdev_nr_sectors(bdev);
338 	zone_info->zone_size_shift = ilog2(zone_info->zone_size);
339 	zone_info->max_zone_append_size =
340 		(u64)queue_max_zone_append_sectors(queue) << SECTOR_SHIFT;
341 	zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors);
342 	if (!IS_ALIGNED(nr_sectors, zone_sectors))
343 		zone_info->nr_zones++;
344 
345 	if (bdev_is_zoned(bdev) && zone_info->max_zone_append_size == 0) {
346 		btrfs_err(fs_info, "zoned: device %pg does not support zone append",
347 			  bdev);
348 		ret = -EINVAL;
349 		goto out;
350 	}
351 
352 	zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
353 	if (!zone_info->seq_zones) {
354 		ret = -ENOMEM;
355 		goto out;
356 	}
357 
358 	zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
359 	if (!zone_info->empty_zones) {
360 		ret = -ENOMEM;
361 		goto out;
362 	}
363 
364 	zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
365 	if (!zones) {
366 		ret = -ENOMEM;
367 		goto out;
368 	}
369 
370 	/* Get zones type */
371 	while (sector < nr_sectors) {
372 		nr_zones = BTRFS_REPORT_NR_ZONES;
373 		ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
374 					  &nr_zones);
375 		if (ret)
376 			goto out;
377 
378 		for (i = 0; i < nr_zones; i++) {
379 			if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
380 				__set_bit(nreported, zone_info->seq_zones);
381 			if (zones[i].cond == BLK_ZONE_COND_EMPTY)
382 				__set_bit(nreported, zone_info->empty_zones);
383 			nreported++;
384 		}
385 		sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
386 	}
387 
388 	if (nreported != zone_info->nr_zones) {
389 		btrfs_err_in_rcu(device->fs_info,
390 				 "inconsistent number of zones on %s (%u/%u)",
391 				 rcu_str_deref(device->name), nreported,
392 				 zone_info->nr_zones);
393 		ret = -EIO;
394 		goto out;
395 	}
396 
397 	/* Validate superblock log */
398 	nr_zones = BTRFS_NR_SB_LOG_ZONES;
399 	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
400 		u32 sb_zone;
401 		u64 sb_wp;
402 		int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
403 
404 		sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
405 		if (sb_zone + 1 >= zone_info->nr_zones)
406 			continue;
407 
408 		sector = sb_zone << (zone_info->zone_size_shift - SECTOR_SHIFT);
409 		ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT,
410 					  &zone_info->sb_zones[sb_pos],
411 					  &nr_zones);
412 		if (ret)
413 			goto out;
414 
415 		if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
416 			btrfs_err_in_rcu(device->fs_info,
417 	"zoned: failed to read super block log zone info at devid %llu zone %u",
418 					 device->devid, sb_zone);
419 			ret = -EUCLEAN;
420 			goto out;
421 		}
422 
423 		/*
424 		 * If zones[0] is conventional, always use the beggining of the
425 		 * zone to record superblock. No need to validate in that case.
426 		 */
427 		if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
428 		    BLK_ZONE_TYPE_CONVENTIONAL)
429 			continue;
430 
431 		ret = sb_write_pointer(device->bdev,
432 				       &zone_info->sb_zones[sb_pos], &sb_wp);
433 		if (ret != -ENOENT && ret) {
434 			btrfs_err_in_rcu(device->fs_info,
435 			"zoned: super block log zone corrupted devid %llu zone %u",
436 					 device->devid, sb_zone);
437 			ret = -EUCLEAN;
438 			goto out;
439 		}
440 	}
441 
442 
443 	kfree(zones);
444 
445 	device->zone_info = zone_info;
446 
447 	switch (bdev_zoned_model(bdev)) {
448 	case BLK_ZONED_HM:
449 		model = "host-managed zoned";
450 		emulated = "";
451 		break;
452 	case BLK_ZONED_HA:
453 		model = "host-aware zoned";
454 		emulated = "";
455 		break;
456 	case BLK_ZONED_NONE:
457 		model = "regular";
458 		emulated = "emulated ";
459 		break;
460 	default:
461 		/* Just in case */
462 		btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
463 				 bdev_zoned_model(bdev),
464 				 rcu_str_deref(device->name));
465 		ret = -EOPNOTSUPP;
466 		goto out_free_zone_info;
467 	}
468 
469 	btrfs_info_in_rcu(fs_info,
470 		"%s block device %s, %u %szones of %llu bytes",
471 		model, rcu_str_deref(device->name), zone_info->nr_zones,
472 		emulated, zone_info->zone_size);
473 
474 	return 0;
475 
476 out:
477 	kfree(zones);
478 out_free_zone_info:
479 	bitmap_free(zone_info->empty_zones);
480 	bitmap_free(zone_info->seq_zones);
481 	kfree(zone_info);
482 	device->zone_info = NULL;
483 
484 	return ret;
485 }
486 
487 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
488 {
489 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
490 
491 	if (!zone_info)
492 		return;
493 
494 	bitmap_free(zone_info->seq_zones);
495 	bitmap_free(zone_info->empty_zones);
496 	kfree(zone_info);
497 	device->zone_info = NULL;
498 }
499 
500 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
501 		       struct blk_zone *zone)
502 {
503 	unsigned int nr_zones = 1;
504 	int ret;
505 
506 	ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
507 	if (ret != 0 || !nr_zones)
508 		return ret ? ret : -EIO;
509 
510 	return 0;
511 }
512 
513 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
514 {
515 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
516 	struct btrfs_device *device;
517 	u64 zoned_devices = 0;
518 	u64 nr_devices = 0;
519 	u64 zone_size = 0;
520 	u64 max_zone_append_size = 0;
521 	const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED);
522 	int ret = 0;
523 
524 	/* Count zoned devices */
525 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
526 		enum blk_zoned_model model;
527 
528 		if (!device->bdev)
529 			continue;
530 
531 		model = bdev_zoned_model(device->bdev);
532 		/*
533 		 * A Host-Managed zoned device must be used as a zoned device.
534 		 * A Host-Aware zoned device and a non-zoned devices can be
535 		 * treated as a zoned device, if ZONED flag is enabled in the
536 		 * superblock.
537 		 */
538 		if (model == BLK_ZONED_HM ||
539 		    (model == BLK_ZONED_HA && incompat_zoned) ||
540 		    (model == BLK_ZONED_NONE && incompat_zoned)) {
541 			struct btrfs_zoned_device_info *zone_info =
542 				device->zone_info;
543 
544 			zone_info = device->zone_info;
545 			zoned_devices++;
546 			if (!zone_size) {
547 				zone_size = zone_info->zone_size;
548 			} else if (zone_info->zone_size != zone_size) {
549 				btrfs_err(fs_info,
550 		"zoned: unequal block device zone sizes: have %llu found %llu",
551 					  device->zone_info->zone_size,
552 					  zone_size);
553 				ret = -EINVAL;
554 				goto out;
555 			}
556 			if (!max_zone_append_size ||
557 			    (zone_info->max_zone_append_size &&
558 			     zone_info->max_zone_append_size < max_zone_append_size))
559 				max_zone_append_size =
560 					zone_info->max_zone_append_size;
561 		}
562 		nr_devices++;
563 	}
564 
565 	if (!zoned_devices && !incompat_zoned)
566 		goto out;
567 
568 	if (!zoned_devices && incompat_zoned) {
569 		/* No zoned block device found on ZONED filesystem */
570 		btrfs_err(fs_info,
571 			  "zoned: no zoned devices found on a zoned filesystem");
572 		ret = -EINVAL;
573 		goto out;
574 	}
575 
576 	if (zoned_devices && !incompat_zoned) {
577 		btrfs_err(fs_info,
578 			  "zoned: mode not enabled but zoned device found");
579 		ret = -EINVAL;
580 		goto out;
581 	}
582 
583 	if (zoned_devices != nr_devices) {
584 		btrfs_err(fs_info,
585 			  "zoned: cannot mix zoned and regular devices");
586 		ret = -EINVAL;
587 		goto out;
588 	}
589 
590 	/*
591 	 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
592 	 * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size,
593 	 * check the alignment here.
594 	 */
595 	if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
596 		btrfs_err(fs_info,
597 			  "zoned: zone size %llu not aligned to stripe %u",
598 			  zone_size, BTRFS_STRIPE_LEN);
599 		ret = -EINVAL;
600 		goto out;
601 	}
602 
603 	if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
604 		btrfs_err(fs_info, "zoned: mixed block groups not supported");
605 		ret = -EINVAL;
606 		goto out;
607 	}
608 
609 	fs_info->zone_size = zone_size;
610 	fs_info->max_zone_append_size = max_zone_append_size;
611 	fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
612 
613 	/*
614 	 * Check mount options here, because we might change fs_info->zoned
615 	 * from fs_info->zone_size.
616 	 */
617 	ret = btrfs_check_mountopts_zoned(fs_info);
618 	if (ret)
619 		goto out;
620 
621 	btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
622 out:
623 	return ret;
624 }
625 
626 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
627 {
628 	if (!btrfs_is_zoned(info))
629 		return 0;
630 
631 	/*
632 	 * Space cache writing is not COWed. Disable that to avoid write errors
633 	 * in sequential zones.
634 	 */
635 	if (btrfs_test_opt(info, SPACE_CACHE)) {
636 		btrfs_err(info, "zoned: space cache v1 is not supported");
637 		return -EINVAL;
638 	}
639 
640 	if (btrfs_test_opt(info, NODATACOW)) {
641 		btrfs_err(info, "zoned: NODATACOW not supported");
642 		return -EINVAL;
643 	}
644 
645 	return 0;
646 }
647 
648 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
649 			   int rw, u64 *bytenr_ret)
650 {
651 	u64 wp;
652 	int ret;
653 
654 	if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
655 		*bytenr_ret = zones[0].start << SECTOR_SHIFT;
656 		return 0;
657 	}
658 
659 	ret = sb_write_pointer(bdev, zones, &wp);
660 	if (ret != -ENOENT && ret < 0)
661 		return ret;
662 
663 	if (rw == WRITE) {
664 		struct blk_zone *reset = NULL;
665 
666 		if (wp == zones[0].start << SECTOR_SHIFT)
667 			reset = &zones[0];
668 		else if (wp == zones[1].start << SECTOR_SHIFT)
669 			reset = &zones[1];
670 
671 		if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
672 			ASSERT(reset->cond == BLK_ZONE_COND_FULL);
673 
674 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
675 					       reset->start, reset->len,
676 					       GFP_NOFS);
677 			if (ret)
678 				return ret;
679 
680 			reset->cond = BLK_ZONE_COND_EMPTY;
681 			reset->wp = reset->start;
682 		}
683 	} else if (ret != -ENOENT) {
684 		/* For READ, we want the precious one */
685 		if (wp == zones[0].start << SECTOR_SHIFT)
686 			wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT;
687 		wp -= BTRFS_SUPER_INFO_SIZE;
688 	}
689 
690 	*bytenr_ret = wp;
691 	return 0;
692 
693 }
694 
695 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
696 			       u64 *bytenr_ret)
697 {
698 	struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
699 	sector_t zone_sectors;
700 	u32 sb_zone;
701 	int ret;
702 	u8 zone_sectors_shift;
703 	sector_t nr_sectors;
704 	u32 nr_zones;
705 
706 	if (!bdev_is_zoned(bdev)) {
707 		*bytenr_ret = btrfs_sb_offset(mirror);
708 		return 0;
709 	}
710 
711 	ASSERT(rw == READ || rw == WRITE);
712 
713 	zone_sectors = bdev_zone_sectors(bdev);
714 	if (!is_power_of_2(zone_sectors))
715 		return -EINVAL;
716 	zone_sectors_shift = ilog2(zone_sectors);
717 	nr_sectors = bdev_nr_sectors(bdev);
718 	nr_zones = nr_sectors >> zone_sectors_shift;
719 
720 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
721 	if (sb_zone + 1 >= nr_zones)
722 		return -ENOENT;
723 
724 	ret = blkdev_report_zones(bdev, sb_zone << zone_sectors_shift,
725 				  BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
726 				  zones);
727 	if (ret < 0)
728 		return ret;
729 	if (ret != BTRFS_NR_SB_LOG_ZONES)
730 		return -EIO;
731 
732 	return sb_log_location(bdev, zones, rw, bytenr_ret);
733 }
734 
735 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
736 			  u64 *bytenr_ret)
737 {
738 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
739 	u32 zone_num;
740 
741 	/*
742 	 * For a zoned filesystem on a non-zoned block device, use the same
743 	 * super block locations as regular filesystem. Doing so, the super
744 	 * block can always be retrieved and the zoned flag of the volume
745 	 * detected from the super block information.
746 	 */
747 	if (!bdev_is_zoned(device->bdev)) {
748 		*bytenr_ret = btrfs_sb_offset(mirror);
749 		return 0;
750 	}
751 
752 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
753 	if (zone_num + 1 >= zinfo->nr_zones)
754 		return -ENOENT;
755 
756 	return sb_log_location(device->bdev,
757 			       &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
758 			       rw, bytenr_ret);
759 }
760 
761 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
762 				  int mirror)
763 {
764 	u32 zone_num;
765 
766 	if (!zinfo)
767 		return false;
768 
769 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
770 	if (zone_num + 1 >= zinfo->nr_zones)
771 		return false;
772 
773 	if (!test_bit(zone_num, zinfo->seq_zones))
774 		return false;
775 
776 	return true;
777 }
778 
779 void btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
780 {
781 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
782 	struct blk_zone *zone;
783 
784 	if (!is_sb_log_zone(zinfo, mirror))
785 		return;
786 
787 	zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
788 	if (zone->cond != BLK_ZONE_COND_FULL) {
789 		if (zone->cond == BLK_ZONE_COND_EMPTY)
790 			zone->cond = BLK_ZONE_COND_IMP_OPEN;
791 
792 		zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
793 
794 		if (zone->wp == zone->start + zone->len)
795 			zone->cond = BLK_ZONE_COND_FULL;
796 
797 		return;
798 	}
799 
800 	zone++;
801 	ASSERT(zone->cond != BLK_ZONE_COND_FULL);
802 	if (zone->cond == BLK_ZONE_COND_EMPTY)
803 		zone->cond = BLK_ZONE_COND_IMP_OPEN;
804 
805 	zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
806 
807 	if (zone->wp == zone->start + zone->len)
808 		zone->cond = BLK_ZONE_COND_FULL;
809 }
810 
811 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
812 {
813 	sector_t zone_sectors;
814 	sector_t nr_sectors;
815 	u8 zone_sectors_shift;
816 	u32 sb_zone;
817 	u32 nr_zones;
818 
819 	zone_sectors = bdev_zone_sectors(bdev);
820 	zone_sectors_shift = ilog2(zone_sectors);
821 	nr_sectors = bdev_nr_sectors(bdev);
822 	nr_zones = nr_sectors >> zone_sectors_shift;
823 
824 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
825 	if (sb_zone + 1 >= nr_zones)
826 		return -ENOENT;
827 
828 	return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
829 				sb_zone << zone_sectors_shift,
830 				zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
831 }
832 
833 /**
834  * btrfs_find_allocatable_zones - find allocatable zones within a given region
835  *
836  * @device:	the device to allocate a region on
837  * @hole_start: the position of the hole to allocate the region
838  * @num_bytes:	size of wanted region
839  * @hole_end:	the end of the hole
840  * @return:	position of allocatable zones
841  *
842  * Allocatable region should not contain any superblock locations.
843  */
844 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
845 				 u64 hole_end, u64 num_bytes)
846 {
847 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
848 	const u8 shift = zinfo->zone_size_shift;
849 	u64 nzones = num_bytes >> shift;
850 	u64 pos = hole_start;
851 	u64 begin, end;
852 	bool have_sb;
853 	int i;
854 
855 	ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
856 	ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
857 
858 	while (pos < hole_end) {
859 		begin = pos >> shift;
860 		end = begin + nzones;
861 
862 		if (end > zinfo->nr_zones)
863 			return hole_end;
864 
865 		/* Check if zones in the region are all empty */
866 		if (btrfs_dev_is_sequential(device, pos) &&
867 		    find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
868 			pos += zinfo->zone_size;
869 			continue;
870 		}
871 
872 		have_sb = false;
873 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
874 			u32 sb_zone;
875 			u64 sb_pos;
876 
877 			sb_zone = sb_zone_number(shift, i);
878 			if (!(end <= sb_zone ||
879 			      sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
880 				have_sb = true;
881 				pos = ((u64)sb_zone + BTRFS_NR_SB_LOG_ZONES) << shift;
882 				break;
883 			}
884 
885 			/* We also need to exclude regular superblock positions */
886 			sb_pos = btrfs_sb_offset(i);
887 			if (!(pos + num_bytes <= sb_pos ||
888 			      sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
889 				have_sb = true;
890 				pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
891 					    zinfo->zone_size);
892 				break;
893 			}
894 		}
895 		if (!have_sb)
896 			break;
897 	}
898 
899 	return pos;
900 }
901 
902 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
903 			    u64 length, u64 *bytes)
904 {
905 	int ret;
906 
907 	*bytes = 0;
908 	ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
909 			       physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
910 			       GFP_NOFS);
911 	if (ret)
912 		return ret;
913 
914 	*bytes = length;
915 	while (length) {
916 		btrfs_dev_set_zone_empty(device, physical);
917 		physical += device->zone_info->zone_size;
918 		length -= device->zone_info->zone_size;
919 	}
920 
921 	return 0;
922 }
923 
924 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
925 {
926 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
927 	const u8 shift = zinfo->zone_size_shift;
928 	unsigned long begin = start >> shift;
929 	unsigned long end = (start + size) >> shift;
930 	u64 pos;
931 	int ret;
932 
933 	ASSERT(IS_ALIGNED(start, zinfo->zone_size));
934 	ASSERT(IS_ALIGNED(size, zinfo->zone_size));
935 
936 	if (end > zinfo->nr_zones)
937 		return -ERANGE;
938 
939 	/* All the zones are conventional */
940 	if (find_next_bit(zinfo->seq_zones, begin, end) == end)
941 		return 0;
942 
943 	/* All the zones are sequential and empty */
944 	if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
945 	    find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
946 		return 0;
947 
948 	for (pos = start; pos < start + size; pos += zinfo->zone_size) {
949 		u64 reset_bytes;
950 
951 		if (!btrfs_dev_is_sequential(device, pos) ||
952 		    btrfs_dev_is_empty_zone(device, pos))
953 			continue;
954 
955 		/* Free regions should be empty */
956 		btrfs_warn_in_rcu(
957 			device->fs_info,
958 		"zoned: resetting device %s (devid %llu) zone %llu for allocation",
959 			rcu_str_deref(device->name), device->devid, pos >> shift);
960 		WARN_ON_ONCE(1);
961 
962 		ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
963 					      &reset_bytes);
964 		if (ret)
965 			return ret;
966 	}
967 
968 	return 0;
969 }
970 
971 /*
972  * Calculate an allocation pointer from the extent allocation information
973  * for a block group consist of conventional zones. It is pointed to the
974  * end of the highest addressed extent in the block group as an allocation
975  * offset.
976  */
977 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
978 				   u64 *offset_ret)
979 {
980 	struct btrfs_fs_info *fs_info = cache->fs_info;
981 	struct btrfs_root *root = fs_info->extent_root;
982 	struct btrfs_path *path;
983 	struct btrfs_key key;
984 	struct btrfs_key found_key;
985 	int ret;
986 	u64 length;
987 
988 	path = btrfs_alloc_path();
989 	if (!path)
990 		return -ENOMEM;
991 
992 	key.objectid = cache->start + cache->length;
993 	key.type = 0;
994 	key.offset = 0;
995 
996 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
997 	/* We should not find the exact match */
998 	if (!ret)
999 		ret = -EUCLEAN;
1000 	if (ret < 0)
1001 		goto out;
1002 
1003 	ret = btrfs_previous_extent_item(root, path, cache->start);
1004 	if (ret) {
1005 		if (ret == 1) {
1006 			ret = 0;
1007 			*offset_ret = 0;
1008 		}
1009 		goto out;
1010 	}
1011 
1012 	btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1013 
1014 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1015 		length = found_key.offset;
1016 	else
1017 		length = fs_info->nodesize;
1018 
1019 	if (!(found_key.objectid >= cache->start &&
1020 	       found_key.objectid + length <= cache->start + cache->length)) {
1021 		ret = -EUCLEAN;
1022 		goto out;
1023 	}
1024 	*offset_ret = found_key.objectid + length - cache->start;
1025 	ret = 0;
1026 
1027 out:
1028 	btrfs_free_path(path);
1029 	return ret;
1030 }
1031 
1032 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1033 {
1034 	struct btrfs_fs_info *fs_info = cache->fs_info;
1035 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1036 	struct extent_map *em;
1037 	struct map_lookup *map;
1038 	struct btrfs_device *device;
1039 	u64 logical = cache->start;
1040 	u64 length = cache->length;
1041 	u64 physical = 0;
1042 	int ret;
1043 	int i;
1044 	unsigned int nofs_flag;
1045 	u64 *alloc_offsets = NULL;
1046 	u64 last_alloc = 0;
1047 	u32 num_sequential = 0, num_conventional = 0;
1048 
1049 	if (!btrfs_is_zoned(fs_info))
1050 		return 0;
1051 
1052 	/* Sanity check */
1053 	if (!IS_ALIGNED(length, fs_info->zone_size)) {
1054 		btrfs_err(fs_info,
1055 		"zoned: block group %llu len %llu unaligned to zone size %llu",
1056 			  logical, length, fs_info->zone_size);
1057 		return -EIO;
1058 	}
1059 
1060 	/* Get the chunk mapping */
1061 	read_lock(&em_tree->lock);
1062 	em = lookup_extent_mapping(em_tree, logical, length);
1063 	read_unlock(&em_tree->lock);
1064 
1065 	if (!em)
1066 		return -EINVAL;
1067 
1068 	map = em->map_lookup;
1069 
1070 	alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1071 	if (!alloc_offsets) {
1072 		free_extent_map(em);
1073 		return -ENOMEM;
1074 	}
1075 
1076 	for (i = 0; i < map->num_stripes; i++) {
1077 		bool is_sequential;
1078 		struct blk_zone zone;
1079 		struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1080 		int dev_replace_is_ongoing = 0;
1081 
1082 		device = map->stripes[i].dev;
1083 		physical = map->stripes[i].physical;
1084 
1085 		if (device->bdev == NULL) {
1086 			alloc_offsets[i] = WP_MISSING_DEV;
1087 			continue;
1088 		}
1089 
1090 		is_sequential = btrfs_dev_is_sequential(device, physical);
1091 		if (is_sequential)
1092 			num_sequential++;
1093 		else
1094 			num_conventional++;
1095 
1096 		if (!is_sequential) {
1097 			alloc_offsets[i] = WP_CONVENTIONAL;
1098 			continue;
1099 		}
1100 
1101 		/*
1102 		 * This zone will be used for allocation, so mark this zone
1103 		 * non-empty.
1104 		 */
1105 		btrfs_dev_clear_zone_empty(device, physical);
1106 
1107 		down_read(&dev_replace->rwsem);
1108 		dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1109 		if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1110 			btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical);
1111 		up_read(&dev_replace->rwsem);
1112 
1113 		/*
1114 		 * The group is mapped to a sequential zone. Get the zone write
1115 		 * pointer to determine the allocation offset within the zone.
1116 		 */
1117 		WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size));
1118 		nofs_flag = memalloc_nofs_save();
1119 		ret = btrfs_get_dev_zone(device, physical, &zone);
1120 		memalloc_nofs_restore(nofs_flag);
1121 		if (ret == -EIO || ret == -EOPNOTSUPP) {
1122 			ret = 0;
1123 			alloc_offsets[i] = WP_MISSING_DEV;
1124 			continue;
1125 		} else if (ret) {
1126 			goto out;
1127 		}
1128 
1129 		if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1130 			ret = -EIO;
1131 			goto out;
1132 		}
1133 
1134 		switch (zone.cond) {
1135 		case BLK_ZONE_COND_OFFLINE:
1136 		case BLK_ZONE_COND_READONLY:
1137 			btrfs_err(fs_info,
1138 		"zoned: offline/readonly zone %llu on device %s (devid %llu)",
1139 				  physical >> device->zone_info->zone_size_shift,
1140 				  rcu_str_deref(device->name), device->devid);
1141 			alloc_offsets[i] = WP_MISSING_DEV;
1142 			break;
1143 		case BLK_ZONE_COND_EMPTY:
1144 			alloc_offsets[i] = 0;
1145 			break;
1146 		case BLK_ZONE_COND_FULL:
1147 			alloc_offsets[i] = fs_info->zone_size;
1148 			break;
1149 		default:
1150 			/* Partially used zone */
1151 			alloc_offsets[i] =
1152 					((zone.wp - zone.start) << SECTOR_SHIFT);
1153 			break;
1154 		}
1155 	}
1156 
1157 	if (num_sequential > 0)
1158 		cache->seq_zone = true;
1159 
1160 	if (num_conventional > 0) {
1161 		/*
1162 		 * Avoid calling calculate_alloc_pointer() for new BG. It
1163 		 * is no use for new BG. It must be always 0.
1164 		 *
1165 		 * Also, we have a lock chain of extent buffer lock ->
1166 		 * chunk mutex.  For new BG, this function is called from
1167 		 * btrfs_make_block_group() which is already taking the
1168 		 * chunk mutex. Thus, we cannot call
1169 		 * calculate_alloc_pointer() which takes extent buffer
1170 		 * locks to avoid deadlock.
1171 		 */
1172 		if (new) {
1173 			cache->alloc_offset = 0;
1174 			goto out;
1175 		}
1176 		ret = calculate_alloc_pointer(cache, &last_alloc);
1177 		if (ret || map->num_stripes == num_conventional) {
1178 			if (!ret)
1179 				cache->alloc_offset = last_alloc;
1180 			else
1181 				btrfs_err(fs_info,
1182 			"zoned: failed to determine allocation offset of bg %llu",
1183 					  cache->start);
1184 			goto out;
1185 		}
1186 	}
1187 
1188 	switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1189 	case 0: /* single */
1190 		cache->alloc_offset = alloc_offsets[0];
1191 		break;
1192 	case BTRFS_BLOCK_GROUP_DUP:
1193 	case BTRFS_BLOCK_GROUP_RAID1:
1194 	case BTRFS_BLOCK_GROUP_RAID0:
1195 	case BTRFS_BLOCK_GROUP_RAID10:
1196 	case BTRFS_BLOCK_GROUP_RAID5:
1197 	case BTRFS_BLOCK_GROUP_RAID6:
1198 		/* non-single profiles are not supported yet */
1199 	default:
1200 		btrfs_err(fs_info, "zoned: profile %s not yet supported",
1201 			  btrfs_bg_type_to_raid_name(map->type));
1202 		ret = -EINVAL;
1203 		goto out;
1204 	}
1205 
1206 out:
1207 	/* An extent is allocated after the write pointer */
1208 	if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1209 		btrfs_err(fs_info,
1210 			  "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1211 			  logical, last_alloc, cache->alloc_offset);
1212 		ret = -EIO;
1213 	}
1214 
1215 	if (!ret)
1216 		cache->meta_write_pointer = cache->alloc_offset + cache->start;
1217 
1218 	kfree(alloc_offsets);
1219 	free_extent_map(em);
1220 
1221 	return ret;
1222 }
1223 
1224 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1225 {
1226 	u64 unusable, free;
1227 
1228 	if (!btrfs_is_zoned(cache->fs_info))
1229 		return;
1230 
1231 	WARN_ON(cache->bytes_super != 0);
1232 	unusable = cache->alloc_offset - cache->used;
1233 	free = cache->length - cache->alloc_offset;
1234 
1235 	/* We only need ->free_space in ALLOC_SEQ block groups */
1236 	cache->last_byte_to_unpin = (u64)-1;
1237 	cache->cached = BTRFS_CACHE_FINISHED;
1238 	cache->free_space_ctl->free_space = free;
1239 	cache->zone_unusable = unusable;
1240 
1241 	/* Should not have any excluded extents. Just in case, though */
1242 	btrfs_free_excluded_extents(cache);
1243 }
1244 
1245 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1246 			    struct extent_buffer *eb)
1247 {
1248 	struct btrfs_fs_info *fs_info = eb->fs_info;
1249 
1250 	if (!btrfs_is_zoned(fs_info) ||
1251 	    btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1252 	    !list_empty(&eb->release_list))
1253 		return;
1254 
1255 	set_extent_buffer_dirty(eb);
1256 	set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1257 			       eb->start + eb->len - 1, EXTENT_DIRTY);
1258 	memzero_extent_buffer(eb, 0, eb->len);
1259 	set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1260 
1261 	spin_lock(&trans->releasing_ebs_lock);
1262 	list_add_tail(&eb->release_list, &trans->releasing_ebs);
1263 	spin_unlock(&trans->releasing_ebs_lock);
1264 	atomic_inc(&eb->refs);
1265 }
1266 
1267 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1268 {
1269 	spin_lock(&trans->releasing_ebs_lock);
1270 	while (!list_empty(&trans->releasing_ebs)) {
1271 		struct extent_buffer *eb;
1272 
1273 		eb = list_first_entry(&trans->releasing_ebs,
1274 				      struct extent_buffer, release_list);
1275 		list_del_init(&eb->release_list);
1276 		free_extent_buffer(eb);
1277 	}
1278 	spin_unlock(&trans->releasing_ebs_lock);
1279 }
1280 
1281 bool btrfs_use_zone_append(struct btrfs_inode *inode, struct extent_map *em)
1282 {
1283 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1284 	struct btrfs_block_group *cache;
1285 	bool ret = false;
1286 
1287 	if (!btrfs_is_zoned(fs_info))
1288 		return false;
1289 
1290 	if (!fs_info->max_zone_append_size)
1291 		return false;
1292 
1293 	if (!is_data_inode(&inode->vfs_inode))
1294 		return false;
1295 
1296 	cache = btrfs_lookup_block_group(fs_info, em->block_start);
1297 	ASSERT(cache);
1298 	if (!cache)
1299 		return false;
1300 
1301 	ret = cache->seq_zone;
1302 	btrfs_put_block_group(cache);
1303 
1304 	return ret;
1305 }
1306 
1307 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1308 				 struct bio *bio)
1309 {
1310 	struct btrfs_ordered_extent *ordered;
1311 	const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1312 
1313 	if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1314 		return;
1315 
1316 	ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1317 	if (WARN_ON(!ordered))
1318 		return;
1319 
1320 	ordered->physical = physical;
1321 	ordered->disk = bio->bi_bdev->bd_disk;
1322 	ordered->partno = bio->bi_bdev->bd_partno;
1323 
1324 	btrfs_put_ordered_extent(ordered);
1325 }
1326 
1327 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1328 {
1329 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1330 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1331 	struct extent_map_tree *em_tree;
1332 	struct extent_map *em;
1333 	struct btrfs_ordered_sum *sum;
1334 	struct block_device *bdev;
1335 	u64 orig_logical = ordered->disk_bytenr;
1336 	u64 *logical = NULL;
1337 	int nr, stripe_len;
1338 
1339 	/* Zoned devices should not have partitions. So, we can assume it is 0 */
1340 	ASSERT(ordered->partno == 0);
1341 	bdev = bdgrab(ordered->disk->part0);
1342 	if (WARN_ON(!bdev))
1343 		return;
1344 
1345 	if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, bdev,
1346 				     ordered->physical, &logical, &nr,
1347 				     &stripe_len)))
1348 		goto out;
1349 
1350 	WARN_ON(nr != 1);
1351 
1352 	if (orig_logical == *logical)
1353 		goto out;
1354 
1355 	ordered->disk_bytenr = *logical;
1356 
1357 	em_tree = &inode->extent_tree;
1358 	write_lock(&em_tree->lock);
1359 	em = search_extent_mapping(em_tree, ordered->file_offset,
1360 				   ordered->num_bytes);
1361 	em->block_start = *logical;
1362 	free_extent_map(em);
1363 	write_unlock(&em_tree->lock);
1364 
1365 	list_for_each_entry(sum, &ordered->list, list) {
1366 		if (*logical < orig_logical)
1367 			sum->bytenr -= orig_logical - *logical;
1368 		else
1369 			sum->bytenr += *logical - orig_logical;
1370 	}
1371 
1372 out:
1373 	kfree(logical);
1374 	bdput(bdev);
1375 }
1376 
1377 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1378 				    struct extent_buffer *eb,
1379 				    struct btrfs_block_group **cache_ret)
1380 {
1381 	struct btrfs_block_group *cache;
1382 	bool ret = true;
1383 
1384 	if (!btrfs_is_zoned(fs_info))
1385 		return true;
1386 
1387 	cache = *cache_ret;
1388 
1389 	if (cache && (eb->start < cache->start ||
1390 		      cache->start + cache->length <= eb->start)) {
1391 		btrfs_put_block_group(cache);
1392 		cache = NULL;
1393 		*cache_ret = NULL;
1394 	}
1395 
1396 	if (!cache)
1397 		cache = btrfs_lookup_block_group(fs_info, eb->start);
1398 
1399 	if (cache) {
1400 		if (cache->meta_write_pointer != eb->start) {
1401 			btrfs_put_block_group(cache);
1402 			cache = NULL;
1403 			ret = false;
1404 		} else {
1405 			cache->meta_write_pointer = eb->start + eb->len;
1406 		}
1407 
1408 		*cache_ret = cache;
1409 	}
1410 
1411 	return ret;
1412 }
1413 
1414 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1415 				     struct extent_buffer *eb)
1416 {
1417 	if (!btrfs_is_zoned(eb->fs_info) || !cache)
1418 		return;
1419 
1420 	ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1421 	cache->meta_write_pointer = eb->start;
1422 }
1423 
1424 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1425 {
1426 	if (!btrfs_dev_is_sequential(device, physical))
1427 		return -EOPNOTSUPP;
1428 
1429 	return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1430 				    length >> SECTOR_SHIFT, GFP_NOFS, 0);
1431 }
1432 
1433 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1434 			  struct blk_zone *zone)
1435 {
1436 	struct btrfs_bio *bbio = NULL;
1437 	u64 mapped_length = PAGE_SIZE;
1438 	unsigned int nofs_flag;
1439 	int nmirrors;
1440 	int i, ret;
1441 
1442 	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1443 			       &mapped_length, &bbio);
1444 	if (ret || !bbio || mapped_length < PAGE_SIZE) {
1445 		btrfs_put_bbio(bbio);
1446 		return -EIO;
1447 	}
1448 
1449 	if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1450 		return -EINVAL;
1451 
1452 	nofs_flag = memalloc_nofs_save();
1453 	nmirrors = (int)bbio->num_stripes;
1454 	for (i = 0; i < nmirrors; i++) {
1455 		u64 physical = bbio->stripes[i].physical;
1456 		struct btrfs_device *dev = bbio->stripes[i].dev;
1457 
1458 		/* Missing device */
1459 		if (!dev->bdev)
1460 			continue;
1461 
1462 		ret = btrfs_get_dev_zone(dev, physical, zone);
1463 		/* Failing device */
1464 		if (ret == -EIO || ret == -EOPNOTSUPP)
1465 			continue;
1466 		break;
1467 	}
1468 	memalloc_nofs_restore(nofs_flag);
1469 
1470 	return ret;
1471 }
1472 
1473 /*
1474  * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1475  * filling zeros between @physical_pos to a write pointer of dev-replace
1476  * source device.
1477  */
1478 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1479 				    u64 physical_start, u64 physical_pos)
1480 {
1481 	struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1482 	struct blk_zone zone;
1483 	u64 length;
1484 	u64 wp;
1485 	int ret;
1486 
1487 	if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1488 		return 0;
1489 
1490 	ret = read_zone_info(fs_info, logical, &zone);
1491 	if (ret)
1492 		return ret;
1493 
1494 	wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1495 
1496 	if (physical_pos == wp)
1497 		return 0;
1498 
1499 	if (physical_pos > wp)
1500 		return -EUCLEAN;
1501 
1502 	length = wp - physical_pos;
1503 	return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1504 }
1505