xref: /openbmc/linux/fs/btrfs/zoned.c (revision e481ff3f)
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 blocks 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 static inline sector_t zone_start_sector(u32 zone_number,
154 					 struct block_device *bdev)
155 {
156 	return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev));
157 }
158 
159 static inline u64 zone_start_physical(u32 zone_number,
160 				      struct btrfs_zoned_device_info *zone_info)
161 {
162 	return (u64)zone_number << zone_info->zone_size_shift;
163 }
164 
165 /*
166  * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block
167  * device into static sized chunks and fake a conventional zone on each of
168  * them.
169  */
170 static int emulate_report_zones(struct btrfs_device *device, u64 pos,
171 				struct blk_zone *zones, unsigned int nr_zones)
172 {
173 	const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT;
174 	sector_t bdev_size = bdev_nr_sectors(device->bdev);
175 	unsigned int i;
176 
177 	pos >>= SECTOR_SHIFT;
178 	for (i = 0; i < nr_zones; i++) {
179 		zones[i].start = i * zone_sectors + pos;
180 		zones[i].len = zone_sectors;
181 		zones[i].capacity = zone_sectors;
182 		zones[i].wp = zones[i].start + zone_sectors;
183 		zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL;
184 		zones[i].cond = BLK_ZONE_COND_NOT_WP;
185 
186 		if (zones[i].wp >= bdev_size) {
187 			i++;
188 			break;
189 		}
190 	}
191 
192 	return i;
193 }
194 
195 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
196 			       struct blk_zone *zones, unsigned int *nr_zones)
197 {
198 	int ret;
199 
200 	if (!*nr_zones)
201 		return 0;
202 
203 	if (!bdev_is_zoned(device->bdev)) {
204 		ret = emulate_report_zones(device, pos, zones, *nr_zones);
205 		*nr_zones = ret;
206 		return 0;
207 	}
208 
209 	ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
210 				  copy_zone_info_cb, zones);
211 	if (ret < 0) {
212 		btrfs_err_in_rcu(device->fs_info,
213 				 "zoned: failed to read zone %llu on %s (devid %llu)",
214 				 pos, rcu_str_deref(device->name),
215 				 device->devid);
216 		return ret;
217 	}
218 	*nr_zones = ret;
219 	if (!ret)
220 		return -EIO;
221 
222 	return 0;
223 }
224 
225 /* The emulated zone size is determined from the size of device extent */
226 static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info)
227 {
228 	struct btrfs_path *path;
229 	struct btrfs_root *root = fs_info->dev_root;
230 	struct btrfs_key key;
231 	struct extent_buffer *leaf;
232 	struct btrfs_dev_extent *dext;
233 	int ret = 0;
234 
235 	key.objectid = 1;
236 	key.type = BTRFS_DEV_EXTENT_KEY;
237 	key.offset = 0;
238 
239 	path = btrfs_alloc_path();
240 	if (!path)
241 		return -ENOMEM;
242 
243 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
244 	if (ret < 0)
245 		goto out;
246 
247 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
248 		ret = btrfs_next_item(root, path);
249 		if (ret < 0)
250 			goto out;
251 		/* No dev extents at all? Not good */
252 		if (ret > 0) {
253 			ret = -EUCLEAN;
254 			goto out;
255 		}
256 	}
257 
258 	leaf = path->nodes[0];
259 	dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
260 	fs_info->zone_size = btrfs_dev_extent_length(leaf, dext);
261 	ret = 0;
262 
263 out:
264 	btrfs_free_path(path);
265 
266 	return ret;
267 }
268 
269 int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
270 {
271 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
272 	struct btrfs_device *device;
273 	int ret = 0;
274 
275 	/* fs_info->zone_size might not set yet. Use the incomapt flag here. */
276 	if (!btrfs_fs_incompat(fs_info, ZONED))
277 		return 0;
278 
279 	mutex_lock(&fs_devices->device_list_mutex);
280 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
281 		/* We can skip reading of zone info for missing devices */
282 		if (!device->bdev)
283 			continue;
284 
285 		ret = btrfs_get_dev_zone_info(device);
286 		if (ret)
287 			break;
288 	}
289 	mutex_unlock(&fs_devices->device_list_mutex);
290 
291 	return ret;
292 }
293 
294 int btrfs_get_dev_zone_info(struct btrfs_device *device)
295 {
296 	struct btrfs_fs_info *fs_info = device->fs_info;
297 	struct btrfs_zoned_device_info *zone_info = NULL;
298 	struct block_device *bdev = device->bdev;
299 	struct request_queue *queue = bdev_get_queue(bdev);
300 	sector_t nr_sectors;
301 	sector_t sector = 0;
302 	struct blk_zone *zones = NULL;
303 	unsigned int i, nreported = 0, nr_zones;
304 	sector_t zone_sectors;
305 	char *model, *emulated;
306 	int ret;
307 
308 	/*
309 	 * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not
310 	 * yet be set.
311 	 */
312 	if (!btrfs_fs_incompat(fs_info, ZONED))
313 		return 0;
314 
315 	if (device->zone_info)
316 		return 0;
317 
318 	zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL);
319 	if (!zone_info)
320 		return -ENOMEM;
321 
322 	if (!bdev_is_zoned(bdev)) {
323 		if (!fs_info->zone_size) {
324 			ret = calculate_emulated_zone_size(fs_info);
325 			if (ret)
326 				goto out;
327 		}
328 
329 		ASSERT(fs_info->zone_size);
330 		zone_sectors = fs_info->zone_size >> SECTOR_SHIFT;
331 	} else {
332 		zone_sectors = bdev_zone_sectors(bdev);
333 	}
334 
335 	/* Check if it's power of 2 (see is_power_of_2) */
336 	ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0);
337 	zone_info->zone_size = zone_sectors << SECTOR_SHIFT;
338 
339 	/* We reject devices with a zone size larger than 8GB */
340 	if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) {
341 		btrfs_err_in_rcu(fs_info,
342 		"zoned: %s: zone size %llu larger than supported maximum %llu",
343 				 rcu_str_deref(device->name),
344 				 zone_info->zone_size, BTRFS_MAX_ZONE_SIZE);
345 		ret = -EINVAL;
346 		goto out;
347 	}
348 
349 	nr_sectors = bdev_nr_sectors(bdev);
350 	zone_info->zone_size_shift = ilog2(zone_info->zone_size);
351 	zone_info->max_zone_append_size =
352 		(u64)queue_max_zone_append_sectors(queue) << SECTOR_SHIFT;
353 	zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors);
354 	if (!IS_ALIGNED(nr_sectors, zone_sectors))
355 		zone_info->nr_zones++;
356 
357 	if (bdev_is_zoned(bdev) && zone_info->max_zone_append_size == 0) {
358 		btrfs_err(fs_info, "zoned: device %pg does not support zone append",
359 			  bdev);
360 		ret = -EINVAL;
361 		goto out;
362 	}
363 
364 	zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
365 	if (!zone_info->seq_zones) {
366 		ret = -ENOMEM;
367 		goto out;
368 	}
369 
370 	zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
371 	if (!zone_info->empty_zones) {
372 		ret = -ENOMEM;
373 		goto out;
374 	}
375 
376 	zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
377 	if (!zones) {
378 		ret = -ENOMEM;
379 		goto out;
380 	}
381 
382 	/* Get zones type */
383 	while (sector < nr_sectors) {
384 		nr_zones = BTRFS_REPORT_NR_ZONES;
385 		ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
386 					  &nr_zones);
387 		if (ret)
388 			goto out;
389 
390 		for (i = 0; i < nr_zones; i++) {
391 			if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
392 				__set_bit(nreported, zone_info->seq_zones);
393 			if (zones[i].cond == BLK_ZONE_COND_EMPTY)
394 				__set_bit(nreported, zone_info->empty_zones);
395 			nreported++;
396 		}
397 		sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
398 	}
399 
400 	if (nreported != zone_info->nr_zones) {
401 		btrfs_err_in_rcu(device->fs_info,
402 				 "inconsistent number of zones on %s (%u/%u)",
403 				 rcu_str_deref(device->name), nreported,
404 				 zone_info->nr_zones);
405 		ret = -EIO;
406 		goto out;
407 	}
408 
409 	/* Validate superblock log */
410 	nr_zones = BTRFS_NR_SB_LOG_ZONES;
411 	for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
412 		u32 sb_zone;
413 		u64 sb_wp;
414 		int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
415 
416 		sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
417 		if (sb_zone + 1 >= zone_info->nr_zones)
418 			continue;
419 
420 		ret = btrfs_get_dev_zones(device,
421 					  zone_start_physical(sb_zone, zone_info),
422 					  &zone_info->sb_zones[sb_pos],
423 					  &nr_zones);
424 		if (ret)
425 			goto out;
426 
427 		if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
428 			btrfs_err_in_rcu(device->fs_info,
429 	"zoned: failed to read super block log zone info at devid %llu zone %u",
430 					 device->devid, sb_zone);
431 			ret = -EUCLEAN;
432 			goto out;
433 		}
434 
435 		/*
436 		 * If zones[0] is conventional, always use the beginning of the
437 		 * zone to record superblock. No need to validate in that case.
438 		 */
439 		if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
440 		    BLK_ZONE_TYPE_CONVENTIONAL)
441 			continue;
442 
443 		ret = sb_write_pointer(device->bdev,
444 				       &zone_info->sb_zones[sb_pos], &sb_wp);
445 		if (ret != -ENOENT && ret) {
446 			btrfs_err_in_rcu(device->fs_info,
447 			"zoned: super block log zone corrupted devid %llu zone %u",
448 					 device->devid, sb_zone);
449 			ret = -EUCLEAN;
450 			goto out;
451 		}
452 	}
453 
454 
455 	kfree(zones);
456 
457 	device->zone_info = zone_info;
458 
459 	switch (bdev_zoned_model(bdev)) {
460 	case BLK_ZONED_HM:
461 		model = "host-managed zoned";
462 		emulated = "";
463 		break;
464 	case BLK_ZONED_HA:
465 		model = "host-aware zoned";
466 		emulated = "";
467 		break;
468 	case BLK_ZONED_NONE:
469 		model = "regular";
470 		emulated = "emulated ";
471 		break;
472 	default:
473 		/* Just in case */
474 		btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
475 				 bdev_zoned_model(bdev),
476 				 rcu_str_deref(device->name));
477 		ret = -EOPNOTSUPP;
478 		goto out_free_zone_info;
479 	}
480 
481 	btrfs_info_in_rcu(fs_info,
482 		"%s block device %s, %u %szones of %llu bytes",
483 		model, rcu_str_deref(device->name), zone_info->nr_zones,
484 		emulated, zone_info->zone_size);
485 
486 	return 0;
487 
488 out:
489 	kfree(zones);
490 out_free_zone_info:
491 	bitmap_free(zone_info->empty_zones);
492 	bitmap_free(zone_info->seq_zones);
493 	kfree(zone_info);
494 	device->zone_info = NULL;
495 
496 	return ret;
497 }
498 
499 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
500 {
501 	struct btrfs_zoned_device_info *zone_info = device->zone_info;
502 
503 	if (!zone_info)
504 		return;
505 
506 	bitmap_free(zone_info->seq_zones);
507 	bitmap_free(zone_info->empty_zones);
508 	kfree(zone_info);
509 	device->zone_info = NULL;
510 }
511 
512 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
513 		       struct blk_zone *zone)
514 {
515 	unsigned int nr_zones = 1;
516 	int ret;
517 
518 	ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
519 	if (ret != 0 || !nr_zones)
520 		return ret ? ret : -EIO;
521 
522 	return 0;
523 }
524 
525 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
526 {
527 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
528 	struct btrfs_device *device;
529 	u64 zoned_devices = 0;
530 	u64 nr_devices = 0;
531 	u64 zone_size = 0;
532 	u64 max_zone_append_size = 0;
533 	const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED);
534 	int ret = 0;
535 
536 	/* Count zoned devices */
537 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
538 		enum blk_zoned_model model;
539 
540 		if (!device->bdev)
541 			continue;
542 
543 		model = bdev_zoned_model(device->bdev);
544 		/*
545 		 * A Host-Managed zoned device must be used as a zoned device.
546 		 * A Host-Aware zoned device and a non-zoned devices can be
547 		 * treated as a zoned device, if ZONED flag is enabled in the
548 		 * superblock.
549 		 */
550 		if (model == BLK_ZONED_HM ||
551 		    (model == BLK_ZONED_HA && incompat_zoned) ||
552 		    (model == BLK_ZONED_NONE && incompat_zoned)) {
553 			struct btrfs_zoned_device_info *zone_info =
554 				device->zone_info;
555 
556 			zone_info = device->zone_info;
557 			zoned_devices++;
558 			if (!zone_size) {
559 				zone_size = zone_info->zone_size;
560 			} else if (zone_info->zone_size != zone_size) {
561 				btrfs_err(fs_info,
562 		"zoned: unequal block device zone sizes: have %llu found %llu",
563 					  device->zone_info->zone_size,
564 					  zone_size);
565 				ret = -EINVAL;
566 				goto out;
567 			}
568 			if (!max_zone_append_size ||
569 			    (zone_info->max_zone_append_size &&
570 			     zone_info->max_zone_append_size < max_zone_append_size))
571 				max_zone_append_size =
572 					zone_info->max_zone_append_size;
573 		}
574 		nr_devices++;
575 	}
576 
577 	if (!zoned_devices && !incompat_zoned)
578 		goto out;
579 
580 	if (!zoned_devices && incompat_zoned) {
581 		/* No zoned block device found on ZONED filesystem */
582 		btrfs_err(fs_info,
583 			  "zoned: no zoned devices found on a zoned filesystem");
584 		ret = -EINVAL;
585 		goto out;
586 	}
587 
588 	if (zoned_devices && !incompat_zoned) {
589 		btrfs_err(fs_info,
590 			  "zoned: mode not enabled but zoned device found");
591 		ret = -EINVAL;
592 		goto out;
593 	}
594 
595 	if (zoned_devices != nr_devices) {
596 		btrfs_err(fs_info,
597 			  "zoned: cannot mix zoned and regular devices");
598 		ret = -EINVAL;
599 		goto out;
600 	}
601 
602 	/*
603 	 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
604 	 * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size,
605 	 * check the alignment here.
606 	 */
607 	if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
608 		btrfs_err(fs_info,
609 			  "zoned: zone size %llu not aligned to stripe %u",
610 			  zone_size, BTRFS_STRIPE_LEN);
611 		ret = -EINVAL;
612 		goto out;
613 	}
614 
615 	if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
616 		btrfs_err(fs_info, "zoned: mixed block groups not supported");
617 		ret = -EINVAL;
618 		goto out;
619 	}
620 
621 	fs_info->zone_size = zone_size;
622 	fs_info->max_zone_append_size = max_zone_append_size;
623 	fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
624 
625 	/*
626 	 * Check mount options here, because we might change fs_info->zoned
627 	 * from fs_info->zone_size.
628 	 */
629 	ret = btrfs_check_mountopts_zoned(fs_info);
630 	if (ret)
631 		goto out;
632 
633 	btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
634 out:
635 	return ret;
636 }
637 
638 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
639 {
640 	if (!btrfs_is_zoned(info))
641 		return 0;
642 
643 	/*
644 	 * Space cache writing is not COWed. Disable that to avoid write errors
645 	 * in sequential zones.
646 	 */
647 	if (btrfs_test_opt(info, SPACE_CACHE)) {
648 		btrfs_err(info, "zoned: space cache v1 is not supported");
649 		return -EINVAL;
650 	}
651 
652 	if (btrfs_test_opt(info, NODATACOW)) {
653 		btrfs_err(info, "zoned: NODATACOW not supported");
654 		return -EINVAL;
655 	}
656 
657 	return 0;
658 }
659 
660 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
661 			   int rw, u64 *bytenr_ret)
662 {
663 	u64 wp;
664 	int ret;
665 
666 	if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
667 		*bytenr_ret = zones[0].start << SECTOR_SHIFT;
668 		return 0;
669 	}
670 
671 	ret = sb_write_pointer(bdev, zones, &wp);
672 	if (ret != -ENOENT && ret < 0)
673 		return ret;
674 
675 	if (rw == WRITE) {
676 		struct blk_zone *reset = NULL;
677 
678 		if (wp == zones[0].start << SECTOR_SHIFT)
679 			reset = &zones[0];
680 		else if (wp == zones[1].start << SECTOR_SHIFT)
681 			reset = &zones[1];
682 
683 		if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
684 			ASSERT(reset->cond == BLK_ZONE_COND_FULL);
685 
686 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
687 					       reset->start, reset->len,
688 					       GFP_NOFS);
689 			if (ret)
690 				return ret;
691 
692 			reset->cond = BLK_ZONE_COND_EMPTY;
693 			reset->wp = reset->start;
694 		}
695 	} else if (ret != -ENOENT) {
696 		/* For READ, we want the precious one */
697 		if (wp == zones[0].start << SECTOR_SHIFT)
698 			wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT;
699 		wp -= BTRFS_SUPER_INFO_SIZE;
700 	}
701 
702 	*bytenr_ret = wp;
703 	return 0;
704 
705 }
706 
707 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
708 			       u64 *bytenr_ret)
709 {
710 	struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
711 	sector_t zone_sectors;
712 	u32 sb_zone;
713 	int ret;
714 	u8 zone_sectors_shift;
715 	sector_t nr_sectors;
716 	u32 nr_zones;
717 
718 	if (!bdev_is_zoned(bdev)) {
719 		*bytenr_ret = btrfs_sb_offset(mirror);
720 		return 0;
721 	}
722 
723 	ASSERT(rw == READ || rw == WRITE);
724 
725 	zone_sectors = bdev_zone_sectors(bdev);
726 	if (!is_power_of_2(zone_sectors))
727 		return -EINVAL;
728 	zone_sectors_shift = ilog2(zone_sectors);
729 	nr_sectors = bdev_nr_sectors(bdev);
730 	nr_zones = nr_sectors >> zone_sectors_shift;
731 
732 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
733 	if (sb_zone + 1 >= nr_zones)
734 		return -ENOENT;
735 
736 	ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
737 				  BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
738 				  zones);
739 	if (ret < 0)
740 		return ret;
741 	if (ret != BTRFS_NR_SB_LOG_ZONES)
742 		return -EIO;
743 
744 	return sb_log_location(bdev, zones, rw, bytenr_ret);
745 }
746 
747 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
748 			  u64 *bytenr_ret)
749 {
750 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
751 	u32 zone_num;
752 
753 	/*
754 	 * For a zoned filesystem on a non-zoned block device, use the same
755 	 * super block locations as regular filesystem. Doing so, the super
756 	 * block can always be retrieved and the zoned flag of the volume
757 	 * detected from the super block information.
758 	 */
759 	if (!bdev_is_zoned(device->bdev)) {
760 		*bytenr_ret = btrfs_sb_offset(mirror);
761 		return 0;
762 	}
763 
764 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
765 	if (zone_num + 1 >= zinfo->nr_zones)
766 		return -ENOENT;
767 
768 	return sb_log_location(device->bdev,
769 			       &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
770 			       rw, bytenr_ret);
771 }
772 
773 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
774 				  int mirror)
775 {
776 	u32 zone_num;
777 
778 	if (!zinfo)
779 		return false;
780 
781 	zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
782 	if (zone_num + 1 >= zinfo->nr_zones)
783 		return false;
784 
785 	if (!test_bit(zone_num, zinfo->seq_zones))
786 		return false;
787 
788 	return true;
789 }
790 
791 void btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
792 {
793 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
794 	struct blk_zone *zone;
795 
796 	if (!is_sb_log_zone(zinfo, mirror))
797 		return;
798 
799 	zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
800 	if (zone->cond != BLK_ZONE_COND_FULL) {
801 		if (zone->cond == BLK_ZONE_COND_EMPTY)
802 			zone->cond = BLK_ZONE_COND_IMP_OPEN;
803 
804 		zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
805 
806 		if (zone->wp == zone->start + zone->len)
807 			zone->cond = BLK_ZONE_COND_FULL;
808 
809 		return;
810 	}
811 
812 	zone++;
813 	ASSERT(zone->cond != BLK_ZONE_COND_FULL);
814 	if (zone->cond == BLK_ZONE_COND_EMPTY)
815 		zone->cond = BLK_ZONE_COND_IMP_OPEN;
816 
817 	zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
818 
819 	if (zone->wp == zone->start + zone->len)
820 		zone->cond = BLK_ZONE_COND_FULL;
821 }
822 
823 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
824 {
825 	sector_t zone_sectors;
826 	sector_t nr_sectors;
827 	u8 zone_sectors_shift;
828 	u32 sb_zone;
829 	u32 nr_zones;
830 
831 	zone_sectors = bdev_zone_sectors(bdev);
832 	zone_sectors_shift = ilog2(zone_sectors);
833 	nr_sectors = bdev_nr_sectors(bdev);
834 	nr_zones = nr_sectors >> zone_sectors_shift;
835 
836 	sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
837 	if (sb_zone + 1 >= nr_zones)
838 		return -ENOENT;
839 
840 	return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
841 				zone_start_sector(sb_zone, bdev),
842 				zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
843 }
844 
845 /**
846  * btrfs_find_allocatable_zones - find allocatable zones within a given region
847  *
848  * @device:	the device to allocate a region on
849  * @hole_start: the position of the hole to allocate the region
850  * @num_bytes:	size of wanted region
851  * @hole_end:	the end of the hole
852  * @return:	position of allocatable zones
853  *
854  * Allocatable region should not contain any superblock locations.
855  */
856 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
857 				 u64 hole_end, u64 num_bytes)
858 {
859 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
860 	const u8 shift = zinfo->zone_size_shift;
861 	u64 nzones = num_bytes >> shift;
862 	u64 pos = hole_start;
863 	u64 begin, end;
864 	bool have_sb;
865 	int i;
866 
867 	ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
868 	ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
869 
870 	while (pos < hole_end) {
871 		begin = pos >> shift;
872 		end = begin + nzones;
873 
874 		if (end > zinfo->nr_zones)
875 			return hole_end;
876 
877 		/* Check if zones in the region are all empty */
878 		if (btrfs_dev_is_sequential(device, pos) &&
879 		    find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
880 			pos += zinfo->zone_size;
881 			continue;
882 		}
883 
884 		have_sb = false;
885 		for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
886 			u32 sb_zone;
887 			u64 sb_pos;
888 
889 			sb_zone = sb_zone_number(shift, i);
890 			if (!(end <= sb_zone ||
891 			      sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
892 				have_sb = true;
893 				pos = zone_start_physical(
894 					sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
895 				break;
896 			}
897 
898 			/* We also need to exclude regular superblock positions */
899 			sb_pos = btrfs_sb_offset(i);
900 			if (!(pos + num_bytes <= sb_pos ||
901 			      sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
902 				have_sb = true;
903 				pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
904 					    zinfo->zone_size);
905 				break;
906 			}
907 		}
908 		if (!have_sb)
909 			break;
910 	}
911 
912 	return pos;
913 }
914 
915 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
916 			    u64 length, u64 *bytes)
917 {
918 	int ret;
919 
920 	*bytes = 0;
921 	ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
922 			       physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
923 			       GFP_NOFS);
924 	if (ret)
925 		return ret;
926 
927 	*bytes = length;
928 	while (length) {
929 		btrfs_dev_set_zone_empty(device, physical);
930 		physical += device->zone_info->zone_size;
931 		length -= device->zone_info->zone_size;
932 	}
933 
934 	return 0;
935 }
936 
937 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
938 {
939 	struct btrfs_zoned_device_info *zinfo = device->zone_info;
940 	const u8 shift = zinfo->zone_size_shift;
941 	unsigned long begin = start >> shift;
942 	unsigned long end = (start + size) >> shift;
943 	u64 pos;
944 	int ret;
945 
946 	ASSERT(IS_ALIGNED(start, zinfo->zone_size));
947 	ASSERT(IS_ALIGNED(size, zinfo->zone_size));
948 
949 	if (end > zinfo->nr_zones)
950 		return -ERANGE;
951 
952 	/* All the zones are conventional */
953 	if (find_next_bit(zinfo->seq_zones, begin, end) == end)
954 		return 0;
955 
956 	/* All the zones are sequential and empty */
957 	if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
958 	    find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
959 		return 0;
960 
961 	for (pos = start; pos < start + size; pos += zinfo->zone_size) {
962 		u64 reset_bytes;
963 
964 		if (!btrfs_dev_is_sequential(device, pos) ||
965 		    btrfs_dev_is_empty_zone(device, pos))
966 			continue;
967 
968 		/* Free regions should be empty */
969 		btrfs_warn_in_rcu(
970 			device->fs_info,
971 		"zoned: resetting device %s (devid %llu) zone %llu for allocation",
972 			rcu_str_deref(device->name), device->devid, pos >> shift);
973 		WARN_ON_ONCE(1);
974 
975 		ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
976 					      &reset_bytes);
977 		if (ret)
978 			return ret;
979 	}
980 
981 	return 0;
982 }
983 
984 /*
985  * Calculate an allocation pointer from the extent allocation information
986  * for a block group consist of conventional zones. It is pointed to the
987  * end of the highest addressed extent in the block group as an allocation
988  * offset.
989  */
990 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
991 				   u64 *offset_ret)
992 {
993 	struct btrfs_fs_info *fs_info = cache->fs_info;
994 	struct btrfs_root *root = fs_info->extent_root;
995 	struct btrfs_path *path;
996 	struct btrfs_key key;
997 	struct btrfs_key found_key;
998 	int ret;
999 	u64 length;
1000 
1001 	path = btrfs_alloc_path();
1002 	if (!path)
1003 		return -ENOMEM;
1004 
1005 	key.objectid = cache->start + cache->length;
1006 	key.type = 0;
1007 	key.offset = 0;
1008 
1009 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1010 	/* We should not find the exact match */
1011 	if (!ret)
1012 		ret = -EUCLEAN;
1013 	if (ret < 0)
1014 		goto out;
1015 
1016 	ret = btrfs_previous_extent_item(root, path, cache->start);
1017 	if (ret) {
1018 		if (ret == 1) {
1019 			ret = 0;
1020 			*offset_ret = 0;
1021 		}
1022 		goto out;
1023 	}
1024 
1025 	btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1026 
1027 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1028 		length = found_key.offset;
1029 	else
1030 		length = fs_info->nodesize;
1031 
1032 	if (!(found_key.objectid >= cache->start &&
1033 	       found_key.objectid + length <= cache->start + cache->length)) {
1034 		ret = -EUCLEAN;
1035 		goto out;
1036 	}
1037 	*offset_ret = found_key.objectid + length - cache->start;
1038 	ret = 0;
1039 
1040 out:
1041 	btrfs_free_path(path);
1042 	return ret;
1043 }
1044 
1045 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1046 {
1047 	struct btrfs_fs_info *fs_info = cache->fs_info;
1048 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1049 	struct extent_map *em;
1050 	struct map_lookup *map;
1051 	struct btrfs_device *device;
1052 	u64 logical = cache->start;
1053 	u64 length = cache->length;
1054 	u64 physical = 0;
1055 	int ret;
1056 	int i;
1057 	unsigned int nofs_flag;
1058 	u64 *alloc_offsets = NULL;
1059 	u64 last_alloc = 0;
1060 	u32 num_sequential = 0, num_conventional = 0;
1061 
1062 	if (!btrfs_is_zoned(fs_info))
1063 		return 0;
1064 
1065 	/* Sanity check */
1066 	if (!IS_ALIGNED(length, fs_info->zone_size)) {
1067 		btrfs_err(fs_info,
1068 		"zoned: block group %llu len %llu unaligned to zone size %llu",
1069 			  logical, length, fs_info->zone_size);
1070 		return -EIO;
1071 	}
1072 
1073 	/* Get the chunk mapping */
1074 	read_lock(&em_tree->lock);
1075 	em = lookup_extent_mapping(em_tree, logical, length);
1076 	read_unlock(&em_tree->lock);
1077 
1078 	if (!em)
1079 		return -EINVAL;
1080 
1081 	map = em->map_lookup;
1082 
1083 	alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1084 	if (!alloc_offsets) {
1085 		free_extent_map(em);
1086 		return -ENOMEM;
1087 	}
1088 
1089 	for (i = 0; i < map->num_stripes; i++) {
1090 		bool is_sequential;
1091 		struct blk_zone zone;
1092 		struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1093 		int dev_replace_is_ongoing = 0;
1094 
1095 		device = map->stripes[i].dev;
1096 		physical = map->stripes[i].physical;
1097 
1098 		if (device->bdev == NULL) {
1099 			alloc_offsets[i] = WP_MISSING_DEV;
1100 			continue;
1101 		}
1102 
1103 		is_sequential = btrfs_dev_is_sequential(device, physical);
1104 		if (is_sequential)
1105 			num_sequential++;
1106 		else
1107 			num_conventional++;
1108 
1109 		if (!is_sequential) {
1110 			alloc_offsets[i] = WP_CONVENTIONAL;
1111 			continue;
1112 		}
1113 
1114 		/*
1115 		 * This zone will be used for allocation, so mark this zone
1116 		 * non-empty.
1117 		 */
1118 		btrfs_dev_clear_zone_empty(device, physical);
1119 
1120 		down_read(&dev_replace->rwsem);
1121 		dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1122 		if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1123 			btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical);
1124 		up_read(&dev_replace->rwsem);
1125 
1126 		/*
1127 		 * The group is mapped to a sequential zone. Get the zone write
1128 		 * pointer to determine the allocation offset within the zone.
1129 		 */
1130 		WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size));
1131 		nofs_flag = memalloc_nofs_save();
1132 		ret = btrfs_get_dev_zone(device, physical, &zone);
1133 		memalloc_nofs_restore(nofs_flag);
1134 		if (ret == -EIO || ret == -EOPNOTSUPP) {
1135 			ret = 0;
1136 			alloc_offsets[i] = WP_MISSING_DEV;
1137 			continue;
1138 		} else if (ret) {
1139 			goto out;
1140 		}
1141 
1142 		if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1143 			btrfs_err_in_rcu(fs_info,
1144 	"zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1145 				zone.start << SECTOR_SHIFT,
1146 				rcu_str_deref(device->name), device->devid);
1147 			ret = -EIO;
1148 			goto out;
1149 		}
1150 
1151 		switch (zone.cond) {
1152 		case BLK_ZONE_COND_OFFLINE:
1153 		case BLK_ZONE_COND_READONLY:
1154 			btrfs_err(fs_info,
1155 		"zoned: offline/readonly zone %llu on device %s (devid %llu)",
1156 				  physical >> device->zone_info->zone_size_shift,
1157 				  rcu_str_deref(device->name), device->devid);
1158 			alloc_offsets[i] = WP_MISSING_DEV;
1159 			break;
1160 		case BLK_ZONE_COND_EMPTY:
1161 			alloc_offsets[i] = 0;
1162 			break;
1163 		case BLK_ZONE_COND_FULL:
1164 			alloc_offsets[i] = fs_info->zone_size;
1165 			break;
1166 		default:
1167 			/* Partially used zone */
1168 			alloc_offsets[i] =
1169 					((zone.wp - zone.start) << SECTOR_SHIFT);
1170 			break;
1171 		}
1172 	}
1173 
1174 	if (num_sequential > 0)
1175 		cache->seq_zone = true;
1176 
1177 	if (num_conventional > 0) {
1178 		/*
1179 		 * Avoid calling calculate_alloc_pointer() for new BG. It
1180 		 * is no use for new BG. It must be always 0.
1181 		 *
1182 		 * Also, we have a lock chain of extent buffer lock ->
1183 		 * chunk mutex.  For new BG, this function is called from
1184 		 * btrfs_make_block_group() which is already taking the
1185 		 * chunk mutex. Thus, we cannot call
1186 		 * calculate_alloc_pointer() which takes extent buffer
1187 		 * locks to avoid deadlock.
1188 		 */
1189 		if (new) {
1190 			cache->alloc_offset = 0;
1191 			goto out;
1192 		}
1193 		ret = calculate_alloc_pointer(cache, &last_alloc);
1194 		if (ret || map->num_stripes == num_conventional) {
1195 			if (!ret)
1196 				cache->alloc_offset = last_alloc;
1197 			else
1198 				btrfs_err(fs_info,
1199 			"zoned: failed to determine allocation offset of bg %llu",
1200 					  cache->start);
1201 			goto out;
1202 		}
1203 	}
1204 
1205 	switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1206 	case 0: /* single */
1207 		if (alloc_offsets[0] == WP_MISSING_DEV) {
1208 			btrfs_err(fs_info,
1209 			"zoned: cannot recover write pointer for zone %llu",
1210 				physical);
1211 			ret = -EIO;
1212 			goto out;
1213 		}
1214 		cache->alloc_offset = alloc_offsets[0];
1215 		break;
1216 	case BTRFS_BLOCK_GROUP_DUP:
1217 	case BTRFS_BLOCK_GROUP_RAID1:
1218 	case BTRFS_BLOCK_GROUP_RAID0:
1219 	case BTRFS_BLOCK_GROUP_RAID10:
1220 	case BTRFS_BLOCK_GROUP_RAID5:
1221 	case BTRFS_BLOCK_GROUP_RAID6:
1222 		/* non-single profiles are not supported yet */
1223 	default:
1224 		btrfs_err(fs_info, "zoned: profile %s not yet supported",
1225 			  btrfs_bg_type_to_raid_name(map->type));
1226 		ret = -EINVAL;
1227 		goto out;
1228 	}
1229 
1230 out:
1231 	if (cache->alloc_offset > fs_info->zone_size) {
1232 		btrfs_err(fs_info,
1233 			"zoned: invalid write pointer %llu in block group %llu",
1234 			cache->alloc_offset, cache->start);
1235 		ret = -EIO;
1236 	}
1237 
1238 	/* An extent is allocated after the write pointer */
1239 	if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1240 		btrfs_err(fs_info,
1241 			  "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1242 			  logical, last_alloc, cache->alloc_offset);
1243 		ret = -EIO;
1244 	}
1245 
1246 	if (!ret)
1247 		cache->meta_write_pointer = cache->alloc_offset + cache->start;
1248 
1249 	kfree(alloc_offsets);
1250 	free_extent_map(em);
1251 
1252 	return ret;
1253 }
1254 
1255 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1256 {
1257 	u64 unusable, free;
1258 
1259 	if (!btrfs_is_zoned(cache->fs_info))
1260 		return;
1261 
1262 	WARN_ON(cache->bytes_super != 0);
1263 	unusable = cache->alloc_offset - cache->used;
1264 	free = cache->length - cache->alloc_offset;
1265 
1266 	/* We only need ->free_space in ALLOC_SEQ block groups */
1267 	cache->last_byte_to_unpin = (u64)-1;
1268 	cache->cached = BTRFS_CACHE_FINISHED;
1269 	cache->free_space_ctl->free_space = free;
1270 	cache->zone_unusable = unusable;
1271 
1272 	/* Should not have any excluded extents. Just in case, though */
1273 	btrfs_free_excluded_extents(cache);
1274 }
1275 
1276 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1277 			    struct extent_buffer *eb)
1278 {
1279 	struct btrfs_fs_info *fs_info = eb->fs_info;
1280 
1281 	if (!btrfs_is_zoned(fs_info) ||
1282 	    btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1283 	    !list_empty(&eb->release_list))
1284 		return;
1285 
1286 	set_extent_buffer_dirty(eb);
1287 	set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1288 			       eb->start + eb->len - 1, EXTENT_DIRTY);
1289 	memzero_extent_buffer(eb, 0, eb->len);
1290 	set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1291 
1292 	spin_lock(&trans->releasing_ebs_lock);
1293 	list_add_tail(&eb->release_list, &trans->releasing_ebs);
1294 	spin_unlock(&trans->releasing_ebs_lock);
1295 	atomic_inc(&eb->refs);
1296 }
1297 
1298 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1299 {
1300 	spin_lock(&trans->releasing_ebs_lock);
1301 	while (!list_empty(&trans->releasing_ebs)) {
1302 		struct extent_buffer *eb;
1303 
1304 		eb = list_first_entry(&trans->releasing_ebs,
1305 				      struct extent_buffer, release_list);
1306 		list_del_init(&eb->release_list);
1307 		free_extent_buffer(eb);
1308 	}
1309 	spin_unlock(&trans->releasing_ebs_lock);
1310 }
1311 
1312 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1313 {
1314 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1315 	struct btrfs_block_group *cache;
1316 	bool ret = false;
1317 
1318 	if (!btrfs_is_zoned(fs_info))
1319 		return false;
1320 
1321 	if (!fs_info->max_zone_append_size)
1322 		return false;
1323 
1324 	if (!is_data_inode(&inode->vfs_inode))
1325 		return false;
1326 
1327 	cache = btrfs_lookup_block_group(fs_info, start);
1328 	ASSERT(cache);
1329 	if (!cache)
1330 		return false;
1331 
1332 	ret = cache->seq_zone;
1333 	btrfs_put_block_group(cache);
1334 
1335 	return ret;
1336 }
1337 
1338 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1339 				 struct bio *bio)
1340 {
1341 	struct btrfs_ordered_extent *ordered;
1342 	const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1343 
1344 	if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1345 		return;
1346 
1347 	ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1348 	if (WARN_ON(!ordered))
1349 		return;
1350 
1351 	ordered->physical = physical;
1352 	ordered->disk = bio->bi_bdev->bd_disk;
1353 	ordered->partno = bio->bi_bdev->bd_partno;
1354 
1355 	btrfs_put_ordered_extent(ordered);
1356 }
1357 
1358 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1359 {
1360 	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1361 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1362 	struct extent_map_tree *em_tree;
1363 	struct extent_map *em;
1364 	struct btrfs_ordered_sum *sum;
1365 	struct block_device *bdev;
1366 	u64 orig_logical = ordered->disk_bytenr;
1367 	u64 *logical = NULL;
1368 	int nr, stripe_len;
1369 
1370 	/* Zoned devices should not have partitions. So, we can assume it is 0 */
1371 	ASSERT(ordered->partno == 0);
1372 	bdev = bdgrab(ordered->disk->part0);
1373 	if (WARN_ON(!bdev))
1374 		return;
1375 
1376 	if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, bdev,
1377 				     ordered->physical, &logical, &nr,
1378 				     &stripe_len)))
1379 		goto out;
1380 
1381 	WARN_ON(nr != 1);
1382 
1383 	if (orig_logical == *logical)
1384 		goto out;
1385 
1386 	ordered->disk_bytenr = *logical;
1387 
1388 	em_tree = &inode->extent_tree;
1389 	write_lock(&em_tree->lock);
1390 	em = search_extent_mapping(em_tree, ordered->file_offset,
1391 				   ordered->num_bytes);
1392 	em->block_start = *logical;
1393 	free_extent_map(em);
1394 	write_unlock(&em_tree->lock);
1395 
1396 	list_for_each_entry(sum, &ordered->list, list) {
1397 		if (*logical < orig_logical)
1398 			sum->bytenr -= orig_logical - *logical;
1399 		else
1400 			sum->bytenr += *logical - orig_logical;
1401 	}
1402 
1403 out:
1404 	kfree(logical);
1405 	bdput(bdev);
1406 }
1407 
1408 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1409 				    struct extent_buffer *eb,
1410 				    struct btrfs_block_group **cache_ret)
1411 {
1412 	struct btrfs_block_group *cache;
1413 	bool ret = true;
1414 
1415 	if (!btrfs_is_zoned(fs_info))
1416 		return true;
1417 
1418 	cache = *cache_ret;
1419 
1420 	if (cache && (eb->start < cache->start ||
1421 		      cache->start + cache->length <= eb->start)) {
1422 		btrfs_put_block_group(cache);
1423 		cache = NULL;
1424 		*cache_ret = NULL;
1425 	}
1426 
1427 	if (!cache)
1428 		cache = btrfs_lookup_block_group(fs_info, eb->start);
1429 
1430 	if (cache) {
1431 		if (cache->meta_write_pointer != eb->start) {
1432 			btrfs_put_block_group(cache);
1433 			cache = NULL;
1434 			ret = false;
1435 		} else {
1436 			cache->meta_write_pointer = eb->start + eb->len;
1437 		}
1438 
1439 		*cache_ret = cache;
1440 	}
1441 
1442 	return ret;
1443 }
1444 
1445 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1446 				     struct extent_buffer *eb)
1447 {
1448 	if (!btrfs_is_zoned(eb->fs_info) || !cache)
1449 		return;
1450 
1451 	ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1452 	cache->meta_write_pointer = eb->start;
1453 }
1454 
1455 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1456 {
1457 	if (!btrfs_dev_is_sequential(device, physical))
1458 		return -EOPNOTSUPP;
1459 
1460 	return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1461 				    length >> SECTOR_SHIFT, GFP_NOFS, 0);
1462 }
1463 
1464 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1465 			  struct blk_zone *zone)
1466 {
1467 	struct btrfs_bio *bbio = NULL;
1468 	u64 mapped_length = PAGE_SIZE;
1469 	unsigned int nofs_flag;
1470 	int nmirrors;
1471 	int i, ret;
1472 
1473 	ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1474 			       &mapped_length, &bbio);
1475 	if (ret || !bbio || mapped_length < PAGE_SIZE) {
1476 		btrfs_put_bbio(bbio);
1477 		return -EIO;
1478 	}
1479 
1480 	if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1481 		return -EINVAL;
1482 
1483 	nofs_flag = memalloc_nofs_save();
1484 	nmirrors = (int)bbio->num_stripes;
1485 	for (i = 0; i < nmirrors; i++) {
1486 		u64 physical = bbio->stripes[i].physical;
1487 		struct btrfs_device *dev = bbio->stripes[i].dev;
1488 
1489 		/* Missing device */
1490 		if (!dev->bdev)
1491 			continue;
1492 
1493 		ret = btrfs_get_dev_zone(dev, physical, zone);
1494 		/* Failing device */
1495 		if (ret == -EIO || ret == -EOPNOTSUPP)
1496 			continue;
1497 		break;
1498 	}
1499 	memalloc_nofs_restore(nofs_flag);
1500 
1501 	return ret;
1502 }
1503 
1504 /*
1505  * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1506  * filling zeros between @physical_pos to a write pointer of dev-replace
1507  * source device.
1508  */
1509 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1510 				    u64 physical_start, u64 physical_pos)
1511 {
1512 	struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1513 	struct blk_zone zone;
1514 	u64 length;
1515 	u64 wp;
1516 	int ret;
1517 
1518 	if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1519 		return 0;
1520 
1521 	ret = read_zone_info(fs_info, logical, &zone);
1522 	if (ret)
1523 		return ret;
1524 
1525 	wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1526 
1527 	if (physical_pos == wp)
1528 		return 0;
1529 
1530 	if (physical_pos > wp)
1531 		return -EUCLEAN;
1532 
1533 	length = wp - physical_pos;
1534 	return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1535 }
1536 
1537 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1538 					    u64 logical, u64 length)
1539 {
1540 	struct btrfs_device *device;
1541 	struct extent_map *em;
1542 	struct map_lookup *map;
1543 
1544 	em = btrfs_get_chunk_map(fs_info, logical, length);
1545 	if (IS_ERR(em))
1546 		return ERR_CAST(em);
1547 
1548 	map = em->map_lookup;
1549 	/* We only support single profile for now */
1550 	ASSERT(map->num_stripes == 1);
1551 	device = map->stripes[0].dev;
1552 
1553 	free_extent_map(em);
1554 
1555 	return device;
1556 }
1557