]> git.ipfire.org Git - thirdparty/linux.git/blob - fs/btrfs/zoned.c
projects / thirdparty / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'for-5.19-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave...
[thirdparty/linux.git] / fs / btrfs / zoned.c
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] * x 0
98 * In use[1] 0 x 0
99 * Full[1] 1 1 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 if (!IS_ALIGNED(nr_sectors, zone_sectors))
419 zone_info->nr_zones++;
420
421 max_active_zones = bdev_max_active_zones(bdev);
422 if (max_active_zones && max_active_zones < BTRFS_MIN_ACTIVE_ZONES) {
423 btrfs_err_in_rcu(fs_info,
424 "zoned: %s: max active zones %u is too small, need at least %u active zones",
425 rcu_str_deref(device->name), max_active_zones,
426 BTRFS_MIN_ACTIVE_ZONES);
427 ret = -EINVAL;
428 goto out;
429 }
430 zone_info->max_active_zones = max_active_zones;
431
432 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
433 if (!zone_info->seq_zones) {
434 ret = -ENOMEM;
435 goto out;
436 }
437
438 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
439 if (!zone_info->empty_zones) {
440 ret = -ENOMEM;
441 goto out;
442 }
443
444 zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
445 if (!zone_info->active_zones) {
446 ret = -ENOMEM;
447 goto out;
448 }
449
450 zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
451 if (!zones) {
452 ret = -ENOMEM;
453 goto out;
454 }
455
456 /*
457 * Enable zone cache only for a zoned device. On a non-zoned device, we
458 * fill the zone info with emulated CONVENTIONAL zones, so no need to
459 * use the cache.
460 */
461 if (populate_cache && bdev_is_zoned(device->bdev)) {
462 zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) *
463 zone_info->nr_zones);
464 if (!zone_info->zone_cache) {
465 btrfs_err_in_rcu(device->fs_info,
466 "zoned: failed to allocate zone cache for %s",
467 rcu_str_deref(device->name));
468 ret = -ENOMEM;
469 goto out;
470 }
471 }
472
473 /* Get zones type */
474 nactive = 0;
475 while (sector < nr_sectors) {
476 nr_zones = BTRFS_REPORT_NR_ZONES;
477 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
478 &nr_zones);
479 if (ret)
480 goto out;
481
482 for (i = 0; i < nr_zones; i++) {
483 if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
484 __set_bit(nreported, zone_info->seq_zones);
485 switch (zones[i].cond) {
486 case BLK_ZONE_COND_EMPTY:
487 __set_bit(nreported, zone_info->empty_zones);
488 break;
489 case BLK_ZONE_COND_IMP_OPEN:
490 case BLK_ZONE_COND_EXP_OPEN:
491 case BLK_ZONE_COND_CLOSED:
492 __set_bit(nreported, zone_info->active_zones);
493 nactive++;
494 break;
495 }
496 nreported++;
497 }
498 sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
499 }
500
501 if (nreported != zone_info->nr_zones) {
502 btrfs_err_in_rcu(device->fs_info,
503 "inconsistent number of zones on %s (%u/%u)",
504 rcu_str_deref(device->name), nreported,
505 zone_info->nr_zones);
506 ret = -EIO;
507 goto out;
508 }
509
510 if (max_active_zones) {
511 if (nactive > max_active_zones) {
512 btrfs_err_in_rcu(device->fs_info,
513 "zoned: %u active zones on %s exceeds max_active_zones %u",
514 nactive, rcu_str_deref(device->name),
515 max_active_zones);
516 ret = -EIO;
517 goto out;
518 }
519 atomic_set(&zone_info->active_zones_left,
520 max_active_zones - nactive);
521 }
522
523 /* Validate superblock log */
524 nr_zones = BTRFS_NR_SB_LOG_ZONES;
525 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
526 u32 sb_zone;
527 u64 sb_wp;
528 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
529
530 sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
531 if (sb_zone + 1 >= zone_info->nr_zones)
532 continue;
533
534 ret = btrfs_get_dev_zones(device,
535 zone_start_physical(sb_zone, zone_info),
536 &zone_info->sb_zones[sb_pos],
537 &nr_zones);
538 if (ret)
539 goto out;
540
541 if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
542 btrfs_err_in_rcu(device->fs_info,
543 "zoned: failed to read super block log zone info at devid %llu zone %u",
544 device->devid, sb_zone);
545 ret = -EUCLEAN;
546 goto out;
547 }
548
549 /*
550 * If zones[0] is conventional, always use the beginning of the
551 * zone to record superblock. No need to validate in that case.
552 */
553 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
554 BLK_ZONE_TYPE_CONVENTIONAL)
555 continue;
556
557 ret = sb_write_pointer(device->bdev,
558 &zone_info->sb_zones[sb_pos], &sb_wp);
559 if (ret != -ENOENT && ret) {
560 btrfs_err_in_rcu(device->fs_info,
561 "zoned: super block log zone corrupted devid %llu zone %u",
562 device->devid, sb_zone);
563 ret = -EUCLEAN;
564 goto out;
565 }
566 }
567
568
569 kfree(zones);
570
571 switch (bdev_zoned_model(bdev)) {
572 case BLK_ZONED_HM:
573 model = "host-managed zoned";
574 emulated = "";
575 break;
576 case BLK_ZONED_HA:
577 model = "host-aware zoned";
578 emulated = "";
579 break;
580 case BLK_ZONED_NONE:
581 model = "regular";
582 emulated = "emulated ";
583 break;
584 default:
585 /* Just in case */
586 btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
587 bdev_zoned_model(bdev),
588 rcu_str_deref(device->name));
589 ret = -EOPNOTSUPP;
590 goto out_free_zone_info;
591 }
592
593 btrfs_info_in_rcu(fs_info,
594 "%s block device %s, %u %szones of %llu bytes",
595 model, rcu_str_deref(device->name), zone_info->nr_zones,
596 emulated, zone_info->zone_size);
597
598 return 0;
599
600 out:
601 kfree(zones);
602 out_free_zone_info:
603 btrfs_destroy_dev_zone_info(device);
604
605 return ret;
606 }
607
608 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
609 {
610 struct btrfs_zoned_device_info *zone_info = device->zone_info;
611
612 if (!zone_info)
613 return;
614
615 bitmap_free(zone_info->active_zones);
616 bitmap_free(zone_info->seq_zones);
617 bitmap_free(zone_info->empty_zones);
618 vfree(zone_info->zone_cache);
619 kfree(zone_info);
620 device->zone_info = NULL;
621 }
622
623 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
624 struct blk_zone *zone)
625 {
626 unsigned int nr_zones = 1;
627 int ret;
628
629 ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
630 if (ret != 0 || !nr_zones)
631 return ret ? ret : -EIO;
632
633 return 0;
634 }
635
636 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
637 {
638 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
639 struct btrfs_device *device;
640 u64 zoned_devices = 0;
641 u64 nr_devices = 0;
642 u64 zone_size = 0;
643 const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED);
644 int ret = 0;
645
646 /* Count zoned devices */
647 list_for_each_entry(device, &fs_devices->devices, dev_list) {
648 enum blk_zoned_model model;
649
650 if (!device->bdev)
651 continue;
652
653 model = bdev_zoned_model(device->bdev);
654 /*
655 * A Host-Managed zoned device must be used as a zoned device.
656 * A Host-Aware zoned device and a non-zoned devices can be
657 * treated as a zoned device, if ZONED flag is enabled in the
658 * superblock.
659 */
660 if (model == BLK_ZONED_HM ||
661 (model == BLK_ZONED_HA && incompat_zoned) ||
662 (model == BLK_ZONED_NONE && incompat_zoned)) {
663 struct btrfs_zoned_device_info *zone_info;
664
665 zone_info = device->zone_info;
666 zoned_devices++;
667 if (!zone_size) {
668 zone_size = zone_info->zone_size;
669 } else if (zone_info->zone_size != zone_size) {
670 btrfs_err(fs_info,
671 "zoned: unequal block device zone sizes: have %llu found %llu",
672 device->zone_info->zone_size,
673 zone_size);
674 ret = -EINVAL;
675 goto out;
676 }
677 }
678 nr_devices++;
679 }
680
681 if (!zoned_devices && !incompat_zoned)
682 goto out;
683
684 if (!zoned_devices && incompat_zoned) {
685 /* No zoned block device found on ZONED filesystem */
686 btrfs_err(fs_info,
687 "zoned: no zoned devices found on a zoned filesystem");
688 ret = -EINVAL;
689 goto out;
690 }
691
692 if (zoned_devices && !incompat_zoned) {
693 btrfs_err(fs_info,
694 "zoned: mode not enabled but zoned device found");
695 ret = -EINVAL;
696 goto out;
697 }
698
699 if (zoned_devices != nr_devices) {
700 btrfs_err(fs_info,
701 "zoned: cannot mix zoned and regular devices");
702 ret = -EINVAL;
703 goto out;
704 }
705
706 /*
707 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
708 * btrfs_create_chunk(). Since we want stripe_len == zone_size,
709 * check the alignment here.
710 */
711 if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
712 btrfs_err(fs_info,
713 "zoned: zone size %llu not aligned to stripe %u",
714 zone_size, BTRFS_STRIPE_LEN);
715 ret = -EINVAL;
716 goto out;
717 }
718
719 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
720 btrfs_err(fs_info, "zoned: mixed block groups not supported");
721 ret = -EINVAL;
722 goto out;
723 }
724
725 fs_info->zone_size = zone_size;
726 fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
727
728 /*
729 * Check mount options here, because we might change fs_info->zoned
730 * from fs_info->zone_size.
731 */
732 ret = btrfs_check_mountopts_zoned(fs_info);
733 if (ret)
734 goto out;
735
736 btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
737 out:
738 return ret;
739 }
740
741 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
742 {
743 if (!btrfs_is_zoned(info))
744 return 0;
745
746 /*
747 * Space cache writing is not COWed. Disable that to avoid write errors
748 * in sequential zones.
749 */
750 if (btrfs_test_opt(info, SPACE_CACHE)) {
751 btrfs_err(info, "zoned: space cache v1 is not supported");
752 return -EINVAL;
753 }
754
755 if (btrfs_test_opt(info, NODATACOW)) {
756 btrfs_err(info, "zoned: NODATACOW not supported");
757 return -EINVAL;
758 }
759
760 return 0;
761 }
762
763 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
764 int rw, u64 *bytenr_ret)
765 {
766 u64 wp;
767 int ret;
768
769 if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
770 *bytenr_ret = zones[0].start << SECTOR_SHIFT;
771 return 0;
772 }
773
774 ret = sb_write_pointer(bdev, zones, &wp);
775 if (ret != -ENOENT && ret < 0)
776 return ret;
777
778 if (rw == WRITE) {
779 struct blk_zone *reset = NULL;
780
781 if (wp == zones[0].start << SECTOR_SHIFT)
782 reset = &zones[0];
783 else if (wp == zones[1].start << SECTOR_SHIFT)
784 reset = &zones[1];
785
786 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
787 ASSERT(sb_zone_is_full(reset));
788
789 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
790 reset->start, reset->len,
791 GFP_NOFS);
792 if (ret)
793 return ret;
794
795 reset->cond = BLK_ZONE_COND_EMPTY;
796 reset->wp = reset->start;
797 }
798 } else if (ret != -ENOENT) {
799 /*
800 * For READ, we want the previous one. Move write pointer to
801 * the end of a zone, if it is at the head of a zone.
802 */
803 u64 zone_end = 0;
804
805 if (wp == zones[0].start << SECTOR_SHIFT)
806 zone_end = zones[1].start + zones[1].capacity;
807 else if (wp == zones[1].start << SECTOR_SHIFT)
808 zone_end = zones[0].start + zones[0].capacity;
809 if (zone_end)
810 wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT,
811 BTRFS_SUPER_INFO_SIZE);
812
813 wp -= BTRFS_SUPER_INFO_SIZE;
814 }
815
816 *bytenr_ret = wp;
817 return 0;
818
819 }
820
821 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
822 u64 *bytenr_ret)
823 {
824 struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
825 sector_t zone_sectors;
826 u32 sb_zone;
827 int ret;
828 u8 zone_sectors_shift;
829 sector_t nr_sectors;
830 u32 nr_zones;
831
832 if (!bdev_is_zoned(bdev)) {
833 *bytenr_ret = btrfs_sb_offset(mirror);
834 return 0;
835 }
836
837 ASSERT(rw == READ || rw == WRITE);
838
839 zone_sectors = bdev_zone_sectors(bdev);
840 if (!is_power_of_2(zone_sectors))
841 return -EINVAL;
842 zone_sectors_shift = ilog2(zone_sectors);
843 nr_sectors = bdev_nr_sectors(bdev);
844 nr_zones = nr_sectors >> zone_sectors_shift;
845
846 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
847 if (sb_zone + 1 >= nr_zones)
848 return -ENOENT;
849
850 ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
851 BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
852 zones);
853 if (ret < 0)
854 return ret;
855 if (ret != BTRFS_NR_SB_LOG_ZONES)
856 return -EIO;
857
858 return sb_log_location(bdev, zones, rw, bytenr_ret);
859 }
860
861 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
862 u64 *bytenr_ret)
863 {
864 struct btrfs_zoned_device_info *zinfo = device->zone_info;
865 u32 zone_num;
866
867 /*
868 * For a zoned filesystem on a non-zoned block device, use the same
869 * super block locations as regular filesystem. Doing so, the super
870 * block can always be retrieved and the zoned flag of the volume
871 * detected from the super block information.
872 */
873 if (!bdev_is_zoned(device->bdev)) {
874 *bytenr_ret = btrfs_sb_offset(mirror);
875 return 0;
876 }
877
878 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
879 if (zone_num + 1 >= zinfo->nr_zones)
880 return -ENOENT;
881
882 return sb_log_location(device->bdev,
883 &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
884 rw, bytenr_ret);
885 }
886
887 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
888 int mirror)
889 {
890 u32 zone_num;
891
892 if (!zinfo)
893 return false;
894
895 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
896 if (zone_num + 1 >= zinfo->nr_zones)
897 return false;
898
899 if (!test_bit(zone_num, zinfo->seq_zones))
900 return false;
901
902 return true;
903 }
904
905 int btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
906 {
907 struct btrfs_zoned_device_info *zinfo = device->zone_info;
908 struct blk_zone *zone;
909 int i;
910
911 if (!is_sb_log_zone(zinfo, mirror))
912 return 0;
913
914 zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
915 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
916 /* Advance the next zone */
917 if (zone->cond == BLK_ZONE_COND_FULL) {
918 zone++;
919 continue;
920 }
921
922 if (zone->cond == BLK_ZONE_COND_EMPTY)
923 zone->cond = BLK_ZONE_COND_IMP_OPEN;
924
925 zone->wp += SUPER_INFO_SECTORS;
926
927 if (sb_zone_is_full(zone)) {
928 /*
929 * No room left to write new superblock. Since
930 * superblock is written with REQ_SYNC, it is safe to
931 * finish the zone now.
932 *
933 * If the write pointer is exactly at the capacity,
934 * explicit ZONE_FINISH is not necessary.
935 */
936 if (zone->wp != zone->start + zone->capacity) {
937 int ret;
938
939 ret = blkdev_zone_mgmt(device->bdev,
940 REQ_OP_ZONE_FINISH, zone->start,
941 zone->len, GFP_NOFS);
942 if (ret)
943 return ret;
944 }
945
946 zone->wp = zone->start + zone->len;
947 zone->cond = BLK_ZONE_COND_FULL;
948 }
949 return 0;
950 }
951
952 /* All the zones are FULL. Should not reach here. */
953 ASSERT(0);
954 return -EIO;
955 }
956
957 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
958 {
959 sector_t zone_sectors;
960 sector_t nr_sectors;
961 u8 zone_sectors_shift;
962 u32 sb_zone;
963 u32 nr_zones;
964
965 zone_sectors = bdev_zone_sectors(bdev);
966 zone_sectors_shift = ilog2(zone_sectors);
967 nr_sectors = bdev_nr_sectors(bdev);
968 nr_zones = nr_sectors >> zone_sectors_shift;
969
970 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
971 if (sb_zone + 1 >= nr_zones)
972 return -ENOENT;
973
974 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
975 zone_start_sector(sb_zone, bdev),
976 zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
977 }
978
979 /**
980 * btrfs_find_allocatable_zones - find allocatable zones within a given region
981 *
982 * @device: the device to allocate a region on
983 * @hole_start: the position of the hole to allocate the region
984 * @num_bytes: size of wanted region
985 * @hole_end: the end of the hole
986 * @return: position of allocatable zones
987 *
988 * Allocatable region should not contain any superblock locations.
989 */
990 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
991 u64 hole_end, u64 num_bytes)
992 {
993 struct btrfs_zoned_device_info *zinfo = device->zone_info;
994 const u8 shift = zinfo->zone_size_shift;
995 u64 nzones = num_bytes >> shift;
996 u64 pos = hole_start;
997 u64 begin, end;
998 bool have_sb;
999 int i;
1000
1001 ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
1002 ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
1003
1004 while (pos < hole_end) {
1005 begin = pos >> shift;
1006 end = begin + nzones;
1007
1008 if (end > zinfo->nr_zones)
1009 return hole_end;
1010
1011 /* Check if zones in the region are all empty */
1012 if (btrfs_dev_is_sequential(device, pos) &&
1013 find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
1014 pos += zinfo->zone_size;
1015 continue;
1016 }
1017
1018 have_sb = false;
1019 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1020 u32 sb_zone;
1021 u64 sb_pos;
1022
1023 sb_zone = sb_zone_number(shift, i);
1024 if (!(end <= sb_zone ||
1025 sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
1026 have_sb = true;
1027 pos = zone_start_physical(
1028 sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
1029 break;
1030 }
1031
1032 /* We also need to exclude regular superblock positions */
1033 sb_pos = btrfs_sb_offset(i);
1034 if (!(pos + num_bytes <= sb_pos ||
1035 sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
1036 have_sb = true;
1037 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
1038 zinfo->zone_size);
1039 break;
1040 }
1041 }
1042 if (!have_sb)
1043 break;
1044 }
1045
1046 return pos;
1047 }
1048
1049 static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos)
1050 {
1051 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1052 unsigned int zno = (pos >> zone_info->zone_size_shift);
1053
1054 /* We can use any number of zones */
1055 if (zone_info->max_active_zones == 0)
1056 return true;
1057
1058 if (!test_bit(zno, zone_info->active_zones)) {
1059 /* Active zone left? */
1060 if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0)
1061 return false;
1062 if (test_and_set_bit(zno, zone_info->active_zones)) {
1063 /* Someone already set the bit */
1064 atomic_inc(&zone_info->active_zones_left);
1065 }
1066 }
1067
1068 return true;
1069 }
1070
1071 static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos)
1072 {
1073 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1074 unsigned int zno = (pos >> zone_info->zone_size_shift);
1075
1076 /* We can use any number of zones */
1077 if (zone_info->max_active_zones == 0)
1078 return;
1079
1080 if (test_and_clear_bit(zno, zone_info->active_zones))
1081 atomic_inc(&zone_info->active_zones_left);
1082 }
1083
1084 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
1085 u64 length, u64 *bytes)
1086 {
1087 int ret;
1088
1089 *bytes = 0;
1090 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
1091 physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
1092 GFP_NOFS);
1093 if (ret)
1094 return ret;
1095
1096 *bytes = length;
1097 while (length) {
1098 btrfs_dev_set_zone_empty(device, physical);
1099 btrfs_dev_clear_active_zone(device, physical);
1100 physical += device->zone_info->zone_size;
1101 length -= device->zone_info->zone_size;
1102 }
1103
1104 return 0;
1105 }
1106
1107 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
1108 {
1109 struct btrfs_zoned_device_info *zinfo = device->zone_info;
1110 const u8 shift = zinfo->zone_size_shift;
1111 unsigned long begin = start >> shift;
1112 unsigned long end = (start + size) >> shift;
1113 u64 pos;
1114 int ret;
1115
1116 ASSERT(IS_ALIGNED(start, zinfo->zone_size));
1117 ASSERT(IS_ALIGNED(size, zinfo->zone_size));
1118
1119 if (end > zinfo->nr_zones)
1120 return -ERANGE;
1121
1122 /* All the zones are conventional */
1123 if (find_next_bit(zinfo->seq_zones, begin, end) == end)
1124 return 0;
1125
1126 /* All the zones are sequential and empty */
1127 if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
1128 find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
1129 return 0;
1130
1131 for (pos = start; pos < start + size; pos += zinfo->zone_size) {
1132 u64 reset_bytes;
1133
1134 if (!btrfs_dev_is_sequential(device, pos) ||
1135 btrfs_dev_is_empty_zone(device, pos))
1136 continue;
1137
1138 /* Free regions should be empty */
1139 btrfs_warn_in_rcu(
1140 device->fs_info,
1141 "zoned: resetting device %s (devid %llu) zone %llu for allocation",
1142 rcu_str_deref(device->name), device->devid, pos >> shift);
1143 WARN_ON_ONCE(1);
1144
1145 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1146 &reset_bytes);
1147 if (ret)
1148 return ret;
1149 }
1150
1151 return 0;
1152 }
1153
1154 /*
1155 * Calculate an allocation pointer from the extent allocation information
1156 * for a block group consist of conventional zones. It is pointed to the
1157 * end of the highest addressed extent in the block group as an allocation
1158 * offset.
1159 */
1160 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1161 u64 *offset_ret)
1162 {
1163 struct btrfs_fs_info *fs_info = cache->fs_info;
1164 struct btrfs_root *root;
1165 struct btrfs_path *path;
1166 struct btrfs_key key;
1167 struct btrfs_key found_key;
1168 int ret;
1169 u64 length;
1170
1171 path = btrfs_alloc_path();
1172 if (!path)
1173 return -ENOMEM;
1174
1175 key.objectid = cache->start + cache->length;
1176 key.type = 0;
1177 key.offset = 0;
1178
1179 root = btrfs_extent_root(fs_info, key.objectid);
1180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1181 /* We should not find the exact match */
1182 if (!ret)
1183 ret = -EUCLEAN;
1184 if (ret < 0)
1185 goto out;
1186
1187 ret = btrfs_previous_extent_item(root, path, cache->start);
1188 if (ret) {
1189 if (ret == 1) {
1190 ret = 0;
1191 *offset_ret = 0;
1192 }
1193 goto out;
1194 }
1195
1196 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1197
1198 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1199 length = found_key.offset;
1200 else
1201 length = fs_info->nodesize;
1202
1203 if (!(found_key.objectid >= cache->start &&
1204 found_key.objectid + length <= cache->start + cache->length)) {
1205 ret = -EUCLEAN;
1206 goto out;
1207 }
1208 *offset_ret = found_key.objectid + length - cache->start;
1209 ret = 0;
1210
1211 out:
1212 btrfs_free_path(path);
1213 return ret;
1214 }
1215
1216 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1217 {
1218 struct btrfs_fs_info *fs_info = cache->fs_info;
1219 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1220 struct extent_map *em;
1221 struct map_lookup *map;
1222 struct btrfs_device *device;
1223 u64 logical = cache->start;
1224 u64 length = cache->length;
1225 int ret;
1226 int i;
1227 unsigned int nofs_flag;
1228 u64 *alloc_offsets = NULL;
1229 u64 *caps = NULL;
1230 u64 *physical = NULL;
1231 unsigned long *active = NULL;
1232 u64 last_alloc = 0;
1233 u32 num_sequential = 0, num_conventional = 0;
1234
1235 if (!btrfs_is_zoned(fs_info))
1236 return 0;
1237
1238 /* Sanity check */
1239 if (!IS_ALIGNED(length, fs_info->zone_size)) {
1240 btrfs_err(fs_info,
1241 "zoned: block group %llu len %llu unaligned to zone size %llu",
1242 logical, length, fs_info->zone_size);
1243 return -EIO;
1244 }
1245
1246 /* Get the chunk mapping */
1247 read_lock(&em_tree->lock);
1248 em = lookup_extent_mapping(em_tree, logical, length);
1249 read_unlock(&em_tree->lock);
1250
1251 if (!em)
1252 return -EINVAL;
1253
1254 map = em->map_lookup;
1255
1256 cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS);
1257 if (!cache->physical_map) {
1258 ret = -ENOMEM;
1259 goto out;
1260 }
1261
1262 alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1263 if (!alloc_offsets) {
1264 ret = -ENOMEM;
1265 goto out;
1266 }
1267
1268 caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS);
1269 if (!caps) {
1270 ret = -ENOMEM;
1271 goto out;
1272 }
1273
1274 physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS);
1275 if (!physical) {
1276 ret = -ENOMEM;
1277 goto out;
1278 }
1279
1280 active = bitmap_zalloc(map->num_stripes, GFP_NOFS);
1281 if (!active) {
1282 ret = -ENOMEM;
1283 goto out;
1284 }
1285
1286 for (i = 0; i < map->num_stripes; i++) {
1287 bool is_sequential;
1288 struct blk_zone zone;
1289 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1290 int dev_replace_is_ongoing = 0;
1291
1292 device = map->stripes[i].dev;
1293 physical[i] = map->stripes[i].physical;
1294
1295 if (device->bdev == NULL) {
1296 alloc_offsets[i] = WP_MISSING_DEV;
1297 continue;
1298 }
1299
1300 is_sequential = btrfs_dev_is_sequential(device, physical[i]);
1301 if (is_sequential)
1302 num_sequential++;
1303 else
1304 num_conventional++;
1305
1306 if (!is_sequential) {
1307 alloc_offsets[i] = WP_CONVENTIONAL;
1308 continue;
1309 }
1310
1311 /*
1312 * This zone will be used for allocation, so mark this zone
1313 * non-empty.
1314 */
1315 btrfs_dev_clear_zone_empty(device, physical[i]);
1316
1317 down_read(&dev_replace->rwsem);
1318 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1319 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1320 btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]);
1321 up_read(&dev_replace->rwsem);
1322
1323 /*
1324 * The group is mapped to a sequential zone. Get the zone write
1325 * pointer to determine the allocation offset within the zone.
1326 */
1327 WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size));
1328 nofs_flag = memalloc_nofs_save();
1329 ret = btrfs_get_dev_zone(device, physical[i], &zone);
1330 memalloc_nofs_restore(nofs_flag);
1331 if (ret == -EIO || ret == -EOPNOTSUPP) {
1332 ret = 0;
1333 alloc_offsets[i] = WP_MISSING_DEV;
1334 continue;
1335 } else if (ret) {
1336 goto out;
1337 }
1338
1339 if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1340 btrfs_err_in_rcu(fs_info,
1341 "zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1342 zone.start << SECTOR_SHIFT,
1343 rcu_str_deref(device->name), device->devid);
1344 ret = -EIO;
1345 goto out;
1346 }
1347
1348 caps[i] = (zone.capacity << SECTOR_SHIFT);
1349
1350 switch (zone.cond) {
1351 case BLK_ZONE_COND_OFFLINE:
1352 case BLK_ZONE_COND_READONLY:
1353 btrfs_err(fs_info,
1354 "zoned: offline/readonly zone %llu on device %s (devid %llu)",
1355 physical[i] >> device->zone_info->zone_size_shift,
1356 rcu_str_deref(device->name), device->devid);
1357 alloc_offsets[i] = WP_MISSING_DEV;
1358 break;
1359 case BLK_ZONE_COND_EMPTY:
1360 alloc_offsets[i] = 0;
1361 break;
1362 case BLK_ZONE_COND_FULL:
1363 alloc_offsets[i] = caps[i];
1364 break;
1365 default:
1366 /* Partially used zone */
1367 alloc_offsets[i] =
1368 ((zone.wp - zone.start) << SECTOR_SHIFT);
1369 __set_bit(i, active);
1370 break;
1371 }
1372
1373 /*
1374 * Consider a zone as active if we can allow any number of
1375 * active zones.
1376 */
1377 if (!device->zone_info->max_active_zones)
1378 __set_bit(i, active);
1379 }
1380
1381 if (num_sequential > 0)
1382 cache->seq_zone = true;
1383
1384 if (num_conventional > 0) {
1385 /*
1386 * Avoid calling calculate_alloc_pointer() for new BG. It
1387 * is no use for new BG. It must be always 0.
1388 *
1389 * Also, we have a lock chain of extent buffer lock ->
1390 * chunk mutex. For new BG, this function is called from
1391 * btrfs_make_block_group() which is already taking the
1392 * chunk mutex. Thus, we cannot call
1393 * calculate_alloc_pointer() which takes extent buffer
1394 * locks to avoid deadlock.
1395 */
1396
1397 /* Zone capacity is always zone size in emulation */
1398 cache->zone_capacity = cache->length;
1399 if (new) {
1400 cache->alloc_offset = 0;
1401 goto out;
1402 }
1403 ret = calculate_alloc_pointer(cache, &last_alloc);
1404 if (ret || map->num_stripes == num_conventional) {
1405 if (!ret)
1406 cache->alloc_offset = last_alloc;
1407 else
1408 btrfs_err(fs_info,
1409 "zoned: failed to determine allocation offset of bg %llu",
1410 cache->start);
1411 goto out;
1412 }
1413 }
1414
1415 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1416 case 0: /* single */
1417 if (alloc_offsets[0] == WP_MISSING_DEV) {
1418 btrfs_err(fs_info,
1419 "zoned: cannot recover write pointer for zone %llu",
1420 physical[0]);
1421 ret = -EIO;
1422 goto out;
1423 }
1424 cache->alloc_offset = alloc_offsets[0];
1425 cache->zone_capacity = caps[0];
1426 cache->zone_is_active = test_bit(0, active);
1427 break;
1428 case BTRFS_BLOCK_GROUP_DUP:
1429 if (map->type & BTRFS_BLOCK_GROUP_DATA) {
1430 btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg");
1431 ret = -EINVAL;
1432 goto out;
1433 }
1434 if (alloc_offsets[0] == WP_MISSING_DEV) {
1435 btrfs_err(fs_info,
1436 "zoned: cannot recover write pointer for zone %llu",
1437 physical[0]);
1438 ret = -EIO;
1439 goto out;
1440 }
1441 if (alloc_offsets[1] == WP_MISSING_DEV) {
1442 btrfs_err(fs_info,
1443 "zoned: cannot recover write pointer for zone %llu",
1444 physical[1]);
1445 ret = -EIO;
1446 goto out;
1447 }
1448 if (alloc_offsets[0] != alloc_offsets[1]) {
1449 btrfs_err(fs_info,
1450 "zoned: write pointer offset mismatch of zones in DUP profile");
1451 ret = -EIO;
1452 goto out;
1453 }
1454 if (test_bit(0, active) != test_bit(1, active)) {
1455 if (!btrfs_zone_activate(cache)) {
1456 ret = -EIO;
1457 goto out;
1458 }
1459 } else {
1460 cache->zone_is_active = test_bit(0, active);
1461 }
1462 cache->alloc_offset = alloc_offsets[0];
1463 cache->zone_capacity = min(caps[0], caps[1]);
1464 break;
1465 case BTRFS_BLOCK_GROUP_RAID1:
1466 case BTRFS_BLOCK_GROUP_RAID0:
1467 case BTRFS_BLOCK_GROUP_RAID10:
1468 case BTRFS_BLOCK_GROUP_RAID5:
1469 case BTRFS_BLOCK_GROUP_RAID6:
1470 /* non-single profiles are not supported yet */
1471 default:
1472 btrfs_err(fs_info, "zoned: profile %s not yet supported",
1473 btrfs_bg_type_to_raid_name(map->type));
1474 ret = -EINVAL;
1475 goto out;
1476 }
1477
1478 if (cache->zone_is_active) {
1479 btrfs_get_block_group(cache);
1480 spin_lock(&fs_info->zone_active_bgs_lock);
1481 list_add_tail(&cache->active_bg_list, &fs_info->zone_active_bgs);
1482 spin_unlock(&fs_info->zone_active_bgs_lock);
1483 }
1484
1485 out:
1486 if (cache->alloc_offset > fs_info->zone_size) {
1487 btrfs_err(fs_info,
1488 "zoned: invalid write pointer %llu in block group %llu",
1489 cache->alloc_offset, cache->start);
1490 ret = -EIO;
1491 }
1492
1493 if (cache->alloc_offset > cache->zone_capacity) {
1494 btrfs_err(fs_info,
1495 "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu",
1496 cache->alloc_offset, cache->zone_capacity,
1497 cache->start);
1498 ret = -EIO;
1499 }
1500
1501 /* An extent is allocated after the write pointer */
1502 if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1503 btrfs_err(fs_info,
1504 "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1505 logical, last_alloc, cache->alloc_offset);
1506 ret = -EIO;
1507 }
1508
1509 if (!ret)
1510 cache->meta_write_pointer = cache->alloc_offset + cache->start;
1511
1512 if (ret) {
1513 kfree(cache->physical_map);
1514 cache->physical_map = NULL;
1515 }
1516 bitmap_free(active);
1517 kfree(physical);
1518 kfree(caps);
1519 kfree(alloc_offsets);
1520 free_extent_map(em);
1521
1522 return ret;
1523 }
1524
1525 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1526 {
1527 u64 unusable, free;
1528
1529 if (!btrfs_is_zoned(cache->fs_info))
1530 return;
1531
1532 WARN_ON(cache->bytes_super != 0);
1533 unusable = (cache->alloc_offset - cache->used) +
1534 (cache->length - cache->zone_capacity);
1535 free = cache->zone_capacity - cache->alloc_offset;
1536
1537 /* We only need ->free_space in ALLOC_SEQ block groups */
1538 cache->last_byte_to_unpin = (u64)-1;
1539 cache->cached = BTRFS_CACHE_FINISHED;
1540 cache->free_space_ctl->free_space = free;
1541 cache->zone_unusable = unusable;
1542 }
1543
1544 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1545 struct extent_buffer *eb)
1546 {
1547 struct btrfs_fs_info *fs_info = eb->fs_info;
1548
1549 if (!btrfs_is_zoned(fs_info) ||
1550 btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1551 !list_empty(&eb->release_list))
1552 return;
1553
1554 set_extent_buffer_dirty(eb);
1555 set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1556 eb->start + eb->len - 1, EXTENT_DIRTY);
1557 memzero_extent_buffer(eb, 0, eb->len);
1558 set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1559
1560 spin_lock(&trans->releasing_ebs_lock);
1561 list_add_tail(&eb->release_list, &trans->releasing_ebs);
1562 spin_unlock(&trans->releasing_ebs_lock);
1563 atomic_inc(&eb->refs);
1564 }
1565
1566 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1567 {
1568 spin_lock(&trans->releasing_ebs_lock);
1569 while (!list_empty(&trans->releasing_ebs)) {
1570 struct extent_buffer *eb;
1571
1572 eb = list_first_entry(&trans->releasing_ebs,
1573 struct extent_buffer, release_list);
1574 list_del_init(&eb->release_list);
1575 free_extent_buffer(eb);
1576 }
1577 spin_unlock(&trans->releasing_ebs_lock);
1578 }
1579
1580 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1581 {
1582 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1583 struct btrfs_block_group *cache;
1584 bool ret = false;
1585
1586 if (!btrfs_is_zoned(fs_info))
1587 return false;
1588
1589 if (!is_data_inode(&inode->vfs_inode))
1590 return false;
1591
1592 /*
1593 * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1594 * extent layout the relocation code has.
1595 * Furthermore we have set aside own block-group from which only the
1596 * relocation "process" can allocate and make sure only one process at a
1597 * time can add pages to an extent that gets relocated, so it's safe to
1598 * use regular REQ_OP_WRITE for this special case.
1599 */
1600 if (btrfs_is_data_reloc_root(inode->root))
1601 return false;
1602
1603 cache = btrfs_lookup_block_group(fs_info, start);
1604 ASSERT(cache);
1605 if (!cache)
1606 return false;
1607
1608 ret = cache->seq_zone;
1609 btrfs_put_block_group(cache);
1610
1611 return ret;
1612 }
1613
1614 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1615 struct bio *bio)
1616 {
1617 struct btrfs_ordered_extent *ordered;
1618 const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1619
1620 if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1621 return;
1622
1623 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1624 if (WARN_ON(!ordered))
1625 return;
1626
1627 ordered->physical = physical;
1628 ordered->bdev = bio->bi_bdev;
1629
1630 btrfs_put_ordered_extent(ordered);
1631 }
1632
1633 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1634 {
1635 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1636 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1637 struct extent_map_tree *em_tree;
1638 struct extent_map *em;
1639 struct btrfs_ordered_sum *sum;
1640 u64 orig_logical = ordered->disk_bytenr;
1641 u64 *logical = NULL;
1642 int nr, stripe_len;
1643
1644 /* Zoned devices should not have partitions. So, we can assume it is 0 */
1645 ASSERT(!bdev_is_partition(ordered->bdev));
1646 if (WARN_ON(!ordered->bdev))
1647 return;
1648
1649 if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev,
1650 ordered->physical, &logical, &nr,
1651 &stripe_len)))
1652 goto out;
1653
1654 WARN_ON(nr != 1);
1655
1656 if (orig_logical == *logical)
1657 goto out;
1658
1659 ordered->disk_bytenr = *logical;
1660
1661 em_tree = &inode->extent_tree;
1662 write_lock(&em_tree->lock);
1663 em = search_extent_mapping(em_tree, ordered->file_offset,
1664 ordered->num_bytes);
1665 em->block_start = *logical;
1666 free_extent_map(em);
1667 write_unlock(&em_tree->lock);
1668
1669 list_for_each_entry(sum, &ordered->list, list) {
1670 if (*logical < orig_logical)
1671 sum->bytenr -= orig_logical - *logical;
1672 else
1673 sum->bytenr += *logical - orig_logical;
1674 }
1675
1676 out:
1677 kfree(logical);
1678 }
1679
1680 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1681 struct extent_buffer *eb,
1682 struct btrfs_block_group **cache_ret)
1683 {
1684 struct btrfs_block_group *cache;
1685 bool ret = true;
1686
1687 if (!btrfs_is_zoned(fs_info))
1688 return true;
1689
1690 cache = btrfs_lookup_block_group(fs_info, eb->start);
1691 if (!cache)
1692 return true;
1693
1694 if (cache->meta_write_pointer != eb->start) {
1695 btrfs_put_block_group(cache);
1696 cache = NULL;
1697 ret = false;
1698 } else {
1699 cache->meta_write_pointer = eb->start + eb->len;
1700 }
1701
1702 *cache_ret = cache;
1703
1704 return ret;
1705 }
1706
1707 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1708 struct extent_buffer *eb)
1709 {
1710 if (!btrfs_is_zoned(eb->fs_info) || !cache)
1711 return;
1712
1713 ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1714 cache->meta_write_pointer = eb->start;
1715 }
1716
1717 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1718 {
1719 if (!btrfs_dev_is_sequential(device, physical))
1720 return -EOPNOTSUPP;
1721
1722 return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1723 length >> SECTOR_SHIFT, GFP_NOFS, 0);
1724 }
1725
1726 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1727 struct blk_zone *zone)
1728 {
1729 struct btrfs_io_context *bioc = NULL;
1730 u64 mapped_length = PAGE_SIZE;
1731 unsigned int nofs_flag;
1732 int nmirrors;
1733 int i, ret;
1734
1735 ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1736 &mapped_length, &bioc);
1737 if (ret || !bioc || mapped_length < PAGE_SIZE) {
1738 ret = -EIO;
1739 goto out_put_bioc;
1740 }
1741
1742 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1743 ret = -EINVAL;
1744 goto out_put_bioc;
1745 }
1746
1747 nofs_flag = memalloc_nofs_save();
1748 nmirrors = (int)bioc->num_stripes;
1749 for (i = 0; i < nmirrors; i++) {
1750 u64 physical = bioc->stripes[i].physical;
1751 struct btrfs_device *dev = bioc->stripes[i].dev;
1752
1753 /* Missing device */
1754 if (!dev->bdev)
1755 continue;
1756
1757 ret = btrfs_get_dev_zone(dev, physical, zone);
1758 /* Failing device */
1759 if (ret == -EIO || ret == -EOPNOTSUPP)
1760 continue;
1761 break;
1762 }
1763 memalloc_nofs_restore(nofs_flag);
1764 out_put_bioc:
1765 btrfs_put_bioc(bioc);
1766 return ret;
1767 }
1768
1769 /*
1770 * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1771 * filling zeros between @physical_pos to a write pointer of dev-replace
1772 * source device.
1773 */
1774 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1775 u64 physical_start, u64 physical_pos)
1776 {
1777 struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1778 struct blk_zone zone;
1779 u64 length;
1780 u64 wp;
1781 int ret;
1782
1783 if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1784 return 0;
1785
1786 ret = read_zone_info(fs_info, logical, &zone);
1787 if (ret)
1788 return ret;
1789
1790 wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1791
1792 if (physical_pos == wp)
1793 return 0;
1794
1795 if (physical_pos > wp)
1796 return -EUCLEAN;
1797
1798 length = wp - physical_pos;
1799 return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1800 }
1801
1802 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1803 u64 logical, u64 length)
1804 {
1805 struct btrfs_device *device;
1806 struct extent_map *em;
1807 struct map_lookup *map;
1808
1809 em = btrfs_get_chunk_map(fs_info, logical, length);
1810 if (IS_ERR(em))
1811 return ERR_CAST(em);
1812
1813 map = em->map_lookup;
1814 /* We only support single profile for now */
1815 device = map->stripes[0].dev;
1816
1817 free_extent_map(em);
1818
1819 return device;
1820 }
1821
1822 /**
1823 * Activate block group and underlying device zones
1824 *
1825 * @block_group: the block group to activate
1826 *
1827 * Return: true on success, false otherwise
1828 */
1829 bool btrfs_zone_activate(struct btrfs_block_group *block_group)
1830 {
1831 struct btrfs_fs_info *fs_info = block_group->fs_info;
1832 struct map_lookup *map;
1833 struct btrfs_device *device;
1834 u64 physical;
1835 bool ret;
1836 int i;
1837
1838 if (!btrfs_is_zoned(block_group->fs_info))
1839 return true;
1840
1841 map = block_group->physical_map;
1842
1843 spin_lock(&block_group->lock);
1844 if (block_group->zone_is_active) {
1845 ret = true;
1846 goto out_unlock;
1847 }
1848
1849 /* No space left */
1850 if (btrfs_zoned_bg_is_full(block_group)) {
1851 ret = false;
1852 goto out_unlock;
1853 }
1854
1855 for (i = 0; i < map->num_stripes; i++) {
1856 device = map->stripes[i].dev;
1857 physical = map->stripes[i].physical;
1858
1859 if (device->zone_info->max_active_zones == 0)
1860 continue;
1861
1862 if (!btrfs_dev_set_active_zone(device, physical)) {
1863 /* Cannot activate the zone */
1864 ret = false;
1865 goto out_unlock;
1866 }
1867 }
1868
1869 /* Successfully activated all the zones */
1870 block_group->zone_is_active = 1;
1871 spin_unlock(&block_group->lock);
1872
1873 /* For the active block group list */
1874 btrfs_get_block_group(block_group);
1875
1876 spin_lock(&fs_info->zone_active_bgs_lock);
1877 list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs);
1878 spin_unlock(&fs_info->zone_active_bgs_lock);
1879
1880 return true;
1881
1882 out_unlock:
1883 spin_unlock(&block_group->lock);
1884 return ret;
1885 }
1886
1887 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
1888 {
1889 struct btrfs_fs_info *fs_info = block_group->fs_info;
1890 struct map_lookup *map;
1891 int ret = 0;
1892 int i;
1893
1894 spin_lock(&block_group->lock);
1895 if (!block_group->zone_is_active) {
1896 spin_unlock(&block_group->lock);
1897 return 0;
1898 }
1899
1900 /* Check if we have unwritten allocated space */
1901 if ((block_group->flags &
1902 (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)) &&
1903 block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
1904 spin_unlock(&block_group->lock);
1905 return -EAGAIN;
1906 }
1907
1908 /*
1909 * If we are sure that the block group is full (= no more room left for
1910 * new allocation) and the IO for the last usable block is completed, we
1911 * don't need to wait for the other IOs. This holds because we ensure
1912 * the sequential IO submissions using the ZONE_APPEND command for data
1913 * and block_group->meta_write_pointer for metadata.
1914 */
1915 if (!fully_written) {
1916 spin_unlock(&block_group->lock);
1917
1918 ret = btrfs_inc_block_group_ro(block_group, false);
1919 if (ret)
1920 return ret;
1921
1922 /* Ensure all writes in this block group finish */
1923 btrfs_wait_block_group_reservations(block_group);
1924 /* No need to wait for NOCOW writers. Zoned mode does not allow that */
1925 btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
1926 block_group->length);
1927
1928 spin_lock(&block_group->lock);
1929
1930 /*
1931 * Bail out if someone already deactivated the block group, or
1932 * allocated space is left in the block group.
1933 */
1934 if (!block_group->zone_is_active) {
1935 spin_unlock(&block_group->lock);
1936 btrfs_dec_block_group_ro(block_group);
1937 return 0;
1938 }
1939
1940 if (block_group->reserved) {
1941 spin_unlock(&block_group->lock);
1942 btrfs_dec_block_group_ro(block_group);
1943 return -EAGAIN;
1944 }
1945 }
1946
1947 block_group->zone_is_active = 0;
1948 block_group->alloc_offset = block_group->zone_capacity;
1949 block_group->free_space_ctl->free_space = 0;
1950 btrfs_clear_treelog_bg(block_group);
1951 btrfs_clear_data_reloc_bg(block_group);
1952 spin_unlock(&block_group->lock);
1953
1954 map = block_group->physical_map;
1955 for (i = 0; i < map->num_stripes; i++) {
1956 struct btrfs_device *device = map->stripes[i].dev;
1957 const u64 physical = map->stripes[i].physical;
1958
1959 if (device->zone_info->max_active_zones == 0)
1960 continue;
1961
1962 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH,
1963 physical >> SECTOR_SHIFT,
1964 device->zone_info->zone_size >> SECTOR_SHIFT,
1965 GFP_NOFS);
1966
1967 if (ret)
1968 return ret;
1969
1970 btrfs_dev_clear_active_zone(device, physical);
1971 }
1972
1973 if (!fully_written)
1974 btrfs_dec_block_group_ro(block_group);
1975
1976 spin_lock(&fs_info->zone_active_bgs_lock);
1977 ASSERT(!list_empty(&block_group->active_bg_list));
1978 list_del_init(&block_group->active_bg_list);
1979 spin_unlock(&fs_info->zone_active_bgs_lock);
1980
1981 /* For active_bg_list */
1982 btrfs_put_block_group(block_group);
1983
1984 return 0;
1985 }
1986
1987 int btrfs_zone_finish(struct btrfs_block_group *block_group)
1988 {
1989 if (!btrfs_is_zoned(block_group->fs_info))
1990 return 0;
1991
1992 return do_zone_finish(block_group, false);
1993 }
1994
1995 bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags)
1996 {
1997 struct btrfs_fs_info *fs_info = fs_devices->fs_info;
1998 struct btrfs_device *device;
1999 bool ret = false;
2000
2001 if (!btrfs_is_zoned(fs_info))
2002 return true;
2003
2004 /* Check if there is a device with active zones left */
2005 mutex_lock(&fs_info->chunk_mutex);
2006 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
2007 struct btrfs_zoned_device_info *zinfo = device->zone_info;
2008
2009 if (!device->bdev)
2010 continue;
2011
2012 if (!zinfo->max_active_zones ||
2013 atomic_read(&zinfo->active_zones_left)) {
2014 ret = true;
2015 break;
2016 }
2017 }
2018 mutex_unlock(&fs_info->chunk_mutex);
2019
2020 return ret;
2021 }
2022
2023 void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length)
2024 {
2025 struct btrfs_block_group *block_group;
2026 u64 min_alloc_bytes;
2027
2028 if (!btrfs_is_zoned(fs_info))
2029 return;
2030
2031 block_group = btrfs_lookup_block_group(fs_info, logical);
2032 ASSERT(block_group);
2033
2034 /* No MIXED_BG on zoned btrfs. */
2035 if (block_group->flags & BTRFS_BLOCK_GROUP_DATA)
2036 min_alloc_bytes = fs_info->sectorsize;
2037 else
2038 min_alloc_bytes = fs_info->nodesize;
2039
2040 /* Bail out if we can allocate more data from this block group. */
2041 if (logical + length + min_alloc_bytes <=
2042 block_group->start + block_group->zone_capacity)
2043 goto out;
2044
2045 do_zone_finish(block_group, true);
2046
2047 out:
2048 btrfs_put_block_group(block_group);
2049 }
2050
2051 static void btrfs_zone_finish_endio_workfn(struct work_struct *work)
2052 {
2053 struct btrfs_block_group *bg =
2054 container_of(work, struct btrfs_block_group, zone_finish_work);
2055
2056 wait_on_extent_buffer_writeback(bg->last_eb);
2057 free_extent_buffer(bg->last_eb);
2058 btrfs_zone_finish_endio(bg->fs_info, bg->start, bg->length);
2059 btrfs_put_block_group(bg);
2060 }
2061
2062 void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg,
2063 struct extent_buffer *eb)
2064 {
2065 if (!bg->seq_zone || eb->start + eb->len * 2 <= bg->start + bg->zone_capacity)
2066 return;
2067
2068 if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) {
2069 btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing",
2070 bg->start);
2071 return;
2072 }
2073
2074 /* For the work */
2075 btrfs_get_block_group(bg);
2076 atomic_inc(&eb->refs);
2077 bg->last_eb = eb;
2078 INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn);
2079 queue_work(system_unbound_wq, &bg->zone_finish_work);
2080 }
2081
2082 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
2083 {
2084 struct btrfs_fs_info *fs_info = bg->fs_info;
2085
2086 spin_lock(&fs_info->relocation_bg_lock);
2087 if (fs_info->data_reloc_bg == bg->start)
2088 fs_info->data_reloc_bg = 0;
2089 spin_unlock(&fs_info->relocation_bg_lock);
2090 }
2091
2092 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
2093 {
2094 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2095 struct btrfs_device *device;
2096
2097 if (!btrfs_is_zoned(fs_info))
2098 return;
2099
2100 mutex_lock(&fs_devices->device_list_mutex);
2101 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2102 if (device->zone_info) {
2103 vfree(device->zone_info->zone_cache);
2104 device->zone_info->zone_cache = NULL;
2105 }
2106 }
2107 mutex_unlock(&fs_devices->device_list_mutex);
2108 }
2109
2110 bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info)
2111 {
2112 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2113 struct btrfs_device *device;
2114 u64 used = 0;
2115 u64 total = 0;
2116 u64 factor;
2117
2118 ASSERT(btrfs_is_zoned(fs_info));
2119
2120 if (fs_info->bg_reclaim_threshold == 0)
2121 return false;
2122
2123 mutex_lock(&fs_devices->device_list_mutex);
2124 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2125 if (!device->bdev)
2126 continue;
2127
2128 total += device->disk_total_bytes;
2129 used += device->bytes_used;
2130 }
2131 mutex_unlock(&fs_devices->device_list_mutex);
2132
2133 factor = div64_u64(used * 100, total);
2134 return factor >= fs_info->bg_reclaim_threshold;
2135 }
2136
2137 void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical,
2138 u64 length)
2139 {
2140 struct btrfs_block_group *block_group;
2141
2142 if (!btrfs_is_zoned(fs_info))
2143 return;
2144
2145 block_group = btrfs_lookup_block_group(fs_info, logical);
2146 /* It should be called on a previous data relocation block group. */
2147 ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA));
2148
2149 spin_lock(&block_group->lock);
2150 if (!block_group->zoned_data_reloc_ongoing)
2151 goto out;
2152
2153 /* All relocation extents are written. */
2154 if (block_group->start + block_group->alloc_offset == logical + length) {
2155 /* Now, release this block group for further allocations. */
2156 block_group->zoned_data_reloc_ongoing = 0;
2157 }
2158
2159 out:
2160 spin_unlock(&block_group->lock);
2161 btrfs_put_block_group(block_group);
2162 }
A mirror of Linus' kernel repository