]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/btrfs/zoned.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
objtool: Add entry UNRET validation
[mirror_ubuntu-jammy-kernel.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/vmalloc.h>
8 #include "ctree.h"
9 #include "volumes.h"
10 #include "zoned.h"
11 #include "rcu-string.h"
12 #include "disk-io.h"
13 #include "block-group.h"
14 #include "transaction.h"
15 #include "dev-replace.h"
16 #include "space-info.h"
17
18 /* Maximum number of zones to report per blkdev_report_zones() call */
19 #define BTRFS_REPORT_NR_ZONES 4096
20 /* Invalid allocation pointer value for missing devices */
21 #define WP_MISSING_DEV ((u64)-1)
22 /* Pseudo write pointer value for conventional zone */
23 #define WP_CONVENTIONAL ((u64)-2)
24
25 /*
26 * Location of the first zone of superblock logging zone pairs.
27 *
28 * - primary superblock: 0B (zone 0)
29 * - first copy: 512G (zone starting at that offset)
30 * - second copy: 4T (zone starting at that offset)
31 */
32 #define BTRFS_SB_LOG_PRIMARY_OFFSET (0ULL)
33 #define BTRFS_SB_LOG_FIRST_OFFSET (512ULL * SZ_1G)
34 #define BTRFS_SB_LOG_SECOND_OFFSET (4096ULL * SZ_1G)
35
36 #define BTRFS_SB_LOG_FIRST_SHIFT const_ilog2(BTRFS_SB_LOG_FIRST_OFFSET)
37 #define BTRFS_SB_LOG_SECOND_SHIFT const_ilog2(BTRFS_SB_LOG_SECOND_OFFSET)
38
39 /* Number of superblock log zones */
40 #define BTRFS_NR_SB_LOG_ZONES 2
41
42 /*
43 * Maximum supported zone size. Currently, SMR disks have a zone size of
44 * 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range. We do not
45 * expect the zone size to become larger than 8GiB in the near future.
46 */
47 #define BTRFS_MAX_ZONE_SIZE SZ_8G
48
49 static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data)
50 {
51 struct blk_zone *zones = data;
52
53 memcpy(&zones[idx], zone, sizeof(*zone));
54
55 return 0;
56 }
57
58 static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones,
59 u64 *wp_ret)
60 {
61 bool empty[BTRFS_NR_SB_LOG_ZONES];
62 bool full[BTRFS_NR_SB_LOG_ZONES];
63 sector_t sector;
64
65 ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL &&
66 zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL);
67
68 empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY);
69 empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY);
70 full[0] = (zones[0].cond == BLK_ZONE_COND_FULL);
71 full[1] = (zones[1].cond == BLK_ZONE_COND_FULL);
72
73 /*
74 * Possible states of log buffer zones
75 *
76 * Empty[0] In use[0] Full[0]
77 * Empty[1] * x 0
78 * In use[1] 0 x 0
79 * Full[1] 1 1 C
80 *
81 * Log position:
82 * *: Special case, no superblock is written
83 * 0: Use write pointer of zones[0]
84 * 1: Use write pointer of zones[1]
85 * C: Compare super blocks from zones[0] and zones[1], use the latest
86 * one determined by generation
87 * x: Invalid state
88 */
89
90 if (empty[0] && empty[1]) {
91 /* Special case to distinguish no superblock to read */
92 *wp_ret = zones[0].start << SECTOR_SHIFT;
93 return -ENOENT;
94 } else if (full[0] && full[1]) {
95 /* Compare two super blocks */
96 struct address_space *mapping = bdev->bd_inode->i_mapping;
97 struct page *page[BTRFS_NR_SB_LOG_ZONES];
98 struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES];
99 int i;
100
101 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
102 u64 bytenr;
103
104 bytenr = ((zones[i].start + zones[i].len)
105 << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE;
106
107 page[i] = read_cache_page_gfp(mapping,
108 bytenr >> PAGE_SHIFT, GFP_NOFS);
109 if (IS_ERR(page[i])) {
110 if (i == 1)
111 btrfs_release_disk_super(super[0]);
112 return PTR_ERR(page[i]);
113 }
114 super[i] = page_address(page[i]);
115 }
116
117 if (super[0]->generation > super[1]->generation)
118 sector = zones[1].start;
119 else
120 sector = zones[0].start;
121
122 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++)
123 btrfs_release_disk_super(super[i]);
124 } else if (!full[0] && (empty[1] || full[1])) {
125 sector = zones[0].wp;
126 } else if (full[0]) {
127 sector = zones[1].wp;
128 } else {
129 return -EUCLEAN;
130 }
131 *wp_ret = sector << SECTOR_SHIFT;
132 return 0;
133 }
134
135 /*
136 * Get the first zone number of the superblock mirror
137 */
138 static inline u32 sb_zone_number(int shift, int mirror)
139 {
140 u64 zone;
141
142 ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX);
143 switch (mirror) {
144 case 0: zone = 0; break;
145 case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break;
146 case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break;
147 }
148
149 ASSERT(zone <= U32_MAX);
150
151 return (u32)zone;
152 }
153
154 static inline sector_t zone_start_sector(u32 zone_number,
155 struct block_device *bdev)
156 {
157 return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev));
158 }
159
160 static inline u64 zone_start_physical(u32 zone_number,
161 struct btrfs_zoned_device_info *zone_info)
162 {
163 return (u64)zone_number << zone_info->zone_size_shift;
164 }
165
166 /*
167 * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block
168 * device into static sized chunks and fake a conventional zone on each of
169 * them.
170 */
171 static int emulate_report_zones(struct btrfs_device *device, u64 pos,
172 struct blk_zone *zones, unsigned int nr_zones)
173 {
174 const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT;
175 sector_t bdev_size = bdev_nr_sectors(device->bdev);
176 unsigned int i;
177
178 pos >>= SECTOR_SHIFT;
179 for (i = 0; i < nr_zones; i++) {
180 zones[i].start = i * zone_sectors + pos;
181 zones[i].len = zone_sectors;
182 zones[i].capacity = zone_sectors;
183 zones[i].wp = zones[i].start + zone_sectors;
184 zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL;
185 zones[i].cond = BLK_ZONE_COND_NOT_WP;
186
187 if (zones[i].wp >= bdev_size) {
188 i++;
189 break;
190 }
191 }
192
193 return i;
194 }
195
196 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
197 struct blk_zone *zones, unsigned int *nr_zones)
198 {
199 struct btrfs_zoned_device_info *zinfo = device->zone_info;
200 u32 zno;
201 int ret;
202
203 if (!*nr_zones)
204 return 0;
205
206 if (!bdev_is_zoned(device->bdev)) {
207 ret = emulate_report_zones(device, pos, zones, *nr_zones);
208 *nr_zones = ret;
209 return 0;
210 }
211
212 /* Check cache */
213 if (zinfo->zone_cache) {
214 unsigned int i;
215
216 ASSERT(IS_ALIGNED(pos, zinfo->zone_size));
217 zno = pos >> zinfo->zone_size_shift;
218 /*
219 * We cannot report zones beyond the zone end. So, it is OK to
220 * cap *nr_zones to at the end.
221 */
222 *nr_zones = min_t(u32, *nr_zones, zinfo->nr_zones - zno);
223
224 for (i = 0; i < *nr_zones; i++) {
225 struct blk_zone *zone_info;
226
227 zone_info = &zinfo->zone_cache[zno + i];
228 if (!zone_info->len)
229 break;
230 }
231
232 if (i == *nr_zones) {
233 /* Cache hit on all the zones */
234 memcpy(zones, zinfo->zone_cache + zno,
235 sizeof(*zinfo->zone_cache) * *nr_zones);
236 return 0;
237 }
238 }
239
240 ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
241 copy_zone_info_cb, zones);
242 if (ret < 0) {
243 btrfs_err_in_rcu(device->fs_info,
244 "zoned: failed to read zone %llu on %s (devid %llu)",
245 pos, rcu_str_deref(device->name),
246 device->devid);
247 return ret;
248 }
249 *nr_zones = ret;
250 if (!ret)
251 return -EIO;
252
253 /* Populate cache */
254 if (zinfo->zone_cache)
255 memcpy(zinfo->zone_cache + zno, zones,
256 sizeof(*zinfo->zone_cache) * *nr_zones);
257
258 return 0;
259 }
260
261 /* The emulated zone size is determined from the size of device extent */
262 static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info)
263 {
264 struct btrfs_path *path;
265 struct btrfs_root *root = fs_info->dev_root;
266 struct btrfs_key key;
267 struct extent_buffer *leaf;
268 struct btrfs_dev_extent *dext;
269 int ret = 0;
270
271 key.objectid = 1;
272 key.type = BTRFS_DEV_EXTENT_KEY;
273 key.offset = 0;
274
275 path = btrfs_alloc_path();
276 if (!path)
277 return -ENOMEM;
278
279 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
280 if (ret < 0)
281 goto out;
282
283 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
284 ret = btrfs_next_leaf(root, path);
285 if (ret < 0)
286 goto out;
287 /* No dev extents at all? Not good */
288 if (ret > 0) {
289 ret = -EUCLEAN;
290 goto out;
291 }
292 }
293
294 leaf = path->nodes[0];
295 dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
296 fs_info->zone_size = btrfs_dev_extent_length(leaf, dext);
297 ret = 0;
298
299 out:
300 btrfs_free_path(path);
301
302 return ret;
303 }
304
305 int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
306 {
307 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
308 struct btrfs_device *device;
309 int ret = 0;
310
311 /* fs_info->zone_size might not set yet. Use the incomapt flag here. */
312 if (!btrfs_fs_incompat(fs_info, ZONED))
313 return 0;
314
315 mutex_lock(&fs_devices->device_list_mutex);
316 list_for_each_entry(device, &fs_devices->devices, dev_list) {
317 /* We can skip reading of zone info for missing devices */
318 if (!device->bdev)
319 continue;
320
321 ret = btrfs_get_dev_zone_info(device, true);
322 if (ret)
323 break;
324 }
325 mutex_unlock(&fs_devices->device_list_mutex);
326
327 return ret;
328 }
329
330 int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache)
331 {
332 struct btrfs_fs_info *fs_info = device->fs_info;
333 struct btrfs_zoned_device_info *zone_info = NULL;
334 struct block_device *bdev = device->bdev;
335 sector_t nr_sectors;
336 sector_t sector = 0;
337 struct blk_zone *zones = NULL;
338 unsigned int i, nreported = 0, nr_zones;
339 sector_t zone_sectors;
340 char *model, *emulated;
341 int ret;
342
343 /*
344 * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not
345 * yet be set.
346 */
347 if (!btrfs_fs_incompat(fs_info, ZONED))
348 return 0;
349
350 if (device->zone_info)
351 return 0;
352
353 zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL);
354 if (!zone_info)
355 return -ENOMEM;
356
357 device->zone_info = zone_info;
358
359 if (!bdev_is_zoned(bdev)) {
360 if (!fs_info->zone_size) {
361 ret = calculate_emulated_zone_size(fs_info);
362 if (ret)
363 goto out;
364 }
365
366 ASSERT(fs_info->zone_size);
367 zone_sectors = fs_info->zone_size >> SECTOR_SHIFT;
368 } else {
369 zone_sectors = bdev_zone_sectors(bdev);
370 }
371
372 /* Check if it's power of 2 (see is_power_of_2) */
373 ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0);
374 zone_info->zone_size = zone_sectors << SECTOR_SHIFT;
375
376 /* We reject devices with a zone size larger than 8GB */
377 if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) {
378 btrfs_err_in_rcu(fs_info,
379 "zoned: %s: zone size %llu larger than supported maximum %llu",
380 rcu_str_deref(device->name),
381 zone_info->zone_size, BTRFS_MAX_ZONE_SIZE);
382 ret = -EINVAL;
383 goto out;
384 }
385
386 nr_sectors = bdev_nr_sectors(bdev);
387 zone_info->zone_size_shift = ilog2(zone_info->zone_size);
388 zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors);
389 if (!IS_ALIGNED(nr_sectors, zone_sectors))
390 zone_info->nr_zones++;
391
392 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
393 if (!zone_info->seq_zones) {
394 ret = -ENOMEM;
395 goto out;
396 }
397
398 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
399 if (!zone_info->empty_zones) {
400 ret = -ENOMEM;
401 goto out;
402 }
403
404 zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
405 if (!zones) {
406 ret = -ENOMEM;
407 goto out;
408 }
409
410 /*
411 * Enable zone cache only for a zoned device. On a non-zoned device, we
412 * fill the zone info with emulated CONVENTIONAL zones, so no need to
413 * use the cache.
414 */
415 if (populate_cache && bdev_is_zoned(device->bdev)) {
416 zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) *
417 zone_info->nr_zones);
418 if (!zone_info->zone_cache) {
419 btrfs_err_in_rcu(device->fs_info,
420 "zoned: failed to allocate zone cache for %s",
421 rcu_str_deref(device->name));
422 ret = -ENOMEM;
423 goto out;
424 }
425 }
426
427 /* Get zones type */
428 while (sector < nr_sectors) {
429 nr_zones = BTRFS_REPORT_NR_ZONES;
430 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
431 &nr_zones);
432 if (ret)
433 goto out;
434
435 for (i = 0; i < nr_zones; i++) {
436 if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
437 __set_bit(nreported, zone_info->seq_zones);
438 if (zones[i].cond == BLK_ZONE_COND_EMPTY)
439 __set_bit(nreported, zone_info->empty_zones);
440 nreported++;
441 }
442 sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
443 }
444
445 if (nreported != zone_info->nr_zones) {
446 btrfs_err_in_rcu(device->fs_info,
447 "inconsistent number of zones on %s (%u/%u)",
448 rcu_str_deref(device->name), nreported,
449 zone_info->nr_zones);
450 ret = -EIO;
451 goto out;
452 }
453
454 /* Validate superblock log */
455 nr_zones = BTRFS_NR_SB_LOG_ZONES;
456 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
457 u32 sb_zone;
458 u64 sb_wp;
459 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
460
461 sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
462 if (sb_zone + 1 >= zone_info->nr_zones)
463 continue;
464
465 ret = btrfs_get_dev_zones(device,
466 zone_start_physical(sb_zone, zone_info),
467 &zone_info->sb_zones[sb_pos],
468 &nr_zones);
469 if (ret)
470 goto out;
471
472 if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
473 btrfs_err_in_rcu(device->fs_info,
474 "zoned: failed to read super block log zone info at devid %llu zone %u",
475 device->devid, sb_zone);
476 ret = -EUCLEAN;
477 goto out;
478 }
479
480 /*
481 * If zones[0] is conventional, always use the beginning of the
482 * zone to record superblock. No need to validate in that case.
483 */
484 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
485 BLK_ZONE_TYPE_CONVENTIONAL)
486 continue;
487
488 ret = sb_write_pointer(device->bdev,
489 &zone_info->sb_zones[sb_pos], &sb_wp);
490 if (ret != -ENOENT && ret) {
491 btrfs_err_in_rcu(device->fs_info,
492 "zoned: super block log zone corrupted devid %llu zone %u",
493 device->devid, sb_zone);
494 ret = -EUCLEAN;
495 goto out;
496 }
497 }
498
499
500 kfree(zones);
501
502 switch (bdev_zoned_model(bdev)) {
503 case BLK_ZONED_HM:
504 model = "host-managed zoned";
505 emulated = "";
506 break;
507 case BLK_ZONED_HA:
508 model = "host-aware zoned";
509 emulated = "";
510 break;
511 case BLK_ZONED_NONE:
512 model = "regular";
513 emulated = "emulated ";
514 break;
515 default:
516 /* Just in case */
517 btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
518 bdev_zoned_model(bdev),
519 rcu_str_deref(device->name));
520 ret = -EOPNOTSUPP;
521 goto out_free_zone_info;
522 }
523
524 btrfs_info_in_rcu(fs_info,
525 "%s block device %s, %u %szones of %llu bytes",
526 model, rcu_str_deref(device->name), zone_info->nr_zones,
527 emulated, zone_info->zone_size);
528
529 return 0;
530
531 out:
532 kfree(zones);
533 out_free_zone_info:
534 btrfs_destroy_dev_zone_info(device);
535
536 return ret;
537 }
538
539 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
540 {
541 struct btrfs_zoned_device_info *zone_info = device->zone_info;
542
543 if (!zone_info)
544 return;
545
546 bitmap_free(zone_info->seq_zones);
547 bitmap_free(zone_info->empty_zones);
548 vfree(zone_info->zone_cache);
549 kfree(zone_info);
550 device->zone_info = NULL;
551 }
552
553 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
554 struct blk_zone *zone)
555 {
556 unsigned int nr_zones = 1;
557 int ret;
558
559 ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
560 if (ret != 0 || !nr_zones)
561 return ret ? ret : -EIO;
562
563 return 0;
564 }
565
566 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
567 {
568 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
569 struct btrfs_device *device;
570 u64 zoned_devices = 0;
571 u64 nr_devices = 0;
572 u64 zone_size = 0;
573 const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED);
574 int ret = 0;
575
576 /* Count zoned devices */
577 list_for_each_entry(device, &fs_devices->devices, dev_list) {
578 enum blk_zoned_model model;
579
580 if (!device->bdev)
581 continue;
582
583 model = bdev_zoned_model(device->bdev);
584 /*
585 * A Host-Managed zoned device must be used as a zoned device.
586 * A Host-Aware zoned device and a non-zoned devices can be
587 * treated as a zoned device, if ZONED flag is enabled in the
588 * superblock.
589 */
590 if (model == BLK_ZONED_HM ||
591 (model == BLK_ZONED_HA && incompat_zoned) ||
592 (model == BLK_ZONED_NONE && incompat_zoned)) {
593 struct btrfs_zoned_device_info *zone_info =
594 device->zone_info;
595
596 zone_info = device->zone_info;
597 zoned_devices++;
598 if (!zone_size) {
599 zone_size = zone_info->zone_size;
600 } else if (zone_info->zone_size != zone_size) {
601 btrfs_err(fs_info,
602 "zoned: unequal block device zone sizes: have %llu found %llu",
603 device->zone_info->zone_size,
604 zone_size);
605 ret = -EINVAL;
606 goto out;
607 }
608 }
609 nr_devices++;
610 }
611
612 if (!zoned_devices && !incompat_zoned)
613 goto out;
614
615 if (!zoned_devices && incompat_zoned) {
616 /* No zoned block device found on ZONED filesystem */
617 btrfs_err(fs_info,
618 "zoned: no zoned devices found on a zoned filesystem");
619 ret = -EINVAL;
620 goto out;
621 }
622
623 if (zoned_devices && !incompat_zoned) {
624 btrfs_err(fs_info,
625 "zoned: mode not enabled but zoned device found");
626 ret = -EINVAL;
627 goto out;
628 }
629
630 if (zoned_devices != nr_devices) {
631 btrfs_err(fs_info,
632 "zoned: cannot mix zoned and regular devices");
633 ret = -EINVAL;
634 goto out;
635 }
636
637 /*
638 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
639 * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size,
640 * check the alignment here.
641 */
642 if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
643 btrfs_err(fs_info,
644 "zoned: zone size %llu not aligned to stripe %u",
645 zone_size, BTRFS_STRIPE_LEN);
646 ret = -EINVAL;
647 goto out;
648 }
649
650 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
651 btrfs_err(fs_info, "zoned: mixed block groups not supported");
652 ret = -EINVAL;
653 goto out;
654 }
655
656 fs_info->zone_size = zone_size;
657 fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
658
659 /*
660 * Check mount options here, because we might change fs_info->zoned
661 * from fs_info->zone_size.
662 */
663 ret = btrfs_check_mountopts_zoned(fs_info);
664 if (ret)
665 goto out;
666
667 btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
668 out:
669 return ret;
670 }
671
672 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
673 {
674 if (!btrfs_is_zoned(info))
675 return 0;
676
677 /*
678 * Space cache writing is not COWed. Disable that to avoid write errors
679 * in sequential zones.
680 */
681 if (btrfs_test_opt(info, SPACE_CACHE)) {
682 btrfs_err(info, "zoned: space cache v1 is not supported");
683 return -EINVAL;
684 }
685
686 if (btrfs_test_opt(info, NODATACOW)) {
687 btrfs_err(info, "zoned: NODATACOW not supported");
688 return -EINVAL;
689 }
690
691 return 0;
692 }
693
694 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
695 int rw, u64 *bytenr_ret)
696 {
697 u64 wp;
698 int ret;
699
700 if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
701 *bytenr_ret = zones[0].start << SECTOR_SHIFT;
702 return 0;
703 }
704
705 ret = sb_write_pointer(bdev, zones, &wp);
706 if (ret != -ENOENT && ret < 0)
707 return ret;
708
709 if (rw == WRITE) {
710 struct blk_zone *reset = NULL;
711
712 if (wp == zones[0].start << SECTOR_SHIFT)
713 reset = &zones[0];
714 else if (wp == zones[1].start << SECTOR_SHIFT)
715 reset = &zones[1];
716
717 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
718 ASSERT(reset->cond == BLK_ZONE_COND_FULL);
719
720 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
721 reset->start, reset->len,
722 GFP_NOFS);
723 if (ret)
724 return ret;
725
726 reset->cond = BLK_ZONE_COND_EMPTY;
727 reset->wp = reset->start;
728 }
729 } else if (ret != -ENOENT) {
730 /* For READ, we want the precious one */
731 if (wp == zones[0].start << SECTOR_SHIFT)
732 wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT;
733 wp -= BTRFS_SUPER_INFO_SIZE;
734 }
735
736 *bytenr_ret = wp;
737 return 0;
738
739 }
740
741 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
742 u64 *bytenr_ret)
743 {
744 struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
745 sector_t zone_sectors;
746 u32 sb_zone;
747 int ret;
748 u8 zone_sectors_shift;
749 sector_t nr_sectors;
750 u32 nr_zones;
751
752 if (!bdev_is_zoned(bdev)) {
753 *bytenr_ret = btrfs_sb_offset(mirror);
754 return 0;
755 }
756
757 ASSERT(rw == READ || rw == WRITE);
758
759 zone_sectors = bdev_zone_sectors(bdev);
760 if (!is_power_of_2(zone_sectors))
761 return -EINVAL;
762 zone_sectors_shift = ilog2(zone_sectors);
763 nr_sectors = bdev_nr_sectors(bdev);
764 nr_zones = nr_sectors >> zone_sectors_shift;
765
766 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
767 if (sb_zone + 1 >= nr_zones)
768 return -ENOENT;
769
770 ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
771 BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
772 zones);
773 if (ret < 0)
774 return ret;
775 if (ret != BTRFS_NR_SB_LOG_ZONES)
776 return -EIO;
777
778 return sb_log_location(bdev, zones, rw, bytenr_ret);
779 }
780
781 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
782 u64 *bytenr_ret)
783 {
784 struct btrfs_zoned_device_info *zinfo = device->zone_info;
785 u32 zone_num;
786
787 /*
788 * For a zoned filesystem on a non-zoned block device, use the same
789 * super block locations as regular filesystem. Doing so, the super
790 * block can always be retrieved and the zoned flag of the volume
791 * detected from the super block information.
792 */
793 if (!bdev_is_zoned(device->bdev)) {
794 *bytenr_ret = btrfs_sb_offset(mirror);
795 return 0;
796 }
797
798 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
799 if (zone_num + 1 >= zinfo->nr_zones)
800 return -ENOENT;
801
802 return sb_log_location(device->bdev,
803 &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
804 rw, bytenr_ret);
805 }
806
807 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
808 int mirror)
809 {
810 u32 zone_num;
811
812 if (!zinfo)
813 return false;
814
815 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
816 if (zone_num + 1 >= zinfo->nr_zones)
817 return false;
818
819 if (!test_bit(zone_num, zinfo->seq_zones))
820 return false;
821
822 return true;
823 }
824
825 void btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
826 {
827 struct btrfs_zoned_device_info *zinfo = device->zone_info;
828 struct blk_zone *zone;
829
830 if (!is_sb_log_zone(zinfo, mirror))
831 return;
832
833 zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
834 if (zone->cond != BLK_ZONE_COND_FULL) {
835 if (zone->cond == BLK_ZONE_COND_EMPTY)
836 zone->cond = BLK_ZONE_COND_IMP_OPEN;
837
838 zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
839
840 if (zone->wp == zone->start + zone->len)
841 zone->cond = BLK_ZONE_COND_FULL;
842
843 return;
844 }
845
846 zone++;
847 ASSERT(zone->cond != BLK_ZONE_COND_FULL);
848 if (zone->cond == BLK_ZONE_COND_EMPTY)
849 zone->cond = BLK_ZONE_COND_IMP_OPEN;
850
851 zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
852
853 if (zone->wp == zone->start + zone->len)
854 zone->cond = BLK_ZONE_COND_FULL;
855 }
856
857 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
858 {
859 sector_t zone_sectors;
860 sector_t nr_sectors;
861 u8 zone_sectors_shift;
862 u32 sb_zone;
863 u32 nr_zones;
864
865 zone_sectors = bdev_zone_sectors(bdev);
866 zone_sectors_shift = ilog2(zone_sectors);
867 nr_sectors = bdev_nr_sectors(bdev);
868 nr_zones = nr_sectors >> zone_sectors_shift;
869
870 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
871 if (sb_zone + 1 >= nr_zones)
872 return -ENOENT;
873
874 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
875 zone_start_sector(sb_zone, bdev),
876 zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
877 }
878
879 /**
880 * btrfs_find_allocatable_zones - find allocatable zones within a given region
881 *
882 * @device: the device to allocate a region on
883 * @hole_start: the position of the hole to allocate the region
884 * @num_bytes: size of wanted region
885 * @hole_end: the end of the hole
886 * @return: position of allocatable zones
887 *
888 * Allocatable region should not contain any superblock locations.
889 */
890 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
891 u64 hole_end, u64 num_bytes)
892 {
893 struct btrfs_zoned_device_info *zinfo = device->zone_info;
894 const u8 shift = zinfo->zone_size_shift;
895 u64 nzones = num_bytes >> shift;
896 u64 pos = hole_start;
897 u64 begin, end;
898 bool have_sb;
899 int i;
900
901 ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
902 ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
903
904 while (pos < hole_end) {
905 begin = pos >> shift;
906 end = begin + nzones;
907
908 if (end > zinfo->nr_zones)
909 return hole_end;
910
911 /* Check if zones in the region are all empty */
912 if (btrfs_dev_is_sequential(device, pos) &&
913 find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
914 pos += zinfo->zone_size;
915 continue;
916 }
917
918 have_sb = false;
919 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
920 u32 sb_zone;
921 u64 sb_pos;
922
923 sb_zone = sb_zone_number(shift, i);
924 if (!(end <= sb_zone ||
925 sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
926 have_sb = true;
927 pos = zone_start_physical(
928 sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
929 break;
930 }
931
932 /* We also need to exclude regular superblock positions */
933 sb_pos = btrfs_sb_offset(i);
934 if (!(pos + num_bytes <= sb_pos ||
935 sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
936 have_sb = true;
937 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
938 zinfo->zone_size);
939 break;
940 }
941 }
942 if (!have_sb)
943 break;
944 }
945
946 return pos;
947 }
948
949 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
950 u64 length, u64 *bytes)
951 {
952 int ret;
953
954 *bytes = 0;
955 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
956 physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
957 GFP_NOFS);
958 if (ret)
959 return ret;
960
961 *bytes = length;
962 while (length) {
963 btrfs_dev_set_zone_empty(device, physical);
964 physical += device->zone_info->zone_size;
965 length -= device->zone_info->zone_size;
966 }
967
968 return 0;
969 }
970
971 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
972 {
973 struct btrfs_zoned_device_info *zinfo = device->zone_info;
974 const u8 shift = zinfo->zone_size_shift;
975 unsigned long begin = start >> shift;
976 unsigned long end = (start + size) >> shift;
977 u64 pos;
978 int ret;
979
980 ASSERT(IS_ALIGNED(start, zinfo->zone_size));
981 ASSERT(IS_ALIGNED(size, zinfo->zone_size));
982
983 if (end > zinfo->nr_zones)
984 return -ERANGE;
985
986 /* All the zones are conventional */
987 if (find_next_bit(zinfo->seq_zones, begin, end) == end)
988 return 0;
989
990 /* All the zones are sequential and empty */
991 if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
992 find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
993 return 0;
994
995 for (pos = start; pos < start + size; pos += zinfo->zone_size) {
996 u64 reset_bytes;
997
998 if (!btrfs_dev_is_sequential(device, pos) ||
999 btrfs_dev_is_empty_zone(device, pos))
1000 continue;
1001
1002 /* Free regions should be empty */
1003 btrfs_warn_in_rcu(
1004 device->fs_info,
1005 "zoned: resetting device %s (devid %llu) zone %llu for allocation",
1006 rcu_str_deref(device->name), device->devid, pos >> shift);
1007 WARN_ON_ONCE(1);
1008
1009 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1010 &reset_bytes);
1011 if (ret)
1012 return ret;
1013 }
1014
1015 return 0;
1016 }
1017
1018 /*
1019 * Calculate an allocation pointer from the extent allocation information
1020 * for a block group consist of conventional zones. It is pointed to the
1021 * end of the highest addressed extent in the block group as an allocation
1022 * offset.
1023 */
1024 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1025 u64 *offset_ret)
1026 {
1027 struct btrfs_fs_info *fs_info = cache->fs_info;
1028 struct btrfs_root *root = fs_info->extent_root;
1029 struct btrfs_path *path;
1030 struct btrfs_key key;
1031 struct btrfs_key found_key;
1032 int ret;
1033 u64 length;
1034
1035 path = btrfs_alloc_path();
1036 if (!path)
1037 return -ENOMEM;
1038
1039 key.objectid = cache->start + cache->length;
1040 key.type = 0;
1041 key.offset = 0;
1042
1043 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1044 /* We should not find the exact match */
1045 if (!ret)
1046 ret = -EUCLEAN;
1047 if (ret < 0)
1048 goto out;
1049
1050 ret = btrfs_previous_extent_item(root, path, cache->start);
1051 if (ret) {
1052 if (ret == 1) {
1053 ret = 0;
1054 *offset_ret = 0;
1055 }
1056 goto out;
1057 }
1058
1059 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1060
1061 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1062 length = found_key.offset;
1063 else
1064 length = fs_info->nodesize;
1065
1066 if (!(found_key.objectid >= cache->start &&
1067 found_key.objectid + length <= cache->start + cache->length)) {
1068 ret = -EUCLEAN;
1069 goto out;
1070 }
1071 *offset_ret = found_key.objectid + length - cache->start;
1072 ret = 0;
1073
1074 out:
1075 btrfs_free_path(path);
1076 return ret;
1077 }
1078
1079 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1080 {
1081 struct btrfs_fs_info *fs_info = cache->fs_info;
1082 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1083 struct extent_map *em;
1084 struct map_lookup *map;
1085 struct btrfs_device *device;
1086 u64 logical = cache->start;
1087 u64 length = cache->length;
1088 u64 physical = 0;
1089 int ret;
1090 int i;
1091 unsigned int nofs_flag;
1092 u64 *alloc_offsets = NULL;
1093 u64 last_alloc = 0;
1094 u32 num_sequential = 0, num_conventional = 0;
1095
1096 if (!btrfs_is_zoned(fs_info))
1097 return 0;
1098
1099 /* Sanity check */
1100 if (!IS_ALIGNED(length, fs_info->zone_size)) {
1101 btrfs_err(fs_info,
1102 "zoned: block group %llu len %llu unaligned to zone size %llu",
1103 logical, length, fs_info->zone_size);
1104 return -EIO;
1105 }
1106
1107 /* Get the chunk mapping */
1108 read_lock(&em_tree->lock);
1109 em = lookup_extent_mapping(em_tree, logical, length);
1110 read_unlock(&em_tree->lock);
1111
1112 if (!em)
1113 return -EINVAL;
1114
1115 map = em->map_lookup;
1116
1117 alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1118 if (!alloc_offsets) {
1119 free_extent_map(em);
1120 return -ENOMEM;
1121 }
1122
1123 for (i = 0; i < map->num_stripes; i++) {
1124 bool is_sequential;
1125 struct blk_zone zone;
1126 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1127 int dev_replace_is_ongoing = 0;
1128
1129 device = map->stripes[i].dev;
1130 physical = map->stripes[i].physical;
1131
1132 if (device->bdev == NULL) {
1133 alloc_offsets[i] = WP_MISSING_DEV;
1134 continue;
1135 }
1136
1137 is_sequential = btrfs_dev_is_sequential(device, physical);
1138 if (is_sequential)
1139 num_sequential++;
1140 else
1141 num_conventional++;
1142
1143 if (!is_sequential) {
1144 alloc_offsets[i] = WP_CONVENTIONAL;
1145 continue;
1146 }
1147
1148 /*
1149 * This zone will be used for allocation, so mark this zone
1150 * non-empty.
1151 */
1152 btrfs_dev_clear_zone_empty(device, physical);
1153
1154 down_read(&dev_replace->rwsem);
1155 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1156 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1157 btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical);
1158 up_read(&dev_replace->rwsem);
1159
1160 /*
1161 * The group is mapped to a sequential zone. Get the zone write
1162 * pointer to determine the allocation offset within the zone.
1163 */
1164 WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size));
1165 nofs_flag = memalloc_nofs_save();
1166 ret = btrfs_get_dev_zone(device, physical, &zone);
1167 memalloc_nofs_restore(nofs_flag);
1168 if (ret == -EIO || ret == -EOPNOTSUPP) {
1169 ret = 0;
1170 alloc_offsets[i] = WP_MISSING_DEV;
1171 continue;
1172 } else if (ret) {
1173 goto out;
1174 }
1175
1176 if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1177 btrfs_err_in_rcu(fs_info,
1178 "zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1179 zone.start << SECTOR_SHIFT,
1180 rcu_str_deref(device->name), device->devid);
1181 ret = -EIO;
1182 goto out;
1183 }
1184
1185 switch (zone.cond) {
1186 case BLK_ZONE_COND_OFFLINE:
1187 case BLK_ZONE_COND_READONLY:
1188 btrfs_err(fs_info,
1189 "zoned: offline/readonly zone %llu on device %s (devid %llu)",
1190 physical >> device->zone_info->zone_size_shift,
1191 rcu_str_deref(device->name), device->devid);
1192 alloc_offsets[i] = WP_MISSING_DEV;
1193 break;
1194 case BLK_ZONE_COND_EMPTY:
1195 alloc_offsets[i] = 0;
1196 break;
1197 case BLK_ZONE_COND_FULL:
1198 alloc_offsets[i] = fs_info->zone_size;
1199 break;
1200 default:
1201 /* Partially used zone */
1202 alloc_offsets[i] =
1203 ((zone.wp - zone.start) << SECTOR_SHIFT);
1204 break;
1205 }
1206 }
1207
1208 if (num_sequential > 0)
1209 cache->seq_zone = true;
1210
1211 if (num_conventional > 0) {
1212 /*
1213 * Avoid calling calculate_alloc_pointer() for new BG. It
1214 * is no use for new BG. It must be always 0.
1215 *
1216 * Also, we have a lock chain of extent buffer lock ->
1217 * chunk mutex. For new BG, this function is called from
1218 * btrfs_make_block_group() which is already taking the
1219 * chunk mutex. Thus, we cannot call
1220 * calculate_alloc_pointer() which takes extent buffer
1221 * locks to avoid deadlock.
1222 */
1223 if (new) {
1224 cache->alloc_offset = 0;
1225 goto out;
1226 }
1227 ret = calculate_alloc_pointer(cache, &last_alloc);
1228 if (ret || map->num_stripes == num_conventional) {
1229 if (!ret)
1230 cache->alloc_offset = last_alloc;
1231 else
1232 btrfs_err(fs_info,
1233 "zoned: failed to determine allocation offset of bg %llu",
1234 cache->start);
1235 goto out;
1236 }
1237 }
1238
1239 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1240 case 0: /* single */
1241 if (alloc_offsets[0] == WP_MISSING_DEV) {
1242 btrfs_err(fs_info,
1243 "zoned: cannot recover write pointer for zone %llu",
1244 physical);
1245 ret = -EIO;
1246 goto out;
1247 }
1248 cache->alloc_offset = alloc_offsets[0];
1249 break;
1250 case BTRFS_BLOCK_GROUP_DUP:
1251 case BTRFS_BLOCK_GROUP_RAID1:
1252 case BTRFS_BLOCK_GROUP_RAID0:
1253 case BTRFS_BLOCK_GROUP_RAID10:
1254 case BTRFS_BLOCK_GROUP_RAID5:
1255 case BTRFS_BLOCK_GROUP_RAID6:
1256 /* non-single profiles are not supported yet */
1257 default:
1258 btrfs_err(fs_info, "zoned: profile %s not yet supported",
1259 btrfs_bg_type_to_raid_name(map->type));
1260 ret = -EINVAL;
1261 goto out;
1262 }
1263
1264 out:
1265 if (cache->alloc_offset > fs_info->zone_size) {
1266 btrfs_err(fs_info,
1267 "zoned: invalid write pointer %llu in block group %llu",
1268 cache->alloc_offset, cache->start);
1269 ret = -EIO;
1270 }
1271
1272 /* An extent is allocated after the write pointer */
1273 if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1274 btrfs_err(fs_info,
1275 "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1276 logical, last_alloc, cache->alloc_offset);
1277 ret = -EIO;
1278 }
1279
1280 if (!ret)
1281 cache->meta_write_pointer = cache->alloc_offset + cache->start;
1282
1283 kfree(alloc_offsets);
1284 free_extent_map(em);
1285
1286 return ret;
1287 }
1288
1289 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1290 {
1291 u64 unusable, free;
1292
1293 if (!btrfs_is_zoned(cache->fs_info))
1294 return;
1295
1296 WARN_ON(cache->bytes_super != 0);
1297 unusable = cache->alloc_offset - cache->used;
1298 free = cache->length - cache->alloc_offset;
1299
1300 /* We only need ->free_space in ALLOC_SEQ block groups */
1301 cache->last_byte_to_unpin = (u64)-1;
1302 cache->cached = BTRFS_CACHE_FINISHED;
1303 cache->free_space_ctl->free_space = free;
1304 cache->zone_unusable = unusable;
1305
1306 /* Should not have any excluded extents. Just in case, though */
1307 btrfs_free_excluded_extents(cache);
1308 }
1309
1310 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1311 struct extent_buffer *eb)
1312 {
1313 struct btrfs_fs_info *fs_info = eb->fs_info;
1314
1315 if (!btrfs_is_zoned(fs_info) ||
1316 btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1317 !list_empty(&eb->release_list))
1318 return;
1319
1320 set_extent_buffer_dirty(eb);
1321 set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1322 eb->start + eb->len - 1, EXTENT_DIRTY);
1323 memzero_extent_buffer(eb, 0, eb->len);
1324 set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1325
1326 spin_lock(&trans->releasing_ebs_lock);
1327 list_add_tail(&eb->release_list, &trans->releasing_ebs);
1328 spin_unlock(&trans->releasing_ebs_lock);
1329 atomic_inc(&eb->refs);
1330 }
1331
1332 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1333 {
1334 spin_lock(&trans->releasing_ebs_lock);
1335 while (!list_empty(&trans->releasing_ebs)) {
1336 struct extent_buffer *eb;
1337
1338 eb = list_first_entry(&trans->releasing_ebs,
1339 struct extent_buffer, release_list);
1340 list_del_init(&eb->release_list);
1341 free_extent_buffer(eb);
1342 }
1343 spin_unlock(&trans->releasing_ebs_lock);
1344 }
1345
1346 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1347 {
1348 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1349 struct btrfs_block_group *cache;
1350 bool ret = false;
1351
1352 if (!btrfs_is_zoned(fs_info))
1353 return false;
1354
1355 if (!is_data_inode(&inode->vfs_inode))
1356 return false;
1357
1358 /*
1359 * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1360 * extent layout the relocation code has.
1361 * Furthermore we have set aside own block-group from which only the
1362 * relocation "process" can allocate and make sure only one process at a
1363 * time can add pages to an extent that gets relocated, so it's safe to
1364 * use regular REQ_OP_WRITE for this special case.
1365 */
1366 if (btrfs_is_data_reloc_root(inode->root))
1367 return false;
1368
1369 cache = btrfs_lookup_block_group(fs_info, start);
1370 ASSERT(cache);
1371 if (!cache)
1372 return false;
1373
1374 ret = cache->seq_zone;
1375 btrfs_put_block_group(cache);
1376
1377 return ret;
1378 }
1379
1380 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1381 struct bio *bio)
1382 {
1383 struct btrfs_ordered_extent *ordered;
1384 const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1385
1386 if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1387 return;
1388
1389 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1390 if (WARN_ON(!ordered))
1391 return;
1392
1393 ordered->physical = physical;
1394 ordered->bdev = bio->bi_bdev;
1395
1396 btrfs_put_ordered_extent(ordered);
1397 }
1398
1399 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1400 {
1401 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1402 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1403 struct extent_map_tree *em_tree;
1404 struct extent_map *em;
1405 struct btrfs_ordered_sum *sum;
1406 u64 orig_logical = ordered->disk_bytenr;
1407 u64 *logical = NULL;
1408 int nr, stripe_len;
1409
1410 /* Zoned devices should not have partitions. So, we can assume it is 0 */
1411 ASSERT(!bdev_is_partition(ordered->bdev));
1412 if (WARN_ON(!ordered->bdev))
1413 return;
1414
1415 if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev,
1416 ordered->physical, &logical, &nr,
1417 &stripe_len)))
1418 goto out;
1419
1420 WARN_ON(nr != 1);
1421
1422 if (orig_logical == *logical)
1423 goto out;
1424
1425 ordered->disk_bytenr = *logical;
1426
1427 em_tree = &inode->extent_tree;
1428 write_lock(&em_tree->lock);
1429 em = search_extent_mapping(em_tree, ordered->file_offset,
1430 ordered->num_bytes);
1431 em->block_start = *logical;
1432 free_extent_map(em);
1433 write_unlock(&em_tree->lock);
1434
1435 list_for_each_entry(sum, &ordered->list, list) {
1436 if (*logical < orig_logical)
1437 sum->bytenr -= orig_logical - *logical;
1438 else
1439 sum->bytenr += *logical - orig_logical;
1440 }
1441
1442 out:
1443 kfree(logical);
1444 }
1445
1446 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1447 struct extent_buffer *eb,
1448 struct btrfs_block_group **cache_ret)
1449 {
1450 struct btrfs_block_group *cache;
1451 bool ret = true;
1452
1453 if (!btrfs_is_zoned(fs_info))
1454 return true;
1455
1456 cache = *cache_ret;
1457
1458 if (cache && (eb->start < cache->start ||
1459 cache->start + cache->length <= eb->start)) {
1460 btrfs_put_block_group(cache);
1461 cache = NULL;
1462 *cache_ret = NULL;
1463 }
1464
1465 if (!cache)
1466 cache = btrfs_lookup_block_group(fs_info, eb->start);
1467
1468 if (cache) {
1469 if (cache->meta_write_pointer != eb->start) {
1470 btrfs_put_block_group(cache);
1471 cache = NULL;
1472 ret = false;
1473 } else {
1474 cache->meta_write_pointer = eb->start + eb->len;
1475 }
1476
1477 *cache_ret = cache;
1478 }
1479
1480 return ret;
1481 }
1482
1483 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1484 struct extent_buffer *eb)
1485 {
1486 if (!btrfs_is_zoned(eb->fs_info) || !cache)
1487 return;
1488
1489 ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1490 cache->meta_write_pointer = eb->start;
1491 }
1492
1493 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1494 {
1495 if (!btrfs_dev_is_sequential(device, physical))
1496 return -EOPNOTSUPP;
1497
1498 return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1499 length >> SECTOR_SHIFT, GFP_NOFS, 0);
1500 }
1501
1502 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1503 struct blk_zone *zone)
1504 {
1505 struct btrfs_bio *bbio = NULL;
1506 u64 mapped_length = PAGE_SIZE;
1507 unsigned int nofs_flag;
1508 int nmirrors;
1509 int i, ret;
1510
1511 ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1512 &mapped_length, &bbio);
1513 if (ret || !bbio || mapped_length < PAGE_SIZE) {
1514 btrfs_put_bbio(bbio);
1515 return -EIO;
1516 }
1517
1518 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1519 return -EINVAL;
1520
1521 nofs_flag = memalloc_nofs_save();
1522 nmirrors = (int)bbio->num_stripes;
1523 for (i = 0; i < nmirrors; i++) {
1524 u64 physical = bbio->stripes[i].physical;
1525 struct btrfs_device *dev = bbio->stripes[i].dev;
1526
1527 /* Missing device */
1528 if (!dev->bdev)
1529 continue;
1530
1531 ret = btrfs_get_dev_zone(dev, physical, zone);
1532 /* Failing device */
1533 if (ret == -EIO || ret == -EOPNOTSUPP)
1534 continue;
1535 break;
1536 }
1537 memalloc_nofs_restore(nofs_flag);
1538
1539 return ret;
1540 }
1541
1542 /*
1543 * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1544 * filling zeros between @physical_pos to a write pointer of dev-replace
1545 * source device.
1546 */
1547 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1548 u64 physical_start, u64 physical_pos)
1549 {
1550 struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1551 struct blk_zone zone;
1552 u64 length;
1553 u64 wp;
1554 int ret;
1555
1556 if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1557 return 0;
1558
1559 ret = read_zone_info(fs_info, logical, &zone);
1560 if (ret)
1561 return ret;
1562
1563 wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1564
1565 if (physical_pos == wp)
1566 return 0;
1567
1568 if (physical_pos > wp)
1569 return -EUCLEAN;
1570
1571 length = wp - physical_pos;
1572 return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1573 }
1574
1575 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1576 u64 logical, u64 length)
1577 {
1578 struct btrfs_device *device;
1579 struct extent_map *em;
1580 struct map_lookup *map;
1581
1582 em = btrfs_get_chunk_map(fs_info, logical, length);
1583 if (IS_ERR(em))
1584 return ERR_CAST(em);
1585
1586 map = em->map_lookup;
1587 /* We only support single profile for now */
1588 ASSERT(map->num_stripes == 1);
1589 device = map->stripes[0].dev;
1590
1591 free_extent_map(em);
1592
1593 return device;
1594 }
1595
1596 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
1597 {
1598 struct btrfs_fs_info *fs_info = bg->fs_info;
1599
1600 spin_lock(&fs_info->relocation_bg_lock);
1601 if (fs_info->data_reloc_bg == bg->start)
1602 fs_info->data_reloc_bg = 0;
1603 spin_unlock(&fs_info->relocation_bg_lock);
1604 }
1605
1606 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
1607 {
1608 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1609 struct btrfs_device *device;
1610
1611 if (!btrfs_is_zoned(fs_info))
1612 return;
1613
1614 mutex_lock(&fs_devices->device_list_mutex);
1615 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1616 if (device->zone_info) {
1617 vfree(device->zone_info->zone_cache);
1618 device->zone_info->zone_cache = NULL;
1619 }
1620 }
1621 mutex_unlock(&fs_devices->device_list_mutex);
1622 }
Ubuntu Jammy Linux kernel mirror