Commit | Line | Data |
---|---|---|
5b316468 NA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
1cd6121f | 3 | #include <linux/bitops.h> |
5b316468 NA |
4 | #include <linux/slab.h> |
5 | #include <linux/blkdev.h> | |
08e11a3d | 6 | #include <linux/sched/mm.h> |
ea6f8ddc | 7 | #include <linux/atomic.h> |
16beac87 | 8 | #include <linux/vmalloc.h> |
5b316468 NA |
9 | #include "ctree.h" |
10 | #include "volumes.h" | |
11 | #include "zoned.h" | |
12 | #include "rcu-string.h" | |
1cd6121f | 13 | #include "disk-io.h" |
08e11a3d | 14 | #include "block-group.h" |
d3575156 | 15 | #include "transaction.h" |
6143c23c | 16 | #include "dev-replace.h" |
7db1c5d1 | 17 | #include "space-info.h" |
5b316468 NA |
18 | |
19 | /* Maximum number of zones to report per blkdev_report_zones() call */ | |
20 | #define BTRFS_REPORT_NR_ZONES 4096 | |
08e11a3d NA |
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) | |
5b316468 | 25 | |
53b74fa9 NA |
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 | ||
12659251 NA |
40 | /* Number of superblock log zones */ |
41 | #define BTRFS_NR_SB_LOG_ZONES 2 | |
42 | ||
ea6f8ddc NA |
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 | ||
53b74fa9 NA |
53 | /* |
54 | * Maximum supported zone size. Currently, SMR disks have a zone size of | |
55 | * 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range. We do not | |
56 | * expect the zone size to become larger than 8GiB in the near future. | |
57 | */ | |
58 | #define BTRFS_MAX_ZONE_SIZE SZ_8G | |
59 | ||
5daaf552 NA |
60 | #define SUPER_INFO_SECTORS ((u64)BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT) |
61 | ||
62 | static inline bool sb_zone_is_full(const struct blk_zone *zone) | |
63 | { | |
64 | return (zone->cond == BLK_ZONE_COND_FULL) || | |
65 | (zone->wp + SUPER_INFO_SECTORS > zone->start + zone->capacity); | |
66 | } | |
67 | ||
5b316468 NA |
68 | static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data) |
69 | { | |
70 | struct blk_zone *zones = data; | |
71 | ||
72 | memcpy(&zones[idx], zone, sizeof(*zone)); | |
73 | ||
74 | return 0; | |
75 | } | |
76 | ||
12659251 NA |
77 | static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones, |
78 | u64 *wp_ret) | |
79 | { | |
80 | bool empty[BTRFS_NR_SB_LOG_ZONES]; | |
81 | bool full[BTRFS_NR_SB_LOG_ZONES]; | |
82 | sector_t sector; | |
5daaf552 | 83 | int i; |
12659251 | 84 | |
5daaf552 NA |
85 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { |
86 | ASSERT(zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL); | |
87 | empty[i] = (zones[i].cond == BLK_ZONE_COND_EMPTY); | |
88 | full[i] = sb_zone_is_full(&zones[i]); | |
89 | } | |
12659251 NA |
90 | |
91 | /* | |
92 | * Possible states of log buffer zones | |
93 | * | |
94 | * Empty[0] In use[0] Full[0] | |
95 | * Empty[1] * x 0 | |
96 | * In use[1] 0 x 0 | |
97 | * Full[1] 1 1 C | |
98 | * | |
99 | * Log position: | |
100 | * *: Special case, no superblock is written | |
101 | * 0: Use write pointer of zones[0] | |
102 | * 1: Use write pointer of zones[1] | |
1a9fd417 | 103 | * C: Compare super blocks from zones[0] and zones[1], use the latest |
12659251 NA |
104 | * one determined by generation |
105 | * x: Invalid state | |
106 | */ | |
107 | ||
108 | if (empty[0] && empty[1]) { | |
109 | /* Special case to distinguish no superblock to read */ | |
110 | *wp_ret = zones[0].start << SECTOR_SHIFT; | |
111 | return -ENOENT; | |
112 | } else if (full[0] && full[1]) { | |
113 | /* Compare two super blocks */ | |
114 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
115 | struct page *page[BTRFS_NR_SB_LOG_ZONES]; | |
116 | struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES]; | |
117 | int i; | |
118 | ||
119 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { | |
120 | u64 bytenr; | |
121 | ||
122 | bytenr = ((zones[i].start + zones[i].len) | |
123 | << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE; | |
124 | ||
125 | page[i] = read_cache_page_gfp(mapping, | |
126 | bytenr >> PAGE_SHIFT, GFP_NOFS); | |
127 | if (IS_ERR(page[i])) { | |
128 | if (i == 1) | |
129 | btrfs_release_disk_super(super[0]); | |
130 | return PTR_ERR(page[i]); | |
131 | } | |
132 | super[i] = page_address(page[i]); | |
133 | } | |
134 | ||
135 | if (super[0]->generation > super[1]->generation) | |
136 | sector = zones[1].start; | |
137 | else | |
138 | sector = zones[0].start; | |
139 | ||
140 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) | |
141 | btrfs_release_disk_super(super[i]); | |
142 | } else if (!full[0] && (empty[1] || full[1])) { | |
143 | sector = zones[0].wp; | |
144 | } else if (full[0]) { | |
145 | sector = zones[1].wp; | |
146 | } else { | |
147 | return -EUCLEAN; | |
148 | } | |
149 | *wp_ret = sector << SECTOR_SHIFT; | |
150 | return 0; | |
151 | } | |
152 | ||
153 | /* | |
53b74fa9 | 154 | * Get the first zone number of the superblock mirror |
12659251 NA |
155 | */ |
156 | static inline u32 sb_zone_number(int shift, int mirror) | |
157 | { | |
53b74fa9 | 158 | u64 zone; |
12659251 | 159 | |
53b74fa9 | 160 | ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX); |
12659251 | 161 | switch (mirror) { |
53b74fa9 NA |
162 | case 0: zone = 0; break; |
163 | case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break; | |
164 | case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break; | |
12659251 NA |
165 | } |
166 | ||
53b74fa9 NA |
167 | ASSERT(zone <= U32_MAX); |
168 | ||
169 | return (u32)zone; | |
12659251 NA |
170 | } |
171 | ||
5b434df8 NA |
172 | static inline sector_t zone_start_sector(u32 zone_number, |
173 | struct block_device *bdev) | |
174 | { | |
175 | return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev)); | |
176 | } | |
177 | ||
178 | static inline u64 zone_start_physical(u32 zone_number, | |
179 | struct btrfs_zoned_device_info *zone_info) | |
180 | { | |
181 | return (u64)zone_number << zone_info->zone_size_shift; | |
182 | } | |
183 | ||
3c9daa09 JT |
184 | /* |
185 | * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block | |
186 | * device into static sized chunks and fake a conventional zone on each of | |
187 | * them. | |
188 | */ | |
189 | static int emulate_report_zones(struct btrfs_device *device, u64 pos, | |
190 | struct blk_zone *zones, unsigned int nr_zones) | |
191 | { | |
192 | const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT; | |
193 | sector_t bdev_size = bdev_nr_sectors(device->bdev); | |
194 | unsigned int i; | |
195 | ||
196 | pos >>= SECTOR_SHIFT; | |
197 | for (i = 0; i < nr_zones; i++) { | |
198 | zones[i].start = i * zone_sectors + pos; | |
199 | zones[i].len = zone_sectors; | |
200 | zones[i].capacity = zone_sectors; | |
201 | zones[i].wp = zones[i].start + zone_sectors; | |
202 | zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL; | |
203 | zones[i].cond = BLK_ZONE_COND_NOT_WP; | |
204 | ||
205 | if (zones[i].wp >= bdev_size) { | |
206 | i++; | |
207 | break; | |
208 | } | |
209 | } | |
210 | ||
211 | return i; | |
212 | } | |
213 | ||
5b316468 NA |
214 | static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos, |
215 | struct blk_zone *zones, unsigned int *nr_zones) | |
216 | { | |
16beac87 NA |
217 | struct btrfs_zoned_device_info *zinfo = device->zone_info; |
218 | u32 zno; | |
5b316468 NA |
219 | int ret; |
220 | ||
221 | if (!*nr_zones) | |
222 | return 0; | |
223 | ||
3c9daa09 JT |
224 | if (!bdev_is_zoned(device->bdev)) { |
225 | ret = emulate_report_zones(device, pos, zones, *nr_zones); | |
226 | *nr_zones = ret; | |
227 | return 0; | |
228 | } | |
229 | ||
16beac87 NA |
230 | /* Check cache */ |
231 | if (zinfo->zone_cache) { | |
232 | unsigned int i; | |
233 | ||
234 | ASSERT(IS_ALIGNED(pos, zinfo->zone_size)); | |
235 | zno = pos >> zinfo->zone_size_shift; | |
236 | /* | |
237 | * We cannot report zones beyond the zone end. So, it is OK to | |
238 | * cap *nr_zones to at the end. | |
239 | */ | |
240 | *nr_zones = min_t(u32, *nr_zones, zinfo->nr_zones - zno); | |
241 | ||
242 | for (i = 0; i < *nr_zones; i++) { | |
243 | struct blk_zone *zone_info; | |
244 | ||
245 | zone_info = &zinfo->zone_cache[zno + i]; | |
246 | if (!zone_info->len) | |
247 | break; | |
248 | } | |
249 | ||
250 | if (i == *nr_zones) { | |
251 | /* Cache hit on all the zones */ | |
252 | memcpy(zones, zinfo->zone_cache + zno, | |
253 | sizeof(*zinfo->zone_cache) * *nr_zones); | |
254 | return 0; | |
255 | } | |
256 | } | |
257 | ||
5b316468 NA |
258 | ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones, |
259 | copy_zone_info_cb, zones); | |
260 | if (ret < 0) { | |
261 | btrfs_err_in_rcu(device->fs_info, | |
262 | "zoned: failed to read zone %llu on %s (devid %llu)", | |
263 | pos, rcu_str_deref(device->name), | |
264 | device->devid); | |
265 | return ret; | |
266 | } | |
267 | *nr_zones = ret; | |
268 | if (!ret) | |
269 | return -EIO; | |
270 | ||
16beac87 NA |
271 | /* Populate cache */ |
272 | if (zinfo->zone_cache) | |
273 | memcpy(zinfo->zone_cache + zno, zones, | |
274 | sizeof(*zinfo->zone_cache) * *nr_zones); | |
275 | ||
5b316468 NA |
276 | return 0; |
277 | } | |
278 | ||
3c9daa09 JT |
279 | /* The emulated zone size is determined from the size of device extent */ |
280 | static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info) | |
281 | { | |
282 | struct btrfs_path *path; | |
283 | struct btrfs_root *root = fs_info->dev_root; | |
284 | struct btrfs_key key; | |
285 | struct extent_buffer *leaf; | |
286 | struct btrfs_dev_extent *dext; | |
287 | int ret = 0; | |
288 | ||
289 | key.objectid = 1; | |
290 | key.type = BTRFS_DEV_EXTENT_KEY; | |
291 | key.offset = 0; | |
292 | ||
293 | path = btrfs_alloc_path(); | |
294 | if (!path) | |
295 | return -ENOMEM; | |
296 | ||
297 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
298 | if (ret < 0) | |
299 | goto out; | |
300 | ||
301 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
ad9a9378 | 302 | ret = btrfs_next_leaf(root, path); |
3c9daa09 JT |
303 | if (ret < 0) |
304 | goto out; | |
305 | /* No dev extents at all? Not good */ | |
306 | if (ret > 0) { | |
307 | ret = -EUCLEAN; | |
308 | goto out; | |
309 | } | |
310 | } | |
311 | ||
312 | leaf = path->nodes[0]; | |
313 | dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); | |
314 | fs_info->zone_size = btrfs_dev_extent_length(leaf, dext); | |
315 | ret = 0; | |
316 | ||
317 | out: | |
318 | btrfs_free_path(path); | |
319 | ||
320 | return ret; | |
321 | } | |
322 | ||
73651042 NA |
323 | int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info) |
324 | { | |
325 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
326 | struct btrfs_device *device; | |
327 | int ret = 0; | |
328 | ||
329 | /* fs_info->zone_size might not set yet. Use the incomapt flag here. */ | |
330 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
331 | return 0; | |
332 | ||
333 | mutex_lock(&fs_devices->device_list_mutex); | |
334 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
335 | /* We can skip reading of zone info for missing devices */ | |
336 | if (!device->bdev) | |
337 | continue; | |
338 | ||
16beac87 | 339 | ret = btrfs_get_dev_zone_info(device, true); |
73651042 NA |
340 | if (ret) |
341 | break; | |
342 | } | |
343 | mutex_unlock(&fs_devices->device_list_mutex); | |
344 | ||
345 | return ret; | |
346 | } | |
347 | ||
16beac87 | 348 | int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache) |
5b316468 | 349 | { |
3c9daa09 | 350 | struct btrfs_fs_info *fs_info = device->fs_info; |
5b316468 NA |
351 | struct btrfs_zoned_device_info *zone_info = NULL; |
352 | struct block_device *bdev = device->bdev; | |
ea6f8ddc NA |
353 | struct request_queue *queue = bdev_get_queue(bdev); |
354 | unsigned int max_active_zones; | |
355 | unsigned int nactive; | |
5b316468 NA |
356 | sector_t nr_sectors; |
357 | sector_t sector = 0; | |
358 | struct blk_zone *zones = NULL; | |
359 | unsigned int i, nreported = 0, nr_zones; | |
d734492a | 360 | sector_t zone_sectors; |
3c9daa09 | 361 | char *model, *emulated; |
5b316468 NA |
362 | int ret; |
363 | ||
3c9daa09 JT |
364 | /* |
365 | * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not | |
366 | * yet be set. | |
367 | */ | |
368 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
5b316468 NA |
369 | return 0; |
370 | ||
371 | if (device->zone_info) | |
372 | return 0; | |
373 | ||
374 | zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL); | |
375 | if (!zone_info) | |
376 | return -ENOMEM; | |
377 | ||
16beac87 NA |
378 | device->zone_info = zone_info; |
379 | ||
3c9daa09 JT |
380 | if (!bdev_is_zoned(bdev)) { |
381 | if (!fs_info->zone_size) { | |
382 | ret = calculate_emulated_zone_size(fs_info); | |
383 | if (ret) | |
384 | goto out; | |
385 | } | |
386 | ||
387 | ASSERT(fs_info->zone_size); | |
388 | zone_sectors = fs_info->zone_size >> SECTOR_SHIFT; | |
389 | } else { | |
390 | zone_sectors = bdev_zone_sectors(bdev); | |
391 | } | |
392 | ||
5b316468 NA |
393 | /* Check if it's power of 2 (see is_power_of_2) */ |
394 | ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0); | |
395 | zone_info->zone_size = zone_sectors << SECTOR_SHIFT; | |
53b74fa9 NA |
396 | |
397 | /* We reject devices with a zone size larger than 8GB */ | |
398 | if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) { | |
399 | btrfs_err_in_rcu(fs_info, | |
400 | "zoned: %s: zone size %llu larger than supported maximum %llu", | |
401 | rcu_str_deref(device->name), | |
402 | zone_info->zone_size, BTRFS_MAX_ZONE_SIZE); | |
403 | ret = -EINVAL; | |
404 | goto out; | |
405 | } | |
406 | ||
407 | nr_sectors = bdev_nr_sectors(bdev); | |
5b316468 NA |
408 | zone_info->zone_size_shift = ilog2(zone_info->zone_size); |
409 | zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors); | |
410 | if (!IS_ALIGNED(nr_sectors, zone_sectors)) | |
411 | zone_info->nr_zones++; | |
412 | ||
ea6f8ddc NA |
413 | max_active_zones = queue_max_active_zones(queue); |
414 | if (max_active_zones && max_active_zones < BTRFS_MIN_ACTIVE_ZONES) { | |
415 | btrfs_err_in_rcu(fs_info, | |
416 | "zoned: %s: max active zones %u is too small, need at least %u active zones", | |
417 | rcu_str_deref(device->name), max_active_zones, | |
418 | BTRFS_MIN_ACTIVE_ZONES); | |
419 | ret = -EINVAL; | |
420 | goto out; | |
421 | } | |
422 | zone_info->max_active_zones = max_active_zones; | |
423 | ||
5b316468 NA |
424 | zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); |
425 | if (!zone_info->seq_zones) { | |
426 | ret = -ENOMEM; | |
427 | goto out; | |
428 | } | |
429 | ||
430 | zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
431 | if (!zone_info->empty_zones) { | |
432 | ret = -ENOMEM; | |
433 | goto out; | |
434 | } | |
435 | ||
ea6f8ddc NA |
436 | zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); |
437 | if (!zone_info->active_zones) { | |
438 | ret = -ENOMEM; | |
439 | goto out; | |
440 | } | |
441 | ||
5b316468 NA |
442 | zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL); |
443 | if (!zones) { | |
444 | ret = -ENOMEM; | |
445 | goto out; | |
446 | } | |
447 | ||
16beac87 NA |
448 | /* |
449 | * Enable zone cache only for a zoned device. On a non-zoned device, we | |
450 | * fill the zone info with emulated CONVENTIONAL zones, so no need to | |
451 | * use the cache. | |
452 | */ | |
453 | if (populate_cache && bdev_is_zoned(device->bdev)) { | |
454 | zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) * | |
455 | zone_info->nr_zones); | |
456 | if (!zone_info->zone_cache) { | |
457 | btrfs_err_in_rcu(device->fs_info, | |
458 | "zoned: failed to allocate zone cache for %s", | |
459 | rcu_str_deref(device->name)); | |
460 | ret = -ENOMEM; | |
461 | goto out; | |
462 | } | |
463 | } | |
464 | ||
5b316468 | 465 | /* Get zones type */ |
ea6f8ddc | 466 | nactive = 0; |
5b316468 NA |
467 | while (sector < nr_sectors) { |
468 | nr_zones = BTRFS_REPORT_NR_ZONES; | |
469 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones, | |
470 | &nr_zones); | |
471 | if (ret) | |
472 | goto out; | |
473 | ||
474 | for (i = 0; i < nr_zones; i++) { | |
475 | if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ) | |
476 | __set_bit(nreported, zone_info->seq_zones); | |
ea6f8ddc NA |
477 | switch (zones[i].cond) { |
478 | case BLK_ZONE_COND_EMPTY: | |
5b316468 | 479 | __set_bit(nreported, zone_info->empty_zones); |
ea6f8ddc NA |
480 | break; |
481 | case BLK_ZONE_COND_IMP_OPEN: | |
482 | case BLK_ZONE_COND_EXP_OPEN: | |
483 | case BLK_ZONE_COND_CLOSED: | |
484 | __set_bit(nreported, zone_info->active_zones); | |
485 | nactive++; | |
486 | break; | |
487 | } | |
5b316468 NA |
488 | nreported++; |
489 | } | |
490 | sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len; | |
491 | } | |
492 | ||
493 | if (nreported != zone_info->nr_zones) { | |
494 | btrfs_err_in_rcu(device->fs_info, | |
495 | "inconsistent number of zones on %s (%u/%u)", | |
496 | rcu_str_deref(device->name), nreported, | |
497 | zone_info->nr_zones); | |
498 | ret = -EIO; | |
499 | goto out; | |
500 | } | |
501 | ||
ea6f8ddc NA |
502 | if (max_active_zones) { |
503 | if (nactive > max_active_zones) { | |
504 | btrfs_err_in_rcu(device->fs_info, | |
505 | "zoned: %u active zones on %s exceeds max_active_zones %u", | |
506 | nactive, rcu_str_deref(device->name), | |
507 | max_active_zones); | |
508 | ret = -EIO; | |
509 | goto out; | |
510 | } | |
511 | atomic_set(&zone_info->active_zones_left, | |
512 | max_active_zones - nactive); | |
513 | } | |
514 | ||
12659251 NA |
515 | /* Validate superblock log */ |
516 | nr_zones = BTRFS_NR_SB_LOG_ZONES; | |
517 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
518 | u32 sb_zone; | |
519 | u64 sb_wp; | |
520 | int sb_pos = BTRFS_NR_SB_LOG_ZONES * i; | |
521 | ||
522 | sb_zone = sb_zone_number(zone_info->zone_size_shift, i); | |
523 | if (sb_zone + 1 >= zone_info->nr_zones) | |
524 | continue; | |
525 | ||
5b434df8 NA |
526 | ret = btrfs_get_dev_zones(device, |
527 | zone_start_physical(sb_zone, zone_info), | |
12659251 NA |
528 | &zone_info->sb_zones[sb_pos], |
529 | &nr_zones); | |
530 | if (ret) | |
531 | goto out; | |
532 | ||
533 | if (nr_zones != BTRFS_NR_SB_LOG_ZONES) { | |
534 | btrfs_err_in_rcu(device->fs_info, | |
535 | "zoned: failed to read super block log zone info at devid %llu zone %u", | |
536 | device->devid, sb_zone); | |
537 | ret = -EUCLEAN; | |
538 | goto out; | |
539 | } | |
540 | ||
541 | /* | |
1a9fd417 | 542 | * If zones[0] is conventional, always use the beginning of the |
12659251 NA |
543 | * zone to record superblock. No need to validate in that case. |
544 | */ | |
545 | if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type == | |
546 | BLK_ZONE_TYPE_CONVENTIONAL) | |
547 | continue; | |
548 | ||
549 | ret = sb_write_pointer(device->bdev, | |
550 | &zone_info->sb_zones[sb_pos], &sb_wp); | |
551 | if (ret != -ENOENT && ret) { | |
552 | btrfs_err_in_rcu(device->fs_info, | |
553 | "zoned: super block log zone corrupted devid %llu zone %u", | |
554 | device->devid, sb_zone); | |
555 | ret = -EUCLEAN; | |
556 | goto out; | |
557 | } | |
558 | } | |
559 | ||
560 | ||
5b316468 NA |
561 | kfree(zones); |
562 | ||
3c9daa09 JT |
563 | switch (bdev_zoned_model(bdev)) { |
564 | case BLK_ZONED_HM: | |
565 | model = "host-managed zoned"; | |
566 | emulated = ""; | |
567 | break; | |
568 | case BLK_ZONED_HA: | |
569 | model = "host-aware zoned"; | |
570 | emulated = ""; | |
571 | break; | |
572 | case BLK_ZONED_NONE: | |
573 | model = "regular"; | |
574 | emulated = "emulated "; | |
575 | break; | |
576 | default: | |
577 | /* Just in case */ | |
578 | btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s", | |
579 | bdev_zoned_model(bdev), | |
580 | rcu_str_deref(device->name)); | |
581 | ret = -EOPNOTSUPP; | |
582 | goto out_free_zone_info; | |
583 | } | |
584 | ||
585 | btrfs_info_in_rcu(fs_info, | |
586 | "%s block device %s, %u %szones of %llu bytes", | |
587 | model, rcu_str_deref(device->name), zone_info->nr_zones, | |
588 | emulated, zone_info->zone_size); | |
5b316468 NA |
589 | |
590 | return 0; | |
591 | ||
592 | out: | |
593 | kfree(zones); | |
3c9daa09 | 594 | out_free_zone_info: |
16beac87 | 595 | btrfs_destroy_dev_zone_info(device); |
5b316468 NA |
596 | |
597 | return ret; | |
598 | } | |
599 | ||
600 | void btrfs_destroy_dev_zone_info(struct btrfs_device *device) | |
601 | { | |
602 | struct btrfs_zoned_device_info *zone_info = device->zone_info; | |
603 | ||
604 | if (!zone_info) | |
605 | return; | |
606 | ||
ea6f8ddc | 607 | bitmap_free(zone_info->active_zones); |
5b316468 NA |
608 | bitmap_free(zone_info->seq_zones); |
609 | bitmap_free(zone_info->empty_zones); | |
16beac87 | 610 | vfree(zone_info->zone_cache); |
5b316468 NA |
611 | kfree(zone_info); |
612 | device->zone_info = NULL; | |
613 | } | |
614 | ||
615 | int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos, | |
616 | struct blk_zone *zone) | |
617 | { | |
618 | unsigned int nr_zones = 1; | |
619 | int ret; | |
620 | ||
621 | ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones); | |
622 | if (ret != 0 || !nr_zones) | |
623 | return ret ? ret : -EIO; | |
624 | ||
625 | return 0; | |
626 | } | |
b70f5097 NA |
627 | |
628 | int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info) | |
629 | { | |
630 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
631 | struct btrfs_device *device; | |
632 | u64 zoned_devices = 0; | |
633 | u64 nr_devices = 0; | |
634 | u64 zone_size = 0; | |
3c9daa09 | 635 | const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED); |
b70f5097 NA |
636 | int ret = 0; |
637 | ||
638 | /* Count zoned devices */ | |
639 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
640 | enum blk_zoned_model model; | |
641 | ||
642 | if (!device->bdev) | |
643 | continue; | |
644 | ||
645 | model = bdev_zoned_model(device->bdev); | |
3c9daa09 JT |
646 | /* |
647 | * A Host-Managed zoned device must be used as a zoned device. | |
648 | * A Host-Aware zoned device and a non-zoned devices can be | |
649 | * treated as a zoned device, if ZONED flag is enabled in the | |
650 | * superblock. | |
651 | */ | |
b70f5097 | 652 | if (model == BLK_ZONED_HM || |
3c9daa09 JT |
653 | (model == BLK_ZONED_HA && incompat_zoned) || |
654 | (model == BLK_ZONED_NONE && incompat_zoned)) { | |
f716fa47 | 655 | struct btrfs_zoned_device_info *zone_info; |
862931c7 NA |
656 | |
657 | zone_info = device->zone_info; | |
b70f5097 NA |
658 | zoned_devices++; |
659 | if (!zone_size) { | |
862931c7 NA |
660 | zone_size = zone_info->zone_size; |
661 | } else if (zone_info->zone_size != zone_size) { | |
b70f5097 NA |
662 | btrfs_err(fs_info, |
663 | "zoned: unequal block device zone sizes: have %llu found %llu", | |
664 | device->zone_info->zone_size, | |
665 | zone_size); | |
666 | ret = -EINVAL; | |
667 | goto out; | |
668 | } | |
669 | } | |
670 | nr_devices++; | |
671 | } | |
672 | ||
673 | if (!zoned_devices && !incompat_zoned) | |
674 | goto out; | |
675 | ||
676 | if (!zoned_devices && incompat_zoned) { | |
677 | /* No zoned block device found on ZONED filesystem */ | |
678 | btrfs_err(fs_info, | |
679 | "zoned: no zoned devices found on a zoned filesystem"); | |
680 | ret = -EINVAL; | |
681 | goto out; | |
682 | } | |
683 | ||
684 | if (zoned_devices && !incompat_zoned) { | |
685 | btrfs_err(fs_info, | |
686 | "zoned: mode not enabled but zoned device found"); | |
687 | ret = -EINVAL; | |
688 | goto out; | |
689 | } | |
690 | ||
691 | if (zoned_devices != nr_devices) { | |
692 | btrfs_err(fs_info, | |
693 | "zoned: cannot mix zoned and regular devices"); | |
694 | ret = -EINVAL; | |
695 | goto out; | |
696 | } | |
697 | ||
698 | /* | |
699 | * stripe_size is always aligned to BTRFS_STRIPE_LEN in | |
f6f39f7a | 700 | * btrfs_create_chunk(). Since we want stripe_len == zone_size, |
b70f5097 NA |
701 | * check the alignment here. |
702 | */ | |
703 | if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) { | |
704 | btrfs_err(fs_info, | |
705 | "zoned: zone size %llu not aligned to stripe %u", | |
706 | zone_size, BTRFS_STRIPE_LEN); | |
707 | ret = -EINVAL; | |
708 | goto out; | |
709 | } | |
710 | ||
a589dde0 NA |
711 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
712 | btrfs_err(fs_info, "zoned: mixed block groups not supported"); | |
713 | ret = -EINVAL; | |
714 | goto out; | |
715 | } | |
716 | ||
b70f5097 | 717 | fs_info->zone_size = zone_size; |
1cd6121f | 718 | fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED; |
b70f5097 | 719 | |
b53429ba JT |
720 | /* |
721 | * Check mount options here, because we might change fs_info->zoned | |
722 | * from fs_info->zone_size. | |
723 | */ | |
724 | ret = btrfs_check_mountopts_zoned(fs_info); | |
725 | if (ret) | |
726 | goto out; | |
727 | ||
b70f5097 NA |
728 | btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size); |
729 | out: | |
730 | return ret; | |
731 | } | |
5d1ab66c NA |
732 | |
733 | int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info) | |
734 | { | |
735 | if (!btrfs_is_zoned(info)) | |
736 | return 0; | |
737 | ||
738 | /* | |
739 | * Space cache writing is not COWed. Disable that to avoid write errors | |
740 | * in sequential zones. | |
741 | */ | |
742 | if (btrfs_test_opt(info, SPACE_CACHE)) { | |
743 | btrfs_err(info, "zoned: space cache v1 is not supported"); | |
744 | return -EINVAL; | |
745 | } | |
746 | ||
d206e9c9 NA |
747 | if (btrfs_test_opt(info, NODATACOW)) { |
748 | btrfs_err(info, "zoned: NODATACOW not supported"); | |
749 | return -EINVAL; | |
750 | } | |
751 | ||
5d1ab66c NA |
752 | return 0; |
753 | } | |
12659251 NA |
754 | |
755 | static int sb_log_location(struct block_device *bdev, struct blk_zone *zones, | |
756 | int rw, u64 *bytenr_ret) | |
757 | { | |
758 | u64 wp; | |
759 | int ret; | |
760 | ||
761 | if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) { | |
762 | *bytenr_ret = zones[0].start << SECTOR_SHIFT; | |
763 | return 0; | |
764 | } | |
765 | ||
766 | ret = sb_write_pointer(bdev, zones, &wp); | |
767 | if (ret != -ENOENT && ret < 0) | |
768 | return ret; | |
769 | ||
770 | if (rw == WRITE) { | |
771 | struct blk_zone *reset = NULL; | |
772 | ||
773 | if (wp == zones[0].start << SECTOR_SHIFT) | |
774 | reset = &zones[0]; | |
775 | else if (wp == zones[1].start << SECTOR_SHIFT) | |
776 | reset = &zones[1]; | |
777 | ||
778 | if (reset && reset->cond != BLK_ZONE_COND_EMPTY) { | |
5daaf552 | 779 | ASSERT(sb_zone_is_full(reset)); |
12659251 NA |
780 | |
781 | ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
782 | reset->start, reset->len, | |
783 | GFP_NOFS); | |
784 | if (ret) | |
785 | return ret; | |
786 | ||
787 | reset->cond = BLK_ZONE_COND_EMPTY; | |
788 | reset->wp = reset->start; | |
789 | } | |
790 | } else if (ret != -ENOENT) { | |
9658b72e NA |
791 | /* |
792 | * For READ, we want the previous one. Move write pointer to | |
793 | * the end of a zone, if it is at the head of a zone. | |
794 | */ | |
795 | u64 zone_end = 0; | |
796 | ||
12659251 | 797 | if (wp == zones[0].start << SECTOR_SHIFT) |
9658b72e NA |
798 | zone_end = zones[1].start + zones[1].capacity; |
799 | else if (wp == zones[1].start << SECTOR_SHIFT) | |
800 | zone_end = zones[0].start + zones[0].capacity; | |
801 | if (zone_end) | |
802 | wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT, | |
803 | BTRFS_SUPER_INFO_SIZE); | |
804 | ||
12659251 NA |
805 | wp -= BTRFS_SUPER_INFO_SIZE; |
806 | } | |
807 | ||
808 | *bytenr_ret = wp; | |
809 | return 0; | |
810 | ||
811 | } | |
812 | ||
813 | int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw, | |
814 | u64 *bytenr_ret) | |
815 | { | |
816 | struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES]; | |
d734492a | 817 | sector_t zone_sectors; |
12659251 NA |
818 | u32 sb_zone; |
819 | int ret; | |
12659251 NA |
820 | u8 zone_sectors_shift; |
821 | sector_t nr_sectors; | |
822 | u32 nr_zones; | |
823 | ||
824 | if (!bdev_is_zoned(bdev)) { | |
825 | *bytenr_ret = btrfs_sb_offset(mirror); | |
826 | return 0; | |
827 | } | |
828 | ||
829 | ASSERT(rw == READ || rw == WRITE); | |
830 | ||
831 | zone_sectors = bdev_zone_sectors(bdev); | |
832 | if (!is_power_of_2(zone_sectors)) | |
833 | return -EINVAL; | |
12659251 | 834 | zone_sectors_shift = ilog2(zone_sectors); |
ac7ac461 | 835 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
836 | nr_zones = nr_sectors >> zone_sectors_shift; |
837 | ||
838 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
839 | if (sb_zone + 1 >= nr_zones) | |
840 | return -ENOENT; | |
841 | ||
5b434df8 | 842 | ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev), |
12659251 NA |
843 | BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb, |
844 | zones); | |
845 | if (ret < 0) | |
846 | return ret; | |
847 | if (ret != BTRFS_NR_SB_LOG_ZONES) | |
848 | return -EIO; | |
849 | ||
850 | return sb_log_location(bdev, zones, rw, bytenr_ret); | |
851 | } | |
852 | ||
853 | int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw, | |
854 | u64 *bytenr_ret) | |
855 | { | |
856 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
857 | u32 zone_num; | |
858 | ||
d6639b35 NA |
859 | /* |
860 | * For a zoned filesystem on a non-zoned block device, use the same | |
861 | * super block locations as regular filesystem. Doing so, the super | |
862 | * block can always be retrieved and the zoned flag of the volume | |
863 | * detected from the super block information. | |
864 | */ | |
865 | if (!bdev_is_zoned(device->bdev)) { | |
12659251 NA |
866 | *bytenr_ret = btrfs_sb_offset(mirror); |
867 | return 0; | |
868 | } | |
869 | ||
870 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
871 | if (zone_num + 1 >= zinfo->nr_zones) | |
872 | return -ENOENT; | |
873 | ||
874 | return sb_log_location(device->bdev, | |
875 | &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror], | |
876 | rw, bytenr_ret); | |
877 | } | |
878 | ||
879 | static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo, | |
880 | int mirror) | |
881 | { | |
882 | u32 zone_num; | |
883 | ||
884 | if (!zinfo) | |
885 | return false; | |
886 | ||
887 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
888 | if (zone_num + 1 >= zinfo->nr_zones) | |
889 | return false; | |
890 | ||
891 | if (!test_bit(zone_num, zinfo->seq_zones)) | |
892 | return false; | |
893 | ||
894 | return true; | |
895 | } | |
896 | ||
8376d9e1 | 897 | int btrfs_advance_sb_log(struct btrfs_device *device, int mirror) |
12659251 NA |
898 | { |
899 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
900 | struct blk_zone *zone; | |
8376d9e1 | 901 | int i; |
12659251 NA |
902 | |
903 | if (!is_sb_log_zone(zinfo, mirror)) | |
8376d9e1 | 904 | return 0; |
12659251 NA |
905 | |
906 | zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror]; | |
8376d9e1 NA |
907 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { |
908 | /* Advance the next zone */ | |
909 | if (zone->cond == BLK_ZONE_COND_FULL) { | |
910 | zone++; | |
911 | continue; | |
912 | } | |
913 | ||
12659251 NA |
914 | if (zone->cond == BLK_ZONE_COND_EMPTY) |
915 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
916 | ||
8376d9e1 NA |
917 | zone->wp += SUPER_INFO_SECTORS; |
918 | ||
919 | if (sb_zone_is_full(zone)) { | |
920 | /* | |
921 | * No room left to write new superblock. Since | |
922 | * superblock is written with REQ_SYNC, it is safe to | |
923 | * finish the zone now. | |
924 | * | |
925 | * If the write pointer is exactly at the capacity, | |
926 | * explicit ZONE_FINISH is not necessary. | |
927 | */ | |
928 | if (zone->wp != zone->start + zone->capacity) { | |
929 | int ret; | |
930 | ||
931 | ret = blkdev_zone_mgmt(device->bdev, | |
932 | REQ_OP_ZONE_FINISH, zone->start, | |
933 | zone->len, GFP_NOFS); | |
934 | if (ret) | |
935 | return ret; | |
936 | } | |
12659251 | 937 | |
8376d9e1 | 938 | zone->wp = zone->start + zone->len; |
12659251 | 939 | zone->cond = BLK_ZONE_COND_FULL; |
8376d9e1 NA |
940 | } |
941 | return 0; | |
12659251 NA |
942 | } |
943 | ||
8376d9e1 NA |
944 | /* All the zones are FULL. Should not reach here. */ |
945 | ASSERT(0); | |
946 | return -EIO; | |
12659251 NA |
947 | } |
948 | ||
949 | int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror) | |
950 | { | |
951 | sector_t zone_sectors; | |
952 | sector_t nr_sectors; | |
953 | u8 zone_sectors_shift; | |
954 | u32 sb_zone; | |
955 | u32 nr_zones; | |
956 | ||
957 | zone_sectors = bdev_zone_sectors(bdev); | |
958 | zone_sectors_shift = ilog2(zone_sectors); | |
ac7ac461 | 959 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
960 | nr_zones = nr_sectors >> zone_sectors_shift; |
961 | ||
962 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
963 | if (sb_zone + 1 >= nr_zones) | |
964 | return -ENOENT; | |
965 | ||
966 | return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
5b434df8 | 967 | zone_start_sector(sb_zone, bdev), |
12659251 NA |
968 | zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS); |
969 | } | |
1cd6121f NA |
970 | |
971 | /** | |
972 | * btrfs_find_allocatable_zones - find allocatable zones within a given region | |
973 | * | |
974 | * @device: the device to allocate a region on | |
975 | * @hole_start: the position of the hole to allocate the region | |
976 | * @num_bytes: size of wanted region | |
977 | * @hole_end: the end of the hole | |
978 | * @return: position of allocatable zones | |
979 | * | |
980 | * Allocatable region should not contain any superblock locations. | |
981 | */ | |
982 | u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start, | |
983 | u64 hole_end, u64 num_bytes) | |
984 | { | |
985 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
986 | const u8 shift = zinfo->zone_size_shift; | |
987 | u64 nzones = num_bytes >> shift; | |
988 | u64 pos = hole_start; | |
989 | u64 begin, end; | |
990 | bool have_sb; | |
991 | int i; | |
992 | ||
993 | ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size)); | |
994 | ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size)); | |
995 | ||
996 | while (pos < hole_end) { | |
997 | begin = pos >> shift; | |
998 | end = begin + nzones; | |
999 | ||
1000 | if (end > zinfo->nr_zones) | |
1001 | return hole_end; | |
1002 | ||
1003 | /* Check if zones in the region are all empty */ | |
1004 | if (btrfs_dev_is_sequential(device, pos) && | |
1005 | find_next_zero_bit(zinfo->empty_zones, end, begin) != end) { | |
1006 | pos += zinfo->zone_size; | |
1007 | continue; | |
1008 | } | |
1009 | ||
1010 | have_sb = false; | |
1011 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1012 | u32 sb_zone; | |
1013 | u64 sb_pos; | |
1014 | ||
1015 | sb_zone = sb_zone_number(shift, i); | |
1016 | if (!(end <= sb_zone || | |
1017 | sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) { | |
1018 | have_sb = true; | |
5b434df8 NA |
1019 | pos = zone_start_physical( |
1020 | sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo); | |
1cd6121f NA |
1021 | break; |
1022 | } | |
1023 | ||
1024 | /* We also need to exclude regular superblock positions */ | |
1025 | sb_pos = btrfs_sb_offset(i); | |
1026 | if (!(pos + num_bytes <= sb_pos || | |
1027 | sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) { | |
1028 | have_sb = true; | |
1029 | pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE, | |
1030 | zinfo->zone_size); | |
1031 | break; | |
1032 | } | |
1033 | } | |
1034 | if (!have_sb) | |
1035 | break; | |
1036 | } | |
1037 | ||
1038 | return pos; | |
1039 | } | |
1040 | ||
afba2bc0 NA |
1041 | static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos) |
1042 | { | |
1043 | struct btrfs_zoned_device_info *zone_info = device->zone_info; | |
1044 | unsigned int zno = (pos >> zone_info->zone_size_shift); | |
1045 | ||
1046 | /* We can use any number of zones */ | |
1047 | if (zone_info->max_active_zones == 0) | |
1048 | return true; | |
1049 | ||
1050 | if (!test_bit(zno, zone_info->active_zones)) { | |
1051 | /* Active zone left? */ | |
1052 | if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0) | |
1053 | return false; | |
1054 | if (test_and_set_bit(zno, zone_info->active_zones)) { | |
1055 | /* Someone already set the bit */ | |
1056 | atomic_inc(&zone_info->active_zones_left); | |
1057 | } | |
1058 | } | |
1059 | ||
1060 | return true; | |
1061 | } | |
1062 | ||
1063 | static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos) | |
1064 | { | |
1065 | struct btrfs_zoned_device_info *zone_info = device->zone_info; | |
1066 | unsigned int zno = (pos >> zone_info->zone_size_shift); | |
1067 | ||
1068 | /* We can use any number of zones */ | |
1069 | if (zone_info->max_active_zones == 0) | |
1070 | return; | |
1071 | ||
1072 | if (test_and_clear_bit(zno, zone_info->active_zones)) | |
1073 | atomic_inc(&zone_info->active_zones_left); | |
1074 | } | |
1075 | ||
1cd6121f NA |
1076 | int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical, |
1077 | u64 length, u64 *bytes) | |
1078 | { | |
1079 | int ret; | |
1080 | ||
1081 | *bytes = 0; | |
1082 | ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET, | |
1083 | physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT, | |
1084 | GFP_NOFS); | |
1085 | if (ret) | |
1086 | return ret; | |
1087 | ||
1088 | *bytes = length; | |
1089 | while (length) { | |
1090 | btrfs_dev_set_zone_empty(device, physical); | |
afba2bc0 | 1091 | btrfs_dev_clear_active_zone(device, physical); |
1cd6121f NA |
1092 | physical += device->zone_info->zone_size; |
1093 | length -= device->zone_info->zone_size; | |
1094 | } | |
1095 | ||
1096 | return 0; | |
1097 | } | |
1098 | ||
1099 | int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size) | |
1100 | { | |
1101 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
1102 | const u8 shift = zinfo->zone_size_shift; | |
1103 | unsigned long begin = start >> shift; | |
1104 | unsigned long end = (start + size) >> shift; | |
1105 | u64 pos; | |
1106 | int ret; | |
1107 | ||
1108 | ASSERT(IS_ALIGNED(start, zinfo->zone_size)); | |
1109 | ASSERT(IS_ALIGNED(size, zinfo->zone_size)); | |
1110 | ||
1111 | if (end > zinfo->nr_zones) | |
1112 | return -ERANGE; | |
1113 | ||
1114 | /* All the zones are conventional */ | |
1115 | if (find_next_bit(zinfo->seq_zones, begin, end) == end) | |
1116 | return 0; | |
1117 | ||
1118 | /* All the zones are sequential and empty */ | |
1119 | if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end && | |
1120 | find_next_zero_bit(zinfo->empty_zones, begin, end) == end) | |
1121 | return 0; | |
1122 | ||
1123 | for (pos = start; pos < start + size; pos += zinfo->zone_size) { | |
1124 | u64 reset_bytes; | |
1125 | ||
1126 | if (!btrfs_dev_is_sequential(device, pos) || | |
1127 | btrfs_dev_is_empty_zone(device, pos)) | |
1128 | continue; | |
1129 | ||
1130 | /* Free regions should be empty */ | |
1131 | btrfs_warn_in_rcu( | |
1132 | device->fs_info, | |
1133 | "zoned: resetting device %s (devid %llu) zone %llu for allocation", | |
1134 | rcu_str_deref(device->name), device->devid, pos >> shift); | |
1135 | WARN_ON_ONCE(1); | |
1136 | ||
1137 | ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size, | |
1138 | &reset_bytes); | |
1139 | if (ret) | |
1140 | return ret; | |
1141 | } | |
1142 | ||
1143 | return 0; | |
1144 | } | |
08e11a3d | 1145 | |
a94794d5 NA |
1146 | /* |
1147 | * Calculate an allocation pointer from the extent allocation information | |
1148 | * for a block group consist of conventional zones. It is pointed to the | |
1149 | * end of the highest addressed extent in the block group as an allocation | |
1150 | * offset. | |
1151 | */ | |
1152 | static int calculate_alloc_pointer(struct btrfs_block_group *cache, | |
1153 | u64 *offset_ret) | |
1154 | { | |
1155 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
29cbcf40 | 1156 | struct btrfs_root *root; |
a94794d5 NA |
1157 | struct btrfs_path *path; |
1158 | struct btrfs_key key; | |
1159 | struct btrfs_key found_key; | |
1160 | int ret; | |
1161 | u64 length; | |
1162 | ||
1163 | path = btrfs_alloc_path(); | |
1164 | if (!path) | |
1165 | return -ENOMEM; | |
1166 | ||
1167 | key.objectid = cache->start + cache->length; | |
1168 | key.type = 0; | |
1169 | key.offset = 0; | |
1170 | ||
29cbcf40 | 1171 | root = btrfs_extent_root(fs_info, key.objectid); |
a94794d5 NA |
1172 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
1173 | /* We should not find the exact match */ | |
1174 | if (!ret) | |
1175 | ret = -EUCLEAN; | |
1176 | if (ret < 0) | |
1177 | goto out; | |
1178 | ||
1179 | ret = btrfs_previous_extent_item(root, path, cache->start); | |
1180 | if (ret) { | |
1181 | if (ret == 1) { | |
1182 | ret = 0; | |
1183 | *offset_ret = 0; | |
1184 | } | |
1185 | goto out; | |
1186 | } | |
1187 | ||
1188 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
1189 | ||
1190 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY) | |
1191 | length = found_key.offset; | |
1192 | else | |
1193 | length = fs_info->nodesize; | |
1194 | ||
1195 | if (!(found_key.objectid >= cache->start && | |
1196 | found_key.objectid + length <= cache->start + cache->length)) { | |
1197 | ret = -EUCLEAN; | |
1198 | goto out; | |
1199 | } | |
1200 | *offset_ret = found_key.objectid + length - cache->start; | |
1201 | ret = 0; | |
1202 | ||
1203 | out: | |
1204 | btrfs_free_path(path); | |
1205 | return ret; | |
1206 | } | |
1207 | ||
1208 | int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) | |
08e11a3d NA |
1209 | { |
1210 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
1211 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1212 | struct extent_map *em; | |
1213 | struct map_lookup *map; | |
1214 | struct btrfs_device *device; | |
1215 | u64 logical = cache->start; | |
1216 | u64 length = cache->length; | |
08e11a3d NA |
1217 | int ret; |
1218 | int i; | |
1219 | unsigned int nofs_flag; | |
1220 | u64 *alloc_offsets = NULL; | |
8eae532b | 1221 | u64 *caps = NULL; |
dbfcc18f | 1222 | u64 *physical = NULL; |
68a384b5 | 1223 | unsigned long *active = NULL; |
a94794d5 | 1224 | u64 last_alloc = 0; |
08e11a3d NA |
1225 | u32 num_sequential = 0, num_conventional = 0; |
1226 | ||
1227 | if (!btrfs_is_zoned(fs_info)) | |
1228 | return 0; | |
1229 | ||
1230 | /* Sanity check */ | |
1231 | if (!IS_ALIGNED(length, fs_info->zone_size)) { | |
1232 | btrfs_err(fs_info, | |
1233 | "zoned: block group %llu len %llu unaligned to zone size %llu", | |
1234 | logical, length, fs_info->zone_size); | |
1235 | return -EIO; | |
1236 | } | |
1237 | ||
1238 | /* Get the chunk mapping */ | |
1239 | read_lock(&em_tree->lock); | |
1240 | em = lookup_extent_mapping(em_tree, logical, length); | |
1241 | read_unlock(&em_tree->lock); | |
1242 | ||
1243 | if (!em) | |
1244 | return -EINVAL; | |
1245 | ||
1246 | map = em->map_lookup; | |
1247 | ||
64259baa | 1248 | cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS); |
dafc340d NA |
1249 | if (!cache->physical_map) { |
1250 | ret = -ENOMEM; | |
1251 | goto out; | |
1252 | } | |
1253 | ||
08e11a3d NA |
1254 | alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS); |
1255 | if (!alloc_offsets) { | |
dafc340d NA |
1256 | ret = -ENOMEM; |
1257 | goto out; | |
08e11a3d NA |
1258 | } |
1259 | ||
8eae532b NA |
1260 | caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS); |
1261 | if (!caps) { | |
1262 | ret = -ENOMEM; | |
1263 | goto out; | |
1264 | } | |
1265 | ||
dbfcc18f JT |
1266 | physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS); |
1267 | if (!physical) { | |
1268 | ret = -ENOMEM; | |
1269 | goto out; | |
1270 | } | |
1271 | ||
68a384b5 NA |
1272 | active = bitmap_zalloc(map->num_stripes, GFP_NOFS); |
1273 | if (!active) { | |
1274 | ret = -ENOMEM; | |
1275 | goto out; | |
1276 | } | |
1277 | ||
08e11a3d NA |
1278 | for (i = 0; i < map->num_stripes; i++) { |
1279 | bool is_sequential; | |
1280 | struct blk_zone zone; | |
6143c23c NA |
1281 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
1282 | int dev_replace_is_ongoing = 0; | |
08e11a3d NA |
1283 | |
1284 | device = map->stripes[i].dev; | |
dbfcc18f | 1285 | physical[i] = map->stripes[i].physical; |
08e11a3d NA |
1286 | |
1287 | if (device->bdev == NULL) { | |
1288 | alloc_offsets[i] = WP_MISSING_DEV; | |
1289 | continue; | |
1290 | } | |
1291 | ||
dbfcc18f | 1292 | is_sequential = btrfs_dev_is_sequential(device, physical[i]); |
08e11a3d NA |
1293 | if (is_sequential) |
1294 | num_sequential++; | |
1295 | else | |
1296 | num_conventional++; | |
1297 | ||
1298 | if (!is_sequential) { | |
1299 | alloc_offsets[i] = WP_CONVENTIONAL; | |
1300 | continue; | |
1301 | } | |
1302 | ||
1303 | /* | |
1304 | * This zone will be used for allocation, so mark this zone | |
1305 | * non-empty. | |
1306 | */ | |
dbfcc18f | 1307 | btrfs_dev_clear_zone_empty(device, physical[i]); |
08e11a3d | 1308 | |
6143c23c NA |
1309 | down_read(&dev_replace->rwsem); |
1310 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
1311 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) | |
dbfcc18f | 1312 | btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]); |
6143c23c NA |
1313 | up_read(&dev_replace->rwsem); |
1314 | ||
08e11a3d NA |
1315 | /* |
1316 | * The group is mapped to a sequential zone. Get the zone write | |
1317 | * pointer to determine the allocation offset within the zone. | |
1318 | */ | |
dbfcc18f | 1319 | WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size)); |
08e11a3d | 1320 | nofs_flag = memalloc_nofs_save(); |
dbfcc18f | 1321 | ret = btrfs_get_dev_zone(device, physical[i], &zone); |
08e11a3d NA |
1322 | memalloc_nofs_restore(nofs_flag); |
1323 | if (ret == -EIO || ret == -EOPNOTSUPP) { | |
1324 | ret = 0; | |
1325 | alloc_offsets[i] = WP_MISSING_DEV; | |
1326 | continue; | |
1327 | } else if (ret) { | |
1328 | goto out; | |
1329 | } | |
1330 | ||
784daf2b | 1331 | if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) { |
47cdfb5e NA |
1332 | btrfs_err_in_rcu(fs_info, |
1333 | "zoned: unexpected conventional zone %llu on device %s (devid %llu)", | |
1334 | zone.start << SECTOR_SHIFT, | |
1335 | rcu_str_deref(device->name), device->devid); | |
784daf2b NA |
1336 | ret = -EIO; |
1337 | goto out; | |
1338 | } | |
1339 | ||
8eae532b NA |
1340 | caps[i] = (zone.capacity << SECTOR_SHIFT); |
1341 | ||
08e11a3d NA |
1342 | switch (zone.cond) { |
1343 | case BLK_ZONE_COND_OFFLINE: | |
1344 | case BLK_ZONE_COND_READONLY: | |
1345 | btrfs_err(fs_info, | |
1346 | "zoned: offline/readonly zone %llu on device %s (devid %llu)", | |
dbfcc18f | 1347 | physical[i] >> device->zone_info->zone_size_shift, |
08e11a3d NA |
1348 | rcu_str_deref(device->name), device->devid); |
1349 | alloc_offsets[i] = WP_MISSING_DEV; | |
1350 | break; | |
1351 | case BLK_ZONE_COND_EMPTY: | |
1352 | alloc_offsets[i] = 0; | |
1353 | break; | |
1354 | case BLK_ZONE_COND_FULL: | |
8eae532b | 1355 | alloc_offsets[i] = caps[i]; |
08e11a3d NA |
1356 | break; |
1357 | default: | |
1358 | /* Partially used zone */ | |
1359 | alloc_offsets[i] = | |
1360 | ((zone.wp - zone.start) << SECTOR_SHIFT); | |
68a384b5 | 1361 | __set_bit(i, active); |
08e11a3d NA |
1362 | break; |
1363 | } | |
68a384b5 NA |
1364 | |
1365 | /* | |
1366 | * Consider a zone as active if we can allow any number of | |
1367 | * active zones. | |
1368 | */ | |
1369 | if (!device->zone_info->max_active_zones) | |
1370 | __set_bit(i, active); | |
08e11a3d NA |
1371 | } |
1372 | ||
08f45559 JT |
1373 | if (num_sequential > 0) |
1374 | cache->seq_zone = true; | |
1375 | ||
08e11a3d NA |
1376 | if (num_conventional > 0) { |
1377 | /* | |
a94794d5 NA |
1378 | * Avoid calling calculate_alloc_pointer() for new BG. It |
1379 | * is no use for new BG. It must be always 0. | |
1380 | * | |
1381 | * Also, we have a lock chain of extent buffer lock -> | |
1382 | * chunk mutex. For new BG, this function is called from | |
1383 | * btrfs_make_block_group() which is already taking the | |
1384 | * chunk mutex. Thus, we cannot call | |
1385 | * calculate_alloc_pointer() which takes extent buffer | |
1386 | * locks to avoid deadlock. | |
08e11a3d | 1387 | */ |
8eae532b NA |
1388 | |
1389 | /* Zone capacity is always zone size in emulation */ | |
1390 | cache->zone_capacity = cache->length; | |
a94794d5 NA |
1391 | if (new) { |
1392 | cache->alloc_offset = 0; | |
1393 | goto out; | |
1394 | } | |
1395 | ret = calculate_alloc_pointer(cache, &last_alloc); | |
1396 | if (ret || map->num_stripes == num_conventional) { | |
1397 | if (!ret) | |
1398 | cache->alloc_offset = last_alloc; | |
1399 | else | |
1400 | btrfs_err(fs_info, | |
1401 | "zoned: failed to determine allocation offset of bg %llu", | |
1402 | cache->start); | |
1403 | goto out; | |
1404 | } | |
08e11a3d NA |
1405 | } |
1406 | ||
1407 | switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
1408 | case 0: /* single */ | |
06e1e7f4 JT |
1409 | if (alloc_offsets[0] == WP_MISSING_DEV) { |
1410 | btrfs_err(fs_info, | |
1411 | "zoned: cannot recover write pointer for zone %llu", | |
dbfcc18f | 1412 | physical[0]); |
06e1e7f4 JT |
1413 | ret = -EIO; |
1414 | goto out; | |
1415 | } | |
08e11a3d | 1416 | cache->alloc_offset = alloc_offsets[0]; |
8eae532b | 1417 | cache->zone_capacity = caps[0]; |
68a384b5 | 1418 | cache->zone_is_active = test_bit(0, active); |
08e11a3d NA |
1419 | break; |
1420 | case BTRFS_BLOCK_GROUP_DUP: | |
265f7237 JT |
1421 | if (map->type & BTRFS_BLOCK_GROUP_DATA) { |
1422 | btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg"); | |
1423 | ret = -EINVAL; | |
1424 | goto out; | |
1425 | } | |
1426 | if (alloc_offsets[0] == WP_MISSING_DEV) { | |
1427 | btrfs_err(fs_info, | |
1428 | "zoned: cannot recover write pointer for zone %llu", | |
1429 | physical[0]); | |
1430 | ret = -EIO; | |
1431 | goto out; | |
1432 | } | |
1433 | if (alloc_offsets[1] == WP_MISSING_DEV) { | |
1434 | btrfs_err(fs_info, | |
1435 | "zoned: cannot recover write pointer for zone %llu", | |
1436 | physical[1]); | |
1437 | ret = -EIO; | |
1438 | goto out; | |
1439 | } | |
1440 | if (alloc_offsets[0] != alloc_offsets[1]) { | |
1441 | btrfs_err(fs_info, | |
1442 | "zoned: write pointer offset mismatch of zones in DUP profile"); | |
1443 | ret = -EIO; | |
1444 | goto out; | |
1445 | } | |
1446 | if (test_bit(0, active) != test_bit(1, active)) { | |
1447 | if (!btrfs_zone_activate(cache)) { | |
1448 | ret = -EIO; | |
1449 | goto out; | |
1450 | } | |
1451 | } else { | |
1452 | cache->zone_is_active = test_bit(0, active); | |
1453 | } | |
1454 | cache->alloc_offset = alloc_offsets[0]; | |
1455 | cache->zone_capacity = min(caps[0], caps[1]); | |
1456 | break; | |
08e11a3d NA |
1457 | case BTRFS_BLOCK_GROUP_RAID1: |
1458 | case BTRFS_BLOCK_GROUP_RAID0: | |
1459 | case BTRFS_BLOCK_GROUP_RAID10: | |
1460 | case BTRFS_BLOCK_GROUP_RAID5: | |
1461 | case BTRFS_BLOCK_GROUP_RAID6: | |
1462 | /* non-single profiles are not supported yet */ | |
1463 | default: | |
1464 | btrfs_err(fs_info, "zoned: profile %s not yet supported", | |
1465 | btrfs_bg_type_to_raid_name(map->type)); | |
1466 | ret = -EINVAL; | |
1467 | goto out; | |
1468 | } | |
1469 | ||
68a384b5 NA |
1470 | if (cache->zone_is_active) { |
1471 | btrfs_get_block_group(cache); | |
1472 | spin_lock(&fs_info->zone_active_bgs_lock); | |
1473 | list_add_tail(&cache->active_bg_list, &fs_info->zone_active_bgs); | |
1474 | spin_unlock(&fs_info->zone_active_bgs_lock); | |
1475 | } | |
1476 | ||
08e11a3d | 1477 | out: |
06e1e7f4 JT |
1478 | if (cache->alloc_offset > fs_info->zone_size) { |
1479 | btrfs_err(fs_info, | |
1480 | "zoned: invalid write pointer %llu in block group %llu", | |
1481 | cache->alloc_offset, cache->start); | |
1482 | ret = -EIO; | |
1483 | } | |
1484 | ||
8eae532b NA |
1485 | if (cache->alloc_offset > cache->zone_capacity) { |
1486 | btrfs_err(fs_info, | |
1487 | "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu", | |
1488 | cache->alloc_offset, cache->zone_capacity, | |
1489 | cache->start); | |
1490 | ret = -EIO; | |
1491 | } | |
1492 | ||
a94794d5 NA |
1493 | /* An extent is allocated after the write pointer */ |
1494 | if (!ret && num_conventional && last_alloc > cache->alloc_offset) { | |
1495 | btrfs_err(fs_info, | |
1496 | "zoned: got wrong write pointer in BG %llu: %llu > %llu", | |
1497 | logical, last_alloc, cache->alloc_offset); | |
1498 | ret = -EIO; | |
1499 | } | |
1500 | ||
0bc09ca1 NA |
1501 | if (!ret) |
1502 | cache->meta_write_pointer = cache->alloc_offset + cache->start; | |
1503 | ||
dafc340d NA |
1504 | if (ret) { |
1505 | kfree(cache->physical_map); | |
1506 | cache->physical_map = NULL; | |
1507 | } | |
68a384b5 | 1508 | bitmap_free(active); |
dbfcc18f | 1509 | kfree(physical); |
8eae532b | 1510 | kfree(caps); |
08e11a3d NA |
1511 | kfree(alloc_offsets); |
1512 | free_extent_map(em); | |
1513 | ||
1514 | return ret; | |
1515 | } | |
169e0da9 NA |
1516 | |
1517 | void btrfs_calc_zone_unusable(struct btrfs_block_group *cache) | |
1518 | { | |
1519 | u64 unusable, free; | |
1520 | ||
1521 | if (!btrfs_is_zoned(cache->fs_info)) | |
1522 | return; | |
1523 | ||
1524 | WARN_ON(cache->bytes_super != 0); | |
98173255 NA |
1525 | unusable = (cache->alloc_offset - cache->used) + |
1526 | (cache->length - cache->zone_capacity); | |
1527 | free = cache->zone_capacity - cache->alloc_offset; | |
169e0da9 NA |
1528 | |
1529 | /* We only need ->free_space in ALLOC_SEQ block groups */ | |
1530 | cache->last_byte_to_unpin = (u64)-1; | |
1531 | cache->cached = BTRFS_CACHE_FINISHED; | |
1532 | cache->free_space_ctl->free_space = free; | |
1533 | cache->zone_unusable = unusable; | |
169e0da9 | 1534 | } |
d3575156 NA |
1535 | |
1536 | void btrfs_redirty_list_add(struct btrfs_transaction *trans, | |
1537 | struct extent_buffer *eb) | |
1538 | { | |
1539 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
1540 | ||
1541 | if (!btrfs_is_zoned(fs_info) || | |
1542 | btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) || | |
1543 | !list_empty(&eb->release_list)) | |
1544 | return; | |
1545 | ||
1546 | set_extent_buffer_dirty(eb); | |
1547 | set_extent_bits_nowait(&trans->dirty_pages, eb->start, | |
1548 | eb->start + eb->len - 1, EXTENT_DIRTY); | |
1549 | memzero_extent_buffer(eb, 0, eb->len); | |
1550 | set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags); | |
1551 | ||
1552 | spin_lock(&trans->releasing_ebs_lock); | |
1553 | list_add_tail(&eb->release_list, &trans->releasing_ebs); | |
1554 | spin_unlock(&trans->releasing_ebs_lock); | |
1555 | atomic_inc(&eb->refs); | |
1556 | } | |
1557 | ||
1558 | void btrfs_free_redirty_list(struct btrfs_transaction *trans) | |
1559 | { | |
1560 | spin_lock(&trans->releasing_ebs_lock); | |
1561 | while (!list_empty(&trans->releasing_ebs)) { | |
1562 | struct extent_buffer *eb; | |
1563 | ||
1564 | eb = list_first_entry(&trans->releasing_ebs, | |
1565 | struct extent_buffer, release_list); | |
1566 | list_del_init(&eb->release_list); | |
1567 | free_extent_buffer(eb); | |
1568 | } | |
1569 | spin_unlock(&trans->releasing_ebs_lock); | |
1570 | } | |
08f45559 | 1571 | |
e380adfc | 1572 | bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start) |
08f45559 JT |
1573 | { |
1574 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1575 | struct btrfs_block_group *cache; | |
1576 | bool ret = false; | |
1577 | ||
1578 | if (!btrfs_is_zoned(fs_info)) | |
1579 | return false; | |
1580 | ||
08f45559 JT |
1581 | if (!is_data_inode(&inode->vfs_inode)) |
1582 | return false; | |
1583 | ||
e6d261e3 JT |
1584 | /* |
1585 | * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the | |
1586 | * extent layout the relocation code has. | |
1587 | * Furthermore we have set aside own block-group from which only the | |
1588 | * relocation "process" can allocate and make sure only one process at a | |
1589 | * time can add pages to an extent that gets relocated, so it's safe to | |
1590 | * use regular REQ_OP_WRITE for this special case. | |
1591 | */ | |
1592 | if (btrfs_is_data_reloc_root(inode->root)) | |
1593 | return false; | |
1594 | ||
e380adfc | 1595 | cache = btrfs_lookup_block_group(fs_info, start); |
08f45559 JT |
1596 | ASSERT(cache); |
1597 | if (!cache) | |
1598 | return false; | |
1599 | ||
1600 | ret = cache->seq_zone; | |
1601 | btrfs_put_block_group(cache); | |
1602 | ||
1603 | return ret; | |
1604 | } | |
d8e3fb10 NA |
1605 | |
1606 | void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset, | |
1607 | struct bio *bio) | |
1608 | { | |
1609 | struct btrfs_ordered_extent *ordered; | |
1610 | const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; | |
1611 | ||
1612 | if (bio_op(bio) != REQ_OP_ZONE_APPEND) | |
1613 | return; | |
1614 | ||
1615 | ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset); | |
1616 | if (WARN_ON(!ordered)) | |
1617 | return; | |
1618 | ||
1619 | ordered->physical = physical; | |
c7c3a6dc | 1620 | ordered->bdev = bio->bi_bdev; |
d8e3fb10 NA |
1621 | |
1622 | btrfs_put_ordered_extent(ordered); | |
1623 | } | |
1624 | ||
1625 | void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered) | |
1626 | { | |
1627 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
1628 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1629 | struct extent_map_tree *em_tree; | |
1630 | struct extent_map *em; | |
1631 | struct btrfs_ordered_sum *sum; | |
d8e3fb10 NA |
1632 | u64 orig_logical = ordered->disk_bytenr; |
1633 | u64 *logical = NULL; | |
1634 | int nr, stripe_len; | |
1635 | ||
1636 | /* Zoned devices should not have partitions. So, we can assume it is 0 */ | |
c7c3a6dc CH |
1637 | ASSERT(!bdev_is_partition(ordered->bdev)); |
1638 | if (WARN_ON(!ordered->bdev)) | |
d8e3fb10 NA |
1639 | return; |
1640 | ||
c7c3a6dc | 1641 | if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev, |
d8e3fb10 NA |
1642 | ordered->physical, &logical, &nr, |
1643 | &stripe_len))) | |
1644 | goto out; | |
1645 | ||
1646 | WARN_ON(nr != 1); | |
1647 | ||
1648 | if (orig_logical == *logical) | |
1649 | goto out; | |
1650 | ||
1651 | ordered->disk_bytenr = *logical; | |
1652 | ||
1653 | em_tree = &inode->extent_tree; | |
1654 | write_lock(&em_tree->lock); | |
1655 | em = search_extent_mapping(em_tree, ordered->file_offset, | |
1656 | ordered->num_bytes); | |
1657 | em->block_start = *logical; | |
1658 | free_extent_map(em); | |
1659 | write_unlock(&em_tree->lock); | |
1660 | ||
1661 | list_for_each_entry(sum, &ordered->list, list) { | |
1662 | if (*logical < orig_logical) | |
1663 | sum->bytenr -= orig_logical - *logical; | |
1664 | else | |
1665 | sum->bytenr += *logical - orig_logical; | |
1666 | } | |
1667 | ||
1668 | out: | |
1669 | kfree(logical); | |
d8e3fb10 | 1670 | } |
0bc09ca1 NA |
1671 | |
1672 | bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info, | |
1673 | struct extent_buffer *eb, | |
1674 | struct btrfs_block_group **cache_ret) | |
1675 | { | |
1676 | struct btrfs_block_group *cache; | |
1677 | bool ret = true; | |
1678 | ||
1679 | if (!btrfs_is_zoned(fs_info)) | |
1680 | return true; | |
1681 | ||
8fdf54fe JT |
1682 | cache = btrfs_lookup_block_group(fs_info, eb->start); |
1683 | if (!cache) | |
1684 | return true; | |
0bc09ca1 | 1685 | |
8fdf54fe | 1686 | if (cache->meta_write_pointer != eb->start) { |
0bc09ca1 NA |
1687 | btrfs_put_block_group(cache); |
1688 | cache = NULL; | |
8fdf54fe JT |
1689 | ret = false; |
1690 | } else { | |
1691 | cache->meta_write_pointer = eb->start + eb->len; | |
0bc09ca1 NA |
1692 | } |
1693 | ||
8fdf54fe | 1694 | *cache_ret = cache; |
0bc09ca1 NA |
1695 | |
1696 | return ret; | |
1697 | } | |
1698 | ||
1699 | void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache, | |
1700 | struct extent_buffer *eb) | |
1701 | { | |
1702 | if (!btrfs_is_zoned(eb->fs_info) || !cache) | |
1703 | return; | |
1704 | ||
1705 | ASSERT(cache->meta_write_pointer == eb->start + eb->len); | |
1706 | cache->meta_write_pointer = eb->start; | |
1707 | } | |
de17addc NA |
1708 | |
1709 | int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length) | |
1710 | { | |
1711 | if (!btrfs_dev_is_sequential(device, physical)) | |
1712 | return -EOPNOTSUPP; | |
1713 | ||
1714 | return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT, | |
1715 | length >> SECTOR_SHIFT, GFP_NOFS, 0); | |
1716 | } | |
7db1c5d1 NA |
1717 | |
1718 | static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical, | |
1719 | struct blk_zone *zone) | |
1720 | { | |
4c664611 | 1721 | struct btrfs_io_context *bioc = NULL; |
7db1c5d1 NA |
1722 | u64 mapped_length = PAGE_SIZE; |
1723 | unsigned int nofs_flag; | |
1724 | int nmirrors; | |
1725 | int i, ret; | |
1726 | ||
1727 | ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical, | |
4c664611 QW |
1728 | &mapped_length, &bioc); |
1729 | if (ret || !bioc || mapped_length < PAGE_SIZE) { | |
1730 | btrfs_put_bioc(bioc); | |
7db1c5d1 NA |
1731 | return -EIO; |
1732 | } | |
1733 | ||
4c664611 | 1734 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
7db1c5d1 NA |
1735 | return -EINVAL; |
1736 | ||
1737 | nofs_flag = memalloc_nofs_save(); | |
4c664611 | 1738 | nmirrors = (int)bioc->num_stripes; |
7db1c5d1 | 1739 | for (i = 0; i < nmirrors; i++) { |
4c664611 QW |
1740 | u64 physical = bioc->stripes[i].physical; |
1741 | struct btrfs_device *dev = bioc->stripes[i].dev; | |
7db1c5d1 NA |
1742 | |
1743 | /* Missing device */ | |
1744 | if (!dev->bdev) | |
1745 | continue; | |
1746 | ||
1747 | ret = btrfs_get_dev_zone(dev, physical, zone); | |
1748 | /* Failing device */ | |
1749 | if (ret == -EIO || ret == -EOPNOTSUPP) | |
1750 | continue; | |
1751 | break; | |
1752 | } | |
1753 | memalloc_nofs_restore(nofs_flag); | |
1754 | ||
1755 | return ret; | |
1756 | } | |
1757 | ||
1758 | /* | |
1759 | * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by | |
1760 | * filling zeros between @physical_pos to a write pointer of dev-replace | |
1761 | * source device. | |
1762 | */ | |
1763 | int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical, | |
1764 | u64 physical_start, u64 physical_pos) | |
1765 | { | |
1766 | struct btrfs_fs_info *fs_info = tgt_dev->fs_info; | |
1767 | struct blk_zone zone; | |
1768 | u64 length; | |
1769 | u64 wp; | |
1770 | int ret; | |
1771 | ||
1772 | if (!btrfs_dev_is_sequential(tgt_dev, physical_pos)) | |
1773 | return 0; | |
1774 | ||
1775 | ret = read_zone_info(fs_info, logical, &zone); | |
1776 | if (ret) | |
1777 | return ret; | |
1778 | ||
1779 | wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT); | |
1780 | ||
1781 | if (physical_pos == wp) | |
1782 | return 0; | |
1783 | ||
1784 | if (physical_pos > wp) | |
1785 | return -EUCLEAN; | |
1786 | ||
1787 | length = wp - physical_pos; | |
1788 | return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length); | |
1789 | } | |
e7ff9e6b JT |
1790 | |
1791 | struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info, | |
1792 | u64 logical, u64 length) | |
1793 | { | |
1794 | struct btrfs_device *device; | |
1795 | struct extent_map *em; | |
1796 | struct map_lookup *map; | |
1797 | ||
1798 | em = btrfs_get_chunk_map(fs_info, logical, length); | |
1799 | if (IS_ERR(em)) | |
1800 | return ERR_CAST(em); | |
1801 | ||
1802 | map = em->map_lookup; | |
1803 | /* We only support single profile for now */ | |
e7ff9e6b JT |
1804 | device = map->stripes[0].dev; |
1805 | ||
1806 | free_extent_map(em); | |
1807 | ||
1808 | return device; | |
1809 | } | |
afba2bc0 NA |
1810 | |
1811 | /** | |
1812 | * Activate block group and underlying device zones | |
1813 | * | |
1814 | * @block_group: the block group to activate | |
1815 | * | |
1816 | * Return: true on success, false otherwise | |
1817 | */ | |
1818 | bool btrfs_zone_activate(struct btrfs_block_group *block_group) | |
1819 | { | |
1820 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
1821 | struct map_lookup *map; | |
1822 | struct btrfs_device *device; | |
1823 | u64 physical; | |
1824 | bool ret; | |
f9a912a3 | 1825 | int i; |
afba2bc0 NA |
1826 | |
1827 | if (!btrfs_is_zoned(block_group->fs_info)) | |
1828 | return true; | |
1829 | ||
1830 | map = block_group->physical_map; | |
afba2bc0 NA |
1831 | |
1832 | spin_lock(&block_group->lock); | |
afba2bc0 NA |
1833 | if (block_group->zone_is_active) { |
1834 | ret = true; | |
1835 | goto out_unlock; | |
1836 | } | |
1837 | ||
54957712 | 1838 | /* No space left */ |
1bfd4767 | 1839 | if (btrfs_zoned_bg_is_full(block_group)) { |
54957712 NA |
1840 | ret = false; |
1841 | goto out_unlock; | |
1842 | } | |
1843 | ||
f9a912a3 JT |
1844 | for (i = 0; i < map->num_stripes; i++) { |
1845 | device = map->stripes[i].dev; | |
1846 | physical = map->stripes[i].physical; | |
afba2bc0 | 1847 | |
f9a912a3 JT |
1848 | if (device->zone_info->max_active_zones == 0) |
1849 | continue; | |
1850 | ||
f9a912a3 JT |
1851 | if (!btrfs_dev_set_active_zone(device, physical)) { |
1852 | /* Cannot activate the zone */ | |
1853 | ret = false; | |
1854 | goto out_unlock; | |
1855 | } | |
f9a912a3 | 1856 | } |
ceb4f608 NA |
1857 | |
1858 | /* Successfully activated all the zones */ | |
1859 | block_group->zone_is_active = 1; | |
afba2bc0 NA |
1860 | spin_unlock(&block_group->lock); |
1861 | ||
ceb4f608 NA |
1862 | /* For the active block group list */ |
1863 | btrfs_get_block_group(block_group); | |
afba2bc0 | 1864 | |
ceb4f608 NA |
1865 | spin_lock(&fs_info->zone_active_bgs_lock); |
1866 | list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs); | |
1867 | spin_unlock(&fs_info->zone_active_bgs_lock); | |
afba2bc0 NA |
1868 | |
1869 | return true; | |
1870 | ||
1871 | out_unlock: | |
1872 | spin_unlock(&block_group->lock); | |
1873 | return ret; | |
1874 | } | |
1875 | ||
d70cbdda | 1876 | static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written) |
afba2bc0 NA |
1877 | { |
1878 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
1879 | struct map_lookup *map; | |
d70cbdda | 1880 | bool need_zone_finish; |
afba2bc0 | 1881 | int ret = 0; |
4dcbb8ab | 1882 | int i; |
afba2bc0 | 1883 | |
afba2bc0 NA |
1884 | spin_lock(&block_group->lock); |
1885 | if (!block_group->zone_is_active) { | |
1886 | spin_unlock(&block_group->lock); | |
1887 | return 0; | |
1888 | } | |
1889 | ||
1890 | /* Check if we have unwritten allocated space */ | |
1891 | if ((block_group->flags & | |
1892 | (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)) && | |
1893 | block_group->alloc_offset > block_group->meta_write_pointer) { | |
1894 | spin_unlock(&block_group->lock); | |
1895 | return -EAGAIN; | |
1896 | } | |
afba2bc0 NA |
1897 | |
1898 | /* | |
d70cbdda NA |
1899 | * If we are sure that the block group is full (= no more room left for |
1900 | * new allocation) and the IO for the last usable block is completed, we | |
1901 | * don't need to wait for the other IOs. This holds because we ensure | |
1902 | * the sequential IO submissions using the ZONE_APPEND command for data | |
1903 | * and block_group->meta_write_pointer for metadata. | |
afba2bc0 | 1904 | */ |
d70cbdda | 1905 | if (!fully_written) { |
afba2bc0 | 1906 | spin_unlock(&block_group->lock); |
afba2bc0 | 1907 | |
d70cbdda NA |
1908 | ret = btrfs_inc_block_group_ro(block_group, false); |
1909 | if (ret) | |
1910 | return ret; | |
1911 | ||
1912 | /* Ensure all writes in this block group finish */ | |
1913 | btrfs_wait_block_group_reservations(block_group); | |
1914 | /* No need to wait for NOCOW writers. Zoned mode does not allow that */ | |
1915 | btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start, | |
1916 | block_group->length); | |
1917 | ||
1918 | spin_lock(&block_group->lock); | |
1919 | ||
1920 | /* | |
1921 | * Bail out if someone already deactivated the block group, or | |
1922 | * allocated space is left in the block group. | |
1923 | */ | |
1924 | if (!block_group->zone_is_active) { | |
1925 | spin_unlock(&block_group->lock); | |
1926 | btrfs_dec_block_group_ro(block_group); | |
1927 | return 0; | |
1928 | } | |
1929 | ||
1930 | if (block_group->reserved) { | |
1931 | spin_unlock(&block_group->lock); | |
1932 | btrfs_dec_block_group_ro(block_group); | |
1933 | return -EAGAIN; | |
1934 | } | |
afba2bc0 NA |
1935 | } |
1936 | ||
d70cbdda NA |
1937 | /* |
1938 | * The block group is not fully allocated, so not fully written yet. We | |
1939 | * need to send ZONE_FINISH command to free up an active zone. | |
1940 | */ | |
1941 | need_zone_finish = !btrfs_zoned_bg_is_full(block_group); | |
1942 | ||
afba2bc0 NA |
1943 | block_group->zone_is_active = 0; |
1944 | block_group->alloc_offset = block_group->zone_capacity; | |
1945 | block_group->free_space_ctl->free_space = 0; | |
1946 | btrfs_clear_treelog_bg(block_group); | |
5911f538 | 1947 | btrfs_clear_data_reloc_bg(block_group); |
afba2bc0 NA |
1948 | spin_unlock(&block_group->lock); |
1949 | ||
d70cbdda | 1950 | map = block_group->physical_map; |
4dcbb8ab | 1951 | for (i = 0; i < map->num_stripes; i++) { |
d70cbdda NA |
1952 | struct btrfs_device *device = map->stripes[i].dev; |
1953 | const u64 physical = map->stripes[i].physical; | |
afba2bc0 | 1954 | |
4dcbb8ab JT |
1955 | if (device->zone_info->max_active_zones == 0) |
1956 | continue; | |
afba2bc0 | 1957 | |
d70cbdda NA |
1958 | if (need_zone_finish) { |
1959 | ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH, | |
1960 | physical >> SECTOR_SHIFT, | |
1961 | device->zone_info->zone_size >> SECTOR_SHIFT, | |
1962 | GFP_NOFS); | |
4dcbb8ab | 1963 | |
d70cbdda NA |
1964 | if (ret) |
1965 | return ret; | |
1966 | } | |
afba2bc0 | 1967 | |
4dcbb8ab | 1968 | btrfs_dev_clear_active_zone(device, physical); |
afba2bc0 | 1969 | } |
d70cbdda NA |
1970 | |
1971 | if (!fully_written) | |
1972 | btrfs_dec_block_group_ro(block_group); | |
afba2bc0 | 1973 | |
4dcbb8ab JT |
1974 | spin_lock(&fs_info->zone_active_bgs_lock); |
1975 | ASSERT(!list_empty(&block_group->active_bg_list)); | |
1976 | list_del_init(&block_group->active_bg_list); | |
1977 | spin_unlock(&fs_info->zone_active_bgs_lock); | |
1978 | ||
1979 | /* For active_bg_list */ | |
1980 | btrfs_put_block_group(block_group); | |
1981 | ||
1982 | return 0; | |
afba2bc0 | 1983 | } |
a85f05e5 | 1984 | |
d70cbdda NA |
1985 | int btrfs_zone_finish(struct btrfs_block_group *block_group) |
1986 | { | |
1987 | if (!btrfs_is_zoned(block_group->fs_info)) | |
1988 | return 0; | |
1989 | ||
1990 | return do_zone_finish(block_group, false); | |
1991 | } | |
1992 | ||
82187d2e | 1993 | bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags) |
a85f05e5 | 1994 | { |
0b9e6676 | 1995 | struct btrfs_fs_info *fs_info = fs_devices->fs_info; |
a85f05e5 NA |
1996 | struct btrfs_device *device; |
1997 | bool ret = false; | |
1998 | ||
0b9e6676 | 1999 | if (!btrfs_is_zoned(fs_info)) |
a85f05e5 NA |
2000 | return true; |
2001 | ||
a85f05e5 | 2002 | /* Check if there is a device with active zones left */ |
0b9e6676 JT |
2003 | mutex_lock(&fs_info->chunk_mutex); |
2004 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
a85f05e5 NA |
2005 | struct btrfs_zoned_device_info *zinfo = device->zone_info; |
2006 | ||
2007 | if (!device->bdev) | |
2008 | continue; | |
2009 | ||
2010 | if (!zinfo->max_active_zones || | |
2011 | atomic_read(&zinfo->active_zones_left)) { | |
2012 | ret = true; | |
2013 | break; | |
2014 | } | |
2015 | } | |
0b9e6676 | 2016 | mutex_unlock(&fs_info->chunk_mutex); |
a85f05e5 NA |
2017 | |
2018 | return ret; | |
2019 | } | |
be1a1d7a NA |
2020 | |
2021 | void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length) | |
2022 | { | |
2023 | struct btrfs_block_group *block_group; | |
8b8a5399 | 2024 | u64 min_alloc_bytes; |
be1a1d7a NA |
2025 | |
2026 | if (!btrfs_is_zoned(fs_info)) | |
2027 | return; | |
2028 | ||
2029 | block_group = btrfs_lookup_block_group(fs_info, logical); | |
2030 | ASSERT(block_group); | |
2031 | ||
8b8a5399 NA |
2032 | /* No MIXED_BG on zoned btrfs. */ |
2033 | if (block_group->flags & BTRFS_BLOCK_GROUP_DATA) | |
2034 | min_alloc_bytes = fs_info->sectorsize; | |
2035 | else | |
2036 | min_alloc_bytes = fs_info->nodesize; | |
2037 | ||
2038 | /* Bail out if we can allocate more data from this block group. */ | |
2039 | if (logical + length + min_alloc_bytes <= | |
2040 | block_group->start + block_group->zone_capacity) | |
be1a1d7a NA |
2041 | goto out; |
2042 | ||
d70cbdda | 2043 | do_zone_finish(block_group, true); |
be1a1d7a NA |
2044 | |
2045 | out: | |
2046 | btrfs_put_block_group(block_group); | |
2047 | } | |
c2707a25 JT |
2048 | |
2049 | void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg) | |
2050 | { | |
2051 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
2052 | ||
2053 | spin_lock(&fs_info->relocation_bg_lock); | |
2054 | if (fs_info->data_reloc_bg == bg->start) | |
2055 | fs_info->data_reloc_bg = 0; | |
2056 | spin_unlock(&fs_info->relocation_bg_lock); | |
2057 | } | |
16beac87 NA |
2058 | |
2059 | void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info) | |
2060 | { | |
2061 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
2062 | struct btrfs_device *device; | |
2063 | ||
2064 | if (!btrfs_is_zoned(fs_info)) | |
2065 | return; | |
2066 | ||
2067 | mutex_lock(&fs_devices->device_list_mutex); | |
2068 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
2069 | if (device->zone_info) { | |
2070 | vfree(device->zone_info->zone_cache); | |
2071 | device->zone_info->zone_cache = NULL; | |
2072 | } | |
2073 | } | |
2074 | mutex_unlock(&fs_devices->device_list_mutex); | |
2075 | } | |
3687fcb0 JT |
2076 | |
2077 | bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info) | |
2078 | { | |
2079 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
2080 | struct btrfs_device *device; | |
2081 | u64 used = 0; | |
2082 | u64 total = 0; | |
2083 | u64 factor; | |
2084 | ||
2085 | ASSERT(btrfs_is_zoned(fs_info)); | |
2086 | ||
2087 | if (fs_info->bg_reclaim_threshold == 0) | |
2088 | return false; | |
2089 | ||
2090 | mutex_lock(&fs_devices->device_list_mutex); | |
2091 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
2092 | if (!device->bdev) | |
2093 | continue; | |
2094 | ||
2095 | total += device->disk_total_bytes; | |
2096 | used += device->bytes_used; | |
2097 | } | |
2098 | mutex_unlock(&fs_devices->device_list_mutex); | |
2099 | ||
2100 | factor = div64_u64(used * 100, total); | |
2101 | return factor >= fs_info->bg_reclaim_threshold; | |
2102 | } |