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> |
5b316468 NA |
7 | #include "ctree.h" |
8 | #include "volumes.h" | |
9 | #include "zoned.h" | |
10 | #include "rcu-string.h" | |
1cd6121f | 11 | #include "disk-io.h" |
08e11a3d | 12 | #include "block-group.h" |
d3575156 | 13 | #include "transaction.h" |
6143c23c | 14 | #include "dev-replace.h" |
7db1c5d1 | 15 | #include "space-info.h" |
5b316468 NA |
16 | |
17 | /* Maximum number of zones to report per blkdev_report_zones() call */ | |
18 | #define BTRFS_REPORT_NR_ZONES 4096 | |
08e11a3d NA |
19 | /* Invalid allocation pointer value for missing devices */ |
20 | #define WP_MISSING_DEV ((u64)-1) | |
21 | /* Pseudo write pointer value for conventional zone */ | |
22 | #define WP_CONVENTIONAL ((u64)-2) | |
5b316468 | 23 | |
53b74fa9 NA |
24 | /* |
25 | * Location of the first zone of superblock logging zone pairs. | |
26 | * | |
27 | * - primary superblock: 0B (zone 0) | |
28 | * - first copy: 512G (zone starting at that offset) | |
29 | * - second copy: 4T (zone starting at that offset) | |
30 | */ | |
31 | #define BTRFS_SB_LOG_PRIMARY_OFFSET (0ULL) | |
32 | #define BTRFS_SB_LOG_FIRST_OFFSET (512ULL * SZ_1G) | |
33 | #define BTRFS_SB_LOG_SECOND_OFFSET (4096ULL * SZ_1G) | |
34 | ||
35 | #define BTRFS_SB_LOG_FIRST_SHIFT const_ilog2(BTRFS_SB_LOG_FIRST_OFFSET) | |
36 | #define BTRFS_SB_LOG_SECOND_SHIFT const_ilog2(BTRFS_SB_LOG_SECOND_OFFSET) | |
37 | ||
12659251 NA |
38 | /* Number of superblock log zones */ |
39 | #define BTRFS_NR_SB_LOG_ZONES 2 | |
40 | ||
53b74fa9 NA |
41 | /* |
42 | * Maximum supported zone size. Currently, SMR disks have a zone size of | |
43 | * 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range. We do not | |
44 | * expect the zone size to become larger than 8GiB in the near future. | |
45 | */ | |
46 | #define BTRFS_MAX_ZONE_SIZE SZ_8G | |
47 | ||
5b316468 NA |
48 | static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data) |
49 | { | |
50 | struct blk_zone *zones = data; | |
51 | ||
52 | memcpy(&zones[idx], zone, sizeof(*zone)); | |
53 | ||
54 | return 0; | |
55 | } | |
56 | ||
12659251 NA |
57 | static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones, |
58 | u64 *wp_ret) | |
59 | { | |
60 | bool empty[BTRFS_NR_SB_LOG_ZONES]; | |
61 | bool full[BTRFS_NR_SB_LOG_ZONES]; | |
62 | sector_t sector; | |
63 | ||
64 | ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL && | |
65 | zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL); | |
66 | ||
67 | empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY); | |
68 | empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY); | |
69 | full[0] = (zones[0].cond == BLK_ZONE_COND_FULL); | |
70 | full[1] = (zones[1].cond == BLK_ZONE_COND_FULL); | |
71 | ||
72 | /* | |
73 | * Possible states of log buffer zones | |
74 | * | |
75 | * Empty[0] In use[0] Full[0] | |
76 | * Empty[1] * x 0 | |
77 | * In use[1] 0 x 0 | |
78 | * Full[1] 1 1 C | |
79 | * | |
80 | * Log position: | |
81 | * *: Special case, no superblock is written | |
82 | * 0: Use write pointer of zones[0] | |
83 | * 1: Use write pointer of zones[1] | |
1a9fd417 | 84 | * C: Compare super blocks from zones[0] and zones[1], use the latest |
12659251 NA |
85 | * one determined by generation |
86 | * x: Invalid state | |
87 | */ | |
88 | ||
89 | if (empty[0] && empty[1]) { | |
90 | /* Special case to distinguish no superblock to read */ | |
91 | *wp_ret = zones[0].start << SECTOR_SHIFT; | |
92 | return -ENOENT; | |
93 | } else if (full[0] && full[1]) { | |
94 | /* Compare two super blocks */ | |
95 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
96 | struct page *page[BTRFS_NR_SB_LOG_ZONES]; | |
97 | struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES]; | |
98 | int i; | |
99 | ||
100 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { | |
101 | u64 bytenr; | |
102 | ||
103 | bytenr = ((zones[i].start + zones[i].len) | |
104 | << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE; | |
105 | ||
106 | page[i] = read_cache_page_gfp(mapping, | |
107 | bytenr >> PAGE_SHIFT, GFP_NOFS); | |
108 | if (IS_ERR(page[i])) { | |
109 | if (i == 1) | |
110 | btrfs_release_disk_super(super[0]); | |
111 | return PTR_ERR(page[i]); | |
112 | } | |
113 | super[i] = page_address(page[i]); | |
114 | } | |
115 | ||
116 | if (super[0]->generation > super[1]->generation) | |
117 | sector = zones[1].start; | |
118 | else | |
119 | sector = zones[0].start; | |
120 | ||
121 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) | |
122 | btrfs_release_disk_super(super[i]); | |
123 | } else if (!full[0] && (empty[1] || full[1])) { | |
124 | sector = zones[0].wp; | |
125 | } else if (full[0]) { | |
126 | sector = zones[1].wp; | |
127 | } else { | |
128 | return -EUCLEAN; | |
129 | } | |
130 | *wp_ret = sector << SECTOR_SHIFT; | |
131 | return 0; | |
132 | } | |
133 | ||
134 | /* | |
53b74fa9 | 135 | * Get the first zone number of the superblock mirror |
12659251 NA |
136 | */ |
137 | static inline u32 sb_zone_number(int shift, int mirror) | |
138 | { | |
53b74fa9 | 139 | u64 zone; |
12659251 | 140 | |
53b74fa9 | 141 | ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX); |
12659251 | 142 | switch (mirror) { |
53b74fa9 NA |
143 | case 0: zone = 0; break; |
144 | case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break; | |
145 | case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break; | |
12659251 NA |
146 | } |
147 | ||
53b74fa9 NA |
148 | ASSERT(zone <= U32_MAX); |
149 | ||
150 | return (u32)zone; | |
12659251 NA |
151 | } |
152 | ||
5b434df8 NA |
153 | static inline sector_t zone_start_sector(u32 zone_number, |
154 | struct block_device *bdev) | |
155 | { | |
156 | return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev)); | |
157 | } | |
158 | ||
159 | static inline u64 zone_start_physical(u32 zone_number, | |
160 | struct btrfs_zoned_device_info *zone_info) | |
161 | { | |
162 | return (u64)zone_number << zone_info->zone_size_shift; | |
163 | } | |
164 | ||
3c9daa09 JT |
165 | /* |
166 | * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block | |
167 | * device into static sized chunks and fake a conventional zone on each of | |
168 | * them. | |
169 | */ | |
170 | static int emulate_report_zones(struct btrfs_device *device, u64 pos, | |
171 | struct blk_zone *zones, unsigned int nr_zones) | |
172 | { | |
173 | const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT; | |
174 | sector_t bdev_size = bdev_nr_sectors(device->bdev); | |
175 | unsigned int i; | |
176 | ||
177 | pos >>= SECTOR_SHIFT; | |
178 | for (i = 0; i < nr_zones; i++) { | |
179 | zones[i].start = i * zone_sectors + pos; | |
180 | zones[i].len = zone_sectors; | |
181 | zones[i].capacity = zone_sectors; | |
182 | zones[i].wp = zones[i].start + zone_sectors; | |
183 | zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL; | |
184 | zones[i].cond = BLK_ZONE_COND_NOT_WP; | |
185 | ||
186 | if (zones[i].wp >= bdev_size) { | |
187 | i++; | |
188 | break; | |
189 | } | |
190 | } | |
191 | ||
192 | return i; | |
193 | } | |
194 | ||
5b316468 NA |
195 | static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos, |
196 | struct blk_zone *zones, unsigned int *nr_zones) | |
197 | { | |
198 | int ret; | |
199 | ||
200 | if (!*nr_zones) | |
201 | return 0; | |
202 | ||
3c9daa09 JT |
203 | if (!bdev_is_zoned(device->bdev)) { |
204 | ret = emulate_report_zones(device, pos, zones, *nr_zones); | |
205 | *nr_zones = ret; | |
206 | return 0; | |
207 | } | |
208 | ||
5b316468 NA |
209 | ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones, |
210 | copy_zone_info_cb, zones); | |
211 | if (ret < 0) { | |
212 | btrfs_err_in_rcu(device->fs_info, | |
213 | "zoned: failed to read zone %llu on %s (devid %llu)", | |
214 | pos, rcu_str_deref(device->name), | |
215 | device->devid); | |
216 | return ret; | |
217 | } | |
218 | *nr_zones = ret; | |
219 | if (!ret) | |
220 | return -EIO; | |
221 | ||
222 | return 0; | |
223 | } | |
224 | ||
3c9daa09 JT |
225 | /* The emulated zone size is determined from the size of device extent */ |
226 | static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info) | |
227 | { | |
228 | struct btrfs_path *path; | |
229 | struct btrfs_root *root = fs_info->dev_root; | |
230 | struct btrfs_key key; | |
231 | struct extent_buffer *leaf; | |
232 | struct btrfs_dev_extent *dext; | |
233 | int ret = 0; | |
234 | ||
235 | key.objectid = 1; | |
236 | key.type = BTRFS_DEV_EXTENT_KEY; | |
237 | key.offset = 0; | |
238 | ||
239 | path = btrfs_alloc_path(); | |
240 | if (!path) | |
241 | return -ENOMEM; | |
242 | ||
243 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
244 | if (ret < 0) | |
245 | goto out; | |
246 | ||
247 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
ad9a9378 | 248 | ret = btrfs_next_leaf(root, path); |
3c9daa09 JT |
249 | if (ret < 0) |
250 | goto out; | |
251 | /* No dev extents at all? Not good */ | |
252 | if (ret > 0) { | |
253 | ret = -EUCLEAN; | |
254 | goto out; | |
255 | } | |
256 | } | |
257 | ||
258 | leaf = path->nodes[0]; | |
259 | dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); | |
260 | fs_info->zone_size = btrfs_dev_extent_length(leaf, dext); | |
261 | ret = 0; | |
262 | ||
263 | out: | |
264 | btrfs_free_path(path); | |
265 | ||
266 | return ret; | |
267 | } | |
268 | ||
73651042 NA |
269 | int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info) |
270 | { | |
271 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
272 | struct btrfs_device *device; | |
273 | int ret = 0; | |
274 | ||
275 | /* fs_info->zone_size might not set yet. Use the incomapt flag here. */ | |
276 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
277 | return 0; | |
278 | ||
279 | mutex_lock(&fs_devices->device_list_mutex); | |
280 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
281 | /* We can skip reading of zone info for missing devices */ | |
282 | if (!device->bdev) | |
283 | continue; | |
284 | ||
285 | ret = btrfs_get_dev_zone_info(device); | |
286 | if (ret) | |
287 | break; | |
288 | } | |
289 | mutex_unlock(&fs_devices->device_list_mutex); | |
290 | ||
291 | return ret; | |
292 | } | |
293 | ||
5b316468 NA |
294 | int btrfs_get_dev_zone_info(struct btrfs_device *device) |
295 | { | |
3c9daa09 | 296 | struct btrfs_fs_info *fs_info = device->fs_info; |
5b316468 NA |
297 | struct btrfs_zoned_device_info *zone_info = NULL; |
298 | struct block_device *bdev = device->bdev; | |
299 | sector_t nr_sectors; | |
300 | sector_t sector = 0; | |
301 | struct blk_zone *zones = NULL; | |
302 | unsigned int i, nreported = 0, nr_zones; | |
d734492a | 303 | sector_t zone_sectors; |
3c9daa09 | 304 | char *model, *emulated; |
5b316468 NA |
305 | int ret; |
306 | ||
3c9daa09 JT |
307 | /* |
308 | * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not | |
309 | * yet be set. | |
310 | */ | |
311 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
5b316468 NA |
312 | return 0; |
313 | ||
314 | if (device->zone_info) | |
315 | return 0; | |
316 | ||
317 | zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL); | |
318 | if (!zone_info) | |
319 | return -ENOMEM; | |
320 | ||
3c9daa09 JT |
321 | if (!bdev_is_zoned(bdev)) { |
322 | if (!fs_info->zone_size) { | |
323 | ret = calculate_emulated_zone_size(fs_info); | |
324 | if (ret) | |
325 | goto out; | |
326 | } | |
327 | ||
328 | ASSERT(fs_info->zone_size); | |
329 | zone_sectors = fs_info->zone_size >> SECTOR_SHIFT; | |
330 | } else { | |
331 | zone_sectors = bdev_zone_sectors(bdev); | |
332 | } | |
333 | ||
5b316468 NA |
334 | /* Check if it's power of 2 (see is_power_of_2) */ |
335 | ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0); | |
336 | zone_info->zone_size = zone_sectors << SECTOR_SHIFT; | |
53b74fa9 NA |
337 | |
338 | /* We reject devices with a zone size larger than 8GB */ | |
339 | if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) { | |
340 | btrfs_err_in_rcu(fs_info, | |
341 | "zoned: %s: zone size %llu larger than supported maximum %llu", | |
342 | rcu_str_deref(device->name), | |
343 | zone_info->zone_size, BTRFS_MAX_ZONE_SIZE); | |
344 | ret = -EINVAL; | |
345 | goto out; | |
346 | } | |
347 | ||
348 | nr_sectors = bdev_nr_sectors(bdev); | |
5b316468 NA |
349 | zone_info->zone_size_shift = ilog2(zone_info->zone_size); |
350 | zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors); | |
351 | if (!IS_ALIGNED(nr_sectors, zone_sectors)) | |
352 | zone_info->nr_zones++; | |
353 | ||
354 | zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
355 | if (!zone_info->seq_zones) { | |
356 | ret = -ENOMEM; | |
357 | goto out; | |
358 | } | |
359 | ||
360 | zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
361 | if (!zone_info->empty_zones) { | |
362 | ret = -ENOMEM; | |
363 | goto out; | |
364 | } | |
365 | ||
366 | zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL); | |
367 | if (!zones) { | |
368 | ret = -ENOMEM; | |
369 | goto out; | |
370 | } | |
371 | ||
372 | /* Get zones type */ | |
373 | while (sector < nr_sectors) { | |
374 | nr_zones = BTRFS_REPORT_NR_ZONES; | |
375 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones, | |
376 | &nr_zones); | |
377 | if (ret) | |
378 | goto out; | |
379 | ||
380 | for (i = 0; i < nr_zones; i++) { | |
381 | if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ) | |
382 | __set_bit(nreported, zone_info->seq_zones); | |
383 | if (zones[i].cond == BLK_ZONE_COND_EMPTY) | |
384 | __set_bit(nreported, zone_info->empty_zones); | |
385 | nreported++; | |
386 | } | |
387 | sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len; | |
388 | } | |
389 | ||
390 | if (nreported != zone_info->nr_zones) { | |
391 | btrfs_err_in_rcu(device->fs_info, | |
392 | "inconsistent number of zones on %s (%u/%u)", | |
393 | rcu_str_deref(device->name), nreported, | |
394 | zone_info->nr_zones); | |
395 | ret = -EIO; | |
396 | goto out; | |
397 | } | |
398 | ||
12659251 NA |
399 | /* Validate superblock log */ |
400 | nr_zones = BTRFS_NR_SB_LOG_ZONES; | |
401 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
402 | u32 sb_zone; | |
403 | u64 sb_wp; | |
404 | int sb_pos = BTRFS_NR_SB_LOG_ZONES * i; | |
405 | ||
406 | sb_zone = sb_zone_number(zone_info->zone_size_shift, i); | |
407 | if (sb_zone + 1 >= zone_info->nr_zones) | |
408 | continue; | |
409 | ||
5b434df8 NA |
410 | ret = btrfs_get_dev_zones(device, |
411 | zone_start_physical(sb_zone, zone_info), | |
12659251 NA |
412 | &zone_info->sb_zones[sb_pos], |
413 | &nr_zones); | |
414 | if (ret) | |
415 | goto out; | |
416 | ||
417 | if (nr_zones != BTRFS_NR_SB_LOG_ZONES) { | |
418 | btrfs_err_in_rcu(device->fs_info, | |
419 | "zoned: failed to read super block log zone info at devid %llu zone %u", | |
420 | device->devid, sb_zone); | |
421 | ret = -EUCLEAN; | |
422 | goto out; | |
423 | } | |
424 | ||
425 | /* | |
1a9fd417 | 426 | * If zones[0] is conventional, always use the beginning of the |
12659251 NA |
427 | * zone to record superblock. No need to validate in that case. |
428 | */ | |
429 | if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type == | |
430 | BLK_ZONE_TYPE_CONVENTIONAL) | |
431 | continue; | |
432 | ||
433 | ret = sb_write_pointer(device->bdev, | |
434 | &zone_info->sb_zones[sb_pos], &sb_wp); | |
435 | if (ret != -ENOENT && ret) { | |
436 | btrfs_err_in_rcu(device->fs_info, | |
437 | "zoned: super block log zone corrupted devid %llu zone %u", | |
438 | device->devid, sb_zone); | |
439 | ret = -EUCLEAN; | |
440 | goto out; | |
441 | } | |
442 | } | |
443 | ||
444 | ||
5b316468 NA |
445 | kfree(zones); |
446 | ||
447 | device->zone_info = zone_info; | |
448 | ||
3c9daa09 JT |
449 | switch (bdev_zoned_model(bdev)) { |
450 | case BLK_ZONED_HM: | |
451 | model = "host-managed zoned"; | |
452 | emulated = ""; | |
453 | break; | |
454 | case BLK_ZONED_HA: | |
455 | model = "host-aware zoned"; | |
456 | emulated = ""; | |
457 | break; | |
458 | case BLK_ZONED_NONE: | |
459 | model = "regular"; | |
460 | emulated = "emulated "; | |
461 | break; | |
462 | default: | |
463 | /* Just in case */ | |
464 | btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s", | |
465 | bdev_zoned_model(bdev), | |
466 | rcu_str_deref(device->name)); | |
467 | ret = -EOPNOTSUPP; | |
468 | goto out_free_zone_info; | |
469 | } | |
470 | ||
471 | btrfs_info_in_rcu(fs_info, | |
472 | "%s block device %s, %u %szones of %llu bytes", | |
473 | model, rcu_str_deref(device->name), zone_info->nr_zones, | |
474 | emulated, zone_info->zone_size); | |
5b316468 NA |
475 | |
476 | return 0; | |
477 | ||
478 | out: | |
479 | kfree(zones); | |
3c9daa09 | 480 | out_free_zone_info: |
5b316468 NA |
481 | bitmap_free(zone_info->empty_zones); |
482 | bitmap_free(zone_info->seq_zones); | |
483 | kfree(zone_info); | |
3c9daa09 | 484 | device->zone_info = NULL; |
5b316468 NA |
485 | |
486 | return ret; | |
487 | } | |
488 | ||
489 | void btrfs_destroy_dev_zone_info(struct btrfs_device *device) | |
490 | { | |
491 | struct btrfs_zoned_device_info *zone_info = device->zone_info; | |
492 | ||
493 | if (!zone_info) | |
494 | return; | |
495 | ||
496 | bitmap_free(zone_info->seq_zones); | |
497 | bitmap_free(zone_info->empty_zones); | |
498 | kfree(zone_info); | |
499 | device->zone_info = NULL; | |
500 | } | |
501 | ||
502 | int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos, | |
503 | struct blk_zone *zone) | |
504 | { | |
505 | unsigned int nr_zones = 1; | |
506 | int ret; | |
507 | ||
508 | ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones); | |
509 | if (ret != 0 || !nr_zones) | |
510 | return ret ? ret : -EIO; | |
511 | ||
512 | return 0; | |
513 | } | |
b70f5097 NA |
514 | |
515 | int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info) | |
516 | { | |
517 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
518 | struct btrfs_device *device; | |
519 | u64 zoned_devices = 0; | |
520 | u64 nr_devices = 0; | |
521 | u64 zone_size = 0; | |
3c9daa09 | 522 | const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED); |
b70f5097 NA |
523 | int ret = 0; |
524 | ||
525 | /* Count zoned devices */ | |
526 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
527 | enum blk_zoned_model model; | |
528 | ||
529 | if (!device->bdev) | |
530 | continue; | |
531 | ||
532 | model = bdev_zoned_model(device->bdev); | |
3c9daa09 JT |
533 | /* |
534 | * A Host-Managed zoned device must be used as a zoned device. | |
535 | * A Host-Aware zoned device and a non-zoned devices can be | |
536 | * treated as a zoned device, if ZONED flag is enabled in the | |
537 | * superblock. | |
538 | */ | |
b70f5097 | 539 | if (model == BLK_ZONED_HM || |
3c9daa09 JT |
540 | (model == BLK_ZONED_HA && incompat_zoned) || |
541 | (model == BLK_ZONED_NONE && incompat_zoned)) { | |
542 | struct btrfs_zoned_device_info *zone_info = | |
543 | device->zone_info; | |
862931c7 NA |
544 | |
545 | zone_info = device->zone_info; | |
b70f5097 NA |
546 | zoned_devices++; |
547 | if (!zone_size) { | |
862931c7 NA |
548 | zone_size = zone_info->zone_size; |
549 | } else if (zone_info->zone_size != zone_size) { | |
b70f5097 NA |
550 | btrfs_err(fs_info, |
551 | "zoned: unequal block device zone sizes: have %llu found %llu", | |
552 | device->zone_info->zone_size, | |
553 | zone_size); | |
554 | ret = -EINVAL; | |
555 | goto out; | |
556 | } | |
557 | } | |
558 | nr_devices++; | |
559 | } | |
560 | ||
561 | if (!zoned_devices && !incompat_zoned) | |
562 | goto out; | |
563 | ||
564 | if (!zoned_devices && incompat_zoned) { | |
565 | /* No zoned block device found on ZONED filesystem */ | |
566 | btrfs_err(fs_info, | |
567 | "zoned: no zoned devices found on a zoned filesystem"); | |
568 | ret = -EINVAL; | |
569 | goto out; | |
570 | } | |
571 | ||
572 | if (zoned_devices && !incompat_zoned) { | |
573 | btrfs_err(fs_info, | |
574 | "zoned: mode not enabled but zoned device found"); | |
575 | ret = -EINVAL; | |
576 | goto out; | |
577 | } | |
578 | ||
579 | if (zoned_devices != nr_devices) { | |
580 | btrfs_err(fs_info, | |
581 | "zoned: cannot mix zoned and regular devices"); | |
582 | ret = -EINVAL; | |
583 | goto out; | |
584 | } | |
585 | ||
586 | /* | |
587 | * stripe_size is always aligned to BTRFS_STRIPE_LEN in | |
f6f39f7a | 588 | * btrfs_create_chunk(). Since we want stripe_len == zone_size, |
b70f5097 NA |
589 | * check the alignment here. |
590 | */ | |
591 | if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) { | |
592 | btrfs_err(fs_info, | |
593 | "zoned: zone size %llu not aligned to stripe %u", | |
594 | zone_size, BTRFS_STRIPE_LEN); | |
595 | ret = -EINVAL; | |
596 | goto out; | |
597 | } | |
598 | ||
a589dde0 NA |
599 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
600 | btrfs_err(fs_info, "zoned: mixed block groups not supported"); | |
601 | ret = -EINVAL; | |
602 | goto out; | |
603 | } | |
604 | ||
b70f5097 | 605 | fs_info->zone_size = zone_size; |
1cd6121f | 606 | fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED; |
b70f5097 | 607 | |
b53429ba JT |
608 | /* |
609 | * Check mount options here, because we might change fs_info->zoned | |
610 | * from fs_info->zone_size. | |
611 | */ | |
612 | ret = btrfs_check_mountopts_zoned(fs_info); | |
613 | if (ret) | |
614 | goto out; | |
615 | ||
b70f5097 NA |
616 | btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size); |
617 | out: | |
618 | return ret; | |
619 | } | |
5d1ab66c NA |
620 | |
621 | int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info) | |
622 | { | |
623 | if (!btrfs_is_zoned(info)) | |
624 | return 0; | |
625 | ||
626 | /* | |
627 | * Space cache writing is not COWed. Disable that to avoid write errors | |
628 | * in sequential zones. | |
629 | */ | |
630 | if (btrfs_test_opt(info, SPACE_CACHE)) { | |
631 | btrfs_err(info, "zoned: space cache v1 is not supported"); | |
632 | return -EINVAL; | |
633 | } | |
634 | ||
d206e9c9 NA |
635 | if (btrfs_test_opt(info, NODATACOW)) { |
636 | btrfs_err(info, "zoned: NODATACOW not supported"); | |
637 | return -EINVAL; | |
638 | } | |
639 | ||
5d1ab66c NA |
640 | return 0; |
641 | } | |
12659251 NA |
642 | |
643 | static int sb_log_location(struct block_device *bdev, struct blk_zone *zones, | |
644 | int rw, u64 *bytenr_ret) | |
645 | { | |
646 | u64 wp; | |
647 | int ret; | |
648 | ||
649 | if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) { | |
650 | *bytenr_ret = zones[0].start << SECTOR_SHIFT; | |
651 | return 0; | |
652 | } | |
653 | ||
654 | ret = sb_write_pointer(bdev, zones, &wp); | |
655 | if (ret != -ENOENT && ret < 0) | |
656 | return ret; | |
657 | ||
658 | if (rw == WRITE) { | |
659 | struct blk_zone *reset = NULL; | |
660 | ||
661 | if (wp == zones[0].start << SECTOR_SHIFT) | |
662 | reset = &zones[0]; | |
663 | else if (wp == zones[1].start << SECTOR_SHIFT) | |
664 | reset = &zones[1]; | |
665 | ||
666 | if (reset && reset->cond != BLK_ZONE_COND_EMPTY) { | |
667 | ASSERT(reset->cond == BLK_ZONE_COND_FULL); | |
668 | ||
669 | ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
670 | reset->start, reset->len, | |
671 | GFP_NOFS); | |
672 | if (ret) | |
673 | return ret; | |
674 | ||
675 | reset->cond = BLK_ZONE_COND_EMPTY; | |
676 | reset->wp = reset->start; | |
677 | } | |
678 | } else if (ret != -ENOENT) { | |
679 | /* For READ, we want the precious one */ | |
680 | if (wp == zones[0].start << SECTOR_SHIFT) | |
681 | wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT; | |
682 | wp -= BTRFS_SUPER_INFO_SIZE; | |
683 | } | |
684 | ||
685 | *bytenr_ret = wp; | |
686 | return 0; | |
687 | ||
688 | } | |
689 | ||
690 | int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw, | |
691 | u64 *bytenr_ret) | |
692 | { | |
693 | struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES]; | |
d734492a | 694 | sector_t zone_sectors; |
12659251 NA |
695 | u32 sb_zone; |
696 | int ret; | |
12659251 NA |
697 | u8 zone_sectors_shift; |
698 | sector_t nr_sectors; | |
699 | u32 nr_zones; | |
700 | ||
701 | if (!bdev_is_zoned(bdev)) { | |
702 | *bytenr_ret = btrfs_sb_offset(mirror); | |
703 | return 0; | |
704 | } | |
705 | ||
706 | ASSERT(rw == READ || rw == WRITE); | |
707 | ||
708 | zone_sectors = bdev_zone_sectors(bdev); | |
709 | if (!is_power_of_2(zone_sectors)) | |
710 | return -EINVAL; | |
12659251 | 711 | zone_sectors_shift = ilog2(zone_sectors); |
ac7ac461 | 712 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
713 | nr_zones = nr_sectors >> zone_sectors_shift; |
714 | ||
715 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
716 | if (sb_zone + 1 >= nr_zones) | |
717 | return -ENOENT; | |
718 | ||
5b434df8 | 719 | ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev), |
12659251 NA |
720 | BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb, |
721 | zones); | |
722 | if (ret < 0) | |
723 | return ret; | |
724 | if (ret != BTRFS_NR_SB_LOG_ZONES) | |
725 | return -EIO; | |
726 | ||
727 | return sb_log_location(bdev, zones, rw, bytenr_ret); | |
728 | } | |
729 | ||
730 | int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw, | |
731 | u64 *bytenr_ret) | |
732 | { | |
733 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
734 | u32 zone_num; | |
735 | ||
d6639b35 NA |
736 | /* |
737 | * For a zoned filesystem on a non-zoned block device, use the same | |
738 | * super block locations as regular filesystem. Doing so, the super | |
739 | * block can always be retrieved and the zoned flag of the volume | |
740 | * detected from the super block information. | |
741 | */ | |
742 | if (!bdev_is_zoned(device->bdev)) { | |
12659251 NA |
743 | *bytenr_ret = btrfs_sb_offset(mirror); |
744 | return 0; | |
745 | } | |
746 | ||
747 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
748 | if (zone_num + 1 >= zinfo->nr_zones) | |
749 | return -ENOENT; | |
750 | ||
751 | return sb_log_location(device->bdev, | |
752 | &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror], | |
753 | rw, bytenr_ret); | |
754 | } | |
755 | ||
756 | static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo, | |
757 | int mirror) | |
758 | { | |
759 | u32 zone_num; | |
760 | ||
761 | if (!zinfo) | |
762 | return false; | |
763 | ||
764 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
765 | if (zone_num + 1 >= zinfo->nr_zones) | |
766 | return false; | |
767 | ||
768 | if (!test_bit(zone_num, zinfo->seq_zones)) | |
769 | return false; | |
770 | ||
771 | return true; | |
772 | } | |
773 | ||
774 | void btrfs_advance_sb_log(struct btrfs_device *device, int mirror) | |
775 | { | |
776 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
777 | struct blk_zone *zone; | |
778 | ||
779 | if (!is_sb_log_zone(zinfo, mirror)) | |
780 | return; | |
781 | ||
782 | zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror]; | |
783 | if (zone->cond != BLK_ZONE_COND_FULL) { | |
784 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
785 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
786 | ||
787 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
788 | ||
789 | if (zone->wp == zone->start + zone->len) | |
790 | zone->cond = BLK_ZONE_COND_FULL; | |
791 | ||
792 | return; | |
793 | } | |
794 | ||
795 | zone++; | |
796 | ASSERT(zone->cond != BLK_ZONE_COND_FULL); | |
797 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
798 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
799 | ||
800 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
801 | ||
802 | if (zone->wp == zone->start + zone->len) | |
803 | zone->cond = BLK_ZONE_COND_FULL; | |
804 | } | |
805 | ||
806 | int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror) | |
807 | { | |
808 | sector_t zone_sectors; | |
809 | sector_t nr_sectors; | |
810 | u8 zone_sectors_shift; | |
811 | u32 sb_zone; | |
812 | u32 nr_zones; | |
813 | ||
814 | zone_sectors = bdev_zone_sectors(bdev); | |
815 | zone_sectors_shift = ilog2(zone_sectors); | |
ac7ac461 | 816 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
817 | nr_zones = nr_sectors >> zone_sectors_shift; |
818 | ||
819 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
820 | if (sb_zone + 1 >= nr_zones) | |
821 | return -ENOENT; | |
822 | ||
823 | return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
5b434df8 | 824 | zone_start_sector(sb_zone, bdev), |
12659251 NA |
825 | zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS); |
826 | } | |
1cd6121f NA |
827 | |
828 | /** | |
829 | * btrfs_find_allocatable_zones - find allocatable zones within a given region | |
830 | * | |
831 | * @device: the device to allocate a region on | |
832 | * @hole_start: the position of the hole to allocate the region | |
833 | * @num_bytes: size of wanted region | |
834 | * @hole_end: the end of the hole | |
835 | * @return: position of allocatable zones | |
836 | * | |
837 | * Allocatable region should not contain any superblock locations. | |
838 | */ | |
839 | u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start, | |
840 | u64 hole_end, u64 num_bytes) | |
841 | { | |
842 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
843 | const u8 shift = zinfo->zone_size_shift; | |
844 | u64 nzones = num_bytes >> shift; | |
845 | u64 pos = hole_start; | |
846 | u64 begin, end; | |
847 | bool have_sb; | |
848 | int i; | |
849 | ||
850 | ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size)); | |
851 | ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size)); | |
852 | ||
853 | while (pos < hole_end) { | |
854 | begin = pos >> shift; | |
855 | end = begin + nzones; | |
856 | ||
857 | if (end > zinfo->nr_zones) | |
858 | return hole_end; | |
859 | ||
860 | /* Check if zones in the region are all empty */ | |
861 | if (btrfs_dev_is_sequential(device, pos) && | |
862 | find_next_zero_bit(zinfo->empty_zones, end, begin) != end) { | |
863 | pos += zinfo->zone_size; | |
864 | continue; | |
865 | } | |
866 | ||
867 | have_sb = false; | |
868 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
869 | u32 sb_zone; | |
870 | u64 sb_pos; | |
871 | ||
872 | sb_zone = sb_zone_number(shift, i); | |
873 | if (!(end <= sb_zone || | |
874 | sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) { | |
875 | have_sb = true; | |
5b434df8 NA |
876 | pos = zone_start_physical( |
877 | sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo); | |
1cd6121f NA |
878 | break; |
879 | } | |
880 | ||
881 | /* We also need to exclude regular superblock positions */ | |
882 | sb_pos = btrfs_sb_offset(i); | |
883 | if (!(pos + num_bytes <= sb_pos || | |
884 | sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) { | |
885 | have_sb = true; | |
886 | pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE, | |
887 | zinfo->zone_size); | |
888 | break; | |
889 | } | |
890 | } | |
891 | if (!have_sb) | |
892 | break; | |
893 | } | |
894 | ||
895 | return pos; | |
896 | } | |
897 | ||
898 | int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical, | |
899 | u64 length, u64 *bytes) | |
900 | { | |
901 | int ret; | |
902 | ||
903 | *bytes = 0; | |
904 | ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET, | |
905 | physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT, | |
906 | GFP_NOFS); | |
907 | if (ret) | |
908 | return ret; | |
909 | ||
910 | *bytes = length; | |
911 | while (length) { | |
912 | btrfs_dev_set_zone_empty(device, physical); | |
913 | physical += device->zone_info->zone_size; | |
914 | length -= device->zone_info->zone_size; | |
915 | } | |
916 | ||
917 | return 0; | |
918 | } | |
919 | ||
920 | int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size) | |
921 | { | |
922 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
923 | const u8 shift = zinfo->zone_size_shift; | |
924 | unsigned long begin = start >> shift; | |
925 | unsigned long end = (start + size) >> shift; | |
926 | u64 pos; | |
927 | int ret; | |
928 | ||
929 | ASSERT(IS_ALIGNED(start, zinfo->zone_size)); | |
930 | ASSERT(IS_ALIGNED(size, zinfo->zone_size)); | |
931 | ||
932 | if (end > zinfo->nr_zones) | |
933 | return -ERANGE; | |
934 | ||
935 | /* All the zones are conventional */ | |
936 | if (find_next_bit(zinfo->seq_zones, begin, end) == end) | |
937 | return 0; | |
938 | ||
939 | /* All the zones are sequential and empty */ | |
940 | if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end && | |
941 | find_next_zero_bit(zinfo->empty_zones, begin, end) == end) | |
942 | return 0; | |
943 | ||
944 | for (pos = start; pos < start + size; pos += zinfo->zone_size) { | |
945 | u64 reset_bytes; | |
946 | ||
947 | if (!btrfs_dev_is_sequential(device, pos) || | |
948 | btrfs_dev_is_empty_zone(device, pos)) | |
949 | continue; | |
950 | ||
951 | /* Free regions should be empty */ | |
952 | btrfs_warn_in_rcu( | |
953 | device->fs_info, | |
954 | "zoned: resetting device %s (devid %llu) zone %llu for allocation", | |
955 | rcu_str_deref(device->name), device->devid, pos >> shift); | |
956 | WARN_ON_ONCE(1); | |
957 | ||
958 | ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size, | |
959 | &reset_bytes); | |
960 | if (ret) | |
961 | return ret; | |
962 | } | |
963 | ||
964 | return 0; | |
965 | } | |
08e11a3d | 966 | |
a94794d5 NA |
967 | /* |
968 | * Calculate an allocation pointer from the extent allocation information | |
969 | * for a block group consist of conventional zones. It is pointed to the | |
970 | * end of the highest addressed extent in the block group as an allocation | |
971 | * offset. | |
972 | */ | |
973 | static int calculate_alloc_pointer(struct btrfs_block_group *cache, | |
974 | u64 *offset_ret) | |
975 | { | |
976 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
977 | struct btrfs_root *root = fs_info->extent_root; | |
978 | struct btrfs_path *path; | |
979 | struct btrfs_key key; | |
980 | struct btrfs_key found_key; | |
981 | int ret; | |
982 | u64 length; | |
983 | ||
984 | path = btrfs_alloc_path(); | |
985 | if (!path) | |
986 | return -ENOMEM; | |
987 | ||
988 | key.objectid = cache->start + cache->length; | |
989 | key.type = 0; | |
990 | key.offset = 0; | |
991 | ||
992 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
993 | /* We should not find the exact match */ | |
994 | if (!ret) | |
995 | ret = -EUCLEAN; | |
996 | if (ret < 0) | |
997 | goto out; | |
998 | ||
999 | ret = btrfs_previous_extent_item(root, path, cache->start); | |
1000 | if (ret) { | |
1001 | if (ret == 1) { | |
1002 | ret = 0; | |
1003 | *offset_ret = 0; | |
1004 | } | |
1005 | goto out; | |
1006 | } | |
1007 | ||
1008 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
1009 | ||
1010 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY) | |
1011 | length = found_key.offset; | |
1012 | else | |
1013 | length = fs_info->nodesize; | |
1014 | ||
1015 | if (!(found_key.objectid >= cache->start && | |
1016 | found_key.objectid + length <= cache->start + cache->length)) { | |
1017 | ret = -EUCLEAN; | |
1018 | goto out; | |
1019 | } | |
1020 | *offset_ret = found_key.objectid + length - cache->start; | |
1021 | ret = 0; | |
1022 | ||
1023 | out: | |
1024 | btrfs_free_path(path); | |
1025 | return ret; | |
1026 | } | |
1027 | ||
1028 | int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) | |
08e11a3d NA |
1029 | { |
1030 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
1031 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
1032 | struct extent_map *em; | |
1033 | struct map_lookup *map; | |
1034 | struct btrfs_device *device; | |
1035 | u64 logical = cache->start; | |
1036 | u64 length = cache->length; | |
1037 | u64 physical = 0; | |
1038 | int ret; | |
1039 | int i; | |
1040 | unsigned int nofs_flag; | |
1041 | u64 *alloc_offsets = NULL; | |
a94794d5 | 1042 | u64 last_alloc = 0; |
08e11a3d NA |
1043 | u32 num_sequential = 0, num_conventional = 0; |
1044 | ||
1045 | if (!btrfs_is_zoned(fs_info)) | |
1046 | return 0; | |
1047 | ||
1048 | /* Sanity check */ | |
1049 | if (!IS_ALIGNED(length, fs_info->zone_size)) { | |
1050 | btrfs_err(fs_info, | |
1051 | "zoned: block group %llu len %llu unaligned to zone size %llu", | |
1052 | logical, length, fs_info->zone_size); | |
1053 | return -EIO; | |
1054 | } | |
1055 | ||
1056 | /* Get the chunk mapping */ | |
1057 | read_lock(&em_tree->lock); | |
1058 | em = lookup_extent_mapping(em_tree, logical, length); | |
1059 | read_unlock(&em_tree->lock); | |
1060 | ||
1061 | if (!em) | |
1062 | return -EINVAL; | |
1063 | ||
1064 | map = em->map_lookup; | |
1065 | ||
1066 | alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS); | |
1067 | if (!alloc_offsets) { | |
1068 | free_extent_map(em); | |
1069 | return -ENOMEM; | |
1070 | } | |
1071 | ||
1072 | for (i = 0; i < map->num_stripes; i++) { | |
1073 | bool is_sequential; | |
1074 | struct blk_zone zone; | |
6143c23c NA |
1075 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
1076 | int dev_replace_is_ongoing = 0; | |
08e11a3d NA |
1077 | |
1078 | device = map->stripes[i].dev; | |
1079 | physical = map->stripes[i].physical; | |
1080 | ||
1081 | if (device->bdev == NULL) { | |
1082 | alloc_offsets[i] = WP_MISSING_DEV; | |
1083 | continue; | |
1084 | } | |
1085 | ||
1086 | is_sequential = btrfs_dev_is_sequential(device, physical); | |
1087 | if (is_sequential) | |
1088 | num_sequential++; | |
1089 | else | |
1090 | num_conventional++; | |
1091 | ||
1092 | if (!is_sequential) { | |
1093 | alloc_offsets[i] = WP_CONVENTIONAL; | |
1094 | continue; | |
1095 | } | |
1096 | ||
1097 | /* | |
1098 | * This zone will be used for allocation, so mark this zone | |
1099 | * non-empty. | |
1100 | */ | |
1101 | btrfs_dev_clear_zone_empty(device, physical); | |
1102 | ||
6143c23c NA |
1103 | down_read(&dev_replace->rwsem); |
1104 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
1105 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) | |
1106 | btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical); | |
1107 | up_read(&dev_replace->rwsem); | |
1108 | ||
08e11a3d NA |
1109 | /* |
1110 | * The group is mapped to a sequential zone. Get the zone write | |
1111 | * pointer to determine the allocation offset within the zone. | |
1112 | */ | |
1113 | WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size)); | |
1114 | nofs_flag = memalloc_nofs_save(); | |
1115 | ret = btrfs_get_dev_zone(device, physical, &zone); | |
1116 | memalloc_nofs_restore(nofs_flag); | |
1117 | if (ret == -EIO || ret == -EOPNOTSUPP) { | |
1118 | ret = 0; | |
1119 | alloc_offsets[i] = WP_MISSING_DEV; | |
1120 | continue; | |
1121 | } else if (ret) { | |
1122 | goto out; | |
1123 | } | |
1124 | ||
784daf2b | 1125 | if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) { |
47cdfb5e NA |
1126 | btrfs_err_in_rcu(fs_info, |
1127 | "zoned: unexpected conventional zone %llu on device %s (devid %llu)", | |
1128 | zone.start << SECTOR_SHIFT, | |
1129 | rcu_str_deref(device->name), device->devid); | |
784daf2b NA |
1130 | ret = -EIO; |
1131 | goto out; | |
1132 | } | |
1133 | ||
08e11a3d NA |
1134 | switch (zone.cond) { |
1135 | case BLK_ZONE_COND_OFFLINE: | |
1136 | case BLK_ZONE_COND_READONLY: | |
1137 | btrfs_err(fs_info, | |
1138 | "zoned: offline/readonly zone %llu on device %s (devid %llu)", | |
1139 | physical >> device->zone_info->zone_size_shift, | |
1140 | rcu_str_deref(device->name), device->devid); | |
1141 | alloc_offsets[i] = WP_MISSING_DEV; | |
1142 | break; | |
1143 | case BLK_ZONE_COND_EMPTY: | |
1144 | alloc_offsets[i] = 0; | |
1145 | break; | |
1146 | case BLK_ZONE_COND_FULL: | |
1147 | alloc_offsets[i] = fs_info->zone_size; | |
1148 | break; | |
1149 | default: | |
1150 | /* Partially used zone */ | |
1151 | alloc_offsets[i] = | |
1152 | ((zone.wp - zone.start) << SECTOR_SHIFT); | |
1153 | break; | |
1154 | } | |
1155 | } | |
1156 | ||
08f45559 JT |
1157 | if (num_sequential > 0) |
1158 | cache->seq_zone = true; | |
1159 | ||
08e11a3d NA |
1160 | if (num_conventional > 0) { |
1161 | /* | |
a94794d5 NA |
1162 | * Avoid calling calculate_alloc_pointer() for new BG. It |
1163 | * is no use for new BG. It must be always 0. | |
1164 | * | |
1165 | * Also, we have a lock chain of extent buffer lock -> | |
1166 | * chunk mutex. For new BG, this function is called from | |
1167 | * btrfs_make_block_group() which is already taking the | |
1168 | * chunk mutex. Thus, we cannot call | |
1169 | * calculate_alloc_pointer() which takes extent buffer | |
1170 | * locks to avoid deadlock. | |
08e11a3d | 1171 | */ |
a94794d5 NA |
1172 | if (new) { |
1173 | cache->alloc_offset = 0; | |
1174 | goto out; | |
1175 | } | |
1176 | ret = calculate_alloc_pointer(cache, &last_alloc); | |
1177 | if (ret || map->num_stripes == num_conventional) { | |
1178 | if (!ret) | |
1179 | cache->alloc_offset = last_alloc; | |
1180 | else | |
1181 | btrfs_err(fs_info, | |
1182 | "zoned: failed to determine allocation offset of bg %llu", | |
1183 | cache->start); | |
1184 | goto out; | |
1185 | } | |
08e11a3d NA |
1186 | } |
1187 | ||
1188 | switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
1189 | case 0: /* single */ | |
06e1e7f4 JT |
1190 | if (alloc_offsets[0] == WP_MISSING_DEV) { |
1191 | btrfs_err(fs_info, | |
1192 | "zoned: cannot recover write pointer for zone %llu", | |
1193 | physical); | |
1194 | ret = -EIO; | |
1195 | goto out; | |
1196 | } | |
08e11a3d NA |
1197 | cache->alloc_offset = alloc_offsets[0]; |
1198 | break; | |
1199 | case BTRFS_BLOCK_GROUP_DUP: | |
1200 | case BTRFS_BLOCK_GROUP_RAID1: | |
1201 | case BTRFS_BLOCK_GROUP_RAID0: | |
1202 | case BTRFS_BLOCK_GROUP_RAID10: | |
1203 | case BTRFS_BLOCK_GROUP_RAID5: | |
1204 | case BTRFS_BLOCK_GROUP_RAID6: | |
1205 | /* non-single profiles are not supported yet */ | |
1206 | default: | |
1207 | btrfs_err(fs_info, "zoned: profile %s not yet supported", | |
1208 | btrfs_bg_type_to_raid_name(map->type)); | |
1209 | ret = -EINVAL; | |
1210 | goto out; | |
1211 | } | |
1212 | ||
1213 | out: | |
06e1e7f4 JT |
1214 | if (cache->alloc_offset > fs_info->zone_size) { |
1215 | btrfs_err(fs_info, | |
1216 | "zoned: invalid write pointer %llu in block group %llu", | |
1217 | cache->alloc_offset, cache->start); | |
1218 | ret = -EIO; | |
1219 | } | |
1220 | ||
a94794d5 NA |
1221 | /* An extent is allocated after the write pointer */ |
1222 | if (!ret && num_conventional && last_alloc > cache->alloc_offset) { | |
1223 | btrfs_err(fs_info, | |
1224 | "zoned: got wrong write pointer in BG %llu: %llu > %llu", | |
1225 | logical, last_alloc, cache->alloc_offset); | |
1226 | ret = -EIO; | |
1227 | } | |
1228 | ||
0bc09ca1 NA |
1229 | if (!ret) |
1230 | cache->meta_write_pointer = cache->alloc_offset + cache->start; | |
1231 | ||
08e11a3d NA |
1232 | kfree(alloc_offsets); |
1233 | free_extent_map(em); | |
1234 | ||
1235 | return ret; | |
1236 | } | |
169e0da9 NA |
1237 | |
1238 | void btrfs_calc_zone_unusable(struct btrfs_block_group *cache) | |
1239 | { | |
1240 | u64 unusable, free; | |
1241 | ||
1242 | if (!btrfs_is_zoned(cache->fs_info)) | |
1243 | return; | |
1244 | ||
1245 | WARN_ON(cache->bytes_super != 0); | |
1246 | unusable = cache->alloc_offset - cache->used; | |
1247 | free = cache->length - cache->alloc_offset; | |
1248 | ||
1249 | /* We only need ->free_space in ALLOC_SEQ block groups */ | |
1250 | cache->last_byte_to_unpin = (u64)-1; | |
1251 | cache->cached = BTRFS_CACHE_FINISHED; | |
1252 | cache->free_space_ctl->free_space = free; | |
1253 | cache->zone_unusable = unusable; | |
1254 | ||
1255 | /* Should not have any excluded extents. Just in case, though */ | |
1256 | btrfs_free_excluded_extents(cache); | |
1257 | } | |
d3575156 NA |
1258 | |
1259 | void btrfs_redirty_list_add(struct btrfs_transaction *trans, | |
1260 | struct extent_buffer *eb) | |
1261 | { | |
1262 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
1263 | ||
1264 | if (!btrfs_is_zoned(fs_info) || | |
1265 | btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) || | |
1266 | !list_empty(&eb->release_list)) | |
1267 | return; | |
1268 | ||
1269 | set_extent_buffer_dirty(eb); | |
1270 | set_extent_bits_nowait(&trans->dirty_pages, eb->start, | |
1271 | eb->start + eb->len - 1, EXTENT_DIRTY); | |
1272 | memzero_extent_buffer(eb, 0, eb->len); | |
1273 | set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags); | |
1274 | ||
1275 | spin_lock(&trans->releasing_ebs_lock); | |
1276 | list_add_tail(&eb->release_list, &trans->releasing_ebs); | |
1277 | spin_unlock(&trans->releasing_ebs_lock); | |
1278 | atomic_inc(&eb->refs); | |
1279 | } | |
1280 | ||
1281 | void btrfs_free_redirty_list(struct btrfs_transaction *trans) | |
1282 | { | |
1283 | spin_lock(&trans->releasing_ebs_lock); | |
1284 | while (!list_empty(&trans->releasing_ebs)) { | |
1285 | struct extent_buffer *eb; | |
1286 | ||
1287 | eb = list_first_entry(&trans->releasing_ebs, | |
1288 | struct extent_buffer, release_list); | |
1289 | list_del_init(&eb->release_list); | |
1290 | free_extent_buffer(eb); | |
1291 | } | |
1292 | spin_unlock(&trans->releasing_ebs_lock); | |
1293 | } | |
08f45559 | 1294 | |
e380adfc | 1295 | bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start) |
08f45559 JT |
1296 | { |
1297 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1298 | struct btrfs_block_group *cache; | |
1299 | bool ret = false; | |
1300 | ||
1301 | if (!btrfs_is_zoned(fs_info)) | |
1302 | return false; | |
1303 | ||
08f45559 JT |
1304 | if (!is_data_inode(&inode->vfs_inode)) |
1305 | return false; | |
1306 | ||
e380adfc | 1307 | cache = btrfs_lookup_block_group(fs_info, start); |
08f45559 JT |
1308 | ASSERT(cache); |
1309 | if (!cache) | |
1310 | return false; | |
1311 | ||
1312 | ret = cache->seq_zone; | |
1313 | btrfs_put_block_group(cache); | |
1314 | ||
1315 | return ret; | |
1316 | } | |
d8e3fb10 NA |
1317 | |
1318 | void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset, | |
1319 | struct bio *bio) | |
1320 | { | |
1321 | struct btrfs_ordered_extent *ordered; | |
1322 | const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; | |
1323 | ||
1324 | if (bio_op(bio) != REQ_OP_ZONE_APPEND) | |
1325 | return; | |
1326 | ||
1327 | ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset); | |
1328 | if (WARN_ON(!ordered)) | |
1329 | return; | |
1330 | ||
1331 | ordered->physical = physical; | |
c7c3a6dc | 1332 | ordered->bdev = bio->bi_bdev; |
d8e3fb10 NA |
1333 | |
1334 | btrfs_put_ordered_extent(ordered); | |
1335 | } | |
1336 | ||
1337 | void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered) | |
1338 | { | |
1339 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
1340 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1341 | struct extent_map_tree *em_tree; | |
1342 | struct extent_map *em; | |
1343 | struct btrfs_ordered_sum *sum; | |
d8e3fb10 NA |
1344 | u64 orig_logical = ordered->disk_bytenr; |
1345 | u64 *logical = NULL; | |
1346 | int nr, stripe_len; | |
1347 | ||
1348 | /* Zoned devices should not have partitions. So, we can assume it is 0 */ | |
c7c3a6dc CH |
1349 | ASSERT(!bdev_is_partition(ordered->bdev)); |
1350 | if (WARN_ON(!ordered->bdev)) | |
d8e3fb10 NA |
1351 | return; |
1352 | ||
c7c3a6dc | 1353 | if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev, |
d8e3fb10 NA |
1354 | ordered->physical, &logical, &nr, |
1355 | &stripe_len))) | |
1356 | goto out; | |
1357 | ||
1358 | WARN_ON(nr != 1); | |
1359 | ||
1360 | if (orig_logical == *logical) | |
1361 | goto out; | |
1362 | ||
1363 | ordered->disk_bytenr = *logical; | |
1364 | ||
1365 | em_tree = &inode->extent_tree; | |
1366 | write_lock(&em_tree->lock); | |
1367 | em = search_extent_mapping(em_tree, ordered->file_offset, | |
1368 | ordered->num_bytes); | |
1369 | em->block_start = *logical; | |
1370 | free_extent_map(em); | |
1371 | write_unlock(&em_tree->lock); | |
1372 | ||
1373 | list_for_each_entry(sum, &ordered->list, list) { | |
1374 | if (*logical < orig_logical) | |
1375 | sum->bytenr -= orig_logical - *logical; | |
1376 | else | |
1377 | sum->bytenr += *logical - orig_logical; | |
1378 | } | |
1379 | ||
1380 | out: | |
1381 | kfree(logical); | |
d8e3fb10 | 1382 | } |
0bc09ca1 NA |
1383 | |
1384 | bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info, | |
1385 | struct extent_buffer *eb, | |
1386 | struct btrfs_block_group **cache_ret) | |
1387 | { | |
1388 | struct btrfs_block_group *cache; | |
1389 | bool ret = true; | |
1390 | ||
1391 | if (!btrfs_is_zoned(fs_info)) | |
1392 | return true; | |
1393 | ||
1394 | cache = *cache_ret; | |
1395 | ||
1396 | if (cache && (eb->start < cache->start || | |
1397 | cache->start + cache->length <= eb->start)) { | |
1398 | btrfs_put_block_group(cache); | |
1399 | cache = NULL; | |
1400 | *cache_ret = NULL; | |
1401 | } | |
1402 | ||
1403 | if (!cache) | |
1404 | cache = btrfs_lookup_block_group(fs_info, eb->start); | |
1405 | ||
1406 | if (cache) { | |
1407 | if (cache->meta_write_pointer != eb->start) { | |
1408 | btrfs_put_block_group(cache); | |
1409 | cache = NULL; | |
1410 | ret = false; | |
1411 | } else { | |
1412 | cache->meta_write_pointer = eb->start + eb->len; | |
1413 | } | |
1414 | ||
1415 | *cache_ret = cache; | |
1416 | } | |
1417 | ||
1418 | return ret; | |
1419 | } | |
1420 | ||
1421 | void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache, | |
1422 | struct extent_buffer *eb) | |
1423 | { | |
1424 | if (!btrfs_is_zoned(eb->fs_info) || !cache) | |
1425 | return; | |
1426 | ||
1427 | ASSERT(cache->meta_write_pointer == eb->start + eb->len); | |
1428 | cache->meta_write_pointer = eb->start; | |
1429 | } | |
de17addc NA |
1430 | |
1431 | int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length) | |
1432 | { | |
1433 | if (!btrfs_dev_is_sequential(device, physical)) | |
1434 | return -EOPNOTSUPP; | |
1435 | ||
1436 | return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT, | |
1437 | length >> SECTOR_SHIFT, GFP_NOFS, 0); | |
1438 | } | |
7db1c5d1 NA |
1439 | |
1440 | static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical, | |
1441 | struct blk_zone *zone) | |
1442 | { | |
1443 | struct btrfs_bio *bbio = NULL; | |
1444 | u64 mapped_length = PAGE_SIZE; | |
1445 | unsigned int nofs_flag; | |
1446 | int nmirrors; | |
1447 | int i, ret; | |
1448 | ||
1449 | ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical, | |
1450 | &mapped_length, &bbio); | |
1451 | if (ret || !bbio || mapped_length < PAGE_SIZE) { | |
1452 | btrfs_put_bbio(bbio); | |
1453 | return -EIO; | |
1454 | } | |
1455 | ||
1456 | if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
1457 | return -EINVAL; | |
1458 | ||
1459 | nofs_flag = memalloc_nofs_save(); | |
1460 | nmirrors = (int)bbio->num_stripes; | |
1461 | for (i = 0; i < nmirrors; i++) { | |
1462 | u64 physical = bbio->stripes[i].physical; | |
1463 | struct btrfs_device *dev = bbio->stripes[i].dev; | |
1464 | ||
1465 | /* Missing device */ | |
1466 | if (!dev->bdev) | |
1467 | continue; | |
1468 | ||
1469 | ret = btrfs_get_dev_zone(dev, physical, zone); | |
1470 | /* Failing device */ | |
1471 | if (ret == -EIO || ret == -EOPNOTSUPP) | |
1472 | continue; | |
1473 | break; | |
1474 | } | |
1475 | memalloc_nofs_restore(nofs_flag); | |
1476 | ||
1477 | return ret; | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by | |
1482 | * filling zeros between @physical_pos to a write pointer of dev-replace | |
1483 | * source device. | |
1484 | */ | |
1485 | int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical, | |
1486 | u64 physical_start, u64 physical_pos) | |
1487 | { | |
1488 | struct btrfs_fs_info *fs_info = tgt_dev->fs_info; | |
1489 | struct blk_zone zone; | |
1490 | u64 length; | |
1491 | u64 wp; | |
1492 | int ret; | |
1493 | ||
1494 | if (!btrfs_dev_is_sequential(tgt_dev, physical_pos)) | |
1495 | return 0; | |
1496 | ||
1497 | ret = read_zone_info(fs_info, logical, &zone); | |
1498 | if (ret) | |
1499 | return ret; | |
1500 | ||
1501 | wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT); | |
1502 | ||
1503 | if (physical_pos == wp) | |
1504 | return 0; | |
1505 | ||
1506 | if (physical_pos > wp) | |
1507 | return -EUCLEAN; | |
1508 | ||
1509 | length = wp - physical_pos; | |
1510 | return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length); | |
1511 | } | |
e7ff9e6b JT |
1512 | |
1513 | struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info, | |
1514 | u64 logical, u64 length) | |
1515 | { | |
1516 | struct btrfs_device *device; | |
1517 | struct extent_map *em; | |
1518 | struct map_lookup *map; | |
1519 | ||
1520 | em = btrfs_get_chunk_map(fs_info, logical, length); | |
1521 | if (IS_ERR(em)) | |
1522 | return ERR_CAST(em); | |
1523 | ||
1524 | map = em->map_lookup; | |
1525 | /* We only support single profile for now */ | |
1526 | ASSERT(map->num_stripes == 1); | |
1527 | device = map->stripes[0].dev; | |
1528 | ||
1529 | free_extent_map(em); | |
1530 | ||
1531 | return device; | |
1532 | } |