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 | |
12659251 NA |
24 | /* Number of superblock log zones */ |
25 | #define BTRFS_NR_SB_LOG_ZONES 2 | |
26 | ||
5b316468 NA |
27 | static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data) |
28 | { | |
29 | struct blk_zone *zones = data; | |
30 | ||
31 | memcpy(&zones[idx], zone, sizeof(*zone)); | |
32 | ||
33 | return 0; | |
34 | } | |
35 | ||
12659251 NA |
36 | static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones, |
37 | u64 *wp_ret) | |
38 | { | |
39 | bool empty[BTRFS_NR_SB_LOG_ZONES]; | |
40 | bool full[BTRFS_NR_SB_LOG_ZONES]; | |
41 | sector_t sector; | |
42 | ||
43 | ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL && | |
44 | zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL); | |
45 | ||
46 | empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY); | |
47 | empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY); | |
48 | full[0] = (zones[0].cond == BLK_ZONE_COND_FULL); | |
49 | full[1] = (zones[1].cond == BLK_ZONE_COND_FULL); | |
50 | ||
51 | /* | |
52 | * Possible states of log buffer zones | |
53 | * | |
54 | * Empty[0] In use[0] Full[0] | |
55 | * Empty[1] * x 0 | |
56 | * In use[1] 0 x 0 | |
57 | * Full[1] 1 1 C | |
58 | * | |
59 | * Log position: | |
60 | * *: Special case, no superblock is written | |
61 | * 0: Use write pointer of zones[0] | |
62 | * 1: Use write pointer of zones[1] | |
63 | * C: Compare super blcoks from zones[0] and zones[1], use the latest | |
64 | * one determined by generation | |
65 | * x: Invalid state | |
66 | */ | |
67 | ||
68 | if (empty[0] && empty[1]) { | |
69 | /* Special case to distinguish no superblock to read */ | |
70 | *wp_ret = zones[0].start << SECTOR_SHIFT; | |
71 | return -ENOENT; | |
72 | } else if (full[0] && full[1]) { | |
73 | /* Compare two super blocks */ | |
74 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
75 | struct page *page[BTRFS_NR_SB_LOG_ZONES]; | |
76 | struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES]; | |
77 | int i; | |
78 | ||
79 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { | |
80 | u64 bytenr; | |
81 | ||
82 | bytenr = ((zones[i].start + zones[i].len) | |
83 | << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE; | |
84 | ||
85 | page[i] = read_cache_page_gfp(mapping, | |
86 | bytenr >> PAGE_SHIFT, GFP_NOFS); | |
87 | if (IS_ERR(page[i])) { | |
88 | if (i == 1) | |
89 | btrfs_release_disk_super(super[0]); | |
90 | return PTR_ERR(page[i]); | |
91 | } | |
92 | super[i] = page_address(page[i]); | |
93 | } | |
94 | ||
95 | if (super[0]->generation > super[1]->generation) | |
96 | sector = zones[1].start; | |
97 | else | |
98 | sector = zones[0].start; | |
99 | ||
100 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) | |
101 | btrfs_release_disk_super(super[i]); | |
102 | } else if (!full[0] && (empty[1] || full[1])) { | |
103 | sector = zones[0].wp; | |
104 | } else if (full[0]) { | |
105 | sector = zones[1].wp; | |
106 | } else { | |
107 | return -EUCLEAN; | |
108 | } | |
109 | *wp_ret = sector << SECTOR_SHIFT; | |
110 | return 0; | |
111 | } | |
112 | ||
113 | /* | |
114 | * The following zones are reserved as the circular buffer on ZONED btrfs. | |
115 | * - The primary superblock: zones 0 and 1 | |
116 | * - The first copy: zones 16 and 17 | |
117 | * - The second copy: zones 1024 or zone at 256GB which is minimum, and | |
118 | * the following one | |
119 | */ | |
120 | static inline u32 sb_zone_number(int shift, int mirror) | |
121 | { | |
122 | ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX); | |
123 | ||
124 | switch (mirror) { | |
125 | case 0: return 0; | |
126 | case 1: return 16; | |
127 | case 2: return min_t(u64, btrfs_sb_offset(mirror) >> shift, 1024); | |
128 | } | |
129 | ||
130 | return 0; | |
131 | } | |
132 | ||
3c9daa09 JT |
133 | /* |
134 | * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block | |
135 | * device into static sized chunks and fake a conventional zone on each of | |
136 | * them. | |
137 | */ | |
138 | static int emulate_report_zones(struct btrfs_device *device, u64 pos, | |
139 | struct blk_zone *zones, unsigned int nr_zones) | |
140 | { | |
141 | const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT; | |
142 | sector_t bdev_size = bdev_nr_sectors(device->bdev); | |
143 | unsigned int i; | |
144 | ||
145 | pos >>= SECTOR_SHIFT; | |
146 | for (i = 0; i < nr_zones; i++) { | |
147 | zones[i].start = i * zone_sectors + pos; | |
148 | zones[i].len = zone_sectors; | |
149 | zones[i].capacity = zone_sectors; | |
150 | zones[i].wp = zones[i].start + zone_sectors; | |
151 | zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL; | |
152 | zones[i].cond = BLK_ZONE_COND_NOT_WP; | |
153 | ||
154 | if (zones[i].wp >= bdev_size) { | |
155 | i++; | |
156 | break; | |
157 | } | |
158 | } | |
159 | ||
160 | return i; | |
161 | } | |
162 | ||
5b316468 NA |
163 | static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos, |
164 | struct blk_zone *zones, unsigned int *nr_zones) | |
165 | { | |
166 | int ret; | |
167 | ||
168 | if (!*nr_zones) | |
169 | return 0; | |
170 | ||
3c9daa09 JT |
171 | if (!bdev_is_zoned(device->bdev)) { |
172 | ret = emulate_report_zones(device, pos, zones, *nr_zones); | |
173 | *nr_zones = ret; | |
174 | return 0; | |
175 | } | |
176 | ||
5b316468 NA |
177 | ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones, |
178 | copy_zone_info_cb, zones); | |
179 | if (ret < 0) { | |
180 | btrfs_err_in_rcu(device->fs_info, | |
181 | "zoned: failed to read zone %llu on %s (devid %llu)", | |
182 | pos, rcu_str_deref(device->name), | |
183 | device->devid); | |
184 | return ret; | |
185 | } | |
186 | *nr_zones = ret; | |
187 | if (!ret) | |
188 | return -EIO; | |
189 | ||
190 | return 0; | |
191 | } | |
192 | ||
3c9daa09 JT |
193 | /* The emulated zone size is determined from the size of device extent */ |
194 | static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info) | |
195 | { | |
196 | struct btrfs_path *path; | |
197 | struct btrfs_root *root = fs_info->dev_root; | |
198 | struct btrfs_key key; | |
199 | struct extent_buffer *leaf; | |
200 | struct btrfs_dev_extent *dext; | |
201 | int ret = 0; | |
202 | ||
203 | key.objectid = 1; | |
204 | key.type = BTRFS_DEV_EXTENT_KEY; | |
205 | key.offset = 0; | |
206 | ||
207 | path = btrfs_alloc_path(); | |
208 | if (!path) | |
209 | return -ENOMEM; | |
210 | ||
211 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
212 | if (ret < 0) | |
213 | goto out; | |
214 | ||
215 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
216 | ret = btrfs_next_item(root, path); | |
217 | if (ret < 0) | |
218 | goto out; | |
219 | /* No dev extents at all? Not good */ | |
220 | if (ret > 0) { | |
221 | ret = -EUCLEAN; | |
222 | goto out; | |
223 | } | |
224 | } | |
225 | ||
226 | leaf = path->nodes[0]; | |
227 | dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); | |
228 | fs_info->zone_size = btrfs_dev_extent_length(leaf, dext); | |
229 | ret = 0; | |
230 | ||
231 | out: | |
232 | btrfs_free_path(path); | |
233 | ||
234 | return ret; | |
235 | } | |
236 | ||
73651042 NA |
237 | int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info) |
238 | { | |
239 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
240 | struct btrfs_device *device; | |
241 | int ret = 0; | |
242 | ||
243 | /* fs_info->zone_size might not set yet. Use the incomapt flag here. */ | |
244 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
245 | return 0; | |
246 | ||
247 | mutex_lock(&fs_devices->device_list_mutex); | |
248 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
249 | /* We can skip reading of zone info for missing devices */ | |
250 | if (!device->bdev) | |
251 | continue; | |
252 | ||
253 | ret = btrfs_get_dev_zone_info(device); | |
254 | if (ret) | |
255 | break; | |
256 | } | |
257 | mutex_unlock(&fs_devices->device_list_mutex); | |
258 | ||
259 | return ret; | |
260 | } | |
261 | ||
5b316468 NA |
262 | int btrfs_get_dev_zone_info(struct btrfs_device *device) |
263 | { | |
3c9daa09 | 264 | struct btrfs_fs_info *fs_info = device->fs_info; |
5b316468 NA |
265 | struct btrfs_zoned_device_info *zone_info = NULL; |
266 | struct block_device *bdev = device->bdev; | |
862931c7 | 267 | struct request_queue *queue = bdev_get_queue(bdev); |
5b316468 NA |
268 | sector_t nr_sectors; |
269 | sector_t sector = 0; | |
270 | struct blk_zone *zones = NULL; | |
271 | unsigned int i, nreported = 0, nr_zones; | |
272 | unsigned int zone_sectors; | |
3c9daa09 | 273 | char *model, *emulated; |
5b316468 NA |
274 | int ret; |
275 | ||
3c9daa09 JT |
276 | /* |
277 | * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not | |
278 | * yet be set. | |
279 | */ | |
280 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
5b316468 NA |
281 | return 0; |
282 | ||
283 | if (device->zone_info) | |
284 | return 0; | |
285 | ||
286 | zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL); | |
287 | if (!zone_info) | |
288 | return -ENOMEM; | |
289 | ||
3c9daa09 JT |
290 | if (!bdev_is_zoned(bdev)) { |
291 | if (!fs_info->zone_size) { | |
292 | ret = calculate_emulated_zone_size(fs_info); | |
293 | if (ret) | |
294 | goto out; | |
295 | } | |
296 | ||
297 | ASSERT(fs_info->zone_size); | |
298 | zone_sectors = fs_info->zone_size >> SECTOR_SHIFT; | |
299 | } else { | |
300 | zone_sectors = bdev_zone_sectors(bdev); | |
301 | } | |
302 | ||
ac7ac461 | 303 | nr_sectors = bdev_nr_sectors(bdev); |
5b316468 NA |
304 | /* Check if it's power of 2 (see is_power_of_2) */ |
305 | ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0); | |
306 | zone_info->zone_size = zone_sectors << SECTOR_SHIFT; | |
307 | zone_info->zone_size_shift = ilog2(zone_info->zone_size); | |
862931c7 NA |
308 | zone_info->max_zone_append_size = |
309 | (u64)queue_max_zone_append_sectors(queue) << SECTOR_SHIFT; | |
5b316468 NA |
310 | zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors); |
311 | if (!IS_ALIGNED(nr_sectors, zone_sectors)) | |
312 | zone_info->nr_zones++; | |
313 | ||
314 | zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
315 | if (!zone_info->seq_zones) { | |
316 | ret = -ENOMEM; | |
317 | goto out; | |
318 | } | |
319 | ||
320 | zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
321 | if (!zone_info->empty_zones) { | |
322 | ret = -ENOMEM; | |
323 | goto out; | |
324 | } | |
325 | ||
326 | zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL); | |
327 | if (!zones) { | |
328 | ret = -ENOMEM; | |
329 | goto out; | |
330 | } | |
331 | ||
332 | /* Get zones type */ | |
333 | while (sector < nr_sectors) { | |
334 | nr_zones = BTRFS_REPORT_NR_ZONES; | |
335 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones, | |
336 | &nr_zones); | |
337 | if (ret) | |
338 | goto out; | |
339 | ||
340 | for (i = 0; i < nr_zones; i++) { | |
341 | if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ) | |
342 | __set_bit(nreported, zone_info->seq_zones); | |
343 | if (zones[i].cond == BLK_ZONE_COND_EMPTY) | |
344 | __set_bit(nreported, zone_info->empty_zones); | |
345 | nreported++; | |
346 | } | |
347 | sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len; | |
348 | } | |
349 | ||
350 | if (nreported != zone_info->nr_zones) { | |
351 | btrfs_err_in_rcu(device->fs_info, | |
352 | "inconsistent number of zones on %s (%u/%u)", | |
353 | rcu_str_deref(device->name), nreported, | |
354 | zone_info->nr_zones); | |
355 | ret = -EIO; | |
356 | goto out; | |
357 | } | |
358 | ||
12659251 NA |
359 | /* Validate superblock log */ |
360 | nr_zones = BTRFS_NR_SB_LOG_ZONES; | |
361 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
362 | u32 sb_zone; | |
363 | u64 sb_wp; | |
364 | int sb_pos = BTRFS_NR_SB_LOG_ZONES * i; | |
365 | ||
366 | sb_zone = sb_zone_number(zone_info->zone_size_shift, i); | |
367 | if (sb_zone + 1 >= zone_info->nr_zones) | |
368 | continue; | |
369 | ||
370 | sector = sb_zone << (zone_info->zone_size_shift - SECTOR_SHIFT); | |
371 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, | |
372 | &zone_info->sb_zones[sb_pos], | |
373 | &nr_zones); | |
374 | if (ret) | |
375 | goto out; | |
376 | ||
377 | if (nr_zones != BTRFS_NR_SB_LOG_ZONES) { | |
378 | btrfs_err_in_rcu(device->fs_info, | |
379 | "zoned: failed to read super block log zone info at devid %llu zone %u", | |
380 | device->devid, sb_zone); | |
381 | ret = -EUCLEAN; | |
382 | goto out; | |
383 | } | |
384 | ||
385 | /* | |
386 | * If zones[0] is conventional, always use the beggining of the | |
387 | * zone to record superblock. No need to validate in that case. | |
388 | */ | |
389 | if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type == | |
390 | BLK_ZONE_TYPE_CONVENTIONAL) | |
391 | continue; | |
392 | ||
393 | ret = sb_write_pointer(device->bdev, | |
394 | &zone_info->sb_zones[sb_pos], &sb_wp); | |
395 | if (ret != -ENOENT && ret) { | |
396 | btrfs_err_in_rcu(device->fs_info, | |
397 | "zoned: super block log zone corrupted devid %llu zone %u", | |
398 | device->devid, sb_zone); | |
399 | ret = -EUCLEAN; | |
400 | goto out; | |
401 | } | |
402 | } | |
403 | ||
404 | ||
5b316468 NA |
405 | kfree(zones); |
406 | ||
407 | device->zone_info = zone_info; | |
408 | ||
3c9daa09 JT |
409 | switch (bdev_zoned_model(bdev)) { |
410 | case BLK_ZONED_HM: | |
411 | model = "host-managed zoned"; | |
412 | emulated = ""; | |
413 | break; | |
414 | case BLK_ZONED_HA: | |
415 | model = "host-aware zoned"; | |
416 | emulated = ""; | |
417 | break; | |
418 | case BLK_ZONED_NONE: | |
419 | model = "regular"; | |
420 | emulated = "emulated "; | |
421 | break; | |
422 | default: | |
423 | /* Just in case */ | |
424 | btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s", | |
425 | bdev_zoned_model(bdev), | |
426 | rcu_str_deref(device->name)); | |
427 | ret = -EOPNOTSUPP; | |
428 | goto out_free_zone_info; | |
429 | } | |
430 | ||
431 | btrfs_info_in_rcu(fs_info, | |
432 | "%s block device %s, %u %szones of %llu bytes", | |
433 | model, rcu_str_deref(device->name), zone_info->nr_zones, | |
434 | emulated, zone_info->zone_size); | |
5b316468 NA |
435 | |
436 | return 0; | |
437 | ||
438 | out: | |
439 | kfree(zones); | |
3c9daa09 | 440 | out_free_zone_info: |
5b316468 NA |
441 | bitmap_free(zone_info->empty_zones); |
442 | bitmap_free(zone_info->seq_zones); | |
443 | kfree(zone_info); | |
3c9daa09 | 444 | device->zone_info = NULL; |
5b316468 NA |
445 | |
446 | return ret; | |
447 | } | |
448 | ||
449 | void btrfs_destroy_dev_zone_info(struct btrfs_device *device) | |
450 | { | |
451 | struct btrfs_zoned_device_info *zone_info = device->zone_info; | |
452 | ||
453 | if (!zone_info) | |
454 | return; | |
455 | ||
456 | bitmap_free(zone_info->seq_zones); | |
457 | bitmap_free(zone_info->empty_zones); | |
458 | kfree(zone_info); | |
459 | device->zone_info = NULL; | |
460 | } | |
461 | ||
462 | int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos, | |
463 | struct blk_zone *zone) | |
464 | { | |
465 | unsigned int nr_zones = 1; | |
466 | int ret; | |
467 | ||
468 | ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones); | |
469 | if (ret != 0 || !nr_zones) | |
470 | return ret ? ret : -EIO; | |
471 | ||
472 | return 0; | |
473 | } | |
b70f5097 NA |
474 | |
475 | int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info) | |
476 | { | |
477 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
478 | struct btrfs_device *device; | |
479 | u64 zoned_devices = 0; | |
480 | u64 nr_devices = 0; | |
481 | u64 zone_size = 0; | |
862931c7 | 482 | u64 max_zone_append_size = 0; |
3c9daa09 | 483 | const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED); |
b70f5097 NA |
484 | int ret = 0; |
485 | ||
486 | /* Count zoned devices */ | |
487 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
488 | enum blk_zoned_model model; | |
489 | ||
490 | if (!device->bdev) | |
491 | continue; | |
492 | ||
493 | model = bdev_zoned_model(device->bdev); | |
3c9daa09 JT |
494 | /* |
495 | * A Host-Managed zoned device must be used as a zoned device. | |
496 | * A Host-Aware zoned device and a non-zoned devices can be | |
497 | * treated as a zoned device, if ZONED flag is enabled in the | |
498 | * superblock. | |
499 | */ | |
b70f5097 | 500 | if (model == BLK_ZONED_HM || |
3c9daa09 JT |
501 | (model == BLK_ZONED_HA && incompat_zoned) || |
502 | (model == BLK_ZONED_NONE && incompat_zoned)) { | |
503 | struct btrfs_zoned_device_info *zone_info = | |
504 | device->zone_info; | |
862931c7 NA |
505 | |
506 | zone_info = device->zone_info; | |
b70f5097 NA |
507 | zoned_devices++; |
508 | if (!zone_size) { | |
862931c7 NA |
509 | zone_size = zone_info->zone_size; |
510 | } else if (zone_info->zone_size != zone_size) { | |
b70f5097 NA |
511 | btrfs_err(fs_info, |
512 | "zoned: unequal block device zone sizes: have %llu found %llu", | |
513 | device->zone_info->zone_size, | |
514 | zone_size); | |
515 | ret = -EINVAL; | |
516 | goto out; | |
517 | } | |
862931c7 NA |
518 | if (!max_zone_append_size || |
519 | (zone_info->max_zone_append_size && | |
520 | zone_info->max_zone_append_size < max_zone_append_size)) | |
521 | max_zone_append_size = | |
522 | zone_info->max_zone_append_size; | |
b70f5097 NA |
523 | } |
524 | nr_devices++; | |
525 | } | |
526 | ||
527 | if (!zoned_devices && !incompat_zoned) | |
528 | goto out; | |
529 | ||
530 | if (!zoned_devices && incompat_zoned) { | |
531 | /* No zoned block device found on ZONED filesystem */ | |
532 | btrfs_err(fs_info, | |
533 | "zoned: no zoned devices found on a zoned filesystem"); | |
534 | ret = -EINVAL; | |
535 | goto out; | |
536 | } | |
537 | ||
538 | if (zoned_devices && !incompat_zoned) { | |
539 | btrfs_err(fs_info, | |
540 | "zoned: mode not enabled but zoned device found"); | |
541 | ret = -EINVAL; | |
542 | goto out; | |
543 | } | |
544 | ||
545 | if (zoned_devices != nr_devices) { | |
546 | btrfs_err(fs_info, | |
547 | "zoned: cannot mix zoned and regular devices"); | |
548 | ret = -EINVAL; | |
549 | goto out; | |
550 | } | |
551 | ||
552 | /* | |
553 | * stripe_size is always aligned to BTRFS_STRIPE_LEN in | |
554 | * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size, | |
555 | * check the alignment here. | |
556 | */ | |
557 | if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) { | |
558 | btrfs_err(fs_info, | |
559 | "zoned: zone size %llu not aligned to stripe %u", | |
560 | zone_size, BTRFS_STRIPE_LEN); | |
561 | ret = -EINVAL; | |
562 | goto out; | |
563 | } | |
564 | ||
a589dde0 NA |
565 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
566 | btrfs_err(fs_info, "zoned: mixed block groups not supported"); | |
567 | ret = -EINVAL; | |
568 | goto out; | |
569 | } | |
570 | ||
b70f5097 | 571 | fs_info->zone_size = zone_size; |
862931c7 | 572 | fs_info->max_zone_append_size = max_zone_append_size; |
1cd6121f | 573 | fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED; |
b70f5097 | 574 | |
b53429ba JT |
575 | /* |
576 | * Check mount options here, because we might change fs_info->zoned | |
577 | * from fs_info->zone_size. | |
578 | */ | |
579 | ret = btrfs_check_mountopts_zoned(fs_info); | |
580 | if (ret) | |
581 | goto out; | |
582 | ||
b70f5097 NA |
583 | btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size); |
584 | out: | |
585 | return ret; | |
586 | } | |
5d1ab66c NA |
587 | |
588 | int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info) | |
589 | { | |
590 | if (!btrfs_is_zoned(info)) | |
591 | return 0; | |
592 | ||
593 | /* | |
594 | * Space cache writing is not COWed. Disable that to avoid write errors | |
595 | * in sequential zones. | |
596 | */ | |
597 | if (btrfs_test_opt(info, SPACE_CACHE)) { | |
598 | btrfs_err(info, "zoned: space cache v1 is not supported"); | |
599 | return -EINVAL; | |
600 | } | |
601 | ||
d206e9c9 NA |
602 | if (btrfs_test_opt(info, NODATACOW)) { |
603 | btrfs_err(info, "zoned: NODATACOW not supported"); | |
604 | return -EINVAL; | |
605 | } | |
606 | ||
5d1ab66c NA |
607 | return 0; |
608 | } | |
12659251 NA |
609 | |
610 | static int sb_log_location(struct block_device *bdev, struct blk_zone *zones, | |
611 | int rw, u64 *bytenr_ret) | |
612 | { | |
613 | u64 wp; | |
614 | int ret; | |
615 | ||
616 | if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) { | |
617 | *bytenr_ret = zones[0].start << SECTOR_SHIFT; | |
618 | return 0; | |
619 | } | |
620 | ||
621 | ret = sb_write_pointer(bdev, zones, &wp); | |
622 | if (ret != -ENOENT && ret < 0) | |
623 | return ret; | |
624 | ||
625 | if (rw == WRITE) { | |
626 | struct blk_zone *reset = NULL; | |
627 | ||
628 | if (wp == zones[0].start << SECTOR_SHIFT) | |
629 | reset = &zones[0]; | |
630 | else if (wp == zones[1].start << SECTOR_SHIFT) | |
631 | reset = &zones[1]; | |
632 | ||
633 | if (reset && reset->cond != BLK_ZONE_COND_EMPTY) { | |
634 | ASSERT(reset->cond == BLK_ZONE_COND_FULL); | |
635 | ||
636 | ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
637 | reset->start, reset->len, | |
638 | GFP_NOFS); | |
639 | if (ret) | |
640 | return ret; | |
641 | ||
642 | reset->cond = BLK_ZONE_COND_EMPTY; | |
643 | reset->wp = reset->start; | |
644 | } | |
645 | } else if (ret != -ENOENT) { | |
646 | /* For READ, we want the precious one */ | |
647 | if (wp == zones[0].start << SECTOR_SHIFT) | |
648 | wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT; | |
649 | wp -= BTRFS_SUPER_INFO_SIZE; | |
650 | } | |
651 | ||
652 | *bytenr_ret = wp; | |
653 | return 0; | |
654 | ||
655 | } | |
656 | ||
657 | int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw, | |
658 | u64 *bytenr_ret) | |
659 | { | |
660 | struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES]; | |
661 | unsigned int zone_sectors; | |
662 | u32 sb_zone; | |
663 | int ret; | |
12659251 NA |
664 | u8 zone_sectors_shift; |
665 | sector_t nr_sectors; | |
666 | u32 nr_zones; | |
667 | ||
668 | if (!bdev_is_zoned(bdev)) { | |
669 | *bytenr_ret = btrfs_sb_offset(mirror); | |
670 | return 0; | |
671 | } | |
672 | ||
673 | ASSERT(rw == READ || rw == WRITE); | |
674 | ||
675 | zone_sectors = bdev_zone_sectors(bdev); | |
676 | if (!is_power_of_2(zone_sectors)) | |
677 | return -EINVAL; | |
12659251 | 678 | zone_sectors_shift = ilog2(zone_sectors); |
ac7ac461 | 679 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
680 | nr_zones = nr_sectors >> zone_sectors_shift; |
681 | ||
682 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
683 | if (sb_zone + 1 >= nr_zones) | |
684 | return -ENOENT; | |
685 | ||
686 | ret = blkdev_report_zones(bdev, sb_zone << zone_sectors_shift, | |
687 | BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb, | |
688 | zones); | |
689 | if (ret < 0) | |
690 | return ret; | |
691 | if (ret != BTRFS_NR_SB_LOG_ZONES) | |
692 | return -EIO; | |
693 | ||
694 | return sb_log_location(bdev, zones, rw, bytenr_ret); | |
695 | } | |
696 | ||
697 | int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw, | |
698 | u64 *bytenr_ret) | |
699 | { | |
700 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
701 | u32 zone_num; | |
702 | ||
d6639b35 NA |
703 | /* |
704 | * For a zoned filesystem on a non-zoned block device, use the same | |
705 | * super block locations as regular filesystem. Doing so, the super | |
706 | * block can always be retrieved and the zoned flag of the volume | |
707 | * detected from the super block information. | |
708 | */ | |
709 | if (!bdev_is_zoned(device->bdev)) { | |
12659251 NA |
710 | *bytenr_ret = btrfs_sb_offset(mirror); |
711 | return 0; | |
712 | } | |
713 | ||
714 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
715 | if (zone_num + 1 >= zinfo->nr_zones) | |
716 | return -ENOENT; | |
717 | ||
718 | return sb_log_location(device->bdev, | |
719 | &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror], | |
720 | rw, bytenr_ret); | |
721 | } | |
722 | ||
723 | static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo, | |
724 | int mirror) | |
725 | { | |
726 | u32 zone_num; | |
727 | ||
728 | if (!zinfo) | |
729 | return false; | |
730 | ||
731 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
732 | if (zone_num + 1 >= zinfo->nr_zones) | |
733 | return false; | |
734 | ||
735 | if (!test_bit(zone_num, zinfo->seq_zones)) | |
736 | return false; | |
737 | ||
738 | return true; | |
739 | } | |
740 | ||
741 | void btrfs_advance_sb_log(struct btrfs_device *device, int mirror) | |
742 | { | |
743 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
744 | struct blk_zone *zone; | |
745 | ||
746 | if (!is_sb_log_zone(zinfo, mirror)) | |
747 | return; | |
748 | ||
749 | zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror]; | |
750 | if (zone->cond != BLK_ZONE_COND_FULL) { | |
751 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
752 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
753 | ||
754 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
755 | ||
756 | if (zone->wp == zone->start + zone->len) | |
757 | zone->cond = BLK_ZONE_COND_FULL; | |
758 | ||
759 | return; | |
760 | } | |
761 | ||
762 | zone++; | |
763 | ASSERT(zone->cond != BLK_ZONE_COND_FULL); | |
764 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
765 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
766 | ||
767 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
768 | ||
769 | if (zone->wp == zone->start + zone->len) | |
770 | zone->cond = BLK_ZONE_COND_FULL; | |
771 | } | |
772 | ||
773 | int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror) | |
774 | { | |
775 | sector_t zone_sectors; | |
776 | sector_t nr_sectors; | |
777 | u8 zone_sectors_shift; | |
778 | u32 sb_zone; | |
779 | u32 nr_zones; | |
780 | ||
781 | zone_sectors = bdev_zone_sectors(bdev); | |
782 | zone_sectors_shift = ilog2(zone_sectors); | |
ac7ac461 | 783 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
784 | nr_zones = nr_sectors >> zone_sectors_shift; |
785 | ||
786 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
787 | if (sb_zone + 1 >= nr_zones) | |
788 | return -ENOENT; | |
789 | ||
790 | return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
791 | sb_zone << zone_sectors_shift, | |
792 | zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS); | |
793 | } | |
1cd6121f NA |
794 | |
795 | /** | |
796 | * btrfs_find_allocatable_zones - find allocatable zones within a given region | |
797 | * | |
798 | * @device: the device to allocate a region on | |
799 | * @hole_start: the position of the hole to allocate the region | |
800 | * @num_bytes: size of wanted region | |
801 | * @hole_end: the end of the hole | |
802 | * @return: position of allocatable zones | |
803 | * | |
804 | * Allocatable region should not contain any superblock locations. | |
805 | */ | |
806 | u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start, | |
807 | u64 hole_end, u64 num_bytes) | |
808 | { | |
809 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
810 | const u8 shift = zinfo->zone_size_shift; | |
811 | u64 nzones = num_bytes >> shift; | |
812 | u64 pos = hole_start; | |
813 | u64 begin, end; | |
814 | bool have_sb; | |
815 | int i; | |
816 | ||
817 | ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size)); | |
818 | ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size)); | |
819 | ||
820 | while (pos < hole_end) { | |
821 | begin = pos >> shift; | |
822 | end = begin + nzones; | |
823 | ||
824 | if (end > zinfo->nr_zones) | |
825 | return hole_end; | |
826 | ||
827 | /* Check if zones in the region are all empty */ | |
828 | if (btrfs_dev_is_sequential(device, pos) && | |
829 | find_next_zero_bit(zinfo->empty_zones, end, begin) != end) { | |
830 | pos += zinfo->zone_size; | |
831 | continue; | |
832 | } | |
833 | ||
834 | have_sb = false; | |
835 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
836 | u32 sb_zone; | |
837 | u64 sb_pos; | |
838 | ||
839 | sb_zone = sb_zone_number(shift, i); | |
840 | if (!(end <= sb_zone || | |
841 | sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) { | |
842 | have_sb = true; | |
843 | pos = ((u64)sb_zone + BTRFS_NR_SB_LOG_ZONES) << shift; | |
844 | break; | |
845 | } | |
846 | ||
847 | /* We also need to exclude regular superblock positions */ | |
848 | sb_pos = btrfs_sb_offset(i); | |
849 | if (!(pos + num_bytes <= sb_pos || | |
850 | sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) { | |
851 | have_sb = true; | |
852 | pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE, | |
853 | zinfo->zone_size); | |
854 | break; | |
855 | } | |
856 | } | |
857 | if (!have_sb) | |
858 | break; | |
859 | } | |
860 | ||
861 | return pos; | |
862 | } | |
863 | ||
864 | int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical, | |
865 | u64 length, u64 *bytes) | |
866 | { | |
867 | int ret; | |
868 | ||
869 | *bytes = 0; | |
870 | ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET, | |
871 | physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT, | |
872 | GFP_NOFS); | |
873 | if (ret) | |
874 | return ret; | |
875 | ||
876 | *bytes = length; | |
877 | while (length) { | |
878 | btrfs_dev_set_zone_empty(device, physical); | |
879 | physical += device->zone_info->zone_size; | |
880 | length -= device->zone_info->zone_size; | |
881 | } | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
886 | int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size) | |
887 | { | |
888 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
889 | const u8 shift = zinfo->zone_size_shift; | |
890 | unsigned long begin = start >> shift; | |
891 | unsigned long end = (start + size) >> shift; | |
892 | u64 pos; | |
893 | int ret; | |
894 | ||
895 | ASSERT(IS_ALIGNED(start, zinfo->zone_size)); | |
896 | ASSERT(IS_ALIGNED(size, zinfo->zone_size)); | |
897 | ||
898 | if (end > zinfo->nr_zones) | |
899 | return -ERANGE; | |
900 | ||
901 | /* All the zones are conventional */ | |
902 | if (find_next_bit(zinfo->seq_zones, begin, end) == end) | |
903 | return 0; | |
904 | ||
905 | /* All the zones are sequential and empty */ | |
906 | if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end && | |
907 | find_next_zero_bit(zinfo->empty_zones, begin, end) == end) | |
908 | return 0; | |
909 | ||
910 | for (pos = start; pos < start + size; pos += zinfo->zone_size) { | |
911 | u64 reset_bytes; | |
912 | ||
913 | if (!btrfs_dev_is_sequential(device, pos) || | |
914 | btrfs_dev_is_empty_zone(device, pos)) | |
915 | continue; | |
916 | ||
917 | /* Free regions should be empty */ | |
918 | btrfs_warn_in_rcu( | |
919 | device->fs_info, | |
920 | "zoned: resetting device %s (devid %llu) zone %llu for allocation", | |
921 | rcu_str_deref(device->name), device->devid, pos >> shift); | |
922 | WARN_ON_ONCE(1); | |
923 | ||
924 | ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size, | |
925 | &reset_bytes); | |
926 | if (ret) | |
927 | return ret; | |
928 | } | |
929 | ||
930 | return 0; | |
931 | } | |
08e11a3d | 932 | |
a94794d5 NA |
933 | /* |
934 | * Calculate an allocation pointer from the extent allocation information | |
935 | * for a block group consist of conventional zones. It is pointed to the | |
936 | * end of the highest addressed extent in the block group as an allocation | |
937 | * offset. | |
938 | */ | |
939 | static int calculate_alloc_pointer(struct btrfs_block_group *cache, | |
940 | u64 *offset_ret) | |
941 | { | |
942 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
943 | struct btrfs_root *root = fs_info->extent_root; | |
944 | struct btrfs_path *path; | |
945 | struct btrfs_key key; | |
946 | struct btrfs_key found_key; | |
947 | int ret; | |
948 | u64 length; | |
949 | ||
950 | path = btrfs_alloc_path(); | |
951 | if (!path) | |
952 | return -ENOMEM; | |
953 | ||
954 | key.objectid = cache->start + cache->length; | |
955 | key.type = 0; | |
956 | key.offset = 0; | |
957 | ||
958 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
959 | /* We should not find the exact match */ | |
960 | if (!ret) | |
961 | ret = -EUCLEAN; | |
962 | if (ret < 0) | |
963 | goto out; | |
964 | ||
965 | ret = btrfs_previous_extent_item(root, path, cache->start); | |
966 | if (ret) { | |
967 | if (ret == 1) { | |
968 | ret = 0; | |
969 | *offset_ret = 0; | |
970 | } | |
971 | goto out; | |
972 | } | |
973 | ||
974 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
975 | ||
976 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY) | |
977 | length = found_key.offset; | |
978 | else | |
979 | length = fs_info->nodesize; | |
980 | ||
981 | if (!(found_key.objectid >= cache->start && | |
982 | found_key.objectid + length <= cache->start + cache->length)) { | |
983 | ret = -EUCLEAN; | |
984 | goto out; | |
985 | } | |
986 | *offset_ret = found_key.objectid + length - cache->start; | |
987 | ret = 0; | |
988 | ||
989 | out: | |
990 | btrfs_free_path(path); | |
991 | return ret; | |
992 | } | |
993 | ||
994 | int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) | |
08e11a3d NA |
995 | { |
996 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
997 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
998 | struct extent_map *em; | |
999 | struct map_lookup *map; | |
1000 | struct btrfs_device *device; | |
1001 | u64 logical = cache->start; | |
1002 | u64 length = cache->length; | |
1003 | u64 physical = 0; | |
1004 | int ret; | |
1005 | int i; | |
1006 | unsigned int nofs_flag; | |
1007 | u64 *alloc_offsets = NULL; | |
a94794d5 | 1008 | u64 last_alloc = 0; |
08e11a3d NA |
1009 | u32 num_sequential = 0, num_conventional = 0; |
1010 | ||
1011 | if (!btrfs_is_zoned(fs_info)) | |
1012 | return 0; | |
1013 | ||
1014 | /* Sanity check */ | |
1015 | if (!IS_ALIGNED(length, fs_info->zone_size)) { | |
1016 | btrfs_err(fs_info, | |
1017 | "zoned: block group %llu len %llu unaligned to zone size %llu", | |
1018 | logical, length, fs_info->zone_size); | |
1019 | return -EIO; | |
1020 | } | |
1021 | ||
1022 | /* Get the chunk mapping */ | |
1023 | read_lock(&em_tree->lock); | |
1024 | em = lookup_extent_mapping(em_tree, logical, length); | |
1025 | read_unlock(&em_tree->lock); | |
1026 | ||
1027 | if (!em) | |
1028 | return -EINVAL; | |
1029 | ||
1030 | map = em->map_lookup; | |
1031 | ||
1032 | alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS); | |
1033 | if (!alloc_offsets) { | |
1034 | free_extent_map(em); | |
1035 | return -ENOMEM; | |
1036 | } | |
1037 | ||
1038 | for (i = 0; i < map->num_stripes; i++) { | |
1039 | bool is_sequential; | |
1040 | struct blk_zone zone; | |
6143c23c NA |
1041 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
1042 | int dev_replace_is_ongoing = 0; | |
08e11a3d NA |
1043 | |
1044 | device = map->stripes[i].dev; | |
1045 | physical = map->stripes[i].physical; | |
1046 | ||
1047 | if (device->bdev == NULL) { | |
1048 | alloc_offsets[i] = WP_MISSING_DEV; | |
1049 | continue; | |
1050 | } | |
1051 | ||
1052 | is_sequential = btrfs_dev_is_sequential(device, physical); | |
1053 | if (is_sequential) | |
1054 | num_sequential++; | |
1055 | else | |
1056 | num_conventional++; | |
1057 | ||
1058 | if (!is_sequential) { | |
1059 | alloc_offsets[i] = WP_CONVENTIONAL; | |
1060 | continue; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * This zone will be used for allocation, so mark this zone | |
1065 | * non-empty. | |
1066 | */ | |
1067 | btrfs_dev_clear_zone_empty(device, physical); | |
1068 | ||
6143c23c NA |
1069 | down_read(&dev_replace->rwsem); |
1070 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
1071 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) | |
1072 | btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical); | |
1073 | up_read(&dev_replace->rwsem); | |
1074 | ||
08e11a3d NA |
1075 | /* |
1076 | * The group is mapped to a sequential zone. Get the zone write | |
1077 | * pointer to determine the allocation offset within the zone. | |
1078 | */ | |
1079 | WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size)); | |
1080 | nofs_flag = memalloc_nofs_save(); | |
1081 | ret = btrfs_get_dev_zone(device, physical, &zone); | |
1082 | memalloc_nofs_restore(nofs_flag); | |
1083 | if (ret == -EIO || ret == -EOPNOTSUPP) { | |
1084 | ret = 0; | |
1085 | alloc_offsets[i] = WP_MISSING_DEV; | |
1086 | continue; | |
1087 | } else if (ret) { | |
1088 | goto out; | |
1089 | } | |
1090 | ||
1091 | switch (zone.cond) { | |
1092 | case BLK_ZONE_COND_OFFLINE: | |
1093 | case BLK_ZONE_COND_READONLY: | |
1094 | btrfs_err(fs_info, | |
1095 | "zoned: offline/readonly zone %llu on device %s (devid %llu)", | |
1096 | physical >> device->zone_info->zone_size_shift, | |
1097 | rcu_str_deref(device->name), device->devid); | |
1098 | alloc_offsets[i] = WP_MISSING_DEV; | |
1099 | break; | |
1100 | case BLK_ZONE_COND_EMPTY: | |
1101 | alloc_offsets[i] = 0; | |
1102 | break; | |
1103 | case BLK_ZONE_COND_FULL: | |
1104 | alloc_offsets[i] = fs_info->zone_size; | |
1105 | break; | |
1106 | default: | |
1107 | /* Partially used zone */ | |
1108 | alloc_offsets[i] = | |
1109 | ((zone.wp - zone.start) << SECTOR_SHIFT); | |
1110 | break; | |
1111 | } | |
1112 | } | |
1113 | ||
08f45559 JT |
1114 | if (num_sequential > 0) |
1115 | cache->seq_zone = true; | |
1116 | ||
08e11a3d NA |
1117 | if (num_conventional > 0) { |
1118 | /* | |
a94794d5 NA |
1119 | * Avoid calling calculate_alloc_pointer() for new BG. It |
1120 | * is no use for new BG. It must be always 0. | |
1121 | * | |
1122 | * Also, we have a lock chain of extent buffer lock -> | |
1123 | * chunk mutex. For new BG, this function is called from | |
1124 | * btrfs_make_block_group() which is already taking the | |
1125 | * chunk mutex. Thus, we cannot call | |
1126 | * calculate_alloc_pointer() which takes extent buffer | |
1127 | * locks to avoid deadlock. | |
08e11a3d | 1128 | */ |
a94794d5 NA |
1129 | if (new) { |
1130 | cache->alloc_offset = 0; | |
1131 | goto out; | |
1132 | } | |
1133 | ret = calculate_alloc_pointer(cache, &last_alloc); | |
1134 | if (ret || map->num_stripes == num_conventional) { | |
1135 | if (!ret) | |
1136 | cache->alloc_offset = last_alloc; | |
1137 | else | |
1138 | btrfs_err(fs_info, | |
1139 | "zoned: failed to determine allocation offset of bg %llu", | |
1140 | cache->start); | |
1141 | goto out; | |
1142 | } | |
08e11a3d NA |
1143 | } |
1144 | ||
1145 | switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
1146 | case 0: /* single */ | |
1147 | cache->alloc_offset = alloc_offsets[0]; | |
1148 | break; | |
1149 | case BTRFS_BLOCK_GROUP_DUP: | |
1150 | case BTRFS_BLOCK_GROUP_RAID1: | |
1151 | case BTRFS_BLOCK_GROUP_RAID0: | |
1152 | case BTRFS_BLOCK_GROUP_RAID10: | |
1153 | case BTRFS_BLOCK_GROUP_RAID5: | |
1154 | case BTRFS_BLOCK_GROUP_RAID6: | |
1155 | /* non-single profiles are not supported yet */ | |
1156 | default: | |
1157 | btrfs_err(fs_info, "zoned: profile %s not yet supported", | |
1158 | btrfs_bg_type_to_raid_name(map->type)); | |
1159 | ret = -EINVAL; | |
1160 | goto out; | |
1161 | } | |
1162 | ||
1163 | out: | |
a94794d5 NA |
1164 | /* An extent is allocated after the write pointer */ |
1165 | if (!ret && num_conventional && last_alloc > cache->alloc_offset) { | |
1166 | btrfs_err(fs_info, | |
1167 | "zoned: got wrong write pointer in BG %llu: %llu > %llu", | |
1168 | logical, last_alloc, cache->alloc_offset); | |
1169 | ret = -EIO; | |
1170 | } | |
1171 | ||
0bc09ca1 NA |
1172 | if (!ret) |
1173 | cache->meta_write_pointer = cache->alloc_offset + cache->start; | |
1174 | ||
08e11a3d NA |
1175 | kfree(alloc_offsets); |
1176 | free_extent_map(em); | |
1177 | ||
1178 | return ret; | |
1179 | } | |
169e0da9 NA |
1180 | |
1181 | void btrfs_calc_zone_unusable(struct btrfs_block_group *cache) | |
1182 | { | |
1183 | u64 unusable, free; | |
1184 | ||
1185 | if (!btrfs_is_zoned(cache->fs_info)) | |
1186 | return; | |
1187 | ||
1188 | WARN_ON(cache->bytes_super != 0); | |
1189 | unusable = cache->alloc_offset - cache->used; | |
1190 | free = cache->length - cache->alloc_offset; | |
1191 | ||
1192 | /* We only need ->free_space in ALLOC_SEQ block groups */ | |
1193 | cache->last_byte_to_unpin = (u64)-1; | |
1194 | cache->cached = BTRFS_CACHE_FINISHED; | |
1195 | cache->free_space_ctl->free_space = free; | |
1196 | cache->zone_unusable = unusable; | |
1197 | ||
1198 | /* Should not have any excluded extents. Just in case, though */ | |
1199 | btrfs_free_excluded_extents(cache); | |
1200 | } | |
d3575156 NA |
1201 | |
1202 | void btrfs_redirty_list_add(struct btrfs_transaction *trans, | |
1203 | struct extent_buffer *eb) | |
1204 | { | |
1205 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
1206 | ||
1207 | if (!btrfs_is_zoned(fs_info) || | |
1208 | btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) || | |
1209 | !list_empty(&eb->release_list)) | |
1210 | return; | |
1211 | ||
1212 | set_extent_buffer_dirty(eb); | |
1213 | set_extent_bits_nowait(&trans->dirty_pages, eb->start, | |
1214 | eb->start + eb->len - 1, EXTENT_DIRTY); | |
1215 | memzero_extent_buffer(eb, 0, eb->len); | |
1216 | set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags); | |
1217 | ||
1218 | spin_lock(&trans->releasing_ebs_lock); | |
1219 | list_add_tail(&eb->release_list, &trans->releasing_ebs); | |
1220 | spin_unlock(&trans->releasing_ebs_lock); | |
1221 | atomic_inc(&eb->refs); | |
1222 | } | |
1223 | ||
1224 | void btrfs_free_redirty_list(struct btrfs_transaction *trans) | |
1225 | { | |
1226 | spin_lock(&trans->releasing_ebs_lock); | |
1227 | while (!list_empty(&trans->releasing_ebs)) { | |
1228 | struct extent_buffer *eb; | |
1229 | ||
1230 | eb = list_first_entry(&trans->releasing_ebs, | |
1231 | struct extent_buffer, release_list); | |
1232 | list_del_init(&eb->release_list); | |
1233 | free_extent_buffer(eb); | |
1234 | } | |
1235 | spin_unlock(&trans->releasing_ebs_lock); | |
1236 | } | |
08f45559 JT |
1237 | |
1238 | bool btrfs_use_zone_append(struct btrfs_inode *inode, struct extent_map *em) | |
1239 | { | |
1240 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1241 | struct btrfs_block_group *cache; | |
1242 | bool ret = false; | |
1243 | ||
1244 | if (!btrfs_is_zoned(fs_info)) | |
1245 | return false; | |
1246 | ||
1247 | if (!fs_info->max_zone_append_size) | |
1248 | return false; | |
1249 | ||
1250 | if (!is_data_inode(&inode->vfs_inode)) | |
1251 | return false; | |
1252 | ||
1253 | cache = btrfs_lookup_block_group(fs_info, em->block_start); | |
1254 | ASSERT(cache); | |
1255 | if (!cache) | |
1256 | return false; | |
1257 | ||
1258 | ret = cache->seq_zone; | |
1259 | btrfs_put_block_group(cache); | |
1260 | ||
1261 | return ret; | |
1262 | } | |
d8e3fb10 NA |
1263 | |
1264 | void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset, | |
1265 | struct bio *bio) | |
1266 | { | |
1267 | struct btrfs_ordered_extent *ordered; | |
1268 | const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; | |
1269 | ||
1270 | if (bio_op(bio) != REQ_OP_ZONE_APPEND) | |
1271 | return; | |
1272 | ||
1273 | ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset); | |
1274 | if (WARN_ON(!ordered)) | |
1275 | return; | |
1276 | ||
1277 | ordered->physical = physical; | |
1278 | ordered->disk = bio->bi_disk; | |
1279 | ordered->partno = bio->bi_partno; | |
1280 | ||
1281 | btrfs_put_ordered_extent(ordered); | |
1282 | } | |
1283 | ||
1284 | void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered) | |
1285 | { | |
1286 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
1287 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1288 | struct extent_map_tree *em_tree; | |
1289 | struct extent_map *em; | |
1290 | struct btrfs_ordered_sum *sum; | |
1291 | struct block_device *bdev; | |
1292 | u64 orig_logical = ordered->disk_bytenr; | |
1293 | u64 *logical = NULL; | |
1294 | int nr, stripe_len; | |
1295 | ||
1296 | /* Zoned devices should not have partitions. So, we can assume it is 0 */ | |
1297 | ASSERT(ordered->partno == 0); | |
1298 | bdev = bdgrab(ordered->disk->part0); | |
1299 | if (WARN_ON(!bdev)) | |
1300 | return; | |
1301 | ||
1302 | if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, bdev, | |
1303 | ordered->physical, &logical, &nr, | |
1304 | &stripe_len))) | |
1305 | goto out; | |
1306 | ||
1307 | WARN_ON(nr != 1); | |
1308 | ||
1309 | if (orig_logical == *logical) | |
1310 | goto out; | |
1311 | ||
1312 | ordered->disk_bytenr = *logical; | |
1313 | ||
1314 | em_tree = &inode->extent_tree; | |
1315 | write_lock(&em_tree->lock); | |
1316 | em = search_extent_mapping(em_tree, ordered->file_offset, | |
1317 | ordered->num_bytes); | |
1318 | em->block_start = *logical; | |
1319 | free_extent_map(em); | |
1320 | write_unlock(&em_tree->lock); | |
1321 | ||
1322 | list_for_each_entry(sum, &ordered->list, list) { | |
1323 | if (*logical < orig_logical) | |
1324 | sum->bytenr -= orig_logical - *logical; | |
1325 | else | |
1326 | sum->bytenr += *logical - orig_logical; | |
1327 | } | |
1328 | ||
1329 | out: | |
1330 | kfree(logical); | |
1331 | bdput(bdev); | |
1332 | } | |
0bc09ca1 NA |
1333 | |
1334 | bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info, | |
1335 | struct extent_buffer *eb, | |
1336 | struct btrfs_block_group **cache_ret) | |
1337 | { | |
1338 | struct btrfs_block_group *cache; | |
1339 | bool ret = true; | |
1340 | ||
1341 | if (!btrfs_is_zoned(fs_info)) | |
1342 | return true; | |
1343 | ||
1344 | cache = *cache_ret; | |
1345 | ||
1346 | if (cache && (eb->start < cache->start || | |
1347 | cache->start + cache->length <= eb->start)) { | |
1348 | btrfs_put_block_group(cache); | |
1349 | cache = NULL; | |
1350 | *cache_ret = NULL; | |
1351 | } | |
1352 | ||
1353 | if (!cache) | |
1354 | cache = btrfs_lookup_block_group(fs_info, eb->start); | |
1355 | ||
1356 | if (cache) { | |
1357 | if (cache->meta_write_pointer != eb->start) { | |
1358 | btrfs_put_block_group(cache); | |
1359 | cache = NULL; | |
1360 | ret = false; | |
1361 | } else { | |
1362 | cache->meta_write_pointer = eb->start + eb->len; | |
1363 | } | |
1364 | ||
1365 | *cache_ret = cache; | |
1366 | } | |
1367 | ||
1368 | return ret; | |
1369 | } | |
1370 | ||
1371 | void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache, | |
1372 | struct extent_buffer *eb) | |
1373 | { | |
1374 | if (!btrfs_is_zoned(eb->fs_info) || !cache) | |
1375 | return; | |
1376 | ||
1377 | ASSERT(cache->meta_write_pointer == eb->start + eb->len); | |
1378 | cache->meta_write_pointer = eb->start; | |
1379 | } | |
de17addc NA |
1380 | |
1381 | int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length) | |
1382 | { | |
1383 | if (!btrfs_dev_is_sequential(device, physical)) | |
1384 | return -EOPNOTSUPP; | |
1385 | ||
1386 | return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT, | |
1387 | length >> SECTOR_SHIFT, GFP_NOFS, 0); | |
1388 | } | |
7db1c5d1 NA |
1389 | |
1390 | static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical, | |
1391 | struct blk_zone *zone) | |
1392 | { | |
1393 | struct btrfs_bio *bbio = NULL; | |
1394 | u64 mapped_length = PAGE_SIZE; | |
1395 | unsigned int nofs_flag; | |
1396 | int nmirrors; | |
1397 | int i, ret; | |
1398 | ||
1399 | ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical, | |
1400 | &mapped_length, &bbio); | |
1401 | if (ret || !bbio || mapped_length < PAGE_SIZE) { | |
1402 | btrfs_put_bbio(bbio); | |
1403 | return -EIO; | |
1404 | } | |
1405 | ||
1406 | if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
1407 | return -EINVAL; | |
1408 | ||
1409 | nofs_flag = memalloc_nofs_save(); | |
1410 | nmirrors = (int)bbio->num_stripes; | |
1411 | for (i = 0; i < nmirrors; i++) { | |
1412 | u64 physical = bbio->stripes[i].physical; | |
1413 | struct btrfs_device *dev = bbio->stripes[i].dev; | |
1414 | ||
1415 | /* Missing device */ | |
1416 | if (!dev->bdev) | |
1417 | continue; | |
1418 | ||
1419 | ret = btrfs_get_dev_zone(dev, physical, zone); | |
1420 | /* Failing device */ | |
1421 | if (ret == -EIO || ret == -EOPNOTSUPP) | |
1422 | continue; | |
1423 | break; | |
1424 | } | |
1425 | memalloc_nofs_restore(nofs_flag); | |
1426 | ||
1427 | return ret; | |
1428 | } | |
1429 | ||
1430 | /* | |
1431 | * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by | |
1432 | * filling zeros between @physical_pos to a write pointer of dev-replace | |
1433 | * source device. | |
1434 | */ | |
1435 | int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical, | |
1436 | u64 physical_start, u64 physical_pos) | |
1437 | { | |
1438 | struct btrfs_fs_info *fs_info = tgt_dev->fs_info; | |
1439 | struct blk_zone zone; | |
1440 | u64 length; | |
1441 | u64 wp; | |
1442 | int ret; | |
1443 | ||
1444 | if (!btrfs_dev_is_sequential(tgt_dev, physical_pos)) | |
1445 | return 0; | |
1446 | ||
1447 | ret = read_zone_info(fs_info, logical, &zone); | |
1448 | if (ret) | |
1449 | return ret; | |
1450 | ||
1451 | wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT); | |
1452 | ||
1453 | if (physical_pos == wp) | |
1454 | return 0; | |
1455 | ||
1456 | if (physical_pos > wp) | |
1457 | return -EUCLEAN; | |
1458 | ||
1459 | length = wp - physical_pos; | |
1460 | return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length); | |
1461 | } |