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" |
5b316468 NA |
15 | |
16 | /* Maximum number of zones to report per blkdev_report_zones() call */ | |
17 | #define BTRFS_REPORT_NR_ZONES 4096 | |
08e11a3d NA |
18 | /* Invalid allocation pointer value for missing devices */ |
19 | #define WP_MISSING_DEV ((u64)-1) | |
20 | /* Pseudo write pointer value for conventional zone */ | |
21 | #define WP_CONVENTIONAL ((u64)-2) | |
5b316468 | 22 | |
12659251 NA |
23 | /* Number of superblock log zones */ |
24 | #define BTRFS_NR_SB_LOG_ZONES 2 | |
25 | ||
5b316468 NA |
26 | static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data) |
27 | { | |
28 | struct blk_zone *zones = data; | |
29 | ||
30 | memcpy(&zones[idx], zone, sizeof(*zone)); | |
31 | ||
32 | return 0; | |
33 | } | |
34 | ||
12659251 NA |
35 | static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones, |
36 | u64 *wp_ret) | |
37 | { | |
38 | bool empty[BTRFS_NR_SB_LOG_ZONES]; | |
39 | bool full[BTRFS_NR_SB_LOG_ZONES]; | |
40 | sector_t sector; | |
41 | ||
42 | ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL && | |
43 | zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL); | |
44 | ||
45 | empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY); | |
46 | empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY); | |
47 | full[0] = (zones[0].cond == BLK_ZONE_COND_FULL); | |
48 | full[1] = (zones[1].cond == BLK_ZONE_COND_FULL); | |
49 | ||
50 | /* | |
51 | * Possible states of log buffer zones | |
52 | * | |
53 | * Empty[0] In use[0] Full[0] | |
54 | * Empty[1] * x 0 | |
55 | * In use[1] 0 x 0 | |
56 | * Full[1] 1 1 C | |
57 | * | |
58 | * Log position: | |
59 | * *: Special case, no superblock is written | |
60 | * 0: Use write pointer of zones[0] | |
61 | * 1: Use write pointer of zones[1] | |
62 | * C: Compare super blcoks from zones[0] and zones[1], use the latest | |
63 | * one determined by generation | |
64 | * x: Invalid state | |
65 | */ | |
66 | ||
67 | if (empty[0] && empty[1]) { | |
68 | /* Special case to distinguish no superblock to read */ | |
69 | *wp_ret = zones[0].start << SECTOR_SHIFT; | |
70 | return -ENOENT; | |
71 | } else if (full[0] && full[1]) { | |
72 | /* Compare two super blocks */ | |
73 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
74 | struct page *page[BTRFS_NR_SB_LOG_ZONES]; | |
75 | struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES]; | |
76 | int i; | |
77 | ||
78 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { | |
79 | u64 bytenr; | |
80 | ||
81 | bytenr = ((zones[i].start + zones[i].len) | |
82 | << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE; | |
83 | ||
84 | page[i] = read_cache_page_gfp(mapping, | |
85 | bytenr >> PAGE_SHIFT, GFP_NOFS); | |
86 | if (IS_ERR(page[i])) { | |
87 | if (i == 1) | |
88 | btrfs_release_disk_super(super[0]); | |
89 | return PTR_ERR(page[i]); | |
90 | } | |
91 | super[i] = page_address(page[i]); | |
92 | } | |
93 | ||
94 | if (super[0]->generation > super[1]->generation) | |
95 | sector = zones[1].start; | |
96 | else | |
97 | sector = zones[0].start; | |
98 | ||
99 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) | |
100 | btrfs_release_disk_super(super[i]); | |
101 | } else if (!full[0] && (empty[1] || full[1])) { | |
102 | sector = zones[0].wp; | |
103 | } else if (full[0]) { | |
104 | sector = zones[1].wp; | |
105 | } else { | |
106 | return -EUCLEAN; | |
107 | } | |
108 | *wp_ret = sector << SECTOR_SHIFT; | |
109 | return 0; | |
110 | } | |
111 | ||
112 | /* | |
113 | * The following zones are reserved as the circular buffer on ZONED btrfs. | |
114 | * - The primary superblock: zones 0 and 1 | |
115 | * - The first copy: zones 16 and 17 | |
116 | * - The second copy: zones 1024 or zone at 256GB which is minimum, and | |
117 | * the following one | |
118 | */ | |
119 | static inline u32 sb_zone_number(int shift, int mirror) | |
120 | { | |
121 | ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX); | |
122 | ||
123 | switch (mirror) { | |
124 | case 0: return 0; | |
125 | case 1: return 16; | |
126 | case 2: return min_t(u64, btrfs_sb_offset(mirror) >> shift, 1024); | |
127 | } | |
128 | ||
129 | return 0; | |
130 | } | |
131 | ||
3c9daa09 JT |
132 | /* |
133 | * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block | |
134 | * device into static sized chunks and fake a conventional zone on each of | |
135 | * them. | |
136 | */ | |
137 | static int emulate_report_zones(struct btrfs_device *device, u64 pos, | |
138 | struct blk_zone *zones, unsigned int nr_zones) | |
139 | { | |
140 | const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT; | |
141 | sector_t bdev_size = bdev_nr_sectors(device->bdev); | |
142 | unsigned int i; | |
143 | ||
144 | pos >>= SECTOR_SHIFT; | |
145 | for (i = 0; i < nr_zones; i++) { | |
146 | zones[i].start = i * zone_sectors + pos; | |
147 | zones[i].len = zone_sectors; | |
148 | zones[i].capacity = zone_sectors; | |
149 | zones[i].wp = zones[i].start + zone_sectors; | |
150 | zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL; | |
151 | zones[i].cond = BLK_ZONE_COND_NOT_WP; | |
152 | ||
153 | if (zones[i].wp >= bdev_size) { | |
154 | i++; | |
155 | break; | |
156 | } | |
157 | } | |
158 | ||
159 | return i; | |
160 | } | |
161 | ||
5b316468 NA |
162 | static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos, |
163 | struct blk_zone *zones, unsigned int *nr_zones) | |
164 | { | |
165 | int ret; | |
166 | ||
167 | if (!*nr_zones) | |
168 | return 0; | |
169 | ||
3c9daa09 JT |
170 | if (!bdev_is_zoned(device->bdev)) { |
171 | ret = emulate_report_zones(device, pos, zones, *nr_zones); | |
172 | *nr_zones = ret; | |
173 | return 0; | |
174 | } | |
175 | ||
5b316468 NA |
176 | ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones, |
177 | copy_zone_info_cb, zones); | |
178 | if (ret < 0) { | |
179 | btrfs_err_in_rcu(device->fs_info, | |
180 | "zoned: failed to read zone %llu on %s (devid %llu)", | |
181 | pos, rcu_str_deref(device->name), | |
182 | device->devid); | |
183 | return ret; | |
184 | } | |
185 | *nr_zones = ret; | |
186 | if (!ret) | |
187 | return -EIO; | |
188 | ||
189 | return 0; | |
190 | } | |
191 | ||
3c9daa09 JT |
192 | /* The emulated zone size is determined from the size of device extent */ |
193 | static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info) | |
194 | { | |
195 | struct btrfs_path *path; | |
196 | struct btrfs_root *root = fs_info->dev_root; | |
197 | struct btrfs_key key; | |
198 | struct extent_buffer *leaf; | |
199 | struct btrfs_dev_extent *dext; | |
200 | int ret = 0; | |
201 | ||
202 | key.objectid = 1; | |
203 | key.type = BTRFS_DEV_EXTENT_KEY; | |
204 | key.offset = 0; | |
205 | ||
206 | path = btrfs_alloc_path(); | |
207 | if (!path) | |
208 | return -ENOMEM; | |
209 | ||
210 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
211 | if (ret < 0) | |
212 | goto out; | |
213 | ||
214 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
215 | ret = btrfs_next_item(root, path); | |
216 | if (ret < 0) | |
217 | goto out; | |
218 | /* No dev extents at all? Not good */ | |
219 | if (ret > 0) { | |
220 | ret = -EUCLEAN; | |
221 | goto out; | |
222 | } | |
223 | } | |
224 | ||
225 | leaf = path->nodes[0]; | |
226 | dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); | |
227 | fs_info->zone_size = btrfs_dev_extent_length(leaf, dext); | |
228 | ret = 0; | |
229 | ||
230 | out: | |
231 | btrfs_free_path(path); | |
232 | ||
233 | return ret; | |
234 | } | |
235 | ||
73651042 NA |
236 | int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info) |
237 | { | |
238 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
239 | struct btrfs_device *device; | |
240 | int ret = 0; | |
241 | ||
242 | /* fs_info->zone_size might not set yet. Use the incomapt flag here. */ | |
243 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
244 | return 0; | |
245 | ||
246 | mutex_lock(&fs_devices->device_list_mutex); | |
247 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
248 | /* We can skip reading of zone info for missing devices */ | |
249 | if (!device->bdev) | |
250 | continue; | |
251 | ||
252 | ret = btrfs_get_dev_zone_info(device); | |
253 | if (ret) | |
254 | break; | |
255 | } | |
256 | mutex_unlock(&fs_devices->device_list_mutex); | |
257 | ||
258 | return ret; | |
259 | } | |
260 | ||
5b316468 NA |
261 | int btrfs_get_dev_zone_info(struct btrfs_device *device) |
262 | { | |
3c9daa09 | 263 | struct btrfs_fs_info *fs_info = device->fs_info; |
5b316468 NA |
264 | struct btrfs_zoned_device_info *zone_info = NULL; |
265 | struct block_device *bdev = device->bdev; | |
862931c7 | 266 | struct request_queue *queue = bdev_get_queue(bdev); |
5b316468 NA |
267 | sector_t nr_sectors; |
268 | sector_t sector = 0; | |
269 | struct blk_zone *zones = NULL; | |
270 | unsigned int i, nreported = 0, nr_zones; | |
271 | unsigned int zone_sectors; | |
3c9daa09 | 272 | char *model, *emulated; |
5b316468 NA |
273 | int ret; |
274 | ||
3c9daa09 JT |
275 | /* |
276 | * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not | |
277 | * yet be set. | |
278 | */ | |
279 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
5b316468 NA |
280 | return 0; |
281 | ||
282 | if (device->zone_info) | |
283 | return 0; | |
284 | ||
285 | zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL); | |
286 | if (!zone_info) | |
287 | return -ENOMEM; | |
288 | ||
3c9daa09 JT |
289 | if (!bdev_is_zoned(bdev)) { |
290 | if (!fs_info->zone_size) { | |
291 | ret = calculate_emulated_zone_size(fs_info); | |
292 | if (ret) | |
293 | goto out; | |
294 | } | |
295 | ||
296 | ASSERT(fs_info->zone_size); | |
297 | zone_sectors = fs_info->zone_size >> SECTOR_SHIFT; | |
298 | } else { | |
299 | zone_sectors = bdev_zone_sectors(bdev); | |
300 | } | |
301 | ||
ac7ac461 | 302 | nr_sectors = bdev_nr_sectors(bdev); |
5b316468 NA |
303 | /* Check if it's power of 2 (see is_power_of_2) */ |
304 | ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0); | |
305 | zone_info->zone_size = zone_sectors << SECTOR_SHIFT; | |
306 | zone_info->zone_size_shift = ilog2(zone_info->zone_size); | |
862931c7 NA |
307 | zone_info->max_zone_append_size = |
308 | (u64)queue_max_zone_append_sectors(queue) << SECTOR_SHIFT; | |
5b316468 NA |
309 | zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors); |
310 | if (!IS_ALIGNED(nr_sectors, zone_sectors)) | |
311 | zone_info->nr_zones++; | |
312 | ||
313 | zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
314 | if (!zone_info->seq_zones) { | |
315 | ret = -ENOMEM; | |
316 | goto out; | |
317 | } | |
318 | ||
319 | zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
320 | if (!zone_info->empty_zones) { | |
321 | ret = -ENOMEM; | |
322 | goto out; | |
323 | } | |
324 | ||
325 | zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL); | |
326 | if (!zones) { | |
327 | ret = -ENOMEM; | |
328 | goto out; | |
329 | } | |
330 | ||
331 | /* Get zones type */ | |
332 | while (sector < nr_sectors) { | |
333 | nr_zones = BTRFS_REPORT_NR_ZONES; | |
334 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones, | |
335 | &nr_zones); | |
336 | if (ret) | |
337 | goto out; | |
338 | ||
339 | for (i = 0; i < nr_zones; i++) { | |
340 | if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ) | |
341 | __set_bit(nreported, zone_info->seq_zones); | |
342 | if (zones[i].cond == BLK_ZONE_COND_EMPTY) | |
343 | __set_bit(nreported, zone_info->empty_zones); | |
344 | nreported++; | |
345 | } | |
346 | sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len; | |
347 | } | |
348 | ||
349 | if (nreported != zone_info->nr_zones) { | |
350 | btrfs_err_in_rcu(device->fs_info, | |
351 | "inconsistent number of zones on %s (%u/%u)", | |
352 | rcu_str_deref(device->name), nreported, | |
353 | zone_info->nr_zones); | |
354 | ret = -EIO; | |
355 | goto out; | |
356 | } | |
357 | ||
12659251 NA |
358 | /* Validate superblock log */ |
359 | nr_zones = BTRFS_NR_SB_LOG_ZONES; | |
360 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
361 | u32 sb_zone; | |
362 | u64 sb_wp; | |
363 | int sb_pos = BTRFS_NR_SB_LOG_ZONES * i; | |
364 | ||
365 | sb_zone = sb_zone_number(zone_info->zone_size_shift, i); | |
366 | if (sb_zone + 1 >= zone_info->nr_zones) | |
367 | continue; | |
368 | ||
369 | sector = sb_zone << (zone_info->zone_size_shift - SECTOR_SHIFT); | |
370 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, | |
371 | &zone_info->sb_zones[sb_pos], | |
372 | &nr_zones); | |
373 | if (ret) | |
374 | goto out; | |
375 | ||
376 | if (nr_zones != BTRFS_NR_SB_LOG_ZONES) { | |
377 | btrfs_err_in_rcu(device->fs_info, | |
378 | "zoned: failed to read super block log zone info at devid %llu zone %u", | |
379 | device->devid, sb_zone); | |
380 | ret = -EUCLEAN; | |
381 | goto out; | |
382 | } | |
383 | ||
384 | /* | |
385 | * If zones[0] is conventional, always use the beggining of the | |
386 | * zone to record superblock. No need to validate in that case. | |
387 | */ | |
388 | if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type == | |
389 | BLK_ZONE_TYPE_CONVENTIONAL) | |
390 | continue; | |
391 | ||
392 | ret = sb_write_pointer(device->bdev, | |
393 | &zone_info->sb_zones[sb_pos], &sb_wp); | |
394 | if (ret != -ENOENT && ret) { | |
395 | btrfs_err_in_rcu(device->fs_info, | |
396 | "zoned: super block log zone corrupted devid %llu zone %u", | |
397 | device->devid, sb_zone); | |
398 | ret = -EUCLEAN; | |
399 | goto out; | |
400 | } | |
401 | } | |
402 | ||
403 | ||
5b316468 NA |
404 | kfree(zones); |
405 | ||
406 | device->zone_info = zone_info; | |
407 | ||
3c9daa09 JT |
408 | switch (bdev_zoned_model(bdev)) { |
409 | case BLK_ZONED_HM: | |
410 | model = "host-managed zoned"; | |
411 | emulated = ""; | |
412 | break; | |
413 | case BLK_ZONED_HA: | |
414 | model = "host-aware zoned"; | |
415 | emulated = ""; | |
416 | break; | |
417 | case BLK_ZONED_NONE: | |
418 | model = "regular"; | |
419 | emulated = "emulated "; | |
420 | break; | |
421 | default: | |
422 | /* Just in case */ | |
423 | btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s", | |
424 | bdev_zoned_model(bdev), | |
425 | rcu_str_deref(device->name)); | |
426 | ret = -EOPNOTSUPP; | |
427 | goto out_free_zone_info; | |
428 | } | |
429 | ||
430 | btrfs_info_in_rcu(fs_info, | |
431 | "%s block device %s, %u %szones of %llu bytes", | |
432 | model, rcu_str_deref(device->name), zone_info->nr_zones, | |
433 | emulated, zone_info->zone_size); | |
5b316468 NA |
434 | |
435 | return 0; | |
436 | ||
437 | out: | |
438 | kfree(zones); | |
3c9daa09 | 439 | out_free_zone_info: |
5b316468 NA |
440 | bitmap_free(zone_info->empty_zones); |
441 | bitmap_free(zone_info->seq_zones); | |
442 | kfree(zone_info); | |
3c9daa09 | 443 | device->zone_info = NULL; |
5b316468 NA |
444 | |
445 | return ret; | |
446 | } | |
447 | ||
448 | void btrfs_destroy_dev_zone_info(struct btrfs_device *device) | |
449 | { | |
450 | struct btrfs_zoned_device_info *zone_info = device->zone_info; | |
451 | ||
452 | if (!zone_info) | |
453 | return; | |
454 | ||
455 | bitmap_free(zone_info->seq_zones); | |
456 | bitmap_free(zone_info->empty_zones); | |
457 | kfree(zone_info); | |
458 | device->zone_info = NULL; | |
459 | } | |
460 | ||
461 | int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos, | |
462 | struct blk_zone *zone) | |
463 | { | |
464 | unsigned int nr_zones = 1; | |
465 | int ret; | |
466 | ||
467 | ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones); | |
468 | if (ret != 0 || !nr_zones) | |
469 | return ret ? ret : -EIO; | |
470 | ||
471 | return 0; | |
472 | } | |
b70f5097 NA |
473 | |
474 | int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info) | |
475 | { | |
476 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
477 | struct btrfs_device *device; | |
478 | u64 zoned_devices = 0; | |
479 | u64 nr_devices = 0; | |
480 | u64 zone_size = 0; | |
862931c7 | 481 | u64 max_zone_append_size = 0; |
3c9daa09 | 482 | const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED); |
b70f5097 NA |
483 | int ret = 0; |
484 | ||
485 | /* Count zoned devices */ | |
486 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
487 | enum blk_zoned_model model; | |
488 | ||
489 | if (!device->bdev) | |
490 | continue; | |
491 | ||
492 | model = bdev_zoned_model(device->bdev); | |
3c9daa09 JT |
493 | /* |
494 | * A Host-Managed zoned device must be used as a zoned device. | |
495 | * A Host-Aware zoned device and a non-zoned devices can be | |
496 | * treated as a zoned device, if ZONED flag is enabled in the | |
497 | * superblock. | |
498 | */ | |
b70f5097 | 499 | if (model == BLK_ZONED_HM || |
3c9daa09 JT |
500 | (model == BLK_ZONED_HA && incompat_zoned) || |
501 | (model == BLK_ZONED_NONE && incompat_zoned)) { | |
502 | struct btrfs_zoned_device_info *zone_info = | |
503 | device->zone_info; | |
862931c7 NA |
504 | |
505 | zone_info = device->zone_info; | |
b70f5097 NA |
506 | zoned_devices++; |
507 | if (!zone_size) { | |
862931c7 NA |
508 | zone_size = zone_info->zone_size; |
509 | } else if (zone_info->zone_size != zone_size) { | |
b70f5097 NA |
510 | btrfs_err(fs_info, |
511 | "zoned: unequal block device zone sizes: have %llu found %llu", | |
512 | device->zone_info->zone_size, | |
513 | zone_size); | |
514 | ret = -EINVAL; | |
515 | goto out; | |
516 | } | |
862931c7 NA |
517 | if (!max_zone_append_size || |
518 | (zone_info->max_zone_append_size && | |
519 | zone_info->max_zone_append_size < max_zone_append_size)) | |
520 | max_zone_append_size = | |
521 | zone_info->max_zone_append_size; | |
b70f5097 NA |
522 | } |
523 | nr_devices++; | |
524 | } | |
525 | ||
526 | if (!zoned_devices && !incompat_zoned) | |
527 | goto out; | |
528 | ||
529 | if (!zoned_devices && incompat_zoned) { | |
530 | /* No zoned block device found on ZONED filesystem */ | |
531 | btrfs_err(fs_info, | |
532 | "zoned: no zoned devices found on a zoned filesystem"); | |
533 | ret = -EINVAL; | |
534 | goto out; | |
535 | } | |
536 | ||
537 | if (zoned_devices && !incompat_zoned) { | |
538 | btrfs_err(fs_info, | |
539 | "zoned: mode not enabled but zoned device found"); | |
540 | ret = -EINVAL; | |
541 | goto out; | |
542 | } | |
543 | ||
544 | if (zoned_devices != nr_devices) { | |
545 | btrfs_err(fs_info, | |
546 | "zoned: cannot mix zoned and regular devices"); | |
547 | ret = -EINVAL; | |
548 | goto out; | |
549 | } | |
550 | ||
551 | /* | |
552 | * stripe_size is always aligned to BTRFS_STRIPE_LEN in | |
553 | * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size, | |
554 | * check the alignment here. | |
555 | */ | |
556 | if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) { | |
557 | btrfs_err(fs_info, | |
558 | "zoned: zone size %llu not aligned to stripe %u", | |
559 | zone_size, BTRFS_STRIPE_LEN); | |
560 | ret = -EINVAL; | |
561 | goto out; | |
562 | } | |
563 | ||
a589dde0 NA |
564 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
565 | btrfs_err(fs_info, "zoned: mixed block groups not supported"); | |
566 | ret = -EINVAL; | |
567 | goto out; | |
568 | } | |
569 | ||
b70f5097 | 570 | fs_info->zone_size = zone_size; |
862931c7 | 571 | fs_info->max_zone_append_size = max_zone_append_size; |
1cd6121f | 572 | fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED; |
b70f5097 | 573 | |
b53429ba JT |
574 | /* |
575 | * Check mount options here, because we might change fs_info->zoned | |
576 | * from fs_info->zone_size. | |
577 | */ | |
578 | ret = btrfs_check_mountopts_zoned(fs_info); | |
579 | if (ret) | |
580 | goto out; | |
581 | ||
b70f5097 NA |
582 | btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size); |
583 | out: | |
584 | return ret; | |
585 | } | |
5d1ab66c NA |
586 | |
587 | int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info) | |
588 | { | |
589 | if (!btrfs_is_zoned(info)) | |
590 | return 0; | |
591 | ||
592 | /* | |
593 | * Space cache writing is not COWed. Disable that to avoid write errors | |
594 | * in sequential zones. | |
595 | */ | |
596 | if (btrfs_test_opt(info, SPACE_CACHE)) { | |
597 | btrfs_err(info, "zoned: space cache v1 is not supported"); | |
598 | return -EINVAL; | |
599 | } | |
600 | ||
d206e9c9 NA |
601 | if (btrfs_test_opt(info, NODATACOW)) { |
602 | btrfs_err(info, "zoned: NODATACOW not supported"); | |
603 | return -EINVAL; | |
604 | } | |
605 | ||
5d1ab66c NA |
606 | return 0; |
607 | } | |
12659251 NA |
608 | |
609 | static int sb_log_location(struct block_device *bdev, struct blk_zone *zones, | |
610 | int rw, u64 *bytenr_ret) | |
611 | { | |
612 | u64 wp; | |
613 | int ret; | |
614 | ||
615 | if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) { | |
616 | *bytenr_ret = zones[0].start << SECTOR_SHIFT; | |
617 | return 0; | |
618 | } | |
619 | ||
620 | ret = sb_write_pointer(bdev, zones, &wp); | |
621 | if (ret != -ENOENT && ret < 0) | |
622 | return ret; | |
623 | ||
624 | if (rw == WRITE) { | |
625 | struct blk_zone *reset = NULL; | |
626 | ||
627 | if (wp == zones[0].start << SECTOR_SHIFT) | |
628 | reset = &zones[0]; | |
629 | else if (wp == zones[1].start << SECTOR_SHIFT) | |
630 | reset = &zones[1]; | |
631 | ||
632 | if (reset && reset->cond != BLK_ZONE_COND_EMPTY) { | |
633 | ASSERT(reset->cond == BLK_ZONE_COND_FULL); | |
634 | ||
635 | ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
636 | reset->start, reset->len, | |
637 | GFP_NOFS); | |
638 | if (ret) | |
639 | return ret; | |
640 | ||
641 | reset->cond = BLK_ZONE_COND_EMPTY; | |
642 | reset->wp = reset->start; | |
643 | } | |
644 | } else if (ret != -ENOENT) { | |
645 | /* For READ, we want the precious one */ | |
646 | if (wp == zones[0].start << SECTOR_SHIFT) | |
647 | wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT; | |
648 | wp -= BTRFS_SUPER_INFO_SIZE; | |
649 | } | |
650 | ||
651 | *bytenr_ret = wp; | |
652 | return 0; | |
653 | ||
654 | } | |
655 | ||
656 | int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw, | |
657 | u64 *bytenr_ret) | |
658 | { | |
659 | struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES]; | |
660 | unsigned int zone_sectors; | |
661 | u32 sb_zone; | |
662 | int ret; | |
12659251 NA |
663 | u8 zone_sectors_shift; |
664 | sector_t nr_sectors; | |
665 | u32 nr_zones; | |
666 | ||
667 | if (!bdev_is_zoned(bdev)) { | |
668 | *bytenr_ret = btrfs_sb_offset(mirror); | |
669 | return 0; | |
670 | } | |
671 | ||
672 | ASSERT(rw == READ || rw == WRITE); | |
673 | ||
674 | zone_sectors = bdev_zone_sectors(bdev); | |
675 | if (!is_power_of_2(zone_sectors)) | |
676 | return -EINVAL; | |
12659251 | 677 | zone_sectors_shift = ilog2(zone_sectors); |
ac7ac461 | 678 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
679 | nr_zones = nr_sectors >> zone_sectors_shift; |
680 | ||
681 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
682 | if (sb_zone + 1 >= nr_zones) | |
683 | return -ENOENT; | |
684 | ||
685 | ret = blkdev_report_zones(bdev, sb_zone << zone_sectors_shift, | |
686 | BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb, | |
687 | zones); | |
688 | if (ret < 0) | |
689 | return ret; | |
690 | if (ret != BTRFS_NR_SB_LOG_ZONES) | |
691 | return -EIO; | |
692 | ||
693 | return sb_log_location(bdev, zones, rw, bytenr_ret); | |
694 | } | |
695 | ||
696 | int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw, | |
697 | u64 *bytenr_ret) | |
698 | { | |
699 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
700 | u32 zone_num; | |
701 | ||
d6639b35 NA |
702 | /* |
703 | * For a zoned filesystem on a non-zoned block device, use the same | |
704 | * super block locations as regular filesystem. Doing so, the super | |
705 | * block can always be retrieved and the zoned flag of the volume | |
706 | * detected from the super block information. | |
707 | */ | |
708 | if (!bdev_is_zoned(device->bdev)) { | |
12659251 NA |
709 | *bytenr_ret = btrfs_sb_offset(mirror); |
710 | return 0; | |
711 | } | |
712 | ||
713 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
714 | if (zone_num + 1 >= zinfo->nr_zones) | |
715 | return -ENOENT; | |
716 | ||
717 | return sb_log_location(device->bdev, | |
718 | &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror], | |
719 | rw, bytenr_ret); | |
720 | } | |
721 | ||
722 | static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo, | |
723 | int mirror) | |
724 | { | |
725 | u32 zone_num; | |
726 | ||
727 | if (!zinfo) | |
728 | return false; | |
729 | ||
730 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
731 | if (zone_num + 1 >= zinfo->nr_zones) | |
732 | return false; | |
733 | ||
734 | if (!test_bit(zone_num, zinfo->seq_zones)) | |
735 | return false; | |
736 | ||
737 | return true; | |
738 | } | |
739 | ||
740 | void btrfs_advance_sb_log(struct btrfs_device *device, int mirror) | |
741 | { | |
742 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
743 | struct blk_zone *zone; | |
744 | ||
745 | if (!is_sb_log_zone(zinfo, mirror)) | |
746 | return; | |
747 | ||
748 | zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror]; | |
749 | if (zone->cond != BLK_ZONE_COND_FULL) { | |
750 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
751 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
752 | ||
753 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
754 | ||
755 | if (zone->wp == zone->start + zone->len) | |
756 | zone->cond = BLK_ZONE_COND_FULL; | |
757 | ||
758 | return; | |
759 | } | |
760 | ||
761 | zone++; | |
762 | ASSERT(zone->cond != BLK_ZONE_COND_FULL); | |
763 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
764 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
765 | ||
766 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
767 | ||
768 | if (zone->wp == zone->start + zone->len) | |
769 | zone->cond = BLK_ZONE_COND_FULL; | |
770 | } | |
771 | ||
772 | int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror) | |
773 | { | |
774 | sector_t zone_sectors; | |
775 | sector_t nr_sectors; | |
776 | u8 zone_sectors_shift; | |
777 | u32 sb_zone; | |
778 | u32 nr_zones; | |
779 | ||
780 | zone_sectors = bdev_zone_sectors(bdev); | |
781 | zone_sectors_shift = ilog2(zone_sectors); | |
ac7ac461 | 782 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
783 | nr_zones = nr_sectors >> zone_sectors_shift; |
784 | ||
785 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
786 | if (sb_zone + 1 >= nr_zones) | |
787 | return -ENOENT; | |
788 | ||
789 | return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
790 | sb_zone << zone_sectors_shift, | |
791 | zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS); | |
792 | } | |
1cd6121f NA |
793 | |
794 | /** | |
795 | * btrfs_find_allocatable_zones - find allocatable zones within a given region | |
796 | * | |
797 | * @device: the device to allocate a region on | |
798 | * @hole_start: the position of the hole to allocate the region | |
799 | * @num_bytes: size of wanted region | |
800 | * @hole_end: the end of the hole | |
801 | * @return: position of allocatable zones | |
802 | * | |
803 | * Allocatable region should not contain any superblock locations. | |
804 | */ | |
805 | u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start, | |
806 | u64 hole_end, u64 num_bytes) | |
807 | { | |
808 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
809 | const u8 shift = zinfo->zone_size_shift; | |
810 | u64 nzones = num_bytes >> shift; | |
811 | u64 pos = hole_start; | |
812 | u64 begin, end; | |
813 | bool have_sb; | |
814 | int i; | |
815 | ||
816 | ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size)); | |
817 | ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size)); | |
818 | ||
819 | while (pos < hole_end) { | |
820 | begin = pos >> shift; | |
821 | end = begin + nzones; | |
822 | ||
823 | if (end > zinfo->nr_zones) | |
824 | return hole_end; | |
825 | ||
826 | /* Check if zones in the region are all empty */ | |
827 | if (btrfs_dev_is_sequential(device, pos) && | |
828 | find_next_zero_bit(zinfo->empty_zones, end, begin) != end) { | |
829 | pos += zinfo->zone_size; | |
830 | continue; | |
831 | } | |
832 | ||
833 | have_sb = false; | |
834 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
835 | u32 sb_zone; | |
836 | u64 sb_pos; | |
837 | ||
838 | sb_zone = sb_zone_number(shift, i); | |
839 | if (!(end <= sb_zone || | |
840 | sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) { | |
841 | have_sb = true; | |
842 | pos = ((u64)sb_zone + BTRFS_NR_SB_LOG_ZONES) << shift; | |
843 | break; | |
844 | } | |
845 | ||
846 | /* We also need to exclude regular superblock positions */ | |
847 | sb_pos = btrfs_sb_offset(i); | |
848 | if (!(pos + num_bytes <= sb_pos || | |
849 | sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) { | |
850 | have_sb = true; | |
851 | pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE, | |
852 | zinfo->zone_size); | |
853 | break; | |
854 | } | |
855 | } | |
856 | if (!have_sb) | |
857 | break; | |
858 | } | |
859 | ||
860 | return pos; | |
861 | } | |
862 | ||
863 | int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical, | |
864 | u64 length, u64 *bytes) | |
865 | { | |
866 | int ret; | |
867 | ||
868 | *bytes = 0; | |
869 | ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET, | |
870 | physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT, | |
871 | GFP_NOFS); | |
872 | if (ret) | |
873 | return ret; | |
874 | ||
875 | *bytes = length; | |
876 | while (length) { | |
877 | btrfs_dev_set_zone_empty(device, physical); | |
878 | physical += device->zone_info->zone_size; | |
879 | length -= device->zone_info->zone_size; | |
880 | } | |
881 | ||
882 | return 0; | |
883 | } | |
884 | ||
885 | int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size) | |
886 | { | |
887 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
888 | const u8 shift = zinfo->zone_size_shift; | |
889 | unsigned long begin = start >> shift; | |
890 | unsigned long end = (start + size) >> shift; | |
891 | u64 pos; | |
892 | int ret; | |
893 | ||
894 | ASSERT(IS_ALIGNED(start, zinfo->zone_size)); | |
895 | ASSERT(IS_ALIGNED(size, zinfo->zone_size)); | |
896 | ||
897 | if (end > zinfo->nr_zones) | |
898 | return -ERANGE; | |
899 | ||
900 | /* All the zones are conventional */ | |
901 | if (find_next_bit(zinfo->seq_zones, begin, end) == end) | |
902 | return 0; | |
903 | ||
904 | /* All the zones are sequential and empty */ | |
905 | if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end && | |
906 | find_next_zero_bit(zinfo->empty_zones, begin, end) == end) | |
907 | return 0; | |
908 | ||
909 | for (pos = start; pos < start + size; pos += zinfo->zone_size) { | |
910 | u64 reset_bytes; | |
911 | ||
912 | if (!btrfs_dev_is_sequential(device, pos) || | |
913 | btrfs_dev_is_empty_zone(device, pos)) | |
914 | continue; | |
915 | ||
916 | /* Free regions should be empty */ | |
917 | btrfs_warn_in_rcu( | |
918 | device->fs_info, | |
919 | "zoned: resetting device %s (devid %llu) zone %llu for allocation", | |
920 | rcu_str_deref(device->name), device->devid, pos >> shift); | |
921 | WARN_ON_ONCE(1); | |
922 | ||
923 | ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size, | |
924 | &reset_bytes); | |
925 | if (ret) | |
926 | return ret; | |
927 | } | |
928 | ||
929 | return 0; | |
930 | } | |
08e11a3d | 931 | |
a94794d5 NA |
932 | /* |
933 | * Calculate an allocation pointer from the extent allocation information | |
934 | * for a block group consist of conventional zones. It is pointed to the | |
935 | * end of the highest addressed extent in the block group as an allocation | |
936 | * offset. | |
937 | */ | |
938 | static int calculate_alloc_pointer(struct btrfs_block_group *cache, | |
939 | u64 *offset_ret) | |
940 | { | |
941 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
942 | struct btrfs_root *root = fs_info->extent_root; | |
943 | struct btrfs_path *path; | |
944 | struct btrfs_key key; | |
945 | struct btrfs_key found_key; | |
946 | int ret; | |
947 | u64 length; | |
948 | ||
949 | path = btrfs_alloc_path(); | |
950 | if (!path) | |
951 | return -ENOMEM; | |
952 | ||
953 | key.objectid = cache->start + cache->length; | |
954 | key.type = 0; | |
955 | key.offset = 0; | |
956 | ||
957 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
958 | /* We should not find the exact match */ | |
959 | if (!ret) | |
960 | ret = -EUCLEAN; | |
961 | if (ret < 0) | |
962 | goto out; | |
963 | ||
964 | ret = btrfs_previous_extent_item(root, path, cache->start); | |
965 | if (ret) { | |
966 | if (ret == 1) { | |
967 | ret = 0; | |
968 | *offset_ret = 0; | |
969 | } | |
970 | goto out; | |
971 | } | |
972 | ||
973 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
974 | ||
975 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY) | |
976 | length = found_key.offset; | |
977 | else | |
978 | length = fs_info->nodesize; | |
979 | ||
980 | if (!(found_key.objectid >= cache->start && | |
981 | found_key.objectid + length <= cache->start + cache->length)) { | |
982 | ret = -EUCLEAN; | |
983 | goto out; | |
984 | } | |
985 | *offset_ret = found_key.objectid + length - cache->start; | |
986 | ret = 0; | |
987 | ||
988 | out: | |
989 | btrfs_free_path(path); | |
990 | return ret; | |
991 | } | |
992 | ||
993 | int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) | |
08e11a3d NA |
994 | { |
995 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
996 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
997 | struct extent_map *em; | |
998 | struct map_lookup *map; | |
999 | struct btrfs_device *device; | |
1000 | u64 logical = cache->start; | |
1001 | u64 length = cache->length; | |
1002 | u64 physical = 0; | |
1003 | int ret; | |
1004 | int i; | |
1005 | unsigned int nofs_flag; | |
1006 | u64 *alloc_offsets = NULL; | |
a94794d5 | 1007 | u64 last_alloc = 0; |
08e11a3d NA |
1008 | u32 num_sequential = 0, num_conventional = 0; |
1009 | ||
1010 | if (!btrfs_is_zoned(fs_info)) | |
1011 | return 0; | |
1012 | ||
1013 | /* Sanity check */ | |
1014 | if (!IS_ALIGNED(length, fs_info->zone_size)) { | |
1015 | btrfs_err(fs_info, | |
1016 | "zoned: block group %llu len %llu unaligned to zone size %llu", | |
1017 | logical, length, fs_info->zone_size); | |
1018 | return -EIO; | |
1019 | } | |
1020 | ||
1021 | /* Get the chunk mapping */ | |
1022 | read_lock(&em_tree->lock); | |
1023 | em = lookup_extent_mapping(em_tree, logical, length); | |
1024 | read_unlock(&em_tree->lock); | |
1025 | ||
1026 | if (!em) | |
1027 | return -EINVAL; | |
1028 | ||
1029 | map = em->map_lookup; | |
1030 | ||
1031 | alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS); | |
1032 | if (!alloc_offsets) { | |
1033 | free_extent_map(em); | |
1034 | return -ENOMEM; | |
1035 | } | |
1036 | ||
1037 | for (i = 0; i < map->num_stripes; i++) { | |
1038 | bool is_sequential; | |
1039 | struct blk_zone zone; | |
6143c23c NA |
1040 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
1041 | int dev_replace_is_ongoing = 0; | |
08e11a3d NA |
1042 | |
1043 | device = map->stripes[i].dev; | |
1044 | physical = map->stripes[i].physical; | |
1045 | ||
1046 | if (device->bdev == NULL) { | |
1047 | alloc_offsets[i] = WP_MISSING_DEV; | |
1048 | continue; | |
1049 | } | |
1050 | ||
1051 | is_sequential = btrfs_dev_is_sequential(device, physical); | |
1052 | if (is_sequential) | |
1053 | num_sequential++; | |
1054 | else | |
1055 | num_conventional++; | |
1056 | ||
1057 | if (!is_sequential) { | |
1058 | alloc_offsets[i] = WP_CONVENTIONAL; | |
1059 | continue; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * This zone will be used for allocation, so mark this zone | |
1064 | * non-empty. | |
1065 | */ | |
1066 | btrfs_dev_clear_zone_empty(device, physical); | |
1067 | ||
6143c23c NA |
1068 | down_read(&dev_replace->rwsem); |
1069 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
1070 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) | |
1071 | btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical); | |
1072 | up_read(&dev_replace->rwsem); | |
1073 | ||
08e11a3d NA |
1074 | /* |
1075 | * The group is mapped to a sequential zone. Get the zone write | |
1076 | * pointer to determine the allocation offset within the zone. | |
1077 | */ | |
1078 | WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size)); | |
1079 | nofs_flag = memalloc_nofs_save(); | |
1080 | ret = btrfs_get_dev_zone(device, physical, &zone); | |
1081 | memalloc_nofs_restore(nofs_flag); | |
1082 | if (ret == -EIO || ret == -EOPNOTSUPP) { | |
1083 | ret = 0; | |
1084 | alloc_offsets[i] = WP_MISSING_DEV; | |
1085 | continue; | |
1086 | } else if (ret) { | |
1087 | goto out; | |
1088 | } | |
1089 | ||
1090 | switch (zone.cond) { | |
1091 | case BLK_ZONE_COND_OFFLINE: | |
1092 | case BLK_ZONE_COND_READONLY: | |
1093 | btrfs_err(fs_info, | |
1094 | "zoned: offline/readonly zone %llu on device %s (devid %llu)", | |
1095 | physical >> device->zone_info->zone_size_shift, | |
1096 | rcu_str_deref(device->name), device->devid); | |
1097 | alloc_offsets[i] = WP_MISSING_DEV; | |
1098 | break; | |
1099 | case BLK_ZONE_COND_EMPTY: | |
1100 | alloc_offsets[i] = 0; | |
1101 | break; | |
1102 | case BLK_ZONE_COND_FULL: | |
1103 | alloc_offsets[i] = fs_info->zone_size; | |
1104 | break; | |
1105 | default: | |
1106 | /* Partially used zone */ | |
1107 | alloc_offsets[i] = | |
1108 | ((zone.wp - zone.start) << SECTOR_SHIFT); | |
1109 | break; | |
1110 | } | |
1111 | } | |
1112 | ||
08f45559 JT |
1113 | if (num_sequential > 0) |
1114 | cache->seq_zone = true; | |
1115 | ||
08e11a3d NA |
1116 | if (num_conventional > 0) { |
1117 | /* | |
a94794d5 NA |
1118 | * Avoid calling calculate_alloc_pointer() for new BG. It |
1119 | * is no use for new BG. It must be always 0. | |
1120 | * | |
1121 | * Also, we have a lock chain of extent buffer lock -> | |
1122 | * chunk mutex. For new BG, this function is called from | |
1123 | * btrfs_make_block_group() which is already taking the | |
1124 | * chunk mutex. Thus, we cannot call | |
1125 | * calculate_alloc_pointer() which takes extent buffer | |
1126 | * locks to avoid deadlock. | |
08e11a3d | 1127 | */ |
a94794d5 NA |
1128 | if (new) { |
1129 | cache->alloc_offset = 0; | |
1130 | goto out; | |
1131 | } | |
1132 | ret = calculate_alloc_pointer(cache, &last_alloc); | |
1133 | if (ret || map->num_stripes == num_conventional) { | |
1134 | if (!ret) | |
1135 | cache->alloc_offset = last_alloc; | |
1136 | else | |
1137 | btrfs_err(fs_info, | |
1138 | "zoned: failed to determine allocation offset of bg %llu", | |
1139 | cache->start); | |
1140 | goto out; | |
1141 | } | |
08e11a3d NA |
1142 | } |
1143 | ||
1144 | switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
1145 | case 0: /* single */ | |
1146 | cache->alloc_offset = alloc_offsets[0]; | |
1147 | break; | |
1148 | case BTRFS_BLOCK_GROUP_DUP: | |
1149 | case BTRFS_BLOCK_GROUP_RAID1: | |
1150 | case BTRFS_BLOCK_GROUP_RAID0: | |
1151 | case BTRFS_BLOCK_GROUP_RAID10: | |
1152 | case BTRFS_BLOCK_GROUP_RAID5: | |
1153 | case BTRFS_BLOCK_GROUP_RAID6: | |
1154 | /* non-single profiles are not supported yet */ | |
1155 | default: | |
1156 | btrfs_err(fs_info, "zoned: profile %s not yet supported", | |
1157 | btrfs_bg_type_to_raid_name(map->type)); | |
1158 | ret = -EINVAL; | |
1159 | goto out; | |
1160 | } | |
1161 | ||
1162 | out: | |
a94794d5 NA |
1163 | /* An extent is allocated after the write pointer */ |
1164 | if (!ret && num_conventional && last_alloc > cache->alloc_offset) { | |
1165 | btrfs_err(fs_info, | |
1166 | "zoned: got wrong write pointer in BG %llu: %llu > %llu", | |
1167 | logical, last_alloc, cache->alloc_offset); | |
1168 | ret = -EIO; | |
1169 | } | |
1170 | ||
0bc09ca1 NA |
1171 | if (!ret) |
1172 | cache->meta_write_pointer = cache->alloc_offset + cache->start; | |
1173 | ||
08e11a3d NA |
1174 | kfree(alloc_offsets); |
1175 | free_extent_map(em); | |
1176 | ||
1177 | return ret; | |
1178 | } | |
169e0da9 NA |
1179 | |
1180 | void btrfs_calc_zone_unusable(struct btrfs_block_group *cache) | |
1181 | { | |
1182 | u64 unusable, free; | |
1183 | ||
1184 | if (!btrfs_is_zoned(cache->fs_info)) | |
1185 | return; | |
1186 | ||
1187 | WARN_ON(cache->bytes_super != 0); | |
1188 | unusable = cache->alloc_offset - cache->used; | |
1189 | free = cache->length - cache->alloc_offset; | |
1190 | ||
1191 | /* We only need ->free_space in ALLOC_SEQ block groups */ | |
1192 | cache->last_byte_to_unpin = (u64)-1; | |
1193 | cache->cached = BTRFS_CACHE_FINISHED; | |
1194 | cache->free_space_ctl->free_space = free; | |
1195 | cache->zone_unusable = unusable; | |
1196 | ||
1197 | /* Should not have any excluded extents. Just in case, though */ | |
1198 | btrfs_free_excluded_extents(cache); | |
1199 | } | |
d3575156 NA |
1200 | |
1201 | void btrfs_redirty_list_add(struct btrfs_transaction *trans, | |
1202 | struct extent_buffer *eb) | |
1203 | { | |
1204 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
1205 | ||
1206 | if (!btrfs_is_zoned(fs_info) || | |
1207 | btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) || | |
1208 | !list_empty(&eb->release_list)) | |
1209 | return; | |
1210 | ||
1211 | set_extent_buffer_dirty(eb); | |
1212 | set_extent_bits_nowait(&trans->dirty_pages, eb->start, | |
1213 | eb->start + eb->len - 1, EXTENT_DIRTY); | |
1214 | memzero_extent_buffer(eb, 0, eb->len); | |
1215 | set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags); | |
1216 | ||
1217 | spin_lock(&trans->releasing_ebs_lock); | |
1218 | list_add_tail(&eb->release_list, &trans->releasing_ebs); | |
1219 | spin_unlock(&trans->releasing_ebs_lock); | |
1220 | atomic_inc(&eb->refs); | |
1221 | } | |
1222 | ||
1223 | void btrfs_free_redirty_list(struct btrfs_transaction *trans) | |
1224 | { | |
1225 | spin_lock(&trans->releasing_ebs_lock); | |
1226 | while (!list_empty(&trans->releasing_ebs)) { | |
1227 | struct extent_buffer *eb; | |
1228 | ||
1229 | eb = list_first_entry(&trans->releasing_ebs, | |
1230 | struct extent_buffer, release_list); | |
1231 | list_del_init(&eb->release_list); | |
1232 | free_extent_buffer(eb); | |
1233 | } | |
1234 | spin_unlock(&trans->releasing_ebs_lock); | |
1235 | } | |
08f45559 JT |
1236 | |
1237 | bool btrfs_use_zone_append(struct btrfs_inode *inode, struct extent_map *em) | |
1238 | { | |
1239 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1240 | struct btrfs_block_group *cache; | |
1241 | bool ret = false; | |
1242 | ||
1243 | if (!btrfs_is_zoned(fs_info)) | |
1244 | return false; | |
1245 | ||
1246 | if (!fs_info->max_zone_append_size) | |
1247 | return false; | |
1248 | ||
1249 | if (!is_data_inode(&inode->vfs_inode)) | |
1250 | return false; | |
1251 | ||
1252 | cache = btrfs_lookup_block_group(fs_info, em->block_start); | |
1253 | ASSERT(cache); | |
1254 | if (!cache) | |
1255 | return false; | |
1256 | ||
1257 | ret = cache->seq_zone; | |
1258 | btrfs_put_block_group(cache); | |
1259 | ||
1260 | return ret; | |
1261 | } | |
d8e3fb10 NA |
1262 | |
1263 | void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset, | |
1264 | struct bio *bio) | |
1265 | { | |
1266 | struct btrfs_ordered_extent *ordered; | |
1267 | const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; | |
1268 | ||
1269 | if (bio_op(bio) != REQ_OP_ZONE_APPEND) | |
1270 | return; | |
1271 | ||
1272 | ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset); | |
1273 | if (WARN_ON(!ordered)) | |
1274 | return; | |
1275 | ||
1276 | ordered->physical = physical; | |
1277 | ordered->disk = bio->bi_disk; | |
1278 | ordered->partno = bio->bi_partno; | |
1279 | ||
1280 | btrfs_put_ordered_extent(ordered); | |
1281 | } | |
1282 | ||
1283 | void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered) | |
1284 | { | |
1285 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
1286 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1287 | struct extent_map_tree *em_tree; | |
1288 | struct extent_map *em; | |
1289 | struct btrfs_ordered_sum *sum; | |
1290 | struct block_device *bdev; | |
1291 | u64 orig_logical = ordered->disk_bytenr; | |
1292 | u64 *logical = NULL; | |
1293 | int nr, stripe_len; | |
1294 | ||
1295 | /* Zoned devices should not have partitions. So, we can assume it is 0 */ | |
1296 | ASSERT(ordered->partno == 0); | |
1297 | bdev = bdgrab(ordered->disk->part0); | |
1298 | if (WARN_ON(!bdev)) | |
1299 | return; | |
1300 | ||
1301 | if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, bdev, | |
1302 | ordered->physical, &logical, &nr, | |
1303 | &stripe_len))) | |
1304 | goto out; | |
1305 | ||
1306 | WARN_ON(nr != 1); | |
1307 | ||
1308 | if (orig_logical == *logical) | |
1309 | goto out; | |
1310 | ||
1311 | ordered->disk_bytenr = *logical; | |
1312 | ||
1313 | em_tree = &inode->extent_tree; | |
1314 | write_lock(&em_tree->lock); | |
1315 | em = search_extent_mapping(em_tree, ordered->file_offset, | |
1316 | ordered->num_bytes); | |
1317 | em->block_start = *logical; | |
1318 | free_extent_map(em); | |
1319 | write_unlock(&em_tree->lock); | |
1320 | ||
1321 | list_for_each_entry(sum, &ordered->list, list) { | |
1322 | if (*logical < orig_logical) | |
1323 | sum->bytenr -= orig_logical - *logical; | |
1324 | else | |
1325 | sum->bytenr += *logical - orig_logical; | |
1326 | } | |
1327 | ||
1328 | out: | |
1329 | kfree(logical); | |
1330 | bdput(bdev); | |
1331 | } | |
0bc09ca1 NA |
1332 | |
1333 | bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info, | |
1334 | struct extent_buffer *eb, | |
1335 | struct btrfs_block_group **cache_ret) | |
1336 | { | |
1337 | struct btrfs_block_group *cache; | |
1338 | bool ret = true; | |
1339 | ||
1340 | if (!btrfs_is_zoned(fs_info)) | |
1341 | return true; | |
1342 | ||
1343 | cache = *cache_ret; | |
1344 | ||
1345 | if (cache && (eb->start < cache->start || | |
1346 | cache->start + cache->length <= eb->start)) { | |
1347 | btrfs_put_block_group(cache); | |
1348 | cache = NULL; | |
1349 | *cache_ret = NULL; | |
1350 | } | |
1351 | ||
1352 | if (!cache) | |
1353 | cache = btrfs_lookup_block_group(fs_info, eb->start); | |
1354 | ||
1355 | if (cache) { | |
1356 | if (cache->meta_write_pointer != eb->start) { | |
1357 | btrfs_put_block_group(cache); | |
1358 | cache = NULL; | |
1359 | ret = false; | |
1360 | } else { | |
1361 | cache->meta_write_pointer = eb->start + eb->len; | |
1362 | } | |
1363 | ||
1364 | *cache_ret = cache; | |
1365 | } | |
1366 | ||
1367 | return ret; | |
1368 | } | |
1369 | ||
1370 | void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache, | |
1371 | struct extent_buffer *eb) | |
1372 | { | |
1373 | if (!btrfs_is_zoned(eb->fs_info) || !cache) | |
1374 | return; | |
1375 | ||
1376 | ASSERT(cache->meta_write_pointer == eb->start + eb->len); | |
1377 | cache->meta_write_pointer = eb->start; | |
1378 | } | |
de17addc NA |
1379 | |
1380 | int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length) | |
1381 | { | |
1382 | if (!btrfs_dev_is_sequential(device, physical)) | |
1383 | return -EOPNOTSUPP; | |
1384 | ||
1385 | return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT, | |
1386 | length >> SECTOR_SHIFT, GFP_NOFS, 0); | |
1387 | } |