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" |
5b316468 NA |
14 | |
15 | /* Maximum number of zones to report per blkdev_report_zones() call */ | |
16 | #define BTRFS_REPORT_NR_ZONES 4096 | |
08e11a3d NA |
17 | /* Invalid allocation pointer value for missing devices */ |
18 | #define WP_MISSING_DEV ((u64)-1) | |
19 | /* Pseudo write pointer value for conventional zone */ | |
20 | #define WP_CONVENTIONAL ((u64)-2) | |
5b316468 | 21 | |
12659251 NA |
22 | /* Number of superblock log zones */ |
23 | #define BTRFS_NR_SB_LOG_ZONES 2 | |
24 | ||
5b316468 NA |
25 | static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data) |
26 | { | |
27 | struct blk_zone *zones = data; | |
28 | ||
29 | memcpy(&zones[idx], zone, sizeof(*zone)); | |
30 | ||
31 | return 0; | |
32 | } | |
33 | ||
12659251 NA |
34 | static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones, |
35 | u64 *wp_ret) | |
36 | { | |
37 | bool empty[BTRFS_NR_SB_LOG_ZONES]; | |
38 | bool full[BTRFS_NR_SB_LOG_ZONES]; | |
39 | sector_t sector; | |
40 | ||
41 | ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL && | |
42 | zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL); | |
43 | ||
44 | empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY); | |
45 | empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY); | |
46 | full[0] = (zones[0].cond == BLK_ZONE_COND_FULL); | |
47 | full[1] = (zones[1].cond == BLK_ZONE_COND_FULL); | |
48 | ||
49 | /* | |
50 | * Possible states of log buffer zones | |
51 | * | |
52 | * Empty[0] In use[0] Full[0] | |
53 | * Empty[1] * x 0 | |
54 | * In use[1] 0 x 0 | |
55 | * Full[1] 1 1 C | |
56 | * | |
57 | * Log position: | |
58 | * *: Special case, no superblock is written | |
59 | * 0: Use write pointer of zones[0] | |
60 | * 1: Use write pointer of zones[1] | |
61 | * C: Compare super blcoks from zones[0] and zones[1], use the latest | |
62 | * one determined by generation | |
63 | * x: Invalid state | |
64 | */ | |
65 | ||
66 | if (empty[0] && empty[1]) { | |
67 | /* Special case to distinguish no superblock to read */ | |
68 | *wp_ret = zones[0].start << SECTOR_SHIFT; | |
69 | return -ENOENT; | |
70 | } else if (full[0] && full[1]) { | |
71 | /* Compare two super blocks */ | |
72 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
73 | struct page *page[BTRFS_NR_SB_LOG_ZONES]; | |
74 | struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES]; | |
75 | int i; | |
76 | ||
77 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { | |
78 | u64 bytenr; | |
79 | ||
80 | bytenr = ((zones[i].start + zones[i].len) | |
81 | << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE; | |
82 | ||
83 | page[i] = read_cache_page_gfp(mapping, | |
84 | bytenr >> PAGE_SHIFT, GFP_NOFS); | |
85 | if (IS_ERR(page[i])) { | |
86 | if (i == 1) | |
87 | btrfs_release_disk_super(super[0]); | |
88 | return PTR_ERR(page[i]); | |
89 | } | |
90 | super[i] = page_address(page[i]); | |
91 | } | |
92 | ||
93 | if (super[0]->generation > super[1]->generation) | |
94 | sector = zones[1].start; | |
95 | else | |
96 | sector = zones[0].start; | |
97 | ||
98 | for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) | |
99 | btrfs_release_disk_super(super[i]); | |
100 | } else if (!full[0] && (empty[1] || full[1])) { | |
101 | sector = zones[0].wp; | |
102 | } else if (full[0]) { | |
103 | sector = zones[1].wp; | |
104 | } else { | |
105 | return -EUCLEAN; | |
106 | } | |
107 | *wp_ret = sector << SECTOR_SHIFT; | |
108 | return 0; | |
109 | } | |
110 | ||
111 | /* | |
112 | * The following zones are reserved as the circular buffer on ZONED btrfs. | |
113 | * - The primary superblock: zones 0 and 1 | |
114 | * - The first copy: zones 16 and 17 | |
115 | * - The second copy: zones 1024 or zone at 256GB which is minimum, and | |
116 | * the following one | |
117 | */ | |
118 | static inline u32 sb_zone_number(int shift, int mirror) | |
119 | { | |
120 | ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX); | |
121 | ||
122 | switch (mirror) { | |
123 | case 0: return 0; | |
124 | case 1: return 16; | |
125 | case 2: return min_t(u64, btrfs_sb_offset(mirror) >> shift, 1024); | |
126 | } | |
127 | ||
128 | return 0; | |
129 | } | |
130 | ||
3c9daa09 JT |
131 | /* |
132 | * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block | |
133 | * device into static sized chunks and fake a conventional zone on each of | |
134 | * them. | |
135 | */ | |
136 | static int emulate_report_zones(struct btrfs_device *device, u64 pos, | |
137 | struct blk_zone *zones, unsigned int nr_zones) | |
138 | { | |
139 | const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT; | |
140 | sector_t bdev_size = bdev_nr_sectors(device->bdev); | |
141 | unsigned int i; | |
142 | ||
143 | pos >>= SECTOR_SHIFT; | |
144 | for (i = 0; i < nr_zones; i++) { | |
145 | zones[i].start = i * zone_sectors + pos; | |
146 | zones[i].len = zone_sectors; | |
147 | zones[i].capacity = zone_sectors; | |
148 | zones[i].wp = zones[i].start + zone_sectors; | |
149 | zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL; | |
150 | zones[i].cond = BLK_ZONE_COND_NOT_WP; | |
151 | ||
152 | if (zones[i].wp >= bdev_size) { | |
153 | i++; | |
154 | break; | |
155 | } | |
156 | } | |
157 | ||
158 | return i; | |
159 | } | |
160 | ||
5b316468 NA |
161 | static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos, |
162 | struct blk_zone *zones, unsigned int *nr_zones) | |
163 | { | |
164 | int ret; | |
165 | ||
166 | if (!*nr_zones) | |
167 | return 0; | |
168 | ||
3c9daa09 JT |
169 | if (!bdev_is_zoned(device->bdev)) { |
170 | ret = emulate_report_zones(device, pos, zones, *nr_zones); | |
171 | *nr_zones = ret; | |
172 | return 0; | |
173 | } | |
174 | ||
5b316468 NA |
175 | ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones, |
176 | copy_zone_info_cb, zones); | |
177 | if (ret < 0) { | |
178 | btrfs_err_in_rcu(device->fs_info, | |
179 | "zoned: failed to read zone %llu on %s (devid %llu)", | |
180 | pos, rcu_str_deref(device->name), | |
181 | device->devid); | |
182 | return ret; | |
183 | } | |
184 | *nr_zones = ret; | |
185 | if (!ret) | |
186 | return -EIO; | |
187 | ||
188 | return 0; | |
189 | } | |
190 | ||
3c9daa09 JT |
191 | /* The emulated zone size is determined from the size of device extent */ |
192 | static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info) | |
193 | { | |
194 | struct btrfs_path *path; | |
195 | struct btrfs_root *root = fs_info->dev_root; | |
196 | struct btrfs_key key; | |
197 | struct extent_buffer *leaf; | |
198 | struct btrfs_dev_extent *dext; | |
199 | int ret = 0; | |
200 | ||
201 | key.objectid = 1; | |
202 | key.type = BTRFS_DEV_EXTENT_KEY; | |
203 | key.offset = 0; | |
204 | ||
205 | path = btrfs_alloc_path(); | |
206 | if (!path) | |
207 | return -ENOMEM; | |
208 | ||
209 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
210 | if (ret < 0) | |
211 | goto out; | |
212 | ||
213 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
214 | ret = btrfs_next_item(root, path); | |
215 | if (ret < 0) | |
216 | goto out; | |
217 | /* No dev extents at all? Not good */ | |
218 | if (ret > 0) { | |
219 | ret = -EUCLEAN; | |
220 | goto out; | |
221 | } | |
222 | } | |
223 | ||
224 | leaf = path->nodes[0]; | |
225 | dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); | |
226 | fs_info->zone_size = btrfs_dev_extent_length(leaf, dext); | |
227 | ret = 0; | |
228 | ||
229 | out: | |
230 | btrfs_free_path(path); | |
231 | ||
232 | return ret; | |
233 | } | |
234 | ||
73651042 NA |
235 | int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info) |
236 | { | |
237 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
238 | struct btrfs_device *device; | |
239 | int ret = 0; | |
240 | ||
241 | /* fs_info->zone_size might not set yet. Use the incomapt flag here. */ | |
242 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
243 | return 0; | |
244 | ||
245 | mutex_lock(&fs_devices->device_list_mutex); | |
246 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
247 | /* We can skip reading of zone info for missing devices */ | |
248 | if (!device->bdev) | |
249 | continue; | |
250 | ||
251 | ret = btrfs_get_dev_zone_info(device); | |
252 | if (ret) | |
253 | break; | |
254 | } | |
255 | mutex_unlock(&fs_devices->device_list_mutex); | |
256 | ||
257 | return ret; | |
258 | } | |
259 | ||
5b316468 NA |
260 | int btrfs_get_dev_zone_info(struct btrfs_device *device) |
261 | { | |
3c9daa09 | 262 | struct btrfs_fs_info *fs_info = device->fs_info; |
5b316468 NA |
263 | struct btrfs_zoned_device_info *zone_info = NULL; |
264 | struct block_device *bdev = device->bdev; | |
862931c7 | 265 | struct request_queue *queue = bdev_get_queue(bdev); |
5b316468 NA |
266 | sector_t nr_sectors; |
267 | sector_t sector = 0; | |
268 | struct blk_zone *zones = NULL; | |
269 | unsigned int i, nreported = 0, nr_zones; | |
270 | unsigned int zone_sectors; | |
3c9daa09 | 271 | char *model, *emulated; |
5b316468 NA |
272 | int ret; |
273 | ||
3c9daa09 JT |
274 | /* |
275 | * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not | |
276 | * yet be set. | |
277 | */ | |
278 | if (!btrfs_fs_incompat(fs_info, ZONED)) | |
5b316468 NA |
279 | return 0; |
280 | ||
281 | if (device->zone_info) | |
282 | return 0; | |
283 | ||
284 | zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL); | |
285 | if (!zone_info) | |
286 | return -ENOMEM; | |
287 | ||
3c9daa09 JT |
288 | if (!bdev_is_zoned(bdev)) { |
289 | if (!fs_info->zone_size) { | |
290 | ret = calculate_emulated_zone_size(fs_info); | |
291 | if (ret) | |
292 | goto out; | |
293 | } | |
294 | ||
295 | ASSERT(fs_info->zone_size); | |
296 | zone_sectors = fs_info->zone_size >> SECTOR_SHIFT; | |
297 | } else { | |
298 | zone_sectors = bdev_zone_sectors(bdev); | |
299 | } | |
300 | ||
ac7ac461 | 301 | nr_sectors = bdev_nr_sectors(bdev); |
5b316468 NA |
302 | /* Check if it's power of 2 (see is_power_of_2) */ |
303 | ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0); | |
304 | zone_info->zone_size = zone_sectors << SECTOR_SHIFT; | |
305 | zone_info->zone_size_shift = ilog2(zone_info->zone_size); | |
862931c7 NA |
306 | zone_info->max_zone_append_size = |
307 | (u64)queue_max_zone_append_sectors(queue) << SECTOR_SHIFT; | |
5b316468 NA |
308 | zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors); |
309 | if (!IS_ALIGNED(nr_sectors, zone_sectors)) | |
310 | zone_info->nr_zones++; | |
311 | ||
312 | zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
313 | if (!zone_info->seq_zones) { | |
314 | ret = -ENOMEM; | |
315 | goto out; | |
316 | } | |
317 | ||
318 | zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); | |
319 | if (!zone_info->empty_zones) { | |
320 | ret = -ENOMEM; | |
321 | goto out; | |
322 | } | |
323 | ||
324 | zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL); | |
325 | if (!zones) { | |
326 | ret = -ENOMEM; | |
327 | goto out; | |
328 | } | |
329 | ||
330 | /* Get zones type */ | |
331 | while (sector < nr_sectors) { | |
332 | nr_zones = BTRFS_REPORT_NR_ZONES; | |
333 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones, | |
334 | &nr_zones); | |
335 | if (ret) | |
336 | goto out; | |
337 | ||
338 | for (i = 0; i < nr_zones; i++) { | |
339 | if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ) | |
340 | __set_bit(nreported, zone_info->seq_zones); | |
341 | if (zones[i].cond == BLK_ZONE_COND_EMPTY) | |
342 | __set_bit(nreported, zone_info->empty_zones); | |
343 | nreported++; | |
344 | } | |
345 | sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len; | |
346 | } | |
347 | ||
348 | if (nreported != zone_info->nr_zones) { | |
349 | btrfs_err_in_rcu(device->fs_info, | |
350 | "inconsistent number of zones on %s (%u/%u)", | |
351 | rcu_str_deref(device->name), nreported, | |
352 | zone_info->nr_zones); | |
353 | ret = -EIO; | |
354 | goto out; | |
355 | } | |
356 | ||
12659251 NA |
357 | /* Validate superblock log */ |
358 | nr_zones = BTRFS_NR_SB_LOG_ZONES; | |
359 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
360 | u32 sb_zone; | |
361 | u64 sb_wp; | |
362 | int sb_pos = BTRFS_NR_SB_LOG_ZONES * i; | |
363 | ||
364 | sb_zone = sb_zone_number(zone_info->zone_size_shift, i); | |
365 | if (sb_zone + 1 >= zone_info->nr_zones) | |
366 | continue; | |
367 | ||
368 | sector = sb_zone << (zone_info->zone_size_shift - SECTOR_SHIFT); | |
369 | ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, | |
370 | &zone_info->sb_zones[sb_pos], | |
371 | &nr_zones); | |
372 | if (ret) | |
373 | goto out; | |
374 | ||
375 | if (nr_zones != BTRFS_NR_SB_LOG_ZONES) { | |
376 | btrfs_err_in_rcu(device->fs_info, | |
377 | "zoned: failed to read super block log zone info at devid %llu zone %u", | |
378 | device->devid, sb_zone); | |
379 | ret = -EUCLEAN; | |
380 | goto out; | |
381 | } | |
382 | ||
383 | /* | |
384 | * If zones[0] is conventional, always use the beggining of the | |
385 | * zone to record superblock. No need to validate in that case. | |
386 | */ | |
387 | if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type == | |
388 | BLK_ZONE_TYPE_CONVENTIONAL) | |
389 | continue; | |
390 | ||
391 | ret = sb_write_pointer(device->bdev, | |
392 | &zone_info->sb_zones[sb_pos], &sb_wp); | |
393 | if (ret != -ENOENT && ret) { | |
394 | btrfs_err_in_rcu(device->fs_info, | |
395 | "zoned: super block log zone corrupted devid %llu zone %u", | |
396 | device->devid, sb_zone); | |
397 | ret = -EUCLEAN; | |
398 | goto out; | |
399 | } | |
400 | } | |
401 | ||
402 | ||
5b316468 NA |
403 | kfree(zones); |
404 | ||
405 | device->zone_info = zone_info; | |
406 | ||
3c9daa09 JT |
407 | switch (bdev_zoned_model(bdev)) { |
408 | case BLK_ZONED_HM: | |
409 | model = "host-managed zoned"; | |
410 | emulated = ""; | |
411 | break; | |
412 | case BLK_ZONED_HA: | |
413 | model = "host-aware zoned"; | |
414 | emulated = ""; | |
415 | break; | |
416 | case BLK_ZONED_NONE: | |
417 | model = "regular"; | |
418 | emulated = "emulated "; | |
419 | break; | |
420 | default: | |
421 | /* Just in case */ | |
422 | btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s", | |
423 | bdev_zoned_model(bdev), | |
424 | rcu_str_deref(device->name)); | |
425 | ret = -EOPNOTSUPP; | |
426 | goto out_free_zone_info; | |
427 | } | |
428 | ||
429 | btrfs_info_in_rcu(fs_info, | |
430 | "%s block device %s, %u %szones of %llu bytes", | |
431 | model, rcu_str_deref(device->name), zone_info->nr_zones, | |
432 | emulated, zone_info->zone_size); | |
5b316468 NA |
433 | |
434 | return 0; | |
435 | ||
436 | out: | |
437 | kfree(zones); | |
3c9daa09 | 438 | out_free_zone_info: |
5b316468 NA |
439 | bitmap_free(zone_info->empty_zones); |
440 | bitmap_free(zone_info->seq_zones); | |
441 | kfree(zone_info); | |
3c9daa09 | 442 | device->zone_info = NULL; |
5b316468 NA |
443 | |
444 | return ret; | |
445 | } | |
446 | ||
447 | void btrfs_destroy_dev_zone_info(struct btrfs_device *device) | |
448 | { | |
449 | struct btrfs_zoned_device_info *zone_info = device->zone_info; | |
450 | ||
451 | if (!zone_info) | |
452 | return; | |
453 | ||
454 | bitmap_free(zone_info->seq_zones); | |
455 | bitmap_free(zone_info->empty_zones); | |
456 | kfree(zone_info); | |
457 | device->zone_info = NULL; | |
458 | } | |
459 | ||
460 | int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos, | |
461 | struct blk_zone *zone) | |
462 | { | |
463 | unsigned int nr_zones = 1; | |
464 | int ret; | |
465 | ||
466 | ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones); | |
467 | if (ret != 0 || !nr_zones) | |
468 | return ret ? ret : -EIO; | |
469 | ||
470 | return 0; | |
471 | } | |
b70f5097 NA |
472 | |
473 | int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info) | |
474 | { | |
475 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
476 | struct btrfs_device *device; | |
477 | u64 zoned_devices = 0; | |
478 | u64 nr_devices = 0; | |
479 | u64 zone_size = 0; | |
862931c7 | 480 | u64 max_zone_append_size = 0; |
3c9daa09 | 481 | const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED); |
b70f5097 NA |
482 | int ret = 0; |
483 | ||
484 | /* Count zoned devices */ | |
485 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
486 | enum blk_zoned_model model; | |
487 | ||
488 | if (!device->bdev) | |
489 | continue; | |
490 | ||
491 | model = bdev_zoned_model(device->bdev); | |
3c9daa09 JT |
492 | /* |
493 | * A Host-Managed zoned device must be used as a zoned device. | |
494 | * A Host-Aware zoned device and a non-zoned devices can be | |
495 | * treated as a zoned device, if ZONED flag is enabled in the | |
496 | * superblock. | |
497 | */ | |
b70f5097 | 498 | if (model == BLK_ZONED_HM || |
3c9daa09 JT |
499 | (model == BLK_ZONED_HA && incompat_zoned) || |
500 | (model == BLK_ZONED_NONE && incompat_zoned)) { | |
501 | struct btrfs_zoned_device_info *zone_info = | |
502 | device->zone_info; | |
862931c7 NA |
503 | |
504 | zone_info = device->zone_info; | |
b70f5097 NA |
505 | zoned_devices++; |
506 | if (!zone_size) { | |
862931c7 NA |
507 | zone_size = zone_info->zone_size; |
508 | } else if (zone_info->zone_size != zone_size) { | |
b70f5097 NA |
509 | btrfs_err(fs_info, |
510 | "zoned: unequal block device zone sizes: have %llu found %llu", | |
511 | device->zone_info->zone_size, | |
512 | zone_size); | |
513 | ret = -EINVAL; | |
514 | goto out; | |
515 | } | |
862931c7 NA |
516 | if (!max_zone_append_size || |
517 | (zone_info->max_zone_append_size && | |
518 | zone_info->max_zone_append_size < max_zone_append_size)) | |
519 | max_zone_append_size = | |
520 | zone_info->max_zone_append_size; | |
b70f5097 NA |
521 | } |
522 | nr_devices++; | |
523 | } | |
524 | ||
525 | if (!zoned_devices && !incompat_zoned) | |
526 | goto out; | |
527 | ||
528 | if (!zoned_devices && incompat_zoned) { | |
529 | /* No zoned block device found on ZONED filesystem */ | |
530 | btrfs_err(fs_info, | |
531 | "zoned: no zoned devices found on a zoned filesystem"); | |
532 | ret = -EINVAL; | |
533 | goto out; | |
534 | } | |
535 | ||
536 | if (zoned_devices && !incompat_zoned) { | |
537 | btrfs_err(fs_info, | |
538 | "zoned: mode not enabled but zoned device found"); | |
539 | ret = -EINVAL; | |
540 | goto out; | |
541 | } | |
542 | ||
543 | if (zoned_devices != nr_devices) { | |
544 | btrfs_err(fs_info, | |
545 | "zoned: cannot mix zoned and regular devices"); | |
546 | ret = -EINVAL; | |
547 | goto out; | |
548 | } | |
549 | ||
550 | /* | |
551 | * stripe_size is always aligned to BTRFS_STRIPE_LEN in | |
552 | * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size, | |
553 | * check the alignment here. | |
554 | */ | |
555 | if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) { | |
556 | btrfs_err(fs_info, | |
557 | "zoned: zone size %llu not aligned to stripe %u", | |
558 | zone_size, BTRFS_STRIPE_LEN); | |
559 | ret = -EINVAL; | |
560 | goto out; | |
561 | } | |
562 | ||
a589dde0 NA |
563 | if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
564 | btrfs_err(fs_info, "zoned: mixed block groups not supported"); | |
565 | ret = -EINVAL; | |
566 | goto out; | |
567 | } | |
568 | ||
b70f5097 | 569 | fs_info->zone_size = zone_size; |
862931c7 | 570 | fs_info->max_zone_append_size = max_zone_append_size; |
1cd6121f | 571 | fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED; |
b70f5097 | 572 | |
b53429ba JT |
573 | /* |
574 | * Check mount options here, because we might change fs_info->zoned | |
575 | * from fs_info->zone_size. | |
576 | */ | |
577 | ret = btrfs_check_mountopts_zoned(fs_info); | |
578 | if (ret) | |
579 | goto out; | |
580 | ||
b70f5097 NA |
581 | btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size); |
582 | out: | |
583 | return ret; | |
584 | } | |
5d1ab66c NA |
585 | |
586 | int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info) | |
587 | { | |
588 | if (!btrfs_is_zoned(info)) | |
589 | return 0; | |
590 | ||
591 | /* | |
592 | * Space cache writing is not COWed. Disable that to avoid write errors | |
593 | * in sequential zones. | |
594 | */ | |
595 | if (btrfs_test_opt(info, SPACE_CACHE)) { | |
596 | btrfs_err(info, "zoned: space cache v1 is not supported"); | |
597 | return -EINVAL; | |
598 | } | |
599 | ||
d206e9c9 NA |
600 | if (btrfs_test_opt(info, NODATACOW)) { |
601 | btrfs_err(info, "zoned: NODATACOW not supported"); | |
602 | return -EINVAL; | |
603 | } | |
604 | ||
5d1ab66c NA |
605 | return 0; |
606 | } | |
12659251 NA |
607 | |
608 | static int sb_log_location(struct block_device *bdev, struct blk_zone *zones, | |
609 | int rw, u64 *bytenr_ret) | |
610 | { | |
611 | u64 wp; | |
612 | int ret; | |
613 | ||
614 | if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) { | |
615 | *bytenr_ret = zones[0].start << SECTOR_SHIFT; | |
616 | return 0; | |
617 | } | |
618 | ||
619 | ret = sb_write_pointer(bdev, zones, &wp); | |
620 | if (ret != -ENOENT && ret < 0) | |
621 | return ret; | |
622 | ||
623 | if (rw == WRITE) { | |
624 | struct blk_zone *reset = NULL; | |
625 | ||
626 | if (wp == zones[0].start << SECTOR_SHIFT) | |
627 | reset = &zones[0]; | |
628 | else if (wp == zones[1].start << SECTOR_SHIFT) | |
629 | reset = &zones[1]; | |
630 | ||
631 | if (reset && reset->cond != BLK_ZONE_COND_EMPTY) { | |
632 | ASSERT(reset->cond == BLK_ZONE_COND_FULL); | |
633 | ||
634 | ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
635 | reset->start, reset->len, | |
636 | GFP_NOFS); | |
637 | if (ret) | |
638 | return ret; | |
639 | ||
640 | reset->cond = BLK_ZONE_COND_EMPTY; | |
641 | reset->wp = reset->start; | |
642 | } | |
643 | } else if (ret != -ENOENT) { | |
644 | /* For READ, we want the precious one */ | |
645 | if (wp == zones[0].start << SECTOR_SHIFT) | |
646 | wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT; | |
647 | wp -= BTRFS_SUPER_INFO_SIZE; | |
648 | } | |
649 | ||
650 | *bytenr_ret = wp; | |
651 | return 0; | |
652 | ||
653 | } | |
654 | ||
655 | int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw, | |
656 | u64 *bytenr_ret) | |
657 | { | |
658 | struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES]; | |
659 | unsigned int zone_sectors; | |
660 | u32 sb_zone; | |
661 | int ret; | |
12659251 NA |
662 | u8 zone_sectors_shift; |
663 | sector_t nr_sectors; | |
664 | u32 nr_zones; | |
665 | ||
666 | if (!bdev_is_zoned(bdev)) { | |
667 | *bytenr_ret = btrfs_sb_offset(mirror); | |
668 | return 0; | |
669 | } | |
670 | ||
671 | ASSERT(rw == READ || rw == WRITE); | |
672 | ||
673 | zone_sectors = bdev_zone_sectors(bdev); | |
674 | if (!is_power_of_2(zone_sectors)) | |
675 | return -EINVAL; | |
12659251 | 676 | zone_sectors_shift = ilog2(zone_sectors); |
ac7ac461 | 677 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
678 | nr_zones = nr_sectors >> zone_sectors_shift; |
679 | ||
680 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
681 | if (sb_zone + 1 >= nr_zones) | |
682 | return -ENOENT; | |
683 | ||
684 | ret = blkdev_report_zones(bdev, sb_zone << zone_sectors_shift, | |
685 | BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb, | |
686 | zones); | |
687 | if (ret < 0) | |
688 | return ret; | |
689 | if (ret != BTRFS_NR_SB_LOG_ZONES) | |
690 | return -EIO; | |
691 | ||
692 | return sb_log_location(bdev, zones, rw, bytenr_ret); | |
693 | } | |
694 | ||
695 | int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw, | |
696 | u64 *bytenr_ret) | |
697 | { | |
698 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
699 | u32 zone_num; | |
700 | ||
d6639b35 NA |
701 | /* |
702 | * For a zoned filesystem on a non-zoned block device, use the same | |
703 | * super block locations as regular filesystem. Doing so, the super | |
704 | * block can always be retrieved and the zoned flag of the volume | |
705 | * detected from the super block information. | |
706 | */ | |
707 | if (!bdev_is_zoned(device->bdev)) { | |
12659251 NA |
708 | *bytenr_ret = btrfs_sb_offset(mirror); |
709 | return 0; | |
710 | } | |
711 | ||
712 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
713 | if (zone_num + 1 >= zinfo->nr_zones) | |
714 | return -ENOENT; | |
715 | ||
716 | return sb_log_location(device->bdev, | |
717 | &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror], | |
718 | rw, bytenr_ret); | |
719 | } | |
720 | ||
721 | static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo, | |
722 | int mirror) | |
723 | { | |
724 | u32 zone_num; | |
725 | ||
726 | if (!zinfo) | |
727 | return false; | |
728 | ||
729 | zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); | |
730 | if (zone_num + 1 >= zinfo->nr_zones) | |
731 | return false; | |
732 | ||
733 | if (!test_bit(zone_num, zinfo->seq_zones)) | |
734 | return false; | |
735 | ||
736 | return true; | |
737 | } | |
738 | ||
739 | void btrfs_advance_sb_log(struct btrfs_device *device, int mirror) | |
740 | { | |
741 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
742 | struct blk_zone *zone; | |
743 | ||
744 | if (!is_sb_log_zone(zinfo, mirror)) | |
745 | return; | |
746 | ||
747 | zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror]; | |
748 | if (zone->cond != BLK_ZONE_COND_FULL) { | |
749 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
750 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
751 | ||
752 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
753 | ||
754 | if (zone->wp == zone->start + zone->len) | |
755 | zone->cond = BLK_ZONE_COND_FULL; | |
756 | ||
757 | return; | |
758 | } | |
759 | ||
760 | zone++; | |
761 | ASSERT(zone->cond != BLK_ZONE_COND_FULL); | |
762 | if (zone->cond == BLK_ZONE_COND_EMPTY) | |
763 | zone->cond = BLK_ZONE_COND_IMP_OPEN; | |
764 | ||
765 | zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT); | |
766 | ||
767 | if (zone->wp == zone->start + zone->len) | |
768 | zone->cond = BLK_ZONE_COND_FULL; | |
769 | } | |
770 | ||
771 | int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror) | |
772 | { | |
773 | sector_t zone_sectors; | |
774 | sector_t nr_sectors; | |
775 | u8 zone_sectors_shift; | |
776 | u32 sb_zone; | |
777 | u32 nr_zones; | |
778 | ||
779 | zone_sectors = bdev_zone_sectors(bdev); | |
780 | zone_sectors_shift = ilog2(zone_sectors); | |
ac7ac461 | 781 | nr_sectors = bdev_nr_sectors(bdev); |
12659251 NA |
782 | nr_zones = nr_sectors >> zone_sectors_shift; |
783 | ||
784 | sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); | |
785 | if (sb_zone + 1 >= nr_zones) | |
786 | return -ENOENT; | |
787 | ||
788 | return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, | |
789 | sb_zone << zone_sectors_shift, | |
790 | zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS); | |
791 | } | |
1cd6121f NA |
792 | |
793 | /** | |
794 | * btrfs_find_allocatable_zones - find allocatable zones within a given region | |
795 | * | |
796 | * @device: the device to allocate a region on | |
797 | * @hole_start: the position of the hole to allocate the region | |
798 | * @num_bytes: size of wanted region | |
799 | * @hole_end: the end of the hole | |
800 | * @return: position of allocatable zones | |
801 | * | |
802 | * Allocatable region should not contain any superblock locations. | |
803 | */ | |
804 | u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start, | |
805 | u64 hole_end, u64 num_bytes) | |
806 | { | |
807 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
808 | const u8 shift = zinfo->zone_size_shift; | |
809 | u64 nzones = num_bytes >> shift; | |
810 | u64 pos = hole_start; | |
811 | u64 begin, end; | |
812 | bool have_sb; | |
813 | int i; | |
814 | ||
815 | ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size)); | |
816 | ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size)); | |
817 | ||
818 | while (pos < hole_end) { | |
819 | begin = pos >> shift; | |
820 | end = begin + nzones; | |
821 | ||
822 | if (end > zinfo->nr_zones) | |
823 | return hole_end; | |
824 | ||
825 | /* Check if zones in the region are all empty */ | |
826 | if (btrfs_dev_is_sequential(device, pos) && | |
827 | find_next_zero_bit(zinfo->empty_zones, end, begin) != end) { | |
828 | pos += zinfo->zone_size; | |
829 | continue; | |
830 | } | |
831 | ||
832 | have_sb = false; | |
833 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
834 | u32 sb_zone; | |
835 | u64 sb_pos; | |
836 | ||
837 | sb_zone = sb_zone_number(shift, i); | |
838 | if (!(end <= sb_zone || | |
839 | sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) { | |
840 | have_sb = true; | |
841 | pos = ((u64)sb_zone + BTRFS_NR_SB_LOG_ZONES) << shift; | |
842 | break; | |
843 | } | |
844 | ||
845 | /* We also need to exclude regular superblock positions */ | |
846 | sb_pos = btrfs_sb_offset(i); | |
847 | if (!(pos + num_bytes <= sb_pos || | |
848 | sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) { | |
849 | have_sb = true; | |
850 | pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE, | |
851 | zinfo->zone_size); | |
852 | break; | |
853 | } | |
854 | } | |
855 | if (!have_sb) | |
856 | break; | |
857 | } | |
858 | ||
859 | return pos; | |
860 | } | |
861 | ||
862 | int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical, | |
863 | u64 length, u64 *bytes) | |
864 | { | |
865 | int ret; | |
866 | ||
867 | *bytes = 0; | |
868 | ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET, | |
869 | physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT, | |
870 | GFP_NOFS); | |
871 | if (ret) | |
872 | return ret; | |
873 | ||
874 | *bytes = length; | |
875 | while (length) { | |
876 | btrfs_dev_set_zone_empty(device, physical); | |
877 | physical += device->zone_info->zone_size; | |
878 | length -= device->zone_info->zone_size; | |
879 | } | |
880 | ||
881 | return 0; | |
882 | } | |
883 | ||
884 | int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size) | |
885 | { | |
886 | struct btrfs_zoned_device_info *zinfo = device->zone_info; | |
887 | const u8 shift = zinfo->zone_size_shift; | |
888 | unsigned long begin = start >> shift; | |
889 | unsigned long end = (start + size) >> shift; | |
890 | u64 pos; | |
891 | int ret; | |
892 | ||
893 | ASSERT(IS_ALIGNED(start, zinfo->zone_size)); | |
894 | ASSERT(IS_ALIGNED(size, zinfo->zone_size)); | |
895 | ||
896 | if (end > zinfo->nr_zones) | |
897 | return -ERANGE; | |
898 | ||
899 | /* All the zones are conventional */ | |
900 | if (find_next_bit(zinfo->seq_zones, begin, end) == end) | |
901 | return 0; | |
902 | ||
903 | /* All the zones are sequential and empty */ | |
904 | if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end && | |
905 | find_next_zero_bit(zinfo->empty_zones, begin, end) == end) | |
906 | return 0; | |
907 | ||
908 | for (pos = start; pos < start + size; pos += zinfo->zone_size) { | |
909 | u64 reset_bytes; | |
910 | ||
911 | if (!btrfs_dev_is_sequential(device, pos) || | |
912 | btrfs_dev_is_empty_zone(device, pos)) | |
913 | continue; | |
914 | ||
915 | /* Free regions should be empty */ | |
916 | btrfs_warn_in_rcu( | |
917 | device->fs_info, | |
918 | "zoned: resetting device %s (devid %llu) zone %llu for allocation", | |
919 | rcu_str_deref(device->name), device->devid, pos >> shift); | |
920 | WARN_ON_ONCE(1); | |
921 | ||
922 | ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size, | |
923 | &reset_bytes); | |
924 | if (ret) | |
925 | return ret; | |
926 | } | |
927 | ||
928 | return 0; | |
929 | } | |
08e11a3d | 930 | |
a94794d5 NA |
931 | /* |
932 | * Calculate an allocation pointer from the extent allocation information | |
933 | * for a block group consist of conventional zones. It is pointed to the | |
934 | * end of the highest addressed extent in the block group as an allocation | |
935 | * offset. | |
936 | */ | |
937 | static int calculate_alloc_pointer(struct btrfs_block_group *cache, | |
938 | u64 *offset_ret) | |
939 | { | |
940 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
941 | struct btrfs_root *root = fs_info->extent_root; | |
942 | struct btrfs_path *path; | |
943 | struct btrfs_key key; | |
944 | struct btrfs_key found_key; | |
945 | int ret; | |
946 | u64 length; | |
947 | ||
948 | path = btrfs_alloc_path(); | |
949 | if (!path) | |
950 | return -ENOMEM; | |
951 | ||
952 | key.objectid = cache->start + cache->length; | |
953 | key.type = 0; | |
954 | key.offset = 0; | |
955 | ||
956 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
957 | /* We should not find the exact match */ | |
958 | if (!ret) | |
959 | ret = -EUCLEAN; | |
960 | if (ret < 0) | |
961 | goto out; | |
962 | ||
963 | ret = btrfs_previous_extent_item(root, path, cache->start); | |
964 | if (ret) { | |
965 | if (ret == 1) { | |
966 | ret = 0; | |
967 | *offset_ret = 0; | |
968 | } | |
969 | goto out; | |
970 | } | |
971 | ||
972 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); | |
973 | ||
974 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY) | |
975 | length = found_key.offset; | |
976 | else | |
977 | length = fs_info->nodesize; | |
978 | ||
979 | if (!(found_key.objectid >= cache->start && | |
980 | found_key.objectid + length <= cache->start + cache->length)) { | |
981 | ret = -EUCLEAN; | |
982 | goto out; | |
983 | } | |
984 | *offset_ret = found_key.objectid + length - cache->start; | |
985 | ret = 0; | |
986 | ||
987 | out: | |
988 | btrfs_free_path(path); | |
989 | return ret; | |
990 | } | |
991 | ||
992 | int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) | |
08e11a3d NA |
993 | { |
994 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
995 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
996 | struct extent_map *em; | |
997 | struct map_lookup *map; | |
998 | struct btrfs_device *device; | |
999 | u64 logical = cache->start; | |
1000 | u64 length = cache->length; | |
1001 | u64 physical = 0; | |
1002 | int ret; | |
1003 | int i; | |
1004 | unsigned int nofs_flag; | |
1005 | u64 *alloc_offsets = NULL; | |
a94794d5 | 1006 | u64 last_alloc = 0; |
08e11a3d NA |
1007 | u32 num_sequential = 0, num_conventional = 0; |
1008 | ||
1009 | if (!btrfs_is_zoned(fs_info)) | |
1010 | return 0; | |
1011 | ||
1012 | /* Sanity check */ | |
1013 | if (!IS_ALIGNED(length, fs_info->zone_size)) { | |
1014 | btrfs_err(fs_info, | |
1015 | "zoned: block group %llu len %llu unaligned to zone size %llu", | |
1016 | logical, length, fs_info->zone_size); | |
1017 | return -EIO; | |
1018 | } | |
1019 | ||
1020 | /* Get the chunk mapping */ | |
1021 | read_lock(&em_tree->lock); | |
1022 | em = lookup_extent_mapping(em_tree, logical, length); | |
1023 | read_unlock(&em_tree->lock); | |
1024 | ||
1025 | if (!em) | |
1026 | return -EINVAL; | |
1027 | ||
1028 | map = em->map_lookup; | |
1029 | ||
1030 | alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS); | |
1031 | if (!alloc_offsets) { | |
1032 | free_extent_map(em); | |
1033 | return -ENOMEM; | |
1034 | } | |
1035 | ||
1036 | for (i = 0; i < map->num_stripes; i++) { | |
1037 | bool is_sequential; | |
1038 | struct blk_zone zone; | |
1039 | ||
1040 | device = map->stripes[i].dev; | |
1041 | physical = map->stripes[i].physical; | |
1042 | ||
1043 | if (device->bdev == NULL) { | |
1044 | alloc_offsets[i] = WP_MISSING_DEV; | |
1045 | continue; | |
1046 | } | |
1047 | ||
1048 | is_sequential = btrfs_dev_is_sequential(device, physical); | |
1049 | if (is_sequential) | |
1050 | num_sequential++; | |
1051 | else | |
1052 | num_conventional++; | |
1053 | ||
1054 | if (!is_sequential) { | |
1055 | alloc_offsets[i] = WP_CONVENTIONAL; | |
1056 | continue; | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * This zone will be used for allocation, so mark this zone | |
1061 | * non-empty. | |
1062 | */ | |
1063 | btrfs_dev_clear_zone_empty(device, physical); | |
1064 | ||
1065 | /* | |
1066 | * The group is mapped to a sequential zone. Get the zone write | |
1067 | * pointer to determine the allocation offset within the zone. | |
1068 | */ | |
1069 | WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size)); | |
1070 | nofs_flag = memalloc_nofs_save(); | |
1071 | ret = btrfs_get_dev_zone(device, physical, &zone); | |
1072 | memalloc_nofs_restore(nofs_flag); | |
1073 | if (ret == -EIO || ret == -EOPNOTSUPP) { | |
1074 | ret = 0; | |
1075 | alloc_offsets[i] = WP_MISSING_DEV; | |
1076 | continue; | |
1077 | } else if (ret) { | |
1078 | goto out; | |
1079 | } | |
1080 | ||
1081 | switch (zone.cond) { | |
1082 | case BLK_ZONE_COND_OFFLINE: | |
1083 | case BLK_ZONE_COND_READONLY: | |
1084 | btrfs_err(fs_info, | |
1085 | "zoned: offline/readonly zone %llu on device %s (devid %llu)", | |
1086 | physical >> device->zone_info->zone_size_shift, | |
1087 | rcu_str_deref(device->name), device->devid); | |
1088 | alloc_offsets[i] = WP_MISSING_DEV; | |
1089 | break; | |
1090 | case BLK_ZONE_COND_EMPTY: | |
1091 | alloc_offsets[i] = 0; | |
1092 | break; | |
1093 | case BLK_ZONE_COND_FULL: | |
1094 | alloc_offsets[i] = fs_info->zone_size; | |
1095 | break; | |
1096 | default: | |
1097 | /* Partially used zone */ | |
1098 | alloc_offsets[i] = | |
1099 | ((zone.wp - zone.start) << SECTOR_SHIFT); | |
1100 | break; | |
1101 | } | |
1102 | } | |
1103 | ||
08f45559 JT |
1104 | if (num_sequential > 0) |
1105 | cache->seq_zone = true; | |
1106 | ||
08e11a3d NA |
1107 | if (num_conventional > 0) { |
1108 | /* | |
a94794d5 NA |
1109 | * Avoid calling calculate_alloc_pointer() for new BG. It |
1110 | * is no use for new BG. It must be always 0. | |
1111 | * | |
1112 | * Also, we have a lock chain of extent buffer lock -> | |
1113 | * chunk mutex. For new BG, this function is called from | |
1114 | * btrfs_make_block_group() which is already taking the | |
1115 | * chunk mutex. Thus, we cannot call | |
1116 | * calculate_alloc_pointer() which takes extent buffer | |
1117 | * locks to avoid deadlock. | |
08e11a3d | 1118 | */ |
a94794d5 NA |
1119 | if (new) { |
1120 | cache->alloc_offset = 0; | |
1121 | goto out; | |
1122 | } | |
1123 | ret = calculate_alloc_pointer(cache, &last_alloc); | |
1124 | if (ret || map->num_stripes == num_conventional) { | |
1125 | if (!ret) | |
1126 | cache->alloc_offset = last_alloc; | |
1127 | else | |
1128 | btrfs_err(fs_info, | |
1129 | "zoned: failed to determine allocation offset of bg %llu", | |
1130 | cache->start); | |
1131 | goto out; | |
1132 | } | |
08e11a3d NA |
1133 | } |
1134 | ||
1135 | switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
1136 | case 0: /* single */ | |
1137 | cache->alloc_offset = alloc_offsets[0]; | |
1138 | break; | |
1139 | case BTRFS_BLOCK_GROUP_DUP: | |
1140 | case BTRFS_BLOCK_GROUP_RAID1: | |
1141 | case BTRFS_BLOCK_GROUP_RAID0: | |
1142 | case BTRFS_BLOCK_GROUP_RAID10: | |
1143 | case BTRFS_BLOCK_GROUP_RAID5: | |
1144 | case BTRFS_BLOCK_GROUP_RAID6: | |
1145 | /* non-single profiles are not supported yet */ | |
1146 | default: | |
1147 | btrfs_err(fs_info, "zoned: profile %s not yet supported", | |
1148 | btrfs_bg_type_to_raid_name(map->type)); | |
1149 | ret = -EINVAL; | |
1150 | goto out; | |
1151 | } | |
1152 | ||
1153 | out: | |
a94794d5 NA |
1154 | /* An extent is allocated after the write pointer */ |
1155 | if (!ret && num_conventional && last_alloc > cache->alloc_offset) { | |
1156 | btrfs_err(fs_info, | |
1157 | "zoned: got wrong write pointer in BG %llu: %llu > %llu", | |
1158 | logical, last_alloc, cache->alloc_offset); | |
1159 | ret = -EIO; | |
1160 | } | |
1161 | ||
0bc09ca1 NA |
1162 | if (!ret) |
1163 | cache->meta_write_pointer = cache->alloc_offset + cache->start; | |
1164 | ||
08e11a3d NA |
1165 | kfree(alloc_offsets); |
1166 | free_extent_map(em); | |
1167 | ||
1168 | return ret; | |
1169 | } | |
169e0da9 NA |
1170 | |
1171 | void btrfs_calc_zone_unusable(struct btrfs_block_group *cache) | |
1172 | { | |
1173 | u64 unusable, free; | |
1174 | ||
1175 | if (!btrfs_is_zoned(cache->fs_info)) | |
1176 | return; | |
1177 | ||
1178 | WARN_ON(cache->bytes_super != 0); | |
1179 | unusable = cache->alloc_offset - cache->used; | |
1180 | free = cache->length - cache->alloc_offset; | |
1181 | ||
1182 | /* We only need ->free_space in ALLOC_SEQ block groups */ | |
1183 | cache->last_byte_to_unpin = (u64)-1; | |
1184 | cache->cached = BTRFS_CACHE_FINISHED; | |
1185 | cache->free_space_ctl->free_space = free; | |
1186 | cache->zone_unusable = unusable; | |
1187 | ||
1188 | /* Should not have any excluded extents. Just in case, though */ | |
1189 | btrfs_free_excluded_extents(cache); | |
1190 | } | |
d3575156 NA |
1191 | |
1192 | void btrfs_redirty_list_add(struct btrfs_transaction *trans, | |
1193 | struct extent_buffer *eb) | |
1194 | { | |
1195 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
1196 | ||
1197 | if (!btrfs_is_zoned(fs_info) || | |
1198 | btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) || | |
1199 | !list_empty(&eb->release_list)) | |
1200 | return; | |
1201 | ||
1202 | set_extent_buffer_dirty(eb); | |
1203 | set_extent_bits_nowait(&trans->dirty_pages, eb->start, | |
1204 | eb->start + eb->len - 1, EXTENT_DIRTY); | |
1205 | memzero_extent_buffer(eb, 0, eb->len); | |
1206 | set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags); | |
1207 | ||
1208 | spin_lock(&trans->releasing_ebs_lock); | |
1209 | list_add_tail(&eb->release_list, &trans->releasing_ebs); | |
1210 | spin_unlock(&trans->releasing_ebs_lock); | |
1211 | atomic_inc(&eb->refs); | |
1212 | } | |
1213 | ||
1214 | void btrfs_free_redirty_list(struct btrfs_transaction *trans) | |
1215 | { | |
1216 | spin_lock(&trans->releasing_ebs_lock); | |
1217 | while (!list_empty(&trans->releasing_ebs)) { | |
1218 | struct extent_buffer *eb; | |
1219 | ||
1220 | eb = list_first_entry(&trans->releasing_ebs, | |
1221 | struct extent_buffer, release_list); | |
1222 | list_del_init(&eb->release_list); | |
1223 | free_extent_buffer(eb); | |
1224 | } | |
1225 | spin_unlock(&trans->releasing_ebs_lock); | |
1226 | } | |
08f45559 JT |
1227 | |
1228 | bool btrfs_use_zone_append(struct btrfs_inode *inode, struct extent_map *em) | |
1229 | { | |
1230 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1231 | struct btrfs_block_group *cache; | |
1232 | bool ret = false; | |
1233 | ||
1234 | if (!btrfs_is_zoned(fs_info)) | |
1235 | return false; | |
1236 | ||
1237 | if (!fs_info->max_zone_append_size) | |
1238 | return false; | |
1239 | ||
1240 | if (!is_data_inode(&inode->vfs_inode)) | |
1241 | return false; | |
1242 | ||
1243 | cache = btrfs_lookup_block_group(fs_info, em->block_start); | |
1244 | ASSERT(cache); | |
1245 | if (!cache) | |
1246 | return false; | |
1247 | ||
1248 | ret = cache->seq_zone; | |
1249 | btrfs_put_block_group(cache); | |
1250 | ||
1251 | return ret; | |
1252 | } | |
d8e3fb10 NA |
1253 | |
1254 | void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset, | |
1255 | struct bio *bio) | |
1256 | { | |
1257 | struct btrfs_ordered_extent *ordered; | |
1258 | const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; | |
1259 | ||
1260 | if (bio_op(bio) != REQ_OP_ZONE_APPEND) | |
1261 | return; | |
1262 | ||
1263 | ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset); | |
1264 | if (WARN_ON(!ordered)) | |
1265 | return; | |
1266 | ||
1267 | ordered->physical = physical; | |
1268 | ordered->disk = bio->bi_disk; | |
1269 | ordered->partno = bio->bi_partno; | |
1270 | ||
1271 | btrfs_put_ordered_extent(ordered); | |
1272 | } | |
1273 | ||
1274 | void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered) | |
1275 | { | |
1276 | struct btrfs_inode *inode = BTRFS_I(ordered->inode); | |
1277 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
1278 | struct extent_map_tree *em_tree; | |
1279 | struct extent_map *em; | |
1280 | struct btrfs_ordered_sum *sum; | |
1281 | struct block_device *bdev; | |
1282 | u64 orig_logical = ordered->disk_bytenr; | |
1283 | u64 *logical = NULL; | |
1284 | int nr, stripe_len; | |
1285 | ||
1286 | /* Zoned devices should not have partitions. So, we can assume it is 0 */ | |
1287 | ASSERT(ordered->partno == 0); | |
1288 | bdev = bdgrab(ordered->disk->part0); | |
1289 | if (WARN_ON(!bdev)) | |
1290 | return; | |
1291 | ||
1292 | if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, bdev, | |
1293 | ordered->physical, &logical, &nr, | |
1294 | &stripe_len))) | |
1295 | goto out; | |
1296 | ||
1297 | WARN_ON(nr != 1); | |
1298 | ||
1299 | if (orig_logical == *logical) | |
1300 | goto out; | |
1301 | ||
1302 | ordered->disk_bytenr = *logical; | |
1303 | ||
1304 | em_tree = &inode->extent_tree; | |
1305 | write_lock(&em_tree->lock); | |
1306 | em = search_extent_mapping(em_tree, ordered->file_offset, | |
1307 | ordered->num_bytes); | |
1308 | em->block_start = *logical; | |
1309 | free_extent_map(em); | |
1310 | write_unlock(&em_tree->lock); | |
1311 | ||
1312 | list_for_each_entry(sum, &ordered->list, list) { | |
1313 | if (*logical < orig_logical) | |
1314 | sum->bytenr -= orig_logical - *logical; | |
1315 | else | |
1316 | sum->bytenr += *logical - orig_logical; | |
1317 | } | |
1318 | ||
1319 | out: | |
1320 | kfree(logical); | |
1321 | bdput(bdev); | |
1322 | } | |
0bc09ca1 NA |
1323 | |
1324 | bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info, | |
1325 | struct extent_buffer *eb, | |
1326 | struct btrfs_block_group **cache_ret) | |
1327 | { | |
1328 | struct btrfs_block_group *cache; | |
1329 | bool ret = true; | |
1330 | ||
1331 | if (!btrfs_is_zoned(fs_info)) | |
1332 | return true; | |
1333 | ||
1334 | cache = *cache_ret; | |
1335 | ||
1336 | if (cache && (eb->start < cache->start || | |
1337 | cache->start + cache->length <= eb->start)) { | |
1338 | btrfs_put_block_group(cache); | |
1339 | cache = NULL; | |
1340 | *cache_ret = NULL; | |
1341 | } | |
1342 | ||
1343 | if (!cache) | |
1344 | cache = btrfs_lookup_block_group(fs_info, eb->start); | |
1345 | ||
1346 | if (cache) { | |
1347 | if (cache->meta_write_pointer != eb->start) { | |
1348 | btrfs_put_block_group(cache); | |
1349 | cache = NULL; | |
1350 | ret = false; | |
1351 | } else { | |
1352 | cache->meta_write_pointer = eb->start + eb->len; | |
1353 | } | |
1354 | ||
1355 | *cache_ret = cache; | |
1356 | } | |
1357 | ||
1358 | return ret; | |
1359 | } | |
1360 | ||
1361 | void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache, | |
1362 | struct extent_buffer *eb) | |
1363 | { | |
1364 | if (!btrfs_is_zoned(eb->fs_info) || !cache) | |
1365 | return; | |
1366 | ||
1367 | ASSERT(cache->meta_write_pointer == eb->start + eb->len); | |
1368 | cache->meta_write_pointer = eb->start; | |
1369 | } |