Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
e93c89c1 SB |
2 | /* |
3 | * Copyright (C) STRATO AG 2012. All rights reserved. | |
e93c89c1 | 4 | */ |
c1d7c514 | 5 | |
e93c89c1 SB |
6 | #include <linux/sched.h> |
7 | #include <linux/bio.h> | |
8 | #include <linux/slab.h> | |
e93c89c1 | 9 | #include <linux/blkdev.h> |
e93c89c1 SB |
10 | #include <linux/kthread.h> |
11 | #include <linux/math64.h> | |
602cbe91 | 12 | #include "misc.h" |
e93c89c1 SB |
13 | #include "ctree.h" |
14 | #include "extent_map.h" | |
15 | #include "disk-io.h" | |
16 | #include "transaction.h" | |
17 | #include "print-tree.h" | |
18 | #include "volumes.h" | |
19 | #include "async-thread.h" | |
20 | #include "check-integrity.h" | |
21 | #include "rcu-string.h" | |
22 | #include "dev-replace.h" | |
49c6f736 | 23 | #include "sysfs.h" |
5b316468 | 24 | #include "zoned.h" |
78ce9fc2 | 25 | #include "block-group.h" |
e93c89c1 | 26 | |
30b3688e QW |
27 | /* |
28 | * Device replace overview | |
29 | * | |
30 | * [Objective] | |
31 | * To copy all extents (both new and on-disk) from source device to target | |
32 | * device, while still keeping the filesystem read-write. | |
33 | * | |
34 | * [Method] | |
35 | * There are two main methods involved: | |
36 | * | |
37 | * - Write duplication | |
38 | * | |
39 | * All new writes will be written to both target and source devices, so even | |
1a9fd417 | 40 | * if replace gets canceled, sources device still contains up-to-date data. |
30b3688e QW |
41 | * |
42 | * Location: handle_ops_on_dev_replace() from __btrfs_map_block() | |
43 | * Start: btrfs_dev_replace_start() | |
44 | * End: btrfs_dev_replace_finishing() | |
45 | * Content: Latest data/metadata | |
46 | * | |
47 | * - Copy existing extents | |
48 | * | |
49 | * This happens by re-using scrub facility, as scrub also iterates through | |
50 | * existing extents from commit root. | |
51 | * | |
52 | * Location: scrub_write_block_to_dev_replace() from | |
53 | * scrub_block_complete() | |
54 | * Content: Data/meta from commit root. | |
55 | * | |
56 | * Due to the content difference, we need to avoid nocow write when dev-replace | |
57 | * is happening. This is done by marking the block group read-only and waiting | |
58 | * for NOCOW writes. | |
59 | * | |
60 | * After replace is done, the finishing part is done by swapping the target and | |
61 | * source devices. | |
62 | * | |
63 | * Location: btrfs_dev_replace_update_device_in_mapping_tree() from | |
64 | * btrfs_dev_replace_finishing() | |
65 | */ | |
66 | ||
e93c89c1 SB |
67 | static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, |
68 | int scrub_ret); | |
e93c89c1 | 69 | static int btrfs_dev_replace_kthread(void *data); |
e93c89c1 SB |
70 | |
71 | int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) | |
72 | { | |
562d7b15 | 73 | struct btrfs_dev_lookup_args args = { .devid = BTRFS_DEV_REPLACE_DEVID }; |
e93c89c1 SB |
74 | struct btrfs_key key; |
75 | struct btrfs_root *dev_root = fs_info->dev_root; | |
76 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
77 | struct extent_buffer *eb; | |
78 | int slot; | |
79 | int ret = 0; | |
80 | struct btrfs_path *path = NULL; | |
81 | int item_size; | |
82 | struct btrfs_dev_replace_item *ptr; | |
83 | u64 src_devid; | |
84 | ||
3cb89497 JB |
85 | if (!dev_root) |
86 | return 0; | |
87 | ||
e93c89c1 SB |
88 | path = btrfs_alloc_path(); |
89 | if (!path) { | |
90 | ret = -ENOMEM; | |
91 | goto out; | |
92 | } | |
93 | ||
94 | key.objectid = 0; | |
95 | key.type = BTRFS_DEV_REPLACE_KEY; | |
96 | key.offset = 0; | |
97 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
98 | if (ret) { | |
99 | no_valid_dev_replace_entry_found: | |
cf89af14 AJ |
100 | /* |
101 | * We don't have a replace item or it's corrupted. If there is | |
102 | * a replace target, fail the mount. | |
103 | */ | |
562d7b15 | 104 | if (btrfs_find_device(fs_info->fs_devices, &args)) { |
cf89af14 AJ |
105 | btrfs_err(fs_info, |
106 | "found replace target device without a valid replace item"); | |
107 | ret = -EUCLEAN; | |
108 | goto out; | |
109 | } | |
e93c89c1 SB |
110 | ret = 0; |
111 | dev_replace->replace_state = | |
27e022a9 | 112 | BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED; |
e93c89c1 SB |
113 | dev_replace->cont_reading_from_srcdev_mode = |
114 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; | |
e93c89c1 SB |
115 | dev_replace->time_started = 0; |
116 | dev_replace->time_stopped = 0; | |
117 | atomic64_set(&dev_replace->num_write_errors, 0); | |
118 | atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); | |
119 | dev_replace->cursor_left = 0; | |
120 | dev_replace->committed_cursor_left = 0; | |
121 | dev_replace->cursor_left_last_write_of_item = 0; | |
122 | dev_replace->cursor_right = 0; | |
123 | dev_replace->srcdev = NULL; | |
124 | dev_replace->tgtdev = NULL; | |
125 | dev_replace->is_valid = 0; | |
126 | dev_replace->item_needs_writeback = 0; | |
127 | goto out; | |
128 | } | |
129 | slot = path->slots[0]; | |
130 | eb = path->nodes[0]; | |
3212fa14 | 131 | item_size = btrfs_item_size(eb, slot); |
e93c89c1 SB |
132 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); |
133 | ||
134 | if (item_size != sizeof(struct btrfs_dev_replace_item)) { | |
efe120a0 FH |
135 | btrfs_warn(fs_info, |
136 | "dev_replace entry found has unexpected size, ignore entry"); | |
e93c89c1 SB |
137 | goto no_valid_dev_replace_entry_found; |
138 | } | |
139 | ||
140 | src_devid = btrfs_dev_replace_src_devid(eb, ptr); | |
141 | dev_replace->cont_reading_from_srcdev_mode = | |
142 | btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); | |
143 | dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); | |
144 | dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); | |
145 | dev_replace->time_stopped = | |
146 | btrfs_dev_replace_time_stopped(eb, ptr); | |
147 | atomic64_set(&dev_replace->num_write_errors, | |
148 | btrfs_dev_replace_num_write_errors(eb, ptr)); | |
149 | atomic64_set(&dev_replace->num_uncorrectable_read_errors, | |
150 | btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); | |
151 | dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); | |
152 | dev_replace->committed_cursor_left = dev_replace->cursor_left; | |
153 | dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; | |
154 | dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); | |
155 | dev_replace->is_valid = 1; | |
156 | ||
157 | dev_replace->item_needs_writeback = 0; | |
158 | switch (dev_replace->replace_state) { | |
159 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
160 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
161 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
cf89af14 AJ |
162 | /* |
163 | * We don't have an active replace item but if there is a | |
164 | * replace target, fail the mount. | |
165 | */ | |
562d7b15 | 166 | if (btrfs_find_device(fs_info->fs_devices, &args)) { |
cf89af14 AJ |
167 | btrfs_err(fs_info, |
168 | "replace devid present without an active replace item"); | |
169 | ret = -EUCLEAN; | |
170 | } else { | |
171 | dev_replace->srcdev = NULL; | |
172 | dev_replace->tgtdev = NULL; | |
173 | } | |
e93c89c1 SB |
174 | break; |
175 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
176 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
562d7b15 JB |
177 | dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices, &args); |
178 | args.devid = src_devid; | |
179 | dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices, &args); | |
180 | ||
e93c89c1 SB |
181 | /* |
182 | * allow 'btrfs dev replace_cancel' if src/tgt device is | |
183 | * missing | |
184 | */ | |
185 | if (!dev_replace->srcdev && | |
0b246afa | 186 | !btrfs_test_opt(fs_info, DEGRADED)) { |
e93c89c1 | 187 | ret = -EIO; |
efe120a0 FH |
188 | btrfs_warn(fs_info, |
189 | "cannot mount because device replace operation is ongoing and"); | |
190 | btrfs_warn(fs_info, | |
191 | "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", | |
192 | src_devid); | |
e93c89c1 SB |
193 | } |
194 | if (!dev_replace->tgtdev && | |
0b246afa | 195 | !btrfs_test_opt(fs_info, DEGRADED)) { |
e93c89c1 | 196 | ret = -EIO; |
efe120a0 FH |
197 | btrfs_warn(fs_info, |
198 | "cannot mount because device replace operation is ongoing and"); | |
199 | btrfs_warn(fs_info, | |
200 | "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", | |
6e71c47a | 201 | BTRFS_DEV_REPLACE_DEVID); |
e93c89c1 SB |
202 | } |
203 | if (dev_replace->tgtdev) { | |
204 | if (dev_replace->srcdev) { | |
205 | dev_replace->tgtdev->total_bytes = | |
206 | dev_replace->srcdev->total_bytes; | |
207 | dev_replace->tgtdev->disk_total_bytes = | |
208 | dev_replace->srcdev->disk_total_bytes; | |
935e5cc9 MX |
209 | dev_replace->tgtdev->commit_total_bytes = |
210 | dev_replace->srcdev->commit_total_bytes; | |
e93c89c1 SB |
211 | dev_replace->tgtdev->bytes_used = |
212 | dev_replace->srcdev->bytes_used; | |
ce7213c7 MX |
213 | dev_replace->tgtdev->commit_bytes_used = |
214 | dev_replace->srcdev->commit_bytes_used; | |
e93c89c1 | 215 | } |
401e29c1 AJ |
216 | set_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
217 | &dev_replace->tgtdev->dev_state); | |
15fc1283 AJ |
218 | |
219 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
220 | dev_replace->tgtdev->io_width = fs_info->sectorsize; | |
221 | dev_replace->tgtdev->io_align = fs_info->sectorsize; | |
222 | dev_replace->tgtdev->sector_size = fs_info->sectorsize; | |
223 | dev_replace->tgtdev->fs_info = fs_info; | |
224 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
225 | &dev_replace->tgtdev->dev_state); | |
e93c89c1 SB |
226 | } |
227 | break; | |
228 | } | |
229 | ||
230 | out: | |
527afb44 | 231 | btrfs_free_path(path); |
e93c89c1 SB |
232 | return ret; |
233 | } | |
234 | ||
d48f39d5 DS |
235 | /* |
236 | * Initialize a new device for device replace target from a given source dev | |
237 | * and path. | |
238 | * | |
239 | * Return 0 and new device in @device_out, otherwise return < 0 | |
240 | */ | |
241 | static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
242 | const char *device_path, | |
243 | struct btrfs_device *srcdev, | |
244 | struct btrfs_device **device_out) | |
245 | { | |
bef16b52 | 246 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
d48f39d5 DS |
247 | struct btrfs_device *device; |
248 | struct block_device *bdev; | |
d48f39d5 DS |
249 | struct rcu_string *name; |
250 | u64 devid = BTRFS_DEV_REPLACE_DEVID; | |
251 | int ret = 0; | |
252 | ||
253 | *device_out = NULL; | |
c6a5d954 | 254 | if (srcdev->fs_devices->seeding) { |
d48f39d5 DS |
255 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); |
256 | return -EINVAL; | |
257 | } | |
258 | ||
259 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
260 | fs_info->bdev_holder); | |
261 | if (IS_ERR(bdev)) { | |
262 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
263 | return PTR_ERR(bdev); | |
264 | } | |
265 | ||
b70f5097 NA |
266 | if (!btrfs_check_device_zone_type(fs_info, bdev)) { |
267 | btrfs_err(fs_info, | |
268 | "dev-replace: zoned type of target device mismatch with filesystem"); | |
269 | ret = -EINVAL; | |
270 | goto error; | |
271 | } | |
272 | ||
ddb93784 | 273 | sync_blockdev(bdev); |
d48f39d5 | 274 | |
bef16b52 | 275 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
d48f39d5 DS |
276 | if (device->bdev == bdev) { |
277 | btrfs_err(fs_info, | |
278 | "target device is in the filesystem!"); | |
279 | ret = -EEXIST; | |
280 | goto error; | |
281 | } | |
282 | } | |
283 | ||
284 | ||
cda00eba | 285 | if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(srcdev)) { |
d48f39d5 DS |
286 | btrfs_err(fs_info, |
287 | "target device is smaller than source device!"); | |
288 | ret = -EINVAL; | |
289 | goto error; | |
290 | } | |
291 | ||
292 | ||
293 | device = btrfs_alloc_device(NULL, &devid, NULL); | |
294 | if (IS_ERR(device)) { | |
295 | ret = PTR_ERR(device); | |
296 | goto error; | |
297 | } | |
298 | ||
299 | name = rcu_string_strdup(device_path, GFP_KERNEL); | |
300 | if (!name) { | |
301 | btrfs_free_device(device); | |
302 | ret = -ENOMEM; | |
303 | goto error; | |
304 | } | |
305 | rcu_assign_pointer(device->name, name); | |
4889bc05 AJ |
306 | ret = lookup_bdev(device_path, &device->devt); |
307 | if (ret) | |
308 | goto error; | |
d48f39d5 | 309 | |
d48f39d5 DS |
310 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
311 | device->generation = 0; | |
312 | device->io_width = fs_info->sectorsize; | |
313 | device->io_align = fs_info->sectorsize; | |
314 | device->sector_size = fs_info->sectorsize; | |
315 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); | |
316 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
317 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
318 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
319 | device->commit_bytes_used = device->bytes_used; | |
320 | device->fs_info = fs_info; | |
321 | device->bdev = bdev; | |
322 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); | |
323 | set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); | |
324 | device->mode = FMODE_EXCL; | |
325 | device->dev_stats_valid = 1; | |
326 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); | |
bef16b52 | 327 | device->fs_devices = fs_devices; |
b0d9e1ea | 328 | |
16beac87 | 329 | ret = btrfs_get_dev_zone_info(device, false); |
5b316468 NA |
330 | if (ret) |
331 | goto error; | |
332 | ||
bef16b52 AJ |
333 | mutex_lock(&fs_devices->device_list_mutex); |
334 | list_add(&device->dev_list, &fs_devices->devices); | |
335 | fs_devices->num_devices++; | |
336 | fs_devices->open_devices++; | |
337 | mutex_unlock(&fs_devices->device_list_mutex); | |
d48f39d5 DS |
338 | |
339 | *device_out = device; | |
340 | return 0; | |
341 | ||
342 | error: | |
343 | blkdev_put(bdev, FMODE_EXCL); | |
344 | return ret; | |
345 | } | |
346 | ||
e93c89c1 SB |
347 | /* |
348 | * called from commit_transaction. Writes changed device replace state to | |
349 | * disk. | |
350 | */ | |
2b584c68 | 351 | int btrfs_run_dev_replace(struct btrfs_trans_handle *trans) |
e93c89c1 | 352 | { |
2b584c68 | 353 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e93c89c1 SB |
354 | int ret; |
355 | struct btrfs_root *dev_root = fs_info->dev_root; | |
356 | struct btrfs_path *path; | |
357 | struct btrfs_key key; | |
358 | struct extent_buffer *eb; | |
359 | struct btrfs_dev_replace_item *ptr; | |
360 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
361 | ||
cb5583dd | 362 | down_read(&dev_replace->rwsem); |
e93c89c1 SB |
363 | if (!dev_replace->is_valid || |
364 | !dev_replace->item_needs_writeback) { | |
cb5583dd | 365 | up_read(&dev_replace->rwsem); |
e93c89c1 SB |
366 | return 0; |
367 | } | |
cb5583dd | 368 | up_read(&dev_replace->rwsem); |
e93c89c1 SB |
369 | |
370 | key.objectid = 0; | |
371 | key.type = BTRFS_DEV_REPLACE_KEY; | |
372 | key.offset = 0; | |
373 | ||
374 | path = btrfs_alloc_path(); | |
375 | if (!path) { | |
376 | ret = -ENOMEM; | |
377 | goto out; | |
378 | } | |
379 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
380 | if (ret < 0) { | |
5d163e0e JM |
381 | btrfs_warn(fs_info, |
382 | "error %d while searching for dev_replace item!", | |
383 | ret); | |
e93c89c1 SB |
384 | goto out; |
385 | } | |
386 | ||
387 | if (ret == 0 && | |
3212fa14 | 388 | btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { |
e93c89c1 SB |
389 | /* |
390 | * need to delete old one and insert a new one. | |
391 | * Since no attempt is made to recover any old state, if the | |
392 | * dev_replace state is 'running', the data on the target | |
393 | * drive is lost. | |
394 | * It would be possible to recover the state: just make sure | |
395 | * that the beginning of the item is never changed and always | |
396 | * contains all the essential information. Then read this | |
397 | * minimal set of information and use it as a base for the | |
398 | * new state. | |
399 | */ | |
400 | ret = btrfs_del_item(trans, dev_root, path); | |
401 | if (ret != 0) { | |
5d163e0e JM |
402 | btrfs_warn(fs_info, |
403 | "delete too small dev_replace item failed %d!", | |
404 | ret); | |
e93c89c1 SB |
405 | goto out; |
406 | } | |
407 | ret = 1; | |
408 | } | |
409 | ||
410 | if (ret == 1) { | |
411 | /* need to insert a new item */ | |
412 | btrfs_release_path(path); | |
413 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
414 | &key, sizeof(*ptr)); | |
415 | if (ret < 0) { | |
5d163e0e JM |
416 | btrfs_warn(fs_info, |
417 | "insert dev_replace item failed %d!", ret); | |
e93c89c1 SB |
418 | goto out; |
419 | } | |
420 | } | |
421 | ||
422 | eb = path->nodes[0]; | |
423 | ptr = btrfs_item_ptr(eb, path->slots[0], | |
424 | struct btrfs_dev_replace_item); | |
425 | ||
cb5583dd | 426 | down_write(&dev_replace->rwsem); |
e93c89c1 SB |
427 | if (dev_replace->srcdev) |
428 | btrfs_set_dev_replace_src_devid(eb, ptr, | |
429 | dev_replace->srcdev->devid); | |
430 | else | |
431 | btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); | |
432 | btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, | |
433 | dev_replace->cont_reading_from_srcdev_mode); | |
434 | btrfs_set_dev_replace_replace_state(eb, ptr, | |
435 | dev_replace->replace_state); | |
436 | btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); | |
437 | btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); | |
438 | btrfs_set_dev_replace_num_write_errors(eb, ptr, | |
439 | atomic64_read(&dev_replace->num_write_errors)); | |
440 | btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, | |
441 | atomic64_read(&dev_replace->num_uncorrectable_read_errors)); | |
442 | dev_replace->cursor_left_last_write_of_item = | |
443 | dev_replace->cursor_left; | |
444 | btrfs_set_dev_replace_cursor_left(eb, ptr, | |
445 | dev_replace->cursor_left_last_write_of_item); | |
446 | btrfs_set_dev_replace_cursor_right(eb, ptr, | |
447 | dev_replace->cursor_right); | |
448 | dev_replace->item_needs_writeback = 0; | |
cb5583dd | 449 | up_write(&dev_replace->rwsem); |
e93c89c1 SB |
450 | |
451 | btrfs_mark_buffer_dirty(eb); | |
452 | ||
453 | out: | |
454 | btrfs_free_path(path); | |
455 | ||
456 | return ret; | |
457 | } | |
458 | ||
3c958bd2 AJ |
459 | static char* btrfs_dev_name(struct btrfs_device *device) |
460 | { | |
acf18c56 | 461 | if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
3c958bd2 AJ |
462 | return "<missing disk>"; |
463 | else | |
464 | return rcu_str_deref(device->name); | |
465 | } | |
466 | ||
78ce9fc2 NA |
467 | static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info, |
468 | struct btrfs_device *src_dev) | |
469 | { | |
470 | struct btrfs_path *path; | |
471 | struct btrfs_key key; | |
472 | struct btrfs_key found_key; | |
473 | struct btrfs_root *root = fs_info->dev_root; | |
474 | struct btrfs_dev_extent *dev_extent = NULL; | |
475 | struct btrfs_block_group *cache; | |
476 | struct btrfs_trans_handle *trans; | |
477 | int ret = 0; | |
478 | u64 chunk_offset; | |
479 | ||
480 | /* Do not use "to_copy" on non zoned filesystem for now */ | |
481 | if (!btrfs_is_zoned(fs_info)) | |
482 | return 0; | |
483 | ||
484 | mutex_lock(&fs_info->chunk_mutex); | |
485 | ||
486 | /* Ensure we don't have pending new block group */ | |
487 | spin_lock(&fs_info->trans_lock); | |
488 | while (fs_info->running_transaction && | |
489 | !list_empty(&fs_info->running_transaction->dev_update_list)) { | |
490 | spin_unlock(&fs_info->trans_lock); | |
491 | mutex_unlock(&fs_info->chunk_mutex); | |
492 | trans = btrfs_attach_transaction(root); | |
493 | if (IS_ERR(trans)) { | |
494 | ret = PTR_ERR(trans); | |
495 | mutex_lock(&fs_info->chunk_mutex); | |
496 | if (ret == -ENOENT) { | |
497 | spin_lock(&fs_info->trans_lock); | |
498 | continue; | |
499 | } else { | |
500 | goto unlock; | |
501 | } | |
502 | } | |
503 | ||
504 | ret = btrfs_commit_transaction(trans); | |
505 | mutex_lock(&fs_info->chunk_mutex); | |
506 | if (ret) | |
507 | goto unlock; | |
508 | ||
509 | spin_lock(&fs_info->trans_lock); | |
510 | } | |
511 | spin_unlock(&fs_info->trans_lock); | |
512 | ||
513 | path = btrfs_alloc_path(); | |
514 | if (!path) { | |
515 | ret = -ENOMEM; | |
516 | goto unlock; | |
517 | } | |
518 | ||
519 | path->reada = READA_FORWARD; | |
520 | path->search_commit_root = 1; | |
521 | path->skip_locking = 1; | |
522 | ||
523 | key.objectid = src_dev->devid; | |
524 | key.type = BTRFS_DEV_EXTENT_KEY; | |
525 | key.offset = 0; | |
526 | ||
527 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
528 | if (ret < 0) | |
529 | goto free_path; | |
530 | if (ret > 0) { | |
531 | if (path->slots[0] >= | |
532 | btrfs_header_nritems(path->nodes[0])) { | |
533 | ret = btrfs_next_leaf(root, path); | |
534 | if (ret < 0) | |
535 | goto free_path; | |
536 | if (ret > 0) { | |
537 | ret = 0; | |
538 | goto free_path; | |
539 | } | |
540 | } else { | |
541 | ret = 0; | |
542 | } | |
543 | } | |
544 | ||
545 | while (1) { | |
546 | struct extent_buffer *leaf = path->nodes[0]; | |
547 | int slot = path->slots[0]; | |
548 | ||
549 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
550 | ||
551 | if (found_key.objectid != src_dev->devid) | |
552 | break; | |
553 | ||
554 | if (found_key.type != BTRFS_DEV_EXTENT_KEY) | |
555 | break; | |
556 | ||
557 | if (found_key.offset < key.offset) | |
558 | break; | |
559 | ||
560 | dev_extent = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
561 | ||
562 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent); | |
563 | ||
564 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
565 | if (!cache) | |
566 | goto skip; | |
567 | ||
568 | spin_lock(&cache->lock); | |
569 | cache->to_copy = 1; | |
570 | spin_unlock(&cache->lock); | |
571 | ||
572 | btrfs_put_block_group(cache); | |
573 | ||
574 | skip: | |
575 | ret = btrfs_next_item(root, path); | |
576 | if (ret != 0) { | |
577 | if (ret > 0) | |
578 | ret = 0; | |
579 | break; | |
580 | } | |
581 | } | |
582 | ||
583 | free_path: | |
584 | btrfs_free_path(path); | |
585 | unlock: | |
586 | mutex_unlock(&fs_info->chunk_mutex); | |
587 | ||
588 | return ret; | |
589 | } | |
590 | ||
591 | bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev, | |
592 | struct btrfs_block_group *cache, | |
593 | u64 physical) | |
594 | { | |
595 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
596 | struct extent_map *em; | |
597 | struct map_lookup *map; | |
598 | u64 chunk_offset = cache->start; | |
599 | int num_extents, cur_extent; | |
600 | int i; | |
601 | ||
602 | /* Do not use "to_copy" on non zoned filesystem for now */ | |
603 | if (!btrfs_is_zoned(fs_info)) | |
604 | return true; | |
605 | ||
606 | spin_lock(&cache->lock); | |
607 | if (cache->removed) { | |
608 | spin_unlock(&cache->lock); | |
609 | return true; | |
610 | } | |
611 | spin_unlock(&cache->lock); | |
612 | ||
613 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); | |
614 | ASSERT(!IS_ERR(em)); | |
615 | map = em->map_lookup; | |
616 | ||
617 | num_extents = cur_extent = 0; | |
618 | for (i = 0; i < map->num_stripes; i++) { | |
619 | /* We have more device extent to copy */ | |
620 | if (srcdev != map->stripes[i].dev) | |
621 | continue; | |
622 | ||
623 | num_extents++; | |
624 | if (physical == map->stripes[i].physical) | |
625 | cur_extent = i; | |
626 | } | |
627 | ||
628 | free_extent_map(em); | |
629 | ||
630 | if (num_extents > 1 && cur_extent < num_extents - 1) { | |
631 | /* | |
632 | * Has more stripes on this device. Keep this block group | |
633 | * readonly until we finish all the stripes. | |
634 | */ | |
635 | return false; | |
636 | } | |
637 | ||
638 | /* Last stripe on this device */ | |
639 | spin_lock(&cache->lock); | |
640 | cache->to_copy = 0; | |
641 | spin_unlock(&cache->lock); | |
642 | ||
643 | return true; | |
644 | } | |
645 | ||
54862d6d | 646 | static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info, |
da353f6b DS |
647 | const char *tgtdev_name, u64 srcdevid, const char *srcdev_name, |
648 | int read_src) | |
e93c89c1 | 649 | { |
2ff7e61e | 650 | struct btrfs_root *root = fs_info->dev_root; |
e93c89c1 | 651 | struct btrfs_trans_handle *trans; |
e93c89c1 SB |
652 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
653 | int ret; | |
654 | struct btrfs_device *tgt_device = NULL; | |
655 | struct btrfs_device *src_device = NULL; | |
656 | ||
a27a94c2 NB |
657 | src_device = btrfs_find_device_by_devspec(fs_info, srcdevid, |
658 | srcdev_name); | |
659 | if (IS_ERR(src_device)) | |
660 | return PTR_ERR(src_device); | |
e93c89c1 | 661 | |
eede2bf3 OS |
662 | if (btrfs_pinned_by_swapfile(fs_info, src_device)) { |
663 | btrfs_warn_in_rcu(fs_info, | |
664 | "cannot replace device %s (devid %llu) due to active swapfile", | |
665 | btrfs_dev_name(src_device), src_device->devid); | |
666 | return -ETXTBSY; | |
667 | } | |
668 | ||
9e271ae2 AJ |
669 | /* |
670 | * Here we commit the transaction to make sure commit_total_bytes | |
671 | * of all the devices are updated. | |
672 | */ | |
673 | trans = btrfs_attach_transaction(root); | |
674 | if (!IS_ERR(trans)) { | |
3a45bb20 | 675 | ret = btrfs_commit_transaction(trans); |
9e271ae2 AJ |
676 | if (ret) |
677 | return ret; | |
678 | } else if (PTR_ERR(trans) != -ENOENT) { | |
679 | return PTR_ERR(trans); | |
680 | } | |
681 | ||
e1e0eb43 NB |
682 | ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name, |
683 | src_device, &tgt_device); | |
684 | if (ret) | |
685 | return ret; | |
686 | ||
78ce9fc2 NA |
687 | ret = mark_block_group_to_copy(fs_info, src_device); |
688 | if (ret) | |
689 | return ret; | |
690 | ||
cb5583dd | 691 | down_write(&dev_replace->rwsem); |
e93c89c1 SB |
692 | switch (dev_replace->replace_state) { |
693 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
694 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
695 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
696 | break; | |
697 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
698 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
5c061471 | 699 | ASSERT(0); |
b5255456 | 700 | ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; |
fa19452a | 701 | up_write(&dev_replace->rwsem); |
e93c89c1 SB |
702 | goto leave; |
703 | } | |
704 | ||
b5255456 | 705 | dev_replace->cont_reading_from_srcdev_mode = read_src; |
e93c89c1 | 706 | dev_replace->srcdev = src_device; |
e93c89c1 SB |
707 | dev_replace->tgtdev = tgt_device; |
708 | ||
fc23c246 | 709 | btrfs_info_in_rcu(fs_info, |
ecaeb14b | 710 | "dev_replace from %s (devid %llu) to %s started", |
3c958bd2 | 711 | btrfs_dev_name(src_device), |
e93c89c1 SB |
712 | src_device->devid, |
713 | rcu_str_deref(tgt_device->name)); | |
714 | ||
e93c89c1 SB |
715 | /* |
716 | * from now on, the writes to the srcdev are all duplicated to | |
717 | * go to the tgtdev as well (refer to btrfs_map_block()). | |
718 | */ | |
719 | dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; | |
a944442c | 720 | dev_replace->time_started = ktime_get_real_seconds(); |
e93c89c1 SB |
721 | dev_replace->cursor_left = 0; |
722 | dev_replace->committed_cursor_left = 0; | |
723 | dev_replace->cursor_left_last_write_of_item = 0; | |
724 | dev_replace->cursor_right = 0; | |
725 | dev_replace->is_valid = 1; | |
726 | dev_replace->item_needs_writeback = 1; | |
7ccefb98 YK |
727 | atomic64_set(&dev_replace->num_write_errors, 0); |
728 | atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); | |
cb5583dd | 729 | up_write(&dev_replace->rwsem); |
e93c89c1 | 730 | |
cd36da2e | 731 | ret = btrfs_sysfs_add_device(tgt_device); |
73416dab | 732 | if (ret) |
ab8d0fc4 | 733 | btrfs_err(fs_info, "kobj add dev failed %d", ret); |
73416dab | 734 | |
6374e57a | 735 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
e93c89c1 | 736 | |
a692e13d FM |
737 | /* |
738 | * Commit dev_replace state and reserve 1 item for it. | |
739 | * This is crucial to ensure we won't miss copying extents for new block | |
740 | * groups that are allocated after we started the device replace, and | |
741 | * must be done after setting up the device replace state. | |
742 | */ | |
f232ab04 | 743 | trans = btrfs_start_transaction(root, 1); |
e93c89c1 SB |
744 | if (IS_ERR(trans)) { |
745 | ret = PTR_ERR(trans); | |
cb5583dd | 746 | down_write(&dev_replace->rwsem); |
5c061471 JM |
747 | dev_replace->replace_state = |
748 | BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED; | |
749 | dev_replace->srcdev = NULL; | |
750 | dev_replace->tgtdev = NULL; | |
fa19452a | 751 | up_write(&dev_replace->rwsem); |
e93c89c1 SB |
752 | goto leave; |
753 | } | |
754 | ||
3a45bb20 | 755 | ret = btrfs_commit_transaction(trans); |
e93c89c1 SB |
756 | WARN_ON(ret); |
757 | ||
758 | /* the disk copy procedure reuses the scrub code */ | |
759 | ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, | |
7cc8e58d | 760 | btrfs_device_get_total_bytes(src_device), |
e93c89c1 SB |
761 | &dev_replace->scrub_progress, 0, 1); |
762 | ||
fc23c246 | 763 | ret = btrfs_dev_replace_finishing(fs_info, ret); |
4cea9037 | 764 | if (ret == -EINPROGRESS) |
b5255456 | 765 | ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS; |
e93c89c1 | 766 | |
2fc9f6ba | 767 | return ret; |
e93c89c1 SB |
768 | |
769 | leave: | |
4f5ad7bd | 770 | btrfs_destroy_dev_replace_tgtdev(tgt_device); |
e93c89c1 SB |
771 | return ret; |
772 | } | |
773 | ||
2ff7e61e | 774 | int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info, |
b5255456 AJ |
775 | struct btrfs_ioctl_dev_replace_args *args) |
776 | { | |
777 | int ret; | |
778 | ||
779 | switch (args->start.cont_reading_from_srcdev_mode) { | |
780 | case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: | |
781 | case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: | |
782 | break; | |
783 | default: | |
784 | return -EINVAL; | |
785 | } | |
786 | ||
787 | if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || | |
788 | args->start.tgtdev_name[0] == '\0') | |
789 | return -EINVAL; | |
790 | ||
2ff7e61e | 791 | ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name, |
b5255456 AJ |
792 | args->start.srcdevid, |
793 | args->start.srcdev_name, | |
794 | args->start.cont_reading_from_srcdev_mode); | |
795 | args->result = ret; | |
796 | /* don't warn if EINPROGRESS, someone else might be running scrub */ | |
53e62fb5 AJ |
797 | if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS || |
798 | ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR) | |
799 | return 0; | |
b5255456 AJ |
800 | |
801 | return ret; | |
802 | } | |
803 | ||
c404e0dc | 804 | /* |
01327610 | 805 | * blocked until all in-flight bios operations are finished. |
c404e0dc MX |
806 | */ |
807 | static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) | |
808 | { | |
c404e0dc | 809 | set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); |
7f8d236a DS |
810 | wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum( |
811 | &fs_info->dev_replace.bio_counter)); | |
c404e0dc MX |
812 | } |
813 | ||
814 | /* | |
815 | * we have removed target device, it is safe to allow new bios request. | |
816 | */ | |
817 | static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) | |
818 | { | |
819 | clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); | |
7f8d236a | 820 | wake_up(&fs_info->dev_replace.replace_wait); |
c404e0dc MX |
821 | } |
822 | ||
4c8f3532 FM |
823 | /* |
824 | * When finishing the device replace, before swapping the source device with the | |
825 | * target device we must update the chunk allocation state in the target device, | |
826 | * as it is empty because replace works by directly copying the chunks and not | |
827 | * through the normal chunk allocation path. | |
828 | */ | |
829 | static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev, | |
830 | struct btrfs_device *tgtdev) | |
831 | { | |
832 | struct extent_state *cached_state = NULL; | |
833 | u64 start = 0; | |
834 | u64 found_start; | |
835 | u64 found_end; | |
836 | int ret = 0; | |
837 | ||
838 | lockdep_assert_held(&srcdev->fs_info->chunk_mutex); | |
839 | ||
840 | while (!find_first_extent_bit(&srcdev->alloc_state, start, | |
841 | &found_start, &found_end, | |
842 | CHUNK_ALLOCATED, &cached_state)) { | |
843 | ret = set_extent_bits(&tgtdev->alloc_state, found_start, | |
844 | found_end, CHUNK_ALLOCATED); | |
845 | if (ret) | |
846 | break; | |
847 | start = found_end + 1; | |
848 | } | |
849 | ||
850 | free_extent_state(cached_state); | |
851 | return ret; | |
852 | } | |
853 | ||
0725c0c9 AJ |
854 | static void btrfs_dev_replace_update_device_in_mapping_tree( |
855 | struct btrfs_fs_info *fs_info, | |
856 | struct btrfs_device *srcdev, | |
857 | struct btrfs_device *tgtdev) | |
858 | { | |
859 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; | |
860 | struct extent_map *em; | |
861 | struct map_lookup *map; | |
862 | u64 start = 0; | |
863 | int i; | |
864 | ||
865 | write_lock(&em_tree->lock); | |
866 | do { | |
867 | em = lookup_extent_mapping(em_tree, start, (u64)-1); | |
868 | if (!em) | |
869 | break; | |
870 | map = em->map_lookup; | |
871 | for (i = 0; i < map->num_stripes; i++) | |
872 | if (srcdev == map->stripes[i].dev) | |
873 | map->stripes[i].dev = tgtdev; | |
874 | start = em->start + em->len; | |
875 | free_extent_map(em); | |
876 | } while (start); | |
877 | write_unlock(&em_tree->lock); | |
878 | } | |
879 | ||
e93c89c1 SB |
880 | static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, |
881 | int scrub_ret) | |
882 | { | |
883 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
884 | struct btrfs_device *tgt_device; | |
885 | struct btrfs_device *src_device; | |
886 | struct btrfs_root *root = fs_info->tree_root; | |
887 | u8 uuid_tmp[BTRFS_UUID_SIZE]; | |
888 | struct btrfs_trans_handle *trans; | |
889 | int ret = 0; | |
890 | ||
891 | /* don't allow cancel or unmount to disturb the finishing procedure */ | |
892 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); | |
893 | ||
cb5583dd | 894 | down_read(&dev_replace->rwsem); |
e93c89c1 SB |
895 | /* was the operation canceled, or is it finished? */ |
896 | if (dev_replace->replace_state != | |
897 | BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { | |
cb5583dd | 898 | up_read(&dev_replace->rwsem); |
e93c89c1 SB |
899 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); |
900 | return 0; | |
901 | } | |
902 | ||
903 | tgt_device = dev_replace->tgtdev; | |
904 | src_device = dev_replace->srcdev; | |
cb5583dd | 905 | up_read(&dev_replace->rwsem); |
e93c89c1 | 906 | |
e93c89c1 SB |
907 | /* |
908 | * flush all outstanding I/O and inode extent mappings before the | |
909 | * copy operation is declared as being finished | |
910 | */ | |
9db4dc24 | 911 | ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false); |
3edb2a68 MX |
912 | if (ret) { |
913 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
914 | return ret; | |
915 | } | |
6374e57a | 916 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
e93c89c1 | 917 | |
debd1c06 NB |
918 | /* |
919 | * We have to use this loop approach because at this point src_device | |
920 | * has to be available for transaction commit to complete, yet new | |
921 | * chunks shouldn't be allocated on the device. | |
922 | */ | |
923 | while (1) { | |
924 | trans = btrfs_start_transaction(root, 0); | |
925 | if (IS_ERR(trans)) { | |
926 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
927 | return PTR_ERR(trans); | |
928 | } | |
929 | ret = btrfs_commit_transaction(trans); | |
930 | WARN_ON(ret); | |
931 | ||
932 | /* Prevent write_all_supers() during the finishing procedure */ | |
933 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
934 | /* Prevent new chunks being allocated on the source device */ | |
935 | mutex_lock(&fs_info->chunk_mutex); | |
936 | ||
937 | if (!list_empty(&src_device->post_commit_list)) { | |
938 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
939 | mutex_unlock(&fs_info->chunk_mutex); | |
940 | } else { | |
941 | break; | |
942 | } | |
e93c89c1 | 943 | } |
e93c89c1 | 944 | |
cb5583dd | 945 | down_write(&dev_replace->rwsem); |
e93c89c1 SB |
946 | dev_replace->replace_state = |
947 | scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED | |
948 | : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; | |
949 | dev_replace->tgtdev = NULL; | |
950 | dev_replace->srcdev = NULL; | |
a944442c | 951 | dev_replace->time_stopped = ktime_get_real_seconds(); |
e93c89c1 SB |
952 | dev_replace->item_needs_writeback = 1; |
953 | ||
4c8f3532 FM |
954 | /* |
955 | * Update allocation state in the new device and replace the old device | |
956 | * with the new one in the mapping tree. | |
957 | */ | |
c404e0dc | 958 | if (!scrub_ret) { |
4c8f3532 FM |
959 | scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device); |
960 | if (scrub_ret) | |
961 | goto error; | |
c404e0dc MX |
962 | btrfs_dev_replace_update_device_in_mapping_tree(fs_info, |
963 | src_device, | |
964 | tgt_device); | |
965 | } else { | |
f9085abf AJ |
966 | if (scrub_ret != -ECANCELED) |
967 | btrfs_err_in_rcu(fs_info, | |
0b246afa | 968 | "btrfs_scrub_dev(%s, %llu, %s) failed %d", |
3c958bd2 | 969 | btrfs_dev_name(src_device), |
0b246afa JM |
970 | src_device->devid, |
971 | rcu_str_deref(tgt_device->name), scrub_ret); | |
4c8f3532 | 972 | error: |
cb5583dd | 973 | up_write(&dev_replace->rwsem); |
0b246afa JM |
974 | mutex_unlock(&fs_info->chunk_mutex); |
975 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
ae6529c3 | 976 | btrfs_rm_dev_replace_blocked(fs_info); |
e93c89c1 | 977 | if (tgt_device) |
4f5ad7bd | 978 | btrfs_destroy_dev_replace_tgtdev(tgt_device); |
ae6529c3 | 979 | btrfs_rm_dev_replace_unblocked(fs_info); |
e93c89c1 SB |
980 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); |
981 | ||
2fc9f6ba | 982 | return scrub_ret; |
e93c89c1 SB |
983 | } |
984 | ||
0b246afa JM |
985 | btrfs_info_in_rcu(fs_info, |
986 | "dev_replace from %s (devid %llu) to %s finished", | |
3c958bd2 | 987 | btrfs_dev_name(src_device), |
0b246afa JM |
988 | src_device->devid, |
989 | rcu_str_deref(tgt_device->name)); | |
401e29c1 | 990 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state); |
e93c89c1 SB |
991 | tgt_device->devid = src_device->devid; |
992 | src_device->devid = BTRFS_DEV_REPLACE_DEVID; | |
e93c89c1 SB |
993 | memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); |
994 | memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); | |
995 | memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); | |
7cc8e58d MX |
996 | btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes); |
997 | btrfs_device_set_disk_total_bytes(tgt_device, | |
998 | src_device->disk_total_bytes); | |
999 | btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used); | |
ce7213c7 | 1000 | tgt_device->commit_bytes_used = src_device->bytes_used; |
88acff64 | 1001 | |
d6507cf1 | 1002 | btrfs_assign_next_active_device(src_device, tgt_device); |
88acff64 | 1003 | |
e93c89c1 | 1004 | list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); |
82372bc8 | 1005 | fs_info->fs_devices->rw_devices++; |
e93c89c1 | 1006 | |
cb5583dd | 1007 | up_write(&dev_replace->rwsem); |
c404e0dc MX |
1008 | btrfs_rm_dev_replace_blocked(fs_info); |
1009 | ||
68a9db5f | 1010 | btrfs_rm_dev_replace_remove_srcdev(src_device); |
1357272f | 1011 | |
c404e0dc MX |
1012 | btrfs_rm_dev_replace_unblocked(fs_info); |
1013 | ||
1e7e1f9e MT |
1014 | /* |
1015 | * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will | |
1016 | * update on-disk dev stats value during commit transaction | |
1017 | */ | |
1018 | atomic_inc(&tgt_device->dev_stats_ccnt); | |
1019 | ||
e93c89c1 SB |
1020 | /* |
1021 | * this is again a consistent state where no dev_replace procedure | |
1022 | * is running, the target device is part of the filesystem, the | |
1023 | * source device is not part of the filesystem anymore and its 1st | |
1024 | * superblock is scratched out so that it is no longer marked to | |
1025 | * belong to this filesystem. | |
1026 | */ | |
0b246afa JM |
1027 | mutex_unlock(&fs_info->chunk_mutex); |
1028 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
e93c89c1 | 1029 | |
084b6e7c | 1030 | /* replace the sysfs entry */ |
53f8a74c | 1031 | btrfs_sysfs_remove_device(src_device); |
668e48af | 1032 | btrfs_sysfs_update_devid(tgt_device); |
313b0858 JB |
1033 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state)) |
1034 | btrfs_scratch_superblocks(fs_info, src_device->bdev, | |
1035 | src_device->name->str); | |
084b6e7c | 1036 | |
e93c89c1 SB |
1037 | /* write back the superblocks */ |
1038 | trans = btrfs_start_transaction(root, 0); | |
1039 | if (!IS_ERR(trans)) | |
3a45bb20 | 1040 | btrfs_commit_transaction(trans); |
e93c89c1 SB |
1041 | |
1042 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
1043 | ||
a466c85e JB |
1044 | btrfs_rm_dev_replace_free_srcdev(src_device); |
1045 | ||
e93c89c1 SB |
1046 | return 0; |
1047 | } | |
1048 | ||
74b595fe DS |
1049 | /* |
1050 | * Read progress of device replace status according to the state and last | |
1051 | * stored position. The value format is the same as for | |
1052 | * btrfs_dev_replace::progress_1000 | |
1053 | */ | |
1054 | static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info) | |
1055 | { | |
1056 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1057 | u64 ret = 0; | |
1058 | ||
1059 | switch (dev_replace->replace_state) { | |
1060 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
1061 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
1062 | ret = 0; | |
1063 | break; | |
1064 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
1065 | ret = 1000; | |
1066 | break; | |
1067 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
1068 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
1069 | ret = div64_u64(dev_replace->cursor_left, | |
1070 | div_u64(btrfs_device_get_total_bytes( | |
1071 | dev_replace->srcdev), 1000)); | |
1072 | break; | |
1073 | } | |
1074 | ||
1075 | return ret; | |
1076 | } | |
1077 | ||
e93c89c1 SB |
1078 | void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, |
1079 | struct btrfs_ioctl_dev_replace_args *args) | |
1080 | { | |
1081 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1082 | ||
cb5583dd | 1083 | down_read(&dev_replace->rwsem); |
e93c89c1 SB |
1084 | /* even if !dev_replace_is_valid, the values are good enough for |
1085 | * the replace_status ioctl */ | |
1086 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
1087 | args->status.replace_state = dev_replace->replace_state; | |
1088 | args->status.time_started = dev_replace->time_started; | |
1089 | args->status.time_stopped = dev_replace->time_stopped; | |
1090 | args->status.num_write_errors = | |
1091 | atomic64_read(&dev_replace->num_write_errors); | |
1092 | args->status.num_uncorrectable_read_errors = | |
1093 | atomic64_read(&dev_replace->num_uncorrectable_read_errors); | |
74b595fe | 1094 | args->status.progress_1000 = btrfs_dev_replace_progress(fs_info); |
cb5583dd | 1095 | up_read(&dev_replace->rwsem); |
e93c89c1 SB |
1096 | } |
1097 | ||
18e67c73 | 1098 | int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) |
e93c89c1 SB |
1099 | { |
1100 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1101 | struct btrfs_device *tgt_device = NULL; | |
8f2ceaa7 | 1102 | struct btrfs_device *src_device = NULL; |
e93c89c1 SB |
1103 | struct btrfs_trans_handle *trans; |
1104 | struct btrfs_root *root = fs_info->tree_root; | |
18e67c73 | 1105 | int result; |
e93c89c1 SB |
1106 | int ret; |
1107 | ||
bc98a42c | 1108 | if (sb_rdonly(fs_info->sb)) |
e649e587 ID |
1109 | return -EROFS; |
1110 | ||
e93c89c1 | 1111 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); |
cb5583dd | 1112 | down_write(&dev_replace->rwsem); |
e93c89c1 SB |
1113 | switch (dev_replace->replace_state) { |
1114 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
1115 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
1116 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
1117 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; | |
cb5583dd | 1118 | up_write(&dev_replace->rwsem); |
d189dd70 | 1119 | break; |
e93c89c1 | 1120 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: |
d189dd70 AJ |
1121 | tgt_device = dev_replace->tgtdev; |
1122 | src_device = dev_replace->srcdev; | |
cb5583dd | 1123 | up_write(&dev_replace->rwsem); |
b47dda2e AJ |
1124 | ret = btrfs_scrub_cancel(fs_info); |
1125 | if (ret < 0) { | |
1126 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; | |
1127 | } else { | |
1128 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
1129 | /* | |
1130 | * btrfs_dev_replace_finishing() will handle the | |
1131 | * cleanup part | |
1132 | */ | |
1133 | btrfs_info_in_rcu(fs_info, | |
1134 | "dev_replace from %s (devid %llu) to %s canceled", | |
1135 | btrfs_dev_name(src_device), src_device->devid, | |
1136 | btrfs_dev_name(tgt_device)); | |
1137 | } | |
d189dd70 | 1138 | break; |
e93c89c1 | 1139 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: |
d189dd70 AJ |
1140 | /* |
1141 | * Scrub doing the replace isn't running so we need to do the | |
1142 | * cleanup step of btrfs_dev_replace_finishing() here | |
1143 | */ | |
e93c89c1 SB |
1144 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; |
1145 | tgt_device = dev_replace->tgtdev; | |
8f2ceaa7 | 1146 | src_device = dev_replace->srcdev; |
e93c89c1 SB |
1147 | dev_replace->tgtdev = NULL; |
1148 | dev_replace->srcdev = NULL; | |
d189dd70 AJ |
1149 | dev_replace->replace_state = |
1150 | BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; | |
1151 | dev_replace->time_stopped = ktime_get_real_seconds(); | |
1152 | dev_replace->item_needs_writeback = 1; | |
e93c89c1 | 1153 | |
cb5583dd | 1154 | up_write(&dev_replace->rwsem); |
8f2ceaa7 | 1155 | |
fe97e2e1 AJ |
1156 | /* Scrub for replace must not be running in suspended state */ |
1157 | ret = btrfs_scrub_cancel(fs_info); | |
1158 | ASSERT(ret != -ENOTCONN); | |
d189dd70 AJ |
1159 | |
1160 | trans = btrfs_start_transaction(root, 0); | |
1161 | if (IS_ERR(trans)) { | |
1162 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
1163 | return PTR_ERR(trans); | |
1164 | } | |
1165 | ret = btrfs_commit_transaction(trans); | |
1166 | WARN_ON(ret); | |
8f2ceaa7 | 1167 | |
d189dd70 AJ |
1168 | btrfs_info_in_rcu(fs_info, |
1169 | "suspended dev_replace from %s (devid %llu) to %s canceled", | |
1170 | btrfs_dev_name(src_device), src_device->devid, | |
1171 | btrfs_dev_name(tgt_device)); | |
1172 | ||
1173 | if (tgt_device) | |
1174 | btrfs_destroy_dev_replace_tgtdev(tgt_device); | |
1175 | break; | |
1176 | default: | |
669e859b | 1177 | up_write(&dev_replace->rwsem); |
d189dd70 AJ |
1178 | result = -EINVAL; |
1179 | } | |
e93c89c1 | 1180 | |
e93c89c1 SB |
1181 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); |
1182 | return result; | |
1183 | } | |
1184 | ||
1185 | void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) | |
1186 | { | |
1187 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1188 | ||
1189 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); | |
cb5583dd DS |
1190 | down_write(&dev_replace->rwsem); |
1191 | ||
e93c89c1 SB |
1192 | switch (dev_replace->replace_state) { |
1193 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
1194 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
1195 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
1196 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
1197 | break; | |
1198 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
1199 | dev_replace->replace_state = | |
1200 | BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; | |
a944442c | 1201 | dev_replace->time_stopped = ktime_get_real_seconds(); |
e93c89c1 | 1202 | dev_replace->item_needs_writeback = 1; |
efe120a0 | 1203 | btrfs_info(fs_info, "suspending dev_replace for unmount"); |
e93c89c1 SB |
1204 | break; |
1205 | } | |
1206 | ||
cb5583dd | 1207 | up_write(&dev_replace->rwsem); |
e93c89c1 SB |
1208 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); |
1209 | } | |
1210 | ||
1211 | /* resume dev_replace procedure that was interrupted by unmount */ | |
1212 | int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) | |
1213 | { | |
1214 | struct task_struct *task; | |
1215 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
1216 | ||
cb5583dd DS |
1217 | down_write(&dev_replace->rwsem); |
1218 | ||
e93c89c1 SB |
1219 | switch (dev_replace->replace_state) { |
1220 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
1221 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
1222 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
cb5583dd | 1223 | up_write(&dev_replace->rwsem); |
e93c89c1 SB |
1224 | return 0; |
1225 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
1226 | break; | |
1227 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
1228 | dev_replace->replace_state = | |
1229 | BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; | |
1230 | break; | |
1231 | } | |
1232 | if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { | |
efe120a0 | 1233 | btrfs_info(fs_info, |
5d163e0e JM |
1234 | "cannot continue dev_replace, tgtdev is missing"); |
1235 | btrfs_info(fs_info, | |
1236 | "you may cancel the operation after 'mount -o degraded'"); | |
0d228ece AJ |
1237 | dev_replace->replace_state = |
1238 | BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; | |
cb5583dd | 1239 | up_write(&dev_replace->rwsem); |
e93c89c1 SB |
1240 | return 0; |
1241 | } | |
cb5583dd | 1242 | up_write(&dev_replace->rwsem); |
e93c89c1 | 1243 | |
010a47bd DS |
1244 | /* |
1245 | * This could collide with a paused balance, but the exclusive op logic | |
1246 | * should never allow both to start and pause. We don't want to allow | |
1247 | * dev-replace to start anyway. | |
1248 | */ | |
c3e1f96c | 1249 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) { |
cb5583dd | 1250 | down_write(&dev_replace->rwsem); |
05c49e6b AJ |
1251 | dev_replace->replace_state = |
1252 | BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; | |
cb5583dd | 1253 | up_write(&dev_replace->rwsem); |
010a47bd DS |
1254 | btrfs_info(fs_info, |
1255 | "cannot resume dev-replace, other exclusive operation running"); | |
1256 | return 0; | |
1257 | } | |
1258 | ||
e93c89c1 | 1259 | task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); |
8c6ffba0 | 1260 | return PTR_ERR_OR_ZERO(task); |
e93c89c1 SB |
1261 | } |
1262 | ||
1263 | static int btrfs_dev_replace_kthread(void *data) | |
1264 | { | |
1265 | struct btrfs_fs_info *fs_info = data; | |
1266 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
e93c89c1 | 1267 | u64 progress; |
00251a52 | 1268 | int ret; |
e93c89c1 | 1269 | |
f1b8a1e8 DS |
1270 | progress = btrfs_dev_replace_progress(fs_info); |
1271 | progress = div_u64(progress, 10); | |
1272 | btrfs_info_in_rcu(fs_info, | |
3c958bd2 AJ |
1273 | "continuing dev_replace from %s (devid %llu) to target %s @%u%%", |
1274 | btrfs_dev_name(dev_replace->srcdev), | |
f1b8a1e8 | 1275 | dev_replace->srcdev->devid, |
3c958bd2 | 1276 | btrfs_dev_name(dev_replace->tgtdev), |
f1b8a1e8 DS |
1277 | (unsigned int)progress); |
1278 | ||
e93c89c1 SB |
1279 | ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, |
1280 | dev_replace->committed_cursor_left, | |
7cc8e58d | 1281 | btrfs_device_get_total_bytes(dev_replace->srcdev), |
e93c89c1 SB |
1282 | &dev_replace->scrub_progress, 0, 1); |
1283 | ret = btrfs_dev_replace_finishing(fs_info, ret); | |
49365e69 | 1284 | WARN_ON(ret && ret != -ECANCELED); |
00251a52 | 1285 | |
c3e1f96c | 1286 | btrfs_exclop_finish(fs_info); |
e93c89c1 SB |
1287 | return 0; |
1288 | } | |
1289 | ||
e1f60a65 | 1290 | int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) |
e93c89c1 SB |
1291 | { |
1292 | if (!dev_replace->is_valid) | |
1293 | return 0; | |
1294 | ||
1295 | switch (dev_replace->replace_state) { | |
1296 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
1297 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
1298 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
1299 | return 0; | |
1300 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
1301 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
1302 | /* | |
1303 | * return true even if tgtdev is missing (this is | |
1304 | * something that can happen if the dev_replace | |
1305 | * procedure is suspended by an umount and then | |
1306 | * the tgtdev is missing (or "btrfs dev scan") was | |
52042d8e | 1307 | * not called and the filesystem is remounted |
e93c89c1 SB |
1308 | * in degraded state. This does not stop the |
1309 | * dev_replace procedure. It needs to be canceled | |
bb7ab3b9 | 1310 | * manually if the cancellation is wanted. |
e93c89c1 SB |
1311 | */ |
1312 | break; | |
1313 | } | |
1314 | return 1; | |
1315 | } | |
1316 | ||
c404e0dc MX |
1317 | void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) |
1318 | { | |
7f8d236a | 1319 | percpu_counter_inc(&fs_info->dev_replace.bio_counter); |
c404e0dc MX |
1320 | } |
1321 | ||
4245215d | 1322 | void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount) |
c404e0dc | 1323 | { |
7f8d236a DS |
1324 | percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount); |
1325 | cond_wake_up_nomb(&fs_info->dev_replace.replace_wait); | |
c404e0dc MX |
1326 | } |
1327 | ||
1328 | void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) | |
1329 | { | |
09dd7a01 | 1330 | while (1) { |
7f8d236a | 1331 | percpu_counter_inc(&fs_info->dev_replace.bio_counter); |
09dd7a01 ZL |
1332 | if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING, |
1333 | &fs_info->fs_state))) | |
1334 | break; | |
1335 | ||
c404e0dc | 1336 | btrfs_bio_counter_dec(fs_info); |
7f8d236a | 1337 | wait_event(fs_info->dev_replace.replace_wait, |
c404e0dc MX |
1338 | !test_bit(BTRFS_FS_STATE_DEV_REPLACING, |
1339 | &fs_info->fs_state)); | |
c404e0dc | 1340 | } |
c404e0dc | 1341 | } |