]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) STRATO AG 2012. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/buffer_head.h> | |
22 | #include <linux/blkdev.h> | |
23 | #include <linux/random.h> | |
24 | #include <linux/iocontext.h> | |
25 | #include <linux/capability.h> | |
26 | #include <linux/kthread.h> | |
27 | #include <linux/math64.h> | |
28 | #include <asm/div64.h> | |
29 | #include "ctree.h" | |
30 | #include "extent_map.h" | |
31 | #include "disk-io.h" | |
32 | #include "transaction.h" | |
33 | #include "print-tree.h" | |
34 | #include "volumes.h" | |
35 | #include "async-thread.h" | |
36 | #include "check-integrity.h" | |
37 | #include "rcu-string.h" | |
38 | #include "dev-replace.h" | |
39 | #include "sysfs.h" | |
40 | ||
41 | static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, | |
42 | int scrub_ret); | |
43 | static void btrfs_dev_replace_update_device_in_mapping_tree( | |
44 | struct btrfs_fs_info *fs_info, | |
45 | struct btrfs_device *srcdev, | |
46 | struct btrfs_device *tgtdev); | |
47 | static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, | |
48 | char *srcdev_name, | |
49 | struct btrfs_device **device); | |
50 | static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info); | |
51 | static int btrfs_dev_replace_kthread(void *data); | |
52 | static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info); | |
53 | ||
54 | ||
55 | int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) | |
56 | { | |
57 | struct btrfs_key key; | |
58 | struct btrfs_root *dev_root = fs_info->dev_root; | |
59 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
60 | struct extent_buffer *eb; | |
61 | int slot; | |
62 | int ret = 0; | |
63 | struct btrfs_path *path = NULL; | |
64 | int item_size; | |
65 | struct btrfs_dev_replace_item *ptr; | |
66 | u64 src_devid; | |
67 | ||
68 | path = btrfs_alloc_path(); | |
69 | if (!path) { | |
70 | ret = -ENOMEM; | |
71 | goto out; | |
72 | } | |
73 | ||
74 | key.objectid = 0; | |
75 | key.type = BTRFS_DEV_REPLACE_KEY; | |
76 | key.offset = 0; | |
77 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
78 | if (ret) { | |
79 | no_valid_dev_replace_entry_found: | |
80 | ret = 0; | |
81 | dev_replace->replace_state = | |
82 | BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED; | |
83 | dev_replace->cont_reading_from_srcdev_mode = | |
84 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; | |
85 | dev_replace->replace_state = 0; | |
86 | dev_replace->time_started = 0; | |
87 | dev_replace->time_stopped = 0; | |
88 | atomic64_set(&dev_replace->num_write_errors, 0); | |
89 | atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); | |
90 | dev_replace->cursor_left = 0; | |
91 | dev_replace->committed_cursor_left = 0; | |
92 | dev_replace->cursor_left_last_write_of_item = 0; | |
93 | dev_replace->cursor_right = 0; | |
94 | dev_replace->srcdev = NULL; | |
95 | dev_replace->tgtdev = NULL; | |
96 | dev_replace->is_valid = 0; | |
97 | dev_replace->item_needs_writeback = 0; | |
98 | goto out; | |
99 | } | |
100 | slot = path->slots[0]; | |
101 | eb = path->nodes[0]; | |
102 | item_size = btrfs_item_size_nr(eb, slot); | |
103 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); | |
104 | ||
105 | if (item_size != sizeof(struct btrfs_dev_replace_item)) { | |
106 | btrfs_warn(fs_info, | |
107 | "dev_replace entry found has unexpected size, ignore entry"); | |
108 | goto no_valid_dev_replace_entry_found; | |
109 | } | |
110 | ||
111 | src_devid = btrfs_dev_replace_src_devid(eb, ptr); | |
112 | dev_replace->cont_reading_from_srcdev_mode = | |
113 | btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); | |
114 | dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); | |
115 | dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); | |
116 | dev_replace->time_stopped = | |
117 | btrfs_dev_replace_time_stopped(eb, ptr); | |
118 | atomic64_set(&dev_replace->num_write_errors, | |
119 | btrfs_dev_replace_num_write_errors(eb, ptr)); | |
120 | atomic64_set(&dev_replace->num_uncorrectable_read_errors, | |
121 | btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); | |
122 | dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); | |
123 | dev_replace->committed_cursor_left = dev_replace->cursor_left; | |
124 | dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; | |
125 | dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); | |
126 | dev_replace->is_valid = 1; | |
127 | ||
128 | dev_replace->item_needs_writeback = 0; | |
129 | switch (dev_replace->replace_state) { | |
130 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
131 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
132 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
133 | dev_replace->srcdev = NULL; | |
134 | dev_replace->tgtdev = NULL; | |
135 | break; | |
136 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
137 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
138 | dev_replace->srcdev = btrfs_find_device(fs_info, src_devid, | |
139 | NULL, NULL); | |
140 | dev_replace->tgtdev = btrfs_find_device(fs_info, | |
141 | BTRFS_DEV_REPLACE_DEVID, | |
142 | NULL, NULL); | |
143 | /* | |
144 | * allow 'btrfs dev replace_cancel' if src/tgt device is | |
145 | * missing | |
146 | */ | |
147 | if (!dev_replace->srcdev && | |
148 | !btrfs_test_opt(dev_root, DEGRADED)) { | |
149 | ret = -EIO; | |
150 | btrfs_warn(fs_info, | |
151 | "cannot mount because device replace operation is ongoing and"); | |
152 | btrfs_warn(fs_info, | |
153 | "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", | |
154 | src_devid); | |
155 | } | |
156 | if (!dev_replace->tgtdev && | |
157 | !btrfs_test_opt(dev_root, DEGRADED)) { | |
158 | ret = -EIO; | |
159 | btrfs_warn(fs_info, | |
160 | "cannot mount because device replace operation is ongoing and"); | |
161 | btrfs_warn(fs_info, | |
162 | "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", | |
163 | BTRFS_DEV_REPLACE_DEVID); | |
164 | } | |
165 | if (dev_replace->tgtdev) { | |
166 | if (dev_replace->srcdev) { | |
167 | dev_replace->tgtdev->total_bytes = | |
168 | dev_replace->srcdev->total_bytes; | |
169 | dev_replace->tgtdev->disk_total_bytes = | |
170 | dev_replace->srcdev->disk_total_bytes; | |
171 | dev_replace->tgtdev->commit_total_bytes = | |
172 | dev_replace->srcdev->commit_total_bytes; | |
173 | dev_replace->tgtdev->bytes_used = | |
174 | dev_replace->srcdev->bytes_used; | |
175 | dev_replace->tgtdev->commit_bytes_used = | |
176 | dev_replace->srcdev->commit_bytes_used; | |
177 | } | |
178 | dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1; | |
179 | btrfs_init_dev_replace_tgtdev_for_resume(fs_info, | |
180 | dev_replace->tgtdev); | |
181 | } | |
182 | break; | |
183 | } | |
184 | ||
185 | out: | |
186 | btrfs_free_path(path); | |
187 | return ret; | |
188 | } | |
189 | ||
190 | /* | |
191 | * called from commit_transaction. Writes changed device replace state to | |
192 | * disk. | |
193 | */ | |
194 | int btrfs_run_dev_replace(struct btrfs_trans_handle *trans, | |
195 | struct btrfs_fs_info *fs_info) | |
196 | { | |
197 | int ret; | |
198 | struct btrfs_root *dev_root = fs_info->dev_root; | |
199 | struct btrfs_path *path; | |
200 | struct btrfs_key key; | |
201 | struct extent_buffer *eb; | |
202 | struct btrfs_dev_replace_item *ptr; | |
203 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
204 | ||
205 | btrfs_dev_replace_lock(dev_replace); | |
206 | if (!dev_replace->is_valid || | |
207 | !dev_replace->item_needs_writeback) { | |
208 | btrfs_dev_replace_unlock(dev_replace); | |
209 | return 0; | |
210 | } | |
211 | btrfs_dev_replace_unlock(dev_replace); | |
212 | ||
213 | key.objectid = 0; | |
214 | key.type = BTRFS_DEV_REPLACE_KEY; | |
215 | key.offset = 0; | |
216 | ||
217 | path = btrfs_alloc_path(); | |
218 | if (!path) { | |
219 | ret = -ENOMEM; | |
220 | goto out; | |
221 | } | |
222 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
223 | if (ret < 0) { | |
224 | btrfs_warn(fs_info, "error %d while searching for dev_replace item!", | |
225 | ret); | |
226 | goto out; | |
227 | } | |
228 | ||
229 | if (ret == 0 && | |
230 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
231 | /* | |
232 | * need to delete old one and insert a new one. | |
233 | * Since no attempt is made to recover any old state, if the | |
234 | * dev_replace state is 'running', the data on the target | |
235 | * drive is lost. | |
236 | * It would be possible to recover the state: just make sure | |
237 | * that the beginning of the item is never changed and always | |
238 | * contains all the essential information. Then read this | |
239 | * minimal set of information and use it as a base for the | |
240 | * new state. | |
241 | */ | |
242 | ret = btrfs_del_item(trans, dev_root, path); | |
243 | if (ret != 0) { | |
244 | btrfs_warn(fs_info, "delete too small dev_replace item failed %d!", | |
245 | ret); | |
246 | goto out; | |
247 | } | |
248 | ret = 1; | |
249 | } | |
250 | ||
251 | if (ret == 1) { | |
252 | /* need to insert a new item */ | |
253 | btrfs_release_path(path); | |
254 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
255 | &key, sizeof(*ptr)); | |
256 | if (ret < 0) { | |
257 | btrfs_warn(fs_info, "insert dev_replace item failed %d!", | |
258 | ret); | |
259 | goto out; | |
260 | } | |
261 | } | |
262 | ||
263 | eb = path->nodes[0]; | |
264 | ptr = btrfs_item_ptr(eb, path->slots[0], | |
265 | struct btrfs_dev_replace_item); | |
266 | ||
267 | btrfs_dev_replace_lock(dev_replace); | |
268 | if (dev_replace->srcdev) | |
269 | btrfs_set_dev_replace_src_devid(eb, ptr, | |
270 | dev_replace->srcdev->devid); | |
271 | else | |
272 | btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); | |
273 | btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, | |
274 | dev_replace->cont_reading_from_srcdev_mode); | |
275 | btrfs_set_dev_replace_replace_state(eb, ptr, | |
276 | dev_replace->replace_state); | |
277 | btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); | |
278 | btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); | |
279 | btrfs_set_dev_replace_num_write_errors(eb, ptr, | |
280 | atomic64_read(&dev_replace->num_write_errors)); | |
281 | btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, | |
282 | atomic64_read(&dev_replace->num_uncorrectable_read_errors)); | |
283 | dev_replace->cursor_left_last_write_of_item = | |
284 | dev_replace->cursor_left; | |
285 | btrfs_set_dev_replace_cursor_left(eb, ptr, | |
286 | dev_replace->cursor_left_last_write_of_item); | |
287 | btrfs_set_dev_replace_cursor_right(eb, ptr, | |
288 | dev_replace->cursor_right); | |
289 | dev_replace->item_needs_writeback = 0; | |
290 | btrfs_dev_replace_unlock(dev_replace); | |
291 | ||
292 | btrfs_mark_buffer_dirty(eb); | |
293 | ||
294 | out: | |
295 | btrfs_free_path(path); | |
296 | ||
297 | return ret; | |
298 | } | |
299 | ||
300 | void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info) | |
301 | { | |
302 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
303 | ||
304 | dev_replace->committed_cursor_left = | |
305 | dev_replace->cursor_left_last_write_of_item; | |
306 | } | |
307 | ||
308 | int btrfs_dev_replace_start(struct btrfs_root *root, | |
309 | struct btrfs_ioctl_dev_replace_args *args) | |
310 | { | |
311 | struct btrfs_trans_handle *trans; | |
312 | struct btrfs_fs_info *fs_info = root->fs_info; | |
313 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
314 | int ret; | |
315 | struct btrfs_device *tgt_device = NULL; | |
316 | struct btrfs_device *src_device = NULL; | |
317 | ||
318 | switch (args->start.cont_reading_from_srcdev_mode) { | |
319 | case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: | |
320 | case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: | |
321 | break; | |
322 | default: | |
323 | return -EINVAL; | |
324 | } | |
325 | ||
326 | if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || | |
327 | args->start.tgtdev_name[0] == '\0') | |
328 | return -EINVAL; | |
329 | ||
330 | /* the disk copy procedure reuses the scrub code */ | |
331 | mutex_lock(&fs_info->volume_mutex); | |
332 | ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid, | |
333 | args->start.srcdev_name, | |
334 | &src_device); | |
335 | if (ret) { | |
336 | mutex_unlock(&fs_info->volume_mutex); | |
337 | return ret; | |
338 | } | |
339 | ||
340 | ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name, | |
341 | src_device, &tgt_device); | |
342 | mutex_unlock(&fs_info->volume_mutex); | |
343 | if (ret) | |
344 | return ret; | |
345 | ||
346 | /* | |
347 | * Here we commit the transaction to make sure commit_total_bytes | |
348 | * of all the devices are updated. | |
349 | */ | |
350 | trans = btrfs_attach_transaction(root); | |
351 | if (!IS_ERR(trans)) { | |
352 | ret = btrfs_commit_transaction(trans, root); | |
353 | if (ret) | |
354 | return ret; | |
355 | } else if (PTR_ERR(trans) != -ENOENT) { | |
356 | return PTR_ERR(trans); | |
357 | } | |
358 | ||
359 | btrfs_dev_replace_lock(dev_replace); | |
360 | switch (dev_replace->replace_state) { | |
361 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
362 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
363 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
364 | break; | |
365 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
366 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
367 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; | |
368 | goto leave; | |
369 | } | |
370 | ||
371 | dev_replace->cont_reading_from_srcdev_mode = | |
372 | args->start.cont_reading_from_srcdev_mode; | |
373 | WARN_ON(!src_device); | |
374 | dev_replace->srcdev = src_device; | |
375 | WARN_ON(!tgt_device); | |
376 | dev_replace->tgtdev = tgt_device; | |
377 | ||
378 | btrfs_info_in_rcu(root->fs_info, | |
379 | "dev_replace from %s (devid %llu) to %s started", | |
380 | src_device->missing ? "<missing disk>" : | |
381 | rcu_str_deref(src_device->name), | |
382 | src_device->devid, | |
383 | rcu_str_deref(tgt_device->name)); | |
384 | ||
385 | /* | |
386 | * from now on, the writes to the srcdev are all duplicated to | |
387 | * go to the tgtdev as well (refer to btrfs_map_block()). | |
388 | */ | |
389 | dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; | |
390 | dev_replace->time_started = get_seconds(); | |
391 | dev_replace->cursor_left = 0; | |
392 | dev_replace->committed_cursor_left = 0; | |
393 | dev_replace->cursor_left_last_write_of_item = 0; | |
394 | dev_replace->cursor_right = 0; | |
395 | dev_replace->is_valid = 1; | |
396 | dev_replace->item_needs_writeback = 1; | |
397 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
398 | btrfs_dev_replace_unlock(dev_replace); | |
399 | ||
400 | ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device); | |
401 | if (ret) | |
402 | btrfs_err(root->fs_info, "kobj add dev failed %d\n", ret); | |
403 | ||
404 | btrfs_wait_ordered_roots(root->fs_info, -1); | |
405 | ||
406 | /* force writing the updated state information to disk */ | |
407 | trans = btrfs_start_transaction(root, 0); | |
408 | if (IS_ERR(trans)) { | |
409 | ret = PTR_ERR(trans); | |
410 | btrfs_dev_replace_lock(dev_replace); | |
411 | goto leave; | |
412 | } | |
413 | ||
414 | ret = btrfs_commit_transaction(trans, root); | |
415 | WARN_ON(ret); | |
416 | ||
417 | /* the disk copy procedure reuses the scrub code */ | |
418 | ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, | |
419 | btrfs_device_get_total_bytes(src_device), | |
420 | &dev_replace->scrub_progress, 0, 1); | |
421 | ||
422 | ret = btrfs_dev_replace_finishing(root->fs_info, ret); | |
423 | /* don't warn if EINPROGRESS, someone else might be running scrub */ | |
424 | if (ret == -EINPROGRESS) { | |
425 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS; | |
426 | ret = 0; | |
427 | } else { | |
428 | WARN_ON(ret); | |
429 | } | |
430 | ||
431 | return ret; | |
432 | ||
433 | leave: | |
434 | dev_replace->srcdev = NULL; | |
435 | dev_replace->tgtdev = NULL; | |
436 | btrfs_dev_replace_unlock(dev_replace); | |
437 | btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); | |
438 | return ret; | |
439 | } | |
440 | ||
441 | /* | |
442 | * blocked until all flighting bios are finished. | |
443 | */ | |
444 | static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) | |
445 | { | |
446 | set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); | |
447 | wait_event(fs_info->replace_wait, !percpu_counter_sum( | |
448 | &fs_info->bio_counter)); | |
449 | } | |
450 | ||
451 | /* | |
452 | * we have removed target device, it is safe to allow new bios request. | |
453 | */ | |
454 | static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) | |
455 | { | |
456 | clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); | |
457 | wake_up(&fs_info->replace_wait); | |
458 | } | |
459 | ||
460 | static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, | |
461 | int scrub_ret) | |
462 | { | |
463 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
464 | struct btrfs_device *tgt_device; | |
465 | struct btrfs_device *src_device; | |
466 | struct btrfs_root *root = fs_info->tree_root; | |
467 | u8 uuid_tmp[BTRFS_UUID_SIZE]; | |
468 | struct btrfs_trans_handle *trans; | |
469 | int ret = 0; | |
470 | ||
471 | /* don't allow cancel or unmount to disturb the finishing procedure */ | |
472 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); | |
473 | ||
474 | btrfs_dev_replace_lock(dev_replace); | |
475 | /* was the operation canceled, or is it finished? */ | |
476 | if (dev_replace->replace_state != | |
477 | BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { | |
478 | btrfs_dev_replace_unlock(dev_replace); | |
479 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
480 | return 0; | |
481 | } | |
482 | ||
483 | tgt_device = dev_replace->tgtdev; | |
484 | src_device = dev_replace->srcdev; | |
485 | btrfs_dev_replace_unlock(dev_replace); | |
486 | ||
487 | /* | |
488 | * flush all outstanding I/O and inode extent mappings before the | |
489 | * copy operation is declared as being finished | |
490 | */ | |
491 | ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1); | |
492 | if (ret) { | |
493 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
494 | return ret; | |
495 | } | |
496 | btrfs_wait_ordered_roots(root->fs_info, -1); | |
497 | ||
498 | trans = btrfs_start_transaction(root, 0); | |
499 | if (IS_ERR(trans)) { | |
500 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
501 | return PTR_ERR(trans); | |
502 | } | |
503 | ret = btrfs_commit_transaction(trans, root); | |
504 | WARN_ON(ret); | |
505 | ||
506 | mutex_lock(&uuid_mutex); | |
507 | /* keep away write_all_supers() during the finishing procedure */ | |
508 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
509 | mutex_lock(&root->fs_info->chunk_mutex); | |
510 | btrfs_dev_replace_lock(dev_replace); | |
511 | dev_replace->replace_state = | |
512 | scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED | |
513 | : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; | |
514 | dev_replace->tgtdev = NULL; | |
515 | dev_replace->srcdev = NULL; | |
516 | dev_replace->time_stopped = get_seconds(); | |
517 | dev_replace->item_needs_writeback = 1; | |
518 | ||
519 | /* replace old device with new one in mapping tree */ | |
520 | if (!scrub_ret) { | |
521 | btrfs_dev_replace_update_device_in_mapping_tree(fs_info, | |
522 | src_device, | |
523 | tgt_device); | |
524 | } else { | |
525 | btrfs_err_in_rcu(root->fs_info, | |
526 | "btrfs_scrub_dev(%s, %llu, %s) failed %d", | |
527 | src_device->missing ? "<missing disk>" : | |
528 | rcu_str_deref(src_device->name), | |
529 | src_device->devid, | |
530 | rcu_str_deref(tgt_device->name), scrub_ret); | |
531 | btrfs_dev_replace_unlock(dev_replace); | |
532 | mutex_unlock(&root->fs_info->chunk_mutex); | |
533 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
534 | mutex_unlock(&uuid_mutex); | |
535 | if (tgt_device) | |
536 | btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); | |
537 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
538 | ||
539 | return scrub_ret; | |
540 | } | |
541 | ||
542 | btrfs_info_in_rcu(root->fs_info, | |
543 | "dev_replace from %s (devid %llu) to %s finished", | |
544 | src_device->missing ? "<missing disk>" : | |
545 | rcu_str_deref(src_device->name), | |
546 | src_device->devid, | |
547 | rcu_str_deref(tgt_device->name)); | |
548 | tgt_device->is_tgtdev_for_dev_replace = 0; | |
549 | tgt_device->devid = src_device->devid; | |
550 | src_device->devid = BTRFS_DEV_REPLACE_DEVID; | |
551 | memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); | |
552 | memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); | |
553 | memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); | |
554 | btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes); | |
555 | btrfs_device_set_disk_total_bytes(tgt_device, | |
556 | src_device->disk_total_bytes); | |
557 | btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used); | |
558 | ASSERT(list_empty(&src_device->resized_list)); | |
559 | tgt_device->commit_total_bytes = src_device->commit_total_bytes; | |
560 | tgt_device->commit_bytes_used = src_device->bytes_used; | |
561 | if (fs_info->sb->s_bdev == src_device->bdev) | |
562 | fs_info->sb->s_bdev = tgt_device->bdev; | |
563 | if (fs_info->fs_devices->latest_bdev == src_device->bdev) | |
564 | fs_info->fs_devices->latest_bdev = tgt_device->bdev; | |
565 | list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); | |
566 | fs_info->fs_devices->rw_devices++; | |
567 | ||
568 | btrfs_dev_replace_unlock(dev_replace); | |
569 | ||
570 | btrfs_rm_dev_replace_blocked(fs_info); | |
571 | ||
572 | btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device); | |
573 | ||
574 | btrfs_rm_dev_replace_unblocked(fs_info); | |
575 | ||
576 | /* | |
577 | * this is again a consistent state where no dev_replace procedure | |
578 | * is running, the target device is part of the filesystem, the | |
579 | * source device is not part of the filesystem anymore and its 1st | |
580 | * superblock is scratched out so that it is no longer marked to | |
581 | * belong to this filesystem. | |
582 | */ | |
583 | mutex_unlock(&root->fs_info->chunk_mutex); | |
584 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
585 | mutex_unlock(&uuid_mutex); | |
586 | ||
587 | /* replace the sysfs entry */ | |
588 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device); | |
589 | btrfs_rm_dev_replace_free_srcdev(fs_info, src_device); | |
590 | ||
591 | /* write back the superblocks */ | |
592 | trans = btrfs_start_transaction(root, 0); | |
593 | if (!IS_ERR(trans)) | |
594 | btrfs_commit_transaction(trans, root); | |
595 | ||
596 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
597 | ||
598 | return 0; | |
599 | } | |
600 | ||
601 | static void btrfs_dev_replace_update_device_in_mapping_tree( | |
602 | struct btrfs_fs_info *fs_info, | |
603 | struct btrfs_device *srcdev, | |
604 | struct btrfs_device *tgtdev) | |
605 | { | |
606 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
607 | struct extent_map *em; | |
608 | struct map_lookup *map; | |
609 | u64 start = 0; | |
610 | int i; | |
611 | ||
612 | write_lock(&em_tree->lock); | |
613 | do { | |
614 | em = lookup_extent_mapping(em_tree, start, (u64)-1); | |
615 | if (!em) | |
616 | break; | |
617 | map = (struct map_lookup *)em->bdev; | |
618 | for (i = 0; i < map->num_stripes; i++) | |
619 | if (srcdev == map->stripes[i].dev) | |
620 | map->stripes[i].dev = tgtdev; | |
621 | start = em->start + em->len; | |
622 | free_extent_map(em); | |
623 | } while (start); | |
624 | write_unlock(&em_tree->lock); | |
625 | } | |
626 | ||
627 | static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, | |
628 | char *srcdev_name, | |
629 | struct btrfs_device **device) | |
630 | { | |
631 | int ret; | |
632 | ||
633 | if (srcdevid) { | |
634 | ret = 0; | |
635 | *device = btrfs_find_device(root->fs_info, srcdevid, NULL, | |
636 | NULL); | |
637 | if (!*device) | |
638 | ret = -ENOENT; | |
639 | } else { | |
640 | ret = btrfs_find_device_missing_or_by_path(root, srcdev_name, | |
641 | device); | |
642 | } | |
643 | return ret; | |
644 | } | |
645 | ||
646 | void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, | |
647 | struct btrfs_ioctl_dev_replace_args *args) | |
648 | { | |
649 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
650 | struct btrfs_device *srcdev; | |
651 | ||
652 | btrfs_dev_replace_lock(dev_replace); | |
653 | /* even if !dev_replace_is_valid, the values are good enough for | |
654 | * the replace_status ioctl */ | |
655 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
656 | args->status.replace_state = dev_replace->replace_state; | |
657 | args->status.time_started = dev_replace->time_started; | |
658 | args->status.time_stopped = dev_replace->time_stopped; | |
659 | args->status.num_write_errors = | |
660 | atomic64_read(&dev_replace->num_write_errors); | |
661 | args->status.num_uncorrectable_read_errors = | |
662 | atomic64_read(&dev_replace->num_uncorrectable_read_errors); | |
663 | switch (dev_replace->replace_state) { | |
664 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
665 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
666 | args->status.progress_1000 = 0; | |
667 | break; | |
668 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
669 | args->status.progress_1000 = 1000; | |
670 | break; | |
671 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
672 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
673 | srcdev = dev_replace->srcdev; | |
674 | args->status.progress_1000 = div_u64(dev_replace->cursor_left, | |
675 | div_u64(btrfs_device_get_total_bytes(srcdev), 1000)); | |
676 | break; | |
677 | } | |
678 | btrfs_dev_replace_unlock(dev_replace); | |
679 | } | |
680 | ||
681 | int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info, | |
682 | struct btrfs_ioctl_dev_replace_args *args) | |
683 | { | |
684 | args->result = __btrfs_dev_replace_cancel(fs_info); | |
685 | return 0; | |
686 | } | |
687 | ||
688 | static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) | |
689 | { | |
690 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
691 | struct btrfs_device *tgt_device = NULL; | |
692 | struct btrfs_trans_handle *trans; | |
693 | struct btrfs_root *root = fs_info->tree_root; | |
694 | u64 result; | |
695 | int ret; | |
696 | ||
697 | if (fs_info->sb->s_flags & MS_RDONLY) | |
698 | return -EROFS; | |
699 | ||
700 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); | |
701 | btrfs_dev_replace_lock(dev_replace); | |
702 | switch (dev_replace->replace_state) { | |
703 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
704 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
705 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
706 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; | |
707 | btrfs_dev_replace_unlock(dev_replace); | |
708 | goto leave; | |
709 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
710 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
711 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
712 | tgt_device = dev_replace->tgtdev; | |
713 | dev_replace->tgtdev = NULL; | |
714 | dev_replace->srcdev = NULL; | |
715 | break; | |
716 | } | |
717 | dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; | |
718 | dev_replace->time_stopped = get_seconds(); | |
719 | dev_replace->item_needs_writeback = 1; | |
720 | btrfs_dev_replace_unlock(dev_replace); | |
721 | btrfs_scrub_cancel(fs_info); | |
722 | ||
723 | trans = btrfs_start_transaction(root, 0); | |
724 | if (IS_ERR(trans)) { | |
725 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
726 | return PTR_ERR(trans); | |
727 | } | |
728 | ret = btrfs_commit_transaction(trans, root); | |
729 | WARN_ON(ret); | |
730 | if (tgt_device) | |
731 | btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); | |
732 | ||
733 | leave: | |
734 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
735 | return result; | |
736 | } | |
737 | ||
738 | void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) | |
739 | { | |
740 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
741 | ||
742 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); | |
743 | btrfs_dev_replace_lock(dev_replace); | |
744 | switch (dev_replace->replace_state) { | |
745 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
746 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
747 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
748 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
749 | break; | |
750 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
751 | dev_replace->replace_state = | |
752 | BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; | |
753 | dev_replace->time_stopped = get_seconds(); | |
754 | dev_replace->item_needs_writeback = 1; | |
755 | btrfs_info(fs_info, "suspending dev_replace for unmount"); | |
756 | break; | |
757 | } | |
758 | ||
759 | btrfs_dev_replace_unlock(dev_replace); | |
760 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
761 | } | |
762 | ||
763 | /* resume dev_replace procedure that was interrupted by unmount */ | |
764 | int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) | |
765 | { | |
766 | struct task_struct *task; | |
767 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
768 | ||
769 | btrfs_dev_replace_lock(dev_replace); | |
770 | switch (dev_replace->replace_state) { | |
771 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
772 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
773 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
774 | btrfs_dev_replace_unlock(dev_replace); | |
775 | return 0; | |
776 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
777 | break; | |
778 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
779 | dev_replace->replace_state = | |
780 | BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; | |
781 | break; | |
782 | } | |
783 | if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { | |
784 | btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing"); | |
785 | btrfs_info(fs_info, | |
786 | "you may cancel the operation after 'mount -o degraded'"); | |
787 | btrfs_dev_replace_unlock(dev_replace); | |
788 | return 0; | |
789 | } | |
790 | btrfs_dev_replace_unlock(dev_replace); | |
791 | ||
792 | WARN_ON(atomic_xchg( | |
793 | &fs_info->mutually_exclusive_operation_running, 1)); | |
794 | task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); | |
795 | return PTR_ERR_OR_ZERO(task); | |
796 | } | |
797 | ||
798 | static int btrfs_dev_replace_kthread(void *data) | |
799 | { | |
800 | struct btrfs_fs_info *fs_info = data; | |
801 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
802 | struct btrfs_ioctl_dev_replace_args *status_args; | |
803 | u64 progress; | |
804 | ||
805 | status_args = kzalloc(sizeof(*status_args), GFP_NOFS); | |
806 | if (status_args) { | |
807 | btrfs_dev_replace_status(fs_info, status_args); | |
808 | progress = status_args->status.progress_1000; | |
809 | kfree(status_args); | |
810 | progress = div_u64(progress, 10); | |
811 | btrfs_info_in_rcu(fs_info, | |
812 | "continuing dev_replace from %s (devid %llu) to %s @%u%%", | |
813 | dev_replace->srcdev->missing ? "<missing disk>" : | |
814 | rcu_str_deref(dev_replace->srcdev->name), | |
815 | dev_replace->srcdev->devid, | |
816 | dev_replace->tgtdev ? | |
817 | rcu_str_deref(dev_replace->tgtdev->name) : | |
818 | "<missing target disk>", | |
819 | (unsigned int)progress); | |
820 | } | |
821 | btrfs_dev_replace_continue_on_mount(fs_info); | |
822 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
823 | ||
824 | return 0; | |
825 | } | |
826 | ||
827 | static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info) | |
828 | { | |
829 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
830 | int ret; | |
831 | ||
832 | ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, | |
833 | dev_replace->committed_cursor_left, | |
834 | btrfs_device_get_total_bytes(dev_replace->srcdev), | |
835 | &dev_replace->scrub_progress, 0, 1); | |
836 | ret = btrfs_dev_replace_finishing(fs_info, ret); | |
837 | WARN_ON(ret); | |
838 | return 0; | |
839 | } | |
840 | ||
841 | int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) | |
842 | { | |
843 | if (!dev_replace->is_valid) | |
844 | return 0; | |
845 | ||
846 | switch (dev_replace->replace_state) { | |
847 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
848 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
849 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
850 | return 0; | |
851 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
852 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
853 | /* | |
854 | * return true even if tgtdev is missing (this is | |
855 | * something that can happen if the dev_replace | |
856 | * procedure is suspended by an umount and then | |
857 | * the tgtdev is missing (or "btrfs dev scan") was | |
858 | * not called and the the filesystem is remounted | |
859 | * in degraded state. This does not stop the | |
860 | * dev_replace procedure. It needs to be canceled | |
861 | * manually if the cancelation is wanted. | |
862 | */ | |
863 | break; | |
864 | } | |
865 | return 1; | |
866 | } | |
867 | ||
868 | void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace) | |
869 | { | |
870 | /* the beginning is just an optimization for the typical case */ | |
871 | if (atomic_read(&dev_replace->nesting_level) == 0) { | |
872 | acquire_lock: | |
873 | /* this is not a nested case where the same thread | |
874 | * is trying to acqurire the same lock twice */ | |
875 | mutex_lock(&dev_replace->lock); | |
876 | mutex_lock(&dev_replace->lock_management_lock); | |
877 | dev_replace->lock_owner = current->pid; | |
878 | atomic_inc(&dev_replace->nesting_level); | |
879 | mutex_unlock(&dev_replace->lock_management_lock); | |
880 | return; | |
881 | } | |
882 | ||
883 | mutex_lock(&dev_replace->lock_management_lock); | |
884 | if (atomic_read(&dev_replace->nesting_level) > 0 && | |
885 | dev_replace->lock_owner == current->pid) { | |
886 | WARN_ON(!mutex_is_locked(&dev_replace->lock)); | |
887 | atomic_inc(&dev_replace->nesting_level); | |
888 | mutex_unlock(&dev_replace->lock_management_lock); | |
889 | return; | |
890 | } | |
891 | ||
892 | mutex_unlock(&dev_replace->lock_management_lock); | |
893 | goto acquire_lock; | |
894 | } | |
895 | ||
896 | void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace) | |
897 | { | |
898 | WARN_ON(!mutex_is_locked(&dev_replace->lock)); | |
899 | mutex_lock(&dev_replace->lock_management_lock); | |
900 | WARN_ON(atomic_read(&dev_replace->nesting_level) < 1); | |
901 | WARN_ON(dev_replace->lock_owner != current->pid); | |
902 | atomic_dec(&dev_replace->nesting_level); | |
903 | if (atomic_read(&dev_replace->nesting_level) == 0) { | |
904 | dev_replace->lock_owner = 0; | |
905 | mutex_unlock(&dev_replace->lock_management_lock); | |
906 | mutex_unlock(&dev_replace->lock); | |
907 | } else { | |
908 | mutex_unlock(&dev_replace->lock_management_lock); | |
909 | } | |
910 | } | |
911 | ||
912 | void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) | |
913 | { | |
914 | percpu_counter_inc(&fs_info->bio_counter); | |
915 | } | |
916 | ||
917 | void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount) | |
918 | { | |
919 | percpu_counter_sub(&fs_info->bio_counter, amount); | |
920 | ||
921 | if (waitqueue_active(&fs_info->replace_wait)) | |
922 | wake_up(&fs_info->replace_wait); | |
923 | } | |
924 | ||
925 | void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) | |
926 | { | |
927 | while (1) { | |
928 | percpu_counter_inc(&fs_info->bio_counter); | |
929 | if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING, | |
930 | &fs_info->fs_state))) | |
931 | break; | |
932 | ||
933 | btrfs_bio_counter_dec(fs_info); | |
934 | wait_event(fs_info->replace_wait, | |
935 | !test_bit(BTRFS_FS_STATE_DEV_REPLACING, | |
936 | &fs_info->fs_state)); | |
937 | } | |
938 | } |