Commit | Line | Data |
---|---|---|
e93c89c1 SB |
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> | |
e93c89c1 SB |
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 | ||
e93c89c1 SB |
40 | static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, |
41 | int scrub_ret); | |
42 | static void btrfs_dev_replace_update_device_in_mapping_tree( | |
43 | struct btrfs_fs_info *fs_info, | |
44 | struct btrfs_device *srcdev, | |
45 | struct btrfs_device *tgtdev); | |
46 | static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, | |
47 | char *srcdev_name, | |
48 | struct btrfs_device **device); | |
49 | static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info); | |
50 | static int btrfs_dev_replace_kthread(void *data); | |
51 | static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info); | |
52 | ||
53 | ||
54 | int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info) | |
55 | { | |
56 | struct btrfs_key key; | |
57 | struct btrfs_root *dev_root = fs_info->dev_root; | |
58 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
59 | struct extent_buffer *eb; | |
60 | int slot; | |
61 | int ret = 0; | |
62 | struct btrfs_path *path = NULL; | |
63 | int item_size; | |
64 | struct btrfs_dev_replace_item *ptr; | |
65 | u64 src_devid; | |
66 | ||
67 | path = btrfs_alloc_path(); | |
68 | if (!path) { | |
69 | ret = -ENOMEM; | |
70 | goto out; | |
71 | } | |
72 | ||
73 | key.objectid = 0; | |
74 | key.type = BTRFS_DEV_REPLACE_KEY; | |
75 | key.offset = 0; | |
76 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
77 | if (ret) { | |
78 | no_valid_dev_replace_entry_found: | |
79 | ret = 0; | |
80 | dev_replace->replace_state = | |
81 | BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED; | |
82 | dev_replace->cont_reading_from_srcdev_mode = | |
83 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS; | |
84 | dev_replace->replace_state = 0; | |
85 | dev_replace->time_started = 0; | |
86 | dev_replace->time_stopped = 0; | |
87 | atomic64_set(&dev_replace->num_write_errors, 0); | |
88 | atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0); | |
89 | dev_replace->cursor_left = 0; | |
90 | dev_replace->committed_cursor_left = 0; | |
91 | dev_replace->cursor_left_last_write_of_item = 0; | |
92 | dev_replace->cursor_right = 0; | |
93 | dev_replace->srcdev = NULL; | |
94 | dev_replace->tgtdev = NULL; | |
95 | dev_replace->is_valid = 0; | |
96 | dev_replace->item_needs_writeback = 0; | |
97 | goto out; | |
98 | } | |
99 | slot = path->slots[0]; | |
100 | eb = path->nodes[0]; | |
101 | item_size = btrfs_item_size_nr(eb, slot); | |
102 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item); | |
103 | ||
104 | if (item_size != sizeof(struct btrfs_dev_replace_item)) { | |
efe120a0 FH |
105 | btrfs_warn(fs_info, |
106 | "dev_replace entry found has unexpected size, ignore entry"); | |
e93c89c1 SB |
107 | goto no_valid_dev_replace_entry_found; |
108 | } | |
109 | ||
110 | src_devid = btrfs_dev_replace_src_devid(eb, ptr); | |
111 | dev_replace->cont_reading_from_srcdev_mode = | |
112 | btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr); | |
113 | dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr); | |
114 | dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr); | |
115 | dev_replace->time_stopped = | |
116 | btrfs_dev_replace_time_stopped(eb, ptr); | |
117 | atomic64_set(&dev_replace->num_write_errors, | |
118 | btrfs_dev_replace_num_write_errors(eb, ptr)); | |
119 | atomic64_set(&dev_replace->num_uncorrectable_read_errors, | |
120 | btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr)); | |
121 | dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr); | |
122 | dev_replace->committed_cursor_left = dev_replace->cursor_left; | |
123 | dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left; | |
124 | dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr); | |
125 | dev_replace->is_valid = 1; | |
126 | ||
127 | dev_replace->item_needs_writeback = 0; | |
128 | switch (dev_replace->replace_state) { | |
129 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
130 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
131 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
132 | dev_replace->srcdev = NULL; | |
133 | dev_replace->tgtdev = NULL; | |
134 | break; | |
135 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
136 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
137 | dev_replace->srcdev = btrfs_find_device(fs_info, src_devid, | |
138 | NULL, NULL); | |
139 | dev_replace->tgtdev = btrfs_find_device(fs_info, | |
140 | BTRFS_DEV_REPLACE_DEVID, | |
141 | NULL, NULL); | |
142 | /* | |
143 | * allow 'btrfs dev replace_cancel' if src/tgt device is | |
144 | * missing | |
145 | */ | |
146 | if (!dev_replace->srcdev && | |
147 | !btrfs_test_opt(dev_root, DEGRADED)) { | |
148 | ret = -EIO; | |
efe120a0 FH |
149 | btrfs_warn(fs_info, |
150 | "cannot mount because device replace operation is ongoing and"); | |
151 | btrfs_warn(fs_info, | |
152 | "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?", | |
153 | src_devid); | |
e93c89c1 SB |
154 | } |
155 | if (!dev_replace->tgtdev && | |
156 | !btrfs_test_opt(dev_root, DEGRADED)) { | |
157 | ret = -EIO; | |
efe120a0 FH |
158 | btrfs_warn(fs_info, |
159 | "cannot mount because device replace operation is ongoing and"); | |
160 | btrfs_warn(fs_info, | |
161 | "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?", | |
6e71c47a | 162 | BTRFS_DEV_REPLACE_DEVID); |
e93c89c1 SB |
163 | } |
164 | if (dev_replace->tgtdev) { | |
165 | if (dev_replace->srcdev) { | |
166 | dev_replace->tgtdev->total_bytes = | |
167 | dev_replace->srcdev->total_bytes; | |
168 | dev_replace->tgtdev->disk_total_bytes = | |
169 | dev_replace->srcdev->disk_total_bytes; | |
170 | dev_replace->tgtdev->bytes_used = | |
171 | dev_replace->srcdev->bytes_used; | |
172 | } | |
173 | dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1; | |
174 | btrfs_init_dev_replace_tgtdev_for_resume(fs_info, | |
175 | dev_replace->tgtdev); | |
176 | } | |
177 | break; | |
178 | } | |
179 | ||
180 | out: | |
181 | if (path) | |
182 | btrfs_free_path(path); | |
183 | return ret; | |
184 | } | |
185 | ||
186 | /* | |
187 | * called from commit_transaction. Writes changed device replace state to | |
188 | * disk. | |
189 | */ | |
190 | int btrfs_run_dev_replace(struct btrfs_trans_handle *trans, | |
191 | struct btrfs_fs_info *fs_info) | |
192 | { | |
193 | int ret; | |
194 | struct btrfs_root *dev_root = fs_info->dev_root; | |
195 | struct btrfs_path *path; | |
196 | struct btrfs_key key; | |
197 | struct extent_buffer *eb; | |
198 | struct btrfs_dev_replace_item *ptr; | |
199 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
200 | ||
201 | btrfs_dev_replace_lock(dev_replace); | |
202 | if (!dev_replace->is_valid || | |
203 | !dev_replace->item_needs_writeback) { | |
204 | btrfs_dev_replace_unlock(dev_replace); | |
205 | return 0; | |
206 | } | |
207 | btrfs_dev_replace_unlock(dev_replace); | |
208 | ||
209 | key.objectid = 0; | |
210 | key.type = BTRFS_DEV_REPLACE_KEY; | |
211 | key.offset = 0; | |
212 | ||
213 | path = btrfs_alloc_path(); | |
214 | if (!path) { | |
215 | ret = -ENOMEM; | |
216 | goto out; | |
217 | } | |
218 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
219 | if (ret < 0) { | |
efe120a0 | 220 | btrfs_warn(fs_info, "error %d while searching for dev_replace item!", |
e93c89c1 SB |
221 | ret); |
222 | goto out; | |
223 | } | |
224 | ||
225 | if (ret == 0 && | |
226 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
227 | /* | |
228 | * need to delete old one and insert a new one. | |
229 | * Since no attempt is made to recover any old state, if the | |
230 | * dev_replace state is 'running', the data on the target | |
231 | * drive is lost. | |
232 | * It would be possible to recover the state: just make sure | |
233 | * that the beginning of the item is never changed and always | |
234 | * contains all the essential information. Then read this | |
235 | * minimal set of information and use it as a base for the | |
236 | * new state. | |
237 | */ | |
238 | ret = btrfs_del_item(trans, dev_root, path); | |
239 | if (ret != 0) { | |
efe120a0 | 240 | btrfs_warn(fs_info, "delete too small dev_replace item failed %d!", |
e93c89c1 SB |
241 | ret); |
242 | goto out; | |
243 | } | |
244 | ret = 1; | |
245 | } | |
246 | ||
247 | if (ret == 1) { | |
248 | /* need to insert a new item */ | |
249 | btrfs_release_path(path); | |
250 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
251 | &key, sizeof(*ptr)); | |
252 | if (ret < 0) { | |
efe120a0 | 253 | btrfs_warn(fs_info, "insert dev_replace item failed %d!", |
e93c89c1 SB |
254 | ret); |
255 | goto out; | |
256 | } | |
257 | } | |
258 | ||
259 | eb = path->nodes[0]; | |
260 | ptr = btrfs_item_ptr(eb, path->slots[0], | |
261 | struct btrfs_dev_replace_item); | |
262 | ||
263 | btrfs_dev_replace_lock(dev_replace); | |
264 | if (dev_replace->srcdev) | |
265 | btrfs_set_dev_replace_src_devid(eb, ptr, | |
266 | dev_replace->srcdev->devid); | |
267 | else | |
268 | btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1); | |
269 | btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr, | |
270 | dev_replace->cont_reading_from_srcdev_mode); | |
271 | btrfs_set_dev_replace_replace_state(eb, ptr, | |
272 | dev_replace->replace_state); | |
273 | btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started); | |
274 | btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped); | |
275 | btrfs_set_dev_replace_num_write_errors(eb, ptr, | |
276 | atomic64_read(&dev_replace->num_write_errors)); | |
277 | btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr, | |
278 | atomic64_read(&dev_replace->num_uncorrectable_read_errors)); | |
279 | dev_replace->cursor_left_last_write_of_item = | |
280 | dev_replace->cursor_left; | |
281 | btrfs_set_dev_replace_cursor_left(eb, ptr, | |
282 | dev_replace->cursor_left_last_write_of_item); | |
283 | btrfs_set_dev_replace_cursor_right(eb, ptr, | |
284 | dev_replace->cursor_right); | |
285 | dev_replace->item_needs_writeback = 0; | |
286 | btrfs_dev_replace_unlock(dev_replace); | |
287 | ||
288 | btrfs_mark_buffer_dirty(eb); | |
289 | ||
290 | out: | |
291 | btrfs_free_path(path); | |
292 | ||
293 | return ret; | |
294 | } | |
295 | ||
296 | void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info) | |
297 | { | |
298 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
299 | ||
300 | dev_replace->committed_cursor_left = | |
301 | dev_replace->cursor_left_last_write_of_item; | |
302 | } | |
303 | ||
e93c89c1 SB |
304 | int btrfs_dev_replace_start(struct btrfs_root *root, |
305 | struct btrfs_ioctl_dev_replace_args *args) | |
306 | { | |
307 | struct btrfs_trans_handle *trans; | |
308 | struct btrfs_fs_info *fs_info = root->fs_info; | |
309 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
310 | int ret; | |
311 | struct btrfs_device *tgt_device = NULL; | |
312 | struct btrfs_device *src_device = NULL; | |
313 | ||
49688107 | 314 | if (btrfs_fs_incompat(fs_info, RAID56)) { |
efe120a0 | 315 | btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6"); |
49688107 SB |
316 | return -EINVAL; |
317 | } | |
318 | ||
e93c89c1 SB |
319 | switch (args->start.cont_reading_from_srcdev_mode) { |
320 | case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS: | |
321 | case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID: | |
322 | break; | |
323 | default: | |
324 | return -EINVAL; | |
325 | } | |
326 | ||
327 | if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') || | |
328 | args->start.tgtdev_name[0] == '\0') | |
329 | return -EINVAL; | |
330 | ||
331 | mutex_lock(&fs_info->volume_mutex); | |
332 | ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name, | |
333 | &tgt_device); | |
334 | if (ret) { | |
efe120a0 | 335 | btrfs_err(fs_info, "target device %s is invalid!", |
e93c89c1 SB |
336 | args->start.tgtdev_name); |
337 | mutex_unlock(&fs_info->volume_mutex); | |
338 | return -EINVAL; | |
339 | } | |
340 | ||
341 | ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid, | |
342 | args->start.srcdev_name, | |
343 | &src_device); | |
344 | mutex_unlock(&fs_info->volume_mutex); | |
345 | if (ret) { | |
346 | ret = -EINVAL; | |
347 | goto leave_no_lock; | |
348 | } | |
349 | ||
350 | if (tgt_device->total_bytes < src_device->total_bytes) { | |
efe120a0 | 351 | btrfs_err(fs_info, "target device is smaller than source device!"); |
e93c89c1 SB |
352 | ret = -EINVAL; |
353 | goto leave_no_lock; | |
354 | } | |
355 | ||
356 | btrfs_dev_replace_lock(dev_replace); | |
357 | switch (dev_replace->replace_state) { | |
358 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
359 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
360 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
361 | break; | |
362 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
363 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
364 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED; | |
365 | goto leave; | |
366 | } | |
367 | ||
368 | dev_replace->cont_reading_from_srcdev_mode = | |
369 | args->start.cont_reading_from_srcdev_mode; | |
370 | WARN_ON(!src_device); | |
371 | dev_replace->srcdev = src_device; | |
372 | WARN_ON(!tgt_device); | |
373 | dev_replace->tgtdev = tgt_device; | |
374 | ||
375 | printk_in_rcu(KERN_INFO | |
efe120a0 | 376 | "BTRFS: dev_replace from %s (devid %llu) to %s started\n", |
e93c89c1 SB |
377 | src_device->missing ? "<missing disk>" : |
378 | rcu_str_deref(src_device->name), | |
379 | src_device->devid, | |
380 | rcu_str_deref(tgt_device->name)); | |
381 | ||
382 | tgt_device->total_bytes = src_device->total_bytes; | |
383 | tgt_device->disk_total_bytes = src_device->disk_total_bytes; | |
384 | tgt_device->bytes_used = src_device->bytes_used; | |
385 | ||
386 | /* | |
387 | * from now on, the writes to the srcdev are all duplicated to | |
388 | * go to the tgtdev as well (refer to btrfs_map_block()). | |
389 | */ | |
390 | dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; | |
4546bcae | 391 | dev_replace->time_started = get_seconds(); |
e93c89c1 SB |
392 | dev_replace->cursor_left = 0; |
393 | dev_replace->committed_cursor_left = 0; | |
394 | dev_replace->cursor_left_last_write_of_item = 0; | |
395 | dev_replace->cursor_right = 0; | |
396 | dev_replace->is_valid = 1; | |
397 | dev_replace->item_needs_writeback = 1; | |
398 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
399 | btrfs_dev_replace_unlock(dev_replace); | |
400 | ||
b0244199 | 401 | btrfs_wait_ordered_roots(root->fs_info, -1); |
e93c89c1 SB |
402 | |
403 | /* force writing the updated state information to disk */ | |
404 | trans = btrfs_start_transaction(root, 0); | |
405 | if (IS_ERR(trans)) { | |
406 | ret = PTR_ERR(trans); | |
407 | btrfs_dev_replace_lock(dev_replace); | |
408 | goto leave; | |
409 | } | |
410 | ||
411 | ret = btrfs_commit_transaction(trans, root); | |
412 | WARN_ON(ret); | |
413 | ||
414 | /* the disk copy procedure reuses the scrub code */ | |
415 | ret = btrfs_scrub_dev(fs_info, src_device->devid, 0, | |
416 | src_device->total_bytes, | |
417 | &dev_replace->scrub_progress, 0, 1); | |
418 | ||
419 | ret = btrfs_dev_replace_finishing(root->fs_info, ret); | |
420 | WARN_ON(ret); | |
421 | ||
422 | return 0; | |
423 | ||
424 | leave: | |
425 | dev_replace->srcdev = NULL; | |
426 | dev_replace->tgtdev = NULL; | |
427 | btrfs_dev_replace_unlock(dev_replace); | |
428 | leave_no_lock: | |
429 | if (tgt_device) | |
430 | btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); | |
431 | return ret; | |
432 | } | |
433 | ||
c404e0dc MX |
434 | /* |
435 | * blocked until all flighting bios are finished. | |
436 | */ | |
437 | static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info) | |
438 | { | |
439 | s64 writers; | |
440 | DEFINE_WAIT(wait); | |
441 | ||
442 | set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); | |
443 | do { | |
444 | prepare_to_wait(&fs_info->replace_wait, &wait, | |
445 | TASK_UNINTERRUPTIBLE); | |
446 | writers = percpu_counter_sum(&fs_info->bio_counter); | |
447 | if (writers) | |
448 | schedule(); | |
449 | finish_wait(&fs_info->replace_wait, &wait); | |
450 | } while (writers); | |
451 | } | |
452 | ||
453 | /* | |
454 | * we have removed target device, it is safe to allow new bios request. | |
455 | */ | |
456 | static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info) | |
457 | { | |
458 | clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state); | |
459 | if (waitqueue_active(&fs_info->replace_wait)) | |
460 | wake_up(&fs_info->replace_wait); | |
461 | } | |
462 | ||
e93c89c1 SB |
463 | static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info, |
464 | int scrub_ret) | |
465 | { | |
466 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
467 | struct btrfs_device *tgt_device; | |
468 | struct btrfs_device *src_device; | |
469 | struct btrfs_root *root = fs_info->tree_root; | |
470 | u8 uuid_tmp[BTRFS_UUID_SIZE]; | |
471 | struct btrfs_trans_handle *trans; | |
472 | int ret = 0; | |
473 | ||
474 | /* don't allow cancel or unmount to disturb the finishing procedure */ | |
475 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); | |
476 | ||
477 | btrfs_dev_replace_lock(dev_replace); | |
478 | /* was the operation canceled, or is it finished? */ | |
479 | if (dev_replace->replace_state != | |
480 | BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) { | |
481 | btrfs_dev_replace_unlock(dev_replace); | |
482 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
483 | return 0; | |
484 | } | |
485 | ||
486 | tgt_device = dev_replace->tgtdev; | |
487 | src_device = dev_replace->srcdev; | |
488 | btrfs_dev_replace_unlock(dev_replace); | |
489 | ||
e93c89c1 SB |
490 | /* |
491 | * flush all outstanding I/O and inode extent mappings before the | |
492 | * copy operation is declared as being finished | |
493 | */ | |
6c255e67 | 494 | ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1); |
3edb2a68 MX |
495 | if (ret) { |
496 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
497 | return ret; | |
498 | } | |
b0244199 | 499 | btrfs_wait_ordered_roots(root->fs_info, -1); |
e93c89c1 SB |
500 | |
501 | trans = btrfs_start_transaction(root, 0); | |
502 | if (IS_ERR(trans)) { | |
503 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
504 | return PTR_ERR(trans); | |
505 | } | |
506 | ret = btrfs_commit_transaction(trans, root); | |
507 | WARN_ON(ret); | |
508 | ||
509 | /* keep away write_all_supers() during the finishing procedure */ | |
391cd9df | 510 | mutex_lock(&root->fs_info->chunk_mutex); |
e93c89c1 SB |
511 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
512 | btrfs_dev_replace_lock(dev_replace); | |
513 | dev_replace->replace_state = | |
514 | scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED | |
515 | : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED; | |
516 | dev_replace->tgtdev = NULL; | |
517 | dev_replace->srcdev = NULL; | |
4546bcae | 518 | dev_replace->time_stopped = get_seconds(); |
e93c89c1 SB |
519 | dev_replace->item_needs_writeback = 1; |
520 | ||
c404e0dc MX |
521 | /* replace old device with new one in mapping tree */ |
522 | if (!scrub_ret) { | |
523 | btrfs_dev_replace_update_device_in_mapping_tree(fs_info, | |
524 | src_device, | |
525 | tgt_device); | |
526 | } else { | |
e93c89c1 | 527 | printk_in_rcu(KERN_ERR |
efe120a0 | 528 | "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n", |
e93c89c1 SB |
529 | src_device->missing ? "<missing disk>" : |
530 | rcu_str_deref(src_device->name), | |
531 | src_device->devid, | |
532 | rcu_str_deref(tgt_device->name), scrub_ret); | |
533 | btrfs_dev_replace_unlock(dev_replace); | |
534 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
391cd9df | 535 | mutex_unlock(&root->fs_info->chunk_mutex); |
e93c89c1 SB |
536 | if (tgt_device) |
537 | btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); | |
538 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
539 | ||
540 | return 0; | |
541 | } | |
542 | ||
543 | printk_in_rcu(KERN_INFO | |
efe120a0 | 544 | "BTRFS: dev_replace from %s (devid %llu) to %s) finished\n", |
e93c89c1 SB |
545 | src_device->missing ? "<missing disk>" : |
546 | rcu_str_deref(src_device->name), | |
547 | src_device->devid, | |
548 | rcu_str_deref(tgt_device->name)); | |
549 | tgt_device->is_tgtdev_for_dev_replace = 0; | |
550 | tgt_device->devid = src_device->devid; | |
551 | src_device->devid = BTRFS_DEV_REPLACE_DEVID; | |
552 | tgt_device->bytes_used = src_device->bytes_used; | |
553 | memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp)); | |
554 | memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid)); | |
555 | memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid)); | |
556 | tgt_device->total_bytes = src_device->total_bytes; | |
557 | tgt_device->disk_total_bytes = src_device->disk_total_bytes; | |
558 | tgt_device->bytes_used = src_device->bytes_used; | |
559 | if (fs_info->sb->s_bdev == src_device->bdev) | |
560 | fs_info->sb->s_bdev = tgt_device->bdev; | |
561 | if (fs_info->fs_devices->latest_bdev == src_device->bdev) | |
562 | fs_info->fs_devices->latest_bdev = tgt_device->bdev; | |
563 | list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list); | |
564 | ||
c404e0dc MX |
565 | btrfs_rm_dev_replace_blocked(fs_info); |
566 | ||
e93c89c1 | 567 | btrfs_rm_dev_replace_srcdev(fs_info, src_device); |
1357272f | 568 | |
c404e0dc MX |
569 | btrfs_rm_dev_replace_unblocked(fs_info); |
570 | ||
e93c89c1 SB |
571 | /* |
572 | * this is again a consistent state where no dev_replace procedure | |
573 | * is running, the target device is part of the filesystem, the | |
574 | * source device is not part of the filesystem anymore and its 1st | |
575 | * superblock is scratched out so that it is no longer marked to | |
576 | * belong to this filesystem. | |
577 | */ | |
578 | btrfs_dev_replace_unlock(dev_replace); | |
579 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
391cd9df | 580 | mutex_unlock(&root->fs_info->chunk_mutex); |
e93c89c1 SB |
581 | |
582 | /* write back the superblocks */ | |
583 | trans = btrfs_start_transaction(root, 0); | |
584 | if (!IS_ERR(trans)) | |
585 | btrfs_commit_transaction(trans, root); | |
586 | ||
587 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
588 | ||
589 | return 0; | |
590 | } | |
591 | ||
592 | static void btrfs_dev_replace_update_device_in_mapping_tree( | |
593 | struct btrfs_fs_info *fs_info, | |
594 | struct btrfs_device *srcdev, | |
595 | struct btrfs_device *tgtdev) | |
596 | { | |
597 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
598 | struct extent_map *em; | |
599 | struct map_lookup *map; | |
600 | u64 start = 0; | |
601 | int i; | |
602 | ||
603 | write_lock(&em_tree->lock); | |
604 | do { | |
605 | em = lookup_extent_mapping(em_tree, start, (u64)-1); | |
606 | if (!em) | |
607 | break; | |
608 | map = (struct map_lookup *)em->bdev; | |
609 | for (i = 0; i < map->num_stripes; i++) | |
610 | if (srcdev == map->stripes[i].dev) | |
611 | map->stripes[i].dev = tgtdev; | |
612 | start = em->start + em->len; | |
613 | free_extent_map(em); | |
614 | } while (start); | |
615 | write_unlock(&em_tree->lock); | |
616 | } | |
617 | ||
618 | static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid, | |
619 | char *srcdev_name, | |
620 | struct btrfs_device **device) | |
621 | { | |
622 | int ret; | |
623 | ||
624 | if (srcdevid) { | |
625 | ret = 0; | |
626 | *device = btrfs_find_device(root->fs_info, srcdevid, NULL, | |
627 | NULL); | |
628 | if (!*device) | |
629 | ret = -ENOENT; | |
630 | } else { | |
631 | ret = btrfs_find_device_missing_or_by_path(root, srcdev_name, | |
632 | device); | |
633 | } | |
634 | return ret; | |
635 | } | |
636 | ||
637 | void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info, | |
638 | struct btrfs_ioctl_dev_replace_args *args) | |
639 | { | |
640 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
641 | ||
642 | btrfs_dev_replace_lock(dev_replace); | |
643 | /* even if !dev_replace_is_valid, the values are good enough for | |
644 | * the replace_status ioctl */ | |
645 | args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
646 | args->status.replace_state = dev_replace->replace_state; | |
647 | args->status.time_started = dev_replace->time_started; | |
648 | args->status.time_stopped = dev_replace->time_stopped; | |
649 | args->status.num_write_errors = | |
650 | atomic64_read(&dev_replace->num_write_errors); | |
651 | args->status.num_uncorrectable_read_errors = | |
652 | atomic64_read(&dev_replace->num_uncorrectable_read_errors); | |
653 | switch (dev_replace->replace_state) { | |
654 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
655 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
656 | args->status.progress_1000 = 0; | |
657 | break; | |
658 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
659 | args->status.progress_1000 = 1000; | |
660 | break; | |
661 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
662 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
663 | args->status.progress_1000 = div64_u64(dev_replace->cursor_left, | |
664 | div64_u64(dev_replace->srcdev->total_bytes, 1000)); | |
665 | break; | |
666 | } | |
667 | btrfs_dev_replace_unlock(dev_replace); | |
668 | } | |
669 | ||
670 | int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info, | |
671 | struct btrfs_ioctl_dev_replace_args *args) | |
672 | { | |
673 | args->result = __btrfs_dev_replace_cancel(fs_info); | |
674 | return 0; | |
675 | } | |
676 | ||
677 | static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info) | |
678 | { | |
679 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
680 | struct btrfs_device *tgt_device = NULL; | |
681 | struct btrfs_trans_handle *trans; | |
682 | struct btrfs_root *root = fs_info->tree_root; | |
683 | u64 result; | |
684 | int ret; | |
685 | ||
e649e587 ID |
686 | if (fs_info->sb->s_flags & MS_RDONLY) |
687 | return -EROFS; | |
688 | ||
e93c89c1 SB |
689 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); |
690 | btrfs_dev_replace_lock(dev_replace); | |
691 | switch (dev_replace->replace_state) { | |
692 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
693 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
694 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
695 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED; | |
696 | btrfs_dev_replace_unlock(dev_replace); | |
697 | goto leave; | |
698 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
699 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
700 | result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR; | |
701 | tgt_device = dev_replace->tgtdev; | |
702 | dev_replace->tgtdev = NULL; | |
703 | dev_replace->srcdev = NULL; | |
704 | break; | |
705 | } | |
706 | dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED; | |
4546bcae | 707 | dev_replace->time_stopped = get_seconds(); |
e93c89c1 SB |
708 | dev_replace->item_needs_writeback = 1; |
709 | btrfs_dev_replace_unlock(dev_replace); | |
710 | btrfs_scrub_cancel(fs_info); | |
711 | ||
712 | trans = btrfs_start_transaction(root, 0); | |
713 | if (IS_ERR(trans)) { | |
714 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
715 | return PTR_ERR(trans); | |
716 | } | |
717 | ret = btrfs_commit_transaction(trans, root); | |
718 | WARN_ON(ret); | |
719 | if (tgt_device) | |
720 | btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device); | |
721 | ||
722 | leave: | |
723 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
724 | return result; | |
725 | } | |
726 | ||
727 | void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info) | |
728 | { | |
729 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
730 | ||
731 | mutex_lock(&dev_replace->lock_finishing_cancel_unmount); | |
732 | btrfs_dev_replace_lock(dev_replace); | |
733 | switch (dev_replace->replace_state) { | |
734 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
735 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
736 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
737 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
738 | break; | |
739 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
740 | dev_replace->replace_state = | |
741 | BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED; | |
4546bcae | 742 | dev_replace->time_stopped = get_seconds(); |
e93c89c1 | 743 | dev_replace->item_needs_writeback = 1; |
efe120a0 | 744 | btrfs_info(fs_info, "suspending dev_replace for unmount"); |
e93c89c1 SB |
745 | break; |
746 | } | |
747 | ||
748 | btrfs_dev_replace_unlock(dev_replace); | |
749 | mutex_unlock(&dev_replace->lock_finishing_cancel_unmount); | |
750 | } | |
751 | ||
752 | /* resume dev_replace procedure that was interrupted by unmount */ | |
753 | int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info) | |
754 | { | |
755 | struct task_struct *task; | |
756 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
757 | ||
758 | btrfs_dev_replace_lock(dev_replace); | |
759 | switch (dev_replace->replace_state) { | |
760 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
761 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
762 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
763 | btrfs_dev_replace_unlock(dev_replace); | |
764 | return 0; | |
765 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
766 | break; | |
767 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
768 | dev_replace->replace_state = | |
769 | BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED; | |
770 | break; | |
771 | } | |
772 | if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) { | |
efe120a0 FH |
773 | btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing"); |
774 | btrfs_info(fs_info, | |
775 | "you may cancel the operation after 'mount -o degraded'"); | |
e93c89c1 SB |
776 | btrfs_dev_replace_unlock(dev_replace); |
777 | return 0; | |
778 | } | |
779 | btrfs_dev_replace_unlock(dev_replace); | |
780 | ||
781 | WARN_ON(atomic_xchg( | |
782 | &fs_info->mutually_exclusive_operation_running, 1)); | |
783 | task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl"); | |
8c6ffba0 | 784 | return PTR_ERR_OR_ZERO(task); |
e93c89c1 SB |
785 | } |
786 | ||
787 | static int btrfs_dev_replace_kthread(void *data) | |
788 | { | |
789 | struct btrfs_fs_info *fs_info = data; | |
790 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
791 | struct btrfs_ioctl_dev_replace_args *status_args; | |
792 | u64 progress; | |
793 | ||
794 | status_args = kzalloc(sizeof(*status_args), GFP_NOFS); | |
795 | if (status_args) { | |
796 | btrfs_dev_replace_status(fs_info, status_args); | |
797 | progress = status_args->status.progress_1000; | |
798 | kfree(status_args); | |
799 | do_div(progress, 10); | |
800 | printk_in_rcu(KERN_INFO | |
efe120a0 FH |
801 | "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n", |
802 | dev_replace->srcdev->missing ? "<missing disk>" : | |
803 | rcu_str_deref(dev_replace->srcdev->name), | |
804 | dev_replace->srcdev->devid, | |
805 | dev_replace->tgtdev ? | |
806 | rcu_str_deref(dev_replace->tgtdev->name) : | |
807 | "<missing target disk>", | |
808 | (unsigned int)progress); | |
e93c89c1 SB |
809 | } |
810 | btrfs_dev_replace_continue_on_mount(fs_info); | |
811 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
812 | ||
813 | return 0; | |
814 | } | |
815 | ||
816 | static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info) | |
817 | { | |
818 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
819 | int ret; | |
820 | ||
821 | ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid, | |
822 | dev_replace->committed_cursor_left, | |
823 | dev_replace->srcdev->total_bytes, | |
824 | &dev_replace->scrub_progress, 0, 1); | |
825 | ret = btrfs_dev_replace_finishing(fs_info, ret); | |
826 | WARN_ON(ret); | |
827 | return 0; | |
828 | } | |
829 | ||
830 | int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace) | |
831 | { | |
832 | if (!dev_replace->is_valid) | |
833 | return 0; | |
834 | ||
835 | switch (dev_replace->replace_state) { | |
836 | case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED: | |
837 | case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED: | |
838 | case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED: | |
839 | return 0; | |
840 | case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED: | |
841 | case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED: | |
842 | /* | |
843 | * return true even if tgtdev is missing (this is | |
844 | * something that can happen if the dev_replace | |
845 | * procedure is suspended by an umount and then | |
846 | * the tgtdev is missing (or "btrfs dev scan") was | |
847 | * not called and the the filesystem is remounted | |
848 | * in degraded state. This does not stop the | |
849 | * dev_replace procedure. It needs to be canceled | |
850 | * manually if the cancelation is wanted. | |
851 | */ | |
852 | break; | |
853 | } | |
854 | return 1; | |
855 | } | |
856 | ||
857 | void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace) | |
858 | { | |
859 | /* the beginning is just an optimization for the typical case */ | |
860 | if (atomic_read(&dev_replace->nesting_level) == 0) { | |
861 | acquire_lock: | |
862 | /* this is not a nested case where the same thread | |
863 | * is trying to acqurire the same lock twice */ | |
864 | mutex_lock(&dev_replace->lock); | |
865 | mutex_lock(&dev_replace->lock_management_lock); | |
866 | dev_replace->lock_owner = current->pid; | |
867 | atomic_inc(&dev_replace->nesting_level); | |
868 | mutex_unlock(&dev_replace->lock_management_lock); | |
869 | return; | |
870 | } | |
871 | ||
872 | mutex_lock(&dev_replace->lock_management_lock); | |
873 | if (atomic_read(&dev_replace->nesting_level) > 0 && | |
874 | dev_replace->lock_owner == current->pid) { | |
875 | WARN_ON(!mutex_is_locked(&dev_replace->lock)); | |
876 | atomic_inc(&dev_replace->nesting_level); | |
877 | mutex_unlock(&dev_replace->lock_management_lock); | |
878 | return; | |
879 | } | |
880 | ||
881 | mutex_unlock(&dev_replace->lock_management_lock); | |
882 | goto acquire_lock; | |
883 | } | |
884 | ||
885 | void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace) | |
886 | { | |
887 | WARN_ON(!mutex_is_locked(&dev_replace->lock)); | |
888 | mutex_lock(&dev_replace->lock_management_lock); | |
889 | WARN_ON(atomic_read(&dev_replace->nesting_level) < 1); | |
890 | WARN_ON(dev_replace->lock_owner != current->pid); | |
891 | atomic_dec(&dev_replace->nesting_level); | |
892 | if (atomic_read(&dev_replace->nesting_level) == 0) { | |
893 | dev_replace->lock_owner = 0; | |
894 | mutex_unlock(&dev_replace->lock_management_lock); | |
895 | mutex_unlock(&dev_replace->lock); | |
896 | } else { | |
897 | mutex_unlock(&dev_replace->lock_management_lock); | |
898 | } | |
899 | } | |
c404e0dc MX |
900 | |
901 | void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info) | |
902 | { | |
903 | percpu_counter_inc(&fs_info->bio_counter); | |
904 | } | |
905 | ||
906 | void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info) | |
907 | { | |
908 | percpu_counter_dec(&fs_info->bio_counter); | |
909 | ||
910 | if (waitqueue_active(&fs_info->replace_wait)) | |
911 | wake_up(&fs_info->replace_wait); | |
912 | } | |
913 | ||
914 | void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info) | |
915 | { | |
916 | DEFINE_WAIT(wait); | |
917 | again: | |
918 | percpu_counter_inc(&fs_info->bio_counter); | |
919 | if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) { | |
920 | btrfs_bio_counter_dec(fs_info); | |
921 | wait_event(fs_info->replace_wait, | |
922 | !test_bit(BTRFS_FS_STATE_DEV_REPLACING, | |
923 | &fs_info->fs_state)); | |
924 | goto again; | |
925 | } | |
926 | ||
927 | } |