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
41static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
42				       int scrub_ret);
43static 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);
47static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
48					 char *srcdev_name,
49					 struct btrfs_device **device);
50static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
51static int btrfs_dev_replace_kthread(void *data);
52static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
53
54
55int 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) {
79no_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
185out:
186	if (path)
187		btrfs_free_path(path);
188	return ret;
189}
190
191/*
192 * called from commit_transaction. Writes changed device replace state to
193 * disk.
194 */
195int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
196			  struct btrfs_fs_info *fs_info)
197{
198	int ret;
199	struct btrfs_root *dev_root = fs_info->dev_root;
200	struct btrfs_path *path;
201	struct btrfs_key key;
202	struct extent_buffer *eb;
203	struct btrfs_dev_replace_item *ptr;
204	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
205
206	btrfs_dev_replace_lock(dev_replace);
207	if (!dev_replace->is_valid ||
208	    !dev_replace->item_needs_writeback) {
209		btrfs_dev_replace_unlock(dev_replace);
210		return 0;
211	}
212	btrfs_dev_replace_unlock(dev_replace);
213
214	key.objectid = 0;
215	key.type = BTRFS_DEV_REPLACE_KEY;
216	key.offset = 0;
217
218	path = btrfs_alloc_path();
219	if (!path) {
220		ret = -ENOMEM;
221		goto out;
222	}
223	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
224	if (ret < 0) {
225		btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
226			ret);
227		goto out;
228	}
229
230	if (ret == 0 &&
231	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
232		/*
233		 * need to delete old one and insert a new one.
234		 * Since no attempt is made to recover any old state, if the
235		 * dev_replace state is 'running', the data on the target
236		 * drive is lost.
237		 * It would be possible to recover the state: just make sure
238		 * that the beginning of the item is never changed and always
239		 * contains all the essential information. Then read this
240		 * minimal set of information and use it as a base for the
241		 * new state.
242		 */
243		ret = btrfs_del_item(trans, dev_root, path);
244		if (ret != 0) {
245			btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
246				ret);
247			goto out;
248		}
249		ret = 1;
250	}
251
252	if (ret == 1) {
253		/* need to insert a new item */
254		btrfs_release_path(path);
255		ret = btrfs_insert_empty_item(trans, dev_root, path,
256					      &key, sizeof(*ptr));
257		if (ret < 0) {
258			btrfs_warn(fs_info, "insert dev_replace item failed %d!",
259				ret);
260			goto out;
261		}
262	}
263
264	eb = path->nodes[0];
265	ptr = btrfs_item_ptr(eb, path->slots[0],
266			     struct btrfs_dev_replace_item);
267
268	btrfs_dev_replace_lock(dev_replace);
269	if (dev_replace->srcdev)
270		btrfs_set_dev_replace_src_devid(eb, ptr,
271			dev_replace->srcdev->devid);
272	else
273		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
274	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
275		dev_replace->cont_reading_from_srcdev_mode);
276	btrfs_set_dev_replace_replace_state(eb, ptr,
277		dev_replace->replace_state);
278	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
279	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
280	btrfs_set_dev_replace_num_write_errors(eb, ptr,
281		atomic64_read(&dev_replace->num_write_errors));
282	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
283		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
284	dev_replace->cursor_left_last_write_of_item =
285		dev_replace->cursor_left;
286	btrfs_set_dev_replace_cursor_left(eb, ptr,
287		dev_replace->cursor_left_last_write_of_item);
288	btrfs_set_dev_replace_cursor_right(eb, ptr,
289		dev_replace->cursor_right);
290	dev_replace->item_needs_writeback = 0;
291	btrfs_dev_replace_unlock(dev_replace);
292
293	btrfs_mark_buffer_dirty(eb);
294
295out:
296	btrfs_free_path(path);
297
298	return ret;
299}
300
301void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
302{
303	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
304
305	dev_replace->committed_cursor_left =
306		dev_replace->cursor_left_last_write_of_item;
307}
308
309int btrfs_dev_replace_start(struct btrfs_root *root,
310			    struct btrfs_ioctl_dev_replace_args *args)
311{
312	struct btrfs_trans_handle *trans;
313	struct btrfs_fs_info *fs_info = root->fs_info;
314	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
315	int ret;
316	struct btrfs_device *tgt_device = NULL;
317	struct btrfs_device *src_device = NULL;
318
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	/*
332	 * Here we commit the transaction to make sure commit_total_bytes
333	 * of all the devices are updated.
334	 */
335	trans = btrfs_attach_transaction(root);
336	if (!IS_ERR(trans)) {
337		ret = btrfs_commit_transaction(trans, root);
338		if (ret)
339			return ret;
340	} else if (PTR_ERR(trans) != -ENOENT) {
341		return PTR_ERR(trans);
342	}
343
344	/* the disk copy procedure reuses the scrub code */
345	mutex_lock(&fs_info->volume_mutex);
346	ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
347					    args->start.srcdev_name,
348					    &src_device);
349	if (ret) {
350		mutex_unlock(&fs_info->volume_mutex);
351		return ret;
352	}
353
354	ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
355					    src_device, &tgt_device);
356	mutex_unlock(&fs_info->volume_mutex);
357	if (ret)
358		return ret;
359
360	btrfs_dev_replace_lock(dev_replace);
361	switch (dev_replace->replace_state) {
362	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
363	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
364	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
365		break;
366	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
367	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
368		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
369		goto leave;
370	}
371
372	dev_replace->cont_reading_from_srcdev_mode =
373		args->start.cont_reading_from_srcdev_mode;
374	WARN_ON(!src_device);
375	dev_replace->srcdev = src_device;
376	WARN_ON(!tgt_device);
377	dev_replace->tgtdev = tgt_device;
378
379	printk_in_rcu(KERN_INFO
380		      "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
381		      src_device->missing ? "<missing disk>" :
382		        rcu_str_deref(src_device->name),
383		      src_device->devid,
384		      rcu_str_deref(tgt_device->name));
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;
391	dev_replace->time_started = get_seconds();
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
401	btrfs_wait_ordered_roots(root->fs_info, -1);
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			      btrfs_device_get_total_bytes(src_device),
417			      &dev_replace->scrub_progress, 0, 1);
418
419	ret = btrfs_dev_replace_finishing(root->fs_info, ret);
420	/* don't warn if EINPROGRESS, someone else might be running scrub */
421	if (ret == -EINPROGRESS) {
422		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
423		ret = 0;
424	} else {
425		WARN_ON(ret);
426	}
427
428	return ret;
429
430leave:
431	dev_replace->srcdev = NULL;
432	dev_replace->tgtdev = NULL;
433	btrfs_dev_replace_unlock(dev_replace);
434	btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
435	return ret;
436}
437
438/*
439 * blocked until all flighting bios are finished.
440 */
441static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
442{
443	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
444	wait_event(fs_info->replace_wait, !percpu_counter_sum(
445		   &fs_info->bio_counter));
446}
447
448/*
449 * we have removed target device, it is safe to allow new bios request.
450 */
451static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
452{
453	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
454	if (waitqueue_active(&fs_info->replace_wait))
455		wake_up(&fs_info->replace_wait);
456}
457
458static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
459				       int scrub_ret)
460{
461	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
462	struct btrfs_device *tgt_device;
463	struct btrfs_device *src_device;
464	struct btrfs_root *root = fs_info->tree_root;
465	u8 uuid_tmp[BTRFS_UUID_SIZE];
466	struct btrfs_trans_handle *trans;
467	int ret = 0;
468
469	/* don't allow cancel or unmount to disturb the finishing procedure */
470	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
471
472	btrfs_dev_replace_lock(dev_replace);
473	/* was the operation canceled, or is it finished? */
474	if (dev_replace->replace_state !=
475	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
476		btrfs_dev_replace_unlock(dev_replace);
477		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
478		return 0;
479	}
480
481	tgt_device = dev_replace->tgtdev;
482	src_device = dev_replace->srcdev;
483	btrfs_dev_replace_unlock(dev_replace);
484
485	/*
486	 * flush all outstanding I/O and inode extent mappings before the
487	 * copy operation is declared as being finished
488	 */
489	ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
490	if (ret) {
491		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
492		return ret;
493	}
494	btrfs_wait_ordered_roots(root->fs_info, -1);
495
496	trans = btrfs_start_transaction(root, 0);
497	if (IS_ERR(trans)) {
498		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
499		return PTR_ERR(trans);
500	}
501	ret = btrfs_commit_transaction(trans, root);
502	WARN_ON(ret);
503
504	mutex_lock(&uuid_mutex);
505	/* keep away write_all_supers() during the finishing procedure */
506	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
507	mutex_lock(&root->fs_info->chunk_mutex);
508	btrfs_dev_replace_lock(dev_replace);
509	dev_replace->replace_state =
510		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
511			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
512	dev_replace->tgtdev = NULL;
513	dev_replace->srcdev = NULL;
514	dev_replace->time_stopped = get_seconds();
515	dev_replace->item_needs_writeback = 1;
516
517	/* replace old device with new one in mapping tree */
518	if (!scrub_ret) {
519		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
520								src_device,
521								tgt_device);
522	} else {
523		printk_in_rcu(KERN_ERR
524			      "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
525			      src_device->missing ? "<missing disk>" :
526			        rcu_str_deref(src_device->name),
527			      src_device->devid,
528			      rcu_str_deref(tgt_device->name), scrub_ret);
529		btrfs_dev_replace_unlock(dev_replace);
530		mutex_unlock(&root->fs_info->chunk_mutex);
531		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
532		mutex_unlock(&uuid_mutex);
533		if (tgt_device)
534			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
535		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
536
537		return scrub_ret;
538	}
539
540	printk_in_rcu(KERN_INFO
541		      "BTRFS: dev_replace from %s (devid %llu) to %s finished\n",
542		      src_device->missing ? "<missing disk>" :
543		        rcu_str_deref(src_device->name),
544		      src_device->devid,
545		      rcu_str_deref(tgt_device->name));
546	tgt_device->is_tgtdev_for_dev_replace = 0;
547	tgt_device->devid = src_device->devid;
548	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
549	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
550	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
551	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
552	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
553	btrfs_device_set_disk_total_bytes(tgt_device,
554					  src_device->disk_total_bytes);
555	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
556	ASSERT(list_empty(&src_device->resized_list));
557	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
558	tgt_device->commit_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	fs_info->fs_devices->rw_devices++;
565
566	btrfs_dev_replace_unlock(dev_replace);
567
568	btrfs_rm_dev_replace_blocked(fs_info);
569
570	btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
571
572	btrfs_rm_dev_replace_unblocked(fs_info);
573
574	/*
575	 * this is again a consistent state where no dev_replace procedure
576	 * is running, the target device is part of the filesystem, the
577	 * source device is not part of the filesystem anymore and its 1st
578	 * superblock is scratched out so that it is no longer marked to
579	 * belong to this filesystem.
580	 */
581	mutex_unlock(&root->fs_info->chunk_mutex);
582	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
583	mutex_unlock(&uuid_mutex);
584
585	/* replace the sysfs entry */
586	btrfs_kobj_rm_device(fs_info, src_device);
587	btrfs_kobj_add_device(fs_info, tgt_device);
588	btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
589
590	/* write back the superblocks */
591	trans = btrfs_start_transaction(root, 0);
592	if (!IS_ERR(trans))
593		btrfs_commit_transaction(trans, root);
594
595	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
596
597	return 0;
598}
599
600static void btrfs_dev_replace_update_device_in_mapping_tree(
601						struct btrfs_fs_info *fs_info,
602						struct btrfs_device *srcdev,
603						struct btrfs_device *tgtdev)
604{
605	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
606	struct extent_map *em;
607	struct map_lookup *map;
608	u64 start = 0;
609	int i;
610
611	write_lock(&em_tree->lock);
612	do {
613		em = lookup_extent_mapping(em_tree, start, (u64)-1);
614		if (!em)
615			break;
616		map = (struct map_lookup *)em->bdev;
617		for (i = 0; i < map->num_stripes; i++)
618			if (srcdev == map->stripes[i].dev)
619				map->stripes[i].dev = tgtdev;
620		start = em->start + em->len;
621		free_extent_map(em);
622	} while (start);
623	write_unlock(&em_tree->lock);
624}
625
626static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
627					 char *srcdev_name,
628					 struct btrfs_device **device)
629{
630	int ret;
631
632	if (srcdevid) {
633		ret = 0;
634		*device = btrfs_find_device(root->fs_info, srcdevid, NULL,
635					    NULL);
636		if (!*device)
637			ret = -ENOENT;
638	} else {
639		ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
640							   device);
641	}
642	return ret;
643}
644
645void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
646			      struct btrfs_ioctl_dev_replace_args *args)
647{
648	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
649	struct btrfs_device *srcdev;
650
651	btrfs_dev_replace_lock(dev_replace);
652	/* even if !dev_replace_is_valid, the values are good enough for
653	 * the replace_status ioctl */
654	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
655	args->status.replace_state = dev_replace->replace_state;
656	args->status.time_started = dev_replace->time_started;
657	args->status.time_stopped = dev_replace->time_stopped;
658	args->status.num_write_errors =
659		atomic64_read(&dev_replace->num_write_errors);
660	args->status.num_uncorrectable_read_errors =
661		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
662	switch (dev_replace->replace_state) {
663	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
664	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
665		args->status.progress_1000 = 0;
666		break;
667	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
668		args->status.progress_1000 = 1000;
669		break;
670	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
671	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
672		srcdev = dev_replace->srcdev;
673		args->status.progress_1000 = div_u64(dev_replace->cursor_left,
674			div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
675		break;
676	}
677	btrfs_dev_replace_unlock(dev_replace);
678}
679
680int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
681			     struct btrfs_ioctl_dev_replace_args *args)
682{
683	args->result = __btrfs_dev_replace_cancel(fs_info);
684	return 0;
685}
686
687static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
688{
689	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
690	struct btrfs_device *tgt_device = NULL;
691	struct btrfs_trans_handle *trans;
692	struct btrfs_root *root = fs_info->tree_root;
693	u64 result;
694	int ret;
695
696	if (fs_info->sb->s_flags & MS_RDONLY)
697		return -EROFS;
698
699	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
700	btrfs_dev_replace_lock(dev_replace);
701	switch (dev_replace->replace_state) {
702	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
703	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
704	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
705		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
706		btrfs_dev_replace_unlock(dev_replace);
707		goto leave;
708	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
709	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
710		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
711		tgt_device = dev_replace->tgtdev;
712		dev_replace->tgtdev = NULL;
713		dev_replace->srcdev = NULL;
714		break;
715	}
716	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
717	dev_replace->time_stopped = get_seconds();
718	dev_replace->item_needs_writeback = 1;
719	btrfs_dev_replace_unlock(dev_replace);
720	btrfs_scrub_cancel(fs_info);
721
722	trans = btrfs_start_transaction(root, 0);
723	if (IS_ERR(trans)) {
724		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
725		return PTR_ERR(trans);
726	}
727	ret = btrfs_commit_transaction(trans, root);
728	WARN_ON(ret);
729	if (tgt_device)
730		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
731
732leave:
733	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
734	return result;
735}
736
737void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
738{
739	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
740
741	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
742	btrfs_dev_replace_lock(dev_replace);
743	switch (dev_replace->replace_state) {
744	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
745	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
746	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
747	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
748		break;
749	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
750		dev_replace->replace_state =
751			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
752		dev_replace->time_stopped = get_seconds();
753		dev_replace->item_needs_writeback = 1;
754		btrfs_info(fs_info, "suspending dev_replace for unmount");
755		break;
756	}
757
758	btrfs_dev_replace_unlock(dev_replace);
759	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
760}
761
762/* resume dev_replace procedure that was interrupted by unmount */
763int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
764{
765	struct task_struct *task;
766	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
767
768	btrfs_dev_replace_lock(dev_replace);
769	switch (dev_replace->replace_state) {
770	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
771	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
772	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
773		btrfs_dev_replace_unlock(dev_replace);
774		return 0;
775	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
776		break;
777	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
778		dev_replace->replace_state =
779			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
780		break;
781	}
782	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
783		btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
784		btrfs_info(fs_info,
785			"you may cancel the operation after 'mount -o degraded'");
786		btrfs_dev_replace_unlock(dev_replace);
787		return 0;
788	}
789	btrfs_dev_replace_unlock(dev_replace);
790
791	WARN_ON(atomic_xchg(
792		&fs_info->mutually_exclusive_operation_running, 1));
793	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
794	return PTR_ERR_OR_ZERO(task);
795}
796
797static int btrfs_dev_replace_kthread(void *data)
798{
799	struct btrfs_fs_info *fs_info = data;
800	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
801	struct btrfs_ioctl_dev_replace_args *status_args;
802	u64 progress;
803
804	status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
805	if (status_args) {
806		btrfs_dev_replace_status(fs_info, status_args);
807		progress = status_args->status.progress_1000;
808		kfree(status_args);
809		progress = div_u64(progress, 10);
810		printk_in_rcu(KERN_INFO
811			"BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
812			dev_replace->srcdev->missing ? "<missing disk>" :
813			rcu_str_deref(dev_replace->srcdev->name),
814			dev_replace->srcdev->devid,
815			dev_replace->tgtdev ?
816			rcu_str_deref(dev_replace->tgtdev->name) :
817			"<missing target disk>",
818			(unsigned int)progress);
819	}
820	btrfs_dev_replace_continue_on_mount(fs_info);
821	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
822
823	return 0;
824}
825
826static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
827{
828	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
829	int ret;
830
831	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
832			      dev_replace->committed_cursor_left,
833			      btrfs_device_get_total_bytes(dev_replace->srcdev),
834			      &dev_replace->scrub_progress, 0, 1);
835	ret = btrfs_dev_replace_finishing(fs_info, ret);
836	WARN_ON(ret);
837	return 0;
838}
839
840int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
841{
842	if (!dev_replace->is_valid)
843		return 0;
844
845	switch (dev_replace->replace_state) {
846	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
847	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
848	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
849		return 0;
850	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
851	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
852		/*
853		 * return true even if tgtdev is missing (this is
854		 * something that can happen if the dev_replace
855		 * procedure is suspended by an umount and then
856		 * the tgtdev is missing (or "btrfs dev scan") was
857		 * not called and the the filesystem is remounted
858		 * in degraded state. This does not stop the
859		 * dev_replace procedure. It needs to be canceled
860		 * manually if the cancelation is wanted.
861		 */
862		break;
863	}
864	return 1;
865}
866
867void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
868{
869	/* the beginning is just an optimization for the typical case */
870	if (atomic_read(&dev_replace->nesting_level) == 0) {
871acquire_lock:
872		/* this is not a nested case where the same thread
873		 * is trying to acqurire the same lock twice */
874		mutex_lock(&dev_replace->lock);
875		mutex_lock(&dev_replace->lock_management_lock);
876		dev_replace->lock_owner = current->pid;
877		atomic_inc(&dev_replace->nesting_level);
878		mutex_unlock(&dev_replace->lock_management_lock);
879		return;
880	}
881
882	mutex_lock(&dev_replace->lock_management_lock);
883	if (atomic_read(&dev_replace->nesting_level) > 0 &&
884	    dev_replace->lock_owner == current->pid) {
885		WARN_ON(!mutex_is_locked(&dev_replace->lock));
886		atomic_inc(&dev_replace->nesting_level);
887		mutex_unlock(&dev_replace->lock_management_lock);
888		return;
889	}
890
891	mutex_unlock(&dev_replace->lock_management_lock);
892	goto acquire_lock;
893}
894
895void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
896{
897	WARN_ON(!mutex_is_locked(&dev_replace->lock));
898	mutex_lock(&dev_replace->lock_management_lock);
899	WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
900	WARN_ON(dev_replace->lock_owner != current->pid);
901	atomic_dec(&dev_replace->nesting_level);
902	if (atomic_read(&dev_replace->nesting_level) == 0) {
903		dev_replace->lock_owner = 0;
904		mutex_unlock(&dev_replace->lock_management_lock);
905		mutex_unlock(&dev_replace->lock);
906	} else {
907		mutex_unlock(&dev_replace->lock_management_lock);
908	}
909}
910
911void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
912{
913	percpu_counter_inc(&fs_info->bio_counter);
914}
915
916void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
917{
918	percpu_counter_sub(&fs_info->bio_counter, amount);
919
920	if (waitqueue_active(&fs_info->replace_wait))
921		wake_up(&fs_info->replace_wait);
922}
923
924void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
925{
926	while (1) {
927		percpu_counter_inc(&fs_info->bio_counter);
928		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
929				     &fs_info->fs_state)))
930			break;
931
932		btrfs_bio_counter_dec(fs_info);
933		wait_event(fs_info->replace_wait,
934			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
935				     &fs_info->fs_state));
936	}
937}
938