xref: /openbmc/linux/fs/btrfs/volumes.c (revision 95298d63)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/bio.h>
8 #include <linux/slab.h>
9 #include <linux/blkdev.h>
10 #include <linux/ratelimit.h>
11 #include <linux/kthread.h>
12 #include <linux/raid/pq.h>
13 #include <linux/semaphore.h>
14 #include <linux/uuid.h>
15 #include <linux/list_sort.h>
16 #include "misc.h"
17 #include "ctree.h"
18 #include "extent_map.h"
19 #include "disk-io.h"
20 #include "transaction.h"
21 #include "print-tree.h"
22 #include "volumes.h"
23 #include "raid56.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
27 #include "dev-replace.h"
28 #include "sysfs.h"
29 #include "tree-checker.h"
30 #include "space-info.h"
31 #include "block-group.h"
32 #include "discard.h"
33 
34 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
35 	[BTRFS_RAID_RAID10] = {
36 		.sub_stripes	= 2,
37 		.dev_stripes	= 1,
38 		.devs_max	= 0,	/* 0 == as many as possible */
39 		.devs_min	= 4,
40 		.tolerated_failures = 1,
41 		.devs_increment	= 2,
42 		.ncopies	= 2,
43 		.nparity        = 0,
44 		.raid_name	= "raid10",
45 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID10,
46 		.mindev_error	= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
47 	},
48 	[BTRFS_RAID_RAID1] = {
49 		.sub_stripes	= 1,
50 		.dev_stripes	= 1,
51 		.devs_max	= 2,
52 		.devs_min	= 2,
53 		.tolerated_failures = 1,
54 		.devs_increment	= 2,
55 		.ncopies	= 2,
56 		.nparity        = 0,
57 		.raid_name	= "raid1",
58 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1,
59 		.mindev_error	= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
60 	},
61 	[BTRFS_RAID_RAID1C3] = {
62 		.sub_stripes	= 1,
63 		.dev_stripes	= 1,
64 		.devs_max	= 3,
65 		.devs_min	= 3,
66 		.tolerated_failures = 2,
67 		.devs_increment	= 3,
68 		.ncopies	= 3,
69 		.nparity        = 0,
70 		.raid_name	= "raid1c3",
71 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C3,
72 		.mindev_error	= BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
73 	},
74 	[BTRFS_RAID_RAID1C4] = {
75 		.sub_stripes	= 1,
76 		.dev_stripes	= 1,
77 		.devs_max	= 4,
78 		.devs_min	= 4,
79 		.tolerated_failures = 3,
80 		.devs_increment	= 4,
81 		.ncopies	= 4,
82 		.nparity        = 0,
83 		.raid_name	= "raid1c4",
84 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C4,
85 		.mindev_error	= BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
86 	},
87 	[BTRFS_RAID_DUP] = {
88 		.sub_stripes	= 1,
89 		.dev_stripes	= 2,
90 		.devs_max	= 1,
91 		.devs_min	= 1,
92 		.tolerated_failures = 0,
93 		.devs_increment	= 1,
94 		.ncopies	= 2,
95 		.nparity        = 0,
96 		.raid_name	= "dup",
97 		.bg_flag	= BTRFS_BLOCK_GROUP_DUP,
98 		.mindev_error	= 0,
99 	},
100 	[BTRFS_RAID_RAID0] = {
101 		.sub_stripes	= 1,
102 		.dev_stripes	= 1,
103 		.devs_max	= 0,
104 		.devs_min	= 2,
105 		.tolerated_failures = 0,
106 		.devs_increment	= 1,
107 		.ncopies	= 1,
108 		.nparity        = 0,
109 		.raid_name	= "raid0",
110 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID0,
111 		.mindev_error	= 0,
112 	},
113 	[BTRFS_RAID_SINGLE] = {
114 		.sub_stripes	= 1,
115 		.dev_stripes	= 1,
116 		.devs_max	= 1,
117 		.devs_min	= 1,
118 		.tolerated_failures = 0,
119 		.devs_increment	= 1,
120 		.ncopies	= 1,
121 		.nparity        = 0,
122 		.raid_name	= "single",
123 		.bg_flag	= 0,
124 		.mindev_error	= 0,
125 	},
126 	[BTRFS_RAID_RAID5] = {
127 		.sub_stripes	= 1,
128 		.dev_stripes	= 1,
129 		.devs_max	= 0,
130 		.devs_min	= 2,
131 		.tolerated_failures = 1,
132 		.devs_increment	= 1,
133 		.ncopies	= 1,
134 		.nparity        = 1,
135 		.raid_name	= "raid5",
136 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID5,
137 		.mindev_error	= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
138 	},
139 	[BTRFS_RAID_RAID6] = {
140 		.sub_stripes	= 1,
141 		.dev_stripes	= 1,
142 		.devs_max	= 0,
143 		.devs_min	= 3,
144 		.tolerated_failures = 2,
145 		.devs_increment	= 1,
146 		.ncopies	= 1,
147 		.nparity        = 2,
148 		.raid_name	= "raid6",
149 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID6,
150 		.mindev_error	= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
151 	},
152 };
153 
154 const char *btrfs_bg_type_to_raid_name(u64 flags)
155 {
156 	const int index = btrfs_bg_flags_to_raid_index(flags);
157 
158 	if (index >= BTRFS_NR_RAID_TYPES)
159 		return NULL;
160 
161 	return btrfs_raid_array[index].raid_name;
162 }
163 
164 /*
165  * Fill @buf with textual description of @bg_flags, no more than @size_buf
166  * bytes including terminating null byte.
167  */
168 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
169 {
170 	int i;
171 	int ret;
172 	char *bp = buf;
173 	u64 flags = bg_flags;
174 	u32 size_bp = size_buf;
175 
176 	if (!flags) {
177 		strcpy(bp, "NONE");
178 		return;
179 	}
180 
181 #define DESCRIBE_FLAG(flag, desc)						\
182 	do {								\
183 		if (flags & (flag)) {					\
184 			ret = snprintf(bp, size_bp, "%s|", (desc));	\
185 			if (ret < 0 || ret >= size_bp)			\
186 				goto out_overflow;			\
187 			size_bp -= ret;					\
188 			bp += ret;					\
189 			flags &= ~(flag);				\
190 		}							\
191 	} while (0)
192 
193 	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
194 	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
195 	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
196 
197 	DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
198 	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
199 		DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
200 			      btrfs_raid_array[i].raid_name);
201 #undef DESCRIBE_FLAG
202 
203 	if (flags) {
204 		ret = snprintf(bp, size_bp, "0x%llx|", flags);
205 		size_bp -= ret;
206 	}
207 
208 	if (size_bp < size_buf)
209 		buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
210 
211 	/*
212 	 * The text is trimmed, it's up to the caller to provide sufficiently
213 	 * large buffer
214 	 */
215 out_overflow:;
216 }
217 
218 static int init_first_rw_device(struct btrfs_trans_handle *trans);
219 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
220 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
221 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
222 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
223 			     enum btrfs_map_op op,
224 			     u64 logical, u64 *length,
225 			     struct btrfs_bio **bbio_ret,
226 			     int mirror_num, int need_raid_map);
227 
228 /*
229  * Device locking
230  * ==============
231  *
232  * There are several mutexes that protect manipulation of devices and low-level
233  * structures like chunks but not block groups, extents or files
234  *
235  * uuid_mutex (global lock)
236  * ------------------------
237  * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
238  * the SCAN_DEV ioctl registration or from mount either implicitly (the first
239  * device) or requested by the device= mount option
240  *
241  * the mutex can be very coarse and can cover long-running operations
242  *
243  * protects: updates to fs_devices counters like missing devices, rw devices,
244  * seeding, structure cloning, opening/closing devices at mount/umount time
245  *
246  * global::fs_devs - add, remove, updates to the global list
247  *
248  * does not protect: manipulation of the fs_devices::devices list!
249  *
250  * btrfs_device::name - renames (write side), read is RCU
251  *
252  * fs_devices::device_list_mutex (per-fs, with RCU)
253  * ------------------------------------------------
254  * protects updates to fs_devices::devices, ie. adding and deleting
255  *
256  * simple list traversal with read-only actions can be done with RCU protection
257  *
258  * may be used to exclude some operations from running concurrently without any
259  * modifications to the list (see write_all_supers)
260  *
261  * balance_mutex
262  * -------------
263  * protects balance structures (status, state) and context accessed from
264  * several places (internally, ioctl)
265  *
266  * chunk_mutex
267  * -----------
268  * protects chunks, adding or removing during allocation, trim or when a new
269  * device is added/removed. Additionally it also protects post_commit_list of
270  * individual devices, since they can be added to the transaction's
271  * post_commit_list only with chunk_mutex held.
272  *
273  * cleaner_mutex
274  * -------------
275  * a big lock that is held by the cleaner thread and prevents running subvolume
276  * cleaning together with relocation or delayed iputs
277  *
278  *
279  * Lock nesting
280  * ============
281  *
282  * uuid_mutex
283  *   device_list_mutex
284  *     chunk_mutex
285  *   balance_mutex
286  *
287  *
288  * Exclusive operations, BTRFS_FS_EXCL_OP
289  * ======================================
290  *
291  * Maintains the exclusivity of the following operations that apply to the
292  * whole filesystem and cannot run in parallel.
293  *
294  * - Balance (*)
295  * - Device add
296  * - Device remove
297  * - Device replace (*)
298  * - Resize
299  *
300  * The device operations (as above) can be in one of the following states:
301  *
302  * - Running state
303  * - Paused state
304  * - Completed state
305  *
306  * Only device operations marked with (*) can go into the Paused state for the
307  * following reasons:
308  *
309  * - ioctl (only Balance can be Paused through ioctl)
310  * - filesystem remounted as read-only
311  * - filesystem unmounted and mounted as read-only
312  * - system power-cycle and filesystem mounted as read-only
313  * - filesystem or device errors leading to forced read-only
314  *
315  * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
316  * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
317  * A device operation in Paused or Running state can be canceled or resumed
318  * either by ioctl (Balance only) or when remounted as read-write.
319  * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
320  * completed.
321  */
322 
323 DEFINE_MUTEX(uuid_mutex);
324 static LIST_HEAD(fs_uuids);
325 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
326 {
327 	return &fs_uuids;
328 }
329 
330 /*
331  * alloc_fs_devices - allocate struct btrfs_fs_devices
332  * @fsid:		if not NULL, copy the UUID to fs_devices::fsid
333  * @metadata_fsid:	if not NULL, copy the UUID to fs_devices::metadata_fsid
334  *
335  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
336  * The returned struct is not linked onto any lists and can be destroyed with
337  * kfree() right away.
338  */
339 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
340 						 const u8 *metadata_fsid)
341 {
342 	struct btrfs_fs_devices *fs_devs;
343 
344 	fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
345 	if (!fs_devs)
346 		return ERR_PTR(-ENOMEM);
347 
348 	mutex_init(&fs_devs->device_list_mutex);
349 
350 	INIT_LIST_HEAD(&fs_devs->devices);
351 	INIT_LIST_HEAD(&fs_devs->alloc_list);
352 	INIT_LIST_HEAD(&fs_devs->fs_list);
353 	if (fsid)
354 		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
355 
356 	if (metadata_fsid)
357 		memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
358 	else if (fsid)
359 		memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
360 
361 	return fs_devs;
362 }
363 
364 void btrfs_free_device(struct btrfs_device *device)
365 {
366 	WARN_ON(!list_empty(&device->post_commit_list));
367 	rcu_string_free(device->name);
368 	extent_io_tree_release(&device->alloc_state);
369 	bio_put(device->flush_bio);
370 	kfree(device);
371 }
372 
373 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
374 {
375 	struct btrfs_device *device;
376 	WARN_ON(fs_devices->opened);
377 	while (!list_empty(&fs_devices->devices)) {
378 		device = list_entry(fs_devices->devices.next,
379 				    struct btrfs_device, dev_list);
380 		list_del(&device->dev_list);
381 		btrfs_free_device(device);
382 	}
383 	kfree(fs_devices);
384 }
385 
386 void __exit btrfs_cleanup_fs_uuids(void)
387 {
388 	struct btrfs_fs_devices *fs_devices;
389 
390 	while (!list_empty(&fs_uuids)) {
391 		fs_devices = list_entry(fs_uuids.next,
392 					struct btrfs_fs_devices, fs_list);
393 		list_del(&fs_devices->fs_list);
394 		free_fs_devices(fs_devices);
395 	}
396 }
397 
398 /*
399  * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
400  * Returned struct is not linked onto any lists and must be destroyed using
401  * btrfs_free_device.
402  */
403 static struct btrfs_device *__alloc_device(void)
404 {
405 	struct btrfs_device *dev;
406 
407 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
408 	if (!dev)
409 		return ERR_PTR(-ENOMEM);
410 
411 	/*
412 	 * Preallocate a bio that's always going to be used for flushing device
413 	 * barriers and matches the device lifespan
414 	 */
415 	dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
416 	if (!dev->flush_bio) {
417 		kfree(dev);
418 		return ERR_PTR(-ENOMEM);
419 	}
420 
421 	INIT_LIST_HEAD(&dev->dev_list);
422 	INIT_LIST_HEAD(&dev->dev_alloc_list);
423 	INIT_LIST_HEAD(&dev->post_commit_list);
424 
425 	atomic_set(&dev->reada_in_flight, 0);
426 	atomic_set(&dev->dev_stats_ccnt, 0);
427 	btrfs_device_data_ordered_init(dev);
428 	INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
429 	INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
430 	extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL);
431 
432 	return dev;
433 }
434 
435 static noinline struct btrfs_fs_devices *find_fsid(
436 		const u8 *fsid, const u8 *metadata_fsid)
437 {
438 	struct btrfs_fs_devices *fs_devices;
439 
440 	ASSERT(fsid);
441 
442 	/* Handle non-split brain cases */
443 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
444 		if (metadata_fsid) {
445 			if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
446 			    && memcmp(metadata_fsid, fs_devices->metadata_uuid,
447 				      BTRFS_FSID_SIZE) == 0)
448 				return fs_devices;
449 		} else {
450 			if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
451 				return fs_devices;
452 		}
453 	}
454 	return NULL;
455 }
456 
457 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
458 				struct btrfs_super_block *disk_super)
459 {
460 
461 	struct btrfs_fs_devices *fs_devices;
462 
463 	/*
464 	 * Handle scanned device having completed its fsid change but
465 	 * belonging to a fs_devices that was created by first scanning
466 	 * a device which didn't have its fsid/metadata_uuid changed
467 	 * at all and the CHANGING_FSID_V2 flag set.
468 	 */
469 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
470 		if (fs_devices->fsid_change &&
471 		    memcmp(disk_super->metadata_uuid, fs_devices->fsid,
472 			   BTRFS_FSID_SIZE) == 0 &&
473 		    memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
474 			   BTRFS_FSID_SIZE) == 0) {
475 			return fs_devices;
476 		}
477 	}
478 	/*
479 	 * Handle scanned device having completed its fsid change but
480 	 * belonging to a fs_devices that was created by a device that
481 	 * has an outdated pair of fsid/metadata_uuid and
482 	 * CHANGING_FSID_V2 flag set.
483 	 */
484 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
485 		if (fs_devices->fsid_change &&
486 		    memcmp(fs_devices->metadata_uuid,
487 			   fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
488 		    memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
489 			   BTRFS_FSID_SIZE) == 0) {
490 			return fs_devices;
491 		}
492 	}
493 
494 	return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
495 }
496 
497 
498 static int
499 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
500 		      int flush, struct block_device **bdev,
501 		      struct btrfs_super_block **disk_super)
502 {
503 	int ret;
504 
505 	*bdev = blkdev_get_by_path(device_path, flags, holder);
506 
507 	if (IS_ERR(*bdev)) {
508 		ret = PTR_ERR(*bdev);
509 		goto error;
510 	}
511 
512 	if (flush)
513 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
514 	ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
515 	if (ret) {
516 		blkdev_put(*bdev, flags);
517 		goto error;
518 	}
519 	invalidate_bdev(*bdev);
520 	*disk_super = btrfs_read_dev_super(*bdev);
521 	if (IS_ERR(*disk_super)) {
522 		ret = PTR_ERR(*disk_super);
523 		blkdev_put(*bdev, flags);
524 		goto error;
525 	}
526 
527 	return 0;
528 
529 error:
530 	*bdev = NULL;
531 	return ret;
532 }
533 
534 static bool device_path_matched(const char *path, struct btrfs_device *device)
535 {
536 	int found;
537 
538 	rcu_read_lock();
539 	found = strcmp(rcu_str_deref(device->name), path);
540 	rcu_read_unlock();
541 
542 	return found == 0;
543 }
544 
545 /*
546  *  Search and remove all stale (devices which are not mounted) devices.
547  *  When both inputs are NULL, it will search and release all stale devices.
548  *  path:	Optional. When provided will it release all unmounted devices
549  *		matching this path only.
550  *  skip_dev:	Optional. Will skip this device when searching for the stale
551  *		devices.
552  *  Return:	0 for success or if @path is NULL.
553  * 		-EBUSY if @path is a mounted device.
554  * 		-ENOENT if @path does not match any device in the list.
555  */
556 static int btrfs_free_stale_devices(const char *path,
557 				     struct btrfs_device *skip_device)
558 {
559 	struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
560 	struct btrfs_device *device, *tmp_device;
561 	int ret = 0;
562 
563 	if (path)
564 		ret = -ENOENT;
565 
566 	list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
567 
568 		mutex_lock(&fs_devices->device_list_mutex);
569 		list_for_each_entry_safe(device, tmp_device,
570 					 &fs_devices->devices, dev_list) {
571 			if (skip_device && skip_device == device)
572 				continue;
573 			if (path && !device->name)
574 				continue;
575 			if (path && !device_path_matched(path, device))
576 				continue;
577 			if (fs_devices->opened) {
578 				/* for an already deleted device return 0 */
579 				if (path && ret != 0)
580 					ret = -EBUSY;
581 				break;
582 			}
583 
584 			/* delete the stale device */
585 			fs_devices->num_devices--;
586 			list_del(&device->dev_list);
587 			btrfs_free_device(device);
588 
589 			ret = 0;
590 			if (fs_devices->num_devices == 0)
591 				break;
592 		}
593 		mutex_unlock(&fs_devices->device_list_mutex);
594 
595 		if (fs_devices->num_devices == 0) {
596 			btrfs_sysfs_remove_fsid(fs_devices);
597 			list_del(&fs_devices->fs_list);
598 			free_fs_devices(fs_devices);
599 		}
600 	}
601 
602 	return ret;
603 }
604 
605 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
606 			struct btrfs_device *device, fmode_t flags,
607 			void *holder)
608 {
609 	struct request_queue *q;
610 	struct block_device *bdev;
611 	struct btrfs_super_block *disk_super;
612 	u64 devid;
613 	int ret;
614 
615 	if (device->bdev)
616 		return -EINVAL;
617 	if (!device->name)
618 		return -EINVAL;
619 
620 	ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
621 				    &bdev, &disk_super);
622 	if (ret)
623 		return ret;
624 
625 	devid = btrfs_stack_device_id(&disk_super->dev_item);
626 	if (devid != device->devid)
627 		goto error_free_page;
628 
629 	if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
630 		goto error_free_page;
631 
632 	device->generation = btrfs_super_generation(disk_super);
633 
634 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
635 		if (btrfs_super_incompat_flags(disk_super) &
636 		    BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
637 			pr_err(
638 		"BTRFS: Invalid seeding and uuid-changed device detected\n");
639 			goto error_free_page;
640 		}
641 
642 		clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
643 		fs_devices->seeding = true;
644 	} else {
645 		if (bdev_read_only(bdev))
646 			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
647 		else
648 			set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
649 	}
650 
651 	q = bdev_get_queue(bdev);
652 	if (!blk_queue_nonrot(q))
653 		fs_devices->rotating = true;
654 
655 	device->bdev = bdev;
656 	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
657 	device->mode = flags;
658 
659 	fs_devices->open_devices++;
660 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
661 	    device->devid != BTRFS_DEV_REPLACE_DEVID) {
662 		fs_devices->rw_devices++;
663 		list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
664 	}
665 	btrfs_release_disk_super(disk_super);
666 
667 	return 0;
668 
669 error_free_page:
670 	btrfs_release_disk_super(disk_super);
671 	blkdev_put(bdev, flags);
672 
673 	return -EINVAL;
674 }
675 
676 /*
677  * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
678  * being created with a disk that has already completed its fsid change. Such
679  * disk can belong to an fs which has its FSID changed or to one which doesn't.
680  * Handle both cases here.
681  */
682 static struct btrfs_fs_devices *find_fsid_inprogress(
683 					struct btrfs_super_block *disk_super)
684 {
685 	struct btrfs_fs_devices *fs_devices;
686 
687 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
688 		if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
689 			   BTRFS_FSID_SIZE) != 0 &&
690 		    memcmp(fs_devices->metadata_uuid, disk_super->fsid,
691 			   BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
692 			return fs_devices;
693 		}
694 	}
695 
696 	return find_fsid(disk_super->fsid, NULL);
697 }
698 
699 
700 static struct btrfs_fs_devices *find_fsid_changed(
701 					struct btrfs_super_block *disk_super)
702 {
703 	struct btrfs_fs_devices *fs_devices;
704 
705 	/*
706 	 * Handles the case where scanned device is part of an fs that had
707 	 * multiple successful changes of FSID but curently device didn't
708 	 * observe it. Meaning our fsid will be different than theirs. We need
709 	 * to handle two subcases :
710 	 *  1 - The fs still continues to have different METADATA/FSID uuids.
711 	 *  2 - The fs is switched back to its original FSID (METADATA/FSID
712 	 *  are equal).
713 	 */
714 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
715 		/* Changed UUIDs */
716 		if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
717 			   BTRFS_FSID_SIZE) != 0 &&
718 		    memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
719 			   BTRFS_FSID_SIZE) == 0 &&
720 		    memcmp(fs_devices->fsid, disk_super->fsid,
721 			   BTRFS_FSID_SIZE) != 0)
722 			return fs_devices;
723 
724 		/* Unchanged UUIDs */
725 		if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
726 			   BTRFS_FSID_SIZE) == 0 &&
727 		    memcmp(fs_devices->fsid, disk_super->metadata_uuid,
728 			   BTRFS_FSID_SIZE) == 0)
729 			return fs_devices;
730 	}
731 
732 	return NULL;
733 }
734 
735 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
736 				struct btrfs_super_block *disk_super)
737 {
738 	struct btrfs_fs_devices *fs_devices;
739 
740 	/*
741 	 * Handle the case where the scanned device is part of an fs whose last
742 	 * metadata UUID change reverted it to the original FSID. At the same
743 	 * time * fs_devices was first created by another constitutent device
744 	 * which didn't fully observe the operation. This results in an
745 	 * btrfs_fs_devices created with metadata/fsid different AND
746 	 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
747 	 * fs_devices equal to the FSID of the disk.
748 	 */
749 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
750 		if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
751 			   BTRFS_FSID_SIZE) != 0 &&
752 		    memcmp(fs_devices->metadata_uuid, disk_super->fsid,
753 			   BTRFS_FSID_SIZE) == 0 &&
754 		    fs_devices->fsid_change)
755 			return fs_devices;
756 	}
757 
758 	return NULL;
759 }
760 /*
761  * Add new device to list of registered devices
762  *
763  * Returns:
764  * device pointer which was just added or updated when successful
765  * error pointer when failed
766  */
767 static noinline struct btrfs_device *device_list_add(const char *path,
768 			   struct btrfs_super_block *disk_super,
769 			   bool *new_device_added)
770 {
771 	struct btrfs_device *device;
772 	struct btrfs_fs_devices *fs_devices = NULL;
773 	struct rcu_string *name;
774 	u64 found_transid = btrfs_super_generation(disk_super);
775 	u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
776 	bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
777 		BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
778 	bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
779 					BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
780 
781 	if (fsid_change_in_progress) {
782 		if (!has_metadata_uuid)
783 			fs_devices = find_fsid_inprogress(disk_super);
784 		else
785 			fs_devices = find_fsid_changed(disk_super);
786 	} else if (has_metadata_uuid) {
787 		fs_devices = find_fsid_with_metadata_uuid(disk_super);
788 	} else {
789 		fs_devices = find_fsid_reverted_metadata(disk_super);
790 		if (!fs_devices)
791 			fs_devices = find_fsid(disk_super->fsid, NULL);
792 	}
793 
794 
795 	if (!fs_devices) {
796 		if (has_metadata_uuid)
797 			fs_devices = alloc_fs_devices(disk_super->fsid,
798 						      disk_super->metadata_uuid);
799 		else
800 			fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
801 
802 		if (IS_ERR(fs_devices))
803 			return ERR_CAST(fs_devices);
804 
805 		fs_devices->fsid_change = fsid_change_in_progress;
806 
807 		mutex_lock(&fs_devices->device_list_mutex);
808 		list_add(&fs_devices->fs_list, &fs_uuids);
809 
810 		device = NULL;
811 	} else {
812 		mutex_lock(&fs_devices->device_list_mutex);
813 		device = btrfs_find_device(fs_devices, devid,
814 				disk_super->dev_item.uuid, NULL, false);
815 
816 		/*
817 		 * If this disk has been pulled into an fs devices created by
818 		 * a device which had the CHANGING_FSID_V2 flag then replace the
819 		 * metadata_uuid/fsid values of the fs_devices.
820 		 */
821 		if (fs_devices->fsid_change &&
822 		    found_transid > fs_devices->latest_generation) {
823 			memcpy(fs_devices->fsid, disk_super->fsid,
824 					BTRFS_FSID_SIZE);
825 
826 			if (has_metadata_uuid)
827 				memcpy(fs_devices->metadata_uuid,
828 				       disk_super->metadata_uuid,
829 				       BTRFS_FSID_SIZE);
830 			else
831 				memcpy(fs_devices->metadata_uuid,
832 				       disk_super->fsid, BTRFS_FSID_SIZE);
833 
834 			fs_devices->fsid_change = false;
835 		}
836 	}
837 
838 	if (!device) {
839 		if (fs_devices->opened) {
840 			mutex_unlock(&fs_devices->device_list_mutex);
841 			return ERR_PTR(-EBUSY);
842 		}
843 
844 		device = btrfs_alloc_device(NULL, &devid,
845 					    disk_super->dev_item.uuid);
846 		if (IS_ERR(device)) {
847 			mutex_unlock(&fs_devices->device_list_mutex);
848 			/* we can safely leave the fs_devices entry around */
849 			return device;
850 		}
851 
852 		name = rcu_string_strdup(path, GFP_NOFS);
853 		if (!name) {
854 			btrfs_free_device(device);
855 			mutex_unlock(&fs_devices->device_list_mutex);
856 			return ERR_PTR(-ENOMEM);
857 		}
858 		rcu_assign_pointer(device->name, name);
859 
860 		list_add_rcu(&device->dev_list, &fs_devices->devices);
861 		fs_devices->num_devices++;
862 
863 		device->fs_devices = fs_devices;
864 		*new_device_added = true;
865 
866 		if (disk_super->label[0])
867 			pr_info(
868 	"BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
869 				disk_super->label, devid, found_transid, path,
870 				current->comm, task_pid_nr(current));
871 		else
872 			pr_info(
873 	"BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
874 				disk_super->fsid, devid, found_transid, path,
875 				current->comm, task_pid_nr(current));
876 
877 	} else if (!device->name || strcmp(device->name->str, path)) {
878 		/*
879 		 * When FS is already mounted.
880 		 * 1. If you are here and if the device->name is NULL that
881 		 *    means this device was missing at time of FS mount.
882 		 * 2. If you are here and if the device->name is different
883 		 *    from 'path' that means either
884 		 *      a. The same device disappeared and reappeared with
885 		 *         different name. or
886 		 *      b. The missing-disk-which-was-replaced, has
887 		 *         reappeared now.
888 		 *
889 		 * We must allow 1 and 2a above. But 2b would be a spurious
890 		 * and unintentional.
891 		 *
892 		 * Further in case of 1 and 2a above, the disk at 'path'
893 		 * would have missed some transaction when it was away and
894 		 * in case of 2a the stale bdev has to be updated as well.
895 		 * 2b must not be allowed at all time.
896 		 */
897 
898 		/*
899 		 * For now, we do allow update to btrfs_fs_device through the
900 		 * btrfs dev scan cli after FS has been mounted.  We're still
901 		 * tracking a problem where systems fail mount by subvolume id
902 		 * when we reject replacement on a mounted FS.
903 		 */
904 		if (!fs_devices->opened && found_transid < device->generation) {
905 			/*
906 			 * That is if the FS is _not_ mounted and if you
907 			 * are here, that means there is more than one
908 			 * disk with same uuid and devid.We keep the one
909 			 * with larger generation number or the last-in if
910 			 * generation are equal.
911 			 */
912 			mutex_unlock(&fs_devices->device_list_mutex);
913 			return ERR_PTR(-EEXIST);
914 		}
915 
916 		/*
917 		 * We are going to replace the device path for a given devid,
918 		 * make sure it's the same device if the device is mounted
919 		 */
920 		if (device->bdev) {
921 			struct block_device *path_bdev;
922 
923 			path_bdev = lookup_bdev(path);
924 			if (IS_ERR(path_bdev)) {
925 				mutex_unlock(&fs_devices->device_list_mutex);
926 				return ERR_CAST(path_bdev);
927 			}
928 
929 			if (device->bdev != path_bdev) {
930 				bdput(path_bdev);
931 				mutex_unlock(&fs_devices->device_list_mutex);
932 				btrfs_warn_in_rcu(device->fs_info,
933 			"duplicate device fsid:devid for %pU:%llu old:%s new:%s",
934 					disk_super->fsid, devid,
935 					rcu_str_deref(device->name), path);
936 				return ERR_PTR(-EEXIST);
937 			}
938 			bdput(path_bdev);
939 			btrfs_info_in_rcu(device->fs_info,
940 				"device fsid %pU devid %llu moved old:%s new:%s",
941 				disk_super->fsid, devid,
942 				rcu_str_deref(device->name), path);
943 		}
944 
945 		name = rcu_string_strdup(path, GFP_NOFS);
946 		if (!name) {
947 			mutex_unlock(&fs_devices->device_list_mutex);
948 			return ERR_PTR(-ENOMEM);
949 		}
950 		rcu_string_free(device->name);
951 		rcu_assign_pointer(device->name, name);
952 		if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
953 			fs_devices->missing_devices--;
954 			clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
955 		}
956 	}
957 
958 	/*
959 	 * Unmount does not free the btrfs_device struct but would zero
960 	 * generation along with most of the other members. So just update
961 	 * it back. We need it to pick the disk with largest generation
962 	 * (as above).
963 	 */
964 	if (!fs_devices->opened) {
965 		device->generation = found_transid;
966 		fs_devices->latest_generation = max_t(u64, found_transid,
967 						fs_devices->latest_generation);
968 	}
969 
970 	fs_devices->total_devices = btrfs_super_num_devices(disk_super);
971 
972 	mutex_unlock(&fs_devices->device_list_mutex);
973 	return device;
974 }
975 
976 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
977 {
978 	struct btrfs_fs_devices *fs_devices;
979 	struct btrfs_device *device;
980 	struct btrfs_device *orig_dev;
981 	int ret = 0;
982 
983 	fs_devices = alloc_fs_devices(orig->fsid, NULL);
984 	if (IS_ERR(fs_devices))
985 		return fs_devices;
986 
987 	mutex_lock(&orig->device_list_mutex);
988 	fs_devices->total_devices = orig->total_devices;
989 
990 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
991 		struct rcu_string *name;
992 
993 		device = btrfs_alloc_device(NULL, &orig_dev->devid,
994 					    orig_dev->uuid);
995 		if (IS_ERR(device)) {
996 			ret = PTR_ERR(device);
997 			goto error;
998 		}
999 
1000 		/*
1001 		 * This is ok to do without rcu read locked because we hold the
1002 		 * uuid mutex so nothing we touch in here is going to disappear.
1003 		 */
1004 		if (orig_dev->name) {
1005 			name = rcu_string_strdup(orig_dev->name->str,
1006 					GFP_KERNEL);
1007 			if (!name) {
1008 				btrfs_free_device(device);
1009 				ret = -ENOMEM;
1010 				goto error;
1011 			}
1012 			rcu_assign_pointer(device->name, name);
1013 		}
1014 
1015 		list_add(&device->dev_list, &fs_devices->devices);
1016 		device->fs_devices = fs_devices;
1017 		fs_devices->num_devices++;
1018 	}
1019 	mutex_unlock(&orig->device_list_mutex);
1020 	return fs_devices;
1021 error:
1022 	mutex_unlock(&orig->device_list_mutex);
1023 	free_fs_devices(fs_devices);
1024 	return ERR_PTR(ret);
1025 }
1026 
1027 /*
1028  * After we have read the system tree and know devids belonging to
1029  * this filesystem, remove the device which does not belong there.
1030  */
1031 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1032 {
1033 	struct btrfs_device *device, *next;
1034 	struct btrfs_device *latest_dev = NULL;
1035 
1036 	mutex_lock(&uuid_mutex);
1037 again:
1038 	/* This is the initialized path, it is safe to release the devices. */
1039 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1040 		if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1041 							&device->dev_state)) {
1042 			if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1043 			     &device->dev_state) &&
1044 			    !test_bit(BTRFS_DEV_STATE_MISSING,
1045 				      &device->dev_state) &&
1046 			     (!latest_dev ||
1047 			      device->generation > latest_dev->generation)) {
1048 				latest_dev = device;
1049 			}
1050 			continue;
1051 		}
1052 
1053 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1054 			/*
1055 			 * In the first step, keep the device which has
1056 			 * the correct fsid and the devid that is used
1057 			 * for the dev_replace procedure.
1058 			 * In the second step, the dev_replace state is
1059 			 * read from the device tree and it is known
1060 			 * whether the procedure is really active or
1061 			 * not, which means whether this device is
1062 			 * used or whether it should be removed.
1063 			 */
1064 			if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1065 						  &device->dev_state)) {
1066 				continue;
1067 			}
1068 		}
1069 		if (device->bdev) {
1070 			blkdev_put(device->bdev, device->mode);
1071 			device->bdev = NULL;
1072 			fs_devices->open_devices--;
1073 		}
1074 		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1075 			list_del_init(&device->dev_alloc_list);
1076 			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1077 			if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1078 				      &device->dev_state))
1079 				fs_devices->rw_devices--;
1080 		}
1081 		list_del_init(&device->dev_list);
1082 		fs_devices->num_devices--;
1083 		btrfs_free_device(device);
1084 	}
1085 
1086 	if (fs_devices->seed) {
1087 		fs_devices = fs_devices->seed;
1088 		goto again;
1089 	}
1090 
1091 	fs_devices->latest_bdev = latest_dev->bdev;
1092 
1093 	mutex_unlock(&uuid_mutex);
1094 }
1095 
1096 static void btrfs_close_bdev(struct btrfs_device *device)
1097 {
1098 	if (!device->bdev)
1099 		return;
1100 
1101 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1102 		sync_blockdev(device->bdev);
1103 		invalidate_bdev(device->bdev);
1104 	}
1105 
1106 	blkdev_put(device->bdev, device->mode);
1107 }
1108 
1109 static void btrfs_close_one_device(struct btrfs_device *device)
1110 {
1111 	struct btrfs_fs_devices *fs_devices = device->fs_devices;
1112 
1113 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1114 	    device->devid != BTRFS_DEV_REPLACE_DEVID) {
1115 		list_del_init(&device->dev_alloc_list);
1116 		fs_devices->rw_devices--;
1117 	}
1118 
1119 	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1120 		fs_devices->missing_devices--;
1121 
1122 	btrfs_close_bdev(device);
1123 	if (device->bdev) {
1124 		fs_devices->open_devices--;
1125 		device->bdev = NULL;
1126 	}
1127 	clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1128 
1129 	device->fs_info = NULL;
1130 	atomic_set(&device->dev_stats_ccnt, 0);
1131 	extent_io_tree_release(&device->alloc_state);
1132 
1133 	/* Verify the device is back in a pristine state  */
1134 	ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1135 	ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1136 	ASSERT(list_empty(&device->dev_alloc_list));
1137 	ASSERT(list_empty(&device->post_commit_list));
1138 	ASSERT(atomic_read(&device->reada_in_flight) == 0);
1139 }
1140 
1141 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1142 {
1143 	struct btrfs_device *device, *tmp;
1144 
1145 	if (--fs_devices->opened > 0)
1146 		return 0;
1147 
1148 	mutex_lock(&fs_devices->device_list_mutex);
1149 	list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1150 		btrfs_close_one_device(device);
1151 	}
1152 	mutex_unlock(&fs_devices->device_list_mutex);
1153 
1154 	WARN_ON(fs_devices->open_devices);
1155 	WARN_ON(fs_devices->rw_devices);
1156 	fs_devices->opened = 0;
1157 	fs_devices->seeding = false;
1158 
1159 	return 0;
1160 }
1161 
1162 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1163 {
1164 	struct btrfs_fs_devices *seed_devices = NULL;
1165 	int ret;
1166 
1167 	mutex_lock(&uuid_mutex);
1168 	ret = close_fs_devices(fs_devices);
1169 	if (!fs_devices->opened) {
1170 		seed_devices = fs_devices->seed;
1171 		fs_devices->seed = NULL;
1172 	}
1173 	mutex_unlock(&uuid_mutex);
1174 
1175 	while (seed_devices) {
1176 		fs_devices = seed_devices;
1177 		seed_devices = fs_devices->seed;
1178 		close_fs_devices(fs_devices);
1179 		free_fs_devices(fs_devices);
1180 	}
1181 	return ret;
1182 }
1183 
1184 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1185 				fmode_t flags, void *holder)
1186 {
1187 	struct btrfs_device *device;
1188 	struct btrfs_device *latest_dev = NULL;
1189 
1190 	flags |= FMODE_EXCL;
1191 
1192 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
1193 		/* Just open everything we can; ignore failures here */
1194 		if (btrfs_open_one_device(fs_devices, device, flags, holder))
1195 			continue;
1196 
1197 		if (!latest_dev ||
1198 		    device->generation > latest_dev->generation)
1199 			latest_dev = device;
1200 	}
1201 	if (fs_devices->open_devices == 0)
1202 		return -EINVAL;
1203 
1204 	fs_devices->opened = 1;
1205 	fs_devices->latest_bdev = latest_dev->bdev;
1206 	fs_devices->total_rw_bytes = 0;
1207 	fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1208 
1209 	return 0;
1210 }
1211 
1212 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1213 {
1214 	struct btrfs_device *dev1, *dev2;
1215 
1216 	dev1 = list_entry(a, struct btrfs_device, dev_list);
1217 	dev2 = list_entry(b, struct btrfs_device, dev_list);
1218 
1219 	if (dev1->devid < dev2->devid)
1220 		return -1;
1221 	else if (dev1->devid > dev2->devid)
1222 		return 1;
1223 	return 0;
1224 }
1225 
1226 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1227 		       fmode_t flags, void *holder)
1228 {
1229 	int ret;
1230 
1231 	lockdep_assert_held(&uuid_mutex);
1232 
1233 	mutex_lock(&fs_devices->device_list_mutex);
1234 	if (fs_devices->opened) {
1235 		fs_devices->opened++;
1236 		ret = 0;
1237 	} else {
1238 		list_sort(NULL, &fs_devices->devices, devid_cmp);
1239 		ret = open_fs_devices(fs_devices, flags, holder);
1240 	}
1241 	mutex_unlock(&fs_devices->device_list_mutex);
1242 
1243 	return ret;
1244 }
1245 
1246 void btrfs_release_disk_super(struct btrfs_super_block *super)
1247 {
1248 	struct page *page = virt_to_page(super);
1249 
1250 	put_page(page);
1251 }
1252 
1253 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1254 						       u64 bytenr)
1255 {
1256 	struct btrfs_super_block *disk_super;
1257 	struct page *page;
1258 	void *p;
1259 	pgoff_t index;
1260 
1261 	/* make sure our super fits in the device */
1262 	if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1263 		return ERR_PTR(-EINVAL);
1264 
1265 	/* make sure our super fits in the page */
1266 	if (sizeof(*disk_super) > PAGE_SIZE)
1267 		return ERR_PTR(-EINVAL);
1268 
1269 	/* make sure our super doesn't straddle pages on disk */
1270 	index = bytenr >> PAGE_SHIFT;
1271 	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1272 		return ERR_PTR(-EINVAL);
1273 
1274 	/* pull in the page with our super */
1275 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1276 
1277 	if (IS_ERR(page))
1278 		return ERR_CAST(page);
1279 
1280 	p = page_address(page);
1281 
1282 	/* align our pointer to the offset of the super block */
1283 	disk_super = p + offset_in_page(bytenr);
1284 
1285 	if (btrfs_super_bytenr(disk_super) != bytenr ||
1286 	    btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1287 		btrfs_release_disk_super(p);
1288 		return ERR_PTR(-EINVAL);
1289 	}
1290 
1291 	if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1292 		disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1293 
1294 	return disk_super;
1295 }
1296 
1297 int btrfs_forget_devices(const char *path)
1298 {
1299 	int ret;
1300 
1301 	mutex_lock(&uuid_mutex);
1302 	ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1303 	mutex_unlock(&uuid_mutex);
1304 
1305 	return ret;
1306 }
1307 
1308 /*
1309  * Look for a btrfs signature on a device. This may be called out of the mount path
1310  * and we are not allowed to call set_blocksize during the scan. The superblock
1311  * is read via pagecache
1312  */
1313 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1314 					   void *holder)
1315 {
1316 	struct btrfs_super_block *disk_super;
1317 	bool new_device_added = false;
1318 	struct btrfs_device *device = NULL;
1319 	struct block_device *bdev;
1320 	u64 bytenr;
1321 
1322 	lockdep_assert_held(&uuid_mutex);
1323 
1324 	/*
1325 	 * we would like to check all the supers, but that would make
1326 	 * a btrfs mount succeed after a mkfs from a different FS.
1327 	 * So, we need to add a special mount option to scan for
1328 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1329 	 */
1330 	bytenr = btrfs_sb_offset(0);
1331 	flags |= FMODE_EXCL;
1332 
1333 	bdev = blkdev_get_by_path(path, flags, holder);
1334 	if (IS_ERR(bdev))
1335 		return ERR_CAST(bdev);
1336 
1337 	disk_super = btrfs_read_disk_super(bdev, bytenr);
1338 	if (IS_ERR(disk_super)) {
1339 		device = ERR_CAST(disk_super);
1340 		goto error_bdev_put;
1341 	}
1342 
1343 	device = device_list_add(path, disk_super, &new_device_added);
1344 	if (!IS_ERR(device)) {
1345 		if (new_device_added)
1346 			btrfs_free_stale_devices(path, device);
1347 	}
1348 
1349 	btrfs_release_disk_super(disk_super);
1350 
1351 error_bdev_put:
1352 	blkdev_put(bdev, flags);
1353 
1354 	return device;
1355 }
1356 
1357 /*
1358  * Try to find a chunk that intersects [start, start + len] range and when one
1359  * such is found, record the end of it in *start
1360  */
1361 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1362 				    u64 len)
1363 {
1364 	u64 physical_start, physical_end;
1365 
1366 	lockdep_assert_held(&device->fs_info->chunk_mutex);
1367 
1368 	if (!find_first_extent_bit(&device->alloc_state, *start,
1369 				   &physical_start, &physical_end,
1370 				   CHUNK_ALLOCATED, NULL)) {
1371 
1372 		if (in_range(physical_start, *start, len) ||
1373 		    in_range(*start, physical_start,
1374 			     physical_end - physical_start)) {
1375 			*start = physical_end + 1;
1376 			return true;
1377 		}
1378 	}
1379 	return false;
1380 }
1381 
1382 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1383 {
1384 	switch (device->fs_devices->chunk_alloc_policy) {
1385 	case BTRFS_CHUNK_ALLOC_REGULAR:
1386 		/*
1387 		 * We don't want to overwrite the superblock on the drive nor
1388 		 * any area used by the boot loader (grub for example), so we
1389 		 * make sure to start at an offset of at least 1MB.
1390 		 */
1391 		return max_t(u64, start, SZ_1M);
1392 	default:
1393 		BUG();
1394 	}
1395 }
1396 
1397 /**
1398  * dev_extent_hole_check - check if specified hole is suitable for allocation
1399  * @device:	the device which we have the hole
1400  * @hole_start: starting position of the hole
1401  * @hole_size:	the size of the hole
1402  * @num_bytes:	the size of the free space that we need
1403  *
1404  * This function may modify @hole_start and @hole_end to reflect the suitable
1405  * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1406  */
1407 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1408 				  u64 *hole_size, u64 num_bytes)
1409 {
1410 	bool changed = false;
1411 	u64 hole_end = *hole_start + *hole_size;
1412 
1413 	/*
1414 	 * Check before we set max_hole_start, otherwise we could end up
1415 	 * sending back this offset anyway.
1416 	 */
1417 	if (contains_pending_extent(device, hole_start, *hole_size)) {
1418 		if (hole_end >= *hole_start)
1419 			*hole_size = hole_end - *hole_start;
1420 		else
1421 			*hole_size = 0;
1422 		changed = true;
1423 	}
1424 
1425 	switch (device->fs_devices->chunk_alloc_policy) {
1426 	case BTRFS_CHUNK_ALLOC_REGULAR:
1427 		/* No extra check */
1428 		break;
1429 	default:
1430 		BUG();
1431 	}
1432 
1433 	return changed;
1434 }
1435 
1436 /*
1437  * find_free_dev_extent_start - find free space in the specified device
1438  * @device:	  the device which we search the free space in
1439  * @num_bytes:	  the size of the free space that we need
1440  * @search_start: the position from which to begin the search
1441  * @start:	  store the start of the free space.
1442  * @len:	  the size of the free space. that we find, or the size
1443  *		  of the max free space if we don't find suitable free space
1444  *
1445  * this uses a pretty simple search, the expectation is that it is
1446  * called very infrequently and that a given device has a small number
1447  * of extents
1448  *
1449  * @start is used to store the start of the free space if we find. But if we
1450  * don't find suitable free space, it will be used to store the start position
1451  * of the max free space.
1452  *
1453  * @len is used to store the size of the free space that we find.
1454  * But if we don't find suitable free space, it is used to store the size of
1455  * the max free space.
1456  *
1457  * NOTE: This function will search *commit* root of device tree, and does extra
1458  * check to ensure dev extents are not double allocated.
1459  * This makes the function safe to allocate dev extents but may not report
1460  * correct usable device space, as device extent freed in current transaction
1461  * is not reported as avaiable.
1462  */
1463 static int find_free_dev_extent_start(struct btrfs_device *device,
1464 				u64 num_bytes, u64 search_start, u64 *start,
1465 				u64 *len)
1466 {
1467 	struct btrfs_fs_info *fs_info = device->fs_info;
1468 	struct btrfs_root *root = fs_info->dev_root;
1469 	struct btrfs_key key;
1470 	struct btrfs_dev_extent *dev_extent;
1471 	struct btrfs_path *path;
1472 	u64 hole_size;
1473 	u64 max_hole_start;
1474 	u64 max_hole_size;
1475 	u64 extent_end;
1476 	u64 search_end = device->total_bytes;
1477 	int ret;
1478 	int slot;
1479 	struct extent_buffer *l;
1480 
1481 	search_start = dev_extent_search_start(device, search_start);
1482 
1483 	path = btrfs_alloc_path();
1484 	if (!path)
1485 		return -ENOMEM;
1486 
1487 	max_hole_start = search_start;
1488 	max_hole_size = 0;
1489 
1490 again:
1491 	if (search_start >= search_end ||
1492 		test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1493 		ret = -ENOSPC;
1494 		goto out;
1495 	}
1496 
1497 	path->reada = READA_FORWARD;
1498 	path->search_commit_root = 1;
1499 	path->skip_locking = 1;
1500 
1501 	key.objectid = device->devid;
1502 	key.offset = search_start;
1503 	key.type = BTRFS_DEV_EXTENT_KEY;
1504 
1505 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1506 	if (ret < 0)
1507 		goto out;
1508 	if (ret > 0) {
1509 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1510 		if (ret < 0)
1511 			goto out;
1512 	}
1513 
1514 	while (1) {
1515 		l = path->nodes[0];
1516 		slot = path->slots[0];
1517 		if (slot >= btrfs_header_nritems(l)) {
1518 			ret = btrfs_next_leaf(root, path);
1519 			if (ret == 0)
1520 				continue;
1521 			if (ret < 0)
1522 				goto out;
1523 
1524 			break;
1525 		}
1526 		btrfs_item_key_to_cpu(l, &key, slot);
1527 
1528 		if (key.objectid < device->devid)
1529 			goto next;
1530 
1531 		if (key.objectid > device->devid)
1532 			break;
1533 
1534 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1535 			goto next;
1536 
1537 		if (key.offset > search_start) {
1538 			hole_size = key.offset - search_start;
1539 			dev_extent_hole_check(device, &search_start, &hole_size,
1540 					      num_bytes);
1541 
1542 			if (hole_size > max_hole_size) {
1543 				max_hole_start = search_start;
1544 				max_hole_size = hole_size;
1545 			}
1546 
1547 			/*
1548 			 * If this free space is greater than which we need,
1549 			 * it must be the max free space that we have found
1550 			 * until now, so max_hole_start must point to the start
1551 			 * of this free space and the length of this free space
1552 			 * is stored in max_hole_size. Thus, we return
1553 			 * max_hole_start and max_hole_size and go back to the
1554 			 * caller.
1555 			 */
1556 			if (hole_size >= num_bytes) {
1557 				ret = 0;
1558 				goto out;
1559 			}
1560 		}
1561 
1562 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1563 		extent_end = key.offset + btrfs_dev_extent_length(l,
1564 								  dev_extent);
1565 		if (extent_end > search_start)
1566 			search_start = extent_end;
1567 next:
1568 		path->slots[0]++;
1569 		cond_resched();
1570 	}
1571 
1572 	/*
1573 	 * At this point, search_start should be the end of
1574 	 * allocated dev extents, and when shrinking the device,
1575 	 * search_end may be smaller than search_start.
1576 	 */
1577 	if (search_end > search_start) {
1578 		hole_size = search_end - search_start;
1579 		if (dev_extent_hole_check(device, &search_start, &hole_size,
1580 					  num_bytes)) {
1581 			btrfs_release_path(path);
1582 			goto again;
1583 		}
1584 
1585 		if (hole_size > max_hole_size) {
1586 			max_hole_start = search_start;
1587 			max_hole_size = hole_size;
1588 		}
1589 	}
1590 
1591 	/* See above. */
1592 	if (max_hole_size < num_bytes)
1593 		ret = -ENOSPC;
1594 	else
1595 		ret = 0;
1596 
1597 out:
1598 	btrfs_free_path(path);
1599 	*start = max_hole_start;
1600 	if (len)
1601 		*len = max_hole_size;
1602 	return ret;
1603 }
1604 
1605 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1606 			 u64 *start, u64 *len)
1607 {
1608 	/* FIXME use last free of some kind */
1609 	return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1610 }
1611 
1612 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1613 			  struct btrfs_device *device,
1614 			  u64 start, u64 *dev_extent_len)
1615 {
1616 	struct btrfs_fs_info *fs_info = device->fs_info;
1617 	struct btrfs_root *root = fs_info->dev_root;
1618 	int ret;
1619 	struct btrfs_path *path;
1620 	struct btrfs_key key;
1621 	struct btrfs_key found_key;
1622 	struct extent_buffer *leaf = NULL;
1623 	struct btrfs_dev_extent *extent = NULL;
1624 
1625 	path = btrfs_alloc_path();
1626 	if (!path)
1627 		return -ENOMEM;
1628 
1629 	key.objectid = device->devid;
1630 	key.offset = start;
1631 	key.type = BTRFS_DEV_EXTENT_KEY;
1632 again:
1633 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1634 	if (ret > 0) {
1635 		ret = btrfs_previous_item(root, path, key.objectid,
1636 					  BTRFS_DEV_EXTENT_KEY);
1637 		if (ret)
1638 			goto out;
1639 		leaf = path->nodes[0];
1640 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1641 		extent = btrfs_item_ptr(leaf, path->slots[0],
1642 					struct btrfs_dev_extent);
1643 		BUG_ON(found_key.offset > start || found_key.offset +
1644 		       btrfs_dev_extent_length(leaf, extent) < start);
1645 		key = found_key;
1646 		btrfs_release_path(path);
1647 		goto again;
1648 	} else if (ret == 0) {
1649 		leaf = path->nodes[0];
1650 		extent = btrfs_item_ptr(leaf, path->slots[0],
1651 					struct btrfs_dev_extent);
1652 	} else {
1653 		btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1654 		goto out;
1655 	}
1656 
1657 	*dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1658 
1659 	ret = btrfs_del_item(trans, root, path);
1660 	if (ret) {
1661 		btrfs_handle_fs_error(fs_info, ret,
1662 				      "Failed to remove dev extent item");
1663 	} else {
1664 		set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1665 	}
1666 out:
1667 	btrfs_free_path(path);
1668 	return ret;
1669 }
1670 
1671 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1672 				  struct btrfs_device *device,
1673 				  u64 chunk_offset, u64 start, u64 num_bytes)
1674 {
1675 	int ret;
1676 	struct btrfs_path *path;
1677 	struct btrfs_fs_info *fs_info = device->fs_info;
1678 	struct btrfs_root *root = fs_info->dev_root;
1679 	struct btrfs_dev_extent *extent;
1680 	struct extent_buffer *leaf;
1681 	struct btrfs_key key;
1682 
1683 	WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1684 	WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1685 	path = btrfs_alloc_path();
1686 	if (!path)
1687 		return -ENOMEM;
1688 
1689 	key.objectid = device->devid;
1690 	key.offset = start;
1691 	key.type = BTRFS_DEV_EXTENT_KEY;
1692 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1693 				      sizeof(*extent));
1694 	if (ret)
1695 		goto out;
1696 
1697 	leaf = path->nodes[0];
1698 	extent = btrfs_item_ptr(leaf, path->slots[0],
1699 				struct btrfs_dev_extent);
1700 	btrfs_set_dev_extent_chunk_tree(leaf, extent,
1701 					BTRFS_CHUNK_TREE_OBJECTID);
1702 	btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1703 					    BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1704 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1705 
1706 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1707 	btrfs_mark_buffer_dirty(leaf);
1708 out:
1709 	btrfs_free_path(path);
1710 	return ret;
1711 }
1712 
1713 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1714 {
1715 	struct extent_map_tree *em_tree;
1716 	struct extent_map *em;
1717 	struct rb_node *n;
1718 	u64 ret = 0;
1719 
1720 	em_tree = &fs_info->mapping_tree;
1721 	read_lock(&em_tree->lock);
1722 	n = rb_last(&em_tree->map.rb_root);
1723 	if (n) {
1724 		em = rb_entry(n, struct extent_map, rb_node);
1725 		ret = em->start + em->len;
1726 	}
1727 	read_unlock(&em_tree->lock);
1728 
1729 	return ret;
1730 }
1731 
1732 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1733 				    u64 *devid_ret)
1734 {
1735 	int ret;
1736 	struct btrfs_key key;
1737 	struct btrfs_key found_key;
1738 	struct btrfs_path *path;
1739 
1740 	path = btrfs_alloc_path();
1741 	if (!path)
1742 		return -ENOMEM;
1743 
1744 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1745 	key.type = BTRFS_DEV_ITEM_KEY;
1746 	key.offset = (u64)-1;
1747 
1748 	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1749 	if (ret < 0)
1750 		goto error;
1751 
1752 	if (ret == 0) {
1753 		/* Corruption */
1754 		btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1755 		ret = -EUCLEAN;
1756 		goto error;
1757 	}
1758 
1759 	ret = btrfs_previous_item(fs_info->chunk_root, path,
1760 				  BTRFS_DEV_ITEMS_OBJECTID,
1761 				  BTRFS_DEV_ITEM_KEY);
1762 	if (ret) {
1763 		*devid_ret = 1;
1764 	} else {
1765 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1766 				      path->slots[0]);
1767 		*devid_ret = found_key.offset + 1;
1768 	}
1769 	ret = 0;
1770 error:
1771 	btrfs_free_path(path);
1772 	return ret;
1773 }
1774 
1775 /*
1776  * the device information is stored in the chunk root
1777  * the btrfs_device struct should be fully filled in
1778  */
1779 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1780 			    struct btrfs_device *device)
1781 {
1782 	int ret;
1783 	struct btrfs_path *path;
1784 	struct btrfs_dev_item *dev_item;
1785 	struct extent_buffer *leaf;
1786 	struct btrfs_key key;
1787 	unsigned long ptr;
1788 
1789 	path = btrfs_alloc_path();
1790 	if (!path)
1791 		return -ENOMEM;
1792 
1793 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1794 	key.type = BTRFS_DEV_ITEM_KEY;
1795 	key.offset = device->devid;
1796 
1797 	ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1798 				      &key, sizeof(*dev_item));
1799 	if (ret)
1800 		goto out;
1801 
1802 	leaf = path->nodes[0];
1803 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1804 
1805 	btrfs_set_device_id(leaf, dev_item, device->devid);
1806 	btrfs_set_device_generation(leaf, dev_item, 0);
1807 	btrfs_set_device_type(leaf, dev_item, device->type);
1808 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1809 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1810 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1811 	btrfs_set_device_total_bytes(leaf, dev_item,
1812 				     btrfs_device_get_disk_total_bytes(device));
1813 	btrfs_set_device_bytes_used(leaf, dev_item,
1814 				    btrfs_device_get_bytes_used(device));
1815 	btrfs_set_device_group(leaf, dev_item, 0);
1816 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1817 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1818 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1819 
1820 	ptr = btrfs_device_uuid(dev_item);
1821 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1822 	ptr = btrfs_device_fsid(dev_item);
1823 	write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1824 			    ptr, BTRFS_FSID_SIZE);
1825 	btrfs_mark_buffer_dirty(leaf);
1826 
1827 	ret = 0;
1828 out:
1829 	btrfs_free_path(path);
1830 	return ret;
1831 }
1832 
1833 /*
1834  * Function to update ctime/mtime for a given device path.
1835  * Mainly used for ctime/mtime based probe like libblkid.
1836  */
1837 static void update_dev_time(const char *path_name)
1838 {
1839 	struct file *filp;
1840 
1841 	filp = filp_open(path_name, O_RDWR, 0);
1842 	if (IS_ERR(filp))
1843 		return;
1844 	file_update_time(filp);
1845 	filp_close(filp, NULL);
1846 }
1847 
1848 static int btrfs_rm_dev_item(struct btrfs_device *device)
1849 {
1850 	struct btrfs_root *root = device->fs_info->chunk_root;
1851 	int ret;
1852 	struct btrfs_path *path;
1853 	struct btrfs_key key;
1854 	struct btrfs_trans_handle *trans;
1855 
1856 	path = btrfs_alloc_path();
1857 	if (!path)
1858 		return -ENOMEM;
1859 
1860 	trans = btrfs_start_transaction(root, 0);
1861 	if (IS_ERR(trans)) {
1862 		btrfs_free_path(path);
1863 		return PTR_ERR(trans);
1864 	}
1865 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1866 	key.type = BTRFS_DEV_ITEM_KEY;
1867 	key.offset = device->devid;
1868 
1869 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1870 	if (ret) {
1871 		if (ret > 0)
1872 			ret = -ENOENT;
1873 		btrfs_abort_transaction(trans, ret);
1874 		btrfs_end_transaction(trans);
1875 		goto out;
1876 	}
1877 
1878 	ret = btrfs_del_item(trans, root, path);
1879 	if (ret) {
1880 		btrfs_abort_transaction(trans, ret);
1881 		btrfs_end_transaction(trans);
1882 	}
1883 
1884 out:
1885 	btrfs_free_path(path);
1886 	if (!ret)
1887 		ret = btrfs_commit_transaction(trans);
1888 	return ret;
1889 }
1890 
1891 /*
1892  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1893  * filesystem. It's up to the caller to adjust that number regarding eg. device
1894  * replace.
1895  */
1896 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1897 		u64 num_devices)
1898 {
1899 	u64 all_avail;
1900 	unsigned seq;
1901 	int i;
1902 
1903 	do {
1904 		seq = read_seqbegin(&fs_info->profiles_lock);
1905 
1906 		all_avail = fs_info->avail_data_alloc_bits |
1907 			    fs_info->avail_system_alloc_bits |
1908 			    fs_info->avail_metadata_alloc_bits;
1909 	} while (read_seqretry(&fs_info->profiles_lock, seq));
1910 
1911 	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1912 		if (!(all_avail & btrfs_raid_array[i].bg_flag))
1913 			continue;
1914 
1915 		if (num_devices < btrfs_raid_array[i].devs_min) {
1916 			int ret = btrfs_raid_array[i].mindev_error;
1917 
1918 			if (ret)
1919 				return ret;
1920 		}
1921 	}
1922 
1923 	return 0;
1924 }
1925 
1926 static struct btrfs_device * btrfs_find_next_active_device(
1927 		struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1928 {
1929 	struct btrfs_device *next_device;
1930 
1931 	list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1932 		if (next_device != device &&
1933 		    !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1934 		    && next_device->bdev)
1935 			return next_device;
1936 	}
1937 
1938 	return NULL;
1939 }
1940 
1941 /*
1942  * Helper function to check if the given device is part of s_bdev / latest_bdev
1943  * and replace it with the provided or the next active device, in the context
1944  * where this function called, there should be always be another device (or
1945  * this_dev) which is active.
1946  */
1947 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1948 				     struct btrfs_device *this_dev)
1949 {
1950 	struct btrfs_fs_info *fs_info = device->fs_info;
1951 	struct btrfs_device *next_device;
1952 
1953 	if (this_dev)
1954 		next_device = this_dev;
1955 	else
1956 		next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1957 								device);
1958 	ASSERT(next_device);
1959 
1960 	if (fs_info->sb->s_bdev &&
1961 			(fs_info->sb->s_bdev == device->bdev))
1962 		fs_info->sb->s_bdev = next_device->bdev;
1963 
1964 	if (fs_info->fs_devices->latest_bdev == device->bdev)
1965 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1966 }
1967 
1968 /*
1969  * Return btrfs_fs_devices::num_devices excluding the device that's being
1970  * currently replaced.
1971  */
1972 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
1973 {
1974 	u64 num_devices = fs_info->fs_devices->num_devices;
1975 
1976 	down_read(&fs_info->dev_replace.rwsem);
1977 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1978 		ASSERT(num_devices > 1);
1979 		num_devices--;
1980 	}
1981 	up_read(&fs_info->dev_replace.rwsem);
1982 
1983 	return num_devices;
1984 }
1985 
1986 static void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
1987 				      struct block_device *bdev,
1988 				      const char *device_path)
1989 {
1990 	struct btrfs_super_block *disk_super;
1991 	int copy_num;
1992 
1993 	if (!bdev)
1994 		return;
1995 
1996 	for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
1997 		struct page *page;
1998 		int ret;
1999 
2000 		disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2001 		if (IS_ERR(disk_super))
2002 			continue;
2003 
2004 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2005 
2006 		page = virt_to_page(disk_super);
2007 		set_page_dirty(page);
2008 		lock_page(page);
2009 		/* write_on_page() unlocks the page */
2010 		ret = write_one_page(page);
2011 		if (ret)
2012 			btrfs_warn(fs_info,
2013 				"error clearing superblock number %d (%d)",
2014 				copy_num, ret);
2015 		btrfs_release_disk_super(disk_super);
2016 
2017 	}
2018 
2019 	/* Notify udev that device has changed */
2020 	btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2021 
2022 	/* Update ctime/mtime for device path for libblkid */
2023 	update_dev_time(device_path);
2024 }
2025 
2026 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2027 		u64 devid)
2028 {
2029 	struct btrfs_device *device;
2030 	struct btrfs_fs_devices *cur_devices;
2031 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2032 	u64 num_devices;
2033 	int ret = 0;
2034 
2035 	mutex_lock(&uuid_mutex);
2036 
2037 	num_devices = btrfs_num_devices(fs_info);
2038 
2039 	ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2040 	if (ret)
2041 		goto out;
2042 
2043 	device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2044 
2045 	if (IS_ERR(device)) {
2046 		if (PTR_ERR(device) == -ENOENT &&
2047 		    strcmp(device_path, "missing") == 0)
2048 			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2049 		else
2050 			ret = PTR_ERR(device);
2051 		goto out;
2052 	}
2053 
2054 	if (btrfs_pinned_by_swapfile(fs_info, device)) {
2055 		btrfs_warn_in_rcu(fs_info,
2056 		  "cannot remove device %s (devid %llu) due to active swapfile",
2057 				  rcu_str_deref(device->name), device->devid);
2058 		ret = -ETXTBSY;
2059 		goto out;
2060 	}
2061 
2062 	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2063 		ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2064 		goto out;
2065 	}
2066 
2067 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2068 	    fs_info->fs_devices->rw_devices == 1) {
2069 		ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2070 		goto out;
2071 	}
2072 
2073 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2074 		mutex_lock(&fs_info->chunk_mutex);
2075 		list_del_init(&device->dev_alloc_list);
2076 		device->fs_devices->rw_devices--;
2077 		mutex_unlock(&fs_info->chunk_mutex);
2078 	}
2079 
2080 	mutex_unlock(&uuid_mutex);
2081 	ret = btrfs_shrink_device(device, 0);
2082 	mutex_lock(&uuid_mutex);
2083 	if (ret)
2084 		goto error_undo;
2085 
2086 	/*
2087 	 * TODO: the superblock still includes this device in its num_devices
2088 	 * counter although write_all_supers() is not locked out. This
2089 	 * could give a filesystem state which requires a degraded mount.
2090 	 */
2091 	ret = btrfs_rm_dev_item(device);
2092 	if (ret)
2093 		goto error_undo;
2094 
2095 	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2096 	btrfs_scrub_cancel_dev(device);
2097 
2098 	/*
2099 	 * the device list mutex makes sure that we don't change
2100 	 * the device list while someone else is writing out all
2101 	 * the device supers. Whoever is writing all supers, should
2102 	 * lock the device list mutex before getting the number of
2103 	 * devices in the super block (super_copy). Conversely,
2104 	 * whoever updates the number of devices in the super block
2105 	 * (super_copy) should hold the device list mutex.
2106 	 */
2107 
2108 	/*
2109 	 * In normal cases the cur_devices == fs_devices. But in case
2110 	 * of deleting a seed device, the cur_devices should point to
2111 	 * its own fs_devices listed under the fs_devices->seed.
2112 	 */
2113 	cur_devices = device->fs_devices;
2114 	mutex_lock(&fs_devices->device_list_mutex);
2115 	list_del_rcu(&device->dev_list);
2116 
2117 	cur_devices->num_devices--;
2118 	cur_devices->total_devices--;
2119 	/* Update total_devices of the parent fs_devices if it's seed */
2120 	if (cur_devices != fs_devices)
2121 		fs_devices->total_devices--;
2122 
2123 	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2124 		cur_devices->missing_devices--;
2125 
2126 	btrfs_assign_next_active_device(device, NULL);
2127 
2128 	if (device->bdev) {
2129 		cur_devices->open_devices--;
2130 		/* remove sysfs entry */
2131 		btrfs_sysfs_remove_devices_dir(fs_devices, device);
2132 	}
2133 
2134 	num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2135 	btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2136 	mutex_unlock(&fs_devices->device_list_mutex);
2137 
2138 	/*
2139 	 * at this point, the device is zero sized and detached from
2140 	 * the devices list.  All that's left is to zero out the old
2141 	 * supers and free the device.
2142 	 */
2143 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2144 		btrfs_scratch_superblocks(fs_info, device->bdev,
2145 					  device->name->str);
2146 
2147 	btrfs_close_bdev(device);
2148 	synchronize_rcu();
2149 	btrfs_free_device(device);
2150 
2151 	if (cur_devices->open_devices == 0) {
2152 		while (fs_devices) {
2153 			if (fs_devices->seed == cur_devices) {
2154 				fs_devices->seed = cur_devices->seed;
2155 				break;
2156 			}
2157 			fs_devices = fs_devices->seed;
2158 		}
2159 		cur_devices->seed = NULL;
2160 		close_fs_devices(cur_devices);
2161 		free_fs_devices(cur_devices);
2162 	}
2163 
2164 out:
2165 	mutex_unlock(&uuid_mutex);
2166 	return ret;
2167 
2168 error_undo:
2169 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2170 		mutex_lock(&fs_info->chunk_mutex);
2171 		list_add(&device->dev_alloc_list,
2172 			 &fs_devices->alloc_list);
2173 		device->fs_devices->rw_devices++;
2174 		mutex_unlock(&fs_info->chunk_mutex);
2175 	}
2176 	goto out;
2177 }
2178 
2179 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2180 {
2181 	struct btrfs_fs_devices *fs_devices;
2182 
2183 	lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2184 
2185 	/*
2186 	 * in case of fs with no seed, srcdev->fs_devices will point
2187 	 * to fs_devices of fs_info. However when the dev being replaced is
2188 	 * a seed dev it will point to the seed's local fs_devices. In short
2189 	 * srcdev will have its correct fs_devices in both the cases.
2190 	 */
2191 	fs_devices = srcdev->fs_devices;
2192 
2193 	list_del_rcu(&srcdev->dev_list);
2194 	list_del(&srcdev->dev_alloc_list);
2195 	fs_devices->num_devices--;
2196 	if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2197 		fs_devices->missing_devices--;
2198 
2199 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2200 		fs_devices->rw_devices--;
2201 
2202 	if (srcdev->bdev)
2203 		fs_devices->open_devices--;
2204 }
2205 
2206 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2207 {
2208 	struct btrfs_fs_info *fs_info = srcdev->fs_info;
2209 	struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2210 
2211 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2212 		/* zero out the old super if it is writable */
2213 		btrfs_scratch_superblocks(fs_info, srcdev->bdev,
2214 					  srcdev->name->str);
2215 	}
2216 
2217 	btrfs_close_bdev(srcdev);
2218 	synchronize_rcu();
2219 	btrfs_free_device(srcdev);
2220 
2221 	/* if this is no devs we rather delete the fs_devices */
2222 	if (!fs_devices->num_devices) {
2223 		struct btrfs_fs_devices *tmp_fs_devices;
2224 
2225 		/*
2226 		 * On a mounted FS, num_devices can't be zero unless it's a
2227 		 * seed. In case of a seed device being replaced, the replace
2228 		 * target added to the sprout FS, so there will be no more
2229 		 * device left under the seed FS.
2230 		 */
2231 		ASSERT(fs_devices->seeding);
2232 
2233 		tmp_fs_devices = fs_info->fs_devices;
2234 		while (tmp_fs_devices) {
2235 			if (tmp_fs_devices->seed == fs_devices) {
2236 				tmp_fs_devices->seed = fs_devices->seed;
2237 				break;
2238 			}
2239 			tmp_fs_devices = tmp_fs_devices->seed;
2240 		}
2241 		fs_devices->seed = NULL;
2242 		close_fs_devices(fs_devices);
2243 		free_fs_devices(fs_devices);
2244 	}
2245 }
2246 
2247 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2248 {
2249 	struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2250 
2251 	mutex_lock(&fs_devices->device_list_mutex);
2252 
2253 	btrfs_sysfs_remove_devices_dir(fs_devices, tgtdev);
2254 
2255 	if (tgtdev->bdev)
2256 		fs_devices->open_devices--;
2257 
2258 	fs_devices->num_devices--;
2259 
2260 	btrfs_assign_next_active_device(tgtdev, NULL);
2261 
2262 	list_del_rcu(&tgtdev->dev_list);
2263 
2264 	mutex_unlock(&fs_devices->device_list_mutex);
2265 
2266 	/*
2267 	 * The update_dev_time() with in btrfs_scratch_superblocks()
2268 	 * may lead to a call to btrfs_show_devname() which will try
2269 	 * to hold device_list_mutex. And here this device
2270 	 * is already out of device list, so we don't have to hold
2271 	 * the device_list_mutex lock.
2272 	 */
2273 	btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2274 				  tgtdev->name->str);
2275 
2276 	btrfs_close_bdev(tgtdev);
2277 	synchronize_rcu();
2278 	btrfs_free_device(tgtdev);
2279 }
2280 
2281 static struct btrfs_device *btrfs_find_device_by_path(
2282 		struct btrfs_fs_info *fs_info, const char *device_path)
2283 {
2284 	int ret = 0;
2285 	struct btrfs_super_block *disk_super;
2286 	u64 devid;
2287 	u8 *dev_uuid;
2288 	struct block_device *bdev;
2289 	struct btrfs_device *device;
2290 
2291 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2292 				    fs_info->bdev_holder, 0, &bdev, &disk_super);
2293 	if (ret)
2294 		return ERR_PTR(ret);
2295 
2296 	devid = btrfs_stack_device_id(&disk_super->dev_item);
2297 	dev_uuid = disk_super->dev_item.uuid;
2298 	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2299 		device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2300 					   disk_super->metadata_uuid, true);
2301 	else
2302 		device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2303 					   disk_super->fsid, true);
2304 
2305 	btrfs_release_disk_super(disk_super);
2306 	if (!device)
2307 		device = ERR_PTR(-ENOENT);
2308 	blkdev_put(bdev, FMODE_READ);
2309 	return device;
2310 }
2311 
2312 /*
2313  * Lookup a device given by device id, or the path if the id is 0.
2314  */
2315 struct btrfs_device *btrfs_find_device_by_devspec(
2316 		struct btrfs_fs_info *fs_info, u64 devid,
2317 		const char *device_path)
2318 {
2319 	struct btrfs_device *device;
2320 
2321 	if (devid) {
2322 		device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2323 					   NULL, true);
2324 		if (!device)
2325 			return ERR_PTR(-ENOENT);
2326 		return device;
2327 	}
2328 
2329 	if (!device_path || !device_path[0])
2330 		return ERR_PTR(-EINVAL);
2331 
2332 	if (strcmp(device_path, "missing") == 0) {
2333 		/* Find first missing device */
2334 		list_for_each_entry(device, &fs_info->fs_devices->devices,
2335 				    dev_list) {
2336 			if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2337 				     &device->dev_state) && !device->bdev)
2338 				return device;
2339 		}
2340 		return ERR_PTR(-ENOENT);
2341 	}
2342 
2343 	return btrfs_find_device_by_path(fs_info, device_path);
2344 }
2345 
2346 /*
2347  * does all the dirty work required for changing file system's UUID.
2348  */
2349 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2350 {
2351 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2352 	struct btrfs_fs_devices *old_devices;
2353 	struct btrfs_fs_devices *seed_devices;
2354 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2355 	struct btrfs_device *device;
2356 	u64 super_flags;
2357 
2358 	lockdep_assert_held(&uuid_mutex);
2359 	if (!fs_devices->seeding)
2360 		return -EINVAL;
2361 
2362 	seed_devices = alloc_fs_devices(NULL, NULL);
2363 	if (IS_ERR(seed_devices))
2364 		return PTR_ERR(seed_devices);
2365 
2366 	old_devices = clone_fs_devices(fs_devices);
2367 	if (IS_ERR(old_devices)) {
2368 		kfree(seed_devices);
2369 		return PTR_ERR(old_devices);
2370 	}
2371 
2372 	list_add(&old_devices->fs_list, &fs_uuids);
2373 
2374 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2375 	seed_devices->opened = 1;
2376 	INIT_LIST_HEAD(&seed_devices->devices);
2377 	INIT_LIST_HEAD(&seed_devices->alloc_list);
2378 	mutex_init(&seed_devices->device_list_mutex);
2379 
2380 	mutex_lock(&fs_devices->device_list_mutex);
2381 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2382 			      synchronize_rcu);
2383 	list_for_each_entry(device, &seed_devices->devices, dev_list)
2384 		device->fs_devices = seed_devices;
2385 
2386 	mutex_lock(&fs_info->chunk_mutex);
2387 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2388 	mutex_unlock(&fs_info->chunk_mutex);
2389 
2390 	fs_devices->seeding = false;
2391 	fs_devices->num_devices = 0;
2392 	fs_devices->open_devices = 0;
2393 	fs_devices->missing_devices = 0;
2394 	fs_devices->rotating = false;
2395 	fs_devices->seed = seed_devices;
2396 
2397 	generate_random_uuid(fs_devices->fsid);
2398 	memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2399 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2400 	mutex_unlock(&fs_devices->device_list_mutex);
2401 
2402 	super_flags = btrfs_super_flags(disk_super) &
2403 		      ~BTRFS_SUPER_FLAG_SEEDING;
2404 	btrfs_set_super_flags(disk_super, super_flags);
2405 
2406 	return 0;
2407 }
2408 
2409 /*
2410  * Store the expected generation for seed devices in device items.
2411  */
2412 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2413 {
2414 	struct btrfs_fs_info *fs_info = trans->fs_info;
2415 	struct btrfs_root *root = fs_info->chunk_root;
2416 	struct btrfs_path *path;
2417 	struct extent_buffer *leaf;
2418 	struct btrfs_dev_item *dev_item;
2419 	struct btrfs_device *device;
2420 	struct btrfs_key key;
2421 	u8 fs_uuid[BTRFS_FSID_SIZE];
2422 	u8 dev_uuid[BTRFS_UUID_SIZE];
2423 	u64 devid;
2424 	int ret;
2425 
2426 	path = btrfs_alloc_path();
2427 	if (!path)
2428 		return -ENOMEM;
2429 
2430 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2431 	key.offset = 0;
2432 	key.type = BTRFS_DEV_ITEM_KEY;
2433 
2434 	while (1) {
2435 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2436 		if (ret < 0)
2437 			goto error;
2438 
2439 		leaf = path->nodes[0];
2440 next_slot:
2441 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2442 			ret = btrfs_next_leaf(root, path);
2443 			if (ret > 0)
2444 				break;
2445 			if (ret < 0)
2446 				goto error;
2447 			leaf = path->nodes[0];
2448 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2449 			btrfs_release_path(path);
2450 			continue;
2451 		}
2452 
2453 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2454 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2455 		    key.type != BTRFS_DEV_ITEM_KEY)
2456 			break;
2457 
2458 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
2459 					  struct btrfs_dev_item);
2460 		devid = btrfs_device_id(leaf, dev_item);
2461 		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2462 				   BTRFS_UUID_SIZE);
2463 		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2464 				   BTRFS_FSID_SIZE);
2465 		device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2466 					   fs_uuid, true);
2467 		BUG_ON(!device); /* Logic error */
2468 
2469 		if (device->fs_devices->seeding) {
2470 			btrfs_set_device_generation(leaf, dev_item,
2471 						    device->generation);
2472 			btrfs_mark_buffer_dirty(leaf);
2473 		}
2474 
2475 		path->slots[0]++;
2476 		goto next_slot;
2477 	}
2478 	ret = 0;
2479 error:
2480 	btrfs_free_path(path);
2481 	return ret;
2482 }
2483 
2484 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2485 {
2486 	struct btrfs_root *root = fs_info->dev_root;
2487 	struct request_queue *q;
2488 	struct btrfs_trans_handle *trans;
2489 	struct btrfs_device *device;
2490 	struct block_device *bdev;
2491 	struct super_block *sb = fs_info->sb;
2492 	struct rcu_string *name;
2493 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2494 	u64 orig_super_total_bytes;
2495 	u64 orig_super_num_devices;
2496 	int seeding_dev = 0;
2497 	int ret = 0;
2498 	bool unlocked = false;
2499 
2500 	if (sb_rdonly(sb) && !fs_devices->seeding)
2501 		return -EROFS;
2502 
2503 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2504 				  fs_info->bdev_holder);
2505 	if (IS_ERR(bdev))
2506 		return PTR_ERR(bdev);
2507 
2508 	if (fs_devices->seeding) {
2509 		seeding_dev = 1;
2510 		down_write(&sb->s_umount);
2511 		mutex_lock(&uuid_mutex);
2512 	}
2513 
2514 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2515 
2516 	mutex_lock(&fs_devices->device_list_mutex);
2517 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
2518 		if (device->bdev == bdev) {
2519 			ret = -EEXIST;
2520 			mutex_unlock(
2521 				&fs_devices->device_list_mutex);
2522 			goto error;
2523 		}
2524 	}
2525 	mutex_unlock(&fs_devices->device_list_mutex);
2526 
2527 	device = btrfs_alloc_device(fs_info, NULL, NULL);
2528 	if (IS_ERR(device)) {
2529 		/* we can safely leave the fs_devices entry around */
2530 		ret = PTR_ERR(device);
2531 		goto error;
2532 	}
2533 
2534 	name = rcu_string_strdup(device_path, GFP_KERNEL);
2535 	if (!name) {
2536 		ret = -ENOMEM;
2537 		goto error_free_device;
2538 	}
2539 	rcu_assign_pointer(device->name, name);
2540 
2541 	trans = btrfs_start_transaction(root, 0);
2542 	if (IS_ERR(trans)) {
2543 		ret = PTR_ERR(trans);
2544 		goto error_free_device;
2545 	}
2546 
2547 	q = bdev_get_queue(bdev);
2548 	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2549 	device->generation = trans->transid;
2550 	device->io_width = fs_info->sectorsize;
2551 	device->io_align = fs_info->sectorsize;
2552 	device->sector_size = fs_info->sectorsize;
2553 	device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2554 					 fs_info->sectorsize);
2555 	device->disk_total_bytes = device->total_bytes;
2556 	device->commit_total_bytes = device->total_bytes;
2557 	device->fs_info = fs_info;
2558 	device->bdev = bdev;
2559 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2560 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2561 	device->mode = FMODE_EXCL;
2562 	device->dev_stats_valid = 1;
2563 	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2564 
2565 	if (seeding_dev) {
2566 		sb->s_flags &= ~SB_RDONLY;
2567 		ret = btrfs_prepare_sprout(fs_info);
2568 		if (ret) {
2569 			btrfs_abort_transaction(trans, ret);
2570 			goto error_trans;
2571 		}
2572 	}
2573 
2574 	device->fs_devices = fs_devices;
2575 
2576 	mutex_lock(&fs_devices->device_list_mutex);
2577 	mutex_lock(&fs_info->chunk_mutex);
2578 	list_add_rcu(&device->dev_list, &fs_devices->devices);
2579 	list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2580 	fs_devices->num_devices++;
2581 	fs_devices->open_devices++;
2582 	fs_devices->rw_devices++;
2583 	fs_devices->total_devices++;
2584 	fs_devices->total_rw_bytes += device->total_bytes;
2585 
2586 	atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2587 
2588 	if (!blk_queue_nonrot(q))
2589 		fs_devices->rotating = true;
2590 
2591 	orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2592 	btrfs_set_super_total_bytes(fs_info->super_copy,
2593 		round_down(orig_super_total_bytes + device->total_bytes,
2594 			   fs_info->sectorsize));
2595 
2596 	orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2597 	btrfs_set_super_num_devices(fs_info->super_copy,
2598 				    orig_super_num_devices + 1);
2599 
2600 	/* add sysfs device entry */
2601 	btrfs_sysfs_add_devices_dir(fs_devices, device);
2602 
2603 	/*
2604 	 * we've got more storage, clear any full flags on the space
2605 	 * infos
2606 	 */
2607 	btrfs_clear_space_info_full(fs_info);
2608 
2609 	mutex_unlock(&fs_info->chunk_mutex);
2610 	mutex_unlock(&fs_devices->device_list_mutex);
2611 
2612 	if (seeding_dev) {
2613 		mutex_lock(&fs_info->chunk_mutex);
2614 		ret = init_first_rw_device(trans);
2615 		mutex_unlock(&fs_info->chunk_mutex);
2616 		if (ret) {
2617 			btrfs_abort_transaction(trans, ret);
2618 			goto error_sysfs;
2619 		}
2620 	}
2621 
2622 	ret = btrfs_add_dev_item(trans, device);
2623 	if (ret) {
2624 		btrfs_abort_transaction(trans, ret);
2625 		goto error_sysfs;
2626 	}
2627 
2628 	if (seeding_dev) {
2629 		ret = btrfs_finish_sprout(trans);
2630 		if (ret) {
2631 			btrfs_abort_transaction(trans, ret);
2632 			goto error_sysfs;
2633 		}
2634 
2635 		btrfs_sysfs_update_sprout_fsid(fs_devices,
2636 				fs_info->fs_devices->fsid);
2637 	}
2638 
2639 	ret = btrfs_commit_transaction(trans);
2640 
2641 	if (seeding_dev) {
2642 		mutex_unlock(&uuid_mutex);
2643 		up_write(&sb->s_umount);
2644 		unlocked = true;
2645 
2646 		if (ret) /* transaction commit */
2647 			return ret;
2648 
2649 		ret = btrfs_relocate_sys_chunks(fs_info);
2650 		if (ret < 0)
2651 			btrfs_handle_fs_error(fs_info, ret,
2652 				    "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2653 		trans = btrfs_attach_transaction(root);
2654 		if (IS_ERR(trans)) {
2655 			if (PTR_ERR(trans) == -ENOENT)
2656 				return 0;
2657 			ret = PTR_ERR(trans);
2658 			trans = NULL;
2659 			goto error_sysfs;
2660 		}
2661 		ret = btrfs_commit_transaction(trans);
2662 	}
2663 
2664 	/*
2665 	 * Now that we have written a new super block to this device, check all
2666 	 * other fs_devices list if device_path alienates any other scanned
2667 	 * device.
2668 	 * We can ignore the return value as it typically returns -EINVAL and
2669 	 * only succeeds if the device was an alien.
2670 	 */
2671 	btrfs_forget_devices(device_path);
2672 
2673 	/* Update ctime/mtime for blkid or udev */
2674 	update_dev_time(device_path);
2675 
2676 	return ret;
2677 
2678 error_sysfs:
2679 	btrfs_sysfs_remove_devices_dir(fs_devices, device);
2680 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2681 	mutex_lock(&fs_info->chunk_mutex);
2682 	list_del_rcu(&device->dev_list);
2683 	list_del(&device->dev_alloc_list);
2684 	fs_info->fs_devices->num_devices--;
2685 	fs_info->fs_devices->open_devices--;
2686 	fs_info->fs_devices->rw_devices--;
2687 	fs_info->fs_devices->total_devices--;
2688 	fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2689 	atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2690 	btrfs_set_super_total_bytes(fs_info->super_copy,
2691 				    orig_super_total_bytes);
2692 	btrfs_set_super_num_devices(fs_info->super_copy,
2693 				    orig_super_num_devices);
2694 	mutex_unlock(&fs_info->chunk_mutex);
2695 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2696 error_trans:
2697 	if (seeding_dev)
2698 		sb->s_flags |= SB_RDONLY;
2699 	if (trans)
2700 		btrfs_end_transaction(trans);
2701 error_free_device:
2702 	btrfs_free_device(device);
2703 error:
2704 	blkdev_put(bdev, FMODE_EXCL);
2705 	if (seeding_dev && !unlocked) {
2706 		mutex_unlock(&uuid_mutex);
2707 		up_write(&sb->s_umount);
2708 	}
2709 	return ret;
2710 }
2711 
2712 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2713 					struct btrfs_device *device)
2714 {
2715 	int ret;
2716 	struct btrfs_path *path;
2717 	struct btrfs_root *root = device->fs_info->chunk_root;
2718 	struct btrfs_dev_item *dev_item;
2719 	struct extent_buffer *leaf;
2720 	struct btrfs_key key;
2721 
2722 	path = btrfs_alloc_path();
2723 	if (!path)
2724 		return -ENOMEM;
2725 
2726 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2727 	key.type = BTRFS_DEV_ITEM_KEY;
2728 	key.offset = device->devid;
2729 
2730 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2731 	if (ret < 0)
2732 		goto out;
2733 
2734 	if (ret > 0) {
2735 		ret = -ENOENT;
2736 		goto out;
2737 	}
2738 
2739 	leaf = path->nodes[0];
2740 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2741 
2742 	btrfs_set_device_id(leaf, dev_item, device->devid);
2743 	btrfs_set_device_type(leaf, dev_item, device->type);
2744 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2745 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2746 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2747 	btrfs_set_device_total_bytes(leaf, dev_item,
2748 				     btrfs_device_get_disk_total_bytes(device));
2749 	btrfs_set_device_bytes_used(leaf, dev_item,
2750 				    btrfs_device_get_bytes_used(device));
2751 	btrfs_mark_buffer_dirty(leaf);
2752 
2753 out:
2754 	btrfs_free_path(path);
2755 	return ret;
2756 }
2757 
2758 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2759 		      struct btrfs_device *device, u64 new_size)
2760 {
2761 	struct btrfs_fs_info *fs_info = device->fs_info;
2762 	struct btrfs_super_block *super_copy = fs_info->super_copy;
2763 	u64 old_total;
2764 	u64 diff;
2765 
2766 	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2767 		return -EACCES;
2768 
2769 	new_size = round_down(new_size, fs_info->sectorsize);
2770 
2771 	mutex_lock(&fs_info->chunk_mutex);
2772 	old_total = btrfs_super_total_bytes(super_copy);
2773 	diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2774 
2775 	if (new_size <= device->total_bytes ||
2776 	    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2777 		mutex_unlock(&fs_info->chunk_mutex);
2778 		return -EINVAL;
2779 	}
2780 
2781 	btrfs_set_super_total_bytes(super_copy,
2782 			round_down(old_total + diff, fs_info->sectorsize));
2783 	device->fs_devices->total_rw_bytes += diff;
2784 
2785 	btrfs_device_set_total_bytes(device, new_size);
2786 	btrfs_device_set_disk_total_bytes(device, new_size);
2787 	btrfs_clear_space_info_full(device->fs_info);
2788 	if (list_empty(&device->post_commit_list))
2789 		list_add_tail(&device->post_commit_list,
2790 			      &trans->transaction->dev_update_list);
2791 	mutex_unlock(&fs_info->chunk_mutex);
2792 
2793 	return btrfs_update_device(trans, device);
2794 }
2795 
2796 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2797 {
2798 	struct btrfs_fs_info *fs_info = trans->fs_info;
2799 	struct btrfs_root *root = fs_info->chunk_root;
2800 	int ret;
2801 	struct btrfs_path *path;
2802 	struct btrfs_key key;
2803 
2804 	path = btrfs_alloc_path();
2805 	if (!path)
2806 		return -ENOMEM;
2807 
2808 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2809 	key.offset = chunk_offset;
2810 	key.type = BTRFS_CHUNK_ITEM_KEY;
2811 
2812 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2813 	if (ret < 0)
2814 		goto out;
2815 	else if (ret > 0) { /* Logic error or corruption */
2816 		btrfs_handle_fs_error(fs_info, -ENOENT,
2817 				      "Failed lookup while freeing chunk.");
2818 		ret = -ENOENT;
2819 		goto out;
2820 	}
2821 
2822 	ret = btrfs_del_item(trans, root, path);
2823 	if (ret < 0)
2824 		btrfs_handle_fs_error(fs_info, ret,
2825 				      "Failed to delete chunk item.");
2826 out:
2827 	btrfs_free_path(path);
2828 	return ret;
2829 }
2830 
2831 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2832 {
2833 	struct btrfs_super_block *super_copy = fs_info->super_copy;
2834 	struct btrfs_disk_key *disk_key;
2835 	struct btrfs_chunk *chunk;
2836 	u8 *ptr;
2837 	int ret = 0;
2838 	u32 num_stripes;
2839 	u32 array_size;
2840 	u32 len = 0;
2841 	u32 cur;
2842 	struct btrfs_key key;
2843 
2844 	mutex_lock(&fs_info->chunk_mutex);
2845 	array_size = btrfs_super_sys_array_size(super_copy);
2846 
2847 	ptr = super_copy->sys_chunk_array;
2848 	cur = 0;
2849 
2850 	while (cur < array_size) {
2851 		disk_key = (struct btrfs_disk_key *)ptr;
2852 		btrfs_disk_key_to_cpu(&key, disk_key);
2853 
2854 		len = sizeof(*disk_key);
2855 
2856 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2857 			chunk = (struct btrfs_chunk *)(ptr + len);
2858 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2859 			len += btrfs_chunk_item_size(num_stripes);
2860 		} else {
2861 			ret = -EIO;
2862 			break;
2863 		}
2864 		if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2865 		    key.offset == chunk_offset) {
2866 			memmove(ptr, ptr + len, array_size - (cur + len));
2867 			array_size -= len;
2868 			btrfs_set_super_sys_array_size(super_copy, array_size);
2869 		} else {
2870 			ptr += len;
2871 			cur += len;
2872 		}
2873 	}
2874 	mutex_unlock(&fs_info->chunk_mutex);
2875 	return ret;
2876 }
2877 
2878 /*
2879  * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2880  * @logical: Logical block offset in bytes.
2881  * @length: Length of extent in bytes.
2882  *
2883  * Return: Chunk mapping or ERR_PTR.
2884  */
2885 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2886 				       u64 logical, u64 length)
2887 {
2888 	struct extent_map_tree *em_tree;
2889 	struct extent_map *em;
2890 
2891 	em_tree = &fs_info->mapping_tree;
2892 	read_lock(&em_tree->lock);
2893 	em = lookup_extent_mapping(em_tree, logical, length);
2894 	read_unlock(&em_tree->lock);
2895 
2896 	if (!em) {
2897 		btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2898 			   logical, length);
2899 		return ERR_PTR(-EINVAL);
2900 	}
2901 
2902 	if (em->start > logical || em->start + em->len < logical) {
2903 		btrfs_crit(fs_info,
2904 			   "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2905 			   logical, length, em->start, em->start + em->len);
2906 		free_extent_map(em);
2907 		return ERR_PTR(-EINVAL);
2908 	}
2909 
2910 	/* callers are responsible for dropping em's ref. */
2911 	return em;
2912 }
2913 
2914 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2915 {
2916 	struct btrfs_fs_info *fs_info = trans->fs_info;
2917 	struct extent_map *em;
2918 	struct map_lookup *map;
2919 	u64 dev_extent_len = 0;
2920 	int i, ret = 0;
2921 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2922 
2923 	em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2924 	if (IS_ERR(em)) {
2925 		/*
2926 		 * This is a logic error, but we don't want to just rely on the
2927 		 * user having built with ASSERT enabled, so if ASSERT doesn't
2928 		 * do anything we still error out.
2929 		 */
2930 		ASSERT(0);
2931 		return PTR_ERR(em);
2932 	}
2933 	map = em->map_lookup;
2934 	mutex_lock(&fs_info->chunk_mutex);
2935 	check_system_chunk(trans, map->type);
2936 	mutex_unlock(&fs_info->chunk_mutex);
2937 
2938 	/*
2939 	 * Take the device list mutex to prevent races with the final phase of
2940 	 * a device replace operation that replaces the device object associated
2941 	 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2942 	 */
2943 	mutex_lock(&fs_devices->device_list_mutex);
2944 	for (i = 0; i < map->num_stripes; i++) {
2945 		struct btrfs_device *device = map->stripes[i].dev;
2946 		ret = btrfs_free_dev_extent(trans, device,
2947 					    map->stripes[i].physical,
2948 					    &dev_extent_len);
2949 		if (ret) {
2950 			mutex_unlock(&fs_devices->device_list_mutex);
2951 			btrfs_abort_transaction(trans, ret);
2952 			goto out;
2953 		}
2954 
2955 		if (device->bytes_used > 0) {
2956 			mutex_lock(&fs_info->chunk_mutex);
2957 			btrfs_device_set_bytes_used(device,
2958 					device->bytes_used - dev_extent_len);
2959 			atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2960 			btrfs_clear_space_info_full(fs_info);
2961 			mutex_unlock(&fs_info->chunk_mutex);
2962 		}
2963 
2964 		ret = btrfs_update_device(trans, device);
2965 		if (ret) {
2966 			mutex_unlock(&fs_devices->device_list_mutex);
2967 			btrfs_abort_transaction(trans, ret);
2968 			goto out;
2969 		}
2970 	}
2971 	mutex_unlock(&fs_devices->device_list_mutex);
2972 
2973 	ret = btrfs_free_chunk(trans, chunk_offset);
2974 	if (ret) {
2975 		btrfs_abort_transaction(trans, ret);
2976 		goto out;
2977 	}
2978 
2979 	trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2980 
2981 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2982 		ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2983 		if (ret) {
2984 			btrfs_abort_transaction(trans, ret);
2985 			goto out;
2986 		}
2987 	}
2988 
2989 	ret = btrfs_remove_block_group(trans, chunk_offset, em);
2990 	if (ret) {
2991 		btrfs_abort_transaction(trans, ret);
2992 		goto out;
2993 	}
2994 
2995 out:
2996 	/* once for us */
2997 	free_extent_map(em);
2998 	return ret;
2999 }
3000 
3001 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3002 {
3003 	struct btrfs_root *root = fs_info->chunk_root;
3004 	struct btrfs_trans_handle *trans;
3005 	struct btrfs_block_group *block_group;
3006 	int ret;
3007 
3008 	/*
3009 	 * Prevent races with automatic removal of unused block groups.
3010 	 * After we relocate and before we remove the chunk with offset
3011 	 * chunk_offset, automatic removal of the block group can kick in,
3012 	 * resulting in a failure when calling btrfs_remove_chunk() below.
3013 	 *
3014 	 * Make sure to acquire this mutex before doing a tree search (dev
3015 	 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3016 	 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3017 	 * we release the path used to search the chunk/dev tree and before
3018 	 * the current task acquires this mutex and calls us.
3019 	 */
3020 	lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3021 
3022 	/* step one, relocate all the extents inside this chunk */
3023 	btrfs_scrub_pause(fs_info);
3024 	ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3025 	btrfs_scrub_continue(fs_info);
3026 	if (ret)
3027 		return ret;
3028 
3029 	block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3030 	if (!block_group)
3031 		return -ENOENT;
3032 	btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3033 	btrfs_put_block_group(block_group);
3034 
3035 	trans = btrfs_start_trans_remove_block_group(root->fs_info,
3036 						     chunk_offset);
3037 	if (IS_ERR(trans)) {
3038 		ret = PTR_ERR(trans);
3039 		btrfs_handle_fs_error(root->fs_info, ret, NULL);
3040 		return ret;
3041 	}
3042 
3043 	/*
3044 	 * step two, delete the device extents and the
3045 	 * chunk tree entries
3046 	 */
3047 	ret = btrfs_remove_chunk(trans, chunk_offset);
3048 	btrfs_end_transaction(trans);
3049 	return ret;
3050 }
3051 
3052 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3053 {
3054 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3055 	struct btrfs_path *path;
3056 	struct extent_buffer *leaf;
3057 	struct btrfs_chunk *chunk;
3058 	struct btrfs_key key;
3059 	struct btrfs_key found_key;
3060 	u64 chunk_type;
3061 	bool retried = false;
3062 	int failed = 0;
3063 	int ret;
3064 
3065 	path = btrfs_alloc_path();
3066 	if (!path)
3067 		return -ENOMEM;
3068 
3069 again:
3070 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3071 	key.offset = (u64)-1;
3072 	key.type = BTRFS_CHUNK_ITEM_KEY;
3073 
3074 	while (1) {
3075 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
3076 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3077 		if (ret < 0) {
3078 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3079 			goto error;
3080 		}
3081 		BUG_ON(ret == 0); /* Corruption */
3082 
3083 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
3084 					  key.type);
3085 		if (ret)
3086 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3087 		if (ret < 0)
3088 			goto error;
3089 		if (ret > 0)
3090 			break;
3091 
3092 		leaf = path->nodes[0];
3093 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3094 
3095 		chunk = btrfs_item_ptr(leaf, path->slots[0],
3096 				       struct btrfs_chunk);
3097 		chunk_type = btrfs_chunk_type(leaf, chunk);
3098 		btrfs_release_path(path);
3099 
3100 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3101 			ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3102 			if (ret == -ENOSPC)
3103 				failed++;
3104 			else
3105 				BUG_ON(ret);
3106 		}
3107 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3108 
3109 		if (found_key.offset == 0)
3110 			break;
3111 		key.offset = found_key.offset - 1;
3112 	}
3113 	ret = 0;
3114 	if (failed && !retried) {
3115 		failed = 0;
3116 		retried = true;
3117 		goto again;
3118 	} else if (WARN_ON(failed && retried)) {
3119 		ret = -ENOSPC;
3120 	}
3121 error:
3122 	btrfs_free_path(path);
3123 	return ret;
3124 }
3125 
3126 /*
3127  * return 1 : allocate a data chunk successfully,
3128  * return <0: errors during allocating a data chunk,
3129  * return 0 : no need to allocate a data chunk.
3130  */
3131 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3132 				      u64 chunk_offset)
3133 {
3134 	struct btrfs_block_group *cache;
3135 	u64 bytes_used;
3136 	u64 chunk_type;
3137 
3138 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3139 	ASSERT(cache);
3140 	chunk_type = cache->flags;
3141 	btrfs_put_block_group(cache);
3142 
3143 	if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3144 		return 0;
3145 
3146 	spin_lock(&fs_info->data_sinfo->lock);
3147 	bytes_used = fs_info->data_sinfo->bytes_used;
3148 	spin_unlock(&fs_info->data_sinfo->lock);
3149 
3150 	if (!bytes_used) {
3151 		struct btrfs_trans_handle *trans;
3152 		int ret;
3153 
3154 		trans =	btrfs_join_transaction(fs_info->tree_root);
3155 		if (IS_ERR(trans))
3156 			return PTR_ERR(trans);
3157 
3158 		ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3159 		btrfs_end_transaction(trans);
3160 		if (ret < 0)
3161 			return ret;
3162 		return 1;
3163 	}
3164 
3165 	return 0;
3166 }
3167 
3168 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3169 			       struct btrfs_balance_control *bctl)
3170 {
3171 	struct btrfs_root *root = fs_info->tree_root;
3172 	struct btrfs_trans_handle *trans;
3173 	struct btrfs_balance_item *item;
3174 	struct btrfs_disk_balance_args disk_bargs;
3175 	struct btrfs_path *path;
3176 	struct extent_buffer *leaf;
3177 	struct btrfs_key key;
3178 	int ret, err;
3179 
3180 	path = btrfs_alloc_path();
3181 	if (!path)
3182 		return -ENOMEM;
3183 
3184 	trans = btrfs_start_transaction(root, 0);
3185 	if (IS_ERR(trans)) {
3186 		btrfs_free_path(path);
3187 		return PTR_ERR(trans);
3188 	}
3189 
3190 	key.objectid = BTRFS_BALANCE_OBJECTID;
3191 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3192 	key.offset = 0;
3193 
3194 	ret = btrfs_insert_empty_item(trans, root, path, &key,
3195 				      sizeof(*item));
3196 	if (ret)
3197 		goto out;
3198 
3199 	leaf = path->nodes[0];
3200 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3201 
3202 	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3203 
3204 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3205 	btrfs_set_balance_data(leaf, item, &disk_bargs);
3206 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3207 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
3208 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3209 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
3210 
3211 	btrfs_set_balance_flags(leaf, item, bctl->flags);
3212 
3213 	btrfs_mark_buffer_dirty(leaf);
3214 out:
3215 	btrfs_free_path(path);
3216 	err = btrfs_commit_transaction(trans);
3217 	if (err && !ret)
3218 		ret = err;
3219 	return ret;
3220 }
3221 
3222 static int del_balance_item(struct btrfs_fs_info *fs_info)
3223 {
3224 	struct btrfs_root *root = fs_info->tree_root;
3225 	struct btrfs_trans_handle *trans;
3226 	struct btrfs_path *path;
3227 	struct btrfs_key key;
3228 	int ret, err;
3229 
3230 	path = btrfs_alloc_path();
3231 	if (!path)
3232 		return -ENOMEM;
3233 
3234 	trans = btrfs_start_transaction(root, 0);
3235 	if (IS_ERR(trans)) {
3236 		btrfs_free_path(path);
3237 		return PTR_ERR(trans);
3238 	}
3239 
3240 	key.objectid = BTRFS_BALANCE_OBJECTID;
3241 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3242 	key.offset = 0;
3243 
3244 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3245 	if (ret < 0)
3246 		goto out;
3247 	if (ret > 0) {
3248 		ret = -ENOENT;
3249 		goto out;
3250 	}
3251 
3252 	ret = btrfs_del_item(trans, root, path);
3253 out:
3254 	btrfs_free_path(path);
3255 	err = btrfs_commit_transaction(trans);
3256 	if (err && !ret)
3257 		ret = err;
3258 	return ret;
3259 }
3260 
3261 /*
3262  * This is a heuristic used to reduce the number of chunks balanced on
3263  * resume after balance was interrupted.
3264  */
3265 static void update_balance_args(struct btrfs_balance_control *bctl)
3266 {
3267 	/*
3268 	 * Turn on soft mode for chunk types that were being converted.
3269 	 */
3270 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3271 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3272 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3273 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3274 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3275 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3276 
3277 	/*
3278 	 * Turn on usage filter if is not already used.  The idea is
3279 	 * that chunks that we have already balanced should be
3280 	 * reasonably full.  Don't do it for chunks that are being
3281 	 * converted - that will keep us from relocating unconverted
3282 	 * (albeit full) chunks.
3283 	 */
3284 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3285 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3286 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3287 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3288 		bctl->data.usage = 90;
3289 	}
3290 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3291 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3292 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3293 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3294 		bctl->sys.usage = 90;
3295 	}
3296 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3297 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3298 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3299 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3300 		bctl->meta.usage = 90;
3301 	}
3302 }
3303 
3304 /*
3305  * Clear the balance status in fs_info and delete the balance item from disk.
3306  */
3307 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3308 {
3309 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3310 	int ret;
3311 
3312 	BUG_ON(!fs_info->balance_ctl);
3313 
3314 	spin_lock(&fs_info->balance_lock);
3315 	fs_info->balance_ctl = NULL;
3316 	spin_unlock(&fs_info->balance_lock);
3317 
3318 	kfree(bctl);
3319 	ret = del_balance_item(fs_info);
3320 	if (ret)
3321 		btrfs_handle_fs_error(fs_info, ret, NULL);
3322 }
3323 
3324 /*
3325  * Balance filters.  Return 1 if chunk should be filtered out
3326  * (should not be balanced).
3327  */
3328 static int chunk_profiles_filter(u64 chunk_type,
3329 				 struct btrfs_balance_args *bargs)
3330 {
3331 	chunk_type = chunk_to_extended(chunk_type) &
3332 				BTRFS_EXTENDED_PROFILE_MASK;
3333 
3334 	if (bargs->profiles & chunk_type)
3335 		return 0;
3336 
3337 	return 1;
3338 }
3339 
3340 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3341 			      struct btrfs_balance_args *bargs)
3342 {
3343 	struct btrfs_block_group *cache;
3344 	u64 chunk_used;
3345 	u64 user_thresh_min;
3346 	u64 user_thresh_max;
3347 	int ret = 1;
3348 
3349 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3350 	chunk_used = cache->used;
3351 
3352 	if (bargs->usage_min == 0)
3353 		user_thresh_min = 0;
3354 	else
3355 		user_thresh_min = div_factor_fine(cache->length,
3356 						  bargs->usage_min);
3357 
3358 	if (bargs->usage_max == 0)
3359 		user_thresh_max = 1;
3360 	else if (bargs->usage_max > 100)
3361 		user_thresh_max = cache->length;
3362 	else
3363 		user_thresh_max = div_factor_fine(cache->length,
3364 						  bargs->usage_max);
3365 
3366 	if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3367 		ret = 0;
3368 
3369 	btrfs_put_block_group(cache);
3370 	return ret;
3371 }
3372 
3373 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3374 		u64 chunk_offset, struct btrfs_balance_args *bargs)
3375 {
3376 	struct btrfs_block_group *cache;
3377 	u64 chunk_used, user_thresh;
3378 	int ret = 1;
3379 
3380 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3381 	chunk_used = cache->used;
3382 
3383 	if (bargs->usage_min == 0)
3384 		user_thresh = 1;
3385 	else if (bargs->usage > 100)
3386 		user_thresh = cache->length;
3387 	else
3388 		user_thresh = div_factor_fine(cache->length, bargs->usage);
3389 
3390 	if (chunk_used < user_thresh)
3391 		ret = 0;
3392 
3393 	btrfs_put_block_group(cache);
3394 	return ret;
3395 }
3396 
3397 static int chunk_devid_filter(struct extent_buffer *leaf,
3398 			      struct btrfs_chunk *chunk,
3399 			      struct btrfs_balance_args *bargs)
3400 {
3401 	struct btrfs_stripe *stripe;
3402 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3403 	int i;
3404 
3405 	for (i = 0; i < num_stripes; i++) {
3406 		stripe = btrfs_stripe_nr(chunk, i);
3407 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3408 			return 0;
3409 	}
3410 
3411 	return 1;
3412 }
3413 
3414 static u64 calc_data_stripes(u64 type, int num_stripes)
3415 {
3416 	const int index = btrfs_bg_flags_to_raid_index(type);
3417 	const int ncopies = btrfs_raid_array[index].ncopies;
3418 	const int nparity = btrfs_raid_array[index].nparity;
3419 
3420 	if (nparity)
3421 		return num_stripes - nparity;
3422 	else
3423 		return num_stripes / ncopies;
3424 }
3425 
3426 /* [pstart, pend) */
3427 static int chunk_drange_filter(struct extent_buffer *leaf,
3428 			       struct btrfs_chunk *chunk,
3429 			       struct btrfs_balance_args *bargs)
3430 {
3431 	struct btrfs_stripe *stripe;
3432 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3433 	u64 stripe_offset;
3434 	u64 stripe_length;
3435 	u64 type;
3436 	int factor;
3437 	int i;
3438 
3439 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3440 		return 0;
3441 
3442 	type = btrfs_chunk_type(leaf, chunk);
3443 	factor = calc_data_stripes(type, num_stripes);
3444 
3445 	for (i = 0; i < num_stripes; i++) {
3446 		stripe = btrfs_stripe_nr(chunk, i);
3447 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3448 			continue;
3449 
3450 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
3451 		stripe_length = btrfs_chunk_length(leaf, chunk);
3452 		stripe_length = div_u64(stripe_length, factor);
3453 
3454 		if (stripe_offset < bargs->pend &&
3455 		    stripe_offset + stripe_length > bargs->pstart)
3456 			return 0;
3457 	}
3458 
3459 	return 1;
3460 }
3461 
3462 /* [vstart, vend) */
3463 static int chunk_vrange_filter(struct extent_buffer *leaf,
3464 			       struct btrfs_chunk *chunk,
3465 			       u64 chunk_offset,
3466 			       struct btrfs_balance_args *bargs)
3467 {
3468 	if (chunk_offset < bargs->vend &&
3469 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3470 		/* at least part of the chunk is inside this vrange */
3471 		return 0;
3472 
3473 	return 1;
3474 }
3475 
3476 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3477 			       struct btrfs_chunk *chunk,
3478 			       struct btrfs_balance_args *bargs)
3479 {
3480 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3481 
3482 	if (bargs->stripes_min <= num_stripes
3483 			&& num_stripes <= bargs->stripes_max)
3484 		return 0;
3485 
3486 	return 1;
3487 }
3488 
3489 static int chunk_soft_convert_filter(u64 chunk_type,
3490 				     struct btrfs_balance_args *bargs)
3491 {
3492 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3493 		return 0;
3494 
3495 	chunk_type = chunk_to_extended(chunk_type) &
3496 				BTRFS_EXTENDED_PROFILE_MASK;
3497 
3498 	if (bargs->target == chunk_type)
3499 		return 1;
3500 
3501 	return 0;
3502 }
3503 
3504 static int should_balance_chunk(struct extent_buffer *leaf,
3505 				struct btrfs_chunk *chunk, u64 chunk_offset)
3506 {
3507 	struct btrfs_fs_info *fs_info = leaf->fs_info;
3508 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3509 	struct btrfs_balance_args *bargs = NULL;
3510 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3511 
3512 	/* type filter */
3513 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3514 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3515 		return 0;
3516 	}
3517 
3518 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3519 		bargs = &bctl->data;
3520 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3521 		bargs = &bctl->sys;
3522 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3523 		bargs = &bctl->meta;
3524 
3525 	/* profiles filter */
3526 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3527 	    chunk_profiles_filter(chunk_type, bargs)) {
3528 		return 0;
3529 	}
3530 
3531 	/* usage filter */
3532 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3533 	    chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3534 		return 0;
3535 	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3536 	    chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3537 		return 0;
3538 	}
3539 
3540 	/* devid filter */
3541 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3542 	    chunk_devid_filter(leaf, chunk, bargs)) {
3543 		return 0;
3544 	}
3545 
3546 	/* drange filter, makes sense only with devid filter */
3547 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3548 	    chunk_drange_filter(leaf, chunk, bargs)) {
3549 		return 0;
3550 	}
3551 
3552 	/* vrange filter */
3553 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3554 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3555 		return 0;
3556 	}
3557 
3558 	/* stripes filter */
3559 	if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3560 	    chunk_stripes_range_filter(leaf, chunk, bargs)) {
3561 		return 0;
3562 	}
3563 
3564 	/* soft profile changing mode */
3565 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3566 	    chunk_soft_convert_filter(chunk_type, bargs)) {
3567 		return 0;
3568 	}
3569 
3570 	/*
3571 	 * limited by count, must be the last filter
3572 	 */
3573 	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3574 		if (bargs->limit == 0)
3575 			return 0;
3576 		else
3577 			bargs->limit--;
3578 	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3579 		/*
3580 		 * Same logic as the 'limit' filter; the minimum cannot be
3581 		 * determined here because we do not have the global information
3582 		 * about the count of all chunks that satisfy the filters.
3583 		 */
3584 		if (bargs->limit_max == 0)
3585 			return 0;
3586 		else
3587 			bargs->limit_max--;
3588 	}
3589 
3590 	return 1;
3591 }
3592 
3593 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3594 {
3595 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3596 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3597 	u64 chunk_type;
3598 	struct btrfs_chunk *chunk;
3599 	struct btrfs_path *path = NULL;
3600 	struct btrfs_key key;
3601 	struct btrfs_key found_key;
3602 	struct extent_buffer *leaf;
3603 	int slot;
3604 	int ret;
3605 	int enospc_errors = 0;
3606 	bool counting = true;
3607 	/* The single value limit and min/max limits use the same bytes in the */
3608 	u64 limit_data = bctl->data.limit;
3609 	u64 limit_meta = bctl->meta.limit;
3610 	u64 limit_sys = bctl->sys.limit;
3611 	u32 count_data = 0;
3612 	u32 count_meta = 0;
3613 	u32 count_sys = 0;
3614 	int chunk_reserved = 0;
3615 
3616 	path = btrfs_alloc_path();
3617 	if (!path) {
3618 		ret = -ENOMEM;
3619 		goto error;
3620 	}
3621 
3622 	/* zero out stat counters */
3623 	spin_lock(&fs_info->balance_lock);
3624 	memset(&bctl->stat, 0, sizeof(bctl->stat));
3625 	spin_unlock(&fs_info->balance_lock);
3626 again:
3627 	if (!counting) {
3628 		/*
3629 		 * The single value limit and min/max limits use the same bytes
3630 		 * in the
3631 		 */
3632 		bctl->data.limit = limit_data;
3633 		bctl->meta.limit = limit_meta;
3634 		bctl->sys.limit = limit_sys;
3635 	}
3636 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3637 	key.offset = (u64)-1;
3638 	key.type = BTRFS_CHUNK_ITEM_KEY;
3639 
3640 	while (1) {
3641 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3642 		    atomic_read(&fs_info->balance_cancel_req)) {
3643 			ret = -ECANCELED;
3644 			goto error;
3645 		}
3646 
3647 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
3648 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3649 		if (ret < 0) {
3650 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3651 			goto error;
3652 		}
3653 
3654 		/*
3655 		 * this shouldn't happen, it means the last relocate
3656 		 * failed
3657 		 */
3658 		if (ret == 0)
3659 			BUG(); /* FIXME break ? */
3660 
3661 		ret = btrfs_previous_item(chunk_root, path, 0,
3662 					  BTRFS_CHUNK_ITEM_KEY);
3663 		if (ret) {
3664 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3665 			ret = 0;
3666 			break;
3667 		}
3668 
3669 		leaf = path->nodes[0];
3670 		slot = path->slots[0];
3671 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3672 
3673 		if (found_key.objectid != key.objectid) {
3674 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3675 			break;
3676 		}
3677 
3678 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3679 		chunk_type = btrfs_chunk_type(leaf, chunk);
3680 
3681 		if (!counting) {
3682 			spin_lock(&fs_info->balance_lock);
3683 			bctl->stat.considered++;
3684 			spin_unlock(&fs_info->balance_lock);
3685 		}
3686 
3687 		ret = should_balance_chunk(leaf, chunk, found_key.offset);
3688 
3689 		btrfs_release_path(path);
3690 		if (!ret) {
3691 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3692 			goto loop;
3693 		}
3694 
3695 		if (counting) {
3696 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3697 			spin_lock(&fs_info->balance_lock);
3698 			bctl->stat.expected++;
3699 			spin_unlock(&fs_info->balance_lock);
3700 
3701 			if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3702 				count_data++;
3703 			else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3704 				count_sys++;
3705 			else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3706 				count_meta++;
3707 
3708 			goto loop;
3709 		}
3710 
3711 		/*
3712 		 * Apply limit_min filter, no need to check if the LIMITS
3713 		 * filter is used, limit_min is 0 by default
3714 		 */
3715 		if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3716 					count_data < bctl->data.limit_min)
3717 				|| ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3718 					count_meta < bctl->meta.limit_min)
3719 				|| ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3720 					count_sys < bctl->sys.limit_min)) {
3721 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3722 			goto loop;
3723 		}
3724 
3725 		if (!chunk_reserved) {
3726 			/*
3727 			 * We may be relocating the only data chunk we have,
3728 			 * which could potentially end up with losing data's
3729 			 * raid profile, so lets allocate an empty one in
3730 			 * advance.
3731 			 */
3732 			ret = btrfs_may_alloc_data_chunk(fs_info,
3733 							 found_key.offset);
3734 			if (ret < 0) {
3735 				mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3736 				goto error;
3737 			} else if (ret == 1) {
3738 				chunk_reserved = 1;
3739 			}
3740 		}
3741 
3742 		ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3743 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3744 		if (ret == -ENOSPC) {
3745 			enospc_errors++;
3746 		} else if (ret == -ETXTBSY) {
3747 			btrfs_info(fs_info,
3748 	   "skipping relocation of block group %llu due to active swapfile",
3749 				   found_key.offset);
3750 			ret = 0;
3751 		} else if (ret) {
3752 			goto error;
3753 		} else {
3754 			spin_lock(&fs_info->balance_lock);
3755 			bctl->stat.completed++;
3756 			spin_unlock(&fs_info->balance_lock);
3757 		}
3758 loop:
3759 		if (found_key.offset == 0)
3760 			break;
3761 		key.offset = found_key.offset - 1;
3762 	}
3763 
3764 	if (counting) {
3765 		btrfs_release_path(path);
3766 		counting = false;
3767 		goto again;
3768 	}
3769 error:
3770 	btrfs_free_path(path);
3771 	if (enospc_errors) {
3772 		btrfs_info(fs_info, "%d enospc errors during balance",
3773 			   enospc_errors);
3774 		if (!ret)
3775 			ret = -ENOSPC;
3776 	}
3777 
3778 	return ret;
3779 }
3780 
3781 /**
3782  * alloc_profile_is_valid - see if a given profile is valid and reduced
3783  * @flags: profile to validate
3784  * @extended: if true @flags is treated as an extended profile
3785  */
3786 static int alloc_profile_is_valid(u64 flags, int extended)
3787 {
3788 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3789 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
3790 
3791 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3792 
3793 	/* 1) check that all other bits are zeroed */
3794 	if (flags & ~mask)
3795 		return 0;
3796 
3797 	/* 2) see if profile is reduced */
3798 	if (flags == 0)
3799 		return !extended; /* "0" is valid for usual profiles */
3800 
3801 	return has_single_bit_set(flags);
3802 }
3803 
3804 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3805 {
3806 	/* cancel requested || normal exit path */
3807 	return atomic_read(&fs_info->balance_cancel_req) ||
3808 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
3809 		 atomic_read(&fs_info->balance_cancel_req) == 0);
3810 }
3811 
3812 /*
3813  * Validate target profile against allowed profiles and return true if it's OK.
3814  * Otherwise print the error message and return false.
3815  */
3816 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3817 		const struct btrfs_balance_args *bargs,
3818 		u64 allowed, const char *type)
3819 {
3820 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3821 		return true;
3822 
3823 	/* Profile is valid and does not have bits outside of the allowed set */
3824 	if (alloc_profile_is_valid(bargs->target, 1) &&
3825 	    (bargs->target & ~allowed) == 0)
3826 		return true;
3827 
3828 	btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3829 			type, btrfs_bg_type_to_raid_name(bargs->target));
3830 	return false;
3831 }
3832 
3833 /*
3834  * Fill @buf with textual description of balance filter flags @bargs, up to
3835  * @size_buf including the terminating null. The output may be trimmed if it
3836  * does not fit into the provided buffer.
3837  */
3838 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3839 				 u32 size_buf)
3840 {
3841 	int ret;
3842 	u32 size_bp = size_buf;
3843 	char *bp = buf;
3844 	u64 flags = bargs->flags;
3845 	char tmp_buf[128] = {'\0'};
3846 
3847 	if (!flags)
3848 		return;
3849 
3850 #define CHECK_APPEND_NOARG(a)						\
3851 	do {								\
3852 		ret = snprintf(bp, size_bp, (a));			\
3853 		if (ret < 0 || ret >= size_bp)				\
3854 			goto out_overflow;				\
3855 		size_bp -= ret;						\
3856 		bp += ret;						\
3857 	} while (0)
3858 
3859 #define CHECK_APPEND_1ARG(a, v1)					\
3860 	do {								\
3861 		ret = snprintf(bp, size_bp, (a), (v1));			\
3862 		if (ret < 0 || ret >= size_bp)				\
3863 			goto out_overflow;				\
3864 		size_bp -= ret;						\
3865 		bp += ret;						\
3866 	} while (0)
3867 
3868 #define CHECK_APPEND_2ARG(a, v1, v2)					\
3869 	do {								\
3870 		ret = snprintf(bp, size_bp, (a), (v1), (v2));		\
3871 		if (ret < 0 || ret >= size_bp)				\
3872 			goto out_overflow;				\
3873 		size_bp -= ret;						\
3874 		bp += ret;						\
3875 	} while (0)
3876 
3877 	if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3878 		CHECK_APPEND_1ARG("convert=%s,",
3879 				  btrfs_bg_type_to_raid_name(bargs->target));
3880 
3881 	if (flags & BTRFS_BALANCE_ARGS_SOFT)
3882 		CHECK_APPEND_NOARG("soft,");
3883 
3884 	if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3885 		btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3886 					    sizeof(tmp_buf));
3887 		CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3888 	}
3889 
3890 	if (flags & BTRFS_BALANCE_ARGS_USAGE)
3891 		CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3892 
3893 	if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3894 		CHECK_APPEND_2ARG("usage=%u..%u,",
3895 				  bargs->usage_min, bargs->usage_max);
3896 
3897 	if (flags & BTRFS_BALANCE_ARGS_DEVID)
3898 		CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3899 
3900 	if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3901 		CHECK_APPEND_2ARG("drange=%llu..%llu,",
3902 				  bargs->pstart, bargs->pend);
3903 
3904 	if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3905 		CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3906 				  bargs->vstart, bargs->vend);
3907 
3908 	if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3909 		CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3910 
3911 	if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3912 		CHECK_APPEND_2ARG("limit=%u..%u,",
3913 				bargs->limit_min, bargs->limit_max);
3914 
3915 	if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3916 		CHECK_APPEND_2ARG("stripes=%u..%u,",
3917 				  bargs->stripes_min, bargs->stripes_max);
3918 
3919 #undef CHECK_APPEND_2ARG
3920 #undef CHECK_APPEND_1ARG
3921 #undef CHECK_APPEND_NOARG
3922 
3923 out_overflow:
3924 
3925 	if (size_bp < size_buf)
3926 		buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3927 	else
3928 		buf[0] = '\0';
3929 }
3930 
3931 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3932 {
3933 	u32 size_buf = 1024;
3934 	char tmp_buf[192] = {'\0'};
3935 	char *buf;
3936 	char *bp;
3937 	u32 size_bp = size_buf;
3938 	int ret;
3939 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3940 
3941 	buf = kzalloc(size_buf, GFP_KERNEL);
3942 	if (!buf)
3943 		return;
3944 
3945 	bp = buf;
3946 
3947 #define CHECK_APPEND_1ARG(a, v1)					\
3948 	do {								\
3949 		ret = snprintf(bp, size_bp, (a), (v1));			\
3950 		if (ret < 0 || ret >= size_bp)				\
3951 			goto out_overflow;				\
3952 		size_bp -= ret;						\
3953 		bp += ret;						\
3954 	} while (0)
3955 
3956 	if (bctl->flags & BTRFS_BALANCE_FORCE)
3957 		CHECK_APPEND_1ARG("%s", "-f ");
3958 
3959 	if (bctl->flags & BTRFS_BALANCE_DATA) {
3960 		describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3961 		CHECK_APPEND_1ARG("-d%s ", tmp_buf);
3962 	}
3963 
3964 	if (bctl->flags & BTRFS_BALANCE_METADATA) {
3965 		describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
3966 		CHECK_APPEND_1ARG("-m%s ", tmp_buf);
3967 	}
3968 
3969 	if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
3970 		describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
3971 		CHECK_APPEND_1ARG("-s%s ", tmp_buf);
3972 	}
3973 
3974 #undef CHECK_APPEND_1ARG
3975 
3976 out_overflow:
3977 
3978 	if (size_bp < size_buf)
3979 		buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
3980 	btrfs_info(fs_info, "balance: %s %s",
3981 		   (bctl->flags & BTRFS_BALANCE_RESUME) ?
3982 		   "resume" : "start", buf);
3983 
3984 	kfree(buf);
3985 }
3986 
3987 /*
3988  * Should be called with balance mutexe held
3989  */
3990 int btrfs_balance(struct btrfs_fs_info *fs_info,
3991 		  struct btrfs_balance_control *bctl,
3992 		  struct btrfs_ioctl_balance_args *bargs)
3993 {
3994 	u64 meta_target, data_target;
3995 	u64 allowed;
3996 	int mixed = 0;
3997 	int ret;
3998 	u64 num_devices;
3999 	unsigned seq;
4000 	bool reducing_redundancy;
4001 	int i;
4002 
4003 	if (btrfs_fs_closing(fs_info) ||
4004 	    atomic_read(&fs_info->balance_pause_req) ||
4005 	    btrfs_should_cancel_balance(fs_info)) {
4006 		ret = -EINVAL;
4007 		goto out;
4008 	}
4009 
4010 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4011 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4012 		mixed = 1;
4013 
4014 	/*
4015 	 * In case of mixed groups both data and meta should be picked,
4016 	 * and identical options should be given for both of them.
4017 	 */
4018 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4019 	if (mixed && (bctl->flags & allowed)) {
4020 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4021 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4022 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4023 			btrfs_err(fs_info,
4024 	  "balance: mixed groups data and metadata options must be the same");
4025 			ret = -EINVAL;
4026 			goto out;
4027 		}
4028 	}
4029 
4030 	/*
4031 	 * rw_devices will not change at the moment, device add/delete/replace
4032 	 * are excluded by EXCL_OP
4033 	 */
4034 	num_devices = fs_info->fs_devices->rw_devices;
4035 
4036 	/*
4037 	 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4038 	 * special bit for it, to make it easier to distinguish.  Thus we need
4039 	 * to set it manually, or balance would refuse the profile.
4040 	 */
4041 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4042 	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4043 		if (num_devices >= btrfs_raid_array[i].devs_min)
4044 			allowed |= btrfs_raid_array[i].bg_flag;
4045 
4046 	if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4047 	    !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4048 	    !validate_convert_profile(fs_info, &bctl->sys,  allowed, "system")) {
4049 		ret = -EINVAL;
4050 		goto out;
4051 	}
4052 
4053 	/*
4054 	 * Allow to reduce metadata or system integrity only if force set for
4055 	 * profiles with redundancy (copies, parity)
4056 	 */
4057 	allowed = 0;
4058 	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4059 		if (btrfs_raid_array[i].ncopies >= 2 ||
4060 		    btrfs_raid_array[i].tolerated_failures >= 1)
4061 			allowed |= btrfs_raid_array[i].bg_flag;
4062 	}
4063 	do {
4064 		seq = read_seqbegin(&fs_info->profiles_lock);
4065 
4066 		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4067 		     (fs_info->avail_system_alloc_bits & allowed) &&
4068 		     !(bctl->sys.target & allowed)) ||
4069 		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4070 		     (fs_info->avail_metadata_alloc_bits & allowed) &&
4071 		     !(bctl->meta.target & allowed)))
4072 			reducing_redundancy = true;
4073 		else
4074 			reducing_redundancy = false;
4075 
4076 		/* if we're not converting, the target field is uninitialized */
4077 		meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4078 			bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4079 		data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4080 			bctl->data.target : fs_info->avail_data_alloc_bits;
4081 	} while (read_seqretry(&fs_info->profiles_lock, seq));
4082 
4083 	if (reducing_redundancy) {
4084 		if (bctl->flags & BTRFS_BALANCE_FORCE) {
4085 			btrfs_info(fs_info,
4086 			   "balance: force reducing metadata redundancy");
4087 		} else {
4088 			btrfs_err(fs_info,
4089 	"balance: reduces metadata redundancy, use --force if you want this");
4090 			ret = -EINVAL;
4091 			goto out;
4092 		}
4093 	}
4094 
4095 	if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4096 		btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4097 		btrfs_warn(fs_info,
4098 	"balance: metadata profile %s has lower redundancy than data profile %s",
4099 				btrfs_bg_type_to_raid_name(meta_target),
4100 				btrfs_bg_type_to_raid_name(data_target));
4101 	}
4102 
4103 	if (fs_info->send_in_progress) {
4104 		btrfs_warn_rl(fs_info,
4105 "cannot run balance while send operations are in progress (%d in progress)",
4106 			      fs_info->send_in_progress);
4107 		ret = -EAGAIN;
4108 		goto out;
4109 	}
4110 
4111 	ret = insert_balance_item(fs_info, bctl);
4112 	if (ret && ret != -EEXIST)
4113 		goto out;
4114 
4115 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4116 		BUG_ON(ret == -EEXIST);
4117 		BUG_ON(fs_info->balance_ctl);
4118 		spin_lock(&fs_info->balance_lock);
4119 		fs_info->balance_ctl = bctl;
4120 		spin_unlock(&fs_info->balance_lock);
4121 	} else {
4122 		BUG_ON(ret != -EEXIST);
4123 		spin_lock(&fs_info->balance_lock);
4124 		update_balance_args(bctl);
4125 		spin_unlock(&fs_info->balance_lock);
4126 	}
4127 
4128 	ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4129 	set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4130 	describe_balance_start_or_resume(fs_info);
4131 	mutex_unlock(&fs_info->balance_mutex);
4132 
4133 	ret = __btrfs_balance(fs_info);
4134 
4135 	mutex_lock(&fs_info->balance_mutex);
4136 	if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4137 		btrfs_info(fs_info, "balance: paused");
4138 	else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req))
4139 		btrfs_info(fs_info, "balance: canceled");
4140 	else
4141 		btrfs_info(fs_info, "balance: ended with status: %d", ret);
4142 
4143 	clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4144 
4145 	if (bargs) {
4146 		memset(bargs, 0, sizeof(*bargs));
4147 		btrfs_update_ioctl_balance_args(fs_info, bargs);
4148 	}
4149 
4150 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4151 	    balance_need_close(fs_info)) {
4152 		reset_balance_state(fs_info);
4153 		clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4154 	}
4155 
4156 	wake_up(&fs_info->balance_wait_q);
4157 
4158 	return ret;
4159 out:
4160 	if (bctl->flags & BTRFS_BALANCE_RESUME)
4161 		reset_balance_state(fs_info);
4162 	else
4163 		kfree(bctl);
4164 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4165 
4166 	return ret;
4167 }
4168 
4169 static int balance_kthread(void *data)
4170 {
4171 	struct btrfs_fs_info *fs_info = data;
4172 	int ret = 0;
4173 
4174 	mutex_lock(&fs_info->balance_mutex);
4175 	if (fs_info->balance_ctl)
4176 		ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4177 	mutex_unlock(&fs_info->balance_mutex);
4178 
4179 	return ret;
4180 }
4181 
4182 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4183 {
4184 	struct task_struct *tsk;
4185 
4186 	mutex_lock(&fs_info->balance_mutex);
4187 	if (!fs_info->balance_ctl) {
4188 		mutex_unlock(&fs_info->balance_mutex);
4189 		return 0;
4190 	}
4191 	mutex_unlock(&fs_info->balance_mutex);
4192 
4193 	if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4194 		btrfs_info(fs_info, "balance: resume skipped");
4195 		return 0;
4196 	}
4197 
4198 	/*
4199 	 * A ro->rw remount sequence should continue with the paused balance
4200 	 * regardless of who pauses it, system or the user as of now, so set
4201 	 * the resume flag.
4202 	 */
4203 	spin_lock(&fs_info->balance_lock);
4204 	fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4205 	spin_unlock(&fs_info->balance_lock);
4206 
4207 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4208 	return PTR_ERR_OR_ZERO(tsk);
4209 }
4210 
4211 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4212 {
4213 	struct btrfs_balance_control *bctl;
4214 	struct btrfs_balance_item *item;
4215 	struct btrfs_disk_balance_args disk_bargs;
4216 	struct btrfs_path *path;
4217 	struct extent_buffer *leaf;
4218 	struct btrfs_key key;
4219 	int ret;
4220 
4221 	path = btrfs_alloc_path();
4222 	if (!path)
4223 		return -ENOMEM;
4224 
4225 	key.objectid = BTRFS_BALANCE_OBJECTID;
4226 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
4227 	key.offset = 0;
4228 
4229 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4230 	if (ret < 0)
4231 		goto out;
4232 	if (ret > 0) { /* ret = -ENOENT; */
4233 		ret = 0;
4234 		goto out;
4235 	}
4236 
4237 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4238 	if (!bctl) {
4239 		ret = -ENOMEM;
4240 		goto out;
4241 	}
4242 
4243 	leaf = path->nodes[0];
4244 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4245 
4246 	bctl->flags = btrfs_balance_flags(leaf, item);
4247 	bctl->flags |= BTRFS_BALANCE_RESUME;
4248 
4249 	btrfs_balance_data(leaf, item, &disk_bargs);
4250 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4251 	btrfs_balance_meta(leaf, item, &disk_bargs);
4252 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4253 	btrfs_balance_sys(leaf, item, &disk_bargs);
4254 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4255 
4256 	/*
4257 	 * This should never happen, as the paused balance state is recovered
4258 	 * during mount without any chance of other exclusive ops to collide.
4259 	 *
4260 	 * This gives the exclusive op status to balance and keeps in paused
4261 	 * state until user intervention (cancel or umount). If the ownership
4262 	 * cannot be assigned, show a message but do not fail. The balance
4263 	 * is in a paused state and must have fs_info::balance_ctl properly
4264 	 * set up.
4265 	 */
4266 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4267 		btrfs_warn(fs_info,
4268 	"balance: cannot set exclusive op status, resume manually");
4269 
4270 	mutex_lock(&fs_info->balance_mutex);
4271 	BUG_ON(fs_info->balance_ctl);
4272 	spin_lock(&fs_info->balance_lock);
4273 	fs_info->balance_ctl = bctl;
4274 	spin_unlock(&fs_info->balance_lock);
4275 	mutex_unlock(&fs_info->balance_mutex);
4276 out:
4277 	btrfs_free_path(path);
4278 	return ret;
4279 }
4280 
4281 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4282 {
4283 	int ret = 0;
4284 
4285 	mutex_lock(&fs_info->balance_mutex);
4286 	if (!fs_info->balance_ctl) {
4287 		mutex_unlock(&fs_info->balance_mutex);
4288 		return -ENOTCONN;
4289 	}
4290 
4291 	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4292 		atomic_inc(&fs_info->balance_pause_req);
4293 		mutex_unlock(&fs_info->balance_mutex);
4294 
4295 		wait_event(fs_info->balance_wait_q,
4296 			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4297 
4298 		mutex_lock(&fs_info->balance_mutex);
4299 		/* we are good with balance_ctl ripped off from under us */
4300 		BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4301 		atomic_dec(&fs_info->balance_pause_req);
4302 	} else {
4303 		ret = -ENOTCONN;
4304 	}
4305 
4306 	mutex_unlock(&fs_info->balance_mutex);
4307 	return ret;
4308 }
4309 
4310 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4311 {
4312 	mutex_lock(&fs_info->balance_mutex);
4313 	if (!fs_info->balance_ctl) {
4314 		mutex_unlock(&fs_info->balance_mutex);
4315 		return -ENOTCONN;
4316 	}
4317 
4318 	/*
4319 	 * A paused balance with the item stored on disk can be resumed at
4320 	 * mount time if the mount is read-write. Otherwise it's still paused
4321 	 * and we must not allow cancelling as it deletes the item.
4322 	 */
4323 	if (sb_rdonly(fs_info->sb)) {
4324 		mutex_unlock(&fs_info->balance_mutex);
4325 		return -EROFS;
4326 	}
4327 
4328 	atomic_inc(&fs_info->balance_cancel_req);
4329 	/*
4330 	 * if we are running just wait and return, balance item is
4331 	 * deleted in btrfs_balance in this case
4332 	 */
4333 	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4334 		mutex_unlock(&fs_info->balance_mutex);
4335 		wait_event(fs_info->balance_wait_q,
4336 			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4337 		mutex_lock(&fs_info->balance_mutex);
4338 	} else {
4339 		mutex_unlock(&fs_info->balance_mutex);
4340 		/*
4341 		 * Lock released to allow other waiters to continue, we'll
4342 		 * reexamine the status again.
4343 		 */
4344 		mutex_lock(&fs_info->balance_mutex);
4345 
4346 		if (fs_info->balance_ctl) {
4347 			reset_balance_state(fs_info);
4348 			clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4349 			btrfs_info(fs_info, "balance: canceled");
4350 		}
4351 	}
4352 
4353 	BUG_ON(fs_info->balance_ctl ||
4354 		test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4355 	atomic_dec(&fs_info->balance_cancel_req);
4356 	mutex_unlock(&fs_info->balance_mutex);
4357 	return 0;
4358 }
4359 
4360 int btrfs_uuid_scan_kthread(void *data)
4361 {
4362 	struct btrfs_fs_info *fs_info = data;
4363 	struct btrfs_root *root = fs_info->tree_root;
4364 	struct btrfs_key key;
4365 	struct btrfs_path *path = NULL;
4366 	int ret = 0;
4367 	struct extent_buffer *eb;
4368 	int slot;
4369 	struct btrfs_root_item root_item;
4370 	u32 item_size;
4371 	struct btrfs_trans_handle *trans = NULL;
4372 	bool closing = false;
4373 
4374 	path = btrfs_alloc_path();
4375 	if (!path) {
4376 		ret = -ENOMEM;
4377 		goto out;
4378 	}
4379 
4380 	key.objectid = 0;
4381 	key.type = BTRFS_ROOT_ITEM_KEY;
4382 	key.offset = 0;
4383 
4384 	while (1) {
4385 		if (btrfs_fs_closing(fs_info)) {
4386 			closing = true;
4387 			break;
4388 		}
4389 		ret = btrfs_search_forward(root, &key, path,
4390 				BTRFS_OLDEST_GENERATION);
4391 		if (ret) {
4392 			if (ret > 0)
4393 				ret = 0;
4394 			break;
4395 		}
4396 
4397 		if (key.type != BTRFS_ROOT_ITEM_KEY ||
4398 		    (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4399 		     key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4400 		    key.objectid > BTRFS_LAST_FREE_OBJECTID)
4401 			goto skip;
4402 
4403 		eb = path->nodes[0];
4404 		slot = path->slots[0];
4405 		item_size = btrfs_item_size_nr(eb, slot);
4406 		if (item_size < sizeof(root_item))
4407 			goto skip;
4408 
4409 		read_extent_buffer(eb, &root_item,
4410 				   btrfs_item_ptr_offset(eb, slot),
4411 				   (int)sizeof(root_item));
4412 		if (btrfs_root_refs(&root_item) == 0)
4413 			goto skip;
4414 
4415 		if (!btrfs_is_empty_uuid(root_item.uuid) ||
4416 		    !btrfs_is_empty_uuid(root_item.received_uuid)) {
4417 			if (trans)
4418 				goto update_tree;
4419 
4420 			btrfs_release_path(path);
4421 			/*
4422 			 * 1 - subvol uuid item
4423 			 * 1 - received_subvol uuid item
4424 			 */
4425 			trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4426 			if (IS_ERR(trans)) {
4427 				ret = PTR_ERR(trans);
4428 				break;
4429 			}
4430 			continue;
4431 		} else {
4432 			goto skip;
4433 		}
4434 update_tree:
4435 		if (!btrfs_is_empty_uuid(root_item.uuid)) {
4436 			ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4437 						  BTRFS_UUID_KEY_SUBVOL,
4438 						  key.objectid);
4439 			if (ret < 0) {
4440 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4441 					ret);
4442 				break;
4443 			}
4444 		}
4445 
4446 		if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4447 			ret = btrfs_uuid_tree_add(trans,
4448 						  root_item.received_uuid,
4449 						 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4450 						  key.objectid);
4451 			if (ret < 0) {
4452 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4453 					ret);
4454 				break;
4455 			}
4456 		}
4457 
4458 skip:
4459 		if (trans) {
4460 			ret = btrfs_end_transaction(trans);
4461 			trans = NULL;
4462 			if (ret)
4463 				break;
4464 		}
4465 
4466 		btrfs_release_path(path);
4467 		if (key.offset < (u64)-1) {
4468 			key.offset++;
4469 		} else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4470 			key.offset = 0;
4471 			key.type = BTRFS_ROOT_ITEM_KEY;
4472 		} else if (key.objectid < (u64)-1) {
4473 			key.offset = 0;
4474 			key.type = BTRFS_ROOT_ITEM_KEY;
4475 			key.objectid++;
4476 		} else {
4477 			break;
4478 		}
4479 		cond_resched();
4480 	}
4481 
4482 out:
4483 	btrfs_free_path(path);
4484 	if (trans && !IS_ERR(trans))
4485 		btrfs_end_transaction(trans);
4486 	if (ret)
4487 		btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4488 	else if (!closing)
4489 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4490 	up(&fs_info->uuid_tree_rescan_sem);
4491 	return 0;
4492 }
4493 
4494 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4495 {
4496 	struct btrfs_trans_handle *trans;
4497 	struct btrfs_root *tree_root = fs_info->tree_root;
4498 	struct btrfs_root *uuid_root;
4499 	struct task_struct *task;
4500 	int ret;
4501 
4502 	/*
4503 	 * 1 - root node
4504 	 * 1 - root item
4505 	 */
4506 	trans = btrfs_start_transaction(tree_root, 2);
4507 	if (IS_ERR(trans))
4508 		return PTR_ERR(trans);
4509 
4510 	uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4511 	if (IS_ERR(uuid_root)) {
4512 		ret = PTR_ERR(uuid_root);
4513 		btrfs_abort_transaction(trans, ret);
4514 		btrfs_end_transaction(trans);
4515 		return ret;
4516 	}
4517 
4518 	fs_info->uuid_root = uuid_root;
4519 
4520 	ret = btrfs_commit_transaction(trans);
4521 	if (ret)
4522 		return ret;
4523 
4524 	down(&fs_info->uuid_tree_rescan_sem);
4525 	task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4526 	if (IS_ERR(task)) {
4527 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
4528 		btrfs_warn(fs_info, "failed to start uuid_scan task");
4529 		up(&fs_info->uuid_tree_rescan_sem);
4530 		return PTR_ERR(task);
4531 	}
4532 
4533 	return 0;
4534 }
4535 
4536 /*
4537  * shrinking a device means finding all of the device extents past
4538  * the new size, and then following the back refs to the chunks.
4539  * The chunk relocation code actually frees the device extent
4540  */
4541 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4542 {
4543 	struct btrfs_fs_info *fs_info = device->fs_info;
4544 	struct btrfs_root *root = fs_info->dev_root;
4545 	struct btrfs_trans_handle *trans;
4546 	struct btrfs_dev_extent *dev_extent = NULL;
4547 	struct btrfs_path *path;
4548 	u64 length;
4549 	u64 chunk_offset;
4550 	int ret;
4551 	int slot;
4552 	int failed = 0;
4553 	bool retried = false;
4554 	struct extent_buffer *l;
4555 	struct btrfs_key key;
4556 	struct btrfs_super_block *super_copy = fs_info->super_copy;
4557 	u64 old_total = btrfs_super_total_bytes(super_copy);
4558 	u64 old_size = btrfs_device_get_total_bytes(device);
4559 	u64 diff;
4560 	u64 start;
4561 
4562 	new_size = round_down(new_size, fs_info->sectorsize);
4563 	start = new_size;
4564 	diff = round_down(old_size - new_size, fs_info->sectorsize);
4565 
4566 	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4567 		return -EINVAL;
4568 
4569 	path = btrfs_alloc_path();
4570 	if (!path)
4571 		return -ENOMEM;
4572 
4573 	path->reada = READA_BACK;
4574 
4575 	trans = btrfs_start_transaction(root, 0);
4576 	if (IS_ERR(trans)) {
4577 		btrfs_free_path(path);
4578 		return PTR_ERR(trans);
4579 	}
4580 
4581 	mutex_lock(&fs_info->chunk_mutex);
4582 
4583 	btrfs_device_set_total_bytes(device, new_size);
4584 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4585 		device->fs_devices->total_rw_bytes -= diff;
4586 		atomic64_sub(diff, &fs_info->free_chunk_space);
4587 	}
4588 
4589 	/*
4590 	 * Once the device's size has been set to the new size, ensure all
4591 	 * in-memory chunks are synced to disk so that the loop below sees them
4592 	 * and relocates them accordingly.
4593 	 */
4594 	if (contains_pending_extent(device, &start, diff)) {
4595 		mutex_unlock(&fs_info->chunk_mutex);
4596 		ret = btrfs_commit_transaction(trans);
4597 		if (ret)
4598 			goto done;
4599 	} else {
4600 		mutex_unlock(&fs_info->chunk_mutex);
4601 		btrfs_end_transaction(trans);
4602 	}
4603 
4604 again:
4605 	key.objectid = device->devid;
4606 	key.offset = (u64)-1;
4607 	key.type = BTRFS_DEV_EXTENT_KEY;
4608 
4609 	do {
4610 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
4611 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4612 		if (ret < 0) {
4613 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4614 			goto done;
4615 		}
4616 
4617 		ret = btrfs_previous_item(root, path, 0, key.type);
4618 		if (ret)
4619 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4620 		if (ret < 0)
4621 			goto done;
4622 		if (ret) {
4623 			ret = 0;
4624 			btrfs_release_path(path);
4625 			break;
4626 		}
4627 
4628 		l = path->nodes[0];
4629 		slot = path->slots[0];
4630 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4631 
4632 		if (key.objectid != device->devid) {
4633 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4634 			btrfs_release_path(path);
4635 			break;
4636 		}
4637 
4638 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4639 		length = btrfs_dev_extent_length(l, dev_extent);
4640 
4641 		if (key.offset + length <= new_size) {
4642 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4643 			btrfs_release_path(path);
4644 			break;
4645 		}
4646 
4647 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4648 		btrfs_release_path(path);
4649 
4650 		/*
4651 		 * We may be relocating the only data chunk we have,
4652 		 * which could potentially end up with losing data's
4653 		 * raid profile, so lets allocate an empty one in
4654 		 * advance.
4655 		 */
4656 		ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4657 		if (ret < 0) {
4658 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4659 			goto done;
4660 		}
4661 
4662 		ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4663 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4664 		if (ret == -ENOSPC) {
4665 			failed++;
4666 		} else if (ret) {
4667 			if (ret == -ETXTBSY) {
4668 				btrfs_warn(fs_info,
4669 		   "could not shrink block group %llu due to active swapfile",
4670 					   chunk_offset);
4671 			}
4672 			goto done;
4673 		}
4674 	} while (key.offset-- > 0);
4675 
4676 	if (failed && !retried) {
4677 		failed = 0;
4678 		retried = true;
4679 		goto again;
4680 	} else if (failed && retried) {
4681 		ret = -ENOSPC;
4682 		goto done;
4683 	}
4684 
4685 	/* Shrinking succeeded, else we would be at "done". */
4686 	trans = btrfs_start_transaction(root, 0);
4687 	if (IS_ERR(trans)) {
4688 		ret = PTR_ERR(trans);
4689 		goto done;
4690 	}
4691 
4692 	mutex_lock(&fs_info->chunk_mutex);
4693 	btrfs_device_set_disk_total_bytes(device, new_size);
4694 	if (list_empty(&device->post_commit_list))
4695 		list_add_tail(&device->post_commit_list,
4696 			      &trans->transaction->dev_update_list);
4697 
4698 	WARN_ON(diff > old_total);
4699 	btrfs_set_super_total_bytes(super_copy,
4700 			round_down(old_total - diff, fs_info->sectorsize));
4701 	mutex_unlock(&fs_info->chunk_mutex);
4702 
4703 	/* Now btrfs_update_device() will change the on-disk size. */
4704 	ret = btrfs_update_device(trans, device);
4705 	if (ret < 0) {
4706 		btrfs_abort_transaction(trans, ret);
4707 		btrfs_end_transaction(trans);
4708 	} else {
4709 		ret = btrfs_commit_transaction(trans);
4710 	}
4711 done:
4712 	btrfs_free_path(path);
4713 	if (ret) {
4714 		mutex_lock(&fs_info->chunk_mutex);
4715 		btrfs_device_set_total_bytes(device, old_size);
4716 		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4717 			device->fs_devices->total_rw_bytes += diff;
4718 		atomic64_add(diff, &fs_info->free_chunk_space);
4719 		mutex_unlock(&fs_info->chunk_mutex);
4720 	}
4721 	return ret;
4722 }
4723 
4724 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4725 			   struct btrfs_key *key,
4726 			   struct btrfs_chunk *chunk, int item_size)
4727 {
4728 	struct btrfs_super_block *super_copy = fs_info->super_copy;
4729 	struct btrfs_disk_key disk_key;
4730 	u32 array_size;
4731 	u8 *ptr;
4732 
4733 	mutex_lock(&fs_info->chunk_mutex);
4734 	array_size = btrfs_super_sys_array_size(super_copy);
4735 	if (array_size + item_size + sizeof(disk_key)
4736 			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4737 		mutex_unlock(&fs_info->chunk_mutex);
4738 		return -EFBIG;
4739 	}
4740 
4741 	ptr = super_copy->sys_chunk_array + array_size;
4742 	btrfs_cpu_key_to_disk(&disk_key, key);
4743 	memcpy(ptr, &disk_key, sizeof(disk_key));
4744 	ptr += sizeof(disk_key);
4745 	memcpy(ptr, chunk, item_size);
4746 	item_size += sizeof(disk_key);
4747 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4748 	mutex_unlock(&fs_info->chunk_mutex);
4749 
4750 	return 0;
4751 }
4752 
4753 /*
4754  * sort the devices in descending order by max_avail, total_avail
4755  */
4756 static int btrfs_cmp_device_info(const void *a, const void *b)
4757 {
4758 	const struct btrfs_device_info *di_a = a;
4759 	const struct btrfs_device_info *di_b = b;
4760 
4761 	if (di_a->max_avail > di_b->max_avail)
4762 		return -1;
4763 	if (di_a->max_avail < di_b->max_avail)
4764 		return 1;
4765 	if (di_a->total_avail > di_b->total_avail)
4766 		return -1;
4767 	if (di_a->total_avail < di_b->total_avail)
4768 		return 1;
4769 	return 0;
4770 }
4771 
4772 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4773 {
4774 	if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4775 		return;
4776 
4777 	btrfs_set_fs_incompat(info, RAID56);
4778 }
4779 
4780 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4781 {
4782 	if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4783 		return;
4784 
4785 	btrfs_set_fs_incompat(info, RAID1C34);
4786 }
4787 
4788 /*
4789  * Structure used internally for __btrfs_alloc_chunk() function.
4790  * Wraps needed parameters.
4791  */
4792 struct alloc_chunk_ctl {
4793 	u64 start;
4794 	u64 type;
4795 	/* Total number of stripes to allocate */
4796 	int num_stripes;
4797 	/* sub_stripes info for map */
4798 	int sub_stripes;
4799 	/* Stripes per device */
4800 	int dev_stripes;
4801 	/* Maximum number of devices to use */
4802 	int devs_max;
4803 	/* Minimum number of devices to use */
4804 	int devs_min;
4805 	/* ndevs has to be a multiple of this */
4806 	int devs_increment;
4807 	/* Number of copies */
4808 	int ncopies;
4809 	/* Number of stripes worth of bytes to store parity information */
4810 	int nparity;
4811 	u64 max_stripe_size;
4812 	u64 max_chunk_size;
4813 	u64 dev_extent_min;
4814 	u64 stripe_size;
4815 	u64 chunk_size;
4816 	int ndevs;
4817 };
4818 
4819 static void init_alloc_chunk_ctl_policy_regular(
4820 				struct btrfs_fs_devices *fs_devices,
4821 				struct alloc_chunk_ctl *ctl)
4822 {
4823 	u64 type = ctl->type;
4824 
4825 	if (type & BTRFS_BLOCK_GROUP_DATA) {
4826 		ctl->max_stripe_size = SZ_1G;
4827 		ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4828 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4829 		/* For larger filesystems, use larger metadata chunks */
4830 		if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4831 			ctl->max_stripe_size = SZ_1G;
4832 		else
4833 			ctl->max_stripe_size = SZ_256M;
4834 		ctl->max_chunk_size = ctl->max_stripe_size;
4835 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4836 		ctl->max_stripe_size = SZ_32M;
4837 		ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4838 		ctl->devs_max = min_t(int, ctl->devs_max,
4839 				      BTRFS_MAX_DEVS_SYS_CHUNK);
4840 	} else {
4841 		BUG();
4842 	}
4843 
4844 	/* We don't want a chunk larger than 10% of writable space */
4845 	ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4846 				  ctl->max_chunk_size);
4847 	ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
4848 }
4849 
4850 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
4851 				 struct alloc_chunk_ctl *ctl)
4852 {
4853 	int index = btrfs_bg_flags_to_raid_index(ctl->type);
4854 
4855 	ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
4856 	ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
4857 	ctl->devs_max = btrfs_raid_array[index].devs_max;
4858 	if (!ctl->devs_max)
4859 		ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
4860 	ctl->devs_min = btrfs_raid_array[index].devs_min;
4861 	ctl->devs_increment = btrfs_raid_array[index].devs_increment;
4862 	ctl->ncopies = btrfs_raid_array[index].ncopies;
4863 	ctl->nparity = btrfs_raid_array[index].nparity;
4864 	ctl->ndevs = 0;
4865 
4866 	switch (fs_devices->chunk_alloc_policy) {
4867 	case BTRFS_CHUNK_ALLOC_REGULAR:
4868 		init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
4869 		break;
4870 	default:
4871 		BUG();
4872 	}
4873 }
4874 
4875 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
4876 			      struct alloc_chunk_ctl *ctl,
4877 			      struct btrfs_device_info *devices_info)
4878 {
4879 	struct btrfs_fs_info *info = fs_devices->fs_info;
4880 	struct btrfs_device *device;
4881 	u64 total_avail;
4882 	u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
4883 	int ret;
4884 	int ndevs = 0;
4885 	u64 max_avail;
4886 	u64 dev_offset;
4887 
4888 	/*
4889 	 * in the first pass through the devices list, we gather information
4890 	 * about the available holes on each device.
4891 	 */
4892 	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4893 		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4894 			WARN(1, KERN_ERR
4895 			       "BTRFS: read-only device in alloc_list\n");
4896 			continue;
4897 		}
4898 
4899 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4900 					&device->dev_state) ||
4901 		    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4902 			continue;
4903 
4904 		if (device->total_bytes > device->bytes_used)
4905 			total_avail = device->total_bytes - device->bytes_used;
4906 		else
4907 			total_avail = 0;
4908 
4909 		/* If there is no space on this device, skip it. */
4910 		if (total_avail < ctl->dev_extent_min)
4911 			continue;
4912 
4913 		ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
4914 					   &max_avail);
4915 		if (ret && ret != -ENOSPC)
4916 			return ret;
4917 
4918 		if (ret == 0)
4919 			max_avail = dev_extent_want;
4920 
4921 		if (max_avail < ctl->dev_extent_min) {
4922 			if (btrfs_test_opt(info, ENOSPC_DEBUG))
4923 				btrfs_debug(info,
4924 			"%s: devid %llu has no free space, have=%llu want=%llu",
4925 					    __func__, device->devid, max_avail,
4926 					    ctl->dev_extent_min);
4927 			continue;
4928 		}
4929 
4930 		if (ndevs == fs_devices->rw_devices) {
4931 			WARN(1, "%s: found more than %llu devices\n",
4932 			     __func__, fs_devices->rw_devices);
4933 			break;
4934 		}
4935 		devices_info[ndevs].dev_offset = dev_offset;
4936 		devices_info[ndevs].max_avail = max_avail;
4937 		devices_info[ndevs].total_avail = total_avail;
4938 		devices_info[ndevs].dev = device;
4939 		++ndevs;
4940 	}
4941 	ctl->ndevs = ndevs;
4942 
4943 	/*
4944 	 * now sort the devices by hole size / available space
4945 	 */
4946 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4947 	     btrfs_cmp_device_info, NULL);
4948 
4949 	return 0;
4950 }
4951 
4952 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
4953 				      struct btrfs_device_info *devices_info)
4954 {
4955 	/* Number of stripes that count for block group size */
4956 	int data_stripes;
4957 
4958 	/*
4959 	 * The primary goal is to maximize the number of stripes, so use as
4960 	 * many devices as possible, even if the stripes are not maximum sized.
4961 	 *
4962 	 * The DUP profile stores more than one stripe per device, the
4963 	 * max_avail is the total size so we have to adjust.
4964 	 */
4965 	ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
4966 				   ctl->dev_stripes);
4967 	ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
4968 
4969 	/* This will have to be fixed for RAID1 and RAID10 over more drives */
4970 	data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
4971 
4972 	/*
4973 	 * Use the number of data stripes to figure out how big this chunk is
4974 	 * really going to be in terms of logical address space, and compare
4975 	 * that answer with the max chunk size. If it's higher, we try to
4976 	 * reduce stripe_size.
4977 	 */
4978 	if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
4979 		/*
4980 		 * Reduce stripe_size, round it up to a 16MB boundary again and
4981 		 * then use it, unless it ends up being even bigger than the
4982 		 * previous value we had already.
4983 		 */
4984 		ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
4985 							data_stripes), SZ_16M),
4986 				       ctl->stripe_size);
4987 	}
4988 
4989 	/* Align to BTRFS_STRIPE_LEN */
4990 	ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
4991 	ctl->chunk_size = ctl->stripe_size * data_stripes;
4992 
4993 	return 0;
4994 }
4995 
4996 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
4997 			      struct alloc_chunk_ctl *ctl,
4998 			      struct btrfs_device_info *devices_info)
4999 {
5000 	struct btrfs_fs_info *info = fs_devices->fs_info;
5001 
5002 	/*
5003 	 * Round down to number of usable stripes, devs_increment can be any
5004 	 * number so we can't use round_down() that requires power of 2, while
5005 	 * rounddown is safe.
5006 	 */
5007 	ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5008 
5009 	if (ctl->ndevs < ctl->devs_min) {
5010 		if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5011 			btrfs_debug(info,
5012 	"%s: not enough devices with free space: have=%d minimum required=%d",
5013 				    __func__, ctl->ndevs, ctl->devs_min);
5014 		}
5015 		return -ENOSPC;
5016 	}
5017 
5018 	ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5019 
5020 	switch (fs_devices->chunk_alloc_policy) {
5021 	case BTRFS_CHUNK_ALLOC_REGULAR:
5022 		return decide_stripe_size_regular(ctl, devices_info);
5023 	default:
5024 		BUG();
5025 	}
5026 }
5027 
5028 static int create_chunk(struct btrfs_trans_handle *trans,
5029 			struct alloc_chunk_ctl *ctl,
5030 			struct btrfs_device_info *devices_info)
5031 {
5032 	struct btrfs_fs_info *info = trans->fs_info;
5033 	struct map_lookup *map = NULL;
5034 	struct extent_map_tree *em_tree;
5035 	struct extent_map *em;
5036 	u64 start = ctl->start;
5037 	u64 type = ctl->type;
5038 	int ret;
5039 	int i;
5040 	int j;
5041 
5042 	map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5043 	if (!map)
5044 		return -ENOMEM;
5045 	map->num_stripes = ctl->num_stripes;
5046 
5047 	for (i = 0; i < ctl->ndevs; ++i) {
5048 		for (j = 0; j < ctl->dev_stripes; ++j) {
5049 			int s = i * ctl->dev_stripes + j;
5050 			map->stripes[s].dev = devices_info[i].dev;
5051 			map->stripes[s].physical = devices_info[i].dev_offset +
5052 						   j * ctl->stripe_size;
5053 		}
5054 	}
5055 	map->stripe_len = BTRFS_STRIPE_LEN;
5056 	map->io_align = BTRFS_STRIPE_LEN;
5057 	map->io_width = BTRFS_STRIPE_LEN;
5058 	map->type = type;
5059 	map->sub_stripes = ctl->sub_stripes;
5060 
5061 	trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5062 
5063 	em = alloc_extent_map();
5064 	if (!em) {
5065 		kfree(map);
5066 		return -ENOMEM;
5067 	}
5068 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5069 	em->map_lookup = map;
5070 	em->start = start;
5071 	em->len = ctl->chunk_size;
5072 	em->block_start = 0;
5073 	em->block_len = em->len;
5074 	em->orig_block_len = ctl->stripe_size;
5075 
5076 	em_tree = &info->mapping_tree;
5077 	write_lock(&em_tree->lock);
5078 	ret = add_extent_mapping(em_tree, em, 0);
5079 	if (ret) {
5080 		write_unlock(&em_tree->lock);
5081 		free_extent_map(em);
5082 		return ret;
5083 	}
5084 	write_unlock(&em_tree->lock);
5085 
5086 	ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5087 	if (ret)
5088 		goto error_del_extent;
5089 
5090 	for (i = 0; i < map->num_stripes; i++) {
5091 		struct btrfs_device *dev = map->stripes[i].dev;
5092 
5093 		btrfs_device_set_bytes_used(dev,
5094 					    dev->bytes_used + ctl->stripe_size);
5095 		if (list_empty(&dev->post_commit_list))
5096 			list_add_tail(&dev->post_commit_list,
5097 				      &trans->transaction->dev_update_list);
5098 	}
5099 
5100 	atomic64_sub(ctl->stripe_size * map->num_stripes,
5101 		     &info->free_chunk_space);
5102 
5103 	free_extent_map(em);
5104 	check_raid56_incompat_flag(info, type);
5105 	check_raid1c34_incompat_flag(info, type);
5106 
5107 	return 0;
5108 
5109 error_del_extent:
5110 	write_lock(&em_tree->lock);
5111 	remove_extent_mapping(em_tree, em);
5112 	write_unlock(&em_tree->lock);
5113 
5114 	/* One for our allocation */
5115 	free_extent_map(em);
5116 	/* One for the tree reference */
5117 	free_extent_map(em);
5118 
5119 	return ret;
5120 }
5121 
5122 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5123 {
5124 	struct btrfs_fs_info *info = trans->fs_info;
5125 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
5126 	struct btrfs_device_info *devices_info = NULL;
5127 	struct alloc_chunk_ctl ctl;
5128 	int ret;
5129 
5130 	lockdep_assert_held(&info->chunk_mutex);
5131 
5132 	if (!alloc_profile_is_valid(type, 0)) {
5133 		ASSERT(0);
5134 		return -EINVAL;
5135 	}
5136 
5137 	if (list_empty(&fs_devices->alloc_list)) {
5138 		if (btrfs_test_opt(info, ENOSPC_DEBUG))
5139 			btrfs_debug(info, "%s: no writable device", __func__);
5140 		return -ENOSPC;
5141 	}
5142 
5143 	if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5144 		btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5145 		ASSERT(0);
5146 		return -EINVAL;
5147 	}
5148 
5149 	ctl.start = find_next_chunk(info);
5150 	ctl.type = type;
5151 	init_alloc_chunk_ctl(fs_devices, &ctl);
5152 
5153 	devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5154 			       GFP_NOFS);
5155 	if (!devices_info)
5156 		return -ENOMEM;
5157 
5158 	ret = gather_device_info(fs_devices, &ctl, devices_info);
5159 	if (ret < 0)
5160 		goto out;
5161 
5162 	ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5163 	if (ret < 0)
5164 		goto out;
5165 
5166 	ret = create_chunk(trans, &ctl, devices_info);
5167 
5168 out:
5169 	kfree(devices_info);
5170 	return ret;
5171 }
5172 
5173 /*
5174  * Chunk allocation falls into two parts. The first part does work
5175  * that makes the new allocated chunk usable, but does not do any operation
5176  * that modifies the chunk tree. The second part does the work that
5177  * requires modifying the chunk tree. This division is important for the
5178  * bootstrap process of adding storage to a seed btrfs.
5179  */
5180 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5181 			     u64 chunk_offset, u64 chunk_size)
5182 {
5183 	struct btrfs_fs_info *fs_info = trans->fs_info;
5184 	struct btrfs_root *extent_root = fs_info->extent_root;
5185 	struct btrfs_root *chunk_root = fs_info->chunk_root;
5186 	struct btrfs_key key;
5187 	struct btrfs_device *device;
5188 	struct btrfs_chunk *chunk;
5189 	struct btrfs_stripe *stripe;
5190 	struct extent_map *em;
5191 	struct map_lookup *map;
5192 	size_t item_size;
5193 	u64 dev_offset;
5194 	u64 stripe_size;
5195 	int i = 0;
5196 	int ret = 0;
5197 
5198 	em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5199 	if (IS_ERR(em))
5200 		return PTR_ERR(em);
5201 
5202 	map = em->map_lookup;
5203 	item_size = btrfs_chunk_item_size(map->num_stripes);
5204 	stripe_size = em->orig_block_len;
5205 
5206 	chunk = kzalloc(item_size, GFP_NOFS);
5207 	if (!chunk) {
5208 		ret = -ENOMEM;
5209 		goto out;
5210 	}
5211 
5212 	/*
5213 	 * Take the device list mutex to prevent races with the final phase of
5214 	 * a device replace operation that replaces the device object associated
5215 	 * with the map's stripes, because the device object's id can change
5216 	 * at any time during that final phase of the device replace operation
5217 	 * (dev-replace.c:btrfs_dev_replace_finishing()).
5218 	 */
5219 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
5220 	for (i = 0; i < map->num_stripes; i++) {
5221 		device = map->stripes[i].dev;
5222 		dev_offset = map->stripes[i].physical;
5223 
5224 		ret = btrfs_update_device(trans, device);
5225 		if (ret)
5226 			break;
5227 		ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5228 					     dev_offset, stripe_size);
5229 		if (ret)
5230 			break;
5231 	}
5232 	if (ret) {
5233 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5234 		goto out;
5235 	}
5236 
5237 	stripe = &chunk->stripe;
5238 	for (i = 0; i < map->num_stripes; i++) {
5239 		device = map->stripes[i].dev;
5240 		dev_offset = map->stripes[i].physical;
5241 
5242 		btrfs_set_stack_stripe_devid(stripe, device->devid);
5243 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
5244 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5245 		stripe++;
5246 	}
5247 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5248 
5249 	btrfs_set_stack_chunk_length(chunk, chunk_size);
5250 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5251 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5252 	btrfs_set_stack_chunk_type(chunk, map->type);
5253 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5254 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5255 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5256 	btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5257 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5258 
5259 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5260 	key.type = BTRFS_CHUNK_ITEM_KEY;
5261 	key.offset = chunk_offset;
5262 
5263 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5264 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5265 		/*
5266 		 * TODO: Cleanup of inserted chunk root in case of
5267 		 * failure.
5268 		 */
5269 		ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5270 	}
5271 
5272 out:
5273 	kfree(chunk);
5274 	free_extent_map(em);
5275 	return ret;
5276 }
5277 
5278 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5279 {
5280 	struct btrfs_fs_info *fs_info = trans->fs_info;
5281 	u64 alloc_profile;
5282 	int ret;
5283 
5284 	alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5285 	ret = btrfs_alloc_chunk(trans, alloc_profile);
5286 	if (ret)
5287 		return ret;
5288 
5289 	alloc_profile = btrfs_system_alloc_profile(fs_info);
5290 	ret = btrfs_alloc_chunk(trans, alloc_profile);
5291 	return ret;
5292 }
5293 
5294 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5295 {
5296 	const int index = btrfs_bg_flags_to_raid_index(map->type);
5297 
5298 	return btrfs_raid_array[index].tolerated_failures;
5299 }
5300 
5301 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5302 {
5303 	struct extent_map *em;
5304 	struct map_lookup *map;
5305 	int readonly = 0;
5306 	int miss_ndevs = 0;
5307 	int i;
5308 
5309 	em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5310 	if (IS_ERR(em))
5311 		return 1;
5312 
5313 	map = em->map_lookup;
5314 	for (i = 0; i < map->num_stripes; i++) {
5315 		if (test_bit(BTRFS_DEV_STATE_MISSING,
5316 					&map->stripes[i].dev->dev_state)) {
5317 			miss_ndevs++;
5318 			continue;
5319 		}
5320 		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5321 					&map->stripes[i].dev->dev_state)) {
5322 			readonly = 1;
5323 			goto end;
5324 		}
5325 	}
5326 
5327 	/*
5328 	 * If the number of missing devices is larger than max errors,
5329 	 * we can not write the data into that chunk successfully, so
5330 	 * set it readonly.
5331 	 */
5332 	if (miss_ndevs > btrfs_chunk_max_errors(map))
5333 		readonly = 1;
5334 end:
5335 	free_extent_map(em);
5336 	return readonly;
5337 }
5338 
5339 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5340 {
5341 	struct extent_map *em;
5342 
5343 	while (1) {
5344 		write_lock(&tree->lock);
5345 		em = lookup_extent_mapping(tree, 0, (u64)-1);
5346 		if (em)
5347 			remove_extent_mapping(tree, em);
5348 		write_unlock(&tree->lock);
5349 		if (!em)
5350 			break;
5351 		/* once for us */
5352 		free_extent_map(em);
5353 		/* once for the tree */
5354 		free_extent_map(em);
5355 	}
5356 }
5357 
5358 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5359 {
5360 	struct extent_map *em;
5361 	struct map_lookup *map;
5362 	int ret;
5363 
5364 	em = btrfs_get_chunk_map(fs_info, logical, len);
5365 	if (IS_ERR(em))
5366 		/*
5367 		 * We could return errors for these cases, but that could get
5368 		 * ugly and we'd probably do the same thing which is just not do
5369 		 * anything else and exit, so return 1 so the callers don't try
5370 		 * to use other copies.
5371 		 */
5372 		return 1;
5373 
5374 	map = em->map_lookup;
5375 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5376 		ret = map->num_stripes;
5377 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5378 		ret = map->sub_stripes;
5379 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5380 		ret = 2;
5381 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5382 		/*
5383 		 * There could be two corrupted data stripes, we need
5384 		 * to loop retry in order to rebuild the correct data.
5385 		 *
5386 		 * Fail a stripe at a time on every retry except the
5387 		 * stripe under reconstruction.
5388 		 */
5389 		ret = map->num_stripes;
5390 	else
5391 		ret = 1;
5392 	free_extent_map(em);
5393 
5394 	down_read(&fs_info->dev_replace.rwsem);
5395 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5396 	    fs_info->dev_replace.tgtdev)
5397 		ret++;
5398 	up_read(&fs_info->dev_replace.rwsem);
5399 
5400 	return ret;
5401 }
5402 
5403 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5404 				    u64 logical)
5405 {
5406 	struct extent_map *em;
5407 	struct map_lookup *map;
5408 	unsigned long len = fs_info->sectorsize;
5409 
5410 	em = btrfs_get_chunk_map(fs_info, logical, len);
5411 
5412 	if (!WARN_ON(IS_ERR(em))) {
5413 		map = em->map_lookup;
5414 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5415 			len = map->stripe_len * nr_data_stripes(map);
5416 		free_extent_map(em);
5417 	}
5418 	return len;
5419 }
5420 
5421 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5422 {
5423 	struct extent_map *em;
5424 	struct map_lookup *map;
5425 	int ret = 0;
5426 
5427 	em = btrfs_get_chunk_map(fs_info, logical, len);
5428 
5429 	if(!WARN_ON(IS_ERR(em))) {
5430 		map = em->map_lookup;
5431 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5432 			ret = 1;
5433 		free_extent_map(em);
5434 	}
5435 	return ret;
5436 }
5437 
5438 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5439 			    struct map_lookup *map, int first,
5440 			    int dev_replace_is_ongoing)
5441 {
5442 	int i;
5443 	int num_stripes;
5444 	int preferred_mirror;
5445 	int tolerance;
5446 	struct btrfs_device *srcdev;
5447 
5448 	ASSERT((map->type &
5449 		 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5450 
5451 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5452 		num_stripes = map->sub_stripes;
5453 	else
5454 		num_stripes = map->num_stripes;
5455 
5456 	preferred_mirror = first + current->pid % num_stripes;
5457 
5458 	if (dev_replace_is_ongoing &&
5459 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5460 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5461 		srcdev = fs_info->dev_replace.srcdev;
5462 	else
5463 		srcdev = NULL;
5464 
5465 	/*
5466 	 * try to avoid the drive that is the source drive for a
5467 	 * dev-replace procedure, only choose it if no other non-missing
5468 	 * mirror is available
5469 	 */
5470 	for (tolerance = 0; tolerance < 2; tolerance++) {
5471 		if (map->stripes[preferred_mirror].dev->bdev &&
5472 		    (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5473 			return preferred_mirror;
5474 		for (i = first; i < first + num_stripes; i++) {
5475 			if (map->stripes[i].dev->bdev &&
5476 			    (tolerance || map->stripes[i].dev != srcdev))
5477 				return i;
5478 		}
5479 	}
5480 
5481 	/* we couldn't find one that doesn't fail.  Just return something
5482 	 * and the io error handling code will clean up eventually
5483 	 */
5484 	return preferred_mirror;
5485 }
5486 
5487 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5488 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5489 {
5490 	int i;
5491 	int again = 1;
5492 
5493 	while (again) {
5494 		again = 0;
5495 		for (i = 0; i < num_stripes - 1; i++) {
5496 			/* Swap if parity is on a smaller index */
5497 			if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5498 				swap(bbio->stripes[i], bbio->stripes[i + 1]);
5499 				swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5500 				again = 1;
5501 			}
5502 		}
5503 	}
5504 }
5505 
5506 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5507 {
5508 	struct btrfs_bio *bbio = kzalloc(
5509 		 /* the size of the btrfs_bio */
5510 		sizeof(struct btrfs_bio) +
5511 		/* plus the variable array for the stripes */
5512 		sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5513 		/* plus the variable array for the tgt dev */
5514 		sizeof(int) * (real_stripes) +
5515 		/*
5516 		 * plus the raid_map, which includes both the tgt dev
5517 		 * and the stripes
5518 		 */
5519 		sizeof(u64) * (total_stripes),
5520 		GFP_NOFS|__GFP_NOFAIL);
5521 
5522 	atomic_set(&bbio->error, 0);
5523 	refcount_set(&bbio->refs, 1);
5524 
5525 	return bbio;
5526 }
5527 
5528 void btrfs_get_bbio(struct btrfs_bio *bbio)
5529 {
5530 	WARN_ON(!refcount_read(&bbio->refs));
5531 	refcount_inc(&bbio->refs);
5532 }
5533 
5534 void btrfs_put_bbio(struct btrfs_bio *bbio)
5535 {
5536 	if (!bbio)
5537 		return;
5538 	if (refcount_dec_and_test(&bbio->refs))
5539 		kfree(bbio);
5540 }
5541 
5542 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5543 /*
5544  * Please note that, discard won't be sent to target device of device
5545  * replace.
5546  */
5547 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5548 					 u64 logical, u64 *length_ret,
5549 					 struct btrfs_bio **bbio_ret)
5550 {
5551 	struct extent_map *em;
5552 	struct map_lookup *map;
5553 	struct btrfs_bio *bbio;
5554 	u64 length = *length_ret;
5555 	u64 offset;
5556 	u64 stripe_nr;
5557 	u64 stripe_nr_end;
5558 	u64 stripe_end_offset;
5559 	u64 stripe_cnt;
5560 	u64 stripe_len;
5561 	u64 stripe_offset;
5562 	u64 num_stripes;
5563 	u32 stripe_index;
5564 	u32 factor = 0;
5565 	u32 sub_stripes = 0;
5566 	u64 stripes_per_dev = 0;
5567 	u32 remaining_stripes = 0;
5568 	u32 last_stripe = 0;
5569 	int ret = 0;
5570 	int i;
5571 
5572 	/* discard always return a bbio */
5573 	ASSERT(bbio_ret);
5574 
5575 	em = btrfs_get_chunk_map(fs_info, logical, length);
5576 	if (IS_ERR(em))
5577 		return PTR_ERR(em);
5578 
5579 	map = em->map_lookup;
5580 	/* we don't discard raid56 yet */
5581 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5582 		ret = -EOPNOTSUPP;
5583 		goto out;
5584 	}
5585 
5586 	offset = logical - em->start;
5587 	length = min_t(u64, em->start + em->len - logical, length);
5588 	*length_ret = length;
5589 
5590 	stripe_len = map->stripe_len;
5591 	/*
5592 	 * stripe_nr counts the total number of stripes we have to stride
5593 	 * to get to this block
5594 	 */
5595 	stripe_nr = div64_u64(offset, stripe_len);
5596 
5597 	/* stripe_offset is the offset of this block in its stripe */
5598 	stripe_offset = offset - stripe_nr * stripe_len;
5599 
5600 	stripe_nr_end = round_up(offset + length, map->stripe_len);
5601 	stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5602 	stripe_cnt = stripe_nr_end - stripe_nr;
5603 	stripe_end_offset = stripe_nr_end * map->stripe_len -
5604 			    (offset + length);
5605 	/*
5606 	 * after this, stripe_nr is the number of stripes on this
5607 	 * device we have to walk to find the data, and stripe_index is
5608 	 * the number of our device in the stripe array
5609 	 */
5610 	num_stripes = 1;
5611 	stripe_index = 0;
5612 	if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5613 			 BTRFS_BLOCK_GROUP_RAID10)) {
5614 		if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5615 			sub_stripes = 1;
5616 		else
5617 			sub_stripes = map->sub_stripes;
5618 
5619 		factor = map->num_stripes / sub_stripes;
5620 		num_stripes = min_t(u64, map->num_stripes,
5621 				    sub_stripes * stripe_cnt);
5622 		stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5623 		stripe_index *= sub_stripes;
5624 		stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5625 					      &remaining_stripes);
5626 		div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5627 		last_stripe *= sub_stripes;
5628 	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5629 				BTRFS_BLOCK_GROUP_DUP)) {
5630 		num_stripes = map->num_stripes;
5631 	} else {
5632 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5633 					&stripe_index);
5634 	}
5635 
5636 	bbio = alloc_btrfs_bio(num_stripes, 0);
5637 	if (!bbio) {
5638 		ret = -ENOMEM;
5639 		goto out;
5640 	}
5641 
5642 	for (i = 0; i < num_stripes; i++) {
5643 		bbio->stripes[i].physical =
5644 			map->stripes[stripe_index].physical +
5645 			stripe_offset + stripe_nr * map->stripe_len;
5646 		bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5647 
5648 		if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5649 				 BTRFS_BLOCK_GROUP_RAID10)) {
5650 			bbio->stripes[i].length = stripes_per_dev *
5651 				map->stripe_len;
5652 
5653 			if (i / sub_stripes < remaining_stripes)
5654 				bbio->stripes[i].length +=
5655 					map->stripe_len;
5656 
5657 			/*
5658 			 * Special for the first stripe and
5659 			 * the last stripe:
5660 			 *
5661 			 * |-------|...|-------|
5662 			 *     |----------|
5663 			 *    off     end_off
5664 			 */
5665 			if (i < sub_stripes)
5666 				bbio->stripes[i].length -=
5667 					stripe_offset;
5668 
5669 			if (stripe_index >= last_stripe &&
5670 			    stripe_index <= (last_stripe +
5671 					     sub_stripes - 1))
5672 				bbio->stripes[i].length -=
5673 					stripe_end_offset;
5674 
5675 			if (i == sub_stripes - 1)
5676 				stripe_offset = 0;
5677 		} else {
5678 			bbio->stripes[i].length = length;
5679 		}
5680 
5681 		stripe_index++;
5682 		if (stripe_index == map->num_stripes) {
5683 			stripe_index = 0;
5684 			stripe_nr++;
5685 		}
5686 	}
5687 
5688 	*bbio_ret = bbio;
5689 	bbio->map_type = map->type;
5690 	bbio->num_stripes = num_stripes;
5691 out:
5692 	free_extent_map(em);
5693 	return ret;
5694 }
5695 
5696 /*
5697  * In dev-replace case, for repair case (that's the only case where the mirror
5698  * is selected explicitly when calling btrfs_map_block), blocks left of the
5699  * left cursor can also be read from the target drive.
5700  *
5701  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5702  * array of stripes.
5703  * For READ, it also needs to be supported using the same mirror number.
5704  *
5705  * If the requested block is not left of the left cursor, EIO is returned. This
5706  * can happen because btrfs_num_copies() returns one more in the dev-replace
5707  * case.
5708  */
5709 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5710 					 u64 logical, u64 length,
5711 					 u64 srcdev_devid, int *mirror_num,
5712 					 u64 *physical)
5713 {
5714 	struct btrfs_bio *bbio = NULL;
5715 	int num_stripes;
5716 	int index_srcdev = 0;
5717 	int found = 0;
5718 	u64 physical_of_found = 0;
5719 	int i;
5720 	int ret = 0;
5721 
5722 	ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5723 				logical, &length, &bbio, 0, 0);
5724 	if (ret) {
5725 		ASSERT(bbio == NULL);
5726 		return ret;
5727 	}
5728 
5729 	num_stripes = bbio->num_stripes;
5730 	if (*mirror_num > num_stripes) {
5731 		/*
5732 		 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5733 		 * that means that the requested area is not left of the left
5734 		 * cursor
5735 		 */
5736 		btrfs_put_bbio(bbio);
5737 		return -EIO;
5738 	}
5739 
5740 	/*
5741 	 * process the rest of the function using the mirror_num of the source
5742 	 * drive. Therefore look it up first.  At the end, patch the device
5743 	 * pointer to the one of the target drive.
5744 	 */
5745 	for (i = 0; i < num_stripes; i++) {
5746 		if (bbio->stripes[i].dev->devid != srcdev_devid)
5747 			continue;
5748 
5749 		/*
5750 		 * In case of DUP, in order to keep it simple, only add the
5751 		 * mirror with the lowest physical address
5752 		 */
5753 		if (found &&
5754 		    physical_of_found <= bbio->stripes[i].physical)
5755 			continue;
5756 
5757 		index_srcdev = i;
5758 		found = 1;
5759 		physical_of_found = bbio->stripes[i].physical;
5760 	}
5761 
5762 	btrfs_put_bbio(bbio);
5763 
5764 	ASSERT(found);
5765 	if (!found)
5766 		return -EIO;
5767 
5768 	*mirror_num = index_srcdev + 1;
5769 	*physical = physical_of_found;
5770 	return ret;
5771 }
5772 
5773 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5774 				      struct btrfs_bio **bbio_ret,
5775 				      struct btrfs_dev_replace *dev_replace,
5776 				      int *num_stripes_ret, int *max_errors_ret)
5777 {
5778 	struct btrfs_bio *bbio = *bbio_ret;
5779 	u64 srcdev_devid = dev_replace->srcdev->devid;
5780 	int tgtdev_indexes = 0;
5781 	int num_stripes = *num_stripes_ret;
5782 	int max_errors = *max_errors_ret;
5783 	int i;
5784 
5785 	if (op == BTRFS_MAP_WRITE) {
5786 		int index_where_to_add;
5787 
5788 		/*
5789 		 * duplicate the write operations while the dev replace
5790 		 * procedure is running. Since the copying of the old disk to
5791 		 * the new disk takes place at run time while the filesystem is
5792 		 * mounted writable, the regular write operations to the old
5793 		 * disk have to be duplicated to go to the new disk as well.
5794 		 *
5795 		 * Note that device->missing is handled by the caller, and that
5796 		 * the write to the old disk is already set up in the stripes
5797 		 * array.
5798 		 */
5799 		index_where_to_add = num_stripes;
5800 		for (i = 0; i < num_stripes; i++) {
5801 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5802 				/* write to new disk, too */
5803 				struct btrfs_bio_stripe *new =
5804 					bbio->stripes + index_where_to_add;
5805 				struct btrfs_bio_stripe *old =
5806 					bbio->stripes + i;
5807 
5808 				new->physical = old->physical;
5809 				new->length = old->length;
5810 				new->dev = dev_replace->tgtdev;
5811 				bbio->tgtdev_map[i] = index_where_to_add;
5812 				index_where_to_add++;
5813 				max_errors++;
5814 				tgtdev_indexes++;
5815 			}
5816 		}
5817 		num_stripes = index_where_to_add;
5818 	} else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5819 		int index_srcdev = 0;
5820 		int found = 0;
5821 		u64 physical_of_found = 0;
5822 
5823 		/*
5824 		 * During the dev-replace procedure, the target drive can also
5825 		 * be used to read data in case it is needed to repair a corrupt
5826 		 * block elsewhere. This is possible if the requested area is
5827 		 * left of the left cursor. In this area, the target drive is a
5828 		 * full copy of the source drive.
5829 		 */
5830 		for (i = 0; i < num_stripes; i++) {
5831 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5832 				/*
5833 				 * In case of DUP, in order to keep it simple,
5834 				 * only add the mirror with the lowest physical
5835 				 * address
5836 				 */
5837 				if (found &&
5838 				    physical_of_found <=
5839 				     bbio->stripes[i].physical)
5840 					continue;
5841 				index_srcdev = i;
5842 				found = 1;
5843 				physical_of_found = bbio->stripes[i].physical;
5844 			}
5845 		}
5846 		if (found) {
5847 			struct btrfs_bio_stripe *tgtdev_stripe =
5848 				bbio->stripes + num_stripes;
5849 
5850 			tgtdev_stripe->physical = physical_of_found;
5851 			tgtdev_stripe->length =
5852 				bbio->stripes[index_srcdev].length;
5853 			tgtdev_stripe->dev = dev_replace->tgtdev;
5854 			bbio->tgtdev_map[index_srcdev] = num_stripes;
5855 
5856 			tgtdev_indexes++;
5857 			num_stripes++;
5858 		}
5859 	}
5860 
5861 	*num_stripes_ret = num_stripes;
5862 	*max_errors_ret = max_errors;
5863 	bbio->num_tgtdevs = tgtdev_indexes;
5864 	*bbio_ret = bbio;
5865 }
5866 
5867 static bool need_full_stripe(enum btrfs_map_op op)
5868 {
5869 	return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5870 }
5871 
5872 /*
5873  * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5874  *		       tuple. This information is used to calculate how big a
5875  *		       particular bio can get before it straddles a stripe.
5876  *
5877  * @fs_info - the filesystem
5878  * @logical - address that we want to figure out the geometry of
5879  * @len	    - the length of IO we are going to perform, starting at @logical
5880  * @op      - type of operation - write or read
5881  * @io_geom - pointer used to return values
5882  *
5883  * Returns < 0 in case a chunk for the given logical address cannot be found,
5884  * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5885  */
5886 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5887 			u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5888 {
5889 	struct extent_map *em;
5890 	struct map_lookup *map;
5891 	u64 offset;
5892 	u64 stripe_offset;
5893 	u64 stripe_nr;
5894 	u64 stripe_len;
5895 	u64 raid56_full_stripe_start = (u64)-1;
5896 	int data_stripes;
5897 	int ret = 0;
5898 
5899 	ASSERT(op != BTRFS_MAP_DISCARD);
5900 
5901 	em = btrfs_get_chunk_map(fs_info, logical, len);
5902 	if (IS_ERR(em))
5903 		return PTR_ERR(em);
5904 
5905 	map = em->map_lookup;
5906 	/* Offset of this logical address in the chunk */
5907 	offset = logical - em->start;
5908 	/* Len of a stripe in a chunk */
5909 	stripe_len = map->stripe_len;
5910 	/* Stripe wher this block falls in */
5911 	stripe_nr = div64_u64(offset, stripe_len);
5912 	/* Offset of stripe in the chunk */
5913 	stripe_offset = stripe_nr * stripe_len;
5914 	if (offset < stripe_offset) {
5915 		btrfs_crit(fs_info,
5916 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5917 			stripe_offset, offset, em->start, logical, stripe_len);
5918 		ret = -EINVAL;
5919 		goto out;
5920 	}
5921 
5922 	/* stripe_offset is the offset of this block in its stripe */
5923 	stripe_offset = offset - stripe_offset;
5924 	data_stripes = nr_data_stripes(map);
5925 
5926 	if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5927 		u64 max_len = stripe_len - stripe_offset;
5928 
5929 		/*
5930 		 * In case of raid56, we need to know the stripe aligned start
5931 		 */
5932 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5933 			unsigned long full_stripe_len = stripe_len * data_stripes;
5934 			raid56_full_stripe_start = offset;
5935 
5936 			/*
5937 			 * Allow a write of a full stripe, but make sure we
5938 			 * don't allow straddling of stripes
5939 			 */
5940 			raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5941 					full_stripe_len);
5942 			raid56_full_stripe_start *= full_stripe_len;
5943 
5944 			/*
5945 			 * For writes to RAID[56], allow a full stripeset across
5946 			 * all disks. For other RAID types and for RAID[56]
5947 			 * reads, just allow a single stripe (on a single disk).
5948 			 */
5949 			if (op == BTRFS_MAP_WRITE) {
5950 				max_len = stripe_len * data_stripes -
5951 					  (offset - raid56_full_stripe_start);
5952 			}
5953 		}
5954 		len = min_t(u64, em->len - offset, max_len);
5955 	} else {
5956 		len = em->len - offset;
5957 	}
5958 
5959 	io_geom->len = len;
5960 	io_geom->offset = offset;
5961 	io_geom->stripe_len = stripe_len;
5962 	io_geom->stripe_nr = stripe_nr;
5963 	io_geom->stripe_offset = stripe_offset;
5964 	io_geom->raid56_stripe_offset = raid56_full_stripe_start;
5965 
5966 out:
5967 	/* once for us */
5968 	free_extent_map(em);
5969 	return ret;
5970 }
5971 
5972 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5973 			     enum btrfs_map_op op,
5974 			     u64 logical, u64 *length,
5975 			     struct btrfs_bio **bbio_ret,
5976 			     int mirror_num, int need_raid_map)
5977 {
5978 	struct extent_map *em;
5979 	struct map_lookup *map;
5980 	u64 stripe_offset;
5981 	u64 stripe_nr;
5982 	u64 stripe_len;
5983 	u32 stripe_index;
5984 	int data_stripes;
5985 	int i;
5986 	int ret = 0;
5987 	int num_stripes;
5988 	int max_errors = 0;
5989 	int tgtdev_indexes = 0;
5990 	struct btrfs_bio *bbio = NULL;
5991 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5992 	int dev_replace_is_ongoing = 0;
5993 	int num_alloc_stripes;
5994 	int patch_the_first_stripe_for_dev_replace = 0;
5995 	u64 physical_to_patch_in_first_stripe = 0;
5996 	u64 raid56_full_stripe_start = (u64)-1;
5997 	struct btrfs_io_geometry geom;
5998 
5999 	ASSERT(bbio_ret);
6000 	ASSERT(op != BTRFS_MAP_DISCARD);
6001 
6002 	ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6003 	if (ret < 0)
6004 		return ret;
6005 
6006 	em = btrfs_get_chunk_map(fs_info, logical, *length);
6007 	ASSERT(!IS_ERR(em));
6008 	map = em->map_lookup;
6009 
6010 	*length = geom.len;
6011 	stripe_len = geom.stripe_len;
6012 	stripe_nr = geom.stripe_nr;
6013 	stripe_offset = geom.stripe_offset;
6014 	raid56_full_stripe_start = geom.raid56_stripe_offset;
6015 	data_stripes = nr_data_stripes(map);
6016 
6017 	down_read(&dev_replace->rwsem);
6018 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6019 	/*
6020 	 * Hold the semaphore for read during the whole operation, write is
6021 	 * requested at commit time but must wait.
6022 	 */
6023 	if (!dev_replace_is_ongoing)
6024 		up_read(&dev_replace->rwsem);
6025 
6026 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6027 	    !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6028 		ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6029 						    dev_replace->srcdev->devid,
6030 						    &mirror_num,
6031 					    &physical_to_patch_in_first_stripe);
6032 		if (ret)
6033 			goto out;
6034 		else
6035 			patch_the_first_stripe_for_dev_replace = 1;
6036 	} else if (mirror_num > map->num_stripes) {
6037 		mirror_num = 0;
6038 	}
6039 
6040 	num_stripes = 1;
6041 	stripe_index = 0;
6042 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6043 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6044 				&stripe_index);
6045 		if (!need_full_stripe(op))
6046 			mirror_num = 1;
6047 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6048 		if (need_full_stripe(op))
6049 			num_stripes = map->num_stripes;
6050 		else if (mirror_num)
6051 			stripe_index = mirror_num - 1;
6052 		else {
6053 			stripe_index = find_live_mirror(fs_info, map, 0,
6054 					    dev_replace_is_ongoing);
6055 			mirror_num = stripe_index + 1;
6056 		}
6057 
6058 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6059 		if (need_full_stripe(op)) {
6060 			num_stripes = map->num_stripes;
6061 		} else if (mirror_num) {
6062 			stripe_index = mirror_num - 1;
6063 		} else {
6064 			mirror_num = 1;
6065 		}
6066 
6067 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6068 		u32 factor = map->num_stripes / map->sub_stripes;
6069 
6070 		stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6071 		stripe_index *= map->sub_stripes;
6072 
6073 		if (need_full_stripe(op))
6074 			num_stripes = map->sub_stripes;
6075 		else if (mirror_num)
6076 			stripe_index += mirror_num - 1;
6077 		else {
6078 			int old_stripe_index = stripe_index;
6079 			stripe_index = find_live_mirror(fs_info, map,
6080 					      stripe_index,
6081 					      dev_replace_is_ongoing);
6082 			mirror_num = stripe_index - old_stripe_index + 1;
6083 		}
6084 
6085 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6086 		if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6087 			/* push stripe_nr back to the start of the full stripe */
6088 			stripe_nr = div64_u64(raid56_full_stripe_start,
6089 					stripe_len * data_stripes);
6090 
6091 			/* RAID[56] write or recovery. Return all stripes */
6092 			num_stripes = map->num_stripes;
6093 			max_errors = nr_parity_stripes(map);
6094 
6095 			*length = map->stripe_len;
6096 			stripe_index = 0;
6097 			stripe_offset = 0;
6098 		} else {
6099 			/*
6100 			 * Mirror #0 or #1 means the original data block.
6101 			 * Mirror #2 is RAID5 parity block.
6102 			 * Mirror #3 is RAID6 Q block.
6103 			 */
6104 			stripe_nr = div_u64_rem(stripe_nr,
6105 					data_stripes, &stripe_index);
6106 			if (mirror_num > 1)
6107 				stripe_index = data_stripes + mirror_num - 2;
6108 
6109 			/* We distribute the parity blocks across stripes */
6110 			div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6111 					&stripe_index);
6112 			if (!need_full_stripe(op) && mirror_num <= 1)
6113 				mirror_num = 1;
6114 		}
6115 	} else {
6116 		/*
6117 		 * after this, stripe_nr is the number of stripes on this
6118 		 * device we have to walk to find the data, and stripe_index is
6119 		 * the number of our device in the stripe array
6120 		 */
6121 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6122 				&stripe_index);
6123 		mirror_num = stripe_index + 1;
6124 	}
6125 	if (stripe_index >= map->num_stripes) {
6126 		btrfs_crit(fs_info,
6127 			   "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6128 			   stripe_index, map->num_stripes);
6129 		ret = -EINVAL;
6130 		goto out;
6131 	}
6132 
6133 	num_alloc_stripes = num_stripes;
6134 	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6135 		if (op == BTRFS_MAP_WRITE)
6136 			num_alloc_stripes <<= 1;
6137 		if (op == BTRFS_MAP_GET_READ_MIRRORS)
6138 			num_alloc_stripes++;
6139 		tgtdev_indexes = num_stripes;
6140 	}
6141 
6142 	bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6143 	if (!bbio) {
6144 		ret = -ENOMEM;
6145 		goto out;
6146 	}
6147 	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
6148 		bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
6149 
6150 	/* build raid_map */
6151 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6152 	    (need_full_stripe(op) || mirror_num > 1)) {
6153 		u64 tmp;
6154 		unsigned rot;
6155 
6156 		bbio->raid_map = (u64 *)((void *)bbio->stripes +
6157 				 sizeof(struct btrfs_bio_stripe) *
6158 				 num_alloc_stripes +
6159 				 sizeof(int) * tgtdev_indexes);
6160 
6161 		/* Work out the disk rotation on this stripe-set */
6162 		div_u64_rem(stripe_nr, num_stripes, &rot);
6163 
6164 		/* Fill in the logical address of each stripe */
6165 		tmp = stripe_nr * data_stripes;
6166 		for (i = 0; i < data_stripes; i++)
6167 			bbio->raid_map[(i+rot) % num_stripes] =
6168 				em->start + (tmp + i) * map->stripe_len;
6169 
6170 		bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6171 		if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6172 			bbio->raid_map[(i+rot+1) % num_stripes] =
6173 				RAID6_Q_STRIPE;
6174 	}
6175 
6176 
6177 	for (i = 0; i < num_stripes; i++) {
6178 		bbio->stripes[i].physical =
6179 			map->stripes[stripe_index].physical +
6180 			stripe_offset +
6181 			stripe_nr * map->stripe_len;
6182 		bbio->stripes[i].dev =
6183 			map->stripes[stripe_index].dev;
6184 		stripe_index++;
6185 	}
6186 
6187 	if (need_full_stripe(op))
6188 		max_errors = btrfs_chunk_max_errors(map);
6189 
6190 	if (bbio->raid_map)
6191 		sort_parity_stripes(bbio, num_stripes);
6192 
6193 	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6194 	    need_full_stripe(op)) {
6195 		handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6196 					  &max_errors);
6197 	}
6198 
6199 	*bbio_ret = bbio;
6200 	bbio->map_type = map->type;
6201 	bbio->num_stripes = num_stripes;
6202 	bbio->max_errors = max_errors;
6203 	bbio->mirror_num = mirror_num;
6204 
6205 	/*
6206 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6207 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
6208 	 * available as a mirror
6209 	 */
6210 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6211 		WARN_ON(num_stripes > 1);
6212 		bbio->stripes[0].dev = dev_replace->tgtdev;
6213 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6214 		bbio->mirror_num = map->num_stripes + 1;
6215 	}
6216 out:
6217 	if (dev_replace_is_ongoing) {
6218 		lockdep_assert_held(&dev_replace->rwsem);
6219 		/* Unlock and let waiting writers proceed */
6220 		up_read(&dev_replace->rwsem);
6221 	}
6222 	free_extent_map(em);
6223 	return ret;
6224 }
6225 
6226 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6227 		      u64 logical, u64 *length,
6228 		      struct btrfs_bio **bbio_ret, int mirror_num)
6229 {
6230 	if (op == BTRFS_MAP_DISCARD)
6231 		return __btrfs_map_block_for_discard(fs_info, logical,
6232 						     length, bbio_ret);
6233 
6234 	return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6235 				 mirror_num, 0);
6236 }
6237 
6238 /* For Scrub/replace */
6239 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6240 		     u64 logical, u64 *length,
6241 		     struct btrfs_bio **bbio_ret)
6242 {
6243 	return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6244 }
6245 
6246 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6247 {
6248 	bio->bi_private = bbio->private;
6249 	bio->bi_end_io = bbio->end_io;
6250 	bio_endio(bio);
6251 
6252 	btrfs_put_bbio(bbio);
6253 }
6254 
6255 static void btrfs_end_bio(struct bio *bio)
6256 {
6257 	struct btrfs_bio *bbio = bio->bi_private;
6258 	int is_orig_bio = 0;
6259 
6260 	if (bio->bi_status) {
6261 		atomic_inc(&bbio->error);
6262 		if (bio->bi_status == BLK_STS_IOERR ||
6263 		    bio->bi_status == BLK_STS_TARGET) {
6264 			unsigned int stripe_index =
6265 				btrfs_io_bio(bio)->stripe_index;
6266 			struct btrfs_device *dev;
6267 
6268 			BUG_ON(stripe_index >= bbio->num_stripes);
6269 			dev = bbio->stripes[stripe_index].dev;
6270 			if (dev->bdev) {
6271 				if (bio_op(bio) == REQ_OP_WRITE)
6272 					btrfs_dev_stat_inc_and_print(dev,
6273 						BTRFS_DEV_STAT_WRITE_ERRS);
6274 				else if (!(bio->bi_opf & REQ_RAHEAD))
6275 					btrfs_dev_stat_inc_and_print(dev,
6276 						BTRFS_DEV_STAT_READ_ERRS);
6277 				if (bio->bi_opf & REQ_PREFLUSH)
6278 					btrfs_dev_stat_inc_and_print(dev,
6279 						BTRFS_DEV_STAT_FLUSH_ERRS);
6280 			}
6281 		}
6282 	}
6283 
6284 	if (bio == bbio->orig_bio)
6285 		is_orig_bio = 1;
6286 
6287 	btrfs_bio_counter_dec(bbio->fs_info);
6288 
6289 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
6290 		if (!is_orig_bio) {
6291 			bio_put(bio);
6292 			bio = bbio->orig_bio;
6293 		}
6294 
6295 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6296 		/* only send an error to the higher layers if it is
6297 		 * beyond the tolerance of the btrfs bio
6298 		 */
6299 		if (atomic_read(&bbio->error) > bbio->max_errors) {
6300 			bio->bi_status = BLK_STS_IOERR;
6301 		} else {
6302 			/*
6303 			 * this bio is actually up to date, we didn't
6304 			 * go over the max number of errors
6305 			 */
6306 			bio->bi_status = BLK_STS_OK;
6307 		}
6308 
6309 		btrfs_end_bbio(bbio, bio);
6310 	} else if (!is_orig_bio) {
6311 		bio_put(bio);
6312 	}
6313 }
6314 
6315 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6316 			      u64 physical, int dev_nr)
6317 {
6318 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6319 	struct btrfs_fs_info *fs_info = bbio->fs_info;
6320 
6321 	bio->bi_private = bbio;
6322 	btrfs_io_bio(bio)->stripe_index = dev_nr;
6323 	bio->bi_end_io = btrfs_end_bio;
6324 	bio->bi_iter.bi_sector = physical >> 9;
6325 	btrfs_debug_in_rcu(fs_info,
6326 	"btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6327 		bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6328 		(unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6329 		dev->devid, bio->bi_iter.bi_size);
6330 	bio_set_dev(bio, dev->bdev);
6331 
6332 	btrfs_bio_counter_inc_noblocked(fs_info);
6333 
6334 	btrfsic_submit_bio(bio);
6335 }
6336 
6337 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6338 {
6339 	atomic_inc(&bbio->error);
6340 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
6341 		/* Should be the original bio. */
6342 		WARN_ON(bio != bbio->orig_bio);
6343 
6344 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6345 		bio->bi_iter.bi_sector = logical >> 9;
6346 		if (atomic_read(&bbio->error) > bbio->max_errors)
6347 			bio->bi_status = BLK_STS_IOERR;
6348 		else
6349 			bio->bi_status = BLK_STS_OK;
6350 		btrfs_end_bbio(bbio, bio);
6351 	}
6352 }
6353 
6354 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6355 			   int mirror_num)
6356 {
6357 	struct btrfs_device *dev;
6358 	struct bio *first_bio = bio;
6359 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6360 	u64 length = 0;
6361 	u64 map_length;
6362 	int ret;
6363 	int dev_nr;
6364 	int total_devs;
6365 	struct btrfs_bio *bbio = NULL;
6366 
6367 	length = bio->bi_iter.bi_size;
6368 	map_length = length;
6369 
6370 	btrfs_bio_counter_inc_blocked(fs_info);
6371 	ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6372 				&map_length, &bbio, mirror_num, 1);
6373 	if (ret) {
6374 		btrfs_bio_counter_dec(fs_info);
6375 		return errno_to_blk_status(ret);
6376 	}
6377 
6378 	total_devs = bbio->num_stripes;
6379 	bbio->orig_bio = first_bio;
6380 	bbio->private = first_bio->bi_private;
6381 	bbio->end_io = first_bio->bi_end_io;
6382 	bbio->fs_info = fs_info;
6383 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6384 
6385 	if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6386 	    ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6387 		/* In this case, map_length has been set to the length of
6388 		   a single stripe; not the whole write */
6389 		if (bio_op(bio) == REQ_OP_WRITE) {
6390 			ret = raid56_parity_write(fs_info, bio, bbio,
6391 						  map_length);
6392 		} else {
6393 			ret = raid56_parity_recover(fs_info, bio, bbio,
6394 						    map_length, mirror_num, 1);
6395 		}
6396 
6397 		btrfs_bio_counter_dec(fs_info);
6398 		return errno_to_blk_status(ret);
6399 	}
6400 
6401 	if (map_length < length) {
6402 		btrfs_crit(fs_info,
6403 			   "mapping failed logical %llu bio len %llu len %llu",
6404 			   logical, length, map_length);
6405 		BUG();
6406 	}
6407 
6408 	for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6409 		dev = bbio->stripes[dev_nr].dev;
6410 		if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6411 						   &dev->dev_state) ||
6412 		    (bio_op(first_bio) == REQ_OP_WRITE &&
6413 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6414 			bbio_error(bbio, first_bio, logical);
6415 			continue;
6416 		}
6417 
6418 		if (dev_nr < total_devs - 1)
6419 			bio = btrfs_bio_clone(first_bio);
6420 		else
6421 			bio = first_bio;
6422 
6423 		submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6424 				  dev_nr);
6425 	}
6426 	btrfs_bio_counter_dec(fs_info);
6427 	return BLK_STS_OK;
6428 }
6429 
6430 /*
6431  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6432  * return NULL.
6433  *
6434  * If devid and uuid are both specified, the match must be exact, otherwise
6435  * only devid is used.
6436  *
6437  * If @seed is true, traverse through the seed devices.
6438  */
6439 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6440 				       u64 devid, u8 *uuid, u8 *fsid,
6441 				       bool seed)
6442 {
6443 	struct btrfs_device *device;
6444 
6445 	while (fs_devices) {
6446 		if (!fsid ||
6447 		    !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6448 			list_for_each_entry(device, &fs_devices->devices,
6449 					    dev_list) {
6450 				if (device->devid == devid &&
6451 				    (!uuid || memcmp(device->uuid, uuid,
6452 						     BTRFS_UUID_SIZE) == 0))
6453 					return device;
6454 			}
6455 		}
6456 		if (seed)
6457 			fs_devices = fs_devices->seed;
6458 		else
6459 			return NULL;
6460 	}
6461 	return NULL;
6462 }
6463 
6464 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6465 					    u64 devid, u8 *dev_uuid)
6466 {
6467 	struct btrfs_device *device;
6468 
6469 	device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6470 	if (IS_ERR(device))
6471 		return device;
6472 
6473 	list_add(&device->dev_list, &fs_devices->devices);
6474 	device->fs_devices = fs_devices;
6475 	fs_devices->num_devices++;
6476 
6477 	set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6478 	fs_devices->missing_devices++;
6479 
6480 	return device;
6481 }
6482 
6483 /**
6484  * btrfs_alloc_device - allocate struct btrfs_device
6485  * @fs_info:	used only for generating a new devid, can be NULL if
6486  *		devid is provided (i.e. @devid != NULL).
6487  * @devid:	a pointer to devid for this device.  If NULL a new devid
6488  *		is generated.
6489  * @uuid:	a pointer to UUID for this device.  If NULL a new UUID
6490  *		is generated.
6491  *
6492  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6493  * on error.  Returned struct is not linked onto any lists and must be
6494  * destroyed with btrfs_free_device.
6495  */
6496 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6497 					const u64 *devid,
6498 					const u8 *uuid)
6499 {
6500 	struct btrfs_device *dev;
6501 	u64 tmp;
6502 
6503 	if (WARN_ON(!devid && !fs_info))
6504 		return ERR_PTR(-EINVAL);
6505 
6506 	dev = __alloc_device();
6507 	if (IS_ERR(dev))
6508 		return dev;
6509 
6510 	if (devid)
6511 		tmp = *devid;
6512 	else {
6513 		int ret;
6514 
6515 		ret = find_next_devid(fs_info, &tmp);
6516 		if (ret) {
6517 			btrfs_free_device(dev);
6518 			return ERR_PTR(ret);
6519 		}
6520 	}
6521 	dev->devid = tmp;
6522 
6523 	if (uuid)
6524 		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6525 	else
6526 		generate_random_uuid(dev->uuid);
6527 
6528 	return dev;
6529 }
6530 
6531 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6532 					u64 devid, u8 *uuid, bool error)
6533 {
6534 	if (error)
6535 		btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6536 			      devid, uuid);
6537 	else
6538 		btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6539 			      devid, uuid);
6540 }
6541 
6542 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6543 {
6544 	int index = btrfs_bg_flags_to_raid_index(type);
6545 	int ncopies = btrfs_raid_array[index].ncopies;
6546 	const int nparity = btrfs_raid_array[index].nparity;
6547 	int data_stripes;
6548 
6549 	if (nparity)
6550 		data_stripes = num_stripes - nparity;
6551 	else
6552 		data_stripes = num_stripes / ncopies;
6553 
6554 	return div_u64(chunk_len, data_stripes);
6555 }
6556 
6557 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6558 			  struct btrfs_chunk *chunk)
6559 {
6560 	struct btrfs_fs_info *fs_info = leaf->fs_info;
6561 	struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6562 	struct map_lookup *map;
6563 	struct extent_map *em;
6564 	u64 logical;
6565 	u64 length;
6566 	u64 devid;
6567 	u8 uuid[BTRFS_UUID_SIZE];
6568 	int num_stripes;
6569 	int ret;
6570 	int i;
6571 
6572 	logical = key->offset;
6573 	length = btrfs_chunk_length(leaf, chunk);
6574 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6575 
6576 	/*
6577 	 * Only need to verify chunk item if we're reading from sys chunk array,
6578 	 * as chunk item in tree block is already verified by tree-checker.
6579 	 */
6580 	if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6581 		ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6582 		if (ret)
6583 			return ret;
6584 	}
6585 
6586 	read_lock(&map_tree->lock);
6587 	em = lookup_extent_mapping(map_tree, logical, 1);
6588 	read_unlock(&map_tree->lock);
6589 
6590 	/* already mapped? */
6591 	if (em && em->start <= logical && em->start + em->len > logical) {
6592 		free_extent_map(em);
6593 		return 0;
6594 	} else if (em) {
6595 		free_extent_map(em);
6596 	}
6597 
6598 	em = alloc_extent_map();
6599 	if (!em)
6600 		return -ENOMEM;
6601 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6602 	if (!map) {
6603 		free_extent_map(em);
6604 		return -ENOMEM;
6605 	}
6606 
6607 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6608 	em->map_lookup = map;
6609 	em->start = logical;
6610 	em->len = length;
6611 	em->orig_start = 0;
6612 	em->block_start = 0;
6613 	em->block_len = em->len;
6614 
6615 	map->num_stripes = num_stripes;
6616 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
6617 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
6618 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6619 	map->type = btrfs_chunk_type(leaf, chunk);
6620 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6621 	map->verified_stripes = 0;
6622 	em->orig_block_len = calc_stripe_length(map->type, em->len,
6623 						map->num_stripes);
6624 	for (i = 0; i < num_stripes; i++) {
6625 		map->stripes[i].physical =
6626 			btrfs_stripe_offset_nr(leaf, chunk, i);
6627 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6628 		read_extent_buffer(leaf, uuid, (unsigned long)
6629 				   btrfs_stripe_dev_uuid_nr(chunk, i),
6630 				   BTRFS_UUID_SIZE);
6631 		map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6632 							devid, uuid, NULL, true);
6633 		if (!map->stripes[i].dev &&
6634 		    !btrfs_test_opt(fs_info, DEGRADED)) {
6635 			free_extent_map(em);
6636 			btrfs_report_missing_device(fs_info, devid, uuid, true);
6637 			return -ENOENT;
6638 		}
6639 		if (!map->stripes[i].dev) {
6640 			map->stripes[i].dev =
6641 				add_missing_dev(fs_info->fs_devices, devid,
6642 						uuid);
6643 			if (IS_ERR(map->stripes[i].dev)) {
6644 				free_extent_map(em);
6645 				btrfs_err(fs_info,
6646 					"failed to init missing dev %llu: %ld",
6647 					devid, PTR_ERR(map->stripes[i].dev));
6648 				return PTR_ERR(map->stripes[i].dev);
6649 			}
6650 			btrfs_report_missing_device(fs_info, devid, uuid, false);
6651 		}
6652 		set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6653 				&(map->stripes[i].dev->dev_state));
6654 
6655 	}
6656 
6657 	write_lock(&map_tree->lock);
6658 	ret = add_extent_mapping(map_tree, em, 0);
6659 	write_unlock(&map_tree->lock);
6660 	if (ret < 0) {
6661 		btrfs_err(fs_info,
6662 			  "failed to add chunk map, start=%llu len=%llu: %d",
6663 			  em->start, em->len, ret);
6664 	}
6665 	free_extent_map(em);
6666 
6667 	return ret;
6668 }
6669 
6670 static void fill_device_from_item(struct extent_buffer *leaf,
6671 				 struct btrfs_dev_item *dev_item,
6672 				 struct btrfs_device *device)
6673 {
6674 	unsigned long ptr;
6675 
6676 	device->devid = btrfs_device_id(leaf, dev_item);
6677 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6678 	device->total_bytes = device->disk_total_bytes;
6679 	device->commit_total_bytes = device->disk_total_bytes;
6680 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6681 	device->commit_bytes_used = device->bytes_used;
6682 	device->type = btrfs_device_type(leaf, dev_item);
6683 	device->io_align = btrfs_device_io_align(leaf, dev_item);
6684 	device->io_width = btrfs_device_io_width(leaf, dev_item);
6685 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6686 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6687 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6688 
6689 	ptr = btrfs_device_uuid(dev_item);
6690 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6691 }
6692 
6693 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6694 						  u8 *fsid)
6695 {
6696 	struct btrfs_fs_devices *fs_devices;
6697 	int ret;
6698 
6699 	lockdep_assert_held(&uuid_mutex);
6700 	ASSERT(fsid);
6701 
6702 	fs_devices = fs_info->fs_devices->seed;
6703 	while (fs_devices) {
6704 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6705 			return fs_devices;
6706 
6707 		fs_devices = fs_devices->seed;
6708 	}
6709 
6710 	fs_devices = find_fsid(fsid, NULL);
6711 	if (!fs_devices) {
6712 		if (!btrfs_test_opt(fs_info, DEGRADED))
6713 			return ERR_PTR(-ENOENT);
6714 
6715 		fs_devices = alloc_fs_devices(fsid, NULL);
6716 		if (IS_ERR(fs_devices))
6717 			return fs_devices;
6718 
6719 		fs_devices->seeding = true;
6720 		fs_devices->opened = 1;
6721 		return fs_devices;
6722 	}
6723 
6724 	fs_devices = clone_fs_devices(fs_devices);
6725 	if (IS_ERR(fs_devices))
6726 		return fs_devices;
6727 
6728 	ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6729 	if (ret) {
6730 		free_fs_devices(fs_devices);
6731 		fs_devices = ERR_PTR(ret);
6732 		goto out;
6733 	}
6734 
6735 	if (!fs_devices->seeding) {
6736 		close_fs_devices(fs_devices);
6737 		free_fs_devices(fs_devices);
6738 		fs_devices = ERR_PTR(-EINVAL);
6739 		goto out;
6740 	}
6741 
6742 	fs_devices->seed = fs_info->fs_devices->seed;
6743 	fs_info->fs_devices->seed = fs_devices;
6744 out:
6745 	return fs_devices;
6746 }
6747 
6748 static int read_one_dev(struct extent_buffer *leaf,
6749 			struct btrfs_dev_item *dev_item)
6750 {
6751 	struct btrfs_fs_info *fs_info = leaf->fs_info;
6752 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6753 	struct btrfs_device *device;
6754 	u64 devid;
6755 	int ret;
6756 	u8 fs_uuid[BTRFS_FSID_SIZE];
6757 	u8 dev_uuid[BTRFS_UUID_SIZE];
6758 
6759 	devid = btrfs_device_id(leaf, dev_item);
6760 	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6761 			   BTRFS_UUID_SIZE);
6762 	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6763 			   BTRFS_FSID_SIZE);
6764 
6765 	if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6766 		fs_devices = open_seed_devices(fs_info, fs_uuid);
6767 		if (IS_ERR(fs_devices))
6768 			return PTR_ERR(fs_devices);
6769 	}
6770 
6771 	device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6772 				   fs_uuid, true);
6773 	if (!device) {
6774 		if (!btrfs_test_opt(fs_info, DEGRADED)) {
6775 			btrfs_report_missing_device(fs_info, devid,
6776 							dev_uuid, true);
6777 			return -ENOENT;
6778 		}
6779 
6780 		device = add_missing_dev(fs_devices, devid, dev_uuid);
6781 		if (IS_ERR(device)) {
6782 			btrfs_err(fs_info,
6783 				"failed to add missing dev %llu: %ld",
6784 				devid, PTR_ERR(device));
6785 			return PTR_ERR(device);
6786 		}
6787 		btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6788 	} else {
6789 		if (!device->bdev) {
6790 			if (!btrfs_test_opt(fs_info, DEGRADED)) {
6791 				btrfs_report_missing_device(fs_info,
6792 						devid, dev_uuid, true);
6793 				return -ENOENT;
6794 			}
6795 			btrfs_report_missing_device(fs_info, devid,
6796 							dev_uuid, false);
6797 		}
6798 
6799 		if (!device->bdev &&
6800 		    !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6801 			/*
6802 			 * this happens when a device that was properly setup
6803 			 * in the device info lists suddenly goes bad.
6804 			 * device->bdev is NULL, and so we have to set
6805 			 * device->missing to one here
6806 			 */
6807 			device->fs_devices->missing_devices++;
6808 			set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6809 		}
6810 
6811 		/* Move the device to its own fs_devices */
6812 		if (device->fs_devices != fs_devices) {
6813 			ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6814 							&device->dev_state));
6815 
6816 			list_move(&device->dev_list, &fs_devices->devices);
6817 			device->fs_devices->num_devices--;
6818 			fs_devices->num_devices++;
6819 
6820 			device->fs_devices->missing_devices--;
6821 			fs_devices->missing_devices++;
6822 
6823 			device->fs_devices = fs_devices;
6824 		}
6825 	}
6826 
6827 	if (device->fs_devices != fs_info->fs_devices) {
6828 		BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6829 		if (device->generation !=
6830 		    btrfs_device_generation(leaf, dev_item))
6831 			return -EINVAL;
6832 	}
6833 
6834 	fill_device_from_item(leaf, dev_item, device);
6835 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6836 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6837 	   !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6838 		device->fs_devices->total_rw_bytes += device->total_bytes;
6839 		atomic64_add(device->total_bytes - device->bytes_used,
6840 				&fs_info->free_chunk_space);
6841 	}
6842 	ret = 0;
6843 	return ret;
6844 }
6845 
6846 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6847 {
6848 	struct btrfs_root *root = fs_info->tree_root;
6849 	struct btrfs_super_block *super_copy = fs_info->super_copy;
6850 	struct extent_buffer *sb;
6851 	struct btrfs_disk_key *disk_key;
6852 	struct btrfs_chunk *chunk;
6853 	u8 *array_ptr;
6854 	unsigned long sb_array_offset;
6855 	int ret = 0;
6856 	u32 num_stripes;
6857 	u32 array_size;
6858 	u32 len = 0;
6859 	u32 cur_offset;
6860 	u64 type;
6861 	struct btrfs_key key;
6862 
6863 	ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6864 	/*
6865 	 * This will create extent buffer of nodesize, superblock size is
6866 	 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6867 	 * overallocate but we can keep it as-is, only the first page is used.
6868 	 */
6869 	sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6870 	if (IS_ERR(sb))
6871 		return PTR_ERR(sb);
6872 	set_extent_buffer_uptodate(sb);
6873 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6874 	/*
6875 	 * The sb extent buffer is artificial and just used to read the system array.
6876 	 * set_extent_buffer_uptodate() call does not properly mark all it's
6877 	 * pages up-to-date when the page is larger: extent does not cover the
6878 	 * whole page and consequently check_page_uptodate does not find all
6879 	 * the page's extents up-to-date (the hole beyond sb),
6880 	 * write_extent_buffer then triggers a WARN_ON.
6881 	 *
6882 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6883 	 * but sb spans only this function. Add an explicit SetPageUptodate call
6884 	 * to silence the warning eg. on PowerPC 64.
6885 	 */
6886 	if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6887 		SetPageUptodate(sb->pages[0]);
6888 
6889 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6890 	array_size = btrfs_super_sys_array_size(super_copy);
6891 
6892 	array_ptr = super_copy->sys_chunk_array;
6893 	sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6894 	cur_offset = 0;
6895 
6896 	while (cur_offset < array_size) {
6897 		disk_key = (struct btrfs_disk_key *)array_ptr;
6898 		len = sizeof(*disk_key);
6899 		if (cur_offset + len > array_size)
6900 			goto out_short_read;
6901 
6902 		btrfs_disk_key_to_cpu(&key, disk_key);
6903 
6904 		array_ptr += len;
6905 		sb_array_offset += len;
6906 		cur_offset += len;
6907 
6908 		if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6909 			btrfs_err(fs_info,
6910 			    "unexpected item type %u in sys_array at offset %u",
6911 				  (u32)key.type, cur_offset);
6912 			ret = -EIO;
6913 			break;
6914 		}
6915 
6916 		chunk = (struct btrfs_chunk *)sb_array_offset;
6917 		/*
6918 		 * At least one btrfs_chunk with one stripe must be present,
6919 		 * exact stripe count check comes afterwards
6920 		 */
6921 		len = btrfs_chunk_item_size(1);
6922 		if (cur_offset + len > array_size)
6923 			goto out_short_read;
6924 
6925 		num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6926 		if (!num_stripes) {
6927 			btrfs_err(fs_info,
6928 			"invalid number of stripes %u in sys_array at offset %u",
6929 				  num_stripes, cur_offset);
6930 			ret = -EIO;
6931 			break;
6932 		}
6933 
6934 		type = btrfs_chunk_type(sb, chunk);
6935 		if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6936 			btrfs_err(fs_info,
6937 			"invalid chunk type %llu in sys_array at offset %u",
6938 				  type, cur_offset);
6939 			ret = -EIO;
6940 			break;
6941 		}
6942 
6943 		len = btrfs_chunk_item_size(num_stripes);
6944 		if (cur_offset + len > array_size)
6945 			goto out_short_read;
6946 
6947 		ret = read_one_chunk(&key, sb, chunk);
6948 		if (ret)
6949 			break;
6950 
6951 		array_ptr += len;
6952 		sb_array_offset += len;
6953 		cur_offset += len;
6954 	}
6955 	clear_extent_buffer_uptodate(sb);
6956 	free_extent_buffer_stale(sb);
6957 	return ret;
6958 
6959 out_short_read:
6960 	btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6961 			len, cur_offset);
6962 	clear_extent_buffer_uptodate(sb);
6963 	free_extent_buffer_stale(sb);
6964 	return -EIO;
6965 }
6966 
6967 /*
6968  * Check if all chunks in the fs are OK for read-write degraded mount
6969  *
6970  * If the @failing_dev is specified, it's accounted as missing.
6971  *
6972  * Return true if all chunks meet the minimal RW mount requirements.
6973  * Return false if any chunk doesn't meet the minimal RW mount requirements.
6974  */
6975 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6976 					struct btrfs_device *failing_dev)
6977 {
6978 	struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6979 	struct extent_map *em;
6980 	u64 next_start = 0;
6981 	bool ret = true;
6982 
6983 	read_lock(&map_tree->lock);
6984 	em = lookup_extent_mapping(map_tree, 0, (u64)-1);
6985 	read_unlock(&map_tree->lock);
6986 	/* No chunk at all? Return false anyway */
6987 	if (!em) {
6988 		ret = false;
6989 		goto out;
6990 	}
6991 	while (em) {
6992 		struct map_lookup *map;
6993 		int missing = 0;
6994 		int max_tolerated;
6995 		int i;
6996 
6997 		map = em->map_lookup;
6998 		max_tolerated =
6999 			btrfs_get_num_tolerated_disk_barrier_failures(
7000 					map->type);
7001 		for (i = 0; i < map->num_stripes; i++) {
7002 			struct btrfs_device *dev = map->stripes[i].dev;
7003 
7004 			if (!dev || !dev->bdev ||
7005 			    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7006 			    dev->last_flush_error)
7007 				missing++;
7008 			else if (failing_dev && failing_dev == dev)
7009 				missing++;
7010 		}
7011 		if (missing > max_tolerated) {
7012 			if (!failing_dev)
7013 				btrfs_warn(fs_info,
7014 	"chunk %llu missing %d devices, max tolerance is %d for writable mount",
7015 				   em->start, missing, max_tolerated);
7016 			free_extent_map(em);
7017 			ret = false;
7018 			goto out;
7019 		}
7020 		next_start = extent_map_end(em);
7021 		free_extent_map(em);
7022 
7023 		read_lock(&map_tree->lock);
7024 		em = lookup_extent_mapping(map_tree, next_start,
7025 					   (u64)(-1) - next_start);
7026 		read_unlock(&map_tree->lock);
7027 	}
7028 out:
7029 	return ret;
7030 }
7031 
7032 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7033 {
7034 	struct btrfs_root *root = fs_info->chunk_root;
7035 	struct btrfs_path *path;
7036 	struct extent_buffer *leaf;
7037 	struct btrfs_key key;
7038 	struct btrfs_key found_key;
7039 	int ret;
7040 	int slot;
7041 	u64 total_dev = 0;
7042 
7043 	path = btrfs_alloc_path();
7044 	if (!path)
7045 		return -ENOMEM;
7046 
7047 	/*
7048 	 * uuid_mutex is needed only if we are mounting a sprout FS
7049 	 * otherwise we don't need it.
7050 	 */
7051 	mutex_lock(&uuid_mutex);
7052 	mutex_lock(&fs_info->chunk_mutex);
7053 
7054 	/*
7055 	 * Read all device items, and then all the chunk items. All
7056 	 * device items are found before any chunk item (their object id
7057 	 * is smaller than the lowest possible object id for a chunk
7058 	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7059 	 */
7060 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7061 	key.offset = 0;
7062 	key.type = 0;
7063 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7064 	if (ret < 0)
7065 		goto error;
7066 	while (1) {
7067 		leaf = path->nodes[0];
7068 		slot = path->slots[0];
7069 		if (slot >= btrfs_header_nritems(leaf)) {
7070 			ret = btrfs_next_leaf(root, path);
7071 			if (ret == 0)
7072 				continue;
7073 			if (ret < 0)
7074 				goto error;
7075 			break;
7076 		}
7077 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
7078 		if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7079 			struct btrfs_dev_item *dev_item;
7080 			dev_item = btrfs_item_ptr(leaf, slot,
7081 						  struct btrfs_dev_item);
7082 			ret = read_one_dev(leaf, dev_item);
7083 			if (ret)
7084 				goto error;
7085 			total_dev++;
7086 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7087 			struct btrfs_chunk *chunk;
7088 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7089 			ret = read_one_chunk(&found_key, leaf, chunk);
7090 			if (ret)
7091 				goto error;
7092 		}
7093 		path->slots[0]++;
7094 	}
7095 
7096 	/*
7097 	 * After loading chunk tree, we've got all device information,
7098 	 * do another round of validation checks.
7099 	 */
7100 	if (total_dev != fs_info->fs_devices->total_devices) {
7101 		btrfs_err(fs_info,
7102 	   "super_num_devices %llu mismatch with num_devices %llu found here",
7103 			  btrfs_super_num_devices(fs_info->super_copy),
7104 			  total_dev);
7105 		ret = -EINVAL;
7106 		goto error;
7107 	}
7108 	if (btrfs_super_total_bytes(fs_info->super_copy) <
7109 	    fs_info->fs_devices->total_rw_bytes) {
7110 		btrfs_err(fs_info,
7111 	"super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7112 			  btrfs_super_total_bytes(fs_info->super_copy),
7113 			  fs_info->fs_devices->total_rw_bytes);
7114 		ret = -EINVAL;
7115 		goto error;
7116 	}
7117 	ret = 0;
7118 error:
7119 	mutex_unlock(&fs_info->chunk_mutex);
7120 	mutex_unlock(&uuid_mutex);
7121 
7122 	btrfs_free_path(path);
7123 	return ret;
7124 }
7125 
7126 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7127 {
7128 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7129 	struct btrfs_device *device;
7130 
7131 	while (fs_devices) {
7132 		mutex_lock(&fs_devices->device_list_mutex);
7133 		list_for_each_entry(device, &fs_devices->devices, dev_list)
7134 			device->fs_info = fs_info;
7135 		mutex_unlock(&fs_devices->device_list_mutex);
7136 
7137 		fs_devices = fs_devices->seed;
7138 	}
7139 }
7140 
7141 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7142 				 const struct btrfs_dev_stats_item *ptr,
7143 				 int index)
7144 {
7145 	u64 val;
7146 
7147 	read_extent_buffer(eb, &val,
7148 			   offsetof(struct btrfs_dev_stats_item, values) +
7149 			    ((unsigned long)ptr) + (index * sizeof(u64)),
7150 			   sizeof(val));
7151 	return val;
7152 }
7153 
7154 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7155 				      struct btrfs_dev_stats_item *ptr,
7156 				      int index, u64 val)
7157 {
7158 	write_extent_buffer(eb, &val,
7159 			    offsetof(struct btrfs_dev_stats_item, values) +
7160 			     ((unsigned long)ptr) + (index * sizeof(u64)),
7161 			    sizeof(val));
7162 }
7163 
7164 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7165 {
7166 	struct btrfs_key key;
7167 	struct btrfs_root *dev_root = fs_info->dev_root;
7168 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7169 	struct extent_buffer *eb;
7170 	int slot;
7171 	int ret = 0;
7172 	struct btrfs_device *device;
7173 	struct btrfs_path *path = NULL;
7174 	int i;
7175 
7176 	path = btrfs_alloc_path();
7177 	if (!path)
7178 		return -ENOMEM;
7179 
7180 	mutex_lock(&fs_devices->device_list_mutex);
7181 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
7182 		int item_size;
7183 		struct btrfs_dev_stats_item *ptr;
7184 
7185 		key.objectid = BTRFS_DEV_STATS_OBJECTID;
7186 		key.type = BTRFS_PERSISTENT_ITEM_KEY;
7187 		key.offset = device->devid;
7188 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7189 		if (ret) {
7190 			for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7191 				btrfs_dev_stat_set(device, i, 0);
7192 			device->dev_stats_valid = 1;
7193 			btrfs_release_path(path);
7194 			continue;
7195 		}
7196 		slot = path->slots[0];
7197 		eb = path->nodes[0];
7198 		item_size = btrfs_item_size_nr(eb, slot);
7199 
7200 		ptr = btrfs_item_ptr(eb, slot,
7201 				     struct btrfs_dev_stats_item);
7202 
7203 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7204 			if (item_size >= (1 + i) * sizeof(__le64))
7205 				btrfs_dev_stat_set(device, i,
7206 					btrfs_dev_stats_value(eb, ptr, i));
7207 			else
7208 				btrfs_dev_stat_set(device, i, 0);
7209 		}
7210 
7211 		device->dev_stats_valid = 1;
7212 		btrfs_dev_stat_print_on_load(device);
7213 		btrfs_release_path(path);
7214 	}
7215 	mutex_unlock(&fs_devices->device_list_mutex);
7216 
7217 	btrfs_free_path(path);
7218 	return ret < 0 ? ret : 0;
7219 }
7220 
7221 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7222 				struct btrfs_device *device)
7223 {
7224 	struct btrfs_fs_info *fs_info = trans->fs_info;
7225 	struct btrfs_root *dev_root = fs_info->dev_root;
7226 	struct btrfs_path *path;
7227 	struct btrfs_key key;
7228 	struct extent_buffer *eb;
7229 	struct btrfs_dev_stats_item *ptr;
7230 	int ret;
7231 	int i;
7232 
7233 	key.objectid = BTRFS_DEV_STATS_OBJECTID;
7234 	key.type = BTRFS_PERSISTENT_ITEM_KEY;
7235 	key.offset = device->devid;
7236 
7237 	path = btrfs_alloc_path();
7238 	if (!path)
7239 		return -ENOMEM;
7240 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7241 	if (ret < 0) {
7242 		btrfs_warn_in_rcu(fs_info,
7243 			"error %d while searching for dev_stats item for device %s",
7244 			      ret, rcu_str_deref(device->name));
7245 		goto out;
7246 	}
7247 
7248 	if (ret == 0 &&
7249 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7250 		/* need to delete old one and insert a new one */
7251 		ret = btrfs_del_item(trans, dev_root, path);
7252 		if (ret != 0) {
7253 			btrfs_warn_in_rcu(fs_info,
7254 				"delete too small dev_stats item for device %s failed %d",
7255 				      rcu_str_deref(device->name), ret);
7256 			goto out;
7257 		}
7258 		ret = 1;
7259 	}
7260 
7261 	if (ret == 1) {
7262 		/* need to insert a new item */
7263 		btrfs_release_path(path);
7264 		ret = btrfs_insert_empty_item(trans, dev_root, path,
7265 					      &key, sizeof(*ptr));
7266 		if (ret < 0) {
7267 			btrfs_warn_in_rcu(fs_info,
7268 				"insert dev_stats item for device %s failed %d",
7269 				rcu_str_deref(device->name), ret);
7270 			goto out;
7271 		}
7272 	}
7273 
7274 	eb = path->nodes[0];
7275 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7276 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7277 		btrfs_set_dev_stats_value(eb, ptr, i,
7278 					  btrfs_dev_stat_read(device, i));
7279 	btrfs_mark_buffer_dirty(eb);
7280 
7281 out:
7282 	btrfs_free_path(path);
7283 	return ret;
7284 }
7285 
7286 /*
7287  * called from commit_transaction. Writes all changed device stats to disk.
7288  */
7289 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7290 {
7291 	struct btrfs_fs_info *fs_info = trans->fs_info;
7292 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7293 	struct btrfs_device *device;
7294 	int stats_cnt;
7295 	int ret = 0;
7296 
7297 	mutex_lock(&fs_devices->device_list_mutex);
7298 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
7299 		stats_cnt = atomic_read(&device->dev_stats_ccnt);
7300 		if (!device->dev_stats_valid || stats_cnt == 0)
7301 			continue;
7302 
7303 
7304 		/*
7305 		 * There is a LOAD-LOAD control dependency between the value of
7306 		 * dev_stats_ccnt and updating the on-disk values which requires
7307 		 * reading the in-memory counters. Such control dependencies
7308 		 * require explicit read memory barriers.
7309 		 *
7310 		 * This memory barriers pairs with smp_mb__before_atomic in
7311 		 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7312 		 * barrier implied by atomic_xchg in
7313 		 * btrfs_dev_stats_read_and_reset
7314 		 */
7315 		smp_rmb();
7316 
7317 		ret = update_dev_stat_item(trans, device);
7318 		if (!ret)
7319 			atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7320 	}
7321 	mutex_unlock(&fs_devices->device_list_mutex);
7322 
7323 	return ret;
7324 }
7325 
7326 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7327 {
7328 	btrfs_dev_stat_inc(dev, index);
7329 	btrfs_dev_stat_print_on_error(dev);
7330 }
7331 
7332 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7333 {
7334 	if (!dev->dev_stats_valid)
7335 		return;
7336 	btrfs_err_rl_in_rcu(dev->fs_info,
7337 		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7338 			   rcu_str_deref(dev->name),
7339 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7340 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7341 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7342 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7343 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7344 }
7345 
7346 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7347 {
7348 	int i;
7349 
7350 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7351 		if (btrfs_dev_stat_read(dev, i) != 0)
7352 			break;
7353 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
7354 		return; /* all values == 0, suppress message */
7355 
7356 	btrfs_info_in_rcu(dev->fs_info,
7357 		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7358 	       rcu_str_deref(dev->name),
7359 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7360 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7361 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7362 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7363 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7364 }
7365 
7366 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7367 			struct btrfs_ioctl_get_dev_stats *stats)
7368 {
7369 	struct btrfs_device *dev;
7370 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7371 	int i;
7372 
7373 	mutex_lock(&fs_devices->device_list_mutex);
7374 	dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7375 				true);
7376 	mutex_unlock(&fs_devices->device_list_mutex);
7377 
7378 	if (!dev) {
7379 		btrfs_warn(fs_info, "get dev_stats failed, device not found");
7380 		return -ENODEV;
7381 	} else if (!dev->dev_stats_valid) {
7382 		btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7383 		return -ENODEV;
7384 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7385 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7386 			if (stats->nr_items > i)
7387 				stats->values[i] =
7388 					btrfs_dev_stat_read_and_reset(dev, i);
7389 			else
7390 				btrfs_dev_stat_set(dev, i, 0);
7391 		}
7392 		btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7393 			   current->comm, task_pid_nr(current));
7394 	} else {
7395 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7396 			if (stats->nr_items > i)
7397 				stats->values[i] = btrfs_dev_stat_read(dev, i);
7398 	}
7399 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7400 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7401 	return 0;
7402 }
7403 
7404 /*
7405  * Update the size and bytes used for each device where it changed.  This is
7406  * delayed since we would otherwise get errors while writing out the
7407  * superblocks.
7408  *
7409  * Must be invoked during transaction commit.
7410  */
7411 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7412 {
7413 	struct btrfs_device *curr, *next;
7414 
7415 	ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7416 
7417 	if (list_empty(&trans->dev_update_list))
7418 		return;
7419 
7420 	/*
7421 	 * We don't need the device_list_mutex here.  This list is owned by the
7422 	 * transaction and the transaction must complete before the device is
7423 	 * released.
7424 	 */
7425 	mutex_lock(&trans->fs_info->chunk_mutex);
7426 	list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7427 				 post_commit_list) {
7428 		list_del_init(&curr->post_commit_list);
7429 		curr->commit_total_bytes = curr->disk_total_bytes;
7430 		curr->commit_bytes_used = curr->bytes_used;
7431 	}
7432 	mutex_unlock(&trans->fs_info->chunk_mutex);
7433 }
7434 
7435 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7436 {
7437 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7438 	while (fs_devices) {
7439 		fs_devices->fs_info = fs_info;
7440 		fs_devices = fs_devices->seed;
7441 	}
7442 }
7443 
7444 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7445 {
7446 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7447 	while (fs_devices) {
7448 		fs_devices->fs_info = NULL;
7449 		fs_devices = fs_devices->seed;
7450 	}
7451 }
7452 
7453 /*
7454  * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7455  */
7456 int btrfs_bg_type_to_factor(u64 flags)
7457 {
7458 	const int index = btrfs_bg_flags_to_raid_index(flags);
7459 
7460 	return btrfs_raid_array[index].ncopies;
7461 }
7462 
7463 
7464 
7465 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7466 				 u64 chunk_offset, u64 devid,
7467 				 u64 physical_offset, u64 physical_len)
7468 {
7469 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7470 	struct extent_map *em;
7471 	struct map_lookup *map;
7472 	struct btrfs_device *dev;
7473 	u64 stripe_len;
7474 	bool found = false;
7475 	int ret = 0;
7476 	int i;
7477 
7478 	read_lock(&em_tree->lock);
7479 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7480 	read_unlock(&em_tree->lock);
7481 
7482 	if (!em) {
7483 		btrfs_err(fs_info,
7484 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7485 			  physical_offset, devid);
7486 		ret = -EUCLEAN;
7487 		goto out;
7488 	}
7489 
7490 	map = em->map_lookup;
7491 	stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7492 	if (physical_len != stripe_len) {
7493 		btrfs_err(fs_info,
7494 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7495 			  physical_offset, devid, em->start, physical_len,
7496 			  stripe_len);
7497 		ret = -EUCLEAN;
7498 		goto out;
7499 	}
7500 
7501 	for (i = 0; i < map->num_stripes; i++) {
7502 		if (map->stripes[i].dev->devid == devid &&
7503 		    map->stripes[i].physical == physical_offset) {
7504 			found = true;
7505 			if (map->verified_stripes >= map->num_stripes) {
7506 				btrfs_err(fs_info,
7507 				"too many dev extents for chunk %llu found",
7508 					  em->start);
7509 				ret = -EUCLEAN;
7510 				goto out;
7511 			}
7512 			map->verified_stripes++;
7513 			break;
7514 		}
7515 	}
7516 	if (!found) {
7517 		btrfs_err(fs_info,
7518 	"dev extent physical offset %llu devid %llu has no corresponding chunk",
7519 			physical_offset, devid);
7520 		ret = -EUCLEAN;
7521 	}
7522 
7523 	/* Make sure no dev extent is beyond device bondary */
7524 	dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7525 	if (!dev) {
7526 		btrfs_err(fs_info, "failed to find devid %llu", devid);
7527 		ret = -EUCLEAN;
7528 		goto out;
7529 	}
7530 
7531 	/* It's possible this device is a dummy for seed device */
7532 	if (dev->disk_total_bytes == 0) {
7533 		dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL,
7534 					NULL, false);
7535 		if (!dev) {
7536 			btrfs_err(fs_info, "failed to find seed devid %llu",
7537 				  devid);
7538 			ret = -EUCLEAN;
7539 			goto out;
7540 		}
7541 	}
7542 
7543 	if (physical_offset + physical_len > dev->disk_total_bytes) {
7544 		btrfs_err(fs_info,
7545 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7546 			  devid, physical_offset, physical_len,
7547 			  dev->disk_total_bytes);
7548 		ret = -EUCLEAN;
7549 		goto out;
7550 	}
7551 out:
7552 	free_extent_map(em);
7553 	return ret;
7554 }
7555 
7556 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7557 {
7558 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7559 	struct extent_map *em;
7560 	struct rb_node *node;
7561 	int ret = 0;
7562 
7563 	read_lock(&em_tree->lock);
7564 	for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7565 		em = rb_entry(node, struct extent_map, rb_node);
7566 		if (em->map_lookup->num_stripes !=
7567 		    em->map_lookup->verified_stripes) {
7568 			btrfs_err(fs_info,
7569 			"chunk %llu has missing dev extent, have %d expect %d",
7570 				  em->start, em->map_lookup->verified_stripes,
7571 				  em->map_lookup->num_stripes);
7572 			ret = -EUCLEAN;
7573 			goto out;
7574 		}
7575 	}
7576 out:
7577 	read_unlock(&em_tree->lock);
7578 	return ret;
7579 }
7580 
7581 /*
7582  * Ensure that all dev extents are mapped to correct chunk, otherwise
7583  * later chunk allocation/free would cause unexpected behavior.
7584  *
7585  * NOTE: This will iterate through the whole device tree, which should be of
7586  * the same size level as the chunk tree.  This slightly increases mount time.
7587  */
7588 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7589 {
7590 	struct btrfs_path *path;
7591 	struct btrfs_root *root = fs_info->dev_root;
7592 	struct btrfs_key key;
7593 	u64 prev_devid = 0;
7594 	u64 prev_dev_ext_end = 0;
7595 	int ret = 0;
7596 
7597 	key.objectid = 1;
7598 	key.type = BTRFS_DEV_EXTENT_KEY;
7599 	key.offset = 0;
7600 
7601 	path = btrfs_alloc_path();
7602 	if (!path)
7603 		return -ENOMEM;
7604 
7605 	path->reada = READA_FORWARD;
7606 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7607 	if (ret < 0)
7608 		goto out;
7609 
7610 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7611 		ret = btrfs_next_item(root, path);
7612 		if (ret < 0)
7613 			goto out;
7614 		/* No dev extents at all? Not good */
7615 		if (ret > 0) {
7616 			ret = -EUCLEAN;
7617 			goto out;
7618 		}
7619 	}
7620 	while (1) {
7621 		struct extent_buffer *leaf = path->nodes[0];
7622 		struct btrfs_dev_extent *dext;
7623 		int slot = path->slots[0];
7624 		u64 chunk_offset;
7625 		u64 physical_offset;
7626 		u64 physical_len;
7627 		u64 devid;
7628 
7629 		btrfs_item_key_to_cpu(leaf, &key, slot);
7630 		if (key.type != BTRFS_DEV_EXTENT_KEY)
7631 			break;
7632 		devid = key.objectid;
7633 		physical_offset = key.offset;
7634 
7635 		dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7636 		chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7637 		physical_len = btrfs_dev_extent_length(leaf, dext);
7638 
7639 		/* Check if this dev extent overlaps with the previous one */
7640 		if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7641 			btrfs_err(fs_info,
7642 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7643 				  devid, physical_offset, prev_dev_ext_end);
7644 			ret = -EUCLEAN;
7645 			goto out;
7646 		}
7647 
7648 		ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7649 					    physical_offset, physical_len);
7650 		if (ret < 0)
7651 			goto out;
7652 		prev_devid = devid;
7653 		prev_dev_ext_end = physical_offset + physical_len;
7654 
7655 		ret = btrfs_next_item(root, path);
7656 		if (ret < 0)
7657 			goto out;
7658 		if (ret > 0) {
7659 			ret = 0;
7660 			break;
7661 		}
7662 	}
7663 
7664 	/* Ensure all chunks have corresponding dev extents */
7665 	ret = verify_chunk_dev_extent_mapping(fs_info);
7666 out:
7667 	btrfs_free_path(path);
7668 	return ret;
7669 }
7670 
7671 /*
7672  * Check whether the given block group or device is pinned by any inode being
7673  * used as a swapfile.
7674  */
7675 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7676 {
7677 	struct btrfs_swapfile_pin *sp;
7678 	struct rb_node *node;
7679 
7680 	spin_lock(&fs_info->swapfile_pins_lock);
7681 	node = fs_info->swapfile_pins.rb_node;
7682 	while (node) {
7683 		sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7684 		if (ptr < sp->ptr)
7685 			node = node->rb_left;
7686 		else if (ptr > sp->ptr)
7687 			node = node->rb_right;
7688 		else
7689 			break;
7690 	}
7691 	spin_unlock(&fs_info->swapfile_pins_lock);
7692 	return node != NULL;
7693 }
7694