xref: /linux/fs/btrfs/volumes.c (revision 4f3c8320c78cdd11c8fdd23c33787407f719322e)
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/sched/mm.h>
8 #include <linux/bio.h>
9 #include <linux/slab.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
17 #include "misc.h"
18 #include "ctree.h"
19 #include "extent_map.h"
20 #include "disk-io.h"
21 #include "transaction.h"
22 #include "print-tree.h"
23 #include "volumes.h"
24 #include "raid56.h"
25 #include "async-thread.h"
26 #include "check-integrity.h"
27 #include "rcu-string.h"
28 #include "dev-replace.h"
29 #include "sysfs.h"
30 #include "tree-checker.h"
31 #include "space-info.h"
32 #include "block-group.h"
33 #include "discard.h"
34 #include "zoned.h"
35 
36 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
37 	[BTRFS_RAID_RAID10] = {
38 		.sub_stripes	= 2,
39 		.dev_stripes	= 1,
40 		.devs_max	= 0,	/* 0 == as many as possible */
41 		.devs_min	= 4,
42 		.tolerated_failures = 1,
43 		.devs_increment	= 2,
44 		.ncopies	= 2,
45 		.nparity        = 0,
46 		.raid_name	= "raid10",
47 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID10,
48 		.mindev_error	= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
49 	},
50 	[BTRFS_RAID_RAID1] = {
51 		.sub_stripes	= 1,
52 		.dev_stripes	= 1,
53 		.devs_max	= 2,
54 		.devs_min	= 2,
55 		.tolerated_failures = 1,
56 		.devs_increment	= 2,
57 		.ncopies	= 2,
58 		.nparity        = 0,
59 		.raid_name	= "raid1",
60 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1,
61 		.mindev_error	= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
62 	},
63 	[BTRFS_RAID_RAID1C3] = {
64 		.sub_stripes	= 1,
65 		.dev_stripes	= 1,
66 		.devs_max	= 3,
67 		.devs_min	= 3,
68 		.tolerated_failures = 2,
69 		.devs_increment	= 3,
70 		.ncopies	= 3,
71 		.nparity        = 0,
72 		.raid_name	= "raid1c3",
73 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C3,
74 		.mindev_error	= BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
75 	},
76 	[BTRFS_RAID_RAID1C4] = {
77 		.sub_stripes	= 1,
78 		.dev_stripes	= 1,
79 		.devs_max	= 4,
80 		.devs_min	= 4,
81 		.tolerated_failures = 3,
82 		.devs_increment	= 4,
83 		.ncopies	= 4,
84 		.nparity        = 0,
85 		.raid_name	= "raid1c4",
86 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C4,
87 		.mindev_error	= BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
88 	},
89 	[BTRFS_RAID_DUP] = {
90 		.sub_stripes	= 1,
91 		.dev_stripes	= 2,
92 		.devs_max	= 1,
93 		.devs_min	= 1,
94 		.tolerated_failures = 0,
95 		.devs_increment	= 1,
96 		.ncopies	= 2,
97 		.nparity        = 0,
98 		.raid_name	= "dup",
99 		.bg_flag	= BTRFS_BLOCK_GROUP_DUP,
100 		.mindev_error	= 0,
101 	},
102 	[BTRFS_RAID_RAID0] = {
103 		.sub_stripes	= 1,
104 		.dev_stripes	= 1,
105 		.devs_max	= 0,
106 		.devs_min	= 2,
107 		.tolerated_failures = 0,
108 		.devs_increment	= 1,
109 		.ncopies	= 1,
110 		.nparity        = 0,
111 		.raid_name	= "raid0",
112 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID0,
113 		.mindev_error	= 0,
114 	},
115 	[BTRFS_RAID_SINGLE] = {
116 		.sub_stripes	= 1,
117 		.dev_stripes	= 1,
118 		.devs_max	= 1,
119 		.devs_min	= 1,
120 		.tolerated_failures = 0,
121 		.devs_increment	= 1,
122 		.ncopies	= 1,
123 		.nparity        = 0,
124 		.raid_name	= "single",
125 		.bg_flag	= 0,
126 		.mindev_error	= 0,
127 	},
128 	[BTRFS_RAID_RAID5] = {
129 		.sub_stripes	= 1,
130 		.dev_stripes	= 1,
131 		.devs_max	= 0,
132 		.devs_min	= 2,
133 		.tolerated_failures = 1,
134 		.devs_increment	= 1,
135 		.ncopies	= 1,
136 		.nparity        = 1,
137 		.raid_name	= "raid5",
138 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID5,
139 		.mindev_error	= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
140 	},
141 	[BTRFS_RAID_RAID6] = {
142 		.sub_stripes	= 1,
143 		.dev_stripes	= 1,
144 		.devs_max	= 0,
145 		.devs_min	= 3,
146 		.tolerated_failures = 2,
147 		.devs_increment	= 1,
148 		.ncopies	= 1,
149 		.nparity        = 2,
150 		.raid_name	= "raid6",
151 		.bg_flag	= BTRFS_BLOCK_GROUP_RAID6,
152 		.mindev_error	= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
153 	},
154 };
155 
156 const char *btrfs_bg_type_to_raid_name(u64 flags)
157 {
158 	const int index = btrfs_bg_flags_to_raid_index(flags);
159 
160 	if (index >= BTRFS_NR_RAID_TYPES)
161 		return NULL;
162 
163 	return btrfs_raid_array[index].raid_name;
164 }
165 
166 /*
167  * Fill @buf with textual description of @bg_flags, no more than @size_buf
168  * bytes including terminating null byte.
169  */
170 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
171 {
172 	int i;
173 	int ret;
174 	char *bp = buf;
175 	u64 flags = bg_flags;
176 	u32 size_bp = size_buf;
177 
178 	if (!flags) {
179 		strcpy(bp, "NONE");
180 		return;
181 	}
182 
183 #define DESCRIBE_FLAG(flag, desc)						\
184 	do {								\
185 		if (flags & (flag)) {					\
186 			ret = snprintf(bp, size_bp, "%s|", (desc));	\
187 			if (ret < 0 || ret >= size_bp)			\
188 				goto out_overflow;			\
189 			size_bp -= ret;					\
190 			bp += ret;					\
191 			flags &= ~(flag);				\
192 		}							\
193 	} while (0)
194 
195 	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
196 	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
197 	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
198 
199 	DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
200 	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
201 		DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
202 			      btrfs_raid_array[i].raid_name);
203 #undef DESCRIBE_FLAG
204 
205 	if (flags) {
206 		ret = snprintf(bp, size_bp, "0x%llx|", flags);
207 		size_bp -= ret;
208 	}
209 
210 	if (size_bp < size_buf)
211 		buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
212 
213 	/*
214 	 * The text is trimmed, it's up to the caller to provide sufficiently
215 	 * large buffer
216 	 */
217 out_overflow:;
218 }
219 
220 static int init_first_rw_device(struct btrfs_trans_handle *trans);
221 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
222 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
223 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
224 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
225 			     enum btrfs_map_op op,
226 			     u64 logical, u64 *length,
227 			     struct btrfs_bio **bbio_ret,
228 			     int mirror_num, int need_raid_map);
229 
230 /*
231  * Device locking
232  * ==============
233  *
234  * There are several mutexes that protect manipulation of devices and low-level
235  * structures like chunks but not block groups, extents or files
236  *
237  * uuid_mutex (global lock)
238  * ------------------------
239  * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
240  * the SCAN_DEV ioctl registration or from mount either implicitly (the first
241  * device) or requested by the device= mount option
242  *
243  * the mutex can be very coarse and can cover long-running operations
244  *
245  * protects: updates to fs_devices counters like missing devices, rw devices,
246  * seeding, structure cloning, opening/closing devices at mount/umount time
247  *
248  * global::fs_devs - add, remove, updates to the global list
249  *
250  * does not protect: manipulation of the fs_devices::devices list in general
251  * but in mount context it could be used to exclude list modifications by eg.
252  * scan ioctl
253  *
254  * btrfs_device::name - renames (write side), read is RCU
255  *
256  * fs_devices::device_list_mutex (per-fs, with RCU)
257  * ------------------------------------------------
258  * protects updates to fs_devices::devices, ie. adding and deleting
259  *
260  * simple list traversal with read-only actions can be done with RCU protection
261  *
262  * may be used to exclude some operations from running concurrently without any
263  * modifications to the list (see write_all_supers)
264  *
265  * Is not required at mount and close times, because our device list is
266  * protected by the uuid_mutex at that point.
267  *
268  * balance_mutex
269  * -------------
270  * protects balance structures (status, state) and context accessed from
271  * several places (internally, ioctl)
272  *
273  * chunk_mutex
274  * -----------
275  * protects chunks, adding or removing during allocation, trim or when a new
276  * device is added/removed. Additionally it also protects post_commit_list of
277  * individual devices, since they can be added to the transaction's
278  * post_commit_list only with chunk_mutex held.
279  *
280  * cleaner_mutex
281  * -------------
282  * a big lock that is held by the cleaner thread and prevents running subvolume
283  * cleaning together with relocation or delayed iputs
284  *
285  *
286  * Lock nesting
287  * ============
288  *
289  * uuid_mutex
290  *   device_list_mutex
291  *     chunk_mutex
292  *   balance_mutex
293  *
294  *
295  * Exclusive operations
296  * ====================
297  *
298  * Maintains the exclusivity of the following operations that apply to the
299  * whole filesystem and cannot run in parallel.
300  *
301  * - Balance (*)
302  * - Device add
303  * - Device remove
304  * - Device replace (*)
305  * - Resize
306  *
307  * The device operations (as above) can be in one of the following states:
308  *
309  * - Running state
310  * - Paused state
311  * - Completed state
312  *
313  * Only device operations marked with (*) can go into the Paused state for the
314  * following reasons:
315  *
316  * - ioctl (only Balance can be Paused through ioctl)
317  * - filesystem remounted as read-only
318  * - filesystem unmounted and mounted as read-only
319  * - system power-cycle and filesystem mounted as read-only
320  * - filesystem or device errors leading to forced read-only
321  *
322  * The status of exclusive operation is set and cleared atomically.
323  * During the course of Paused state, fs_info::exclusive_operation remains set.
324  * A device operation in Paused or Running state can be canceled or resumed
325  * either by ioctl (Balance only) or when remounted as read-write.
326  * The exclusive status is cleared when the device operation is canceled or
327  * completed.
328  */
329 
330 DEFINE_MUTEX(uuid_mutex);
331 static LIST_HEAD(fs_uuids);
332 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
333 {
334 	return &fs_uuids;
335 }
336 
337 /*
338  * alloc_fs_devices - allocate struct btrfs_fs_devices
339  * @fsid:		if not NULL, copy the UUID to fs_devices::fsid
340  * @metadata_fsid:	if not NULL, copy the UUID to fs_devices::metadata_fsid
341  *
342  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
343  * The returned struct is not linked onto any lists and can be destroyed with
344  * kfree() right away.
345  */
346 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
347 						 const u8 *metadata_fsid)
348 {
349 	struct btrfs_fs_devices *fs_devs;
350 
351 	fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
352 	if (!fs_devs)
353 		return ERR_PTR(-ENOMEM);
354 
355 	mutex_init(&fs_devs->device_list_mutex);
356 
357 	INIT_LIST_HEAD(&fs_devs->devices);
358 	INIT_LIST_HEAD(&fs_devs->alloc_list);
359 	INIT_LIST_HEAD(&fs_devs->fs_list);
360 	INIT_LIST_HEAD(&fs_devs->seed_list);
361 	if (fsid)
362 		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
363 
364 	if (metadata_fsid)
365 		memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
366 	else if (fsid)
367 		memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
368 
369 	return fs_devs;
370 }
371 
372 void btrfs_free_device(struct btrfs_device *device)
373 {
374 	WARN_ON(!list_empty(&device->post_commit_list));
375 	rcu_string_free(device->name);
376 	extent_io_tree_release(&device->alloc_state);
377 	bio_put(device->flush_bio);
378 	btrfs_destroy_dev_zone_info(device);
379 	kfree(device);
380 }
381 
382 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
383 {
384 	struct btrfs_device *device;
385 	WARN_ON(fs_devices->opened);
386 	while (!list_empty(&fs_devices->devices)) {
387 		device = list_entry(fs_devices->devices.next,
388 				    struct btrfs_device, dev_list);
389 		list_del(&device->dev_list);
390 		btrfs_free_device(device);
391 	}
392 	kfree(fs_devices);
393 }
394 
395 void __exit btrfs_cleanup_fs_uuids(void)
396 {
397 	struct btrfs_fs_devices *fs_devices;
398 
399 	while (!list_empty(&fs_uuids)) {
400 		fs_devices = list_entry(fs_uuids.next,
401 					struct btrfs_fs_devices, fs_list);
402 		list_del(&fs_devices->fs_list);
403 		free_fs_devices(fs_devices);
404 	}
405 }
406 
407 /*
408  * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
409  * Returned struct is not linked onto any lists and must be destroyed using
410  * btrfs_free_device.
411  */
412 static struct btrfs_device *__alloc_device(struct btrfs_fs_info *fs_info)
413 {
414 	struct btrfs_device *dev;
415 
416 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
417 	if (!dev)
418 		return ERR_PTR(-ENOMEM);
419 
420 	/*
421 	 * Preallocate a bio that's always going to be used for flushing device
422 	 * barriers and matches the device lifespan
423 	 */
424 	dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
425 	if (!dev->flush_bio) {
426 		kfree(dev);
427 		return ERR_PTR(-ENOMEM);
428 	}
429 
430 	INIT_LIST_HEAD(&dev->dev_list);
431 	INIT_LIST_HEAD(&dev->dev_alloc_list);
432 	INIT_LIST_HEAD(&dev->post_commit_list);
433 
434 	atomic_set(&dev->reada_in_flight, 0);
435 	atomic_set(&dev->dev_stats_ccnt, 0);
436 	btrfs_device_data_ordered_init(dev, fs_info);
437 	INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
438 	INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
439 	extent_io_tree_init(fs_info, &dev->alloc_state,
440 			    IO_TREE_DEVICE_ALLOC_STATE, NULL);
441 
442 	return dev;
443 }
444 
445 static noinline struct btrfs_fs_devices *find_fsid(
446 		const u8 *fsid, const u8 *metadata_fsid)
447 {
448 	struct btrfs_fs_devices *fs_devices;
449 
450 	ASSERT(fsid);
451 
452 	/* Handle non-split brain cases */
453 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
454 		if (metadata_fsid) {
455 			if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
456 			    && memcmp(metadata_fsid, fs_devices->metadata_uuid,
457 				      BTRFS_FSID_SIZE) == 0)
458 				return fs_devices;
459 		} else {
460 			if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
461 				return fs_devices;
462 		}
463 	}
464 	return NULL;
465 }
466 
467 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
468 				struct btrfs_super_block *disk_super)
469 {
470 
471 	struct btrfs_fs_devices *fs_devices;
472 
473 	/*
474 	 * Handle scanned device having completed its fsid change but
475 	 * belonging to a fs_devices that was created by first scanning
476 	 * a device which didn't have its fsid/metadata_uuid changed
477 	 * at all and the CHANGING_FSID_V2 flag set.
478 	 */
479 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
480 		if (fs_devices->fsid_change &&
481 		    memcmp(disk_super->metadata_uuid, fs_devices->fsid,
482 			   BTRFS_FSID_SIZE) == 0 &&
483 		    memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
484 			   BTRFS_FSID_SIZE) == 0) {
485 			return fs_devices;
486 		}
487 	}
488 	/*
489 	 * Handle scanned device having completed its fsid change but
490 	 * belonging to a fs_devices that was created by a device that
491 	 * has an outdated pair of fsid/metadata_uuid and
492 	 * CHANGING_FSID_V2 flag set.
493 	 */
494 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
495 		if (fs_devices->fsid_change &&
496 		    memcmp(fs_devices->metadata_uuid,
497 			   fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
498 		    memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
499 			   BTRFS_FSID_SIZE) == 0) {
500 			return fs_devices;
501 		}
502 	}
503 
504 	return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
505 }
506 
507 
508 static int
509 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
510 		      int flush, struct block_device **bdev,
511 		      struct btrfs_super_block **disk_super)
512 {
513 	int ret;
514 
515 	*bdev = blkdev_get_by_path(device_path, flags, holder);
516 
517 	if (IS_ERR(*bdev)) {
518 		ret = PTR_ERR(*bdev);
519 		goto error;
520 	}
521 
522 	if (flush)
523 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
524 	ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
525 	if (ret) {
526 		blkdev_put(*bdev, flags);
527 		goto error;
528 	}
529 	invalidate_bdev(*bdev);
530 	*disk_super = btrfs_read_dev_super(*bdev);
531 	if (IS_ERR(*disk_super)) {
532 		ret = PTR_ERR(*disk_super);
533 		blkdev_put(*bdev, flags);
534 		goto error;
535 	}
536 
537 	return 0;
538 
539 error:
540 	*bdev = NULL;
541 	return ret;
542 }
543 
544 static bool device_path_matched(const char *path, struct btrfs_device *device)
545 {
546 	int found;
547 
548 	rcu_read_lock();
549 	found = strcmp(rcu_str_deref(device->name), path);
550 	rcu_read_unlock();
551 
552 	return found == 0;
553 }
554 
555 /*
556  *  Search and remove all stale (devices which are not mounted) devices.
557  *  When both inputs are NULL, it will search and release all stale devices.
558  *  path:	Optional. When provided will it release all unmounted devices
559  *		matching this path only.
560  *  skip_dev:	Optional. Will skip this device when searching for the stale
561  *		devices.
562  *  Return:	0 for success or if @path is NULL.
563  * 		-EBUSY if @path is a mounted device.
564  * 		-ENOENT if @path does not match any device in the list.
565  */
566 static int btrfs_free_stale_devices(const char *path,
567 				     struct btrfs_device *skip_device)
568 {
569 	struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
570 	struct btrfs_device *device, *tmp_device;
571 	int ret = 0;
572 
573 	if (path)
574 		ret = -ENOENT;
575 
576 	list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
577 
578 		mutex_lock(&fs_devices->device_list_mutex);
579 		list_for_each_entry_safe(device, tmp_device,
580 					 &fs_devices->devices, dev_list) {
581 			if (skip_device && skip_device == device)
582 				continue;
583 			if (path && !device->name)
584 				continue;
585 			if (path && !device_path_matched(path, device))
586 				continue;
587 			if (fs_devices->opened) {
588 				/* for an already deleted device return 0 */
589 				if (path && ret != 0)
590 					ret = -EBUSY;
591 				break;
592 			}
593 
594 			/* delete the stale device */
595 			fs_devices->num_devices--;
596 			list_del(&device->dev_list);
597 			btrfs_free_device(device);
598 
599 			ret = 0;
600 		}
601 		mutex_unlock(&fs_devices->device_list_mutex);
602 
603 		if (fs_devices->num_devices == 0) {
604 			btrfs_sysfs_remove_fsid(fs_devices);
605 			list_del(&fs_devices->fs_list);
606 			free_fs_devices(fs_devices);
607 		}
608 	}
609 
610 	return ret;
611 }
612 
613 /*
614  * This is only used on mount, and we are protected from competing things
615  * messing with our fs_devices by the uuid_mutex, thus we do not need the
616  * fs_devices->device_list_mutex here.
617  */
618 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
619 			struct btrfs_device *device, fmode_t flags,
620 			void *holder)
621 {
622 	struct request_queue *q;
623 	struct block_device *bdev;
624 	struct btrfs_super_block *disk_super;
625 	u64 devid;
626 	int ret;
627 
628 	if (device->bdev)
629 		return -EINVAL;
630 	if (!device->name)
631 		return -EINVAL;
632 
633 	ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
634 				    &bdev, &disk_super);
635 	if (ret)
636 		return ret;
637 
638 	devid = btrfs_stack_device_id(&disk_super->dev_item);
639 	if (devid != device->devid)
640 		goto error_free_page;
641 
642 	if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
643 		goto error_free_page;
644 
645 	device->generation = btrfs_super_generation(disk_super);
646 
647 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
648 		if (btrfs_super_incompat_flags(disk_super) &
649 		    BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
650 			pr_err(
651 		"BTRFS: Invalid seeding and uuid-changed device detected\n");
652 			goto error_free_page;
653 		}
654 
655 		clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
656 		fs_devices->seeding = true;
657 	} else {
658 		if (bdev_read_only(bdev))
659 			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
660 		else
661 			set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
662 	}
663 
664 	q = bdev_get_queue(bdev);
665 	if (!blk_queue_nonrot(q))
666 		fs_devices->rotating = true;
667 
668 	device->bdev = bdev;
669 	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
670 	device->mode = flags;
671 
672 	ret = btrfs_get_dev_zone_info(device);
673 	if (ret != 0)
674 		goto error_free_page;
675 
676 	fs_devices->open_devices++;
677 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
678 	    device->devid != BTRFS_DEV_REPLACE_DEVID) {
679 		fs_devices->rw_devices++;
680 		list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
681 	}
682 	btrfs_release_disk_super(disk_super);
683 
684 	return 0;
685 
686 error_free_page:
687 	btrfs_release_disk_super(disk_super);
688 	blkdev_put(bdev, flags);
689 
690 	return -EINVAL;
691 }
692 
693 /*
694  * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
695  * being created with a disk that has already completed its fsid change. Such
696  * disk can belong to an fs which has its FSID changed or to one which doesn't.
697  * Handle both cases here.
698  */
699 static struct btrfs_fs_devices *find_fsid_inprogress(
700 					struct btrfs_super_block *disk_super)
701 {
702 	struct btrfs_fs_devices *fs_devices;
703 
704 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
705 		if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
706 			   BTRFS_FSID_SIZE) != 0 &&
707 		    memcmp(fs_devices->metadata_uuid, disk_super->fsid,
708 			   BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
709 			return fs_devices;
710 		}
711 	}
712 
713 	return find_fsid(disk_super->fsid, NULL);
714 }
715 
716 
717 static struct btrfs_fs_devices *find_fsid_changed(
718 					struct btrfs_super_block *disk_super)
719 {
720 	struct btrfs_fs_devices *fs_devices;
721 
722 	/*
723 	 * Handles the case where scanned device is part of an fs that had
724 	 * multiple successful changes of FSID but curently device didn't
725 	 * observe it. Meaning our fsid will be different than theirs. We need
726 	 * to handle two subcases :
727 	 *  1 - The fs still continues to have different METADATA/FSID uuids.
728 	 *  2 - The fs is switched back to its original FSID (METADATA/FSID
729 	 *  are equal).
730 	 */
731 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
732 		/* Changed UUIDs */
733 		if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
734 			   BTRFS_FSID_SIZE) != 0 &&
735 		    memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
736 			   BTRFS_FSID_SIZE) == 0 &&
737 		    memcmp(fs_devices->fsid, disk_super->fsid,
738 			   BTRFS_FSID_SIZE) != 0)
739 			return fs_devices;
740 
741 		/* Unchanged UUIDs */
742 		if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
743 			   BTRFS_FSID_SIZE) == 0 &&
744 		    memcmp(fs_devices->fsid, disk_super->metadata_uuid,
745 			   BTRFS_FSID_SIZE) == 0)
746 			return fs_devices;
747 	}
748 
749 	return NULL;
750 }
751 
752 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
753 				struct btrfs_super_block *disk_super)
754 {
755 	struct btrfs_fs_devices *fs_devices;
756 
757 	/*
758 	 * Handle the case where the scanned device is part of an fs whose last
759 	 * metadata UUID change reverted it to the original FSID. At the same
760 	 * time * fs_devices was first created by another constitutent device
761 	 * which didn't fully observe the operation. This results in an
762 	 * btrfs_fs_devices created with metadata/fsid different AND
763 	 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
764 	 * fs_devices equal to the FSID of the disk.
765 	 */
766 	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
767 		if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
768 			   BTRFS_FSID_SIZE) != 0 &&
769 		    memcmp(fs_devices->metadata_uuid, disk_super->fsid,
770 			   BTRFS_FSID_SIZE) == 0 &&
771 		    fs_devices->fsid_change)
772 			return fs_devices;
773 	}
774 
775 	return NULL;
776 }
777 /*
778  * Add new device to list of registered devices
779  *
780  * Returns:
781  * device pointer which was just added or updated when successful
782  * error pointer when failed
783  */
784 static noinline struct btrfs_device *device_list_add(const char *path,
785 			   struct btrfs_super_block *disk_super,
786 			   bool *new_device_added)
787 {
788 	struct btrfs_device *device;
789 	struct btrfs_fs_devices *fs_devices = NULL;
790 	struct rcu_string *name;
791 	u64 found_transid = btrfs_super_generation(disk_super);
792 	u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
793 	bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
794 		BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
795 	bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
796 					BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
797 
798 	if (fsid_change_in_progress) {
799 		if (!has_metadata_uuid)
800 			fs_devices = find_fsid_inprogress(disk_super);
801 		else
802 			fs_devices = find_fsid_changed(disk_super);
803 	} else if (has_metadata_uuid) {
804 		fs_devices = find_fsid_with_metadata_uuid(disk_super);
805 	} else {
806 		fs_devices = find_fsid_reverted_metadata(disk_super);
807 		if (!fs_devices)
808 			fs_devices = find_fsid(disk_super->fsid, NULL);
809 	}
810 
811 
812 	if (!fs_devices) {
813 		if (has_metadata_uuid)
814 			fs_devices = alloc_fs_devices(disk_super->fsid,
815 						      disk_super->metadata_uuid);
816 		else
817 			fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
818 
819 		if (IS_ERR(fs_devices))
820 			return ERR_CAST(fs_devices);
821 
822 		fs_devices->fsid_change = fsid_change_in_progress;
823 
824 		mutex_lock(&fs_devices->device_list_mutex);
825 		list_add(&fs_devices->fs_list, &fs_uuids);
826 
827 		device = NULL;
828 	} else {
829 		mutex_lock(&fs_devices->device_list_mutex);
830 		device = btrfs_find_device(fs_devices, devid,
831 				disk_super->dev_item.uuid, NULL);
832 
833 		/*
834 		 * If this disk has been pulled into an fs devices created by
835 		 * a device which had the CHANGING_FSID_V2 flag then replace the
836 		 * metadata_uuid/fsid values of the fs_devices.
837 		 */
838 		if (fs_devices->fsid_change &&
839 		    found_transid > fs_devices->latest_generation) {
840 			memcpy(fs_devices->fsid, disk_super->fsid,
841 					BTRFS_FSID_SIZE);
842 
843 			if (has_metadata_uuid)
844 				memcpy(fs_devices->metadata_uuid,
845 				       disk_super->metadata_uuid,
846 				       BTRFS_FSID_SIZE);
847 			else
848 				memcpy(fs_devices->metadata_uuid,
849 				       disk_super->fsid, BTRFS_FSID_SIZE);
850 
851 			fs_devices->fsid_change = false;
852 		}
853 	}
854 
855 	if (!device) {
856 		if (fs_devices->opened) {
857 			mutex_unlock(&fs_devices->device_list_mutex);
858 			return ERR_PTR(-EBUSY);
859 		}
860 
861 		device = btrfs_alloc_device(NULL, &devid,
862 					    disk_super->dev_item.uuid);
863 		if (IS_ERR(device)) {
864 			mutex_unlock(&fs_devices->device_list_mutex);
865 			/* we can safely leave the fs_devices entry around */
866 			return device;
867 		}
868 
869 		name = rcu_string_strdup(path, GFP_NOFS);
870 		if (!name) {
871 			btrfs_free_device(device);
872 			mutex_unlock(&fs_devices->device_list_mutex);
873 			return ERR_PTR(-ENOMEM);
874 		}
875 		rcu_assign_pointer(device->name, name);
876 
877 		list_add_rcu(&device->dev_list, &fs_devices->devices);
878 		fs_devices->num_devices++;
879 
880 		device->fs_devices = fs_devices;
881 		*new_device_added = true;
882 
883 		if (disk_super->label[0])
884 			pr_info(
885 	"BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
886 				disk_super->label, devid, found_transid, path,
887 				current->comm, task_pid_nr(current));
888 		else
889 			pr_info(
890 	"BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
891 				disk_super->fsid, devid, found_transid, path,
892 				current->comm, task_pid_nr(current));
893 
894 	} else if (!device->name || strcmp(device->name->str, path)) {
895 		/*
896 		 * When FS is already mounted.
897 		 * 1. If you are here and if the device->name is NULL that
898 		 *    means this device was missing at time of FS mount.
899 		 * 2. If you are here and if the device->name is different
900 		 *    from 'path' that means either
901 		 *      a. The same device disappeared and reappeared with
902 		 *         different name. or
903 		 *      b. The missing-disk-which-was-replaced, has
904 		 *         reappeared now.
905 		 *
906 		 * We must allow 1 and 2a above. But 2b would be a spurious
907 		 * and unintentional.
908 		 *
909 		 * Further in case of 1 and 2a above, the disk at 'path'
910 		 * would have missed some transaction when it was away and
911 		 * in case of 2a the stale bdev has to be updated as well.
912 		 * 2b must not be allowed at all time.
913 		 */
914 
915 		/*
916 		 * For now, we do allow update to btrfs_fs_device through the
917 		 * btrfs dev scan cli after FS has been mounted.  We're still
918 		 * tracking a problem where systems fail mount by subvolume id
919 		 * when we reject replacement on a mounted FS.
920 		 */
921 		if (!fs_devices->opened && found_transid < device->generation) {
922 			/*
923 			 * That is if the FS is _not_ mounted and if you
924 			 * are here, that means there is more than one
925 			 * disk with same uuid and devid.We keep the one
926 			 * with larger generation number or the last-in if
927 			 * generation are equal.
928 			 */
929 			mutex_unlock(&fs_devices->device_list_mutex);
930 			return ERR_PTR(-EEXIST);
931 		}
932 
933 		/*
934 		 * We are going to replace the device path for a given devid,
935 		 * make sure it's the same device if the device is mounted
936 		 */
937 		if (device->bdev) {
938 			int error;
939 			dev_t path_dev;
940 
941 			error = lookup_bdev(path, &path_dev);
942 			if (error) {
943 				mutex_unlock(&fs_devices->device_list_mutex);
944 				return ERR_PTR(error);
945 			}
946 
947 			if (device->bdev->bd_dev != path_dev) {
948 				mutex_unlock(&fs_devices->device_list_mutex);
949 				/*
950 				 * device->fs_info may not be reliable here, so
951 				 * pass in a NULL instead. This avoids a
952 				 * possible use-after-free when the fs_info and
953 				 * fs_info->sb are already torn down.
954 				 */
955 				btrfs_warn_in_rcu(NULL,
956 	"duplicate device %s devid %llu generation %llu scanned by %s (%d)",
957 						  path, devid, found_transid,
958 						  current->comm,
959 						  task_pid_nr(current));
960 				return ERR_PTR(-EEXIST);
961 			}
962 			btrfs_info_in_rcu(device->fs_info,
963 	"devid %llu device path %s changed to %s scanned by %s (%d)",
964 					  devid, rcu_str_deref(device->name),
965 					  path, current->comm,
966 					  task_pid_nr(current));
967 		}
968 
969 		name = rcu_string_strdup(path, GFP_NOFS);
970 		if (!name) {
971 			mutex_unlock(&fs_devices->device_list_mutex);
972 			return ERR_PTR(-ENOMEM);
973 		}
974 		rcu_string_free(device->name);
975 		rcu_assign_pointer(device->name, name);
976 		if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
977 			fs_devices->missing_devices--;
978 			clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
979 		}
980 	}
981 
982 	/*
983 	 * Unmount does not free the btrfs_device struct but would zero
984 	 * generation along with most of the other members. So just update
985 	 * it back. We need it to pick the disk with largest generation
986 	 * (as above).
987 	 */
988 	if (!fs_devices->opened) {
989 		device->generation = found_transid;
990 		fs_devices->latest_generation = max_t(u64, found_transid,
991 						fs_devices->latest_generation);
992 	}
993 
994 	fs_devices->total_devices = btrfs_super_num_devices(disk_super);
995 
996 	mutex_unlock(&fs_devices->device_list_mutex);
997 	return device;
998 }
999 
1000 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1001 {
1002 	struct btrfs_fs_devices *fs_devices;
1003 	struct btrfs_device *device;
1004 	struct btrfs_device *orig_dev;
1005 	int ret = 0;
1006 
1007 	fs_devices = alloc_fs_devices(orig->fsid, NULL);
1008 	if (IS_ERR(fs_devices))
1009 		return fs_devices;
1010 
1011 	mutex_lock(&orig->device_list_mutex);
1012 	fs_devices->total_devices = orig->total_devices;
1013 
1014 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1015 		struct rcu_string *name;
1016 
1017 		device = btrfs_alloc_device(NULL, &orig_dev->devid,
1018 					    orig_dev->uuid);
1019 		if (IS_ERR(device)) {
1020 			ret = PTR_ERR(device);
1021 			goto error;
1022 		}
1023 
1024 		/*
1025 		 * This is ok to do without rcu read locked because we hold the
1026 		 * uuid mutex so nothing we touch in here is going to disappear.
1027 		 */
1028 		if (orig_dev->name) {
1029 			name = rcu_string_strdup(orig_dev->name->str,
1030 					GFP_KERNEL);
1031 			if (!name) {
1032 				btrfs_free_device(device);
1033 				ret = -ENOMEM;
1034 				goto error;
1035 			}
1036 			rcu_assign_pointer(device->name, name);
1037 		}
1038 
1039 		list_add(&device->dev_list, &fs_devices->devices);
1040 		device->fs_devices = fs_devices;
1041 		fs_devices->num_devices++;
1042 	}
1043 	mutex_unlock(&orig->device_list_mutex);
1044 	return fs_devices;
1045 error:
1046 	mutex_unlock(&orig->device_list_mutex);
1047 	free_fs_devices(fs_devices);
1048 	return ERR_PTR(ret);
1049 }
1050 
1051 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1052 				      struct btrfs_device **latest_dev)
1053 {
1054 	struct btrfs_device *device, *next;
1055 
1056 	/* This is the initialized path, it is safe to release the devices. */
1057 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1058 		if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1059 			if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1060 				      &device->dev_state) &&
1061 			    !test_bit(BTRFS_DEV_STATE_MISSING,
1062 				      &device->dev_state) &&
1063 			    (!*latest_dev ||
1064 			     device->generation > (*latest_dev)->generation)) {
1065 				*latest_dev = device;
1066 			}
1067 			continue;
1068 		}
1069 
1070 		/*
1071 		 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1072 		 * in btrfs_init_dev_replace() so just continue.
1073 		 */
1074 		if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1075 			continue;
1076 
1077 		if (device->bdev) {
1078 			blkdev_put(device->bdev, device->mode);
1079 			device->bdev = NULL;
1080 			fs_devices->open_devices--;
1081 		}
1082 		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1083 			list_del_init(&device->dev_alloc_list);
1084 			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1085 		}
1086 		list_del_init(&device->dev_list);
1087 		fs_devices->num_devices--;
1088 		btrfs_free_device(device);
1089 	}
1090 
1091 }
1092 
1093 /*
1094  * After we have read the system tree and know devids belonging to this
1095  * filesystem, remove the device which does not belong there.
1096  */
1097 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1098 {
1099 	struct btrfs_device *latest_dev = NULL;
1100 	struct btrfs_fs_devices *seed_dev;
1101 
1102 	mutex_lock(&uuid_mutex);
1103 	__btrfs_free_extra_devids(fs_devices, &latest_dev);
1104 
1105 	list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1106 		__btrfs_free_extra_devids(seed_dev, &latest_dev);
1107 
1108 	fs_devices->latest_bdev = latest_dev->bdev;
1109 
1110 	mutex_unlock(&uuid_mutex);
1111 }
1112 
1113 static void btrfs_close_bdev(struct btrfs_device *device)
1114 {
1115 	if (!device->bdev)
1116 		return;
1117 
1118 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1119 		sync_blockdev(device->bdev);
1120 		invalidate_bdev(device->bdev);
1121 	}
1122 
1123 	blkdev_put(device->bdev, device->mode);
1124 }
1125 
1126 static void btrfs_close_one_device(struct btrfs_device *device)
1127 {
1128 	struct btrfs_fs_devices *fs_devices = device->fs_devices;
1129 
1130 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1131 	    device->devid != BTRFS_DEV_REPLACE_DEVID) {
1132 		list_del_init(&device->dev_alloc_list);
1133 		fs_devices->rw_devices--;
1134 	}
1135 
1136 	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1137 		fs_devices->missing_devices--;
1138 
1139 	btrfs_close_bdev(device);
1140 	if (device->bdev) {
1141 		fs_devices->open_devices--;
1142 		device->bdev = NULL;
1143 	}
1144 	clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1145 	btrfs_destroy_dev_zone_info(device);
1146 
1147 	device->fs_info = NULL;
1148 	atomic_set(&device->dev_stats_ccnt, 0);
1149 	extent_io_tree_release(&device->alloc_state);
1150 
1151 	/* Verify the device is back in a pristine state  */
1152 	ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1153 	ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1154 	ASSERT(list_empty(&device->dev_alloc_list));
1155 	ASSERT(list_empty(&device->post_commit_list));
1156 	ASSERT(atomic_read(&device->reada_in_flight) == 0);
1157 }
1158 
1159 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1160 {
1161 	struct btrfs_device *device, *tmp;
1162 
1163 	lockdep_assert_held(&uuid_mutex);
1164 
1165 	if (--fs_devices->opened > 0)
1166 		return;
1167 
1168 	list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1169 		btrfs_close_one_device(device);
1170 
1171 	WARN_ON(fs_devices->open_devices);
1172 	WARN_ON(fs_devices->rw_devices);
1173 	fs_devices->opened = 0;
1174 	fs_devices->seeding = false;
1175 	fs_devices->fs_info = NULL;
1176 }
1177 
1178 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1179 {
1180 	LIST_HEAD(list);
1181 	struct btrfs_fs_devices *tmp;
1182 
1183 	mutex_lock(&uuid_mutex);
1184 	close_fs_devices(fs_devices);
1185 	if (!fs_devices->opened)
1186 		list_splice_init(&fs_devices->seed_list, &list);
1187 
1188 	list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1189 		close_fs_devices(fs_devices);
1190 		list_del(&fs_devices->seed_list);
1191 		free_fs_devices(fs_devices);
1192 	}
1193 	mutex_unlock(&uuid_mutex);
1194 }
1195 
1196 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1197 				fmode_t flags, void *holder)
1198 {
1199 	struct btrfs_device *device;
1200 	struct btrfs_device *latest_dev = NULL;
1201 	struct btrfs_device *tmp_device;
1202 
1203 	flags |= FMODE_EXCL;
1204 
1205 	list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1206 				 dev_list) {
1207 		int ret;
1208 
1209 		ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1210 		if (ret == 0 &&
1211 		    (!latest_dev || device->generation > latest_dev->generation)) {
1212 			latest_dev = device;
1213 		} else if (ret == -ENODATA) {
1214 			fs_devices->num_devices--;
1215 			list_del(&device->dev_list);
1216 			btrfs_free_device(device);
1217 		}
1218 	}
1219 	if (fs_devices->open_devices == 0)
1220 		return -EINVAL;
1221 
1222 	fs_devices->opened = 1;
1223 	fs_devices->latest_bdev = latest_dev->bdev;
1224 	fs_devices->total_rw_bytes = 0;
1225 	fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1226 	fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1227 
1228 	return 0;
1229 }
1230 
1231 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1232 {
1233 	struct btrfs_device *dev1, *dev2;
1234 
1235 	dev1 = list_entry(a, struct btrfs_device, dev_list);
1236 	dev2 = list_entry(b, struct btrfs_device, dev_list);
1237 
1238 	if (dev1->devid < dev2->devid)
1239 		return -1;
1240 	else if (dev1->devid > dev2->devid)
1241 		return 1;
1242 	return 0;
1243 }
1244 
1245 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1246 		       fmode_t flags, void *holder)
1247 {
1248 	int ret;
1249 
1250 	lockdep_assert_held(&uuid_mutex);
1251 	/*
1252 	 * The device_list_mutex cannot be taken here in case opening the
1253 	 * underlying device takes further locks like bd_mutex.
1254 	 *
1255 	 * We also don't need the lock here as this is called during mount and
1256 	 * exclusion is provided by uuid_mutex
1257 	 */
1258 
1259 	if (fs_devices->opened) {
1260 		fs_devices->opened++;
1261 		ret = 0;
1262 	} else {
1263 		list_sort(NULL, &fs_devices->devices, devid_cmp);
1264 		ret = open_fs_devices(fs_devices, flags, holder);
1265 	}
1266 
1267 	return ret;
1268 }
1269 
1270 void btrfs_release_disk_super(struct btrfs_super_block *super)
1271 {
1272 	struct page *page = virt_to_page(super);
1273 
1274 	put_page(page);
1275 }
1276 
1277 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1278 						       u64 bytenr, u64 bytenr_orig)
1279 {
1280 	struct btrfs_super_block *disk_super;
1281 	struct page *page;
1282 	void *p;
1283 	pgoff_t index;
1284 
1285 	/* make sure our super fits in the device */
1286 	if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1287 		return ERR_PTR(-EINVAL);
1288 
1289 	/* make sure our super fits in the page */
1290 	if (sizeof(*disk_super) > PAGE_SIZE)
1291 		return ERR_PTR(-EINVAL);
1292 
1293 	/* make sure our super doesn't straddle pages on disk */
1294 	index = bytenr >> PAGE_SHIFT;
1295 	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1296 		return ERR_PTR(-EINVAL);
1297 
1298 	/* pull in the page with our super */
1299 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1300 
1301 	if (IS_ERR(page))
1302 		return ERR_CAST(page);
1303 
1304 	p = page_address(page);
1305 
1306 	/* align our pointer to the offset of the super block */
1307 	disk_super = p + offset_in_page(bytenr);
1308 
1309 	if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1310 	    btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1311 		btrfs_release_disk_super(p);
1312 		return ERR_PTR(-EINVAL);
1313 	}
1314 
1315 	if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1316 		disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1317 
1318 	return disk_super;
1319 }
1320 
1321 int btrfs_forget_devices(const char *path)
1322 {
1323 	int ret;
1324 
1325 	mutex_lock(&uuid_mutex);
1326 	ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1327 	mutex_unlock(&uuid_mutex);
1328 
1329 	return ret;
1330 }
1331 
1332 /*
1333  * Look for a btrfs signature on a device. This may be called out of the mount path
1334  * and we are not allowed to call set_blocksize during the scan. The superblock
1335  * is read via pagecache
1336  */
1337 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1338 					   void *holder)
1339 {
1340 	struct btrfs_super_block *disk_super;
1341 	bool new_device_added = false;
1342 	struct btrfs_device *device = NULL;
1343 	struct block_device *bdev;
1344 	u64 bytenr, bytenr_orig;
1345 	int ret;
1346 
1347 	lockdep_assert_held(&uuid_mutex);
1348 
1349 	/*
1350 	 * we would like to check all the supers, but that would make
1351 	 * a btrfs mount succeed after a mkfs from a different FS.
1352 	 * So, we need to add a special mount option to scan for
1353 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1354 	 */
1355 	flags |= FMODE_EXCL;
1356 
1357 	bdev = blkdev_get_by_path(path, flags, holder);
1358 	if (IS_ERR(bdev))
1359 		return ERR_CAST(bdev);
1360 
1361 	bytenr_orig = btrfs_sb_offset(0);
1362 	ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1363 	if (ret)
1364 		return ERR_PTR(ret);
1365 
1366 	disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1367 	if (IS_ERR(disk_super)) {
1368 		device = ERR_CAST(disk_super);
1369 		goto error_bdev_put;
1370 	}
1371 
1372 	device = device_list_add(path, disk_super, &new_device_added);
1373 	if (!IS_ERR(device)) {
1374 		if (new_device_added)
1375 			btrfs_free_stale_devices(path, device);
1376 	}
1377 
1378 	btrfs_release_disk_super(disk_super);
1379 
1380 error_bdev_put:
1381 	blkdev_put(bdev, flags);
1382 
1383 	return device;
1384 }
1385 
1386 /*
1387  * Try to find a chunk that intersects [start, start + len] range and when one
1388  * such is found, record the end of it in *start
1389  */
1390 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1391 				    u64 len)
1392 {
1393 	u64 physical_start, physical_end;
1394 
1395 	lockdep_assert_held(&device->fs_info->chunk_mutex);
1396 
1397 	if (!find_first_extent_bit(&device->alloc_state, *start,
1398 				   &physical_start, &physical_end,
1399 				   CHUNK_ALLOCATED, NULL)) {
1400 
1401 		if (in_range(physical_start, *start, len) ||
1402 		    in_range(*start, physical_start,
1403 			     physical_end - physical_start)) {
1404 			*start = physical_end + 1;
1405 			return true;
1406 		}
1407 	}
1408 	return false;
1409 }
1410 
1411 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1412 {
1413 	switch (device->fs_devices->chunk_alloc_policy) {
1414 	case BTRFS_CHUNK_ALLOC_REGULAR:
1415 		/*
1416 		 * We don't want to overwrite the superblock on the drive nor
1417 		 * any area used by the boot loader (grub for example), so we
1418 		 * make sure to start at an offset of at least 1MB.
1419 		 */
1420 		return max_t(u64, start, SZ_1M);
1421 	default:
1422 		BUG();
1423 	}
1424 }
1425 
1426 /**
1427  * dev_extent_hole_check - check if specified hole is suitable for allocation
1428  * @device:	the device which we have the hole
1429  * @hole_start: starting position of the hole
1430  * @hole_size:	the size of the hole
1431  * @num_bytes:	the size of the free space that we need
1432  *
1433  * This function may modify @hole_start and @hole_end to reflect the suitable
1434  * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1435  */
1436 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1437 				  u64 *hole_size, u64 num_bytes)
1438 {
1439 	bool changed = false;
1440 	u64 hole_end = *hole_start + *hole_size;
1441 
1442 	/*
1443 	 * Check before we set max_hole_start, otherwise we could end up
1444 	 * sending back this offset anyway.
1445 	 */
1446 	if (contains_pending_extent(device, hole_start, *hole_size)) {
1447 		if (hole_end >= *hole_start)
1448 			*hole_size = hole_end - *hole_start;
1449 		else
1450 			*hole_size = 0;
1451 		changed = true;
1452 	}
1453 
1454 	switch (device->fs_devices->chunk_alloc_policy) {
1455 	case BTRFS_CHUNK_ALLOC_REGULAR:
1456 		/* No extra check */
1457 		break;
1458 	default:
1459 		BUG();
1460 	}
1461 
1462 	return changed;
1463 }
1464 
1465 /*
1466  * find_free_dev_extent_start - find free space in the specified device
1467  * @device:	  the device which we search the free space in
1468  * @num_bytes:	  the size of the free space that we need
1469  * @search_start: the position from which to begin the search
1470  * @start:	  store the start of the free space.
1471  * @len:	  the size of the free space. that we find, or the size
1472  *		  of the max free space if we don't find suitable free space
1473  *
1474  * this uses a pretty simple search, the expectation is that it is
1475  * called very infrequently and that a given device has a small number
1476  * of extents
1477  *
1478  * @start is used to store the start of the free space if we find. But if we
1479  * don't find suitable free space, it will be used to store the start position
1480  * of the max free space.
1481  *
1482  * @len is used to store the size of the free space that we find.
1483  * But if we don't find suitable free space, it is used to store the size of
1484  * the max free space.
1485  *
1486  * NOTE: This function will search *commit* root of device tree, and does extra
1487  * check to ensure dev extents are not double allocated.
1488  * This makes the function safe to allocate dev extents but may not report
1489  * correct usable device space, as device extent freed in current transaction
1490  * is not reported as avaiable.
1491  */
1492 static int find_free_dev_extent_start(struct btrfs_device *device,
1493 				u64 num_bytes, u64 search_start, u64 *start,
1494 				u64 *len)
1495 {
1496 	struct btrfs_fs_info *fs_info = device->fs_info;
1497 	struct btrfs_root *root = fs_info->dev_root;
1498 	struct btrfs_key key;
1499 	struct btrfs_dev_extent *dev_extent;
1500 	struct btrfs_path *path;
1501 	u64 hole_size;
1502 	u64 max_hole_start;
1503 	u64 max_hole_size;
1504 	u64 extent_end;
1505 	u64 search_end = device->total_bytes;
1506 	int ret;
1507 	int slot;
1508 	struct extent_buffer *l;
1509 
1510 	search_start = dev_extent_search_start(device, search_start);
1511 
1512 	path = btrfs_alloc_path();
1513 	if (!path)
1514 		return -ENOMEM;
1515 
1516 	max_hole_start = search_start;
1517 	max_hole_size = 0;
1518 
1519 again:
1520 	if (search_start >= search_end ||
1521 		test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1522 		ret = -ENOSPC;
1523 		goto out;
1524 	}
1525 
1526 	path->reada = READA_FORWARD;
1527 	path->search_commit_root = 1;
1528 	path->skip_locking = 1;
1529 
1530 	key.objectid = device->devid;
1531 	key.offset = search_start;
1532 	key.type = BTRFS_DEV_EXTENT_KEY;
1533 
1534 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1535 	if (ret < 0)
1536 		goto out;
1537 	if (ret > 0) {
1538 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1539 		if (ret < 0)
1540 			goto out;
1541 	}
1542 
1543 	while (1) {
1544 		l = path->nodes[0];
1545 		slot = path->slots[0];
1546 		if (slot >= btrfs_header_nritems(l)) {
1547 			ret = btrfs_next_leaf(root, path);
1548 			if (ret == 0)
1549 				continue;
1550 			if (ret < 0)
1551 				goto out;
1552 
1553 			break;
1554 		}
1555 		btrfs_item_key_to_cpu(l, &key, slot);
1556 
1557 		if (key.objectid < device->devid)
1558 			goto next;
1559 
1560 		if (key.objectid > device->devid)
1561 			break;
1562 
1563 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1564 			goto next;
1565 
1566 		if (key.offset > search_start) {
1567 			hole_size = key.offset - search_start;
1568 			dev_extent_hole_check(device, &search_start, &hole_size,
1569 					      num_bytes);
1570 
1571 			if (hole_size > max_hole_size) {
1572 				max_hole_start = search_start;
1573 				max_hole_size = hole_size;
1574 			}
1575 
1576 			/*
1577 			 * If this free space is greater than which we need,
1578 			 * it must be the max free space that we have found
1579 			 * until now, so max_hole_start must point to the start
1580 			 * of this free space and the length of this free space
1581 			 * is stored in max_hole_size. Thus, we return
1582 			 * max_hole_start and max_hole_size and go back to the
1583 			 * caller.
1584 			 */
1585 			if (hole_size >= num_bytes) {
1586 				ret = 0;
1587 				goto out;
1588 			}
1589 		}
1590 
1591 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1592 		extent_end = key.offset + btrfs_dev_extent_length(l,
1593 								  dev_extent);
1594 		if (extent_end > search_start)
1595 			search_start = extent_end;
1596 next:
1597 		path->slots[0]++;
1598 		cond_resched();
1599 	}
1600 
1601 	/*
1602 	 * At this point, search_start should be the end of
1603 	 * allocated dev extents, and when shrinking the device,
1604 	 * search_end may be smaller than search_start.
1605 	 */
1606 	if (search_end > search_start) {
1607 		hole_size = search_end - search_start;
1608 		if (dev_extent_hole_check(device, &search_start, &hole_size,
1609 					  num_bytes)) {
1610 			btrfs_release_path(path);
1611 			goto again;
1612 		}
1613 
1614 		if (hole_size > max_hole_size) {
1615 			max_hole_start = search_start;
1616 			max_hole_size = hole_size;
1617 		}
1618 	}
1619 
1620 	/* See above. */
1621 	if (max_hole_size < num_bytes)
1622 		ret = -ENOSPC;
1623 	else
1624 		ret = 0;
1625 
1626 out:
1627 	btrfs_free_path(path);
1628 	*start = max_hole_start;
1629 	if (len)
1630 		*len = max_hole_size;
1631 	return ret;
1632 }
1633 
1634 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1635 			 u64 *start, u64 *len)
1636 {
1637 	/* FIXME use last free of some kind */
1638 	return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1639 }
1640 
1641 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1642 			  struct btrfs_device *device,
1643 			  u64 start, u64 *dev_extent_len)
1644 {
1645 	struct btrfs_fs_info *fs_info = device->fs_info;
1646 	struct btrfs_root *root = fs_info->dev_root;
1647 	int ret;
1648 	struct btrfs_path *path;
1649 	struct btrfs_key key;
1650 	struct btrfs_key found_key;
1651 	struct extent_buffer *leaf = NULL;
1652 	struct btrfs_dev_extent *extent = NULL;
1653 
1654 	path = btrfs_alloc_path();
1655 	if (!path)
1656 		return -ENOMEM;
1657 
1658 	key.objectid = device->devid;
1659 	key.offset = start;
1660 	key.type = BTRFS_DEV_EXTENT_KEY;
1661 again:
1662 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1663 	if (ret > 0) {
1664 		ret = btrfs_previous_item(root, path, key.objectid,
1665 					  BTRFS_DEV_EXTENT_KEY);
1666 		if (ret)
1667 			goto out;
1668 		leaf = path->nodes[0];
1669 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1670 		extent = btrfs_item_ptr(leaf, path->slots[0],
1671 					struct btrfs_dev_extent);
1672 		BUG_ON(found_key.offset > start || found_key.offset +
1673 		       btrfs_dev_extent_length(leaf, extent) < start);
1674 		key = found_key;
1675 		btrfs_release_path(path);
1676 		goto again;
1677 	} else if (ret == 0) {
1678 		leaf = path->nodes[0];
1679 		extent = btrfs_item_ptr(leaf, path->slots[0],
1680 					struct btrfs_dev_extent);
1681 	} else {
1682 		btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1683 		goto out;
1684 	}
1685 
1686 	*dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1687 
1688 	ret = btrfs_del_item(trans, root, path);
1689 	if (ret) {
1690 		btrfs_handle_fs_error(fs_info, ret,
1691 				      "Failed to remove dev extent item");
1692 	} else {
1693 		set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1694 	}
1695 out:
1696 	btrfs_free_path(path);
1697 	return ret;
1698 }
1699 
1700 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1701 				  struct btrfs_device *device,
1702 				  u64 chunk_offset, u64 start, u64 num_bytes)
1703 {
1704 	int ret;
1705 	struct btrfs_path *path;
1706 	struct btrfs_fs_info *fs_info = device->fs_info;
1707 	struct btrfs_root *root = fs_info->dev_root;
1708 	struct btrfs_dev_extent *extent;
1709 	struct extent_buffer *leaf;
1710 	struct btrfs_key key;
1711 
1712 	WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1713 	WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1714 	path = btrfs_alloc_path();
1715 	if (!path)
1716 		return -ENOMEM;
1717 
1718 	key.objectid = device->devid;
1719 	key.offset = start;
1720 	key.type = BTRFS_DEV_EXTENT_KEY;
1721 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1722 				      sizeof(*extent));
1723 	if (ret)
1724 		goto out;
1725 
1726 	leaf = path->nodes[0];
1727 	extent = btrfs_item_ptr(leaf, path->slots[0],
1728 				struct btrfs_dev_extent);
1729 	btrfs_set_dev_extent_chunk_tree(leaf, extent,
1730 					BTRFS_CHUNK_TREE_OBJECTID);
1731 	btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1732 					    BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1733 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1734 
1735 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1736 	btrfs_mark_buffer_dirty(leaf);
1737 out:
1738 	btrfs_free_path(path);
1739 	return ret;
1740 }
1741 
1742 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1743 {
1744 	struct extent_map_tree *em_tree;
1745 	struct extent_map *em;
1746 	struct rb_node *n;
1747 	u64 ret = 0;
1748 
1749 	em_tree = &fs_info->mapping_tree;
1750 	read_lock(&em_tree->lock);
1751 	n = rb_last(&em_tree->map.rb_root);
1752 	if (n) {
1753 		em = rb_entry(n, struct extent_map, rb_node);
1754 		ret = em->start + em->len;
1755 	}
1756 	read_unlock(&em_tree->lock);
1757 
1758 	return ret;
1759 }
1760 
1761 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1762 				    u64 *devid_ret)
1763 {
1764 	int ret;
1765 	struct btrfs_key key;
1766 	struct btrfs_key found_key;
1767 	struct btrfs_path *path;
1768 
1769 	path = btrfs_alloc_path();
1770 	if (!path)
1771 		return -ENOMEM;
1772 
1773 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1774 	key.type = BTRFS_DEV_ITEM_KEY;
1775 	key.offset = (u64)-1;
1776 
1777 	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1778 	if (ret < 0)
1779 		goto error;
1780 
1781 	if (ret == 0) {
1782 		/* Corruption */
1783 		btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1784 		ret = -EUCLEAN;
1785 		goto error;
1786 	}
1787 
1788 	ret = btrfs_previous_item(fs_info->chunk_root, path,
1789 				  BTRFS_DEV_ITEMS_OBJECTID,
1790 				  BTRFS_DEV_ITEM_KEY);
1791 	if (ret) {
1792 		*devid_ret = 1;
1793 	} else {
1794 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1795 				      path->slots[0]);
1796 		*devid_ret = found_key.offset + 1;
1797 	}
1798 	ret = 0;
1799 error:
1800 	btrfs_free_path(path);
1801 	return ret;
1802 }
1803 
1804 /*
1805  * the device information is stored in the chunk root
1806  * the btrfs_device struct should be fully filled in
1807  */
1808 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1809 			    struct btrfs_device *device)
1810 {
1811 	int ret;
1812 	struct btrfs_path *path;
1813 	struct btrfs_dev_item *dev_item;
1814 	struct extent_buffer *leaf;
1815 	struct btrfs_key key;
1816 	unsigned long ptr;
1817 
1818 	path = btrfs_alloc_path();
1819 	if (!path)
1820 		return -ENOMEM;
1821 
1822 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1823 	key.type = BTRFS_DEV_ITEM_KEY;
1824 	key.offset = device->devid;
1825 
1826 	ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1827 				      &key, sizeof(*dev_item));
1828 	if (ret)
1829 		goto out;
1830 
1831 	leaf = path->nodes[0];
1832 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1833 
1834 	btrfs_set_device_id(leaf, dev_item, device->devid);
1835 	btrfs_set_device_generation(leaf, dev_item, 0);
1836 	btrfs_set_device_type(leaf, dev_item, device->type);
1837 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1838 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1839 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1840 	btrfs_set_device_total_bytes(leaf, dev_item,
1841 				     btrfs_device_get_disk_total_bytes(device));
1842 	btrfs_set_device_bytes_used(leaf, dev_item,
1843 				    btrfs_device_get_bytes_used(device));
1844 	btrfs_set_device_group(leaf, dev_item, 0);
1845 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1846 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1847 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1848 
1849 	ptr = btrfs_device_uuid(dev_item);
1850 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1851 	ptr = btrfs_device_fsid(dev_item);
1852 	write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1853 			    ptr, BTRFS_FSID_SIZE);
1854 	btrfs_mark_buffer_dirty(leaf);
1855 
1856 	ret = 0;
1857 out:
1858 	btrfs_free_path(path);
1859 	return ret;
1860 }
1861 
1862 /*
1863  * Function to update ctime/mtime for a given device path.
1864  * Mainly used for ctime/mtime based probe like libblkid.
1865  */
1866 static void update_dev_time(const char *path_name)
1867 {
1868 	struct file *filp;
1869 
1870 	filp = filp_open(path_name, O_RDWR, 0);
1871 	if (IS_ERR(filp))
1872 		return;
1873 	file_update_time(filp);
1874 	filp_close(filp, NULL);
1875 }
1876 
1877 static int btrfs_rm_dev_item(struct btrfs_device *device)
1878 {
1879 	struct btrfs_root *root = device->fs_info->chunk_root;
1880 	int ret;
1881 	struct btrfs_path *path;
1882 	struct btrfs_key key;
1883 	struct btrfs_trans_handle *trans;
1884 
1885 	path = btrfs_alloc_path();
1886 	if (!path)
1887 		return -ENOMEM;
1888 
1889 	trans = btrfs_start_transaction(root, 0);
1890 	if (IS_ERR(trans)) {
1891 		btrfs_free_path(path);
1892 		return PTR_ERR(trans);
1893 	}
1894 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1895 	key.type = BTRFS_DEV_ITEM_KEY;
1896 	key.offset = device->devid;
1897 
1898 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1899 	if (ret) {
1900 		if (ret > 0)
1901 			ret = -ENOENT;
1902 		btrfs_abort_transaction(trans, ret);
1903 		btrfs_end_transaction(trans);
1904 		goto out;
1905 	}
1906 
1907 	ret = btrfs_del_item(trans, root, path);
1908 	if (ret) {
1909 		btrfs_abort_transaction(trans, ret);
1910 		btrfs_end_transaction(trans);
1911 	}
1912 
1913 out:
1914 	btrfs_free_path(path);
1915 	if (!ret)
1916 		ret = btrfs_commit_transaction(trans);
1917 	return ret;
1918 }
1919 
1920 /*
1921  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1922  * filesystem. It's up to the caller to adjust that number regarding eg. device
1923  * replace.
1924  */
1925 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1926 		u64 num_devices)
1927 {
1928 	u64 all_avail;
1929 	unsigned seq;
1930 	int i;
1931 
1932 	do {
1933 		seq = read_seqbegin(&fs_info->profiles_lock);
1934 
1935 		all_avail = fs_info->avail_data_alloc_bits |
1936 			    fs_info->avail_system_alloc_bits |
1937 			    fs_info->avail_metadata_alloc_bits;
1938 	} while (read_seqretry(&fs_info->profiles_lock, seq));
1939 
1940 	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1941 		if (!(all_avail & btrfs_raid_array[i].bg_flag))
1942 			continue;
1943 
1944 		if (num_devices < btrfs_raid_array[i].devs_min) {
1945 			int ret = btrfs_raid_array[i].mindev_error;
1946 
1947 			if (ret)
1948 				return ret;
1949 		}
1950 	}
1951 
1952 	return 0;
1953 }
1954 
1955 static struct btrfs_device * btrfs_find_next_active_device(
1956 		struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1957 {
1958 	struct btrfs_device *next_device;
1959 
1960 	list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1961 		if (next_device != device &&
1962 		    !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1963 		    && next_device->bdev)
1964 			return next_device;
1965 	}
1966 
1967 	return NULL;
1968 }
1969 
1970 /*
1971  * Helper function to check if the given device is part of s_bdev / latest_bdev
1972  * and replace it with the provided or the next active device, in the context
1973  * where this function called, there should be always be another device (or
1974  * this_dev) which is active.
1975  */
1976 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1977 					    struct btrfs_device *next_device)
1978 {
1979 	struct btrfs_fs_info *fs_info = device->fs_info;
1980 
1981 	if (!next_device)
1982 		next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1983 							    device);
1984 	ASSERT(next_device);
1985 
1986 	if (fs_info->sb->s_bdev &&
1987 			(fs_info->sb->s_bdev == device->bdev))
1988 		fs_info->sb->s_bdev = next_device->bdev;
1989 
1990 	if (fs_info->fs_devices->latest_bdev == device->bdev)
1991 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1992 }
1993 
1994 /*
1995  * Return btrfs_fs_devices::num_devices excluding the device that's being
1996  * currently replaced.
1997  */
1998 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
1999 {
2000 	u64 num_devices = fs_info->fs_devices->num_devices;
2001 
2002 	down_read(&fs_info->dev_replace.rwsem);
2003 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2004 		ASSERT(num_devices > 1);
2005 		num_devices--;
2006 	}
2007 	up_read(&fs_info->dev_replace.rwsem);
2008 
2009 	return num_devices;
2010 }
2011 
2012 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2013 			       struct block_device *bdev,
2014 			       const char *device_path)
2015 {
2016 	struct btrfs_super_block *disk_super;
2017 	int copy_num;
2018 
2019 	if (!bdev)
2020 		return;
2021 
2022 	for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2023 		struct page *page;
2024 		int ret;
2025 
2026 		disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2027 		if (IS_ERR(disk_super))
2028 			continue;
2029 
2030 		if (bdev_is_zoned(bdev)) {
2031 			btrfs_reset_sb_log_zones(bdev, copy_num);
2032 			continue;
2033 		}
2034 
2035 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2036 
2037 		page = virt_to_page(disk_super);
2038 		set_page_dirty(page);
2039 		lock_page(page);
2040 		/* write_on_page() unlocks the page */
2041 		ret = write_one_page(page);
2042 		if (ret)
2043 			btrfs_warn(fs_info,
2044 				"error clearing superblock number %d (%d)",
2045 				copy_num, ret);
2046 		btrfs_release_disk_super(disk_super);
2047 
2048 	}
2049 
2050 	/* Notify udev that device has changed */
2051 	btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2052 
2053 	/* Update ctime/mtime for device path for libblkid */
2054 	update_dev_time(device_path);
2055 }
2056 
2057 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2058 		    u64 devid)
2059 {
2060 	struct btrfs_device *device;
2061 	struct btrfs_fs_devices *cur_devices;
2062 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2063 	u64 num_devices;
2064 	int ret = 0;
2065 
2066 	mutex_lock(&uuid_mutex);
2067 
2068 	num_devices = btrfs_num_devices(fs_info);
2069 
2070 	ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2071 	if (ret)
2072 		goto out;
2073 
2074 	device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2075 
2076 	if (IS_ERR(device)) {
2077 		if (PTR_ERR(device) == -ENOENT &&
2078 		    strcmp(device_path, "missing") == 0)
2079 			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2080 		else
2081 			ret = PTR_ERR(device);
2082 		goto out;
2083 	}
2084 
2085 	if (btrfs_pinned_by_swapfile(fs_info, device)) {
2086 		btrfs_warn_in_rcu(fs_info,
2087 		  "cannot remove device %s (devid %llu) due to active swapfile",
2088 				  rcu_str_deref(device->name), device->devid);
2089 		ret = -ETXTBSY;
2090 		goto out;
2091 	}
2092 
2093 	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2094 		ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2095 		goto out;
2096 	}
2097 
2098 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2099 	    fs_info->fs_devices->rw_devices == 1) {
2100 		ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2101 		goto out;
2102 	}
2103 
2104 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2105 		mutex_lock(&fs_info->chunk_mutex);
2106 		list_del_init(&device->dev_alloc_list);
2107 		device->fs_devices->rw_devices--;
2108 		mutex_unlock(&fs_info->chunk_mutex);
2109 	}
2110 
2111 	mutex_unlock(&uuid_mutex);
2112 	ret = btrfs_shrink_device(device, 0);
2113 	if (!ret)
2114 		btrfs_reada_remove_dev(device);
2115 	mutex_lock(&uuid_mutex);
2116 	if (ret)
2117 		goto error_undo;
2118 
2119 	/*
2120 	 * TODO: the superblock still includes this device in its num_devices
2121 	 * counter although write_all_supers() is not locked out. This
2122 	 * could give a filesystem state which requires a degraded mount.
2123 	 */
2124 	ret = btrfs_rm_dev_item(device);
2125 	if (ret)
2126 		goto error_undo;
2127 
2128 	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2129 	btrfs_scrub_cancel_dev(device);
2130 
2131 	/*
2132 	 * the device list mutex makes sure that we don't change
2133 	 * the device list while someone else is writing out all
2134 	 * the device supers. Whoever is writing all supers, should
2135 	 * lock the device list mutex before getting the number of
2136 	 * devices in the super block (super_copy). Conversely,
2137 	 * whoever updates the number of devices in the super block
2138 	 * (super_copy) should hold the device list mutex.
2139 	 */
2140 
2141 	/*
2142 	 * In normal cases the cur_devices == fs_devices. But in case
2143 	 * of deleting a seed device, the cur_devices should point to
2144 	 * its own fs_devices listed under the fs_devices->seed.
2145 	 */
2146 	cur_devices = device->fs_devices;
2147 	mutex_lock(&fs_devices->device_list_mutex);
2148 	list_del_rcu(&device->dev_list);
2149 
2150 	cur_devices->num_devices--;
2151 	cur_devices->total_devices--;
2152 	/* Update total_devices of the parent fs_devices if it's seed */
2153 	if (cur_devices != fs_devices)
2154 		fs_devices->total_devices--;
2155 
2156 	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2157 		cur_devices->missing_devices--;
2158 
2159 	btrfs_assign_next_active_device(device, NULL);
2160 
2161 	if (device->bdev) {
2162 		cur_devices->open_devices--;
2163 		/* remove sysfs entry */
2164 		btrfs_sysfs_remove_device(device);
2165 	}
2166 
2167 	num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2168 	btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2169 	mutex_unlock(&fs_devices->device_list_mutex);
2170 
2171 	/*
2172 	 * at this point, the device is zero sized and detached from
2173 	 * the devices list.  All that's left is to zero out the old
2174 	 * supers and free the device.
2175 	 */
2176 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2177 		btrfs_scratch_superblocks(fs_info, device->bdev,
2178 					  device->name->str);
2179 
2180 	btrfs_close_bdev(device);
2181 	synchronize_rcu();
2182 	btrfs_free_device(device);
2183 
2184 	if (cur_devices->open_devices == 0) {
2185 		list_del_init(&cur_devices->seed_list);
2186 		close_fs_devices(cur_devices);
2187 		free_fs_devices(cur_devices);
2188 	}
2189 
2190 out:
2191 	mutex_unlock(&uuid_mutex);
2192 	return ret;
2193 
2194 error_undo:
2195 	btrfs_reada_undo_remove_dev(device);
2196 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2197 		mutex_lock(&fs_info->chunk_mutex);
2198 		list_add(&device->dev_alloc_list,
2199 			 &fs_devices->alloc_list);
2200 		device->fs_devices->rw_devices++;
2201 		mutex_unlock(&fs_info->chunk_mutex);
2202 	}
2203 	goto out;
2204 }
2205 
2206 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2207 {
2208 	struct btrfs_fs_devices *fs_devices;
2209 
2210 	lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2211 
2212 	/*
2213 	 * in case of fs with no seed, srcdev->fs_devices will point
2214 	 * to fs_devices of fs_info. However when the dev being replaced is
2215 	 * a seed dev it will point to the seed's local fs_devices. In short
2216 	 * srcdev will have its correct fs_devices in both the cases.
2217 	 */
2218 	fs_devices = srcdev->fs_devices;
2219 
2220 	list_del_rcu(&srcdev->dev_list);
2221 	list_del(&srcdev->dev_alloc_list);
2222 	fs_devices->num_devices--;
2223 	if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2224 		fs_devices->missing_devices--;
2225 
2226 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2227 		fs_devices->rw_devices--;
2228 
2229 	if (srcdev->bdev)
2230 		fs_devices->open_devices--;
2231 }
2232 
2233 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2234 {
2235 	struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2236 
2237 	mutex_lock(&uuid_mutex);
2238 
2239 	btrfs_close_bdev(srcdev);
2240 	synchronize_rcu();
2241 	btrfs_free_device(srcdev);
2242 
2243 	/* if this is no devs we rather delete the fs_devices */
2244 	if (!fs_devices->num_devices) {
2245 		/*
2246 		 * On a mounted FS, num_devices can't be zero unless it's a
2247 		 * seed. In case of a seed device being replaced, the replace
2248 		 * target added to the sprout FS, so there will be no more
2249 		 * device left under the seed FS.
2250 		 */
2251 		ASSERT(fs_devices->seeding);
2252 
2253 		list_del_init(&fs_devices->seed_list);
2254 		close_fs_devices(fs_devices);
2255 		free_fs_devices(fs_devices);
2256 	}
2257 	mutex_unlock(&uuid_mutex);
2258 }
2259 
2260 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2261 {
2262 	struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2263 
2264 	mutex_lock(&fs_devices->device_list_mutex);
2265 
2266 	btrfs_sysfs_remove_device(tgtdev);
2267 
2268 	if (tgtdev->bdev)
2269 		fs_devices->open_devices--;
2270 
2271 	fs_devices->num_devices--;
2272 
2273 	btrfs_assign_next_active_device(tgtdev, NULL);
2274 
2275 	list_del_rcu(&tgtdev->dev_list);
2276 
2277 	mutex_unlock(&fs_devices->device_list_mutex);
2278 
2279 	/*
2280 	 * The update_dev_time() with in btrfs_scratch_superblocks()
2281 	 * may lead to a call to btrfs_show_devname() which will try
2282 	 * to hold device_list_mutex. And here this device
2283 	 * is already out of device list, so we don't have to hold
2284 	 * the device_list_mutex lock.
2285 	 */
2286 	btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2287 				  tgtdev->name->str);
2288 
2289 	btrfs_close_bdev(tgtdev);
2290 	synchronize_rcu();
2291 	btrfs_free_device(tgtdev);
2292 }
2293 
2294 static struct btrfs_device *btrfs_find_device_by_path(
2295 		struct btrfs_fs_info *fs_info, const char *device_path)
2296 {
2297 	int ret = 0;
2298 	struct btrfs_super_block *disk_super;
2299 	u64 devid;
2300 	u8 *dev_uuid;
2301 	struct block_device *bdev;
2302 	struct btrfs_device *device;
2303 
2304 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2305 				    fs_info->bdev_holder, 0, &bdev, &disk_super);
2306 	if (ret)
2307 		return ERR_PTR(ret);
2308 
2309 	devid = btrfs_stack_device_id(&disk_super->dev_item);
2310 	dev_uuid = disk_super->dev_item.uuid;
2311 	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2312 		device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2313 					   disk_super->metadata_uuid);
2314 	else
2315 		device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2316 					   disk_super->fsid);
2317 
2318 	btrfs_release_disk_super(disk_super);
2319 	if (!device)
2320 		device = ERR_PTR(-ENOENT);
2321 	blkdev_put(bdev, FMODE_READ);
2322 	return device;
2323 }
2324 
2325 /*
2326  * Lookup a device given by device id, or the path if the id is 0.
2327  */
2328 struct btrfs_device *btrfs_find_device_by_devspec(
2329 		struct btrfs_fs_info *fs_info, u64 devid,
2330 		const char *device_path)
2331 {
2332 	struct btrfs_device *device;
2333 
2334 	if (devid) {
2335 		device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2336 					   NULL);
2337 		if (!device)
2338 			return ERR_PTR(-ENOENT);
2339 		return device;
2340 	}
2341 
2342 	if (!device_path || !device_path[0])
2343 		return ERR_PTR(-EINVAL);
2344 
2345 	if (strcmp(device_path, "missing") == 0) {
2346 		/* Find first missing device */
2347 		list_for_each_entry(device, &fs_info->fs_devices->devices,
2348 				    dev_list) {
2349 			if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2350 				     &device->dev_state) && !device->bdev)
2351 				return device;
2352 		}
2353 		return ERR_PTR(-ENOENT);
2354 	}
2355 
2356 	return btrfs_find_device_by_path(fs_info, device_path);
2357 }
2358 
2359 /*
2360  * does all the dirty work required for changing file system's UUID.
2361  */
2362 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2363 {
2364 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2365 	struct btrfs_fs_devices *old_devices;
2366 	struct btrfs_fs_devices *seed_devices;
2367 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2368 	struct btrfs_device *device;
2369 	u64 super_flags;
2370 
2371 	lockdep_assert_held(&uuid_mutex);
2372 	if (!fs_devices->seeding)
2373 		return -EINVAL;
2374 
2375 	/*
2376 	 * Private copy of the seed devices, anchored at
2377 	 * fs_info->fs_devices->seed_list
2378 	 */
2379 	seed_devices = alloc_fs_devices(NULL, NULL);
2380 	if (IS_ERR(seed_devices))
2381 		return PTR_ERR(seed_devices);
2382 
2383 	/*
2384 	 * It's necessary to retain a copy of the original seed fs_devices in
2385 	 * fs_uuids so that filesystems which have been seeded can successfully
2386 	 * reference the seed device from open_seed_devices. This also supports
2387 	 * multiple fs seed.
2388 	 */
2389 	old_devices = clone_fs_devices(fs_devices);
2390 	if (IS_ERR(old_devices)) {
2391 		kfree(seed_devices);
2392 		return PTR_ERR(old_devices);
2393 	}
2394 
2395 	list_add(&old_devices->fs_list, &fs_uuids);
2396 
2397 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2398 	seed_devices->opened = 1;
2399 	INIT_LIST_HEAD(&seed_devices->devices);
2400 	INIT_LIST_HEAD(&seed_devices->alloc_list);
2401 	mutex_init(&seed_devices->device_list_mutex);
2402 
2403 	mutex_lock(&fs_devices->device_list_mutex);
2404 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2405 			      synchronize_rcu);
2406 	list_for_each_entry(device, &seed_devices->devices, dev_list)
2407 		device->fs_devices = seed_devices;
2408 
2409 	fs_devices->seeding = false;
2410 	fs_devices->num_devices = 0;
2411 	fs_devices->open_devices = 0;
2412 	fs_devices->missing_devices = 0;
2413 	fs_devices->rotating = false;
2414 	list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2415 
2416 	generate_random_uuid(fs_devices->fsid);
2417 	memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2418 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2419 	mutex_unlock(&fs_devices->device_list_mutex);
2420 
2421 	super_flags = btrfs_super_flags(disk_super) &
2422 		      ~BTRFS_SUPER_FLAG_SEEDING;
2423 	btrfs_set_super_flags(disk_super, super_flags);
2424 
2425 	return 0;
2426 }
2427 
2428 /*
2429  * Store the expected generation for seed devices in device items.
2430  */
2431 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2432 {
2433 	struct btrfs_fs_info *fs_info = trans->fs_info;
2434 	struct btrfs_root *root = fs_info->chunk_root;
2435 	struct btrfs_path *path;
2436 	struct extent_buffer *leaf;
2437 	struct btrfs_dev_item *dev_item;
2438 	struct btrfs_device *device;
2439 	struct btrfs_key key;
2440 	u8 fs_uuid[BTRFS_FSID_SIZE];
2441 	u8 dev_uuid[BTRFS_UUID_SIZE];
2442 	u64 devid;
2443 	int ret;
2444 
2445 	path = btrfs_alloc_path();
2446 	if (!path)
2447 		return -ENOMEM;
2448 
2449 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2450 	key.offset = 0;
2451 	key.type = BTRFS_DEV_ITEM_KEY;
2452 
2453 	while (1) {
2454 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2455 		if (ret < 0)
2456 			goto error;
2457 
2458 		leaf = path->nodes[0];
2459 next_slot:
2460 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2461 			ret = btrfs_next_leaf(root, path);
2462 			if (ret > 0)
2463 				break;
2464 			if (ret < 0)
2465 				goto error;
2466 			leaf = path->nodes[0];
2467 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2468 			btrfs_release_path(path);
2469 			continue;
2470 		}
2471 
2472 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2473 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2474 		    key.type != BTRFS_DEV_ITEM_KEY)
2475 			break;
2476 
2477 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
2478 					  struct btrfs_dev_item);
2479 		devid = btrfs_device_id(leaf, dev_item);
2480 		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2481 				   BTRFS_UUID_SIZE);
2482 		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2483 				   BTRFS_FSID_SIZE);
2484 		device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2485 					   fs_uuid);
2486 		BUG_ON(!device); /* Logic error */
2487 
2488 		if (device->fs_devices->seeding) {
2489 			btrfs_set_device_generation(leaf, dev_item,
2490 						    device->generation);
2491 			btrfs_mark_buffer_dirty(leaf);
2492 		}
2493 
2494 		path->slots[0]++;
2495 		goto next_slot;
2496 	}
2497 	ret = 0;
2498 error:
2499 	btrfs_free_path(path);
2500 	return ret;
2501 }
2502 
2503 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2504 {
2505 	struct btrfs_root *root = fs_info->dev_root;
2506 	struct request_queue *q;
2507 	struct btrfs_trans_handle *trans;
2508 	struct btrfs_device *device;
2509 	struct block_device *bdev;
2510 	struct super_block *sb = fs_info->sb;
2511 	struct rcu_string *name;
2512 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2513 	u64 orig_super_total_bytes;
2514 	u64 orig_super_num_devices;
2515 	int seeding_dev = 0;
2516 	int ret = 0;
2517 	bool locked = false;
2518 
2519 	if (sb_rdonly(sb) && !fs_devices->seeding)
2520 		return -EROFS;
2521 
2522 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2523 				  fs_info->bdev_holder);
2524 	if (IS_ERR(bdev))
2525 		return PTR_ERR(bdev);
2526 
2527 	if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2528 		ret = -EINVAL;
2529 		goto error;
2530 	}
2531 
2532 	if (fs_devices->seeding) {
2533 		seeding_dev = 1;
2534 		down_write(&sb->s_umount);
2535 		mutex_lock(&uuid_mutex);
2536 		locked = true;
2537 	}
2538 
2539 	sync_blockdev(bdev);
2540 
2541 	rcu_read_lock();
2542 	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2543 		if (device->bdev == bdev) {
2544 			ret = -EEXIST;
2545 			rcu_read_unlock();
2546 			goto error;
2547 		}
2548 	}
2549 	rcu_read_unlock();
2550 
2551 	device = btrfs_alloc_device(fs_info, NULL, NULL);
2552 	if (IS_ERR(device)) {
2553 		/* we can safely leave the fs_devices entry around */
2554 		ret = PTR_ERR(device);
2555 		goto error;
2556 	}
2557 
2558 	name = rcu_string_strdup(device_path, GFP_KERNEL);
2559 	if (!name) {
2560 		ret = -ENOMEM;
2561 		goto error_free_device;
2562 	}
2563 	rcu_assign_pointer(device->name, name);
2564 
2565 	device->fs_info = fs_info;
2566 	device->bdev = bdev;
2567 
2568 	ret = btrfs_get_dev_zone_info(device);
2569 	if (ret)
2570 		goto error_free_device;
2571 
2572 	trans = btrfs_start_transaction(root, 0);
2573 	if (IS_ERR(trans)) {
2574 		ret = PTR_ERR(trans);
2575 		goto error_free_zone;
2576 	}
2577 
2578 	q = bdev_get_queue(bdev);
2579 	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2580 	device->generation = trans->transid;
2581 	device->io_width = fs_info->sectorsize;
2582 	device->io_align = fs_info->sectorsize;
2583 	device->sector_size = fs_info->sectorsize;
2584 	device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2585 					 fs_info->sectorsize);
2586 	device->disk_total_bytes = device->total_bytes;
2587 	device->commit_total_bytes = device->total_bytes;
2588 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2589 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2590 	device->mode = FMODE_EXCL;
2591 	device->dev_stats_valid = 1;
2592 	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2593 
2594 	if (seeding_dev) {
2595 		btrfs_clear_sb_rdonly(sb);
2596 		ret = btrfs_prepare_sprout(fs_info);
2597 		if (ret) {
2598 			btrfs_abort_transaction(trans, ret);
2599 			goto error_trans;
2600 		}
2601 	}
2602 
2603 	device->fs_devices = fs_devices;
2604 
2605 	mutex_lock(&fs_devices->device_list_mutex);
2606 	mutex_lock(&fs_info->chunk_mutex);
2607 	list_add_rcu(&device->dev_list, &fs_devices->devices);
2608 	list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2609 	fs_devices->num_devices++;
2610 	fs_devices->open_devices++;
2611 	fs_devices->rw_devices++;
2612 	fs_devices->total_devices++;
2613 	fs_devices->total_rw_bytes += device->total_bytes;
2614 
2615 	atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2616 
2617 	if (!blk_queue_nonrot(q))
2618 		fs_devices->rotating = true;
2619 
2620 	orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2621 	btrfs_set_super_total_bytes(fs_info->super_copy,
2622 		round_down(orig_super_total_bytes + device->total_bytes,
2623 			   fs_info->sectorsize));
2624 
2625 	orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2626 	btrfs_set_super_num_devices(fs_info->super_copy,
2627 				    orig_super_num_devices + 1);
2628 
2629 	/*
2630 	 * we've got more storage, clear any full flags on the space
2631 	 * infos
2632 	 */
2633 	btrfs_clear_space_info_full(fs_info);
2634 
2635 	mutex_unlock(&fs_info->chunk_mutex);
2636 
2637 	/* Add sysfs device entry */
2638 	btrfs_sysfs_add_device(device);
2639 
2640 	mutex_unlock(&fs_devices->device_list_mutex);
2641 
2642 	if (seeding_dev) {
2643 		mutex_lock(&fs_info->chunk_mutex);
2644 		ret = init_first_rw_device(trans);
2645 		mutex_unlock(&fs_info->chunk_mutex);
2646 		if (ret) {
2647 			btrfs_abort_transaction(trans, ret);
2648 			goto error_sysfs;
2649 		}
2650 	}
2651 
2652 	ret = btrfs_add_dev_item(trans, device);
2653 	if (ret) {
2654 		btrfs_abort_transaction(trans, ret);
2655 		goto error_sysfs;
2656 	}
2657 
2658 	if (seeding_dev) {
2659 		ret = btrfs_finish_sprout(trans);
2660 		if (ret) {
2661 			btrfs_abort_transaction(trans, ret);
2662 			goto error_sysfs;
2663 		}
2664 
2665 		/*
2666 		 * fs_devices now represents the newly sprouted filesystem and
2667 		 * its fsid has been changed by btrfs_prepare_sprout
2668 		 */
2669 		btrfs_sysfs_update_sprout_fsid(fs_devices);
2670 	}
2671 
2672 	ret = btrfs_commit_transaction(trans);
2673 
2674 	if (seeding_dev) {
2675 		mutex_unlock(&uuid_mutex);
2676 		up_write(&sb->s_umount);
2677 		locked = false;
2678 
2679 		if (ret) /* transaction commit */
2680 			return ret;
2681 
2682 		ret = btrfs_relocate_sys_chunks(fs_info);
2683 		if (ret < 0)
2684 			btrfs_handle_fs_error(fs_info, ret,
2685 				    "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2686 		trans = btrfs_attach_transaction(root);
2687 		if (IS_ERR(trans)) {
2688 			if (PTR_ERR(trans) == -ENOENT)
2689 				return 0;
2690 			ret = PTR_ERR(trans);
2691 			trans = NULL;
2692 			goto error_sysfs;
2693 		}
2694 		ret = btrfs_commit_transaction(trans);
2695 	}
2696 
2697 	/*
2698 	 * Now that we have written a new super block to this device, check all
2699 	 * other fs_devices list if device_path alienates any other scanned
2700 	 * device.
2701 	 * We can ignore the return value as it typically returns -EINVAL and
2702 	 * only succeeds if the device was an alien.
2703 	 */
2704 	btrfs_forget_devices(device_path);
2705 
2706 	/* Update ctime/mtime for blkid or udev */
2707 	update_dev_time(device_path);
2708 
2709 	return ret;
2710 
2711 error_sysfs:
2712 	btrfs_sysfs_remove_device(device);
2713 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2714 	mutex_lock(&fs_info->chunk_mutex);
2715 	list_del_rcu(&device->dev_list);
2716 	list_del(&device->dev_alloc_list);
2717 	fs_info->fs_devices->num_devices--;
2718 	fs_info->fs_devices->open_devices--;
2719 	fs_info->fs_devices->rw_devices--;
2720 	fs_info->fs_devices->total_devices--;
2721 	fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2722 	atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2723 	btrfs_set_super_total_bytes(fs_info->super_copy,
2724 				    orig_super_total_bytes);
2725 	btrfs_set_super_num_devices(fs_info->super_copy,
2726 				    orig_super_num_devices);
2727 	mutex_unlock(&fs_info->chunk_mutex);
2728 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2729 error_trans:
2730 	if (seeding_dev)
2731 		btrfs_set_sb_rdonly(sb);
2732 	if (trans)
2733 		btrfs_end_transaction(trans);
2734 error_free_zone:
2735 	btrfs_destroy_dev_zone_info(device);
2736 error_free_device:
2737 	btrfs_free_device(device);
2738 error:
2739 	blkdev_put(bdev, FMODE_EXCL);
2740 	if (locked) {
2741 		mutex_unlock(&uuid_mutex);
2742 		up_write(&sb->s_umount);
2743 	}
2744 	return ret;
2745 }
2746 
2747 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2748 					struct btrfs_device *device)
2749 {
2750 	int ret;
2751 	struct btrfs_path *path;
2752 	struct btrfs_root *root = device->fs_info->chunk_root;
2753 	struct btrfs_dev_item *dev_item;
2754 	struct extent_buffer *leaf;
2755 	struct btrfs_key key;
2756 
2757 	path = btrfs_alloc_path();
2758 	if (!path)
2759 		return -ENOMEM;
2760 
2761 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2762 	key.type = BTRFS_DEV_ITEM_KEY;
2763 	key.offset = device->devid;
2764 
2765 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2766 	if (ret < 0)
2767 		goto out;
2768 
2769 	if (ret > 0) {
2770 		ret = -ENOENT;
2771 		goto out;
2772 	}
2773 
2774 	leaf = path->nodes[0];
2775 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2776 
2777 	btrfs_set_device_id(leaf, dev_item, device->devid);
2778 	btrfs_set_device_type(leaf, dev_item, device->type);
2779 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2780 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2781 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2782 	btrfs_set_device_total_bytes(leaf, dev_item,
2783 				     btrfs_device_get_disk_total_bytes(device));
2784 	btrfs_set_device_bytes_used(leaf, dev_item,
2785 				    btrfs_device_get_bytes_used(device));
2786 	btrfs_mark_buffer_dirty(leaf);
2787 
2788 out:
2789 	btrfs_free_path(path);
2790 	return ret;
2791 }
2792 
2793 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2794 		      struct btrfs_device *device, u64 new_size)
2795 {
2796 	struct btrfs_fs_info *fs_info = device->fs_info;
2797 	struct btrfs_super_block *super_copy = fs_info->super_copy;
2798 	u64 old_total;
2799 	u64 diff;
2800 
2801 	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2802 		return -EACCES;
2803 
2804 	new_size = round_down(new_size, fs_info->sectorsize);
2805 
2806 	mutex_lock(&fs_info->chunk_mutex);
2807 	old_total = btrfs_super_total_bytes(super_copy);
2808 	diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2809 
2810 	if (new_size <= device->total_bytes ||
2811 	    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2812 		mutex_unlock(&fs_info->chunk_mutex);
2813 		return -EINVAL;
2814 	}
2815 
2816 	btrfs_set_super_total_bytes(super_copy,
2817 			round_down(old_total + diff, fs_info->sectorsize));
2818 	device->fs_devices->total_rw_bytes += diff;
2819 
2820 	btrfs_device_set_total_bytes(device, new_size);
2821 	btrfs_device_set_disk_total_bytes(device, new_size);
2822 	btrfs_clear_space_info_full(device->fs_info);
2823 	if (list_empty(&device->post_commit_list))
2824 		list_add_tail(&device->post_commit_list,
2825 			      &trans->transaction->dev_update_list);
2826 	mutex_unlock(&fs_info->chunk_mutex);
2827 
2828 	return btrfs_update_device(trans, device);
2829 }
2830 
2831 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2832 {
2833 	struct btrfs_fs_info *fs_info = trans->fs_info;
2834 	struct btrfs_root *root = fs_info->chunk_root;
2835 	int ret;
2836 	struct btrfs_path *path;
2837 	struct btrfs_key key;
2838 
2839 	path = btrfs_alloc_path();
2840 	if (!path)
2841 		return -ENOMEM;
2842 
2843 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2844 	key.offset = chunk_offset;
2845 	key.type = BTRFS_CHUNK_ITEM_KEY;
2846 
2847 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2848 	if (ret < 0)
2849 		goto out;
2850 	else if (ret > 0) { /* Logic error or corruption */
2851 		btrfs_handle_fs_error(fs_info, -ENOENT,
2852 				      "Failed lookup while freeing chunk.");
2853 		ret = -ENOENT;
2854 		goto out;
2855 	}
2856 
2857 	ret = btrfs_del_item(trans, root, path);
2858 	if (ret < 0)
2859 		btrfs_handle_fs_error(fs_info, ret,
2860 				      "Failed to delete chunk item.");
2861 out:
2862 	btrfs_free_path(path);
2863 	return ret;
2864 }
2865 
2866 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2867 {
2868 	struct btrfs_super_block *super_copy = fs_info->super_copy;
2869 	struct btrfs_disk_key *disk_key;
2870 	struct btrfs_chunk *chunk;
2871 	u8 *ptr;
2872 	int ret = 0;
2873 	u32 num_stripes;
2874 	u32 array_size;
2875 	u32 len = 0;
2876 	u32 cur;
2877 	struct btrfs_key key;
2878 
2879 	mutex_lock(&fs_info->chunk_mutex);
2880 	array_size = btrfs_super_sys_array_size(super_copy);
2881 
2882 	ptr = super_copy->sys_chunk_array;
2883 	cur = 0;
2884 
2885 	while (cur < array_size) {
2886 		disk_key = (struct btrfs_disk_key *)ptr;
2887 		btrfs_disk_key_to_cpu(&key, disk_key);
2888 
2889 		len = sizeof(*disk_key);
2890 
2891 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2892 			chunk = (struct btrfs_chunk *)(ptr + len);
2893 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2894 			len += btrfs_chunk_item_size(num_stripes);
2895 		} else {
2896 			ret = -EIO;
2897 			break;
2898 		}
2899 		if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2900 		    key.offset == chunk_offset) {
2901 			memmove(ptr, ptr + len, array_size - (cur + len));
2902 			array_size -= len;
2903 			btrfs_set_super_sys_array_size(super_copy, array_size);
2904 		} else {
2905 			ptr += len;
2906 			cur += len;
2907 		}
2908 	}
2909 	mutex_unlock(&fs_info->chunk_mutex);
2910 	return ret;
2911 }
2912 
2913 /*
2914  * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2915  * @logical: Logical block offset in bytes.
2916  * @length: Length of extent in bytes.
2917  *
2918  * Return: Chunk mapping or ERR_PTR.
2919  */
2920 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2921 				       u64 logical, u64 length)
2922 {
2923 	struct extent_map_tree *em_tree;
2924 	struct extent_map *em;
2925 
2926 	em_tree = &fs_info->mapping_tree;
2927 	read_lock(&em_tree->lock);
2928 	em = lookup_extent_mapping(em_tree, logical, length);
2929 	read_unlock(&em_tree->lock);
2930 
2931 	if (!em) {
2932 		btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2933 			   logical, length);
2934 		return ERR_PTR(-EINVAL);
2935 	}
2936 
2937 	if (em->start > logical || em->start + em->len < logical) {
2938 		btrfs_crit(fs_info,
2939 			   "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2940 			   logical, length, em->start, em->start + em->len);
2941 		free_extent_map(em);
2942 		return ERR_PTR(-EINVAL);
2943 	}
2944 
2945 	/* callers are responsible for dropping em's ref. */
2946 	return em;
2947 }
2948 
2949 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2950 {
2951 	struct btrfs_fs_info *fs_info = trans->fs_info;
2952 	struct extent_map *em;
2953 	struct map_lookup *map;
2954 	u64 dev_extent_len = 0;
2955 	int i, ret = 0;
2956 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2957 
2958 	em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2959 	if (IS_ERR(em)) {
2960 		/*
2961 		 * This is a logic error, but we don't want to just rely on the
2962 		 * user having built with ASSERT enabled, so if ASSERT doesn't
2963 		 * do anything we still error out.
2964 		 */
2965 		ASSERT(0);
2966 		return PTR_ERR(em);
2967 	}
2968 	map = em->map_lookup;
2969 	mutex_lock(&fs_info->chunk_mutex);
2970 	check_system_chunk(trans, map->type);
2971 	mutex_unlock(&fs_info->chunk_mutex);
2972 
2973 	/*
2974 	 * Take the device list mutex to prevent races with the final phase of
2975 	 * a device replace operation that replaces the device object associated
2976 	 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2977 	 */
2978 	mutex_lock(&fs_devices->device_list_mutex);
2979 	for (i = 0; i < map->num_stripes; i++) {
2980 		struct btrfs_device *device = map->stripes[i].dev;
2981 		ret = btrfs_free_dev_extent(trans, device,
2982 					    map->stripes[i].physical,
2983 					    &dev_extent_len);
2984 		if (ret) {
2985 			mutex_unlock(&fs_devices->device_list_mutex);
2986 			btrfs_abort_transaction(trans, ret);
2987 			goto out;
2988 		}
2989 
2990 		if (device->bytes_used > 0) {
2991 			mutex_lock(&fs_info->chunk_mutex);
2992 			btrfs_device_set_bytes_used(device,
2993 					device->bytes_used - dev_extent_len);
2994 			atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2995 			btrfs_clear_space_info_full(fs_info);
2996 			mutex_unlock(&fs_info->chunk_mutex);
2997 		}
2998 
2999 		ret = btrfs_update_device(trans, device);
3000 		if (ret) {
3001 			mutex_unlock(&fs_devices->device_list_mutex);
3002 			btrfs_abort_transaction(trans, ret);
3003 			goto out;
3004 		}
3005 	}
3006 	mutex_unlock(&fs_devices->device_list_mutex);
3007 
3008 	ret = btrfs_free_chunk(trans, chunk_offset);
3009 	if (ret) {
3010 		btrfs_abort_transaction(trans, ret);
3011 		goto out;
3012 	}
3013 
3014 	trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3015 
3016 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3017 		ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3018 		if (ret) {
3019 			btrfs_abort_transaction(trans, ret);
3020 			goto out;
3021 		}
3022 	}
3023 
3024 	ret = btrfs_remove_block_group(trans, chunk_offset, em);
3025 	if (ret) {
3026 		btrfs_abort_transaction(trans, ret);
3027 		goto out;
3028 	}
3029 
3030 out:
3031 	/* once for us */
3032 	free_extent_map(em);
3033 	return ret;
3034 }
3035 
3036 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3037 {
3038 	struct btrfs_root *root = fs_info->chunk_root;
3039 	struct btrfs_trans_handle *trans;
3040 	struct btrfs_block_group *block_group;
3041 	int ret;
3042 
3043 	/*
3044 	 * Prevent races with automatic removal of unused block groups.
3045 	 * After we relocate and before we remove the chunk with offset
3046 	 * chunk_offset, automatic removal of the block group can kick in,
3047 	 * resulting in a failure when calling btrfs_remove_chunk() below.
3048 	 *
3049 	 * Make sure to acquire this mutex before doing a tree search (dev
3050 	 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3051 	 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3052 	 * we release the path used to search the chunk/dev tree and before
3053 	 * the current task acquires this mutex and calls us.
3054 	 */
3055 	lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3056 
3057 	/* step one, relocate all the extents inside this chunk */
3058 	btrfs_scrub_pause(fs_info);
3059 	ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3060 	btrfs_scrub_continue(fs_info);
3061 	if (ret)
3062 		return ret;
3063 
3064 	block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3065 	if (!block_group)
3066 		return -ENOENT;
3067 	btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3068 	btrfs_put_block_group(block_group);
3069 
3070 	trans = btrfs_start_trans_remove_block_group(root->fs_info,
3071 						     chunk_offset);
3072 	if (IS_ERR(trans)) {
3073 		ret = PTR_ERR(trans);
3074 		btrfs_handle_fs_error(root->fs_info, ret, NULL);
3075 		return ret;
3076 	}
3077 
3078 	/*
3079 	 * step two, delete the device extents and the
3080 	 * chunk tree entries
3081 	 */
3082 	ret = btrfs_remove_chunk(trans, chunk_offset);
3083 	btrfs_end_transaction(trans);
3084 	return ret;
3085 }
3086 
3087 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3088 {
3089 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3090 	struct btrfs_path *path;
3091 	struct extent_buffer *leaf;
3092 	struct btrfs_chunk *chunk;
3093 	struct btrfs_key key;
3094 	struct btrfs_key found_key;
3095 	u64 chunk_type;
3096 	bool retried = false;
3097 	int failed = 0;
3098 	int ret;
3099 
3100 	path = btrfs_alloc_path();
3101 	if (!path)
3102 		return -ENOMEM;
3103 
3104 again:
3105 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3106 	key.offset = (u64)-1;
3107 	key.type = BTRFS_CHUNK_ITEM_KEY;
3108 
3109 	while (1) {
3110 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
3111 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3112 		if (ret < 0) {
3113 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3114 			goto error;
3115 		}
3116 		BUG_ON(ret == 0); /* Corruption */
3117 
3118 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
3119 					  key.type);
3120 		if (ret)
3121 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3122 		if (ret < 0)
3123 			goto error;
3124 		if (ret > 0)
3125 			break;
3126 
3127 		leaf = path->nodes[0];
3128 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3129 
3130 		chunk = btrfs_item_ptr(leaf, path->slots[0],
3131 				       struct btrfs_chunk);
3132 		chunk_type = btrfs_chunk_type(leaf, chunk);
3133 		btrfs_release_path(path);
3134 
3135 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3136 			ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3137 			if (ret == -ENOSPC)
3138 				failed++;
3139 			else
3140 				BUG_ON(ret);
3141 		}
3142 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3143 
3144 		if (found_key.offset == 0)
3145 			break;
3146 		key.offset = found_key.offset - 1;
3147 	}
3148 	ret = 0;
3149 	if (failed && !retried) {
3150 		failed = 0;
3151 		retried = true;
3152 		goto again;
3153 	} else if (WARN_ON(failed && retried)) {
3154 		ret = -ENOSPC;
3155 	}
3156 error:
3157 	btrfs_free_path(path);
3158 	return ret;
3159 }
3160 
3161 /*
3162  * return 1 : allocate a data chunk successfully,
3163  * return <0: errors during allocating a data chunk,
3164  * return 0 : no need to allocate a data chunk.
3165  */
3166 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3167 				      u64 chunk_offset)
3168 {
3169 	struct btrfs_block_group *cache;
3170 	u64 bytes_used;
3171 	u64 chunk_type;
3172 
3173 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3174 	ASSERT(cache);
3175 	chunk_type = cache->flags;
3176 	btrfs_put_block_group(cache);
3177 
3178 	if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3179 		return 0;
3180 
3181 	spin_lock(&fs_info->data_sinfo->lock);
3182 	bytes_used = fs_info->data_sinfo->bytes_used;
3183 	spin_unlock(&fs_info->data_sinfo->lock);
3184 
3185 	if (!bytes_used) {
3186 		struct btrfs_trans_handle *trans;
3187 		int ret;
3188 
3189 		trans =	btrfs_join_transaction(fs_info->tree_root);
3190 		if (IS_ERR(trans))
3191 			return PTR_ERR(trans);
3192 
3193 		ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3194 		btrfs_end_transaction(trans);
3195 		if (ret < 0)
3196 			return ret;
3197 		return 1;
3198 	}
3199 
3200 	return 0;
3201 }
3202 
3203 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3204 			       struct btrfs_balance_control *bctl)
3205 {
3206 	struct btrfs_root *root = fs_info->tree_root;
3207 	struct btrfs_trans_handle *trans;
3208 	struct btrfs_balance_item *item;
3209 	struct btrfs_disk_balance_args disk_bargs;
3210 	struct btrfs_path *path;
3211 	struct extent_buffer *leaf;
3212 	struct btrfs_key key;
3213 	int ret, err;
3214 
3215 	path = btrfs_alloc_path();
3216 	if (!path)
3217 		return -ENOMEM;
3218 
3219 	trans = btrfs_start_transaction(root, 0);
3220 	if (IS_ERR(trans)) {
3221 		btrfs_free_path(path);
3222 		return PTR_ERR(trans);
3223 	}
3224 
3225 	key.objectid = BTRFS_BALANCE_OBJECTID;
3226 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3227 	key.offset = 0;
3228 
3229 	ret = btrfs_insert_empty_item(trans, root, path, &key,
3230 				      sizeof(*item));
3231 	if (ret)
3232 		goto out;
3233 
3234 	leaf = path->nodes[0];
3235 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3236 
3237 	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3238 
3239 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3240 	btrfs_set_balance_data(leaf, item, &disk_bargs);
3241 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3242 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
3243 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3244 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
3245 
3246 	btrfs_set_balance_flags(leaf, item, bctl->flags);
3247 
3248 	btrfs_mark_buffer_dirty(leaf);
3249 out:
3250 	btrfs_free_path(path);
3251 	err = btrfs_commit_transaction(trans);
3252 	if (err && !ret)
3253 		ret = err;
3254 	return ret;
3255 }
3256 
3257 static int del_balance_item(struct btrfs_fs_info *fs_info)
3258 {
3259 	struct btrfs_root *root = fs_info->tree_root;
3260 	struct btrfs_trans_handle *trans;
3261 	struct btrfs_path *path;
3262 	struct btrfs_key key;
3263 	int ret, err;
3264 
3265 	path = btrfs_alloc_path();
3266 	if (!path)
3267 		return -ENOMEM;
3268 
3269 	trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3270 	if (IS_ERR(trans)) {
3271 		btrfs_free_path(path);
3272 		return PTR_ERR(trans);
3273 	}
3274 
3275 	key.objectid = BTRFS_BALANCE_OBJECTID;
3276 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3277 	key.offset = 0;
3278 
3279 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3280 	if (ret < 0)
3281 		goto out;
3282 	if (ret > 0) {
3283 		ret = -ENOENT;
3284 		goto out;
3285 	}
3286 
3287 	ret = btrfs_del_item(trans, root, path);
3288 out:
3289 	btrfs_free_path(path);
3290 	err = btrfs_commit_transaction(trans);
3291 	if (err && !ret)
3292 		ret = err;
3293 	return ret;
3294 }
3295 
3296 /*
3297  * This is a heuristic used to reduce the number of chunks balanced on
3298  * resume after balance was interrupted.
3299  */
3300 static void update_balance_args(struct btrfs_balance_control *bctl)
3301 {
3302 	/*
3303 	 * Turn on soft mode for chunk types that were being converted.
3304 	 */
3305 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3306 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3307 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3308 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3309 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3310 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3311 
3312 	/*
3313 	 * Turn on usage filter if is not already used.  The idea is
3314 	 * that chunks that we have already balanced should be
3315 	 * reasonably full.  Don't do it for chunks that are being
3316 	 * converted - that will keep us from relocating unconverted
3317 	 * (albeit full) chunks.
3318 	 */
3319 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3320 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3321 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3322 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3323 		bctl->data.usage = 90;
3324 	}
3325 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3326 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3327 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3328 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3329 		bctl->sys.usage = 90;
3330 	}
3331 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3332 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3333 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3334 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3335 		bctl->meta.usage = 90;
3336 	}
3337 }
3338 
3339 /*
3340  * Clear the balance status in fs_info and delete the balance item from disk.
3341  */
3342 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3343 {
3344 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3345 	int ret;
3346 
3347 	BUG_ON(!fs_info->balance_ctl);
3348 
3349 	spin_lock(&fs_info->balance_lock);
3350 	fs_info->balance_ctl = NULL;
3351 	spin_unlock(&fs_info->balance_lock);
3352 
3353 	kfree(bctl);
3354 	ret = del_balance_item(fs_info);
3355 	if (ret)
3356 		btrfs_handle_fs_error(fs_info, ret, NULL);
3357 }
3358 
3359 /*
3360  * Balance filters.  Return 1 if chunk should be filtered out
3361  * (should not be balanced).
3362  */
3363 static int chunk_profiles_filter(u64 chunk_type,
3364 				 struct btrfs_balance_args *bargs)
3365 {
3366 	chunk_type = chunk_to_extended(chunk_type) &
3367 				BTRFS_EXTENDED_PROFILE_MASK;
3368 
3369 	if (bargs->profiles & chunk_type)
3370 		return 0;
3371 
3372 	return 1;
3373 }
3374 
3375 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3376 			      struct btrfs_balance_args *bargs)
3377 {
3378 	struct btrfs_block_group *cache;
3379 	u64 chunk_used;
3380 	u64 user_thresh_min;
3381 	u64 user_thresh_max;
3382 	int ret = 1;
3383 
3384 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3385 	chunk_used = cache->used;
3386 
3387 	if (bargs->usage_min == 0)
3388 		user_thresh_min = 0;
3389 	else
3390 		user_thresh_min = div_factor_fine(cache->length,
3391 						  bargs->usage_min);
3392 
3393 	if (bargs->usage_max == 0)
3394 		user_thresh_max = 1;
3395 	else if (bargs->usage_max > 100)
3396 		user_thresh_max = cache->length;
3397 	else
3398 		user_thresh_max = div_factor_fine(cache->length,
3399 						  bargs->usage_max);
3400 
3401 	if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3402 		ret = 0;
3403 
3404 	btrfs_put_block_group(cache);
3405 	return ret;
3406 }
3407 
3408 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3409 		u64 chunk_offset, struct btrfs_balance_args *bargs)
3410 {
3411 	struct btrfs_block_group *cache;
3412 	u64 chunk_used, user_thresh;
3413 	int ret = 1;
3414 
3415 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3416 	chunk_used = cache->used;
3417 
3418 	if (bargs->usage_min == 0)
3419 		user_thresh = 1;
3420 	else if (bargs->usage > 100)
3421 		user_thresh = cache->length;
3422 	else
3423 		user_thresh = div_factor_fine(cache->length, bargs->usage);
3424 
3425 	if (chunk_used < user_thresh)
3426 		ret = 0;
3427 
3428 	btrfs_put_block_group(cache);
3429 	return ret;
3430 }
3431 
3432 static int chunk_devid_filter(struct extent_buffer *leaf,
3433 			      struct btrfs_chunk *chunk,
3434 			      struct btrfs_balance_args *bargs)
3435 {
3436 	struct btrfs_stripe *stripe;
3437 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3438 	int i;
3439 
3440 	for (i = 0; i < num_stripes; i++) {
3441 		stripe = btrfs_stripe_nr(chunk, i);
3442 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3443 			return 0;
3444 	}
3445 
3446 	return 1;
3447 }
3448 
3449 static u64 calc_data_stripes(u64 type, int num_stripes)
3450 {
3451 	const int index = btrfs_bg_flags_to_raid_index(type);
3452 	const int ncopies = btrfs_raid_array[index].ncopies;
3453 	const int nparity = btrfs_raid_array[index].nparity;
3454 
3455 	if (nparity)
3456 		return num_stripes - nparity;
3457 	else
3458 		return num_stripes / ncopies;
3459 }
3460 
3461 /* [pstart, pend) */
3462 static int chunk_drange_filter(struct extent_buffer *leaf,
3463 			       struct btrfs_chunk *chunk,
3464 			       struct btrfs_balance_args *bargs)
3465 {
3466 	struct btrfs_stripe *stripe;
3467 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3468 	u64 stripe_offset;
3469 	u64 stripe_length;
3470 	u64 type;
3471 	int factor;
3472 	int i;
3473 
3474 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3475 		return 0;
3476 
3477 	type = btrfs_chunk_type(leaf, chunk);
3478 	factor = calc_data_stripes(type, num_stripes);
3479 
3480 	for (i = 0; i < num_stripes; i++) {
3481 		stripe = btrfs_stripe_nr(chunk, i);
3482 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3483 			continue;
3484 
3485 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
3486 		stripe_length = btrfs_chunk_length(leaf, chunk);
3487 		stripe_length = div_u64(stripe_length, factor);
3488 
3489 		if (stripe_offset < bargs->pend &&
3490 		    stripe_offset + stripe_length > bargs->pstart)
3491 			return 0;
3492 	}
3493 
3494 	return 1;
3495 }
3496 
3497 /* [vstart, vend) */
3498 static int chunk_vrange_filter(struct extent_buffer *leaf,
3499 			       struct btrfs_chunk *chunk,
3500 			       u64 chunk_offset,
3501 			       struct btrfs_balance_args *bargs)
3502 {
3503 	if (chunk_offset < bargs->vend &&
3504 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3505 		/* at least part of the chunk is inside this vrange */
3506 		return 0;
3507 
3508 	return 1;
3509 }
3510 
3511 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3512 			       struct btrfs_chunk *chunk,
3513 			       struct btrfs_balance_args *bargs)
3514 {
3515 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3516 
3517 	if (bargs->stripes_min <= num_stripes
3518 			&& num_stripes <= bargs->stripes_max)
3519 		return 0;
3520 
3521 	return 1;
3522 }
3523 
3524 static int chunk_soft_convert_filter(u64 chunk_type,
3525 				     struct btrfs_balance_args *bargs)
3526 {
3527 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3528 		return 0;
3529 
3530 	chunk_type = chunk_to_extended(chunk_type) &
3531 				BTRFS_EXTENDED_PROFILE_MASK;
3532 
3533 	if (bargs->target == chunk_type)
3534 		return 1;
3535 
3536 	return 0;
3537 }
3538 
3539 static int should_balance_chunk(struct extent_buffer *leaf,
3540 				struct btrfs_chunk *chunk, u64 chunk_offset)
3541 {
3542 	struct btrfs_fs_info *fs_info = leaf->fs_info;
3543 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3544 	struct btrfs_balance_args *bargs = NULL;
3545 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3546 
3547 	/* type filter */
3548 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3549 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3550 		return 0;
3551 	}
3552 
3553 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3554 		bargs = &bctl->data;
3555 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3556 		bargs = &bctl->sys;
3557 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3558 		bargs = &bctl->meta;
3559 
3560 	/* profiles filter */
3561 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3562 	    chunk_profiles_filter(chunk_type, bargs)) {
3563 		return 0;
3564 	}
3565 
3566 	/* usage filter */
3567 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3568 	    chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3569 		return 0;
3570 	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3571 	    chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3572 		return 0;
3573 	}
3574 
3575 	/* devid filter */
3576 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3577 	    chunk_devid_filter(leaf, chunk, bargs)) {
3578 		return 0;
3579 	}
3580 
3581 	/* drange filter, makes sense only with devid filter */
3582 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3583 	    chunk_drange_filter(leaf, chunk, bargs)) {
3584 		return 0;
3585 	}
3586 
3587 	/* vrange filter */
3588 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3589 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3590 		return 0;
3591 	}
3592 
3593 	/* stripes filter */
3594 	if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3595 	    chunk_stripes_range_filter(leaf, chunk, bargs)) {
3596 		return 0;
3597 	}
3598 
3599 	/* soft profile changing mode */
3600 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3601 	    chunk_soft_convert_filter(chunk_type, bargs)) {
3602 		return 0;
3603 	}
3604 
3605 	/*
3606 	 * limited by count, must be the last filter
3607 	 */
3608 	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3609 		if (bargs->limit == 0)
3610 			return 0;
3611 		else
3612 			bargs->limit--;
3613 	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3614 		/*
3615 		 * Same logic as the 'limit' filter; the minimum cannot be
3616 		 * determined here because we do not have the global information
3617 		 * about the count of all chunks that satisfy the filters.
3618 		 */
3619 		if (bargs->limit_max == 0)
3620 			return 0;
3621 		else
3622 			bargs->limit_max--;
3623 	}
3624 
3625 	return 1;
3626 }
3627 
3628 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3629 {
3630 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3631 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3632 	u64 chunk_type;
3633 	struct btrfs_chunk *chunk;
3634 	struct btrfs_path *path = NULL;
3635 	struct btrfs_key key;
3636 	struct btrfs_key found_key;
3637 	struct extent_buffer *leaf;
3638 	int slot;
3639 	int ret;
3640 	int enospc_errors = 0;
3641 	bool counting = true;
3642 	/* The single value limit and min/max limits use the same bytes in the */
3643 	u64 limit_data = bctl->data.limit;
3644 	u64 limit_meta = bctl->meta.limit;
3645 	u64 limit_sys = bctl->sys.limit;
3646 	u32 count_data = 0;
3647 	u32 count_meta = 0;
3648 	u32 count_sys = 0;
3649 	int chunk_reserved = 0;
3650 
3651 	path = btrfs_alloc_path();
3652 	if (!path) {
3653 		ret = -ENOMEM;
3654 		goto error;
3655 	}
3656 
3657 	/* zero out stat counters */
3658 	spin_lock(&fs_info->balance_lock);
3659 	memset(&bctl->stat, 0, sizeof(bctl->stat));
3660 	spin_unlock(&fs_info->balance_lock);
3661 again:
3662 	if (!counting) {
3663 		/*
3664 		 * The single value limit and min/max limits use the same bytes
3665 		 * in the
3666 		 */
3667 		bctl->data.limit = limit_data;
3668 		bctl->meta.limit = limit_meta;
3669 		bctl->sys.limit = limit_sys;
3670 	}
3671 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3672 	key.offset = (u64)-1;
3673 	key.type = BTRFS_CHUNK_ITEM_KEY;
3674 
3675 	while (1) {
3676 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3677 		    atomic_read(&fs_info->balance_cancel_req)) {
3678 			ret = -ECANCELED;
3679 			goto error;
3680 		}
3681 
3682 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
3683 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3684 		if (ret < 0) {
3685 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3686 			goto error;
3687 		}
3688 
3689 		/*
3690 		 * this shouldn't happen, it means the last relocate
3691 		 * failed
3692 		 */
3693 		if (ret == 0)
3694 			BUG(); /* FIXME break ? */
3695 
3696 		ret = btrfs_previous_item(chunk_root, path, 0,
3697 					  BTRFS_CHUNK_ITEM_KEY);
3698 		if (ret) {
3699 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3700 			ret = 0;
3701 			break;
3702 		}
3703 
3704 		leaf = path->nodes[0];
3705 		slot = path->slots[0];
3706 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3707 
3708 		if (found_key.objectid != key.objectid) {
3709 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3710 			break;
3711 		}
3712 
3713 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3714 		chunk_type = btrfs_chunk_type(leaf, chunk);
3715 
3716 		if (!counting) {
3717 			spin_lock(&fs_info->balance_lock);
3718 			bctl->stat.considered++;
3719 			spin_unlock(&fs_info->balance_lock);
3720 		}
3721 
3722 		ret = should_balance_chunk(leaf, chunk, found_key.offset);
3723 
3724 		btrfs_release_path(path);
3725 		if (!ret) {
3726 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3727 			goto loop;
3728 		}
3729 
3730 		if (counting) {
3731 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3732 			spin_lock(&fs_info->balance_lock);
3733 			bctl->stat.expected++;
3734 			spin_unlock(&fs_info->balance_lock);
3735 
3736 			if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3737 				count_data++;
3738 			else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3739 				count_sys++;
3740 			else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3741 				count_meta++;
3742 
3743 			goto loop;
3744 		}
3745 
3746 		/*
3747 		 * Apply limit_min filter, no need to check if the LIMITS
3748 		 * filter is used, limit_min is 0 by default
3749 		 */
3750 		if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3751 					count_data < bctl->data.limit_min)
3752 				|| ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3753 					count_meta < bctl->meta.limit_min)
3754 				|| ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3755 					count_sys < bctl->sys.limit_min)) {
3756 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3757 			goto loop;
3758 		}
3759 
3760 		if (!chunk_reserved) {
3761 			/*
3762 			 * We may be relocating the only data chunk we have,
3763 			 * which could potentially end up with losing data's
3764 			 * raid profile, so lets allocate an empty one in
3765 			 * advance.
3766 			 */
3767 			ret = btrfs_may_alloc_data_chunk(fs_info,
3768 							 found_key.offset);
3769 			if (ret < 0) {
3770 				mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3771 				goto error;
3772 			} else if (ret == 1) {
3773 				chunk_reserved = 1;
3774 			}
3775 		}
3776 
3777 		ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3778 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3779 		if (ret == -ENOSPC) {
3780 			enospc_errors++;
3781 		} else if (ret == -ETXTBSY) {
3782 			btrfs_info(fs_info,
3783 	   "skipping relocation of block group %llu due to active swapfile",
3784 				   found_key.offset);
3785 			ret = 0;
3786 		} else if (ret) {
3787 			goto error;
3788 		} else {
3789 			spin_lock(&fs_info->balance_lock);
3790 			bctl->stat.completed++;
3791 			spin_unlock(&fs_info->balance_lock);
3792 		}
3793 loop:
3794 		if (found_key.offset == 0)
3795 			break;
3796 		key.offset = found_key.offset - 1;
3797 	}
3798 
3799 	if (counting) {
3800 		btrfs_release_path(path);
3801 		counting = false;
3802 		goto again;
3803 	}
3804 error:
3805 	btrfs_free_path(path);
3806 	if (enospc_errors) {
3807 		btrfs_info(fs_info, "%d enospc errors during balance",
3808 			   enospc_errors);
3809 		if (!ret)
3810 			ret = -ENOSPC;
3811 	}
3812 
3813 	return ret;
3814 }
3815 
3816 /**
3817  * alloc_profile_is_valid - see if a given profile is valid and reduced
3818  * @flags: profile to validate
3819  * @extended: if true @flags is treated as an extended profile
3820  */
3821 static int alloc_profile_is_valid(u64 flags, int extended)
3822 {
3823 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3824 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
3825 
3826 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3827 
3828 	/* 1) check that all other bits are zeroed */
3829 	if (flags & ~mask)
3830 		return 0;
3831 
3832 	/* 2) see if profile is reduced */
3833 	if (flags == 0)
3834 		return !extended; /* "0" is valid for usual profiles */
3835 
3836 	return has_single_bit_set(flags);
3837 }
3838 
3839 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3840 {
3841 	/* cancel requested || normal exit path */
3842 	return atomic_read(&fs_info->balance_cancel_req) ||
3843 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
3844 		 atomic_read(&fs_info->balance_cancel_req) == 0);
3845 }
3846 
3847 /*
3848  * Validate target profile against allowed profiles and return true if it's OK.
3849  * Otherwise print the error message and return false.
3850  */
3851 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3852 		const struct btrfs_balance_args *bargs,
3853 		u64 allowed, const char *type)
3854 {
3855 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3856 		return true;
3857 
3858 	/* Profile is valid and does not have bits outside of the allowed set */
3859 	if (alloc_profile_is_valid(bargs->target, 1) &&
3860 	    (bargs->target & ~allowed) == 0)
3861 		return true;
3862 
3863 	btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3864 			type, btrfs_bg_type_to_raid_name(bargs->target));
3865 	return false;
3866 }
3867 
3868 /*
3869  * Fill @buf with textual description of balance filter flags @bargs, up to
3870  * @size_buf including the terminating null. The output may be trimmed if it
3871  * does not fit into the provided buffer.
3872  */
3873 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3874 				 u32 size_buf)
3875 {
3876 	int ret;
3877 	u32 size_bp = size_buf;
3878 	char *bp = buf;
3879 	u64 flags = bargs->flags;
3880 	char tmp_buf[128] = {'\0'};
3881 
3882 	if (!flags)
3883 		return;
3884 
3885 #define CHECK_APPEND_NOARG(a)						\
3886 	do {								\
3887 		ret = snprintf(bp, size_bp, (a));			\
3888 		if (ret < 0 || ret >= size_bp)				\
3889 			goto out_overflow;				\
3890 		size_bp -= ret;						\
3891 		bp += ret;						\
3892 	} while (0)
3893 
3894 #define CHECK_APPEND_1ARG(a, v1)					\
3895 	do {								\
3896 		ret = snprintf(bp, size_bp, (a), (v1));			\
3897 		if (ret < 0 || ret >= size_bp)				\
3898 			goto out_overflow;				\
3899 		size_bp -= ret;						\
3900 		bp += ret;						\
3901 	} while (0)
3902 
3903 #define CHECK_APPEND_2ARG(a, v1, v2)					\
3904 	do {								\
3905 		ret = snprintf(bp, size_bp, (a), (v1), (v2));		\
3906 		if (ret < 0 || ret >= size_bp)				\
3907 			goto out_overflow;				\
3908 		size_bp -= ret;						\
3909 		bp += ret;						\
3910 	} while (0)
3911 
3912 	if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3913 		CHECK_APPEND_1ARG("convert=%s,",
3914 				  btrfs_bg_type_to_raid_name(bargs->target));
3915 
3916 	if (flags & BTRFS_BALANCE_ARGS_SOFT)
3917 		CHECK_APPEND_NOARG("soft,");
3918 
3919 	if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3920 		btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3921 					    sizeof(tmp_buf));
3922 		CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3923 	}
3924 
3925 	if (flags & BTRFS_BALANCE_ARGS_USAGE)
3926 		CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3927 
3928 	if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3929 		CHECK_APPEND_2ARG("usage=%u..%u,",
3930 				  bargs->usage_min, bargs->usage_max);
3931 
3932 	if (flags & BTRFS_BALANCE_ARGS_DEVID)
3933 		CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3934 
3935 	if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3936 		CHECK_APPEND_2ARG("drange=%llu..%llu,",
3937 				  bargs->pstart, bargs->pend);
3938 
3939 	if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3940 		CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3941 				  bargs->vstart, bargs->vend);
3942 
3943 	if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3944 		CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3945 
3946 	if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3947 		CHECK_APPEND_2ARG("limit=%u..%u,",
3948 				bargs->limit_min, bargs->limit_max);
3949 
3950 	if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3951 		CHECK_APPEND_2ARG("stripes=%u..%u,",
3952 				  bargs->stripes_min, bargs->stripes_max);
3953 
3954 #undef CHECK_APPEND_2ARG
3955 #undef CHECK_APPEND_1ARG
3956 #undef CHECK_APPEND_NOARG
3957 
3958 out_overflow:
3959 
3960 	if (size_bp < size_buf)
3961 		buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3962 	else
3963 		buf[0] = '\0';
3964 }
3965 
3966 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3967 {
3968 	u32 size_buf = 1024;
3969 	char tmp_buf[192] = {'\0'};
3970 	char *buf;
3971 	char *bp;
3972 	u32 size_bp = size_buf;
3973 	int ret;
3974 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3975 
3976 	buf = kzalloc(size_buf, GFP_KERNEL);
3977 	if (!buf)
3978 		return;
3979 
3980 	bp = buf;
3981 
3982 #define CHECK_APPEND_1ARG(a, v1)					\
3983 	do {								\
3984 		ret = snprintf(bp, size_bp, (a), (v1));			\
3985 		if (ret < 0 || ret >= size_bp)				\
3986 			goto out_overflow;				\
3987 		size_bp -= ret;						\
3988 		bp += ret;						\
3989 	} while (0)
3990 
3991 	if (bctl->flags & BTRFS_BALANCE_FORCE)
3992 		CHECK_APPEND_1ARG("%s", "-f ");
3993 
3994 	if (bctl->flags & BTRFS_BALANCE_DATA) {
3995 		describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3996 		CHECK_APPEND_1ARG("-d%s ", tmp_buf);
3997 	}
3998 
3999 	if (bctl->flags & BTRFS_BALANCE_METADATA) {
4000 		describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4001 		CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4002 	}
4003 
4004 	if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4005 		describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4006 		CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4007 	}
4008 
4009 #undef CHECK_APPEND_1ARG
4010 
4011 out_overflow:
4012 
4013 	if (size_bp < size_buf)
4014 		buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4015 	btrfs_info(fs_info, "balance: %s %s",
4016 		   (bctl->flags & BTRFS_BALANCE_RESUME) ?
4017 		   "resume" : "start", buf);
4018 
4019 	kfree(buf);
4020 }
4021 
4022 /*
4023  * Should be called with balance mutexe held
4024  */
4025 int btrfs_balance(struct btrfs_fs_info *fs_info,
4026 		  struct btrfs_balance_control *bctl,
4027 		  struct btrfs_ioctl_balance_args *bargs)
4028 {
4029 	u64 meta_target, data_target;
4030 	u64 allowed;
4031 	int mixed = 0;
4032 	int ret;
4033 	u64 num_devices;
4034 	unsigned seq;
4035 	bool reducing_redundancy;
4036 	int i;
4037 
4038 	if (btrfs_fs_closing(fs_info) ||
4039 	    atomic_read(&fs_info->balance_pause_req) ||
4040 	    btrfs_should_cancel_balance(fs_info)) {
4041 		ret = -EINVAL;
4042 		goto out;
4043 	}
4044 
4045 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4046 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4047 		mixed = 1;
4048 
4049 	/*
4050 	 * In case of mixed groups both data and meta should be picked,
4051 	 * and identical options should be given for both of them.
4052 	 */
4053 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4054 	if (mixed && (bctl->flags & allowed)) {
4055 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4056 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4057 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4058 			btrfs_err(fs_info,
4059 	  "balance: mixed groups data and metadata options must be the same");
4060 			ret = -EINVAL;
4061 			goto out;
4062 		}
4063 	}
4064 
4065 	/*
4066 	 * rw_devices will not change at the moment, device add/delete/replace
4067 	 * are exclusive
4068 	 */
4069 	num_devices = fs_info->fs_devices->rw_devices;
4070 
4071 	/*
4072 	 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4073 	 * special bit for it, to make it easier to distinguish.  Thus we need
4074 	 * to set it manually, or balance would refuse the profile.
4075 	 */
4076 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4077 	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4078 		if (num_devices >= btrfs_raid_array[i].devs_min)
4079 			allowed |= btrfs_raid_array[i].bg_flag;
4080 
4081 	if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4082 	    !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4083 	    !validate_convert_profile(fs_info, &bctl->sys,  allowed, "system")) {
4084 		ret = -EINVAL;
4085 		goto out;
4086 	}
4087 
4088 	/*
4089 	 * Allow to reduce metadata or system integrity only if force set for
4090 	 * profiles with redundancy (copies, parity)
4091 	 */
4092 	allowed = 0;
4093 	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4094 		if (btrfs_raid_array[i].ncopies >= 2 ||
4095 		    btrfs_raid_array[i].tolerated_failures >= 1)
4096 			allowed |= btrfs_raid_array[i].bg_flag;
4097 	}
4098 	do {
4099 		seq = read_seqbegin(&fs_info->profiles_lock);
4100 
4101 		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4102 		     (fs_info->avail_system_alloc_bits & allowed) &&
4103 		     !(bctl->sys.target & allowed)) ||
4104 		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4105 		     (fs_info->avail_metadata_alloc_bits & allowed) &&
4106 		     !(bctl->meta.target & allowed)))
4107 			reducing_redundancy = true;
4108 		else
4109 			reducing_redundancy = false;
4110 
4111 		/* if we're not converting, the target field is uninitialized */
4112 		meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4113 			bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4114 		data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4115 			bctl->data.target : fs_info->avail_data_alloc_bits;
4116 	} while (read_seqretry(&fs_info->profiles_lock, seq));
4117 
4118 	if (reducing_redundancy) {
4119 		if (bctl->flags & BTRFS_BALANCE_FORCE) {
4120 			btrfs_info(fs_info,
4121 			   "balance: force reducing metadata redundancy");
4122 		} else {
4123 			btrfs_err(fs_info,
4124 	"balance: reduces metadata redundancy, use --force if you want this");
4125 			ret = -EINVAL;
4126 			goto out;
4127 		}
4128 	}
4129 
4130 	if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4131 		btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4132 		btrfs_warn(fs_info,
4133 	"balance: metadata profile %s has lower redundancy than data profile %s",
4134 				btrfs_bg_type_to_raid_name(meta_target),
4135 				btrfs_bg_type_to_raid_name(data_target));
4136 	}
4137 
4138 	if (fs_info->send_in_progress) {
4139 		btrfs_warn_rl(fs_info,
4140 "cannot run balance while send operations are in progress (%d in progress)",
4141 			      fs_info->send_in_progress);
4142 		ret = -EAGAIN;
4143 		goto out;
4144 	}
4145 
4146 	ret = insert_balance_item(fs_info, bctl);
4147 	if (ret && ret != -EEXIST)
4148 		goto out;
4149 
4150 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4151 		BUG_ON(ret == -EEXIST);
4152 		BUG_ON(fs_info->balance_ctl);
4153 		spin_lock(&fs_info->balance_lock);
4154 		fs_info->balance_ctl = bctl;
4155 		spin_unlock(&fs_info->balance_lock);
4156 	} else {
4157 		BUG_ON(ret != -EEXIST);
4158 		spin_lock(&fs_info->balance_lock);
4159 		update_balance_args(bctl);
4160 		spin_unlock(&fs_info->balance_lock);
4161 	}
4162 
4163 	ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4164 	set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4165 	describe_balance_start_or_resume(fs_info);
4166 	mutex_unlock(&fs_info->balance_mutex);
4167 
4168 	ret = __btrfs_balance(fs_info);
4169 
4170 	mutex_lock(&fs_info->balance_mutex);
4171 	if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4172 		btrfs_info(fs_info, "balance: paused");
4173 	/*
4174 	 * Balance can be canceled by:
4175 	 *
4176 	 * - Regular cancel request
4177 	 *   Then ret == -ECANCELED and balance_cancel_req > 0
4178 	 *
4179 	 * - Fatal signal to "btrfs" process
4180 	 *   Either the signal caught by wait_reserve_ticket() and callers
4181 	 *   got -EINTR, or caught by btrfs_should_cancel_balance() and
4182 	 *   got -ECANCELED.
4183 	 *   Either way, in this case balance_cancel_req = 0, and
4184 	 *   ret == -EINTR or ret == -ECANCELED.
4185 	 *
4186 	 * So here we only check the return value to catch canceled balance.
4187 	 */
4188 	else if (ret == -ECANCELED || ret == -EINTR)
4189 		btrfs_info(fs_info, "balance: canceled");
4190 	else
4191 		btrfs_info(fs_info, "balance: ended with status: %d", ret);
4192 
4193 	clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4194 
4195 	if (bargs) {
4196 		memset(bargs, 0, sizeof(*bargs));
4197 		btrfs_update_ioctl_balance_args(fs_info, bargs);
4198 	}
4199 
4200 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4201 	    balance_need_close(fs_info)) {
4202 		reset_balance_state(fs_info);
4203 		btrfs_exclop_finish(fs_info);
4204 	}
4205 
4206 	wake_up(&fs_info->balance_wait_q);
4207 
4208 	return ret;
4209 out:
4210 	if (bctl->flags & BTRFS_BALANCE_RESUME)
4211 		reset_balance_state(fs_info);
4212 	else
4213 		kfree(bctl);
4214 	btrfs_exclop_finish(fs_info);
4215 
4216 	return ret;
4217 }
4218 
4219 static int balance_kthread(void *data)
4220 {
4221 	struct btrfs_fs_info *fs_info = data;
4222 	int ret = 0;
4223 
4224 	mutex_lock(&fs_info->balance_mutex);
4225 	if (fs_info->balance_ctl)
4226 		ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4227 	mutex_unlock(&fs_info->balance_mutex);
4228 
4229 	return ret;
4230 }
4231 
4232 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4233 {
4234 	struct task_struct *tsk;
4235 
4236 	mutex_lock(&fs_info->balance_mutex);
4237 	if (!fs_info->balance_ctl) {
4238 		mutex_unlock(&fs_info->balance_mutex);
4239 		return 0;
4240 	}
4241 	mutex_unlock(&fs_info->balance_mutex);
4242 
4243 	if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4244 		btrfs_info(fs_info, "balance: resume skipped");
4245 		return 0;
4246 	}
4247 
4248 	/*
4249 	 * A ro->rw remount sequence should continue with the paused balance
4250 	 * regardless of who pauses it, system or the user as of now, so set
4251 	 * the resume flag.
4252 	 */
4253 	spin_lock(&fs_info->balance_lock);
4254 	fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4255 	spin_unlock(&fs_info->balance_lock);
4256 
4257 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4258 	return PTR_ERR_OR_ZERO(tsk);
4259 }
4260 
4261 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4262 {
4263 	struct btrfs_balance_control *bctl;
4264 	struct btrfs_balance_item *item;
4265 	struct btrfs_disk_balance_args disk_bargs;
4266 	struct btrfs_path *path;
4267 	struct extent_buffer *leaf;
4268 	struct btrfs_key key;
4269 	int ret;
4270 
4271 	path = btrfs_alloc_path();
4272 	if (!path)
4273 		return -ENOMEM;
4274 
4275 	key.objectid = BTRFS_BALANCE_OBJECTID;
4276 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
4277 	key.offset = 0;
4278 
4279 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4280 	if (ret < 0)
4281 		goto out;
4282 	if (ret > 0) { /* ret = -ENOENT; */
4283 		ret = 0;
4284 		goto out;
4285 	}
4286 
4287 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4288 	if (!bctl) {
4289 		ret = -ENOMEM;
4290 		goto out;
4291 	}
4292 
4293 	leaf = path->nodes[0];
4294 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4295 
4296 	bctl->flags = btrfs_balance_flags(leaf, item);
4297 	bctl->flags |= BTRFS_BALANCE_RESUME;
4298 
4299 	btrfs_balance_data(leaf, item, &disk_bargs);
4300 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4301 	btrfs_balance_meta(leaf, item, &disk_bargs);
4302 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4303 	btrfs_balance_sys(leaf, item, &disk_bargs);
4304 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4305 
4306 	/*
4307 	 * This should never happen, as the paused balance state is recovered
4308 	 * during mount without any chance of other exclusive ops to collide.
4309 	 *
4310 	 * This gives the exclusive op status to balance and keeps in paused
4311 	 * state until user intervention (cancel or umount). If the ownership
4312 	 * cannot be assigned, show a message but do not fail. The balance
4313 	 * is in a paused state and must have fs_info::balance_ctl properly
4314 	 * set up.
4315 	 */
4316 	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4317 		btrfs_warn(fs_info,
4318 	"balance: cannot set exclusive op status, resume manually");
4319 
4320 	mutex_lock(&fs_info->balance_mutex);
4321 	BUG_ON(fs_info->balance_ctl);
4322 	spin_lock(&fs_info->balance_lock);
4323 	fs_info->balance_ctl = bctl;
4324 	spin_unlock(&fs_info->balance_lock);
4325 	mutex_unlock(&fs_info->balance_mutex);
4326 out:
4327 	btrfs_free_path(path);
4328 	return ret;
4329 }
4330 
4331 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4332 {
4333 	int ret = 0;
4334 
4335 	mutex_lock(&fs_info->balance_mutex);
4336 	if (!fs_info->balance_ctl) {
4337 		mutex_unlock(&fs_info->balance_mutex);
4338 		return -ENOTCONN;
4339 	}
4340 
4341 	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4342 		atomic_inc(&fs_info->balance_pause_req);
4343 		mutex_unlock(&fs_info->balance_mutex);
4344 
4345 		wait_event(fs_info->balance_wait_q,
4346 			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4347 
4348 		mutex_lock(&fs_info->balance_mutex);
4349 		/* we are good with balance_ctl ripped off from under us */
4350 		BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4351 		atomic_dec(&fs_info->balance_pause_req);
4352 	} else {
4353 		ret = -ENOTCONN;
4354 	}
4355 
4356 	mutex_unlock(&fs_info->balance_mutex);
4357 	return ret;
4358 }
4359 
4360 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4361 {
4362 	mutex_lock(&fs_info->balance_mutex);
4363 	if (!fs_info->balance_ctl) {
4364 		mutex_unlock(&fs_info->balance_mutex);
4365 		return -ENOTCONN;
4366 	}
4367 
4368 	/*
4369 	 * A paused balance with the item stored on disk can be resumed at
4370 	 * mount time if the mount is read-write. Otherwise it's still paused
4371 	 * and we must not allow cancelling as it deletes the item.
4372 	 */
4373 	if (sb_rdonly(fs_info->sb)) {
4374 		mutex_unlock(&fs_info->balance_mutex);
4375 		return -EROFS;
4376 	}
4377 
4378 	atomic_inc(&fs_info->balance_cancel_req);
4379 	/*
4380 	 * if we are running just wait and return, balance item is
4381 	 * deleted in btrfs_balance in this case
4382 	 */
4383 	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4384 		mutex_unlock(&fs_info->balance_mutex);
4385 		wait_event(fs_info->balance_wait_q,
4386 			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4387 		mutex_lock(&fs_info->balance_mutex);
4388 	} else {
4389 		mutex_unlock(&fs_info->balance_mutex);
4390 		/*
4391 		 * Lock released to allow other waiters to continue, we'll
4392 		 * reexamine the status again.
4393 		 */
4394 		mutex_lock(&fs_info->balance_mutex);
4395 
4396 		if (fs_info->balance_ctl) {
4397 			reset_balance_state(fs_info);
4398 			btrfs_exclop_finish(fs_info);
4399 			btrfs_info(fs_info, "balance: canceled");
4400 		}
4401 	}
4402 
4403 	BUG_ON(fs_info->balance_ctl ||
4404 		test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4405 	atomic_dec(&fs_info->balance_cancel_req);
4406 	mutex_unlock(&fs_info->balance_mutex);
4407 	return 0;
4408 }
4409 
4410 int btrfs_uuid_scan_kthread(void *data)
4411 {
4412 	struct btrfs_fs_info *fs_info = data;
4413 	struct btrfs_root *root = fs_info->tree_root;
4414 	struct btrfs_key key;
4415 	struct btrfs_path *path = NULL;
4416 	int ret = 0;
4417 	struct extent_buffer *eb;
4418 	int slot;
4419 	struct btrfs_root_item root_item;
4420 	u32 item_size;
4421 	struct btrfs_trans_handle *trans = NULL;
4422 	bool closing = false;
4423 
4424 	path = btrfs_alloc_path();
4425 	if (!path) {
4426 		ret = -ENOMEM;
4427 		goto out;
4428 	}
4429 
4430 	key.objectid = 0;
4431 	key.type = BTRFS_ROOT_ITEM_KEY;
4432 	key.offset = 0;
4433 
4434 	while (1) {
4435 		if (btrfs_fs_closing(fs_info)) {
4436 			closing = true;
4437 			break;
4438 		}
4439 		ret = btrfs_search_forward(root, &key, path,
4440 				BTRFS_OLDEST_GENERATION);
4441 		if (ret) {
4442 			if (ret > 0)
4443 				ret = 0;
4444 			break;
4445 		}
4446 
4447 		if (key.type != BTRFS_ROOT_ITEM_KEY ||
4448 		    (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4449 		     key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4450 		    key.objectid > BTRFS_LAST_FREE_OBJECTID)
4451 			goto skip;
4452 
4453 		eb = path->nodes[0];
4454 		slot = path->slots[0];
4455 		item_size = btrfs_item_size_nr(eb, slot);
4456 		if (item_size < sizeof(root_item))
4457 			goto skip;
4458 
4459 		read_extent_buffer(eb, &root_item,
4460 				   btrfs_item_ptr_offset(eb, slot),
4461 				   (int)sizeof(root_item));
4462 		if (btrfs_root_refs(&root_item) == 0)
4463 			goto skip;
4464 
4465 		if (!btrfs_is_empty_uuid(root_item.uuid) ||
4466 		    !btrfs_is_empty_uuid(root_item.received_uuid)) {
4467 			if (trans)
4468 				goto update_tree;
4469 
4470 			btrfs_release_path(path);
4471 			/*
4472 			 * 1 - subvol uuid item
4473 			 * 1 - received_subvol uuid item
4474 			 */
4475 			trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4476 			if (IS_ERR(trans)) {
4477 				ret = PTR_ERR(trans);
4478 				break;
4479 			}
4480 			continue;
4481 		} else {
4482 			goto skip;
4483 		}
4484 update_tree:
4485 		btrfs_release_path(path);
4486 		if (!btrfs_is_empty_uuid(root_item.uuid)) {
4487 			ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4488 						  BTRFS_UUID_KEY_SUBVOL,
4489 						  key.objectid);
4490 			if (ret < 0) {
4491 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4492 					ret);
4493 				break;
4494 			}
4495 		}
4496 
4497 		if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4498 			ret = btrfs_uuid_tree_add(trans,
4499 						  root_item.received_uuid,
4500 						 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4501 						  key.objectid);
4502 			if (ret < 0) {
4503 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4504 					ret);
4505 				break;
4506 			}
4507 		}
4508 
4509 skip:
4510 		btrfs_release_path(path);
4511 		if (trans) {
4512 			ret = btrfs_end_transaction(trans);
4513 			trans = NULL;
4514 			if (ret)
4515 				break;
4516 		}
4517 
4518 		if (key.offset < (u64)-1) {
4519 			key.offset++;
4520 		} else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4521 			key.offset = 0;
4522 			key.type = BTRFS_ROOT_ITEM_KEY;
4523 		} else if (key.objectid < (u64)-1) {
4524 			key.offset = 0;
4525 			key.type = BTRFS_ROOT_ITEM_KEY;
4526 			key.objectid++;
4527 		} else {
4528 			break;
4529 		}
4530 		cond_resched();
4531 	}
4532 
4533 out:
4534 	btrfs_free_path(path);
4535 	if (trans && !IS_ERR(trans))
4536 		btrfs_end_transaction(trans);
4537 	if (ret)
4538 		btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4539 	else if (!closing)
4540 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4541 	up(&fs_info->uuid_tree_rescan_sem);
4542 	return 0;
4543 }
4544 
4545 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4546 {
4547 	struct btrfs_trans_handle *trans;
4548 	struct btrfs_root *tree_root = fs_info->tree_root;
4549 	struct btrfs_root *uuid_root;
4550 	struct task_struct *task;
4551 	int ret;
4552 
4553 	/*
4554 	 * 1 - root node
4555 	 * 1 - root item
4556 	 */
4557 	trans = btrfs_start_transaction(tree_root, 2);
4558 	if (IS_ERR(trans))
4559 		return PTR_ERR(trans);
4560 
4561 	uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4562 	if (IS_ERR(uuid_root)) {
4563 		ret = PTR_ERR(uuid_root);
4564 		btrfs_abort_transaction(trans, ret);
4565 		btrfs_end_transaction(trans);
4566 		return ret;
4567 	}
4568 
4569 	fs_info->uuid_root = uuid_root;
4570 
4571 	ret = btrfs_commit_transaction(trans);
4572 	if (ret)
4573 		return ret;
4574 
4575 	down(&fs_info->uuid_tree_rescan_sem);
4576 	task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4577 	if (IS_ERR(task)) {
4578 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
4579 		btrfs_warn(fs_info, "failed to start uuid_scan task");
4580 		up(&fs_info->uuid_tree_rescan_sem);
4581 		return PTR_ERR(task);
4582 	}
4583 
4584 	return 0;
4585 }
4586 
4587 /*
4588  * shrinking a device means finding all of the device extents past
4589  * the new size, and then following the back refs to the chunks.
4590  * The chunk relocation code actually frees the device extent
4591  */
4592 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4593 {
4594 	struct btrfs_fs_info *fs_info = device->fs_info;
4595 	struct btrfs_root *root = fs_info->dev_root;
4596 	struct btrfs_trans_handle *trans;
4597 	struct btrfs_dev_extent *dev_extent = NULL;
4598 	struct btrfs_path *path;
4599 	u64 length;
4600 	u64 chunk_offset;
4601 	int ret;
4602 	int slot;
4603 	int failed = 0;
4604 	bool retried = false;
4605 	struct extent_buffer *l;
4606 	struct btrfs_key key;
4607 	struct btrfs_super_block *super_copy = fs_info->super_copy;
4608 	u64 old_total = btrfs_super_total_bytes(super_copy);
4609 	u64 old_size = btrfs_device_get_total_bytes(device);
4610 	u64 diff;
4611 	u64 start;
4612 
4613 	new_size = round_down(new_size, fs_info->sectorsize);
4614 	start = new_size;
4615 	diff = round_down(old_size - new_size, fs_info->sectorsize);
4616 
4617 	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4618 		return -EINVAL;
4619 
4620 	path = btrfs_alloc_path();
4621 	if (!path)
4622 		return -ENOMEM;
4623 
4624 	path->reada = READA_BACK;
4625 
4626 	trans = btrfs_start_transaction(root, 0);
4627 	if (IS_ERR(trans)) {
4628 		btrfs_free_path(path);
4629 		return PTR_ERR(trans);
4630 	}
4631 
4632 	mutex_lock(&fs_info->chunk_mutex);
4633 
4634 	btrfs_device_set_total_bytes(device, new_size);
4635 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4636 		device->fs_devices->total_rw_bytes -= diff;
4637 		atomic64_sub(diff, &fs_info->free_chunk_space);
4638 	}
4639 
4640 	/*
4641 	 * Once the device's size has been set to the new size, ensure all
4642 	 * in-memory chunks are synced to disk so that the loop below sees them
4643 	 * and relocates them accordingly.
4644 	 */
4645 	if (contains_pending_extent(device, &start, diff)) {
4646 		mutex_unlock(&fs_info->chunk_mutex);
4647 		ret = btrfs_commit_transaction(trans);
4648 		if (ret)
4649 			goto done;
4650 	} else {
4651 		mutex_unlock(&fs_info->chunk_mutex);
4652 		btrfs_end_transaction(trans);
4653 	}
4654 
4655 again:
4656 	key.objectid = device->devid;
4657 	key.offset = (u64)-1;
4658 	key.type = BTRFS_DEV_EXTENT_KEY;
4659 
4660 	do {
4661 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
4662 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4663 		if (ret < 0) {
4664 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4665 			goto done;
4666 		}
4667 
4668 		ret = btrfs_previous_item(root, path, 0, key.type);
4669 		if (ret)
4670 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4671 		if (ret < 0)
4672 			goto done;
4673 		if (ret) {
4674 			ret = 0;
4675 			btrfs_release_path(path);
4676 			break;
4677 		}
4678 
4679 		l = path->nodes[0];
4680 		slot = path->slots[0];
4681 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4682 
4683 		if (key.objectid != device->devid) {
4684 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4685 			btrfs_release_path(path);
4686 			break;
4687 		}
4688 
4689 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4690 		length = btrfs_dev_extent_length(l, dev_extent);
4691 
4692 		if (key.offset + length <= new_size) {
4693 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4694 			btrfs_release_path(path);
4695 			break;
4696 		}
4697 
4698 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4699 		btrfs_release_path(path);
4700 
4701 		/*
4702 		 * We may be relocating the only data chunk we have,
4703 		 * which could potentially end up with losing data's
4704 		 * raid profile, so lets allocate an empty one in
4705 		 * advance.
4706 		 */
4707 		ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4708 		if (ret < 0) {
4709 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4710 			goto done;
4711 		}
4712 
4713 		ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4714 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4715 		if (ret == -ENOSPC) {
4716 			failed++;
4717 		} else if (ret) {
4718 			if (ret == -ETXTBSY) {
4719 				btrfs_warn(fs_info,
4720 		   "could not shrink block group %llu due to active swapfile",
4721 					   chunk_offset);
4722 			}
4723 			goto done;
4724 		}
4725 	} while (key.offset-- > 0);
4726 
4727 	if (failed && !retried) {
4728 		failed = 0;
4729 		retried = true;
4730 		goto again;
4731 	} else if (failed && retried) {
4732 		ret = -ENOSPC;
4733 		goto done;
4734 	}
4735 
4736 	/* Shrinking succeeded, else we would be at "done". */
4737 	trans = btrfs_start_transaction(root, 0);
4738 	if (IS_ERR(trans)) {
4739 		ret = PTR_ERR(trans);
4740 		goto done;
4741 	}
4742 
4743 	mutex_lock(&fs_info->chunk_mutex);
4744 	/* Clear all state bits beyond the shrunk device size */
4745 	clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4746 			  CHUNK_STATE_MASK);
4747 
4748 	btrfs_device_set_disk_total_bytes(device, new_size);
4749 	if (list_empty(&device->post_commit_list))
4750 		list_add_tail(&device->post_commit_list,
4751 			      &trans->transaction->dev_update_list);
4752 
4753 	WARN_ON(diff > old_total);
4754 	btrfs_set_super_total_bytes(super_copy,
4755 			round_down(old_total - diff, fs_info->sectorsize));
4756 	mutex_unlock(&fs_info->chunk_mutex);
4757 
4758 	/* Now btrfs_update_device() will change the on-disk size. */
4759 	ret = btrfs_update_device(trans, device);
4760 	if (ret < 0) {
4761 		btrfs_abort_transaction(trans, ret);
4762 		btrfs_end_transaction(trans);
4763 	} else {
4764 		ret = btrfs_commit_transaction(trans);
4765 	}
4766 done:
4767 	btrfs_free_path(path);
4768 	if (ret) {
4769 		mutex_lock(&fs_info->chunk_mutex);
4770 		btrfs_device_set_total_bytes(device, old_size);
4771 		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4772 			device->fs_devices->total_rw_bytes += diff;
4773 		atomic64_add(diff, &fs_info->free_chunk_space);
4774 		mutex_unlock(&fs_info->chunk_mutex);
4775 	}
4776 	return ret;
4777 }
4778 
4779 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4780 			   struct btrfs_key *key,
4781 			   struct btrfs_chunk *chunk, int item_size)
4782 {
4783 	struct btrfs_super_block *super_copy = fs_info->super_copy;
4784 	struct btrfs_disk_key disk_key;
4785 	u32 array_size;
4786 	u8 *ptr;
4787 
4788 	mutex_lock(&fs_info->chunk_mutex);
4789 	array_size = btrfs_super_sys_array_size(super_copy);
4790 	if (array_size + item_size + sizeof(disk_key)
4791 			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4792 		mutex_unlock(&fs_info->chunk_mutex);
4793 		return -EFBIG;
4794 	}
4795 
4796 	ptr = super_copy->sys_chunk_array + array_size;
4797 	btrfs_cpu_key_to_disk(&disk_key, key);
4798 	memcpy(ptr, &disk_key, sizeof(disk_key));
4799 	ptr += sizeof(disk_key);
4800 	memcpy(ptr, chunk, item_size);
4801 	item_size += sizeof(disk_key);
4802 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4803 	mutex_unlock(&fs_info->chunk_mutex);
4804 
4805 	return 0;
4806 }
4807 
4808 /*
4809  * sort the devices in descending order by max_avail, total_avail
4810  */
4811 static int btrfs_cmp_device_info(const void *a, const void *b)
4812 {
4813 	const struct btrfs_device_info *di_a = a;
4814 	const struct btrfs_device_info *di_b = b;
4815 
4816 	if (di_a->max_avail > di_b->max_avail)
4817 		return -1;
4818 	if (di_a->max_avail < di_b->max_avail)
4819 		return 1;
4820 	if (di_a->total_avail > di_b->total_avail)
4821 		return -1;
4822 	if (di_a->total_avail < di_b->total_avail)
4823 		return 1;
4824 	return 0;
4825 }
4826 
4827 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4828 {
4829 	if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4830 		return;
4831 
4832 	btrfs_set_fs_incompat(info, RAID56);
4833 }
4834 
4835 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4836 {
4837 	if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4838 		return;
4839 
4840 	btrfs_set_fs_incompat(info, RAID1C34);
4841 }
4842 
4843 /*
4844  * Structure used internally for __btrfs_alloc_chunk() function.
4845  * Wraps needed parameters.
4846  */
4847 struct alloc_chunk_ctl {
4848 	u64 start;
4849 	u64 type;
4850 	/* Total number of stripes to allocate */
4851 	int num_stripes;
4852 	/* sub_stripes info for map */
4853 	int sub_stripes;
4854 	/* Stripes per device */
4855 	int dev_stripes;
4856 	/* Maximum number of devices to use */
4857 	int devs_max;
4858 	/* Minimum number of devices to use */
4859 	int devs_min;
4860 	/* ndevs has to be a multiple of this */
4861 	int devs_increment;
4862 	/* Number of copies */
4863 	int ncopies;
4864 	/* Number of stripes worth of bytes to store parity information */
4865 	int nparity;
4866 	u64 max_stripe_size;
4867 	u64 max_chunk_size;
4868 	u64 dev_extent_min;
4869 	u64 stripe_size;
4870 	u64 chunk_size;
4871 	int ndevs;
4872 };
4873 
4874 static void init_alloc_chunk_ctl_policy_regular(
4875 				struct btrfs_fs_devices *fs_devices,
4876 				struct alloc_chunk_ctl *ctl)
4877 {
4878 	u64 type = ctl->type;
4879 
4880 	if (type & BTRFS_BLOCK_GROUP_DATA) {
4881 		ctl->max_stripe_size = SZ_1G;
4882 		ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4883 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4884 		/* For larger filesystems, use larger metadata chunks */
4885 		if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4886 			ctl->max_stripe_size = SZ_1G;
4887 		else
4888 			ctl->max_stripe_size = SZ_256M;
4889 		ctl->max_chunk_size = ctl->max_stripe_size;
4890 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4891 		ctl->max_stripe_size = SZ_32M;
4892 		ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4893 		ctl->devs_max = min_t(int, ctl->devs_max,
4894 				      BTRFS_MAX_DEVS_SYS_CHUNK);
4895 	} else {
4896 		BUG();
4897 	}
4898 
4899 	/* We don't want a chunk larger than 10% of writable space */
4900 	ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4901 				  ctl->max_chunk_size);
4902 	ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
4903 }
4904 
4905 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
4906 				 struct alloc_chunk_ctl *ctl)
4907 {
4908 	int index = btrfs_bg_flags_to_raid_index(ctl->type);
4909 
4910 	ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
4911 	ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
4912 	ctl->devs_max = btrfs_raid_array[index].devs_max;
4913 	if (!ctl->devs_max)
4914 		ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
4915 	ctl->devs_min = btrfs_raid_array[index].devs_min;
4916 	ctl->devs_increment = btrfs_raid_array[index].devs_increment;
4917 	ctl->ncopies = btrfs_raid_array[index].ncopies;
4918 	ctl->nparity = btrfs_raid_array[index].nparity;
4919 	ctl->ndevs = 0;
4920 
4921 	switch (fs_devices->chunk_alloc_policy) {
4922 	case BTRFS_CHUNK_ALLOC_REGULAR:
4923 		init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
4924 		break;
4925 	default:
4926 		BUG();
4927 	}
4928 }
4929 
4930 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
4931 			      struct alloc_chunk_ctl *ctl,
4932 			      struct btrfs_device_info *devices_info)
4933 {
4934 	struct btrfs_fs_info *info = fs_devices->fs_info;
4935 	struct btrfs_device *device;
4936 	u64 total_avail;
4937 	u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
4938 	int ret;
4939 	int ndevs = 0;
4940 	u64 max_avail;
4941 	u64 dev_offset;
4942 
4943 	/*
4944 	 * in the first pass through the devices list, we gather information
4945 	 * about the available holes on each device.
4946 	 */
4947 	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4948 		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4949 			WARN(1, KERN_ERR
4950 			       "BTRFS: read-only device in alloc_list\n");
4951 			continue;
4952 		}
4953 
4954 		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4955 					&device->dev_state) ||
4956 		    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4957 			continue;
4958 
4959 		if (device->total_bytes > device->bytes_used)
4960 			total_avail = device->total_bytes - device->bytes_used;
4961 		else
4962 			total_avail = 0;
4963 
4964 		/* If there is no space on this device, skip it. */
4965 		if (total_avail < ctl->dev_extent_min)
4966 			continue;
4967 
4968 		ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
4969 					   &max_avail);
4970 		if (ret && ret != -ENOSPC)
4971 			return ret;
4972 
4973 		if (ret == 0)
4974 			max_avail = dev_extent_want;
4975 
4976 		if (max_avail < ctl->dev_extent_min) {
4977 			if (btrfs_test_opt(info, ENOSPC_DEBUG))
4978 				btrfs_debug(info,
4979 			"%s: devid %llu has no free space, have=%llu want=%llu",
4980 					    __func__, device->devid, max_avail,
4981 					    ctl->dev_extent_min);
4982 			continue;
4983 		}
4984 
4985 		if (ndevs == fs_devices->rw_devices) {
4986 			WARN(1, "%s: found more than %llu devices\n",
4987 			     __func__, fs_devices->rw_devices);
4988 			break;
4989 		}
4990 		devices_info[ndevs].dev_offset = dev_offset;
4991 		devices_info[ndevs].max_avail = max_avail;
4992 		devices_info[ndevs].total_avail = total_avail;
4993 		devices_info[ndevs].dev = device;
4994 		++ndevs;
4995 	}
4996 	ctl->ndevs = ndevs;
4997 
4998 	/*
4999 	 * now sort the devices by hole size / available space
5000 	 */
5001 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5002 	     btrfs_cmp_device_info, NULL);
5003 
5004 	return 0;
5005 }
5006 
5007 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5008 				      struct btrfs_device_info *devices_info)
5009 {
5010 	/* Number of stripes that count for block group size */
5011 	int data_stripes;
5012 
5013 	/*
5014 	 * The primary goal is to maximize the number of stripes, so use as
5015 	 * many devices as possible, even if the stripes are not maximum sized.
5016 	 *
5017 	 * The DUP profile stores more than one stripe per device, the
5018 	 * max_avail is the total size so we have to adjust.
5019 	 */
5020 	ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5021 				   ctl->dev_stripes);
5022 	ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5023 
5024 	/* This will have to be fixed for RAID1 and RAID10 over more drives */
5025 	data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5026 
5027 	/*
5028 	 * Use the number of data stripes to figure out how big this chunk is
5029 	 * really going to be in terms of logical address space, and compare
5030 	 * that answer with the max chunk size. If it's higher, we try to
5031 	 * reduce stripe_size.
5032 	 */
5033 	if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5034 		/*
5035 		 * Reduce stripe_size, round it up to a 16MB boundary again and
5036 		 * then use it, unless it ends up being even bigger than the
5037 		 * previous value we had already.
5038 		 */
5039 		ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5040 							data_stripes), SZ_16M),
5041 				       ctl->stripe_size);
5042 	}
5043 
5044 	/* Align to BTRFS_STRIPE_LEN */
5045 	ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5046 	ctl->chunk_size = ctl->stripe_size * data_stripes;
5047 
5048 	return 0;
5049 }
5050 
5051 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5052 			      struct alloc_chunk_ctl *ctl,
5053 			      struct btrfs_device_info *devices_info)
5054 {
5055 	struct btrfs_fs_info *info = fs_devices->fs_info;
5056 
5057 	/*
5058 	 * Round down to number of usable stripes, devs_increment can be any
5059 	 * number so we can't use round_down() that requires power of 2, while
5060 	 * rounddown is safe.
5061 	 */
5062 	ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5063 
5064 	if (ctl->ndevs < ctl->devs_min) {
5065 		if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5066 			btrfs_debug(info,
5067 	"%s: not enough devices with free space: have=%d minimum required=%d",
5068 				    __func__, ctl->ndevs, ctl->devs_min);
5069 		}
5070 		return -ENOSPC;
5071 	}
5072 
5073 	ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5074 
5075 	switch (fs_devices->chunk_alloc_policy) {
5076 	case BTRFS_CHUNK_ALLOC_REGULAR:
5077 		return decide_stripe_size_regular(ctl, devices_info);
5078 	default:
5079 		BUG();
5080 	}
5081 }
5082 
5083 static int create_chunk(struct btrfs_trans_handle *trans,
5084 			struct alloc_chunk_ctl *ctl,
5085 			struct btrfs_device_info *devices_info)
5086 {
5087 	struct btrfs_fs_info *info = trans->fs_info;
5088 	struct map_lookup *map = NULL;
5089 	struct extent_map_tree *em_tree;
5090 	struct extent_map *em;
5091 	u64 start = ctl->start;
5092 	u64 type = ctl->type;
5093 	int ret;
5094 	int i;
5095 	int j;
5096 
5097 	map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5098 	if (!map)
5099 		return -ENOMEM;
5100 	map->num_stripes = ctl->num_stripes;
5101 
5102 	for (i = 0; i < ctl->ndevs; ++i) {
5103 		for (j = 0; j < ctl->dev_stripes; ++j) {
5104 			int s = i * ctl->dev_stripes + j;
5105 			map->stripes[s].dev = devices_info[i].dev;
5106 			map->stripes[s].physical = devices_info[i].dev_offset +
5107 						   j * ctl->stripe_size;
5108 		}
5109 	}
5110 	map->stripe_len = BTRFS_STRIPE_LEN;
5111 	map->io_align = BTRFS_STRIPE_LEN;
5112 	map->io_width = BTRFS_STRIPE_LEN;
5113 	map->type = type;
5114 	map->sub_stripes = ctl->sub_stripes;
5115 
5116 	trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5117 
5118 	em = alloc_extent_map();
5119 	if (!em) {
5120 		kfree(map);
5121 		return -ENOMEM;
5122 	}
5123 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5124 	em->map_lookup = map;
5125 	em->start = start;
5126 	em->len = ctl->chunk_size;
5127 	em->block_start = 0;
5128 	em->block_len = em->len;
5129 	em->orig_block_len = ctl->stripe_size;
5130 
5131 	em_tree = &info->mapping_tree;
5132 	write_lock(&em_tree->lock);
5133 	ret = add_extent_mapping(em_tree, em, 0);
5134 	if (ret) {
5135 		write_unlock(&em_tree->lock);
5136 		free_extent_map(em);
5137 		return ret;
5138 	}
5139 	write_unlock(&em_tree->lock);
5140 
5141 	ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5142 	if (ret)
5143 		goto error_del_extent;
5144 
5145 	for (i = 0; i < map->num_stripes; i++) {
5146 		struct btrfs_device *dev = map->stripes[i].dev;
5147 
5148 		btrfs_device_set_bytes_used(dev,
5149 					    dev->bytes_used + ctl->stripe_size);
5150 		if (list_empty(&dev->post_commit_list))
5151 			list_add_tail(&dev->post_commit_list,
5152 				      &trans->transaction->dev_update_list);
5153 	}
5154 
5155 	atomic64_sub(ctl->stripe_size * map->num_stripes,
5156 		     &info->free_chunk_space);
5157 
5158 	free_extent_map(em);
5159 	check_raid56_incompat_flag(info, type);
5160 	check_raid1c34_incompat_flag(info, type);
5161 
5162 	return 0;
5163 
5164 error_del_extent:
5165 	write_lock(&em_tree->lock);
5166 	remove_extent_mapping(em_tree, em);
5167 	write_unlock(&em_tree->lock);
5168 
5169 	/* One for our allocation */
5170 	free_extent_map(em);
5171 	/* One for the tree reference */
5172 	free_extent_map(em);
5173 
5174 	return ret;
5175 }
5176 
5177 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5178 {
5179 	struct btrfs_fs_info *info = trans->fs_info;
5180 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
5181 	struct btrfs_device_info *devices_info = NULL;
5182 	struct alloc_chunk_ctl ctl;
5183 	int ret;
5184 
5185 	lockdep_assert_held(&info->chunk_mutex);
5186 
5187 	if (!alloc_profile_is_valid(type, 0)) {
5188 		ASSERT(0);
5189 		return -EINVAL;
5190 	}
5191 
5192 	if (list_empty(&fs_devices->alloc_list)) {
5193 		if (btrfs_test_opt(info, ENOSPC_DEBUG))
5194 			btrfs_debug(info, "%s: no writable device", __func__);
5195 		return -ENOSPC;
5196 	}
5197 
5198 	if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5199 		btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5200 		ASSERT(0);
5201 		return -EINVAL;
5202 	}
5203 
5204 	ctl.start = find_next_chunk(info);
5205 	ctl.type = type;
5206 	init_alloc_chunk_ctl(fs_devices, &ctl);
5207 
5208 	devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5209 			       GFP_NOFS);
5210 	if (!devices_info)
5211 		return -ENOMEM;
5212 
5213 	ret = gather_device_info(fs_devices, &ctl, devices_info);
5214 	if (ret < 0)
5215 		goto out;
5216 
5217 	ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5218 	if (ret < 0)
5219 		goto out;
5220 
5221 	ret = create_chunk(trans, &ctl, devices_info);
5222 
5223 out:
5224 	kfree(devices_info);
5225 	return ret;
5226 }
5227 
5228 /*
5229  * Chunk allocation falls into two parts. The first part does work
5230  * that makes the new allocated chunk usable, but does not do any operation
5231  * that modifies the chunk tree. The second part does the work that
5232  * requires modifying the chunk tree. This division is important for the
5233  * bootstrap process of adding storage to a seed btrfs.
5234  */
5235 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5236 			     u64 chunk_offset, u64 chunk_size)
5237 {
5238 	struct btrfs_fs_info *fs_info = trans->fs_info;
5239 	struct btrfs_root *extent_root = fs_info->extent_root;
5240 	struct btrfs_root *chunk_root = fs_info->chunk_root;
5241 	struct btrfs_key key;
5242 	struct btrfs_device *device;
5243 	struct btrfs_chunk *chunk;
5244 	struct btrfs_stripe *stripe;
5245 	struct extent_map *em;
5246 	struct map_lookup *map;
5247 	size_t item_size;
5248 	u64 dev_offset;
5249 	u64 stripe_size;
5250 	int i = 0;
5251 	int ret = 0;
5252 
5253 	em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5254 	if (IS_ERR(em))
5255 		return PTR_ERR(em);
5256 
5257 	map = em->map_lookup;
5258 	item_size = btrfs_chunk_item_size(map->num_stripes);
5259 	stripe_size = em->orig_block_len;
5260 
5261 	chunk = kzalloc(item_size, GFP_NOFS);
5262 	if (!chunk) {
5263 		ret = -ENOMEM;
5264 		goto out;
5265 	}
5266 
5267 	/*
5268 	 * Take the device list mutex to prevent races with the final phase of
5269 	 * a device replace operation that replaces the device object associated
5270 	 * with the map's stripes, because the device object's id can change
5271 	 * at any time during that final phase of the device replace operation
5272 	 * (dev-replace.c:btrfs_dev_replace_finishing()).
5273 	 */
5274 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
5275 	for (i = 0; i < map->num_stripes; i++) {
5276 		device = map->stripes[i].dev;
5277 		dev_offset = map->stripes[i].physical;
5278 
5279 		ret = btrfs_update_device(trans, device);
5280 		if (ret)
5281 			break;
5282 		ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5283 					     dev_offset, stripe_size);
5284 		if (ret)
5285 			break;
5286 	}
5287 	if (ret) {
5288 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5289 		goto out;
5290 	}
5291 
5292 	stripe = &chunk->stripe;
5293 	for (i = 0; i < map->num_stripes; i++) {
5294 		device = map->stripes[i].dev;
5295 		dev_offset = map->stripes[i].physical;
5296 
5297 		btrfs_set_stack_stripe_devid(stripe, device->devid);
5298 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
5299 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5300 		stripe++;
5301 	}
5302 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5303 
5304 	btrfs_set_stack_chunk_length(chunk, chunk_size);
5305 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5306 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5307 	btrfs_set_stack_chunk_type(chunk, map->type);
5308 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5309 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5310 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5311 	btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5312 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5313 
5314 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5315 	key.type = BTRFS_CHUNK_ITEM_KEY;
5316 	key.offset = chunk_offset;
5317 
5318 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5319 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5320 		/*
5321 		 * TODO: Cleanup of inserted chunk root in case of
5322 		 * failure.
5323 		 */
5324 		ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5325 	}
5326 
5327 out:
5328 	kfree(chunk);
5329 	free_extent_map(em);
5330 	return ret;
5331 }
5332 
5333 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5334 {
5335 	struct btrfs_fs_info *fs_info = trans->fs_info;
5336 	u64 alloc_profile;
5337 	int ret;
5338 
5339 	alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5340 	ret = btrfs_alloc_chunk(trans, alloc_profile);
5341 	if (ret)
5342 		return ret;
5343 
5344 	alloc_profile = btrfs_system_alloc_profile(fs_info);
5345 	ret = btrfs_alloc_chunk(trans, alloc_profile);
5346 	return ret;
5347 }
5348 
5349 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5350 {
5351 	const int index = btrfs_bg_flags_to_raid_index(map->type);
5352 
5353 	return btrfs_raid_array[index].tolerated_failures;
5354 }
5355 
5356 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5357 {
5358 	struct extent_map *em;
5359 	struct map_lookup *map;
5360 	int readonly = 0;
5361 	int miss_ndevs = 0;
5362 	int i;
5363 
5364 	em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5365 	if (IS_ERR(em))
5366 		return 1;
5367 
5368 	map = em->map_lookup;
5369 	for (i = 0; i < map->num_stripes; i++) {
5370 		if (test_bit(BTRFS_DEV_STATE_MISSING,
5371 					&map->stripes[i].dev->dev_state)) {
5372 			miss_ndevs++;
5373 			continue;
5374 		}
5375 		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5376 					&map->stripes[i].dev->dev_state)) {
5377 			readonly = 1;
5378 			goto end;
5379 		}
5380 	}
5381 
5382 	/*
5383 	 * If the number of missing devices is larger than max errors,
5384 	 * we can not write the data into that chunk successfully, so
5385 	 * set it readonly.
5386 	 */
5387 	if (miss_ndevs > btrfs_chunk_max_errors(map))
5388 		readonly = 1;
5389 end:
5390 	free_extent_map(em);
5391 	return readonly;
5392 }
5393 
5394 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5395 {
5396 	struct extent_map *em;
5397 
5398 	while (1) {
5399 		write_lock(&tree->lock);
5400 		em = lookup_extent_mapping(tree, 0, (u64)-1);
5401 		if (em)
5402 			remove_extent_mapping(tree, em);
5403 		write_unlock(&tree->lock);
5404 		if (!em)
5405 			break;
5406 		/* once for us */
5407 		free_extent_map(em);
5408 		/* once for the tree */
5409 		free_extent_map(em);
5410 	}
5411 }
5412 
5413 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5414 {
5415 	struct extent_map *em;
5416 	struct map_lookup *map;
5417 	int ret;
5418 
5419 	em = btrfs_get_chunk_map(fs_info, logical, len);
5420 	if (IS_ERR(em))
5421 		/*
5422 		 * We could return errors for these cases, but that could get
5423 		 * ugly and we'd probably do the same thing which is just not do
5424 		 * anything else and exit, so return 1 so the callers don't try
5425 		 * to use other copies.
5426 		 */
5427 		return 1;
5428 
5429 	map = em->map_lookup;
5430 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5431 		ret = map->num_stripes;
5432 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5433 		ret = map->sub_stripes;
5434 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5435 		ret = 2;
5436 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5437 		/*
5438 		 * There could be two corrupted data stripes, we need
5439 		 * to loop retry in order to rebuild the correct data.
5440 		 *
5441 		 * Fail a stripe at a time on every retry except the
5442 		 * stripe under reconstruction.
5443 		 */
5444 		ret = map->num_stripes;
5445 	else
5446 		ret = 1;
5447 	free_extent_map(em);
5448 
5449 	down_read(&fs_info->dev_replace.rwsem);
5450 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5451 	    fs_info->dev_replace.tgtdev)
5452 		ret++;
5453 	up_read(&fs_info->dev_replace.rwsem);
5454 
5455 	return ret;
5456 }
5457 
5458 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5459 				    u64 logical)
5460 {
5461 	struct extent_map *em;
5462 	struct map_lookup *map;
5463 	unsigned long len = fs_info->sectorsize;
5464 
5465 	em = btrfs_get_chunk_map(fs_info, logical, len);
5466 
5467 	if (!WARN_ON(IS_ERR(em))) {
5468 		map = em->map_lookup;
5469 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5470 			len = map->stripe_len * nr_data_stripes(map);
5471 		free_extent_map(em);
5472 	}
5473 	return len;
5474 }
5475 
5476 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5477 {
5478 	struct extent_map *em;
5479 	struct map_lookup *map;
5480 	int ret = 0;
5481 
5482 	em = btrfs_get_chunk_map(fs_info, logical, len);
5483 
5484 	if(!WARN_ON(IS_ERR(em))) {
5485 		map = em->map_lookup;
5486 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5487 			ret = 1;
5488 		free_extent_map(em);
5489 	}
5490 	return ret;
5491 }
5492 
5493 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5494 			    struct map_lookup *map, int first,
5495 			    int dev_replace_is_ongoing)
5496 {
5497 	int i;
5498 	int num_stripes;
5499 	int preferred_mirror;
5500 	int tolerance;
5501 	struct btrfs_device *srcdev;
5502 
5503 	ASSERT((map->type &
5504 		 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5505 
5506 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5507 		num_stripes = map->sub_stripes;
5508 	else
5509 		num_stripes = map->num_stripes;
5510 
5511 	switch (fs_info->fs_devices->read_policy) {
5512 	default:
5513 		/* Shouldn't happen, just warn and use pid instead of failing */
5514 		btrfs_warn_rl(fs_info,
5515 			      "unknown read_policy type %u, reset to pid",
5516 			      fs_info->fs_devices->read_policy);
5517 		fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5518 		fallthrough;
5519 	case BTRFS_READ_POLICY_PID:
5520 		preferred_mirror = first + (current->pid % num_stripes);
5521 		break;
5522 	}
5523 
5524 	if (dev_replace_is_ongoing &&
5525 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5526 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5527 		srcdev = fs_info->dev_replace.srcdev;
5528 	else
5529 		srcdev = NULL;
5530 
5531 	/*
5532 	 * try to avoid the drive that is the source drive for a
5533 	 * dev-replace procedure, only choose it if no other non-missing
5534 	 * mirror is available
5535 	 */
5536 	for (tolerance = 0; tolerance < 2; tolerance++) {
5537 		if (map->stripes[preferred_mirror].dev->bdev &&
5538 		    (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5539 			return preferred_mirror;
5540 		for (i = first; i < first + num_stripes; i++) {
5541 			if (map->stripes[i].dev->bdev &&
5542 			    (tolerance || map->stripes[i].dev != srcdev))
5543 				return i;
5544 		}
5545 	}
5546 
5547 	/* we couldn't find one that doesn't fail.  Just return something
5548 	 * and the io error handling code will clean up eventually
5549 	 */
5550 	return preferred_mirror;
5551 }
5552 
5553 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5554 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5555 {
5556 	int i;
5557 	int again = 1;
5558 
5559 	while (again) {
5560 		again = 0;
5561 		for (i = 0; i < num_stripes - 1; i++) {
5562 			/* Swap if parity is on a smaller index */
5563 			if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5564 				swap(bbio->stripes[i], bbio->stripes[i + 1]);
5565 				swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5566 				again = 1;
5567 			}
5568 		}
5569 	}
5570 }
5571 
5572 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5573 {
5574 	struct btrfs_bio *bbio = kzalloc(
5575 		 /* the size of the btrfs_bio */
5576 		sizeof(struct btrfs_bio) +
5577 		/* plus the variable array for the stripes */
5578 		sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5579 		/* plus the variable array for the tgt dev */
5580 		sizeof(int) * (real_stripes) +
5581 		/*
5582 		 * plus the raid_map, which includes both the tgt dev
5583 		 * and the stripes
5584 		 */
5585 		sizeof(u64) * (total_stripes),
5586 		GFP_NOFS|__GFP_NOFAIL);
5587 
5588 	atomic_set(&bbio->error, 0);
5589 	refcount_set(&bbio->refs, 1);
5590 
5591 	bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5592 	bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5593 
5594 	return bbio;
5595 }
5596 
5597 void btrfs_get_bbio(struct btrfs_bio *bbio)
5598 {
5599 	WARN_ON(!refcount_read(&bbio->refs));
5600 	refcount_inc(&bbio->refs);
5601 }
5602 
5603 void btrfs_put_bbio(struct btrfs_bio *bbio)
5604 {
5605 	if (!bbio)
5606 		return;
5607 	if (refcount_dec_and_test(&bbio->refs))
5608 		kfree(bbio);
5609 }
5610 
5611 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5612 /*
5613  * Please note that, discard won't be sent to target device of device
5614  * replace.
5615  */
5616 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5617 					 u64 logical, u64 *length_ret,
5618 					 struct btrfs_bio **bbio_ret)
5619 {
5620 	struct extent_map *em;
5621 	struct map_lookup *map;
5622 	struct btrfs_bio *bbio;
5623 	u64 length = *length_ret;
5624 	u64 offset;
5625 	u64 stripe_nr;
5626 	u64 stripe_nr_end;
5627 	u64 stripe_end_offset;
5628 	u64 stripe_cnt;
5629 	u64 stripe_len;
5630 	u64 stripe_offset;
5631 	u64 num_stripes;
5632 	u32 stripe_index;
5633 	u32 factor = 0;
5634 	u32 sub_stripes = 0;
5635 	u64 stripes_per_dev = 0;
5636 	u32 remaining_stripes = 0;
5637 	u32 last_stripe = 0;
5638 	int ret = 0;
5639 	int i;
5640 
5641 	/* discard always return a bbio */
5642 	ASSERT(bbio_ret);
5643 
5644 	em = btrfs_get_chunk_map(fs_info, logical, length);
5645 	if (IS_ERR(em))
5646 		return PTR_ERR(em);
5647 
5648 	map = em->map_lookup;
5649 	/* we don't discard raid56 yet */
5650 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5651 		ret = -EOPNOTSUPP;
5652 		goto out;
5653 	}
5654 
5655 	offset = logical - em->start;
5656 	length = min_t(u64, em->start + em->len - logical, length);
5657 	*length_ret = length;
5658 
5659 	stripe_len = map->stripe_len;
5660 	/*
5661 	 * stripe_nr counts the total number of stripes we have to stride
5662 	 * to get to this block
5663 	 */
5664 	stripe_nr = div64_u64(offset, stripe_len);
5665 
5666 	/* stripe_offset is the offset of this block in its stripe */
5667 	stripe_offset = offset - stripe_nr * stripe_len;
5668 
5669 	stripe_nr_end = round_up(offset + length, map->stripe_len);
5670 	stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5671 	stripe_cnt = stripe_nr_end - stripe_nr;
5672 	stripe_end_offset = stripe_nr_end * map->stripe_len -
5673 			    (offset + length);
5674 	/*
5675 	 * after this, stripe_nr is the number of stripes on this
5676 	 * device we have to walk to find the data, and stripe_index is
5677 	 * the number of our device in the stripe array
5678 	 */
5679 	num_stripes = 1;
5680 	stripe_index = 0;
5681 	if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5682 			 BTRFS_BLOCK_GROUP_RAID10)) {
5683 		if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5684 			sub_stripes = 1;
5685 		else
5686 			sub_stripes = map->sub_stripes;
5687 
5688 		factor = map->num_stripes / sub_stripes;
5689 		num_stripes = min_t(u64, map->num_stripes,
5690 				    sub_stripes * stripe_cnt);
5691 		stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5692 		stripe_index *= sub_stripes;
5693 		stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5694 					      &remaining_stripes);
5695 		div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5696 		last_stripe *= sub_stripes;
5697 	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5698 				BTRFS_BLOCK_GROUP_DUP)) {
5699 		num_stripes = map->num_stripes;
5700 	} else {
5701 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5702 					&stripe_index);
5703 	}
5704 
5705 	bbio = alloc_btrfs_bio(num_stripes, 0);
5706 	if (!bbio) {
5707 		ret = -ENOMEM;
5708 		goto out;
5709 	}
5710 
5711 	for (i = 0; i < num_stripes; i++) {
5712 		bbio->stripes[i].physical =
5713 			map->stripes[stripe_index].physical +
5714 			stripe_offset + stripe_nr * map->stripe_len;
5715 		bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5716 
5717 		if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5718 				 BTRFS_BLOCK_GROUP_RAID10)) {
5719 			bbio->stripes[i].length = stripes_per_dev *
5720 				map->stripe_len;
5721 
5722 			if (i / sub_stripes < remaining_stripes)
5723 				bbio->stripes[i].length +=
5724 					map->stripe_len;
5725 
5726 			/*
5727 			 * Special for the first stripe and
5728 			 * the last stripe:
5729 			 *
5730 			 * |-------|...|-------|
5731 			 *     |----------|
5732 			 *    off     end_off
5733 			 */
5734 			if (i < sub_stripes)
5735 				bbio->stripes[i].length -=
5736 					stripe_offset;
5737 
5738 			if (stripe_index >= last_stripe &&
5739 			    stripe_index <= (last_stripe +
5740 					     sub_stripes - 1))
5741 				bbio->stripes[i].length -=
5742 					stripe_end_offset;
5743 
5744 			if (i == sub_stripes - 1)
5745 				stripe_offset = 0;
5746 		} else {
5747 			bbio->stripes[i].length = length;
5748 		}
5749 
5750 		stripe_index++;
5751 		if (stripe_index == map->num_stripes) {
5752 			stripe_index = 0;
5753 			stripe_nr++;
5754 		}
5755 	}
5756 
5757 	*bbio_ret = bbio;
5758 	bbio->map_type = map->type;
5759 	bbio->num_stripes = num_stripes;
5760 out:
5761 	free_extent_map(em);
5762 	return ret;
5763 }
5764 
5765 /*
5766  * In dev-replace case, for repair case (that's the only case where the mirror
5767  * is selected explicitly when calling btrfs_map_block), blocks left of the
5768  * left cursor can also be read from the target drive.
5769  *
5770  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5771  * array of stripes.
5772  * For READ, it also needs to be supported using the same mirror number.
5773  *
5774  * If the requested block is not left of the left cursor, EIO is returned. This
5775  * can happen because btrfs_num_copies() returns one more in the dev-replace
5776  * case.
5777  */
5778 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5779 					 u64 logical, u64 length,
5780 					 u64 srcdev_devid, int *mirror_num,
5781 					 u64 *physical)
5782 {
5783 	struct btrfs_bio *bbio = NULL;
5784 	int num_stripes;
5785 	int index_srcdev = 0;
5786 	int found = 0;
5787 	u64 physical_of_found = 0;
5788 	int i;
5789 	int ret = 0;
5790 
5791 	ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5792 				logical, &length, &bbio, 0, 0);
5793 	if (ret) {
5794 		ASSERT(bbio == NULL);
5795 		return ret;
5796 	}
5797 
5798 	num_stripes = bbio->num_stripes;
5799 	if (*mirror_num > num_stripes) {
5800 		/*
5801 		 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5802 		 * that means that the requested area is not left of the left
5803 		 * cursor
5804 		 */
5805 		btrfs_put_bbio(bbio);
5806 		return -EIO;
5807 	}
5808 
5809 	/*
5810 	 * process the rest of the function using the mirror_num of the source
5811 	 * drive. Therefore look it up first.  At the end, patch the device
5812 	 * pointer to the one of the target drive.
5813 	 */
5814 	for (i = 0; i < num_stripes; i++) {
5815 		if (bbio->stripes[i].dev->devid != srcdev_devid)
5816 			continue;
5817 
5818 		/*
5819 		 * In case of DUP, in order to keep it simple, only add the
5820 		 * mirror with the lowest physical address
5821 		 */
5822 		if (found &&
5823 		    physical_of_found <= bbio->stripes[i].physical)
5824 			continue;
5825 
5826 		index_srcdev = i;
5827 		found = 1;
5828 		physical_of_found = bbio->stripes[i].physical;
5829 	}
5830 
5831 	btrfs_put_bbio(bbio);
5832 
5833 	ASSERT(found);
5834 	if (!found)
5835 		return -EIO;
5836 
5837 	*mirror_num = index_srcdev + 1;
5838 	*physical = physical_of_found;
5839 	return ret;
5840 }
5841 
5842 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5843 				      struct btrfs_bio **bbio_ret,
5844 				      struct btrfs_dev_replace *dev_replace,
5845 				      int *num_stripes_ret, int *max_errors_ret)
5846 {
5847 	struct btrfs_bio *bbio = *bbio_ret;
5848 	u64 srcdev_devid = dev_replace->srcdev->devid;
5849 	int tgtdev_indexes = 0;
5850 	int num_stripes = *num_stripes_ret;
5851 	int max_errors = *max_errors_ret;
5852 	int i;
5853 
5854 	if (op == BTRFS_MAP_WRITE) {
5855 		int index_where_to_add;
5856 
5857 		/*
5858 		 * duplicate the write operations while the dev replace
5859 		 * procedure is running. Since the copying of the old disk to
5860 		 * the new disk takes place at run time while the filesystem is
5861 		 * mounted writable, the regular write operations to the old
5862 		 * disk have to be duplicated to go to the new disk as well.
5863 		 *
5864 		 * Note that device->missing is handled by the caller, and that
5865 		 * the write to the old disk is already set up in the stripes
5866 		 * array.
5867 		 */
5868 		index_where_to_add = num_stripes;
5869 		for (i = 0; i < num_stripes; i++) {
5870 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5871 				/* write to new disk, too */
5872 				struct btrfs_bio_stripe *new =
5873 					bbio->stripes + index_where_to_add;
5874 				struct btrfs_bio_stripe *old =
5875 					bbio->stripes + i;
5876 
5877 				new->physical = old->physical;
5878 				new->length = old->length;
5879 				new->dev = dev_replace->tgtdev;
5880 				bbio->tgtdev_map[i] = index_where_to_add;
5881 				index_where_to_add++;
5882 				max_errors++;
5883 				tgtdev_indexes++;
5884 			}
5885 		}
5886 		num_stripes = index_where_to_add;
5887 	} else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5888 		int index_srcdev = 0;
5889 		int found = 0;
5890 		u64 physical_of_found = 0;
5891 
5892 		/*
5893 		 * During the dev-replace procedure, the target drive can also
5894 		 * be used to read data in case it is needed to repair a corrupt
5895 		 * block elsewhere. This is possible if the requested area is
5896 		 * left of the left cursor. In this area, the target drive is a
5897 		 * full copy of the source drive.
5898 		 */
5899 		for (i = 0; i < num_stripes; i++) {
5900 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5901 				/*
5902 				 * In case of DUP, in order to keep it simple,
5903 				 * only add the mirror with the lowest physical
5904 				 * address
5905 				 */
5906 				if (found &&
5907 				    physical_of_found <=
5908 				     bbio->stripes[i].physical)
5909 					continue;
5910 				index_srcdev = i;
5911 				found = 1;
5912 				physical_of_found = bbio->stripes[i].physical;
5913 			}
5914 		}
5915 		if (found) {
5916 			struct btrfs_bio_stripe *tgtdev_stripe =
5917 				bbio->stripes + num_stripes;
5918 
5919 			tgtdev_stripe->physical = physical_of_found;
5920 			tgtdev_stripe->length =
5921 				bbio->stripes[index_srcdev].length;
5922 			tgtdev_stripe->dev = dev_replace->tgtdev;
5923 			bbio->tgtdev_map[index_srcdev] = num_stripes;
5924 
5925 			tgtdev_indexes++;
5926 			num_stripes++;
5927 		}
5928 	}
5929 
5930 	*num_stripes_ret = num_stripes;
5931 	*max_errors_ret = max_errors;
5932 	bbio->num_tgtdevs = tgtdev_indexes;
5933 	*bbio_ret = bbio;
5934 }
5935 
5936 static bool need_full_stripe(enum btrfs_map_op op)
5937 {
5938 	return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5939 }
5940 
5941 /*
5942  * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5943  *		       tuple. This information is used to calculate how big a
5944  *		       particular bio can get before it straddles a stripe.
5945  *
5946  * @fs_info - the filesystem
5947  * @logical - address that we want to figure out the geometry of
5948  * @len	    - the length of IO we are going to perform, starting at @logical
5949  * @op      - type of operation - write or read
5950  * @io_geom - pointer used to return values
5951  *
5952  * Returns < 0 in case a chunk for the given logical address cannot be found,
5953  * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5954  */
5955 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5956 			u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5957 {
5958 	struct extent_map *em;
5959 	struct map_lookup *map;
5960 	u64 offset;
5961 	u64 stripe_offset;
5962 	u64 stripe_nr;
5963 	u64 stripe_len;
5964 	u64 raid56_full_stripe_start = (u64)-1;
5965 	int data_stripes;
5966 	int ret = 0;
5967 
5968 	ASSERT(op != BTRFS_MAP_DISCARD);
5969 
5970 	em = btrfs_get_chunk_map(fs_info, logical, len);
5971 	if (IS_ERR(em))
5972 		return PTR_ERR(em);
5973 
5974 	map = em->map_lookup;
5975 	/* Offset of this logical address in the chunk */
5976 	offset = logical - em->start;
5977 	/* Len of a stripe in a chunk */
5978 	stripe_len = map->stripe_len;
5979 	/* Stripe wher this block falls in */
5980 	stripe_nr = div64_u64(offset, stripe_len);
5981 	/* Offset of stripe in the chunk */
5982 	stripe_offset = stripe_nr * stripe_len;
5983 	if (offset < stripe_offset) {
5984 		btrfs_crit(fs_info,
5985 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5986 			stripe_offset, offset, em->start, logical, stripe_len);
5987 		ret = -EINVAL;
5988 		goto out;
5989 	}
5990 
5991 	/* stripe_offset is the offset of this block in its stripe */
5992 	stripe_offset = offset - stripe_offset;
5993 	data_stripes = nr_data_stripes(map);
5994 
5995 	if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5996 		u64 max_len = stripe_len - stripe_offset;
5997 
5998 		/*
5999 		 * In case of raid56, we need to know the stripe aligned start
6000 		 */
6001 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6002 			unsigned long full_stripe_len = stripe_len * data_stripes;
6003 			raid56_full_stripe_start = offset;
6004 
6005 			/*
6006 			 * Allow a write of a full stripe, but make sure we
6007 			 * don't allow straddling of stripes
6008 			 */
6009 			raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6010 					full_stripe_len);
6011 			raid56_full_stripe_start *= full_stripe_len;
6012 
6013 			/*
6014 			 * For writes to RAID[56], allow a full stripeset across
6015 			 * all disks. For other RAID types and for RAID[56]
6016 			 * reads, just allow a single stripe (on a single disk).
6017 			 */
6018 			if (op == BTRFS_MAP_WRITE) {
6019 				max_len = stripe_len * data_stripes -
6020 					  (offset - raid56_full_stripe_start);
6021 			}
6022 		}
6023 		len = min_t(u64, em->len - offset, max_len);
6024 	} else {
6025 		len = em->len - offset;
6026 	}
6027 
6028 	io_geom->len = len;
6029 	io_geom->offset = offset;
6030 	io_geom->stripe_len = stripe_len;
6031 	io_geom->stripe_nr = stripe_nr;
6032 	io_geom->stripe_offset = stripe_offset;
6033 	io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6034 
6035 out:
6036 	/* once for us */
6037 	free_extent_map(em);
6038 	return ret;
6039 }
6040 
6041 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6042 			     enum btrfs_map_op op,
6043 			     u64 logical, u64 *length,
6044 			     struct btrfs_bio **bbio_ret,
6045 			     int mirror_num, int need_raid_map)
6046 {
6047 	struct extent_map *em;
6048 	struct map_lookup *map;
6049 	u64 stripe_offset;
6050 	u64 stripe_nr;
6051 	u64 stripe_len;
6052 	u32 stripe_index;
6053 	int data_stripes;
6054 	int i;
6055 	int ret = 0;
6056 	int num_stripes;
6057 	int max_errors = 0;
6058 	int tgtdev_indexes = 0;
6059 	struct btrfs_bio *bbio = NULL;
6060 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6061 	int dev_replace_is_ongoing = 0;
6062 	int num_alloc_stripes;
6063 	int patch_the_first_stripe_for_dev_replace = 0;
6064 	u64 physical_to_patch_in_first_stripe = 0;
6065 	u64 raid56_full_stripe_start = (u64)-1;
6066 	struct btrfs_io_geometry geom;
6067 
6068 	ASSERT(bbio_ret);
6069 	ASSERT(op != BTRFS_MAP_DISCARD);
6070 
6071 	ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6072 	if (ret < 0)
6073 		return ret;
6074 
6075 	em = btrfs_get_chunk_map(fs_info, logical, *length);
6076 	ASSERT(!IS_ERR(em));
6077 	map = em->map_lookup;
6078 
6079 	*length = geom.len;
6080 	stripe_len = geom.stripe_len;
6081 	stripe_nr = geom.stripe_nr;
6082 	stripe_offset = geom.stripe_offset;
6083 	raid56_full_stripe_start = geom.raid56_stripe_offset;
6084 	data_stripes = nr_data_stripes(map);
6085 
6086 	down_read(&dev_replace->rwsem);
6087 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6088 	/*
6089 	 * Hold the semaphore for read during the whole operation, write is
6090 	 * requested at commit time but must wait.
6091 	 */
6092 	if (!dev_replace_is_ongoing)
6093 		up_read(&dev_replace->rwsem);
6094 
6095 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6096 	    !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6097 		ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6098 						    dev_replace->srcdev->devid,
6099 						    &mirror_num,
6100 					    &physical_to_patch_in_first_stripe);
6101 		if (ret)
6102 			goto out;
6103 		else
6104 			patch_the_first_stripe_for_dev_replace = 1;
6105 	} else if (mirror_num > map->num_stripes) {
6106 		mirror_num = 0;
6107 	}
6108 
6109 	num_stripes = 1;
6110 	stripe_index = 0;
6111 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6112 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6113 				&stripe_index);
6114 		if (!need_full_stripe(op))
6115 			mirror_num = 1;
6116 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6117 		if (need_full_stripe(op))
6118 			num_stripes = map->num_stripes;
6119 		else if (mirror_num)
6120 			stripe_index = mirror_num - 1;
6121 		else {
6122 			stripe_index = find_live_mirror(fs_info, map, 0,
6123 					    dev_replace_is_ongoing);
6124 			mirror_num = stripe_index + 1;
6125 		}
6126 
6127 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6128 		if (need_full_stripe(op)) {
6129 			num_stripes = map->num_stripes;
6130 		} else if (mirror_num) {
6131 			stripe_index = mirror_num - 1;
6132 		} else {
6133 			mirror_num = 1;
6134 		}
6135 
6136 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6137 		u32 factor = map->num_stripes / map->sub_stripes;
6138 
6139 		stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6140 		stripe_index *= map->sub_stripes;
6141 
6142 		if (need_full_stripe(op))
6143 			num_stripes = map->sub_stripes;
6144 		else if (mirror_num)
6145 			stripe_index += mirror_num - 1;
6146 		else {
6147 			int old_stripe_index = stripe_index;
6148 			stripe_index = find_live_mirror(fs_info, map,
6149 					      stripe_index,
6150 					      dev_replace_is_ongoing);
6151 			mirror_num = stripe_index - old_stripe_index + 1;
6152 		}
6153 
6154 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6155 		if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6156 			/* push stripe_nr back to the start of the full stripe */
6157 			stripe_nr = div64_u64(raid56_full_stripe_start,
6158 					stripe_len * data_stripes);
6159 
6160 			/* RAID[56] write or recovery. Return all stripes */
6161 			num_stripes = map->num_stripes;
6162 			max_errors = nr_parity_stripes(map);
6163 
6164 			*length = map->stripe_len;
6165 			stripe_index = 0;
6166 			stripe_offset = 0;
6167 		} else {
6168 			/*
6169 			 * Mirror #0 or #1 means the original data block.
6170 			 * Mirror #2 is RAID5 parity block.
6171 			 * Mirror #3 is RAID6 Q block.
6172 			 */
6173 			stripe_nr = div_u64_rem(stripe_nr,
6174 					data_stripes, &stripe_index);
6175 			if (mirror_num > 1)
6176 				stripe_index = data_stripes + mirror_num - 2;
6177 
6178 			/* We distribute the parity blocks across stripes */
6179 			div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6180 					&stripe_index);
6181 			if (!need_full_stripe(op) && mirror_num <= 1)
6182 				mirror_num = 1;
6183 		}
6184 	} else {
6185 		/*
6186 		 * after this, stripe_nr is the number of stripes on this
6187 		 * device we have to walk to find the data, and stripe_index is
6188 		 * the number of our device in the stripe array
6189 		 */
6190 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6191 				&stripe_index);
6192 		mirror_num = stripe_index + 1;
6193 	}
6194 	if (stripe_index >= map->num_stripes) {
6195 		btrfs_crit(fs_info,
6196 			   "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6197 			   stripe_index, map->num_stripes);
6198 		ret = -EINVAL;
6199 		goto out;
6200 	}
6201 
6202 	num_alloc_stripes = num_stripes;
6203 	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6204 		if (op == BTRFS_MAP_WRITE)
6205 			num_alloc_stripes <<= 1;
6206 		if (op == BTRFS_MAP_GET_READ_MIRRORS)
6207 			num_alloc_stripes++;
6208 		tgtdev_indexes = num_stripes;
6209 	}
6210 
6211 	bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6212 	if (!bbio) {
6213 		ret = -ENOMEM;
6214 		goto out;
6215 	}
6216 
6217 	for (i = 0; i < num_stripes; i++) {
6218 		bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6219 			stripe_offset + stripe_nr * map->stripe_len;
6220 		bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6221 		stripe_index++;
6222 	}
6223 
6224 	/* build raid_map */
6225 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6226 	    (need_full_stripe(op) || mirror_num > 1)) {
6227 		u64 tmp;
6228 		unsigned rot;
6229 
6230 		/* Work out the disk rotation on this stripe-set */
6231 		div_u64_rem(stripe_nr, num_stripes, &rot);
6232 
6233 		/* Fill in the logical address of each stripe */
6234 		tmp = stripe_nr * data_stripes;
6235 		for (i = 0; i < data_stripes; i++)
6236 			bbio->raid_map[(i+rot) % num_stripes] =
6237 				em->start + (tmp + i) * map->stripe_len;
6238 
6239 		bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6240 		if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6241 			bbio->raid_map[(i+rot+1) % num_stripes] =
6242 				RAID6_Q_STRIPE;
6243 
6244 		sort_parity_stripes(bbio, num_stripes);
6245 	}
6246 
6247 	if (need_full_stripe(op))
6248 		max_errors = btrfs_chunk_max_errors(map);
6249 
6250 	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6251 	    need_full_stripe(op)) {
6252 		handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6253 					  &max_errors);
6254 	}
6255 
6256 	*bbio_ret = bbio;
6257 	bbio->map_type = map->type;
6258 	bbio->num_stripes = num_stripes;
6259 	bbio->max_errors = max_errors;
6260 	bbio->mirror_num = mirror_num;
6261 
6262 	/*
6263 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6264 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
6265 	 * available as a mirror
6266 	 */
6267 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6268 		WARN_ON(num_stripes > 1);
6269 		bbio->stripes[0].dev = dev_replace->tgtdev;
6270 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6271 		bbio->mirror_num = map->num_stripes + 1;
6272 	}
6273 out:
6274 	if (dev_replace_is_ongoing) {
6275 		lockdep_assert_held(&dev_replace->rwsem);
6276 		/* Unlock and let waiting writers proceed */
6277 		up_read(&dev_replace->rwsem);
6278 	}
6279 	free_extent_map(em);
6280 	return ret;
6281 }
6282 
6283 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6284 		      u64 logical, u64 *length,
6285 		      struct btrfs_bio **bbio_ret, int mirror_num)
6286 {
6287 	if (op == BTRFS_MAP_DISCARD)
6288 		return __btrfs_map_block_for_discard(fs_info, logical,
6289 						     length, bbio_ret);
6290 
6291 	return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6292 				 mirror_num, 0);
6293 }
6294 
6295 /* For Scrub/replace */
6296 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6297 		     u64 logical, u64 *length,
6298 		     struct btrfs_bio **bbio_ret)
6299 {
6300 	return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6301 }
6302 
6303 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6304 {
6305 	bio->bi_private = bbio->private;
6306 	bio->bi_end_io = bbio->end_io;
6307 	bio_endio(bio);
6308 
6309 	btrfs_put_bbio(bbio);
6310 }
6311 
6312 static void btrfs_end_bio(struct bio *bio)
6313 {
6314 	struct btrfs_bio *bbio = bio->bi_private;
6315 	int is_orig_bio = 0;
6316 
6317 	if (bio->bi_status) {
6318 		atomic_inc(&bbio->error);
6319 		if (bio->bi_status == BLK_STS_IOERR ||
6320 		    bio->bi_status == BLK_STS_TARGET) {
6321 			struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6322 
6323 			ASSERT(dev->bdev);
6324 			if (bio_op(bio) == REQ_OP_WRITE)
6325 				btrfs_dev_stat_inc_and_print(dev,
6326 						BTRFS_DEV_STAT_WRITE_ERRS);
6327 			else if (!(bio->bi_opf & REQ_RAHEAD))
6328 				btrfs_dev_stat_inc_and_print(dev,
6329 						BTRFS_DEV_STAT_READ_ERRS);
6330 			if (bio->bi_opf & REQ_PREFLUSH)
6331 				btrfs_dev_stat_inc_and_print(dev,
6332 						BTRFS_DEV_STAT_FLUSH_ERRS);
6333 		}
6334 	}
6335 
6336 	if (bio == bbio->orig_bio)
6337 		is_orig_bio = 1;
6338 
6339 	btrfs_bio_counter_dec(bbio->fs_info);
6340 
6341 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
6342 		if (!is_orig_bio) {
6343 			bio_put(bio);
6344 			bio = bbio->orig_bio;
6345 		}
6346 
6347 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6348 		/* only send an error to the higher layers if it is
6349 		 * beyond the tolerance of the btrfs bio
6350 		 */
6351 		if (atomic_read(&bbio->error) > bbio->max_errors) {
6352 			bio->bi_status = BLK_STS_IOERR;
6353 		} else {
6354 			/*
6355 			 * this bio is actually up to date, we didn't
6356 			 * go over the max number of errors
6357 			 */
6358 			bio->bi_status = BLK_STS_OK;
6359 		}
6360 
6361 		btrfs_end_bbio(bbio, bio);
6362 	} else if (!is_orig_bio) {
6363 		bio_put(bio);
6364 	}
6365 }
6366 
6367 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6368 			      u64 physical, struct btrfs_device *dev)
6369 {
6370 	struct btrfs_fs_info *fs_info = bbio->fs_info;
6371 
6372 	bio->bi_private = bbio;
6373 	btrfs_io_bio(bio)->device = dev;
6374 	bio->bi_end_io = btrfs_end_bio;
6375 	bio->bi_iter.bi_sector = physical >> 9;
6376 	btrfs_debug_in_rcu(fs_info,
6377 	"btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6378 		bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6379 		(unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6380 		dev->devid, bio->bi_iter.bi_size);
6381 	bio_set_dev(bio, dev->bdev);
6382 
6383 	btrfs_bio_counter_inc_noblocked(fs_info);
6384 
6385 	btrfsic_submit_bio(bio);
6386 }
6387 
6388 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6389 {
6390 	atomic_inc(&bbio->error);
6391 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
6392 		/* Should be the original bio. */
6393 		WARN_ON(bio != bbio->orig_bio);
6394 
6395 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6396 		bio->bi_iter.bi_sector = logical >> 9;
6397 		if (atomic_read(&bbio->error) > bbio->max_errors)
6398 			bio->bi_status = BLK_STS_IOERR;
6399 		else
6400 			bio->bi_status = BLK_STS_OK;
6401 		btrfs_end_bbio(bbio, bio);
6402 	}
6403 }
6404 
6405 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6406 			   int mirror_num)
6407 {
6408 	struct btrfs_device *dev;
6409 	struct bio *first_bio = bio;
6410 	u64 logical = bio->bi_iter.bi_sector << 9;
6411 	u64 length = 0;
6412 	u64 map_length;
6413 	int ret;
6414 	int dev_nr;
6415 	int total_devs;
6416 	struct btrfs_bio *bbio = NULL;
6417 
6418 	length = bio->bi_iter.bi_size;
6419 	map_length = length;
6420 
6421 	btrfs_bio_counter_inc_blocked(fs_info);
6422 	ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6423 				&map_length, &bbio, mirror_num, 1);
6424 	if (ret) {
6425 		btrfs_bio_counter_dec(fs_info);
6426 		return errno_to_blk_status(ret);
6427 	}
6428 
6429 	total_devs = bbio->num_stripes;
6430 	bbio->orig_bio = first_bio;
6431 	bbio->private = first_bio->bi_private;
6432 	bbio->end_io = first_bio->bi_end_io;
6433 	bbio->fs_info = fs_info;
6434 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6435 
6436 	if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6437 	    ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6438 		/* In this case, map_length has been set to the length of
6439 		   a single stripe; not the whole write */
6440 		if (bio_op(bio) == REQ_OP_WRITE) {
6441 			ret = raid56_parity_write(fs_info, bio, bbio,
6442 						  map_length);
6443 		} else {
6444 			ret = raid56_parity_recover(fs_info, bio, bbio,
6445 						    map_length, mirror_num, 1);
6446 		}
6447 
6448 		btrfs_bio_counter_dec(fs_info);
6449 		return errno_to_blk_status(ret);
6450 	}
6451 
6452 	if (map_length < length) {
6453 		btrfs_crit(fs_info,
6454 			   "mapping failed logical %llu bio len %llu len %llu",
6455 			   logical, length, map_length);
6456 		BUG();
6457 	}
6458 
6459 	for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6460 		dev = bbio->stripes[dev_nr].dev;
6461 		if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6462 						   &dev->dev_state) ||
6463 		    (bio_op(first_bio) == REQ_OP_WRITE &&
6464 		    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6465 			bbio_error(bbio, first_bio, logical);
6466 			continue;
6467 		}
6468 
6469 		if (dev_nr < total_devs - 1)
6470 			bio = btrfs_bio_clone(first_bio);
6471 		else
6472 			bio = first_bio;
6473 
6474 		submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6475 	}
6476 	btrfs_bio_counter_dec(fs_info);
6477 	return BLK_STS_OK;
6478 }
6479 
6480 /*
6481  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6482  * return NULL.
6483  *
6484  * If devid and uuid are both specified, the match must be exact, otherwise
6485  * only devid is used.
6486  *
6487  * If @seed is true, traverse through the seed devices.
6488  */
6489 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6490 				       u64 devid, u8 *uuid, u8 *fsid)
6491 {
6492 	struct btrfs_device *device;
6493 	struct btrfs_fs_devices *seed_devs;
6494 
6495 	if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6496 		list_for_each_entry(device, &fs_devices->devices, dev_list) {
6497 			if (device->devid == devid &&
6498 			    (!uuid || memcmp(device->uuid, uuid,
6499 					     BTRFS_UUID_SIZE) == 0))
6500 				return device;
6501 		}
6502 	}
6503 
6504 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6505 		if (!fsid ||
6506 		    !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6507 			list_for_each_entry(device, &seed_devs->devices,
6508 					    dev_list) {
6509 				if (device->devid == devid &&
6510 				    (!uuid || memcmp(device->uuid, uuid,
6511 						     BTRFS_UUID_SIZE) == 0))
6512 					return device;
6513 			}
6514 		}
6515 	}
6516 
6517 	return NULL;
6518 }
6519 
6520 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6521 					    u64 devid, u8 *dev_uuid)
6522 {
6523 	struct btrfs_device *device;
6524 	unsigned int nofs_flag;
6525 
6526 	/*
6527 	 * We call this under the chunk_mutex, so we want to use NOFS for this
6528 	 * allocation, however we don't want to change btrfs_alloc_device() to
6529 	 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6530 	 * places.
6531 	 */
6532 	nofs_flag = memalloc_nofs_save();
6533 	device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6534 	memalloc_nofs_restore(nofs_flag);
6535 	if (IS_ERR(device))
6536 		return device;
6537 
6538 	list_add(&device->dev_list, &fs_devices->devices);
6539 	device->fs_devices = fs_devices;
6540 	fs_devices->num_devices++;
6541 
6542 	set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6543 	fs_devices->missing_devices++;
6544 
6545 	return device;
6546 }
6547 
6548 /**
6549  * btrfs_alloc_device - allocate struct btrfs_device
6550  * @fs_info:	used only for generating a new devid, can be NULL if
6551  *		devid is provided (i.e. @devid != NULL).
6552  * @devid:	a pointer to devid for this device.  If NULL a new devid
6553  *		is generated.
6554  * @uuid:	a pointer to UUID for this device.  If NULL a new UUID
6555  *		is generated.
6556  *
6557  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6558  * on error.  Returned struct is not linked onto any lists and must be
6559  * destroyed with btrfs_free_device.
6560  */
6561 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6562 					const u64 *devid,
6563 					const u8 *uuid)
6564 {
6565 	struct btrfs_device *dev;
6566 	u64 tmp;
6567 
6568 	if (WARN_ON(!devid && !fs_info))
6569 		return ERR_PTR(-EINVAL);
6570 
6571 	dev = __alloc_device(fs_info);
6572 	if (IS_ERR(dev))
6573 		return dev;
6574 
6575 	if (devid)
6576 		tmp = *devid;
6577 	else {
6578 		int ret;
6579 
6580 		ret = find_next_devid(fs_info, &tmp);
6581 		if (ret) {
6582 			btrfs_free_device(dev);
6583 			return ERR_PTR(ret);
6584 		}
6585 	}
6586 	dev->devid = tmp;
6587 
6588 	if (uuid)
6589 		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6590 	else
6591 		generate_random_uuid(dev->uuid);
6592 
6593 	return dev;
6594 }
6595 
6596 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6597 					u64 devid, u8 *uuid, bool error)
6598 {
6599 	if (error)
6600 		btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6601 			      devid, uuid);
6602 	else
6603 		btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6604 			      devid, uuid);
6605 }
6606 
6607 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6608 {
6609 	int index = btrfs_bg_flags_to_raid_index(type);
6610 	int ncopies = btrfs_raid_array[index].ncopies;
6611 	const int nparity = btrfs_raid_array[index].nparity;
6612 	int data_stripes;
6613 
6614 	if (nparity)
6615 		data_stripes = num_stripes - nparity;
6616 	else
6617 		data_stripes = num_stripes / ncopies;
6618 
6619 	return div_u64(chunk_len, data_stripes);
6620 }
6621 
6622 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6623 			  struct btrfs_chunk *chunk)
6624 {
6625 	struct btrfs_fs_info *fs_info = leaf->fs_info;
6626 	struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6627 	struct map_lookup *map;
6628 	struct extent_map *em;
6629 	u64 logical;
6630 	u64 length;
6631 	u64 devid;
6632 	u8 uuid[BTRFS_UUID_SIZE];
6633 	int num_stripes;
6634 	int ret;
6635 	int i;
6636 
6637 	logical = key->offset;
6638 	length = btrfs_chunk_length(leaf, chunk);
6639 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6640 
6641 	/*
6642 	 * Only need to verify chunk item if we're reading from sys chunk array,
6643 	 * as chunk item in tree block is already verified by tree-checker.
6644 	 */
6645 	if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6646 		ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6647 		if (ret)
6648 			return ret;
6649 	}
6650 
6651 	read_lock(&map_tree->lock);
6652 	em = lookup_extent_mapping(map_tree, logical, 1);
6653 	read_unlock(&map_tree->lock);
6654 
6655 	/* already mapped? */
6656 	if (em && em->start <= logical && em->start + em->len > logical) {
6657 		free_extent_map(em);
6658 		return 0;
6659 	} else if (em) {
6660 		free_extent_map(em);
6661 	}
6662 
6663 	em = alloc_extent_map();
6664 	if (!em)
6665 		return -ENOMEM;
6666 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6667 	if (!map) {
6668 		free_extent_map(em);
6669 		return -ENOMEM;
6670 	}
6671 
6672 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6673 	em->map_lookup = map;
6674 	em->start = logical;
6675 	em->len = length;
6676 	em->orig_start = 0;
6677 	em->block_start = 0;
6678 	em->block_len = em->len;
6679 
6680 	map->num_stripes = num_stripes;
6681 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
6682 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
6683 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6684 	map->type = btrfs_chunk_type(leaf, chunk);
6685 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6686 	map->verified_stripes = 0;
6687 	em->orig_block_len = calc_stripe_length(map->type, em->len,
6688 						map->num_stripes);
6689 	for (i = 0; i < num_stripes; i++) {
6690 		map->stripes[i].physical =
6691 			btrfs_stripe_offset_nr(leaf, chunk, i);
6692 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6693 		read_extent_buffer(leaf, uuid, (unsigned long)
6694 				   btrfs_stripe_dev_uuid_nr(chunk, i),
6695 				   BTRFS_UUID_SIZE);
6696 		map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6697 							devid, uuid, NULL);
6698 		if (!map->stripes[i].dev &&
6699 		    !btrfs_test_opt(fs_info, DEGRADED)) {
6700 			free_extent_map(em);
6701 			btrfs_report_missing_device(fs_info, devid, uuid, true);
6702 			return -ENOENT;
6703 		}
6704 		if (!map->stripes[i].dev) {
6705 			map->stripes[i].dev =
6706 				add_missing_dev(fs_info->fs_devices, devid,
6707 						uuid);
6708 			if (IS_ERR(map->stripes[i].dev)) {
6709 				free_extent_map(em);
6710 				btrfs_err(fs_info,
6711 					"failed to init missing dev %llu: %ld",
6712 					devid, PTR_ERR(map->stripes[i].dev));
6713 				return PTR_ERR(map->stripes[i].dev);
6714 			}
6715 			btrfs_report_missing_device(fs_info, devid, uuid, false);
6716 		}
6717 		set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6718 				&(map->stripes[i].dev->dev_state));
6719 
6720 	}
6721 
6722 	write_lock(&map_tree->lock);
6723 	ret = add_extent_mapping(map_tree, em, 0);
6724 	write_unlock(&map_tree->lock);
6725 	if (ret < 0) {
6726 		btrfs_err(fs_info,
6727 			  "failed to add chunk map, start=%llu len=%llu: %d",
6728 			  em->start, em->len, ret);
6729 	}
6730 	free_extent_map(em);
6731 
6732 	return ret;
6733 }
6734 
6735 static void fill_device_from_item(struct extent_buffer *leaf,
6736 				 struct btrfs_dev_item *dev_item,
6737 				 struct btrfs_device *device)
6738 {
6739 	unsigned long ptr;
6740 
6741 	device->devid = btrfs_device_id(leaf, dev_item);
6742 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6743 	device->total_bytes = device->disk_total_bytes;
6744 	device->commit_total_bytes = device->disk_total_bytes;
6745 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6746 	device->commit_bytes_used = device->bytes_used;
6747 	device->type = btrfs_device_type(leaf, dev_item);
6748 	device->io_align = btrfs_device_io_align(leaf, dev_item);
6749 	device->io_width = btrfs_device_io_width(leaf, dev_item);
6750 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6751 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6752 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6753 
6754 	ptr = btrfs_device_uuid(dev_item);
6755 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6756 }
6757 
6758 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6759 						  u8 *fsid)
6760 {
6761 	struct btrfs_fs_devices *fs_devices;
6762 	int ret;
6763 
6764 	lockdep_assert_held(&uuid_mutex);
6765 	ASSERT(fsid);
6766 
6767 	/* This will match only for multi-device seed fs */
6768 	list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6769 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6770 			return fs_devices;
6771 
6772 
6773 	fs_devices = find_fsid(fsid, NULL);
6774 	if (!fs_devices) {
6775 		if (!btrfs_test_opt(fs_info, DEGRADED))
6776 			return ERR_PTR(-ENOENT);
6777 
6778 		fs_devices = alloc_fs_devices(fsid, NULL);
6779 		if (IS_ERR(fs_devices))
6780 			return fs_devices;
6781 
6782 		fs_devices->seeding = true;
6783 		fs_devices->opened = 1;
6784 		return fs_devices;
6785 	}
6786 
6787 	/*
6788 	 * Upon first call for a seed fs fsid, just create a private copy of the
6789 	 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6790 	 */
6791 	fs_devices = clone_fs_devices(fs_devices);
6792 	if (IS_ERR(fs_devices))
6793 		return fs_devices;
6794 
6795 	ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6796 	if (ret) {
6797 		free_fs_devices(fs_devices);
6798 		return ERR_PTR(ret);
6799 	}
6800 
6801 	if (!fs_devices->seeding) {
6802 		close_fs_devices(fs_devices);
6803 		free_fs_devices(fs_devices);
6804 		return ERR_PTR(-EINVAL);
6805 	}
6806 
6807 	list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6808 
6809 	return fs_devices;
6810 }
6811 
6812 static int read_one_dev(struct extent_buffer *leaf,
6813 			struct btrfs_dev_item *dev_item)
6814 {
6815 	struct btrfs_fs_info *fs_info = leaf->fs_info;
6816 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6817 	struct btrfs_device *device;
6818 	u64 devid;
6819 	int ret;
6820 	u8 fs_uuid[BTRFS_FSID_SIZE];
6821 	u8 dev_uuid[BTRFS_UUID_SIZE];
6822 
6823 	devid = btrfs_device_id(leaf, dev_item);
6824 	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6825 			   BTRFS_UUID_SIZE);
6826 	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6827 			   BTRFS_FSID_SIZE);
6828 
6829 	if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6830 		fs_devices = open_seed_devices(fs_info, fs_uuid);
6831 		if (IS_ERR(fs_devices))
6832 			return PTR_ERR(fs_devices);
6833 	}
6834 
6835 	device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6836 				   fs_uuid);
6837 	if (!device) {
6838 		if (!btrfs_test_opt(fs_info, DEGRADED)) {
6839 			btrfs_report_missing_device(fs_info, devid,
6840 							dev_uuid, true);
6841 			return -ENOENT;
6842 		}
6843 
6844 		device = add_missing_dev(fs_devices, devid, dev_uuid);
6845 		if (IS_ERR(device)) {
6846 			btrfs_err(fs_info,
6847 				"failed to add missing dev %llu: %ld",
6848 				devid, PTR_ERR(device));
6849 			return PTR_ERR(device);
6850 		}
6851 		btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6852 	} else {
6853 		if (!device->bdev) {
6854 			if (!btrfs_test_opt(fs_info, DEGRADED)) {
6855 				btrfs_report_missing_device(fs_info,
6856 						devid, dev_uuid, true);
6857 				return -ENOENT;
6858 			}
6859 			btrfs_report_missing_device(fs_info, devid,
6860 							dev_uuid, false);
6861 		}
6862 
6863 		if (!device->bdev &&
6864 		    !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6865 			/*
6866 			 * this happens when a device that was properly setup
6867 			 * in the device info lists suddenly goes bad.
6868 			 * device->bdev is NULL, and so we have to set
6869 			 * device->missing to one here
6870 			 */
6871 			device->fs_devices->missing_devices++;
6872 			set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6873 		}
6874 
6875 		/* Move the device to its own fs_devices */
6876 		if (device->fs_devices != fs_devices) {
6877 			ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6878 							&device->dev_state));
6879 
6880 			list_move(&device->dev_list, &fs_devices->devices);
6881 			device->fs_devices->num_devices--;
6882 			fs_devices->num_devices++;
6883 
6884 			device->fs_devices->missing_devices--;
6885 			fs_devices->missing_devices++;
6886 
6887 			device->fs_devices = fs_devices;
6888 		}
6889 	}
6890 
6891 	if (device->fs_devices != fs_info->fs_devices) {
6892 		BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6893 		if (device->generation !=
6894 		    btrfs_device_generation(leaf, dev_item))
6895 			return -EINVAL;
6896 	}
6897 
6898 	fill_device_from_item(leaf, dev_item, device);
6899 	if (device->bdev) {
6900 		u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
6901 
6902 		if (device->total_bytes > max_total_bytes) {
6903 			btrfs_err(fs_info,
6904 			"device total_bytes should be at most %llu but found %llu",
6905 				  max_total_bytes, device->total_bytes);
6906 			return -EINVAL;
6907 		}
6908 	}
6909 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6910 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6911 	   !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6912 		device->fs_devices->total_rw_bytes += device->total_bytes;
6913 		atomic64_add(device->total_bytes - device->bytes_used,
6914 				&fs_info->free_chunk_space);
6915 	}
6916 	ret = 0;
6917 	return ret;
6918 }
6919 
6920 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6921 {
6922 	struct btrfs_root *root = fs_info->tree_root;
6923 	struct btrfs_super_block *super_copy = fs_info->super_copy;
6924 	struct extent_buffer *sb;
6925 	struct btrfs_disk_key *disk_key;
6926 	struct btrfs_chunk *chunk;
6927 	u8 *array_ptr;
6928 	unsigned long sb_array_offset;
6929 	int ret = 0;
6930 	u32 num_stripes;
6931 	u32 array_size;
6932 	u32 len = 0;
6933 	u32 cur_offset;
6934 	u64 type;
6935 	struct btrfs_key key;
6936 
6937 	ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6938 	/*
6939 	 * This will create extent buffer of nodesize, superblock size is
6940 	 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6941 	 * overallocate but we can keep it as-is, only the first page is used.
6942 	 */
6943 	sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
6944 					  root->root_key.objectid, 0);
6945 	if (IS_ERR(sb))
6946 		return PTR_ERR(sb);
6947 	set_extent_buffer_uptodate(sb);
6948 	/*
6949 	 * The sb extent buffer is artificial and just used to read the system array.
6950 	 * set_extent_buffer_uptodate() call does not properly mark all it's
6951 	 * pages up-to-date when the page is larger: extent does not cover the
6952 	 * whole page and consequently check_page_uptodate does not find all
6953 	 * the page's extents up-to-date (the hole beyond sb),
6954 	 * write_extent_buffer then triggers a WARN_ON.
6955 	 *
6956 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6957 	 * but sb spans only this function. Add an explicit SetPageUptodate call
6958 	 * to silence the warning eg. on PowerPC 64.
6959 	 */
6960 	if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6961 		SetPageUptodate(sb->pages[0]);
6962 
6963 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6964 	array_size = btrfs_super_sys_array_size(super_copy);
6965 
6966 	array_ptr = super_copy->sys_chunk_array;
6967 	sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6968 	cur_offset = 0;
6969 
6970 	while (cur_offset < array_size) {
6971 		disk_key = (struct btrfs_disk_key *)array_ptr;
6972 		len = sizeof(*disk_key);
6973 		if (cur_offset + len > array_size)
6974 			goto out_short_read;
6975 
6976 		btrfs_disk_key_to_cpu(&key, disk_key);
6977 
6978 		array_ptr += len;
6979 		sb_array_offset += len;
6980 		cur_offset += len;
6981 
6982 		if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6983 			btrfs_err(fs_info,
6984 			    "unexpected item type %u in sys_array at offset %u",
6985 				  (u32)key.type, cur_offset);
6986 			ret = -EIO;
6987 			break;
6988 		}
6989 
6990 		chunk = (struct btrfs_chunk *)sb_array_offset;
6991 		/*
6992 		 * At least one btrfs_chunk with one stripe must be present,
6993 		 * exact stripe count check comes afterwards
6994 		 */
6995 		len = btrfs_chunk_item_size(1);
6996 		if (cur_offset + len > array_size)
6997 			goto out_short_read;
6998 
6999 		num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7000 		if (!num_stripes) {
7001 			btrfs_err(fs_info,
7002 			"invalid number of stripes %u in sys_array at offset %u",
7003 				  num_stripes, cur_offset);
7004 			ret = -EIO;
7005 			break;
7006 		}
7007 
7008 		type = btrfs_chunk_type(sb, chunk);
7009 		if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7010 			btrfs_err(fs_info,
7011 			"invalid chunk type %llu in sys_array at offset %u",
7012 				  type, cur_offset);
7013 			ret = -EIO;
7014 			break;
7015 		}
7016 
7017 		len = btrfs_chunk_item_size(num_stripes);
7018 		if (cur_offset + len > array_size)
7019 			goto out_short_read;
7020 
7021 		ret = read_one_chunk(&key, sb, chunk);
7022 		if (ret)
7023 			break;
7024 
7025 		array_ptr += len;
7026 		sb_array_offset += len;
7027 		cur_offset += len;
7028 	}
7029 	clear_extent_buffer_uptodate(sb);
7030 	free_extent_buffer_stale(sb);
7031 	return ret;
7032 
7033 out_short_read:
7034 	btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7035 			len, cur_offset);
7036 	clear_extent_buffer_uptodate(sb);
7037 	free_extent_buffer_stale(sb);
7038 	return -EIO;
7039 }
7040 
7041 /*
7042  * Check if all chunks in the fs are OK for read-write degraded mount
7043  *
7044  * If the @failing_dev is specified, it's accounted as missing.
7045  *
7046  * Return true if all chunks meet the minimal RW mount requirements.
7047  * Return false if any chunk doesn't meet the minimal RW mount requirements.
7048  */
7049 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7050 					struct btrfs_device *failing_dev)
7051 {
7052 	struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7053 	struct extent_map *em;
7054 	u64 next_start = 0;
7055 	bool ret = true;
7056 
7057 	read_lock(&map_tree->lock);
7058 	em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7059 	read_unlock(&map_tree->lock);
7060 	/* No chunk at all? Return false anyway */
7061 	if (!em) {
7062 		ret = false;
7063 		goto out;
7064 	}
7065 	while (em) {
7066 		struct map_lookup *map;
7067 		int missing = 0;
7068 		int max_tolerated;
7069 		int i;
7070 
7071 		map = em->map_lookup;
7072 		max_tolerated =
7073 			btrfs_get_num_tolerated_disk_barrier_failures(
7074 					map->type);
7075 		for (i = 0; i < map->num_stripes; i++) {
7076 			struct btrfs_device *dev = map->stripes[i].dev;
7077 
7078 			if (!dev || !dev->bdev ||
7079 			    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7080 			    dev->last_flush_error)
7081 				missing++;
7082 			else if (failing_dev && failing_dev == dev)
7083 				missing++;
7084 		}
7085 		if (missing > max_tolerated) {
7086 			if (!failing_dev)
7087 				btrfs_warn(fs_info,
7088 	"chunk %llu missing %d devices, max tolerance is %d for writable mount",
7089 				   em->start, missing, max_tolerated);
7090 			free_extent_map(em);
7091 			ret = false;
7092 			goto out;
7093 		}
7094 		next_start = extent_map_end(em);
7095 		free_extent_map(em);
7096 
7097 		read_lock(&map_tree->lock);
7098 		em = lookup_extent_mapping(map_tree, next_start,
7099 					   (u64)(-1) - next_start);
7100 		read_unlock(&map_tree->lock);
7101 	}
7102 out:
7103 	return ret;
7104 }
7105 
7106 static void readahead_tree_node_children(struct extent_buffer *node)
7107 {
7108 	int i;
7109 	const int nr_items = btrfs_header_nritems(node);
7110 
7111 	for (i = 0; i < nr_items; i++)
7112 		btrfs_readahead_node_child(node, i);
7113 }
7114 
7115 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7116 {
7117 	struct btrfs_root *root = fs_info->chunk_root;
7118 	struct btrfs_path *path;
7119 	struct extent_buffer *leaf;
7120 	struct btrfs_key key;
7121 	struct btrfs_key found_key;
7122 	int ret;
7123 	int slot;
7124 	u64 total_dev = 0;
7125 	u64 last_ra_node = 0;
7126 
7127 	path = btrfs_alloc_path();
7128 	if (!path)
7129 		return -ENOMEM;
7130 
7131 	/*
7132 	 * uuid_mutex is needed only if we are mounting a sprout FS
7133 	 * otherwise we don't need it.
7134 	 */
7135 	mutex_lock(&uuid_mutex);
7136 
7137 	/*
7138 	 * It is possible for mount and umount to race in such a way that
7139 	 * we execute this code path, but open_fs_devices failed to clear
7140 	 * total_rw_bytes. We certainly want it cleared before reading the
7141 	 * device items, so clear it here.
7142 	 */
7143 	fs_info->fs_devices->total_rw_bytes = 0;
7144 
7145 	/*
7146 	 * Read all device items, and then all the chunk items. All
7147 	 * device items are found before any chunk item (their object id
7148 	 * is smaller than the lowest possible object id for a chunk
7149 	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7150 	 */
7151 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7152 	key.offset = 0;
7153 	key.type = 0;
7154 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7155 	if (ret < 0)
7156 		goto error;
7157 	while (1) {
7158 		struct extent_buffer *node;
7159 
7160 		leaf = path->nodes[0];
7161 		slot = path->slots[0];
7162 		if (slot >= btrfs_header_nritems(leaf)) {
7163 			ret = btrfs_next_leaf(root, path);
7164 			if (ret == 0)
7165 				continue;
7166 			if (ret < 0)
7167 				goto error;
7168 			break;
7169 		}
7170 		/*
7171 		 * The nodes on level 1 are not locked but we don't need to do
7172 		 * that during mount time as nothing else can access the tree
7173 		 */
7174 		node = path->nodes[1];
7175 		if (node) {
7176 			if (last_ra_node != node->start) {
7177 				readahead_tree_node_children(node);
7178 				last_ra_node = node->start;
7179 			}
7180 		}
7181 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
7182 		if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7183 			struct btrfs_dev_item *dev_item;
7184 			dev_item = btrfs_item_ptr(leaf, slot,
7185 						  struct btrfs_dev_item);
7186 			ret = read_one_dev(leaf, dev_item);
7187 			if (ret)
7188 				goto error;
7189 			total_dev++;
7190 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7191 			struct btrfs_chunk *chunk;
7192 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7193 			mutex_lock(&fs_info->chunk_mutex);
7194 			ret = read_one_chunk(&found_key, leaf, chunk);
7195 			mutex_unlock(&fs_info->chunk_mutex);
7196 			if (ret)
7197 				goto error;
7198 		}
7199 		path->slots[0]++;
7200 	}
7201 
7202 	/*
7203 	 * After loading chunk tree, we've got all device information,
7204 	 * do another round of validation checks.
7205 	 */
7206 	if (total_dev != fs_info->fs_devices->total_devices) {
7207 		btrfs_err(fs_info,
7208 	   "super_num_devices %llu mismatch with num_devices %llu found here",
7209 			  btrfs_super_num_devices(fs_info->super_copy),
7210 			  total_dev);
7211 		ret = -EINVAL;
7212 		goto error;
7213 	}
7214 	if (btrfs_super_total_bytes(fs_info->super_copy) <
7215 	    fs_info->fs_devices->total_rw_bytes) {
7216 		btrfs_err(fs_info,
7217 	"super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7218 			  btrfs_super_total_bytes(fs_info->super_copy),
7219 			  fs_info->fs_devices->total_rw_bytes);
7220 		ret = -EINVAL;
7221 		goto error;
7222 	}
7223 	ret = 0;
7224 error:
7225 	mutex_unlock(&uuid_mutex);
7226 
7227 	btrfs_free_path(path);
7228 	return ret;
7229 }
7230 
7231 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7232 {
7233 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7234 	struct btrfs_device *device;
7235 
7236 	fs_devices->fs_info = fs_info;
7237 
7238 	mutex_lock(&fs_devices->device_list_mutex);
7239 	list_for_each_entry(device, &fs_devices->devices, dev_list)
7240 		device->fs_info = fs_info;
7241 
7242 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7243 		list_for_each_entry(device, &seed_devs->devices, dev_list)
7244 			device->fs_info = fs_info;
7245 
7246 		seed_devs->fs_info = fs_info;
7247 	}
7248 	mutex_unlock(&fs_devices->device_list_mutex);
7249 }
7250 
7251 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7252 				 const struct btrfs_dev_stats_item *ptr,
7253 				 int index)
7254 {
7255 	u64 val;
7256 
7257 	read_extent_buffer(eb, &val,
7258 			   offsetof(struct btrfs_dev_stats_item, values) +
7259 			    ((unsigned long)ptr) + (index * sizeof(u64)),
7260 			   sizeof(val));
7261 	return val;
7262 }
7263 
7264 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7265 				      struct btrfs_dev_stats_item *ptr,
7266 				      int index, u64 val)
7267 {
7268 	write_extent_buffer(eb, &val,
7269 			    offsetof(struct btrfs_dev_stats_item, values) +
7270 			     ((unsigned long)ptr) + (index * sizeof(u64)),
7271 			    sizeof(val));
7272 }
7273 
7274 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7275 				       struct btrfs_path *path)
7276 {
7277 	struct btrfs_dev_stats_item *ptr;
7278 	struct extent_buffer *eb;
7279 	struct btrfs_key key;
7280 	int item_size;
7281 	int i, ret, slot;
7282 
7283 	key.objectid = BTRFS_DEV_STATS_OBJECTID;
7284 	key.type = BTRFS_PERSISTENT_ITEM_KEY;
7285 	key.offset = device->devid;
7286 	ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7287 	if (ret) {
7288 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7289 			btrfs_dev_stat_set(device, i, 0);
7290 		device->dev_stats_valid = 1;
7291 		btrfs_release_path(path);
7292 		return ret < 0 ? ret : 0;
7293 	}
7294 	slot = path->slots[0];
7295 	eb = path->nodes[0];
7296 	item_size = btrfs_item_size_nr(eb, slot);
7297 
7298 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7299 
7300 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7301 		if (item_size >= (1 + i) * sizeof(__le64))
7302 			btrfs_dev_stat_set(device, i,
7303 					   btrfs_dev_stats_value(eb, ptr, i));
7304 		else
7305 			btrfs_dev_stat_set(device, i, 0);
7306 	}
7307 
7308 	device->dev_stats_valid = 1;
7309 	btrfs_dev_stat_print_on_load(device);
7310 	btrfs_release_path(path);
7311 
7312 	return 0;
7313 }
7314 
7315 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7316 {
7317 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7318 	struct btrfs_device *device;
7319 	struct btrfs_path *path = NULL;
7320 	int ret = 0;
7321 
7322 	path = btrfs_alloc_path();
7323 	if (!path)
7324 		return -ENOMEM;
7325 
7326 	mutex_lock(&fs_devices->device_list_mutex);
7327 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
7328 		ret = btrfs_device_init_dev_stats(device, path);
7329 		if (ret)
7330 			goto out;
7331 	}
7332 	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7333 		list_for_each_entry(device, &seed_devs->devices, dev_list) {
7334 			ret = btrfs_device_init_dev_stats(device, path);
7335 			if (ret)
7336 				goto out;
7337 		}
7338 	}
7339 out:
7340 	mutex_unlock(&fs_devices->device_list_mutex);
7341 
7342 	btrfs_free_path(path);
7343 	return ret;
7344 }
7345 
7346 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7347 				struct btrfs_device *device)
7348 {
7349 	struct btrfs_fs_info *fs_info = trans->fs_info;
7350 	struct btrfs_root *dev_root = fs_info->dev_root;
7351 	struct btrfs_path *path;
7352 	struct btrfs_key key;
7353 	struct extent_buffer *eb;
7354 	struct btrfs_dev_stats_item *ptr;
7355 	int ret;
7356 	int i;
7357 
7358 	key.objectid = BTRFS_DEV_STATS_OBJECTID;
7359 	key.type = BTRFS_PERSISTENT_ITEM_KEY;
7360 	key.offset = device->devid;
7361 
7362 	path = btrfs_alloc_path();
7363 	if (!path)
7364 		return -ENOMEM;
7365 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7366 	if (ret < 0) {
7367 		btrfs_warn_in_rcu(fs_info,
7368 			"error %d while searching for dev_stats item for device %s",
7369 			      ret, rcu_str_deref(device->name));
7370 		goto out;
7371 	}
7372 
7373 	if (ret == 0 &&
7374 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7375 		/* need to delete old one and insert a new one */
7376 		ret = btrfs_del_item(trans, dev_root, path);
7377 		if (ret != 0) {
7378 			btrfs_warn_in_rcu(fs_info,
7379 				"delete too small dev_stats item for device %s failed %d",
7380 				      rcu_str_deref(device->name), ret);
7381 			goto out;
7382 		}
7383 		ret = 1;
7384 	}
7385 
7386 	if (ret == 1) {
7387 		/* need to insert a new item */
7388 		btrfs_release_path(path);
7389 		ret = btrfs_insert_empty_item(trans, dev_root, path,
7390 					      &key, sizeof(*ptr));
7391 		if (ret < 0) {
7392 			btrfs_warn_in_rcu(fs_info,
7393 				"insert dev_stats item for device %s failed %d",
7394 				rcu_str_deref(device->name), ret);
7395 			goto out;
7396 		}
7397 	}
7398 
7399 	eb = path->nodes[0];
7400 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7401 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7402 		btrfs_set_dev_stats_value(eb, ptr, i,
7403 					  btrfs_dev_stat_read(device, i));
7404 	btrfs_mark_buffer_dirty(eb);
7405 
7406 out:
7407 	btrfs_free_path(path);
7408 	return ret;
7409 }
7410 
7411 /*
7412  * called from commit_transaction. Writes all changed device stats to disk.
7413  */
7414 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7415 {
7416 	struct btrfs_fs_info *fs_info = trans->fs_info;
7417 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7418 	struct btrfs_device *device;
7419 	int stats_cnt;
7420 	int ret = 0;
7421 
7422 	mutex_lock(&fs_devices->device_list_mutex);
7423 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
7424 		stats_cnt = atomic_read(&device->dev_stats_ccnt);
7425 		if (!device->dev_stats_valid || stats_cnt == 0)
7426 			continue;
7427 
7428 
7429 		/*
7430 		 * There is a LOAD-LOAD control dependency between the value of
7431 		 * dev_stats_ccnt and updating the on-disk values which requires
7432 		 * reading the in-memory counters. Such control dependencies
7433 		 * require explicit read memory barriers.
7434 		 *
7435 		 * This memory barriers pairs with smp_mb__before_atomic in
7436 		 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7437 		 * barrier implied by atomic_xchg in
7438 		 * btrfs_dev_stats_read_and_reset
7439 		 */
7440 		smp_rmb();
7441 
7442 		ret = update_dev_stat_item(trans, device);
7443 		if (!ret)
7444 			atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7445 	}
7446 	mutex_unlock(&fs_devices->device_list_mutex);
7447 
7448 	return ret;
7449 }
7450 
7451 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7452 {
7453 	btrfs_dev_stat_inc(dev, index);
7454 	btrfs_dev_stat_print_on_error(dev);
7455 }
7456 
7457 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7458 {
7459 	if (!dev->dev_stats_valid)
7460 		return;
7461 	btrfs_err_rl_in_rcu(dev->fs_info,
7462 		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7463 			   rcu_str_deref(dev->name),
7464 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7465 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7466 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7467 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7468 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7469 }
7470 
7471 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7472 {
7473 	int i;
7474 
7475 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7476 		if (btrfs_dev_stat_read(dev, i) != 0)
7477 			break;
7478 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
7479 		return; /* all values == 0, suppress message */
7480 
7481 	btrfs_info_in_rcu(dev->fs_info,
7482 		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7483 	       rcu_str_deref(dev->name),
7484 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7485 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7486 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7487 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7488 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7489 }
7490 
7491 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7492 			struct btrfs_ioctl_get_dev_stats *stats)
7493 {
7494 	struct btrfs_device *dev;
7495 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7496 	int i;
7497 
7498 	mutex_lock(&fs_devices->device_list_mutex);
7499 	dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7500 	mutex_unlock(&fs_devices->device_list_mutex);
7501 
7502 	if (!dev) {
7503 		btrfs_warn(fs_info, "get dev_stats failed, device not found");
7504 		return -ENODEV;
7505 	} else if (!dev->dev_stats_valid) {
7506 		btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7507 		return -ENODEV;
7508 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7509 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7510 			if (stats->nr_items > i)
7511 				stats->values[i] =
7512 					btrfs_dev_stat_read_and_reset(dev, i);
7513 			else
7514 				btrfs_dev_stat_set(dev, i, 0);
7515 		}
7516 		btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7517 			   current->comm, task_pid_nr(current));
7518 	} else {
7519 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7520 			if (stats->nr_items > i)
7521 				stats->values[i] = btrfs_dev_stat_read(dev, i);
7522 	}
7523 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7524 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7525 	return 0;
7526 }
7527 
7528 /*
7529  * Update the size and bytes used for each device where it changed.  This is
7530  * delayed since we would otherwise get errors while writing out the
7531  * superblocks.
7532  *
7533  * Must be invoked during transaction commit.
7534  */
7535 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7536 {
7537 	struct btrfs_device *curr, *next;
7538 
7539 	ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7540 
7541 	if (list_empty(&trans->dev_update_list))
7542 		return;
7543 
7544 	/*
7545 	 * We don't need the device_list_mutex here.  This list is owned by the
7546 	 * transaction and the transaction must complete before the device is
7547 	 * released.
7548 	 */
7549 	mutex_lock(&trans->fs_info->chunk_mutex);
7550 	list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7551 				 post_commit_list) {
7552 		list_del_init(&curr->post_commit_list);
7553 		curr->commit_total_bytes = curr->disk_total_bytes;
7554 		curr->commit_bytes_used = curr->bytes_used;
7555 	}
7556 	mutex_unlock(&trans->fs_info->chunk_mutex);
7557 }
7558 
7559 /*
7560  * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7561  */
7562 int btrfs_bg_type_to_factor(u64 flags)
7563 {
7564 	const int index = btrfs_bg_flags_to_raid_index(flags);
7565 
7566 	return btrfs_raid_array[index].ncopies;
7567 }
7568 
7569 
7570 
7571 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7572 				 u64 chunk_offset, u64 devid,
7573 				 u64 physical_offset, u64 physical_len)
7574 {
7575 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7576 	struct extent_map *em;
7577 	struct map_lookup *map;
7578 	struct btrfs_device *dev;
7579 	u64 stripe_len;
7580 	bool found = false;
7581 	int ret = 0;
7582 	int i;
7583 
7584 	read_lock(&em_tree->lock);
7585 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7586 	read_unlock(&em_tree->lock);
7587 
7588 	if (!em) {
7589 		btrfs_err(fs_info,
7590 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7591 			  physical_offset, devid);
7592 		ret = -EUCLEAN;
7593 		goto out;
7594 	}
7595 
7596 	map = em->map_lookup;
7597 	stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7598 	if (physical_len != stripe_len) {
7599 		btrfs_err(fs_info,
7600 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7601 			  physical_offset, devid, em->start, physical_len,
7602 			  stripe_len);
7603 		ret = -EUCLEAN;
7604 		goto out;
7605 	}
7606 
7607 	for (i = 0; i < map->num_stripes; i++) {
7608 		if (map->stripes[i].dev->devid == devid &&
7609 		    map->stripes[i].physical == physical_offset) {
7610 			found = true;
7611 			if (map->verified_stripes >= map->num_stripes) {
7612 				btrfs_err(fs_info,
7613 				"too many dev extents for chunk %llu found",
7614 					  em->start);
7615 				ret = -EUCLEAN;
7616 				goto out;
7617 			}
7618 			map->verified_stripes++;
7619 			break;
7620 		}
7621 	}
7622 	if (!found) {
7623 		btrfs_err(fs_info,
7624 	"dev extent physical offset %llu devid %llu has no corresponding chunk",
7625 			physical_offset, devid);
7626 		ret = -EUCLEAN;
7627 	}
7628 
7629 	/* Make sure no dev extent is beyond device bondary */
7630 	dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7631 	if (!dev) {
7632 		btrfs_err(fs_info, "failed to find devid %llu", devid);
7633 		ret = -EUCLEAN;
7634 		goto out;
7635 	}
7636 
7637 	if (physical_offset + physical_len > dev->disk_total_bytes) {
7638 		btrfs_err(fs_info,
7639 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7640 			  devid, physical_offset, physical_len,
7641 			  dev->disk_total_bytes);
7642 		ret = -EUCLEAN;
7643 		goto out;
7644 	}
7645 out:
7646 	free_extent_map(em);
7647 	return ret;
7648 }
7649 
7650 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7651 {
7652 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7653 	struct extent_map *em;
7654 	struct rb_node *node;
7655 	int ret = 0;
7656 
7657 	read_lock(&em_tree->lock);
7658 	for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7659 		em = rb_entry(node, struct extent_map, rb_node);
7660 		if (em->map_lookup->num_stripes !=
7661 		    em->map_lookup->verified_stripes) {
7662 			btrfs_err(fs_info,
7663 			"chunk %llu has missing dev extent, have %d expect %d",
7664 				  em->start, em->map_lookup->verified_stripes,
7665 				  em->map_lookup->num_stripes);
7666 			ret = -EUCLEAN;
7667 			goto out;
7668 		}
7669 	}
7670 out:
7671 	read_unlock(&em_tree->lock);
7672 	return ret;
7673 }
7674 
7675 /*
7676  * Ensure that all dev extents are mapped to correct chunk, otherwise
7677  * later chunk allocation/free would cause unexpected behavior.
7678  *
7679  * NOTE: This will iterate through the whole device tree, which should be of
7680  * the same size level as the chunk tree.  This slightly increases mount time.
7681  */
7682 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7683 {
7684 	struct btrfs_path *path;
7685 	struct btrfs_root *root = fs_info->dev_root;
7686 	struct btrfs_key key;
7687 	u64 prev_devid = 0;
7688 	u64 prev_dev_ext_end = 0;
7689 	int ret = 0;
7690 
7691 	/*
7692 	 * We don't have a dev_root because we mounted with ignorebadroots and
7693 	 * failed to load the root, so we want to skip the verification in this
7694 	 * case for sure.
7695 	 *
7696 	 * However if the dev root is fine, but the tree itself is corrupted
7697 	 * we'd still fail to mount.  This verification is only to make sure
7698 	 * writes can happen safely, so instead just bypass this check
7699 	 * completely in the case of IGNOREBADROOTS.
7700 	 */
7701 	if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7702 		return 0;
7703 
7704 	key.objectid = 1;
7705 	key.type = BTRFS_DEV_EXTENT_KEY;
7706 	key.offset = 0;
7707 
7708 	path = btrfs_alloc_path();
7709 	if (!path)
7710 		return -ENOMEM;
7711 
7712 	path->reada = READA_FORWARD;
7713 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7714 	if (ret < 0)
7715 		goto out;
7716 
7717 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7718 		ret = btrfs_next_item(root, path);
7719 		if (ret < 0)
7720 			goto out;
7721 		/* No dev extents at all? Not good */
7722 		if (ret > 0) {
7723 			ret = -EUCLEAN;
7724 			goto out;
7725 		}
7726 	}
7727 	while (1) {
7728 		struct extent_buffer *leaf = path->nodes[0];
7729 		struct btrfs_dev_extent *dext;
7730 		int slot = path->slots[0];
7731 		u64 chunk_offset;
7732 		u64 physical_offset;
7733 		u64 physical_len;
7734 		u64 devid;
7735 
7736 		btrfs_item_key_to_cpu(leaf, &key, slot);
7737 		if (key.type != BTRFS_DEV_EXTENT_KEY)
7738 			break;
7739 		devid = key.objectid;
7740 		physical_offset = key.offset;
7741 
7742 		dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7743 		chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7744 		physical_len = btrfs_dev_extent_length(leaf, dext);
7745 
7746 		/* Check if this dev extent overlaps with the previous one */
7747 		if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7748 			btrfs_err(fs_info,
7749 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7750 				  devid, physical_offset, prev_dev_ext_end);
7751 			ret = -EUCLEAN;
7752 			goto out;
7753 		}
7754 
7755 		ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7756 					    physical_offset, physical_len);
7757 		if (ret < 0)
7758 			goto out;
7759 		prev_devid = devid;
7760 		prev_dev_ext_end = physical_offset + physical_len;
7761 
7762 		ret = btrfs_next_item(root, path);
7763 		if (ret < 0)
7764 			goto out;
7765 		if (ret > 0) {
7766 			ret = 0;
7767 			break;
7768 		}
7769 	}
7770 
7771 	/* Ensure all chunks have corresponding dev extents */
7772 	ret = verify_chunk_dev_extent_mapping(fs_info);
7773 out:
7774 	btrfs_free_path(path);
7775 	return ret;
7776 }
7777 
7778 /*
7779  * Check whether the given block group or device is pinned by any inode being
7780  * used as a swapfile.
7781  */
7782 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7783 {
7784 	struct btrfs_swapfile_pin *sp;
7785 	struct rb_node *node;
7786 
7787 	spin_lock(&fs_info->swapfile_pins_lock);
7788 	node = fs_info->swapfile_pins.rb_node;
7789 	while (node) {
7790 		sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7791 		if (ptr < sp->ptr)
7792 			node = node->rb_left;
7793 		else if (ptr > sp->ptr)
7794 			node = node->rb_right;
7795 		else
7796 			break;
7797 	}
7798 	spin_unlock(&fs_info->swapfile_pins_lock);
7799 	return node != NULL;
7800 }
7801