brintos / linux-shallow public Read only

1// SPDX-License-Identifier: GPL-2.02/*3 * Copyright (C) 2007 Oracle.  All rights reserved.4 */5 6#include <linux/sched.h>7#include <linux/sched/mm.h>8#include <linux/slab.h>9#include <linux/ratelimit.h>10#include <linux/kthread.h>11#include <linux/semaphore.h>12#include <linux/uuid.h>13#include <linux/list_sort.h>14#include <linux/namei.h>15#include "misc.h"16#include "ctree.h"17#include "disk-io.h"18#include "transaction.h"19#include "volumes.h"20#include "raid56.h"21#include "rcu-string.h"22#include "dev-replace.h"23#include "sysfs.h"24#include "tree-checker.h"25#include "space-info.h"26#include "block-group.h"27#include "discard.h"28#include "zoned.h"29#include "fs.h"30#include "accessors.h"31#include "uuid-tree.h"32#include "ioctl.h"33#include "relocation.h"34#include "scrub.h"35#include "super.h"36#include "raid-stripe-tree.h"37 38#define BTRFS_BLOCK_GROUP_STRIPE_MASK	(BTRFS_BLOCK_GROUP_RAID0 | \39					 BTRFS_BLOCK_GROUP_RAID10 | \40					 BTRFS_BLOCK_GROUP_RAID56_MASK)41 42struct btrfs_io_geometry {43	u32 stripe_index;44	u32 stripe_nr;45	int mirror_num;46	int num_stripes;47	u64 stripe_offset;48	u64 raid56_full_stripe_start;49	int max_errors;50	enum btrfs_map_op op;51};52 53const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {54	[BTRFS_RAID_RAID10] = {55		.sub_stripes	= 2,56		.dev_stripes	= 1,57		.devs_max	= 0,	/* 0 == as many as possible */58		.devs_min	= 2,59		.tolerated_failures = 1,60		.devs_increment	= 2,61		.ncopies	= 2,62		.nparity        = 0,63		.raid_name	= "raid10",64		.bg_flag	= BTRFS_BLOCK_GROUP_RAID10,65		.mindev_error	= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,66	},67	[BTRFS_RAID_RAID1] = {68		.sub_stripes	= 1,69		.dev_stripes	= 1,70		.devs_max	= 2,71		.devs_min	= 2,72		.tolerated_failures = 1,73		.devs_increment	= 2,74		.ncopies	= 2,75		.nparity        = 0,76		.raid_name	= "raid1",77		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1,78		.mindev_error	= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,79	},80	[BTRFS_RAID_RAID1C3] = {81		.sub_stripes	= 1,82		.dev_stripes	= 1,83		.devs_max	= 3,84		.devs_min	= 3,85		.tolerated_failures = 2,86		.devs_increment	= 3,87		.ncopies	= 3,88		.nparity        = 0,89		.raid_name	= "raid1c3",90		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C3,91		.mindev_error	= BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,92	},93	[BTRFS_RAID_RAID1C4] = {94		.sub_stripes	= 1,95		.dev_stripes	= 1,96		.devs_max	= 4,97		.devs_min	= 4,98		.tolerated_failures = 3,99		.devs_increment	= 4,100		.ncopies	= 4,101		.nparity        = 0,102		.raid_name	= "raid1c4",103		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C4,104		.mindev_error	= BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,105	},106	[BTRFS_RAID_DUP] = {107		.sub_stripes	= 1,108		.dev_stripes	= 2,109		.devs_max	= 1,110		.devs_min	= 1,111		.tolerated_failures = 0,112		.devs_increment	= 1,113		.ncopies	= 2,114		.nparity        = 0,115		.raid_name	= "dup",116		.bg_flag	= BTRFS_BLOCK_GROUP_DUP,117		.mindev_error	= 0,118	},119	[BTRFS_RAID_RAID0] = {120		.sub_stripes	= 1,121		.dev_stripes	= 1,122		.devs_max	= 0,123		.devs_min	= 1,124		.tolerated_failures = 0,125		.devs_increment	= 1,126		.ncopies	= 1,127		.nparity        = 0,128		.raid_name	= "raid0",129		.bg_flag	= BTRFS_BLOCK_GROUP_RAID0,130		.mindev_error	= 0,131	},132	[BTRFS_RAID_SINGLE] = {133		.sub_stripes	= 1,134		.dev_stripes	= 1,135		.devs_max	= 1,136		.devs_min	= 1,137		.tolerated_failures = 0,138		.devs_increment	= 1,139		.ncopies	= 1,140		.nparity        = 0,141		.raid_name	= "single",142		.bg_flag	= 0,143		.mindev_error	= 0,144	},145	[BTRFS_RAID_RAID5] = {146		.sub_stripes	= 1,147		.dev_stripes	= 1,148		.devs_max	= 0,149		.devs_min	= 2,150		.tolerated_failures = 1,151		.devs_increment	= 1,152		.ncopies	= 1,153		.nparity        = 1,154		.raid_name	= "raid5",155		.bg_flag	= BTRFS_BLOCK_GROUP_RAID5,156		.mindev_error	= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,157	},158	[BTRFS_RAID_RAID6] = {159		.sub_stripes	= 1,160		.dev_stripes	= 1,161		.devs_max	= 0,162		.devs_min	= 3,163		.tolerated_failures = 2,164		.devs_increment	= 1,165		.ncopies	= 1,166		.nparity        = 2,167		.raid_name	= "raid6",168		.bg_flag	= BTRFS_BLOCK_GROUP_RAID6,169		.mindev_error	= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,170	},171};172 173/*174 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which175 * can be used as index to access btrfs_raid_array[].176 */177enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)178{179	const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);180 181	if (!profile)182		return BTRFS_RAID_SINGLE;183 184	return BTRFS_BG_FLAG_TO_INDEX(profile);185}186 187const char *btrfs_bg_type_to_raid_name(u64 flags)188{189	const int index = btrfs_bg_flags_to_raid_index(flags);190 191	if (index >= BTRFS_NR_RAID_TYPES)192		return NULL;193 194	return btrfs_raid_array[index].raid_name;195}196 197int btrfs_nr_parity_stripes(u64 type)198{199	enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);200 201	return btrfs_raid_array[index].nparity;202}203 204/*205 * Fill @buf with textual description of @bg_flags, no more than @size_buf206 * bytes including terminating null byte.207 */208void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)209{210	int i;211	int ret;212	char *bp = buf;213	u64 flags = bg_flags;214	u32 size_bp = size_buf;215 216	if (!flags) {217		strcpy(bp, "NONE");218		return;219	}220 221#define DESCRIBE_FLAG(flag, desc)						\222	do {								\223		if (flags & (flag)) {					\224			ret = snprintf(bp, size_bp, "%s|", (desc));	\225			if (ret < 0 || ret >= size_bp)			\226				goto out_overflow;			\227			size_bp -= ret;					\228			bp += ret;					\229			flags &= ~(flag);				\230		}							\231	} while (0)232 233	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");234	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");235	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");236 237	DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");238	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)239		DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,240			      btrfs_raid_array[i].raid_name);241#undef DESCRIBE_FLAG242 243	if (flags) {244		ret = snprintf(bp, size_bp, "0x%llx|", flags);245		size_bp -= ret;246	}247 248	if (size_bp < size_buf)249		buf[size_buf - size_bp - 1] = '\0'; /* remove last | */250 251	/*252	 * The text is trimmed, it's up to the caller to provide sufficiently253	 * large buffer254	 */255out_overflow:;256}257 258static int init_first_rw_device(struct btrfs_trans_handle *trans);259static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);260static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);261 262/*263 * Device locking264 * ==============265 *266 * There are several mutexes that protect manipulation of devices and low-level267 * structures like chunks but not block groups, extents or files268 *269 * uuid_mutex (global lock)270 * ------------------------271 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from272 * the SCAN_DEV ioctl registration or from mount either implicitly (the first273 * device) or requested by the device= mount option274 *275 * the mutex can be very coarse and can cover long-running operations276 *277 * protects: updates to fs_devices counters like missing devices, rw devices,278 * seeding, structure cloning, opening/closing devices at mount/umount time279 *280 * global::fs_devs - add, remove, updates to the global list281 *282 * does not protect: manipulation of the fs_devices::devices list in general283 * but in mount context it could be used to exclude list modifications by eg.284 * scan ioctl285 *286 * btrfs_device::name - renames (write side), read is RCU287 *288 * fs_devices::device_list_mutex (per-fs, with RCU)289 * ------------------------------------------------290 * protects updates to fs_devices::devices, ie. adding and deleting291 *292 * simple list traversal with read-only actions can be done with RCU protection293 *294 * may be used to exclude some operations from running concurrently without any295 * modifications to the list (see write_all_supers)296 *297 * Is not required at mount and close times, because our device list is298 * protected by the uuid_mutex at that point.299 *300 * balance_mutex301 * -------------302 * protects balance structures (status, state) and context accessed from303 * several places (internally, ioctl)304 *305 * chunk_mutex306 * -----------307 * protects chunks, adding or removing during allocation, trim or when a new308 * device is added/removed. Additionally it also protects post_commit_list of309 * individual devices, since they can be added to the transaction's310 * post_commit_list only with chunk_mutex held.311 *312 * cleaner_mutex313 * -------------314 * a big lock that is held by the cleaner thread and prevents running subvolume315 * cleaning together with relocation or delayed iputs316 *317 *318 * Lock nesting319 * ============320 *321 * uuid_mutex322 *   device_list_mutex323 *     chunk_mutex324 *   balance_mutex325 *326 *327 * Exclusive operations328 * ====================329 *330 * Maintains the exclusivity of the following operations that apply to the331 * whole filesystem and cannot run in parallel.332 *333 * - Balance (*)334 * - Device add335 * - Device remove336 * - Device replace (*)337 * - Resize338 *339 * The device operations (as above) can be in one of the following states:340 *341 * - Running state342 * - Paused state343 * - Completed state344 *345 * Only device operations marked with (*) can go into the Paused state for the346 * following reasons:347 *348 * - ioctl (only Balance can be Paused through ioctl)349 * - filesystem remounted as read-only350 * - filesystem unmounted and mounted as read-only351 * - system power-cycle and filesystem mounted as read-only352 * - filesystem or device errors leading to forced read-only353 *354 * The status of exclusive operation is set and cleared atomically.355 * During the course of Paused state, fs_info::exclusive_operation remains set.356 * A device operation in Paused or Running state can be canceled or resumed357 * either by ioctl (Balance only) or when remounted as read-write.358 * The exclusive status is cleared when the device operation is canceled or359 * completed.360 */361 362DEFINE_MUTEX(uuid_mutex);363static LIST_HEAD(fs_uuids);364struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)365{366	return &fs_uuids;367}368 369/*370 * Allocate new btrfs_fs_devices structure identified by a fsid.371 *372 * @fsid:    if not NULL, copy the UUID to fs_devices::fsid and to373 *           fs_devices::metadata_fsid374 *375 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().376 * The returned struct is not linked onto any lists and can be destroyed with377 * kfree() right away.378 */379static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)380{381	struct btrfs_fs_devices *fs_devs;382 383	fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);384	if (!fs_devs)385		return ERR_PTR(-ENOMEM);386 387	mutex_init(&fs_devs->device_list_mutex);388 389	INIT_LIST_HEAD(&fs_devs->devices);390	INIT_LIST_HEAD(&fs_devs->alloc_list);391	INIT_LIST_HEAD(&fs_devs->fs_list);392	INIT_LIST_HEAD(&fs_devs->seed_list);393 394	if (fsid) {395		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);396		memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);397	}398 399	return fs_devs;400}401 402static void btrfs_free_device(struct btrfs_device *device)403{404	WARN_ON(!list_empty(&device->post_commit_list));405	rcu_string_free(device->name);406	extent_io_tree_release(&device->alloc_state);407	btrfs_destroy_dev_zone_info(device);408	kfree(device);409}410 411static void free_fs_devices(struct btrfs_fs_devices *fs_devices)412{413	struct btrfs_device *device;414 415	WARN_ON(fs_devices->opened);416	while (!list_empty(&fs_devices->devices)) {417		device = list_entry(fs_devices->devices.next,418				    struct btrfs_device, dev_list);419		list_del(&device->dev_list);420		btrfs_free_device(device);421	}422	kfree(fs_devices);423}424 425void __exit btrfs_cleanup_fs_uuids(void)426{427	struct btrfs_fs_devices *fs_devices;428 429	while (!list_empty(&fs_uuids)) {430		fs_devices = list_entry(fs_uuids.next,431					struct btrfs_fs_devices, fs_list);432		list_del(&fs_devices->fs_list);433		free_fs_devices(fs_devices);434	}435}436 437static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices,438				  const u8 *fsid, const u8 *metadata_fsid)439{440	if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0)441		return false;442 443	if (!metadata_fsid)444		return true;445 446	if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0)447		return false;448 449	return true;450}451 452static noinline struct btrfs_fs_devices *find_fsid(453		const u8 *fsid, const u8 *metadata_fsid)454{455	struct btrfs_fs_devices *fs_devices;456 457	ASSERT(fsid);458 459	/* Handle non-split brain cases */460	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {461		if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid))462			return fs_devices;463	}464	return NULL;465}466 467static int468btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder,469		      int flush, struct file **bdev_file,470		      struct btrfs_super_block **disk_super)471{472	struct block_device *bdev;473	int ret;474 475	*bdev_file = bdev_file_open_by_path(device_path, flags, holder, NULL);476 477	if (IS_ERR(*bdev_file)) {478		ret = PTR_ERR(*bdev_file);479		btrfs_err(NULL, "failed to open device for path %s with flags 0x%x: %d",480			  device_path, flags, ret);481		goto error;482	}483	bdev = file_bdev(*bdev_file);484 485	if (flush)486		sync_blockdev(bdev);487	if (holder) {488		ret = set_blocksize(*bdev_file, BTRFS_BDEV_BLOCKSIZE);489		if (ret) {490			fput(*bdev_file);491			goto error;492		}493	}494	invalidate_bdev(bdev);495	*disk_super = btrfs_read_dev_super(bdev);496	if (IS_ERR(*disk_super)) {497		ret = PTR_ERR(*disk_super);498		fput(*bdev_file);499		goto error;500	}501 502	return 0;503 504error:505	*disk_super = NULL;506	*bdev_file = NULL;507	return ret;508}509 510/*511 *  Search and remove all stale devices (which are not mounted).  When both512 *  inputs are NULL, it will search and release all stale devices.513 *514 *  @devt:         Optional. When provided will it release all unmounted devices515 *                 matching this devt only.516 *  @skip_device:  Optional. Will skip this device when searching for the stale517 *                 devices.518 *519 *  Return:	0 for success or if @devt is 0.520 *		-EBUSY if @devt is a mounted device.521 *		-ENOENT if @devt does not match any device in the list.522 */523static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)524{525	struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;526	struct btrfs_device *device, *tmp_device;527	int ret;528	bool freed = false;529 530	lockdep_assert_held(&uuid_mutex);531 532	/* Return good status if there is no instance of devt. */533	ret = 0;534	list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {535 536		mutex_lock(&fs_devices->device_list_mutex);537		list_for_each_entry_safe(device, tmp_device,538					 &fs_devices->devices, dev_list) {539			if (skip_device && skip_device == device)540				continue;541			if (devt && devt != device->devt)542				continue;543			if (fs_devices->opened) {544				if (devt)545					ret = -EBUSY;546				break;547			}548 549			/* delete the stale device */550			fs_devices->num_devices--;551			list_del(&device->dev_list);552			btrfs_free_device(device);553 554			freed = true;555		}556		mutex_unlock(&fs_devices->device_list_mutex);557 558		if (fs_devices->num_devices == 0) {559			btrfs_sysfs_remove_fsid(fs_devices);560			list_del(&fs_devices->fs_list);561			free_fs_devices(fs_devices);562		}563	}564 565	/* If there is at least one freed device return 0. */566	if (freed)567		return 0;568 569	return ret;570}571 572static struct btrfs_fs_devices *find_fsid_by_device(573					struct btrfs_super_block *disk_super,574					dev_t devt, bool *same_fsid_diff_dev)575{576	struct btrfs_fs_devices *fsid_fs_devices;577	struct btrfs_fs_devices *devt_fs_devices;578	const bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &579					BTRFS_FEATURE_INCOMPAT_METADATA_UUID);580	bool found_by_devt = false;581 582	/* Find the fs_device by the usual method, if found use it. */583	fsid_fs_devices = find_fsid(disk_super->fsid,584		    has_metadata_uuid ? disk_super->metadata_uuid : NULL);585 586	/* The temp_fsid feature is supported only with single device filesystem. */587	if (btrfs_super_num_devices(disk_super) != 1)588		return fsid_fs_devices;589 590	/*591	 * A seed device is an integral component of the sprout device, which592	 * functions as a multi-device filesystem. So, temp-fsid feature is593	 * not supported.594	 */595	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)596		return fsid_fs_devices;597 598	/* Try to find a fs_devices by matching devt. */599	list_for_each_entry(devt_fs_devices, &fs_uuids, fs_list) {600		struct btrfs_device *device;601 602		list_for_each_entry(device, &devt_fs_devices->devices, dev_list) {603			if (device->devt == devt) {604				found_by_devt = true;605				break;606			}607		}608		if (found_by_devt)609			break;610	}611 612	if (found_by_devt) {613		/* Existing device. */614		if (fsid_fs_devices == NULL) {615			if (devt_fs_devices->opened == 0) {616				/* Stale device. */617				return NULL;618			} else {619				/* temp_fsid is mounting a subvol. */620				return devt_fs_devices;621			}622		} else {623			/* Regular or temp_fsid device mounting a subvol. */624			return devt_fs_devices;625		}626	} else {627		/* New device. */628		if (fsid_fs_devices == NULL) {629			return NULL;630		} else {631			/* sb::fsid is already used create a new temp_fsid. */632			*same_fsid_diff_dev = true;633			return NULL;634		}635	}636 637	/* Not reached. */638}639 640/*641 * This is only used on mount, and we are protected from competing things642 * messing with our fs_devices by the uuid_mutex, thus we do not need the643 * fs_devices->device_list_mutex here.644 */645static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,646			struct btrfs_device *device, blk_mode_t flags,647			void *holder)648{649	struct file *bdev_file;650	struct btrfs_super_block *disk_super;651	u64 devid;652	int ret;653 654	if (device->bdev)655		return -EINVAL;656	if (!device->name)657		return -EINVAL;658 659	ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,660				    &bdev_file, &disk_super);661	if (ret)662		return ret;663 664	devid = btrfs_stack_device_id(&disk_super->dev_item);665	if (devid != device->devid)666		goto error_free_page;667 668	if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))669		goto error_free_page;670 671	device->generation = btrfs_super_generation(disk_super);672 673	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {674		if (btrfs_super_incompat_flags(disk_super) &675		    BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {676			pr_err(677		"BTRFS: Invalid seeding and uuid-changed device detected\n");678			goto error_free_page;679		}680 681		clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);682		fs_devices->seeding = true;683	} else {684		if (bdev_read_only(file_bdev(bdev_file)))685			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);686		else687			set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);688	}689 690	if (!bdev_nonrot(file_bdev(bdev_file)))691		fs_devices->rotating = true;692 693	if (bdev_max_discard_sectors(file_bdev(bdev_file)))694		fs_devices->discardable = true;695 696	device->bdev_file = bdev_file;697	device->bdev = file_bdev(bdev_file);698	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);699 700	if (device->devt != device->bdev->bd_dev) {701		btrfs_warn(NULL,702			   "device %s maj:min changed from %d:%d to %d:%d",703			   device->name->str, MAJOR(device->devt),704			   MINOR(device->devt), MAJOR(device->bdev->bd_dev),705			   MINOR(device->bdev->bd_dev));706 707		device->devt = device->bdev->bd_dev;708	}709 710	fs_devices->open_devices++;711	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&712	    device->devid != BTRFS_DEV_REPLACE_DEVID) {713		fs_devices->rw_devices++;714		list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);715	}716	btrfs_release_disk_super(disk_super);717 718	return 0;719 720error_free_page:721	btrfs_release_disk_super(disk_super);722	fput(bdev_file);723 724	return -EINVAL;725}726 727const u8 *btrfs_sb_fsid_ptr(const struct btrfs_super_block *sb)728{729	bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) &730				  BTRFS_FEATURE_INCOMPAT_METADATA_UUID);731 732	return has_metadata_uuid ? sb->metadata_uuid : sb->fsid;733}734 735/*736 * Add new device to list of registered devices737 *738 * Returns:739 * device pointer which was just added or updated when successful740 * error pointer when failed741 */742static noinline struct btrfs_device *device_list_add(const char *path,743			   struct btrfs_super_block *disk_super,744			   bool *new_device_added)745{746	struct btrfs_device *device;747	struct btrfs_fs_devices *fs_devices = NULL;748	struct rcu_string *name;749	u64 found_transid = btrfs_super_generation(disk_super);750	u64 devid = btrfs_stack_device_id(&disk_super->dev_item);751	dev_t path_devt;752	int error;753	bool same_fsid_diff_dev = false;754	bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &755		BTRFS_FEATURE_INCOMPAT_METADATA_UUID);756 757	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {758		btrfs_err(NULL,759"device %s has incomplete metadata_uuid change, please use btrfstune to complete",760			  path);761		return ERR_PTR(-EAGAIN);762	}763 764	error = lookup_bdev(path, &path_devt);765	if (error) {766		btrfs_err(NULL, "failed to lookup block device for path %s: %d",767			  path, error);768		return ERR_PTR(error);769	}770 771	fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev);772 773	if (!fs_devices) {774		fs_devices = alloc_fs_devices(disk_super->fsid);775		if (IS_ERR(fs_devices))776			return ERR_CAST(fs_devices);777 778		if (has_metadata_uuid)779			memcpy(fs_devices->metadata_uuid,780			       disk_super->metadata_uuid, BTRFS_FSID_SIZE);781 782		if (same_fsid_diff_dev) {783			generate_random_uuid(fs_devices->fsid);784			fs_devices->temp_fsid = true;785		pr_info("BTRFS: device %s (%d:%d) using temp-fsid %pU\n",786				path, MAJOR(path_devt), MINOR(path_devt),787				fs_devices->fsid);788		}789 790		mutex_lock(&fs_devices->device_list_mutex);791		list_add(&fs_devices->fs_list, &fs_uuids);792 793		device = NULL;794	} else {795		struct btrfs_dev_lookup_args args = {796			.devid = devid,797			.uuid = disk_super->dev_item.uuid,798		};799 800		mutex_lock(&fs_devices->device_list_mutex);801		device = btrfs_find_device(fs_devices, &args);802 803		if (found_transid > fs_devices->latest_generation) {804			memcpy(fs_devices->fsid, disk_super->fsid,805					BTRFS_FSID_SIZE);806			memcpy(fs_devices->metadata_uuid,807			       btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE);808		}809	}810 811	if (!device) {812		unsigned int nofs_flag;813 814		if (fs_devices->opened) {815			btrfs_err(NULL,816"device %s (%d:%d) belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",817				  path, MAJOR(path_devt), MINOR(path_devt),818				  fs_devices->fsid, current->comm,819				  task_pid_nr(current));820			mutex_unlock(&fs_devices->device_list_mutex);821			return ERR_PTR(-EBUSY);822		}823 824		nofs_flag = memalloc_nofs_save();825		device = btrfs_alloc_device(NULL, &devid,826					    disk_super->dev_item.uuid, path);827		memalloc_nofs_restore(nofs_flag);828		if (IS_ERR(device)) {829			mutex_unlock(&fs_devices->device_list_mutex);830			/* we can safely leave the fs_devices entry around */831			return device;832		}833 834		device->devt = path_devt;835 836		list_add_rcu(&device->dev_list, &fs_devices->devices);837		fs_devices->num_devices++;838 839		device->fs_devices = fs_devices;840		*new_device_added = true;841 842		if (disk_super->label[0])843			pr_info(844"BTRFS: device label %s devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",845				disk_super->label, devid, found_transid, path,846				MAJOR(path_devt), MINOR(path_devt),847				current->comm, task_pid_nr(current));848		else849			pr_info(850"BTRFS: device fsid %pU devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",851				disk_super->fsid, devid, found_transid, path,852				MAJOR(path_devt), MINOR(path_devt),853				current->comm, task_pid_nr(current));854 855	} else if (!device->name || strcmp(device->name->str, path)) {856		/*857		 * When FS is already mounted.858		 * 1. If you are here and if the device->name is NULL that859		 *    means this device was missing at time of FS mount.860		 * 2. If you are here and if the device->name is different861		 *    from 'path' that means either862		 *      a. The same device disappeared and reappeared with863		 *         different name. or864		 *      b. The missing-disk-which-was-replaced, has865		 *         reappeared now.866		 *867		 * We must allow 1 and 2a above. But 2b would be a spurious868		 * and unintentional.869		 *870		 * Further in case of 1 and 2a above, the disk at 'path'871		 * would have missed some transaction when it was away and872		 * in case of 2a the stale bdev has to be updated as well.873		 * 2b must not be allowed at all time.874		 */875 876		/*877		 * For now, we do allow update to btrfs_fs_device through the878		 * btrfs dev scan cli after FS has been mounted.  We're still879		 * tracking a problem where systems fail mount by subvolume id880		 * when we reject replacement on a mounted FS.881		 */882		if (!fs_devices->opened && found_transid < device->generation) {883			/*884			 * That is if the FS is _not_ mounted and if you885			 * are here, that means there is more than one886			 * disk with same uuid and devid.We keep the one887			 * with larger generation number or the last-in if888			 * generation are equal.889			 */890			mutex_unlock(&fs_devices->device_list_mutex);891			btrfs_err(NULL,892"device %s already registered with a higher generation, found %llu expect %llu",893				  path, found_transid, device->generation);894			return ERR_PTR(-EEXIST);895		}896 897		/*898		 * We are going to replace the device path for a given devid,899		 * make sure it's the same device if the device is mounted900		 *901		 * NOTE: the device->fs_info may not be reliable here so pass902		 * in a NULL to message helpers instead. This avoids a possible903		 * use-after-free when the fs_info and fs_info->sb are already904		 * torn down.905		 */906		if (device->bdev) {907			if (device->devt != path_devt) {908				mutex_unlock(&fs_devices->device_list_mutex);909				btrfs_warn_in_rcu(NULL,910	"duplicate device %s devid %llu generation %llu scanned by %s (%d)",911						  path, devid, found_transid,912						  current->comm,913						  task_pid_nr(current));914				return ERR_PTR(-EEXIST);915			}916			btrfs_info_in_rcu(NULL,917	"devid %llu device path %s changed to %s scanned by %s (%d)",918					  devid, btrfs_dev_name(device),919					  path, current->comm,920					  task_pid_nr(current));921		}922 923		name = rcu_string_strdup(path, GFP_NOFS);924		if (!name) {925			mutex_unlock(&fs_devices->device_list_mutex);926			return ERR_PTR(-ENOMEM);927		}928		rcu_string_free(device->name);929		rcu_assign_pointer(device->name, name);930		if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {931			fs_devices->missing_devices--;932			clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);933		}934		device->devt = path_devt;935	}936 937	/*938	 * Unmount does not free the btrfs_device struct but would zero939	 * generation along with most of the other members. So just update940	 * it back. We need it to pick the disk with largest generation941	 * (as above).942	 */943	if (!fs_devices->opened) {944		device->generation = found_transid;945		fs_devices->latest_generation = max_t(u64, found_transid,946						fs_devices->latest_generation);947	}948 949	fs_devices->total_devices = btrfs_super_num_devices(disk_super);950 951	mutex_unlock(&fs_devices->device_list_mutex);952	return device;953}954 955static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)956{957	struct btrfs_fs_devices *fs_devices;958	struct btrfs_device *device;959	struct btrfs_device *orig_dev;960	int ret = 0;961 962	lockdep_assert_held(&uuid_mutex);963 964	fs_devices = alloc_fs_devices(orig->fsid);965	if (IS_ERR(fs_devices))966		return fs_devices;967 968	fs_devices->total_devices = orig->total_devices;969 970	list_for_each_entry(orig_dev, &orig->devices, dev_list) {971		const char *dev_path = NULL;972 973		/*974		 * This is ok to do without RCU read locked because we hold the975		 * uuid mutex so nothing we touch in here is going to disappear.976		 */977		if (orig_dev->name)978			dev_path = orig_dev->name->str;979 980		device = btrfs_alloc_device(NULL, &orig_dev->devid,981					    orig_dev->uuid, dev_path);982		if (IS_ERR(device)) {983			ret = PTR_ERR(device);984			goto error;985		}986 987		if (orig_dev->zone_info) {988			struct btrfs_zoned_device_info *zone_info;989 990			zone_info = btrfs_clone_dev_zone_info(orig_dev);991			if (!zone_info) {992				btrfs_free_device(device);993				ret = -ENOMEM;994				goto error;995			}996			device->zone_info = zone_info;997		}998 999		list_add(&device->dev_list, &fs_devices->devices);1000		device->fs_devices = fs_devices;1001		fs_devices->num_devices++;1002	}1003	return fs_devices;1004error:1005	free_fs_devices(fs_devices);1006	return ERR_PTR(ret);1007}1008 1009static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,1010				      struct btrfs_device **latest_dev)1011{1012	struct btrfs_device *device, *next;1013 1014	/* This is the initialized path, it is safe to release the devices. */1015	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {1016		if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {1017			if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,1018				      &device->dev_state) &&1019			    !test_bit(BTRFS_DEV_STATE_MISSING,1020				      &device->dev_state) &&1021			    (!*latest_dev ||1022			     device->generation > (*latest_dev)->generation)) {1023				*latest_dev = device;1024			}1025			continue;1026		}1027 1028		/*1029		 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,1030		 * in btrfs_init_dev_replace() so just continue.1031		 */1032		if (device->devid == BTRFS_DEV_REPLACE_DEVID)1033			continue;1034 1035		if (device->bdev_file) {1036			fput(device->bdev_file);1037			device->bdev = NULL;1038			device->bdev_file = NULL;1039			fs_devices->open_devices--;1040		}1041		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {1042			list_del_init(&device->dev_alloc_list);1043			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);1044			fs_devices->rw_devices--;1045		}1046		list_del_init(&device->dev_list);1047		fs_devices->num_devices--;1048		btrfs_free_device(device);1049	}1050 1051}1052 1053/*1054 * After we have read the system tree and know devids belonging to this1055 * filesystem, remove the device which does not belong there.1056 */1057void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)1058{1059	struct btrfs_device *latest_dev = NULL;1060	struct btrfs_fs_devices *seed_dev;1061 1062	mutex_lock(&uuid_mutex);1063	__btrfs_free_extra_devids(fs_devices, &latest_dev);1064 1065	list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)1066		__btrfs_free_extra_devids(seed_dev, &latest_dev);1067 1068	fs_devices->latest_dev = latest_dev;1069 1070	mutex_unlock(&uuid_mutex);1071}1072 1073static void btrfs_close_bdev(struct btrfs_device *device)1074{1075	if (!device->bdev)1076		return;1077 1078	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {1079		sync_blockdev(device->bdev);1080		invalidate_bdev(device->bdev);1081	}1082 1083	fput(device->bdev_file);1084}1085 1086static void btrfs_close_one_device(struct btrfs_device *device)1087{1088	struct btrfs_fs_devices *fs_devices = device->fs_devices;1089 1090	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&1091	    device->devid != BTRFS_DEV_REPLACE_DEVID) {1092		list_del_init(&device->dev_alloc_list);1093		fs_devices->rw_devices--;1094	}1095 1096	if (device->devid == BTRFS_DEV_REPLACE_DEVID)1097		clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);1098 1099	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {1100		clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);1101		fs_devices->missing_devices--;1102	}1103 1104	btrfs_close_bdev(device);1105	if (device->bdev) {1106		fs_devices->open_devices--;1107		device->bdev = NULL;1108		device->bdev_file = NULL;1109	}1110	clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);1111	btrfs_destroy_dev_zone_info(device);1112 1113	device->fs_info = NULL;1114	atomic_set(&device->dev_stats_ccnt, 0);1115	extent_io_tree_release(&device->alloc_state);1116 1117	/*1118	 * Reset the flush error record. We might have a transient flush error1119	 * in this mount, and if so we aborted the current transaction and set1120	 * the fs to an error state, guaranteeing no super blocks can be further1121	 * committed. However that error might be transient and if we unmount the1122	 * filesystem and mount it again, we should allow the mount to succeed1123	 * (btrfs_check_rw_degradable() should not fail) - if after mounting the1124	 * filesystem again we still get flush errors, then we will again abort1125	 * any transaction and set the error state, guaranteeing no commits of1126	 * unsafe super blocks.1127	 */1128	device->last_flush_error = 0;1129 1130	/* Verify the device is back in a pristine state  */1131	WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));1132	WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));1133	WARN_ON(!list_empty(&device->dev_alloc_list));1134	WARN_ON(!list_empty(&device->post_commit_list));1135}1136 1137static void close_fs_devices(struct btrfs_fs_devices *fs_devices)1138{1139	struct btrfs_device *device, *tmp;1140 1141	lockdep_assert_held(&uuid_mutex);1142 1143	if (--fs_devices->opened > 0)1144		return;1145 1146	list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)1147		btrfs_close_one_device(device);1148 1149	WARN_ON(fs_devices->open_devices);1150	WARN_ON(fs_devices->rw_devices);1151	fs_devices->opened = 0;1152	fs_devices->seeding = false;1153	fs_devices->fs_info = NULL;1154}1155 1156void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)1157{1158	LIST_HEAD(list);1159	struct btrfs_fs_devices *tmp;1160 1161	mutex_lock(&uuid_mutex);1162	close_fs_devices(fs_devices);1163	if (!fs_devices->opened) {1164		list_splice_init(&fs_devices->seed_list, &list);1165 1166		/*1167		 * If the struct btrfs_fs_devices is not assembled with any1168		 * other device, it can be re-initialized during the next mount1169		 * without the needing device-scan step. Therefore, it can be1170		 * fully freed.1171		 */1172		if (fs_devices->num_devices == 1) {1173			list_del(&fs_devices->fs_list);1174			free_fs_devices(fs_devices);1175		}1176	}1177 1178 1179	list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {1180		close_fs_devices(fs_devices);1181		list_del(&fs_devices->seed_list);1182		free_fs_devices(fs_devices);1183	}1184	mutex_unlock(&uuid_mutex);1185}1186 1187static int open_fs_devices(struct btrfs_fs_devices *fs_devices,1188				blk_mode_t flags, void *holder)1189{1190	struct btrfs_device *device;1191	struct btrfs_device *latest_dev = NULL;1192	struct btrfs_device *tmp_device;1193	int ret = 0;1194 1195	list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,1196				 dev_list) {1197		int ret2;1198 1199		ret2 = btrfs_open_one_device(fs_devices, device, flags, holder);1200		if (ret2 == 0 &&1201		    (!latest_dev || device->generation > latest_dev->generation)) {1202			latest_dev = device;1203		} else if (ret2 == -ENODATA) {1204			fs_devices->num_devices--;1205			list_del(&device->dev_list);1206			btrfs_free_device(device);1207		}1208		if (ret == 0 && ret2 != 0)1209			ret = ret2;1210	}1211 1212	if (fs_devices->open_devices == 0) {1213		if (ret)1214			return ret;1215		return -EINVAL;1216	}1217 1218	fs_devices->opened = 1;1219	fs_devices->latest_dev = latest_dev;1220	fs_devices->total_rw_bytes = 0;1221	fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;1222	fs_devices->read_policy = BTRFS_READ_POLICY_PID;1223 1224	return 0;1225}1226 1227static int devid_cmp(void *priv, const struct list_head *a,1228		     const struct list_head *b)1229{1230	const struct btrfs_device *dev1, *dev2;1231 1232	dev1 = list_entry(a, struct btrfs_device, dev_list);1233	dev2 = list_entry(b, struct btrfs_device, dev_list);1234 1235	if (dev1->devid < dev2->devid)1236		return -1;1237	else if (dev1->devid > dev2->devid)1238		return 1;1239	return 0;1240}1241 1242int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,1243		       blk_mode_t flags, void *holder)1244{1245	int ret;1246 1247	lockdep_assert_held(&uuid_mutex);1248	/*1249	 * The device_list_mutex cannot be taken here in case opening the1250	 * underlying device takes further locks like open_mutex.1251	 *1252	 * We also don't need the lock here as this is called during mount and1253	 * exclusion is provided by uuid_mutex1254	 */1255 1256	if (fs_devices->opened) {1257		fs_devices->opened++;1258		ret = 0;1259	} else {1260		list_sort(NULL, &fs_devices->devices, devid_cmp);1261		ret = open_fs_devices(fs_devices, flags, holder);1262	}1263 1264	return ret;1265}1266 1267void btrfs_release_disk_super(struct btrfs_super_block *super)1268{1269	struct page *page = virt_to_page(super);1270 1271	put_page(page);1272}1273 1274static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,1275						       u64 bytenr, u64 bytenr_orig)1276{1277	struct btrfs_super_block *disk_super;1278	struct page *page;1279	void *p;1280	pgoff_t index;1281 1282	/* make sure our super fits in the device */1283	if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))1284		return ERR_PTR(-EINVAL);1285 1286	/* make sure our super fits in the page */1287	if (sizeof(*disk_super) > PAGE_SIZE)1288		return ERR_PTR(-EINVAL);1289 1290	/* make sure our super doesn't straddle pages on disk */1291	index = bytenr >> PAGE_SHIFT;1292	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)1293		return ERR_PTR(-EINVAL);1294 1295	/* pull in the page with our super */1296	page = read_cache_page_gfp(bdev->bd_mapping, index, GFP_KERNEL);1297 1298	if (IS_ERR(page))1299		return ERR_CAST(page);1300 1301	p = page_address(page);1302 1303	/* align our pointer to the offset of the super block */1304	disk_super = p + offset_in_page(bytenr);1305 1306	if (btrfs_super_bytenr(disk_super) != bytenr_orig ||1307	    btrfs_super_magic(disk_super) != BTRFS_MAGIC) {1308		btrfs_release_disk_super(p);1309		return ERR_PTR(-EINVAL);1310	}1311 1312	if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])1313		disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;1314 1315	return disk_super;1316}1317 1318int btrfs_forget_devices(dev_t devt)1319{1320	int ret;1321 1322	mutex_lock(&uuid_mutex);1323	ret = btrfs_free_stale_devices(devt, NULL);1324	mutex_unlock(&uuid_mutex);1325 1326	return ret;1327}1328 1329static bool btrfs_skip_registration(struct btrfs_super_block *disk_super,1330				    const char *path, dev_t devt,1331				    bool mount_arg_dev)1332{1333	struct btrfs_fs_devices *fs_devices;1334 1335	/*1336	 * Do not skip device registration for mounted devices with matching1337	 * maj:min but different paths. Booting without initrd relies on1338	 * /dev/root initially, later replaced with the actual root device.1339	 * A successful scan ensures grub2-probe selects the correct device.1340	 */1341	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {1342		struct btrfs_device *device;1343 1344		mutex_lock(&fs_devices->device_list_mutex);1345 1346		if (!fs_devices->opened) {1347			mutex_unlock(&fs_devices->device_list_mutex);1348			continue;1349		}1350 1351		list_for_each_entry(device, &fs_devices->devices, dev_list) {1352			if (device->bdev && (device->bdev->bd_dev == devt) &&1353			    strcmp(device->name->str, path) != 0) {1354				mutex_unlock(&fs_devices->device_list_mutex);1355 1356				/* Do not skip registration. */1357				return false;1358			}1359		}1360		mutex_unlock(&fs_devices->device_list_mutex);1361	}1362 1363	if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 &&1364	    !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING))1365		return true;1366 1367	return false;1368}1369 1370/*1371 * Look for a btrfs signature on a device. This may be called out of the mount path1372 * and we are not allowed to call set_blocksize during the scan. The superblock1373 * is read via pagecache.1374 *1375 * With @mount_arg_dev it's a scan during mount time that will always register1376 * the device or return an error. Multi-device and seeding devices are registered1377 * in both cases.1378 */1379struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,1380					   bool mount_arg_dev)1381{1382	struct btrfs_super_block *disk_super;1383	bool new_device_added = false;1384	struct btrfs_device *device = NULL;1385	struct file *bdev_file;1386	u64 bytenr;1387	dev_t devt;1388	int ret;1389 1390	lockdep_assert_held(&uuid_mutex);1391 1392	/*1393	 * Avoid an exclusive open here, as the systemd-udev may initiate the1394	 * device scan which may race with the user's mount or mkfs command,1395	 * resulting in failure.1396	 * Since the device scan is solely for reading purposes, there is no1397	 * need for an exclusive open. Additionally, the devices are read again1398	 * during the mount process. It is ok to get some inconsistent1399	 * values temporarily, as the device paths of the fsid are the only1400	 * required information for assembling the volume.1401	 */1402	bdev_file = bdev_file_open_by_path(path, flags, NULL, NULL);1403	if (IS_ERR(bdev_file))1404		return ERR_CAST(bdev_file);1405 1406	/*1407	 * We would like to check all the super blocks, but doing so would1408	 * allow a mount to succeed after a mkfs from a different filesystem.1409	 * Currently, recovery from a bad primary btrfs superblock is done1410	 * using the userspace command 'btrfs check --super'.1411	 */1412	ret = btrfs_sb_log_location_bdev(file_bdev(bdev_file), 0, READ, &bytenr);1413	if (ret) {1414		device = ERR_PTR(ret);1415		goto error_bdev_put;1416	}1417 1418	disk_super = btrfs_read_disk_super(file_bdev(bdev_file), bytenr,1419					   btrfs_sb_offset(0));1420	if (IS_ERR(disk_super)) {1421		device = ERR_CAST(disk_super);1422		goto error_bdev_put;1423	}1424 1425	devt = file_bdev(bdev_file)->bd_dev;1426	if (btrfs_skip_registration(disk_super, path, devt, mount_arg_dev)) {1427		pr_debug("BTRFS: skip registering single non-seed device %s (%d:%d)\n",1428			  path, MAJOR(devt), MINOR(devt));1429 1430		btrfs_free_stale_devices(devt, NULL);1431 1432		device = NULL;1433		goto free_disk_super;1434	}1435 1436	device = device_list_add(path, disk_super, &new_device_added);1437	if (!IS_ERR(device) && new_device_added)1438		btrfs_free_stale_devices(device->devt, device);1439 1440free_disk_super:1441	btrfs_release_disk_super(disk_super);1442 1443error_bdev_put:1444	fput(bdev_file);1445 1446	return device;1447}1448 1449/*1450 * Try to find a chunk that intersects [start, start + len] range and when one1451 * such is found, record the end of it in *start1452 */1453static bool contains_pending_extent(struct btrfs_device *device, u64 *start,1454				    u64 len)1455{1456	u64 physical_start, physical_end;1457 1458	lockdep_assert_held(&device->fs_info->chunk_mutex);1459 1460	if (find_first_extent_bit(&device->alloc_state, *start,1461				  &physical_start, &physical_end,1462				  CHUNK_ALLOCATED, NULL)) {1463 1464		if (in_range(physical_start, *start, len) ||1465		    in_range(*start, physical_start,1466			     physical_end + 1 - physical_start)) {1467			*start = physical_end + 1;1468			return true;1469		}1470	}1471	return false;1472}1473 1474static u64 dev_extent_search_start(struct btrfs_device *device)1475{1476	switch (device->fs_devices->chunk_alloc_policy) {1477	case BTRFS_CHUNK_ALLOC_REGULAR:1478		return BTRFS_DEVICE_RANGE_RESERVED;1479	case BTRFS_CHUNK_ALLOC_ZONED:1480		/*1481		 * We don't care about the starting region like regular1482		 * allocator, because we anyway use/reserve the first two zones1483		 * for superblock logging.1484		 */1485		return 0;1486	default:1487		BUG();1488	}1489}1490 1491static bool dev_extent_hole_check_zoned(struct btrfs_device *device,1492					u64 *hole_start, u64 *hole_size,1493					u64 num_bytes)1494{1495	u64 zone_size = device->zone_info->zone_size;1496	u64 pos;1497	int ret;1498	bool changed = false;1499 1500	ASSERT(IS_ALIGNED(*hole_start, zone_size));1501 1502	while (*hole_size > 0) {1503		pos = btrfs_find_allocatable_zones(device, *hole_start,1504						   *hole_start + *hole_size,1505						   num_bytes);1506		if (pos != *hole_start) {1507			*hole_size = *hole_start + *hole_size - pos;1508			*hole_start = pos;1509			changed = true;1510			if (*hole_size < num_bytes)1511				break;1512		}1513 1514		ret = btrfs_ensure_empty_zones(device, pos, num_bytes);1515 1516		/* Range is ensured to be empty */1517		if (!ret)1518			return changed;1519 1520		/* Given hole range was invalid (outside of device) */1521		if (ret == -ERANGE) {1522			*hole_start += *hole_size;1523			*hole_size = 0;1524			return true;1525		}1526 1527		*hole_start += zone_size;1528		*hole_size -= zone_size;1529		changed = true;1530	}1531 1532	return changed;1533}1534 1535/*1536 * Check if specified hole is suitable for allocation.1537 *1538 * @device:	the device which we have the hole1539 * @hole_start: starting position of the hole1540 * @hole_size:	the size of the hole1541 * @num_bytes:	the size of the free space that we need1542 *1543 * This function may modify @hole_start and @hole_size to reflect the suitable1544 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.1545 */1546static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,1547				  u64 *hole_size, u64 num_bytes)1548{1549	bool changed = false;1550	u64 hole_end = *hole_start + *hole_size;1551 1552	for (;;) {1553		/*1554		 * Check before we set max_hole_start, otherwise we could end up1555		 * sending back this offset anyway.1556		 */1557		if (contains_pending_extent(device, hole_start, *hole_size)) {1558			if (hole_end >= *hole_start)1559				*hole_size = hole_end - *hole_start;1560			else1561				*hole_size = 0;1562			changed = true;1563		}1564 1565		switch (device->fs_devices->chunk_alloc_policy) {1566		case BTRFS_CHUNK_ALLOC_REGULAR:1567			/* No extra check */1568			break;1569		case BTRFS_CHUNK_ALLOC_ZONED:1570			if (dev_extent_hole_check_zoned(device, hole_start,1571							hole_size, num_bytes)) {1572				changed = true;1573				/*1574				 * The changed hole can contain pending extent.1575				 * Loop again to check that.1576				 */1577				continue;1578			}1579			break;1580		default:1581			BUG();1582		}1583 1584		break;1585	}1586 1587	return changed;1588}1589 1590/*1591 * Find free space in the specified device.1592 *1593 * @device:	  the device which we search the free space in1594 * @num_bytes:	  the size of the free space that we need1595 * @search_start: the position from which to begin the search1596 * @start:	  store the start of the free space.1597 * @len:	  the size of the free space. that we find, or the size1598 *		  of the max free space if we don't find suitable free space1599 *1600 * This does a pretty simple search, the expectation is that it is called very1601 * infrequently and that a given device has a small number of extents.1602 *1603 * @start is used to store the start of the free space if we find. But if we1604 * don't find suitable free space, it will be used to store the start position1605 * of the max free space.1606 *1607 * @len is used to store the size of the free space that we find.1608 * But if we don't find suitable free space, it is used to store the size of1609 * the max free space.1610 *1611 * NOTE: This function will search *commit* root of device tree, and does extra1612 * check to ensure dev extents are not double allocated.1613 * This makes the function safe to allocate dev extents but may not report1614 * correct usable device space, as device extent freed in current transaction1615 * is not reported as available.1616 */1617static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,1618				u64 *start, u64 *len)1619{1620	struct btrfs_fs_info *fs_info = device->fs_info;1621	struct btrfs_root *root = fs_info->dev_root;1622	struct btrfs_key key;1623	struct btrfs_dev_extent *dev_extent;1624	struct btrfs_path *path;1625	u64 search_start;1626	u64 hole_size;1627	u64 max_hole_start;1628	u64 max_hole_size = 0;1629	u64 extent_end;1630	u64 search_end = device->total_bytes;1631	int ret;1632	int slot;1633	struct extent_buffer *l;1634 1635	search_start = dev_extent_search_start(device);1636	max_hole_start = search_start;1637 1638	WARN_ON(device->zone_info &&1639		!IS_ALIGNED(num_bytes, device->zone_info->zone_size));1640 1641	path = btrfs_alloc_path();1642	if (!path) {1643		ret = -ENOMEM;1644		goto out;1645	}1646again:1647	if (search_start >= search_end ||1648		test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {1649		ret = -ENOSPC;1650		goto out;1651	}1652 1653	path->reada = READA_FORWARD;1654	path->search_commit_root = 1;1655	path->skip_locking = 1;1656 1657	key.objectid = device->devid;1658	key.offset = search_start;1659	key.type = BTRFS_DEV_EXTENT_KEY;1660 1661	ret = btrfs_search_backwards(root, &key, path);1662	if (ret < 0)1663		goto out;1664 1665	while (search_start < search_end) {1666		l = path->nodes[0];1667		slot = path->slots[0];1668		if (slot >= btrfs_header_nritems(l)) {1669			ret = btrfs_next_leaf(root, path);1670			if (ret == 0)1671				continue;1672			if (ret < 0)1673				goto out;1674 1675			break;1676		}1677		btrfs_item_key_to_cpu(l, &key, slot);1678 1679		if (key.objectid < device->devid)1680			goto next;1681 1682		if (key.objectid > device->devid)1683			break;1684 1685		if (key.type != BTRFS_DEV_EXTENT_KEY)1686			goto next;1687 1688		if (key.offset > search_end)1689			break;1690 1691		if (key.offset > search_start) {1692			hole_size = key.offset - search_start;1693			dev_extent_hole_check(device, &search_start, &hole_size,1694					      num_bytes);1695 1696			if (hole_size > max_hole_size) {1697				max_hole_start = search_start;1698				max_hole_size = hole_size;1699			}1700 1701			/*1702			 * If this free space is greater than which we need,1703			 * it must be the max free space that we have found1704			 * until now, so max_hole_start must point to the start1705			 * of this free space and the length of this free space1706			 * is stored in max_hole_size. Thus, we return1707			 * max_hole_start and max_hole_size and go back to the1708			 * caller.1709			 */1710			if (hole_size >= num_bytes) {1711				ret = 0;1712				goto out;1713			}1714		}1715 1716		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);1717		extent_end = key.offset + btrfs_dev_extent_length(l,1718								  dev_extent);1719		if (extent_end > search_start)1720			search_start = extent_end;1721next:1722		path->slots[0]++;1723		cond_resched();1724	}1725 1726	/*1727	 * At this point, search_start should be the end of1728	 * allocated dev extents, and when shrinking the device,1729	 * search_end may be smaller than search_start.1730	 */1731	if (search_end > search_start) {1732		hole_size = search_end - search_start;1733		if (dev_extent_hole_check(device, &search_start, &hole_size,1734					  num_bytes)) {1735			btrfs_release_path(path);1736			goto again;1737		}1738 1739		if (hole_size > max_hole_size) {1740			max_hole_start = search_start;1741			max_hole_size = hole_size;1742		}1743	}1744 1745	/* See above. */1746	if (max_hole_size < num_bytes)1747		ret = -ENOSPC;1748	else1749		ret = 0;1750 1751	ASSERT(max_hole_start + max_hole_size <= search_end);1752out:1753	btrfs_free_path(path);1754	*start = max_hole_start;1755	if (len)1756		*len = max_hole_size;1757	return ret;1758}1759 1760static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,1761			  struct btrfs_device *device,1762			  u64 start, u64 *dev_extent_len)1763{1764	struct btrfs_fs_info *fs_info = device->fs_info;1765	struct btrfs_root *root = fs_info->dev_root;1766	int ret;1767	struct btrfs_path *path;1768	struct btrfs_key key;1769	struct btrfs_key found_key;1770	struct extent_buffer *leaf = NULL;1771	struct btrfs_dev_extent *extent = NULL;1772 1773	path = btrfs_alloc_path();1774	if (!path)1775		return -ENOMEM;1776 1777	key.objectid = device->devid;1778	key.offset = start;1779	key.type = BTRFS_DEV_EXTENT_KEY;1780again:1781	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);1782	if (ret > 0) {1783		ret = btrfs_previous_item(root, path, key.objectid,1784					  BTRFS_DEV_EXTENT_KEY);1785		if (ret)1786			goto out;1787		leaf = path->nodes[0];1788		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);1789		extent = btrfs_item_ptr(leaf, path->slots[0],1790					struct btrfs_dev_extent);1791		BUG_ON(found_key.offset > start || found_key.offset +1792		       btrfs_dev_extent_length(leaf, extent) < start);1793		key = found_key;1794		btrfs_release_path(path);1795		goto again;1796	} else if (ret == 0) {1797		leaf = path->nodes[0];1798		extent = btrfs_item_ptr(leaf, path->slots[0],1799					struct btrfs_dev_extent);1800	} else {1801		goto out;1802	}1803 1804	*dev_extent_len = btrfs_dev_extent_length(leaf, extent);1805 1806	ret = btrfs_del_item(trans, root, path);1807	if (ret == 0)1808		set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);1809out:1810	btrfs_free_path(path);1811	return ret;1812}1813 1814static u64 find_next_chunk(struct btrfs_fs_info *fs_info)1815{1816	struct rb_node *n;1817	u64 ret = 0;1818 1819	read_lock(&fs_info->mapping_tree_lock);1820	n = rb_last(&fs_info->mapping_tree.rb_root);1821	if (n) {1822		struct btrfs_chunk_map *map;1823 1824		map = rb_entry(n, struct btrfs_chunk_map, rb_node);1825		ret = map->start + map->chunk_len;1826	}1827	read_unlock(&fs_info->mapping_tree_lock);1828 1829	return ret;1830}1831 1832static noinline int find_next_devid(struct btrfs_fs_info *fs_info,1833				    u64 *devid_ret)1834{1835	int ret;1836	struct btrfs_key key;1837	struct btrfs_key found_key;1838	struct btrfs_path *path;1839 1840	path = btrfs_alloc_path();1841	if (!path)1842		return -ENOMEM;1843 1844	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;1845	key.type = BTRFS_DEV_ITEM_KEY;1846	key.offset = (u64)-1;1847 1848	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);1849	if (ret < 0)1850		goto error;1851 1852	if (ret == 0) {1853		/* Corruption */1854		btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");1855		ret = -EUCLEAN;1856		goto error;1857	}1858 1859	ret = btrfs_previous_item(fs_info->chunk_root, path,1860				  BTRFS_DEV_ITEMS_OBJECTID,1861				  BTRFS_DEV_ITEM_KEY);1862	if (ret) {1863		*devid_ret = 1;1864	} else {1865		btrfs_item_key_to_cpu(path->nodes[0], &found_key,1866				      path->slots[0]);1867		*devid_ret = found_key.offset + 1;1868	}1869	ret = 0;1870error:1871	btrfs_free_path(path);1872	return ret;1873}1874 1875/*1876 * the device information is stored in the chunk root1877 * the btrfs_device struct should be fully filled in1878 */1879static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,1880			    struct btrfs_device *device)1881{1882	int ret;1883	struct btrfs_path *path;1884	struct btrfs_dev_item *dev_item;1885	struct extent_buffer *leaf;1886	struct btrfs_key key;1887	unsigned long ptr;1888 1889	path = btrfs_alloc_path();1890	if (!path)1891		return -ENOMEM;1892 1893	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;1894	key.type = BTRFS_DEV_ITEM_KEY;1895	key.offset = device->devid;1896 1897	btrfs_reserve_chunk_metadata(trans, true);1898	ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,1899				      &key, sizeof(*dev_item));1900	btrfs_trans_release_chunk_metadata(trans);1901	if (ret)1902		goto out;1903 1904	leaf = path->nodes[0];1905	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);1906 1907	btrfs_set_device_id(leaf, dev_item, device->devid);1908	btrfs_set_device_generation(leaf, dev_item, 0);1909	btrfs_set_device_type(leaf, dev_item, device->type);1910	btrfs_set_device_io_align(leaf, dev_item, device->io_align);1911	btrfs_set_device_io_width(leaf, dev_item, device->io_width);1912	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);1913	btrfs_set_device_total_bytes(leaf, dev_item,1914				     btrfs_device_get_disk_total_bytes(device));1915	btrfs_set_device_bytes_used(leaf, dev_item,1916				    btrfs_device_get_bytes_used(device));1917	btrfs_set_device_group(leaf, dev_item, 0);1918	btrfs_set_device_seek_speed(leaf, dev_item, 0);1919	btrfs_set_device_bandwidth(leaf, dev_item, 0);1920	btrfs_set_device_start_offset(leaf, dev_item, 0);1921 1922	ptr = btrfs_device_uuid(dev_item);1923	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);1924	ptr = btrfs_device_fsid(dev_item);1925	write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,1926			    ptr, BTRFS_FSID_SIZE);1927	btrfs_mark_buffer_dirty(trans, leaf);1928 1929	ret = 0;1930out:1931	btrfs_free_path(path);1932	return ret;1933}1934 1935/*1936 * Function to update ctime/mtime for a given device path.1937 * Mainly used for ctime/mtime based probe like libblkid.1938 *1939 * We don't care about errors here, this is just to be kind to userspace.1940 */1941static void update_dev_time(const char *device_path)1942{1943	struct path path;1944	int ret;1945 1946	ret = kern_path(device_path, LOOKUP_FOLLOW, &path);1947	if (ret)1948		return;1949 1950	inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);1951	path_put(&path);1952}1953 1954static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,1955			     struct btrfs_device *device)1956{1957	struct btrfs_root *root = device->fs_info->chunk_root;1958	int ret;1959	struct btrfs_path *path;1960	struct btrfs_key key;1961 1962	path = btrfs_alloc_path();1963	if (!path)1964		return -ENOMEM;1965 1966	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;1967	key.type = BTRFS_DEV_ITEM_KEY;1968	key.offset = device->devid;1969 1970	btrfs_reserve_chunk_metadata(trans, false);1971	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);1972	btrfs_trans_release_chunk_metadata(trans);1973	if (ret) {1974		if (ret > 0)1975			ret = -ENOENT;1976		goto out;1977	}1978 1979	ret = btrfs_del_item(trans, root, path);1980out:1981	btrfs_free_path(path);1982	return ret;1983}1984 1985/*1986 * Verify that @num_devices satisfies the RAID profile constraints in the whole1987 * filesystem. It's up to the caller to adjust that number regarding eg. device1988 * replace.1989 */1990static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,1991		u64 num_devices)1992{1993	u64 all_avail;1994	unsigned seq;1995	int i;1996 1997	do {1998		seq = read_seqbegin(&fs_info->profiles_lock);1999 2000		all_avail = fs_info->avail_data_alloc_bits |2001			    fs_info->avail_system_alloc_bits |2002			    fs_info->avail_metadata_alloc_bits;2003	} while (read_seqretry(&fs_info->profiles_lock, seq));2004 2005	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {2006		if (!(all_avail & btrfs_raid_array[i].bg_flag))2007			continue;2008 2009		if (num_devices < btrfs_raid_array[i].devs_min)2010			return btrfs_raid_array[i].mindev_error;2011	}2012 2013	return 0;2014}2015 2016static struct btrfs_device * btrfs_find_next_active_device(2017		struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)2018{2019	struct btrfs_device *next_device;2020 2021	list_for_each_entry(next_device, &fs_devs->devices, dev_list) {2022		if (next_device != device &&2023		    !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)2024		    && next_device->bdev)2025			return next_device;2026	}2027 2028	return NULL;2029}2030 2031/*2032 * Helper function to check if the given device is part of s_bdev / latest_dev2033 * and replace it with the provided or the next active device, in the context2034 * where this function called, there should be always be another device (or2035 * this_dev) which is active.2036 */2037void __cold btrfs_assign_next_active_device(struct btrfs_device *device,2038					    struct btrfs_device *next_device)2039{2040	struct btrfs_fs_info *fs_info = device->fs_info;2041 2042	if (!next_device)2043		next_device = btrfs_find_next_active_device(fs_info->fs_devices,2044							    device);2045	ASSERT(next_device);2046 2047	if (fs_info->sb->s_bdev &&2048			(fs_info->sb->s_bdev == device->bdev))2049		fs_info->sb->s_bdev = next_device->bdev;2050 2051	if (fs_info->fs_devices->latest_dev->bdev == device->bdev)2052		fs_info->fs_devices->latest_dev = next_device;2053}2054 2055/*2056 * Return btrfs_fs_devices::num_devices excluding the device that's being2057 * currently replaced.2058 */2059static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)2060{2061	u64 num_devices = fs_info->fs_devices->num_devices;2062 2063	down_read(&fs_info->dev_replace.rwsem);2064	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {2065		ASSERT(num_devices > 1);2066		num_devices--;2067	}2068	up_read(&fs_info->dev_replace.rwsem);2069 2070	return num_devices;2071}2072 2073static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,2074				     struct block_device *bdev, int copy_num)2075{2076	struct btrfs_super_block *disk_super;2077	const size_t len = sizeof(disk_super->magic);2078	const u64 bytenr = btrfs_sb_offset(copy_num);2079	int ret;2080 2081	disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);2082	if (IS_ERR(disk_super))2083		return;2084 2085	memset(&disk_super->magic, 0, len);2086	folio_mark_dirty(virt_to_folio(disk_super));2087	btrfs_release_disk_super(disk_super);2088 2089	ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);2090	if (ret)2091		btrfs_warn(fs_info, "error clearing superblock number %d (%d)",2092			copy_num, ret);2093}2094 2095void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device)2096{2097	int copy_num;2098	struct block_device *bdev = device->bdev;2099 2100	if (!bdev)2101		return;2102 2103	for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {2104		if (bdev_is_zoned(bdev))2105			btrfs_reset_sb_log_zones(bdev, copy_num);2106		else2107			btrfs_scratch_superblock(fs_info, bdev, copy_num);2108	}2109 2110	/* Notify udev that device has changed */2111	btrfs_kobject_uevent(bdev, KOBJ_CHANGE);2112 2113	/* Update ctime/mtime for device path for libblkid */2114	update_dev_time(device->name->str);2115}2116 2117int btrfs_rm_device(struct btrfs_fs_info *fs_info,2118		    struct btrfs_dev_lookup_args *args,2119		    struct file **bdev_file)2120{2121	struct btrfs_trans_handle *trans;2122	struct btrfs_device *device;2123	struct btrfs_fs_devices *cur_devices;2124	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;2125	u64 num_devices;2126	int ret = 0;2127 2128	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {2129		btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");2130		return -EINVAL;2131	}2132 2133	/*2134	 * The device list in fs_devices is accessed without locks (neither2135	 * uuid_mutex nor device_list_mutex) as it won't change on a mounted2136	 * filesystem and another device rm cannot run.2137	 */2138	num_devices = btrfs_num_devices(fs_info);2139 2140	ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);2141	if (ret)2142		return ret;2143 2144	device = btrfs_find_device(fs_info->fs_devices, args);2145	if (!device) {2146		if (args->missing)2147			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;2148		else2149			ret = -ENOENT;2150		return ret;2151	}2152 2153	if (btrfs_pinned_by_swapfile(fs_info, device)) {2154		btrfs_warn_in_rcu(fs_info,2155		  "cannot remove device %s (devid %llu) due to active swapfile",2156				  btrfs_dev_name(device), device->devid);2157		return -ETXTBSY;2158	}2159 2160	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))2161		return BTRFS_ERROR_DEV_TGT_REPLACE;2162 2163	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&2164	    fs_info->fs_devices->rw_devices == 1)2165		return BTRFS_ERROR_DEV_ONLY_WRITABLE;2166 2167	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {2168		mutex_lock(&fs_info->chunk_mutex);2169		list_del_init(&device->dev_alloc_list);2170		device->fs_devices->rw_devices--;2171		mutex_unlock(&fs_info->chunk_mutex);2172	}2173 2174	ret = btrfs_shrink_device(device, 0);2175	if (ret)2176		goto error_undo;2177 2178	trans = btrfs_start_transaction(fs_info->chunk_root, 0);2179	if (IS_ERR(trans)) {2180		ret = PTR_ERR(trans);2181		goto error_undo;2182	}2183 2184	ret = btrfs_rm_dev_item(trans, device);2185	if (ret) {2186		/* Any error in dev item removal is critical */2187		btrfs_crit(fs_info,2188			   "failed to remove device item for devid %llu: %d",2189			   device->devid, ret);2190		btrfs_abort_transaction(trans, ret);2191		btrfs_end_transaction(trans);2192		return ret;2193	}2194 2195	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);2196	btrfs_scrub_cancel_dev(device);2197 2198	/*2199	 * the device list mutex makes sure that we don't change2200	 * the device list while someone else is writing out all2201	 * the device supers. Whoever is writing all supers, should2202	 * lock the device list mutex before getting the number of2203	 * devices in the super block (super_copy). Conversely,2204	 * whoever updates the number of devices in the super block2205	 * (super_copy) should hold the device list mutex.2206	 */2207 2208	/*2209	 * In normal cases the cur_devices == fs_devices. But in case2210	 * of deleting a seed device, the cur_devices should point to2211	 * its own fs_devices listed under the fs_devices->seed_list.2212	 */2213	cur_devices = device->fs_devices;2214	mutex_lock(&fs_devices->device_list_mutex);2215	list_del_rcu(&device->dev_list);2216 2217	cur_devices->num_devices--;2218	cur_devices->total_devices--;2219	/* Update total_devices of the parent fs_devices if it's seed */2220	if (cur_devices != fs_devices)2221		fs_devices->total_devices--;2222 2223	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))2224		cur_devices->missing_devices--;2225 2226	btrfs_assign_next_active_device(device, NULL);2227 2228	if (device->bdev_file) {2229		cur_devices->open_devices--;2230		/* remove sysfs entry */2231		btrfs_sysfs_remove_device(device);2232	}2233 2234	num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;2235	btrfs_set_super_num_devices(fs_info->super_copy, num_devices);2236	mutex_unlock(&fs_devices->device_list_mutex);2237 2238	/*2239	 * At this point, the device is zero sized and detached from the2240	 * devices list.  All that's left is to zero out the old supers and2241	 * free the device.2242	 *2243	 * We cannot call btrfs_close_bdev() here because we're holding the sb2244	 * write lock, and fput() on the block device will pull in the2245	 * ->open_mutex on the block device and it's dependencies.  Instead2246	 *  just flush the device and let the caller do the final bdev_release.2247	 */2248	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {2249		btrfs_scratch_superblocks(fs_info, device);2250		if (device->bdev) {2251			sync_blockdev(device->bdev);2252			invalidate_bdev(device->bdev);2253		}2254	}2255 2256	*bdev_file = device->bdev_file;2257	synchronize_rcu();2258	btrfs_free_device(device);2259 2260	/*2261	 * This can happen if cur_devices is the private seed devices list.  We2262	 * cannot call close_fs_devices() here because it expects the uuid_mutex2263	 * to be held, but in fact we don't need that for the private2264	 * seed_devices, we can simply decrement cur_devices->opened and then2265	 * remove it from our list and free the fs_devices.2266	 */2267	if (cur_devices->num_devices == 0) {2268		list_del_init(&cur_devices->seed_list);2269		ASSERT(cur_devices->opened == 1);2270		cur_devices->opened--;2271		free_fs_devices(cur_devices);2272	}2273 2274	ret = btrfs_commit_transaction(trans);2275 2276	return ret;2277 2278error_undo:2279	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {2280		mutex_lock(&fs_info->chunk_mutex);2281		list_add(&device->dev_alloc_list,2282			 &fs_devices->alloc_list);2283		device->fs_devices->rw_devices++;2284		mutex_unlock(&fs_info->chunk_mutex);2285	}2286	return ret;2287}2288 2289void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)2290{2291	struct btrfs_fs_devices *fs_devices;2292 2293	lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);2294 2295	/*2296	 * in case of fs with no seed, srcdev->fs_devices will point2297	 * to fs_devices of fs_info. However when the dev being replaced is2298	 * a seed dev it will point to the seed's local fs_devices. In short2299	 * srcdev will have its correct fs_devices in both the cases.2300	 */2301	fs_devices = srcdev->fs_devices;2302 2303	list_del_rcu(&srcdev->dev_list);2304	list_del(&srcdev->dev_alloc_list);2305	fs_devices->num_devices--;2306	if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))2307		fs_devices->missing_devices--;2308 2309	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))2310		fs_devices->rw_devices--;2311 2312	if (srcdev->bdev)2313		fs_devices->open_devices--;2314}2315 2316void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)2317{2318	struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;2319 2320	mutex_lock(&uuid_mutex);2321 2322	btrfs_close_bdev(srcdev);2323	synchronize_rcu();2324	btrfs_free_device(srcdev);2325 2326	/* if this is no devs we rather delete the fs_devices */2327	if (!fs_devices->num_devices) {2328		/*2329		 * On a mounted FS, num_devices can't be zero unless it's a2330		 * seed. In case of a seed device being replaced, the replace2331		 * target added to the sprout FS, so there will be no more2332		 * device left under the seed FS.2333		 */2334		ASSERT(fs_devices->seeding);2335 2336		list_del_init(&fs_devices->seed_list);2337		close_fs_devices(fs_devices);2338		free_fs_devices(fs_devices);2339	}2340	mutex_unlock(&uuid_mutex);2341}2342 2343void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)2344{2345	struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;2346 2347	mutex_lock(&fs_devices->device_list_mutex);2348 2349	btrfs_sysfs_remove_device(tgtdev);2350 2351	if (tgtdev->bdev)2352		fs_devices->open_devices--;2353 2354	fs_devices->num_devices--;2355 2356	btrfs_assign_next_active_device(tgtdev, NULL);2357 2358	list_del_rcu(&tgtdev->dev_list);2359 2360	mutex_unlock(&fs_devices->device_list_mutex);2361 2362	btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev);2363 2364	btrfs_close_bdev(tgtdev);2365	synchronize_rcu();2366	btrfs_free_device(tgtdev);2367}2368 2369/*2370 * Populate args from device at path.2371 *2372 * @fs_info:	the filesystem2373 * @args:	the args to populate2374 * @path:	the path to the device2375 *2376 * This will read the super block of the device at @path and populate @args with2377 * the devid, fsid, and uuid.  This is meant to be used for ioctls that need to2378 * lookup a device to operate on, but need to do it before we take any locks.2379 * This properly handles the special case of "missing" that a user may pass in,2380 * and does some basic sanity checks.  The caller must make sure that @path is2381 * properly NUL terminated before calling in, and must call2382 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and2383 * uuid buffers.2384 *2385 * Return: 0 for success, -errno for failure2386 */2387int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,2388				 struct btrfs_dev_lookup_args *args,2389				 const char *path)2390{2391	struct btrfs_super_block *disk_super;2392	struct file *bdev_file;2393	int ret;2394 2395	if (!path || !path[0])2396		return -EINVAL;2397	if (!strcmp(path, "missing")) {2398		args->missing = true;2399		return 0;2400	}2401 2402	args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);2403	args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);2404	if (!args->uuid || !args->fsid) {2405		btrfs_put_dev_args_from_path(args);2406		return -ENOMEM;2407	}2408 2409	ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,2410				    &bdev_file, &disk_super);2411	if (ret) {2412		btrfs_put_dev_args_from_path(args);2413		return ret;2414	}2415 2416	args->devid = btrfs_stack_device_id(&disk_super->dev_item);2417	memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);2418	if (btrfs_fs_incompat(fs_info, METADATA_UUID))2419		memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);2420	else2421		memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);2422	btrfs_release_disk_super(disk_super);2423	fput(bdev_file);2424	return 0;2425}2426 2427/*2428 * Only use this jointly with btrfs_get_dev_args_from_path() because we will2429 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables2430 * that don't need to be freed.2431 */2432void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)2433{2434	kfree(args->uuid);2435	kfree(args->fsid);2436	args->uuid = NULL;2437	args->fsid = NULL;2438}2439 2440struct btrfs_device *btrfs_find_device_by_devspec(2441		struct btrfs_fs_info *fs_info, u64 devid,2442		const char *device_path)2443{2444	BTRFS_DEV_LOOKUP_ARGS(args);2445	struct btrfs_device *device;2446	int ret;2447 2448	if (devid) {2449		args.devid = devid;2450		device = btrfs_find_device(fs_info->fs_devices, &args);2451		if (!device)2452			return ERR_PTR(-ENOENT);2453		return device;2454	}2455 2456	ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);2457	if (ret)2458		return ERR_PTR(ret);2459	device = btrfs_find_device(fs_info->fs_devices, &args);2460	btrfs_put_dev_args_from_path(&args);2461	if (!device)2462		return ERR_PTR(-ENOENT);2463	return device;2464}2465 2466static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)2467{2468	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;2469	struct btrfs_fs_devices *old_devices;2470	struct btrfs_fs_devices *seed_devices;2471 2472	lockdep_assert_held(&uuid_mutex);2473	if (!fs_devices->seeding)2474		return ERR_PTR(-EINVAL);2475 2476	/*2477	 * Private copy of the seed devices, anchored at2478	 * fs_info->fs_devices->seed_list2479	 */2480	seed_devices = alloc_fs_devices(NULL);2481	if (IS_ERR(seed_devices))2482		return seed_devices;2483 2484	/*2485	 * It's necessary to retain a copy of the original seed fs_devices in2486	 * fs_uuids so that filesystems which have been seeded can successfully2487	 * reference the seed device from open_seed_devices. This also supports2488	 * multiple fs seed.2489	 */2490	old_devices = clone_fs_devices(fs_devices);2491	if (IS_ERR(old_devices)) {2492		kfree(seed_devices);2493		return old_devices;2494	}2495 2496	list_add(&old_devices->fs_list, &fs_uuids);2497 2498	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));2499	seed_devices->opened = 1;2500	INIT_LIST_HEAD(&seed_devices->devices);2501	INIT_LIST_HEAD(&seed_devices->alloc_list);2502	mutex_init(&seed_devices->device_list_mutex);2503 2504	return seed_devices;2505}2506 2507/*2508 * Splice seed devices into the sprout fs_devices.2509 * Generate a new fsid for the sprouted read-write filesystem.2510 */2511static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,2512			       struct btrfs_fs_devices *seed_devices)2513{2514	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;2515	struct btrfs_super_block *disk_super = fs_info->super_copy;2516	struct btrfs_device *device;2517	u64 super_flags;2518 2519	/*2520	 * We are updating the fsid, the thread leading to device_list_add()2521	 * could race, so uuid_mutex is needed.2522	 */2523	lockdep_assert_held(&uuid_mutex);2524 2525	/*2526	 * The threads listed below may traverse dev_list but can do that without2527	 * device_list_mutex:2528	 * - All device ops and balance - as we are in btrfs_exclop_start.2529	 * - Various dev_list readers - are using RCU.2530	 * - btrfs_ioctl_fitrim() - is using RCU.2531	 *2532	 * For-read threads as below are using device_list_mutex:2533	 * - Readonly scrub btrfs_scrub_dev()2534	 * - Readonly scrub btrfs_scrub_progress()2535	 * - btrfs_get_dev_stats()2536	 */2537	lockdep_assert_held(&fs_devices->device_list_mutex);2538 2539	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,2540			      synchronize_rcu);2541	list_for_each_entry(device, &seed_devices->devices, dev_list)2542		device->fs_devices = seed_devices;2543 2544	fs_devices->seeding = false;2545	fs_devices->num_devices = 0;2546	fs_devices->open_devices = 0;2547	fs_devices->missing_devices = 0;2548	fs_devices->rotating = false;2549	list_add(&seed_devices->seed_list, &fs_devices->seed_list);2550 2551	generate_random_uuid(fs_devices->fsid);2552	memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);2553	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);2554 2555	super_flags = btrfs_super_flags(disk_super) &2556		      ~BTRFS_SUPER_FLAG_SEEDING;2557	btrfs_set_super_flags(disk_super, super_flags);2558}2559 2560/*2561 * Store the expected generation for seed devices in device items.2562 */2563static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)2564{2565	BTRFS_DEV_LOOKUP_ARGS(args);2566	struct btrfs_fs_info *fs_info = trans->fs_info;2567	struct btrfs_root *root = fs_info->chunk_root;2568	struct btrfs_path *path;2569	struct extent_buffer *leaf;2570	struct btrfs_dev_item *dev_item;2571	struct btrfs_device *device;2572	struct btrfs_key key;2573	u8 fs_uuid[BTRFS_FSID_SIZE];2574	u8 dev_uuid[BTRFS_UUID_SIZE];2575	int ret;2576 2577	path = btrfs_alloc_path();2578	if (!path)2579		return -ENOMEM;2580 2581	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;2582	key.offset = 0;2583	key.type = BTRFS_DEV_ITEM_KEY;2584 2585	while (1) {2586		btrfs_reserve_chunk_metadata(trans, false);2587		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);2588		btrfs_trans_release_chunk_metadata(trans);2589		if (ret < 0)2590			goto error;2591 2592		leaf = path->nodes[0];2593next_slot:2594		if (path->slots[0] >= btrfs_header_nritems(leaf)) {2595			ret = btrfs_next_leaf(root, path);2596			if (ret > 0)2597				break;2598			if (ret < 0)2599				goto error;2600			leaf = path->nodes[0];2601			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);2602			btrfs_release_path(path);2603			continue;2604		}2605 2606		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);2607		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||2608		    key.type != BTRFS_DEV_ITEM_KEY)2609			break;2610 2611		dev_item = btrfs_item_ptr(leaf, path->slots[0],2612					  struct btrfs_dev_item);2613		args.devid = btrfs_device_id(leaf, dev_item);2614		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),2615				   BTRFS_UUID_SIZE);2616		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),2617				   BTRFS_FSID_SIZE);2618		args.uuid = dev_uuid;2619		args.fsid = fs_uuid;2620		device = btrfs_find_device(fs_info->fs_devices, &args);2621		BUG_ON(!device); /* Logic error */2622 2623		if (device->fs_devices->seeding) {2624			btrfs_set_device_generation(leaf, dev_item,2625						    device->generation);2626			btrfs_mark_buffer_dirty(trans, leaf);2627		}2628 2629		path->slots[0]++;2630		goto next_slot;2631	}2632	ret = 0;2633error:2634	btrfs_free_path(path);2635	return ret;2636}2637 2638int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)2639{2640	struct btrfs_root *root = fs_info->dev_root;2641	struct btrfs_trans_handle *trans;2642	struct btrfs_device *device;2643	struct file *bdev_file;2644	struct super_block *sb = fs_info->sb;2645	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;2646	struct btrfs_fs_devices *seed_devices = NULL;2647	u64 orig_super_total_bytes;2648	u64 orig_super_num_devices;2649	int ret = 0;2650	bool seeding_dev = false;2651	bool locked = false;2652 2653	if (sb_rdonly(sb) && !fs_devices->seeding)2654		return -EROFS;2655 2656	bdev_file = bdev_file_open_by_path(device_path, BLK_OPEN_WRITE,2657					fs_info->bdev_holder, NULL);2658	if (IS_ERR(bdev_file))2659		return PTR_ERR(bdev_file);2660 2661	if (!btrfs_check_device_zone_type(fs_info, file_bdev(bdev_file))) {2662		ret = -EINVAL;2663		goto error;2664	}2665 2666	if (fs_devices->seeding) {2667		seeding_dev = true;2668		down_write(&sb->s_umount);2669		mutex_lock(&uuid_mutex);2670		locked = true;2671	}2672 2673	sync_blockdev(file_bdev(bdev_file));2674 2675	rcu_read_lock();2676	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {2677		if (device->bdev == file_bdev(bdev_file)) {2678			ret = -EEXIST;2679			rcu_read_unlock();2680			goto error;2681		}2682	}2683	rcu_read_unlock();2684 2685	device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);2686	if (IS_ERR(device)) {2687		/* we can safely leave the fs_devices entry around */2688		ret = PTR_ERR(device);2689		goto error;2690	}2691 2692	device->fs_info = fs_info;2693	device->bdev_file = bdev_file;2694	device->bdev = file_bdev(bdev_file);2695	ret = lookup_bdev(device_path, &device->devt);2696	if (ret)2697		goto error_free_device;2698 2699	ret = btrfs_get_dev_zone_info(device, false);2700	if (ret)2701		goto error_free_device;2702 2703	trans = btrfs_start_transaction(root, 0);2704	if (IS_ERR(trans)) {2705		ret = PTR_ERR(trans);2706		goto error_free_zone;2707	}2708 2709	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);2710	device->generation = trans->transid;2711	device->io_width = fs_info->sectorsize;2712	device->io_align = fs_info->sectorsize;2713	device->sector_size = fs_info->sectorsize;2714	device->total_bytes =2715		round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize);2716	device->disk_total_bytes = device->total_bytes;2717	device->commit_total_bytes = device->total_bytes;2718	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);2719	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);2720	device->dev_stats_valid = 1;2721	set_blocksize(device->bdev_file, BTRFS_BDEV_BLOCKSIZE);2722 2723	if (seeding_dev) {2724		btrfs_clear_sb_rdonly(sb);2725 2726		/* GFP_KERNEL allocation must not be under device_list_mutex */2727		seed_devices = btrfs_init_sprout(fs_info);2728		if (IS_ERR(seed_devices)) {2729			ret = PTR_ERR(seed_devices);2730			btrfs_abort_transaction(trans, ret);2731			goto error_trans;2732		}2733	}2734 2735	mutex_lock(&fs_devices->device_list_mutex);2736	if (seeding_dev) {2737		btrfs_setup_sprout(fs_info, seed_devices);2738		btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,2739						device);2740	}2741 2742	device->fs_devices = fs_devices;2743 2744	mutex_lock(&fs_info->chunk_mutex);2745	list_add_rcu(&device->dev_list, &fs_devices->devices);2746	list_add(&device->dev_alloc_list, &fs_devices->alloc_list);2747	fs_devices->num_devices++;2748	fs_devices->open_devices++;2749	fs_devices->rw_devices++;2750	fs_devices->total_devices++;2751	fs_devices->total_rw_bytes += device->total_bytes;2752 2753	atomic64_add(device->total_bytes, &fs_info->free_chunk_space);2754 2755	if (!bdev_nonrot(device->bdev))2756		fs_devices->rotating = true;2757 2758	orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);2759	btrfs_set_super_total_bytes(fs_info->super_copy,2760		round_down(orig_super_total_bytes + device->total_bytes,2761			   fs_info->sectorsize));2762 2763	orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);2764	btrfs_set_super_num_devices(fs_info->super_copy,2765				    orig_super_num_devices + 1);2766 2767	/*2768	 * we've got more storage, clear any full flags on the space2769	 * infos2770	 */2771	btrfs_clear_space_info_full(fs_info);2772 2773	mutex_unlock(&fs_info->chunk_mutex);2774 2775	/* Add sysfs device entry */2776	btrfs_sysfs_add_device(device);2777 2778	mutex_unlock(&fs_devices->device_list_mutex);2779 2780	if (seeding_dev) {2781		mutex_lock(&fs_info->chunk_mutex);2782		ret = init_first_rw_device(trans);2783		mutex_unlock(&fs_info->chunk_mutex);2784		if (ret) {2785			btrfs_abort_transaction(trans, ret);2786			goto error_sysfs;2787		}2788	}2789 2790	ret = btrfs_add_dev_item(trans, device);2791	if (ret) {2792		btrfs_abort_transaction(trans, ret);2793		goto error_sysfs;2794	}2795 2796	if (seeding_dev) {2797		ret = btrfs_finish_sprout(trans);2798		if (ret) {2799			btrfs_abort_transaction(trans, ret);2800			goto error_sysfs;2801		}2802 2803		/*2804		 * fs_devices now represents the newly sprouted filesystem and2805		 * its fsid has been changed by btrfs_sprout_splice().2806		 */2807		btrfs_sysfs_update_sprout_fsid(fs_devices);2808	}2809 2810	ret = btrfs_commit_transaction(trans);2811 2812	if (seeding_dev) {2813		mutex_unlock(&uuid_mutex);2814		up_write(&sb->s_umount);2815		locked = false;2816 2817		if (ret) /* transaction commit */2818			return ret;2819 2820		ret = btrfs_relocate_sys_chunks(fs_info);2821		if (ret < 0)2822			btrfs_handle_fs_error(fs_info, ret,2823				    "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");2824		trans = btrfs_attach_transaction(root);2825		if (IS_ERR(trans)) {2826			if (PTR_ERR(trans) == -ENOENT)2827				return 0;2828			ret = PTR_ERR(trans);2829			trans = NULL;2830			goto error_sysfs;2831		}2832		ret = btrfs_commit_transaction(trans);2833	}2834 2835	/*2836	 * Now that we have written a new super block to this device, check all2837	 * other fs_devices list if device_path alienates any other scanned2838	 * device.2839	 * We can ignore the return value as it typically returns -EINVAL and2840	 * only succeeds if the device was an alien.2841	 */2842	btrfs_forget_devices(device->devt);2843 2844	/* Update ctime/mtime for blkid or udev */2845	update_dev_time(device_path);2846 2847	return ret;2848 2849error_sysfs:2850	btrfs_sysfs_remove_device(device);2851	mutex_lock(&fs_info->fs_devices->device_list_mutex);2852	mutex_lock(&fs_info->chunk_mutex);2853	list_del_rcu(&device->dev_list);2854	list_del(&device->dev_alloc_list);2855	fs_info->fs_devices->num_devices--;2856	fs_info->fs_devices->open_devices--;2857	fs_info->fs_devices->rw_devices--;2858	fs_info->fs_devices->total_devices--;2859	fs_info->fs_devices->total_rw_bytes -= device->total_bytes;2860	atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);2861	btrfs_set_super_total_bytes(fs_info->super_copy,2862				    orig_super_total_bytes);2863	btrfs_set_super_num_devices(fs_info->super_copy,2864				    orig_super_num_devices);2865	mutex_unlock(&fs_info->chunk_mutex);2866	mutex_unlock(&fs_info->fs_devices->device_list_mutex);2867error_trans:2868	if (seeding_dev)2869		btrfs_set_sb_rdonly(sb);2870	if (trans)2871		btrfs_end_transaction(trans);2872error_free_zone:2873	btrfs_destroy_dev_zone_info(device);2874error_free_device:2875	btrfs_free_device(device);2876error:2877	fput(bdev_file);2878	if (locked) {2879		mutex_unlock(&uuid_mutex);2880		up_write(&sb->s_umount);2881	}2882	return ret;2883}2884 2885static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,2886					struct btrfs_device *device)2887{2888	int ret;2889	struct btrfs_path *path;2890	struct btrfs_root *root = device->fs_info->chunk_root;2891	struct btrfs_dev_item *dev_item;2892	struct extent_buffer *leaf;2893	struct btrfs_key key;2894 2895	path = btrfs_alloc_path();2896	if (!path)2897		return -ENOMEM;2898 2899	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;2900	key.type = BTRFS_DEV_ITEM_KEY;2901	key.offset = device->devid;2902 2903	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);2904	if (ret < 0)2905		goto out;2906 2907	if (ret > 0) {2908		ret = -ENOENT;2909		goto out;2910	}2911 2912	leaf = path->nodes[0];2913	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);2914 2915	btrfs_set_device_id(leaf, dev_item, device->devid);2916	btrfs_set_device_type(leaf, dev_item, device->type);2917	btrfs_set_device_io_align(leaf, dev_item, device->io_align);2918	btrfs_set_device_io_width(leaf, dev_item, device->io_width);2919	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);2920	btrfs_set_device_total_bytes(leaf, dev_item,2921				     btrfs_device_get_disk_total_bytes(device));2922	btrfs_set_device_bytes_used(leaf, dev_item,2923				    btrfs_device_get_bytes_used(device));2924	btrfs_mark_buffer_dirty(trans, leaf);2925 2926out:2927	btrfs_free_path(path);2928	return ret;2929}2930 2931int btrfs_grow_device(struct btrfs_trans_handle *trans,2932		      struct btrfs_device *device, u64 new_size)2933{2934	struct btrfs_fs_info *fs_info = device->fs_info;2935	struct btrfs_super_block *super_copy = fs_info->super_copy;2936	u64 old_total;2937	u64 diff;2938	int ret;2939 2940	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))2941		return -EACCES;2942 2943	new_size = round_down(new_size, fs_info->sectorsize);2944 2945	mutex_lock(&fs_info->chunk_mutex);2946	old_total = btrfs_super_total_bytes(super_copy);2947	diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);2948 2949	if (new_size <= device->total_bytes ||2950	    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {2951		mutex_unlock(&fs_info->chunk_mutex);2952		return -EINVAL;2953	}2954 2955	btrfs_set_super_total_bytes(super_copy,2956			round_down(old_total + diff, fs_info->sectorsize));2957	device->fs_devices->total_rw_bytes += diff;2958	atomic64_add(diff, &fs_info->free_chunk_space);2959 2960	btrfs_device_set_total_bytes(device, new_size);2961	btrfs_device_set_disk_total_bytes(device, new_size);2962	btrfs_clear_space_info_full(device->fs_info);2963	if (list_empty(&device->post_commit_list))2964		list_add_tail(&device->post_commit_list,2965			      &trans->transaction->dev_update_list);2966	mutex_unlock(&fs_info->chunk_mutex);2967 2968	btrfs_reserve_chunk_metadata(trans, false);2969	ret = btrfs_update_device(trans, device);2970	btrfs_trans_release_chunk_metadata(trans);2971 2972	return ret;2973}2974 2975static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)2976{2977	struct btrfs_fs_info *fs_info = trans->fs_info;2978	struct btrfs_root *root = fs_info->chunk_root;2979	int ret;2980	struct btrfs_path *path;2981	struct btrfs_key key;2982 2983	path = btrfs_alloc_path();2984	if (!path)2985		return -ENOMEM;2986 2987	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;2988	key.offset = chunk_offset;2989	key.type = BTRFS_CHUNK_ITEM_KEY;2990 2991	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);2992	if (ret < 0)2993		goto out;2994	else if (ret > 0) { /* Logic error or corruption */2995		btrfs_err(fs_info, "failed to lookup chunk %llu when freeing",2996			  chunk_offset);2997		btrfs_abort_transaction(trans, -ENOENT);2998		ret = -EUCLEAN;2999		goto out;3000	}3001 3002	ret = btrfs_del_item(trans, root, path);3003	if (ret < 0) {3004		btrfs_err(fs_info, "failed to delete chunk %llu item", chunk_offset);3005		btrfs_abort_transaction(trans, ret);3006		goto out;3007	}3008out:3009	btrfs_free_path(path);3010	return ret;3011}3012 3013static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)3014{3015	struct btrfs_super_block *super_copy = fs_info->super_copy;3016	struct btrfs_disk_key *disk_key;3017	struct btrfs_chunk *chunk;3018	u8 *ptr;3019	int ret = 0;3020	u32 num_stripes;3021	u32 array_size;3022	u32 len = 0;3023	u32 cur;3024	struct btrfs_key key;3025 3026	lockdep_assert_held(&fs_info->chunk_mutex);3027	array_size = btrfs_super_sys_array_size(super_copy);3028 3029	ptr = super_copy->sys_chunk_array;3030	cur = 0;3031 3032	while (cur < array_size) {3033		disk_key = (struct btrfs_disk_key *)ptr;3034		btrfs_disk_key_to_cpu(&key, disk_key);3035 3036		len = sizeof(*disk_key);3037 3038		if (key.type == BTRFS_CHUNK_ITEM_KEY) {3039			chunk = (struct btrfs_chunk *)(ptr + len);3040			num_stripes = btrfs_stack_chunk_num_stripes(chunk);3041			len += btrfs_chunk_item_size(num_stripes);3042		} else {3043			ret = -EIO;3044			break;3045		}3046		if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&3047		    key.offset == chunk_offset) {3048			memmove(ptr, ptr + len, array_size - (cur + len));3049			array_size -= len;3050			btrfs_set_super_sys_array_size(super_copy, array_size);3051		} else {3052			ptr += len;3053			cur += len;3054		}3055	}3056	return ret;3057}3058 3059struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,3060						    u64 logical, u64 length)3061{3062	struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node;3063	struct rb_node *prev = NULL;3064	struct rb_node *orig_prev;3065	struct btrfs_chunk_map *map;3066	struct btrfs_chunk_map *prev_map = NULL;3067 3068	while (node) {3069		map = rb_entry(node, struct btrfs_chunk_map, rb_node);3070		prev = node;3071		prev_map = map;3072 3073		if (logical < map->start) {3074			node = node->rb_left;3075		} else if (logical >= map->start + map->chunk_len) {3076			node = node->rb_right;3077		} else {3078			refcount_inc(&map->refs);3079			return map;3080		}3081	}3082 3083	if (!prev)3084		return NULL;3085 3086	orig_prev = prev;3087	while (prev && logical >= prev_map->start + prev_map->chunk_len) {3088		prev = rb_next(prev);3089		prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);3090	}3091 3092	if (!prev) {3093		prev = orig_prev;3094		prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);3095		while (prev && logical < prev_map->start) {3096			prev = rb_prev(prev);3097			prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);3098		}3099	}3100 3101	if (prev) {3102		u64 end = logical + length;3103 3104		/*3105		 * Caller can pass a U64_MAX length when it wants to get any3106		 * chunk starting at an offset of 'logical' or higher, so deal3107		 * with underflow by resetting the end offset to U64_MAX.3108		 */3109		if (end < logical)3110			end = U64_MAX;3111 3112		if (end > prev_map->start &&3113		    logical < prev_map->start + prev_map->chunk_len) {3114			refcount_inc(&prev_map->refs);3115			return prev_map;3116		}3117	}3118 3119	return NULL;3120}3121 3122struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,3123					     u64 logical, u64 length)3124{3125	struct btrfs_chunk_map *map;3126 3127	read_lock(&fs_info->mapping_tree_lock);3128	map = btrfs_find_chunk_map_nolock(fs_info, logical, length);3129	read_unlock(&fs_info->mapping_tree_lock);3130 3131	return map;3132}3133 3134/*3135 * Find the mapping containing the given logical extent.3136 *3137 * @logical: Logical block offset in bytes.3138 * @length: Length of extent in bytes.3139 *3140 * Return: Chunk mapping or ERR_PTR.3141 */3142struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,3143					    u64 logical, u64 length)3144{3145	struct btrfs_chunk_map *map;3146 3147	map = btrfs_find_chunk_map(fs_info, logical, length);3148 3149	if (unlikely(!map)) {3150		btrfs_crit(fs_info,3151			   "unable to find chunk map for logical %llu length %llu",3152			   logical, length);3153		return ERR_PTR(-EINVAL);3154	}3155 3156	if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) {3157		btrfs_crit(fs_info,3158			   "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",3159			   logical, logical + length, map->start,3160			   map->start + map->chunk_len);3161		btrfs_free_chunk_map(map);3162		return ERR_PTR(-EINVAL);3163	}3164 3165	/* Callers are responsible for dropping the reference. */3166	return map;3167}3168 3169static int remove_chunk_item(struct btrfs_trans_handle *trans,3170			     struct btrfs_chunk_map *map, u64 chunk_offset)3171{3172	int i;3173 3174	/*3175	 * Removing chunk items and updating the device items in the chunks btree3176	 * requires holding the chunk_mutex.3177	 * See the comment at btrfs_chunk_alloc() for the details.3178	 */3179	lockdep_assert_held(&trans->fs_info->chunk_mutex);3180 3181	for (i = 0; i < map->num_stripes; i++) {3182		int ret;3183 3184		ret = btrfs_update_device(trans, map->stripes[i].dev);3185		if (ret)3186			return ret;3187	}3188 3189	return btrfs_free_chunk(trans, chunk_offset);3190}3191 3192int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)3193{3194	struct btrfs_fs_info *fs_info = trans->fs_info;3195	struct btrfs_chunk_map *map;3196	u64 dev_extent_len = 0;3197	int i, ret = 0;3198	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;3199 3200	map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);3201	if (IS_ERR(map)) {3202		/*3203		 * This is a logic error, but we don't want to just rely on the3204		 * user having built with ASSERT enabled, so if ASSERT doesn't3205		 * do anything we still error out.3206		 */3207		ASSERT(0);3208		return PTR_ERR(map);3209	}3210 3211	/*3212	 * First delete the device extent items from the devices btree.3213	 * We take the device_list_mutex to avoid racing with the finishing phase3214	 * of a device replace operation. See the comment below before acquiring3215	 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex3216	 * because that can result in a deadlock when deleting the device extent3217	 * items from the devices btree - COWing an extent buffer from the btree3218	 * may result in allocating a new metadata chunk, which would attempt to3219	 * lock again fs_info->chunk_mutex.3220	 */3221	mutex_lock(&fs_devices->device_list_mutex);3222	for (i = 0; i < map->num_stripes; i++) {3223		struct btrfs_device *device = map->stripes[i].dev;3224		ret = btrfs_free_dev_extent(trans, device,3225					    map->stripes[i].physical,3226					    &dev_extent_len);3227		if (ret) {3228			mutex_unlock(&fs_devices->device_list_mutex);3229			btrfs_abort_transaction(trans, ret);3230			goto out;3231		}3232 3233		if (device->bytes_used > 0) {3234			mutex_lock(&fs_info->chunk_mutex);3235			btrfs_device_set_bytes_used(device,3236					device->bytes_used - dev_extent_len);3237			atomic64_add(dev_extent_len, &fs_info->free_chunk_space);3238			btrfs_clear_space_info_full(fs_info);3239			mutex_unlock(&fs_info->chunk_mutex);3240		}3241	}3242	mutex_unlock(&fs_devices->device_list_mutex);3243 3244	/*3245	 * We acquire fs_info->chunk_mutex for 2 reasons:3246	 *3247	 * 1) Just like with the first phase of the chunk allocation, we must3248	 *    reserve system space, do all chunk btree updates and deletions, and3249	 *    update the system chunk array in the superblock while holding this3250	 *    mutex. This is for similar reasons as explained on the comment at3251	 *    the top of btrfs_chunk_alloc();3252	 *3253	 * 2) Prevent races with the final phase of a device replace operation3254	 *    that replaces the device object associated with the map's stripes,3255	 *    because the device object's id can change at any time during that3256	 *    final phase of the device replace operation3257	 *    (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the3258	 *    replaced device and then see it with an ID of3259	 *    BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating3260	 *    the device item, which does not exists on the chunk btree.3261	 *    The finishing phase of device replace acquires both the3262	 *    device_list_mutex and the chunk_mutex, in that order, so we are3263	 *    safe by just acquiring the chunk_mutex.3264	 */3265	trans->removing_chunk = true;3266	mutex_lock(&fs_info->chunk_mutex);3267 3268	check_system_chunk(trans, map->type);3269 3270	ret = remove_chunk_item(trans, map, chunk_offset);3271	/*3272	 * Normally we should not get -ENOSPC since we reserved space before3273	 * through the call to check_system_chunk().3274	 *3275	 * Despite our system space_info having enough free space, we may not3276	 * be able to allocate extents from its block groups, because all have3277	 * an incompatible profile, which will force us to allocate a new system3278	 * block group with the right profile, or right after we called3279	 * check_system_space() above, a scrub turned the only system block group3280	 * with enough free space into RO mode.3281	 * This is explained with more detail at do_chunk_alloc().3282	 *3283	 * So if we get -ENOSPC, allocate a new system chunk and retry once.3284	 */3285	if (ret == -ENOSPC) {3286		const u64 sys_flags = btrfs_system_alloc_profile(fs_info);3287		struct btrfs_block_group *sys_bg;3288 3289		sys_bg = btrfs_create_chunk(trans, sys_flags);3290		if (IS_ERR(sys_bg)) {3291			ret = PTR_ERR(sys_bg);3292			btrfs_abort_transaction(trans, ret);3293			goto out;3294		}3295 3296		ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);3297		if (ret) {3298			btrfs_abort_transaction(trans, ret);3299			goto out;3300		}3301 3302		ret = remove_chunk_item(trans, map, chunk_offset);3303		if (ret) {3304			btrfs_abort_transaction(trans, ret);3305			goto out;3306		}3307	} else if (ret) {3308		btrfs_abort_transaction(trans, ret);3309		goto out;3310	}3311 3312	trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len);3313 3314	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {3315		ret = btrfs_del_sys_chunk(fs_info, chunk_offset);3316		if (ret) {3317			btrfs_abort_transaction(trans, ret);3318			goto out;3319		}3320	}3321 3322	mutex_unlock(&fs_info->chunk_mutex);3323	trans->removing_chunk = false;3324 3325	/*3326	 * We are done with chunk btree updates and deletions, so release the3327	 * system space we previously reserved (with check_system_chunk()).3328	 */3329	btrfs_trans_release_chunk_metadata(trans);3330 3331	ret = btrfs_remove_block_group(trans, map);3332	if (ret) {3333		btrfs_abort_transaction(trans, ret);3334		goto out;3335	}3336 3337out:3338	if (trans->removing_chunk) {3339		mutex_unlock(&fs_info->chunk_mutex);3340		trans->removing_chunk = false;3341	}3342	/* once for us */3343	btrfs_free_chunk_map(map);3344	return ret;3345}3346 3347int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)3348{3349	struct btrfs_root *root = fs_info->chunk_root;3350	struct btrfs_trans_handle *trans;3351	struct btrfs_block_group *block_group;3352	u64 length;3353	int ret;3354 3355	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {3356		btrfs_err(fs_info,3357			  "relocate: not supported on extent tree v2 yet");3358		return -EINVAL;3359	}3360 3361	/*3362	 * Prevent races with automatic removal of unused block groups.3363	 * After we relocate and before we remove the chunk with offset3364	 * chunk_offset, automatic removal of the block group can kick in,3365	 * resulting in a failure when calling btrfs_remove_chunk() below.3366	 *3367	 * Make sure to acquire this mutex before doing a tree search (dev3368	 * or chunk trees) to find chunks. Otherwise the cleaner kthread might3369	 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after3370	 * we release the path used to search the chunk/dev tree and before3371	 * the current task acquires this mutex and calls us.3372	 */3373	lockdep_assert_held(&fs_info->reclaim_bgs_lock);3374 3375	/* step one, relocate all the extents inside this chunk */3376	btrfs_scrub_pause(fs_info);3377	ret = btrfs_relocate_block_group(fs_info, chunk_offset);3378	btrfs_scrub_continue(fs_info);3379	if (ret) {3380		/*3381		 * If we had a transaction abort, stop all running scrubs.3382		 * See transaction.c:cleanup_transaction() why we do it here.3383		 */3384		if (BTRFS_FS_ERROR(fs_info))3385			btrfs_scrub_cancel(fs_info);3386		return ret;3387	}3388 3389	block_group = btrfs_lookup_block_group(fs_info, chunk_offset);3390	if (!block_group)3391		return -ENOENT;3392	btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);3393	length = block_group->length;3394	btrfs_put_block_group(block_group);3395 3396	/*3397	 * On a zoned file system, discard the whole block group, this will3398	 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If3399	 * resetting the zone fails, don't treat it as a fatal problem from the3400	 * filesystem's point of view.3401	 */3402	if (btrfs_is_zoned(fs_info)) {3403		ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);3404		if (ret)3405			btrfs_info(fs_info,3406				"failed to reset zone %llu after relocation",3407				chunk_offset);3408	}3409 3410	trans = btrfs_start_trans_remove_block_group(root->fs_info,3411						     chunk_offset);3412	if (IS_ERR(trans)) {3413		ret = PTR_ERR(trans);3414		btrfs_handle_fs_error(root->fs_info, ret, NULL);3415		return ret;3416	}3417 3418	/*3419	 * step two, delete the device extents and the3420	 * chunk tree entries3421	 */3422	ret = btrfs_remove_chunk(trans, chunk_offset);3423	btrfs_end_transaction(trans);3424	return ret;3425}3426 3427static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)3428{3429	struct btrfs_root *chunk_root = fs_info->chunk_root;3430	struct btrfs_path *path;3431	struct extent_buffer *leaf;3432	struct btrfs_chunk *chunk;3433	struct btrfs_key key;3434	struct btrfs_key found_key;3435	u64 chunk_type;3436	bool retried = false;3437	int failed = 0;3438	int ret;3439 3440	path = btrfs_alloc_path();3441	if (!path)3442		return -ENOMEM;3443 3444again:3445	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;3446	key.offset = (u64)-1;3447	key.type = BTRFS_CHUNK_ITEM_KEY;3448 3449	while (1) {3450		mutex_lock(&fs_info->reclaim_bgs_lock);3451		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);3452		if (ret < 0) {3453			mutex_unlock(&fs_info->reclaim_bgs_lock);3454			goto error;3455		}3456		if (ret == 0) {3457			/*3458			 * On the first search we would find chunk tree with3459			 * offset -1, which is not possible. On subsequent3460			 * loops this would find an existing item on an invalid3461			 * offset (one less than the previous one, wrong3462			 * alignment and size).3463			 */3464			ret = -EUCLEAN;3465			mutex_unlock(&fs_info->reclaim_bgs_lock);3466			goto error;3467		}3468 3469		ret = btrfs_previous_item(chunk_root, path, key.objectid,3470					  key.type);3471		if (ret)3472			mutex_unlock(&fs_info->reclaim_bgs_lock);3473		if (ret < 0)3474			goto error;3475		if (ret > 0)3476			break;3477 3478		leaf = path->nodes[0];3479		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);3480 3481		chunk = btrfs_item_ptr(leaf, path->slots[0],3482				       struct btrfs_chunk);3483		chunk_type = btrfs_chunk_type(leaf, chunk);3484		btrfs_release_path(path);3485 3486		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {3487			ret = btrfs_relocate_chunk(fs_info, found_key.offset);3488			if (ret == -ENOSPC)3489				failed++;3490			else3491				BUG_ON(ret);3492		}3493		mutex_unlock(&fs_info->reclaim_bgs_lock);3494 3495		if (found_key.offset == 0)3496			break;3497		key.offset = found_key.offset - 1;3498	}3499	ret = 0;3500	if (failed && !retried) {3501		failed = 0;3502		retried = true;3503		goto again;3504	} else if (WARN_ON(failed && retried)) {3505		ret = -ENOSPC;3506	}3507error:3508	btrfs_free_path(path);3509	return ret;3510}3511 3512/*3513 * return 1 : allocate a data chunk successfully,3514 * return <0: errors during allocating a data chunk,3515 * return 0 : no need to allocate a data chunk.3516 */3517static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,3518				      u64 chunk_offset)3519{3520	struct btrfs_block_group *cache;3521	u64 bytes_used;3522	u64 chunk_type;3523 3524	cache = btrfs_lookup_block_group(fs_info, chunk_offset);3525	ASSERT(cache);3526	chunk_type = cache->flags;3527	btrfs_put_block_group(cache);3528 3529	if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))3530		return 0;3531 3532	spin_lock(&fs_info->data_sinfo->lock);3533	bytes_used = fs_info->data_sinfo->bytes_used;3534	spin_unlock(&fs_info->data_sinfo->lock);3535 3536	if (!bytes_used) {3537		struct btrfs_trans_handle *trans;3538		int ret;3539 3540		trans =	btrfs_join_transaction(fs_info->tree_root);3541		if (IS_ERR(trans))3542			return PTR_ERR(trans);3543 3544		ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);3545		btrfs_end_transaction(trans);3546		if (ret < 0)3547			return ret;3548		return 1;3549	}3550 3551	return 0;3552}3553 3554static void btrfs_disk_balance_args_to_cpu(struct btrfs_balance_args *cpu,3555					   const struct btrfs_disk_balance_args *disk)3556{3557	memset(cpu, 0, sizeof(*cpu));3558 3559	cpu->profiles = le64_to_cpu(disk->profiles);3560	cpu->usage = le64_to_cpu(disk->usage);3561	cpu->devid = le64_to_cpu(disk->devid);3562	cpu->pstart = le64_to_cpu(disk->pstart);3563	cpu->pend = le64_to_cpu(disk->pend);3564	cpu->vstart = le64_to_cpu(disk->vstart);3565	cpu->vend = le64_to_cpu(disk->vend);3566	cpu->target = le64_to_cpu(disk->target);3567	cpu->flags = le64_to_cpu(disk->flags);3568	cpu->limit = le64_to_cpu(disk->limit);3569	cpu->stripes_min = le32_to_cpu(disk->stripes_min);3570	cpu->stripes_max = le32_to_cpu(disk->stripes_max);3571}3572 3573static void btrfs_cpu_balance_args_to_disk(struct btrfs_disk_balance_args *disk,3574					   const struct btrfs_balance_args *cpu)3575{3576	memset(disk, 0, sizeof(*disk));3577 3578	disk->profiles = cpu_to_le64(cpu->profiles);3579	disk->usage = cpu_to_le64(cpu->usage);3580	disk->devid = cpu_to_le64(cpu->devid);3581	disk->pstart = cpu_to_le64(cpu->pstart);3582	disk->pend = cpu_to_le64(cpu->pend);3583	disk->vstart = cpu_to_le64(cpu->vstart);3584	disk->vend = cpu_to_le64(cpu->vend);3585	disk->target = cpu_to_le64(cpu->target);3586	disk->flags = cpu_to_le64(cpu->flags);3587	disk->limit = cpu_to_le64(cpu->limit);3588	disk->stripes_min = cpu_to_le32(cpu->stripes_min);3589	disk->stripes_max = cpu_to_le32(cpu->stripes_max);3590}3591 3592static int insert_balance_item(struct btrfs_fs_info *fs_info,3593			       struct btrfs_balance_control *bctl)3594{3595	struct btrfs_root *root = fs_info->tree_root;3596	struct btrfs_trans_handle *trans;3597	struct btrfs_balance_item *item;3598	struct btrfs_disk_balance_args disk_bargs;3599	struct btrfs_path *path;3600	struct extent_buffer *leaf;3601	struct btrfs_key key;3602	int ret, err;3603 3604	path = btrfs_alloc_path();3605	if (!path)3606		return -ENOMEM;3607 3608	trans = btrfs_start_transaction(root, 0);3609	if (IS_ERR(trans)) {3610		btrfs_free_path(path);3611		return PTR_ERR(trans);3612	}3613 3614	key.objectid = BTRFS_BALANCE_OBJECTID;3615	key.type = BTRFS_TEMPORARY_ITEM_KEY;3616	key.offset = 0;3617 3618	ret = btrfs_insert_empty_item(trans, root, path, &key,3619				      sizeof(*item));3620	if (ret)3621		goto out;3622 3623	leaf = path->nodes[0];3624	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);3625 3626	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));3627 3628	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);3629	btrfs_set_balance_data(leaf, item, &disk_bargs);3630	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);3631	btrfs_set_balance_meta(leaf, item, &disk_bargs);3632	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);3633	btrfs_set_balance_sys(leaf, item, &disk_bargs);3634 3635	btrfs_set_balance_flags(leaf, item, bctl->flags);3636 3637	btrfs_mark_buffer_dirty(trans, leaf);3638out:3639	btrfs_free_path(path);3640	err = btrfs_commit_transaction(trans);3641	if (err && !ret)3642		ret = err;3643	return ret;3644}3645 3646static int del_balance_item(struct btrfs_fs_info *fs_info)3647{3648	struct btrfs_root *root = fs_info->tree_root;3649	struct btrfs_trans_handle *trans;3650	struct btrfs_path *path;3651	struct btrfs_key key;3652	int ret, err;3653 3654	path = btrfs_alloc_path();3655	if (!path)3656		return -ENOMEM;3657 3658	trans = btrfs_start_transaction_fallback_global_rsv(root, 0);3659	if (IS_ERR(trans)) {3660		btrfs_free_path(path);3661		return PTR_ERR(trans);3662	}3663 3664	key.objectid = BTRFS_BALANCE_OBJECTID;3665	key.type = BTRFS_TEMPORARY_ITEM_KEY;3666	key.offset = 0;3667 3668	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);3669	if (ret < 0)3670		goto out;3671	if (ret > 0) {3672		ret = -ENOENT;3673		goto out;3674	}3675 3676	ret = btrfs_del_item(trans, root, path);3677out:3678	btrfs_free_path(path);3679	err = btrfs_commit_transaction(trans);3680	if (err && !ret)3681		ret = err;3682	return ret;3683}3684 3685/*3686 * This is a heuristic used to reduce the number of chunks balanced on3687 * resume after balance was interrupted.3688 */3689static void update_balance_args(struct btrfs_balance_control *bctl)3690{3691	/*3692	 * Turn on soft mode for chunk types that were being converted.3693	 */3694	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)3695		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;3696	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)3697		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;3698	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)3699		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;3700 3701	/*3702	 * Turn on usage filter if is not already used.  The idea is3703	 * that chunks that we have already balanced should be3704	 * reasonably full.  Don't do it for chunks that are being3705	 * converted - that will keep us from relocating unconverted3706	 * (albeit full) chunks.3707	 */3708	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&3709	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&3710	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {3711		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;3712		bctl->data.usage = 90;3713	}3714	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&3715	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&3716	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {3717		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;3718		bctl->sys.usage = 90;3719	}3720	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&3721	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&3722	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {3723		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;3724		bctl->meta.usage = 90;3725	}3726}3727 3728/*3729 * Clear the balance status in fs_info and delete the balance item from disk.3730 */3731static void reset_balance_state(struct btrfs_fs_info *fs_info)3732{3733	struct btrfs_balance_control *bctl = fs_info->balance_ctl;3734	int ret;3735 3736	ASSERT(fs_info->balance_ctl);3737 3738	spin_lock(&fs_info->balance_lock);3739	fs_info->balance_ctl = NULL;3740	spin_unlock(&fs_info->balance_lock);3741 3742	kfree(bctl);3743	ret = del_balance_item(fs_info);3744	if (ret)3745		btrfs_handle_fs_error(fs_info, ret, NULL);3746}3747 3748/*3749 * Balance filters.  Return 1 if chunk should be filtered out3750 * (should not be balanced).3751 */3752static int chunk_profiles_filter(u64 chunk_type,3753				 struct btrfs_balance_args *bargs)3754{3755	chunk_type = chunk_to_extended(chunk_type) &3756				BTRFS_EXTENDED_PROFILE_MASK;3757 3758	if (bargs->profiles & chunk_type)3759		return 0;3760 3761	return 1;3762}3763 3764static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,3765			      struct btrfs_balance_args *bargs)3766{3767	struct btrfs_block_group *cache;3768	u64 chunk_used;3769	u64 user_thresh_min;3770	u64 user_thresh_max;3771	int ret = 1;3772 3773	cache = btrfs_lookup_block_group(fs_info, chunk_offset);3774	chunk_used = cache->used;3775 3776	if (bargs->usage_min == 0)3777		user_thresh_min = 0;3778	else3779		user_thresh_min = mult_perc(cache->length, bargs->usage_min);3780 3781	if (bargs->usage_max == 0)3782		user_thresh_max = 1;3783	else if (bargs->usage_max > 100)3784		user_thresh_max = cache->length;3785	else3786		user_thresh_max = mult_perc(cache->length, bargs->usage_max);3787 3788	if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)3789		ret = 0;3790 3791	btrfs_put_block_group(cache);3792	return ret;3793}3794 3795static int chunk_usage_filter(struct btrfs_fs_info *fs_info,3796		u64 chunk_offset, struct btrfs_balance_args *bargs)3797{3798	struct btrfs_block_group *cache;3799	u64 chunk_used, user_thresh;3800	int ret = 1;3801 3802	cache = btrfs_lookup_block_group(fs_info, chunk_offset);3803	chunk_used = cache->used;3804 3805	if (bargs->usage_min == 0)3806		user_thresh = 1;3807	else if (bargs->usage > 100)3808		user_thresh = cache->length;3809	else3810		user_thresh = mult_perc(cache->length, bargs->usage);3811 3812	if (chunk_used < user_thresh)3813		ret = 0;3814 3815	btrfs_put_block_group(cache);3816	return ret;3817}3818 3819static int chunk_devid_filter(struct extent_buffer *leaf,3820			      struct btrfs_chunk *chunk,3821			      struct btrfs_balance_args *bargs)3822{3823	struct btrfs_stripe *stripe;3824	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);3825	int i;3826 3827	for (i = 0; i < num_stripes; i++) {3828		stripe = btrfs_stripe_nr(chunk, i);3829		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)3830			return 0;3831	}3832 3833	return 1;3834}3835 3836static u64 calc_data_stripes(u64 type, int num_stripes)3837{3838	const int index = btrfs_bg_flags_to_raid_index(type);3839	const int ncopies = btrfs_raid_array[index].ncopies;3840	const int nparity = btrfs_raid_array[index].nparity;3841 3842	return (num_stripes - nparity) / ncopies;3843}3844 3845/* [pstart, pend) */3846static int chunk_drange_filter(struct extent_buffer *leaf,3847			       struct btrfs_chunk *chunk,3848			       struct btrfs_balance_args *bargs)3849{3850	struct btrfs_stripe *stripe;3851	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);3852	u64 stripe_offset;3853	u64 stripe_length;3854	u64 type;3855	int factor;3856	int i;3857 3858	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))3859		return 0;3860 3861	type = btrfs_chunk_type(leaf, chunk);3862	factor = calc_data_stripes(type, num_stripes);3863 3864	for (i = 0; i < num_stripes; i++) {3865		stripe = btrfs_stripe_nr(chunk, i);3866		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)3867			continue;3868 3869		stripe_offset = btrfs_stripe_offset(leaf, stripe);3870		stripe_length = btrfs_chunk_length(leaf, chunk);3871		stripe_length = div_u64(stripe_length, factor);3872 3873		if (stripe_offset < bargs->pend &&3874		    stripe_offset + stripe_length > bargs->pstart)3875			return 0;3876	}3877 3878	return 1;3879}3880 3881/* [vstart, vend) */3882static int chunk_vrange_filter(struct extent_buffer *leaf,3883			       struct btrfs_chunk *chunk,3884			       u64 chunk_offset,3885			       struct btrfs_balance_args *bargs)3886{3887	if (chunk_offset < bargs->vend &&3888	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)3889		/* at least part of the chunk is inside this vrange */3890		return 0;3891 3892	return 1;3893}3894 3895static int chunk_stripes_range_filter(struct extent_buffer *leaf,3896			       struct btrfs_chunk *chunk,3897			       struct btrfs_balance_args *bargs)3898{3899	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);3900 3901	if (bargs->stripes_min <= num_stripes3902			&& num_stripes <= bargs->stripes_max)3903		return 0;3904 3905	return 1;3906}3907 3908static int chunk_soft_convert_filter(u64 chunk_type,3909				     struct btrfs_balance_args *bargs)3910{3911	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))3912		return 0;3913 3914	chunk_type = chunk_to_extended(chunk_type) &3915				BTRFS_EXTENDED_PROFILE_MASK;3916 3917	if (bargs->target == chunk_type)3918		return 1;3919 3920	return 0;3921}3922 3923static int should_balance_chunk(struct extent_buffer *leaf,3924				struct btrfs_chunk *chunk, u64 chunk_offset)3925{3926	struct btrfs_fs_info *fs_info = leaf->fs_info;3927	struct btrfs_balance_control *bctl = fs_info->balance_ctl;3928	struct btrfs_balance_args *bargs = NULL;3929	u64 chunk_type = btrfs_chunk_type(leaf, chunk);3930 3931	/* type filter */3932	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &3933	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {3934		return 0;3935	}3936 3937	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)3938		bargs = &bctl->data;3939	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)3940		bargs = &bctl->sys;3941	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)3942		bargs = &bctl->meta;3943 3944	/* profiles filter */3945	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&3946	    chunk_profiles_filter(chunk_type, bargs)) {3947		return 0;3948	}3949 3950	/* usage filter */3951	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&3952	    chunk_usage_filter(fs_info, chunk_offset, bargs)) {3953		return 0;3954	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&3955	    chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {3956		return 0;3957	}3958 3959	/* devid filter */3960	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&3961	    chunk_devid_filter(leaf, chunk, bargs)) {3962		return 0;3963	}3964 3965	/* drange filter, makes sense only with devid filter */3966	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&3967	    chunk_drange_filter(leaf, chunk, bargs)) {3968		return 0;3969	}3970 3971	/* vrange filter */3972	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&3973	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {3974		return 0;3975	}3976 3977	/* stripes filter */3978	if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&3979	    chunk_stripes_range_filter(leaf, chunk, bargs)) {3980		return 0;3981	}3982 3983	/* soft profile changing mode */3984	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&3985	    chunk_soft_convert_filter(chunk_type, bargs)) {3986		return 0;3987	}3988 3989	/*3990	 * limited by count, must be the last filter3991	 */3992	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {3993		if (bargs->limit == 0)3994			return 0;3995		else3996			bargs->limit--;3997	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {3998		/*3999		 * Same logic as the 'limit' filter; the minimum cannot be4000		 * determined here because we do not have the global information4001		 * about the count of all chunks that satisfy the filters.4002		 */4003		if (bargs->limit_max == 0)4004			return 0;4005		else4006			bargs->limit_max--;4007	}4008 4009	return 1;4010}4011 4012static int __btrfs_balance(struct btrfs_fs_info *fs_info)4013{4014	struct btrfs_balance_control *bctl = fs_info->balance_ctl;4015	struct btrfs_root *chunk_root = fs_info->chunk_root;4016	u64 chunk_type;4017	struct btrfs_chunk *chunk;4018	struct btrfs_path *path = NULL;4019	struct btrfs_key key;4020	struct btrfs_key found_key;4021	struct extent_buffer *leaf;4022	int slot;4023	int ret;4024	int enospc_errors = 0;4025	bool counting = true;4026	/* The single value limit and min/max limits use the same bytes in the */4027	u64 limit_data = bctl->data.limit;4028	u64 limit_meta = bctl->meta.limit;4029	u64 limit_sys = bctl->sys.limit;4030	u32 count_data = 0;4031	u32 count_meta = 0;4032	u32 count_sys = 0;4033	int chunk_reserved = 0;4034 4035	path = btrfs_alloc_path();4036	if (!path) {4037		ret = -ENOMEM;4038		goto error;4039	}4040 4041	/* zero out stat counters */4042	spin_lock(&fs_info->balance_lock);4043	memset(&bctl->stat, 0, sizeof(bctl->stat));4044	spin_unlock(&fs_info->balance_lock);4045again:4046	if (!counting) {4047		/*4048		 * The single value limit and min/max limits use the same bytes4049		 * in the4050		 */4051		bctl->data.limit = limit_data;4052		bctl->meta.limit = limit_meta;4053		bctl->sys.limit = limit_sys;4054	}4055	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;4056	key.offset = (u64)-1;4057	key.type = BTRFS_CHUNK_ITEM_KEY;4058 4059	while (1) {4060		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||4061		    atomic_read(&fs_info->balance_cancel_req)) {4062			ret = -ECANCELED;4063			goto error;4064		}4065 4066		mutex_lock(&fs_info->reclaim_bgs_lock);4067		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);4068		if (ret < 0) {4069			mutex_unlock(&fs_info->reclaim_bgs_lock);4070			goto error;4071		}4072 4073		/*4074		 * this shouldn't happen, it means the last relocate4075		 * failed4076		 */4077		if (ret == 0)4078			BUG(); /* FIXME break ? */4079 4080		ret = btrfs_previous_item(chunk_root, path, 0,4081					  BTRFS_CHUNK_ITEM_KEY);4082		if (ret) {4083			mutex_unlock(&fs_info->reclaim_bgs_lock);4084			ret = 0;4085			break;4086		}4087 4088		leaf = path->nodes[0];4089		slot = path->slots[0];4090		btrfs_item_key_to_cpu(leaf, &found_key, slot);4091 4092		if (found_key.objectid != key.objectid) {4093			mutex_unlock(&fs_info->reclaim_bgs_lock);4094			break;4095		}4096 4097		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);4098		chunk_type = btrfs_chunk_type(leaf, chunk);4099 4100		if (!counting) {4101			spin_lock(&fs_info->balance_lock);4102			bctl->stat.considered++;4103			spin_unlock(&fs_info->balance_lock);4104		}4105 4106		ret = should_balance_chunk(leaf, chunk, found_key.offset);4107 4108		btrfs_release_path(path);4109		if (!ret) {4110			mutex_unlock(&fs_info->reclaim_bgs_lock);4111			goto loop;4112		}4113 4114		if (counting) {4115			mutex_unlock(&fs_info->reclaim_bgs_lock);4116			spin_lock(&fs_info->balance_lock);4117			bctl->stat.expected++;4118			spin_unlock(&fs_info->balance_lock);4119 4120			if (chunk_type & BTRFS_BLOCK_GROUP_DATA)4121				count_data++;4122			else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)4123				count_sys++;4124			else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)4125				count_meta++;4126 4127			goto loop;4128		}4129 4130		/*4131		 * Apply limit_min filter, no need to check if the LIMITS4132		 * filter is used, limit_min is 0 by default4133		 */4134		if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&4135					count_data < bctl->data.limit_min)4136				|| ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&4137					count_meta < bctl->meta.limit_min)4138				|| ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&4139					count_sys < bctl->sys.limit_min)) {4140			mutex_unlock(&fs_info->reclaim_bgs_lock);4141			goto loop;4142		}4143 4144		if (!chunk_reserved) {4145			/*4146			 * We may be relocating the only data chunk we have,4147			 * which could potentially end up with losing data's4148			 * raid profile, so lets allocate an empty one in4149			 * advance.4150			 */4151			ret = btrfs_may_alloc_data_chunk(fs_info,4152							 found_key.offset);4153			if (ret < 0) {4154				mutex_unlock(&fs_info->reclaim_bgs_lock);4155				goto error;4156			} else if (ret == 1) {4157				chunk_reserved = 1;4158			}4159		}4160 4161		ret = btrfs_relocate_chunk(fs_info, found_key.offset);4162		mutex_unlock(&fs_info->reclaim_bgs_lock);4163		if (ret == -ENOSPC) {4164			enospc_errors++;4165		} else if (ret == -ETXTBSY) {4166			btrfs_info(fs_info,4167	   "skipping relocation of block group %llu due to active swapfile",4168				   found_key.offset);4169			ret = 0;4170		} else if (ret) {4171			goto error;4172		} else {4173			spin_lock(&fs_info->balance_lock);4174			bctl->stat.completed++;4175			spin_unlock(&fs_info->balance_lock);4176		}4177loop:4178		if (found_key.offset == 0)4179			break;4180		key.offset = found_key.offset - 1;4181	}4182 4183	if (counting) {4184		btrfs_release_path(path);4185		counting = false;4186		goto again;4187	}4188error:4189	btrfs_free_path(path);4190	if (enospc_errors) {4191		btrfs_info(fs_info, "%d enospc errors during balance",4192			   enospc_errors);4193		if (!ret)4194			ret = -ENOSPC;4195	}4196 4197	return ret;4198}4199 4200/*4201 * See if a given profile is valid and reduced.4202 *4203 * @flags:     profile to validate4204 * @extended:  if true @flags is treated as an extended profile4205 */4206static int alloc_profile_is_valid(u64 flags, int extended)4207{4208	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :4209			       BTRFS_BLOCK_GROUP_PROFILE_MASK);4210 4211	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;4212 4213	/* 1) check that all other bits are zeroed */4214	if (flags & ~mask)4215		return 0;4216 4217	/* 2) see if profile is reduced */4218	if (flags == 0)4219		return !extended; /* "0" is valid for usual profiles */4220 4221	return has_single_bit_set(flags);4222}4223 4224/*4225 * Validate target profile against allowed profiles and return true if it's OK.4226 * Otherwise print the error message and return false.4227 */4228static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,4229		const struct btrfs_balance_args *bargs,4230		u64 allowed, const char *type)4231{4232	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))4233		return true;4234 4235	/* Profile is valid and does not have bits outside of the allowed set */4236	if (alloc_profile_is_valid(bargs->target, 1) &&4237	    (bargs->target & ~allowed) == 0)4238		return true;4239 4240	btrfs_err(fs_info, "balance: invalid convert %s profile %s",4241			type, btrfs_bg_type_to_raid_name(bargs->target));4242	return false;4243}4244 4245/*4246 * Fill @buf with textual description of balance filter flags @bargs, up to4247 * @size_buf including the terminating null. The output may be trimmed if it4248 * does not fit into the provided buffer.4249 */4250static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,4251				 u32 size_buf)4252{4253	int ret;4254	u32 size_bp = size_buf;4255	char *bp = buf;4256	u64 flags = bargs->flags;4257	char tmp_buf[128] = {'\0'};4258 4259	if (!flags)4260		return;4261 4262#define CHECK_APPEND_NOARG(a)						\4263	do {								\4264		ret = snprintf(bp, size_bp, (a));			\4265		if (ret < 0 || ret >= size_bp)				\4266			goto out_overflow;				\4267		size_bp -= ret;						\4268		bp += ret;						\4269	} while (0)4270 4271#define CHECK_APPEND_1ARG(a, v1)					\4272	do {								\4273		ret = snprintf(bp, size_bp, (a), (v1));			\4274		if (ret < 0 || ret >= size_bp)				\4275			goto out_overflow;				\4276		size_bp -= ret;						\4277		bp += ret;						\4278	} while (0)4279 4280#define CHECK_APPEND_2ARG(a, v1, v2)					\4281	do {								\4282		ret = snprintf(bp, size_bp, (a), (v1), (v2));		\4283		if (ret < 0 || ret >= size_bp)				\4284			goto out_overflow;				\4285		size_bp -= ret;						\4286		bp += ret;						\4287	} while (0)4288 4289	if (flags & BTRFS_BALANCE_ARGS_CONVERT)4290		CHECK_APPEND_1ARG("convert=%s,",4291				  btrfs_bg_type_to_raid_name(bargs->target));4292 4293	if (flags & BTRFS_BALANCE_ARGS_SOFT)4294		CHECK_APPEND_NOARG("soft,");4295 4296	if (flags & BTRFS_BALANCE_ARGS_PROFILES) {4297		btrfs_describe_block_groups(bargs->profiles, tmp_buf,4298					    sizeof(tmp_buf));4299		CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);4300	}4301 4302	if (flags & BTRFS_BALANCE_ARGS_USAGE)4303		CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);4304 4305	if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)4306		CHECK_APPEND_2ARG("usage=%u..%u,",4307				  bargs->usage_min, bargs->usage_max);4308 4309	if (flags & BTRFS_BALANCE_ARGS_DEVID)4310		CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);4311 4312	if (flags & BTRFS_BALANCE_ARGS_DRANGE)4313		CHECK_APPEND_2ARG("drange=%llu..%llu,",4314				  bargs->pstart, bargs->pend);4315 4316	if (flags & BTRFS_BALANCE_ARGS_VRANGE)4317		CHECK_APPEND_2ARG("vrange=%llu..%llu,",4318				  bargs->vstart, bargs->vend);4319 4320	if (flags & BTRFS_BALANCE_ARGS_LIMIT)4321		CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);4322 4323	if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)4324		CHECK_APPEND_2ARG("limit=%u..%u,",4325				bargs->limit_min, bargs->limit_max);4326 4327	if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)4328		CHECK_APPEND_2ARG("stripes=%u..%u,",4329				  bargs->stripes_min, bargs->stripes_max);4330 4331#undef CHECK_APPEND_2ARG4332#undef CHECK_APPEND_1ARG4333#undef CHECK_APPEND_NOARG4334 4335out_overflow:4336 4337	if (size_bp < size_buf)4338		buf[size_buf - size_bp - 1] = '\0'; /* remove last , */4339	else4340		buf[0] = '\0';4341}4342 4343static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)4344{4345	u32 size_buf = 1024;4346	char tmp_buf[192] = {'\0'};4347	char *buf;4348	char *bp;4349	u32 size_bp = size_buf;4350	int ret;4351	struct btrfs_balance_control *bctl = fs_info->balance_ctl;4352 4353	buf = kzalloc(size_buf, GFP_KERNEL);4354	if (!buf)4355		return;4356 4357	bp = buf;4358 4359#define CHECK_APPEND_1ARG(a, v1)					\4360	do {								\4361		ret = snprintf(bp, size_bp, (a), (v1));			\4362		if (ret < 0 || ret >= size_bp)				\4363			goto out_overflow;				\4364		size_bp -= ret;						\4365		bp += ret;						\4366	} while (0)4367 4368	if (bctl->flags & BTRFS_BALANCE_FORCE)4369		CHECK_APPEND_1ARG("%s", "-f ");4370 4371	if (bctl->flags & BTRFS_BALANCE_DATA) {4372		describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));4373		CHECK_APPEND_1ARG("-d%s ", tmp_buf);4374	}4375 4376	if (bctl->flags & BTRFS_BALANCE_METADATA) {4377		describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));4378		CHECK_APPEND_1ARG("-m%s ", tmp_buf);4379	}4380 4381	if (bctl->flags & BTRFS_BALANCE_SYSTEM) {4382		describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));4383		CHECK_APPEND_1ARG("-s%s ", tmp_buf);4384	}4385 4386#undef CHECK_APPEND_1ARG4387 4388out_overflow:4389 4390	if (size_bp < size_buf)4391		buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */4392	btrfs_info(fs_info, "balance: %s %s",4393		   (bctl->flags & BTRFS_BALANCE_RESUME) ?4394		   "resume" : "start", buf);4395 4396	kfree(buf);4397}4398 4399/*4400 * Should be called with balance mutexe held4401 */4402int btrfs_balance(struct btrfs_fs_info *fs_info,4403		  struct btrfs_balance_control *bctl,4404		  struct btrfs_ioctl_balance_args *bargs)4405{4406	u64 meta_target, data_target;4407	u64 allowed;4408	int mixed = 0;4409	int ret;4410	u64 num_devices;4411	unsigned seq;4412	bool reducing_redundancy;4413	bool paused = false;4414	int i;4415 4416	if (btrfs_fs_closing(fs_info) ||4417	    atomic_read(&fs_info->balance_pause_req) ||4418	    btrfs_should_cancel_balance(fs_info)) {4419		ret = -EINVAL;4420		goto out;4421	}4422 4423	allowed = btrfs_super_incompat_flags(fs_info->super_copy);4424	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)4425		mixed = 1;4426 4427	/*4428	 * In case of mixed groups both data and meta should be picked,4429	 * and identical options should be given for both of them.4430	 */4431	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;4432	if (mixed && (bctl->flags & allowed)) {4433		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||4434		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||4435		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {4436			btrfs_err(fs_info,4437	  "balance: mixed groups data and metadata options must be the same");4438			ret = -EINVAL;4439			goto out;4440		}4441	}4442 4443	/*4444	 * rw_devices will not change at the moment, device add/delete/replace4445	 * are exclusive4446	 */4447	num_devices = fs_info->fs_devices->rw_devices;4448 4449	/*4450	 * SINGLE profile on-disk has no profile bit, but in-memory we have a4451	 * special bit for it, to make it easier to distinguish.  Thus we need4452	 * to set it manually, or balance would refuse the profile.4453	 */4454	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;4455	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)4456		if (num_devices >= btrfs_raid_array[i].devs_min)4457			allowed |= btrfs_raid_array[i].bg_flag;4458 4459	if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||4460	    !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||4461	    !validate_convert_profile(fs_info, &bctl->sys,  allowed, "system")) {4462		ret = -EINVAL;4463		goto out;4464	}4465 4466	/*4467	 * Allow to reduce metadata or system integrity only if force set for4468	 * profiles with redundancy (copies, parity)4469	 */4470	allowed = 0;4471	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {4472		if (btrfs_raid_array[i].ncopies >= 2 ||4473		    btrfs_raid_array[i].tolerated_failures >= 1)4474			allowed |= btrfs_raid_array[i].bg_flag;4475	}4476	do {4477		seq = read_seqbegin(&fs_info->profiles_lock);4478 4479		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&4480		     (fs_info->avail_system_alloc_bits & allowed) &&4481		     !(bctl->sys.target & allowed)) ||4482		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&4483		     (fs_info->avail_metadata_alloc_bits & allowed) &&4484		     !(bctl->meta.target & allowed)))4485			reducing_redundancy = true;4486		else4487			reducing_redundancy = false;4488 4489		/* if we're not converting, the target field is uninitialized */4490		meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?4491			bctl->meta.target : fs_info->avail_metadata_alloc_bits;4492		data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?4493			bctl->data.target : fs_info->avail_data_alloc_bits;4494	} while (read_seqretry(&fs_info->profiles_lock, seq));4495 4496	if (reducing_redundancy) {4497		if (bctl->flags & BTRFS_BALANCE_FORCE) {4498			btrfs_info(fs_info,4499			   "balance: force reducing metadata redundancy");4500		} else {4501			btrfs_err(fs_info,4502	"balance: reduces metadata redundancy, use --force if you want this");4503			ret = -EINVAL;4504			goto out;4505		}4506	}4507 4508	if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <4509		btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {4510		btrfs_warn(fs_info,4511	"balance: metadata profile %s has lower redundancy than data profile %s",4512				btrfs_bg_type_to_raid_name(meta_target),4513				btrfs_bg_type_to_raid_name(data_target));4514	}4515 4516	ret = insert_balance_item(fs_info, bctl);4517	if (ret && ret != -EEXIST)4518		goto out;4519 4520	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {4521		BUG_ON(ret == -EEXIST);4522		BUG_ON(fs_info->balance_ctl);4523		spin_lock(&fs_info->balance_lock);4524		fs_info->balance_ctl = bctl;4525		spin_unlock(&fs_info->balance_lock);4526	} else {4527		BUG_ON(ret != -EEXIST);4528		spin_lock(&fs_info->balance_lock);4529		update_balance_args(bctl);4530		spin_unlock(&fs_info->balance_lock);4531	}4532 4533	ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));4534	set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);4535	describe_balance_start_or_resume(fs_info);4536	mutex_unlock(&fs_info->balance_mutex);4537 4538	ret = __btrfs_balance(fs_info);4539 4540	mutex_lock(&fs_info->balance_mutex);4541	if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {4542		btrfs_info(fs_info, "balance: paused");4543		btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);4544		paused = true;4545	}4546	/*4547	 * Balance can be canceled by:4548	 *4549	 * - Regular cancel request4550	 *   Then ret == -ECANCELED and balance_cancel_req > 04551	 *4552	 * - Fatal signal to "btrfs" process4553	 *   Either the signal caught by wait_reserve_ticket() and callers4554	 *   got -EINTR, or caught by btrfs_should_cancel_balance() and4555	 *   got -ECANCELED.4556	 *   Either way, in this case balance_cancel_req = 0, and4557	 *   ret == -EINTR or ret == -ECANCELED.4558	 *4559	 * So here we only check the return value to catch canceled balance.4560	 */4561	else if (ret == -ECANCELED || ret == -EINTR)4562		btrfs_info(fs_info, "balance: canceled");4563	else4564		btrfs_info(fs_info, "balance: ended with status: %d", ret);4565 4566	clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);4567 4568	if (bargs) {4569		memset(bargs, 0, sizeof(*bargs));4570		btrfs_update_ioctl_balance_args(fs_info, bargs);4571	}4572 4573	/* We didn't pause, we can clean everything up. */4574	if (!paused) {4575		reset_balance_state(fs_info);4576		btrfs_exclop_finish(fs_info);4577	}4578 4579	wake_up(&fs_info->balance_wait_q);4580 4581	return ret;4582out:4583	if (bctl->flags & BTRFS_BALANCE_RESUME)4584		reset_balance_state(fs_info);4585	else4586		kfree(bctl);4587	btrfs_exclop_finish(fs_info);4588 4589	return ret;4590}4591 4592static int balance_kthread(void *data)4593{4594	struct btrfs_fs_info *fs_info = data;4595	int ret = 0;4596 4597	sb_start_write(fs_info->sb);4598	mutex_lock(&fs_info->balance_mutex);4599	if (fs_info->balance_ctl)4600		ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);4601	mutex_unlock(&fs_info->balance_mutex);4602	sb_end_write(fs_info->sb);4603 4604	return ret;4605}4606 4607int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)4608{4609	struct task_struct *tsk;4610 4611	mutex_lock(&fs_info->balance_mutex);4612	if (!fs_info->balance_ctl) {4613		mutex_unlock(&fs_info->balance_mutex);4614		return 0;4615	}4616	mutex_unlock(&fs_info->balance_mutex);4617 4618	if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {4619		btrfs_info(fs_info, "balance: resume skipped");4620		return 0;4621	}4622 4623	spin_lock(&fs_info->super_lock);4624	ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);4625	fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;4626	spin_unlock(&fs_info->super_lock);4627	/*4628	 * A ro->rw remount sequence should continue with the paused balance4629	 * regardless of who pauses it, system or the user as of now, so set4630	 * the resume flag.4631	 */4632	spin_lock(&fs_info->balance_lock);4633	fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;4634	spin_unlock(&fs_info->balance_lock);4635 4636	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");4637	return PTR_ERR_OR_ZERO(tsk);4638}4639 4640int btrfs_recover_balance(struct btrfs_fs_info *fs_info)4641{4642	struct btrfs_balance_control *bctl;4643	struct btrfs_balance_item *item;4644	struct btrfs_disk_balance_args disk_bargs;4645	struct btrfs_path *path;4646	struct extent_buffer *leaf;4647	struct btrfs_key key;4648	int ret;4649 4650	path = btrfs_alloc_path();4651	if (!path)4652		return -ENOMEM;4653 4654	key.objectid = BTRFS_BALANCE_OBJECTID;4655	key.type = BTRFS_TEMPORARY_ITEM_KEY;4656	key.offset = 0;4657 4658	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);4659	if (ret < 0)4660		goto out;4661	if (ret > 0) { /* ret = -ENOENT; */4662		ret = 0;4663		goto out;4664	}4665 4666	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);4667	if (!bctl) {4668		ret = -ENOMEM;4669		goto out;4670	}4671 4672	leaf = path->nodes[0];4673	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);4674 4675	bctl->flags = btrfs_balance_flags(leaf, item);4676	bctl->flags |= BTRFS_BALANCE_RESUME;4677 4678	btrfs_balance_data(leaf, item, &disk_bargs);4679	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);4680	btrfs_balance_meta(leaf, item, &disk_bargs);4681	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);4682	btrfs_balance_sys(leaf, item, &disk_bargs);4683	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);4684 4685	/*4686	 * This should never happen, as the paused balance state is recovered4687	 * during mount without any chance of other exclusive ops to collide.4688	 *4689	 * This gives the exclusive op status to balance and keeps in paused4690	 * state until user intervention (cancel or umount). If the ownership4691	 * cannot be assigned, show a message but do not fail. The balance4692	 * is in a paused state and must have fs_info::balance_ctl properly4693	 * set up.4694	 */4695	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))4696		btrfs_warn(fs_info,4697	"balance: cannot set exclusive op status, resume manually");4698 4699	btrfs_release_path(path);4700 4701	mutex_lock(&fs_info->balance_mutex);4702	BUG_ON(fs_info->balance_ctl);4703	spin_lock(&fs_info->balance_lock);4704	fs_info->balance_ctl = bctl;4705	spin_unlock(&fs_info->balance_lock);4706	mutex_unlock(&fs_info->balance_mutex);4707out:4708	btrfs_free_path(path);4709	return ret;4710}4711 4712int btrfs_pause_balance(struct btrfs_fs_info *fs_info)4713{4714	int ret = 0;4715 4716	mutex_lock(&fs_info->balance_mutex);4717	if (!fs_info->balance_ctl) {4718		mutex_unlock(&fs_info->balance_mutex);4719		return -ENOTCONN;4720	}4721 4722	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {4723		atomic_inc(&fs_info->balance_pause_req);4724		mutex_unlock(&fs_info->balance_mutex);4725 4726		wait_event(fs_info->balance_wait_q,4727			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));4728 4729		mutex_lock(&fs_info->balance_mutex);4730		/* we are good with balance_ctl ripped off from under us */4731		BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));4732		atomic_dec(&fs_info->balance_pause_req);4733	} else {4734		ret = -ENOTCONN;4735	}4736 4737	mutex_unlock(&fs_info->balance_mutex);4738	return ret;4739}4740 4741int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)4742{4743	mutex_lock(&fs_info->balance_mutex);4744	if (!fs_info->balance_ctl) {4745		mutex_unlock(&fs_info->balance_mutex);4746		return -ENOTCONN;4747	}4748 4749	/*4750	 * A paused balance with the item stored on disk can be resumed at4751	 * mount time if the mount is read-write. Otherwise it's still paused4752	 * and we must not allow cancelling as it deletes the item.4753	 */4754	if (sb_rdonly(fs_info->sb)) {4755		mutex_unlock(&fs_info->balance_mutex);4756		return -EROFS;4757	}4758 4759	atomic_inc(&fs_info->balance_cancel_req);4760	/*4761	 * if we are running just wait and return, balance item is4762	 * deleted in btrfs_balance in this case4763	 */4764	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {4765		mutex_unlock(&fs_info->balance_mutex);4766		wait_event(fs_info->balance_wait_q,4767			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));4768		mutex_lock(&fs_info->balance_mutex);4769	} else {4770		mutex_unlock(&fs_info->balance_mutex);4771		/*4772		 * Lock released to allow other waiters to continue, we'll4773		 * reexamine the status again.4774		 */4775		mutex_lock(&fs_info->balance_mutex);4776 4777		if (fs_info->balance_ctl) {4778			reset_balance_state(fs_info);4779			btrfs_exclop_finish(fs_info);4780			btrfs_info(fs_info, "balance: canceled");4781		}4782	}4783 4784	ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));4785	atomic_dec(&fs_info->balance_cancel_req);4786	mutex_unlock(&fs_info->balance_mutex);4787	return 0;4788}4789 4790/*4791 * shrinking a device means finding all of the device extents past4792 * the new size, and then following the back refs to the chunks.4793 * The chunk relocation code actually frees the device extent4794 */4795int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)4796{4797	struct btrfs_fs_info *fs_info = device->fs_info;4798	struct btrfs_root *root = fs_info->dev_root;4799	struct btrfs_trans_handle *trans;4800	struct btrfs_dev_extent *dev_extent = NULL;4801	struct btrfs_path *path;4802	u64 length;4803	u64 chunk_offset;4804	int ret;4805	int slot;4806	int failed = 0;4807	bool retried = false;4808	struct extent_buffer *l;4809	struct btrfs_key key;4810	struct btrfs_super_block *super_copy = fs_info->super_copy;4811	u64 old_total = btrfs_super_total_bytes(super_copy);4812	u64 old_size = btrfs_device_get_total_bytes(device);4813	u64 diff;4814	u64 start;4815	u64 free_diff = 0;4816 4817	new_size = round_down(new_size, fs_info->sectorsize);4818	start = new_size;4819	diff = round_down(old_size - new_size, fs_info->sectorsize);4820 4821	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))4822		return -EINVAL;4823 4824	path = btrfs_alloc_path();4825	if (!path)4826		return -ENOMEM;4827 4828	path->reada = READA_BACK;4829 4830	trans = btrfs_start_transaction(root, 0);4831	if (IS_ERR(trans)) {4832		btrfs_free_path(path);4833		return PTR_ERR(trans);4834	}4835 4836	mutex_lock(&fs_info->chunk_mutex);4837 4838	btrfs_device_set_total_bytes(device, new_size);4839	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {4840		device->fs_devices->total_rw_bytes -= diff;4841 4842		/*4843		 * The new free_chunk_space is new_size - used, so we have to4844		 * subtract the delta of the old free_chunk_space which included4845		 * old_size - used.  If used > new_size then just subtract this4846		 * entire device's free space.4847		 */4848		if (device->bytes_used < new_size)4849			free_diff = (old_size - device->bytes_used) -4850				    (new_size - device->bytes_used);4851		else4852			free_diff = old_size - device->bytes_used;4853		atomic64_sub(free_diff, &fs_info->free_chunk_space);4854	}4855 4856	/*4857	 * Once the device's size has been set to the new size, ensure all4858	 * in-memory chunks are synced to disk so that the loop below sees them4859	 * and relocates them accordingly.4860	 */4861	if (contains_pending_extent(device, &start, diff)) {4862		mutex_unlock(&fs_info->chunk_mutex);4863		ret = btrfs_commit_transaction(trans);4864		if (ret)4865			goto done;4866	} else {4867		mutex_unlock(&fs_info->chunk_mutex);4868		btrfs_end_transaction(trans);4869	}4870 4871again:4872	key.objectid = device->devid;4873	key.offset = (u64)-1;4874	key.type = BTRFS_DEV_EXTENT_KEY;4875 4876	do {4877		mutex_lock(&fs_info->reclaim_bgs_lock);4878		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);4879		if (ret < 0) {4880			mutex_unlock(&fs_info->reclaim_bgs_lock);4881			goto done;4882		}4883 4884		ret = btrfs_previous_item(root, path, 0, key.type);4885		if (ret) {4886			mutex_unlock(&fs_info->reclaim_bgs_lock);4887			if (ret < 0)4888				goto done;4889			ret = 0;4890			btrfs_release_path(path);4891			break;4892		}4893 4894		l = path->nodes[0];4895		slot = path->slots[0];4896		btrfs_item_key_to_cpu(l, &key, path->slots[0]);4897 4898		if (key.objectid != device->devid) {4899			mutex_unlock(&fs_info->reclaim_bgs_lock);4900			btrfs_release_path(path);4901			break;4902		}4903 4904		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);4905		length = btrfs_dev_extent_length(l, dev_extent);4906 4907		if (key.offset + length <= new_size) {4908			mutex_unlock(&fs_info->reclaim_bgs_lock);4909			btrfs_release_path(path);4910			break;4911		}4912 4913		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);4914		btrfs_release_path(path);4915 4916		/*4917		 * We may be relocating the only data chunk we have,4918		 * which could potentially end up with losing data's4919		 * raid profile, so lets allocate an empty one in4920		 * advance.4921		 */4922		ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);4923		if (ret < 0) {4924			mutex_unlock(&fs_info->reclaim_bgs_lock);4925			goto done;4926		}4927 4928		ret = btrfs_relocate_chunk(fs_info, chunk_offset);4929		mutex_unlock(&fs_info->reclaim_bgs_lock);4930		if (ret == -ENOSPC) {4931			failed++;4932		} else if (ret) {4933			if (ret == -ETXTBSY) {4934				btrfs_warn(fs_info,4935		   "could not shrink block group %llu due to active swapfile",4936					   chunk_offset);4937			}4938			goto done;4939		}4940	} while (key.offset-- > 0);4941 4942	if (failed && !retried) {4943		failed = 0;4944		retried = true;4945		goto again;4946	} else if (failed && retried) {4947		ret = -ENOSPC;4948		goto done;4949	}4950 4951	/* Shrinking succeeded, else we would be at "done". */4952	trans = btrfs_start_transaction(root, 0);4953	if (IS_ERR(trans)) {4954		ret = PTR_ERR(trans);4955		goto done;4956	}4957 4958	mutex_lock(&fs_info->chunk_mutex);4959	/* Clear all state bits beyond the shrunk device size */4960	clear_extent_bits(&device->alloc_state, new_size, (u64)-1,4961			  CHUNK_STATE_MASK);4962 4963	btrfs_device_set_disk_total_bytes(device, new_size);4964	if (list_empty(&device->post_commit_list))4965		list_add_tail(&device->post_commit_list,4966			      &trans->transaction->dev_update_list);4967 4968	WARN_ON(diff > old_total);4969	btrfs_set_super_total_bytes(super_copy,4970			round_down(old_total - diff, fs_info->sectorsize));4971	mutex_unlock(&fs_info->chunk_mutex);4972 4973	btrfs_reserve_chunk_metadata(trans, false);4974	/* Now btrfs_update_device() will change the on-disk size. */4975	ret = btrfs_update_device(trans, device);4976	btrfs_trans_release_chunk_metadata(trans);4977	if (ret < 0) {4978		btrfs_abort_transaction(trans, ret);4979		btrfs_end_transaction(trans);4980	} else {4981		ret = btrfs_commit_transaction(trans);4982	}4983done:4984	btrfs_free_path(path);4985	if (ret) {4986		mutex_lock(&fs_info->chunk_mutex);4987		btrfs_device_set_total_bytes(device, old_size);4988		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {4989			device->fs_devices->total_rw_bytes += diff;4990			atomic64_add(free_diff, &fs_info->free_chunk_space);4991		}4992		mutex_unlock(&fs_info->chunk_mutex);4993	}4994	return ret;4995}4996 4997static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,4998			   struct btrfs_key *key,4999			   struct btrfs_chunk *chunk, int item_size)5000{5001	struct btrfs_super_block *super_copy = fs_info->super_copy;5002	struct btrfs_disk_key disk_key;5003	u32 array_size;5004	u8 *ptr;5005 5006	lockdep_assert_held(&fs_info->chunk_mutex);5007 5008	array_size = btrfs_super_sys_array_size(super_copy);5009	if (array_size + item_size + sizeof(disk_key)5010			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)5011		return -EFBIG;5012 5013	ptr = super_copy->sys_chunk_array + array_size;5014	btrfs_cpu_key_to_disk(&disk_key, key);5015	memcpy(ptr, &disk_key, sizeof(disk_key));5016	ptr += sizeof(disk_key);5017	memcpy(ptr, chunk, item_size);5018	item_size += sizeof(disk_key);5019	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);5020 5021	return 0;5022}5023 5024/*5025 * sort the devices in descending order by max_avail, total_avail5026 */5027static int btrfs_cmp_device_info(const void *a, const void *b)5028{5029	const struct btrfs_device_info *di_a = a;5030	const struct btrfs_device_info *di_b = b;5031 5032	if (di_a->max_avail > di_b->max_avail)5033		return -1;5034	if (di_a->max_avail < di_b->max_avail)5035		return 1;5036	if (di_a->total_avail > di_b->total_avail)5037		return -1;5038	if (di_a->total_avail < di_b->total_avail)5039		return 1;5040	return 0;5041}5042 5043static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)5044{5045	if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))5046		return;5047 5048	btrfs_set_fs_incompat(info, RAID56);5049}5050 5051static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)5052{5053	if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))5054		return;5055 5056	btrfs_set_fs_incompat(info, RAID1C34);5057}5058 5059/*5060 * Structure used internally for btrfs_create_chunk() function.5061 * Wraps needed parameters.5062 */5063struct alloc_chunk_ctl {5064	u64 start;5065	u64 type;5066	/* Total number of stripes to allocate */5067	int num_stripes;5068	/* sub_stripes info for map */5069	int sub_stripes;5070	/* Stripes per device */5071	int dev_stripes;5072	/* Maximum number of devices to use */5073	int devs_max;5074	/* Minimum number of devices to use */5075	int devs_min;5076	/* ndevs has to be a multiple of this */5077	int devs_increment;5078	/* Number of copies */5079	int ncopies;5080	/* Number of stripes worth of bytes to store parity information */5081	int nparity;5082	u64 max_stripe_size;5083	u64 max_chunk_size;5084	u64 dev_extent_min;5085	u64 stripe_size;5086	u64 chunk_size;5087	int ndevs;5088};5089 5090static void init_alloc_chunk_ctl_policy_regular(5091				struct btrfs_fs_devices *fs_devices,5092				struct alloc_chunk_ctl *ctl)5093{5094	struct btrfs_space_info *space_info;5095 5096	space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);5097	ASSERT(space_info);5098 5099	ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);5100	ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G);5101 5102	if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)5103		ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);5104 5105	/* We don't want a chunk larger than 10% of writable space */5106	ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),5107				  ctl->max_chunk_size);5108	ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);5109}5110 5111static void init_alloc_chunk_ctl_policy_zoned(5112				      struct btrfs_fs_devices *fs_devices,5113				      struct alloc_chunk_ctl *ctl)5114{5115	u64 zone_size = fs_devices->fs_info->zone_size;5116	u64 limit;5117	int min_num_stripes = ctl->devs_min * ctl->dev_stripes;5118	int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;5119	u64 min_chunk_size = min_data_stripes * zone_size;5120	u64 type = ctl->type;5121 5122	ctl->max_stripe_size = zone_size;5123	if (type & BTRFS_BLOCK_GROUP_DATA) {5124		ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,5125						 zone_size);5126	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {5127		ctl->max_chunk_size = ctl->max_stripe_size;5128	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {5129		ctl->max_chunk_size = 2 * ctl->max_stripe_size;5130		ctl->devs_max = min_t(int, ctl->devs_max,5131				      BTRFS_MAX_DEVS_SYS_CHUNK);5132	} else {5133		BUG();5134	}5135 5136	/* We don't want a chunk larger than 10% of writable space */5137	limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),5138			       zone_size),5139		    min_chunk_size);5140	ctl->max_chunk_size = min(limit, ctl->max_chunk_size);5141	ctl->dev_extent_min = zone_size * ctl->dev_stripes;5142}5143 5144static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,5145				 struct alloc_chunk_ctl *ctl)5146{5147	int index = btrfs_bg_flags_to_raid_index(ctl->type);5148 5149	ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;5150	ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;5151	ctl->devs_max = btrfs_raid_array[index].devs_max;5152	if (!ctl->devs_max)5153		ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);5154	ctl->devs_min = btrfs_raid_array[index].devs_min;5155	ctl->devs_increment = btrfs_raid_array[index].devs_increment;5156	ctl->ncopies = btrfs_raid_array[index].ncopies;5157	ctl->nparity = btrfs_raid_array[index].nparity;5158	ctl->ndevs = 0;5159 5160	switch (fs_devices->chunk_alloc_policy) {5161	case BTRFS_CHUNK_ALLOC_REGULAR:5162		init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);5163		break;5164	case BTRFS_CHUNK_ALLOC_ZONED:5165		init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);5166		break;5167	default:5168		BUG();5169	}5170}5171 5172static int gather_device_info(struct btrfs_fs_devices *fs_devices,5173			      struct alloc_chunk_ctl *ctl,5174			      struct btrfs_device_info *devices_info)5175{5176	struct btrfs_fs_info *info = fs_devices->fs_info;5177	struct btrfs_device *device;5178	u64 total_avail;5179	u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;5180	int ret;5181	int ndevs = 0;5182	u64 max_avail;5183	u64 dev_offset;5184 5185	/*5186	 * in the first pass through the devices list, we gather information5187	 * about the available holes on each device.5188	 */5189	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {5190		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {5191			WARN(1, KERN_ERR5192			       "BTRFS: read-only device in alloc_list\n");5193			continue;5194		}5195 5196		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,5197					&device->dev_state) ||5198		    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))5199			continue;5200 5201		if (device->total_bytes > device->bytes_used)5202			total_avail = device->total_bytes - device->bytes_used;5203		else5204			total_avail = 0;5205 5206		/* If there is no space on this device, skip it. */5207		if (total_avail < ctl->dev_extent_min)5208			continue;5209 5210		ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,5211					   &max_avail);5212		if (ret && ret != -ENOSPC)5213			return ret;5214 5215		if (ret == 0)5216			max_avail = dev_extent_want;5217 5218		if (max_avail < ctl->dev_extent_min) {5219			if (btrfs_test_opt(info, ENOSPC_DEBUG))5220				btrfs_debug(info,5221			"%s: devid %llu has no free space, have=%llu want=%llu",5222					    __func__, device->devid, max_avail,5223					    ctl->dev_extent_min);5224			continue;5225		}5226 5227		if (ndevs == fs_devices->rw_devices) {5228			WARN(1, "%s: found more than %llu devices\n",5229			     __func__, fs_devices->rw_devices);5230			break;5231		}5232		devices_info[ndevs].dev_offset = dev_offset;5233		devices_info[ndevs].max_avail = max_avail;5234		devices_info[ndevs].total_avail = total_avail;5235		devices_info[ndevs].dev = device;5236		++ndevs;5237	}5238	ctl->ndevs = ndevs;5239 5240	/*5241	 * now sort the devices by hole size / available space5242	 */5243	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),5244	     btrfs_cmp_device_info, NULL);5245 5246	return 0;5247}5248 5249static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,5250				      struct btrfs_device_info *devices_info)5251{5252	/* Number of stripes that count for block group size */5253	int data_stripes;5254 5255	/*5256	 * The primary goal is to maximize the number of stripes, so use as5257	 * many devices as possible, even if the stripes are not maximum sized.5258	 *5259	 * The DUP profile stores more than one stripe per device, the5260	 * max_avail is the total size so we have to adjust.5261	 */5262	ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,5263				   ctl->dev_stripes);5264	ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;5265 5266	/* This will have to be fixed for RAID1 and RAID10 over more drives */5267	data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;5268 5269	/*5270	 * Use the number of data stripes to figure out how big this chunk is5271	 * really going to be in terms of logical address space, and compare5272	 * that answer with the max chunk size. If it's higher, we try to5273	 * reduce stripe_size.5274	 */5275	if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {5276		/*5277		 * Reduce stripe_size, round it up to a 16MB boundary again and5278		 * then use it, unless it ends up being even bigger than the5279		 * previous value we had already.5280		 */5281		ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,5282							data_stripes), SZ_16M),5283				       ctl->stripe_size);5284	}5285 5286	/* Stripe size should not go beyond 1G. */5287	ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);5288 5289	/* Align to BTRFS_STRIPE_LEN */5290	ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);5291	ctl->chunk_size = ctl->stripe_size * data_stripes;5292 5293	return 0;5294}5295 5296static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,5297				    struct btrfs_device_info *devices_info)5298{5299	u64 zone_size = devices_info[0].dev->zone_info->zone_size;5300	/* Number of stripes that count for block group size */5301	int data_stripes;5302 5303	/*5304	 * It should hold because:5305	 *    dev_extent_min == dev_extent_want == zone_size * dev_stripes5306	 */5307	ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);5308 5309	ctl->stripe_size = zone_size;5310	ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;5311	data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;5312 5313	/* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */5314	if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {5315		ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,5316					     ctl->stripe_size) + ctl->nparity,5317				     ctl->dev_stripes);5318		ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;5319		data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;5320		ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);5321	}5322 5323	ctl->chunk_size = ctl->stripe_size * data_stripes;5324 5325	return 0;5326}5327 5328static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,5329			      struct alloc_chunk_ctl *ctl,5330			      struct btrfs_device_info *devices_info)5331{5332	struct btrfs_fs_info *info = fs_devices->fs_info;5333 5334	/*5335	 * Round down to number of usable stripes, devs_increment can be any5336	 * number so we can't use round_down() that requires power of 2, while5337	 * rounddown is safe.5338	 */5339	ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);5340 5341	if (ctl->ndevs < ctl->devs_min) {5342		if (btrfs_test_opt(info, ENOSPC_DEBUG)) {5343			btrfs_debug(info,5344	"%s: not enough devices with free space: have=%d minimum required=%d",5345				    __func__, ctl->ndevs, ctl->devs_min);5346		}5347		return -ENOSPC;5348	}5349 5350	ctl->ndevs = min(ctl->ndevs, ctl->devs_max);5351 5352	switch (fs_devices->chunk_alloc_policy) {5353	case BTRFS_CHUNK_ALLOC_REGULAR:5354		return decide_stripe_size_regular(ctl, devices_info);5355	case BTRFS_CHUNK_ALLOC_ZONED:5356		return decide_stripe_size_zoned(ctl, devices_info);5357	default:5358		BUG();5359	}5360}5361 5362static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits)5363{5364	for (int i = 0; i < map->num_stripes; i++) {5365		struct btrfs_io_stripe *stripe = &map->stripes[i];5366		struct btrfs_device *device = stripe->dev;5367 5368		set_extent_bit(&device->alloc_state, stripe->physical,5369			       stripe->physical + map->stripe_size - 1,5370			       bits | EXTENT_NOWAIT, NULL);5371	}5372}5373 5374static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits)5375{5376	for (int i = 0; i < map->num_stripes; i++) {5377		struct btrfs_io_stripe *stripe = &map->stripes[i];5378		struct btrfs_device *device = stripe->dev;5379 5380		__clear_extent_bit(&device->alloc_state, stripe->physical,5381				   stripe->physical + map->stripe_size - 1,5382				   bits | EXTENT_NOWAIT,5383				   NULL, NULL);5384	}5385}5386 5387void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)5388{5389	write_lock(&fs_info->mapping_tree_lock);5390	rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);5391	RB_CLEAR_NODE(&map->rb_node);5392	chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);5393	write_unlock(&fs_info->mapping_tree_lock);5394 5395	/* Once for the tree reference. */5396	btrfs_free_chunk_map(map);5397}5398 5399EXPORT_FOR_TESTS5400int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)5401{5402	struct rb_node **p;5403	struct rb_node *parent = NULL;5404	bool leftmost = true;5405 5406	write_lock(&fs_info->mapping_tree_lock);5407	p = &fs_info->mapping_tree.rb_root.rb_node;5408	while (*p) {5409		struct btrfs_chunk_map *entry;5410 5411		parent = *p;5412		entry = rb_entry(parent, struct btrfs_chunk_map, rb_node);5413 5414		if (map->start < entry->start) {5415			p = &(*p)->rb_left;5416		} else if (map->start > entry->start) {5417			p = &(*p)->rb_right;5418			leftmost = false;5419		} else {5420			write_unlock(&fs_info->mapping_tree_lock);5421			return -EEXIST;5422		}5423	}5424	rb_link_node(&map->rb_node, parent, p);5425	rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost);5426	chunk_map_device_set_bits(map, CHUNK_ALLOCATED);5427	chunk_map_device_clear_bits(map, CHUNK_TRIMMED);5428	write_unlock(&fs_info->mapping_tree_lock);5429 5430	return 0;5431}5432 5433EXPORT_FOR_TESTS5434struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp)5435{5436	struct btrfs_chunk_map *map;5437 5438	map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp);5439	if (!map)5440		return NULL;5441 5442	refcount_set(&map->refs, 1);5443	RB_CLEAR_NODE(&map->rb_node);5444 5445	return map;5446}5447 5448static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,5449			struct alloc_chunk_ctl *ctl,5450			struct btrfs_device_info *devices_info)5451{5452	struct btrfs_fs_info *info = trans->fs_info;5453	struct btrfs_chunk_map *map;5454	struct btrfs_block_group *block_group;5455	u64 start = ctl->start;5456	u64 type = ctl->type;5457	int ret;5458 5459	map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS);5460	if (!map)5461		return ERR_PTR(-ENOMEM);5462 5463	map->start = start;5464	map->chunk_len = ctl->chunk_size;5465	map->stripe_size = ctl->stripe_size;5466	map->type = type;5467	map->io_align = BTRFS_STRIPE_LEN;5468	map->io_width = BTRFS_STRIPE_LEN;5469	map->sub_stripes = ctl->sub_stripes;5470	map->num_stripes = ctl->num_stripes;5471 5472	for (int i = 0; i < ctl->ndevs; i++) {5473		for (int j = 0; j < ctl->dev_stripes; j++) {5474			int s = i * ctl->dev_stripes + j;5475			map->stripes[s].dev = devices_info[i].dev;5476			map->stripes[s].physical = devices_info[i].dev_offset +5477						   j * ctl->stripe_size;5478		}5479	}5480 5481	trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);5482 5483	ret = btrfs_add_chunk_map(info, map);5484	if (ret) {5485		btrfs_free_chunk_map(map);5486		return ERR_PTR(ret);5487	}5488 5489	block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);5490	if (IS_ERR(block_group)) {5491		btrfs_remove_chunk_map(info, map);5492		return block_group;5493	}5494 5495	for (int i = 0; i < map->num_stripes; i++) {5496		struct btrfs_device *dev = map->stripes[i].dev;5497 5498		btrfs_device_set_bytes_used(dev,5499					    dev->bytes_used + ctl->stripe_size);5500		if (list_empty(&dev->post_commit_list))5501			list_add_tail(&dev->post_commit_list,5502				      &trans->transaction->dev_update_list);5503	}5504 5505	atomic64_sub(ctl->stripe_size * map->num_stripes,5506		     &info->free_chunk_space);5507 5508	check_raid56_incompat_flag(info, type);5509	check_raid1c34_incompat_flag(info, type);5510 5511	return block_group;5512}5513 5514struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,5515					    u64 type)5516{5517	struct btrfs_fs_info *info = trans->fs_info;5518	struct btrfs_fs_devices *fs_devices = info->fs_devices;5519	struct btrfs_device_info *devices_info = NULL;5520	struct alloc_chunk_ctl ctl;5521	struct btrfs_block_group *block_group;5522	int ret;5523 5524	lockdep_assert_held(&info->chunk_mutex);5525 5526	if (!alloc_profile_is_valid(type, 0)) {5527		ASSERT(0);5528		return ERR_PTR(-EINVAL);5529	}5530 5531	if (list_empty(&fs_devices->alloc_list)) {5532		if (btrfs_test_opt(info, ENOSPC_DEBUG))5533			btrfs_debug(info, "%s: no writable device", __func__);5534		return ERR_PTR(-ENOSPC);5535	}5536 5537	if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {5538		btrfs_err(info, "invalid chunk type 0x%llx requested", type);5539		ASSERT(0);5540		return ERR_PTR(-EINVAL);5541	}5542 5543	ctl.start = find_next_chunk(info);5544	ctl.type = type;5545	init_alloc_chunk_ctl(fs_devices, &ctl);5546 5547	devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),5548			       GFP_NOFS);5549	if (!devices_info)5550		return ERR_PTR(-ENOMEM);5551 5552	ret = gather_device_info(fs_devices, &ctl, devices_info);5553	if (ret < 0) {5554		block_group = ERR_PTR(ret);5555		goto out;5556	}5557 5558	ret = decide_stripe_size(fs_devices, &ctl, devices_info);5559	if (ret < 0) {5560		block_group = ERR_PTR(ret);5561		goto out;5562	}5563 5564	block_group = create_chunk(trans, &ctl, devices_info);5565 5566out:5567	kfree(devices_info);5568	return block_group;5569}5570 5571/*5572 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the5573 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system5574 * chunks.5575 *5576 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation5577 * phases.5578 */5579int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,5580				     struct btrfs_block_group *bg)5581{5582	struct btrfs_fs_info *fs_info = trans->fs_info;5583	struct btrfs_root *chunk_root = fs_info->chunk_root;5584	struct btrfs_key key;5585	struct btrfs_chunk *chunk;5586	struct btrfs_stripe *stripe;5587	struct btrfs_chunk_map *map;5588	size_t item_size;5589	int i;5590	int ret;5591 5592	/*5593	 * We take the chunk_mutex for 2 reasons:5594	 *5595	 * 1) Updates and insertions in the chunk btree must be done while holding5596	 *    the chunk_mutex, as well as updating the system chunk array in the5597	 *    superblock. See the comment on top of btrfs_chunk_alloc() for the5598	 *    details;5599	 *5600	 * 2) To prevent races with the final phase of a device replace operation5601	 *    that replaces the device object associated with the map's stripes,5602	 *    because the device object's id can change at any time during that5603	 *    final phase of the device replace operation5604	 *    (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the5605	 *    replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,5606	 *    which would cause a failure when updating the device item, which does5607	 *    not exists, or persisting a stripe of the chunk item with such ID.5608	 *    Here we can't use the device_list_mutex because our caller already5609	 *    has locked the chunk_mutex, and the final phase of device replace5610	 *    acquires both mutexes - first the device_list_mutex and then the5611	 *    chunk_mutex. Using any of those two mutexes protects us from a5612	 *    concurrent device replace.5613	 */5614	lockdep_assert_held(&fs_info->chunk_mutex);5615 5616	map = btrfs_get_chunk_map(fs_info, bg->start, bg->length);5617	if (IS_ERR(map)) {5618		ret = PTR_ERR(map);5619		btrfs_abort_transaction(trans, ret);5620		return ret;5621	}5622 5623	item_size = btrfs_chunk_item_size(map->num_stripes);5624 5625	chunk = kzalloc(item_size, GFP_NOFS);5626	if (!chunk) {5627		ret = -ENOMEM;5628		btrfs_abort_transaction(trans, ret);5629		goto out;5630	}5631 5632	for (i = 0; i < map->num_stripes; i++) {5633		struct btrfs_device *device = map->stripes[i].dev;5634 5635		ret = btrfs_update_device(trans, device);5636		if (ret)5637			goto out;5638	}5639 5640	stripe = &chunk->stripe;5641	for (i = 0; i < map->num_stripes; i++) {5642		struct btrfs_device *device = map->stripes[i].dev;5643		const u64 dev_offset = map->stripes[i].physical;5644 5645		btrfs_set_stack_stripe_devid(stripe, device->devid);5646		btrfs_set_stack_stripe_offset(stripe, dev_offset);5647		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);5648		stripe++;5649	}5650 5651	btrfs_set_stack_chunk_length(chunk, bg->length);5652	btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);5653	btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);5654	btrfs_set_stack_chunk_type(chunk, map->type);5655	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);5656	btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);5657	btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);5658	btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);5659	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);5660 5661	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;5662	key.type = BTRFS_CHUNK_ITEM_KEY;5663	key.offset = bg->start;5664 5665	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);5666	if (ret)5667		goto out;5668 5669	set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);5670 5671	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {5672		ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);5673		if (ret)5674			goto out;5675	}5676 5677out:5678	kfree(chunk);5679	btrfs_free_chunk_map(map);5680	return ret;5681}5682 5683static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)5684{5685	struct btrfs_fs_info *fs_info = trans->fs_info;5686	u64 alloc_profile;5687	struct btrfs_block_group *meta_bg;5688	struct btrfs_block_group *sys_bg;5689 5690	/*5691	 * When adding a new device for sprouting, the seed device is read-only5692	 * so we must first allocate a metadata and a system chunk. But before5693	 * adding the block group items to the extent, device and chunk btrees,5694	 * we must first:5695	 *5696	 * 1) Create both chunks without doing any changes to the btrees, as5697	 *    otherwise we would get -ENOSPC since the block groups from the5698	 *    seed device are read-only;5699	 *5700	 * 2) Add the device item for the new sprout device - finishing the setup5701	 *    of a new block group requires updating the device item in the chunk5702	 *    btree, so it must exist when we attempt to do it. The previous step5703	 *    ensures this does not fail with -ENOSPC.5704	 *5705	 * After that we can add the block group items to their btrees:5706	 * update existing device item in the chunk btree, add a new block group5707	 * item to the extent btree, add a new chunk item to the chunk btree and5708	 * finally add the new device extent items to the devices btree.5709	 */5710 5711	alloc_profile = btrfs_metadata_alloc_profile(fs_info);5712	meta_bg = btrfs_create_chunk(trans, alloc_profile);5713	if (IS_ERR(meta_bg))5714		return PTR_ERR(meta_bg);5715 5716	alloc_profile = btrfs_system_alloc_profile(fs_info);5717	sys_bg = btrfs_create_chunk(trans, alloc_profile);5718	if (IS_ERR(sys_bg))5719		return PTR_ERR(sys_bg);5720 5721	return 0;5722}5723 5724static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map)5725{5726	const int index = btrfs_bg_flags_to_raid_index(map->type);5727 5728	return btrfs_raid_array[index].tolerated_failures;5729}5730 5731bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)5732{5733	struct btrfs_chunk_map *map;5734	int miss_ndevs = 0;5735	int i;5736	bool ret = true;5737 5738	map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);5739	if (IS_ERR(map))5740		return false;5741 5742	for (i = 0; i < map->num_stripes; i++) {5743		if (test_bit(BTRFS_DEV_STATE_MISSING,5744					&map->stripes[i].dev->dev_state)) {5745			miss_ndevs++;5746			continue;5747		}5748		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,5749					&map->stripes[i].dev->dev_state)) {5750			ret = false;5751			goto end;5752		}5753	}5754 5755	/*5756	 * If the number of missing devices is larger than max errors, we can5757	 * not write the data into that chunk successfully.5758	 */5759	if (miss_ndevs > btrfs_chunk_max_errors(map))5760		ret = false;5761end:5762	btrfs_free_chunk_map(map);5763	return ret;5764}5765 5766void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info)5767{5768	write_lock(&fs_info->mapping_tree_lock);5769	while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) {5770		struct btrfs_chunk_map *map;5771		struct rb_node *node;5772 5773		node = rb_first_cached(&fs_info->mapping_tree);5774		map = rb_entry(node, struct btrfs_chunk_map, rb_node);5775		rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);5776		RB_CLEAR_NODE(&map->rb_node);5777		chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);5778		/* Once for the tree ref. */5779		btrfs_free_chunk_map(map);5780		cond_resched_rwlock_write(&fs_info->mapping_tree_lock);5781	}5782	write_unlock(&fs_info->mapping_tree_lock);5783}5784 5785static int btrfs_chunk_map_num_copies(const struct btrfs_chunk_map *map)5786{5787	enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(map->type);5788 5789	if (map->type & BTRFS_BLOCK_GROUP_RAID5)5790		return 2;5791 5792	/*5793	 * There could be two corrupted data stripes, we need to loop retry in5794	 * order to rebuild the correct data.5795	 *5796	 * Fail a stripe at a time on every retry except the stripe under5797	 * reconstruction.5798	 */5799	if (map->type & BTRFS_BLOCK_GROUP_RAID6)5800		return map->num_stripes;5801 5802	/* Non-RAID56, use their ncopies from btrfs_raid_array. */5803	return btrfs_raid_array[index].ncopies;5804}5805 5806int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)5807{5808	struct btrfs_chunk_map *map;5809	int ret;5810 5811	map = btrfs_get_chunk_map(fs_info, logical, len);5812	if (IS_ERR(map))5813		/*5814		 * We could return errors for these cases, but that could get5815		 * ugly and we'd probably do the same thing which is just not do5816		 * anything else and exit, so return 1 so the callers don't try5817		 * to use other copies.5818		 */5819		return 1;5820 5821	ret = btrfs_chunk_map_num_copies(map);5822	btrfs_free_chunk_map(map);5823	return ret;5824}5825 5826unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,5827				    u64 logical)5828{5829	struct btrfs_chunk_map *map;5830	unsigned long len = fs_info->sectorsize;5831 5832	if (!btrfs_fs_incompat(fs_info, RAID56))5833		return len;5834 5835	map = btrfs_get_chunk_map(fs_info, logical, len);5836 5837	if (!WARN_ON(IS_ERR(map))) {5838		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)5839			len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));5840		btrfs_free_chunk_map(map);5841	}5842	return len;5843}5844 5845int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)5846{5847	struct btrfs_chunk_map *map;5848	int ret = 0;5849 5850	if (!btrfs_fs_incompat(fs_info, RAID56))5851		return 0;5852 5853	map = btrfs_get_chunk_map(fs_info, logical, len);5854 5855	if (!WARN_ON(IS_ERR(map))) {5856		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)5857			ret = 1;5858		btrfs_free_chunk_map(map);5859	}5860	return ret;5861}5862 5863static int find_live_mirror(struct btrfs_fs_info *fs_info,5864			    struct btrfs_chunk_map *map, int first,5865			    int dev_replace_is_ongoing)5866{5867	const enum btrfs_read_policy policy = READ_ONCE(fs_info->fs_devices->read_policy);5868	int i;5869	int num_stripes;5870	int preferred_mirror;5871	int tolerance;5872	struct btrfs_device *srcdev;5873 5874	ASSERT((map->type &5875		 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));5876 5877	if (map->type & BTRFS_BLOCK_GROUP_RAID10)5878		num_stripes = map->sub_stripes;5879	else5880		num_stripes = map->num_stripes;5881 5882	switch (policy) {5883	default:5884		/* Shouldn't happen, just warn and use pid instead of failing */5885		btrfs_warn_rl(fs_info, "unknown read_policy type %u, reset to pid",5886			      policy);5887		WRITE_ONCE(fs_info->fs_devices->read_policy, BTRFS_READ_POLICY_PID);5888		fallthrough;5889	case BTRFS_READ_POLICY_PID:5890		preferred_mirror = first + (current->pid % num_stripes);5891		break;5892	}5893 5894	if (dev_replace_is_ongoing &&5895	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==5896	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)5897		srcdev = fs_info->dev_replace.srcdev;5898	else5899		srcdev = NULL;5900 5901	/*5902	 * try to avoid the drive that is the source drive for a5903	 * dev-replace procedure, only choose it if no other non-missing5904	 * mirror is available5905	 */5906	for (tolerance = 0; tolerance < 2; tolerance++) {5907		if (map->stripes[preferred_mirror].dev->bdev &&5908		    (tolerance || map->stripes[preferred_mirror].dev != srcdev))5909			return preferred_mirror;5910		for (i = first; i < first + num_stripes; i++) {5911			if (map->stripes[i].dev->bdev &&5912			    (tolerance || map->stripes[i].dev != srcdev))5913				return i;5914		}5915	}5916 5917	/* we couldn't find one that doesn't fail.  Just return something5918	 * and the io error handling code will clean up eventually5919	 */5920	return preferred_mirror;5921}5922 5923static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,5924						       u64 logical,5925						       u16 total_stripes)5926{5927	struct btrfs_io_context *bioc;5928 5929	bioc = kzalloc(5930		 /* The size of btrfs_io_context */5931		sizeof(struct btrfs_io_context) +5932		/* Plus the variable array for the stripes */5933		sizeof(struct btrfs_io_stripe) * (total_stripes),5934		GFP_NOFS);5935 5936	if (!bioc)5937		return NULL;5938 5939	refcount_set(&bioc->refs, 1);5940 5941	bioc->fs_info = fs_info;5942	bioc->replace_stripe_src = -1;5943	bioc->full_stripe_logical = (u64)-1;5944	bioc->logical = logical;5945 5946	return bioc;5947}5948 5949void btrfs_get_bioc(struct btrfs_io_context *bioc)5950{5951	WARN_ON(!refcount_read(&bioc->refs));5952	refcount_inc(&bioc->refs);5953}5954 5955void btrfs_put_bioc(struct btrfs_io_context *bioc)5956{5957	if (!bioc)5958		return;5959	if (refcount_dec_and_test(&bioc->refs))5960		kfree(bioc);5961}5962 5963/*5964 * Please note that, discard won't be sent to target device of device5965 * replace.5966 */5967struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,5968					       u64 logical, u64 *length_ret,5969					       u32 *num_stripes)5970{5971	struct btrfs_chunk_map *map;5972	struct btrfs_discard_stripe *stripes;5973	u64 length = *length_ret;5974	u64 offset;5975	u32 stripe_nr;5976	u32 stripe_nr_end;5977	u32 stripe_cnt;5978	u64 stripe_end_offset;5979	u64 stripe_offset;5980	u32 stripe_index;5981	u32 factor = 0;5982	u32 sub_stripes = 0;5983	u32 stripes_per_dev = 0;5984	u32 remaining_stripes = 0;5985	u32 last_stripe = 0;5986	int ret;5987	int i;5988 5989	map = btrfs_get_chunk_map(fs_info, logical, length);5990	if (IS_ERR(map))5991		return ERR_CAST(map);5992 5993	/* we don't discard raid56 yet */5994	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {5995		ret = -EOPNOTSUPP;5996		goto out_free_map;5997	}5998 5999	offset = logical - map->start;6000	length = min_t(u64, map->start + map->chunk_len - logical, length);6001	*length_ret = length;6002 6003	/*6004	 * stripe_nr counts the total number of stripes we have to stride6005	 * to get to this block6006	 */6007	stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;6008 6009	/* stripe_offset is the offset of this block in its stripe */6010	stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);6011 6012	stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>6013			BTRFS_STRIPE_LEN_SHIFT;6014	stripe_cnt = stripe_nr_end - stripe_nr;6015	stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -6016			    (offset + length);6017	/*6018	 * after this, stripe_nr is the number of stripes on this6019	 * device we have to walk to find the data, and stripe_index is6020	 * the number of our device in the stripe array6021	 */6022	*num_stripes = 1;6023	stripe_index = 0;6024	if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |6025			 BTRFS_BLOCK_GROUP_RAID10)) {6026		if (map->type & BTRFS_BLOCK_GROUP_RAID0)6027			sub_stripes = 1;6028		else6029			sub_stripes = map->sub_stripes;6030 6031		factor = map->num_stripes / sub_stripes;6032		*num_stripes = min_t(u64, map->num_stripes,6033				    sub_stripes * stripe_cnt);6034		stripe_index = stripe_nr % factor;6035		stripe_nr /= factor;6036		stripe_index *= sub_stripes;6037 6038		remaining_stripes = stripe_cnt % factor;6039		stripes_per_dev = stripe_cnt / factor;6040		last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;6041	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |6042				BTRFS_BLOCK_GROUP_DUP)) {6043		*num_stripes = map->num_stripes;6044	} else {6045		stripe_index = stripe_nr % map->num_stripes;6046		stripe_nr /= map->num_stripes;6047	}6048 6049	stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);6050	if (!stripes) {6051		ret = -ENOMEM;6052		goto out_free_map;6053	}6054 6055	for (i = 0; i < *num_stripes; i++) {6056		stripes[i].physical =6057			map->stripes[stripe_index].physical +6058			stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);6059		stripes[i].dev = map->stripes[stripe_index].dev;6060 6061		if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |6062				 BTRFS_BLOCK_GROUP_RAID10)) {6063			stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);6064 6065			if (i / sub_stripes < remaining_stripes)6066				stripes[i].length += BTRFS_STRIPE_LEN;6067 6068			/*6069			 * Special for the first stripe and6070			 * the last stripe:6071			 *6072			 * |-------|...|-------|6073			 *     |----------|6074			 *    off     end_off6075			 */6076			if (i < sub_stripes)6077				stripes[i].length -= stripe_offset;6078 6079			if (stripe_index >= last_stripe &&6080			    stripe_index <= (last_stripe +6081					     sub_stripes - 1))6082				stripes[i].length -= stripe_end_offset;6083 6084			if (i == sub_stripes - 1)6085				stripe_offset = 0;6086		} else {6087			stripes[i].length = length;6088		}6089 6090		stripe_index++;6091		if (stripe_index == map->num_stripes) {6092			stripe_index = 0;6093			stripe_nr++;6094		}6095	}6096 6097	btrfs_free_chunk_map(map);6098	return stripes;6099out_free_map:6100	btrfs_free_chunk_map(map);6101	return ERR_PTR(ret);6102}6103 6104static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)6105{6106	struct btrfs_block_group *cache;6107	bool ret;6108 6109	/* Non zoned filesystem does not use "to_copy" flag */6110	if (!btrfs_is_zoned(fs_info))6111		return false;6112 6113	cache = btrfs_lookup_block_group(fs_info, logical);6114 6115	ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);6116 6117	btrfs_put_block_group(cache);6118	return ret;6119}6120 6121static void handle_ops_on_dev_replace(struct btrfs_io_context *bioc,6122				      struct btrfs_dev_replace *dev_replace,6123				      u64 logical,6124				      struct btrfs_io_geometry *io_geom)6125{6126	u64 srcdev_devid = dev_replace->srcdev->devid;6127	/*6128	 * At this stage, num_stripes is still the real number of stripes,6129	 * excluding the duplicated stripes.6130	 */6131	int num_stripes = io_geom->num_stripes;6132	int max_errors = io_geom->max_errors;6133	int nr_extra_stripes = 0;6134	int i;6135 6136	/*6137	 * A block group which has "to_copy" set will eventually be copied by6138	 * the dev-replace process. We can avoid cloning IO here.6139	 */6140	if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))6141		return;6142 6143	/*6144	 * Duplicate the write operations while the dev-replace procedure is6145	 * running. Since the copying of the old disk to the new disk takes6146	 * place at run time while the filesystem is mounted writable, the6147	 * regular write operations to the old disk have to be duplicated to go6148	 * to the new disk as well.6149	 *6150	 * Note that device->missing is handled by the caller, and that the6151	 * write to the old disk is already set up in the stripes array.6152	 */6153	for (i = 0; i < num_stripes; i++) {6154		struct btrfs_io_stripe *old = &bioc->stripes[i];6155		struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];6156 6157		if (old->dev->devid != srcdev_devid)6158			continue;6159 6160		new->physical = old->physical;6161		new->dev = dev_replace->tgtdev;6162		if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)6163			bioc->replace_stripe_src = i;6164		nr_extra_stripes++;6165	}6166 6167	/* We can only have at most 2 extra nr_stripes (for DUP). */6168	ASSERT(nr_extra_stripes <= 2);6169	/*6170	 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for6171	 * replace.6172	 * If we have 2 extra stripes, only choose the one with smaller physical.6173	 */6174	if (io_geom->op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {6175		struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];6176		struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];6177 6178		/* Only DUP can have two extra stripes. */6179		ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);6180 6181		/*6182		 * Swap the last stripe stripes and reduce @nr_extra_stripes.6183		 * The extra stripe would still be there, but won't be accessed.6184		 */6185		if (first->physical > second->physical) {6186			swap(second->physical, first->physical);6187			swap(second->dev, first->dev);6188			nr_extra_stripes--;6189		}6190	}6191 6192	io_geom->num_stripes = num_stripes + nr_extra_stripes;6193	io_geom->max_errors = max_errors + nr_extra_stripes;6194	bioc->replace_nr_stripes = nr_extra_stripes;6195}6196 6197static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, u64 offset,6198			    struct btrfs_io_geometry *io_geom)6199{6200	/*6201	 * Stripe_nr is the stripe where this block falls.  stripe_offset is6202	 * the offset of this block in its stripe.6203	 */6204	io_geom->stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;6205	io_geom->stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;6206	ASSERT(io_geom->stripe_offset < U32_MAX);6207 6208	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {6209		unsigned long full_stripe_len =6210			btrfs_stripe_nr_to_offset(nr_data_stripes(map));6211 6212		/*6213		 * For full stripe start, we use previously calculated6214		 * @stripe_nr. Align it to nr_data_stripes, then multiply with6215		 * STRIPE_LEN.6216		 *6217		 * By this we can avoid u64 division completely.  And we have6218		 * to go rounddown(), not round_down(), as nr_data_stripes is6219		 * not ensured to be power of 2.6220		 */6221		io_geom->raid56_full_stripe_start = btrfs_stripe_nr_to_offset(6222			rounddown(io_geom->stripe_nr, nr_data_stripes(map)));6223 6224		ASSERT(io_geom->raid56_full_stripe_start + full_stripe_len > offset);6225		ASSERT(io_geom->raid56_full_stripe_start <= offset);6226		/*6227		 * For writes to RAID56, allow to write a full stripe set, but6228		 * no straddling of stripe sets.6229		 */6230		if (io_geom->op == BTRFS_MAP_WRITE)6231			return full_stripe_len - (offset - io_geom->raid56_full_stripe_start);6232	}6233 6234	/*6235	 * For other RAID types and for RAID56 reads, allow a single stripe (on6236	 * a single disk).6237	 */6238	if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)6239		return BTRFS_STRIPE_LEN - io_geom->stripe_offset;6240	return U64_MAX;6241}6242 6243static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical,6244			 u64 *length, struct btrfs_io_stripe *dst,6245			 struct btrfs_chunk_map *map,6246			 struct btrfs_io_geometry *io_geom)6247{6248	dst->dev = map->stripes[io_geom->stripe_index].dev;6249 6250	if (io_geom->op == BTRFS_MAP_READ &&6251	    btrfs_need_stripe_tree_update(fs_info, map->type))6252		return btrfs_get_raid_extent_offset(fs_info, logical, length,6253						    map->type,6254						    io_geom->stripe_index, dst);6255 6256	dst->physical = map->stripes[io_geom->stripe_index].physical +6257			io_geom->stripe_offset +6258			btrfs_stripe_nr_to_offset(io_geom->stripe_nr);6259	return 0;6260}6261 6262static bool is_single_device_io(struct btrfs_fs_info *fs_info,6263				const struct btrfs_io_stripe *smap,6264				const struct btrfs_chunk_map *map,6265				int num_alloc_stripes,6266				enum btrfs_map_op op, int mirror_num)6267{6268	if (!smap)6269		return false;6270 6271	if (num_alloc_stripes != 1)6272		return false;6273 6274	if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ)6275		return false;6276 6277	if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)6278		return false;6279 6280	return true;6281}6282 6283static void map_blocks_raid0(const struct btrfs_chunk_map *map,6284			     struct btrfs_io_geometry *io_geom)6285{6286	io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;6287	io_geom->stripe_nr /= map->num_stripes;6288	if (io_geom->op == BTRFS_MAP_READ)6289		io_geom->mirror_num = 1;6290}6291 6292static void map_blocks_raid1(struct btrfs_fs_info *fs_info,6293			     struct btrfs_chunk_map *map,6294			     struct btrfs_io_geometry *io_geom,6295			     bool dev_replace_is_ongoing)6296{6297	if (io_geom->op != BTRFS_MAP_READ) {6298		io_geom->num_stripes = map->num_stripes;6299		return;6300	}6301 6302	if (io_geom->mirror_num) {6303		io_geom->stripe_index = io_geom->mirror_num - 1;6304		return;6305	}6306 6307	io_geom->stripe_index = find_live_mirror(fs_info, map, 0,6308						 dev_replace_is_ongoing);6309	io_geom->mirror_num = io_geom->stripe_index + 1;6310}6311 6312static void map_blocks_dup(const struct btrfs_chunk_map *map,6313			   struct btrfs_io_geometry *io_geom)6314{6315	if (io_geom->op != BTRFS_MAP_READ) {6316		io_geom->num_stripes = map->num_stripes;6317		return;6318	}6319 6320	if (io_geom->mirror_num) {6321		io_geom->stripe_index = io_geom->mirror_num - 1;6322		return;6323	}6324 6325	io_geom->mirror_num = 1;6326}6327 6328static void map_blocks_raid10(struct btrfs_fs_info *fs_info,6329			      struct btrfs_chunk_map *map,6330			      struct btrfs_io_geometry *io_geom,6331			      bool dev_replace_is_ongoing)6332{6333	u32 factor = map->num_stripes / map->sub_stripes;6334	int old_stripe_index;6335 6336	io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes;6337	io_geom->stripe_nr /= factor;6338 6339	if (io_geom->op != BTRFS_MAP_READ) {6340		io_geom->num_stripes = map->sub_stripes;6341		return;6342	}6343 6344	if (io_geom->mirror_num) {6345		io_geom->stripe_index += io_geom->mirror_num - 1;6346		return;6347	}6348 6349	old_stripe_index = io_geom->stripe_index;6350	io_geom->stripe_index = find_live_mirror(fs_info, map,6351						 io_geom->stripe_index,6352						 dev_replace_is_ongoing);6353	io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1;6354}6355 6356static void map_blocks_raid56_write(struct btrfs_chunk_map *map,6357				    struct btrfs_io_geometry *io_geom,6358				    u64 logical, u64 *length)6359{6360	int data_stripes = nr_data_stripes(map);6361 6362	/*6363	 * Needs full stripe mapping.6364	 *6365	 * Push stripe_nr back to the start of the full stripe For those cases6366	 * needing a full stripe, @stripe_nr is the full stripe number.6367	 *6368	 * Originally we go raid56_full_stripe_start / full_stripe_len, but6369	 * that can be expensive.  Here we just divide @stripe_nr with6370	 * @data_stripes.6371	 */6372	io_geom->stripe_nr /= data_stripes;6373 6374	/* RAID[56] write or recovery. Return all stripes */6375	io_geom->num_stripes = map->num_stripes;6376	io_geom->max_errors = btrfs_chunk_max_errors(map);6377 6378	/* Return the length to the full stripe end. */6379	*length = min(logical + *length,6380		      io_geom->raid56_full_stripe_start + map->start +6381		      btrfs_stripe_nr_to_offset(data_stripes)) -6382		logical;6383	io_geom->stripe_index = 0;6384	io_geom->stripe_offset = 0;6385}6386 6387static void map_blocks_raid56_read(struct btrfs_chunk_map *map,6388				   struct btrfs_io_geometry *io_geom)6389{6390	int data_stripes = nr_data_stripes(map);6391 6392	ASSERT(io_geom->mirror_num <= 1);6393	/* Just grab the data stripe directly. */6394	io_geom->stripe_index = io_geom->stripe_nr % data_stripes;6395	io_geom->stripe_nr /= data_stripes;6396 6397	/* We distribute the parity blocks across stripes. */6398	io_geom->stripe_index =6399		(io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes;6400 6401	if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1)6402		io_geom->mirror_num = 1;6403}6404 6405static void map_blocks_single(const struct btrfs_chunk_map *map,6406			      struct btrfs_io_geometry *io_geom)6407{6408	io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;6409	io_geom->stripe_nr /= map->num_stripes;6410	io_geom->mirror_num = io_geom->stripe_index + 1;6411}6412 6413/*6414 * Map one logical range to one or more physical ranges.6415 *6416 * @length:		(Mandatory) mapped length of this run.6417 *			One logical range can be split into different segments6418 *			due to factors like zones and RAID0/5/6/10 stripe6419 *			boundaries.6420 *6421 * @bioc_ret:		(Mandatory) returned btrfs_io_context structure.6422 *			which has one or more physical ranges (btrfs_io_stripe)6423 *			recorded inside.6424 *			Caller should call btrfs_put_bioc() to free it after use.6425 *6426 * @smap:		(Optional) single physical range optimization.6427 *			If the map request can be fulfilled by one single6428 *			physical range, and this is parameter is not NULL,6429 *			then @bioc_ret would be NULL, and @smap would be6430 *			updated.6431 *6432 * @mirror_num_ret:	(Mandatory) returned mirror number if the original6433 *			value is 0.6434 *6435 *			Mirror number 0 means to choose any live mirrors.6436 *6437 *			For non-RAID56 profiles, non-zero mirror_num means6438 *			the Nth mirror. (e.g. mirror_num 1 means the first6439 *			copy).6440 *6441 *			For RAID56 profile, mirror 1 means rebuild from P and6442 *			the remaining data stripes.6443 *6444 *			For RAID6 profile, mirror > 2 means mark another6445 *			data/P stripe error and rebuild from the remaining6446 *			stripes..6447 */6448int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,6449		    u64 logical, u64 *length,6450		    struct btrfs_io_context **bioc_ret,6451		    struct btrfs_io_stripe *smap, int *mirror_num_ret)6452{6453	struct btrfs_chunk_map *map;6454	struct btrfs_io_geometry io_geom = { 0 };6455	u64 map_offset;6456	int ret = 0;6457	int num_copies;6458	struct btrfs_io_context *bioc = NULL;6459	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;6460	int dev_replace_is_ongoing = 0;6461	u16 num_alloc_stripes;6462	u64 max_len;6463 6464	ASSERT(bioc_ret);6465 6466	io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);6467	io_geom.num_stripes = 1;6468	io_geom.stripe_index = 0;6469	io_geom.op = op;6470 6471	map = btrfs_get_chunk_map(fs_info, logical, *length);6472	if (IS_ERR(map))6473		return PTR_ERR(map);6474 6475	num_copies = btrfs_chunk_map_num_copies(map);6476	if (io_geom.mirror_num > num_copies)6477		return -EINVAL;6478 6479	map_offset = logical - map->start;6480	io_geom.raid56_full_stripe_start = (u64)-1;6481	max_len = btrfs_max_io_len(map, map_offset, &io_geom);6482	*length = min_t(u64, map->chunk_len - map_offset, max_len);6483 6484	down_read(&dev_replace->rwsem);6485	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);6486	/*6487	 * Hold the semaphore for read during the whole operation, write is6488	 * requested at commit time but must wait.6489	 */6490	if (!dev_replace_is_ongoing)6491		up_read(&dev_replace->rwsem);6492 6493	switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {6494	case BTRFS_BLOCK_GROUP_RAID0:6495		map_blocks_raid0(map, &io_geom);6496		break;6497	case BTRFS_BLOCK_GROUP_RAID1:6498	case BTRFS_BLOCK_GROUP_RAID1C3:6499	case BTRFS_BLOCK_GROUP_RAID1C4:6500		map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing);6501		break;6502	case BTRFS_BLOCK_GROUP_DUP:6503		map_blocks_dup(map, &io_geom);6504		break;6505	case BTRFS_BLOCK_GROUP_RAID10:6506		map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing);6507		break;6508	case BTRFS_BLOCK_GROUP_RAID5:6509	case BTRFS_BLOCK_GROUP_RAID6:6510		if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)6511			map_blocks_raid56_write(map, &io_geom, logical, length);6512		else6513			map_blocks_raid56_read(map, &io_geom);6514		break;6515	default:6516		/*6517		 * After this, stripe_nr is the number of stripes on this6518		 * device we have to walk to find the data, and stripe_index is6519		 * the number of our device in the stripe array6520		 */6521		map_blocks_single(map, &io_geom);6522		break;6523	}6524	if (io_geom.stripe_index >= map->num_stripes) {6525		btrfs_crit(fs_info,6526			   "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",6527			   io_geom.stripe_index, map->num_stripes);6528		ret = -EINVAL;6529		goto out;6530	}6531 6532	num_alloc_stripes = io_geom.num_stripes;6533	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&6534	    op != BTRFS_MAP_READ)6535		/*6536		 * For replace case, we need to add extra stripes for extra6537		 * duplicated stripes.6538		 *6539		 * For both WRITE and GET_READ_MIRRORS, we may have at most6540		 * 2 more stripes (DUP types, otherwise 1).6541		 */6542		num_alloc_stripes += 2;6543 6544	/*6545	 * If this I/O maps to a single device, try to return the device and6546	 * physical block information on the stack instead of allocating an6547	 * I/O context structure.6548	 */6549	if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op,6550				io_geom.mirror_num)) {6551		ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom);6552		if (mirror_num_ret)6553			*mirror_num_ret = io_geom.mirror_num;6554		*bioc_ret = NULL;6555		goto out;6556	}6557 6558	bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes);6559	if (!bioc) {6560		ret = -ENOMEM;6561		goto out;6562	}6563	bioc->map_type = map->type;6564 6565	/*6566	 * For RAID56 full map, we need to make sure the stripes[] follows the6567	 * rule that data stripes are all ordered, then followed with P and Q6568	 * (if we have).6569	 *6570	 * It's still mostly the same as other profiles, just with extra rotation.6571	 */6572	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&6573	    (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) {6574		/*6575		 * For RAID56 @stripe_nr is already the number of full stripes6576		 * before us, which is also the rotation value (needs to modulo6577		 * with num_stripes).6578		 *6579		 * In this case, we just add @stripe_nr with @i, then do the6580		 * modulo, to reduce one modulo call.6581		 */6582		bioc->full_stripe_logical = map->start +6583			btrfs_stripe_nr_to_offset(io_geom.stripe_nr *6584						  nr_data_stripes(map));6585		for (int i = 0; i < io_geom.num_stripes; i++) {6586			struct btrfs_io_stripe *dst = &bioc->stripes[i];6587			u32 stripe_index;6588 6589			stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes;6590			dst->dev = map->stripes[stripe_index].dev;6591			dst->physical =6592				map->stripes[stripe_index].physical +6593				io_geom.stripe_offset +6594				btrfs_stripe_nr_to_offset(io_geom.stripe_nr);6595		}6596	} else {6597		/*6598		 * For all other non-RAID56 profiles, just copy the target6599		 * stripe into the bioc.6600		 */6601		for (int i = 0; i < io_geom.num_stripes; i++) {6602			ret = set_io_stripe(fs_info, logical, length,6603					    &bioc->stripes[i], map, &io_geom);6604			if (ret < 0)6605				break;6606			io_geom.stripe_index++;6607		}6608	}6609 6610	if (ret) {6611		*bioc_ret = NULL;6612		btrfs_put_bioc(bioc);6613		goto out;6614	}6615 6616	if (op != BTRFS_MAP_READ)6617		io_geom.max_errors = btrfs_chunk_max_errors(map);6618 6619	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&6620	    op != BTRFS_MAP_READ) {6621		handle_ops_on_dev_replace(bioc, dev_replace, logical, &io_geom);6622	}6623 6624	*bioc_ret = bioc;6625	bioc->num_stripes = io_geom.num_stripes;6626	bioc->max_errors = io_geom.max_errors;6627	bioc->mirror_num = io_geom.mirror_num;6628 6629out:6630	if (dev_replace_is_ongoing) {6631		lockdep_assert_held(&dev_replace->rwsem);6632		/* Unlock and let waiting writers proceed */6633		up_read(&dev_replace->rwsem);6634	}6635	btrfs_free_chunk_map(map);6636	return ret;6637}6638 6639static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,6640				      const struct btrfs_fs_devices *fs_devices)6641{6642	if (args->fsid == NULL)6643		return true;6644	if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)6645		return true;6646	return false;6647}6648 6649static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,6650				  const struct btrfs_device *device)6651{6652	if (args->missing) {6653		if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&6654		    !device->bdev)6655			return true;6656		return false;6657	}6658 6659	if (device->devid != args->devid)6660		return false;6661	if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)6662		return false;6663	return true;6664}6665 6666/*6667 * Find a device specified by @devid or @uuid in the list of @fs_devices, or6668 * return NULL.6669 *6670 * If devid and uuid are both specified, the match must be exact, otherwise6671 * only devid is used.6672 */6673struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,6674				       const struct btrfs_dev_lookup_args *args)6675{6676	struct btrfs_device *device;6677	struct btrfs_fs_devices *seed_devs;6678 6679	if (dev_args_match_fs_devices(args, fs_devices)) {6680		list_for_each_entry(device, &fs_devices->devices, dev_list) {6681			if (dev_args_match_device(args, device))6682				return device;6683		}6684	}6685 6686	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {6687		if (!dev_args_match_fs_devices(args, seed_devs))6688			continue;6689		list_for_each_entry(device, &seed_devs->devices, dev_list) {6690			if (dev_args_match_device(args, device))6691				return device;6692		}6693	}6694 6695	return NULL;6696}6697 6698static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,6699					    u64 devid, u8 *dev_uuid)6700{6701	struct btrfs_device *device;6702	unsigned int nofs_flag;6703 6704	/*6705	 * We call this under the chunk_mutex, so we want to use NOFS for this6706	 * allocation, however we don't want to change btrfs_alloc_device() to6707	 * always do NOFS because we use it in a lot of other GFP_KERNEL safe6708	 * places.6709	 */6710 6711	nofs_flag = memalloc_nofs_save();6712	device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);6713	memalloc_nofs_restore(nofs_flag);6714	if (IS_ERR(device))6715		return device;6716 6717	list_add(&device->dev_list, &fs_devices->devices);6718	device->fs_devices = fs_devices;6719	fs_devices->num_devices++;6720 6721	set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);6722	fs_devices->missing_devices++;6723 6724	return device;6725}6726 6727/*6728 * Allocate new device struct, set up devid and UUID.6729 *6730 * @fs_info:	used only for generating a new devid, can be NULL if6731 *		devid is provided (i.e. @devid != NULL).6732 * @devid:	a pointer to devid for this device.  If NULL a new devid6733 *		is generated.6734 * @uuid:	a pointer to UUID for this device.  If NULL a new UUID6735 *		is generated.6736 * @path:	a pointer to device path if available, NULL otherwise.6737 *6738 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()6739 * on error.  Returned struct is not linked onto any lists and must be6740 * destroyed with btrfs_free_device.6741 */6742struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,6743					const u64 *devid, const u8 *uuid,6744					const char *path)6745{6746	struct btrfs_device *dev;6747	u64 tmp;6748 6749	if (WARN_ON(!devid && !fs_info))6750		return ERR_PTR(-EINVAL);6751 6752	dev = kzalloc(sizeof(*dev), GFP_KERNEL);6753	if (!dev)6754		return ERR_PTR(-ENOMEM);6755 6756	INIT_LIST_HEAD(&dev->dev_list);6757	INIT_LIST_HEAD(&dev->dev_alloc_list);6758	INIT_LIST_HEAD(&dev->post_commit_list);6759 6760	atomic_set(&dev->dev_stats_ccnt, 0);6761	btrfs_device_data_ordered_init(dev);6762	extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);6763 6764	if (devid)6765		tmp = *devid;6766	else {6767		int ret;6768 6769		ret = find_next_devid(fs_info, &tmp);6770		if (ret) {6771			btrfs_free_device(dev);6772			return ERR_PTR(ret);6773		}6774	}6775	dev->devid = tmp;6776 6777	if (uuid)6778		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);6779	else6780		generate_random_uuid(dev->uuid);6781 6782	if (path) {6783		struct rcu_string *name;6784 6785		name = rcu_string_strdup(path, GFP_KERNEL);6786		if (!name) {6787			btrfs_free_device(dev);6788			return ERR_PTR(-ENOMEM);6789		}6790		rcu_assign_pointer(dev->name, name);6791	}6792 6793	return dev;6794}6795 6796static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,6797					u64 devid, u8 *uuid, bool error)6798{6799	if (error)6800		btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",6801			      devid, uuid);6802	else6803		btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",6804			      devid, uuid);6805}6806 6807u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map)6808{6809	const int data_stripes = calc_data_stripes(map->type, map->num_stripes);6810 6811	return div_u64(map->chunk_len, data_stripes);6812}6813 6814#if BITS_PER_LONG == 326815/*6816 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE6817 * can't be accessed on 32bit systems.6818 *6819 * This function do mount time check to reject the fs if it already has6820 * metadata chunk beyond that limit.6821 */6822static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,6823				  u64 logical, u64 length, u64 type)6824{6825	if (!(type & BTRFS_BLOCK_GROUP_METADATA))6826		return 0;6827 6828	if (logical + length < MAX_LFS_FILESIZE)6829		return 0;6830 6831	btrfs_err_32bit_limit(fs_info);6832	return -EOVERFLOW;6833}6834 6835/*6836 * This is to give early warning for any metadata chunk reaching6837 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.6838 * Although we can still access the metadata, it's not going to be possible6839 * once the limit is reached.6840 */6841static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,6842				  u64 logical, u64 length, u64 type)6843{6844	if (!(type & BTRFS_BLOCK_GROUP_METADATA))6845		return;6846 6847	if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)6848		return;6849 6850	btrfs_warn_32bit_limit(fs_info);6851}6852#endif6853 6854static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,6855						  u64 devid, u8 *uuid)6856{6857	struct btrfs_device *dev;6858 6859	if (!btrfs_test_opt(fs_info, DEGRADED)) {6860		btrfs_report_missing_device(fs_info, devid, uuid, true);6861		return ERR_PTR(-ENOENT);6862	}6863 6864	dev = add_missing_dev(fs_info->fs_devices, devid, uuid);6865	if (IS_ERR(dev)) {6866		btrfs_err(fs_info, "failed to init missing device %llu: %ld",6867			  devid, PTR_ERR(dev));6868		return dev;6869	}6870	btrfs_report_missing_device(fs_info, devid, uuid, false);6871 6872	return dev;6873}6874 6875static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,6876			  struct btrfs_chunk *chunk)6877{6878	BTRFS_DEV_LOOKUP_ARGS(args);6879	struct btrfs_fs_info *fs_info = leaf->fs_info;6880	struct btrfs_chunk_map *map;6881	u64 logical;6882	u64 length;6883	u64 devid;6884	u64 type;6885	u8 uuid[BTRFS_UUID_SIZE];6886	int index;6887	int num_stripes;6888	int ret;6889	int i;6890 6891	logical = key->offset;6892	length = btrfs_chunk_length(leaf, chunk);6893	type = btrfs_chunk_type(leaf, chunk);6894	index = btrfs_bg_flags_to_raid_index(type);6895	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);6896 6897#if BITS_PER_LONG == 326898	ret = check_32bit_meta_chunk(fs_info, logical, length, type);6899	if (ret < 0)6900		return ret;6901	warn_32bit_meta_chunk(fs_info, logical, length, type);6902#endif6903 6904	/*6905	 * Only need to verify chunk item if we're reading from sys chunk array,6906	 * as chunk item in tree block is already verified by tree-checker.6907	 */6908	if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {6909		ret = btrfs_check_chunk_valid(leaf, chunk, logical);6910		if (ret)6911			return ret;6912	}6913 6914	map = btrfs_find_chunk_map(fs_info, logical, 1);6915 6916	/* already mapped? */6917	if (map && map->start <= logical && map->start + map->chunk_len > logical) {6918		btrfs_free_chunk_map(map);6919		return 0;6920	} else if (map) {6921		btrfs_free_chunk_map(map);6922	}6923 6924	map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS);6925	if (!map)6926		return -ENOMEM;6927 6928	map->start = logical;6929	map->chunk_len = length;6930	map->num_stripes = num_stripes;6931	map->io_width = btrfs_chunk_io_width(leaf, chunk);6932	map->io_align = btrfs_chunk_io_align(leaf, chunk);6933	map->type = type;6934	/*6935	 * We can't use the sub_stripes value, as for profiles other than6936	 * RAID10, they may have 0 as sub_stripes for filesystems created by6937	 * older mkfs (<v5.4).6938	 * In that case, it can cause divide-by-zero errors later.6939	 * Since currently sub_stripes is fixed for each profile, let's6940	 * use the trusted value instead.6941	 */6942	map->sub_stripes = btrfs_raid_array[index].sub_stripes;6943	map->verified_stripes = 0;6944	map->stripe_size = btrfs_calc_stripe_length(map);6945	for (i = 0; i < num_stripes; i++) {6946		map->stripes[i].physical =6947			btrfs_stripe_offset_nr(leaf, chunk, i);6948		devid = btrfs_stripe_devid_nr(leaf, chunk, i);6949		args.devid = devid;6950		read_extent_buffer(leaf, uuid, (unsigned long)6951				   btrfs_stripe_dev_uuid_nr(chunk, i),6952				   BTRFS_UUID_SIZE);6953		args.uuid = uuid;6954		map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);6955		if (!map->stripes[i].dev) {6956			map->stripes[i].dev = handle_missing_device(fs_info,6957								    devid, uuid);6958			if (IS_ERR(map->stripes[i].dev)) {6959				ret = PTR_ERR(map->stripes[i].dev);6960				btrfs_free_chunk_map(map);6961				return ret;6962			}6963		}6964 6965		set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,6966				&(map->stripes[i].dev->dev_state));6967	}6968 6969	ret = btrfs_add_chunk_map(fs_info, map);6970	if (ret < 0) {6971		btrfs_err(fs_info,6972			  "failed to add chunk map, start=%llu len=%llu: %d",6973			  map->start, map->chunk_len, ret);6974	}6975 6976	return ret;6977}6978 6979static void fill_device_from_item(struct extent_buffer *leaf,6980				 struct btrfs_dev_item *dev_item,6981				 struct btrfs_device *device)6982{6983	unsigned long ptr;6984 6985	device->devid = btrfs_device_id(leaf, dev_item);6986	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);6987	device->total_bytes = device->disk_total_bytes;6988	device->commit_total_bytes = device->disk_total_bytes;6989	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);6990	device->commit_bytes_used = device->bytes_used;6991	device->type = btrfs_device_type(leaf, dev_item);6992	device->io_align = btrfs_device_io_align(leaf, dev_item);6993	device->io_width = btrfs_device_io_width(leaf, dev_item);6994	device->sector_size = btrfs_device_sector_size(leaf, dev_item);6995	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);6996	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);6997 6998	ptr = btrfs_device_uuid(dev_item);6999	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);7000}7001 7002static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,7003						  u8 *fsid)7004{7005	struct btrfs_fs_devices *fs_devices;7006	int ret;7007 7008	lockdep_assert_held(&uuid_mutex);7009	ASSERT(fsid);7010 7011	/* This will match only for multi-device seed fs */7012	list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)7013		if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))7014			return fs_devices;7015 7016 7017	fs_devices = find_fsid(fsid, NULL);7018	if (!fs_devices) {7019		if (!btrfs_test_opt(fs_info, DEGRADED))7020			return ERR_PTR(-ENOENT);7021 7022		fs_devices = alloc_fs_devices(fsid);7023		if (IS_ERR(fs_devices))7024			return fs_devices;7025 7026		fs_devices->seeding = true;7027		fs_devices->opened = 1;7028		return fs_devices;7029	}7030 7031	/*7032	 * Upon first call for a seed fs fsid, just create a private copy of the7033	 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list7034	 */7035	fs_devices = clone_fs_devices(fs_devices);7036	if (IS_ERR(fs_devices))7037		return fs_devices;7038 7039	ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);7040	if (ret) {7041		free_fs_devices(fs_devices);7042		return ERR_PTR(ret);7043	}7044 7045	if (!fs_devices->seeding) {7046		close_fs_devices(fs_devices);7047		free_fs_devices(fs_devices);7048		return ERR_PTR(-EINVAL);7049	}7050 7051	list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);7052 7053	return fs_devices;7054}7055 7056static int read_one_dev(struct extent_buffer *leaf,7057			struct btrfs_dev_item *dev_item)7058{7059	BTRFS_DEV_LOOKUP_ARGS(args);7060	struct btrfs_fs_info *fs_info = leaf->fs_info;7061	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;7062	struct btrfs_device *device;7063	u64 devid;7064	int ret;7065	u8 fs_uuid[BTRFS_FSID_SIZE];7066	u8 dev_uuid[BTRFS_UUID_SIZE];7067 7068	devid = btrfs_device_id(leaf, dev_item);7069	args.devid = devid;7070	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),7071			   BTRFS_UUID_SIZE);7072	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),7073			   BTRFS_FSID_SIZE);7074	args.uuid = dev_uuid;7075	args.fsid = fs_uuid;7076 7077	if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {7078		fs_devices = open_seed_devices(fs_info, fs_uuid);7079		if (IS_ERR(fs_devices))7080			return PTR_ERR(fs_devices);7081	}7082 7083	device = btrfs_find_device(fs_info->fs_devices, &args);7084	if (!device) {7085		if (!btrfs_test_opt(fs_info, DEGRADED)) {7086			btrfs_report_missing_device(fs_info, devid,7087							dev_uuid, true);7088			return -ENOENT;7089		}7090 7091		device = add_missing_dev(fs_devices, devid, dev_uuid);7092		if (IS_ERR(device)) {7093			btrfs_err(fs_info,7094				"failed to add missing dev %llu: %ld",7095				devid, PTR_ERR(device));7096			return PTR_ERR(device);7097		}7098		btrfs_report_missing_device(fs_info, devid, dev_uuid, false);7099	} else {7100		if (!device->bdev) {7101			if (!btrfs_test_opt(fs_info, DEGRADED)) {7102				btrfs_report_missing_device(fs_info,7103						devid, dev_uuid, true);7104				return -ENOENT;7105			}7106			btrfs_report_missing_device(fs_info, devid,7107							dev_uuid, false);7108		}7109 7110		if (!device->bdev &&7111		    !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {7112			/*7113			 * this happens when a device that was properly setup7114			 * in the device info lists suddenly goes bad.7115			 * device->bdev is NULL, and so we have to set7116			 * device->missing to one here7117			 */7118			device->fs_devices->missing_devices++;7119			set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);7120		}7121 7122		/* Move the device to its own fs_devices */7123		if (device->fs_devices != fs_devices) {7124			ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,7125							&device->dev_state));7126 7127			list_move(&device->dev_list, &fs_devices->devices);7128			device->fs_devices->num_devices--;7129			fs_devices->num_devices++;7130 7131			device->fs_devices->missing_devices--;7132			fs_devices->missing_devices++;7133 7134			device->fs_devices = fs_devices;7135		}7136	}7137 7138	if (device->fs_devices != fs_info->fs_devices) {7139		BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));7140		if (device->generation !=7141		    btrfs_device_generation(leaf, dev_item))7142			return -EINVAL;7143	}7144 7145	fill_device_from_item(leaf, dev_item, device);7146	if (device->bdev) {7147		u64 max_total_bytes = bdev_nr_bytes(device->bdev);7148 7149		if (device->total_bytes > max_total_bytes) {7150			btrfs_err(fs_info,7151			"device total_bytes should be at most %llu but found %llu",7152				  max_total_bytes, device->total_bytes);7153			return -EINVAL;7154		}7155	}7156	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);7157	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&7158	   !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {7159		device->fs_devices->total_rw_bytes += device->total_bytes;7160		atomic64_add(device->total_bytes - device->bytes_used,7161				&fs_info->free_chunk_space);7162	}7163	ret = 0;7164	return ret;7165}7166 7167int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)7168{7169	struct btrfs_super_block *super_copy = fs_info->super_copy;7170	struct extent_buffer *sb;7171	struct btrfs_disk_key *disk_key;7172	struct btrfs_chunk *chunk;7173	u8 *array_ptr;7174	unsigned long sb_array_offset;7175	int ret = 0;7176	u32 num_stripes;7177	u32 array_size;7178	u32 len = 0;7179	u32 cur_offset;7180	u64 type;7181	struct btrfs_key key;7182 7183	ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);7184 7185	/*7186	 * We allocated a dummy extent, just to use extent buffer accessors.7187	 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but7188	 * that's fine, we will not go beyond system chunk array anyway.7189	 */7190	sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);7191	if (!sb)7192		return -ENOMEM;7193	set_extent_buffer_uptodate(sb);7194 7195	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);7196	array_size = btrfs_super_sys_array_size(super_copy);7197 7198	array_ptr = super_copy->sys_chunk_array;7199	sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);7200	cur_offset = 0;7201 7202	while (cur_offset < array_size) {7203		disk_key = (struct btrfs_disk_key *)array_ptr;7204		len = sizeof(*disk_key);7205		if (cur_offset + len > array_size)7206			goto out_short_read;7207 7208		btrfs_disk_key_to_cpu(&key, disk_key);7209 7210		array_ptr += len;7211		sb_array_offset += len;7212		cur_offset += len;7213 7214		if (key.type != BTRFS_CHUNK_ITEM_KEY) {7215			btrfs_err(fs_info,7216			    "unexpected item type %u in sys_array at offset %u",7217				  (u32)key.type, cur_offset);7218			ret = -EIO;7219			break;7220		}7221 7222		chunk = (struct btrfs_chunk *)sb_array_offset;7223		/*7224		 * At least one btrfs_chunk with one stripe must be present,7225		 * exact stripe count check comes afterwards7226		 */7227		len = btrfs_chunk_item_size(1);7228		if (cur_offset + len > array_size)7229			goto out_short_read;7230 7231		num_stripes = btrfs_chunk_num_stripes(sb, chunk);7232		if (!num_stripes) {7233			btrfs_err(fs_info,7234			"invalid number of stripes %u in sys_array at offset %u",7235				  num_stripes, cur_offset);7236			ret = -EIO;7237			break;7238		}7239 7240		type = btrfs_chunk_type(sb, chunk);7241		if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {7242			btrfs_err(fs_info,7243			"invalid chunk type %llu in sys_array at offset %u",7244				  type, cur_offset);7245			ret = -EIO;7246			break;7247		}7248 7249		len = btrfs_chunk_item_size(num_stripes);7250		if (cur_offset + len > array_size)7251			goto out_short_read;7252 7253		ret = read_one_chunk(&key, sb, chunk);7254		if (ret)7255			break;7256 7257		array_ptr += len;7258		sb_array_offset += len;7259		cur_offset += len;7260	}7261	clear_extent_buffer_uptodate(sb);7262	free_extent_buffer_stale(sb);7263	return ret;7264 7265out_short_read:7266	btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",7267			len, cur_offset);7268	clear_extent_buffer_uptodate(sb);7269	free_extent_buffer_stale(sb);7270	return -EIO;7271}7272 7273/*7274 * Check if all chunks in the fs are OK for read-write degraded mount7275 *7276 * If the @failing_dev is specified, it's accounted as missing.7277 *7278 * Return true if all chunks meet the minimal RW mount requirements.7279 * Return false if any chunk doesn't meet the minimal RW mount requirements.7280 */7281bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,7282					struct btrfs_device *failing_dev)7283{7284	struct btrfs_chunk_map *map;7285	u64 next_start;7286	bool ret = true;7287 7288	map = btrfs_find_chunk_map(fs_info, 0, U64_MAX);7289	/* No chunk at all? Return false anyway */7290	if (!map) {7291		ret = false;7292		goto out;7293	}7294	while (map) {7295		int missing = 0;7296		int max_tolerated;7297		int i;7298 7299		max_tolerated =7300			btrfs_get_num_tolerated_disk_barrier_failures(7301					map->type);7302		for (i = 0; i < map->num_stripes; i++) {7303			struct btrfs_device *dev = map->stripes[i].dev;7304 7305			if (!dev || !dev->bdev ||7306			    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||7307			    dev->last_flush_error)7308				missing++;7309			else if (failing_dev && failing_dev == dev)7310				missing++;7311		}7312		if (missing > max_tolerated) {7313			if (!failing_dev)7314				btrfs_warn(fs_info,7315	"chunk %llu missing %d devices, max tolerance is %d for writable mount",7316				   map->start, missing, max_tolerated);7317			btrfs_free_chunk_map(map);7318			ret = false;7319			goto out;7320		}7321		next_start = map->start + map->chunk_len;7322		btrfs_free_chunk_map(map);7323 7324		map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start);7325	}7326out:7327	return ret;7328}7329 7330static void readahead_tree_node_children(struct extent_buffer *node)7331{7332	int i;7333	const int nr_items = btrfs_header_nritems(node);7334 7335	for (i = 0; i < nr_items; i++)7336		btrfs_readahead_node_child(node, i);7337}7338 7339int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)7340{7341	struct btrfs_root *root = fs_info->chunk_root;7342	struct btrfs_path *path;7343	struct extent_buffer *leaf;7344	struct btrfs_key key;7345	struct btrfs_key found_key;7346	int ret;7347	int slot;7348	int iter_ret = 0;7349	u64 total_dev = 0;7350	u64 last_ra_node = 0;7351 7352	path = btrfs_alloc_path();7353	if (!path)7354		return -ENOMEM;7355 7356	/*7357	 * uuid_mutex is needed only if we are mounting a sprout FS7358	 * otherwise we don't need it.7359	 */7360	mutex_lock(&uuid_mutex);7361 7362	/*7363	 * It is possible for mount and umount to race in such a way that7364	 * we execute this code path, but open_fs_devices failed to clear7365	 * total_rw_bytes. We certainly want it cleared before reading the7366	 * device items, so clear it here.7367	 */7368	fs_info->fs_devices->total_rw_bytes = 0;7369 7370	/*7371	 * Lockdep complains about possible circular locking dependency between7372	 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores7373	 * used for freeze procection of a fs (struct super_block.s_writers),7374	 * which we take when starting a transaction, and extent buffers of the7375	 * chunk tree if we call read_one_dev() while holding a lock on an7376	 * extent buffer of the chunk tree. Since we are mounting the filesystem7377	 * and at this point there can't be any concurrent task modifying the7378	 * chunk tree, to keep it simple, just skip locking on the chunk tree.7379	 */7380	ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));7381	path->skip_locking = 1;7382 7383	/*7384	 * Read all device items, and then all the chunk items. All7385	 * device items are found before any chunk item (their object id7386	 * is smaller than the lowest possible object id for a chunk7387	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).7388	 */7389	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;7390	key.offset = 0;7391	key.type = 0;7392	btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {7393		struct extent_buffer *node = path->nodes[1];7394 7395		leaf = path->nodes[0];7396		slot = path->slots[0];7397 7398		if (node) {7399			if (last_ra_node != node->start) {7400				readahead_tree_node_children(node);7401				last_ra_node = node->start;7402			}7403		}7404		if (found_key.type == BTRFS_DEV_ITEM_KEY) {7405			struct btrfs_dev_item *dev_item;7406			dev_item = btrfs_item_ptr(leaf, slot,7407						  struct btrfs_dev_item);7408			ret = read_one_dev(leaf, dev_item);7409			if (ret)7410				goto error;7411			total_dev++;7412		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {7413			struct btrfs_chunk *chunk;7414 7415			/*7416			 * We are only called at mount time, so no need to take7417			 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,7418			 * we always lock first fs_info->chunk_mutex before7419			 * acquiring any locks on the chunk tree. This is a7420			 * requirement for chunk allocation, see the comment on7421			 * top of btrfs_chunk_alloc() for details.7422			 */7423			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);7424			ret = read_one_chunk(&found_key, leaf, chunk);7425			if (ret)7426				goto error;7427		}7428	}7429	/* Catch error found during iteration */7430	if (iter_ret < 0) {7431		ret = iter_ret;7432		goto error;7433	}7434 7435	/*7436	 * After loading chunk tree, we've got all device information,7437	 * do another round of validation checks.7438	 */7439	if (total_dev != fs_info->fs_devices->total_devices) {7440		btrfs_warn(fs_info,7441"super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",7442			  btrfs_super_num_devices(fs_info->super_copy),7443			  total_dev);7444		fs_info->fs_devices->total_devices = total_dev;7445		btrfs_set_super_num_devices(fs_info->super_copy, total_dev);7446	}7447	if (btrfs_super_total_bytes(fs_info->super_copy) <7448	    fs_info->fs_devices->total_rw_bytes) {7449		btrfs_err(fs_info,7450	"super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",7451			  btrfs_super_total_bytes(fs_info->super_copy),7452			  fs_info->fs_devices->total_rw_bytes);7453		ret = -EINVAL;7454		goto error;7455	}7456	ret = 0;7457error:7458	mutex_unlock(&uuid_mutex);7459 7460	btrfs_free_path(path);7461	return ret;7462}7463 7464int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)7465{7466	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;7467	struct btrfs_device *device;7468	int ret = 0;7469 7470	fs_devices->fs_info = fs_info;7471 7472	mutex_lock(&fs_devices->device_list_mutex);7473	list_for_each_entry(device, &fs_devices->devices, dev_list)7474		device->fs_info = fs_info;7475 7476	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {7477		list_for_each_entry(device, &seed_devs->devices, dev_list) {7478			device->fs_info = fs_info;7479			ret = btrfs_get_dev_zone_info(device, false);7480			if (ret)7481				break;7482		}7483 7484		seed_devs->fs_info = fs_info;7485	}7486	mutex_unlock(&fs_devices->device_list_mutex);7487 7488	return ret;7489}7490 7491static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,7492				 const struct btrfs_dev_stats_item *ptr,7493				 int index)7494{7495	u64 val;7496 7497	read_extent_buffer(eb, &val,7498			   offsetof(struct btrfs_dev_stats_item, values) +7499			    ((unsigned long)ptr) + (index * sizeof(u64)),7500			   sizeof(val));7501	return val;7502}7503 7504static void btrfs_set_dev_stats_value(struct extent_buffer *eb,7505				      struct btrfs_dev_stats_item *ptr,7506				      int index, u64 val)7507{7508	write_extent_buffer(eb, &val,7509			    offsetof(struct btrfs_dev_stats_item, values) +7510			     ((unsigned long)ptr) + (index * sizeof(u64)),7511			    sizeof(val));7512}7513 7514static int btrfs_device_init_dev_stats(struct btrfs_device *device,7515				       struct btrfs_path *path)7516{7517	struct btrfs_dev_stats_item *ptr;7518	struct extent_buffer *eb;7519	struct btrfs_key key;7520	int item_size;7521	int i, ret, slot;7522 7523	if (!device->fs_info->dev_root)7524		return 0;7525 7526	key.objectid = BTRFS_DEV_STATS_OBJECTID;7527	key.type = BTRFS_PERSISTENT_ITEM_KEY;7528	key.offset = device->devid;7529	ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);7530	if (ret) {7531		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)7532			btrfs_dev_stat_set(device, i, 0);7533		device->dev_stats_valid = 1;7534		btrfs_release_path(path);7535		return ret < 0 ? ret : 0;7536	}7537	slot = path->slots[0];7538	eb = path->nodes[0];7539	item_size = btrfs_item_size(eb, slot);7540 7541	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);7542 7543	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {7544		if (item_size >= (1 + i) * sizeof(__le64))7545			btrfs_dev_stat_set(device, i,7546					   btrfs_dev_stats_value(eb, ptr, i));7547		else7548			btrfs_dev_stat_set(device, i, 0);7549	}7550 7551	device->dev_stats_valid = 1;7552	btrfs_dev_stat_print_on_load(device);7553	btrfs_release_path(path);7554 7555	return 0;7556}7557 7558int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)7559{7560	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;7561	struct btrfs_device *device;7562	struct btrfs_path *path = NULL;7563	int ret = 0;7564 7565	path = btrfs_alloc_path();7566	if (!path)7567		return -ENOMEM;7568 7569	mutex_lock(&fs_devices->device_list_mutex);7570	list_for_each_entry(device, &fs_devices->devices, dev_list) {7571		ret = btrfs_device_init_dev_stats(device, path);7572		if (ret)7573			goto out;7574	}7575	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {7576		list_for_each_entry(device, &seed_devs->devices, dev_list) {7577			ret = btrfs_device_init_dev_stats(device, path);7578			if (ret)7579				goto out;7580		}7581	}7582out:7583	mutex_unlock(&fs_devices->device_list_mutex);7584 7585	btrfs_free_path(path);7586	return ret;7587}7588 7589static int update_dev_stat_item(struct btrfs_trans_handle *trans,7590				struct btrfs_device *device)7591{7592	struct btrfs_fs_info *fs_info = trans->fs_info;7593	struct btrfs_root *dev_root = fs_info->dev_root;7594	struct btrfs_path *path;7595	struct btrfs_key key;7596	struct extent_buffer *eb;7597	struct btrfs_dev_stats_item *ptr;7598	int ret;7599	int i;7600 7601	key.objectid = BTRFS_DEV_STATS_OBJECTID;7602	key.type = BTRFS_PERSISTENT_ITEM_KEY;7603	key.offset = device->devid;7604 7605	path = btrfs_alloc_path();7606	if (!path)7607		return -ENOMEM;7608	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);7609	if (ret < 0) {7610		btrfs_warn_in_rcu(fs_info,7611			"error %d while searching for dev_stats item for device %s",7612				  ret, btrfs_dev_name(device));7613		goto out;7614	}7615 7616	if (ret == 0 &&7617	    btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {7618		/* need to delete old one and insert a new one */7619		ret = btrfs_del_item(trans, dev_root, path);7620		if (ret != 0) {7621			btrfs_warn_in_rcu(fs_info,7622				"delete too small dev_stats item for device %s failed %d",7623					  btrfs_dev_name(device), ret);7624			goto out;7625		}7626		ret = 1;7627	}7628 7629	if (ret == 1) {7630		/* need to insert a new item */7631		btrfs_release_path(path);7632		ret = btrfs_insert_empty_item(trans, dev_root, path,7633					      &key, sizeof(*ptr));7634		if (ret < 0) {7635			btrfs_warn_in_rcu(fs_info,7636				"insert dev_stats item for device %s failed %d",7637				btrfs_dev_name(device), ret);7638			goto out;7639		}7640	}7641 7642	eb = path->nodes[0];7643	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);7644	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)7645		btrfs_set_dev_stats_value(eb, ptr, i,7646					  btrfs_dev_stat_read(device, i));7647	btrfs_mark_buffer_dirty(trans, eb);7648 7649out:7650	btrfs_free_path(path);7651	return ret;7652}7653 7654/*7655 * called from commit_transaction. Writes all changed device stats to disk.7656 */7657int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)7658{7659	struct btrfs_fs_info *fs_info = trans->fs_info;7660	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;7661	struct btrfs_device *device;7662	int stats_cnt;7663	int ret = 0;7664 7665	mutex_lock(&fs_devices->device_list_mutex);7666	list_for_each_entry(device, &fs_devices->devices, dev_list) {7667		stats_cnt = atomic_read(&device->dev_stats_ccnt);7668		if (!device->dev_stats_valid || stats_cnt == 0)7669			continue;7670 7671 7672		/*7673		 * There is a LOAD-LOAD control dependency between the value of7674		 * dev_stats_ccnt and updating the on-disk values which requires7675		 * reading the in-memory counters. Such control dependencies7676		 * require explicit read memory barriers.7677		 *7678		 * This memory barriers pairs with smp_mb__before_atomic in7679		 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full7680		 * barrier implied by atomic_xchg in7681		 * btrfs_dev_stats_read_and_reset7682		 */7683		smp_rmb();7684 7685		ret = update_dev_stat_item(trans, device);7686		if (!ret)7687			atomic_sub(stats_cnt, &device->dev_stats_ccnt);7688	}7689	mutex_unlock(&fs_devices->device_list_mutex);7690 7691	return ret;7692}7693 7694void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)7695{7696	btrfs_dev_stat_inc(dev, index);7697 7698	if (!dev->dev_stats_valid)7699		return;7700	btrfs_err_rl_in_rcu(dev->fs_info,7701		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",7702			   btrfs_dev_name(dev),7703			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),7704			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),7705			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),7706			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),7707			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));7708}7709 7710static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)7711{7712	int i;7713 7714	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)7715		if (btrfs_dev_stat_read(dev, i) != 0)7716			break;7717	if (i == BTRFS_DEV_STAT_VALUES_MAX)7718		return; /* all values == 0, suppress message */7719 7720	btrfs_info_in_rcu(dev->fs_info,7721		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",7722	       btrfs_dev_name(dev),7723	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),7724	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),7725	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),7726	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),7727	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));7728}7729 7730int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,7731			struct btrfs_ioctl_get_dev_stats *stats)7732{7733	BTRFS_DEV_LOOKUP_ARGS(args);7734	struct btrfs_device *dev;7735	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;7736	int i;7737 7738	mutex_lock(&fs_devices->device_list_mutex);7739	args.devid = stats->devid;7740	dev = btrfs_find_device(fs_info->fs_devices, &args);7741	mutex_unlock(&fs_devices->device_list_mutex);7742 7743	if (!dev) {7744		btrfs_warn(fs_info, "get dev_stats failed, device not found");7745		return -ENODEV;7746	} else if (!dev->dev_stats_valid) {7747		btrfs_warn(fs_info, "get dev_stats failed, not yet valid");7748		return -ENODEV;7749	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {7750		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {7751			if (stats->nr_items > i)7752				stats->values[i] =7753					btrfs_dev_stat_read_and_reset(dev, i);7754			else7755				btrfs_dev_stat_set(dev, i, 0);7756		}7757		btrfs_info(fs_info, "device stats zeroed by %s (%d)",7758			   current->comm, task_pid_nr(current));7759	} else {7760		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)7761			if (stats->nr_items > i)7762				stats->values[i] = btrfs_dev_stat_read(dev, i);7763	}7764	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)7765		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;7766	return 0;7767}7768 7769/*7770 * Update the size and bytes used for each device where it changed.  This is7771 * delayed since we would otherwise get errors while writing out the7772 * superblocks.7773 *7774 * Must be invoked during transaction commit.7775 */7776void btrfs_commit_device_sizes(struct btrfs_transaction *trans)7777{7778	struct btrfs_device *curr, *next;7779 7780	ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);7781 7782	if (list_empty(&trans->dev_update_list))7783		return;7784 7785	/*7786	 * We don't need the device_list_mutex here.  This list is owned by the7787	 * transaction and the transaction must complete before the device is7788	 * released.7789	 */7790	mutex_lock(&trans->fs_info->chunk_mutex);7791	list_for_each_entry_safe(curr, next, &trans->dev_update_list,7792				 post_commit_list) {7793		list_del_init(&curr->post_commit_list);7794		curr->commit_total_bytes = curr->disk_total_bytes;7795		curr->commit_bytes_used = curr->bytes_used;7796	}7797	mutex_unlock(&trans->fs_info->chunk_mutex);7798}7799 7800/*7801 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.7802 */7803int btrfs_bg_type_to_factor(u64 flags)7804{7805	const int index = btrfs_bg_flags_to_raid_index(flags);7806 7807	return btrfs_raid_array[index].ncopies;7808}7809 7810 7811 7812static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,7813				 u64 chunk_offset, u64 devid,7814				 u64 physical_offset, u64 physical_len)7815{7816	struct btrfs_dev_lookup_args args = { .devid = devid };7817	struct btrfs_chunk_map *map;7818	struct btrfs_device *dev;7819	u64 stripe_len;7820	bool found = false;7821	int ret = 0;7822	int i;7823 7824	map = btrfs_find_chunk_map(fs_info, chunk_offset, 1);7825	if (!map) {7826		btrfs_err(fs_info,7827"dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",7828			  physical_offset, devid);7829		ret = -EUCLEAN;7830		goto out;7831	}7832 7833	stripe_len = btrfs_calc_stripe_length(map);7834	if (physical_len != stripe_len) {7835		btrfs_err(fs_info,7836"dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",7837			  physical_offset, devid, map->start, physical_len,7838			  stripe_len);7839		ret = -EUCLEAN;7840		goto out;7841	}7842 7843	/*7844	 * Very old mkfs.btrfs (before v4.1) will not respect the reserved7845	 * space. Although kernel can handle it without problem, better to warn7846	 * the users.7847	 */7848	if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)7849		btrfs_warn(fs_info,7850		"devid %llu physical %llu len %llu inside the reserved space",7851			   devid, physical_offset, physical_len);7852 7853	for (i = 0; i < map->num_stripes; i++) {7854		if (map->stripes[i].dev->devid == devid &&7855		    map->stripes[i].physical == physical_offset) {7856			found = true;7857			if (map->verified_stripes >= map->num_stripes) {7858				btrfs_err(fs_info,7859				"too many dev extents for chunk %llu found",7860					  map->start);7861				ret = -EUCLEAN;7862				goto out;7863			}7864			map->verified_stripes++;7865			break;7866		}7867	}7868	if (!found) {7869		btrfs_err(fs_info,7870	"dev extent physical offset %llu devid %llu has no corresponding chunk",7871			physical_offset, devid);7872		ret = -EUCLEAN;7873	}7874 7875	/* Make sure no dev extent is beyond device boundary */7876	dev = btrfs_find_device(fs_info->fs_devices, &args);7877	if (!dev) {7878		btrfs_err(fs_info, "failed to find devid %llu", devid);7879		ret = -EUCLEAN;7880		goto out;7881	}7882 7883	if (physical_offset + physical_len > dev->disk_total_bytes) {7884		btrfs_err(fs_info,7885"dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",7886			  devid, physical_offset, physical_len,7887			  dev->disk_total_bytes);7888		ret = -EUCLEAN;7889		goto out;7890	}7891 7892	if (dev->zone_info) {7893		u64 zone_size = dev->zone_info->zone_size;7894 7895		if (!IS_ALIGNED(physical_offset, zone_size) ||7896		    !IS_ALIGNED(physical_len, zone_size)) {7897			btrfs_err(fs_info,7898"zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",7899				  devid, physical_offset, physical_len);7900			ret = -EUCLEAN;7901			goto out;7902		}7903	}7904 7905out:7906	btrfs_free_chunk_map(map);7907	return ret;7908}7909 7910static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)7911{7912	struct rb_node *node;7913	int ret = 0;7914 7915	read_lock(&fs_info->mapping_tree_lock);7916	for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) {7917		struct btrfs_chunk_map *map;7918 7919		map = rb_entry(node, struct btrfs_chunk_map, rb_node);7920		if (map->num_stripes != map->verified_stripes) {7921			btrfs_err(fs_info,7922			"chunk %llu has missing dev extent, have %d expect %d",7923				  map->start, map->verified_stripes, map->num_stripes);7924			ret = -EUCLEAN;7925			goto out;7926		}7927	}7928out:7929	read_unlock(&fs_info->mapping_tree_lock);7930	return ret;7931}7932 7933/*7934 * Ensure that all dev extents are mapped to correct chunk, otherwise7935 * later chunk allocation/free would cause unexpected behavior.7936 *7937 * NOTE: This will iterate through the whole device tree, which should be of7938 * the same size level as the chunk tree.  This slightly increases mount time.7939 */7940int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)7941{7942	struct btrfs_path *path;7943	struct btrfs_root *root = fs_info->dev_root;7944	struct btrfs_key key;7945	u64 prev_devid = 0;7946	u64 prev_dev_ext_end = 0;7947	int ret = 0;7948 7949	/*7950	 * We don't have a dev_root because we mounted with ignorebadroots and7951	 * failed to load the root, so we want to skip the verification in this7952	 * case for sure.7953	 *7954	 * However if the dev root is fine, but the tree itself is corrupted7955	 * we'd still fail to mount.  This verification is only to make sure7956	 * writes can happen safely, so instead just bypass this check7957	 * completely in the case of IGNOREBADROOTS.7958	 */7959	if (btrfs_test_opt(fs_info, IGNOREBADROOTS))7960		return 0;7961 7962	key.objectid = 1;7963	key.type = BTRFS_DEV_EXTENT_KEY;7964	key.offset = 0;7965 7966	path = btrfs_alloc_path();7967	if (!path)7968		return -ENOMEM;7969 7970	path->reada = READA_FORWARD;7971	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);7972	if (ret < 0)7973		goto out;7974 7975	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {7976		ret = btrfs_next_leaf(root, path);7977		if (ret < 0)7978			goto out;7979		/* No dev extents at all? Not good */7980		if (ret > 0) {7981			ret = -EUCLEAN;7982			goto out;7983		}7984	}7985	while (1) {7986		struct extent_buffer *leaf = path->nodes[0];7987		struct btrfs_dev_extent *dext;7988		int slot = path->slots[0];7989		u64 chunk_offset;7990		u64 physical_offset;7991		u64 physical_len;7992		u64 devid;7993 7994		btrfs_item_key_to_cpu(leaf, &key, slot);7995		if (key.type != BTRFS_DEV_EXTENT_KEY)7996			break;7997		devid = key.objectid;7998		physical_offset = key.offset;7999 8000		dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);8001		chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);8002		physical_len = btrfs_dev_extent_length(leaf, dext);8003 8004		/* Check if this dev extent overlaps with the previous one */8005		if (devid == prev_devid && physical_offset < prev_dev_ext_end) {8006			btrfs_err(fs_info,8007"dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",8008				  devid, physical_offset, prev_dev_ext_end);8009			ret = -EUCLEAN;8010			goto out;8011		}8012 8013		ret = verify_one_dev_extent(fs_info, chunk_offset, devid,8014					    physical_offset, physical_len);8015		if (ret < 0)8016			goto out;8017		prev_devid = devid;8018		prev_dev_ext_end = physical_offset + physical_len;8019 8020		ret = btrfs_next_item(root, path);8021		if (ret < 0)8022			goto out;8023		if (ret > 0) {8024			ret = 0;8025			break;8026		}8027	}8028 8029	/* Ensure all chunks have corresponding dev extents */8030	ret = verify_chunk_dev_extent_mapping(fs_info);8031out:8032	btrfs_free_path(path);8033	return ret;8034}8035 8036/*8037 * Check whether the given block group or device is pinned by any inode being8038 * used as a swapfile.8039 */8040bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)8041{8042	struct btrfs_swapfile_pin *sp;8043	struct rb_node *node;8044 8045	spin_lock(&fs_info->swapfile_pins_lock);8046	node = fs_info->swapfile_pins.rb_node;8047	while (node) {8048		sp = rb_entry(node, struct btrfs_swapfile_pin, node);8049		if (ptr < sp->ptr)8050			node = node->rb_left;8051		else if (ptr > sp->ptr)8052			node = node->rb_right;8053		else8054			break;8055	}8056	spin_unlock(&fs_info->swapfile_pins_lock);8057	return node != NULL;8058}8059 8060static int relocating_repair_kthread(void *data)8061{8062	struct btrfs_block_group *cache = data;8063	struct btrfs_fs_info *fs_info = cache->fs_info;8064	u64 target;8065	int ret = 0;8066 8067	target = cache->start;8068	btrfs_put_block_group(cache);8069 8070	sb_start_write(fs_info->sb);8071	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {8072		btrfs_info(fs_info,8073			   "zoned: skip relocating block group %llu to repair: EBUSY",8074			   target);8075		sb_end_write(fs_info->sb);8076		return -EBUSY;8077	}8078 8079	mutex_lock(&fs_info->reclaim_bgs_lock);8080 8081	/* Ensure block group still exists */8082	cache = btrfs_lookup_block_group(fs_info, target);8083	if (!cache)8084		goto out;8085 8086	if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))8087		goto out;8088 8089	ret = btrfs_may_alloc_data_chunk(fs_info, target);8090	if (ret < 0)8091		goto out;8092 8093	btrfs_info(fs_info,8094		   "zoned: relocating block group %llu to repair IO failure",8095		   target);8096	ret = btrfs_relocate_chunk(fs_info, target);8097 8098out:8099	if (cache)8100		btrfs_put_block_group(cache);8101	mutex_unlock(&fs_info->reclaim_bgs_lock);8102	btrfs_exclop_finish(fs_info);8103	sb_end_write(fs_info->sb);8104 8105	return ret;8106}8107 8108bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)8109{8110	struct btrfs_block_group *cache;8111 8112	if (!btrfs_is_zoned(fs_info))8113		return false;8114 8115	/* Do not attempt to repair in degraded state */8116	if (btrfs_test_opt(fs_info, DEGRADED))8117		return true;8118 8119	cache = btrfs_lookup_block_group(fs_info, logical);8120	if (!cache)8121		return true;8122 8123	if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {8124		btrfs_put_block_group(cache);8125		return true;8126	}8127 8128	kthread_run(relocating_repair_kthread, cache,8129		    "btrfs-relocating-repair");8130 8131	return true;8132}8133 8134static void map_raid56_repair_block(struct btrfs_io_context *bioc,8135				    struct btrfs_io_stripe *smap,8136				    u64 logical)8137{8138	int data_stripes = nr_bioc_data_stripes(bioc);8139	int i;8140 8141	for (i = 0; i < data_stripes; i++) {8142		u64 stripe_start = bioc->full_stripe_logical +8143				   btrfs_stripe_nr_to_offset(i);8144 8145		if (logical >= stripe_start &&8146		    logical < stripe_start + BTRFS_STRIPE_LEN)8147			break;8148	}8149	ASSERT(i < data_stripes);8150	smap->dev = bioc->stripes[i].dev;8151	smap->physical = bioc->stripes[i].physical +8152			((logical - bioc->full_stripe_logical) &8153			 BTRFS_STRIPE_LEN_MASK);8154}8155 8156/*8157 * Map a repair write into a single device.8158 *8159 * A repair write is triggered by read time repair or scrub, which would only8160 * update the contents of a single device.8161 * Not update any other mirrors nor go through RMW path.8162 *8163 * Callers should ensure:8164 *8165 * - Call btrfs_bio_counter_inc_blocked() first8166 * - The range does not cross stripe boundary8167 * - Has a valid @mirror_num passed in.8168 */8169int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,8170			   struct btrfs_io_stripe *smap, u64 logical,8171			   u32 length, int mirror_num)8172{8173	struct btrfs_io_context *bioc = NULL;8174	u64 map_length = length;8175	int mirror_ret = mirror_num;8176	int ret;8177 8178	ASSERT(mirror_num > 0);8179 8180	ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,8181			      &bioc, smap, &mirror_ret);8182	if (ret < 0)8183		return ret;8184 8185	/* The map range should not cross stripe boundary. */8186	ASSERT(map_length >= length);8187 8188	/* Already mapped to single stripe. */8189	if (!bioc)8190		goto out;8191 8192	/* Map the RAID56 multi-stripe writes to a single one. */8193	if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {8194		map_raid56_repair_block(bioc, smap, logical);8195		goto out;8196	}8197 8198	ASSERT(mirror_num <= bioc->num_stripes);8199	smap->dev = bioc->stripes[mirror_num - 1].dev;8200	smap->physical = bioc->stripes[mirror_num - 1].physical;8201out:8202	btrfs_put_bioc(bioc);8203	ASSERT(smap->dev);8204	return 0;8205}8206