1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2007 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/sched.h>
   7#include <linux/sched/mm.h>
   8#include <linux/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 "extent_map.h"
  18#include "disk-io.h"
  19#include "transaction.h"
  20#include "print-tree.h"
  21#include "volumes.h"
  22#include "raid56.h"
  23#include "rcu-string.h"
  24#include "dev-replace.h"
  25#include "sysfs.h"
  26#include "tree-checker.h"
  27#include "space-info.h"
  28#include "block-group.h"
  29#include "discard.h"
  30#include "zoned.h"
  31#include "fs.h"
  32#include "accessors.h"
  33#include "uuid-tree.h"
  34#include "ioctl.h"
  35#include "relocation.h"
  36#include "scrub.h"
  37#include "super.h"
  38#include "raid-stripe-tree.h"
  39
  40#define BTRFS_BLOCK_GROUP_STRIPE_MASK	(BTRFS_BLOCK_GROUP_RAID0 | \
  41					 BTRFS_BLOCK_GROUP_RAID10 | \
  42					 BTRFS_BLOCK_GROUP_RAID56_MASK)
  43
  44struct btrfs_io_geometry {
  45	u32 stripe_index;
  46	u32 stripe_nr;
  47	int mirror_num;
  48	int num_stripes;
  49	u64 stripe_offset;
  50	u64 raid56_full_stripe_start;
  51	int max_errors;
  52	enum btrfs_map_op op;
  53};
  54
  55const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
  56	[BTRFS_RAID_RAID10] = {
  57		.sub_stripes	= 2,
  58		.dev_stripes	= 1,
  59		.devs_max	= 0,	/* 0 == as many as possible */
  60		.devs_min	= 2,
  61		.tolerated_failures = 1,
  62		.devs_increment	= 2,
  63		.ncopies	= 2,
  64		.nparity        = 0,
  65		.raid_name	= "raid10",
  66		.bg_flag	= BTRFS_BLOCK_GROUP_RAID10,
  67		.mindev_error	= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
  68	},
  69	[BTRFS_RAID_RAID1] = {
  70		.sub_stripes	= 1,
  71		.dev_stripes	= 1,
  72		.devs_max	= 2,
  73		.devs_min	= 2,
  74		.tolerated_failures = 1,
  75		.devs_increment	= 2,
  76		.ncopies	= 2,
  77		.nparity        = 0,
  78		.raid_name	= "raid1",
  79		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1,
  80		.mindev_error	= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
  81	},
  82	[BTRFS_RAID_RAID1C3] = {
  83		.sub_stripes	= 1,
  84		.dev_stripes	= 1,
  85		.devs_max	= 3,
  86		.devs_min	= 3,
  87		.tolerated_failures = 2,
  88		.devs_increment	= 3,
  89		.ncopies	= 3,
  90		.nparity        = 0,
  91		.raid_name	= "raid1c3",
  92		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C3,
  93		.mindev_error	= BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
  94	},
  95	[BTRFS_RAID_RAID1C4] = {
  96		.sub_stripes	= 1,
  97		.dev_stripes	= 1,
  98		.devs_max	= 4,
  99		.devs_min	= 4,
 100		.tolerated_failures = 3,
 101		.devs_increment	= 4,
 102		.ncopies	= 4,
 103		.nparity        = 0,
 104		.raid_name	= "raid1c4",
 105		.bg_flag	= BTRFS_BLOCK_GROUP_RAID1C4,
 106		.mindev_error	= BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
 107	},
 108	[BTRFS_RAID_DUP] = {
 109		.sub_stripes	= 1,
 110		.dev_stripes	= 2,
 111		.devs_max	= 1,
 112		.devs_min	= 1,
 113		.tolerated_failures = 0,
 114		.devs_increment	= 1,
 115		.ncopies	= 2,
 116		.nparity        = 0,
 117		.raid_name	= "dup",
 118		.bg_flag	= BTRFS_BLOCK_GROUP_DUP,
 119		.mindev_error	= 0,
 120	},
 121	[BTRFS_RAID_RAID0] = {
 122		.sub_stripes	= 1,
 123		.dev_stripes	= 1,
 124		.devs_max	= 0,
 125		.devs_min	= 1,
 126		.tolerated_failures = 0,
 127		.devs_increment	= 1,
 128		.ncopies	= 1,
 129		.nparity        = 0,
 130		.raid_name	= "raid0",
 131		.bg_flag	= BTRFS_BLOCK_GROUP_RAID0,
 132		.mindev_error	= 0,
 133	},
 134	[BTRFS_RAID_SINGLE] = {
 135		.sub_stripes	= 1,
 136		.dev_stripes	= 1,
 137		.devs_max	= 1,
 138		.devs_min	= 1,
 139		.tolerated_failures = 0,
 140		.devs_increment	= 1,
 141		.ncopies	= 1,
 142		.nparity        = 0,
 143		.raid_name	= "single",
 144		.bg_flag	= 0,
 145		.mindev_error	= 0,
 146	},
 147	[BTRFS_RAID_RAID5] = {
 148		.sub_stripes	= 1,
 149		.dev_stripes	= 1,
 150		.devs_max	= 0,
 151		.devs_min	= 2,
 152		.tolerated_failures = 1,
 153		.devs_increment	= 1,
 154		.ncopies	= 1,
 155		.nparity        = 1,
 156		.raid_name	= "raid5",
 157		.bg_flag	= BTRFS_BLOCK_GROUP_RAID5,
 158		.mindev_error	= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
 159	},
 160	[BTRFS_RAID_RAID6] = {
 161		.sub_stripes	= 1,
 162		.dev_stripes	= 1,
 163		.devs_max	= 0,
 164		.devs_min	= 3,
 165		.tolerated_failures = 2,
 166		.devs_increment	= 1,
 167		.ncopies	= 1,
 168		.nparity        = 2,
 169		.raid_name	= "raid6",
 170		.bg_flag	= BTRFS_BLOCK_GROUP_RAID6,
 171		.mindev_error	= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
 172	},
 173};
 174
 175/*
 176 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
 177 * can be used as index to access btrfs_raid_array[].
 178 */
 179enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
 180{
 181	const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);
 182
 183	if (!profile)
 184		return BTRFS_RAID_SINGLE;
 185
 186	return BTRFS_BG_FLAG_TO_INDEX(profile);
 187}
 188
 189const char *btrfs_bg_type_to_raid_name(u64 flags)
 190{
 191	const int index = btrfs_bg_flags_to_raid_index(flags);
 192
 193	if (index >= BTRFS_NR_RAID_TYPES)
 194		return NULL;
 195
 196	return btrfs_raid_array[index].raid_name;
 197}
 198
 199int btrfs_nr_parity_stripes(u64 type)
 200{
 201	enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);
 202
 203	return btrfs_raid_array[index].nparity;
 204}
 205
 206/*
 207 * Fill @buf with textual description of @bg_flags, no more than @size_buf
 208 * bytes including terminating null byte.
 209 */
 210void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
 211{
 212	int i;
 213	int ret;
 214	char *bp = buf;
 215	u64 flags = bg_flags;
 216	u32 size_bp = size_buf;
 217
 218	if (!flags) {
 219		strcpy(bp, "NONE");
 220		return;
 221	}
 222
 223#define DESCRIBE_FLAG(flag, desc)						\
 224	do {								\
 225		if (flags & (flag)) {					\
 226			ret = snprintf(bp, size_bp, "%s|", (desc));	\
 227			if (ret < 0 || ret >= size_bp)			\
 228				goto out_overflow;			\
 229			size_bp -= ret;					\
 230			bp += ret;					\
 231			flags &= ~(flag);				\
 232		}							\
 233	} while (0)
 234
 235	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
 236	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
 237	DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
 238
 239	DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
 240	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
 241		DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
 242			      btrfs_raid_array[i].raid_name);
 243#undef DESCRIBE_FLAG
 244
 245	if (flags) {
 246		ret = snprintf(bp, size_bp, "0x%llx|", flags);
 247		size_bp -= ret;
 248	}
 249
 250	if (size_bp < size_buf)
 251		buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
 252
 253	/*
 254	 * The text is trimmed, it's up to the caller to provide sufficiently
 255	 * large buffer
 256	 */
 257out_overflow:;
 258}
 259
 260static int init_first_rw_device(struct btrfs_trans_handle *trans);
 261static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
 262static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
 263
 264/*
 265 * Device locking
 266 * ==============
 267 *
 268 * There are several mutexes that protect manipulation of devices and low-level
 269 * structures like chunks but not block groups, extents or files
 270 *
 271 * uuid_mutex (global lock)
 272 * ------------------------
 273 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
 274 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
 275 * device) or requested by the device= mount option
 276 *
 277 * the mutex can be very coarse and can cover long-running operations
 278 *
 279 * protects: updates to fs_devices counters like missing devices, rw devices,
 280 * seeding, structure cloning, opening/closing devices at mount/umount time
 281 *
 282 * global::fs_devs - add, remove, updates to the global list
 283 *
 284 * does not protect: manipulation of the fs_devices::devices list in general
 285 * but in mount context it could be used to exclude list modifications by eg.
 286 * scan ioctl
 287 *
 288 * btrfs_device::name - renames (write side), read is RCU
 289 *
 290 * fs_devices::device_list_mutex (per-fs, with RCU)
 291 * ------------------------------------------------
 292 * protects updates to fs_devices::devices, ie. adding and deleting
 293 *
 294 * simple list traversal with read-only actions can be done with RCU protection
 295 *
 296 * may be used to exclude some operations from running concurrently without any
 297 * modifications to the list (see write_all_supers)
 298 *
 299 * Is not required at mount and close times, because our device list is
 300 * protected by the uuid_mutex at that point.
 301 *
 302 * balance_mutex
 303 * -------------
 304 * protects balance structures (status, state) and context accessed from
 305 * several places (internally, ioctl)
 306 *
 307 * chunk_mutex
 308 * -----------
 309 * protects chunks, adding or removing during allocation, trim or when a new
 310 * device is added/removed. Additionally it also protects post_commit_list of
 311 * individual devices, since they can be added to the transaction's
 312 * post_commit_list only with chunk_mutex held.
 313 *
 314 * cleaner_mutex
 315 * -------------
 316 * a big lock that is held by the cleaner thread and prevents running subvolume
 317 * cleaning together with relocation or delayed iputs
 318 *
 319 *
 320 * Lock nesting
 321 * ============
 322 *
 323 * uuid_mutex
 324 *   device_list_mutex
 325 *     chunk_mutex
 326 *   balance_mutex
 327 *
 328 *
 329 * Exclusive operations
 330 * ====================
 331 *
 332 * Maintains the exclusivity of the following operations that apply to the
 333 * whole filesystem and cannot run in parallel.
 334 *
 335 * - Balance (*)
 336 * - Device add
 337 * - Device remove
 338 * - Device replace (*)
 339 * - Resize
 340 *
 341 * The device operations (as above) can be in one of the following states:
 342 *
 343 * - Running state
 344 * - Paused state
 345 * - Completed state
 346 *
 347 * Only device operations marked with (*) can go into the Paused state for the
 348 * following reasons:
 349 *
 350 * - ioctl (only Balance can be Paused through ioctl)
 351 * - filesystem remounted as read-only
 352 * - filesystem unmounted and mounted as read-only
 353 * - system power-cycle and filesystem mounted as read-only
 354 * - filesystem or device errors leading to forced read-only
 355 *
 356 * The status of exclusive operation is set and cleared atomically.
 357 * During the course of Paused state, fs_info::exclusive_operation remains set.
 358 * A device operation in Paused or Running state can be canceled or resumed
 359 * either by ioctl (Balance only) or when remounted as read-write.
 360 * The exclusive status is cleared when the device operation is canceled or
 361 * completed.
 362 */
 363
 364DEFINE_MUTEX(uuid_mutex);
 365static LIST_HEAD(fs_uuids);
 366struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
 367{
 368	return &fs_uuids;
 369}
 370
 371/*
 372 * Allocate new btrfs_fs_devices structure identified by a fsid.
 373 *
 374 * @fsid:    if not NULL, copy the UUID to fs_devices::fsid and to
 375 *           fs_devices::metadata_fsid
 376 *
 377 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
 378 * The returned struct is not linked onto any lists and can be destroyed with
 379 * kfree() right away.
 380 */
 381static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
 382{
 383	struct btrfs_fs_devices *fs_devs;
 384
 385	fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
 386	if (!fs_devs)
 387		return ERR_PTR(-ENOMEM);
 388
 389	mutex_init(&fs_devs->device_list_mutex);
 390
 391	INIT_LIST_HEAD(&fs_devs->devices);
 392	INIT_LIST_HEAD(&fs_devs->alloc_list);
 393	INIT_LIST_HEAD(&fs_devs->fs_list);
 394	INIT_LIST_HEAD(&fs_devs->seed_list);
 395
 396	if (fsid) {
 397		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
 398		memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
 399	}
 400
 401	return fs_devs;
 402}
 403
 404static void btrfs_free_device(struct btrfs_device *device)
 405{
 406	WARN_ON(!list_empty(&device->post_commit_list));
 407	rcu_string_free(device->name);
 408	extent_io_tree_release(&device->alloc_state);
 409	btrfs_destroy_dev_zone_info(device);
 410	kfree(device);
 411}
 412
 413static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
 414{
 415	struct btrfs_device *device;
 416
 417	WARN_ON(fs_devices->opened);
 418	while (!list_empty(&fs_devices->devices)) {
 419		device = list_entry(fs_devices->devices.next,
 420				    struct btrfs_device, dev_list);
 421		list_del(&device->dev_list);
 422		btrfs_free_device(device);
 423	}
 424	kfree(fs_devices);
 425}
 426
 427void __exit btrfs_cleanup_fs_uuids(void)
 428{
 429	struct btrfs_fs_devices *fs_devices;
 430
 431	while (!list_empty(&fs_uuids)) {
 432		fs_devices = list_entry(fs_uuids.next,
 433					struct btrfs_fs_devices, fs_list);
 434		list_del(&fs_devices->fs_list);
 435		free_fs_devices(fs_devices);
 436	}
 437}
 438
 439static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices,
 440				  const u8 *fsid, const u8 *metadata_fsid)
 441{
 442	if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0)
 443		return false;
 444
 445	if (!metadata_fsid)
 446		return true;
 447
 448	if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0)
 449		return false;
 450
 451	return true;
 452}
 453
 454static noinline struct btrfs_fs_devices *find_fsid(
 455		const u8 *fsid, const u8 *metadata_fsid)
 456{
 457	struct btrfs_fs_devices *fs_devices;
 458
 459	ASSERT(fsid);
 460
 461	/* Handle non-split brain cases */
 462	list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
 463		if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid))
 464			return fs_devices;
 465	}
 466	return NULL;
 467}
 468
 469static int
 470btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder,
 471		      int flush, struct bdev_handle **bdev_handle,
 472		      struct btrfs_super_block **disk_super)
 473{
 474	struct block_device *bdev;
 475	int ret;
 476
 477	*bdev_handle = bdev_open_by_path(device_path, flags, holder, NULL);
 478
 479	if (IS_ERR(*bdev_handle)) {
 480		ret = PTR_ERR(*bdev_handle);
 481		goto error;
 482	}
 483	bdev = (*bdev_handle)->bdev;
 484
 485	if (flush)
 486		sync_blockdev(bdev);
 487	ret = set_blocksize(bdev, BTRFS_BDEV_BLOCKSIZE);
 488	if (ret) {
 489		bdev_release(*bdev_handle);
 490		goto error;
 491	}
 492	invalidate_bdev(bdev);
 493	*disk_super = btrfs_read_dev_super(bdev);
 494	if (IS_ERR(*disk_super)) {
 495		ret = PTR_ERR(*disk_super);
 496		bdev_release(*bdev_handle);
 497		goto error;
 498	}
 499
 500	return 0;
 501
 502error:
 503	*bdev_handle = NULL;
 504	return ret;
 505}
 506
 507/*
 508 *  Search and remove all stale devices (which are not mounted).  When both
 509 *  inputs are NULL, it will search and release all stale devices.
 510 *
 511 *  @devt:         Optional. When provided will it release all unmounted devices
 512 *                 matching this devt only.
 513 *  @skip_device:  Optional. Will skip this device when searching for the stale
 514 *                 devices.
 515 *
 516 *  Return:	0 for success or if @devt is 0.
 517 *		-EBUSY if @devt is a mounted device.
 518 *		-ENOENT if @devt does not match any device in the list.
 519 */
 520static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
 521{
 522	struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
 523	struct btrfs_device *device, *tmp_device;
 524	int ret;
 525	bool freed = false;
 526
 527	lockdep_assert_held(&uuid_mutex);
 528
 529	/* Return good status if there is no instance of devt. */
 530	ret = 0;
 531	list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
 532
 533		mutex_lock(&fs_devices->device_list_mutex);
 534		list_for_each_entry_safe(device, tmp_device,
 535					 &fs_devices->devices, dev_list) {
 536			if (skip_device && skip_device == device)
 537				continue;
 538			if (devt && devt != device->devt)
 539				continue;
 540			if (fs_devices->opened) {
 541				if (devt)
 542					ret = -EBUSY;
 543				break;
 544			}
 545
 546			/* delete the stale device */
 547			fs_devices->num_devices--;
 548			list_del(&device->dev_list);
 549			btrfs_free_device(device);
 550
 551			freed = true;
 552		}
 553		mutex_unlock(&fs_devices->device_list_mutex);
 554
 555		if (fs_devices->num_devices == 0) {
 556			btrfs_sysfs_remove_fsid(fs_devices);
 557			list_del(&fs_devices->fs_list);
 558			free_fs_devices(fs_devices);
 559		}
 560	}
 561
 562	/* If there is at least one freed device return 0. */
 563	if (freed)
 564		return 0;
 565
 566	return ret;
 567}
 568
 569static struct btrfs_fs_devices *find_fsid_by_device(
 570					struct btrfs_super_block *disk_super,
 571					dev_t devt, bool *same_fsid_diff_dev)
 572{
 573	struct btrfs_fs_devices *fsid_fs_devices;
 574	struct btrfs_fs_devices *devt_fs_devices;
 575	const bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
 576					BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
 577	bool found_by_devt = false;
 578
 579	/* Find the fs_device by the usual method, if found use it. */
 580	fsid_fs_devices = find_fsid(disk_super->fsid,
 581		    has_metadata_uuid ? disk_super->metadata_uuid : NULL);
 582
 583	/* The temp_fsid feature is supported only with single device filesystem. */
 584	if (btrfs_super_num_devices(disk_super) != 1)
 585		return fsid_fs_devices;
 586
 587	/*
 588	 * A seed device is an integral component of the sprout device, which
 589	 * functions as a multi-device filesystem. So, temp-fsid feature is
 590	 * not supported.
 591	 */
 592	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)
 593		return fsid_fs_devices;
 594
 595	/* Try to find a fs_devices by matching devt. */
 596	list_for_each_entry(devt_fs_devices, &fs_uuids, fs_list) {
 597		struct btrfs_device *device;
 598
 599		list_for_each_entry(device, &devt_fs_devices->devices, dev_list) {
 600			if (device->devt == devt) {
 601				found_by_devt = true;
 602				break;
 603			}
 604		}
 605		if (found_by_devt)
 606			break;
 607	}
 608
 609	if (found_by_devt) {
 610		/* Existing device. */
 611		if (fsid_fs_devices == NULL) {
 612			if (devt_fs_devices->opened == 0) {
 613				/* Stale device. */
 614				return NULL;
 615			} else {
 616				/* temp_fsid is mounting a subvol. */
 617				return devt_fs_devices;
 618			}
 619		} else {
 620			/* Regular or temp_fsid device mounting a subvol. */
 621			return devt_fs_devices;
 622		}
 623	} else {
 624		/* New device. */
 625		if (fsid_fs_devices == NULL) {
 626			return NULL;
 627		} else {
 628			/* sb::fsid is already used create a new temp_fsid. */
 629			*same_fsid_diff_dev = true;
 630			return NULL;
 631		}
 632	}
 633
 634	/* Not reached. */
 635}
 636
 637/*
 638 * This is only used on mount, and we are protected from competing things
 639 * messing with our fs_devices by the uuid_mutex, thus we do not need the
 640 * fs_devices->device_list_mutex here.
 641 */
 642static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
 643			struct btrfs_device *device, blk_mode_t flags,
 644			void *holder)
 645{
 646	struct bdev_handle *bdev_handle;
 647	struct btrfs_super_block *disk_super;
 648	u64 devid;
 649	int ret;
 650
 651	if (device->bdev)
 652		return -EINVAL;
 653	if (!device->name)
 654		return -EINVAL;
 655
 656	ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 657				    &bdev_handle, &disk_super);
 658	if (ret)
 659		return ret;
 660
 661	devid = btrfs_stack_device_id(&disk_super->dev_item);
 662	if (devid != device->devid)
 663		goto error_free_page;
 664
 665	if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
 666		goto error_free_page;
 667
 668	device->generation = btrfs_super_generation(disk_super);
 669
 670	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 671		if (btrfs_super_incompat_flags(disk_super) &
 672		    BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
 673			pr_err(
 674		"BTRFS: Invalid seeding and uuid-changed device detected\n");
 675			goto error_free_page;
 676		}
 677
 678		clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 679		fs_devices->seeding = true;
 680	} else {
 681		if (bdev_read_only(bdev_handle->bdev))
 682			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 683		else
 684			set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 685	}
 686
 687	if (!bdev_nonrot(bdev_handle->bdev))
 688		fs_devices->rotating = true;
 689
 690	if (bdev_max_discard_sectors(bdev_handle->bdev))
 691		fs_devices->discardable = true;
 692
 693	device->bdev_handle = bdev_handle;
 694	device->bdev = bdev_handle->bdev;
 695	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
 696
 697	fs_devices->open_devices++;
 698	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
 699	    device->devid != BTRFS_DEV_REPLACE_DEVID) {
 700		fs_devices->rw_devices++;
 701		list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
 702	}
 703	btrfs_release_disk_super(disk_super);
 704
 705	return 0;
 706
 707error_free_page:
 708	btrfs_release_disk_super(disk_super);
 709	bdev_release(bdev_handle);
 710
 711	return -EINVAL;
 712}
 713
 714u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb)
 715{
 716	bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) &
 717				  BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
 718
 719	return has_metadata_uuid ? sb->metadata_uuid : sb->fsid;
 720}
 721
 722/*
 723 * Add new device to list of registered devices
 724 *
 725 * Returns:
 726 * device pointer which was just added or updated when successful
 727 * error pointer when failed
 728 */
 729static noinline struct btrfs_device *device_list_add(const char *path,
 730			   struct btrfs_super_block *disk_super,
 731			   bool *new_device_added)
 732{
 733	struct btrfs_device *device;
 734	struct btrfs_fs_devices *fs_devices = NULL;
 735	struct rcu_string *name;
 736	u64 found_transid = btrfs_super_generation(disk_super);
 737	u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
 738	dev_t path_devt;
 739	int error;
 740	bool same_fsid_diff_dev = false;
 741	bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
 742		BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
 743
 744	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
 745		btrfs_err(NULL,
 746"device %s has incomplete metadata_uuid change, please use btrfstune to complete",
 747			  path);
 748		return ERR_PTR(-EAGAIN);
 749	}
 750
 751	error = lookup_bdev(path, &path_devt);
 752	if (error) {
 753		btrfs_err(NULL, "failed to lookup block device for path %s: %d",
 754			  path, error);
 755		return ERR_PTR(error);
 756	}
 757
 758	fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev);
 759
 760	if (!fs_devices) {
 761		fs_devices = alloc_fs_devices(disk_super->fsid);
 762		if (IS_ERR(fs_devices))
 763			return ERR_CAST(fs_devices);
 764
 765		if (has_metadata_uuid)
 766			memcpy(fs_devices->metadata_uuid,
 767			       disk_super->metadata_uuid, BTRFS_FSID_SIZE);
 768
 769		if (same_fsid_diff_dev) {
 770			generate_random_uuid(fs_devices->fsid);
 771			fs_devices->temp_fsid = true;
 772			pr_info("BTRFS: device %s using temp-fsid %pU\n",
 773				path, fs_devices->fsid);
 774		}
 775
 776		mutex_lock(&fs_devices->device_list_mutex);
 777		list_add(&fs_devices->fs_list, &fs_uuids);
 778
 779		device = NULL;
 780	} else {
 781		struct btrfs_dev_lookup_args args = {
 782			.devid = devid,
 783			.uuid = disk_super->dev_item.uuid,
 784		};
 785
 786		mutex_lock(&fs_devices->device_list_mutex);
 787		device = btrfs_find_device(fs_devices, &args);
 788
 789		if (found_transid > fs_devices->latest_generation) {
 790			memcpy(fs_devices->fsid, disk_super->fsid,
 791					BTRFS_FSID_SIZE);
 792			memcpy(fs_devices->metadata_uuid,
 793			       btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE);
 794		}
 795	}
 796
 797	if (!device) {
 798		unsigned int nofs_flag;
 799
 800		if (fs_devices->opened) {
 801			btrfs_err(NULL,
 802"device %s belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",
 803				  path, fs_devices->fsid, current->comm,
 804				  task_pid_nr(current));
 805			mutex_unlock(&fs_devices->device_list_mutex);
 806			return ERR_PTR(-EBUSY);
 807		}
 808
 809		nofs_flag = memalloc_nofs_save();
 810		device = btrfs_alloc_device(NULL, &devid,
 811					    disk_super->dev_item.uuid, path);
 812		memalloc_nofs_restore(nofs_flag);
 813		if (IS_ERR(device)) {
 814			mutex_unlock(&fs_devices->device_list_mutex);
 815			/* we can safely leave the fs_devices entry around */
 816			return device;
 817		}
 818
 819		device->devt = path_devt;
 820
 821		list_add_rcu(&device->dev_list, &fs_devices->devices);
 822		fs_devices->num_devices++;
 823
 824		device->fs_devices = fs_devices;
 825		*new_device_added = true;
 826
 827		if (disk_super->label[0])
 828			pr_info(
 829	"BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
 830				disk_super->label, devid, found_transid, path,
 831				current->comm, task_pid_nr(current));
 832		else
 833			pr_info(
 834	"BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
 835				disk_super->fsid, devid, found_transid, path,
 836				current->comm, task_pid_nr(current));
 837
 838	} else if (!device->name || strcmp(device->name->str, path)) {
 839		/*
 840		 * When FS is already mounted.
 841		 * 1. If you are here and if the device->name is NULL that
 842		 *    means this device was missing at time of FS mount.
 843		 * 2. If you are here and if the device->name is different
 844		 *    from 'path' that means either
 845		 *      a. The same device disappeared and reappeared with
 846		 *         different name. or
 847		 *      b. The missing-disk-which-was-replaced, has
 848		 *         reappeared now.
 849		 *
 850		 * We must allow 1 and 2a above. But 2b would be a spurious
 851		 * and unintentional.
 852		 *
 853		 * Further in case of 1 and 2a above, the disk at 'path'
 854		 * would have missed some transaction when it was away and
 855		 * in case of 2a the stale bdev has to be updated as well.
 856		 * 2b must not be allowed at all time.
 857		 */
 858
 859		/*
 860		 * For now, we do allow update to btrfs_fs_device through the
 861		 * btrfs dev scan cli after FS has been mounted.  We're still
 862		 * tracking a problem where systems fail mount by subvolume id
 863		 * when we reject replacement on a mounted FS.
 864		 */
 865		if (!fs_devices->opened && found_transid < device->generation) {
 866			/*
 867			 * That is if the FS is _not_ mounted and if you
 868			 * are here, that means there is more than one
 869			 * disk with same uuid and devid.We keep the one
 870			 * with larger generation number or the last-in if
 871			 * generation are equal.
 872			 */
 873			mutex_unlock(&fs_devices->device_list_mutex);
 874			btrfs_err(NULL,
 875"device %s already registered with a higher generation, found %llu expect %llu",
 876				  path, found_transid, device->generation);
 877			return ERR_PTR(-EEXIST);
 878		}
 879
 880		/*
 881		 * We are going to replace the device path for a given devid,
 882		 * make sure it's the same device if the device is mounted
 883		 *
 884		 * NOTE: the device->fs_info may not be reliable here so pass
 885		 * in a NULL to message helpers instead. This avoids a possible
 886		 * use-after-free when the fs_info and fs_info->sb are already
 887		 * torn down.
 888		 */
 889		if (device->bdev) {
 890			if (device->devt != path_devt) {
 891				mutex_unlock(&fs_devices->device_list_mutex);
 892				btrfs_warn_in_rcu(NULL,
 893	"duplicate device %s devid %llu generation %llu scanned by %s (%d)",
 894						  path, devid, found_transid,
 895						  current->comm,
 896						  task_pid_nr(current));
 897				return ERR_PTR(-EEXIST);
 898			}
 899			btrfs_info_in_rcu(NULL,
 900	"devid %llu device path %s changed to %s scanned by %s (%d)",
 901					  devid, btrfs_dev_name(device),
 902					  path, current->comm,
 903					  task_pid_nr(current));
 904		}
 905
 906		name = rcu_string_strdup(path, GFP_NOFS);
 907		if (!name) {
 908			mutex_unlock(&fs_devices->device_list_mutex);
 909			return ERR_PTR(-ENOMEM);
 910		}
 911		rcu_string_free(device->name);
 912		rcu_assign_pointer(device->name, name);
 913		if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
 914			fs_devices->missing_devices--;
 915			clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
 916		}
 917		device->devt = path_devt;
 918	}
 919
 920	/*
 921	 * Unmount does not free the btrfs_device struct but would zero
 922	 * generation along with most of the other members. So just update
 923	 * it back. We need it to pick the disk with largest generation
 924	 * (as above).
 925	 */
 926	if (!fs_devices->opened) {
 927		device->generation = found_transid;
 928		fs_devices->latest_generation = max_t(u64, found_transid,
 929						fs_devices->latest_generation);
 930	}
 931
 932	fs_devices->total_devices = btrfs_super_num_devices(disk_super);
 933
 934	mutex_unlock(&fs_devices->device_list_mutex);
 935	return device;
 936}
 937
 938static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 939{
 940	struct btrfs_fs_devices *fs_devices;
 941	struct btrfs_device *device;
 942	struct btrfs_device *orig_dev;
 943	int ret = 0;
 944
 945	lockdep_assert_held(&uuid_mutex);
 946
 947	fs_devices = alloc_fs_devices(orig->fsid);
 948	if (IS_ERR(fs_devices))
 949		return fs_devices;
 950
 951	fs_devices->total_devices = orig->total_devices;
 952
 953	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 954		const char *dev_path = NULL;
 955
 956		/*
 957		 * This is ok to do without RCU read locked because we hold the
 958		 * uuid mutex so nothing we touch in here is going to disappear.
 959		 */
 960		if (orig_dev->name)
 961			dev_path = orig_dev->name->str;
 962
 963		device = btrfs_alloc_device(NULL, &orig_dev->devid,
 964					    orig_dev->uuid, dev_path);
 965		if (IS_ERR(device)) {
 966			ret = PTR_ERR(device);
 967			goto error;
 968		}
 969
 970		if (orig_dev->zone_info) {
 971			struct btrfs_zoned_device_info *zone_info;
 972
 973			zone_info = btrfs_clone_dev_zone_info(orig_dev);
 974			if (!zone_info) {
 975				btrfs_free_device(device);
 976				ret = -ENOMEM;
 977				goto error;
 978			}
 979			device->zone_info = zone_info;
 980		}
 981
 982		list_add(&device->dev_list, &fs_devices->devices);
 983		device->fs_devices = fs_devices;
 984		fs_devices->num_devices++;
 985	}
 986	return fs_devices;
 987error:
 988	free_fs_devices(fs_devices);
 989	return ERR_PTR(ret);
 990}
 991
 992static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
 993				      struct btrfs_device **latest_dev)
 994{
 995	struct btrfs_device *device, *next;
 996
 997	/* This is the initialized path, it is safe to release the devices. */
 998	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 999		if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1000			if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1001				      &device->dev_state) &&
1002			    !test_bit(BTRFS_DEV_STATE_MISSING,
1003				      &device->dev_state) &&
1004			    (!*latest_dev ||
1005			     device->generation > (*latest_dev)->generation)) {
1006				*latest_dev = device;
1007			}
1008			continue;
1009		}
1010
1011		/*
1012		 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1013		 * in btrfs_init_dev_replace() so just continue.
1014		 */
1015		if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1016			continue;
1017
1018		if (device->bdev_handle) {
1019			bdev_release(device->bdev_handle);
1020			device->bdev = NULL;
1021			device->bdev_handle = NULL;
1022			fs_devices->open_devices--;
1023		}
1024		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1025			list_del_init(&device->dev_alloc_list);
1026			clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1027			fs_devices->rw_devices--;
1028		}
1029		list_del_init(&device->dev_list);
1030		fs_devices->num_devices--;
1031		btrfs_free_device(device);
1032	}
1033
1034}
1035
1036/*
1037 * After we have read the system tree and know devids belonging to this
1038 * filesystem, remove the device which does not belong there.
1039 */
1040void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1041{
1042	struct btrfs_device *latest_dev = NULL;
1043	struct btrfs_fs_devices *seed_dev;
1044
1045	mutex_lock(&uuid_mutex);
1046	__btrfs_free_extra_devids(fs_devices, &latest_dev);
1047
1048	list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1049		__btrfs_free_extra_devids(seed_dev, &latest_dev);
1050
1051	fs_devices->latest_dev = latest_dev;
1052
1053	mutex_unlock(&uuid_mutex);
1054}
1055
1056static void btrfs_close_bdev(struct btrfs_device *device)
1057{
1058	if (!device->bdev)
1059		return;
1060
1061	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1062		sync_blockdev(device->bdev);
1063		invalidate_bdev(device->bdev);
1064	}
1065
1066	bdev_release(device->bdev_handle);
1067}
1068
1069static void btrfs_close_one_device(struct btrfs_device *device)
1070{
1071	struct btrfs_fs_devices *fs_devices = device->fs_devices;
1072
1073	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1074	    device->devid != BTRFS_DEV_REPLACE_DEVID) {
1075		list_del_init(&device->dev_alloc_list);
1076		fs_devices->rw_devices--;
1077	}
1078
1079	if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1080		clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1081
1082	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1083		clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1084		fs_devices->missing_devices--;
1085	}
1086
1087	btrfs_close_bdev(device);
1088	if (device->bdev) {
1089		fs_devices->open_devices--;
1090		device->bdev = NULL;
1091	}
1092	clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1093	btrfs_destroy_dev_zone_info(device);
1094
1095	device->fs_info = NULL;
1096	atomic_set(&device->dev_stats_ccnt, 0);
1097	extent_io_tree_release(&device->alloc_state);
1098
1099	/*
1100	 * Reset the flush error record. We might have a transient flush error
1101	 * in this mount, and if so we aborted the current transaction and set
1102	 * the fs to an error state, guaranteeing no super blocks can be further
1103	 * committed. However that error might be transient and if we unmount the
1104	 * filesystem and mount it again, we should allow the mount to succeed
1105	 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1106	 * filesystem again we still get flush errors, then we will again abort
1107	 * any transaction and set the error state, guaranteeing no commits of
1108	 * unsafe super blocks.
1109	 */
1110	device->last_flush_error = 0;
1111
1112	/* Verify the device is back in a pristine state  */
1113	WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1114	WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1115	WARN_ON(!list_empty(&device->dev_alloc_list));
1116	WARN_ON(!list_empty(&device->post_commit_list));
1117}
1118
1119static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1120{
1121	struct btrfs_device *device, *tmp;
1122
1123	lockdep_assert_held(&uuid_mutex);
1124
1125	if (--fs_devices->opened > 0)
1126		return;
1127
1128	list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1129		btrfs_close_one_device(device);
1130
1131	WARN_ON(fs_devices->open_devices);
1132	WARN_ON(fs_devices->rw_devices);
1133	fs_devices->opened = 0;
1134	fs_devices->seeding = false;
1135	fs_devices->fs_info = NULL;
1136}
1137
1138void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1139{
1140	LIST_HEAD(list);
1141	struct btrfs_fs_devices *tmp;
1142
1143	mutex_lock(&uuid_mutex);
1144	close_fs_devices(fs_devices);
1145	if (!fs_devices->opened) {
1146		list_splice_init(&fs_devices->seed_list, &list);
1147
1148		/*
1149		 * If the struct btrfs_fs_devices is not assembled with any
1150		 * other device, it can be re-initialized during the next mount
1151		 * without the needing device-scan step. Therefore, it can be
1152		 * fully freed.
1153		 */
1154		if (fs_devices->num_devices == 1) {
1155			list_del(&fs_devices->fs_list);
1156			free_fs_devices(fs_devices);
1157		}
1158	}
1159
1160
1161	list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1162		close_fs_devices(fs_devices);
1163		list_del(&fs_devices->seed_list);
1164		free_fs_devices(fs_devices);
1165	}
1166	mutex_unlock(&uuid_mutex);
1167}
1168
1169static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1170				blk_mode_t flags, void *holder)
1171{
1172	struct btrfs_device *device;
1173	struct btrfs_device *latest_dev = NULL;
1174	struct btrfs_device *tmp_device;
1175
1176	list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1177				 dev_list) {
1178		int ret;
1179
1180		ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1181		if (ret == 0 &&
1182		    (!latest_dev || device->generation > latest_dev->generation)) {
1183			latest_dev = device;
1184		} else if (ret == -ENODATA) {
1185			fs_devices->num_devices--;
1186			list_del(&device->dev_list);
1187			btrfs_free_device(device);
1188		}
1189	}
1190	if (fs_devices->open_devices == 0)
1191		return -EINVAL;
1192
1193	fs_devices->opened = 1;
1194	fs_devices->latest_dev = latest_dev;
1195	fs_devices->total_rw_bytes = 0;
1196	fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1197	fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1198
1199	return 0;
1200}
1201
1202static int devid_cmp(void *priv, const struct list_head *a,
1203		     const struct list_head *b)
1204{
1205	const struct btrfs_device *dev1, *dev2;
1206
1207	dev1 = list_entry(a, struct btrfs_device, dev_list);
1208	dev2 = list_entry(b, struct btrfs_device, dev_list);
1209
1210	if (dev1->devid < dev2->devid)
1211		return -1;
1212	else if (dev1->devid > dev2->devid)
1213		return 1;
1214	return 0;
1215}
1216
1217int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1218		       blk_mode_t flags, void *holder)
1219{
1220	int ret;
1221
1222	lockdep_assert_held(&uuid_mutex);
1223	/*
1224	 * The device_list_mutex cannot be taken here in case opening the
1225	 * underlying device takes further locks like open_mutex.
1226	 *
1227	 * We also don't need the lock here as this is called during mount and
1228	 * exclusion is provided by uuid_mutex
1229	 */
1230
1231	if (fs_devices->opened) {
1232		fs_devices->opened++;
1233		ret = 0;
1234	} else {
1235		list_sort(NULL, &fs_devices->devices, devid_cmp);
1236		ret = open_fs_devices(fs_devices, flags, holder);
1237	}
1238
1239	return ret;
1240}
1241
1242void btrfs_release_disk_super(struct btrfs_super_block *super)
1243{
1244	struct page *page = virt_to_page(super);
1245
1246	put_page(page);
1247}
1248
1249static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1250						       u64 bytenr, u64 bytenr_orig)
1251{
1252	struct btrfs_super_block *disk_super;
1253	struct page *page;
1254	void *p;
1255	pgoff_t index;
1256
1257	/* make sure our super fits in the device */
1258	if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1259		return ERR_PTR(-EINVAL);
1260
1261	/* make sure our super fits in the page */
1262	if (sizeof(*disk_super) > PAGE_SIZE)
1263		return ERR_PTR(-EINVAL);
1264
1265	/* make sure our super doesn't straddle pages on disk */
1266	index = bytenr >> PAGE_SHIFT;
1267	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1268		return ERR_PTR(-EINVAL);
1269
1270	/* pull in the page with our super */
1271	page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1272
1273	if (IS_ERR(page))
1274		return ERR_CAST(page);
1275
1276	p = page_address(page);
1277
1278	/* align our pointer to the offset of the super block */
1279	disk_super = p + offset_in_page(bytenr);
1280
1281	if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1282	    btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1283		btrfs_release_disk_super(p);
1284		return ERR_PTR(-EINVAL);
1285	}
1286
1287	if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1288		disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1289
1290	return disk_super;
1291}
1292
1293int btrfs_forget_devices(dev_t devt)
1294{
1295	int ret;
1296
1297	mutex_lock(&uuid_mutex);
1298	ret = btrfs_free_stale_devices(devt, NULL);
1299	mutex_unlock(&uuid_mutex);
1300
1301	return ret;
1302}
1303
1304/*
1305 * Look for a btrfs signature on a device. This may be called out of the mount path
1306 * and we are not allowed to call set_blocksize during the scan. The superblock
1307 * is read via pagecache.
1308 *
1309 * With @mount_arg_dev it's a scan during mount time that will always register
1310 * the device or return an error. Multi-device and seeding devices are registered
1311 * in both cases.
1312 */
1313struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
1314					   bool mount_arg_dev)
1315{
1316	struct btrfs_super_block *disk_super;
1317	bool new_device_added = false;
1318	struct btrfs_device *device = NULL;
1319	struct bdev_handle *bdev_handle;
1320	u64 bytenr, bytenr_orig;
1321	int ret;
1322
1323	lockdep_assert_held(&uuid_mutex);
1324
1325	/*
1326	 * we would like to check all the supers, but that would make
1327	 * a btrfs mount succeed after a mkfs from a different FS.
1328	 * So, we need to add a special mount option to scan for
1329	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1330	 */
1331
1332	/*
1333	 * Avoid an exclusive open here, as the systemd-udev may initiate the
1334	 * device scan which may race with the user's mount or mkfs command,
1335	 * resulting in failure.
1336	 * Since the device scan is solely for reading purposes, there is no
1337	 * need for an exclusive open. Additionally, the devices are read again
1338	 * during the mount process. It is ok to get some inconsistent
1339	 * values temporarily, as the device paths of the fsid are the only
1340	 * required information for assembling the volume.
1341	 */
1342	bdev_handle = bdev_open_by_path(path, flags, NULL, NULL);
1343	if (IS_ERR(bdev_handle))
1344		return ERR_CAST(bdev_handle);
1345
1346	bytenr_orig = btrfs_sb_offset(0);
1347	ret = btrfs_sb_log_location_bdev(bdev_handle->bdev, 0, READ, &bytenr);
1348	if (ret) {
1349		device = ERR_PTR(ret);
1350		goto error_bdev_put;
1351	}
1352
1353	disk_super = btrfs_read_disk_super(bdev_handle->bdev, bytenr,
1354					   bytenr_orig);
1355	if (IS_ERR(disk_super)) {
1356		device = ERR_CAST(disk_super);
1357		goto error_bdev_put;
1358	}
1359
1360	if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 &&
1361	    !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)) {
1362		dev_t devt;
1363
1364		ret = lookup_bdev(path, &devt);
1365		if (ret)
1366			btrfs_warn(NULL, "lookup bdev failed for path %s: %d",
1367				   path, ret);
1368		else
1369			btrfs_free_stale_devices(devt, NULL);
1370
1371		pr_debug("BTRFS: skip registering single non-seed device %s\n", path);
1372		device = NULL;
1373		goto free_disk_super;
1374	}
1375
1376	device = device_list_add(path, disk_super, &new_device_added);
1377	if (!IS_ERR(device) && new_device_added)
1378		btrfs_free_stale_devices(device->devt, device);
1379
1380free_disk_super:
1381	btrfs_release_disk_super(disk_super);
1382
1383error_bdev_put:
1384	bdev_release(bdev_handle);
1385
1386	return device;
1387}
1388
1389/*
1390 * Try to find a chunk that intersects [start, start + len] range and when one
1391 * such is found, record the end of it in *start
1392 */
1393static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1394				    u64 len)
1395{
1396	u64 physical_start, physical_end;
1397
1398	lockdep_assert_held(&device->fs_info->chunk_mutex);
1399
1400	if (find_first_extent_bit(&device->alloc_state, *start,
1401				  &physical_start, &physical_end,
1402				  CHUNK_ALLOCATED, NULL)) {
1403
1404		if (in_range(physical_start, *start, len) ||
1405		    in_range(*start, physical_start,
1406			     physical_end - physical_start)) {
1407			*start = physical_end + 1;
1408			return true;
1409		}
1410	}
1411	return false;
1412}
1413
1414static u64 dev_extent_search_start(struct btrfs_device *device)
1415{
1416	switch (device->fs_devices->chunk_alloc_policy) {
1417	case BTRFS_CHUNK_ALLOC_REGULAR:
1418		return BTRFS_DEVICE_RANGE_RESERVED;
1419	case BTRFS_CHUNK_ALLOC_ZONED:
1420		/*
1421		 * We don't care about the starting region like regular
1422		 * allocator, because we anyway use/reserve the first two zones
1423		 * for superblock logging.
1424		 */
1425		return 0;
1426	default:
1427		BUG();
1428	}
1429}
1430
1431static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1432					u64 *hole_start, u64 *hole_size,
1433					u64 num_bytes)
1434{
1435	u64 zone_size = device->zone_info->zone_size;
1436	u64 pos;
1437	int ret;
1438	bool changed = false;
1439
1440	ASSERT(IS_ALIGNED(*hole_start, zone_size));
1441
1442	while (*hole_size > 0) {
1443		pos = btrfs_find_allocatable_zones(device, *hole_start,
1444						   *hole_start + *hole_size,
1445						   num_bytes);
1446		if (pos != *hole_start) {
1447			*hole_size = *hole_start + *hole_size - pos;
1448			*hole_start = pos;
1449			changed = true;
1450			if (*hole_size < num_bytes)
1451				break;
1452		}
1453
1454		ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1455
1456		/* Range is ensured to be empty */
1457		if (!ret)
1458			return changed;
1459
1460		/* Given hole range was invalid (outside of device) */
1461		if (ret == -ERANGE) {
1462			*hole_start += *hole_size;
1463			*hole_size = 0;
1464			return true;
1465		}
1466
1467		*hole_start += zone_size;
1468		*hole_size -= zone_size;
1469		changed = true;
1470	}
1471
1472	return changed;
1473}
1474
1475/*
1476 * Check if specified hole is suitable for allocation.
1477 *
1478 * @device:	the device which we have the hole
1479 * @hole_start: starting position of the hole
1480 * @hole_size:	the size of the hole
1481 * @num_bytes:	the size of the free space that we need
1482 *
1483 * This function may modify @hole_start and @hole_size to reflect the suitable
1484 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1485 */
1486static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1487				  u64 *hole_size, u64 num_bytes)
1488{
1489	bool changed = false;
1490	u64 hole_end = *hole_start + *hole_size;
1491
1492	for (;;) {
1493		/*
1494		 * Check before we set max_hole_start, otherwise we could end up
1495		 * sending back this offset anyway.
1496		 */
1497		if (contains_pending_extent(device, hole_start, *hole_size)) {
1498			if (hole_end >= *hole_start)
1499				*hole_size = hole_end - *hole_start;
1500			else
1501				*hole_size = 0;
1502			changed = true;
1503		}
1504
1505		switch (device->fs_devices->chunk_alloc_policy) {
1506		case BTRFS_CHUNK_ALLOC_REGULAR:
1507			/* No extra check */
1508			break;
1509		case BTRFS_CHUNK_ALLOC_ZONED:
1510			if (dev_extent_hole_check_zoned(device, hole_start,
1511							hole_size, num_bytes)) {
1512				changed = true;
1513				/*
1514				 * The changed hole can contain pending extent.
1515				 * Loop again to check that.
1516				 */
1517				continue;
1518			}
1519			break;
1520		default:
1521			BUG();
1522		}
1523
1524		break;
1525	}
1526
1527	return changed;
1528}
1529
1530/*
1531 * Find free space in the specified device.
1532 *
1533 * @device:	  the device which we search the free space in
1534 * @num_bytes:	  the size of the free space that we need
1535 * @search_start: the position from which to begin the search
1536 * @start:	  store the start of the free space.
1537 * @len:	  the size of the free space. that we find, or the size
1538 *		  of the max free space if we don't find suitable free space
1539 *
1540 * This does a pretty simple search, the expectation is that it is called very
1541 * infrequently and that a given device has a small number of extents.
1542 *
1543 * @start is used to store the start of the free space if we find. But if we
1544 * don't find suitable free space, it will be used to store the start position
1545 * of the max free space.
1546 *
1547 * @len is used to store the size of the free space that we find.
1548 * But if we don't find suitable free space, it is used to store the size of
1549 * the max free space.
1550 *
1551 * NOTE: This function will search *commit* root of device tree, and does extra
1552 * check to ensure dev extents are not double allocated.
1553 * This makes the function safe to allocate dev extents but may not report
1554 * correct usable device space, as device extent freed in current transaction
1555 * is not reported as available.
1556 */
1557static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1558				u64 *start, u64 *len)
1559{
1560	struct btrfs_fs_info *fs_info = device->fs_info;
1561	struct btrfs_root *root = fs_info->dev_root;
1562	struct btrfs_key key;
1563	struct btrfs_dev_extent *dev_extent;
1564	struct btrfs_path *path;
1565	u64 search_start;
1566	u64 hole_size;
1567	u64 max_hole_start;
1568	u64 max_hole_size = 0;
1569	u64 extent_end;
1570	u64 search_end = device->total_bytes;
1571	int ret;
1572	int slot;
1573	struct extent_buffer *l;
1574
1575	search_start = dev_extent_search_start(device);
1576	max_hole_start = search_start;
1577
1578	WARN_ON(device->zone_info &&
1579		!IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1580
1581	path = btrfs_alloc_path();
1582	if (!path) {
1583		ret = -ENOMEM;
1584		goto out;
1585	}
1586again:
1587	if (search_start >= search_end ||
1588		test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1589		ret = -ENOSPC;
1590		goto out;
1591	}
1592
1593	path->reada = READA_FORWARD;
1594	path->search_commit_root = 1;
1595	path->skip_locking = 1;
1596
1597	key.objectid = device->devid;
1598	key.offset = search_start;
1599	key.type = BTRFS_DEV_EXTENT_KEY;
1600
1601	ret = btrfs_search_backwards(root, &key, path);
1602	if (ret < 0)
1603		goto out;
1604
1605	while (search_start < search_end) {
1606		l = path->nodes[0];
1607		slot = path->slots[0];
1608		if (slot >= btrfs_header_nritems(l)) {
1609			ret = btrfs_next_leaf(root, path);
1610			if (ret == 0)
1611				continue;
1612			if (ret < 0)
1613				goto out;
1614
1615			break;
1616		}
1617		btrfs_item_key_to_cpu(l, &key, slot);
1618
1619		if (key.objectid < device->devid)
1620			goto next;
1621
1622		if (key.objectid > device->devid)
1623			break;
1624
1625		if (key.type != BTRFS_DEV_EXTENT_KEY)
1626			goto next;
1627
1628		if (key.offset > search_end)
1629			break;
1630
1631		if (key.offset > search_start) {
1632			hole_size = key.offset - search_start;
1633			dev_extent_hole_check(device, &search_start, &hole_size,
1634					      num_bytes);
1635
1636			if (hole_size > max_hole_size) {
1637				max_hole_start = search_start;
1638				max_hole_size = hole_size;
1639			}
1640
1641			/*
1642			 * If this free space is greater than which we need,
1643			 * it must be the max free space that we have found
1644			 * until now, so max_hole_start must point to the start
1645			 * of this free space and the length of this free space
1646			 * is stored in max_hole_size. Thus, we return
1647			 * max_hole_start and max_hole_size and go back to the
1648			 * caller.
1649			 */
1650			if (hole_size >= num_bytes) {
1651				ret = 0;
1652				goto out;
1653			}
1654		}
1655
1656		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1657		extent_end = key.offset + btrfs_dev_extent_length(l,
1658								  dev_extent);
1659		if (extent_end > search_start)
1660			search_start = extent_end;
1661next:
1662		path->slots[0]++;
1663		cond_resched();
1664	}
1665
1666	/*
1667	 * At this point, search_start should be the end of
1668	 * allocated dev extents, and when shrinking the device,
1669	 * search_end may be smaller than search_start.
1670	 */
1671	if (search_end > search_start) {
1672		hole_size = search_end - search_start;
1673		if (dev_extent_hole_check(device, &search_start, &hole_size,
1674					  num_bytes)) {
1675			btrfs_release_path(path);
1676			goto again;
1677		}
1678
1679		if (hole_size > max_hole_size) {
1680			max_hole_start = search_start;
1681			max_hole_size = hole_size;
1682		}
1683	}
1684
1685	/* See above. */
1686	if (max_hole_size < num_bytes)
1687		ret = -ENOSPC;
1688	else
1689		ret = 0;
1690
1691	ASSERT(max_hole_start + max_hole_size <= search_end);
1692out:
1693	btrfs_free_path(path);
1694	*start = max_hole_start;
1695	if (len)
1696		*len = max_hole_size;
1697	return ret;
1698}
1699
1700static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1701			  struct btrfs_device *device,
1702			  u64 start, u64 *dev_extent_len)
1703{
1704	struct btrfs_fs_info *fs_info = device->fs_info;
1705	struct btrfs_root *root = fs_info->dev_root;
1706	int ret;
1707	struct btrfs_path *path;
1708	struct btrfs_key key;
1709	struct btrfs_key found_key;
1710	struct extent_buffer *leaf = NULL;
1711	struct btrfs_dev_extent *extent = NULL;
1712
1713	path = btrfs_alloc_path();
1714	if (!path)
1715		return -ENOMEM;
1716
1717	key.objectid = device->devid;
1718	key.offset = start;
1719	key.type = BTRFS_DEV_EXTENT_KEY;
1720again:
1721	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1722	if (ret > 0) {
1723		ret = btrfs_previous_item(root, path, key.objectid,
1724					  BTRFS_DEV_EXTENT_KEY);
1725		if (ret)
1726			goto out;
1727		leaf = path->nodes[0];
1728		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1729		extent = btrfs_item_ptr(leaf, path->slots[0],
1730					struct btrfs_dev_extent);
1731		BUG_ON(found_key.offset > start || found_key.offset +
1732		       btrfs_dev_extent_length(leaf, extent) < start);
1733		key = found_key;
1734		btrfs_release_path(path);
1735		goto again;
1736	} else if (ret == 0) {
1737		leaf = path->nodes[0];
1738		extent = btrfs_item_ptr(leaf, path->slots[0],
1739					struct btrfs_dev_extent);
1740	} else {
1741		goto out;
1742	}
1743
1744	*dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1745
1746	ret = btrfs_del_item(trans, root, path);
1747	if (ret == 0)
1748		set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1749out:
1750	btrfs_free_path(path);
1751	return ret;
1752}
1753
1754static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1755{
1756	struct rb_node *n;
1757	u64 ret = 0;
1758
1759	read_lock(&fs_info->mapping_tree_lock);
1760	n = rb_last(&fs_info->mapping_tree.rb_root);
1761	if (n) {
1762		struct btrfs_chunk_map *map;
1763
1764		map = rb_entry(n, struct btrfs_chunk_map, rb_node);
1765		ret = map->start + map->chunk_len;
1766	}
1767	read_unlock(&fs_info->mapping_tree_lock);
1768
1769	return ret;
1770}
1771
1772static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1773				    u64 *devid_ret)
1774{
1775	int ret;
1776	struct btrfs_key key;
1777	struct btrfs_key found_key;
1778	struct btrfs_path *path;
1779
1780	path = btrfs_alloc_path();
1781	if (!path)
1782		return -ENOMEM;
1783
1784	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1785	key.type = BTRFS_DEV_ITEM_KEY;
1786	key.offset = (u64)-1;
1787
1788	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1789	if (ret < 0)
1790		goto error;
1791
1792	if (ret == 0) {
1793		/* Corruption */
1794		btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1795		ret = -EUCLEAN;
1796		goto error;
1797	}
1798
1799	ret = btrfs_previous_item(fs_info->chunk_root, path,
1800				  BTRFS_DEV_ITEMS_OBJECTID,
1801				  BTRFS_DEV_ITEM_KEY);
1802	if (ret) {
1803		*devid_ret = 1;
1804	} else {
1805		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1806				      path->slots[0]);
1807		*devid_ret = found_key.offset + 1;
1808	}
1809	ret = 0;
1810error:
1811	btrfs_free_path(path);
1812	return ret;
1813}
1814
1815/*
1816 * the device information is stored in the chunk root
1817 * the btrfs_device struct should be fully filled in
1818 */
1819static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1820			    struct btrfs_device *device)
1821{
1822	int ret;
1823	struct btrfs_path *path;
1824	struct btrfs_dev_item *dev_item;
1825	struct extent_buffer *leaf;
1826	struct btrfs_key key;
1827	unsigned long ptr;
1828
1829	path = btrfs_alloc_path();
1830	if (!path)
1831		return -ENOMEM;
1832
1833	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1834	key.type = BTRFS_DEV_ITEM_KEY;
1835	key.offset = device->devid;
1836
1837	btrfs_reserve_chunk_metadata(trans, true);
1838	ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1839				      &key, sizeof(*dev_item));
1840	btrfs_trans_release_chunk_metadata(trans);
1841	if (ret)
1842		goto out;
1843
1844	leaf = path->nodes[0];
1845	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1846
1847	btrfs_set_device_id(leaf, dev_item, device->devid);
1848	btrfs_set_device_generation(leaf, dev_item, 0);
1849	btrfs_set_device_type(leaf, dev_item, device->type);
1850	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1851	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1852	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1853	btrfs_set_device_total_bytes(leaf, dev_item,
1854				     btrfs_device_get_disk_total_bytes(device));
1855	btrfs_set_device_bytes_used(leaf, dev_item,
1856				    btrfs_device_get_bytes_used(device));
1857	btrfs_set_device_group(leaf, dev_item, 0);
1858	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1859	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1860	btrfs_set_device_start_offset(leaf, dev_item, 0);
1861
1862	ptr = btrfs_device_uuid(dev_item);
1863	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1864	ptr = btrfs_device_fsid(dev_item);
1865	write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1866			    ptr, BTRFS_FSID_SIZE);
1867	btrfs_mark_buffer_dirty(trans, leaf);
1868
1869	ret = 0;
1870out:
1871	btrfs_free_path(path);
1872	return ret;
1873}
1874
1875/*
1876 * Function to update ctime/mtime for a given device path.
1877 * Mainly used for ctime/mtime based probe like libblkid.
1878 *
1879 * We don't care about errors here, this is just to be kind to userspace.
1880 */
1881static void update_dev_time(const char *device_path)
1882{
1883	struct path path;
1884	int ret;
1885
1886	ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1887	if (ret)
1888		return;
1889
1890	inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);
1891	path_put(&path);
1892}
1893
1894static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1895			     struct btrfs_device *device)
1896{
1897	struct btrfs_root *root = device->fs_info->chunk_root;
1898	int ret;
1899	struct btrfs_path *path;
1900	struct btrfs_key key;
1901
1902	path = btrfs_alloc_path();
1903	if (!path)
1904		return -ENOMEM;
1905
1906	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1907	key.type = BTRFS_DEV_ITEM_KEY;
1908	key.offset = device->devid;
1909
1910	btrfs_reserve_chunk_metadata(trans, false);
1911	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1912	btrfs_trans_release_chunk_metadata(trans);
1913	if (ret) {
1914		if (ret > 0)
1915			ret = -ENOENT;
1916		goto out;
1917	}
1918
1919	ret = btrfs_del_item(trans, root, path);
1920out:
1921	btrfs_free_path(path);
1922	return ret;
1923}
1924
1925/*
1926 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1927 * filesystem. It's up to the caller to adjust that number regarding eg. device
1928 * replace.
1929 */
1930static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1931		u64 num_devices)
1932{
1933	u64 all_avail;
1934	unsigned seq;
1935	int i;
1936
1937	do {
1938		seq = read_seqbegin(&fs_info->profiles_lock);
1939
1940		all_avail = fs_info->avail_data_alloc_bits |
1941			    fs_info->avail_system_alloc_bits |
1942			    fs_info->avail_metadata_alloc_bits;
1943	} while (read_seqretry(&fs_info->profiles_lock, seq));
1944
1945	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1946		if (!(all_avail & btrfs_raid_array[i].bg_flag))
1947			continue;
1948
1949		if (num_devices < btrfs_raid_array[i].devs_min)
1950			return btrfs_raid_array[i].mindev_error;
1951	}
1952
1953	return 0;
1954}
1955
1956static struct btrfs_device * btrfs_find_next_active_device(
1957		struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1958{
1959	struct btrfs_device *next_device;
1960
1961	list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1962		if (next_device != device &&
1963		    !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1964		    && next_device->bdev)
1965			return next_device;
1966	}
1967
1968	return NULL;
1969}
1970
1971/*
1972 * Helper function to check if the given device is part of s_bdev / latest_dev
1973 * and replace it with the provided or the next active device, in the context
1974 * where this function called, there should be always be another device (or
1975 * this_dev) which is active.
1976 */
1977void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1978					    struct btrfs_device *next_device)
1979{
1980	struct btrfs_fs_info *fs_info = device->fs_info;
1981
1982	if (!next_device)
1983		next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1984							    device);
1985	ASSERT(next_device);
1986
1987	if (fs_info->sb->s_bdev &&
1988			(fs_info->sb->s_bdev == device->bdev))
1989		fs_info->sb->s_bdev = next_device->bdev;
1990
1991	if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
1992		fs_info->fs_devices->latest_dev = next_device;
1993}
1994
1995/*
1996 * Return btrfs_fs_devices::num_devices excluding the device that's being
1997 * currently replaced.
1998 */
1999static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2000{
2001	u64 num_devices = fs_info->fs_devices->num_devices;
2002
2003	down_read(&fs_info->dev_replace.rwsem);
2004	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2005		ASSERT(num_devices > 1);
2006		num_devices--;
2007	}
2008	up_read(&fs_info->dev_replace.rwsem);
2009
2010	return num_devices;
2011}
2012
2013static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2014				     struct block_device *bdev, int copy_num)
2015{
2016	struct btrfs_super_block *disk_super;
2017	const size_t len = sizeof(disk_super->magic);
2018	const u64 bytenr = btrfs_sb_offset(copy_num);
2019	int ret;
2020
2021	disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2022	if (IS_ERR(disk_super))
2023		return;
2024
2025	memset(&disk_super->magic, 0, len);
2026	folio_mark_dirty(virt_to_folio(disk_super));
2027	btrfs_release_disk_super(disk_super);
2028
2029	ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2030	if (ret)
2031		btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2032			copy_num, ret);
2033}
2034
2035void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2036			       struct block_device *bdev,
2037			       const char *device_path)
2038{
2039	int copy_num;
2040
2041	if (!bdev)
2042		return;
2043
2044	for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2045		if (bdev_is_zoned(bdev))
2046			btrfs_reset_sb_log_zones(bdev, copy_num);
2047		else
2048			btrfs_scratch_superblock(fs_info, bdev, copy_num);
2049	}
2050
2051	/* Notify udev that device has changed */
2052	btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2053
2054	/* Update ctime/mtime for device path for libblkid */
2055	update_dev_time(device_path);
2056}
2057
2058int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2059		    struct btrfs_dev_lookup_args *args,
2060		    struct bdev_handle **bdev_handle)
2061{
2062	struct btrfs_trans_handle *trans;
2063	struct btrfs_device *device;
2064	struct btrfs_fs_devices *cur_devices;
2065	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2066	u64 num_devices;
2067	int ret = 0;
2068
2069	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2070		btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2071		return -EINVAL;
2072	}
2073
2074	/*
2075	 * The device list in fs_devices is accessed without locks (neither
2076	 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2077	 * filesystem and another device rm cannot run.
2078	 */
2079	num_devices = btrfs_num_devices(fs_info);
2080
2081	ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2082	if (ret)
2083		return ret;
2084
2085	device = btrfs_find_device(fs_info->fs_devices, args);
2086	if (!device) {
2087		if (args->missing)
2088			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2089		else
2090			ret = -ENOENT;
2091		return ret;
2092	}
2093
2094	if (btrfs_pinned_by_swapfile(fs_info, device)) {
2095		btrfs_warn_in_rcu(fs_info,
2096		  "cannot remove device %s (devid %llu) due to active swapfile",
2097				  btrfs_dev_name(device), device->devid);
2098		return -ETXTBSY;
2099	}
2100
2101	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2102		return BTRFS_ERROR_DEV_TGT_REPLACE;
2103
2104	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2105	    fs_info->fs_devices->rw_devices == 1)
2106		return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2107
2108	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2109		mutex_lock(&fs_info->chunk_mutex);
2110		list_del_init(&device->dev_alloc_list);
2111		device->fs_devices->rw_devices--;
2112		mutex_unlock(&fs_info->chunk_mutex);
2113	}
2114
2115	ret = btrfs_shrink_device(device, 0);
2116	if (ret)
2117		goto error_undo;
2118
2119	trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2120	if (IS_ERR(trans)) {
2121		ret = PTR_ERR(trans);
2122		goto error_undo;
2123	}
2124
2125	ret = btrfs_rm_dev_item(trans, device);
2126	if (ret) {
2127		/* Any error in dev item removal is critical */
2128		btrfs_crit(fs_info,
2129			   "failed to remove device item for devid %llu: %d",
2130			   device->devid, ret);
2131		btrfs_abort_transaction(trans, ret);
2132		btrfs_end_transaction(trans);
2133		return ret;
2134	}
2135
2136	clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2137	btrfs_scrub_cancel_dev(device);
2138
2139	/*
2140	 * the device list mutex makes sure that we don't change
2141	 * the device list while someone else is writing out all
2142	 * the device supers. Whoever is writing all supers, should
2143	 * lock the device list mutex before getting the number of
2144	 * devices in the super block (super_copy). Conversely,
2145	 * whoever updates the number of devices in the super block
2146	 * (super_copy) should hold the device list mutex.
2147	 */
2148
2149	/*
2150	 * In normal cases the cur_devices == fs_devices. But in case
2151	 * of deleting a seed device, the cur_devices should point to
2152	 * its own fs_devices listed under the fs_devices->seed_list.
2153	 */
2154	cur_devices = device->fs_devices;
2155	mutex_lock(&fs_devices->device_list_mutex);
2156	list_del_rcu(&device->dev_list);
2157
2158	cur_devices->num_devices--;
2159	cur_devices->total_devices--;
2160	/* Update total_devices of the parent fs_devices if it's seed */
2161	if (cur_devices != fs_devices)
2162		fs_devices->total_devices--;
2163
2164	if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2165		cur_devices->missing_devices--;
2166
2167	btrfs_assign_next_active_device(device, NULL);
2168
2169	if (device->bdev_handle) {
2170		cur_devices->open_devices--;
2171		/* remove sysfs entry */
2172		btrfs_sysfs_remove_device(device);
2173	}
2174
2175	num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2176	btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2177	mutex_unlock(&fs_devices->device_list_mutex);
2178
2179	/*
2180	 * At this point, the device is zero sized and detached from the
2181	 * devices list.  All that's left is to zero out the old supers and
2182	 * free the device.
2183	 *
2184	 * We cannot call btrfs_close_bdev() here because we're holding the sb
2185	 * write lock, and bdev_release() will pull in the ->open_mutex on
2186	 * the block device and it's dependencies.  Instead just flush the
2187	 * device and let the caller do the final bdev_release.
2188	 */
2189	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2190		btrfs_scratch_superblocks(fs_info, device->bdev,
2191					  device->name->str);
2192		if (device->bdev) {
2193			sync_blockdev(device->bdev);
2194			invalidate_bdev(device->bdev);
2195		}
2196	}
2197
2198	*bdev_handle = device->bdev_handle;
2199	synchronize_rcu();
2200	btrfs_free_device(device);
2201
2202	/*
2203	 * This can happen if cur_devices is the private seed devices list.  We
2204	 * cannot call close_fs_devices() here because it expects the uuid_mutex
2205	 * to be held, but in fact we don't need that for the private
2206	 * seed_devices, we can simply decrement cur_devices->opened and then
2207	 * remove it from our list and free the fs_devices.
2208	 */
2209	if (cur_devices->num_devices == 0) {
2210		list_del_init(&cur_devices->seed_list);
2211		ASSERT(cur_devices->opened == 1);
2212		cur_devices->opened--;
2213		free_fs_devices(cur_devices);
2214	}
2215
2216	ret = btrfs_commit_transaction(trans);
2217
2218	return ret;
2219
2220error_undo:
2221	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2222		mutex_lock(&fs_info->chunk_mutex);
2223		list_add(&device->dev_alloc_list,
2224			 &fs_devices->alloc_list);
2225		device->fs_devices->rw_devices++;
2226		mutex_unlock(&fs_info->chunk_mutex);
2227	}
2228	return ret;
2229}
2230
2231void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2232{
2233	struct btrfs_fs_devices *fs_devices;
2234
2235	lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2236
2237	/*
2238	 * in case of fs with no seed, srcdev->fs_devices will point
2239	 * to fs_devices of fs_info. However when the dev being replaced is
2240	 * a seed dev it will point to the seed's local fs_devices. In short
2241	 * srcdev will have its correct fs_devices in both the cases.
2242	 */
2243	fs_devices = srcdev->fs_devices;
2244
2245	list_del_rcu(&srcdev->dev_list);
2246	list_del(&srcdev->dev_alloc_list);
2247	fs_devices->num_devices--;
2248	if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2249		fs_devices->missing_devices--;
2250
2251	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2252		fs_devices->rw_devices--;
2253
2254	if (srcdev->bdev)
2255		fs_devices->open_devices--;
2256}
2257
2258void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2259{
2260	struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2261
2262	mutex_lock(&uuid_mutex);
2263
2264	btrfs_close_bdev(srcdev);
2265	synchronize_rcu();
2266	btrfs_free_device(srcdev);
2267
2268	/* if this is no devs we rather delete the fs_devices */
2269	if (!fs_devices->num_devices) {
2270		/*
2271		 * On a mounted FS, num_devices can't be zero unless it's a
2272		 * seed. In case of a seed device being replaced, the replace
2273		 * target added to the sprout FS, so there will be no more
2274		 * device left under the seed FS.
2275		 */
2276		ASSERT(fs_devices->seeding);
2277
2278		list_del_init(&fs_devices->seed_list);
2279		close_fs_devices(fs_devices);
2280		free_fs_devices(fs_devices);
2281	}
2282	mutex_unlock(&uuid_mutex);
2283}
2284
2285void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2286{
2287	struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2288
2289	mutex_lock(&fs_devices->device_list_mutex);
2290
2291	btrfs_sysfs_remove_device(tgtdev);
2292
2293	if (tgtdev->bdev)
2294		fs_devices->open_devices--;
2295
2296	fs_devices->num_devices--;
2297
2298	btrfs_assign_next_active_device(tgtdev, NULL);
2299
2300	list_del_rcu(&tgtdev->dev_list);
2301
2302	mutex_unlock(&fs_devices->device_list_mutex);
2303
2304	btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2305				  tgtdev->name->str);
2306
2307	btrfs_close_bdev(tgtdev);
2308	synchronize_rcu();
2309	btrfs_free_device(tgtdev);
2310}
2311
2312/*
2313 * Populate args from device at path.
2314 *
2315 * @fs_info:	the filesystem
2316 * @args:	the args to populate
2317 * @path:	the path to the device
2318 *
2319 * This will read the super block of the device at @path and populate @args with
2320 * the devid, fsid, and uuid.  This is meant to be used for ioctls that need to
2321 * lookup a device to operate on, but need to do it before we take any locks.
2322 * This properly handles the special case of "missing" that a user may pass in,
2323 * and does some basic sanity checks.  The caller must make sure that @path is
2324 * properly NUL terminated before calling in, and must call
2325 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2326 * uuid buffers.
2327 *
2328 * Return: 0 for success, -errno for failure
2329 */
2330int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2331				 struct btrfs_dev_lookup_args *args,
2332				 const char *path)
2333{
2334	struct btrfs_super_block *disk_super;
2335	struct bdev_handle *bdev_handle;
2336	int ret;
2337
2338	if (!path || !path[0])
2339		return -EINVAL;
2340	if (!strcmp(path, "missing")) {
2341		args->missing = true;
2342		return 0;
2343	}
2344
2345	args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2346	args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2347	if (!args->uuid || !args->fsid) {
2348		btrfs_put_dev_args_from_path(args);
2349		return -ENOMEM;
2350	}
2351
2352	ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2353				    &bdev_handle, &disk_super);
2354	if (ret) {
2355		btrfs_put_dev_args_from_path(args);
2356		return ret;
2357	}
2358
2359	args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2360	memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2361	if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2362		memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2363	else
2364		memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2365	btrfs_release_disk_super(disk_super);
2366	bdev_release(bdev_handle);
2367	return 0;
2368}
2369
2370/*
2371 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2372 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2373 * that don't need to be freed.
2374 */
2375void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2376{
2377	kfree(args->uuid);
2378	kfree(args->fsid);
2379	args->uuid = NULL;
2380	args->fsid = NULL;
2381}
2382
2383struct btrfs_device *btrfs_find_device_by_devspec(
2384		struct btrfs_fs_info *fs_info, u64 devid,
2385		const char *device_path)
2386{
2387	BTRFS_DEV_LOOKUP_ARGS(args);
2388	struct btrfs_device *device;
2389	int ret;
2390
2391	if (devid) {
2392		args.devid = devid;
2393		device = btrfs_find_device(fs_info->fs_devices, &args);
2394		if (!device)
2395			return ERR_PTR(-ENOENT);
2396		return device;
2397	}
2398
2399	ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2400	if (ret)
2401		return ERR_PTR(ret);
2402	device = btrfs_find_device(fs_info->fs_devices, &args);
2403	btrfs_put_dev_args_from_path(&args);
2404	if (!device)
2405		return ERR_PTR(-ENOENT);
2406	return device;
2407}
2408
2409static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2410{
2411	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2412	struct btrfs_fs_devices *old_devices;
2413	struct btrfs_fs_devices *seed_devices;
2414
2415	lockdep_assert_held(&uuid_mutex);
2416	if (!fs_devices->seeding)
2417		return ERR_PTR(-EINVAL);
2418
2419	/*
2420	 * Private copy of the seed devices, anchored at
2421	 * fs_info->fs_devices->seed_list
2422	 */
2423	seed_devices = alloc_fs_devices(NULL);
2424	if (IS_ERR(seed_devices))
2425		return seed_devices;
2426
2427	/*
2428	 * It's necessary to retain a copy of the original seed fs_devices in
2429	 * fs_uuids so that filesystems which have been seeded can successfully
2430	 * reference the seed device from open_seed_devices. This also supports
2431	 * multiple fs seed.
2432	 */
2433	old_devices = clone_fs_devices(fs_devices);
2434	if (IS_ERR(old_devices)) {
2435		kfree(seed_devices);
2436		return old_devices;
2437	}
2438
2439	list_add(&old_devices->fs_list, &fs_uuids);
2440
2441	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2442	seed_devices->opened = 1;
2443	INIT_LIST_HEAD(&seed_devices->devices);
2444	INIT_LIST_HEAD(&seed_devices->alloc_list);
2445	mutex_init(&seed_devices->device_list_mutex);
2446
2447	return seed_devices;
2448}
2449
2450/*
2451 * Splice seed devices into the sprout fs_devices.
2452 * Generate a new fsid for the sprouted read-write filesystem.
2453 */
2454static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2455			       struct btrfs_fs_devices *seed_devices)
2456{
2457	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2458	struct btrfs_super_block *disk_super = fs_info->super_copy;
2459	struct btrfs_device *device;
2460	u64 super_flags;
2461
2462	/*
2463	 * We are updating the fsid, the thread leading to device_list_add()
2464	 * could race, so uuid_mutex is needed.
2465	 */
2466	lockdep_assert_held(&uuid_mutex);
2467
2468	/*
2469	 * The threads listed below may traverse dev_list but can do that without
2470	 * device_list_mutex:
2471	 * - All device ops and balance - as we are in btrfs_exclop_start.
2472	 * - Various dev_list readers - are using RCU.
2473	 * - btrfs_ioctl_fitrim() - is using RCU.
2474	 *
2475	 * For-read threads as below are using device_list_mutex:
2476	 * - Readonly scrub btrfs_scrub_dev()
2477	 * - Readonly scrub btrfs_scrub_progress()
2478	 * - btrfs_get_dev_stats()
2479	 */
2480	lockdep_assert_held(&fs_devices->device_list_mutex);
2481
2482	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2483			      synchronize_rcu);
2484	list_for_each_entry(device, &seed_devices->devices, dev_list)
2485		device->fs_devices = seed_devices;
2486
2487	fs_devices->seeding = false;
2488	fs_devices->num_devices = 0;
2489	fs_devices->open_devices = 0;
2490	fs_devices->missing_devices = 0;
2491	fs_devices->rotating = false;
2492	list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2493
2494	generate_random_uuid(fs_devices->fsid);
2495	memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2496	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2497
2498	super_flags = btrfs_super_flags(disk_super) &
2499		      ~BTRFS_SUPER_FLAG_SEEDING;
2500	btrfs_set_super_flags(disk_super, super_flags);
2501}
2502
2503/*
2504 * Store the expected generation for seed devices in device items.
2505 */
2506static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2507{
2508	BTRFS_DEV_LOOKUP_ARGS(args);
2509	struct btrfs_fs_info *fs_info = trans->fs_info;
2510	struct btrfs_root *root = fs_info->chunk_root;
2511	struct btrfs_path *path;
2512	struct extent_buffer *leaf;
2513	struct btrfs_dev_item *dev_item;
2514	struct btrfs_device *device;
2515	struct btrfs_key key;
2516	u8 fs_uuid[BTRFS_FSID_SIZE];
2517	u8 dev_uuid[BTRFS_UUID_SIZE];
2518	int ret;
2519
2520	path = btrfs_alloc_path();
2521	if (!path)
2522		return -ENOMEM;
2523
2524	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2525	key.offset = 0;
2526	key.type = BTRFS_DEV_ITEM_KEY;
2527
2528	while (1) {
2529		btrfs_reserve_chunk_metadata(trans, false);
2530		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2531		btrfs_trans_release_chunk_metadata(trans);
2532		if (ret < 0)
2533			goto error;
2534
2535		leaf = path->nodes[0];
2536next_slot:
2537		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2538			ret = btrfs_next_leaf(root, path);
2539			if (ret > 0)
2540				break;
2541			if (ret < 0)
2542				goto error;
2543			leaf = path->nodes[0];
2544			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2545			btrfs_release_path(path);
2546			continue;
2547		}
2548
2549		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2550		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2551		    key.type != BTRFS_DEV_ITEM_KEY)
2552			break;
2553
2554		dev_item = btrfs_item_ptr(leaf, path->slots[0],
2555					  struct btrfs_dev_item);
2556		args.devid = btrfs_device_id(leaf, dev_item);
2557		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2558				   BTRFS_UUID_SIZE);
2559		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2560				   BTRFS_FSID_SIZE);
2561		args.uuid = dev_uuid;
2562		args.fsid = fs_uuid;
2563		device = btrfs_find_device(fs_info->fs_devices, &args);
2564		BUG_ON(!device); /* Logic error */
2565
2566		if (device->fs_devices->seeding) {
2567			btrfs_set_device_generation(leaf, dev_item,
2568						    device->generation);
2569			btrfs_mark_buffer_dirty(trans, leaf);
2570		}
2571
2572		path->slots[0]++;
2573		goto next_slot;
2574	}
2575	ret = 0;
2576error:
2577	btrfs_free_path(path);
2578	return ret;
2579}
2580
2581int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2582{
2583	struct btrfs_root *root = fs_info->dev_root;
2584	struct btrfs_trans_handle *trans;
2585	struct btrfs_device *device;
2586	struct bdev_handle *bdev_handle;
2587	struct super_block *sb = fs_info->sb;
2588	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2589	struct btrfs_fs_devices *seed_devices = NULL;
2590	u64 orig_super_total_bytes;
2591	u64 orig_super_num_devices;
2592	int ret = 0;
2593	bool seeding_dev = false;
2594	bool locked = false;
2595
2596	if (sb_rdonly(sb) && !fs_devices->seeding)
2597		return -EROFS;
2598
2599	bdev_handle = bdev_open_by_path(device_path, BLK_OPEN_WRITE,
2600					fs_info->bdev_holder, NULL);
2601	if (IS_ERR(bdev_handle))
2602		return PTR_ERR(bdev_handle);
2603
2604	if (!btrfs_check_device_zone_type(fs_info, bdev_handle->bdev)) {
2605		ret = -EINVAL;
2606		goto error;
2607	}
2608
2609	if (fs_devices->seeding) {
2610		seeding_dev = true;
2611		down_write(&sb->s_umount);
2612		mutex_lock(&uuid_mutex);
2613		locked = true;
2614	}
2615
2616	sync_blockdev(bdev_handle->bdev);
2617
2618	rcu_read_lock();
2619	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2620		if (device->bdev == bdev_handle->bdev) {
2621			ret = -EEXIST;
2622			rcu_read_unlock();
2623			goto error;
2624		}
2625	}
2626	rcu_read_unlock();
2627
2628	device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2629	if (IS_ERR(device)) {
2630		/* we can safely leave the fs_devices entry around */
2631		ret = PTR_ERR(device);
2632		goto error;
2633	}
2634
2635	device->fs_info = fs_info;
2636	device->bdev_handle = bdev_handle;
2637	device->bdev = bdev_handle->bdev;
2638	ret = lookup_bdev(device_path, &device->devt);
2639	if (ret)
2640		goto error_free_device;
2641
2642	ret = btrfs_get_dev_zone_info(device, false);
2643	if (ret)
2644		goto error_free_device;
2645
2646	trans = btrfs_start_transaction(root, 0);
2647	if (IS_ERR(trans)) {
2648		ret = PTR_ERR(trans);
2649		goto error_free_zone;
2650	}
2651
2652	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2653	device->generation = trans->transid;
2654	device->io_width = fs_info->sectorsize;
2655	device->io_align = fs_info->sectorsize;
2656	device->sector_size = fs_info->sectorsize;
2657	device->total_bytes =
2658		round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize);
2659	device->disk_total_bytes = device->total_bytes;
2660	device->commit_total_bytes = device->total_bytes;
2661	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2662	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2663	device->dev_stats_valid = 1;
2664	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2665
2666	if (seeding_dev) {
2667		btrfs_clear_sb_rdonly(sb);
2668
2669		/* GFP_KERNEL allocation must not be under device_list_mutex */
2670		seed_devices = btrfs_init_sprout(fs_info);
2671		if (IS_ERR(seed_devices)) {
2672			ret = PTR_ERR(seed_devices);
2673			btrfs_abort_transaction(trans, ret);
2674			goto error_trans;
2675		}
2676	}
2677
2678	mutex_lock(&fs_devices->device_list_mutex);
2679	if (seeding_dev) {
2680		btrfs_setup_sprout(fs_info, seed_devices);
2681		btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2682						device);
2683	}
2684
2685	device->fs_devices = fs_devices;
2686
2687	mutex_lock(&fs_info->chunk_mutex);
2688	list_add_rcu(&device->dev_list, &fs_devices->devices);
2689	list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2690	fs_devices->num_devices++;
2691	fs_devices->open_devices++;
2692	fs_devices->rw_devices++;
2693	fs_devices->total_devices++;
2694	fs_devices->total_rw_bytes += device->total_bytes;
2695
2696	atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2697
2698	if (!bdev_nonrot(device->bdev))
2699		fs_devices->rotating = true;
2700
2701	orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2702	btrfs_set_super_total_bytes(fs_info->super_copy,
2703		round_down(orig_super_total_bytes + device->total_bytes,
2704			   fs_info->sectorsize));
2705
2706	orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2707	btrfs_set_super_num_devices(fs_info->super_copy,
2708				    orig_super_num_devices + 1);
2709
2710	/*
2711	 * we've got more storage, clear any full flags on the space
2712	 * infos
2713	 */
2714	btrfs_clear_space_info_full(fs_info);
2715
2716	mutex_unlock(&fs_info->chunk_mutex);
2717
2718	/* Add sysfs device entry */
2719	btrfs_sysfs_add_device(device);
2720
2721	mutex_unlock(&fs_devices->device_list_mutex);
2722
2723	if (seeding_dev) {
2724		mutex_lock(&fs_info->chunk_mutex);
2725		ret = init_first_rw_device(trans);
2726		mutex_unlock(&fs_info->chunk_mutex);
2727		if (ret) {
2728			btrfs_abort_transaction(trans, ret);
2729			goto error_sysfs;
2730		}
2731	}
2732
2733	ret = btrfs_add_dev_item(trans, device);
2734	if (ret) {
2735		btrfs_abort_transaction(trans, ret);
2736		goto error_sysfs;
2737	}
2738
2739	if (seeding_dev) {
2740		ret = btrfs_finish_sprout(trans);
2741		if (ret) {
2742			btrfs_abort_transaction(trans, ret);
2743			goto error_sysfs;
2744		}
2745
2746		/*
2747		 * fs_devices now represents the newly sprouted filesystem and
2748		 * its fsid has been changed by btrfs_sprout_splice().
2749		 */
2750		btrfs_sysfs_update_sprout_fsid(fs_devices);
2751	}
2752
2753	ret = btrfs_commit_transaction(trans);
2754
2755	if (seeding_dev) {
2756		mutex_unlock(&uuid_mutex);
2757		up_write(&sb->s_umount);
2758		locked = false;
2759
2760		if (ret) /* transaction commit */
2761			return ret;
2762
2763		ret = btrfs_relocate_sys_chunks(fs_info);
2764		if (ret < 0)
2765			btrfs_handle_fs_error(fs_info, ret,
2766				    "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2767		trans = btrfs_attach_transaction(root);
2768		if (IS_ERR(trans)) {
2769			if (PTR_ERR(trans) == -ENOENT)
2770				return 0;
2771			ret = PTR_ERR(trans);
2772			trans = NULL;
2773			goto error_sysfs;
2774		}
2775		ret = btrfs_commit_transaction(trans);
2776	}
2777
2778	/*
2779	 * Now that we have written a new super block to this device, check all
2780	 * other fs_devices list if device_path alienates any other scanned
2781	 * device.
2782	 * We can ignore the return value as it typically returns -EINVAL and
2783	 * only succeeds if the device was an alien.
2784	 */
2785	btrfs_forget_devices(device->devt);
2786
2787	/* Update ctime/mtime for blkid or udev */
2788	update_dev_time(device_path);
2789
2790	return ret;
2791
2792error_sysfs:
2793	btrfs_sysfs_remove_device(device);
2794	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2795	mutex_lock(&fs_info->chunk_mutex);
2796	list_del_rcu(&device->dev_list);
2797	list_del(&device->dev_alloc_list);
2798	fs_info->fs_devices->num_devices--;
2799	fs_info->fs_devices->open_devices--;
2800	fs_info->fs_devices->rw_devices--;
2801	fs_info->fs_devices->total_devices--;
2802	fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2803	atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2804	btrfs_set_super_total_bytes(fs_info->super_copy,
2805				    orig_super_total_bytes);
2806	btrfs_set_super_num_devices(fs_info->super_copy,
2807				    orig_super_num_devices);
2808	mutex_unlock(&fs_info->chunk_mutex);
2809	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2810error_trans:
2811	if (seeding_dev)
2812		btrfs_set_sb_rdonly(sb);
2813	if (trans)
2814		btrfs_end_transaction(trans);
2815error_free_zone:
2816	btrfs_destroy_dev_zone_info(device);
2817error_free_device:
2818	btrfs_free_device(device);
2819error:
2820	bdev_release(bdev_handle);
2821	if (locked) {
2822		mutex_unlock(&uuid_mutex);
2823		up_write(&sb->s_umount);
2824	}
2825	return ret;
2826}
2827
2828static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2829					struct btrfs_device *device)
2830{
2831	int ret;
2832	struct btrfs_path *path;
2833	struct btrfs_root *root = device->fs_info->chunk_root;
2834	struct btrfs_dev_item *dev_item;
2835	struct extent_buffer *leaf;
2836	struct btrfs_key key;
2837
2838	path = btrfs_alloc_path();
2839	if (!path)
2840		return -ENOMEM;
2841
2842	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2843	key.type = BTRFS_DEV_ITEM_KEY;
2844	key.offset = device->devid;
2845
2846	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2847	if (ret < 0)
2848		goto out;
2849
2850	if (ret > 0) {
2851		ret = -ENOENT;
2852		goto out;
2853	}
2854
2855	leaf = path->nodes[0];
2856	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2857
2858	btrfs_set_device_id(leaf, dev_item, device->devid);
2859	btrfs_set_device_type(leaf, dev_item, device->type);
2860	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2861	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2862	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2863	btrfs_set_device_total_bytes(leaf, dev_item,
2864				     btrfs_device_get_disk_total_bytes(device));
2865	btrfs_set_device_bytes_used(leaf, dev_item,
2866				    btrfs_device_get_bytes_used(device));
2867	btrfs_mark_buffer_dirty(trans, leaf);
2868
2869out:
2870	btrfs_free_path(path);
2871	return ret;
2872}
2873
2874int btrfs_grow_device(struct btrfs_trans_handle *trans,
2875		      struct btrfs_device *device, u64 new_size)
2876{
2877	struct btrfs_fs_info *fs_info = device->fs_info;
2878	struct btrfs_super_block *super_copy = fs_info->super_copy;
2879	u64 old_total;
2880	u64 diff;
2881	int ret;
2882
2883	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2884		return -EACCES;
2885
2886	new_size = round_down(new_size, fs_info->sectorsize);
2887
2888	mutex_lock(&fs_info->chunk_mutex);
2889	old_total = btrfs_super_total_bytes(super_copy);
2890	diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2891
2892	if (new_size <= device->total_bytes ||
2893	    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2894		mutex_unlock(&fs_info->chunk_mutex);
2895		return -EINVAL;
2896	}
2897
2898	btrfs_set_super_total_bytes(super_copy,
2899			round_down(old_total + diff, fs_info->sectorsize));
2900	device->fs_devices->total_rw_bytes += diff;
2901	atomic64_add(diff, &fs_info->free_chunk_space);
2902
2903	btrfs_device_set_total_bytes(device, new_size);
2904	btrfs_device_set_disk_total_bytes(device, new_size);
2905	btrfs_clear_space_info_full(device->fs_info);
2906	if (list_empty(&device->post_commit_list))
2907		list_add_tail(&device->post_commit_list,
2908			      &trans->transaction->dev_update_list);
2909	mutex_unlock(&fs_info->chunk_mutex);
2910
2911	btrfs_reserve_chunk_metadata(trans, false);
2912	ret = btrfs_update_device(trans, device);
2913	btrfs_trans_release_chunk_metadata(trans);
2914
2915	return ret;
2916}
2917
2918static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2919{
2920	struct btrfs_fs_info *fs_info = trans->fs_info;
2921	struct btrfs_root *root = fs_info->chunk_root;
2922	int ret;
2923	struct btrfs_path *path;
2924	struct btrfs_key key;
2925
2926	path = btrfs_alloc_path();
2927	if (!path)
2928		return -ENOMEM;
2929
2930	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2931	key.offset = chunk_offset;
2932	key.type = BTRFS_CHUNK_ITEM_KEY;
2933
2934	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2935	if (ret < 0)
2936		goto out;
2937	else if (ret > 0) { /* Logic error or corruption */
2938		btrfs_handle_fs_error(fs_info, -ENOENT,
2939				      "Failed lookup while freeing chunk.");
2940		ret = -ENOENT;
2941		goto out;
2942	}
2943
2944	ret = btrfs_del_item(trans, root, path);
2945	if (ret < 0)
2946		btrfs_handle_fs_error(fs_info, ret,
2947				      "Failed to delete chunk item.");
2948out:
2949	btrfs_free_path(path);
2950	return ret;
2951}
2952
2953static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2954{
2955	struct btrfs_super_block *super_copy = fs_info->super_copy;
2956	struct btrfs_disk_key *disk_key;
2957	struct btrfs_chunk *chunk;
2958	u8 *ptr;
2959	int ret = 0;
2960	u32 num_stripes;
2961	u32 array_size;
2962	u32 len = 0;
2963	u32 cur;
2964	struct btrfs_key key;
2965
2966	lockdep_assert_held(&fs_info->chunk_mutex);
2967	array_size = btrfs_super_sys_array_size(super_copy);
2968
2969	ptr = super_copy->sys_chunk_array;
2970	cur = 0;
2971
2972	while (cur < array_size) {
2973		disk_key = (struct btrfs_disk_key *)ptr;
2974		btrfs_disk_key_to_cpu(&key, disk_key);
2975
2976		len = sizeof(*disk_key);
2977
2978		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2979			chunk = (struct btrfs_chunk *)(ptr + len);
2980			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2981			len += btrfs_chunk_item_size(num_stripes);
2982		} else {
2983			ret = -EIO;
2984			break;
2985		}
2986		if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2987		    key.offset == chunk_offset) {
2988			memmove(ptr, ptr + len, array_size - (cur + len));
2989			array_size -= len;
2990			btrfs_set_super_sys_array_size(super_copy, array_size);
2991		} else {
2992			ptr += len;
2993			cur += len;
2994		}
2995	}
2996	return ret;
2997}
2998
2999struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
3000						    u64 logical, u64 length)
3001{
3002	struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node;
3003	struct rb_node *prev = NULL;
3004	struct rb_node *orig_prev;
3005	struct btrfs_chunk_map *map;
3006	struct btrfs_chunk_map *prev_map = NULL;
3007
3008	while (node) {
3009		map = rb_entry(node, struct btrfs_chunk_map, rb_node);
3010		prev = node;
3011		prev_map = map;
3012
3013		if (logical < map->start) {
3014			node = node->rb_left;
3015		} else if (logical >= map->start + map->chunk_len) {
3016			node = node->rb_right;
3017		} else {
3018			refcount_inc(&map->refs);
3019			return map;
3020		}
3021	}
3022
3023	if (!prev)
3024		return NULL;
3025
3026	orig_prev = prev;
3027	while (prev && logical >= prev_map->start + prev_map->chunk_len) {
3028		prev = rb_next(prev);
3029		prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3030	}
3031
3032	if (!prev) {
3033		prev = orig_prev;
3034		prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3035		while (prev && logical < prev_map->start) {
3036			prev = rb_prev(prev);
3037			prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3038		}
3039	}
3040
3041	if (prev) {
3042		u64 end = logical + length;
3043
3044		/*
3045		 * Caller can pass a U64_MAX length when it wants to get any
3046		 * chunk starting at an offset of 'logical' or higher, so deal
3047		 * with underflow by resetting the end offset to U64_MAX.
3048		 */
3049		if (end < logical)
3050			end = U64_MAX;
3051
3052		if (end > prev_map->start &&
3053		    logical < prev_map->start + prev_map->chunk_len) {
3054			refcount_inc(&prev_map->refs);
3055			return prev_map;
3056		}
3057	}
3058
3059	return NULL;
3060}
3061
3062struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
3063					     u64 logical, u64 length)
3064{
3065	struct btrfs_chunk_map *map;
3066
3067	read_lock(&fs_info->mapping_tree_lock);
3068	map = btrfs_find_chunk_map_nolock(fs_info, logical, length);
3069	read_unlock(&fs_info->mapping_tree_lock);
3070
3071	return map;
3072}
3073
3074/*
3075 * Find the mapping containing the given logical extent.
3076 *
3077 * @logical: Logical block offset in bytes.
3078 * @length: Length of extent in bytes.
3079 *
3080 * Return: Chunk mapping or ERR_PTR.
3081 */
3082struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3083					    u64 logical, u64 length)
3084{
3085	struct btrfs_chunk_map *map;
3086
3087	map = btrfs_find_chunk_map(fs_info, logical, length);
3088
3089	if (unlikely(!map)) {
3090		btrfs_crit(fs_info,
3091			   "unable to find chunk map for logical %llu length %llu",
3092			   logical, length);
3093		return ERR_PTR(-EINVAL);
3094	}
3095
3096	if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) {
3097		btrfs_crit(fs_info,
3098			   "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",
3099			   logical, logical + length, map->start,
3100			   map->start + map->chunk_len);
3101		btrfs_free_chunk_map(map);
3102		return ERR_PTR(-EINVAL);
3103	}
3104
3105	/* Callers are responsible for dropping the reference. */
3106	return map;
3107}
3108
3109static int remove_chunk_item(struct btrfs_trans_handle *trans,
3110			     struct btrfs_chunk_map *map, u64 chunk_offset)
3111{
3112	int i;
3113
3114	/*
3115	 * Removing chunk items and updating the device items in the chunks btree
3116	 * requires holding the chunk_mutex.
3117	 * See the comment at btrfs_chunk_alloc() for the details.
3118	 */
3119	lockdep_assert_held(&trans->fs_info->chunk_mutex);
3120
3121	for (i = 0; i < map->num_stripes; i++) {
3122		int ret;
3123
3124		ret = btrfs_update_device(trans, map->stripes[i].dev);
3125		if (ret)
3126			return ret;
3127	}
3128
3129	return btrfs_free_chunk(trans, chunk_offset);
3130}
3131
3132int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3133{
3134	struct btrfs_fs_info *fs_info = trans->fs_info;
3135	struct btrfs_chunk_map *map;
3136	u64 dev_extent_len = 0;
3137	int i, ret = 0;
3138	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3139
3140	map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3141	if (IS_ERR(map)) {
3142		/*
3143		 * This is a logic error, but we don't want to just rely on the
3144		 * user having built with ASSERT enabled, so if ASSERT doesn't
3145		 * do anything we still error out.
3146		 */
3147		ASSERT(0);
3148		return PTR_ERR(map);
3149	}
3150
3151	/*
3152	 * First delete the device extent items from the devices btree.
3153	 * We take the device_list_mutex to avoid racing with the finishing phase
3154	 * of a device replace operation. See the comment below before acquiring
3155	 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3156	 * because that can result in a deadlock when deleting the device extent
3157	 * items from the devices btree - COWing an extent buffer from the btree
3158	 * may result in allocating a new metadata chunk, which would attempt to
3159	 * lock again fs_info->chunk_mutex.
3160	 */
3161	mutex_lock(&fs_devices->device_list_mutex);
3162	for (i = 0; i < map->num_stripes; i++) {
3163		struct btrfs_device *device = map->stripes[i].dev;
3164		ret = btrfs_free_dev_extent(trans, device,
3165					    map->stripes[i].physical,
3166					    &dev_extent_len);
3167		if (ret) {
3168			mutex_unlock(&fs_devices->device_list_mutex);
3169			btrfs_abort_transaction(trans, ret);
3170			goto out;
3171		}
3172
3173		if (device->bytes_used > 0) {
3174			mutex_lock(&fs_info->chunk_mutex);
3175			btrfs_device_set_bytes_used(device,
3176					device->bytes_used - dev_extent_len);
3177			atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3178			btrfs_clear_space_info_full(fs_info);
3179			mutex_unlock(&fs_info->chunk_mutex);
3180		}
3181	}
3182	mutex_unlock(&fs_devices->device_list_mutex);
3183
3184	/*
3185	 * We acquire fs_info->chunk_mutex for 2 reasons:
3186	 *
3187	 * 1) Just like with the first phase of the chunk allocation, we must
3188	 *    reserve system space, do all chunk btree updates and deletions, and
3189	 *    update the system chunk array in the superblock while holding this
3190	 *    mutex. This is for similar reasons as explained on the comment at
3191	 *    the top of btrfs_chunk_alloc();
3192	 *
3193	 * 2) Prevent races with the final phase of a device replace operation
3194	 *    that replaces the device object associated with the map's stripes,
3195	 *    because the device object's id can change at any time during that
3196	 *    final phase of the device replace operation
3197	 *    (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3198	 *    replaced device and then see it with an ID of
3199	 *    BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3200	 *    the device item, which does not exists on the chunk btree.
3201	 *    The finishing phase of device replace acquires both the
3202	 *    device_list_mutex and the chunk_mutex, in that order, so we are
3203	 *    safe by just acquiring the chunk_mutex.
3204	 */
3205	trans->removing_chunk = true;
3206	mutex_lock(&fs_info->chunk_mutex);
3207
3208	check_system_chunk(trans, map->type);
3209
3210	ret = remove_chunk_item(trans, map, chunk_offset);
3211	/*
3212	 * Normally we should not get -ENOSPC since we reserved space before
3213	 * through the call to check_system_chunk().
3214	 *
3215	 * Despite our system space_info having enough free space, we may not
3216	 * be able to allocate extents from its block groups, because all have
3217	 * an incompatible profile, which will force us to allocate a new system
3218	 * block group with the right profile, or right after we called
3219	 * check_system_space() above, a scrub turned the only system block group
3220	 * with enough free space into RO mode.
3221	 * This is explained with more detail at do_chunk_alloc().
3222	 *
3223	 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3224	 */
3225	if (ret == -ENOSPC) {
3226		const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3227		struct btrfs_block_group *sys_bg;
3228
3229		sys_bg = btrfs_create_chunk(trans, sys_flags);
3230		if (IS_ERR(sys_bg)) {
3231			ret = PTR_ERR(sys_bg);
3232			btrfs_abort_transaction(trans, ret);
3233			goto out;
3234		}
3235
3236		ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3237		if (ret) {
3238			btrfs_abort_transaction(trans, ret);
3239			goto out;
3240		}
3241
3242		ret = remove_chunk_item(trans, map, chunk_offset);
3243		if (ret) {
3244			btrfs_abort_transaction(trans, ret);
3245			goto out;
3246		}
3247	} else if (ret) {
3248		btrfs_abort_transaction(trans, ret);
3249		goto out;
3250	}
3251
3252	trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len);
3253
3254	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3255		ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3256		if (ret) {
3257			btrfs_abort_transaction(trans, ret);
3258			goto out;
3259		}
3260	}
3261
3262	mutex_unlock(&fs_info->chunk_mutex);
3263	trans->removing_chunk = false;
3264
3265	/*
3266	 * We are done with chunk btree updates and deletions, so release the
3267	 * system space we previously reserved (with check_system_chunk()).
3268	 */
3269	btrfs_trans_release_chunk_metadata(trans);
3270
3271	ret = btrfs_remove_block_group(trans, map);
3272	if (ret) {
3273		btrfs_abort_transaction(trans, ret);
3274		goto out;
3275	}
3276
3277out:
3278	if (trans->removing_chunk) {
3279		mutex_unlock(&fs_info->chunk_mutex);
3280		trans->removing_chunk = false;
3281	}
3282	/* once for us */
3283	btrfs_free_chunk_map(map);
3284	return ret;
3285}
3286
3287int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3288{
3289	struct btrfs_root *root = fs_info->chunk_root;
3290	struct btrfs_trans_handle *trans;
3291	struct btrfs_block_group *block_group;
3292	u64 length;
3293	int ret;
3294
3295	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3296		btrfs_err(fs_info,
3297			  "relocate: not supported on extent tree v2 yet");
3298		return -EINVAL;
3299	}
3300
3301	/*
3302	 * Prevent races with automatic removal of unused block groups.
3303	 * After we relocate and before we remove the chunk with offset
3304	 * chunk_offset, automatic removal of the block group can kick in,
3305	 * resulting in a failure when calling btrfs_remove_chunk() below.
3306	 *
3307	 * Make sure to acquire this mutex before doing a tree search (dev
3308	 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3309	 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3310	 * we release the path used to search the chunk/dev tree and before
3311	 * the current task acquires this mutex and calls us.
3312	 */
3313	lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3314
3315	/* step one, relocate all the extents inside this chunk */
3316	btrfs_scrub_pause(fs_info);
3317	ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3318	btrfs_scrub_continue(fs_info);
3319	if (ret) {
3320		/*
3321		 * If we had a transaction abort, stop all running scrubs.
3322		 * See transaction.c:cleanup_transaction() why we do it here.
3323		 */
3324		if (BTRFS_FS_ERROR(fs_info))
3325			btrfs_scrub_cancel(fs_info);
3326		return ret;
3327	}
3328
3329	block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3330	if (!block_group)
3331		return -ENOENT;
3332	btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3333	length = block_group->length;
3334	btrfs_put_block_group(block_group);
3335
3336	/*
3337	 * On a zoned file system, discard the whole block group, this will
3338	 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3339	 * resetting the zone fails, don't treat it as a fatal problem from the
3340	 * filesystem's point of view.
3341	 */
3342	if (btrfs_is_zoned(fs_info)) {
3343		ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3344		if (ret)
3345			btrfs_info(fs_info,
3346				"failed to reset zone %llu after relocation",
3347				chunk_offset);
3348	}
3349
3350	trans = btrfs_start_trans_remove_block_group(root->fs_info,
3351						     chunk_offset);
3352	if (IS_ERR(trans)) {
3353		ret = PTR_ERR(trans);
3354		btrfs_handle_fs_error(root->fs_info, ret, NULL);
3355		return ret;
3356	}
3357
3358	/*
3359	 * step two, delete the device extents and the
3360	 * chunk tree entries
3361	 */
3362	ret = btrfs_remove_chunk(trans, chunk_offset);
3363	btrfs_end_transaction(trans);
3364	return ret;
3365}
3366
3367static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3368{
3369	struct btrfs_root *chunk_root = fs_info->chunk_root;
3370	struct btrfs_path *path;
3371	struct extent_buffer *leaf;
3372	struct btrfs_chunk *chunk;
3373	struct btrfs_key key;
3374	struct btrfs_key found_key;
3375	u64 chunk_type;
3376	bool retried = false;
3377	int failed = 0;
3378	int ret;
3379
3380	path = btrfs_alloc_path();
3381	if (!path)
3382		return -ENOMEM;
3383
3384again:
3385	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3386	key.offset = (u64)-1;
3387	key.type = BTRFS_CHUNK_ITEM_KEY;
3388
3389	while (1) {
3390		mutex_lock(&fs_info->reclaim_bgs_lock);
3391		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3392		if (ret < 0) {
3393			mutex_unlock(&fs_info->reclaim_bgs_lock);
3394			goto error;
3395		}
3396		BUG_ON(ret == 0); /* Corruption */
3397
3398		ret = btrfs_previous_item(chunk_root, path, key.objectid,
3399					  key.type);
3400		if (ret)
3401			mutex_unlock(&fs_info->reclaim_bgs_lock);
3402		if (ret < 0)
3403			goto error;
3404		if (ret > 0)
3405			break;
3406
3407		leaf = path->nodes[0];
3408		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3409
3410		chunk = btrfs_item_ptr(leaf, path->slots[0],
3411				       struct btrfs_chunk);
3412		chunk_type = btrfs_chunk_type(leaf, chunk);
3413		btrfs_release_path(path);
3414
3415		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3416			ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3417			if (ret == -ENOSPC)
3418				failed++;
3419			else
3420				BUG_ON(ret);
3421		}
3422		mutex_unlock(&fs_info->reclaim_bgs_lock);
3423
3424		if (found_key.offset == 0)
3425			break;
3426		key.offset = found_key.offset - 1;
3427	}
3428	ret = 0;
3429	if (failed && !retried) {
3430		failed = 0;
3431		retried = true;
3432		goto again;
3433	} else if (WARN_ON(failed && retried)) {
3434		ret = -ENOSPC;
3435	}
3436error:
3437	btrfs_free_path(path);
3438	return ret;
3439}
3440
3441/*
3442 * return 1 : allocate a data chunk successfully,
3443 * return <0: errors during allocating a data chunk,
3444 * return 0 : no need to allocate a data chunk.
3445 */
3446static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3447				      u64 chunk_offset)
3448{
3449	struct btrfs_block_group *cache;
3450	u64 bytes_used;
3451	u64 chunk_type;
3452
3453	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3454	ASSERT(cache);
3455	chunk_type = cache->flags;
3456	btrfs_put_block_group(cache);
3457
3458	if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3459		return 0;
3460
3461	spin_lock(&fs_info->data_sinfo->lock);
3462	bytes_used = fs_info->data_sinfo->bytes_used;
3463	spin_unlock(&fs_info->data_sinfo->lock);
3464
3465	if (!bytes_used) {
3466		struct btrfs_trans_handle *trans;
3467		int ret;
3468
3469		trans =	btrfs_join_transaction(fs_info->tree_root);
3470		if (IS_ERR(trans))
3471			return PTR_ERR(trans);
3472
3473		ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3474		btrfs_end_transaction(trans);
3475		if (ret < 0)
3476			return ret;
3477		return 1;
3478	}
3479
3480	return 0;
3481}
3482
3483static int insert_balance_item(struct btrfs_fs_info *fs_info,
3484			       struct btrfs_balance_control *bctl)
3485{
3486	struct btrfs_root *root = fs_info->tree_root;
3487	struct btrfs_trans_handle *trans;
3488	struct btrfs_balance_item *item;
3489	struct btrfs_disk_balance_args disk_bargs;
3490	struct btrfs_path *path;
3491	struct extent_buffer *leaf;
3492	struct btrfs_key key;
3493	int ret, err;
3494
3495	path = btrfs_alloc_path();
3496	if (!path)
3497		return -ENOMEM;
3498
3499	trans = btrfs_start_transaction(root, 0);
3500	if (IS_ERR(trans)) {
3501		btrfs_free_path(path);
3502		return PTR_ERR(trans);
3503	}
3504
3505	key.objectid = BTRFS_BALANCE_OBJECTID;
3506	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3507	key.offset = 0;
3508
3509	ret = btrfs_insert_empty_item(trans, root, path, &key,
3510				      sizeof(*item));
3511	if (ret)
3512		goto out;
3513
3514	leaf = path->nodes[0];
3515	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3516
3517	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3518
3519	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3520	btrfs_set_balance_data(leaf, item, &disk_bargs);
3521	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3522	btrfs_set_balance_meta(leaf, item, &disk_bargs);
3523	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3524	btrfs_set_balance_sys(leaf, item, &disk_bargs);
3525
3526	btrfs_set_balance_flags(leaf, item, bctl->flags);
3527
3528	btrfs_mark_buffer_dirty(trans, leaf);
3529out:
3530	btrfs_free_path(path);
3531	err = btrfs_commit_transaction(trans);
3532	if (err && !ret)
3533		ret = err;
3534	return ret;
3535}
3536
3537static int del_balance_item(struct btrfs_fs_info *fs_info)
3538{
3539	struct btrfs_root *root = fs_info->tree_root;
3540	struct btrfs_trans_handle *trans;
3541	struct btrfs_path *path;
3542	struct btrfs_key key;
3543	int ret, err;
3544
3545	path = btrfs_alloc_path();
3546	if (!path)
3547		return -ENOMEM;
3548
3549	trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3550	if (IS_ERR(trans)) {
3551		btrfs_free_path(path);
3552		return PTR_ERR(trans);
3553	}
3554
3555	key.objectid = BTRFS_BALANCE_OBJECTID;
3556	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3557	key.offset = 0;
3558
3559	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3560	if (ret < 0)
3561		goto out;
3562	if (ret > 0) {
3563		ret = -ENOENT;
3564		goto out;
3565	}
3566
3567	ret = btrfs_del_item(trans, root, path);
3568out:
3569	btrfs_free_path(path);
3570	err = btrfs_commit_transaction(trans);
3571	if (err && !ret)
3572		ret = err;
3573	return ret;
3574}
3575
3576/*
3577 * This is a heuristic used to reduce the number of chunks balanced on
3578 * resume after balance was interrupted.
3579 */
3580static void update_balance_args(struct btrfs_balance_control *bctl)
3581{
3582	/*
3583	 * Turn on soft mode for chunk types that were being converted.
3584	 */
3585	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3586		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3587	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3588		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3589	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3590		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3591
3592	/*
3593	 * Turn on usage filter if is not already used.  The idea is
3594	 * that chunks that we have already balanced should be
3595	 * reasonably full.  Don't do it for chunks that are being
3596	 * converted - that will keep us from relocating unconverted
3597	 * (albeit full) chunks.
3598	 */
3599	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3600	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3601	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3602		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3603		bctl->data.usage = 90;
3604	}
3605	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3606	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3607	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3608		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3609		bctl->sys.usage = 90;
3610	}
3611	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3612	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3613	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3614		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3615		bctl->meta.usage = 90;
3616	}
3617}
3618
3619/*
3620 * Clear the balance status in fs_info and delete the balance item from disk.
3621 */
3622static void reset_balance_state(struct btrfs_fs_info *fs_info)
3623{
3624	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3625	int ret;
3626
3627	BUG_ON(!fs_info->balance_ctl);
3628
3629	spin_lock(&fs_info->balance_lock);
3630	fs_info->balance_ctl = NULL;
3631	spin_unlock(&fs_info->balance_lock);
3632
3633	kfree(bctl);
3634	ret = del_balance_item(fs_info);
3635	if (ret)
3636		btrfs_handle_fs_error(fs_info, ret, NULL);
3637}
3638
3639/*
3640 * Balance filters.  Return 1 if chunk should be filtered out
3641 * (should not be balanced).
3642 */
3643static int chunk_profiles_filter(u64 chunk_type,
3644				 struct btrfs_balance_args *bargs)
3645{
3646	chunk_type = chunk_to_extended(chunk_type) &
3647				BTRFS_EXTENDED_PROFILE_MASK;
3648
3649	if (bargs->profiles & chunk_type)
3650		return 0;
3651
3652	return 1;
3653}
3654
3655static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3656			      struct btrfs_balance_args *bargs)
3657{
3658	struct btrfs_block_group *cache;
3659	u64 chunk_used;
3660	u64 user_thresh_min;
3661	u64 user_thresh_max;
3662	int ret = 1;
3663
3664	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3665	chunk_used = cache->used;
3666
3667	if (bargs->usage_min == 0)
3668		user_thresh_min = 0;
3669	else
3670		user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3671
3672	if (bargs->usage_max == 0)
3673		user_thresh_max = 1;
3674	else if (bargs->usage_max > 100)
3675		user_thresh_max = cache->length;
3676	else
3677		user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3678
3679	if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3680		ret = 0;
3681
3682	btrfs_put_block_group(cache);
3683	return ret;
3684}
3685
3686static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3687		u64 chunk_offset, struct btrfs_balance_args *bargs)
3688{
3689	struct btrfs_block_group *cache;
3690	u64 chunk_used, user_thresh;
3691	int ret = 1;
3692
3693	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3694	chunk_used = cache->used;
3695
3696	if (bargs->usage_min == 0)
3697		user_thresh = 1;
3698	else if (bargs->usage > 100)
3699		user_thresh = cache->length;
3700	else
3701		user_thresh = mult_perc(cache->length, bargs->usage);
3702
3703	if (chunk_used < user_thresh)
3704		ret = 0;
3705
3706	btrfs_put_block_group(cache);
3707	return ret;
3708}
3709
3710static int chunk_devid_filter(struct extent_buffer *leaf,
3711			      struct btrfs_chunk *chunk,
3712			      struct btrfs_balance_args *bargs)
3713{
3714	struct btrfs_stripe *stripe;
3715	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3716	int i;
3717
3718	for (i = 0; i < num_stripes; i++) {
3719		stripe = btrfs_stripe_nr(chunk, i);
3720		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3721			return 0;
3722	}
3723
3724	return 1;
3725}
3726
3727static u64 calc_data_stripes(u64 type, int num_stripes)
3728{
3729	const int index = btrfs_bg_flags_to_raid_index(type);
3730	const int ncopies = btrfs_raid_array[index].ncopies;
3731	const int nparity = btrfs_raid_array[index].nparity;
3732
3733	return (num_stripes - nparity) / ncopies;
3734}
3735
3736/* [pstart, pend) */
3737static int chunk_drange_filter(struct extent_buffer *leaf,
3738			       struct btrfs_chunk *chunk,
3739			       struct btrfs_balance_args *bargs)
3740{
3741	struct btrfs_stripe *stripe;
3742	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3743	u64 stripe_offset;
3744	u64 stripe_length;
3745	u64 type;
3746	int factor;
3747	int i;
3748
3749	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3750		return 0;
3751
3752	type = btrfs_chunk_type(leaf, chunk);
3753	factor = calc_data_stripes(type, num_stripes);
3754
3755	for (i = 0; i < num_stripes; i++) {
3756		stripe = btrfs_stripe_nr(chunk, i);
3757		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3758			continue;
3759
3760		stripe_offset = btrfs_stripe_offset(leaf, stripe);
3761		stripe_length = btrfs_chunk_length(leaf, chunk);
3762		stripe_length = div_u64(stripe_length, factor);
3763
3764		if (stripe_offset < bargs->pend &&
3765		    stripe_offset + stripe_length > bargs->pstart)
3766			return 0;
3767	}
3768
3769	return 1;
3770}
3771
3772/* [vstart, vend) */
3773static int chunk_vrange_filter(struct extent_buffer *leaf,
3774			       struct btrfs_chunk *chunk,
3775			       u64 chunk_offset,
3776			       struct btrfs_balance_args *bargs)
3777{
3778	if (chunk_offset < bargs->vend &&
3779	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3780		/* at least part of the chunk is inside this vrange */
3781		return 0;
3782
3783	return 1;
3784}
3785
3786static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3787			       struct btrfs_chunk *chunk,
3788			       struct btrfs_balance_args *bargs)
3789{
3790	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3791
3792	if (bargs->stripes_min <= num_stripes
3793			&& num_stripes <= bargs->stripes_max)
3794		return 0;
3795
3796	return 1;
3797}
3798
3799static int chunk_soft_convert_filter(u64 chunk_type,
3800				     struct btrfs_balance_args *bargs)
3801{
3802	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3803		return 0;
3804
3805	chunk_type = chunk_to_extended(chunk_type) &
3806				BTRFS_EXTENDED_PROFILE_MASK;
3807
3808	if (bargs->target == chunk_type)
3809		return 1;
3810
3811	return 0;
3812}
3813
3814static int should_balance_chunk(struct extent_buffer *leaf,
3815				struct btrfs_chunk *chunk, u64 chunk_offset)
3816{
3817	struct btrfs_fs_info *fs_info = leaf->fs_info;
3818	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3819	struct btrfs_balance_args *bargs = NULL;
3820	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3821
3822	/* type filter */
3823	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3824	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3825		return 0;
3826	}
3827
3828	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3829		bargs = &bctl->data;
3830	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3831		bargs = &bctl->sys;
3832	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3833		bargs = &bctl->meta;
3834
3835	/* profiles filter */
3836	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3837	    chunk_profiles_filter(chunk_type, bargs)) {
3838		return 0;
3839	}
3840
3841	/* usage filter */
3842	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3843	    chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3844		return 0;
3845	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3846	    chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3847		return 0;
3848	}
3849
3850	/* devid filter */
3851	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3852	    chunk_devid_filter(leaf, chunk, bargs)) {
3853		return 0;
3854	}
3855
3856	/* drange filter, makes sense only with devid filter */
3857	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3858	    chunk_drange_filter(leaf, chunk, bargs)) {
3859		return 0;
3860	}
3861
3862	/* vrange filter */
3863	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3864	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3865		return 0;
3866	}
3867
3868	/* stripes filter */
3869	if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3870	    chunk_stripes_range_filter(leaf, chunk, bargs)) {
3871		return 0;
3872	}
3873
3874	/* soft profile changing mode */
3875	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3876	    chunk_soft_convert_filter(chunk_type, bargs)) {
3877		return 0;
3878	}
3879
3880	/*
3881	 * limited by count, must be the last filter
3882	 */
3883	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3884		if (bargs->limit == 0)
3885			return 0;
3886		else
3887			bargs->limit--;
3888	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3889		/*
3890		 * Same logic as the 'limit' filter; the minimum cannot be
3891		 * determined here because we do not have the global information
3892		 * about the count of all chunks that satisfy the filters.
3893		 */
3894		if (bargs->limit_max == 0)
3895			return 0;
3896		else
3897			bargs->limit_max--;
3898	}
3899
3900	return 1;
3901}
3902
3903static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3904{
3905	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3906	struct btrfs_root *chunk_root = fs_info->chunk_root;
3907	u64 chunk_type;
3908	struct btrfs_chunk *chunk;
3909	struct btrfs_path *path = NULL;
3910	struct btrfs_key key;
3911	struct btrfs_key found_key;
3912	struct extent_buffer *leaf;
3913	int slot;
3914	int ret;
3915	int enospc_errors = 0;
3916	bool counting = true;
3917	/* The single value limit and min/max limits use the same bytes in the */
3918	u64 limit_data = bctl->data.limit;
3919	u64 limit_meta = bctl->meta.limit;
3920	u64 limit_sys = bctl->sys.limit;
3921	u32 count_data = 0;
3922	u32 count_meta = 0;
3923	u32 count_sys = 0;
3924	int chunk_reserved = 0;
3925
3926	path = btrfs_alloc_path();
3927	if (!path) {
3928		ret = -ENOMEM;
3929		goto error;
3930	}
3931
3932	/* zero out stat counters */
3933	spin_lock(&fs_info->balance_lock);
3934	memset(&bctl->stat, 0, sizeof(bctl->stat));
3935	spin_unlock(&fs_info->balance_lock);
3936again:
3937	if (!counting) {
3938		/*
3939		 * The single value limit and min/max limits use the same bytes
3940		 * in the
3941		 */
3942		bctl->data.limit = limit_data;
3943		bctl->meta.limit = limit_meta;
3944		bctl->sys.limit = limit_sys;
3945	}
3946	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3947	key.offset = (u64)-1;
3948	key.type = BTRFS_CHUNK_ITEM_KEY;
3949
3950	while (1) {
3951		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3952		    atomic_read(&fs_info->balance_cancel_req)) {
3953			ret = -ECANCELED;
3954			goto error;
3955		}
3956
3957		mutex_lock(&fs_info->reclaim_bgs_lock);
3958		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3959		if (ret < 0) {
3960			mutex_unlock(&fs_info->reclaim_bgs_lock);
3961			goto error;
3962		}
3963
3964		/*
3965		 * this shouldn't happen, it means the last relocate
3966		 * failed
3967		 */
3968		if (ret == 0)
3969			BUG(); /* FIXME break ? */
3970
3971		ret = btrfs_previous_item(chunk_root, path, 0,
3972					  BTRFS_CHUNK_ITEM_KEY);
3973		if (ret) {
3974			mutex_unlock(&fs_info->reclaim_bgs_lock);
3975			ret = 0;
3976			break;
3977		}
3978
3979		leaf = path->nodes[0];
3980		slot = path->slots[0];
3981		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3982
3983		if (found_key.objectid != key.objectid) {
3984			mutex_unlock(&fs_info->reclaim_bgs_lock);
3985			break;
3986		}
3987
3988		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3989		chunk_type = btrfs_chunk_type(leaf, chunk);
3990
3991		if (!counting) {
3992			spin_lock(&fs_info->balance_lock);
3993			bctl->stat.considered++;
3994			spin_unlock(&fs_info->balance_lock);
3995		}
3996
3997		ret = should_balance_chunk(leaf, chunk, found_key.offset);
3998
3999		btrfs_release_path(path);
4000		if (!ret) {
4001			mutex_unlock(&fs_info->reclaim_bgs_lock);
4002			goto loop;
4003		}
4004
4005		if (counting) {
4006			mutex_unlock(&fs_info->reclaim_bgs_lock);
4007			spin_lock(&fs_info->balance_lock);
4008			bctl->stat.expected++;
4009			spin_unlock(&fs_info->balance_lock);
4010
4011			if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
4012				count_data++;
4013			else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
4014				count_sys++;
4015			else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
4016				count_meta++;
4017
4018			goto loop;
4019		}
4020
4021		/*
4022		 * Apply limit_min filter, no need to check if the LIMITS
4023		 * filter is used, limit_min is 0 by default
4024		 */
4025		if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
4026					count_data < bctl->data.limit_min)
4027				|| ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
4028					count_meta < bctl->meta.limit_min)
4029				|| ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
4030					count_sys < bctl->sys.limit_min)) {
4031			mutex_unlock(&fs_info->reclaim_bgs_lock);
4032			goto loop;
4033		}
4034
4035		if (!chunk_reserved) {
4036			/*
4037			 * We may be relocating the only data chunk we have,
4038			 * which could potentially end up with losing data's
4039			 * raid profile, so lets allocate an empty one in
4040			 * advance.
4041			 */
4042			ret = btrfs_may_alloc_data_chunk(fs_info,
4043							 found_key.offset);
4044			if (ret < 0) {
4045				mutex_unlock(&fs_info->reclaim_bgs_lock);
4046				goto error;
4047			} else if (ret == 1) {
4048				chunk_reserved = 1;
4049			}
4050		}
4051
4052		ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4053		mutex_unlock(&fs_info->reclaim_bgs_lock);
4054		if (ret == -ENOSPC) {
4055			enospc_errors++;
4056		} else if (ret == -ETXTBSY) {
4057			btrfs_info(fs_info,
4058	   "skipping relocation of block group %llu due to active swapfile",
4059				   found_key.offset);
4060			ret = 0;
4061		} else if (ret) {
4062			goto error;
4063		} else {
4064			spin_lock(&fs_info->balance_lock);
4065			bctl->stat.completed++;
4066			spin_unlock(&fs_info->balance_lock);
4067		}
4068loop:
4069		if (found_key.offset == 0)
4070			break;
4071		key.offset = found_key.offset - 1;
4072	}
4073
4074	if (counting) {
4075		btrfs_release_path(path);
4076		counting = false;
4077		goto again;
4078	}
4079error:
4080	btrfs_free_path(path);
4081	if (enospc_errors) {
4082		btrfs_info(fs_info, "%d enospc errors during balance",
4083			   enospc_errors);
4084		if (!ret)
4085			ret = -ENOSPC;
4086	}
4087
4088	return ret;
4089}
4090
4091/*
4092 * See if a given profile is valid and reduced.
4093 *
4094 * @flags:     profile to validate
4095 * @extended:  if true @flags is treated as an extended profile
4096 */
4097static int alloc_profile_is_valid(u64 flags, int extended)
4098{
4099	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4100			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
4101
4102	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4103
4104	/* 1) check that all other bits are zeroed */
4105	if (flags & ~mask)
4106		return 0;
4107
4108	/* 2) see if profile is reduced */
4109	if (flags == 0)
4110		return !extended; /* "0" is valid for usual profiles */
4111
4112	return has_single_bit_set(flags);
4113}
4114
4115/*
4116 * Validate target profile against allowed profiles and return true if it's OK.
4117 * Otherwise print the error message and return false.
4118 */
4119static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4120		const struct btrfs_balance_args *bargs,
4121		u64 allowed, const char *type)
4122{
4123	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4124		return true;
4125
4126	/* Profile is valid and does not have bits outside of the allowed set */
4127	if (alloc_profile_is_valid(bargs->target, 1) &&
4128	    (bargs->target & ~allowed) == 0)
4129		return true;
4130
4131	btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4132			type, btrfs_bg_type_to_raid_name(bargs->target));
4133	return false;
4134}
4135
4136/*
4137 * Fill @buf with textual description of balance filter flags @bargs, up to
4138 * @size_buf including the terminating null. The output may be trimmed if it
4139 * does not fit into the provided buffer.
4140 */
4141static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4142				 u32 size_buf)
4143{
4144	int ret;
4145	u32 size_bp = size_buf;
4146	char *bp = buf;
4147	u64 flags = bargs->flags;
4148	char tmp_buf[128] = {'\0'};
4149
4150	if (!flags)
4151		return;
4152
4153#define CHECK_APPEND_NOARG(a)						\
4154	do {								\
4155		ret = snprintf(bp, size_bp, (a));			\
4156		if (ret < 0 || ret >= size_bp)				\
4157			goto out_overflow;				\
4158		size_bp -= ret;						\
4159		bp += ret;						\
4160	} while (0)
4161
4162#define CHECK_APPEND_1ARG(a, v1)					\
4163	do {								\
4164		ret = snprintf(bp, size_bp, (a), (v1));			\
4165		if (ret < 0 || ret >= size_bp)				\
4166			goto out_overflow;				\
4167		size_bp -= ret;						\
4168		bp += ret;						\
4169	} while (0)
4170
4171#define CHECK_APPEND_2ARG(a, v1, v2)					\
4172	do {								\
4173		ret = snprintf(bp, size_bp, (a), (v1), (v2));		\
4174		if (ret < 0 || ret >= size_bp)				\
4175			goto out_overflow;				\
4176		size_bp -= ret;						\
4177		bp += ret;						\
4178	} while (0)
4179
4180	if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4181		CHECK_APPEND_1ARG("convert=%s,",
4182				  btrfs_bg_type_to_raid_name(bargs->target));
4183
4184	if (flags & BTRFS_BALANCE_ARGS_SOFT)
4185		CHECK_APPEND_NOARG("soft,");
4186
4187	if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4188		btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4189					    sizeof(tmp_buf));
4190		CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4191	}
4192
4193	if (flags & BTRFS_BALANCE_ARGS_USAGE)
4194		CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4195
4196	if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4197		CHECK_APPEND_2ARG("usage=%u..%u,",
4198				  bargs->usage_min, bargs->usage_max);
4199
4200	if (flags & BTRFS_BALANCE_ARGS_DEVID)
4201		CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4202
4203	if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4204		CHECK_APPEND_2ARG("drange=%llu..%llu,",
4205				  bargs->pstart, bargs->pend);
4206
4207	if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4208		CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4209				  bargs->vstart, bargs->vend);
4210
4211	if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4212		CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4213
4214	if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4215		CHECK_APPEND_2ARG("limit=%u..%u,",
4216				bargs->limit_min, bargs->limit_max);
4217
4218	if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4219		CHECK_APPEND_2ARG("stripes=%u..%u,",
4220				  bargs->stripes_min, bargs->stripes_max);
4221
4222#undef CHECK_APPEND_2ARG
4223#undef CHECK_APPEND_1ARG
4224#undef CHECK_APPEND_NOARG
4225
4226out_overflow:
4227
4228	if (size_bp < size_buf)
4229		buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4230	else
4231		buf[0] = '\0';
4232}
4233
4234static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4235{
4236	u32 size_buf = 1024;
4237	char tmp_buf[192] = {'\0'};
4238	char *buf;
4239	char *bp;
4240	u32 size_bp = size_buf;
4241	int ret;
4242	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4243
4244	buf = kzalloc(size_buf, GFP_KERNEL);
4245	if (!buf)
4246		return;
4247
4248	bp = buf;
4249
4250#define CHECK_APPEND_1ARG(a, v1)					\
4251	do {								\
4252		ret = snprintf(bp, size_bp, (a), (v1));			\
4253		if (ret < 0 || ret >= size_bp)				\
4254			goto out_overflow;				\
4255		size_bp -= ret;						\
4256		bp += ret;						\
4257	} while (0)
4258
4259	if (bctl->flags & BTRFS_BALANCE_FORCE)
4260		CHECK_APPEND_1ARG("%s", "-f ");
4261
4262	if (bctl->flags & BTRFS_BALANCE_DATA) {
4263		describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4264		CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4265	}
4266
4267	if (bctl->flags & BTRFS_BALANCE_METADATA) {
4268		describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4269		CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4270	}
4271
4272	if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4273		describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4274		CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4275	}
4276
4277#undef CHECK_APPEND_1ARG
4278
4279out_overflow:
4280
4281	if (size_bp < size_buf)
4282		buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4283	btrfs_info(fs_info, "balance: %s %s",
4284		   (bctl->flags & BTRFS_BALANCE_RESUME) ?
4285		   "resume" : "start", buf);
4286
4287	kfree(buf);
4288}
4289
4290/*
4291 * Should be called with balance mutexe held
4292 */
4293int btrfs_balance(struct btrfs_fs_info *fs_info,
4294		  struct btrfs_balance_control *bctl,
4295		  struct btrfs_ioctl_balance_args *bargs)
4296{
4297	u64 meta_target, data_target;
4298	u64 allowed;
4299	int mixed = 0;
4300	int ret;
4301	u64 num_devices;
4302	unsigned seq;
4303	bool reducing_redundancy;
4304	bool paused = false;
4305	int i;
4306
4307	if (btrfs_fs_closing(fs_info) ||
4308	    atomic_read(&fs_info->balance_pause_req) ||
4309	    btrfs_should_cancel_balance(fs_info)) {
4310		ret = -EINVAL;
4311		goto out;
4312	}
4313
4314	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4315	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4316		mixed = 1;
4317
4318	/*
4319	 * In case of mixed groups both data and meta should be picked,
4320	 * and identical options should be given for both of them.
4321	 */
4322	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4323	if (mixed && (bctl->flags & allowed)) {
4324		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4325		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4326		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4327			btrfs_err(fs_info,
4328	  "balance: mixed groups data and metadata options must be the same");
4329			ret = -EINVAL;
4330			goto out;
4331		}
4332	}
4333
4334	/*
4335	 * rw_devices will not change at the moment, device add/delete/replace
4336	 * are exclusive
4337	 */
4338	num_devices = fs_info->fs_devices->rw_devices;
4339
4340	/*
4341	 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4342	 * special bit for it, to make it easier to distinguish.  Thus we need
4343	 * to set it manually, or balance would refuse the profile.
4344	 */
4345	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4346	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4347		if (num_devices >= btrfs_raid_array[i].devs_min)
4348			allowed |= btrfs_raid_array[i].bg_flag;
4349
4350	if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4351	    !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4352	    !validate_convert_profile(fs_info, &bctl->sys,  allowed, "system")) {
4353		ret = -EINVAL;
4354		goto out;
4355	}
4356
4357	/*
4358	 * Allow to reduce metadata or system integrity only if force set for
4359	 * profiles with redundancy (copies, parity)
4360	 */
4361	allowed = 0;
4362	for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4363		if (btrfs_raid_array[i].ncopies >= 2 ||
4364		    btrfs_raid_array[i].tolerated_failures >= 1)
4365			allowed |= btrfs_raid_array[i].bg_flag;
4366	}
4367	do {
4368		seq = read_seqbegin(&fs_info->profiles_lock);
4369
4370		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4371		     (fs_info->avail_system_alloc_bits & allowed) &&
4372		     !(bctl->sys.target & allowed)) ||
4373		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4374		     (fs_info->avail_metadata_alloc_bits & allowed) &&
4375		     !(bctl->meta.target & allowed)))
4376			reducing_redundancy = true;
4377		else
4378			reducing_redundancy = false;
4379
4380		/* if we're not converting, the target field is uninitialized */
4381		meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4382			bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4383		data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4384			bctl->data.target : fs_info->avail_data_alloc_bits;
4385	} while (read_seqretry(&fs_info->profiles_lock, seq));
4386
4387	if (reducing_redundancy) {
4388		if (bctl->flags & BTRFS_BALANCE_FORCE) {
4389			btrfs_info(fs_info,
4390			   "balance: force reducing metadata redundancy");
4391		} else {
4392			btrfs_err(fs_info,
4393	"balance: reduces metadata redundancy, use --force if you want this");
4394			ret = -EINVAL;
4395			goto out;
4396		}
4397	}
4398
4399	if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4400		btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4401		btrfs_warn(fs_info,
4402	"balance: metadata profile %s has lower redundancy than data profile %s",
4403				btrfs_bg_type_to_raid_name(meta_target),
4404				btrfs_bg_type_to_raid_name(data_target));
4405	}
4406
4407	ret = insert_balance_item(fs_info, bctl);
4408	if (ret && ret != -EEXIST)
4409		goto out;
4410
4411	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4412		BUG_ON(ret == -EEXIST);
4413		BUG_ON(fs_info->balance_ctl);
4414		spin_lock(&fs_info->balance_lock);
4415		fs_info->balance_ctl = bctl;
4416		spin_unlock(&fs_info->balance_lock);
4417	} else {
4418		BUG_ON(ret != -EEXIST);
4419		spin_lock(&fs_info->balance_lock);
4420		update_balance_args(bctl);
4421		spin_unlock(&fs_info->balance_lock);
4422	}
4423
4424	ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4425	set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4426	describe_balance_start_or_resume(fs_info);
4427	mutex_unlock(&fs_info->balance_mutex);
4428
4429	ret = __btrfs_balance(fs_info);
4430
4431	mutex_lock(&fs_info->balance_mutex);
4432	if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4433		btrfs_info(fs_info, "balance: paused");
4434		btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4435		paused = true;
4436	}
4437	/*
4438	 * Balance can be canceled by:
4439	 *
4440	 * - Regular cancel request
4441	 *   Then ret == -ECANCELED and balance_cancel_req > 0
4442	 *
4443	 * - Fatal signal to "btrfs" process
4444	 *   Either the signal caught by wait_reserve_ticket() and callers
4445	 *   got -EINTR, or caught by btrfs_should_cancel_balance() and
4446	 *   got -ECANCELED.
4447	 *   Either way, in this case balance_cancel_req = 0, and
4448	 *   ret == -EINTR or ret == -ECANCELED.
4449	 *
4450	 * So here we only check the return value to catch canceled balance.
4451	 */
4452	else if (ret == -ECANCELED || ret == -EINTR)
4453		btrfs_info(fs_info, "balance: canceled");
4454	else
4455		btrfs_info(fs_info, "balance: ended with status: %d", ret);
4456
4457	clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4458
4459	if (bargs) {
4460		memset(bargs, 0, sizeof(*bargs));
4461		btrfs_update_ioctl_balance_args(fs_info, bargs);
4462	}
4463
4464	/* We didn't pause, we can clean everything up. */
4465	if (!paused) {
4466		reset_balance_state(fs_info);
4467		btrfs_exclop_finish(fs_info);
4468	}
4469
4470	wake_up(&fs_info->balance_wait_q);
4471
4472	return ret;
4473out:
4474	if (bctl->flags & BTRFS_BALANCE_RESUME)
4475		reset_balance_state(fs_info);
4476	else
4477		kfree(bctl);
4478	btrfs_exclop_finish(fs_info);
4479
4480	return ret;
4481}
4482
4483static int balance_kthread(void *data)
4484{
4485	struct btrfs_fs_info *fs_info = data;
4486	int ret = 0;
4487
4488	sb_start_write(fs_info->sb);
4489	mutex_lock(&fs_info->balance_mutex);
4490	if (fs_info->balance_ctl)
4491		ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4492	mutex_unlock(&fs_info->balance_mutex);
4493	sb_end_write(fs_info->sb);
4494
4495	return ret;
4496}
4497
4498int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4499{
4500	struct task_struct *tsk;
4501
4502	mutex_lock(&fs_info->balance_mutex);
4503	if (!fs_info->balance_ctl) {
4504		mutex_unlock(&fs_info->balance_mutex);
4505		return 0;
4506	}
4507	mutex_unlock(&fs_info->balance_mutex);
4508
4509	if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4510		btrfs_info(fs_info, "balance: resume skipped");
4511		return 0;
4512	}
4513
4514	spin_lock(&fs_info->super_lock);
4515	ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4516	fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4517	spin_unlock(&fs_info->super_lock);
4518	/*
4519	 * A ro->rw remount sequence should continue with the paused balance
4520	 * regardless of who pauses it, system or the user as of now, so set
4521	 * the resume flag.
4522	 */
4523	spin_lock(&fs_info->balance_lock);
4524	fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4525	spin_unlock(&fs_info->balance_lock);
4526
4527	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4528	return PTR_ERR_OR_ZERO(tsk);
4529}
4530
4531int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4532{
4533	struct btrfs_balance_control *bctl;
4534	struct btrfs_balance_item *item;
4535	struct btrfs_disk_balance_args disk_bargs;
4536	struct btrfs_path *path;
4537	struct extent_buffer *leaf;
4538	struct btrfs_key key;
4539	int ret;
4540
4541	path = btrfs_alloc_path();
4542	if (!path)
4543		return -ENOMEM;
4544
4545	key.objectid = BTRFS_BALANCE_OBJECTID;
4546	key.type = BTRFS_TEMPORARY_ITEM_KEY;
4547	key.offset = 0;
4548
4549	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4550	if (ret < 0)
4551		goto out;
4552	if (ret > 0) { /* ret = -ENOENT; */
4553		ret = 0;
4554		goto out;
4555	}
4556
4557	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4558	if (!bctl) {
4559		ret = -ENOMEM;
4560		goto out;
4561	}
4562
4563	leaf = path->nodes[0];
4564	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4565
4566	bctl->flags = btrfs_balance_flags(leaf, item);
4567	bctl->flags |= BTRFS_BALANCE_RESUME;
4568
4569	btrfs_balance_data(leaf, item, &disk_bargs);
4570	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4571	btrfs_balance_meta(leaf, item, &disk_bargs);
4572	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4573	btrfs_balance_sys(leaf, item, &disk_bargs);
4574	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4575
4576	/*
4577	 * This should never happen, as the paused balance state is recovered
4578	 * during mount without any chance of other exclusive ops to collide.
4579	 *
4580	 * This gives the exclusive op status to balance and keeps in paused
4581	 * state until user intervention (cancel or umount). If the ownership
4582	 * cannot be assigned, show a message but do not fail. The balance
4583	 * is in a paused state and must have fs_info::balance_ctl properly
4584	 * set up.
4585	 */
4586	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4587		btrfs_warn(fs_info,
4588	"balance: cannot set exclusive op status, resume manually");
4589
4590	btrfs_release_path(path);
4591
4592	mutex_lock(&fs_info->balance_mutex);
4593	BUG_ON(fs_info->balance_ctl);
4594	spin_lock(&fs_info->balance_lock);
4595	fs_info->balance_ctl = bctl;
4596	spin_unlock(&fs_info->balance_lock);
4597	mutex_unlock(&fs_info->balance_mutex);
4598out:
4599	btrfs_free_path(path);
4600	return ret;
4601}
4602
4603int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4604{
4605	int ret = 0;
4606
4607	mutex_lock(&fs_info->balance_mutex);
4608	if (!fs_info->balance_ctl) {
4609		mutex_unlock(&fs_info->balance_mutex);
4610		return -ENOTCONN;
4611	}
4612
4613	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4614		atomic_inc(&fs_info->balance_pause_req);
4615		mutex_unlock(&fs_info->balance_mutex);
4616
4617		wait_event(fs_info->balance_wait_q,
4618			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4619
4620		mutex_lock(&fs_info->balance_mutex);
4621		/* we are good with balance_ctl ripped off from under us */
4622		BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4623		atomic_dec(&fs_info->balance_pause_req);
4624	} else {
4625		ret = -ENOTCONN;
4626	}
4627
4628	mutex_unlock(&fs_info->balance_mutex);
4629	return ret;
4630}
4631
4632int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4633{
4634	mutex_lock(&fs_info->balance_mutex);
4635	if (!fs_info->balance_ctl) {
4636		mutex_unlock(&fs_info->balance_mutex);
4637		return -ENOTCONN;
4638	}
4639
4640	/*
4641	 * A paused balance with the item stored on disk can be resumed at
4642	 * mount time if the mount is read-write. Otherwise it's still paused
4643	 * and we must not allow cancelling as it deletes the item.
4644	 */
4645	if (sb_rdonly(fs_info->sb)) {
4646		mutex_unlock(&fs_info->balance_mutex);
4647		return -EROFS;
4648	}
4649
4650	atomic_inc(&fs_info->balance_cancel_req);
4651	/*
4652	 * if we are running just wait and return, balance item is
4653	 * deleted in btrfs_balance in this case
4654	 */
4655	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4656		mutex_unlock(&fs_info->balance_mutex);
4657		wait_event(fs_info->balance_wait_q,
4658			   !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4659		mutex_lock(&fs_info->balance_mutex);
4660	} else {
4661		mutex_unlock(&fs_info->balance_mutex);
4662		/*
4663		 * Lock released to allow other waiters to continue, we'll
4664		 * reexamine the status again.
4665		 */
4666		mutex_lock(&fs_info->balance_mutex);
4667
4668		if (fs_info->balance_ctl) {
4669			reset_balance_state(fs_info);
4670			btrfs_exclop_finish(fs_info);
4671			btrfs_info(fs_info, "balance: canceled");
4672		}
4673	}
4674
4675	ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4676	atomic_dec(&fs_info->balance_cancel_req);
4677	mutex_unlock(&fs_info->balance_mutex);
4678	return 0;
4679}
4680
4681int btrfs_uuid_scan_kthread(void *data)
4682{
4683	struct btrfs_fs_info *fs_info = data;
4684	struct btrfs_root *root = fs_info->tree_root;
4685	struct btrfs_key key;
4686	struct btrfs_path *path = NULL;
4687	int ret = 0;
4688	struct extent_buffer *eb;
4689	int slot;
4690	struct btrfs_root_item root_item;
4691	u32 item_size;
4692	struct btrfs_trans_handle *trans = NULL;
4693	bool closing = false;
4694
4695	path = btrfs_alloc_path();
4696	if (!path) {
4697		ret = -ENOMEM;
4698		goto out;
4699	}
4700
4701	key.objectid = 0;
4702	key.type = BTRFS_ROOT_ITEM_KEY;
4703	key.offset = 0;
4704
4705	while (1) {
4706		if (btrfs_fs_closing(fs_info)) {
4707			closing = true;
4708			break;
4709		}
4710		ret = btrfs_search_forward(root, &key, path,
4711				BTRFS_OLDEST_GENERATION);
4712		if (ret) {
4713			if (ret > 0)
4714				ret = 0;
4715			break;
4716		}
4717
4718		if (key.type != BTRFS_ROOT_ITEM_KEY ||
4719		    (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4720		     key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4721		    key.objectid > BTRFS_LAST_FREE_OBJECTID)
4722			goto skip;
4723
4724		eb = path->nodes[0];
4725		slot = path->slots[0];
4726		item_size = btrfs_item_size(eb, slot);
4727		if (item_size < sizeof(root_item))
4728			goto skip;
4729
4730		read_extent_buffer(eb, &root_item,
4731				   btrfs_item_ptr_offset(eb, slot),
4732				   (int)sizeof(root_item));
4733		if (btrfs_root_refs(&root_item) == 0)
4734			goto skip;
4735
4736		if (!btrfs_is_empty_uuid(root_item.uuid) ||
4737		    !btrfs_is_empty_uuid(root_item.received_uuid)) {
4738			if (trans)
4739				goto update_tree;
4740
4741			btrfs_release_path(path);
4742			/*
4743			 * 1 - subvol uuid item
4744			 * 1 - received_subvol uuid item
4745			 */
4746			trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4747			if (IS_ERR(trans)) {
4748				ret = PTR_ERR(trans);
4749				break;
4750			}
4751			continue;
4752		} else {
4753			goto skip;
4754		}
4755update_tree:
4756		btrfs_release_path(path);
4757		if (!btrfs_is_empty_uuid(root_item.uuid)) {
4758			ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4759						  BTRFS_UUID_KEY_SUBVOL,
4760						  key.objectid);
4761			if (ret < 0) {
4762				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4763					ret);
4764				break;
4765			}
4766		}
4767
4768		if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4769			ret = btrfs_uuid_tree_add(trans,
4770						  root_item.received_uuid,
4771						 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4772						  key.objectid);
4773			if (ret < 0) {
4774				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4775					ret);
4776				break;
4777			}
4778		}
4779
4780skip:
4781		btrfs_release_path(path);
4782		if (trans) {
4783			ret = btrfs_end_transaction(trans);
4784			trans = NULL;
4785			if (ret)
4786				break;
4787		}
4788
4789		if (key.offset < (u64)-1) {
4790			key.offset++;
4791		} else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4792			key.offset = 0;
4793			key.type = BTRFS_ROOT_ITEM_KEY;
4794		} else if (key.objectid < (u64)-1) {
4795			key.offset = 0;
4796			key.type = BTRFS_ROOT_ITEM_KEY;
4797			key.objectid++;
4798		} else {
4799			break;
4800		}
4801		cond_resched();
4802	}
4803
4804out:
4805	btrfs_free_path(path);
4806	if (trans && !IS_ERR(trans))
4807		btrfs_end_transaction(trans);
4808	if (ret)
4809		btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4810	else if (!closing)
4811		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4812	up(&fs_info->uuid_tree_rescan_sem);
4813	return 0;
4814}
4815
4816int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4817{
4818	struct btrfs_trans_handle *trans;
4819	struct btrfs_root *tree_root = fs_info->tree_root;
4820	struct btrfs_root *uuid_root;
4821	struct task_struct *task;
4822	int ret;
4823
4824	/*
4825	 * 1 - root node
4826	 * 1 - root item
4827	 */
4828	trans = btrfs_start_transaction(tree_root, 2);
4829	if (IS_ERR(trans))
4830		return PTR_ERR(trans);
4831
4832	uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4833	if (IS_ERR(uuid_root)) {
4834		ret = PTR_ERR(uuid_root);
4835		btrfs_abort_transaction(trans, ret);
4836		btrfs_end_transaction(trans);
4837		return ret;
4838	}
4839
4840	fs_info->uuid_root = uuid_root;
4841
4842	ret = btrfs_commit_transaction(trans);
4843	if (ret)
4844		return ret;
4845
4846	down(&fs_info->uuid_tree_rescan_sem);
4847	task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4848	if (IS_ERR(task)) {
4849		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
4850		btrfs_warn(fs_info, "failed to start uuid_scan task");
4851		up(&fs_info->uuid_tree_rescan_sem);
4852		return PTR_ERR(task);
4853	}
4854
4855	return 0;
4856}
4857
4858/*
4859 * shrinking a device means finding all of the device extents past
4860 * the new size, and then following the back refs to the chunks.
4861 * The chunk relocation code actually frees the device extent
4862 */
4863int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4864{
4865	struct btrfs_fs_info *fs_info = device->fs_info;
4866	struct btrfs_root *root = fs_info->dev_root;
4867	struct btrfs_trans_handle *trans;
4868	struct btrfs_dev_extent *dev_extent = NULL;
4869	struct btrfs_path *path;
4870	u64 length;
4871	u64 chunk_offset;
4872	int ret;
4873	int slot;
4874	int failed = 0;
4875	bool retried = false;
4876	struct extent_buffer *l;
4877	struct btrfs_key key;
4878	struct btrfs_super_block *super_copy = fs_info->super_copy;
4879	u64 old_total = btrfs_super_total_bytes(super_copy);
4880	u64 old_size = btrfs_device_get_total_bytes(device);
4881	u64 diff;
4882	u64 start;
4883	u64 free_diff = 0;
4884
4885	new_size = round_down(new_size, fs_info->sectorsize);
4886	start = new_size;
4887	diff = round_down(old_size - new_size, fs_info->sectorsize);
4888
4889	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4890		return -EINVAL;
4891
4892	path = btrfs_alloc_path();
4893	if (!path)
4894		return -ENOMEM;
4895
4896	path->reada = READA_BACK;
4897
4898	trans = btrfs_start_transaction(root, 0);
4899	if (IS_ERR(trans)) {
4900		btrfs_free_path(path);
4901		return PTR_ERR(trans);
4902	}
4903
4904	mutex_lock(&fs_info->chunk_mutex);
4905
4906	btrfs_device_set_total_bytes(device, new_size);
4907	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4908		device->fs_devices->total_rw_bytes -= diff;
4909
4910		/*
4911		 * The new free_chunk_space is new_size - used, so we have to
4912		 * subtract the delta of the old free_chunk_space which included
4913		 * old_size - used.  If used > new_size then just subtract this
4914		 * entire device's free space.
4915		 */
4916		if (device->bytes_used < new_size)
4917			free_diff = (old_size - device->bytes_used) -
4918				    (new_size - device->bytes_used);
4919		else
4920			free_diff = old_size - device->bytes_used;
4921		atomic64_sub(free_diff, &fs_info->free_chunk_space);
4922	}
4923
4924	/*
4925	 * Once the device's size has been set to the new size, ensure all
4926	 * in-memory chunks are synced to disk so that the loop below sees them
4927	 * and relocates them accordingly.
4928	 */
4929	if (contains_pending_extent(device, &start, diff)) {
4930		mutex_unlock(&fs_info->chunk_mutex);
4931		ret = btrfs_commit_transaction(trans);
4932		if (ret)
4933			goto done;
4934	} else {
4935		mutex_unlock(&fs_info->chunk_mutex);
4936		btrfs_end_transaction(trans);
4937	}
4938
4939again:
4940	key.objectid = device->devid;
4941	key.offset = (u64)-1;
4942	key.type = BTRFS_DEV_EXTENT_KEY;
4943
4944	do {
4945		mutex_lock(&fs_info->reclaim_bgs_lock);
4946		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4947		if (ret < 0) {
4948			mutex_unlock(&fs_info->reclaim_bgs_lock);
4949			goto done;
4950		}
4951
4952		ret = btrfs_previous_item(root, path, 0, key.type);
4953		if (ret) {
4954			mutex_unlock(&fs_info->reclaim_bgs_lock);
4955			if (ret < 0)
4956				goto done;
4957			ret = 0;
4958			btrfs_release_path(path);
4959			break;
4960		}
4961
4962		l = path->nodes[0];
4963		slot = path->slots[0];
4964		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4965
4966		if (key.objectid != device->devid) {
4967			mutex_unlock(&fs_info->reclaim_bgs_lock);
4968			btrfs_release_path(path);
4969			break;
4970		}
4971
4972		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4973		length = btrfs_dev_extent_length(l, dev_extent);
4974
4975		if (key.offset + length <= new_size) {
4976			mutex_unlock(&fs_info->reclaim_bgs_lock);
4977			btrfs_release_path(path);
4978			break;
4979		}
4980
4981		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4982		btrfs_release_path(path);
4983
4984		/*
4985		 * We may be relocating the only data chunk we have,
4986		 * which could potentially end up with losing data's
4987		 * raid profile, so lets allocate an empty one in
4988		 * advance.
4989		 */
4990		ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4991		if (ret < 0) {
4992			mutex_unlock(&fs_info->reclaim_bgs_lock);
4993			goto done;
4994		}
4995
4996		ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4997		mutex_unlock(&fs_info->reclaim_bgs_lock);
4998		if (ret == -ENOSPC) {
4999			failed++;
5000		} else if (ret) {
5001			if (ret == -ETXTBSY) {
5002				btrfs_warn(fs_info,
5003		   "could not shrink block group %llu due to active swapfile",
5004					   chunk_offset);
5005			}
5006			goto done;
5007		}
5008	} while (key.offset-- > 0);
5009
5010	if (failed && !retried) {
5011		failed = 0;
5012		retried = true;
5013		goto again;
5014	} else if (failed && retried) {
5015		ret = -ENOSPC;
5016		goto done;
5017	}
5018
5019	/* Shrinking succeeded, else we would be at "done". */
5020	trans = btrfs_start_transaction(root, 0);
5021	if (IS_ERR(trans)) {
5022		ret = PTR_ERR(trans);
5023		goto done;
5024	}
5025
5026	mutex_lock(&fs_info->chunk_mutex);
5027	/* Clear all state bits beyond the shrunk device size */
5028	clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
5029			  CHUNK_STATE_MASK);
5030
5031	btrfs_device_set_disk_total_bytes(device, new_size);
5032	if (list_empty(&device->post_commit_list))
5033		list_add_tail(&device->post_commit_list,
5034			      &trans->transaction->dev_update_list);
5035
5036	WARN_ON(diff > old_total);
5037	btrfs_set_super_total_bytes(super_copy,
5038			round_down(old_total - diff, fs_info->sectorsize));
5039	mutex_unlock(&fs_info->chunk_mutex);
5040
5041	btrfs_reserve_chunk_metadata(trans, false);
5042	/* Now btrfs_update_device() will change the on-disk size. */
5043	ret = btrfs_update_device(trans, device);
5044	btrfs_trans_release_chunk_metadata(trans);
5045	if (ret < 0) {
5046		btrfs_abort_transaction(trans, ret);
5047		btrfs_end_transaction(trans);
5048	} else {
5049		ret = btrfs_commit_transaction(trans);
5050	}
5051done:
5052	btrfs_free_path(path);
5053	if (ret) {
5054		mutex_lock(&fs_info->chunk_mutex);
5055		btrfs_device_set_total_bytes(device, old_size);
5056		if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5057			device->fs_devices->total_rw_bytes += diff;
5058			atomic64_add(free_diff, &fs_info->free_chunk_space);
5059		}
5060		mutex_unlock(&fs_info->chunk_mutex);
5061	}
5062	return ret;
5063}
5064
5065static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5066			   struct btrfs_key *key,
5067			   struct btrfs_chunk *chunk, int item_size)
5068{
5069	struct btrfs_super_block *super_copy = fs_info->super_copy;
5070	struct btrfs_disk_key disk_key;
5071	u32 array_size;
5072	u8 *ptr;
5073
5074	lockdep_assert_held(&fs_info->chunk_mutex);
5075
5076	array_size = btrfs_super_sys_array_size(super_copy);
5077	if (array_size + item_size + sizeof(disk_key)
5078			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5079		return -EFBIG;
5080
5081	ptr = super_copy->sys_chunk_array + array_size;
5082	btrfs_cpu_key_to_disk(&disk_key, key);
5083	memcpy(ptr, &disk_key, sizeof(disk_key));
5084	ptr += sizeof(disk_key);
5085	memcpy(ptr, chunk, item_size);
5086	item_size += sizeof(disk_key);
5087	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5088
5089	return 0;
5090}
5091
5092/*
5093 * sort the devices in descending order by max_avail, total_avail
5094 */
5095static int btrfs_cmp_device_info(const void *a, const void *b)
5096{
5097	const struct btrfs_device_info *di_a = a;
5098	const struct btrfs_device_info *di_b = b;
5099
5100	if (di_a->max_avail > di_b->max_avail)
5101		return -1;
5102	if (di_a->max_avail < di_b->max_avail)
5103		return 1;
5104	if (di_a->total_avail > di_b->total_avail)
5105		return -1;
5106	if (di_a->total_avail < di_b->total_avail)
5107		return 1;
5108	return 0;
5109}
5110
5111static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5112{
5113	if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5114		return;
5115
5116	btrfs_set_fs_incompat(info, RAID56);
5117}
5118
5119static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5120{
5121	if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5122		return;
5123
5124	btrfs_set_fs_incompat(info, RAID1C34);
5125}
5126
5127/*
5128 * Structure used internally for btrfs_create_chunk() function.
5129 * Wraps needed parameters.
5130 */
5131struct alloc_chunk_ctl {
5132	u64 start;
5133	u64 type;
5134	/* Total number of stripes to allocate */
5135	int num_stripes;
5136	/* sub_stripes info for map */
5137	int sub_stripes;
5138	/* Stripes per device */
5139	int dev_stripes;
5140	/* Maximum number of devices to use */
5141	int devs_max;
5142	/* Minimum number of devices to use */
5143	int devs_min;
5144	/* ndevs has to be a multiple of this */
5145	int devs_increment;
5146	/* Number of copies */
5147	int ncopies;
5148	/* Number of stripes worth of bytes to store parity information */
5149	int nparity;
5150	u64 max_stripe_size;
5151	u64 max_chunk_size;
5152	u64 dev_extent_min;
5153	u64 stripe_size;
5154	u64 chunk_size;
5155	int ndevs;
5156};
5157
5158static void init_alloc_chunk_ctl_policy_regular(
5159				struct btrfs_fs_devices *fs_devices,
5160				struct alloc_chunk_ctl *ctl)
5161{
5162	struct btrfs_space_info *space_info;
5163
5164	space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5165	ASSERT(space_info);
5166
5167	ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5168	ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G);
5169
5170	if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5171		ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5172
5173	/* We don't want a chunk larger than 10% of writable space */
5174	ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5175				  ctl->max_chunk_size);
5176	ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5177}
5178
5179static void init_alloc_chunk_ctl_policy_zoned(
5180				      struct btrfs_fs_devices *fs_devices,
5181				      struct alloc_chunk_ctl *ctl)
5182{
5183	u64 zone_size = fs_devices->fs_info->zone_size;
5184	u64 limit;
5185	int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5186	int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5187	u64 min_chunk_size = min_data_stripes * zone_size;
5188	u64 type = ctl->type;
5189
5190	ctl->max_stripe_size = zone_size;
5191	if (type & BTRFS_BLOCK_GROUP_DATA) {
5192		ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5193						 zone_size);
5194	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5195		ctl->max_chunk_size = ctl->max_stripe_size;
5196	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5197		ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5198		ctl->devs_max = min_t(int, ctl->devs_max,
5199				      BTRFS_MAX_DEVS_SYS_CHUNK);
5200	} else {
5201		BUG();
5202	}
5203
5204	/* We don't want a chunk larger than 10% of writable space */
5205	limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5206			       zone_size),
5207		    min_chunk_size);
5208	ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5209	ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5210}
5211
5212static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5213				 struct alloc_chunk_ctl *ctl)
5214{
5215	int index = btrfs_bg_flags_to_raid_index(ctl->type);
5216
5217	ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5218	ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5219	ctl->devs_max = btrfs_raid_array[index].devs_max;
5220	if (!ctl->devs_max)
5221		ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5222	ctl->devs_min = btrfs_raid_array[index].devs_min;
5223	ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5224	ctl->ncopies = btrfs_raid_array[index].ncopies;
5225	ctl->nparity = btrfs_raid_array[index].nparity;
5226	ctl->ndevs = 0;
5227
5228	switch (fs_devices->chunk_alloc_policy) {
5229	case BTRFS_CHUNK_ALLOC_REGULAR:
5230		init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5231		break;
5232	case BTRFS_CHUNK_ALLOC_ZONED:
5233		init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5234		break;
5235	default:
5236		BUG();
5237	}
5238}
5239
5240static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5241			      struct alloc_chunk_ctl *ctl,
5242			      struct btrfs_device_info *devices_info)
5243{
5244	struct btrfs_fs_info *info = fs_devices->fs_info;
5245	struct btrfs_device *device;
5246	u64 total_avail;
5247	u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5248	int ret;
5249	int ndevs = 0;
5250	u64 max_avail;
5251	u64 dev_offset;
5252
5253	/*
5254	 * in the first pass through the devices list, we gather information
5255	 * about the available holes on each device.
5256	 */
5257	list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5258		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5259			WARN(1, KERN_ERR
5260			       "BTRFS: read-only device in alloc_list\n");
5261			continue;
5262		}
5263
5264		if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5265					&device->dev_state) ||
5266		    test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5267			continue;
5268
5269		if (device->total_bytes > device->bytes_used)
5270			total_avail = device->total_bytes - device->bytes_used;
5271		else
5272			total_avail = 0;
5273
5274		/* If there is no space on this device, skip it. */
5275		if (total_avail < ctl->dev_extent_min)
5276			continue;
5277
5278		ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5279					   &max_avail);
5280		if (ret && ret != -ENOSPC)
5281			return ret;
5282
5283		if (ret == 0)
5284			max_avail = dev_extent_want;
5285
5286		if (max_avail < ctl->dev_extent_min) {
5287			if (btrfs_test_opt(info, ENOSPC_DEBUG))
5288				btrfs_debug(info,
5289			"%s: devid %llu has no free space, have=%llu want=%llu",
5290					    __func__, device->devid, max_avail,
5291					    ctl->dev_extent_min);
5292			continue;
5293		}
5294
5295		if (ndevs == fs_devices->rw_devices) {
5296			WARN(1, "%s: found more than %llu devices\n",
5297			     __func__, fs_devices->rw_devices);
5298			break;
5299		}
5300		devices_info[ndevs].dev_offset = dev_offset;
5301		devices_info[ndevs].max_avail = max_avail;
5302		devices_info[ndevs].total_avail = total_avail;
5303		devices_info[ndevs].dev = device;
5304		++ndevs;
5305	}
5306	ctl->ndevs = ndevs;
5307
5308	/*
5309	 * now sort the devices by hole size / available space
5310	 */
5311	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5312	     btrfs_cmp_device_info, NULL);
5313
5314	return 0;
5315}
5316
5317static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5318				      struct btrfs_device_info *devices_info)
5319{
5320	/* Number of stripes that count for block group size */
5321	int data_stripes;
5322
5323	/*
5324	 * The primary goal is to maximize the number of stripes, so use as
5325	 * many devices as possible, even if the stripes are not maximum sized.
5326	 *
5327	 * The DUP profile stores more than one stripe per device, the
5328	 * max_avail is the total size so we have to adjust.
5329	 */
5330	ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5331				   ctl->dev_stripes);
5332	ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5333
5334	/* This will have to be fixed for RAID1 and RAID10 over more drives */
5335	data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5336
5337	/*
5338	 * Use the number of data stripes to figure out how big this chunk is
5339	 * really going to be in terms of logical address space, and compare
5340	 * that answer with the max chunk size. If it's higher, we try to
5341	 * reduce stripe_size.
5342	 */
5343	if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5344		/*
5345		 * Reduce stripe_size, round it up to a 16MB boundary again and
5346		 * then use it, unless it ends up being even bigger than the
5347		 * previous value we had already.
5348		 */
5349		ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5350							data_stripes), SZ_16M),
5351				       ctl->stripe_size);
5352	}
5353
5354	/* Stripe size should not go beyond 1G. */
5355	ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5356
5357	/* Align to BTRFS_STRIPE_LEN */
5358	ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5359	ctl->chunk_size = ctl->stripe_size * data_stripes;
5360
5361	return 0;
5362}
5363
5364static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5365				    struct btrfs_device_info *devices_info)
5366{
5367	u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5368	/* Number of stripes that count for block group size */
5369	int data_stripes;
5370
5371	/*
5372	 * It should hold because:
5373	 *    dev_extent_min == dev_extent_want == zone_size * dev_stripes
5374	 */
5375	ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5376
5377	ctl->stripe_size = zone_size;
5378	ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5379	data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5380
5381	/* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5382	if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5383		ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5384					     ctl->stripe_size) + ctl->nparity,
5385				     ctl->dev_stripes);
5386		ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5387		data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5388		ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5389	}
5390
5391	ctl->chunk_size = ctl->stripe_size * data_stripes;
5392
5393	return 0;
5394}
5395
5396static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5397			      struct alloc_chunk_ctl *ctl,
5398			      struct btrfs_device_info *devices_info)
5399{
5400	struct btrfs_fs_info *info = fs_devices->fs_info;
5401
5402	/*
5403	 * Round down to number of usable stripes, devs_increment can be any
5404	 * number so we can't use round_down() that requires power of 2, while
5405	 * rounddown is safe.
5406	 */
5407	ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5408
5409	if (ctl->ndevs < ctl->devs_min) {
5410		if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5411			btrfs_debug(info,
5412	"%s: not enough devices with free space: have=%d minimum required=%d",
5413				    __func__, ctl->ndevs, ctl->devs_min);
5414		}
5415		return -ENOSPC;
5416	}
5417
5418	ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5419
5420	switch (fs_devices->chunk_alloc_policy) {
5421	case BTRFS_CHUNK_ALLOC_REGULAR:
5422		return decide_stripe_size_regular(ctl, devices_info);
5423	case BTRFS_CHUNK_ALLOC_ZONED:
5424		return decide_stripe_size_zoned(ctl, devices_info);
5425	default:
5426		BUG();
5427	}
5428}
5429
5430static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits)
5431{
5432	for (int i = 0; i < map->num_stripes; i++) {
5433		struct btrfs_io_stripe *stripe = &map->stripes[i];
5434		struct btrfs_device *device = stripe->dev;
5435
5436		set_extent_bit(&device->alloc_state, stripe->physical,
5437			       stripe->physical + map->stripe_size - 1,
5438			       bits | EXTENT_NOWAIT, NULL);
5439	}
5440}
5441
5442static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits)
5443{
5444	for (int i = 0; i < map->num_stripes; i++) {
5445		struct btrfs_io_stripe *stripe = &map->stripes[i];
5446		struct btrfs_device *device = stripe->dev;
5447
5448		__clear_extent_bit(&device->alloc_state, stripe->physical,
5449				   stripe->physical + map->stripe_size - 1,
5450				   bits | EXTENT_NOWAIT,
5451				   NULL, NULL);
5452	}
5453}
5454
5455void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5456{
5457	write_lock(&fs_info->mapping_tree_lock);
5458	rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5459	RB_CLEAR_NODE(&map->rb_node);
5460	chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5461	write_unlock(&fs_info->mapping_tree_lock);
5462
5463	/* Once for the tree reference. */
5464	btrfs_free_chunk_map(map);
5465}
5466
5467EXPORT_FOR_TESTS
5468int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5469{
5470	struct rb_node **p;
5471	struct rb_node *parent = NULL;
5472	bool leftmost = true;
5473
5474	write_lock(&fs_info->mapping_tree_lock);
5475	p = &fs_info->mapping_tree.rb_root.rb_node;
5476	while (*p) {
5477		struct btrfs_chunk_map *entry;
5478
5479		parent = *p;
5480		entry = rb_entry(parent, struct btrfs_chunk_map, rb_node);
5481
5482		if (map->start < entry->start) {
5483			p = &(*p)->rb_left;
5484		} else if (map->start > entry->start) {
5485			p = &(*p)->rb_right;
5486			leftmost = false;
5487		} else {
5488			write_unlock(&fs_info->mapping_tree_lock);
5489			return -EEXIST;
5490		}
5491	}
5492	rb_link_node(&map->rb_node, parent, p);
5493	rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost);
5494	chunk_map_device_set_bits(map, CHUNK_ALLOCATED);
5495	chunk_map_device_clear_bits(map, CHUNK_TRIMMED);
5496	write_unlock(&fs_info->mapping_tree_lock);
5497
5498	return 0;
5499}
5500
5501EXPORT_FOR_TESTS
5502struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp)
5503{
5504	struct btrfs_chunk_map *map;
5505
5506	map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp);
5507	if (!map)
5508		return NULL;
5509
5510	refcount_set(&map->refs, 1);
5511	RB_CLEAR_NODE(&map->rb_node);
5512
5513	return map;
5514}
5515
5516struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp)
5517{
5518	const int size = btrfs_chunk_map_size(map->num_stripes);
5519	struct btrfs_chunk_map *clone;
5520
5521	clone = kmemdup(map, size, gfp);
5522	if (!clone)
5523		return NULL;
5524
5525	refcount_set(&clone->refs, 1);
5526	RB_CLEAR_NODE(&clone->rb_node);
5527
5528	return clone;
5529}
5530
5531static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5532			struct alloc_chunk_ctl *ctl,
5533			struct btrfs_device_info *devices_info)
5534{
5535	struct btrfs_fs_info *info = trans->fs_info;
5536	struct btrfs_chunk_map *map;
5537	struct btrfs_block_group *block_group;
5538	u64 start = ctl->start;
5539	u64 type = ctl->type;
5540	int ret;
5541	int i;
5542	int j;
5543
5544	map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS);
5545	if (!map)
5546		return ERR_PTR(-ENOMEM);
5547
5548	map->start = start;
5549	map->chunk_len = ctl->chunk_size;
5550	map->stripe_size = ctl->stripe_size;
5551	map->type = type;
5552	map->io_align = BTRFS_STRIPE_LEN;
5553	map->io_width = BTRFS_STRIPE_LEN;
5554	map->sub_stripes = ctl->sub_stripes;
5555	map->num_stripes = ctl->num_stripes;
5556
5557	for (i = 0; i < ctl->ndevs; ++i) {
5558		for (j = 0; j < ctl->dev_stripes; ++j) {
5559			int s = i * ctl->dev_stripes + j;
5560			map->stripes[s].dev = devices_info[i].dev;
5561			map->stripes[s].physical = devices_info[i].dev_offset +
5562						   j * ctl->stripe_size;
5563		}
5564	}
5565
5566	trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5567
5568	ret = btrfs_add_chunk_map(info, map);
5569	if (ret) {
5570		btrfs_free_chunk_map(map);
5571		return ERR_PTR(ret);
5572	}
5573
5574	block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5575	if (IS_ERR(block_group)) {
5576		btrfs_remove_chunk_map(info, map);
5577		return block_group;
5578	}
5579
5580	for (int i = 0; i < map->num_stripes; i++) {
5581		struct btrfs_device *dev = map->stripes[i].dev;
5582
5583		btrfs_device_set_bytes_used(dev,
5584					    dev->bytes_used + ctl->stripe_size);
5585		if (list_empty(&dev->post_commit_list))
5586			list_add_tail(&dev->post_commit_list,
5587				      &trans->transaction->dev_update_list);
5588	}
5589
5590	atomic64_sub(ctl->stripe_size * map->num_stripes,
5591		     &info->free_chunk_space);
5592
5593	check_raid56_incompat_flag(info, type);
5594	check_raid1c34_incompat_flag(info, type);
5595
5596	return block_group;
5597}
5598
5599struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5600					    u64 type)
5601{
5602	struct btrfs_fs_info *info = trans->fs_info;
5603	struct btrfs_fs_devices *fs_devices = info->fs_devices;
5604	struct btrfs_device_info *devices_info = NULL;
5605	struct alloc_chunk_ctl ctl;
5606	struct btrfs_block_group *block_group;
5607	int ret;
5608
5609	lockdep_assert_held(&info->chunk_mutex);
5610
5611	if (!alloc_profile_is_valid(type, 0)) {
5612		ASSERT(0);
5613		return ERR_PTR(-EINVAL);
5614	}
5615
5616	if (list_empty(&fs_devices->alloc_list)) {
5617		if (btrfs_test_opt(info, ENOSPC_DEBUG))
5618			btrfs_debug(info, "%s: no writable device", __func__);
5619		return ERR_PTR(-ENOSPC);
5620	}
5621
5622	if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5623		btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5624		ASSERT(0);
5625		return ERR_PTR(-EINVAL);
5626	}
5627
5628	ctl.start = find_next_chunk(info);
5629	ctl.type = type;
5630	init_alloc_chunk_ctl(fs_devices, &ctl);
5631
5632	devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5633			       GFP_NOFS);
5634	if (!devices_info)
5635		return ERR_PTR(-ENOMEM);
5636
5637	ret = gather_device_info(fs_devices, &ctl, devices_info);
5638	if (ret < 0) {
5639		block_group = ERR_PTR(ret);
5640		goto out;
5641	}
5642
5643	ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5644	if (ret < 0) {
5645		block_group = ERR_PTR(ret);
5646		goto out;
5647	}
5648
5649	block_group = create_chunk(trans, &ctl, devices_info);
5650
5651out:
5652	kfree(devices_info);
5653	return block_group;
5654}
5655
5656/*
5657 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5658 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5659 * chunks.
5660 *
5661 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5662 * phases.
5663 */
5664int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5665				     struct btrfs_block_group *bg)
5666{
5667	struct btrfs_fs_info *fs_info = trans->fs_info;
5668	struct btrfs_root *chunk_root = fs_info->chunk_root;
5669	struct btrfs_key key;
5670	struct btrfs_chunk *chunk;
5671	struct btrfs_stripe *stripe;
5672	struct btrfs_chunk_map *map;
5673	size_t item_size;
5674	int i;
5675	int ret;
5676
5677	/*
5678	 * We take the chunk_mutex for 2 reasons:
5679	 *
5680	 * 1) Updates and insertions in the chunk btree must be done while holding
5681	 *    the chunk_mutex, as well as updating the system chunk array in the
5682	 *    superblock. See the comment on top of btrfs_chunk_alloc() for the
5683	 *    details;
5684	 *
5685	 * 2) To prevent races with the final phase of a device replace operation
5686	 *    that replaces the device object associated with the map's stripes,
5687	 *    because the device object's id can change at any time during that
5688	 *    final phase of the device replace operation
5689	 *    (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5690	 *    replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5691	 *    which would cause a failure when updating the device item, which does
5692	 *    not exists, or persisting a stripe of the chunk item with such ID.
5693	 *    Here we can't use the device_list_mutex because our caller already
5694	 *    has locked the chunk_mutex, and the final phase of device replace
5695	 *    acquires both mutexes - first the device_list_mutex and then the
5696	 *    chunk_mutex. Using any of those two mutexes protects us from a
5697	 *    concurrent device replace.
5698	 */
5699	lockdep_assert_held(&fs_info->chunk_mutex);
5700
5701	map = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5702	if (IS_ERR(map)) {
5703		ret = PTR_ERR(map);
5704		btrfs_abort_transaction(trans, ret);
5705		return ret;
5706	}
5707
5708	item_size = btrfs_chunk_item_size(map->num_stripes);
5709
5710	chunk = kzalloc(item_size, GFP_NOFS);
5711	if (!chunk) {
5712		ret = -ENOMEM;
5713		btrfs_abort_transaction(trans, ret);
5714		goto out;
5715	}
5716
5717	for (i = 0; i < map->num_stripes; i++) {
5718		struct btrfs_device *device = map->stripes[i].dev;
5719
5720		ret = btrfs_update_device(trans, device);
5721		if (ret)
5722			goto out;
5723	}
5724
5725	stripe = &chunk->stripe;
5726	for (i = 0; i < map->num_stripes; i++) {
5727		struct btrfs_device *device = map->stripes[i].dev;
5728		const u64 dev_offset = map->stripes[i].physical;
5729
5730		btrfs_set_stack_stripe_devid(stripe, device->devid);
5731		btrfs_set_stack_stripe_offset(stripe, dev_offset);
5732		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5733		stripe++;
5734	}
5735
5736	btrfs_set_stack_chunk_length(chunk, bg->length);
5737	btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5738	btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5739	btrfs_set_stack_chunk_type(chunk, map->type);
5740	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5741	btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5742	btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5743	btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5744	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5745
5746	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5747	key.type = BTRFS_CHUNK_ITEM_KEY;
5748	key.offset = bg->start;
5749
5750	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5751	if (ret)
5752		goto out;
5753
5754	set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5755
5756	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5757		ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5758		if (ret)
5759			goto out;
5760	}
5761
5762out:
5763	kfree(chunk);
5764	btrfs_free_chunk_map(map);
5765	return ret;
5766}
5767
5768static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5769{
5770	struct btrfs_fs_info *fs_info = trans->fs_info;
5771	u64 alloc_profile;
5772	struct btrfs_block_group *meta_bg;
5773	struct btrfs_block_group *sys_bg;
5774
5775	/*
5776	 * When adding a new device for sprouting, the seed device is read-only
5777	 * so we must first allocate a metadata and a system chunk. But before
5778	 * adding the block group items to the extent, device and chunk btrees,
5779	 * we must first:
5780	 *
5781	 * 1) Create both chunks without doing any changes to the btrees, as
5782	 *    otherwise we would get -ENOSPC since the block groups from the
5783	 *    seed device are read-only;
5784	 *
5785	 * 2) Add the device item for the new sprout device - finishing the setup
5786	 *    of a new block group requires updating the device item in the chunk
5787	 *    btree, so it must exist when we attempt to do it. The previous step
5788	 *    ensures this does not fail with -ENOSPC.
5789	 *
5790	 * After that we can add the block group items to their btrees:
5791	 * update existing device item in the chunk btree, add a new block group
5792	 * item to the extent btree, add a new chunk item to the chunk btree and
5793	 * finally add the new device extent items to the devices btree.
5794	 */
5795
5796	alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5797	meta_bg = btrfs_create_chunk(trans, alloc_profile);
5798	if (IS_ERR(meta_bg))
5799		return PTR_ERR(meta_bg);
5800
5801	alloc_profile = btrfs_system_alloc_profile(fs_info);
5802	sys_bg = btrfs_create_chunk(trans, alloc_profile);
5803	if (IS_ERR(sys_bg))
5804		return PTR_ERR(sys_bg);
5805
5806	return 0;
5807}
5808
5809static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map)
5810{
5811	const int index = btrfs_bg_flags_to_raid_index(map->type);
5812
5813	return btrfs_raid_array[index].tolerated_failures;
5814}
5815
5816bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5817{
5818	struct btrfs_chunk_map *map;
5819	int miss_ndevs = 0;
5820	int i;
5821	bool ret = true;
5822
5823	map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5824	if (IS_ERR(map))
5825		return false;
5826
5827	for (i = 0; i < map->num_stripes; i++) {
5828		if (test_bit(BTRFS_DEV_STATE_MISSING,
5829					&map->stripes[i].dev->dev_state)) {
5830			miss_ndevs++;
5831			continue;
5832		}
5833		if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5834					&map->stripes[i].dev->dev_state)) {
5835			ret = false;
5836			goto end;
5837		}
5838	}
5839
5840	/*
5841	 * If the number of missing devices is larger than max errors, we can
5842	 * not write the data into that chunk successfully.
5843	 */
5844	if (miss_ndevs > btrfs_chunk_max_errors(map))
5845		ret = false;
5846end:
5847	btrfs_free_chunk_map(map);
5848	return ret;
5849}
5850
5851void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info)
5852{
5853	write_lock(&fs_info->mapping_tree_lock);
5854	while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) {
5855		struct btrfs_chunk_map *map;
5856		struct rb_node *node;
5857
5858		node = rb_first_cached(&fs_info->mapping_tree);
5859		map = rb_entry(node, struct btrfs_chunk_map, rb_node);
5860		rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5861		RB_CLEAR_NODE(&map->rb_node);
5862		chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5863		/* Once for the tree ref. */
5864		btrfs_free_chunk_map(map);
5865		cond_resched_rwlock_write(&fs_info->mapping_tree_lock);
5866	}
5867	write_unlock(&fs_info->mapping_tree_lock);
5868}
5869
5870int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5871{
5872	struct btrfs_chunk_map *map;
5873	enum btrfs_raid_types index;
5874	int ret = 1;
5875
5876	map = btrfs_get_chunk_map(fs_info, logical, len);
5877	if (IS_ERR(map))
5878		/*
5879		 * We could return errors for these cases, but that could get
5880		 * ugly and we'd probably do the same thing which is just not do
5881		 * anything else and exit, so return 1 so the callers don't try
5882		 * to use other copies.
5883		 */
5884		return 1;
5885
5886	index = btrfs_bg_flags_to_raid_index(map->type);
5887
5888	/* Non-RAID56, use their ncopies from btrfs_raid_array. */
5889	if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5890		ret = btrfs_raid_array[index].ncopies;
5891	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5892		ret = 2;
5893	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5894		/*
5895		 * There could be two corrupted data stripes, we need
5896		 * to loop retry in order to rebuild the correct data.
5897		 *
5898		 * Fail a stripe at a time on every retry except the
5899		 * stripe under reconstruction.
5900		 */
5901		ret = map->num_stripes;
5902	btrfs_free_chunk_map(map);
5903	return ret;
5904}
5905
5906unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5907				    u64 logical)
5908{
5909	struct btrfs_chunk_map *map;
5910	unsigned long len = fs_info->sectorsize;
5911
5912	if (!btrfs_fs_incompat(fs_info, RAID56))
5913		return len;
5914
5915	map = btrfs_get_chunk_map(fs_info, logical, len);
5916
5917	if (!WARN_ON(IS_ERR(map))) {
5918		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5919			len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
5920		btrfs_free_chunk_map(map);
5921	}
5922	return len;
5923}
5924
5925int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5926{
5927	struct btrfs_chunk_map *map;
5928	int ret = 0;
5929
5930	if (!btrfs_fs_incompat(fs_info, RAID56))
5931		return 0;
5932
5933	map = btrfs_get_chunk_map(fs_info, logical, len);
5934
5935	if (!WARN_ON(IS_ERR(map))) {
5936		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5937			ret = 1;
5938		btrfs_free_chunk_map(map);
5939	}
5940	return ret;
5941}
5942
5943static int find_live_mirror(struct btrfs_fs_info *fs_info,
5944			    struct btrfs_chunk_map *map, int first,
5945			    int dev_replace_is_ongoing)
5946{
5947	int i;
5948	int num_stripes;
5949	int preferred_mirror;
5950	int tolerance;
5951	struct btrfs_device *srcdev;
5952
5953	ASSERT((map->type &
5954		 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5955
5956	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5957		num_stripes = map->sub_stripes;
5958	else
5959		num_stripes = map->num_stripes;
5960
5961	switch (fs_info->fs_devices->read_policy) {
5962	default:
5963		/* Shouldn't happen, just warn and use pid instead of failing */
5964		btrfs_warn_rl(fs_info,
5965			      "unknown read_policy type %u, reset to pid",
5966			      fs_info->fs_devices->read_policy);
5967		fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5968		fallthrough;
5969	case BTRFS_READ_POLICY_PID:
5970		preferred_mirror = first + (current->pid % num_stripes);
5971		break;
5972	}
5973
5974	if (dev_replace_is_ongoing &&
5975	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5976	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5977		srcdev = fs_info->dev_replace.srcdev;
5978	else
5979		srcdev = NULL;
5980
5981	/*
5982	 * try to avoid the drive that is the source drive for a
5983	 * dev-replace procedure, only choose it if no other non-missing
5984	 * mirror is available
5985	 */
5986	for (tolerance = 0; tolerance < 2; tolerance++) {
5987		if (map->stripes[preferred_mirror].dev->bdev &&
5988		    (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5989			return preferred_mirror;
5990		for (i = first; i < first + num_stripes; i++) {
5991			if (map->stripes[i].dev->bdev &&
5992			    (tolerance || map->stripes[i].dev != srcdev))
5993				return i;
5994		}
5995	}
5996
5997	/* we couldn't find one that doesn't fail.  Just return something
5998	 * and the io error handling code will clean up eventually
5999	 */
6000	return preferred_mirror;
6001}
6002
6003static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
6004						       u64 logical,
6005						       u16 total_stripes)
6006{
6007	struct btrfs_io_context *bioc;
6008
6009	bioc = kzalloc(
6010		 /* The size of btrfs_io_context */
6011		sizeof(struct btrfs_io_context) +
6012		/* Plus the variable array for the stripes */
6013		sizeof(struct btrfs_io_stripe) * (total_stripes),
6014		GFP_NOFS);
6015
6016	if (!bioc)
6017		return NULL;
6018
6019	refcount_set(&bioc->refs, 1);
6020
6021	bioc->fs_info = fs_info;
6022	bioc->replace_stripe_src = -1;
6023	bioc->full_stripe_logical = (u64)-1;
6024	bioc->logical = logical;
6025
6026	return bioc;
6027}
6028
6029void btrfs_get_bioc(struct btrfs_io_context *bioc)
6030{
6031	WARN_ON(!refcount_read(&bioc->refs));
6032	refcount_inc(&bioc->refs);
6033}
6034
6035void btrfs_put_bioc(struct btrfs_io_context *bioc)
6036{
6037	if (!bioc)
6038		return;
6039	if (refcount_dec_and_test(&bioc->refs))
6040		kfree(bioc);
6041}
6042
6043/*
6044 * Please note that, discard won't be sent to target device of device
6045 * replace.
6046 */
6047struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
6048					       u64 logical, u64 *length_ret,
6049					       u32 *num_stripes)
6050{
6051	struct btrfs_chunk_map *map;
6052	struct btrfs_discard_stripe *stripes;
6053	u64 length = *length_ret;
6054	u64 offset;
6055	u32 stripe_nr;
6056	u32 stripe_nr_end;
6057	u32 stripe_cnt;
6058	u64 stripe_end_offset;
6059	u64 stripe_offset;
6060	u32 stripe_index;
6061	u32 factor = 0;
6062	u32 sub_stripes = 0;
6063	u32 stripes_per_dev = 0;
6064	u32 remaining_stripes = 0;
6065	u32 last_stripe = 0;
6066	int ret;
6067	int i;
6068
6069	map = btrfs_get_chunk_map(fs_info, logical, length);
6070	if (IS_ERR(map))
6071		return ERR_CAST(map);
6072
6073	/* we don't discard raid56 yet */
6074	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6075		ret = -EOPNOTSUPP;
6076		goto out_free_map;
6077	}
6078
6079	offset = logical - map->start;
6080	length = min_t(u64, map->start + map->chunk_len - logical, length);
6081	*length_ret = length;
6082
6083	/*
6084	 * stripe_nr counts the total number of stripes we have to stride
6085	 * to get to this block
6086	 */
6087	stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6088
6089	/* stripe_offset is the offset of this block in its stripe */
6090	stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
6091
6092	stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
6093			BTRFS_STRIPE_LEN_SHIFT;
6094	stripe_cnt = stripe_nr_end - stripe_nr;
6095	stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
6096			    (offset + length);
6097	/*
6098	 * after this, stripe_nr is the number of stripes on this
6099	 * device we have to walk to find the data, and stripe_index is
6100	 * the number of our device in the stripe array
6101	 */
6102	*num_stripes = 1;
6103	stripe_index = 0;
6104	if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6105			 BTRFS_BLOCK_GROUP_RAID10)) {
6106		if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6107			sub_stripes = 1;
6108		else
6109			sub_stripes = map->sub_stripes;
6110
6111		factor = map->num_stripes / sub_stripes;
6112		*num_stripes = min_t(u64, map->num_stripes,
6113				    sub_stripes * stripe_cnt);
6114		stripe_index = stripe_nr % factor;
6115		stripe_nr /= factor;
6116		stripe_index *= sub_stripes;
6117
6118		remaining_stripes = stripe_cnt % factor;
6119		stripes_per_dev = stripe_cnt / factor;
6120		last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6121	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6122				BTRFS_BLOCK_GROUP_DUP)) {
6123		*num_stripes = map->num_stripes;
6124	} else {
6125		stripe_index = stripe_nr % map->num_stripes;
6126		stripe_nr /= map->num_stripes;
6127	}
6128
6129	stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6130	if (!stripes) {
6131		ret = -ENOMEM;
6132		goto out_free_map;
6133	}
6134
6135	for (i = 0; i < *num_stripes; i++) {
6136		stripes[i].physical =
6137			map->stripes[stripe_index].physical +
6138			stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6139		stripes[i].dev = map->stripes[stripe_index].dev;
6140
6141		if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6142				 BTRFS_BLOCK_GROUP_RAID10)) {
6143			stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6144
6145			if (i / sub_stripes < remaining_stripes)
6146				stripes[i].length += BTRFS_STRIPE_LEN;
6147
6148			/*
6149			 * Special for the first stripe and
6150			 * the last stripe:
6151			 *
6152			 * |-------|...|-------|
6153			 *     |----------|
6154			 *    off     end_off
6155			 */
6156			if (i < sub_stripes)
6157				stripes[i].length -= stripe_offset;
6158
6159			if (stripe_index >= last_stripe &&
6160			    stripe_index <= (last_stripe +
6161					     sub_stripes - 1))
6162				stripes[i].length -= stripe_end_offset;
6163
6164			if (i == sub_stripes - 1)
6165				stripe_offset = 0;
6166		} else {
6167			stripes[i].length = length;
6168		}
6169
6170		stripe_index++;
6171		if (stripe_index == map->num_stripes) {
6172			stripe_index = 0;
6173			stripe_nr++;
6174		}
6175	}
6176
6177	btrfs_free_chunk_map(map);
6178	return stripes;
6179out_free_map:
6180	btrfs_free_chunk_map(map);
6181	return ERR_PTR(ret);
6182}
6183
6184static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6185{
6186	struct btrfs_block_group *cache;
6187	bool ret;
6188
6189	/* Non zoned filesystem does not use "to_copy" flag */
6190	if (!btrfs_is_zoned(fs_info))
6191		return false;
6192
6193	cache = btrfs_lookup_block_group(fs_info, logical);
6194
6195	ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6196
6197	btrfs_put_block_group(cache);
6198	return ret;
6199}
6200
6201static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6202				      struct btrfs_io_context *bioc,
6203				      struct btrfs_dev_replace *dev_replace,
6204				      u64 logical,
6205				      int *num_stripes_ret, int *max_errors_ret)
6206{
6207	u64 srcdev_devid = dev_replace->srcdev->devid;
6208	/*
6209	 * At this stage, num_stripes is still the real number of stripes,
6210	 * excluding the duplicated stripes.
6211	 */
6212	int num_stripes = *num_stripes_ret;
6213	int nr_extra_stripes = 0;
6214	int max_errors = *max_errors_ret;
6215	int i;
6216
6217	/*
6218	 * A block group which has "to_copy" set will eventually be copied by
6219	 * the dev-replace process. We can avoid cloning IO here.
6220	 */
6221	if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6222		return;
6223
6224	/*
6225	 * Duplicate the write operations while the dev-replace procedure is
6226	 * running. Since the copying of the old disk to the new disk takes
6227	 * place at run time while the filesystem is mounted writable, the
6228	 * regular write operations to the old disk have to be duplicated to go
6229	 * to the new disk as well.
6230	 *
6231	 * Note that device->missing is handled by the caller, and that the
6232	 * write to the old disk is already set up in the stripes array.
6233	 */
6234	for (i = 0; i < num_stripes; i++) {
6235		struct btrfs_io_stripe *old = &bioc->stripes[i];
6236		struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6237
6238		if (old->dev->devid != srcdev_devid)
6239			continue;
6240
6241		new->physical = old->physical;
6242		new->dev = dev_replace->tgtdev;
6243		if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6244			bioc->replace_stripe_src = i;
6245		nr_extra_stripes++;
6246	}
6247
6248	/* We can only have at most 2 extra nr_stripes (for DUP). */
6249	ASSERT(nr_extra_stripes <= 2);
6250	/*
6251	 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6252	 * replace.
6253	 * If we have 2 extra stripes, only choose the one with smaller physical.
6254	 */
6255	if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6256		struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6257		struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6258
6259		/* Only DUP can have two extra stripes. */
6260		ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6261
6262		/*
6263		 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6264		 * The extra stripe would still be there, but won't be accessed.
6265		 */
6266		if (first->physical > second->physical) {
6267			swap(second->physical, first->physical);
6268			swap(second->dev, first->dev);
6269			nr_extra_stripes--;
6270		}
6271	}
6272
6273	*num_stripes_ret = num_stripes + nr_extra_stripes;
6274	*max_errors_ret = max_errors + nr_extra_stripes;
6275	bioc->replace_nr_stripes = nr_extra_stripes;
6276}
6277
6278static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, u64 offset,
6279			    struct btrfs_io_geometry *io_geom)
6280{
6281	/*
6282	 * Stripe_nr is the stripe where this block falls.  stripe_offset is
6283	 * the offset of this block in its stripe.
6284	 */
6285	io_geom->stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6286	io_geom->stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6287	ASSERT(io_geom->stripe_offset < U32_MAX);
6288
6289	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6290		unsigned long full_stripe_len =
6291			btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6292
6293		/*
6294		 * For full stripe start, we use previously calculated
6295		 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6296		 * STRIPE_LEN.
6297		 *
6298		 * By this we can avoid u64 division completely.  And we have
6299		 * to go rounddown(), not round_down(), as nr_data_stripes is
6300		 * not ensured to be power of 2.
6301		 */
6302		io_geom->raid56_full_stripe_start = btrfs_stripe_nr_to_offset(
6303			rounddown(io_geom->stripe_nr, nr_data_stripes(map)));
6304
6305		ASSERT(io_geom->raid56_full_stripe_start + full_stripe_len > offset);
6306		ASSERT(io_geom->raid56_full_stripe_start <= offset);
6307		/*
6308		 * For writes to RAID56, allow to write a full stripe set, but
6309		 * no straddling of stripe sets.
6310		 */
6311		if (io_geom->op == BTRFS_MAP_WRITE)
6312			return full_stripe_len - (offset - io_geom->raid56_full_stripe_start);
6313	}
6314
6315	/*
6316	 * For other RAID types and for RAID56 reads, allow a single stripe (on
6317	 * a single disk).
6318	 */
6319	if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6320		return BTRFS_STRIPE_LEN - io_geom->stripe_offset;
6321	return U64_MAX;
6322}
6323
6324static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical,
6325			 u64 *length, struct btrfs_io_stripe *dst,
6326			 struct btrfs_chunk_map *map,
6327			 struct btrfs_io_geometry *io_geom)
6328{
6329	dst->dev = map->stripes[io_geom->stripe_index].dev;
6330
6331	if (io_geom->op == BTRFS_MAP_READ &&
6332	    btrfs_need_stripe_tree_update(fs_info, map->type))
6333		return btrfs_get_raid_extent_offset(fs_info, logical, length,
6334						    map->type,
6335						    io_geom->stripe_index, dst);
6336
6337	dst->physical = map->stripes[io_geom->stripe_index].physical +
6338			io_geom->stripe_offset +
6339			btrfs_stripe_nr_to_offset(io_geom->stripe_nr);
6340	return 0;
6341}
6342
6343static bool is_single_device_io(struct btrfs_fs_info *fs_info,
6344				const struct btrfs_io_stripe *smap,
6345				const struct btrfs_chunk_map *map,
6346				int num_alloc_stripes,
6347				enum btrfs_map_op op, int mirror_num)
6348{
6349	if (!smap)
6350		return false;
6351
6352	if (num_alloc_stripes != 1)
6353		return false;
6354
6355	if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ)
6356		return false;
6357
6358	if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)
6359		return false;
6360
6361	return true;
6362}
6363
6364static void map_blocks_raid0(const struct btrfs_chunk_map *map,
6365			     struct btrfs_io_geometry *io_geom)
6366{
6367	io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6368	io_geom->stripe_nr /= map->num_stripes;
6369	if (io_geom->op == BTRFS_MAP_READ)
6370		io_geom->mirror_num = 1;
6371}
6372
6373static void map_blocks_raid1(struct btrfs_fs_info *fs_info,
6374			     struct btrfs_chunk_map *map,
6375			     struct btrfs_io_geometry *io_geom,
6376			     bool dev_replace_is_ongoing)
6377{
6378	if (io_geom->op != BTRFS_MAP_READ) {
6379		io_geom->num_stripes = map->num_stripes;
6380		return;
6381	}
6382
6383	if (io_geom->mirror_num) {
6384		io_geom->stripe_index = io_geom->mirror_num - 1;
6385		return;
6386	}
6387
6388	io_geom->stripe_index = find_live_mirror(fs_info, map, 0,
6389						 dev_replace_is_ongoing);
6390	io_geom->mirror_num = io_geom->stripe_index + 1;
6391}
6392
6393static void map_blocks_dup(const struct btrfs_chunk_map *map,
6394			   struct btrfs_io_geometry *io_geom)
6395{
6396	if (io_geom->op != BTRFS_MAP_READ) {
6397		io_geom->num_stripes = map->num_stripes;
6398		return;
6399	}
6400
6401	if (io_geom->mirror_num) {
6402		io_geom->stripe_index = io_geom->mirror_num - 1;
6403		return;
6404	}
6405
6406	io_geom->mirror_num = 1;
6407}
6408
6409static void map_blocks_raid10(struct btrfs_fs_info *fs_info,
6410			      struct btrfs_chunk_map *map,
6411			      struct btrfs_io_geometry *io_geom,
6412			      bool dev_replace_is_ongoing)
6413{
6414	u32 factor = map->num_stripes / map->sub_stripes;
6415	int old_stripe_index;
6416
6417	io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes;
6418	io_geom->stripe_nr /= factor;
6419
6420	if (io_geom->op != BTRFS_MAP_READ) {
6421		io_geom->num_stripes = map->sub_stripes;
6422		return;
6423	}
6424
6425	if (io_geom->mirror_num) {
6426		io_geom->stripe_index += io_geom->mirror_num - 1;
6427		return;
6428	}
6429
6430	old_stripe_index = io_geom->stripe_index;
6431	io_geom->stripe_index = find_live_mirror(fs_info, map,
6432						 io_geom->stripe_index,
6433						 dev_replace_is_ongoing);
6434	io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1;
6435}
6436
6437static void map_blocks_raid56_write(struct btrfs_chunk_map *map,
6438				    struct btrfs_io_geometry *io_geom,
6439				    u64 logical, u64 *length)
6440{
6441	int data_stripes = nr_data_stripes(map);
6442
6443	/*
6444	 * Needs full stripe mapping.
6445	 *
6446	 * Push stripe_nr back to the start of the full stripe For those cases
6447	 * needing a full stripe, @stripe_nr is the full stripe number.
6448	 *
6449	 * Originally we go raid56_full_stripe_start / full_stripe_len, but
6450	 * that can be expensive.  Here we just divide @stripe_nr with
6451	 * @data_stripes.
6452	 */
6453	io_geom->stripe_nr /= data_stripes;
6454
6455	/* RAID[56] write or recovery. Return all stripes */
6456	io_geom->num_stripes = map->num_stripes;
6457	io_geom->max_errors = btrfs_chunk_max_errors(map);
6458
6459	/* Return the length to the full stripe end. */
6460	*length = min(logical + *length,
6461		      io_geom->raid56_full_stripe_start + map->start +
6462		      btrfs_stripe_nr_to_offset(data_stripes)) -
6463		logical;
6464	io_geom->stripe_index = 0;
6465	io_geom->stripe_offset = 0;
6466}
6467
6468static void map_blocks_raid56_read(struct btrfs_chunk_map *map,
6469				   struct btrfs_io_geometry *io_geom)
6470{
6471	int data_stripes = nr_data_stripes(map);
6472
6473	ASSERT(io_geom->mirror_num <= 1);
6474	/* Just grab the data stripe directly. */
6475	io_geom->stripe_index = io_geom->stripe_nr % data_stripes;
6476	io_geom->stripe_nr /= data_stripes;
6477
6478	/* We distribute the parity blocks across stripes. */
6479	io_geom->stripe_index =
6480		(io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes;
6481
6482	if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1)
6483		io_geom->mirror_num = 1;
6484}
6485
6486static void map_blocks_single(const struct btrfs_chunk_map *map,
6487			      struct btrfs_io_geometry *io_geom)
6488{
6489	io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6490	io_geom->stripe_nr /= map->num_stripes;
6491	io_geom->mirror_num = io_geom->stripe_index + 1;
6492}
6493
6494/*
6495 * Map one logical range to one or more physical ranges.
6496 *
6497 * @length:		(Mandatory) mapped length of this run.
6498 *			One logical range can be split into different segments
6499 *			due to factors like zones and RAID0/5/6/10 stripe
6500 *			boundaries.
6501 *
6502 * @bioc_ret:		(Mandatory) returned btrfs_io_context structure.
6503 *			which has one or more physical ranges (btrfs_io_stripe)
6504 *			recorded inside.
6505 *			Caller should call btrfs_put_bioc() to free it after use.
6506 *
6507 * @smap:		(Optional) single physical range optimization.
6508 *			If the map request can be fulfilled by one single
6509 *			physical range, and this is parameter is not NULL,
6510 *			then @bioc_ret would be NULL, and @smap would be
6511 *			updated.
6512 *
6513 * @mirror_num_ret:	(Mandatory) returned mirror number if the original
6514 *			value is 0.
6515 *
6516 *			Mirror number 0 means to choose any live mirrors.
6517 *
6518 *			For non-RAID56 profiles, non-zero mirror_num means
6519 *			the Nth mirror. (e.g. mirror_num 1 means the first
6520 *			copy).
6521 *
6522 *			For RAID56 profile, mirror 1 means rebuild from P and
6523 *			the remaining data stripes.
6524 *
6525 *			For RAID6 profile, mirror > 2 means mark another
6526 *			data/P stripe error and rebuild from the remaining
6527 *			stripes..
6528 */
6529int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6530		    u64 logical, u64 *length,
6531		    struct btrfs_io_context **bioc_ret,
6532		    struct btrfs_io_stripe *smap, int *mirror_num_ret)
6533{
6534	struct btrfs_chunk_map *map;
6535	struct btrfs_io_geometry io_geom = { 0 };
6536	u64 map_offset;
6537	int i;
6538	int ret = 0;
6539	int num_copies;
6540	struct btrfs_io_context *bioc = NULL;
6541	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6542	int dev_replace_is_ongoing = 0;
6543	u16 num_alloc_stripes;
6544	u64 max_len;
6545
6546	ASSERT(bioc_ret);
6547
6548	io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6549	io_geom.num_stripes = 1;
6550	io_geom.stripe_index = 0;
6551	io_geom.op = op;
6552
6553	num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6554	if (io_geom.mirror_num > num_copies)
6555		return -EINVAL;
6556
6557	map = btrfs_get_chunk_map(fs_info, logical, *length);
6558	if (IS_ERR(map))
6559		return PTR_ERR(map);
6560
6561	map_offset = logical - map->start;
6562	io_geom.raid56_full_stripe_start = (u64)-1;
6563	max_len = btrfs_max_io_len(map, map_offset, &io_geom);
6564	*length = min_t(u64, map->chunk_len - map_offset, max_len);
6565
6566	down_read(&dev_replace->rwsem);
6567	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6568	/*
6569	 * Hold the semaphore for read during the whole operation, write is
6570	 * requested at commit time but must wait.
6571	 */
6572	if (!dev_replace_is_ongoing)
6573		up_read(&dev_replace->rwsem);
6574
6575	switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6576	case BTRFS_BLOCK_GROUP_RAID0:
6577		map_blocks_raid0(map, &io_geom);
6578		break;
6579	case BTRFS_BLOCK_GROUP_RAID1:
6580	case BTRFS_BLOCK_GROUP_RAID1C3:
6581	case BTRFS_BLOCK_GROUP_RAID1C4:
6582		map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing);
6583		break;
6584	case BTRFS_BLOCK_GROUP_DUP:
6585		map_blocks_dup(map, &io_geom);
6586		break;
6587	case BTRFS_BLOCK_GROUP_RAID10:
6588		map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing);
6589		break;
6590	case BTRFS_BLOCK_GROUP_RAID5:
6591	case BTRFS_BLOCK_GROUP_RAID6:
6592		if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)
6593			map_blocks_raid56_write(map, &io_geom, logical, length);
6594		else
6595			map_blocks_raid56_read(map, &io_geom);
6596		break;
6597	default:
6598		/*
6599		 * After this, stripe_nr is the number of stripes on this
6600		 * device we have to walk to find the data, and stripe_index is
6601		 * the number of our device in the stripe array
6602		 */
6603		map_blocks_single(map, &io_geom);
6604		break;
6605	}
6606	if (io_geom.stripe_index >= map->num_stripes) {
6607		btrfs_crit(fs_info,
6608			   "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6609			   io_geom.stripe_index, map->num_stripes);
6610		ret = -EINVAL;
6611		goto out;
6612	}
6613
6614	num_alloc_stripes = io_geom.num_stripes;
6615	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6616	    op != BTRFS_MAP_READ)
6617		/*
6618		 * For replace case, we need to add extra stripes for extra
6619		 * duplicated stripes.
6620		 *
6621		 * For both WRITE and GET_READ_MIRRORS, we may have at most
6622		 * 2 more stripes (DUP types, otherwise 1).
6623		 */
6624		num_alloc_stripes += 2;
6625
6626	/*
6627	 * If this I/O maps to a single device, try to return the device and
6628	 * physical block information on the stack instead of allocating an
6629	 * I/O context structure.
6630	 */
6631	if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op,
6632				io_geom.mirror_num)) {
6633		ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom);
6634		if (mirror_num_ret)
6635			*mirror_num_ret = io_geom.mirror_num;
6636		*bioc_ret = NULL;
6637		goto out;
6638	}
6639
6640	bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes);
6641	if (!bioc) {
6642		ret = -ENOMEM;
6643		goto out;
6644	}
6645	bioc->map_type = map->type;
6646
6647	/*
6648	 * For RAID56 full map, we need to make sure the stripes[] follows the
6649	 * rule that data stripes are all ordered, then followed with P and Q
6650	 * (if we have).
6651	 *
6652	 * It's still mostly the same as other profiles, just with extra rotation.
6653	 */
6654	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
6655	    (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) {
6656		/*
6657		 * For RAID56 @stripe_nr is already the number of full stripes
6658		 * before us, which is also the rotation value (needs to modulo
6659		 * with num_stripes).
6660		 *
6661		 * In this case, we just add @stripe_nr with @i, then do the
6662		 * modulo, to reduce one modulo call.
6663		 */
6664		bioc->full_stripe_logical = map->start +
6665			btrfs_stripe_nr_to_offset(io_geom.stripe_nr *
6666						  nr_data_stripes(map));
6667		for (int i = 0; i < io_geom.num_stripes; i++) {
6668			struct btrfs_io_stripe *dst = &bioc->stripes[i];
6669			u32 stripe_index;
6670
6671			stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes;
6672			dst->dev = map->stripes[stripe_index].dev;
6673			dst->physical =
6674				map->stripes[stripe_index].physical +
6675				io_geom.stripe_offset +
6676				btrfs_stripe_nr_to_offset(io_geom.stripe_nr);
6677		}
6678	} else {
6679		/*
6680		 * For all other non-RAID56 profiles, just copy the target
6681		 * stripe into the bioc.
6682		 */
6683		for (i = 0; i < io_geom.num_stripes; i++) {
6684			ret = set_io_stripe(fs_info, logical, length,
6685					    &bioc->stripes[i], map, &io_geom);
6686			if (ret < 0)
6687				break;
6688			io_geom.stripe_index++;
6689		}
6690	}
6691
6692	if (ret) {
6693		*bioc_ret = NULL;
6694		btrfs_put_bioc(bioc);
6695		goto out;
6696	}
6697
6698	if (op != BTRFS_MAP_READ)
6699		io_geom.max_errors = btrfs_chunk_max_errors(map);
6700
6701	if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6702	    op != BTRFS_MAP_READ) {
6703		handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6704					  &io_geom.num_stripes, &io_geom.max_errors);
6705	}
6706
6707	*bioc_ret = bioc;
6708	bioc->num_stripes = io_geom.num_stripes;
6709	bioc->max_errors = io_geom.max_errors;
6710	bioc->mirror_num = io_geom.mirror_num;
6711
6712out:
6713	if (dev_replace_is_ongoing) {
6714		lockdep_assert_held(&dev_replace->rwsem);
6715		/* Unlock and let waiting writers proceed */
6716		up_read(&dev_replace->rwsem);
6717	}
6718	btrfs_free_chunk_map(map);
6719	return ret;
6720}
6721
6722static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6723				      const struct btrfs_fs_devices *fs_devices)
6724{
6725	if (args->fsid == NULL)
6726		return true;
6727	if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6728		return true;
6729	return false;
6730}
6731
6732static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6733				  const struct btrfs_device *device)
6734{
6735	if (args->missing) {
6736		if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6737		    !device->bdev)
6738			return true;
6739		return false;
6740	}
6741
6742	if (device->devid != args->devid)
6743		return false;
6744	if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6745		return false;
6746	return true;
6747}
6748
6749/*
6750 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6751 * return NULL.
6752 *
6753 * If devid and uuid are both specified, the match must be exact, otherwise
6754 * only devid is used.
6755 */
6756struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6757				       const struct btrfs_dev_lookup_args *args)
6758{
6759	struct btrfs_device *device;
6760	struct btrfs_fs_devices *seed_devs;
6761
6762	if (dev_args_match_fs_devices(args, fs_devices)) {
6763		list_for_each_entry(device, &fs_devices->devices, dev_list) {
6764			if (dev_args_match_device(args, device))
6765				return device;
6766		}
6767	}
6768
6769	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6770		if (!dev_args_match_fs_devices(args, seed_devs))
6771			continue;
6772		list_for_each_entry(device, &seed_devs->devices, dev_list) {
6773			if (dev_args_match_device(args, device))
6774				return device;
6775		}
6776	}
6777
6778	return NULL;
6779}
6780
6781static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6782					    u64 devid, u8 *dev_uuid)
6783{
6784	struct btrfs_device *device;
6785	unsigned int nofs_flag;
6786
6787	/*
6788	 * We call this under the chunk_mutex, so we want to use NOFS for this
6789	 * allocation, however we don't want to change btrfs_alloc_device() to
6790	 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6791	 * places.
6792	 */
6793
6794	nofs_flag = memalloc_nofs_save();
6795	device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6796	memalloc_nofs_restore(nofs_flag);
6797	if (IS_ERR(device))
6798		return device;
6799
6800	list_add(&device->dev_list, &fs_devices->devices);
6801	device->fs_devices = fs_devices;
6802	fs_devices->num_devices++;
6803
6804	set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6805	fs_devices->missing_devices++;
6806
6807	return device;
6808}
6809
6810/*
6811 * Allocate new device struct, set up devid and UUID.
6812 *
6813 * @fs_info:	used only for generating a new devid, can be NULL if
6814 *		devid is provided (i.e. @devid != NULL).
6815 * @devid:	a pointer to devid for this device.  If NULL a new devid
6816 *		is generated.
6817 * @uuid:	a pointer to UUID for this device.  If NULL a new UUID
6818 *		is generated.
6819 * @path:	a pointer to device path if available, NULL otherwise.
6820 *
6821 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6822 * on error.  Returned struct is not linked onto any lists and must be
6823 * destroyed with btrfs_free_device.
6824 */
6825struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6826					const u64 *devid, const u8 *uuid,
6827					const char *path)
6828{
6829	struct btrfs_device *dev;
6830	u64 tmp;
6831
6832	if (WARN_ON(!devid && !fs_info))
6833		return ERR_PTR(-EINVAL);
6834
6835	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6836	if (!dev)
6837		return ERR_PTR(-ENOMEM);
6838
6839	INIT_LIST_HEAD(&dev->dev_list);
6840	INIT_LIST_HEAD(&dev->dev_alloc_list);
6841	INIT_LIST_HEAD(&dev->post_commit_list);
6842
6843	atomic_set(&dev->dev_stats_ccnt, 0);
6844	btrfs_device_data_ordered_init(dev);
6845	extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6846
6847	if (devid)
6848		tmp = *devid;
6849	else {
6850		int ret;
6851
6852		ret = find_next_devid(fs_info, &tmp);
6853		if (ret) {
6854			btrfs_free_device(dev);
6855			return ERR_PTR(ret);
6856		}
6857	}
6858	dev->devid = tmp;
6859
6860	if (uuid)
6861		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6862	else
6863		generate_random_uuid(dev->uuid);
6864
6865	if (path) {
6866		struct rcu_string *name;
6867
6868		name = rcu_string_strdup(path, GFP_KERNEL);
6869		if (!name) {
6870			btrfs_free_device(dev);
6871			return ERR_PTR(-ENOMEM);
6872		}
6873		rcu_assign_pointer(dev->name, name);
6874	}
6875
6876	return dev;
6877}
6878
6879static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6880					u64 devid, u8 *uuid, bool error)
6881{
6882	if (error)
6883		btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6884			      devid, uuid);
6885	else
6886		btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6887			      devid, uuid);
6888}
6889
6890u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map)
6891{
6892	const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6893
6894	return div_u64(map->chunk_len, data_stripes);
6895}
6896
6897#if BITS_PER_LONG == 32
6898/*
6899 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6900 * can't be accessed on 32bit systems.
6901 *
6902 * This function do mount time check to reject the fs if it already has
6903 * metadata chunk beyond that limit.
6904 */
6905static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6906				  u64 logical, u64 length, u64 type)
6907{
6908	if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6909		return 0;
6910
6911	if (logical + length < MAX_LFS_FILESIZE)
6912		return 0;
6913
6914	btrfs_err_32bit_limit(fs_info);
6915	return -EOVERFLOW;
6916}
6917
6918/*
6919 * This is to give early warning for any metadata chunk reaching
6920 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6921 * Although we can still access the metadata, it's not going to be possible
6922 * once the limit is reached.
6923 */
6924static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6925				  u64 logical, u64 length, u64 type)
6926{
6927	if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6928		return;
6929
6930	if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6931		return;
6932
6933	btrfs_warn_32bit_limit(fs_info);
6934}
6935#endif
6936
6937static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6938						  u64 devid, u8 *uuid)
6939{
6940	struct btrfs_device *dev;
6941
6942	if (!btrfs_test_opt(fs_info, DEGRADED)) {
6943		btrfs_report_missing_device(fs_info, devid, uuid, true);
6944		return ERR_PTR(-ENOENT);
6945	}
6946
6947	dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6948	if (IS_ERR(dev)) {
6949		btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6950			  devid, PTR_ERR(dev));
6951		return dev;
6952	}
6953	btrfs_report_missing_device(fs_info, devid, uuid, false);
6954
6955	return dev;
6956}
6957
6958static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6959			  struct btrfs_chunk *chunk)
6960{
6961	BTRFS_DEV_LOOKUP_ARGS(args);
6962	struct btrfs_fs_info *fs_info = leaf->fs_info;
6963	struct btrfs_chunk_map *map;
6964	u64 logical;
6965	u64 length;
6966	u64 devid;
6967	u64 type;
6968	u8 uuid[BTRFS_UUID_SIZE];
6969	int index;
6970	int num_stripes;
6971	int ret;
6972	int i;
6973
6974	logical = key->offset;
6975	length = btrfs_chunk_length(leaf, chunk);
6976	type = btrfs_chunk_type(leaf, chunk);
6977	index = btrfs_bg_flags_to_raid_index(type);
6978	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6979
6980#if BITS_PER_LONG == 32
6981	ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6982	if (ret < 0)
6983		return ret;
6984	warn_32bit_meta_chunk(fs_info, logical, length, type);
6985#endif
6986
6987	/*
6988	 * Only need to verify chunk item if we're reading from sys chunk array,
6989	 * as chunk item in tree block is already verified by tree-checker.
6990	 */
6991	if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6992		ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6993		if (ret)
6994			return ret;
6995	}
6996
6997	map = btrfs_find_chunk_map(fs_info, logical, 1);
6998
6999	/* already mapped? */
7000	if (map && map->start <= logical && map->start + map->chunk_len > logical) {
7001		btrfs_free_chunk_map(map);
7002		return 0;
7003	} else if (map) {
7004		btrfs_free_chunk_map(map);
7005	}
7006
7007	map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS);
7008	if (!map)
7009		return -ENOMEM;
7010
7011	map->start = logical;
7012	map->chunk_len = length;
7013	map->num_stripes = num_stripes;
7014	map->io_width = btrfs_chunk_io_width(leaf, chunk);
7015	map->io_align = btrfs_chunk_io_align(leaf, chunk);
7016	map->type = type;
7017	/*
7018	 * We can't use the sub_stripes value, as for profiles other than
7019	 * RAID10, they may have 0 as sub_stripes for filesystems created by
7020	 * older mkfs (<v5.4).
7021	 * In that case, it can cause divide-by-zero errors later.
7022	 * Since currently sub_stripes is fixed for each profile, let's
7023	 * use the trusted value instead.
7024	 */
7025	map->sub_stripes = btrfs_raid_array[index].sub_stripes;
7026	map->verified_stripes = 0;
7027	map->stripe_size = btrfs_calc_stripe_length(map);
7028	for (i = 0; i < num_stripes; i++) {
7029		map->stripes[i].physical =
7030			btrfs_stripe_offset_nr(leaf, chunk, i);
7031		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7032		args.devid = devid;
7033		read_extent_buffer(leaf, uuid, (unsigned long)
7034				   btrfs_stripe_dev_uuid_nr(chunk, i),
7035				   BTRFS_UUID_SIZE);
7036		args.uuid = uuid;
7037		map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7038		if (!map->stripes[i].dev) {
7039			map->stripes[i].dev = handle_missing_device(fs_info,
7040								    devid, uuid);
7041			if (IS_ERR(map->stripes[i].dev)) {
7042				ret = PTR_ERR(map->stripes[i].dev);
7043				btrfs_free_chunk_map(map);
7044				return ret;
7045			}
7046		}
7047
7048		set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7049				&(map->stripes[i].dev->dev_state));
7050	}
7051
7052	ret = btrfs_add_chunk_map(fs_info, map);
7053	if (ret < 0) {
7054		btrfs_err(fs_info,
7055			  "failed to add chunk map, start=%llu len=%llu: %d",
7056			  map->start, map->chunk_len, ret);
7057	}
7058
7059	return ret;
7060}
7061
7062static void fill_device_from_item(struct extent_buffer *leaf,
7063				 struct btrfs_dev_item *dev_item,
7064				 struct btrfs_device *device)
7065{
7066	unsigned long ptr;
7067
7068	device->devid = btrfs_device_id(leaf, dev_item);
7069	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7070	device->total_bytes = device->disk_total_bytes;
7071	device->commit_total_bytes = device->disk_total_bytes;
7072	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7073	device->commit_bytes_used = device->bytes_used;
7074	device->type = btrfs_device_type(leaf, dev_item);
7075	device->io_align = btrfs_device_io_align(leaf, dev_item);
7076	device->io_width = btrfs_device_io_width(leaf, dev_item);
7077	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7078	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7079	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7080
7081	ptr = btrfs_device_uuid(dev_item);
7082	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7083}
7084
7085static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7086						  u8 *fsid)
7087{
7088	struct btrfs_fs_devices *fs_devices;
7089	int ret;
7090
7091	lockdep_assert_held(&uuid_mutex);
7092	ASSERT(fsid);
7093
7094	/* This will match only for multi-device seed fs */
7095	list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7096		if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7097			return fs_devices;
7098
7099
7100	fs_devices = find_fsid(fsid, NULL);
7101	if (!fs_devices) {
7102		if (!btrfs_test_opt(fs_info, DEGRADED))
7103			return ERR_PTR(-ENOENT);
7104
7105		fs_devices = alloc_fs_devices(fsid);
7106		if (IS_ERR(fs_devices))
7107			return fs_devices;
7108
7109		fs_devices->seeding = true;
7110		fs_devices->opened = 1;
7111		return fs_devices;
7112	}
7113
7114	/*
7115	 * Upon first call for a seed fs fsid, just create a private copy of the
7116	 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7117	 */
7118	fs_devices = clone_fs_devices(fs_devices);
7119	if (IS_ERR(fs_devices))
7120		return fs_devices;
7121
7122	ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
7123	if (ret) {
7124		free_fs_devices(fs_devices);
7125		return ERR_PTR(ret);
7126	}
7127
7128	if (!fs_devices->seeding) {
7129		close_fs_devices(fs_devices);
7130		free_fs_devices(fs_devices);
7131		return ERR_PTR(-EINVAL);
7132	}
7133
7134	list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7135
7136	return fs_devices;
7137}
7138
7139static int read_one_dev(struct extent_buffer *leaf,
7140			struct btrfs_dev_item *dev_item)
7141{
7142	BTRFS_DEV_LOOKUP_ARGS(args);
7143	struct btrfs_fs_info *fs_info = leaf->fs_info;
7144	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7145	struct btrfs_device *device;
7146	u64 devid;
7147	int ret;
7148	u8 fs_uuid[BTRFS_FSID_SIZE];
7149	u8 dev_uuid[BTRFS_UUID_SIZE];
7150
7151	devid = btrfs_device_id(leaf, dev_item);
7152	args.devid = devid;
7153	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7154			   BTRFS_UUID_SIZE);
7155	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7156			   BTRFS_FSID_SIZE);
7157	args.uuid = dev_uuid;
7158	args.fsid = fs_uuid;
7159
7160	if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7161		fs_devices = open_seed_devices(fs_info, fs_uuid);
7162		if (IS_ERR(fs_devices))
7163			return PTR_ERR(fs_devices);
7164	}
7165
7166	device = btrfs_find_device(fs_info->fs_devices, &args);
7167	if (!device) {
7168		if (!btrfs_test_opt(fs_info, DEGRADED)) {
7169			btrfs_report_missing_device(fs_info, devid,
7170							dev_uuid, true);
7171			return -ENOENT;
7172		}
7173
7174		device = add_missing_dev(fs_devices, devid, dev_uuid);
7175		if (IS_ERR(device)) {
7176			btrfs_err(fs_info,
7177				"failed to add missing dev %llu: %ld",
7178				devid, PTR_ERR(device));
7179			return PTR_ERR(device);
7180		}
7181		btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7182	} else {
7183		if (!device->bdev) {
7184			if (!btrfs_test_opt(fs_info, DEGRADED)) {
7185				btrfs_report_missing_device(fs_info,
7186						devid, dev_uuid, true);
7187				return -ENOENT;
7188			}
7189			btrfs_report_missing_device(fs_info, devid,
7190							dev_uuid, false);
7191		}
7192
7193		if (!device->bdev &&
7194		    !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7195			/*
7196			 * this happens when a device that was properly setup
7197			 * in the device info lists suddenly goes bad.
7198			 * device->bdev is NULL, and so we have to set
7199			 * device->missing to one here
7200			 */
7201			device->fs_devices->missing_devices++;
7202			set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7203		}
7204
7205		/* Move the device to its own fs_devices */
7206		if (device->fs_devices != fs_devices) {
7207			ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7208							&device->dev_state));
7209
7210			list_move(&device->dev_list, &fs_devices->devices);
7211			device->fs_devices->num_devices--;
7212			fs_devices->num_devices++;
7213
7214			device->fs_devices->missing_devices--;
7215			fs_devices->missing_devices++;
7216
7217			device->fs_devices = fs_devices;
7218		}
7219	}
7220
7221	if (device->fs_devices != fs_info->fs_devices) {
7222		BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7223		if (device->generation !=
7224		    btrfs_device_generation(leaf, dev_item))
7225			return -EINVAL;
7226	}
7227
7228	fill_device_from_item(leaf, dev_item, device);
7229	if (device->bdev) {
7230		u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7231
7232		if (device->total_bytes > max_total_bytes) {
7233			btrfs_err(fs_info,
7234			"device total_bytes should be at most %llu but found %llu",
7235				  max_total_bytes, device->total_bytes);
7236			return -EINVAL;
7237		}
7238	}
7239	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7240	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7241	   !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7242		device->fs_devices->total_rw_bytes += device->total_bytes;
7243		atomic64_add(device->total_bytes - device->bytes_used,
7244				&fs_info->free_chunk_space);
7245	}
7246	ret = 0;
7247	return ret;
7248}
7249
7250int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7251{
7252	struct btrfs_super_block *super_copy = fs_info->super_copy;
7253	struct extent_buffer *sb;
7254	struct btrfs_disk_key *disk_key;
7255	struct btrfs_chunk *chunk;
7256	u8 *array_ptr;
7257	unsigned long sb_array_offset;
7258	int ret = 0;
7259	u32 num_stripes;
7260	u32 array_size;
7261	u32 len = 0;
7262	u32 cur_offset;
7263	u64 type;
7264	struct btrfs_key key;
7265
7266	ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7267
7268	/*
7269	 * We allocated a dummy extent, just to use extent buffer accessors.
7270	 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7271	 * that's fine, we will not go beyond system chunk array anyway.
7272	 */
7273	sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7274	if (!sb)
7275		return -ENOMEM;
7276	set_extent_buffer_uptodate(sb);
7277
7278	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7279	array_size = btrfs_super_sys_array_size(super_copy);
7280
7281	array_ptr = super_copy->sys_chunk_array;
7282	sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7283	cur_offset = 0;
7284
7285	while (cur_offset < array_size) {
7286		disk_key = (struct btrfs_disk_key *)array_ptr;
7287		len = sizeof(*disk_key);
7288		if (cur_offset + len > array_size)
7289			goto out_short_read;
7290
7291		btrfs_disk_key_to_cpu(&key, disk_key);
7292
7293		array_ptr += len;
7294		sb_array_offset += len;
7295		cur_offset += len;
7296
7297		if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7298			btrfs_err(fs_info,
7299			    "unexpected item type %u in sys_array at offset %u",
7300				  (u32)key.type, cur_offset);
7301			ret = -EIO;
7302			break;
7303		}
7304
7305		chunk = (struct btrfs_chunk *)sb_array_offset;
7306		/*
7307		 * At least one btrfs_chunk with one stripe must be present,
7308		 * exact stripe count check comes afterwards
7309		 */
7310		len = btrfs_chunk_item_size(1);
7311		if (cur_offset + len > array_size)
7312			goto out_short_read;
7313
7314		num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7315		if (!num_stripes) {
7316			btrfs_err(fs_info,
7317			"invalid number of stripes %u in sys_array at offset %u",
7318				  num_stripes, cur_offset);
7319			ret = -EIO;
7320			break;
7321		}
7322
7323		type = btrfs_chunk_type(sb, chunk);
7324		if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7325			btrfs_err(fs_info,
7326			"invalid chunk type %llu in sys_array at offset %u",
7327				  type, cur_offset);
7328			ret = -EIO;
7329			break;
7330		}
7331
7332		len = btrfs_chunk_item_size(num_stripes);
7333		if (cur_offset + len > array_size)
7334			goto out_short_read;
7335
7336		ret = read_one_chunk(&key, sb, chunk);
7337		if (ret)
7338			break;
7339
7340		array_ptr += len;
7341		sb_array_offset += len;
7342		cur_offset += len;
7343	}
7344	clear_extent_buffer_uptodate(sb);
7345	free_extent_buffer_stale(sb);
7346	return ret;
7347
7348out_short_read:
7349	btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7350			len, cur_offset);
7351	clear_extent_buffer_uptodate(sb);
7352	free_extent_buffer_stale(sb);
7353	return -EIO;
7354}
7355
7356/*
7357 * Check if all chunks in the fs are OK for read-write degraded mount
7358 *
7359 * If the @failing_dev is specified, it's accounted as missing.
7360 *
7361 * Return true if all chunks meet the minimal RW mount requirements.
7362 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7363 */
7364bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7365					struct btrfs_device *failing_dev)
7366{
7367	struct btrfs_chunk_map *map;
7368	u64 next_start;
7369	bool ret = true;
7370
7371	map = btrfs_find_chunk_map(fs_info, 0, U64_MAX);
7372	/* No chunk at all? Return false anyway */
7373	if (!map) {
7374		ret = false;
7375		goto out;
7376	}
7377	while (map) {
7378		int missing = 0;
7379		int max_tolerated;
7380		int i;
7381
7382		max_tolerated =
7383			btrfs_get_num_tolerated_disk_barrier_failures(
7384					map->type);
7385		for (i = 0; i < map->num_stripes; i++) {
7386			struct btrfs_device *dev = map->stripes[i].dev;
7387
7388			if (!dev || !dev->bdev ||
7389			    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7390			    dev->last_flush_error)
7391				missing++;
7392			else if (failing_dev && failing_dev == dev)
7393				missing++;
7394		}
7395		if (missing > max_tolerated) {
7396			if (!failing_dev)
7397				btrfs_warn(fs_info,
7398	"chunk %llu missing %d devices, max tolerance is %d for writable mount",
7399				   map->start, missing, max_tolerated);
7400			btrfs_free_chunk_map(map);
7401			ret = false;
7402			goto out;
7403		}
7404		next_start = map->start + map->chunk_len;
7405		btrfs_free_chunk_map(map);
7406
7407		map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start);
7408	}
7409out:
7410	return ret;
7411}
7412
7413static void readahead_tree_node_children(struct extent_buffer *node)
7414{
7415	int i;
7416	const int nr_items = btrfs_header_nritems(node);
7417
7418	for (i = 0; i < nr_items; i++)
7419		btrfs_readahead_node_child(node, i);
7420}
7421
7422int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7423{
7424	struct btrfs_root *root = fs_info->chunk_root;
7425	struct btrfs_path *path;
7426	struct extent_buffer *leaf;
7427	struct btrfs_key key;
7428	struct btrfs_key found_key;
7429	int ret;
7430	int slot;
7431	int iter_ret = 0;
7432	u64 total_dev = 0;
7433	u64 last_ra_node = 0;
7434
7435	path = btrfs_alloc_path();
7436	if (!path)
7437		return -ENOMEM;
7438
7439	/*
7440	 * uuid_mutex is needed only if we are mounting a sprout FS
7441	 * otherwise we don't need it.
7442	 */
7443	mutex_lock(&uuid_mutex);
7444
7445	/*
7446	 * It is possible for mount and umount to race in such a way that
7447	 * we execute this code path, but open_fs_devices failed to clear
7448	 * total_rw_bytes. We certainly want it cleared before reading the
7449	 * device items, so clear it here.
7450	 */
7451	fs_info->fs_devices->total_rw_bytes = 0;
7452
7453	/*
7454	 * Lockdep complains about possible circular locking dependency between
7455	 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7456	 * used for freeze procection of a fs (struct super_block.s_writers),
7457	 * which we take when starting a transaction, and extent buffers of the
7458	 * chunk tree if we call read_one_dev() while holding a lock on an
7459	 * extent buffer of the chunk tree. Since we are mounting the filesystem
7460	 * and at this point there can't be any concurrent task modifying the
7461	 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7462	 */
7463	ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7464	path->skip_locking = 1;
7465
7466	/*
7467	 * Read all device items, and then all the chunk items. All
7468	 * device items are found before any chunk item (their object id
7469	 * is smaller than the lowest possible object id for a chunk
7470	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7471	 */
7472	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7473	key.offset = 0;
7474	key.type = 0;
7475	btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7476		struct extent_buffer *node = path->nodes[1];
7477
7478		leaf = path->nodes[0];
7479		slot = path->slots[0];
7480
7481		if (node) {
7482			if (last_ra_node != node->start) {
7483				readahead_tree_node_children(node);
7484				last_ra_node = node->start;
7485			}
7486		}
7487		if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7488			struct btrfs_dev_item *dev_item;
7489			dev_item = btrfs_item_ptr(leaf, slot,
7490						  struct btrfs_dev_item);
7491			ret = read_one_dev(leaf, dev_item);
7492			if (ret)
7493				goto error;
7494			total_dev++;
7495		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7496			struct btrfs_chunk *chunk;
7497
7498			/*
7499			 * We are only called at mount time, so no need to take
7500			 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7501			 * we always lock first fs_info->chunk_mutex before
7502			 * acquiring any locks on the chunk tree. This is a
7503			 * requirement for chunk allocation, see the comment on
7504			 * top of btrfs_chunk_alloc() for details.
7505			 */
7506			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7507			ret = read_one_chunk(&found_key, leaf, chunk);
7508			if (ret)
7509				goto error;
7510		}
7511	}
7512	/* Catch error found during iteration */
7513	if (iter_ret < 0) {
7514		ret = iter_ret;
7515		goto error;
7516	}
7517
7518	/*
7519	 * After loading chunk tree, we've got all device information,
7520	 * do another round of validation checks.
7521	 */
7522	if (total_dev != fs_info->fs_devices->total_devices) {
7523		btrfs_warn(fs_info,
7524"super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7525			  btrfs_super_num_devices(fs_info->super_copy),
7526			  total_dev);
7527		fs_info->fs_devices->total_devices = total_dev;
7528		btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7529	}
7530	if (btrfs_super_total_bytes(fs_info->super_copy) <
7531	    fs_info->fs_devices->total_rw_bytes) {
7532		btrfs_err(fs_info,
7533	"super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7534			  btrfs_super_total_bytes(fs_info->super_copy),
7535			  fs_info->fs_devices->total_rw_bytes);
7536		ret = -EINVAL;
7537		goto error;
7538	}
7539	ret = 0;
7540error:
7541	mutex_unlock(&uuid_mutex);
7542
7543	btrfs_free_path(path);
7544	return ret;
7545}
7546
7547int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7548{
7549	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7550	struct btrfs_device *device;
7551	int ret = 0;
7552
7553	fs_devices->fs_info = fs_info;
7554
7555	mutex_lock(&fs_devices->device_list_mutex);
7556	list_for_each_entry(device, &fs_devices->devices, dev_list)
7557		device->fs_info = fs_info;
7558
7559	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7560		list_for_each_entry(device, &seed_devs->devices, dev_list) {
7561			device->fs_info = fs_info;
7562			ret = btrfs_get_dev_zone_info(device, false);
7563			if (ret)
7564				break;
7565		}
7566
7567		seed_devs->fs_info = fs_info;
7568	}
7569	mutex_unlock(&fs_devices->device_list_mutex);
7570
7571	return ret;
7572}
7573
7574static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7575				 const struct btrfs_dev_stats_item *ptr,
7576				 int index)
7577{
7578	u64 val;
7579
7580	read_extent_buffer(eb, &val,
7581			   offsetof(struct btrfs_dev_stats_item, values) +
7582			    ((unsigned long)ptr) + (index * sizeof(u64)),
7583			   sizeof(val));
7584	return val;
7585}
7586
7587static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7588				      struct btrfs_dev_stats_item *ptr,
7589				      int index, u64 val)
7590{
7591	write_extent_buffer(eb, &val,
7592			    offsetof(struct btrfs_dev_stats_item, values) +
7593			     ((unsigned long)ptr) + (index * sizeof(u64)),
7594			    sizeof(val));
7595}
7596
7597static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7598				       struct btrfs_path *path)
7599{
7600	struct btrfs_dev_stats_item *ptr;
7601	struct extent_buffer *eb;
7602	struct btrfs_key key;
7603	int item_size;
7604	int i, ret, slot;
7605
7606	if (!device->fs_info->dev_root)
7607		return 0;
7608
7609	key.objectid = BTRFS_DEV_STATS_OBJECTID;
7610	key.type = BTRFS_PERSISTENT_ITEM_KEY;
7611	key.offset = device->devid;
7612	ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7613	if (ret) {
7614		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7615			btrfs_dev_stat_set(device, i, 0);
7616		device->dev_stats_valid = 1;
7617		btrfs_release_path(path);
7618		return ret < 0 ? ret : 0;
7619	}
7620	slot = path->slots[0];
7621	eb = path->nodes[0];
7622	item_size = btrfs_item_size(eb, slot);
7623
7624	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7625
7626	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7627		if (item_size >= (1 + i) * sizeof(__le64))
7628			btrfs_dev_stat_set(device, i,
7629					   btrfs_dev_stats_value(eb, ptr, i));
7630		else
7631			btrfs_dev_stat_set(device, i, 0);
7632	}
7633
7634	device->dev_stats_valid = 1;
7635	btrfs_dev_stat_print_on_load(device);
7636	btrfs_release_path(path);
7637
7638	return 0;
7639}
7640
7641int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7642{
7643	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7644	struct btrfs_device *device;
7645	struct btrfs_path *path = NULL;
7646	int ret = 0;
7647
7648	path = btrfs_alloc_path();
7649	if (!path)
7650		return -ENOMEM;
7651
7652	mutex_lock(&fs_devices->device_list_mutex);
7653	list_for_each_entry(device, &fs_devices->devices, dev_list) {
7654		ret = btrfs_device_init_dev_stats(device, path);
7655		if (ret)
7656			goto out;
7657	}
7658	list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7659		list_for_each_entry(device, &seed_devs->devices, dev_list) {
7660			ret = btrfs_device_init_dev_stats(device, path);
7661			if (ret)
7662				goto out;
7663		}
7664	}
7665out:
7666	mutex_unlock(&fs_devices->device_list_mutex);
7667
7668	btrfs_free_path(path);
7669	return ret;
7670}
7671
7672static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7673				struct btrfs_device *device)
7674{
7675	struct btrfs_fs_info *fs_info = trans->fs_info;
7676	struct btrfs_root *dev_root = fs_info->dev_root;
7677	struct btrfs_path *path;
7678	struct btrfs_key key;
7679	struct extent_buffer *eb;
7680	struct btrfs_dev_stats_item *ptr;
7681	int ret;
7682	int i;
7683
7684	key.objectid = BTRFS_DEV_STATS_OBJECTID;
7685	key.type = BTRFS_PERSISTENT_ITEM_KEY;
7686	key.offset = device->devid;
7687
7688	path = btrfs_alloc_path();
7689	if (!path)
7690		return -ENOMEM;
7691	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7692	if (ret < 0) {
7693		btrfs_warn_in_rcu(fs_info,
7694			"error %d while searching for dev_stats item for device %s",
7695				  ret, btrfs_dev_name(device));
7696		goto out;
7697	}
7698
7699	if (ret == 0 &&
7700	    btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7701		/* need to delete old one and insert a new one */
7702		ret = btrfs_del_item(trans, dev_root, path);
7703		if (ret != 0) {
7704			btrfs_warn_in_rcu(fs_info,
7705				"delete too small dev_stats item for device %s failed %d",
7706					  btrfs_dev_name(device), ret);
7707			goto out;
7708		}
7709		ret = 1;
7710	}
7711
7712	if (ret == 1) {
7713		/* need to insert a new item */
7714		btrfs_release_path(path);
7715		ret = btrfs_insert_empty_item(trans, dev_root, path,
7716					      &key, sizeof(*ptr));
7717		if (ret < 0) {
7718			btrfs_warn_in_rcu(fs_info,
7719				"insert dev_stats item for device %s failed %d",
7720				btrfs_dev_name(device), ret);
7721			goto out;
7722		}
7723	}
7724
7725	eb = path->nodes[0];
7726	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7727	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7728		btrfs_set_dev_stats_value(eb, ptr, i,
7729					  btrfs_dev_stat_read(device, i));
7730	btrfs_mark_buffer_dirty(trans, eb);
7731
7732out:
7733	btrfs_free_path(path);
7734	return ret;
7735}
7736
7737/*
7738 * called from commit_transaction. Writes all changed device stats to disk.
7739 */
7740int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7741{
7742	struct btrfs_fs_info *fs_info = trans->fs_info;
7743	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7744	struct btrfs_device *device;
7745	int stats_cnt;
7746	int ret = 0;
7747
7748	mutex_lock(&fs_devices->device_list_mutex);
7749	list_for_each_entry(device, &fs_devices->devices, dev_list) {
7750		stats_cnt = atomic_read(&device->dev_stats_ccnt);
7751		if (!device->dev_stats_valid || stats_cnt == 0)
7752			continue;
7753
7754
7755		/*
7756		 * There is a LOAD-LOAD control dependency between the value of
7757		 * dev_stats_ccnt and updating the on-disk values which requires
7758		 * reading the in-memory counters. Such control dependencies
7759		 * require explicit read memory barriers.
7760		 *
7761		 * This memory barriers pairs with smp_mb__before_atomic in
7762		 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7763		 * barrier implied by atomic_xchg in
7764		 * btrfs_dev_stats_read_and_reset
7765		 */
7766		smp_rmb();
7767
7768		ret = update_dev_stat_item(trans, device);
7769		if (!ret)
7770			atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7771	}
7772	mutex_unlock(&fs_devices->device_list_mutex);
7773
7774	return ret;
7775}
7776
7777void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7778{
7779	btrfs_dev_stat_inc(dev, index);
7780
7781	if (!dev->dev_stats_valid)
7782		return;
7783	btrfs_err_rl_in_rcu(dev->fs_info,
7784		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7785			   btrfs_dev_name(dev),
7786			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7787			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7788			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7789			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7790			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7791}
7792
7793static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7794{
7795	int i;
7796
7797	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7798		if (btrfs_dev_stat_read(dev, i) != 0)
7799			break;
7800	if (i == BTRFS_DEV_STAT_VALUES_MAX)
7801		return; /* all values == 0, suppress message */
7802
7803	btrfs_info_in_rcu(dev->fs_info,
7804		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7805	       btrfs_dev_name(dev),
7806	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7807	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7808	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7809	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7810	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7811}
7812
7813int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7814			struct btrfs_ioctl_get_dev_stats *stats)
7815{
7816	BTRFS_DEV_LOOKUP_ARGS(args);
7817	struct btrfs_device *dev;
7818	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7819	int i;
7820
7821	mutex_lock(&fs_devices->device_list_mutex);
7822	args.devid = stats->devid;
7823	dev = btrfs_find_device(fs_info->fs_devices, &args);
7824	mutex_unlock(&fs_devices->device_list_mutex);
7825
7826	if (!dev) {
7827		btrfs_warn(fs_info, "get dev_stats failed, device not found");
7828		return -ENODEV;
7829	} else if (!dev->dev_stats_valid) {
7830		btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7831		return -ENODEV;
7832	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7833		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7834			if (stats->nr_items > i)
7835				stats->values[i] =
7836					btrfs_dev_stat_read_and_reset(dev, i);
7837			else
7838				btrfs_dev_stat_set(dev, i, 0);
7839		}
7840		btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7841			   current->comm, task_pid_nr(current));
7842	} else {
7843		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7844			if (stats->nr_items > i)
7845				stats->values[i] = btrfs_dev_stat_read(dev, i);
7846	}
7847	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7848		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7849	return 0;
7850}
7851
7852/*
7853 * Update the size and bytes used for each device where it changed.  This is
7854 * delayed since we would otherwise get errors while writing out the
7855 * superblocks.
7856 *
7857 * Must be invoked during transaction commit.
7858 */
7859void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7860{
7861	struct btrfs_device *curr, *next;
7862
7863	ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7864
7865	if (list_empty(&trans->dev_update_list))
7866		return;
7867
7868	/*
7869	 * We don't need the device_list_mutex here.  This list is owned by the
7870	 * transaction and the transaction must complete before the device is
7871	 * released.
7872	 */
7873	mutex_lock(&trans->fs_info->chunk_mutex);
7874	list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7875				 post_commit_list) {
7876		list_del_init(&curr->post_commit_list);
7877		curr->commit_total_bytes = curr->disk_total_bytes;
7878		curr->commit_bytes_used = curr->bytes_used;
7879	}
7880	mutex_unlock(&trans->fs_info->chunk_mutex);
7881}
7882
7883/*
7884 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7885 */
7886int btrfs_bg_type_to_factor(u64 flags)
7887{
7888	const int index = btrfs_bg_flags_to_raid_index(flags);
7889
7890	return btrfs_raid_array[index].ncopies;
7891}
7892
7893
7894
7895static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7896				 u64 chunk_offset, u64 devid,
7897				 u64 physical_offset, u64 physical_len)
7898{
7899	struct btrfs_dev_lookup_args args = { .devid = devid };
7900	struct btrfs_chunk_map *map;
7901	struct btrfs_device *dev;
7902	u64 stripe_len;
7903	bool found = false;
7904	int ret = 0;
7905	int i;
7906
7907	map = btrfs_find_chunk_map(fs_info, chunk_offset, 1);
7908	if (!map) {
7909		btrfs_err(fs_info,
7910"dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7911			  physical_offset, devid);
7912		ret = -EUCLEAN;
7913		goto out;
7914	}
7915
7916	stripe_len = btrfs_calc_stripe_length(map);
7917	if (physical_len != stripe_len) {
7918		btrfs_err(fs_info,
7919"dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7920			  physical_offset, devid, map->start, physical_len,
7921			  stripe_len);
7922		ret = -EUCLEAN;
7923		goto out;
7924	}
7925
7926	/*
7927	 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7928	 * space. Although kernel can handle it without problem, better to warn
7929	 * the users.
7930	 */
7931	if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7932		btrfs_warn(fs_info,
7933		"devid %llu physical %llu len %llu inside the reserved space",
7934			   devid, physical_offset, physical_len);
7935
7936	for (i = 0; i < map->num_stripes; i++) {
7937		if (map->stripes[i].dev->devid == devid &&
7938		    map->stripes[i].physical == physical_offset) {
7939			found = true;
7940			if (map->verified_stripes >= map->num_stripes) {
7941				btrfs_err(fs_info,
7942				"too many dev extents for chunk %llu found",
7943					  map->start);
7944				ret = -EUCLEAN;
7945				goto out;
7946			}
7947			map->verified_stripes++;
7948			break;
7949		}
7950	}
7951	if (!found) {
7952		btrfs_err(fs_info,
7953	"dev extent physical offset %llu devid %llu has no corresponding chunk",
7954			physical_offset, devid);
7955		ret = -EUCLEAN;
7956	}
7957
7958	/* Make sure no dev extent is beyond device boundary */
7959	dev = btrfs_find_device(fs_info->fs_devices, &args);
7960	if (!dev) {
7961		btrfs_err(fs_info, "failed to find devid %llu", devid);
7962		ret = -EUCLEAN;
7963		goto out;
7964	}
7965
7966	if (physical_offset + physical_len > dev->disk_total_bytes) {
7967		btrfs_err(fs_info,
7968"dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7969			  devid, physical_offset, physical_len,
7970			  dev->disk_total_bytes);
7971		ret = -EUCLEAN;
7972		goto out;
7973	}
7974
7975	if (dev->zone_info) {
7976		u64 zone_size = dev->zone_info->zone_size;
7977
7978		if (!IS_ALIGNED(physical_offset, zone_size) ||
7979		    !IS_ALIGNED(physical_len, zone_size)) {
7980			btrfs_err(fs_info,
7981"zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7982				  devid, physical_offset, physical_len);
7983			ret = -EUCLEAN;
7984			goto out;
7985		}
7986	}
7987
7988out:
7989	btrfs_free_chunk_map(map);
7990	return ret;
7991}
7992
7993static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7994{
7995	struct rb_node *node;
7996	int ret = 0;
7997
7998	read_lock(&fs_info->mapping_tree_lock);
7999	for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) {
8000		struct btrfs_chunk_map *map;
8001
8002		map = rb_entry(node, struct btrfs_chunk_map, rb_node);
8003		if (map->num_stripes != map->verified_stripes) {
8004			btrfs_err(fs_info,
8005			"chunk %llu has missing dev extent, have %d expect %d",
8006				  map->start, map->verified_stripes, map->num_stripes);
8007			ret = -EUCLEAN;
8008			goto out;
8009		}
8010	}
8011out:
8012	read_unlock(&fs_info->mapping_tree_lock);
8013	return ret;
8014}
8015
8016/*
8017 * Ensure that all dev extents are mapped to correct chunk, otherwise
8018 * later chunk allocation/free would cause unexpected behavior.
8019 *
8020 * NOTE: This will iterate through the whole device tree, which should be of
8021 * the same size level as the chunk tree.  This slightly increases mount time.
8022 */
8023int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8024{
8025	struct btrfs_path *path;
8026	struct btrfs_root *root = fs_info->dev_root;
8027	struct btrfs_key key;
8028	u64 prev_devid = 0;
8029	u64 prev_dev_ext_end = 0;
8030	int ret = 0;
8031
8032	/*
8033	 * We don't have a dev_root because we mounted with ignorebadroots and
8034	 * failed to load the root, so we want to skip the verification in this
8035	 * case for sure.
8036	 *
8037	 * However if the dev root is fine, but the tree itself is corrupted
8038	 * we'd still fail to mount.  This verification is only to make sure
8039	 * writes can happen safely, so instead just bypass this check
8040	 * completely in the case of IGNOREBADROOTS.
8041	 */
8042	if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8043		return 0;
8044
8045	key.objectid = 1;
8046	key.type = BTRFS_DEV_EXTENT_KEY;
8047	key.offset = 0;
8048
8049	path = btrfs_alloc_path();
8050	if (!path)
8051		return -ENOMEM;
8052
8053	path->reada = READA_FORWARD;
8054	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8055	if (ret < 0)
8056		goto out;
8057
8058	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8059		ret = btrfs_next_leaf(root, path);
8060		if (ret < 0)
8061			goto out;
8062		/* No dev extents at all? Not good */
8063		if (ret > 0) {
8064			ret = -EUCLEAN;
8065			goto out;
8066		}
8067	}
8068	while (1) {
8069		struct extent_buffer *leaf = path->nodes[0];
8070		struct btrfs_dev_extent *dext;
8071		int slot = path->slots[0];
8072		u64 chunk_offset;
8073		u64 physical_offset;
8074		u64 physical_len;
8075		u64 devid;
8076
8077		btrfs_item_key_to_cpu(leaf, &key, slot);
8078		if (key.type != BTRFS_DEV_EXTENT_KEY)
8079			break;
8080		devid = key.objectid;
8081		physical_offset = key.offset;
8082
8083		dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8084		chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8085		physical_len = btrfs_dev_extent_length(leaf, dext);
8086
8087		/* Check if this dev extent overlaps with the previous one */
8088		if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8089			btrfs_err(fs_info,
8090"dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8091				  devid, physical_offset, prev_dev_ext_end);
8092			ret = -EUCLEAN;
8093			goto out;
8094		}
8095
8096		ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8097					    physical_offset, physical_len);
8098		if (ret < 0)
8099			goto out;
8100		prev_devid = devid;
8101		prev_dev_ext_end = physical_offset + physical_len;
8102
8103		ret = btrfs_next_item(root, path);
8104		if (ret < 0)
8105			goto out;
8106		if (ret > 0) {
8107			ret = 0;
8108			break;
8109		}
8110	}
8111
8112	/* Ensure all chunks have corresponding dev extents */
8113	ret = verify_chunk_dev_extent_mapping(fs_info);
8114out:
8115	btrfs_free_path(path);
8116	return ret;
8117}
8118
8119/*
8120 * Check whether the given block group or device is pinned by any inode being
8121 * used as a swapfile.
8122 */
8123bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8124{
8125	struct btrfs_swapfile_pin *sp;
8126	struct rb_node *node;
8127
8128	spin_lock(&fs_info->swapfile_pins_lock);
8129	node = fs_info->swapfile_pins.rb_node;
8130	while (node) {
8131		sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8132		if (ptr < sp->ptr)
8133			node = node->rb_left;
8134		else if (ptr > sp->ptr)
8135			node = node->rb_right;
8136		else
8137			break;
8138	}
8139	spin_unlock(&fs_info->swapfile_pins_lock);
8140	return node != NULL;
8141}
8142
8143static int relocating_repair_kthread(void *data)
8144{
8145	struct btrfs_block_group *cache = data;
8146	struct btrfs_fs_info *fs_info = cache->fs_info;
8147	u64 target;
8148	int ret = 0;
8149
8150	target = cache->start;
8151	btrfs_put_block_group(cache);
8152
8153	sb_start_write(fs_info->sb);
8154	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8155		btrfs_info(fs_info,
8156			   "zoned: skip relocating block group %llu to repair: EBUSY",
8157			   target);
8158		sb_end_write(fs_info->sb);
8159		return -EBUSY;
8160	}
8161
8162	mutex_lock(&fs_info->reclaim_bgs_lock);
8163
8164	/* Ensure block group still exists */
8165	cache = btrfs_lookup_block_group(fs_info, target);
8166	if (!cache)
8167		goto out;
8168
8169	if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
8170		goto out;
8171
8172	ret = btrfs_may_alloc_data_chunk(fs_info, target);
8173	if (ret < 0)
8174		goto out;
8175
8176	btrfs_info(fs_info,
8177		   "zoned: relocating block group %llu to repair IO failure",
8178		   target);
8179	ret = btrfs_relocate_chunk(fs_info, target);
8180
8181out:
8182	if (cache)
8183		btrfs_put_block_group(cache);
8184	mutex_unlock(&fs_info->reclaim_bgs_lock);
8185	btrfs_exclop_finish(fs_info);
8186	sb_end_write(fs_info->sb);
8187
8188	return ret;
8189}
8190
8191bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8192{
8193	struct btrfs_block_group *cache;
8194
8195	if (!btrfs_is_zoned(fs_info))
8196		return false;
8197
8198	/* Do not attempt to repair in degraded state */
8199	if (btrfs_test_opt(fs_info, DEGRADED))
8200		return true;
8201
8202	cache = btrfs_lookup_block_group(fs_info, logical);
8203	if (!cache)
8204		return true;
8205
8206	if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8207		btrfs_put_block_group(cache);
8208		return true;
8209	}
8210
8211	kthread_run(relocating_repair_kthread, cache,
8212		    "btrfs-relocating-repair");
8213
8214	return true;
8215}
8216
8217static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8218				    struct btrfs_io_stripe *smap,
8219				    u64 logical)
8220{
8221	int data_stripes = nr_bioc_data_stripes(bioc);
8222	int i;
8223
8224	for (i = 0; i < data_stripes; i++) {
8225		u64 stripe_start = bioc->full_stripe_logical +
8226				   btrfs_stripe_nr_to_offset(i);
8227
8228		if (logical >= stripe_start &&
8229		    logical < stripe_start + BTRFS_STRIPE_LEN)
8230			break;
8231	}
8232	ASSERT(i < data_stripes);
8233	smap->dev = bioc->stripes[i].dev;
8234	smap->physical = bioc->stripes[i].physical +
8235			((logical - bioc->full_stripe_logical) &
8236			 BTRFS_STRIPE_LEN_MASK);
8237}
8238
8239/*
8240 * Map a repair write into a single device.
8241 *
8242 * A repair write is triggered by read time repair or scrub, which would only
8243 * update the contents of a single device.
8244 * Not update any other mirrors nor go through RMW path.
8245 *
8246 * Callers should ensure:
8247 *
8248 * - Call btrfs_bio_counter_inc_blocked() first
8249 * - The range does not cross stripe boundary
8250 * - Has a valid @mirror_num passed in.
8251 */
8252int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8253			   struct btrfs_io_stripe *smap, u64 logical,
8254			   u32 length, int mirror_num)
8255{
8256	struct btrfs_io_context *bioc = NULL;
8257	u64 map_length = length;
8258	int mirror_ret = mirror_num;
8259	int ret;
8260
8261	ASSERT(mirror_num > 0);
8262
8263	ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8264			      &bioc, smap, &mirror_ret);
8265	if (ret < 0)
8266		return ret;
8267
8268	/* The map range should not cross stripe boundary. */
8269	ASSERT(map_length >= length);
8270
8271	/* Already mapped to single stripe. */
8272	if (!bioc)
8273		goto out;
8274
8275	/* Map the RAID56 multi-stripe writes to a single one. */
8276	if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8277		map_raid56_repair_block(bioc, smap, logical);
8278		goto out;
8279	}
8280
8281	ASSERT(mirror_num <= bioc->num_stripes);
8282	smap->dev = bioc->stripes[mirror_num - 1].dev;
8283	smap->physical = bioc->stripes[mirror_num - 1].physical;
8284out:
8285	btrfs_put_bioc(bioc);
8286	ASSERT(smap->dev);
8287	return 0;
8288}