1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2007 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/kernel.h>
   7#include <linux/bio.h>
 
   8#include <linux/file.h>
   9#include <linux/fs.h>
  10#include <linux/fsnotify.h>
  11#include <linux/pagemap.h>
  12#include <linux/highmem.h>
  13#include <linux/time.h>
 
  14#include <linux/string.h>
  15#include <linux/backing-dev.h>
  16#include <linux/mount.h>
 
  17#include <linux/namei.h>
 
  18#include <linux/writeback.h>
  19#include <linux/compat.h>
 
  20#include <linux/security.h>
  21#include <linux/xattr.h>
  22#include <linux/mm.h>
  23#include <linux/slab.h>
  24#include <linux/blkdev.h>
  25#include <linux/uuid.h>
  26#include <linux/btrfs.h>
  27#include <linux/uaccess.h>
  28#include <linux/iversion.h>
  29#include "ctree.h"
  30#include "disk-io.h"
  31#include "transaction.h"
  32#include "btrfs_inode.h"
  33#include "print-tree.h"
  34#include "volumes.h"
  35#include "locking.h"
  36#include "inode-map.h"
  37#include "backref.h"
  38#include "rcu-string.h"
  39#include "send.h"
  40#include "dev-replace.h"
  41#include "props.h"
  42#include "sysfs.h"
  43#include "qgroup.h"
  44#include "tree-log.h"
  45#include "compression.h"
  46#include "space-info.h"
  47#include "delalloc-space.h"
  48#include "block-group.h"
  49
  50#ifdef CONFIG_64BIT
  51/* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  52 * structures are incorrect, as the timespec structure from userspace
  53 * is 4 bytes too small. We define these alternatives here to teach
  54 * the kernel about the 32-bit struct packing.
  55 */
  56struct btrfs_ioctl_timespec_32 {
  57	__u64 sec;
  58	__u32 nsec;
  59} __attribute__ ((__packed__));
  60
  61struct btrfs_ioctl_received_subvol_args_32 {
  62	char	uuid[BTRFS_UUID_SIZE];	/* in */
  63	__u64	stransid;		/* in */
  64	__u64	rtransid;		/* out */
  65	struct btrfs_ioctl_timespec_32 stime; /* in */
  66	struct btrfs_ioctl_timespec_32 rtime; /* out */
  67	__u64	flags;			/* in */
  68	__u64	reserved[16];		/* in */
  69} __attribute__ ((__packed__));
  70
  71#define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  72				struct btrfs_ioctl_received_subvol_args_32)
  73#endif
  74
  75#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  76struct btrfs_ioctl_send_args_32 {
  77	__s64 send_fd;			/* in */
  78	__u64 clone_sources_count;	/* in */
  79	compat_uptr_t clone_sources;	/* in */
  80	__u64 parent_root;		/* in */
  81	__u64 flags;			/* in */
  82	__u64 reserved[4];		/* in */
  83} __attribute__ ((__packed__));
  84
  85#define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  86			       struct btrfs_ioctl_send_args_32)
  87#endif
  88
  89static int btrfs_clone(struct inode *src, struct inode *inode,
  90		       u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  91		       int no_time_update);
  92
  93/* Mask out flags that are inappropriate for the given type of inode. */
  94static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  95		unsigned int flags)
  96{
  97	if (S_ISDIR(inode->i_mode))
  98		return flags;
  99	else if (S_ISREG(inode->i_mode))
 100		return flags & ~FS_DIRSYNC_FL;
 101	else
 102		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 103}
 104
 105/*
 106 * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 107 * ioctl.
 108 */
 109static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 110{
 111	unsigned int iflags = 0;
 112
 113	if (flags & BTRFS_INODE_SYNC)
 114		iflags |= FS_SYNC_FL;
 115	if (flags & BTRFS_INODE_IMMUTABLE)
 116		iflags |= FS_IMMUTABLE_FL;
 117	if (flags & BTRFS_INODE_APPEND)
 118		iflags |= FS_APPEND_FL;
 119	if (flags & BTRFS_INODE_NODUMP)
 120		iflags |= FS_NODUMP_FL;
 121	if (flags & BTRFS_INODE_NOATIME)
 122		iflags |= FS_NOATIME_FL;
 123	if (flags & BTRFS_INODE_DIRSYNC)
 124		iflags |= FS_DIRSYNC_FL;
 125	if (flags & BTRFS_INODE_NODATACOW)
 126		iflags |= FS_NOCOW_FL;
 127
 128	if (flags & BTRFS_INODE_NOCOMPRESS)
 129		iflags |= FS_NOCOMP_FL;
 130	else if (flags & BTRFS_INODE_COMPRESS)
 131		iflags |= FS_COMPR_FL;
 132
 133	return iflags;
 134}
 135
 136/*
 137 * Update inode->i_flags based on the btrfs internal flags.
 138 */
 139void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 140{
 141	struct btrfs_inode *binode = BTRFS_I(inode);
 142	unsigned int new_fl = 0;
 143
 144	if (binode->flags & BTRFS_INODE_SYNC)
 145		new_fl |= S_SYNC;
 146	if (binode->flags & BTRFS_INODE_IMMUTABLE)
 147		new_fl |= S_IMMUTABLE;
 148	if (binode->flags & BTRFS_INODE_APPEND)
 149		new_fl |= S_APPEND;
 150	if (binode->flags & BTRFS_INODE_NOATIME)
 151		new_fl |= S_NOATIME;
 152	if (binode->flags & BTRFS_INODE_DIRSYNC)
 153		new_fl |= S_DIRSYNC;
 154
 155	set_mask_bits(&inode->i_flags,
 156		      S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157		      new_fl);
 158}
 159
 160static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 161{
 162	struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 163	unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 164
 165	if (copy_to_user(arg, &flags, sizeof(flags)))
 166		return -EFAULT;
 167	return 0;
 168}
 169
 170/* Check if @flags are a supported and valid set of FS_*_FL flags */
 171static int check_fsflags(unsigned int flags)
 172{
 173	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 174		      FS_NOATIME_FL | FS_NODUMP_FL | \
 175		      FS_SYNC_FL | FS_DIRSYNC_FL | \
 176		      FS_NOCOMP_FL | FS_COMPR_FL |
 177		      FS_NOCOW_FL))
 178		return -EOPNOTSUPP;
 179
 180	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 181		return -EINVAL;
 182
 183	return 0;
 184}
 185
 186static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 187{
 188	struct inode *inode = file_inode(file);
 189	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 190	struct btrfs_inode *binode = BTRFS_I(inode);
 191	struct btrfs_root *root = binode->root;
 192	struct btrfs_trans_handle *trans;
 193	unsigned int fsflags, old_fsflags;
 194	int ret;
 195	const char *comp = NULL;
 196	u32 binode_flags = binode->flags;
 
 197
 198	if (!inode_owner_or_capable(inode))
 199		return -EPERM;
 200
 201	if (btrfs_root_readonly(root))
 202		return -EROFS;
 203
 204	if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 205		return -EFAULT;
 206
 207	ret = check_fsflags(fsflags);
 208	if (ret)
 209		return ret;
 210
 211	ret = mnt_want_write_file(file);
 212	if (ret)
 213		return ret;
 214
 215	inode_lock(inode);
 216
 217	fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 218	old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 219	ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 220	if (ret)
 221		goto out_unlock;
 
 
 
 
 
 
 
 222
 223	if (fsflags & FS_SYNC_FL)
 224		binode_flags |= BTRFS_INODE_SYNC;
 225	else
 226		binode_flags &= ~BTRFS_INODE_SYNC;
 227	if (fsflags & FS_IMMUTABLE_FL)
 228		binode_flags |= BTRFS_INODE_IMMUTABLE;
 229	else
 230		binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 231	if (fsflags & FS_APPEND_FL)
 232		binode_flags |= BTRFS_INODE_APPEND;
 233	else
 234		binode_flags &= ~BTRFS_INODE_APPEND;
 235	if (fsflags & FS_NODUMP_FL)
 236		binode_flags |= BTRFS_INODE_NODUMP;
 237	else
 238		binode_flags &= ~BTRFS_INODE_NODUMP;
 239	if (fsflags & FS_NOATIME_FL)
 240		binode_flags |= BTRFS_INODE_NOATIME;
 241	else
 242		binode_flags &= ~BTRFS_INODE_NOATIME;
 243	if (fsflags & FS_DIRSYNC_FL)
 244		binode_flags |= BTRFS_INODE_DIRSYNC;
 245	else
 246		binode_flags &= ~BTRFS_INODE_DIRSYNC;
 247	if (fsflags & FS_NOCOW_FL) {
 248		if (S_ISREG(inode->i_mode)) {
 249			/*
 250			 * It's safe to turn csums off here, no extents exist.
 251			 * Otherwise we want the flag to reflect the real COW
 252			 * status of the file and will not set it.
 253			 */
 254			if (inode->i_size == 0)
 255				binode_flags |= BTRFS_INODE_NODATACOW |
 256						BTRFS_INODE_NODATASUM;
 257		} else {
 258			binode_flags |= BTRFS_INODE_NODATACOW;
 259		}
 260	} else {
 261		/*
 262		 * Revert back under same assumptions as above
 263		 */
 264		if (S_ISREG(inode->i_mode)) {
 265			if (inode->i_size == 0)
 266				binode_flags &= ~(BTRFS_INODE_NODATACOW |
 267						  BTRFS_INODE_NODATASUM);
 268		} else {
 269			binode_flags &= ~BTRFS_INODE_NODATACOW;
 270		}
 271	}
 272
 273	/*
 274	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 275	 * flag may be changed automatically if compression code won't make
 276	 * things smaller.
 277	 */
 278	if (fsflags & FS_NOCOMP_FL) {
 279		binode_flags &= ~BTRFS_INODE_COMPRESS;
 280		binode_flags |= BTRFS_INODE_NOCOMPRESS;
 281	} else if (fsflags & FS_COMPR_FL) {
 282
 283		if (IS_SWAPFILE(inode)) {
 284			ret = -ETXTBSY;
 285			goto out_unlock;
 286		}
 287
 288		binode_flags |= BTRFS_INODE_COMPRESS;
 289		binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 290
 291		comp = btrfs_compress_type2str(fs_info->compress_type);
 292		if (!comp || comp[0] == 0)
 293			comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 294	} else {
 295		binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 296	}
 297
 298	/*
 299	 * 1 for inode item
 300	 * 2 for properties
 301	 */
 302	trans = btrfs_start_transaction(root, 3);
 303	if (IS_ERR(trans)) {
 304		ret = PTR_ERR(trans);
 305		goto out_unlock;
 306	}
 307
 308	if (comp) {
 309		ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 310				     strlen(comp), 0);
 311		if (ret) {
 312			btrfs_abort_transaction(trans, ret);
 313			goto out_end_trans;
 314		}
 315	} else {
 316		ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 317				     0, 0);
 318		if (ret && ret != -ENODATA) {
 319			btrfs_abort_transaction(trans, ret);
 320			goto out_end_trans;
 321		}
 322	}
 323
 324	binode->flags = binode_flags;
 325	btrfs_sync_inode_flags_to_i_flags(inode);
 326	inode_inc_iversion(inode);
 327	inode->i_ctime = current_time(inode);
 328	ret = btrfs_update_inode(trans, root, inode);
 329
 330 out_end_trans:
 331	btrfs_end_transaction(trans);
 332 out_unlock:
 333	inode_unlock(inode);
 334	mnt_drop_write_file(file);
 335	return ret;
 336}
 337
 338/*
 339 * Translate btrfs internal inode flags to xflags as expected by the
 340 * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 341 * silently dropped.
 342 */
 343static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 344{
 345	unsigned int xflags = 0;
 346
 347	if (flags & BTRFS_INODE_APPEND)
 348		xflags |= FS_XFLAG_APPEND;
 349	if (flags & BTRFS_INODE_IMMUTABLE)
 350		xflags |= FS_XFLAG_IMMUTABLE;
 351	if (flags & BTRFS_INODE_NOATIME)
 352		xflags |= FS_XFLAG_NOATIME;
 353	if (flags & BTRFS_INODE_NODUMP)
 354		xflags |= FS_XFLAG_NODUMP;
 355	if (flags & BTRFS_INODE_SYNC)
 356		xflags |= FS_XFLAG_SYNC;
 357
 358	return xflags;
 359}
 360
 361/* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 362static int check_xflags(unsigned int flags)
 363{
 364	if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 365		      FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 366		return -EOPNOTSUPP;
 367	return 0;
 368}
 369
 370/*
 371 * Set the xflags from the internal inode flags. The remaining items of fsxattr
 372 * are zeroed.
 373 */
 374static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 375{
 376	struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 377	struct fsxattr fa;
 378
 379	simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 380	if (copy_to_user(arg, &fa, sizeof(fa)))
 381		return -EFAULT;
 382
 383	return 0;
 384}
 385
 386static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 387{
 388	struct inode *inode = file_inode(file);
 389	struct btrfs_inode *binode = BTRFS_I(inode);
 390	struct btrfs_root *root = binode->root;
 391	struct btrfs_trans_handle *trans;
 392	struct fsxattr fa, old_fa;
 393	unsigned old_flags;
 394	unsigned old_i_flags;
 395	int ret = 0;
 396
 397	if (!inode_owner_or_capable(inode))
 398		return -EPERM;
 399
 400	if (btrfs_root_readonly(root))
 401		return -EROFS;
 402
 403	if (copy_from_user(&fa, arg, sizeof(fa)))
 404		return -EFAULT;
 405
 406	ret = check_xflags(fa.fsx_xflags);
 407	if (ret)
 408		return ret;
 409
 410	if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 411		return -EOPNOTSUPP;
 412
 413	ret = mnt_want_write_file(file);
 414	if (ret)
 415		return ret;
 416
 417	inode_lock(inode);
 418
 419	old_flags = binode->flags;
 420	old_i_flags = inode->i_flags;
 421
 422	simple_fill_fsxattr(&old_fa,
 423			    btrfs_inode_flags_to_xflags(binode->flags));
 424	ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 425	if (ret)
 426		goto out_unlock;
 427
 428	if (fa.fsx_xflags & FS_XFLAG_SYNC)
 429		binode->flags |= BTRFS_INODE_SYNC;
 430	else
 431		binode->flags &= ~BTRFS_INODE_SYNC;
 432	if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 433		binode->flags |= BTRFS_INODE_IMMUTABLE;
 434	else
 435		binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 436	if (fa.fsx_xflags & FS_XFLAG_APPEND)
 437		binode->flags |= BTRFS_INODE_APPEND;
 438	else
 439		binode->flags &= ~BTRFS_INODE_APPEND;
 440	if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 441		binode->flags |= BTRFS_INODE_NODUMP;
 442	else
 443		binode->flags &= ~BTRFS_INODE_NODUMP;
 444	if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 445		binode->flags |= BTRFS_INODE_NOATIME;
 446	else
 447		binode->flags &= ~BTRFS_INODE_NOATIME;
 448
 449	/* 1 item for the inode */
 450	trans = btrfs_start_transaction(root, 1);
 451	if (IS_ERR(trans)) {
 452		ret = PTR_ERR(trans);
 453		goto out_unlock;
 454	}
 455
 456	btrfs_sync_inode_flags_to_i_flags(inode);
 457	inode_inc_iversion(inode);
 458	inode->i_ctime = current_time(inode);
 459	ret = btrfs_update_inode(trans, root, inode);
 460
 461	btrfs_end_transaction(trans);
 462
 463out_unlock:
 464	if (ret) {
 465		binode->flags = old_flags;
 466		inode->i_flags = old_i_flags;
 467	}
 468
 
 469	inode_unlock(inode);
 470	mnt_drop_write_file(file);
 471
 472	return ret;
 473}
 474
 475static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 476{
 477	struct inode *inode = file_inode(file);
 478
 479	return put_user(inode->i_generation, arg);
 480}
 481
 482static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 483{
 484	struct inode *inode = file_inode(file);
 485	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 486	struct btrfs_device *device;
 487	struct request_queue *q;
 488	struct fstrim_range range;
 489	u64 minlen = ULLONG_MAX;
 490	u64 num_devices = 0;
 
 491	int ret;
 492
 493	if (!capable(CAP_SYS_ADMIN))
 494		return -EPERM;
 495
 496	/*
 497	 * If the fs is mounted with nologreplay, which requires it to be
 498	 * mounted in RO mode as well, we can not allow discard on free space
 499	 * inside block groups, because log trees refer to extents that are not
 500	 * pinned in a block group's free space cache (pinning the extents is
 501	 * precisely the first phase of replaying a log tree).
 502	 */
 503	if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 504		return -EROFS;
 505
 506	rcu_read_lock();
 507	list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 508				dev_list) {
 509		if (!device->bdev)
 510			continue;
 511		q = bdev_get_queue(device->bdev);
 512		if (blk_queue_discard(q)) {
 513			num_devices++;
 514			minlen = min_t(u64, q->limits.discard_granularity,
 515				     minlen);
 516		}
 517	}
 518	rcu_read_unlock();
 519
 520	if (!num_devices)
 521		return -EOPNOTSUPP;
 522	if (copy_from_user(&range, arg, sizeof(range)))
 523		return -EFAULT;
 524
 525	/*
 526	 * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 527	 * block group is in the logical address space, which can be any
 528	 * sectorsize aligned bytenr in  the range [0, U64_MAX].
 529	 */
 530	if (range.len < fs_info->sb->s_blocksize)
 531		return -EINVAL;
 532
 
 533	range.minlen = max(range.minlen, minlen);
 534	ret = btrfs_trim_fs(fs_info, &range);
 535	if (ret < 0)
 536		return ret;
 537
 538	if (copy_to_user(arg, &range, sizeof(range)))
 539		return -EFAULT;
 540
 541	return 0;
 542}
 543
 544int btrfs_is_empty_uuid(u8 *uuid)
 545{
 546	int i;
 547
 548	for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 549		if (uuid[i])
 550			return 0;
 551	}
 552	return 1;
 553}
 554
 555static noinline int create_subvol(struct inode *dir,
 556				  struct dentry *dentry,
 557				  const char *name, int namelen,
 558				  u64 *async_transid,
 559				  struct btrfs_qgroup_inherit *inherit)
 560{
 561	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 562	struct btrfs_trans_handle *trans;
 563	struct btrfs_key key;
 564	struct btrfs_root_item *root_item;
 565	struct btrfs_inode_item *inode_item;
 566	struct extent_buffer *leaf;
 567	struct btrfs_root *root = BTRFS_I(dir)->root;
 568	struct btrfs_root *new_root;
 569	struct btrfs_block_rsv block_rsv;
 570	struct timespec64 cur_time = current_time(dir);
 571	struct inode *inode;
 572	int ret;
 573	int err;
 574	u64 objectid;
 575	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 576	u64 index = 0;
 
 577	uuid_le new_uuid;
 578
 579	root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 580	if (!root_item)
 581		return -ENOMEM;
 582
 583	ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 584	if (ret)
 585		goto fail_free;
 586
 587	/*
 588	 * Don't create subvolume whose level is not zero. Or qgroup will be
 589	 * screwed up since it assumes subvolume qgroup's level to be 0.
 590	 */
 591	if (btrfs_qgroup_level(objectid)) {
 592		ret = -ENOSPC;
 593		goto fail_free;
 594	}
 595
 596	btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 597	/*
 598	 * The same as the snapshot creation, please see the comment
 599	 * of create_snapshot().
 600	 */
 601	ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 
 602	if (ret)
 603		goto fail_free;
 604
 605	trans = btrfs_start_transaction(root, 0);
 606	if (IS_ERR(trans)) {
 607		ret = PTR_ERR(trans);
 608		btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 609		goto fail_free;
 610	}
 611	trans->block_rsv = &block_rsv;
 612	trans->bytes_reserved = block_rsv.size;
 613
 614	ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 615	if (ret)
 616		goto fail;
 617
 618	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 619	if (IS_ERR(leaf)) {
 620		ret = PTR_ERR(leaf);
 621		goto fail;
 622	}
 623
 
 
 
 
 
 
 
 
 624	btrfs_mark_buffer_dirty(leaf);
 625
 626	inode_item = &root_item->inode;
 627	btrfs_set_stack_inode_generation(inode_item, 1);
 628	btrfs_set_stack_inode_size(inode_item, 3);
 629	btrfs_set_stack_inode_nlink(inode_item, 1);
 630	btrfs_set_stack_inode_nbytes(inode_item,
 631				     fs_info->nodesize);
 632	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 633
 634	btrfs_set_root_flags(root_item, 0);
 635	btrfs_set_root_limit(root_item, 0);
 636	btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 637
 638	btrfs_set_root_bytenr(root_item, leaf->start);
 639	btrfs_set_root_generation(root_item, trans->transid);
 640	btrfs_set_root_level(root_item, 0);
 641	btrfs_set_root_refs(root_item, 1);
 642	btrfs_set_root_used(root_item, leaf->len);
 643	btrfs_set_root_last_snapshot(root_item, 0);
 644
 645	btrfs_set_root_generation_v2(root_item,
 646			btrfs_root_generation(root_item));
 647	uuid_le_gen(&new_uuid);
 648	memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 649	btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 650	btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 651	root_item->ctime = root_item->otime;
 652	btrfs_set_root_ctransid(root_item, trans->transid);
 653	btrfs_set_root_otransid(root_item, trans->transid);
 654
 655	btrfs_tree_unlock(leaf);
 656	free_extent_buffer(leaf);
 657	leaf = NULL;
 658
 659	btrfs_set_root_dirid(root_item, new_dirid);
 660
 661	key.objectid = objectid;
 662	key.offset = 0;
 663	key.type = BTRFS_ROOT_ITEM_KEY;
 664	ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 665				root_item);
 666	if (ret)
 667		goto fail;
 668
 669	key.offset = (u64)-1;
 670	new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 671	if (IS_ERR(new_root)) {
 672		ret = PTR_ERR(new_root);
 673		btrfs_abort_transaction(trans, ret);
 674		goto fail;
 675	}
 676
 677	btrfs_record_root_in_trans(trans, new_root);
 678
 679	ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 680	if (ret) {
 681		/* We potentially lose an unused inode item here */
 682		btrfs_abort_transaction(trans, ret);
 683		goto fail;
 684	}
 685
 686	mutex_lock(&new_root->objectid_mutex);
 687	new_root->highest_objectid = new_dirid;
 688	mutex_unlock(&new_root->objectid_mutex);
 689
 690	/*
 691	 * insert the directory item
 692	 */
 693	ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 694	if (ret) {
 695		btrfs_abort_transaction(trans, ret);
 696		goto fail;
 697	}
 698
 699	ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 
 700				    BTRFS_FT_DIR, index);
 701	if (ret) {
 702		btrfs_abort_transaction(trans, ret);
 703		goto fail;
 704	}
 705
 706	btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 707	ret = btrfs_update_inode(trans, root, dir);
 708	BUG_ON(ret);
 709
 710	ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 
 711				 btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 712	BUG_ON(ret);
 713
 714	ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 715				  BTRFS_UUID_KEY_SUBVOL, objectid);
 716	if (ret)
 717		btrfs_abort_transaction(trans, ret);
 718
 719fail:
 720	kfree(root_item);
 721	trans->block_rsv = NULL;
 722	trans->bytes_reserved = 0;
 723	btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 724
 725	if (async_transid) {
 726		*async_transid = trans->transid;
 727		err = btrfs_commit_transaction_async(trans, 1);
 728		if (err)
 729			err = btrfs_commit_transaction(trans);
 730	} else {
 731		err = btrfs_commit_transaction(trans);
 732	}
 733	if (err && !ret)
 734		ret = err;
 735
 736	if (!ret) {
 737		inode = btrfs_lookup_dentry(dir, dentry);
 738		if (IS_ERR(inode))
 739			return PTR_ERR(inode);
 740		d_instantiate(dentry, inode);
 741	}
 742	return ret;
 743
 744fail_free:
 745	kfree(root_item);
 746	return ret;
 747}
 748
 749static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 750			   struct dentry *dentry,
 751			   u64 *async_transid, bool readonly,
 752			   struct btrfs_qgroup_inherit *inherit)
 753{
 754	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 755	struct inode *inode;
 756	struct btrfs_pending_snapshot *pending_snapshot;
 757	struct btrfs_trans_handle *trans;
 758	int ret;
 759	bool snapshot_force_cow = false;
 760
 761	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 762		return -EINVAL;
 763
 764	if (atomic_read(&root->nr_swapfiles)) {
 765		btrfs_warn(fs_info,
 766			   "cannot snapshot subvolume with active swapfile");
 767		return -ETXTBSY;
 768	}
 769
 770	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 771	if (!pending_snapshot)
 772		return -ENOMEM;
 773
 774	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 775			GFP_KERNEL);
 776	pending_snapshot->path = btrfs_alloc_path();
 777	if (!pending_snapshot->root_item || !pending_snapshot->path) {
 778		ret = -ENOMEM;
 779		goto free_pending;
 780	}
 781
 782	/*
 783	 * Force new buffered writes to reserve space even when NOCOW is
 784	 * possible. This is to avoid later writeback (running dealloc) to
 785	 * fallback to COW mode and unexpectedly fail with ENOSPC.
 786	 */
 787	atomic_inc(&root->will_be_snapshotted);
 788	smp_mb__after_atomic();
 789	/* wait for no snapshot writes */
 790	wait_event(root->subv_writers->wait,
 791		   percpu_counter_sum(&root->subv_writers->counter) == 0);
 792
 793	ret = btrfs_start_delalloc_snapshot(root);
 794	if (ret)
 795		goto dec_and_free;
 796
 797	/*
 798	 * All previous writes have started writeback in NOCOW mode, so now
 799	 * we force future writes to fallback to COW mode during snapshot
 800	 * creation.
 801	 */
 802	atomic_inc(&root->snapshot_force_cow);
 803	snapshot_force_cow = true;
 804
 805	btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 806
 807	btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 808			     BTRFS_BLOCK_RSV_TEMP);
 809	/*
 810	 * 1 - parent dir inode
 811	 * 2 - dir entries
 812	 * 1 - root item
 813	 * 2 - root ref/backref
 814	 * 1 - root of snapshot
 815	 * 1 - UUID item
 816	 */
 817	ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 818					&pending_snapshot->block_rsv, 8,
 
 819					false);
 820	if (ret)
 821		goto dec_and_free;
 822
 823	pending_snapshot->dentry = dentry;
 824	pending_snapshot->root = root;
 825	pending_snapshot->readonly = readonly;
 826	pending_snapshot->dir = dir;
 827	pending_snapshot->inherit = inherit;
 828
 829	trans = btrfs_start_transaction(root, 0);
 830	if (IS_ERR(trans)) {
 831		ret = PTR_ERR(trans);
 832		goto fail;
 833	}
 834
 835	spin_lock(&fs_info->trans_lock);
 836	list_add(&pending_snapshot->list,
 837		 &trans->transaction->pending_snapshots);
 838	spin_unlock(&fs_info->trans_lock);
 839	if (async_transid) {
 840		*async_transid = trans->transid;
 841		ret = btrfs_commit_transaction_async(trans, 1);
 842		if (ret)
 843			ret = btrfs_commit_transaction(trans);
 844	} else {
 845		ret = btrfs_commit_transaction(trans);
 846	}
 847	if (ret)
 848		goto fail;
 849
 850	ret = pending_snapshot->error;
 851	if (ret)
 852		goto fail;
 853
 854	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 855	if (ret)
 856		goto fail;
 857
 858	inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 859	if (IS_ERR(inode)) {
 860		ret = PTR_ERR(inode);
 861		goto fail;
 862	}
 863
 864	d_instantiate(dentry, inode);
 865	ret = 0;
 866fail:
 867	btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 868dec_and_free:
 869	if (snapshot_force_cow)
 870		atomic_dec(&root->snapshot_force_cow);
 871	if (atomic_dec_and_test(&root->will_be_snapshotted))
 872		wake_up_var(&root->will_be_snapshotted);
 873free_pending:
 874	kfree(pending_snapshot->root_item);
 875	btrfs_free_path(pending_snapshot->path);
 876	kfree(pending_snapshot);
 877
 878	return ret;
 879}
 880
 881/*  copy of may_delete in fs/namei.c()
 882 *	Check whether we can remove a link victim from directory dir, check
 883 *  whether the type of victim is right.
 884 *  1. We can't do it if dir is read-only (done in permission())
 885 *  2. We should have write and exec permissions on dir
 886 *  3. We can't remove anything from append-only dir
 887 *  4. We can't do anything with immutable dir (done in permission())
 888 *  5. If the sticky bit on dir is set we should either
 889 *	a. be owner of dir, or
 890 *	b. be owner of victim, or
 891 *	c. have CAP_FOWNER capability
 892 *  6. If the victim is append-only or immutable we can't do anything with
 893 *     links pointing to it.
 894 *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 895 *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 896 *  9. We can't remove a root or mountpoint.
 897 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 898 *     nfs_async_unlink().
 899 */
 900
 901static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 902{
 903	int error;
 904
 905	if (d_really_is_negative(victim))
 906		return -ENOENT;
 907
 908	BUG_ON(d_inode(victim->d_parent) != dir);
 909	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 910
 911	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 912	if (error)
 913		return error;
 914	if (IS_APPEND(dir))
 915		return -EPERM;
 916	if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 917	    IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 918		return -EPERM;
 919	if (isdir) {
 920		if (!d_is_dir(victim))
 921			return -ENOTDIR;
 922		if (IS_ROOT(victim))
 923			return -EBUSY;
 924	} else if (d_is_dir(victim))
 925		return -EISDIR;
 926	if (IS_DEADDIR(dir))
 927		return -ENOENT;
 928	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 929		return -EBUSY;
 930	return 0;
 931}
 932
 933/* copy of may_create in fs/namei.c() */
 934static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 935{
 936	if (d_really_is_positive(child))
 937		return -EEXIST;
 938	if (IS_DEADDIR(dir))
 939		return -ENOENT;
 940	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 941}
 942
 943/*
 944 * Create a new subvolume below @parent.  This is largely modeled after
 945 * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 946 * inside this filesystem so it's quite a bit simpler.
 947 */
 948static noinline int btrfs_mksubvol(const struct path *parent,
 949				   const char *name, int namelen,
 950				   struct btrfs_root *snap_src,
 951				   u64 *async_transid, bool readonly,
 952				   struct btrfs_qgroup_inherit *inherit)
 953{
 954	struct inode *dir = d_inode(parent->dentry);
 955	struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 956	struct dentry *dentry;
 957	int error;
 958
 959	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 960	if (error == -EINTR)
 961		return error;
 962
 963	dentry = lookup_one_len(name, parent->dentry, namelen);
 964	error = PTR_ERR(dentry);
 965	if (IS_ERR(dentry))
 966		goto out_unlock;
 967
 968	error = btrfs_may_create(dir, dentry);
 969	if (error)
 970		goto out_dput;
 971
 972	/*
 973	 * even if this name doesn't exist, we may get hash collisions.
 974	 * check for them now when we can safely fail
 975	 */
 976	error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 977					       dir->i_ino, name,
 978					       namelen);
 979	if (error)
 980		goto out_dput;
 981
 982	down_read(&fs_info->subvol_sem);
 983
 984	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 985		goto out_up_read;
 986
 987	if (snap_src) {
 988		error = create_snapshot(snap_src, dir, dentry,
 989					async_transid, readonly, inherit);
 990	} else {
 991		error = create_subvol(dir, dentry, name, namelen,
 992				      async_transid, inherit);
 993	}
 994	if (!error)
 995		fsnotify_mkdir(dir, dentry);
 996out_up_read:
 997	up_read(&fs_info->subvol_sem);
 998out_dput:
 999	dput(dentry);
1000out_unlock:
1001	inode_unlock(dir);
1002	return error;
1003}
1004
1005/*
1006 * When we're defragging a range, we don't want to kick it off again
1007 * if it is really just waiting for delalloc to send it down.
1008 * If we find a nice big extent or delalloc range for the bytes in the
1009 * file you want to defrag, we return 0 to let you know to skip this
1010 * part of the file
1011 */
1012static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1013{
1014	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1015	struct extent_map *em = NULL;
1016	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1017	u64 end;
1018
1019	read_lock(&em_tree->lock);
1020	em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1021	read_unlock(&em_tree->lock);
1022
1023	if (em) {
1024		end = extent_map_end(em);
1025		free_extent_map(em);
1026		if (end - offset > thresh)
1027			return 0;
1028	}
1029	/* if we already have a nice delalloc here, just stop */
1030	thresh /= 2;
1031	end = count_range_bits(io_tree, &offset, offset + thresh,
1032			       thresh, EXTENT_DELALLOC, 1);
1033	if (end >= thresh)
1034		return 0;
1035	return 1;
1036}
1037
1038/*
1039 * helper function to walk through a file and find extents
1040 * newer than a specific transid, and smaller than thresh.
1041 *
1042 * This is used by the defragging code to find new and small
1043 * extents
1044 */
1045static int find_new_extents(struct btrfs_root *root,
1046			    struct inode *inode, u64 newer_than,
1047			    u64 *off, u32 thresh)
1048{
1049	struct btrfs_path *path;
1050	struct btrfs_key min_key;
1051	struct extent_buffer *leaf;
1052	struct btrfs_file_extent_item *extent;
1053	int type;
1054	int ret;
1055	u64 ino = btrfs_ino(BTRFS_I(inode));
1056
1057	path = btrfs_alloc_path();
1058	if (!path)
1059		return -ENOMEM;
1060
1061	min_key.objectid = ino;
1062	min_key.type = BTRFS_EXTENT_DATA_KEY;
1063	min_key.offset = *off;
1064
1065	while (1) {
1066		ret = btrfs_search_forward(root, &min_key, path, newer_than);
1067		if (ret != 0)
1068			goto none;
1069process_slot:
1070		if (min_key.objectid != ino)
1071			goto none;
1072		if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1073			goto none;
1074
1075		leaf = path->nodes[0];
1076		extent = btrfs_item_ptr(leaf, path->slots[0],
1077					struct btrfs_file_extent_item);
1078
1079		type = btrfs_file_extent_type(leaf, extent);
1080		if (type == BTRFS_FILE_EXTENT_REG &&
1081		    btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1082		    check_defrag_in_cache(inode, min_key.offset, thresh)) {
1083			*off = min_key.offset;
1084			btrfs_free_path(path);
1085			return 0;
1086		}
1087
1088		path->slots[0]++;
1089		if (path->slots[0] < btrfs_header_nritems(leaf)) {
1090			btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1091			goto process_slot;
1092		}
1093
1094		if (min_key.offset == (u64)-1)
1095			goto none;
1096
1097		min_key.offset++;
1098		btrfs_release_path(path);
1099	}
1100none:
1101	btrfs_free_path(path);
1102	return -ENOENT;
1103}
1104
1105static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1106{
1107	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1108	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1109	struct extent_map *em;
1110	u64 len = PAGE_SIZE;
1111
1112	/*
1113	 * hopefully we have this extent in the tree already, try without
1114	 * the full extent lock
1115	 */
1116	read_lock(&em_tree->lock);
1117	em = lookup_extent_mapping(em_tree, start, len);
1118	read_unlock(&em_tree->lock);
1119
1120	if (!em) {
1121		struct extent_state *cached = NULL;
1122		u64 end = start + len - 1;
1123
1124		/* get the big lock and read metadata off disk */
1125		lock_extent_bits(io_tree, start, end, &cached);
1126		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1127		unlock_extent_cached(io_tree, start, end, &cached);
1128
1129		if (IS_ERR(em))
1130			return NULL;
1131	}
1132
1133	return em;
1134}
1135
1136static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1137{
1138	struct extent_map *next;
1139	bool ret = true;
1140
1141	/* this is the last extent */
1142	if (em->start + em->len >= i_size_read(inode))
1143		return false;
1144
1145	next = defrag_lookup_extent(inode, em->start + em->len);
1146	if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1147		ret = false;
1148	else if ((em->block_start + em->block_len == next->block_start) &&
1149		 (em->block_len > SZ_128K && next->block_len > SZ_128K))
1150		ret = false;
1151
1152	free_extent_map(next);
1153	return ret;
1154}
1155
1156static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1157			       u64 *last_len, u64 *skip, u64 *defrag_end,
1158			       int compress)
1159{
1160	struct extent_map *em;
1161	int ret = 1;
1162	bool next_mergeable = true;
1163	bool prev_mergeable = true;
1164
1165	/*
1166	 * make sure that once we start defragging an extent, we keep on
1167	 * defragging it
1168	 */
1169	if (start < *defrag_end)
1170		return 1;
1171
1172	*skip = 0;
1173
1174	em = defrag_lookup_extent(inode, start);
1175	if (!em)
1176		return 0;
1177
1178	/* this will cover holes, and inline extents */
1179	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1180		ret = 0;
1181		goto out;
1182	}
1183
1184	if (!*defrag_end)
1185		prev_mergeable = false;
1186
1187	next_mergeable = defrag_check_next_extent(inode, em);
1188	/*
1189	 * we hit a real extent, if it is big or the next extent is not a
1190	 * real extent, don't bother defragging it
1191	 */
1192	if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1193	    (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1194		ret = 0;
1195out:
1196	/*
1197	 * last_len ends up being a counter of how many bytes we've defragged.
1198	 * every time we choose not to defrag an extent, we reset *last_len
1199	 * so that the next tiny extent will force a defrag.
1200	 *
1201	 * The end result of this is that tiny extents before a single big
1202	 * extent will force at least part of that big extent to be defragged.
1203	 */
1204	if (ret) {
1205		*defrag_end = extent_map_end(em);
1206	} else {
1207		*last_len = 0;
1208		*skip = extent_map_end(em);
1209		*defrag_end = 0;
1210	}
1211
1212	free_extent_map(em);
1213	return ret;
1214}
1215
1216/*
1217 * it doesn't do much good to defrag one or two pages
1218 * at a time.  This pulls in a nice chunk of pages
1219 * to COW and defrag.
1220 *
1221 * It also makes sure the delalloc code has enough
1222 * dirty data to avoid making new small extents as part
1223 * of the defrag
1224 *
1225 * It's a good idea to start RA on this range
1226 * before calling this.
1227 */
1228static int cluster_pages_for_defrag(struct inode *inode,
1229				    struct page **pages,
1230				    unsigned long start_index,
1231				    unsigned long num_pages)
1232{
1233	unsigned long file_end;
1234	u64 isize = i_size_read(inode);
1235	u64 page_start;
1236	u64 page_end;
1237	u64 page_cnt;
1238	int ret;
1239	int i;
1240	int i_done;
1241	struct btrfs_ordered_extent *ordered;
1242	struct extent_state *cached_state = NULL;
1243	struct extent_io_tree *tree;
1244	struct extent_changeset *data_reserved = NULL;
1245	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1246
1247	file_end = (isize - 1) >> PAGE_SHIFT;
1248	if (!isize || start_index > file_end)
1249		return 0;
1250
1251	page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1252
1253	ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1254			start_index << PAGE_SHIFT,
1255			page_cnt << PAGE_SHIFT);
1256	if (ret)
1257		return ret;
1258	i_done = 0;
1259	tree = &BTRFS_I(inode)->io_tree;
1260
1261	/* step one, lock all the pages */
1262	for (i = 0; i < page_cnt; i++) {
1263		struct page *page;
1264again:
1265		page = find_or_create_page(inode->i_mapping,
1266					   start_index + i, mask);
1267		if (!page)
1268			break;
1269
1270		page_start = page_offset(page);
1271		page_end = page_start + PAGE_SIZE - 1;
1272		while (1) {
1273			lock_extent_bits(tree, page_start, page_end,
1274					 &cached_state);
1275			ordered = btrfs_lookup_ordered_extent(inode,
1276							      page_start);
1277			unlock_extent_cached(tree, page_start, page_end,
1278					     &cached_state);
1279			if (!ordered)
1280				break;
1281
1282			unlock_page(page);
1283			btrfs_start_ordered_extent(inode, ordered, 1);
1284			btrfs_put_ordered_extent(ordered);
1285			lock_page(page);
1286			/*
1287			 * we unlocked the page above, so we need check if
1288			 * it was released or not.
1289			 */
1290			if (page->mapping != inode->i_mapping) {
1291				unlock_page(page);
1292				put_page(page);
1293				goto again;
1294			}
1295		}
1296
1297		if (!PageUptodate(page)) {
1298			btrfs_readpage(NULL, page);
1299			lock_page(page);
1300			if (!PageUptodate(page)) {
1301				unlock_page(page);
1302				put_page(page);
1303				ret = -EIO;
1304				break;
1305			}
1306		}
1307
1308		if (page->mapping != inode->i_mapping) {
1309			unlock_page(page);
1310			put_page(page);
1311			goto again;
1312		}
1313
1314		pages[i] = page;
1315		i_done++;
1316	}
1317	if (!i_done || ret)
1318		goto out;
1319
1320	if (!(inode->i_sb->s_flags & SB_ACTIVE))
1321		goto out;
1322
1323	/*
1324	 * so now we have a nice long stream of locked
1325	 * and up to date pages, lets wait on them
1326	 */
1327	for (i = 0; i < i_done; i++)
1328		wait_on_page_writeback(pages[i]);
1329
1330	page_start = page_offset(pages[0]);
1331	page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1332
1333	lock_extent_bits(&BTRFS_I(inode)->io_tree,
1334			 page_start, page_end - 1, &cached_state);
1335	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1336			  page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1337			  EXTENT_DEFRAG, 0, 0, &cached_state);
 
1338
1339	if (i_done != page_cnt) {
1340		spin_lock(&BTRFS_I(inode)->lock);
1341		btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1342		spin_unlock(&BTRFS_I(inode)->lock);
1343		btrfs_delalloc_release_space(inode, data_reserved,
1344				start_index << PAGE_SHIFT,
1345				(page_cnt - i_done) << PAGE_SHIFT, true);
1346	}
1347
1348
1349	set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1350			  &cached_state);
1351
1352	unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1353			     page_start, page_end - 1, &cached_state);
1354
1355	for (i = 0; i < i_done; i++) {
1356		clear_page_dirty_for_io(pages[i]);
1357		ClearPageChecked(pages[i]);
1358		set_page_extent_mapped(pages[i]);
1359		set_page_dirty(pages[i]);
1360		unlock_page(pages[i]);
1361		put_page(pages[i]);
1362	}
1363	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
 
1364	extent_changeset_free(data_reserved);
1365	return i_done;
1366out:
1367	for (i = 0; i < i_done; i++) {
1368		unlock_page(pages[i]);
1369		put_page(pages[i]);
1370	}
1371	btrfs_delalloc_release_space(inode, data_reserved,
1372			start_index << PAGE_SHIFT,
1373			page_cnt << PAGE_SHIFT, true);
1374	btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
 
1375	extent_changeset_free(data_reserved);
1376	return ret;
1377
1378}
1379
1380int btrfs_defrag_file(struct inode *inode, struct file *file,
1381		      struct btrfs_ioctl_defrag_range_args *range,
1382		      u64 newer_than, unsigned long max_to_defrag)
1383{
1384	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1385	struct btrfs_root *root = BTRFS_I(inode)->root;
1386	struct file_ra_state *ra = NULL;
1387	unsigned long last_index;
1388	u64 isize = i_size_read(inode);
1389	u64 last_len = 0;
1390	u64 skip = 0;
1391	u64 defrag_end = 0;
1392	u64 newer_off = range->start;
1393	unsigned long i;
1394	unsigned long ra_index = 0;
1395	int ret;
1396	int defrag_count = 0;
1397	int compress_type = BTRFS_COMPRESS_ZLIB;
1398	u32 extent_thresh = range->extent_thresh;
1399	unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1400	unsigned long cluster = max_cluster;
1401	u64 new_align = ~((u64)SZ_128K - 1);
1402	struct page **pages = NULL;
1403	bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1404
1405	if (isize == 0)
1406		return 0;
1407
1408	if (range->start >= isize)
1409		return -EINVAL;
1410
1411	if (do_compress) {
1412		if (range->compress_type > BTRFS_COMPRESS_TYPES)
1413			return -EINVAL;
1414		if (range->compress_type)
1415			compress_type = range->compress_type;
1416	}
1417
1418	if (extent_thresh == 0)
1419		extent_thresh = SZ_256K;
1420
1421	/*
1422	 * If we were not given a file, allocate a readahead context. As
1423	 * readahead is just an optimization, defrag will work without it so
1424	 * we don't error out.
1425	 */
1426	if (!file) {
1427		ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1428		if (ra)
1429			file_ra_state_init(ra, inode->i_mapping);
1430	} else {
1431		ra = &file->f_ra;
1432	}
1433
1434	pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1435	if (!pages) {
1436		ret = -ENOMEM;
1437		goto out_ra;
1438	}
1439
1440	/* find the last page to defrag */
1441	if (range->start + range->len > range->start) {
1442		last_index = min_t(u64, isize - 1,
1443			 range->start + range->len - 1) >> PAGE_SHIFT;
1444	} else {
1445		last_index = (isize - 1) >> PAGE_SHIFT;
1446	}
1447
1448	if (newer_than) {
1449		ret = find_new_extents(root, inode, newer_than,
1450				       &newer_off, SZ_64K);
1451		if (!ret) {
1452			range->start = newer_off;
1453			/*
1454			 * we always align our defrag to help keep
1455			 * the extents in the file evenly spaced
1456			 */
1457			i = (newer_off & new_align) >> PAGE_SHIFT;
1458		} else
1459			goto out_ra;
1460	} else {
1461		i = range->start >> PAGE_SHIFT;
1462	}
1463	if (!max_to_defrag)
1464		max_to_defrag = last_index - i + 1;
1465
1466	/*
1467	 * make writeback starts from i, so the defrag range can be
1468	 * written sequentially.
1469	 */
1470	if (i < inode->i_mapping->writeback_index)
1471		inode->i_mapping->writeback_index = i;
1472
1473	while (i <= last_index && defrag_count < max_to_defrag &&
1474	       (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1475		/*
1476		 * make sure we stop running if someone unmounts
1477		 * the FS
1478		 */
1479		if (!(inode->i_sb->s_flags & SB_ACTIVE))
1480			break;
1481
1482		if (btrfs_defrag_cancelled(fs_info)) {
1483			btrfs_debug(fs_info, "defrag_file cancelled");
1484			ret = -EAGAIN;
1485			break;
1486		}
1487
1488		if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1489					 extent_thresh, &last_len, &skip,
1490					 &defrag_end, do_compress)){
1491			unsigned long next;
1492			/*
1493			 * the should_defrag function tells us how much to skip
1494			 * bump our counter by the suggested amount
1495			 */
1496			next = DIV_ROUND_UP(skip, PAGE_SIZE);
1497			i = max(i + 1, next);
1498			continue;
1499		}
1500
1501		if (!newer_than) {
1502			cluster = (PAGE_ALIGN(defrag_end) >>
1503				   PAGE_SHIFT) - i;
1504			cluster = min(cluster, max_cluster);
1505		} else {
1506			cluster = max_cluster;
1507		}
1508
1509		if (i + cluster > ra_index) {
1510			ra_index = max(i, ra_index);
1511			if (ra)
1512				page_cache_sync_readahead(inode->i_mapping, ra,
1513						file, ra_index, cluster);
1514			ra_index += cluster;
1515		}
1516
1517		inode_lock(inode);
1518		if (IS_SWAPFILE(inode)) {
1519			ret = -ETXTBSY;
1520		} else {
1521			if (do_compress)
1522				BTRFS_I(inode)->defrag_compress = compress_type;
1523			ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1524		}
1525		if (ret < 0) {
1526			inode_unlock(inode);
1527			goto out_ra;
1528		}
1529
1530		defrag_count += ret;
1531		balance_dirty_pages_ratelimited(inode->i_mapping);
1532		inode_unlock(inode);
1533
1534		if (newer_than) {
1535			if (newer_off == (u64)-1)
1536				break;
1537
1538			if (ret > 0)
1539				i += ret;
1540
1541			newer_off = max(newer_off + 1,
1542					(u64)i << PAGE_SHIFT);
1543
1544			ret = find_new_extents(root, inode, newer_than,
1545					       &newer_off, SZ_64K);
1546			if (!ret) {
1547				range->start = newer_off;
1548				i = (newer_off & new_align) >> PAGE_SHIFT;
1549			} else {
1550				break;
1551			}
1552		} else {
1553			if (ret > 0) {
1554				i += ret;
1555				last_len += ret << PAGE_SHIFT;
1556			} else {
1557				i++;
1558				last_len = 0;
1559			}
1560		}
1561	}
1562
1563	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1564		filemap_flush(inode->i_mapping);
1565		if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1566			     &BTRFS_I(inode)->runtime_flags))
1567			filemap_flush(inode->i_mapping);
1568	}
1569
1570	if (range->compress_type == BTRFS_COMPRESS_LZO) {
1571		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1572	} else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1573		btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1574	}
1575
1576	ret = defrag_count;
1577
1578out_ra:
1579	if (do_compress) {
1580		inode_lock(inode);
1581		BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1582		inode_unlock(inode);
1583	}
1584	if (!file)
1585		kfree(ra);
1586	kfree(pages);
1587	return ret;
1588}
1589
1590static noinline int btrfs_ioctl_resize(struct file *file,
1591					void __user *arg)
1592{
1593	struct inode *inode = file_inode(file);
1594	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1595	u64 new_size;
1596	u64 old_size;
1597	u64 devid = 1;
1598	struct btrfs_root *root = BTRFS_I(inode)->root;
1599	struct btrfs_ioctl_vol_args *vol_args;
1600	struct btrfs_trans_handle *trans;
1601	struct btrfs_device *device = NULL;
1602	char *sizestr;
1603	char *retptr;
1604	char *devstr = NULL;
1605	int ret = 0;
1606	int mod = 0;
1607
1608	if (!capable(CAP_SYS_ADMIN))
1609		return -EPERM;
1610
1611	ret = mnt_want_write_file(file);
1612	if (ret)
1613		return ret;
1614
1615	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1616		mnt_drop_write_file(file);
1617		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1618	}
1619
 
1620	vol_args = memdup_user(arg, sizeof(*vol_args));
1621	if (IS_ERR(vol_args)) {
1622		ret = PTR_ERR(vol_args);
1623		goto out;
1624	}
1625
1626	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1627
1628	sizestr = vol_args->name;
1629	devstr = strchr(sizestr, ':');
1630	if (devstr) {
1631		sizestr = devstr + 1;
1632		*devstr = '\0';
1633		devstr = vol_args->name;
1634		ret = kstrtoull(devstr, 10, &devid);
1635		if (ret)
1636			goto out_free;
1637		if (!devid) {
1638			ret = -EINVAL;
1639			goto out_free;
1640		}
1641		btrfs_info(fs_info, "resizing devid %llu", devid);
1642	}
1643
1644	device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1645	if (!device) {
1646		btrfs_info(fs_info, "resizer unable to find device %llu",
1647			   devid);
1648		ret = -ENODEV;
1649		goto out_free;
1650	}
1651
1652	if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1653		btrfs_info(fs_info,
1654			   "resizer unable to apply on readonly device %llu",
1655		       devid);
1656		ret = -EPERM;
1657		goto out_free;
1658	}
1659
1660	if (!strcmp(sizestr, "max"))
1661		new_size = device->bdev->bd_inode->i_size;
1662	else {
1663		if (sizestr[0] == '-') {
1664			mod = -1;
1665			sizestr++;
1666		} else if (sizestr[0] == '+') {
1667			mod = 1;
1668			sizestr++;
1669		}
1670		new_size = memparse(sizestr, &retptr);
1671		if (*retptr != '\0' || new_size == 0) {
1672			ret = -EINVAL;
1673			goto out_free;
1674		}
1675	}
1676
1677	if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1678		ret = -EPERM;
1679		goto out_free;
1680	}
1681
1682	old_size = btrfs_device_get_total_bytes(device);
1683
1684	if (mod < 0) {
1685		if (new_size > old_size) {
1686			ret = -EINVAL;
1687			goto out_free;
1688		}
1689		new_size = old_size - new_size;
1690	} else if (mod > 0) {
1691		if (new_size > ULLONG_MAX - old_size) {
1692			ret = -ERANGE;
1693			goto out_free;
1694		}
1695		new_size = old_size + new_size;
1696	}
1697
1698	if (new_size < SZ_256M) {
1699		ret = -EINVAL;
1700		goto out_free;
1701	}
1702	if (new_size > device->bdev->bd_inode->i_size) {
1703		ret = -EFBIG;
1704		goto out_free;
1705	}
1706
1707	new_size = round_down(new_size, fs_info->sectorsize);
1708
1709	btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1710			  rcu_str_deref(device->name), new_size);
1711
1712	if (new_size > old_size) {
1713		trans = btrfs_start_transaction(root, 0);
1714		if (IS_ERR(trans)) {
1715			ret = PTR_ERR(trans);
1716			goto out_free;
1717		}
1718		ret = btrfs_grow_device(trans, device, new_size);
1719		btrfs_commit_transaction(trans);
1720	} else if (new_size < old_size) {
1721		ret = btrfs_shrink_device(device, new_size);
1722	} /* equal, nothing need to do */
1723
1724out_free:
1725	kfree(vol_args);
1726out:
 
1727	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1728	mnt_drop_write_file(file);
1729	return ret;
1730}
1731
1732static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1733				const char *name, unsigned long fd, int subvol,
1734				u64 *transid, bool readonly,
1735				struct btrfs_qgroup_inherit *inherit)
1736{
1737	int namelen;
1738	int ret = 0;
1739
1740	if (!S_ISDIR(file_inode(file)->i_mode))
1741		return -ENOTDIR;
1742
1743	ret = mnt_want_write_file(file);
1744	if (ret)
1745		goto out;
1746
1747	namelen = strlen(name);
1748	if (strchr(name, '/')) {
1749		ret = -EINVAL;
1750		goto out_drop_write;
1751	}
1752
1753	if (name[0] == '.' &&
1754	   (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1755		ret = -EEXIST;
1756		goto out_drop_write;
1757	}
1758
1759	if (subvol) {
1760		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1761				     NULL, transid, readonly, inherit);
1762	} else {
1763		struct fd src = fdget(fd);
1764		struct inode *src_inode;
1765		if (!src.file) {
1766			ret = -EINVAL;
1767			goto out_drop_write;
1768		}
1769
1770		src_inode = file_inode(src.file);
1771		if (src_inode->i_sb != file_inode(file)->i_sb) {
1772			btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1773				   "Snapshot src from another FS");
1774			ret = -EXDEV;
1775		} else if (!inode_owner_or_capable(src_inode)) {
1776			/*
1777			 * Subvolume creation is not restricted, but snapshots
1778			 * are limited to own subvolumes only
1779			 */
1780			ret = -EPERM;
1781		} else {
1782			ret = btrfs_mksubvol(&file->f_path, name, namelen,
1783					     BTRFS_I(src_inode)->root,
1784					     transid, readonly, inherit);
1785		}
1786		fdput(src);
1787	}
1788out_drop_write:
1789	mnt_drop_write_file(file);
1790out:
1791	return ret;
1792}
1793
1794static noinline int btrfs_ioctl_snap_create(struct file *file,
1795					    void __user *arg, int subvol)
1796{
1797	struct btrfs_ioctl_vol_args *vol_args;
1798	int ret;
1799
1800	if (!S_ISDIR(file_inode(file)->i_mode))
1801		return -ENOTDIR;
1802
1803	vol_args = memdup_user(arg, sizeof(*vol_args));
1804	if (IS_ERR(vol_args))
1805		return PTR_ERR(vol_args);
1806	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1807
1808	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1809					      vol_args->fd, subvol,
1810					      NULL, false, NULL);
1811
1812	kfree(vol_args);
1813	return ret;
1814}
1815
1816static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1817					       void __user *arg, int subvol)
1818{
1819	struct btrfs_ioctl_vol_args_v2 *vol_args;
1820	int ret;
1821	u64 transid = 0;
1822	u64 *ptr = NULL;
1823	bool readonly = false;
1824	struct btrfs_qgroup_inherit *inherit = NULL;
1825
1826	if (!S_ISDIR(file_inode(file)->i_mode))
1827		return -ENOTDIR;
1828
1829	vol_args = memdup_user(arg, sizeof(*vol_args));
1830	if (IS_ERR(vol_args))
1831		return PTR_ERR(vol_args);
1832	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1833
1834	if (vol_args->flags &
1835	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1836	      BTRFS_SUBVOL_QGROUP_INHERIT)) {
1837		ret = -EOPNOTSUPP;
1838		goto free_args;
1839	}
1840
1841	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1842		struct inode *inode = file_inode(file);
1843		struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1844
1845		btrfs_warn(fs_info,
1846"SNAP_CREATE_V2 ioctl with CREATE_ASYNC is deprecated and will be removed in kernel 5.7");
1847
1848		ptr = &transid;
1849	}
1850	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1851		readonly = true;
1852	if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1853		if (vol_args->size > PAGE_SIZE) {
1854			ret = -EINVAL;
1855			goto free_args;
1856		}
1857		inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1858		if (IS_ERR(inherit)) {
1859			ret = PTR_ERR(inherit);
1860			goto free_args;
1861		}
1862	}
1863
1864	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1865					      vol_args->fd, subvol, ptr,
1866					      readonly, inherit);
1867	if (ret)
1868		goto free_inherit;
1869
1870	if (ptr && copy_to_user(arg +
1871				offsetof(struct btrfs_ioctl_vol_args_v2,
1872					transid),
1873				ptr, sizeof(*ptr)))
1874		ret = -EFAULT;
1875
1876free_inherit:
1877	kfree(inherit);
1878free_args:
1879	kfree(vol_args);
1880	return ret;
1881}
1882
1883static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1884						void __user *arg)
1885{
1886	struct inode *inode = file_inode(file);
1887	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1888	struct btrfs_root *root = BTRFS_I(inode)->root;
1889	int ret = 0;
1890	u64 flags = 0;
1891
1892	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1893		return -EINVAL;
1894
1895	down_read(&fs_info->subvol_sem);
1896	if (btrfs_root_readonly(root))
1897		flags |= BTRFS_SUBVOL_RDONLY;
1898	up_read(&fs_info->subvol_sem);
1899
1900	if (copy_to_user(arg, &flags, sizeof(flags)))
1901		ret = -EFAULT;
1902
1903	return ret;
1904}
1905
1906static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1907					      void __user *arg)
1908{
1909	struct inode *inode = file_inode(file);
1910	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1911	struct btrfs_root *root = BTRFS_I(inode)->root;
1912	struct btrfs_trans_handle *trans;
1913	u64 root_flags;
1914	u64 flags;
1915	int ret = 0;
1916
1917	if (!inode_owner_or_capable(inode))
1918		return -EPERM;
1919
1920	ret = mnt_want_write_file(file);
1921	if (ret)
1922		goto out;
1923
1924	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1925		ret = -EINVAL;
1926		goto out_drop_write;
1927	}
1928
1929	if (copy_from_user(&flags, arg, sizeof(flags))) {
1930		ret = -EFAULT;
1931		goto out_drop_write;
1932	}
1933
1934	if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1935		ret = -EINVAL;
1936		goto out_drop_write;
1937	}
1938
1939	if (flags & ~BTRFS_SUBVOL_RDONLY) {
1940		ret = -EOPNOTSUPP;
1941		goto out_drop_write;
1942	}
1943
1944	down_write(&fs_info->subvol_sem);
1945
1946	/* nothing to do */
1947	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1948		goto out_drop_sem;
1949
1950	root_flags = btrfs_root_flags(&root->root_item);
1951	if (flags & BTRFS_SUBVOL_RDONLY) {
1952		btrfs_set_root_flags(&root->root_item,
1953				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1954	} else {
1955		/*
1956		 * Block RO -> RW transition if this subvolume is involved in
1957		 * send
1958		 */
1959		spin_lock(&root->root_item_lock);
1960		if (root->send_in_progress == 0) {
1961			btrfs_set_root_flags(&root->root_item,
1962				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1963			spin_unlock(&root->root_item_lock);
1964		} else {
1965			spin_unlock(&root->root_item_lock);
1966			btrfs_warn(fs_info,
1967				   "Attempt to set subvolume %llu read-write during send",
1968				   root->root_key.objectid);
1969			ret = -EPERM;
1970			goto out_drop_sem;
1971		}
1972	}
1973
1974	trans = btrfs_start_transaction(root, 1);
1975	if (IS_ERR(trans)) {
1976		ret = PTR_ERR(trans);
1977		goto out_reset;
1978	}
1979
1980	ret = btrfs_update_root(trans, fs_info->tree_root,
1981				&root->root_key, &root->root_item);
1982	if (ret < 0) {
1983		btrfs_end_transaction(trans);
1984		goto out_reset;
1985	}
1986
1987	ret = btrfs_commit_transaction(trans);
1988
1989out_reset:
1990	if (ret)
1991		btrfs_set_root_flags(&root->root_item, root_flags);
1992out_drop_sem:
1993	up_write(&fs_info->subvol_sem);
1994out_drop_write:
1995	mnt_drop_write_file(file);
1996out:
1997	return ret;
1998}
1999
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2000static noinline int key_in_sk(struct btrfs_key *key,
2001			      struct btrfs_ioctl_search_key *sk)
2002{
2003	struct btrfs_key test;
2004	int ret;
2005
2006	test.objectid = sk->min_objectid;
2007	test.type = sk->min_type;
2008	test.offset = sk->min_offset;
2009
2010	ret = btrfs_comp_cpu_keys(key, &test);
2011	if (ret < 0)
2012		return 0;
2013
2014	test.objectid = sk->max_objectid;
2015	test.type = sk->max_type;
2016	test.offset = sk->max_offset;
2017
2018	ret = btrfs_comp_cpu_keys(key, &test);
2019	if (ret > 0)
2020		return 0;
2021	return 1;
2022}
2023
2024static noinline int copy_to_sk(struct btrfs_path *path,
2025			       struct btrfs_key *key,
2026			       struct btrfs_ioctl_search_key *sk,
2027			       size_t *buf_size,
2028			       char __user *ubuf,
2029			       unsigned long *sk_offset,
2030			       int *num_found)
2031{
2032	u64 found_transid;
2033	struct extent_buffer *leaf;
2034	struct btrfs_ioctl_search_header sh;
2035	struct btrfs_key test;
2036	unsigned long item_off;
2037	unsigned long item_len;
2038	int nritems;
2039	int i;
2040	int slot;
2041	int ret = 0;
2042
2043	leaf = path->nodes[0];
2044	slot = path->slots[0];
2045	nritems = btrfs_header_nritems(leaf);
2046
2047	if (btrfs_header_generation(leaf) > sk->max_transid) {
2048		i = nritems;
2049		goto advance_key;
2050	}
2051	found_transid = btrfs_header_generation(leaf);
2052
2053	for (i = slot; i < nritems; i++) {
2054		item_off = btrfs_item_ptr_offset(leaf, i);
2055		item_len = btrfs_item_size_nr(leaf, i);
2056
2057		btrfs_item_key_to_cpu(leaf, key, i);
2058		if (!key_in_sk(key, sk))
2059			continue;
2060
2061		if (sizeof(sh) + item_len > *buf_size) {
2062			if (*num_found) {
2063				ret = 1;
2064				goto out;
2065			}
2066
2067			/*
2068			 * return one empty item back for v1, which does not
2069			 * handle -EOVERFLOW
2070			 */
2071
2072			*buf_size = sizeof(sh) + item_len;
2073			item_len = 0;
2074			ret = -EOVERFLOW;
2075		}
2076
2077		if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2078			ret = 1;
2079			goto out;
2080		}
2081
2082		sh.objectid = key->objectid;
2083		sh.offset = key->offset;
2084		sh.type = key->type;
2085		sh.len = item_len;
2086		sh.transid = found_transid;
2087
2088		/* copy search result header */
2089		if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2090			ret = -EFAULT;
2091			goto out;
2092		}
2093
2094		*sk_offset += sizeof(sh);
2095
2096		if (item_len) {
2097			char __user *up = ubuf + *sk_offset;
2098			/* copy the item */
2099			if (read_extent_buffer_to_user(leaf, up,
2100						       item_off, item_len)) {
2101				ret = -EFAULT;
2102				goto out;
2103			}
2104
2105			*sk_offset += item_len;
2106		}
2107		(*num_found)++;
2108
2109		if (ret) /* -EOVERFLOW from above */
2110			goto out;
2111
2112		if (*num_found >= sk->nr_items) {
2113			ret = 1;
2114			goto out;
2115		}
2116	}
2117advance_key:
2118	ret = 0;
2119	test.objectid = sk->max_objectid;
2120	test.type = sk->max_type;
2121	test.offset = sk->max_offset;
2122	if (btrfs_comp_cpu_keys(key, &test) >= 0)
2123		ret = 1;
2124	else if (key->offset < (u64)-1)
2125		key->offset++;
2126	else if (key->type < (u8)-1) {
2127		key->offset = 0;
2128		key->type++;
2129	} else if (key->objectid < (u64)-1) {
2130		key->offset = 0;
2131		key->type = 0;
2132		key->objectid++;
2133	} else
2134		ret = 1;
2135out:
2136	/*
2137	 *  0: all items from this leaf copied, continue with next
2138	 *  1: * more items can be copied, but unused buffer is too small
2139	 *     * all items were found
2140	 *     Either way, it will stops the loop which iterates to the next
2141	 *     leaf
2142	 *  -EOVERFLOW: item was to large for buffer
2143	 *  -EFAULT: could not copy extent buffer back to userspace
2144	 */
2145	return ret;
2146}
2147
2148static noinline int search_ioctl(struct inode *inode,
2149				 struct btrfs_ioctl_search_key *sk,
2150				 size_t *buf_size,
2151				 char __user *ubuf)
2152{
2153	struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2154	struct btrfs_root *root;
2155	struct btrfs_key key;
2156	struct btrfs_path *path;
2157	int ret;
2158	int num_found = 0;
2159	unsigned long sk_offset = 0;
2160
2161	if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2162		*buf_size = sizeof(struct btrfs_ioctl_search_header);
2163		return -EOVERFLOW;
2164	}
2165
2166	path = btrfs_alloc_path();
2167	if (!path)
2168		return -ENOMEM;
2169
2170	if (sk->tree_id == 0) {
2171		/* search the root of the inode that was passed */
2172		root = BTRFS_I(inode)->root;
2173	} else {
2174		key.objectid = sk->tree_id;
2175		key.type = BTRFS_ROOT_ITEM_KEY;
2176		key.offset = (u64)-1;
2177		root = btrfs_read_fs_root_no_name(info, &key);
2178		if (IS_ERR(root)) {
2179			btrfs_free_path(path);
2180			return PTR_ERR(root);
2181		}
2182	}
2183
2184	key.objectid = sk->min_objectid;
2185	key.type = sk->min_type;
2186	key.offset = sk->min_offset;
2187
2188	while (1) {
2189		ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2190		if (ret != 0) {
2191			if (ret > 0)
2192				ret = 0;
2193			goto err;
2194		}
2195		ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2196				 &sk_offset, &num_found);
2197		btrfs_release_path(path);
2198		if (ret)
2199			break;
2200
2201	}
2202	if (ret > 0)
2203		ret = 0;
2204err:
2205	sk->nr_items = num_found;
2206	btrfs_free_path(path);
2207	return ret;
2208}
2209
2210static noinline int btrfs_ioctl_tree_search(struct file *file,
2211					   void __user *argp)
2212{
2213	struct btrfs_ioctl_search_args __user *uargs;
2214	struct btrfs_ioctl_search_key sk;
2215	struct inode *inode;
2216	int ret;
2217	size_t buf_size;
2218
2219	if (!capable(CAP_SYS_ADMIN))
2220		return -EPERM;
2221
2222	uargs = (struct btrfs_ioctl_search_args __user *)argp;
2223
2224	if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2225		return -EFAULT;
2226
2227	buf_size = sizeof(uargs->buf);
2228
2229	inode = file_inode(file);
2230	ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2231
2232	/*
2233	 * In the origin implementation an overflow is handled by returning a
2234	 * search header with a len of zero, so reset ret.
2235	 */
2236	if (ret == -EOVERFLOW)
2237		ret = 0;
2238
2239	if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2240		ret = -EFAULT;
2241	return ret;
2242}
2243
2244static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2245					       void __user *argp)
2246{
2247	struct btrfs_ioctl_search_args_v2 __user *uarg;
2248	struct btrfs_ioctl_search_args_v2 args;
2249	struct inode *inode;
2250	int ret;
2251	size_t buf_size;
2252	const size_t buf_limit = SZ_16M;
2253
2254	if (!capable(CAP_SYS_ADMIN))
2255		return -EPERM;
2256
2257	/* copy search header and buffer size */
2258	uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2259	if (copy_from_user(&args, uarg, sizeof(args)))
2260		return -EFAULT;
2261
2262	buf_size = args.buf_size;
2263
2264	/* limit result size to 16MB */
2265	if (buf_size > buf_limit)
2266		buf_size = buf_limit;
2267
2268	inode = file_inode(file);
2269	ret = search_ioctl(inode, &args.key, &buf_size,
2270			   (char __user *)(&uarg->buf[0]));
2271	if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2272		ret = -EFAULT;
2273	else if (ret == -EOVERFLOW &&
2274		copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2275		ret = -EFAULT;
2276
2277	return ret;
2278}
2279
2280/*
2281 * Search INODE_REFs to identify path name of 'dirid' directory
2282 * in a 'tree_id' tree. and sets path name to 'name'.
2283 */
2284static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2285				u64 tree_id, u64 dirid, char *name)
2286{
2287	struct btrfs_root *root;
2288	struct btrfs_key key;
2289	char *ptr;
2290	int ret = -1;
2291	int slot;
2292	int len;
2293	int total_len = 0;
2294	struct btrfs_inode_ref *iref;
2295	struct extent_buffer *l;
2296	struct btrfs_path *path;
2297
2298	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2299		name[0]='\0';
2300		return 0;
2301	}
2302
2303	path = btrfs_alloc_path();
2304	if (!path)
2305		return -ENOMEM;
2306
2307	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2308
2309	key.objectid = tree_id;
2310	key.type = BTRFS_ROOT_ITEM_KEY;
2311	key.offset = (u64)-1;
2312	root = btrfs_read_fs_root_no_name(info, &key);
2313	if (IS_ERR(root)) {
2314		ret = PTR_ERR(root);
 
2315		goto out;
2316	}
2317
2318	key.objectid = dirid;
2319	key.type = BTRFS_INODE_REF_KEY;
2320	key.offset = (u64)-1;
2321
2322	while (1) {
2323		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2324		if (ret < 0)
2325			goto out;
2326		else if (ret > 0) {
2327			ret = btrfs_previous_item(root, path, dirid,
2328						  BTRFS_INODE_REF_KEY);
2329			if (ret < 0)
2330				goto out;
2331			else if (ret > 0) {
2332				ret = -ENOENT;
2333				goto out;
2334			}
2335		}
2336
2337		l = path->nodes[0];
2338		slot = path->slots[0];
2339		btrfs_item_key_to_cpu(l, &key, slot);
2340
2341		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2342		len = btrfs_inode_ref_name_len(l, iref);
2343		ptr -= len + 1;
2344		total_len += len + 1;
2345		if (ptr < name) {
2346			ret = -ENAMETOOLONG;
2347			goto out;
2348		}
2349
2350		*(ptr + len) = '/';
2351		read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2352
2353		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2354			break;
2355
2356		btrfs_release_path(path);
2357		key.objectid = key.offset;
2358		key.offset = (u64)-1;
2359		dirid = key.objectid;
2360	}
2361	memmove(name, ptr, total_len);
2362	name[total_len] = '\0';
2363	ret = 0;
2364out:
2365	btrfs_free_path(path);
2366	return ret;
2367}
2368
2369static int btrfs_search_path_in_tree_user(struct inode *inode,
2370				struct btrfs_ioctl_ino_lookup_user_args *args)
2371{
2372	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2373	struct super_block *sb = inode->i_sb;
2374	struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2375	u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2376	u64 dirid = args->dirid;
2377	unsigned long item_off;
2378	unsigned long item_len;
2379	struct btrfs_inode_ref *iref;
2380	struct btrfs_root_ref *rref;
2381	struct btrfs_root *root;
2382	struct btrfs_path *path;
2383	struct btrfs_key key, key2;
2384	struct extent_buffer *leaf;
2385	struct inode *temp_inode;
2386	char *ptr;
2387	int slot;
2388	int len;
2389	int total_len = 0;
2390	int ret;
2391
2392	path = btrfs_alloc_path();
2393	if (!path)
2394		return -ENOMEM;
2395
2396	/*
2397	 * If the bottom subvolume does not exist directly under upper_limit,
2398	 * construct the path in from the bottom up.
2399	 */
2400	if (dirid != upper_limit.objectid) {
2401		ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2402
2403		key.objectid = treeid;
2404		key.type = BTRFS_ROOT_ITEM_KEY;
2405		key.offset = (u64)-1;
2406		root = btrfs_read_fs_root_no_name(fs_info, &key);
2407		if (IS_ERR(root)) {
2408			ret = PTR_ERR(root);
2409			goto out;
2410		}
2411
2412		key.objectid = dirid;
2413		key.type = BTRFS_INODE_REF_KEY;
2414		key.offset = (u64)-1;
2415		while (1) {
2416			ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2417			if (ret < 0) {
2418				goto out;
2419			} else if (ret > 0) {
2420				ret = btrfs_previous_item(root, path, dirid,
2421							  BTRFS_INODE_REF_KEY);
2422				if (ret < 0) {
2423					goto out;
2424				} else if (ret > 0) {
2425					ret = -ENOENT;
2426					goto out;
2427				}
2428			}
2429
2430			leaf = path->nodes[0];
2431			slot = path->slots[0];
2432			btrfs_item_key_to_cpu(leaf, &key, slot);
2433
2434			iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2435			len = btrfs_inode_ref_name_len(leaf, iref);
2436			ptr -= len + 1;
2437			total_len += len + 1;
2438			if (ptr < args->path) {
2439				ret = -ENAMETOOLONG;
2440				goto out;
2441			}
2442
2443			*(ptr + len) = '/';
2444			read_extent_buffer(leaf, ptr,
2445					(unsigned long)(iref + 1), len);
2446
2447			/* Check the read+exec permission of this directory */
2448			ret = btrfs_previous_item(root, path, dirid,
2449						  BTRFS_INODE_ITEM_KEY);
2450			if (ret < 0) {
2451				goto out;
2452			} else if (ret > 0) {
2453				ret = -ENOENT;
2454				goto out;
2455			}
2456
2457			leaf = path->nodes[0];
2458			slot = path->slots[0];
2459			btrfs_item_key_to_cpu(leaf, &key2, slot);
2460			if (key2.objectid != dirid) {
2461				ret = -ENOENT;
2462				goto out;
2463			}
2464
2465			temp_inode = btrfs_iget(sb, &key2, root, NULL);
2466			if (IS_ERR(temp_inode)) {
2467				ret = PTR_ERR(temp_inode);
2468				goto out;
2469			}
2470			ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2471			iput(temp_inode);
2472			if (ret) {
2473				ret = -EACCES;
2474				goto out;
2475			}
2476
2477			if (key.offset == upper_limit.objectid)
2478				break;
2479			if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2480				ret = -EACCES;
2481				goto out;
2482			}
2483
2484			btrfs_release_path(path);
2485			key.objectid = key.offset;
2486			key.offset = (u64)-1;
2487			dirid = key.objectid;
2488		}
2489
2490		memmove(args->path, ptr, total_len);
2491		args->path[total_len] = '\0';
2492		btrfs_release_path(path);
2493	}
2494
2495	/* Get the bottom subvolume's name from ROOT_REF */
2496	root = fs_info->tree_root;
2497	key.objectid = treeid;
2498	key.type = BTRFS_ROOT_REF_KEY;
2499	key.offset = args->treeid;
2500	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2501	if (ret < 0) {
2502		goto out;
2503	} else if (ret > 0) {
2504		ret = -ENOENT;
2505		goto out;
2506	}
2507
2508	leaf = path->nodes[0];
2509	slot = path->slots[0];
2510	btrfs_item_key_to_cpu(leaf, &key, slot);
2511
2512	item_off = btrfs_item_ptr_offset(leaf, slot);
2513	item_len = btrfs_item_size_nr(leaf, slot);
2514	/* Check if dirid in ROOT_REF corresponds to passed dirid */
2515	rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2516	if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2517		ret = -EINVAL;
2518		goto out;
2519	}
2520
2521	/* Copy subvolume's name */
2522	item_off += sizeof(struct btrfs_root_ref);
2523	item_len -= sizeof(struct btrfs_root_ref);
2524	read_extent_buffer(leaf, args->name, item_off, item_len);
2525	args->name[item_len] = 0;
2526
2527out:
2528	btrfs_free_path(path);
2529	return ret;
2530}
2531
2532static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2533					   void __user *argp)
2534{
2535	struct btrfs_ioctl_ino_lookup_args *args;
2536	struct inode *inode;
2537	int ret = 0;
2538
2539	args = memdup_user(argp, sizeof(*args));
2540	if (IS_ERR(args))
2541		return PTR_ERR(args);
2542
2543	inode = file_inode(file);
2544
2545	/*
2546	 * Unprivileged query to obtain the containing subvolume root id. The
2547	 * path is reset so it's consistent with btrfs_search_path_in_tree.
2548	 */
2549	if (args->treeid == 0)
2550		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2551
2552	if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2553		args->name[0] = 0;
2554		goto out;
2555	}
2556
2557	if (!capable(CAP_SYS_ADMIN)) {
2558		ret = -EPERM;
2559		goto out;
2560	}
2561
2562	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2563					args->treeid, args->objectid,
2564					args->name);
2565
2566out:
2567	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2568		ret = -EFAULT;
2569
2570	kfree(args);
2571	return ret;
2572}
2573
2574/*
2575 * Version of ino_lookup ioctl (unprivileged)
2576 *
2577 * The main differences from ino_lookup ioctl are:
2578 *
2579 *   1. Read + Exec permission will be checked using inode_permission() during
2580 *      path construction. -EACCES will be returned in case of failure.
2581 *   2. Path construction will be stopped at the inode number which corresponds
2582 *      to the fd with which this ioctl is called. If constructed path does not
2583 *      exist under fd's inode, -EACCES will be returned.
2584 *   3. The name of bottom subvolume is also searched and filled.
2585 */
2586static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2587{
2588	struct btrfs_ioctl_ino_lookup_user_args *args;
2589	struct inode *inode;
2590	int ret;
2591
2592	args = memdup_user(argp, sizeof(*args));
2593	if (IS_ERR(args))
2594		return PTR_ERR(args);
2595
2596	inode = file_inode(file);
2597
2598	if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2599	    BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2600		/*
2601		 * The subvolume does not exist under fd with which this is
2602		 * called
2603		 */
2604		kfree(args);
2605		return -EACCES;
2606	}
2607
2608	ret = btrfs_search_path_in_tree_user(inode, args);
2609
2610	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2611		ret = -EFAULT;
2612
2613	kfree(args);
2614	return ret;
2615}
2616
2617/* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2618static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2619{
2620	struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2621	struct btrfs_fs_info *fs_info;
2622	struct btrfs_root *root;
2623	struct btrfs_path *path;
2624	struct btrfs_key key;
2625	struct btrfs_root_item *root_item;
2626	struct btrfs_root_ref *rref;
2627	struct extent_buffer *leaf;
2628	unsigned long item_off;
2629	unsigned long item_len;
2630	struct inode *inode;
2631	int slot;
2632	int ret = 0;
2633
2634	path = btrfs_alloc_path();
2635	if (!path)
2636		return -ENOMEM;
2637
2638	subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2639	if (!subvol_info) {
2640		btrfs_free_path(path);
2641		return -ENOMEM;
2642	}
2643
2644	inode = file_inode(file);
2645	fs_info = BTRFS_I(inode)->root->fs_info;
2646
2647	/* Get root_item of inode's subvolume */
2648	key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2649	key.type = BTRFS_ROOT_ITEM_KEY;
2650	key.offset = (u64)-1;
2651	root = btrfs_read_fs_root_no_name(fs_info, &key);
2652	if (IS_ERR(root)) {
2653		ret = PTR_ERR(root);
2654		goto out;
2655	}
2656	root_item = &root->root_item;
2657
2658	subvol_info->treeid = key.objectid;
2659
2660	subvol_info->generation = btrfs_root_generation(root_item);
2661	subvol_info->flags = btrfs_root_flags(root_item);
2662
2663	memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2664	memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2665						    BTRFS_UUID_SIZE);
2666	memcpy(subvol_info->received_uuid, root_item->received_uuid,
2667						    BTRFS_UUID_SIZE);
2668
2669	subvol_info->ctransid = btrfs_root_ctransid(root_item);
2670	subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2671	subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2672
2673	subvol_info->otransid = btrfs_root_otransid(root_item);
2674	subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2675	subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2676
2677	subvol_info->stransid = btrfs_root_stransid(root_item);
2678	subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2679	subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2680
2681	subvol_info->rtransid = btrfs_root_rtransid(root_item);
2682	subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2683	subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2684
2685	if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2686		/* Search root tree for ROOT_BACKREF of this subvolume */
2687		root = fs_info->tree_root;
2688
2689		key.type = BTRFS_ROOT_BACKREF_KEY;
2690		key.offset = 0;
2691		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2692		if (ret < 0) {
2693			goto out;
2694		} else if (path->slots[0] >=
2695			   btrfs_header_nritems(path->nodes[0])) {
2696			ret = btrfs_next_leaf(root, path);
2697			if (ret < 0) {
2698				goto out;
2699			} else if (ret > 0) {
2700				ret = -EUCLEAN;
2701				goto out;
2702			}
2703		}
2704
2705		leaf = path->nodes[0];
2706		slot = path->slots[0];
2707		btrfs_item_key_to_cpu(leaf, &key, slot);
2708		if (key.objectid == subvol_info->treeid &&
2709		    key.type == BTRFS_ROOT_BACKREF_KEY) {
2710			subvol_info->parent_id = key.offset;
2711
2712			rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2713			subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2714
2715			item_off = btrfs_item_ptr_offset(leaf, slot)
2716					+ sizeof(struct btrfs_root_ref);
2717			item_len = btrfs_item_size_nr(leaf, slot)
2718					- sizeof(struct btrfs_root_ref);
2719			read_extent_buffer(leaf, subvol_info->name,
2720					   item_off, item_len);
2721		} else {
2722			ret = -ENOENT;
2723			goto out;
2724		}
2725	}
2726
2727	if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2728		ret = -EFAULT;
2729
2730out:
2731	btrfs_free_path(path);
2732	kzfree(subvol_info);
2733	return ret;
2734}
2735
2736/*
2737 * Return ROOT_REF information of the subvolume containing this inode
2738 * except the subvolume name.
2739 */
2740static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2741{
2742	struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2743	struct btrfs_root_ref *rref;
2744	struct btrfs_root *root;
2745	struct btrfs_path *path;
2746	struct btrfs_key key;
2747	struct extent_buffer *leaf;
2748	struct inode *inode;
2749	u64 objectid;
2750	int slot;
2751	int ret;
2752	u8 found;
2753
2754	path = btrfs_alloc_path();
2755	if (!path)
2756		return -ENOMEM;
2757
2758	rootrefs = memdup_user(argp, sizeof(*rootrefs));
2759	if (IS_ERR(rootrefs)) {
2760		btrfs_free_path(path);
2761		return PTR_ERR(rootrefs);
2762	}
2763
2764	inode = file_inode(file);
2765	root = BTRFS_I(inode)->root->fs_info->tree_root;
2766	objectid = BTRFS_I(inode)->root->root_key.objectid;
2767
2768	key.objectid = objectid;
2769	key.type = BTRFS_ROOT_REF_KEY;
2770	key.offset = rootrefs->min_treeid;
2771	found = 0;
2772
2773	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2774	if (ret < 0) {
2775		goto out;
2776	} else if (path->slots[0] >=
2777		   btrfs_header_nritems(path->nodes[0])) {
2778		ret = btrfs_next_leaf(root, path);
2779		if (ret < 0) {
2780			goto out;
2781		} else if (ret > 0) {
2782			ret = -EUCLEAN;
2783			goto out;
2784		}
2785	}
2786	while (1) {
2787		leaf = path->nodes[0];
2788		slot = path->slots[0];
2789
2790		btrfs_item_key_to_cpu(leaf, &key, slot);
2791		if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2792			ret = 0;
2793			goto out;
2794		}
2795
2796		if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2797			ret = -EOVERFLOW;
2798			goto out;
2799		}
2800
2801		rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2802		rootrefs->rootref[found].treeid = key.offset;
2803		rootrefs->rootref[found].dirid =
2804				  btrfs_root_ref_dirid(leaf, rref);
2805		found++;
2806
2807		ret = btrfs_next_item(root, path);
2808		if (ret < 0) {
2809			goto out;
2810		} else if (ret > 0) {
2811			ret = -EUCLEAN;
2812			goto out;
2813		}
2814	}
2815
2816out:
2817	if (!ret || ret == -EOVERFLOW) {
2818		rootrefs->num_items = found;
2819		/* update min_treeid for next search */
2820		if (found)
2821			rootrefs->min_treeid =
2822				rootrefs->rootref[found - 1].treeid + 1;
2823		if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2824			ret = -EFAULT;
2825	}
2826
2827	kfree(rootrefs);
2828	btrfs_free_path(path);
2829
2830	return ret;
2831}
2832
2833static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2834					     void __user *arg)
2835{
2836	struct dentry *parent = file->f_path.dentry;
2837	struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2838	struct dentry *dentry;
2839	struct inode *dir = d_inode(parent);
2840	struct inode *inode;
2841	struct btrfs_root *root = BTRFS_I(dir)->root;
2842	struct btrfs_root *dest = NULL;
2843	struct btrfs_ioctl_vol_args *vol_args;
 
 
 
 
2844	int namelen;
 
2845	int err = 0;
2846
2847	if (!S_ISDIR(dir->i_mode))
2848		return -ENOTDIR;
2849
2850	vol_args = memdup_user(arg, sizeof(*vol_args));
2851	if (IS_ERR(vol_args))
2852		return PTR_ERR(vol_args);
2853
2854	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2855	namelen = strlen(vol_args->name);
2856	if (strchr(vol_args->name, '/') ||
2857	    strncmp(vol_args->name, "..", namelen) == 0) {
2858		err = -EINVAL;
2859		goto out;
2860	}
2861
2862	err = mnt_want_write_file(file);
2863	if (err)
2864		goto out;
2865
2866
2867	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2868	if (err == -EINTR)
2869		goto out_drop_write;
2870	dentry = lookup_one_len(vol_args->name, parent, namelen);
2871	if (IS_ERR(dentry)) {
2872		err = PTR_ERR(dentry);
2873		goto out_unlock_dir;
2874	}
2875
2876	if (d_really_is_negative(dentry)) {
2877		err = -ENOENT;
2878		goto out_dput;
2879	}
2880
2881	inode = d_inode(dentry);
2882	dest = BTRFS_I(inode)->root;
2883	if (!capable(CAP_SYS_ADMIN)) {
2884		/*
2885		 * Regular user.  Only allow this with a special mount
2886		 * option, when the user has write+exec access to the
2887		 * subvol root, and when rmdir(2) would have been
2888		 * allowed.
2889		 *
2890		 * Note that this is _not_ check that the subvol is
2891		 * empty or doesn't contain data that we wouldn't
2892		 * otherwise be able to delete.
2893		 *
2894		 * Users who want to delete empty subvols should try
2895		 * rmdir(2).
2896		 */
2897		err = -EPERM;
2898		if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2899			goto out_dput;
2900
2901		/*
2902		 * Do not allow deletion if the parent dir is the same
2903		 * as the dir to be deleted.  That means the ioctl
2904		 * must be called on the dentry referencing the root
2905		 * of the subvol, not a random directory contained
2906		 * within it.
2907		 */
2908		err = -EINVAL;
2909		if (root == dest)
2910			goto out_dput;
2911
2912		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2913		if (err)
2914			goto out_dput;
2915	}
2916
2917	/* check if subvolume may be deleted by a user */
2918	err = btrfs_may_delete(dir, dentry, 1);
2919	if (err)
2920		goto out_dput;
2921
2922	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2923		err = -EINVAL;
2924		goto out_dput;
2925	}
2926
2927	inode_lock(inode);
2928	err = btrfs_delete_subvolume(dir, dentry);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2929	inode_unlock(inode);
2930	if (!err) {
2931		fsnotify_rmdir(dir, dentry);
 
2932		d_delete(dentry);
2933	}
2934
 
 
 
 
 
 
2935out_dput:
2936	dput(dentry);
2937out_unlock_dir:
2938	inode_unlock(dir);
2939out_drop_write:
2940	mnt_drop_write_file(file);
2941out:
2942	kfree(vol_args);
2943	return err;
2944}
2945
2946static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2947{
2948	struct inode *inode = file_inode(file);
2949	struct btrfs_root *root = BTRFS_I(inode)->root;
2950	struct btrfs_ioctl_defrag_range_args *range;
2951	int ret;
2952
2953	ret = mnt_want_write_file(file);
2954	if (ret)
2955		return ret;
2956
2957	if (btrfs_root_readonly(root)) {
2958		ret = -EROFS;
2959		goto out;
2960	}
2961
2962	switch (inode->i_mode & S_IFMT) {
2963	case S_IFDIR:
2964		if (!capable(CAP_SYS_ADMIN)) {
2965			ret = -EPERM;
2966			goto out;
2967		}
2968		ret = btrfs_defrag_root(root);
2969		break;
2970	case S_IFREG:
2971		/*
2972		 * Note that this does not check the file descriptor for write
2973		 * access. This prevents defragmenting executables that are
2974		 * running and allows defrag on files open in read-only mode.
2975		 */
2976		if (!capable(CAP_SYS_ADMIN) &&
2977		    inode_permission(inode, MAY_WRITE)) {
2978			ret = -EPERM;
2979			goto out;
2980		}
2981
2982		range = kzalloc(sizeof(*range), GFP_KERNEL);
2983		if (!range) {
2984			ret = -ENOMEM;
2985			goto out;
2986		}
2987
2988		if (argp) {
2989			if (copy_from_user(range, argp,
2990					   sizeof(*range))) {
2991				ret = -EFAULT;
2992				kfree(range);
2993				goto out;
2994			}
2995			/* compression requires us to start the IO */
2996			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2997				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2998				range->extent_thresh = (u32)-1;
2999			}
3000		} else {
3001			/* the rest are all set to zero by kzalloc */
3002			range->len = (u64)-1;
3003		}
3004		ret = btrfs_defrag_file(file_inode(file), file,
3005					range, BTRFS_OLDEST_GENERATION, 0);
3006		if (ret > 0)
3007			ret = 0;
3008		kfree(range);
3009		break;
3010	default:
3011		ret = -EINVAL;
3012	}
3013out:
3014	mnt_drop_write_file(file);
3015	return ret;
3016}
3017
3018static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3019{
3020	struct btrfs_ioctl_vol_args *vol_args;
3021	int ret;
3022
3023	if (!capable(CAP_SYS_ADMIN))
3024		return -EPERM;
3025
3026	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3027		return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3028
 
3029	vol_args = memdup_user(arg, sizeof(*vol_args));
3030	if (IS_ERR(vol_args)) {
3031		ret = PTR_ERR(vol_args);
3032		goto out;
3033	}
3034
3035	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3036	ret = btrfs_init_new_device(fs_info, vol_args->name);
3037
3038	if (!ret)
3039		btrfs_info(fs_info, "disk added %s", vol_args->name);
3040
3041	kfree(vol_args);
3042out:
 
3043	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3044	return ret;
3045}
3046
3047static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3048{
3049	struct inode *inode = file_inode(file);
3050	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3051	struct btrfs_ioctl_vol_args_v2 *vol_args;
3052	int ret;
3053
3054	if (!capable(CAP_SYS_ADMIN))
3055		return -EPERM;
3056
3057	ret = mnt_want_write_file(file);
3058	if (ret)
3059		return ret;
3060
3061	vol_args = memdup_user(arg, sizeof(*vol_args));
3062	if (IS_ERR(vol_args)) {
3063		ret = PTR_ERR(vol_args);
3064		goto err_drop;
3065	}
3066
3067	/* Check for compatibility reject unknown flags */
3068	if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3069		ret = -EOPNOTSUPP;
3070		goto out;
3071	}
3072
3073	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3074		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3075		goto out;
3076	}
3077
3078	if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3079		ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3080	} else {
3081		vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3082		ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3083	}
3084	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3085
3086	if (!ret) {
3087		if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3088			btrfs_info(fs_info, "device deleted: id %llu",
3089					vol_args->devid);
3090		else
3091			btrfs_info(fs_info, "device deleted: %s",
3092					vol_args->name);
3093	}
3094out:
3095	kfree(vol_args);
3096err_drop:
3097	mnt_drop_write_file(file);
3098	return ret;
3099}
3100
3101static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3102{
3103	struct inode *inode = file_inode(file);
3104	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3105	struct btrfs_ioctl_vol_args *vol_args;
3106	int ret;
3107
3108	if (!capable(CAP_SYS_ADMIN))
3109		return -EPERM;
3110
3111	ret = mnt_want_write_file(file);
3112	if (ret)
3113		return ret;
3114
3115	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3116		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3117		goto out_drop_write;
3118	}
3119
3120	vol_args = memdup_user(arg, sizeof(*vol_args));
3121	if (IS_ERR(vol_args)) {
3122		ret = PTR_ERR(vol_args);
3123		goto out;
3124	}
3125
3126	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3127	ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3128
3129	if (!ret)
3130		btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3131	kfree(vol_args);
3132out:
3133	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3134out_drop_write:
3135	mnt_drop_write_file(file);
3136
3137	return ret;
3138}
3139
3140static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3141				void __user *arg)
3142{
3143	struct btrfs_ioctl_fs_info_args *fi_args;
3144	struct btrfs_device *device;
3145	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3146	int ret = 0;
3147
3148	fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3149	if (!fi_args)
3150		return -ENOMEM;
3151
3152	rcu_read_lock();
3153	fi_args->num_devices = fs_devices->num_devices;
3154
3155	list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3156		if (device->devid > fi_args->max_id)
3157			fi_args->max_id = device->devid;
3158	}
3159	rcu_read_unlock();
3160
3161	memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3162	fi_args->nodesize = fs_info->nodesize;
3163	fi_args->sectorsize = fs_info->sectorsize;
3164	fi_args->clone_alignment = fs_info->sectorsize;
3165
3166	if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3167		ret = -EFAULT;
3168
3169	kfree(fi_args);
3170	return ret;
3171}
3172
3173static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3174				 void __user *arg)
3175{
3176	struct btrfs_ioctl_dev_info_args *di_args;
3177	struct btrfs_device *dev;
3178	int ret = 0;
3179	char *s_uuid = NULL;
3180
3181	di_args = memdup_user(arg, sizeof(*di_args));
3182	if (IS_ERR(di_args))
3183		return PTR_ERR(di_args);
3184
3185	if (!btrfs_is_empty_uuid(di_args->uuid))
3186		s_uuid = di_args->uuid;
3187
3188	rcu_read_lock();
3189	dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3190				NULL, true);
3191
3192	if (!dev) {
3193		ret = -ENODEV;
3194		goto out;
3195	}
3196
3197	di_args->devid = dev->devid;
3198	di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3199	di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3200	memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3201	if (dev->name) {
3202		strncpy(di_args->path, rcu_str_deref(dev->name),
3203				sizeof(di_args->path) - 1);
 
 
3204		di_args->path[sizeof(di_args->path) - 1] = 0;
3205	} else {
3206		di_args->path[0] = '\0';
3207	}
3208
3209out:
3210	rcu_read_unlock();
3211	if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3212		ret = -EFAULT;
3213
3214	kfree(di_args);
3215	return ret;
3216}
3217
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3218static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3219				       struct inode *inode2, u64 loff2, u64 len)
3220{
3221	unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3222	unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3223}
3224
3225static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3226				     struct inode *inode2, u64 loff2, u64 len)
 
3227{
 
 
3228	if (inode1 < inode2) {
3229		swap(inode1, inode2);
3230		swap(loff1, loff2);
3231	} else if (inode1 == inode2 && loff2 < loff1) {
3232		swap(loff1, loff2);
3233	}
3234	lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3235	lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3236}
3237
3238static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
3239				   struct inode *dst, u64 dst_loff)
 
3240{
3241	int ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3242
3243	/*
3244	 * Lock destination range to serialize with concurrent readpages() and
3245	 * source range to serialize with relocation.
 
3246	 */
3247	btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3248	ret = btrfs_clone(src, dst, loff, len, len, dst_loff, 1);
3249	btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
 
 
 
 
 
3250
 
 
 
 
 
3251	return ret;
3252}
3253
3254#define BTRFS_MAX_DEDUPE_LEN	SZ_16M
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3255
3256static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3257			     struct inode *dst, u64 dst_loff)
3258{
3259	int ret;
3260	u64 i, tail_len, chunk_count;
3261	struct btrfs_root *root_dst = BTRFS_I(dst)->root;
 
 
 
3262
3263	spin_lock(&root_dst->root_item_lock);
3264	if (root_dst->send_in_progress) {
3265		btrfs_warn_rl(root_dst->fs_info,
3266"cannot deduplicate to root %llu while send operations are using it (%d in progress)",
3267			      root_dst->root_key.objectid,
3268			      root_dst->send_in_progress);
3269		spin_unlock(&root_dst->root_item_lock);
3270		return -EAGAIN;
3271	}
3272	root_dst->dedupe_in_progress++;
3273	spin_unlock(&root_dst->root_item_lock);
3274
3275	tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3276	chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3277
3278	for (i = 0; i < chunk_count; i++) {
3279		ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3280					      dst, dst_loff);
3281		if (ret)
3282			goto out;
3283
3284		loff += BTRFS_MAX_DEDUPE_LEN;
3285		dst_loff += BTRFS_MAX_DEDUPE_LEN;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3286	}
3287
3288	if (tail_len > 0)
3289		ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3290					      dst_loff);
3291out:
3292	spin_lock(&root_dst->root_item_lock);
3293	root_dst->dedupe_in_progress--;
3294	spin_unlock(&root_dst->root_item_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3295
3296	return ret;
3297}
3298
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3299static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3300				     struct inode *inode,
3301				     u64 endoff,
3302				     const u64 destoff,
3303				     const u64 olen,
3304				     int no_time_update)
3305{
3306	struct btrfs_root *root = BTRFS_I(inode)->root;
3307	int ret;
3308
3309	inode_inc_iversion(inode);
3310	if (!no_time_update)
3311		inode->i_mtime = inode->i_ctime = current_time(inode);
3312	/*
3313	 * We round up to the block size at eof when determining which
3314	 * extents to clone above, but shouldn't round up the file size.
3315	 */
3316	if (endoff > destoff + olen)
3317		endoff = destoff + olen;
3318	if (endoff > inode->i_size)
3319		btrfs_i_size_write(BTRFS_I(inode), endoff);
3320
3321	ret = btrfs_update_inode(trans, root, inode);
3322	if (ret) {
3323		btrfs_abort_transaction(trans, ret);
3324		btrfs_end_transaction(trans);
3325		goto out;
3326	}
3327	ret = btrfs_end_transaction(trans);
3328out:
3329	return ret;
3330}
3331
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3332/*
3333 * Make sure we do not end up inserting an inline extent into a file that has
3334 * already other (non-inline) extents. If a file has an inline extent it can
3335 * not have any other extents and the (single) inline extent must start at the
3336 * file offset 0. Failing to respect these rules will lead to file corruption,
3337 * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3338 *
3339 * We can have extents that have been already written to disk or we can have
3340 * dirty ranges still in delalloc, in which case the extent maps and items are
3341 * created only when we run delalloc, and the delalloc ranges might fall outside
3342 * the range we are currently locking in the inode's io tree. So we check the
3343 * inode's i_size because of that (i_size updates are done while holding the
3344 * i_mutex, which we are holding here).
3345 * We also check to see if the inode has a size not greater than "datal" but has
3346 * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3347 * protected against such concurrent fallocate calls by the i_mutex).
3348 *
3349 * If the file has no extents but a size greater than datal, do not allow the
3350 * copy because we would need turn the inline extent into a non-inline one (even
3351 * with NO_HOLES enabled). If we find our destination inode only has one inline
3352 * extent, just overwrite it with the source inline extent if its size is less
3353 * than the source extent's size, or we could copy the source inline extent's
3354 * data into the destination inode's inline extent if the later is greater then
3355 * the former.
3356 */
3357static int clone_copy_inline_extent(struct inode *dst,
3358				    struct btrfs_trans_handle *trans,
3359				    struct btrfs_path *path,
3360				    struct btrfs_key *new_key,
3361				    const u64 drop_start,
3362				    const u64 datal,
3363				    const u64 skip,
3364				    const u64 size,
3365				    char *inline_data)
3366{
3367	struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3368	struct btrfs_root *root = BTRFS_I(dst)->root;
3369	const u64 aligned_end = ALIGN(new_key->offset + datal,
3370				      fs_info->sectorsize);
3371	int ret;
3372	struct btrfs_key key;
3373
3374	if (new_key->offset > 0)
3375		return -EOPNOTSUPP;
3376
3377	key.objectid = btrfs_ino(BTRFS_I(dst));
3378	key.type = BTRFS_EXTENT_DATA_KEY;
3379	key.offset = 0;
3380	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3381	if (ret < 0) {
3382		return ret;
3383	} else if (ret > 0) {
3384		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3385			ret = btrfs_next_leaf(root, path);
3386			if (ret < 0)
3387				return ret;
3388			else if (ret > 0)
3389				goto copy_inline_extent;
3390		}
3391		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3392		if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3393		    key.type == BTRFS_EXTENT_DATA_KEY) {
3394			ASSERT(key.offset > 0);
3395			return -EOPNOTSUPP;
3396		}
3397	} else if (i_size_read(dst) <= datal) {
3398		struct btrfs_file_extent_item *ei;
3399		u64 ext_len;
3400
3401		/*
3402		 * If the file size is <= datal, make sure there are no other
3403		 * extents following (can happen do to an fallocate call with
3404		 * the flag FALLOC_FL_KEEP_SIZE).
3405		 */
3406		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3407				    struct btrfs_file_extent_item);
3408		/*
3409		 * If it's an inline extent, it can not have other extents
3410		 * following it.
3411		 */
3412		if (btrfs_file_extent_type(path->nodes[0], ei) ==
3413		    BTRFS_FILE_EXTENT_INLINE)
3414			goto copy_inline_extent;
3415
3416		ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3417		if (ext_len > aligned_end)
3418			return -EOPNOTSUPP;
3419
3420		ret = btrfs_next_item(root, path);
3421		if (ret < 0) {
3422			return ret;
3423		} else if (ret == 0) {
3424			btrfs_item_key_to_cpu(path->nodes[0], &key,
3425					      path->slots[0]);
3426			if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3427			    key.type == BTRFS_EXTENT_DATA_KEY)
3428				return -EOPNOTSUPP;
3429		}
3430	}
3431
3432copy_inline_extent:
3433	/*
3434	 * We have no extent items, or we have an extent at offset 0 which may
3435	 * or may not be inlined. All these cases are dealt the same way.
3436	 */
3437	if (i_size_read(dst) > datal) {
3438		/*
3439		 * If the destination inode has an inline extent...
3440		 * This would require copying the data from the source inline
3441		 * extent into the beginning of the destination's inline extent.
3442		 * But this is really complex, both extents can be compressed
3443		 * or just one of them, which would require decompressing and
3444		 * re-compressing data (which could increase the new compressed
3445		 * size, not allowing the compressed data to fit anymore in an
3446		 * inline extent).
3447		 * So just don't support this case for now (it should be rare,
3448		 * we are not really saving space when cloning inline extents).
3449		 */
3450		return -EOPNOTSUPP;
3451	}
3452
3453	btrfs_release_path(path);
3454	ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3455	if (ret)
3456		return ret;
3457	ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3458	if (ret)
3459		return ret;
3460
3461	if (skip) {
3462		const u32 start = btrfs_file_extent_calc_inline_size(0);
3463
3464		memmove(inline_data + start, inline_data + start + skip, datal);
3465	}
3466
3467	write_extent_buffer(path->nodes[0], inline_data,
3468			    btrfs_item_ptr_offset(path->nodes[0],
3469						  path->slots[0]),
3470			    size);
3471	inode_add_bytes(dst, datal);
3472	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags);
3473
3474	return 0;
3475}
3476
3477/**
3478 * btrfs_clone() - clone a range from inode file to another
3479 *
3480 * @src: Inode to clone from
3481 * @inode: Inode to clone to
3482 * @off: Offset within source to start clone from
3483 * @olen: Original length, passed by user, of range to clone
3484 * @olen_aligned: Block-aligned value of olen
3485 * @destoff: Offset within @inode to start clone
3486 * @no_time_update: Whether to update mtime/ctime on the target inode
3487 */
3488static int btrfs_clone(struct inode *src, struct inode *inode,
3489		       const u64 off, const u64 olen, const u64 olen_aligned,
3490		       const u64 destoff, int no_time_update)
3491{
3492	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3493	struct btrfs_root *root = BTRFS_I(inode)->root;
3494	struct btrfs_path *path = NULL;
3495	struct extent_buffer *leaf;
3496	struct btrfs_trans_handle *trans;
3497	char *buf = NULL;
3498	struct btrfs_key key;
3499	u32 nritems;
3500	int slot;
3501	int ret;
3502	const u64 len = olen_aligned;
3503	u64 last_dest_end = destoff;
3504
3505	ret = -ENOMEM;
3506	buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3507	if (!buf)
3508		return ret;
3509
3510	path = btrfs_alloc_path();
3511	if (!path) {
3512		kvfree(buf);
3513		return ret;
3514	}
3515
3516	path->reada = READA_FORWARD;
3517	/* clone data */
3518	key.objectid = btrfs_ino(BTRFS_I(src));
3519	key.type = BTRFS_EXTENT_DATA_KEY;
3520	key.offset = off;
3521
3522	while (1) {
3523		u64 next_key_min_offset = key.offset + 1;
3524		struct btrfs_file_extent_item *extent;
3525		int type;
3526		u32 size;
3527		struct btrfs_key new_key;
3528		u64 disko = 0, diskl = 0;
3529		u64 datao = 0, datal = 0;
3530		u8 comp;
3531		u64 drop_start;
3532
3533		/*
3534		 * note the key will change type as we walk through the
3535		 * tree.
3536		 */
3537		path->leave_spinning = 1;
3538		ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3539				0, 0);
3540		if (ret < 0)
3541			goto out;
3542		/*
3543		 * First search, if no extent item that starts at offset off was
3544		 * found but the previous item is an extent item, it's possible
3545		 * it might overlap our target range, therefore process it.
3546		 */
3547		if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3548			btrfs_item_key_to_cpu(path->nodes[0], &key,
3549					      path->slots[0] - 1);
3550			if (key.type == BTRFS_EXTENT_DATA_KEY)
3551				path->slots[0]--;
3552		}
3553
3554		nritems = btrfs_header_nritems(path->nodes[0]);
3555process_slot:
3556		if (path->slots[0] >= nritems) {
3557			ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3558			if (ret < 0)
3559				goto out;
3560			if (ret > 0)
3561				break;
3562			nritems = btrfs_header_nritems(path->nodes[0]);
3563		}
3564		leaf = path->nodes[0];
3565		slot = path->slots[0];
3566
3567		btrfs_item_key_to_cpu(leaf, &key, slot);
3568		if (key.type > BTRFS_EXTENT_DATA_KEY ||
3569		    key.objectid != btrfs_ino(BTRFS_I(src)))
3570			break;
3571
3572		ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
3573
3574		extent = btrfs_item_ptr(leaf, slot,
3575					struct btrfs_file_extent_item);
3576		comp = btrfs_file_extent_compression(leaf, extent);
3577		type = btrfs_file_extent_type(leaf, extent);
3578		if (type == BTRFS_FILE_EXTENT_REG ||
3579		    type == BTRFS_FILE_EXTENT_PREALLOC) {
3580			disko = btrfs_file_extent_disk_bytenr(leaf, extent);
3581			diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
3582			datao = btrfs_file_extent_offset(leaf, extent);
3583			datal = btrfs_file_extent_num_bytes(leaf, extent);
3584		} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3585			/* Take upper bound, may be compressed */
3586			datal = btrfs_file_extent_ram_bytes(leaf, extent);
3587		}
3588
3589		/*
3590		 * The first search might have left us at an extent item that
3591		 * ends before our target range's start, can happen if we have
3592		 * holes and NO_HOLES feature enabled.
3593		 */
3594		if (key.offset + datal <= off) {
3595			path->slots[0]++;
3596			goto process_slot;
3597		} else if (key.offset >= off + len) {
3598			break;
3599		}
3600		next_key_min_offset = key.offset + datal;
3601		size = btrfs_item_size_nr(leaf, slot);
3602		read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
3603				   size);
3604
3605		btrfs_release_path(path);
3606		path->leave_spinning = 0;
3607
3608		memcpy(&new_key, &key, sizeof(new_key));
3609		new_key.objectid = btrfs_ino(BTRFS_I(inode));
3610		if (off <= key.offset)
3611			new_key.offset = key.offset + destoff - off;
3612		else
3613			new_key.offset = destoff;
3614
3615		/*
3616		 * Deal with a hole that doesn't have an extent item that
3617		 * represents it (NO_HOLES feature enabled).
3618		 * This hole is either in the middle of the cloning range or at
3619		 * the beginning (fully overlaps it or partially overlaps it).
3620		 */
3621		if (new_key.offset != last_dest_end)
3622			drop_start = last_dest_end;
3623		else
3624			drop_start = new_key.offset;
3625
3626		if (type == BTRFS_FILE_EXTENT_REG ||
3627		    type == BTRFS_FILE_EXTENT_PREALLOC) {
3628			struct btrfs_clone_extent_info clone_info;
3629
3630			/*
3631			 *    a  | --- range to clone ---|  b
3632			 * | ------------- extent ------------- |
 
3633			 */
3634
3635			/* Subtract range b */
3636			if (key.offset + datal > off + len)
3637				datal = off + len - key.offset;
3638
3639			/* Subtract range a */
3640			if (off > key.offset) {
3641				datao += off - key.offset;
3642				datal -= off - key.offset;
3643			}
 
 
 
 
 
3644
3645			clone_info.disk_offset = disko;
3646			clone_info.disk_len = diskl;
3647			clone_info.data_offset = datao;
3648			clone_info.data_len = datal;
3649			clone_info.file_offset = new_key.offset;
3650			clone_info.extent_buf = buf;
3651			clone_info.item_size = size;
3652			ret = btrfs_punch_hole_range(inode, path,
3653						     drop_start,
3654						     new_key.offset + datal - 1,
3655						     &clone_info, &trans);
3656			if (ret)
3657				goto out;
3658		} else if (type == BTRFS_FILE_EXTENT_INLINE) {
3659			u64 skip = 0;
3660			u64 trim = 0;
3661
3662			if (off > key.offset) {
3663				skip = off - key.offset;
3664				new_key.offset += skip;
3665			}
3666
3667			if (key.offset + datal > off + len)
3668				trim = key.offset + datal - (off + len);
 
 
 
 
3669
3670			if (comp && (skip || trim)) {
3671				ret = -EINVAL;
3672				goto out;
3673			}
3674			size -= skip + trim;
3675			datal -= skip + trim;
 
 
 
 
 
3676
3677			/*
3678			 * If our extent is inline, we know we will drop or
3679			 * adjust at most 1 extent item in the destination root.
3680			 *
3681			 * 1 - adjusting old extent (we may have to split it)
3682			 * 1 - add new extent
3683			 * 1 - inode update
3684			 */
3685			trans = btrfs_start_transaction(root, 3);
3686			if (IS_ERR(trans)) {
3687				ret = PTR_ERR(trans);
3688				goto out;
3689			}
3690
3691			ret = clone_copy_inline_extent(inode, trans, path,
3692						       &new_key, drop_start,
3693						       datal, skip, size, buf);
3694			if (ret) {
3695				if (ret != -EOPNOTSUPP)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3696					btrfs_abort_transaction(trans, ret);
3697				btrfs_end_transaction(trans);
3698				goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3699			}
3700		}
3701
3702		btrfs_release_path(path);
 
 
 
 
3703
3704		last_dest_end = ALIGN(new_key.offset + datal,
3705				      fs_info->sectorsize);
3706		ret = clone_finish_inode_update(trans, inode, last_dest_end,
3707						destoff, olen, no_time_update);
3708		if (ret)
3709			goto out;
3710		if (new_key.offset + datal >= destoff + len)
3711			break;
3712
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3713		btrfs_release_path(path);
3714		key.offset = next_key_min_offset;
3715
3716		if (fatal_signal_pending(current)) {
3717			ret = -EINTR;
3718			goto out;
3719		}
3720	}
3721	ret = 0;
3722
3723	if (last_dest_end < destoff + len) {
3724		struct btrfs_clone_extent_info clone_info = { 0 };
3725		/*
3726		 * We have an implicit hole (NO_HOLES feature is enabled) that
3727		 * fully or partially overlaps our cloning range at its end.
3728		 */
3729		btrfs_release_path(path);
3730		path->leave_spinning = 0;
3731
3732		/*
3733		 * We are dealing with a hole and our clone_info already has a
3734		 * disk_offset of 0, we only need to fill the data length and
3735		 * file offset.
3736		 */
3737		clone_info.data_len = destoff + len - last_dest_end;
3738		clone_info.file_offset = last_dest_end;
3739		ret = btrfs_punch_hole_range(inode, path,
3740					     last_dest_end, destoff + len - 1,
3741					     &clone_info, &trans);
3742		if (ret)
3743			goto out;
3744
 
 
 
 
 
 
 
 
 
 
 
3745		ret = clone_finish_inode_update(trans, inode, destoff + len,
3746						destoff, olen, no_time_update);
3747	}
3748
3749out:
3750	btrfs_free_path(path);
3751	kvfree(buf);
3752	return ret;
3753}
3754
3755static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3756					u64 off, u64 olen, u64 destoff)
3757{
3758	struct inode *inode = file_inode(file);
3759	struct inode *src = file_inode(file_src);
3760	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 
3761	int ret;
3762	u64 len = olen;
3763	u64 bs = fs_info->sb->s_blocksize;
 
3764
3765	/*
3766	 * TODO:
3767	 * - split compressed inline extents.  annoying: we need to
3768	 *   decompress into destination's address_space (the file offset
3769	 *   may change, so source mapping won't do), then recompress (or
3770	 *   otherwise reinsert) a subrange.
3771	 *
3772	 * - split destination inode's inline extents.  The inline extents can
3773	 *   be either compressed or non-compressed.
3774	 */
3775
3776	/*
3777	 * VFS's generic_remap_file_range_prep() protects us from cloning the
3778	 * eof block into the middle of a file, which would result in corruption
3779	 * if the file size is not blocksize aligned. So we don't need to check
3780	 * for that case here.
3781	 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3782	if (off + len == src->i_size)
3783		len = ALIGN(src->i_size, bs) - off;
3784
3785	if (destoff > inode->i_size) {
3786		const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
 
 
 
 
 
 
 
 
 
 
 
 
 
3787
 
3788		ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3789		if (ret)
3790			return ret;
3791		/*
3792		 * We may have truncated the last block if the inode's size is
3793		 * not sector size aligned, so we need to wait for writeback to
3794		 * complete before proceeding further, otherwise we can race
3795		 * with cloning and attempt to increment a reference to an
3796		 * extent that no longer exists (writeback completed right after
3797		 * we found the previous extent covering eof and before we
3798		 * attempted to increment its reference count).
3799		 */
3800		ret = btrfs_wait_ordered_range(inode, wb_start,
3801					       destoff - wb_start);
3802		if (ret)
3803			return ret;
3804	}
3805
3806	/*
3807	 * Lock destination range to serialize with concurrent readpages() and
3808	 * source range to serialize with relocation.
 
 
 
3809	 */
3810	btrfs_double_extent_lock(src, off, inode, destoff, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3811	ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3812	btrfs_double_extent_unlock(src, off, inode, destoff, len);
 
 
 
 
 
 
 
 
3813	/*
3814	 * Truncate page cache pages so that future reads will see the cloned
3815	 * data immediately and not the previous data.
3816	 */
3817	truncate_inode_pages_range(&inode->i_data,
3818				round_down(destoff, PAGE_SIZE),
3819				round_up(destoff + len, PAGE_SIZE) - 1);
3820
3821	return ret;
3822}
3823
3824static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
3825				       struct file *file_out, loff_t pos_out,
3826				       loff_t *len, unsigned int remap_flags)
3827{
3828	struct inode *inode_in = file_inode(file_in);
3829	struct inode *inode_out = file_inode(file_out);
3830	u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
3831	bool same_inode = inode_out == inode_in;
3832	u64 wb_len;
3833	int ret;
3834
3835	if (!(remap_flags & REMAP_FILE_DEDUP)) {
3836		struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
3837
3838		if (btrfs_root_readonly(root_out))
3839			return -EROFS;
3840
3841		if (file_in->f_path.mnt != file_out->f_path.mnt ||
3842		    inode_in->i_sb != inode_out->i_sb)
3843			return -EXDEV;
3844	}
3845
3846	/* don't make the dst file partly checksummed */
3847	if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
3848	    (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
3849		return -EINVAL;
3850	}
3851
3852	/*
3853	 * Now that the inodes are locked, we need to start writeback ourselves
3854	 * and can not rely on the writeback from the VFS's generic helper
3855	 * generic_remap_file_range_prep() because:
3856	 *
3857	 * 1) For compression we must call filemap_fdatawrite_range() range
3858	 *    twice (btrfs_fdatawrite_range() does it for us), and the generic
3859	 *    helper only calls it once;
3860	 *
3861	 * 2) filemap_fdatawrite_range(), called by the generic helper only
3862	 *    waits for the writeback to complete, i.e. for IO to be done, and
3863	 *    not for the ordered extents to complete. We need to wait for them
3864	 *    to complete so that new file extent items are in the fs tree.
3865	 */
3866	if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
3867		wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
3868	else
3869		wb_len = ALIGN(*len, bs);
3870
3871	/*
3872	 * Since we don't lock ranges, wait for ongoing lockless dio writes (as
3873	 * any in progress could create its ordered extents after we wait for
3874	 * existing ordered extents below).
3875	 */
3876	inode_dio_wait(inode_in);
3877	if (!same_inode)
3878		inode_dio_wait(inode_out);
3879
3880	/*
3881	 * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
3882	 *
3883	 * Btrfs' back references do not have a block level granularity, they
3884	 * work at the whole extent level.
3885	 * NOCOW buffered write without data space reserved may not be able
3886	 * to fall back to CoW due to lack of data space, thus could cause
3887	 * data loss.
3888	 *
3889	 * Here we take a shortcut by flushing the whole inode, so that all
3890	 * nocow write should reach disk as nocow before we increase the
3891	 * reference of the extent. We could do better by only flushing NOCOW
3892	 * data, but that needs extra accounting.
3893	 *
3894	 * Also we don't need to check ASYNC_EXTENT, as async extent will be
3895	 * CoWed anyway, not affecting nocow part.
3896	 */
3897	ret = filemap_flush(inode_in->i_mapping);
3898	if (ret < 0)
3899		return ret;
3900
3901	ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
3902				       wb_len);
3903	if (ret < 0)
3904		return ret;
3905	ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
3906				       wb_len);
3907	if (ret < 0)
3908		return ret;
3909
3910	return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
3911					    len, remap_flags);
3912}
3913
3914loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
3915		struct file *dst_file, loff_t destoff, loff_t len,
3916		unsigned int remap_flags)
3917{
3918	struct inode *src_inode = file_inode(src_file);
3919	struct inode *dst_inode = file_inode(dst_file);
3920	bool same_inode = dst_inode == src_inode;
3921	int ret;
3922
3923	if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
3924		return -EINVAL;
3925
3926	if (same_inode)
3927		inode_lock(src_inode);
3928	else
3929		lock_two_nondirectories(src_inode, dst_inode);
3930
3931	ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
3932					  &len, remap_flags);
3933	if (ret < 0 || len == 0)
3934		goto out_unlock;
3935
3936	if (remap_flags & REMAP_FILE_DEDUP)
3937		ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
3938	else
3939		ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
3940
3941out_unlock:
3942	if (same_inode)
3943		inode_unlock(src_inode);
3944	else
3945		unlock_two_nondirectories(src_inode, dst_inode);
3946
3947	return ret < 0 ? ret : len;
3948}
3949
3950static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3951{
3952	struct inode *inode = file_inode(file);
3953	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3954	struct btrfs_root *root = BTRFS_I(inode)->root;
3955	struct btrfs_root *new_root;
3956	struct btrfs_dir_item *di;
3957	struct btrfs_trans_handle *trans;
3958	struct btrfs_path *path;
3959	struct btrfs_key location;
3960	struct btrfs_disk_key disk_key;
3961	u64 objectid = 0;
3962	u64 dir_id;
3963	int ret;
3964
3965	if (!capable(CAP_SYS_ADMIN))
3966		return -EPERM;
3967
3968	ret = mnt_want_write_file(file);
3969	if (ret)
3970		return ret;
3971
3972	if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3973		ret = -EFAULT;
3974		goto out;
3975	}
3976
3977	if (!objectid)
3978		objectid = BTRFS_FS_TREE_OBJECTID;
3979
3980	location.objectid = objectid;
3981	location.type = BTRFS_ROOT_ITEM_KEY;
3982	location.offset = (u64)-1;
3983
3984	new_root = btrfs_read_fs_root_no_name(fs_info, &location);
3985	if (IS_ERR(new_root)) {
3986		ret = PTR_ERR(new_root);
3987		goto out;
3988	}
3989	if (!is_fstree(new_root->root_key.objectid)) {
3990		ret = -ENOENT;
3991		goto out;
3992	}
3993
3994	path = btrfs_alloc_path();
3995	if (!path) {
3996		ret = -ENOMEM;
3997		goto out;
3998	}
3999	path->leave_spinning = 1;
4000
4001	trans = btrfs_start_transaction(root, 1);
4002	if (IS_ERR(trans)) {
4003		btrfs_free_path(path);
4004		ret = PTR_ERR(trans);
4005		goto out;
4006	}
4007
4008	dir_id = btrfs_super_root_dir(fs_info->super_copy);
4009	di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4010				   dir_id, "default", 7, 1);
4011	if (IS_ERR_OR_NULL(di)) {
4012		btrfs_free_path(path);
4013		btrfs_end_transaction(trans);
4014		btrfs_err(fs_info,
4015			  "Umm, you don't have the default diritem, this isn't going to work");
4016		ret = -ENOENT;
4017		goto out;
4018	}
4019
4020	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4021	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4022	btrfs_mark_buffer_dirty(path->nodes[0]);
4023	btrfs_free_path(path);
4024
4025	btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4026	btrfs_end_transaction(trans);
4027out:
4028	mnt_drop_write_file(file);
4029	return ret;
4030}
4031
4032static void get_block_group_info(struct list_head *groups_list,
4033				 struct btrfs_ioctl_space_info *space)
4034{
4035	struct btrfs_block_group_cache *block_group;
4036
4037	space->total_bytes = 0;
4038	space->used_bytes = 0;
4039	space->flags = 0;
4040	list_for_each_entry(block_group, groups_list, list) {
4041		space->flags = block_group->flags;
4042		space->total_bytes += block_group->key.offset;
4043		space->used_bytes +=
4044			btrfs_block_group_used(&block_group->item);
4045	}
4046}
4047
4048static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4049				   void __user *arg)
4050{
4051	struct btrfs_ioctl_space_args space_args;
4052	struct btrfs_ioctl_space_info space;
4053	struct btrfs_ioctl_space_info *dest;
4054	struct btrfs_ioctl_space_info *dest_orig;
4055	struct btrfs_ioctl_space_info __user *user_dest;
4056	struct btrfs_space_info *info;
4057	static const u64 types[] = {
4058		BTRFS_BLOCK_GROUP_DATA,
4059		BTRFS_BLOCK_GROUP_SYSTEM,
4060		BTRFS_BLOCK_GROUP_METADATA,
4061		BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4062	};
4063	int num_types = 4;
4064	int alloc_size;
4065	int ret = 0;
4066	u64 slot_count = 0;
4067	int i, c;
4068
4069	if (copy_from_user(&space_args,
4070			   (struct btrfs_ioctl_space_args __user *)arg,
4071			   sizeof(space_args)))
4072		return -EFAULT;
4073
4074	for (i = 0; i < num_types; i++) {
4075		struct btrfs_space_info *tmp;
4076
4077		info = NULL;
4078		rcu_read_lock();
4079		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4080					list) {
4081			if (tmp->flags == types[i]) {
4082				info = tmp;
4083				break;
4084			}
4085		}
4086		rcu_read_unlock();
4087
4088		if (!info)
4089			continue;
4090
4091		down_read(&info->groups_sem);
4092		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4093			if (!list_empty(&info->block_groups[c]))
4094				slot_count++;
4095		}
4096		up_read(&info->groups_sem);
4097	}
4098
4099	/*
4100	 * Global block reserve, exported as a space_info
4101	 */
4102	slot_count++;
4103
4104	/* space_slots == 0 means they are asking for a count */
4105	if (space_args.space_slots == 0) {
4106		space_args.total_spaces = slot_count;
4107		goto out;
4108	}
4109
4110	slot_count = min_t(u64, space_args.space_slots, slot_count);
4111
4112	alloc_size = sizeof(*dest) * slot_count;
4113
4114	/* we generally have at most 6 or so space infos, one for each raid
4115	 * level.  So, a whole page should be more than enough for everyone
4116	 */
4117	if (alloc_size > PAGE_SIZE)
4118		return -ENOMEM;
4119
4120	space_args.total_spaces = 0;
4121	dest = kmalloc(alloc_size, GFP_KERNEL);
4122	if (!dest)
4123		return -ENOMEM;
4124	dest_orig = dest;
4125
4126	/* now we have a buffer to copy into */
4127	for (i = 0; i < num_types; i++) {
4128		struct btrfs_space_info *tmp;
4129
4130		if (!slot_count)
4131			break;
4132
4133		info = NULL;
4134		rcu_read_lock();
4135		list_for_each_entry_rcu(tmp, &fs_info->space_info,
4136					list) {
4137			if (tmp->flags == types[i]) {
4138				info = tmp;
4139				break;
4140			}
4141		}
4142		rcu_read_unlock();
4143
4144		if (!info)
4145			continue;
4146		down_read(&info->groups_sem);
4147		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4148			if (!list_empty(&info->block_groups[c])) {
4149				get_block_group_info(&info->block_groups[c],
4150						     &space);
4151				memcpy(dest, &space, sizeof(space));
4152				dest++;
4153				space_args.total_spaces++;
4154				slot_count--;
4155			}
4156			if (!slot_count)
4157				break;
4158		}
4159		up_read(&info->groups_sem);
4160	}
4161
4162	/*
4163	 * Add global block reserve
4164	 */
4165	if (slot_count) {
4166		struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4167
4168		spin_lock(&block_rsv->lock);
4169		space.total_bytes = block_rsv->size;
4170		space.used_bytes = block_rsv->size - block_rsv->reserved;
4171		spin_unlock(&block_rsv->lock);
4172		space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4173		memcpy(dest, &space, sizeof(space));
4174		space_args.total_spaces++;
4175	}
4176
4177	user_dest = (struct btrfs_ioctl_space_info __user *)
4178		(arg + sizeof(struct btrfs_ioctl_space_args));
4179
4180	if (copy_to_user(user_dest, dest_orig, alloc_size))
4181		ret = -EFAULT;
4182
4183	kfree(dest_orig);
4184out:
4185	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4186		ret = -EFAULT;
4187
4188	return ret;
4189}
4190
4191static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4192					    void __user *argp)
4193{
4194	struct btrfs_trans_handle *trans;
4195	u64 transid;
4196	int ret;
4197
4198	trans = btrfs_attach_transaction_barrier(root);
4199	if (IS_ERR(trans)) {
4200		if (PTR_ERR(trans) != -ENOENT)
4201			return PTR_ERR(trans);
4202
4203		/* No running transaction, don't bother */
4204		transid = root->fs_info->last_trans_committed;
4205		goto out;
4206	}
4207	transid = trans->transid;
4208	ret = btrfs_commit_transaction_async(trans, 0);
4209	if (ret) {
4210		btrfs_end_transaction(trans);
4211		return ret;
4212	}
4213out:
4214	if (argp)
4215		if (copy_to_user(argp, &transid, sizeof(transid)))
4216			return -EFAULT;
4217	return 0;
4218}
4219
4220static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4221					   void __user *argp)
4222{
4223	u64 transid;
4224
4225	if (argp) {
4226		if (copy_from_user(&transid, argp, sizeof(transid)))
4227			return -EFAULT;
4228	} else {
4229		transid = 0;  /* current trans */
4230	}
4231	return btrfs_wait_for_commit(fs_info, transid);
4232}
4233
4234static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4235{
4236	struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4237	struct btrfs_ioctl_scrub_args *sa;
4238	int ret;
4239
4240	if (!capable(CAP_SYS_ADMIN))
4241		return -EPERM;
4242
4243	sa = memdup_user(arg, sizeof(*sa));
4244	if (IS_ERR(sa))
4245		return PTR_ERR(sa);
4246
4247	if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4248		ret = mnt_want_write_file(file);
4249		if (ret)
4250			goto out;
4251	}
4252
4253	ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4254			      &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4255			      0);
4256
4257	if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4258		ret = -EFAULT;
4259
4260	if (!(sa->flags & BTRFS_SCRUB_READONLY))
4261		mnt_drop_write_file(file);
4262out:
4263	kfree(sa);
4264	return ret;
4265}
4266
4267static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4268{
4269	if (!capable(CAP_SYS_ADMIN))
4270		return -EPERM;
4271
4272	return btrfs_scrub_cancel(fs_info);
4273}
4274
4275static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4276				       void __user *arg)
4277{
4278	struct btrfs_ioctl_scrub_args *sa;
4279	int ret;
4280
4281	if (!capable(CAP_SYS_ADMIN))
4282		return -EPERM;
4283
4284	sa = memdup_user(arg, sizeof(*sa));
4285	if (IS_ERR(sa))
4286		return PTR_ERR(sa);
4287
4288	ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4289
4290	if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4291		ret = -EFAULT;
4292
4293	kfree(sa);
4294	return ret;
4295}
4296
4297static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4298				      void __user *arg)
4299{
4300	struct btrfs_ioctl_get_dev_stats *sa;
4301	int ret;
4302
4303	sa = memdup_user(arg, sizeof(*sa));
4304	if (IS_ERR(sa))
4305		return PTR_ERR(sa);
4306
4307	if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4308		kfree(sa);
4309		return -EPERM;
4310	}
4311
4312	ret = btrfs_get_dev_stats(fs_info, sa);
4313
4314	if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4315		ret = -EFAULT;
4316
4317	kfree(sa);
4318	return ret;
4319}
4320
4321static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4322				    void __user *arg)
4323{
4324	struct btrfs_ioctl_dev_replace_args *p;
4325	int ret;
4326
4327	if (!capable(CAP_SYS_ADMIN))
4328		return -EPERM;
4329
4330	p = memdup_user(arg, sizeof(*p));
4331	if (IS_ERR(p))
4332		return PTR_ERR(p);
4333
4334	switch (p->cmd) {
4335	case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4336		if (sb_rdonly(fs_info->sb)) {
4337			ret = -EROFS;
4338			goto out;
4339		}
4340		if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4341			ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4342		} else {
4343			ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4344			clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4345		}
4346		break;
4347	case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4348		btrfs_dev_replace_status(fs_info, p);
4349		ret = 0;
4350		break;
4351	case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4352		p->result = btrfs_dev_replace_cancel(fs_info);
4353		ret = 0;
4354		break;
4355	default:
4356		ret = -EINVAL;
4357		break;
4358	}
4359
4360	if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4361		ret = -EFAULT;
4362out:
4363	kfree(p);
4364	return ret;
4365}
4366
4367static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4368{
4369	int ret = 0;
4370	int i;
4371	u64 rel_ptr;
4372	int size;
4373	struct btrfs_ioctl_ino_path_args *ipa = NULL;
4374	struct inode_fs_paths *ipath = NULL;
4375	struct btrfs_path *path;
4376
4377	if (!capable(CAP_DAC_READ_SEARCH))
4378		return -EPERM;
4379
4380	path = btrfs_alloc_path();
4381	if (!path) {
4382		ret = -ENOMEM;
4383		goto out;
4384	}
4385
4386	ipa = memdup_user(arg, sizeof(*ipa));
4387	if (IS_ERR(ipa)) {
4388		ret = PTR_ERR(ipa);
4389		ipa = NULL;
4390		goto out;
4391	}
4392
4393	size = min_t(u32, ipa->size, 4096);
4394	ipath = init_ipath(size, root, path);
4395	if (IS_ERR(ipath)) {
4396		ret = PTR_ERR(ipath);
4397		ipath = NULL;
4398		goto out;
4399	}
4400
4401	ret = paths_from_inode(ipa->inum, ipath);
4402	if (ret < 0)
4403		goto out;
4404
4405	for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4406		rel_ptr = ipath->fspath->val[i] -
4407			  (u64)(unsigned long)ipath->fspath->val;
4408		ipath->fspath->val[i] = rel_ptr;
4409	}
4410
4411	ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4412			   ipath->fspath, size);
4413	if (ret) {
4414		ret = -EFAULT;
4415		goto out;
4416	}
4417
4418out:
4419	btrfs_free_path(path);
4420	free_ipath(ipath);
4421	kfree(ipa);
4422
4423	return ret;
4424}
4425
4426static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4427{
4428	struct btrfs_data_container *inodes = ctx;
4429	const size_t c = 3 * sizeof(u64);
4430
4431	if (inodes->bytes_left >= c) {
4432		inodes->bytes_left -= c;
4433		inodes->val[inodes->elem_cnt] = inum;
4434		inodes->val[inodes->elem_cnt + 1] = offset;
4435		inodes->val[inodes->elem_cnt + 2] = root;
4436		inodes->elem_cnt += 3;
4437	} else {
4438		inodes->bytes_missing += c - inodes->bytes_left;
4439		inodes->bytes_left = 0;
4440		inodes->elem_missed += 3;
4441	}
4442
4443	return 0;
4444}
4445
4446static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4447					void __user *arg, int version)
4448{
4449	int ret = 0;
4450	int size;
4451	struct btrfs_ioctl_logical_ino_args *loi;
4452	struct btrfs_data_container *inodes = NULL;
4453	struct btrfs_path *path = NULL;
4454	bool ignore_offset;
4455
4456	if (!capable(CAP_SYS_ADMIN))
4457		return -EPERM;
4458
4459	loi = memdup_user(arg, sizeof(*loi));
4460	if (IS_ERR(loi))
4461		return PTR_ERR(loi);
4462
4463	if (version == 1) {
4464		ignore_offset = false;
4465		size = min_t(u32, loi->size, SZ_64K);
4466	} else {
4467		/* All reserved bits must be 0 for now */
4468		if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4469			ret = -EINVAL;
4470			goto out_loi;
4471		}
4472		/* Only accept flags we have defined so far */
4473		if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4474			ret = -EINVAL;
4475			goto out_loi;
4476		}
4477		ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4478		size = min_t(u32, loi->size, SZ_16M);
4479	}
4480
4481	path = btrfs_alloc_path();
4482	if (!path) {
4483		ret = -ENOMEM;
4484		goto out;
4485	}
4486
4487	inodes = init_data_container(size);
4488	if (IS_ERR(inodes)) {
4489		ret = PTR_ERR(inodes);
4490		inodes = NULL;
4491		goto out;
4492	}
4493
4494	ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4495					  build_ino_list, inodes, ignore_offset);
4496	if (ret == -EINVAL)
4497		ret = -ENOENT;
4498	if (ret < 0)
4499		goto out;
4500
4501	ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4502			   size);
4503	if (ret)
4504		ret = -EFAULT;
4505
4506out:
4507	btrfs_free_path(path);
4508	kvfree(inodes);
4509out_loi:
4510	kfree(loi);
4511
4512	return ret;
4513}
4514
4515void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4516			       struct btrfs_ioctl_balance_args *bargs)
4517{
4518	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4519
4520	bargs->flags = bctl->flags;
4521
4522	if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4523		bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4524	if (atomic_read(&fs_info->balance_pause_req))
4525		bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4526	if (atomic_read(&fs_info->balance_cancel_req))
4527		bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4528
4529	memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4530	memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4531	memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4532
4533	spin_lock(&fs_info->balance_lock);
4534	memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4535	spin_unlock(&fs_info->balance_lock);
 
 
 
 
4536}
4537
4538static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4539{
4540	struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4541	struct btrfs_fs_info *fs_info = root->fs_info;
4542	struct btrfs_ioctl_balance_args *bargs;
4543	struct btrfs_balance_control *bctl;
4544	bool need_unlock; /* for mut. excl. ops lock */
4545	int ret;
4546
4547	if (!capable(CAP_SYS_ADMIN))
4548		return -EPERM;
4549
4550	ret = mnt_want_write_file(file);
4551	if (ret)
4552		return ret;
4553
4554again:
4555	if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
 
4556		mutex_lock(&fs_info->balance_mutex);
4557		need_unlock = true;
4558		goto locked;
4559	}
4560
4561	/*
4562	 * mut. excl. ops lock is locked.  Three possibilities:
4563	 *   (1) some other op is running
4564	 *   (2) balance is running
4565	 *   (3) balance is paused -- special case (think resume)
4566	 */
4567	mutex_lock(&fs_info->balance_mutex);
4568	if (fs_info->balance_ctl) {
4569		/* this is either (2) or (3) */
4570		if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4571			mutex_unlock(&fs_info->balance_mutex);
4572			/*
4573			 * Lock released to allow other waiters to continue,
4574			 * we'll reexamine the status again.
4575			 */
4576			mutex_lock(&fs_info->balance_mutex);
4577
4578			if (fs_info->balance_ctl &&
4579			    !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4580				/* this is (3) */
4581				need_unlock = false;
4582				goto locked;
4583			}
4584
4585			mutex_unlock(&fs_info->balance_mutex);
 
4586			goto again;
4587		} else {
4588			/* this is (2) */
4589			mutex_unlock(&fs_info->balance_mutex);
4590			ret = -EINPROGRESS;
4591			goto out;
4592		}
4593	} else {
4594		/* this is (1) */
4595		mutex_unlock(&fs_info->balance_mutex);
4596		ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4597		goto out;
4598	}
4599
4600locked:
4601	BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4602
4603	if (arg) {
4604		bargs = memdup_user(arg, sizeof(*bargs));
4605		if (IS_ERR(bargs)) {
4606			ret = PTR_ERR(bargs);
4607			goto out_unlock;
4608		}
4609
4610		if (bargs->flags & BTRFS_BALANCE_RESUME) {
4611			if (!fs_info->balance_ctl) {
4612				ret = -ENOTCONN;
4613				goto out_bargs;
4614			}
4615
4616			bctl = fs_info->balance_ctl;
4617			spin_lock(&fs_info->balance_lock);
4618			bctl->flags |= BTRFS_BALANCE_RESUME;
4619			spin_unlock(&fs_info->balance_lock);
4620
4621			goto do_balance;
4622		}
4623	} else {
4624		bargs = NULL;
4625	}
4626
4627	if (fs_info->balance_ctl) {
4628		ret = -EINPROGRESS;
4629		goto out_bargs;
4630	}
4631
4632	bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4633	if (!bctl) {
4634		ret = -ENOMEM;
4635		goto out_bargs;
4636	}
4637
 
4638	if (arg) {
4639		memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4640		memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4641		memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4642
4643		bctl->flags = bargs->flags;
4644	} else {
4645		/* balance everything - no filters */
4646		bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4647	}
4648
4649	if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4650		ret = -EINVAL;
4651		goto out_bctl;
4652	}
4653
4654do_balance:
4655	/*
4656	 * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4657	 * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4658	 * restriper was paused all the way until unmount, in free_fs_info.
4659	 * The flag should be cleared after reset_balance_state.
4660	 */
4661	need_unlock = false;
4662
4663	ret = btrfs_balance(fs_info, bctl, bargs);
4664	bctl = NULL;
4665
4666	if ((ret == 0 || ret == -ECANCELED) && arg) {
4667		if (copy_to_user(arg, bargs, sizeof(*bargs)))
4668			ret = -EFAULT;
4669	}
4670
4671out_bctl:
4672	kfree(bctl);
4673out_bargs:
4674	kfree(bargs);
4675out_unlock:
4676	mutex_unlock(&fs_info->balance_mutex);
 
4677	if (need_unlock)
4678		clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4679out:
4680	mnt_drop_write_file(file);
4681	return ret;
4682}
4683
4684static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4685{
4686	if (!capable(CAP_SYS_ADMIN))
4687		return -EPERM;
4688
4689	switch (cmd) {
4690	case BTRFS_BALANCE_CTL_PAUSE:
4691		return btrfs_pause_balance(fs_info);
4692	case BTRFS_BALANCE_CTL_CANCEL:
4693		return btrfs_cancel_balance(fs_info);
4694	}
4695
4696	return -EINVAL;
4697}
4698
4699static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4700					 void __user *arg)
4701{
4702	struct btrfs_ioctl_balance_args *bargs;
4703	int ret = 0;
4704
4705	if (!capable(CAP_SYS_ADMIN))
4706		return -EPERM;
4707
4708	mutex_lock(&fs_info->balance_mutex);
4709	if (!fs_info->balance_ctl) {
4710		ret = -ENOTCONN;
4711		goto out;
4712	}
4713
4714	bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4715	if (!bargs) {
4716		ret = -ENOMEM;
4717		goto out;
4718	}
4719
4720	btrfs_update_ioctl_balance_args(fs_info, bargs);
4721
4722	if (copy_to_user(arg, bargs, sizeof(*bargs)))
4723		ret = -EFAULT;
4724
4725	kfree(bargs);
4726out:
4727	mutex_unlock(&fs_info->balance_mutex);
4728	return ret;
4729}
4730
4731static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4732{
4733	struct inode *inode = file_inode(file);
4734	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4735	struct btrfs_ioctl_quota_ctl_args *sa;
 
4736	int ret;
 
4737
4738	if (!capable(CAP_SYS_ADMIN))
4739		return -EPERM;
4740
4741	ret = mnt_want_write_file(file);
4742	if (ret)
4743		return ret;
4744
4745	sa = memdup_user(arg, sizeof(*sa));
4746	if (IS_ERR(sa)) {
4747		ret = PTR_ERR(sa);
4748		goto drop_write;
4749	}
4750
4751	down_write(&fs_info->subvol_sem);
 
 
 
 
 
4752
4753	switch (sa->cmd) {
4754	case BTRFS_QUOTA_CTL_ENABLE:
4755		ret = btrfs_quota_enable(fs_info);
4756		break;
4757	case BTRFS_QUOTA_CTL_DISABLE:
4758		ret = btrfs_quota_disable(fs_info);
4759		break;
4760	default:
4761		ret = -EINVAL;
4762		break;
4763	}
4764
 
 
 
 
4765	kfree(sa);
4766	up_write(&fs_info->subvol_sem);
4767drop_write:
4768	mnt_drop_write_file(file);
4769	return ret;
4770}
4771
4772static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4773{
4774	struct inode *inode = file_inode(file);
4775	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4776	struct btrfs_root *root = BTRFS_I(inode)->root;
4777	struct btrfs_ioctl_qgroup_assign_args *sa;
4778	struct btrfs_trans_handle *trans;
4779	int ret;
4780	int err;
4781
4782	if (!capable(CAP_SYS_ADMIN))
4783		return -EPERM;
4784
4785	ret = mnt_want_write_file(file);
4786	if (ret)
4787		return ret;
4788
4789	sa = memdup_user(arg, sizeof(*sa));
4790	if (IS_ERR(sa)) {
4791		ret = PTR_ERR(sa);
4792		goto drop_write;
4793	}
4794
4795	trans = btrfs_join_transaction(root);
4796	if (IS_ERR(trans)) {
4797		ret = PTR_ERR(trans);
4798		goto out;
4799	}
4800
4801	if (sa->assign) {
4802		ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
 
4803	} else {
4804		ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
 
4805	}
4806
4807	/* update qgroup status and info */
4808	err = btrfs_run_qgroups(trans);
4809	if (err < 0)
4810		btrfs_handle_fs_error(fs_info, err,
4811				      "failed to update qgroup status and info");
4812	err = btrfs_end_transaction(trans);
4813	if (err && !ret)
4814		ret = err;
4815
4816out:
4817	kfree(sa);
4818drop_write:
4819	mnt_drop_write_file(file);
4820	return ret;
4821}
4822
4823static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4824{
4825	struct inode *inode = file_inode(file);
 
4826	struct btrfs_root *root = BTRFS_I(inode)->root;
4827	struct btrfs_ioctl_qgroup_create_args *sa;
4828	struct btrfs_trans_handle *trans;
4829	int ret;
4830	int err;
4831
4832	if (!capable(CAP_SYS_ADMIN))
4833		return -EPERM;
4834
4835	ret = mnt_want_write_file(file);
4836	if (ret)
4837		return ret;
4838
4839	sa = memdup_user(arg, sizeof(*sa));
4840	if (IS_ERR(sa)) {
4841		ret = PTR_ERR(sa);
4842		goto drop_write;
4843	}
4844
4845	if (!sa->qgroupid) {
4846		ret = -EINVAL;
4847		goto out;
4848	}
4849
4850	trans = btrfs_join_transaction(root);
4851	if (IS_ERR(trans)) {
4852		ret = PTR_ERR(trans);
4853		goto out;
4854	}
4855
4856	if (sa->create) {
4857		ret = btrfs_create_qgroup(trans, sa->qgroupid);
4858	} else {
4859		ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4860	}
4861
4862	err = btrfs_end_transaction(trans);
4863	if (err && !ret)
4864		ret = err;
4865
4866out:
4867	kfree(sa);
4868drop_write:
4869	mnt_drop_write_file(file);
4870	return ret;
4871}
4872
4873static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4874{
4875	struct inode *inode = file_inode(file);
 
4876	struct btrfs_root *root = BTRFS_I(inode)->root;
4877	struct btrfs_ioctl_qgroup_limit_args *sa;
4878	struct btrfs_trans_handle *trans;
4879	int ret;
4880	int err;
4881	u64 qgroupid;
4882
4883	if (!capable(CAP_SYS_ADMIN))
4884		return -EPERM;
4885
4886	ret = mnt_want_write_file(file);
4887	if (ret)
4888		return ret;
4889
4890	sa = memdup_user(arg, sizeof(*sa));
4891	if (IS_ERR(sa)) {
4892		ret = PTR_ERR(sa);
4893		goto drop_write;
4894	}
4895
4896	trans = btrfs_join_transaction(root);
4897	if (IS_ERR(trans)) {
4898		ret = PTR_ERR(trans);
4899		goto out;
4900	}
4901
4902	qgroupid = sa->qgroupid;
4903	if (!qgroupid) {
4904		/* take the current subvol as qgroup */
4905		qgroupid = root->root_key.objectid;
4906	}
4907
4908	ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4909
4910	err = btrfs_end_transaction(trans);
4911	if (err && !ret)
4912		ret = err;
4913
4914out:
4915	kfree(sa);
4916drop_write:
4917	mnt_drop_write_file(file);
4918	return ret;
4919}
4920
4921static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4922{
4923	struct inode *inode = file_inode(file);
4924	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4925	struct btrfs_ioctl_quota_rescan_args *qsa;
4926	int ret;
4927
4928	if (!capable(CAP_SYS_ADMIN))
4929		return -EPERM;
4930
4931	ret = mnt_want_write_file(file);
4932	if (ret)
4933		return ret;
4934
4935	qsa = memdup_user(arg, sizeof(*qsa));
4936	if (IS_ERR(qsa)) {
4937		ret = PTR_ERR(qsa);
4938		goto drop_write;
4939	}
4940
4941	if (qsa->flags) {
4942		ret = -EINVAL;
4943		goto out;
4944	}
4945
4946	ret = btrfs_qgroup_rescan(fs_info);
4947
4948out:
4949	kfree(qsa);
4950drop_write:
4951	mnt_drop_write_file(file);
4952	return ret;
4953}
4954
4955static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
4956{
4957	struct inode *inode = file_inode(file);
4958	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4959	struct btrfs_ioctl_quota_rescan_args *qsa;
4960	int ret = 0;
4961
4962	if (!capable(CAP_SYS_ADMIN))
4963		return -EPERM;
4964
4965	qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4966	if (!qsa)
4967		return -ENOMEM;
4968
4969	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4970		qsa->flags = 1;
4971		qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4972	}
4973
4974	if (copy_to_user(arg, qsa, sizeof(*qsa)))
4975		ret = -EFAULT;
4976
4977	kfree(qsa);
4978	return ret;
4979}
4980
4981static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
4982{
4983	struct inode *inode = file_inode(file);
4984	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4985
4986	if (!capable(CAP_SYS_ADMIN))
4987		return -EPERM;
4988
4989	return btrfs_qgroup_wait_for_completion(fs_info, true);
4990}
4991
4992static long _btrfs_ioctl_set_received_subvol(struct file *file,
4993					    struct btrfs_ioctl_received_subvol_args *sa)
4994{
4995	struct inode *inode = file_inode(file);
4996	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4997	struct btrfs_root *root = BTRFS_I(inode)->root;
4998	struct btrfs_root_item *root_item = &root->root_item;
4999	struct btrfs_trans_handle *trans;
5000	struct timespec64 ct = current_time(inode);
5001	int ret = 0;
5002	int received_uuid_changed;
5003
5004	if (!inode_owner_or_capable(inode))
5005		return -EPERM;
5006
5007	ret = mnt_want_write_file(file);
5008	if (ret < 0)
5009		return ret;
5010
5011	down_write(&fs_info->subvol_sem);
5012
5013	if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5014		ret = -EINVAL;
5015		goto out;
5016	}
5017
5018	if (btrfs_root_readonly(root)) {
5019		ret = -EROFS;
5020		goto out;
5021	}
5022
5023	/*
5024	 * 1 - root item
5025	 * 2 - uuid items (received uuid + subvol uuid)
5026	 */
5027	trans = btrfs_start_transaction(root, 3);
5028	if (IS_ERR(trans)) {
5029		ret = PTR_ERR(trans);
5030		trans = NULL;
5031		goto out;
5032	}
5033
5034	sa->rtransid = trans->transid;
5035	sa->rtime.sec = ct.tv_sec;
5036	sa->rtime.nsec = ct.tv_nsec;
5037
5038	received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5039				       BTRFS_UUID_SIZE);
5040	if (received_uuid_changed &&
5041	    !btrfs_is_empty_uuid(root_item->received_uuid)) {
5042		ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
 
5043					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5044					  root->root_key.objectid);
5045		if (ret && ret != -ENOENT) {
5046		        btrfs_abort_transaction(trans, ret);
5047		        btrfs_end_transaction(trans);
5048		        goto out;
5049		}
5050	}
5051	memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5052	btrfs_set_root_stransid(root_item, sa->stransid);
5053	btrfs_set_root_rtransid(root_item, sa->rtransid);
5054	btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5055	btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5056	btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5057	btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5058
5059	ret = btrfs_update_root(trans, fs_info->tree_root,
5060				&root->root_key, &root->root_item);
5061	if (ret < 0) {
5062		btrfs_end_transaction(trans);
5063		goto out;
5064	}
5065	if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5066		ret = btrfs_uuid_tree_add(trans, sa->uuid,
5067					  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5068					  root->root_key.objectid);
5069		if (ret < 0 && ret != -EEXIST) {
5070			btrfs_abort_transaction(trans, ret);
5071			btrfs_end_transaction(trans);
5072			goto out;
5073		}
5074	}
5075	ret = btrfs_commit_transaction(trans);
5076out:
5077	up_write(&fs_info->subvol_sem);
5078	mnt_drop_write_file(file);
5079	return ret;
5080}
5081
5082#ifdef CONFIG_64BIT
5083static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5084						void __user *arg)
5085{
5086	struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5087	struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5088	int ret = 0;
5089
5090	args32 = memdup_user(arg, sizeof(*args32));
5091	if (IS_ERR(args32))
5092		return PTR_ERR(args32);
5093
5094	args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5095	if (!args64) {
5096		ret = -ENOMEM;
5097		goto out;
5098	}
5099
5100	memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5101	args64->stransid = args32->stransid;
5102	args64->rtransid = args32->rtransid;
5103	args64->stime.sec = args32->stime.sec;
5104	args64->stime.nsec = args32->stime.nsec;
5105	args64->rtime.sec = args32->rtime.sec;
5106	args64->rtime.nsec = args32->rtime.nsec;
5107	args64->flags = args32->flags;
5108
5109	ret = _btrfs_ioctl_set_received_subvol(file, args64);
5110	if (ret)
5111		goto out;
5112
5113	memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5114	args32->stransid = args64->stransid;
5115	args32->rtransid = args64->rtransid;
5116	args32->stime.sec = args64->stime.sec;
5117	args32->stime.nsec = args64->stime.nsec;
5118	args32->rtime.sec = args64->rtime.sec;
5119	args32->rtime.nsec = args64->rtime.nsec;
5120	args32->flags = args64->flags;
5121
5122	ret = copy_to_user(arg, args32, sizeof(*args32));
5123	if (ret)
5124		ret = -EFAULT;
5125
5126out:
5127	kfree(args32);
5128	kfree(args64);
5129	return ret;
5130}
5131#endif
5132
5133static long btrfs_ioctl_set_received_subvol(struct file *file,
5134					    void __user *arg)
5135{
5136	struct btrfs_ioctl_received_subvol_args *sa = NULL;
5137	int ret = 0;
5138
5139	sa = memdup_user(arg, sizeof(*sa));
5140	if (IS_ERR(sa))
5141		return PTR_ERR(sa);
5142
5143	ret = _btrfs_ioctl_set_received_subvol(file, sa);
5144
5145	if (ret)
5146		goto out;
5147
5148	ret = copy_to_user(arg, sa, sizeof(*sa));
5149	if (ret)
5150		ret = -EFAULT;
5151
5152out:
5153	kfree(sa);
5154	return ret;
5155}
5156
5157static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5158{
5159	struct inode *inode = file_inode(file);
5160	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5161	size_t len;
5162	int ret;
5163	char label[BTRFS_LABEL_SIZE];
5164
5165	spin_lock(&fs_info->super_lock);
5166	memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5167	spin_unlock(&fs_info->super_lock);
5168
5169	len = strnlen(label, BTRFS_LABEL_SIZE);
5170
5171	if (len == BTRFS_LABEL_SIZE) {
5172		btrfs_warn(fs_info,
5173			   "label is too long, return the first %zu bytes",
5174			   --len);
5175	}
5176
5177	ret = copy_to_user(arg, label, len);
5178
5179	return ret ? -EFAULT : 0;
5180}
5181
5182static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5183{
5184	struct inode *inode = file_inode(file);
5185	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5186	struct btrfs_root *root = BTRFS_I(inode)->root;
5187	struct btrfs_super_block *super_block = fs_info->super_copy;
5188	struct btrfs_trans_handle *trans;
5189	char label[BTRFS_LABEL_SIZE];
5190	int ret;
5191
5192	if (!capable(CAP_SYS_ADMIN))
5193		return -EPERM;
5194
5195	if (copy_from_user(label, arg, sizeof(label)))
5196		return -EFAULT;
5197
5198	if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5199		btrfs_err(fs_info,
5200			  "unable to set label with more than %d bytes",
5201			  BTRFS_LABEL_SIZE - 1);
5202		return -EINVAL;
5203	}
5204
5205	ret = mnt_want_write_file(file);
5206	if (ret)
5207		return ret;
5208
5209	trans = btrfs_start_transaction(root, 0);
5210	if (IS_ERR(trans)) {
5211		ret = PTR_ERR(trans);
5212		goto out_unlock;
5213	}
5214
5215	spin_lock(&fs_info->super_lock);
5216	strcpy(super_block->label, label);
5217	spin_unlock(&fs_info->super_lock);
5218	ret = btrfs_commit_transaction(trans);
5219
5220out_unlock:
5221	mnt_drop_write_file(file);
5222	return ret;
5223}
5224
5225#define INIT_FEATURE_FLAGS(suffix) \
5226	{ .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5227	  .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5228	  .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5229
5230int btrfs_ioctl_get_supported_features(void __user *arg)
5231{
5232	static const struct btrfs_ioctl_feature_flags features[3] = {
5233		INIT_FEATURE_FLAGS(SUPP),
5234		INIT_FEATURE_FLAGS(SAFE_SET),
5235		INIT_FEATURE_FLAGS(SAFE_CLEAR)
5236	};
5237
5238	if (copy_to_user(arg, &features, sizeof(features)))
5239		return -EFAULT;
5240
5241	return 0;
5242}
5243
5244static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5245{
5246	struct inode *inode = file_inode(file);
5247	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5248	struct btrfs_super_block *super_block = fs_info->super_copy;
5249	struct btrfs_ioctl_feature_flags features;
5250
5251	features.compat_flags = btrfs_super_compat_flags(super_block);
5252	features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5253	features.incompat_flags = btrfs_super_incompat_flags(super_block);
5254
5255	if (copy_to_user(arg, &features, sizeof(features)))
5256		return -EFAULT;
5257
5258	return 0;
5259}
5260
5261static int check_feature_bits(struct btrfs_fs_info *fs_info,
5262			      enum btrfs_feature_set set,
5263			      u64 change_mask, u64 flags, u64 supported_flags,
5264			      u64 safe_set, u64 safe_clear)
5265{
5266	const char *type = btrfs_feature_set_name(set);
5267	char *names;
5268	u64 disallowed, unsupported;
5269	u64 set_mask = flags & change_mask;
5270	u64 clear_mask = ~flags & change_mask;
5271
5272	unsupported = set_mask & ~supported_flags;
5273	if (unsupported) {
5274		names = btrfs_printable_features(set, unsupported);
5275		if (names) {
5276			btrfs_warn(fs_info,
5277				   "this kernel does not support the %s feature bit%s",
5278				   names, strchr(names, ',') ? "s" : "");
5279			kfree(names);
5280		} else
5281			btrfs_warn(fs_info,
5282				   "this kernel does not support %s bits 0x%llx",
5283				   type, unsupported);
5284		return -EOPNOTSUPP;
5285	}
5286
5287	disallowed = set_mask & ~safe_set;
5288	if (disallowed) {
5289		names = btrfs_printable_features(set, disallowed);
5290		if (names) {
5291			btrfs_warn(fs_info,
5292				   "can't set the %s feature bit%s while mounted",
5293				   names, strchr(names, ',') ? "s" : "");
5294			kfree(names);
5295		} else
5296			btrfs_warn(fs_info,
5297				   "can't set %s bits 0x%llx while mounted",
5298				   type, disallowed);
5299		return -EPERM;
5300	}
5301
5302	disallowed = clear_mask & ~safe_clear;
5303	if (disallowed) {
5304		names = btrfs_printable_features(set, disallowed);
5305		if (names) {
5306			btrfs_warn(fs_info,
5307				   "can't clear the %s feature bit%s while mounted",
5308				   names, strchr(names, ',') ? "s" : "");
5309			kfree(names);
5310		} else
5311			btrfs_warn(fs_info,
5312				   "can't clear %s bits 0x%llx while mounted",
5313				   type, disallowed);
5314		return -EPERM;
5315	}
5316
5317	return 0;
5318}
5319
5320#define check_feature(fs_info, change_mask, flags, mask_base)	\
5321check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,	\
5322		   BTRFS_FEATURE_ ## mask_base ## _SUPP,	\
5323		   BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,	\
5324		   BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5325
5326static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5327{
5328	struct inode *inode = file_inode(file);
5329	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5330	struct btrfs_root *root = BTRFS_I(inode)->root;
5331	struct btrfs_super_block *super_block = fs_info->super_copy;
5332	struct btrfs_ioctl_feature_flags flags[2];
5333	struct btrfs_trans_handle *trans;
5334	u64 newflags;
5335	int ret;
5336
5337	if (!capable(CAP_SYS_ADMIN))
5338		return -EPERM;
5339
5340	if (copy_from_user(flags, arg, sizeof(flags)))
5341		return -EFAULT;
5342
5343	/* Nothing to do */
5344	if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5345	    !flags[0].incompat_flags)
5346		return 0;
5347
5348	ret = check_feature(fs_info, flags[0].compat_flags,
5349			    flags[1].compat_flags, COMPAT);
5350	if (ret)
5351		return ret;
5352
5353	ret = check_feature(fs_info, flags[0].compat_ro_flags,
5354			    flags[1].compat_ro_flags, COMPAT_RO);
5355	if (ret)
5356		return ret;
5357
5358	ret = check_feature(fs_info, flags[0].incompat_flags,
5359			    flags[1].incompat_flags, INCOMPAT);
5360	if (ret)
5361		return ret;
5362
5363	ret = mnt_want_write_file(file);
5364	if (ret)
5365		return ret;
5366
5367	trans = btrfs_start_transaction(root, 0);
5368	if (IS_ERR(trans)) {
5369		ret = PTR_ERR(trans);
5370		goto out_drop_write;
5371	}
5372
5373	spin_lock(&fs_info->super_lock);
5374	newflags = btrfs_super_compat_flags(super_block);
5375	newflags |= flags[0].compat_flags & flags[1].compat_flags;
5376	newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5377	btrfs_set_super_compat_flags(super_block, newflags);
5378
5379	newflags = btrfs_super_compat_ro_flags(super_block);
5380	newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5381	newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5382	btrfs_set_super_compat_ro_flags(super_block, newflags);
5383
5384	newflags = btrfs_super_incompat_flags(super_block);
5385	newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5386	newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5387	btrfs_set_super_incompat_flags(super_block, newflags);
5388	spin_unlock(&fs_info->super_lock);
5389
5390	ret = btrfs_commit_transaction(trans);
5391out_drop_write:
5392	mnt_drop_write_file(file);
5393
5394	return ret;
5395}
5396
5397static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5398{
5399	struct btrfs_ioctl_send_args *arg;
5400	int ret;
5401
5402	if (compat) {
5403#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5404		struct btrfs_ioctl_send_args_32 args32;
5405
5406		ret = copy_from_user(&args32, argp, sizeof(args32));
5407		if (ret)
5408			return -EFAULT;
5409		arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5410		if (!arg)
5411			return -ENOMEM;
5412		arg->send_fd = args32.send_fd;
5413		arg->clone_sources_count = args32.clone_sources_count;
5414		arg->clone_sources = compat_ptr(args32.clone_sources);
5415		arg->parent_root = args32.parent_root;
5416		arg->flags = args32.flags;
5417		memcpy(arg->reserved, args32.reserved,
5418		       sizeof(args32.reserved));
5419#else
5420		return -ENOTTY;
5421#endif
5422	} else {
5423		arg = memdup_user(argp, sizeof(*arg));
5424		if (IS_ERR(arg))
5425			return PTR_ERR(arg);
5426	}
5427	ret = btrfs_ioctl_send(file, arg);
5428	kfree(arg);
5429	return ret;
5430}
5431
5432long btrfs_ioctl(struct file *file, unsigned int
5433		cmd, unsigned long arg)
5434{
5435	struct inode *inode = file_inode(file);
5436	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5437	struct btrfs_root *root = BTRFS_I(inode)->root;
5438	void __user *argp = (void __user *)arg;
5439
5440	switch (cmd) {
5441	case FS_IOC_GETFLAGS:
5442		return btrfs_ioctl_getflags(file, argp);
5443	case FS_IOC_SETFLAGS:
5444		return btrfs_ioctl_setflags(file, argp);
5445	case FS_IOC_GETVERSION:
5446		return btrfs_ioctl_getversion(file, argp);
5447	case FS_IOC_GETFSLABEL:
5448		return btrfs_ioctl_get_fslabel(file, argp);
5449	case FS_IOC_SETFSLABEL:
5450		return btrfs_ioctl_set_fslabel(file, argp);
5451	case FITRIM:
5452		return btrfs_ioctl_fitrim(file, argp);
5453	case BTRFS_IOC_SNAP_CREATE:
5454		return btrfs_ioctl_snap_create(file, argp, 0);
5455	case BTRFS_IOC_SNAP_CREATE_V2:
5456		return btrfs_ioctl_snap_create_v2(file, argp, 0);
5457	case BTRFS_IOC_SUBVOL_CREATE:
5458		return btrfs_ioctl_snap_create(file, argp, 1);
5459	case BTRFS_IOC_SUBVOL_CREATE_V2:
5460		return btrfs_ioctl_snap_create_v2(file, argp, 1);
5461	case BTRFS_IOC_SNAP_DESTROY:
5462		return btrfs_ioctl_snap_destroy(file, argp);
5463	case BTRFS_IOC_SUBVOL_GETFLAGS:
5464		return btrfs_ioctl_subvol_getflags(file, argp);
5465	case BTRFS_IOC_SUBVOL_SETFLAGS:
5466		return btrfs_ioctl_subvol_setflags(file, argp);
5467	case BTRFS_IOC_DEFAULT_SUBVOL:
5468		return btrfs_ioctl_default_subvol(file, argp);
5469	case BTRFS_IOC_DEFRAG:
5470		return btrfs_ioctl_defrag(file, NULL);
5471	case BTRFS_IOC_DEFRAG_RANGE:
5472		return btrfs_ioctl_defrag(file, argp);
5473	case BTRFS_IOC_RESIZE:
5474		return btrfs_ioctl_resize(file, argp);
5475	case BTRFS_IOC_ADD_DEV:
5476		return btrfs_ioctl_add_dev(fs_info, argp);
5477	case BTRFS_IOC_RM_DEV:
5478		return btrfs_ioctl_rm_dev(file, argp);
5479	case BTRFS_IOC_RM_DEV_V2:
5480		return btrfs_ioctl_rm_dev_v2(file, argp);
5481	case BTRFS_IOC_FS_INFO:
5482		return btrfs_ioctl_fs_info(fs_info, argp);
5483	case BTRFS_IOC_DEV_INFO:
5484		return btrfs_ioctl_dev_info(fs_info, argp);
5485	case BTRFS_IOC_BALANCE:
5486		return btrfs_ioctl_balance(file, NULL);
5487	case BTRFS_IOC_TREE_SEARCH:
5488		return btrfs_ioctl_tree_search(file, argp);
5489	case BTRFS_IOC_TREE_SEARCH_V2:
5490		return btrfs_ioctl_tree_search_v2(file, argp);
5491	case BTRFS_IOC_INO_LOOKUP:
5492		return btrfs_ioctl_ino_lookup(file, argp);
5493	case BTRFS_IOC_INO_PATHS:
5494		return btrfs_ioctl_ino_to_path(root, argp);
5495	case BTRFS_IOC_LOGICAL_INO:
5496		return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5497	case BTRFS_IOC_LOGICAL_INO_V2:
5498		return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5499	case BTRFS_IOC_SPACE_INFO:
5500		return btrfs_ioctl_space_info(fs_info, argp);
5501	case BTRFS_IOC_SYNC: {
5502		int ret;
5503
5504		ret = btrfs_start_delalloc_roots(fs_info, -1);
5505		if (ret)
5506			return ret;
5507		ret = btrfs_sync_fs(inode->i_sb, 1);
5508		/*
5509		 * The transaction thread may want to do more work,
5510		 * namely it pokes the cleaner kthread that will start
5511		 * processing uncleaned subvols.
5512		 */
5513		wake_up_process(fs_info->transaction_kthread);
5514		return ret;
5515	}
5516	case BTRFS_IOC_START_SYNC:
5517		return btrfs_ioctl_start_sync(root, argp);
5518	case BTRFS_IOC_WAIT_SYNC:
5519		return btrfs_ioctl_wait_sync(fs_info, argp);
5520	case BTRFS_IOC_SCRUB:
5521		return btrfs_ioctl_scrub(file, argp);
5522	case BTRFS_IOC_SCRUB_CANCEL:
5523		return btrfs_ioctl_scrub_cancel(fs_info);
5524	case BTRFS_IOC_SCRUB_PROGRESS:
5525		return btrfs_ioctl_scrub_progress(fs_info, argp);
5526	case BTRFS_IOC_BALANCE_V2:
5527		return btrfs_ioctl_balance(file, argp);
5528	case BTRFS_IOC_BALANCE_CTL:
5529		return btrfs_ioctl_balance_ctl(fs_info, arg);
5530	case BTRFS_IOC_BALANCE_PROGRESS:
5531		return btrfs_ioctl_balance_progress(fs_info, argp);
5532	case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5533		return btrfs_ioctl_set_received_subvol(file, argp);
5534#ifdef CONFIG_64BIT
5535	case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5536		return btrfs_ioctl_set_received_subvol_32(file, argp);
5537#endif
5538	case BTRFS_IOC_SEND:
5539		return _btrfs_ioctl_send(file, argp, false);
5540#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5541	case BTRFS_IOC_SEND_32:
5542		return _btrfs_ioctl_send(file, argp, true);
5543#endif
5544	case BTRFS_IOC_GET_DEV_STATS:
5545		return btrfs_ioctl_get_dev_stats(fs_info, argp);
5546	case BTRFS_IOC_QUOTA_CTL:
5547		return btrfs_ioctl_quota_ctl(file, argp);
5548	case BTRFS_IOC_QGROUP_ASSIGN:
5549		return btrfs_ioctl_qgroup_assign(file, argp);
5550	case BTRFS_IOC_QGROUP_CREATE:
5551		return btrfs_ioctl_qgroup_create(file, argp);
5552	case BTRFS_IOC_QGROUP_LIMIT:
5553		return btrfs_ioctl_qgroup_limit(file, argp);
5554	case BTRFS_IOC_QUOTA_RESCAN:
5555		return btrfs_ioctl_quota_rescan(file, argp);
5556	case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5557		return btrfs_ioctl_quota_rescan_status(file, argp);
5558	case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5559		return btrfs_ioctl_quota_rescan_wait(file, argp);
5560	case BTRFS_IOC_DEV_REPLACE:
5561		return btrfs_ioctl_dev_replace(fs_info, argp);
 
 
 
 
5562	case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5563		return btrfs_ioctl_get_supported_features(argp);
5564	case BTRFS_IOC_GET_FEATURES:
5565		return btrfs_ioctl_get_features(file, argp);
5566	case BTRFS_IOC_SET_FEATURES:
5567		return btrfs_ioctl_set_features(file, argp);
5568	case FS_IOC_FSGETXATTR:
5569		return btrfs_ioctl_fsgetxattr(file, argp);
5570	case FS_IOC_FSSETXATTR:
5571		return btrfs_ioctl_fssetxattr(file, argp);
5572	case BTRFS_IOC_GET_SUBVOL_INFO:
5573		return btrfs_ioctl_get_subvol_info(file, argp);
5574	case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5575		return btrfs_ioctl_get_subvol_rootref(file, argp);
5576	case BTRFS_IOC_INO_LOOKUP_USER:
5577		return btrfs_ioctl_ino_lookup_user(file, argp);
5578	}
5579
5580	return -ENOTTY;
5581}
5582
5583#ifdef CONFIG_COMPAT
5584long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5585{
5586	/*
5587	 * These all access 32-bit values anyway so no further
5588	 * handling is necessary.
5589	 */
5590	switch (cmd) {
5591	case FS_IOC32_GETFLAGS:
5592		cmd = FS_IOC_GETFLAGS;
5593		break;
5594	case FS_IOC32_SETFLAGS:
5595		cmd = FS_IOC_SETFLAGS;
5596		break;
5597	case FS_IOC32_GETVERSION:
5598		cmd = FS_IOC_GETVERSION;
5599		break;
5600	}
5601
5602	return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5603}
5604#endif