fs/btrfs/ioctl.c

   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 <linux/fileattr.h>
  30 #include <linux/fsverity.h>
  31 #include <linux/sched/xacct.h>
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "export.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "print-tree.h"
  38 #include "volumes.h"
  39 #include "locking.h"
  40 #include "backref.h"
  41 #include "rcu-string.h"
  42 #include "send.h"
  43 #include "dev-replace.h"
  44 #include "props.h"
  45 #include "sysfs.h"
  46 #include "qgroup.h"
  47 #include "tree-log.h"
  48 #include "compression.h"
  49 #include "space-info.h"
  50 #include "delalloc-space.h"
  51 #include "block-group.h"
  52 #include "subpage.h"
  53 #include "fs.h"
  54 #include "accessors.h"
  55 #include "extent-tree.h"
  56 #include "root-tree.h"
  57 #include "defrag.h"
  58 #include "dir-item.h"
  59 #include "uuid-tree.h"
  60 #include "ioctl.h"
  61 #include "file.h"
  62 #include "scrub.h"
  63 #include "super.h"
  64
  65 #ifdef CONFIG_64BIT
  66 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  67  * structures are incorrect, as the timespec structure from userspace
  68  * is 4 bytes too small. We define these alternatives here to teach
  69  * the kernel about the 32-bit struct packing.
  70  */
  71 struct btrfs_ioctl_timespec_32 {
  72         __u64 sec;
  73         __u32 nsec;
  74 } __attribute__ ((__packed__));
  75
  76 struct btrfs_ioctl_received_subvol_args_32 {
  77         char    uuid[BTRFS_UUID_SIZE];  /* in */
  78         __u64   stransid;               /* in */
  79         __u64   rtransid;               /* out */
  80         struct btrfs_ioctl_timespec_32 stime; /* in */
  81         struct btrfs_ioctl_timespec_32 rtime; /* out */
  82         __u64   flags;                  /* in */
  83         __u64   reserved[16];           /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  87                                 struct btrfs_ioctl_received_subvol_args_32)
  88 #endif
  89
  90 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  91 struct btrfs_ioctl_send_args_32 {
  92         __s64 send_fd;                  /* in */
  93         __u64 clone_sources_count;      /* in */
  94         compat_uptr_t clone_sources;    /* in */
  95         __u64 parent_root;              /* in */
  96         __u64 flags;                    /* in */
  97         __u32 version;                  /* in */
  98         __u8  reserved[28];             /* in */
  99 } __attribute__ ((__packed__));
 100
 101 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
 102                                struct btrfs_ioctl_send_args_32)
 103
 104 struct btrfs_ioctl_encoded_io_args_32 {
 105         compat_uptr_t iov;
 106         compat_ulong_t iovcnt;
 107         __s64 offset;
 108         __u64 flags;
 109         __u64 len;
 110         __u64 unencoded_len;
 111         __u64 unencoded_offset;
 112         __u32 compression;
 113         __u32 encryption;
 114         __u8 reserved[64];
 115 };
 116
 117 #define BTRFS_IOC_ENCODED_READ_32 _IOR(BTRFS_IOCTL_MAGIC, 64, \
 118                                        struct btrfs_ioctl_encoded_io_args_32)
 119 #define BTRFS_IOC_ENCODED_WRITE_32 _IOW(BTRFS_IOCTL_MAGIC, 64, \
 120                                         struct btrfs_ioctl_encoded_io_args_32)
 121 #endif
 122
 123 /* Mask out flags that are inappropriate for the given type of inode. */
 124 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
 125                 unsigned int flags)
 126 {
 127         if (S_ISDIR(inode->i_mode))
 128                 return flags;
 129         else if (S_ISREG(inode->i_mode))
 130                 return flags & ~FS_DIRSYNC_FL;
 131         else
 132                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 133 }
 134
 135 /*
 136  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 137  * ioctl.
 138  */
 139 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
 140 {
 141         unsigned int iflags = 0;
 142         u32 flags = binode->flags;
 143         u32 ro_flags = binode->ro_flags;
 144
 145         if (flags & BTRFS_INODE_SYNC)
 146                 iflags |= FS_SYNC_FL;
 147         if (flags & BTRFS_INODE_IMMUTABLE)
 148                 iflags |= FS_IMMUTABLE_FL;
 149         if (flags & BTRFS_INODE_APPEND)
 150                 iflags |= FS_APPEND_FL;
 151         if (flags & BTRFS_INODE_NODUMP)
 152                 iflags |= FS_NODUMP_FL;
 153         if (flags & BTRFS_INODE_NOATIME)
 154                 iflags |= FS_NOATIME_FL;
 155         if (flags & BTRFS_INODE_DIRSYNC)
 156                 iflags |= FS_DIRSYNC_FL;
 157         if (flags & BTRFS_INODE_NODATACOW)
 158                 iflags |= FS_NOCOW_FL;
 159         if (ro_flags & BTRFS_INODE_RO_VERITY)
 160                 iflags |= FS_VERITY_FL;
 161
 162         if (flags & BTRFS_INODE_NOCOMPRESS)
 163                 iflags |= FS_NOCOMP_FL;
 164         else if (flags & BTRFS_INODE_COMPRESS)
 165                 iflags |= FS_COMPR_FL;
 166
 167         return iflags;
 168 }
 169
 170 /*
 171  * Update inode->i_flags based on the btrfs internal flags.
 172  */
 173 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 174 {
 175         struct btrfs_inode *binode = BTRFS_I(inode);
 176         unsigned int new_fl = 0;
 177
 178         if (binode->flags & BTRFS_INODE_SYNC)
 179                 new_fl |= S_SYNC;
 180         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 181                 new_fl |= S_IMMUTABLE;
 182         if (binode->flags & BTRFS_INODE_APPEND)
 183                 new_fl |= S_APPEND;
 184         if (binode->flags & BTRFS_INODE_NOATIME)
 185                 new_fl |= S_NOATIME;
 186         if (binode->flags & BTRFS_INODE_DIRSYNC)
 187                 new_fl |= S_DIRSYNC;
 188         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
 189                 new_fl |= S_VERITY;
 190
 191         set_mask_bits(&inode->i_flags,
 192                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
 193                       S_VERITY, new_fl);
 194 }
 195
 196 /*
 197  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 198  * the old and new flags are not conflicting
 199  */
 200 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 201 {
 202         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 203                       FS_NOATIME_FL | FS_NODUMP_FL | \
 204                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 205                       FS_NOCOMP_FL | FS_COMPR_FL |
 206                       FS_NOCOW_FL))
 207                 return -EOPNOTSUPP;
 208
 209         /* COMPR and NOCOMP on new/old are valid */
 210         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 211                 return -EINVAL;
 212
 213         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 214                 return -EINVAL;
 215
 216         /* NOCOW and compression options are mutually exclusive */
 217         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 218                 return -EINVAL;
 219         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 220                 return -EINVAL;
 221
 222         return 0;
 223 }
 224
 225 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 226                                     unsigned int flags)
 227 {
 228         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 229                 return -EPERM;
 230
 231         return 0;
 232 }
 233
 234 /*
 235  * Set flags/xflags from the internal inode flags. The remaining items of
 236  * fsxattr are zeroed.
 237  */
 238 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 239 {
 240         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
 241
 242         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
 243         return 0;
 244 }
 245
 246 int btrfs_fileattr_set(struct mnt_idmap *idmap,
 247                        struct dentry *dentry, struct fileattr *fa)
 248 {
 249         struct inode *inode = d_inode(dentry);
 250         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 251         struct btrfs_inode *binode = BTRFS_I(inode);
 252         struct btrfs_root *root = binode->root;
 253         struct btrfs_trans_handle *trans;
 254         unsigned int fsflags, old_fsflags;
 255         int ret;
 256         const char *comp = NULL;
 257         u32 binode_flags;
 258
 259         if (btrfs_root_readonly(root))
 260                 return -EROFS;
 261
 262         if (fileattr_has_fsx(fa))
 263                 return -EOPNOTSUPP;
 264
 265         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
 266         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
 267         ret = check_fsflags(old_fsflags, fsflags);
 268         if (ret)
 269                 return ret;
 270
 271         ret = check_fsflags_compatible(fs_info, fsflags);
 272         if (ret)
 273                 return ret;
 274
 275         binode_flags = binode->flags;
 276         if (fsflags & FS_SYNC_FL)
 277                 binode_flags |= BTRFS_INODE_SYNC;
 278         else
 279                 binode_flags &= ~BTRFS_INODE_SYNC;
 280         if (fsflags & FS_IMMUTABLE_FL)
 281                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 282         else
 283                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 284         if (fsflags & FS_APPEND_FL)
 285                 binode_flags |= BTRFS_INODE_APPEND;
 286         else
 287                 binode_flags &= ~BTRFS_INODE_APPEND;
 288         if (fsflags & FS_NODUMP_FL)
 289                 binode_flags |= BTRFS_INODE_NODUMP;
 290         else
 291                 binode_flags &= ~BTRFS_INODE_NODUMP;
 292         if (fsflags & FS_NOATIME_FL)
 293                 binode_flags |= BTRFS_INODE_NOATIME;
 294         else
 295                 binode_flags &= ~BTRFS_INODE_NOATIME;
 296
 297         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
 298         if (!fa->flags_valid) {
 299                 /* 1 item for the inode */
 300                 trans = btrfs_start_transaction(root, 1);
 301                 if (IS_ERR(trans))
 302                         return PTR_ERR(trans);
 303                 goto update_flags;
 304         }
 305
 306         if (fsflags & FS_DIRSYNC_FL)
 307                 binode_flags |= BTRFS_INODE_DIRSYNC;
 308         else
 309                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 310         if (fsflags & FS_NOCOW_FL) {
 311                 if (S_ISREG(inode->i_mode)) {
 312                         /*
 313                          * It's safe to turn csums off here, no extents exist.
 314                          * Otherwise we want the flag to reflect the real COW
 315                          * status of the file and will not set it.
 316                          */
 317                         if (inode->i_size == 0)
 318                                 binode_flags |= BTRFS_INODE_NODATACOW |
 319                                                 BTRFS_INODE_NODATASUM;
 320                 } else {
 321                         binode_flags |= BTRFS_INODE_NODATACOW;
 322                 }
 323         } else {
 324                 /*
 325                  * Revert back under same assumptions as above
 326                  */
 327                 if (S_ISREG(inode->i_mode)) {
 328                         if (inode->i_size == 0)
 329                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 330                                                   BTRFS_INODE_NODATASUM);
 331                 } else {
 332                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 333                 }
 334         }
 335
 336         /*
 337          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 338          * flag may be changed automatically if compression code won't make
 339          * things smaller.
 340          */
 341         if (fsflags & FS_NOCOMP_FL) {
 342                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 343                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 344         } else if (fsflags & FS_COMPR_FL) {
 345
 346                 if (IS_SWAPFILE(inode))
 347                         return -ETXTBSY;
 348
 349                 binode_flags |= BTRFS_INODE_COMPRESS;
 350                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 351
 352                 comp = btrfs_compress_type2str(fs_info->compress_type);
 353                 if (!comp || comp[0] == 0)
 354                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 355         } else {
 356                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 357         }
 358
 359         /*
 360          * 1 for inode item
 361          * 2 for properties
 362          */
 363         trans = btrfs_start_transaction(root, 3);
 364         if (IS_ERR(trans))
 365                 return PTR_ERR(trans);
 366
 367         if (comp) {
 368                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 369                                      strlen(comp), 0);
 370                 if (ret) {
 371                         btrfs_abort_transaction(trans, ret);
 372                         goto out_end_trans;
 373                 }
 374         } else {
 375                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 376                                      0, 0);
 377                 if (ret && ret != -ENODATA) {
 378                         btrfs_abort_transaction(trans, ret);
 379                         goto out_end_trans;
 380                 }
 381         }
 382
 383 update_flags:
 384         binode->flags = binode_flags;
 385         btrfs_sync_inode_flags_to_i_flags(inode);
 386         inode_inc_iversion(inode);
 387         inode_set_ctime_current(inode);
 388         ret = btrfs_update_inode(trans, BTRFS_I(inode));
 389
 390  out_end_trans:
 391         btrfs_end_transaction(trans);
 392         return ret;
 393 }
 394
 395 /*
 396  * Start exclusive operation @type, return true on success
 397  */
 398 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 399                         enum btrfs_exclusive_operation type)
 400 {
 401         bool ret = false;
 402
 403         spin_lock(&fs_info->super_lock);
 404         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
 405                 fs_info->exclusive_operation = type;
 406                 ret = true;
 407         }
 408         spin_unlock(&fs_info->super_lock);
 409
 410         return ret;
 411 }
 412
 413 /*
 414  * Conditionally allow to enter the exclusive operation in case it's compatible
 415  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
 416  * btrfs_exclop_finish.
 417  *
 418  * Compatibility:
 419  * - the same type is already running
 420  * - when trying to add a device and balance has been paused
 421  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
 422  *   must check the condition first that would allow none -> @type
 423  */
 424 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
 425                                  enum btrfs_exclusive_operation type)
 426 {
 427         spin_lock(&fs_info->super_lock);
 428         if (fs_info->exclusive_operation == type ||
 429             (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED &&
 430              type == BTRFS_EXCLOP_DEV_ADD))
 431                 return true;
 432
 433         spin_unlock(&fs_info->super_lock);
 434         return false;
 435 }
 436
 437 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
 438 {
 439         spin_unlock(&fs_info->super_lock);
 440 }
 441
 442 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 443 {
 444         spin_lock(&fs_info->super_lock);
 445         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 446         spin_unlock(&fs_info->super_lock);
 447         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 448 }
 449
 450 void btrfs_exclop_balance(struct btrfs_fs_info *fs_info,
 451                           enum btrfs_exclusive_operation op)
 452 {
 453         switch (op) {
 454         case BTRFS_EXCLOP_BALANCE_PAUSED:
 455                 spin_lock(&fs_info->super_lock);
 456                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE ||
 457                        fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD ||
 458                        fs_info->exclusive_operation == BTRFS_EXCLOP_NONE ||
 459                        fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 460                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED;
 461                 spin_unlock(&fs_info->super_lock);
 462                 break;
 463         case BTRFS_EXCLOP_BALANCE:
 464                 spin_lock(&fs_info->super_lock);
 465                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 466                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
 467                 spin_unlock(&fs_info->super_lock);
 468                 break;
 469         default:
 470                 btrfs_warn(fs_info,
 471                         "invalid exclop balance operation %d requested", op);
 472         }
 473 }
 474
 475 static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg)
 476 {
 477         return put_user(inode->i_generation, arg);
 478 }
 479
 480 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 481                                         void __user *arg)
 482 {
 483         struct btrfs_device *device;
 484         struct fstrim_range range;
 485         u64 minlen = ULLONG_MAX;
 486         u64 num_devices = 0;
 487         int ret;
 488
 489         if (!capable(CAP_SYS_ADMIN))
 490                 return -EPERM;
 491
 492         /*
 493          * btrfs_trim_block_group() depends on space cache, which is not
 494          * available in zoned filesystem. So, disallow fitrim on a zoned
 495          * filesystem for now.
 496          */
 497         if (btrfs_is_zoned(fs_info))
 498                 return -EOPNOTSUPP;
 499
 500         /*
 501          * If the fs is mounted with nologreplay, which requires it to be
 502          * mounted in RO mode as well, we can not allow discard on free space
 503          * inside block groups, because log trees refer to extents that are not
 504          * pinned in a block group's free space cache (pinning the extents is
 505          * precisely the first phase of replaying a log tree).
 506          */
 507         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 508                 return -EROFS;
 509
 510         rcu_read_lock();
 511         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 512                                 dev_list) {
 513                 if (!device->bdev || !bdev_max_discard_sectors(device->bdev))
 514                         continue;
 515                 num_devices++;
 516                 minlen = min_t(u64, bdev_discard_granularity(device->bdev),
 517                                     minlen);
 518         }
 519         rcu_read_unlock();
 520
 521         if (!num_devices)
 522                 return -EOPNOTSUPP;
 523         if (copy_from_user(&range, arg, sizeof(range)))
 524                 return -EFAULT;
 525
 526         /*
 527          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 528          * block group is in the logical address space, which can be any
 529          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 530          */
 531         if (range.len < fs_info->sb->s_blocksize)
 532                 return -EINVAL;
 533
 534         range.minlen = max(range.minlen, minlen);
 535         ret = btrfs_trim_fs(fs_info, &range);
 536         if (ret < 0)
 537                 return ret;
 538
 539         if (copy_to_user(arg, &range, sizeof(range)))
 540                 return -EFAULT;
 541
 542         return 0;
 543 }
 544
 545 int __pure btrfs_is_empty_uuid(u8 *uuid)
 546 {
 547         int i;
 548
 549         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 550                 if (uuid[i])
 551                         return 0;
 552         }
 553         return 1;
 554 }
 555
 556 /*
 557  * Calculate the number of transaction items to reserve for creating a subvolume
 558  * or snapshot, not including the inode, directory entries, or parent directory.
 559  */
 560 static unsigned int create_subvol_num_items(struct btrfs_qgroup_inherit *inherit)
 561 {
 562         /*
 563          * 1 to add root block
 564          * 1 to add root item
 565          * 1 to add root ref
 566          * 1 to add root backref
 567          * 1 to add UUID item
 568          * 1 to add qgroup info
 569          * 1 to add qgroup limit
 570          *
 571          * Ideally the last two would only be accounted if qgroups are enabled,
 572          * but that can change between now and the time we would insert them.
 573          */
 574         unsigned int num_items = 7;
 575
 576         if (inherit) {
 577                 /* 2 to add qgroup relations for each inherited qgroup */
 578                 num_items += 2 * inherit->num_qgroups;
 579         }
 580         return num_items;
 581 }
 582
 583 static noinline int create_subvol(struct mnt_idmap *idmap,
 584                                   struct inode *dir, struct dentry *dentry,
 585                                   struct btrfs_qgroup_inherit *inherit)
 586 {
 587         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 588         struct btrfs_trans_handle *trans;
 589         struct btrfs_key key;
 590         struct btrfs_root_item *root_item;
 591         struct btrfs_inode_item *inode_item;
 592         struct extent_buffer *leaf;
 593         struct btrfs_root *root = BTRFS_I(dir)->root;
 594         struct btrfs_root *new_root;
 595         struct btrfs_block_rsv block_rsv;
 596         struct timespec64 cur_time = current_time(dir);
 597         struct btrfs_new_inode_args new_inode_args = {
 598                 .dir = dir,
 599                 .dentry = dentry,
 600                 .subvol = true,
 601         };
 602         unsigned int trans_num_items;
 603         int ret;
 604         dev_t anon_dev;
 605         u64 objectid;
 606
 607         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 608         if (!root_item)
 609                 return -ENOMEM;
 610
 611         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 612         if (ret)
 613                 goto out_root_item;
 614
 615         /*
 616          * Don't create subvolume whose level is not zero. Or qgroup will be
 617          * screwed up since it assumes subvolume qgroup's level to be 0.
 618          */
 619         if (btrfs_qgroup_level(objectid)) {
 620                 ret = -ENOSPC;
 621                 goto out_root_item;
 622         }
 623
 624         ret = get_anon_bdev(&anon_dev);
 625         if (ret < 0)
 626                 goto out_root_item;
 627
 628         new_inode_args.inode = btrfs_new_subvol_inode(idmap, dir);
 629         if (!new_inode_args.inode) {
 630                 ret = -ENOMEM;
 631                 goto out_anon_dev;
 632         }
 633         ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
 634         if (ret)
 635                 goto out_inode;
 636         trans_num_items += create_subvol_num_items(inherit);
 637
 638         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 639         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 640                                                trans_num_items, false);
 641         if (ret)
 642                 goto out_new_inode_args;
 643
 644         trans = btrfs_start_transaction(root, 0);
 645         if (IS_ERR(trans)) {
 646                 ret = PTR_ERR(trans);
 647                 btrfs_subvolume_release_metadata(root, &block_rsv);
 648                 goto out_new_inode_args;
 649         }
 650         trans->block_rsv = &block_rsv;
 651         trans->bytes_reserved = block_rsv.size;
 652         /* Tree log can't currently deal with an inode which is a new root. */
 653         btrfs_set_log_full_commit(trans);
 654
 655         ret = btrfs_qgroup_inherit(trans, 0, objectid, root->root_key.objectid, inherit);
 656         if (ret)
 657                 goto out;
 658
 659         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 660                                       0, BTRFS_NESTING_NORMAL);
 661         if (IS_ERR(leaf)) {
 662                 ret = PTR_ERR(leaf);
 663                 goto out;
 664         }
 665
 666         btrfs_mark_buffer_dirty(trans, leaf);
 667
 668         inode_item = &root_item->inode;
 669         btrfs_set_stack_inode_generation(inode_item, 1);
 670         btrfs_set_stack_inode_size(inode_item, 3);
 671         btrfs_set_stack_inode_nlink(inode_item, 1);
 672         btrfs_set_stack_inode_nbytes(inode_item,
 673                                      fs_info->nodesize);
 674         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 675
 676         btrfs_set_root_flags(root_item, 0);
 677         btrfs_set_root_limit(root_item, 0);
 678         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 679
 680         btrfs_set_root_bytenr(root_item, leaf->start);
 681         btrfs_set_root_generation(root_item, trans->transid);
 682         btrfs_set_root_level(root_item, 0);
 683         btrfs_set_root_refs(root_item, 1);
 684         btrfs_set_root_used(root_item, leaf->len);
 685         btrfs_set_root_last_snapshot(root_item, 0);
 686
 687         btrfs_set_root_generation_v2(root_item,
 688                         btrfs_root_generation(root_item));
 689         generate_random_guid(root_item->uuid);
 690         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 691         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 692         root_item->ctime = root_item->otime;
 693         btrfs_set_root_ctransid(root_item, trans->transid);
 694         btrfs_set_root_otransid(root_item, trans->transid);
 695
 696         btrfs_tree_unlock(leaf);
 697
 698         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 699
 700         key.objectid = objectid;
 701         key.offset = 0;
 702         key.type = BTRFS_ROOT_ITEM_KEY;
 703         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 704                                 root_item);
 705         if (ret) {
 706                 /*
 707                  * Since we don't abort the transaction in this case, free the
 708                  * tree block so that we don't leak space and leave the
 709                  * filesystem in an inconsistent state (an extent item in the
 710                  * extent tree with a backreference for a root that does not
 711                  * exists).
 712                  */
 713                 btrfs_tree_lock(leaf);
 714                 btrfs_clear_buffer_dirty(trans, leaf);
 715                 btrfs_tree_unlock(leaf);
 716                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 717                 free_extent_buffer(leaf);
 718                 goto out;
 719         }
 720
 721         free_extent_buffer(leaf);
 722         leaf = NULL;
 723
 724         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 725         if (IS_ERR(new_root)) {
 726                 ret = PTR_ERR(new_root);
 727                 btrfs_abort_transaction(trans, ret);
 728                 goto out;
 729         }
 730         /* anon_dev is owned by new_root now. */
 731         anon_dev = 0;
 732         BTRFS_I(new_inode_args.inode)->root = new_root;
 733         /* ... and new_root is owned by new_inode_args.inode now. */
 734
 735         ret = btrfs_record_root_in_trans(trans, new_root);
 736         if (ret) {
 737                 btrfs_abort_transaction(trans, ret);
 738                 goto out;
 739         }
 740
 741         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 742                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 743         if (ret) {
 744                 btrfs_abort_transaction(trans, ret);
 745                 goto out;
 746         }
 747
 748         ret = btrfs_create_new_inode(trans, &new_inode_args);
 749         if (ret) {
 750                 btrfs_abort_transaction(trans, ret);
 751                 goto out;
 752         }
 753
 754         d_instantiate_new(dentry, new_inode_args.inode);
 755         new_inode_args.inode = NULL;
 756
 757 out:
 758         trans->block_rsv = NULL;
 759         trans->bytes_reserved = 0;
 760         btrfs_subvolume_release_metadata(root, &block_rsv);
 761
 762         btrfs_end_transaction(trans);
 763 out_new_inode_args:
 764         btrfs_new_inode_args_destroy(&new_inode_args);
 765 out_inode:
 766         iput(new_inode_args.inode);
 767 out_anon_dev:
 768         if (anon_dev)
 769                 free_anon_bdev(anon_dev);
 770 out_root_item:
 771         kfree(root_item);
 772         return ret;
 773 }
 774
 775 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 776                            struct dentry *dentry, bool readonly,
 777                            struct btrfs_qgroup_inherit *inherit)
 778 {
 779         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 780         struct inode *inode;
 781         struct btrfs_pending_snapshot *pending_snapshot;
 782         unsigned int trans_num_items;
 783         struct btrfs_trans_handle *trans;
 784         int ret;
 785
 786         /* We do not support snapshotting right now. */
 787         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 788                 btrfs_warn(fs_info,
 789                            "extent tree v2 doesn't support snapshotting yet");
 790                 return -EOPNOTSUPP;
 791         }
 792
 793         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 794                 return -EINVAL;
 795
 796         if (atomic_read(&root->nr_swapfiles)) {
 797                 btrfs_warn(fs_info,
 798                            "cannot snapshot subvolume with active swapfile");
 799                 return -ETXTBSY;
 800         }
 801
 802         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 803         if (!pending_snapshot)
 804                 return -ENOMEM;
 805
 806         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 807         if (ret < 0)
 808                 goto free_pending;
 809         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 810                         GFP_KERNEL);
 811         pending_snapshot->path = btrfs_alloc_path();
 812         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 813                 ret = -ENOMEM;
 814                 goto free_pending;
 815         }
 816
 817         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 818                              BTRFS_BLOCK_RSV_TEMP);
 819         /*
 820          * 1 to add dir item
 821          * 1 to add dir index
 822          * 1 to update parent inode item
 823          */
 824         trans_num_items = create_subvol_num_items(inherit) + 3;
 825         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 826                                                &pending_snapshot->block_rsv,
 827                                                trans_num_items, false);
 828         if (ret)
 829                 goto free_pending;
 830
 831         pending_snapshot->dentry = dentry;
 832         pending_snapshot->root = root;
 833         pending_snapshot->readonly = readonly;
 834         pending_snapshot->dir = dir;
 835         pending_snapshot->inherit = inherit;
 836
 837         trans = btrfs_start_transaction(root, 0);
 838         if (IS_ERR(trans)) {
 839                 ret = PTR_ERR(trans);
 840                 goto fail;
 841         }
 842
 843         trans->pending_snapshot = pending_snapshot;
 844
 845         ret = btrfs_commit_transaction(trans);
 846         if (ret)
 847                 goto fail;
 848
 849         ret = pending_snapshot->error;
 850         if (ret)
 851                 goto fail;
 852
 853         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 854         if (ret)
 855                 goto fail;
 856
 857         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 858         if (IS_ERR(inode)) {
 859                 ret = PTR_ERR(inode);
 860                 goto fail;
 861         }
 862
 863         d_instantiate(dentry, inode);
 864         ret = 0;
 865         pending_snapshot->anon_dev = 0;
 866 fail:
 867         /* Prevent double freeing of anon_dev */
 868         if (ret && pending_snapshot->snap)
 869                 pending_snapshot->snap->anon_dev = 0;
 870         btrfs_put_root(pending_snapshot->snap);
 871         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 872 free_pending:
 873         if (pending_snapshot->anon_dev)
 874                 free_anon_bdev(pending_snapshot->anon_dev);
 875         kfree(pending_snapshot->root_item);
 876         btrfs_free_path(pending_snapshot->path);
 877         kfree(pending_snapshot);
 878
 879         return ret;
 880 }
 881
 882 /*  copy of may_delete in fs/namei.c()
 883  *      Check whether we can remove a link victim from directory dir, check
 884  *  whether the type of victim is right.
 885  *  1. We can't do it if dir is read-only (done in permission())
 886  *  2. We should have write and exec permissions on dir
 887  *  3. We can't remove anything from append-only dir
 888  *  4. We can't do anything with immutable dir (done in permission())
 889  *  5. If the sticky bit on dir is set we should either
 890  *      a. be owner of dir, or
 891  *      b. be owner of victim, or
 892  *      c. have CAP_FOWNER capability
 893  *  6. If the victim is append-only or immutable we can't do anything with
 894  *     links pointing to it.
 895  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 896  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 897  *  9. We can't remove a root or mountpoint.
 898  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 899  *     nfs_async_unlink().
 900  */
 901
 902 static int btrfs_may_delete(struct mnt_idmap *idmap,
 903                             struct inode *dir, struct dentry *victim, int isdir)
 904 {
 905         int error;
 906
 907         if (d_really_is_negative(victim))
 908                 return -ENOENT;
 909
 910         BUG_ON(d_inode(victim->d_parent) != dir);
 911         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 912
 913         error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
 914         if (error)
 915                 return error;
 916         if (IS_APPEND(dir))
 917                 return -EPERM;
 918         if (check_sticky(idmap, dir, d_inode(victim)) ||
 919             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 920             IS_SWAPFILE(d_inode(victim)))
 921                 return -EPERM;
 922         if (isdir) {
 923                 if (!d_is_dir(victim))
 924                         return -ENOTDIR;
 925                 if (IS_ROOT(victim))
 926                         return -EBUSY;
 927         } else if (d_is_dir(victim))
 928                 return -EISDIR;
 929         if (IS_DEADDIR(dir))
 930                 return -ENOENT;
 931         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 932                 return -EBUSY;
 933         return 0;
 934 }
 935
 936 /* copy of may_create in fs/namei.c() */
 937 static inline int btrfs_may_create(struct mnt_idmap *idmap,
 938                                    struct inode *dir, struct dentry *child)
 939 {
 940         if (d_really_is_positive(child))
 941                 return -EEXIST;
 942         if (IS_DEADDIR(dir))
 943                 return -ENOENT;
 944         if (!fsuidgid_has_mapping(dir->i_sb, idmap))
 945                 return -EOVERFLOW;
 946         return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
 947 }
 948
 949 /*
 950  * Create a new subvolume below @parent.  This is largely modeled after
 951  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 952  * inside this filesystem so it's quite a bit simpler.
 953  */
 954 static noinline int btrfs_mksubvol(const struct path *parent,
 955                                    struct mnt_idmap *idmap,
 956                                    const char *name, int namelen,
 957                                    struct btrfs_root *snap_src,
 958                                    bool readonly,
 959                                    struct btrfs_qgroup_inherit *inherit)
 960 {
 961         struct inode *dir = d_inode(parent->dentry);
 962         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 963         struct dentry *dentry;
 964         struct fscrypt_str name_str = FSTR_INIT((char *)name, namelen);
 965         int error;
 966
 967         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 968         if (error == -EINTR)
 969                 return error;
 970
 971         dentry = lookup_one(idmap, name, parent->dentry, namelen);
 972         error = PTR_ERR(dentry);
 973         if (IS_ERR(dentry))
 974                 goto out_unlock;
 975
 976         error = btrfs_may_create(idmap, dir, dentry);
 977         if (error)
 978                 goto out_dput;
 979
 980         /*
 981          * even if this name doesn't exist, we may get hash collisions.
 982          * check for them now when we can safely fail
 983          */
 984         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 985                                                dir->i_ino, &name_str);
 986         if (error)
 987                 goto out_dput;
 988
 989         down_read(&fs_info->subvol_sem);
 990
 991         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 992                 goto out_up_read;
 993
 994         if (snap_src)
 995                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 996         else
 997                 error = create_subvol(idmap, dir, dentry, inherit);
 998
 999         if (!error)
1000                 fsnotify_mkdir(dir, dentry);
1001 out_up_read:
1002         up_read(&fs_info->subvol_sem);
1003 out_dput:
1004         dput(dentry);
1005 out_unlock:
1006         btrfs_inode_unlock(BTRFS_I(dir), 0);
1007         return error;
1008 }
1009
1010 static noinline int btrfs_mksnapshot(const struct path *parent,
1011                                    struct mnt_idmap *idmap,
1012                                    const char *name, int namelen,
1013                                    struct btrfs_root *root,
1014                                    bool readonly,
1015                                    struct btrfs_qgroup_inherit *inherit)
1016 {
1017         int ret;
1018         bool snapshot_force_cow = false;
1019
1020         /*
1021          * Force new buffered writes to reserve space even when NOCOW is
1022          * possible. This is to avoid later writeback (running dealloc) to
1023          * fallback to COW mode and unexpectedly fail with ENOSPC.
1024          */
1025         btrfs_drew_read_lock(&root->snapshot_lock);
1026
1027         ret = btrfs_start_delalloc_snapshot(root, false);
1028         if (ret)
1029                 goto out;
1030
1031         /*
1032          * All previous writes have started writeback in NOCOW mode, so now
1033          * we force future writes to fallback to COW mode during snapshot
1034          * creation.
1035          */
1036         atomic_inc(&root->snapshot_force_cow);
1037         snapshot_force_cow = true;
1038
1039         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1040
1041         ret = btrfs_mksubvol(parent, idmap, name, namelen,
1042                              root, readonly, inherit);
1043 out:
1044         if (snapshot_force_cow)
1045                 atomic_dec(&root->snapshot_force_cow);
1046         btrfs_drew_read_unlock(&root->snapshot_lock);
1047         return ret;
1048 }
1049
1050 /*
1051  * Try to start exclusive operation @type or cancel it if it's running.
1052  *
1053  * Return:
1054  *   0        - normal mode, newly claimed op started
1055  *  >0        - normal mode, something else is running,
1056  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1057  * ECANCELED  - cancel mode, successful cancel
1058  * ENOTCONN   - cancel mode, operation not running anymore
1059  */
1060 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1061                         enum btrfs_exclusive_operation type, bool cancel)
1062 {
1063         if (!cancel) {
1064                 /* Start normal op */
1065                 if (!btrfs_exclop_start(fs_info, type))
1066                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1067                 /* Exclusive operation is now claimed */
1068                 return 0;
1069         }
1070
1071         /* Cancel running op */
1072         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1073                 /*
1074                  * This blocks any exclop finish from setting it to NONE, so we
1075                  * request cancellation. Either it runs and we will wait for it,
1076                  * or it has finished and no waiting will happen.
1077                  */
1078                 atomic_inc(&fs_info->reloc_cancel_req);
1079                 btrfs_exclop_start_unlock(fs_info);
1080
1081                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1082                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1083                                     TASK_INTERRUPTIBLE);
1084
1085                 return -ECANCELED;
1086         }
1087
1088         /* Something else is running or none */
1089         return -ENOTCONN;
1090 }
1091
1092 static noinline int btrfs_ioctl_resize(struct file *file,
1093                                         void __user *arg)
1094 {
1095         BTRFS_DEV_LOOKUP_ARGS(args);
1096         struct inode *inode = file_inode(file);
1097         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1098         u64 new_size;
1099         u64 old_size;
1100         u64 devid = 1;
1101         struct btrfs_root *root = BTRFS_I(inode)->root;
1102         struct btrfs_ioctl_vol_args *vol_args;
1103         struct btrfs_trans_handle *trans;
1104         struct btrfs_device *device = NULL;
1105         char *sizestr;
1106         char *retptr;
1107         char *devstr = NULL;
1108         int ret = 0;
1109         int mod = 0;
1110         bool cancel;
1111
1112         if (!capable(CAP_SYS_ADMIN))
1113                 return -EPERM;
1114
1115         ret = mnt_want_write_file(file);
1116         if (ret)
1117                 return ret;
1118
1119         /*
1120          * Read the arguments before checking exclusivity to be able to
1121          * distinguish regular resize and cancel
1122          */
1123         vol_args = memdup_user(arg, sizeof(*vol_args));
1124         if (IS_ERR(vol_args)) {
1125                 ret = PTR_ERR(vol_args);
1126                 goto out_drop;
1127         }
1128         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1129         sizestr = vol_args->name;
1130         cancel = (strcmp("cancel", sizestr) == 0);
1131         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1132         if (ret)
1133                 goto out_free;
1134         /* Exclusive operation is now claimed */
1135
1136         devstr = strchr(sizestr, ':');
1137         if (devstr) {
1138                 sizestr = devstr + 1;
1139                 *devstr = '\0';
1140                 devstr = vol_args->name;
1141                 ret = kstrtoull(devstr, 10, &devid);
1142                 if (ret)
1143                         goto out_finish;
1144                 if (!devid) {
1145                         ret = -EINVAL;
1146                         goto out_finish;
1147                 }
1148                 btrfs_info(fs_info, "resizing devid %llu", devid);
1149         }
1150
1151         args.devid = devid;
1152         device = btrfs_find_device(fs_info->fs_devices, &args);
1153         if (!device) {
1154                 btrfs_info(fs_info, "resizer unable to find device %llu",
1155                            devid);
1156                 ret = -ENODEV;
1157                 goto out_finish;
1158         }
1159
1160         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1161                 btrfs_info(fs_info,
1162                            "resizer unable to apply on readonly device %llu",
1163                        devid);
1164                 ret = -EPERM;
1165                 goto out_finish;
1166         }
1167
1168         if (!strcmp(sizestr, "max"))
1169                 new_size = bdev_nr_bytes(device->bdev);
1170         else {
1171                 if (sizestr[0] == '-') {
1172                         mod = -1;
1173                         sizestr++;
1174                 } else if (sizestr[0] == '+') {
1175                         mod = 1;
1176                         sizestr++;
1177                 }
1178                 new_size = memparse(sizestr, &retptr);
1179                 if (*retptr != '\0' || new_size == 0) {
1180                         ret = -EINVAL;
1181                         goto out_finish;
1182                 }
1183         }
1184
1185         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1186                 ret = -EPERM;
1187                 goto out_finish;
1188         }
1189
1190         old_size = btrfs_device_get_total_bytes(device);
1191
1192         if (mod < 0) {
1193                 if (new_size > old_size) {
1194                         ret = -EINVAL;
1195                         goto out_finish;
1196                 }
1197                 new_size = old_size - new_size;
1198         } else if (mod > 0) {
1199                 if (new_size > ULLONG_MAX - old_size) {
1200                         ret = -ERANGE;
1201                         goto out_finish;
1202                 }
1203                 new_size = old_size + new_size;
1204         }
1205
1206         if (new_size < SZ_256M) {
1207                 ret = -EINVAL;
1208                 goto out_finish;
1209         }
1210         if (new_size > bdev_nr_bytes(device->bdev)) {
1211                 ret = -EFBIG;
1212                 goto out_finish;
1213         }
1214
1215         new_size = round_down(new_size, fs_info->sectorsize);
1216
1217         if (new_size > old_size) {
1218                 trans = btrfs_start_transaction(root, 0);
1219                 if (IS_ERR(trans)) {
1220                         ret = PTR_ERR(trans);
1221                         goto out_finish;
1222                 }
1223                 ret = btrfs_grow_device(trans, device, new_size);
1224                 btrfs_commit_transaction(trans);
1225         } else if (new_size < old_size) {
1226                 ret = btrfs_shrink_device(device, new_size);
1227         } /* equal, nothing need to do */
1228
1229         if (ret == 0 && new_size != old_size)
1230                 btrfs_info_in_rcu(fs_info,
1231                         "resize device %s (devid %llu) from %llu to %llu",
1232                         btrfs_dev_name(device), device->devid,
1233                         old_size, new_size);
1234 out_finish:
1235         btrfs_exclop_finish(fs_info);
1236 out_free:
1237         kfree(vol_args);
1238 out_drop:
1239         mnt_drop_write_file(file);
1240         return ret;
1241 }
1242
1243 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1244                                 struct mnt_idmap *idmap,
1245                                 const char *name, unsigned long fd, int subvol,
1246                                 bool readonly,
1247                                 struct btrfs_qgroup_inherit *inherit)
1248 {
1249         int namelen;
1250         int ret = 0;
1251
1252         if (!S_ISDIR(file_inode(file)->i_mode))
1253                 return -ENOTDIR;
1254
1255         ret = mnt_want_write_file(file);
1256         if (ret)
1257                 goto out;
1258
1259         namelen = strlen(name);
1260         if (strchr(name, '/')) {
1261                 ret = -EINVAL;
1262                 goto out_drop_write;
1263         }
1264
1265         if (name[0] == '.' &&
1266            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1267                 ret = -EEXIST;
1268                 goto out_drop_write;
1269         }
1270
1271         if (subvol) {
1272                 ret = btrfs_mksubvol(&file->f_path, idmap, name,
1273                                      namelen, NULL, readonly, inherit);
1274         } else {
1275                 struct fd src = fdget(fd);
1276                 struct inode *src_inode;
1277                 if (!src.file) {
1278                         ret = -EINVAL;
1279                         goto out_drop_write;
1280                 }
1281
1282                 src_inode = file_inode(src.file);
1283                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1284                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1285                                    "Snapshot src from another FS");
1286                         ret = -EXDEV;
1287                 } else if (!inode_owner_or_capable(idmap, src_inode)) {
1288                         /*
1289                          * Subvolume creation is not restricted, but snapshots
1290                          * are limited to own subvolumes only
1291                          */
1292                         ret = -EPERM;
1293                 } else {
1294                         ret = btrfs_mksnapshot(&file->f_path, idmap,
1295                                                name, namelen,
1296                                                BTRFS_I(src_inode)->root,
1297                                                readonly, inherit);
1298                 }
1299                 fdput(src);
1300         }
1301 out_drop_write:
1302         mnt_drop_write_file(file);
1303 out:
1304         return ret;
1305 }
1306
1307 static noinline int btrfs_ioctl_snap_create(struct file *file,
1308                                             void __user *arg, int subvol)
1309 {
1310         struct btrfs_ioctl_vol_args *vol_args;
1311         int ret;
1312
1313         if (!S_ISDIR(file_inode(file)->i_mode))
1314                 return -ENOTDIR;
1315
1316         vol_args = memdup_user(arg, sizeof(*vol_args));
1317         if (IS_ERR(vol_args))
1318                 return PTR_ERR(vol_args);
1319         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1320
1321         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1322                                         vol_args->name, vol_args->fd, subvol,
1323                                         false, NULL);
1324
1325         kfree(vol_args);
1326         return ret;
1327 }
1328
1329 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1330                                                void __user *arg, int subvol)
1331 {
1332         struct btrfs_ioctl_vol_args_v2 *vol_args;
1333         int ret;
1334         bool readonly = false;
1335         struct btrfs_qgroup_inherit *inherit = NULL;
1336
1337         if (!S_ISDIR(file_inode(file)->i_mode))
1338                 return -ENOTDIR;
1339
1340         vol_args = memdup_user(arg, sizeof(*vol_args));
1341         if (IS_ERR(vol_args))
1342                 return PTR_ERR(vol_args);
1343         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1344
1345         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1346                 ret = -EOPNOTSUPP;
1347                 goto free_args;
1348         }
1349
1350         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1351                 readonly = true;
1352         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1353                 u64 nums;
1354
1355                 if (vol_args->size < sizeof(*inherit) ||
1356                     vol_args->size > PAGE_SIZE) {
1357                         ret = -EINVAL;
1358                         goto free_args;
1359                 }
1360                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1361                 if (IS_ERR(inherit)) {
1362                         ret = PTR_ERR(inherit);
1363                         goto free_args;
1364                 }
1365
1366                 if (inherit->num_qgroups > PAGE_SIZE ||
1367                     inherit->num_ref_copies > PAGE_SIZE ||
1368                     inherit->num_excl_copies > PAGE_SIZE) {
1369                         ret = -EINVAL;
1370                         goto free_inherit;
1371                 }
1372
1373                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1374                        2 * inherit->num_excl_copies;
1375                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1376                         ret = -EINVAL;
1377                         goto free_inherit;
1378                 }
1379         }
1380
1381         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1382                                         vol_args->name, vol_args->fd, subvol,
1383                                         readonly, inherit);
1384         if (ret)
1385                 goto free_inherit;
1386 free_inherit:
1387         kfree(inherit);
1388 free_args:
1389         kfree(vol_args);
1390         return ret;
1391 }
1392
1393 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
1394                                                 void __user *arg)
1395 {
1396         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1397         struct btrfs_root *root = BTRFS_I(inode)->root;
1398         int ret = 0;
1399         u64 flags = 0;
1400
1401         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1402                 return -EINVAL;
1403
1404         down_read(&fs_info->subvol_sem);
1405         if (btrfs_root_readonly(root))
1406                 flags |= BTRFS_SUBVOL_RDONLY;
1407         up_read(&fs_info->subvol_sem);
1408
1409         if (copy_to_user(arg, &flags, sizeof(flags)))
1410                 ret = -EFAULT;
1411
1412         return ret;
1413 }
1414
1415 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1416                                               void __user *arg)
1417 {
1418         struct inode *inode = file_inode(file);
1419         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1420         struct btrfs_root *root = BTRFS_I(inode)->root;
1421         struct btrfs_trans_handle *trans;
1422         u64 root_flags;
1423         u64 flags;
1424         int ret = 0;
1425
1426         if (!inode_owner_or_capable(file_mnt_idmap(file), inode))
1427                 return -EPERM;
1428
1429         ret = mnt_want_write_file(file);
1430         if (ret)
1431                 goto out;
1432
1433         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1434                 ret = -EINVAL;
1435                 goto out_drop_write;
1436         }
1437
1438         if (copy_from_user(&flags, arg, sizeof(flags))) {
1439                 ret = -EFAULT;
1440                 goto out_drop_write;
1441         }
1442
1443         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1444                 ret = -EOPNOTSUPP;
1445                 goto out_drop_write;
1446         }
1447
1448         down_write(&fs_info->subvol_sem);
1449
1450         /* nothing to do */
1451         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1452                 goto out_drop_sem;
1453
1454         root_flags = btrfs_root_flags(&root->root_item);
1455         if (flags & BTRFS_SUBVOL_RDONLY) {
1456                 btrfs_set_root_flags(&root->root_item,
1457                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1458         } else {
1459                 /*
1460                  * Block RO -> RW transition if this subvolume is involved in
1461                  * send
1462                  */
1463                 spin_lock(&root->root_item_lock);
1464                 if (root->send_in_progress == 0) {
1465                         btrfs_set_root_flags(&root->root_item,
1466                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1467                         spin_unlock(&root->root_item_lock);
1468                 } else {
1469                         spin_unlock(&root->root_item_lock);
1470                         btrfs_warn(fs_info,
1471                                    "Attempt to set subvolume %llu read-write during send",
1472                                    root->root_key.objectid);
1473                         ret = -EPERM;
1474                         goto out_drop_sem;
1475                 }
1476         }
1477
1478         trans = btrfs_start_transaction(root, 1);
1479         if (IS_ERR(trans)) {
1480                 ret = PTR_ERR(trans);
1481                 goto out_reset;
1482         }
1483
1484         ret = btrfs_update_root(trans, fs_info->tree_root,
1485                                 &root->root_key, &root->root_item);
1486         if (ret < 0) {
1487                 btrfs_end_transaction(trans);
1488                 goto out_reset;
1489         }
1490
1491         ret = btrfs_commit_transaction(trans);
1492
1493 out_reset:
1494         if (ret)
1495                 btrfs_set_root_flags(&root->root_item, root_flags);
1496 out_drop_sem:
1497         up_write(&fs_info->subvol_sem);
1498 out_drop_write:
1499         mnt_drop_write_file(file);
1500 out:
1501         return ret;
1502 }
1503
1504 static noinline int key_in_sk(struct btrfs_key *key,
1505                               struct btrfs_ioctl_search_key *sk)
1506 {
1507         struct btrfs_key test;
1508         int ret;
1509
1510         test.objectid = sk->min_objectid;
1511         test.type = sk->min_type;
1512         test.offset = sk->min_offset;
1513
1514         ret = btrfs_comp_cpu_keys(key, &test);
1515         if (ret < 0)
1516                 return 0;
1517
1518         test.objectid = sk->max_objectid;
1519         test.type = sk->max_type;
1520         test.offset = sk->max_offset;
1521
1522         ret = btrfs_comp_cpu_keys(key, &test);
1523         if (ret > 0)
1524                 return 0;
1525         return 1;
1526 }
1527
1528 static noinline int copy_to_sk(struct btrfs_path *path,
1529                                struct btrfs_key *key,
1530                                struct btrfs_ioctl_search_key *sk,
1531                                size_t *buf_size,
1532                                char __user *ubuf,
1533                                unsigned long *sk_offset,
1534                                int *num_found)
1535 {
1536         u64 found_transid;
1537         struct extent_buffer *leaf;
1538         struct btrfs_ioctl_search_header sh;
1539         struct btrfs_key test;
1540         unsigned long item_off;
1541         unsigned long item_len;
1542         int nritems;
1543         int i;
1544         int slot;
1545         int ret = 0;
1546
1547         leaf = path->nodes[0];
1548         slot = path->slots[0];
1549         nritems = btrfs_header_nritems(leaf);
1550
1551         if (btrfs_header_generation(leaf) > sk->max_transid) {
1552                 i = nritems;
1553                 goto advance_key;
1554         }
1555         found_transid = btrfs_header_generation(leaf);
1556
1557         for (i = slot; i < nritems; i++) {
1558                 item_off = btrfs_item_ptr_offset(leaf, i);
1559                 item_len = btrfs_item_size(leaf, i);
1560
1561                 btrfs_item_key_to_cpu(leaf, key, i);
1562                 if (!key_in_sk(key, sk))
1563                         continue;
1564
1565                 if (sizeof(sh) + item_len > *buf_size) {
1566                         if (*num_found) {
1567                                 ret = 1;
1568                                 goto out;
1569                         }
1570
1571                         /*
1572                          * return one empty item back for v1, which does not
1573                          * handle -EOVERFLOW
1574                          */
1575
1576                         *buf_size = sizeof(sh) + item_len;
1577                         item_len = 0;
1578                         ret = -EOVERFLOW;
1579                 }
1580
1581                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1582                         ret = 1;
1583                         goto out;
1584                 }
1585
1586                 sh.objectid = key->objectid;
1587                 sh.offset = key->offset;
1588                 sh.type = key->type;
1589                 sh.len = item_len;
1590                 sh.transid = found_transid;
1591
1592                 /*
1593                  * Copy search result header. If we fault then loop again so we
1594                  * can fault in the pages and -EFAULT there if there's a
1595                  * problem. Otherwise we'll fault and then copy the buffer in
1596                  * properly this next time through
1597                  */
1598                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
1599                         ret = 0;
1600                         goto out;
1601                 }
1602
1603                 *sk_offset += sizeof(sh);
1604
1605                 if (item_len) {
1606                         char __user *up = ubuf + *sk_offset;
1607                         /*
1608                          * Copy the item, same behavior as above, but reset the
1609                          * * sk_offset so we copy the full thing again.
1610                          */
1611                         if (read_extent_buffer_to_user_nofault(leaf, up,
1612                                                 item_off, item_len)) {
1613                                 ret = 0;
1614                                 *sk_offset -= sizeof(sh);
1615                                 goto out;
1616                         }
1617
1618                         *sk_offset += item_len;
1619                 }
1620                 (*num_found)++;
1621
1622                 if (ret) /* -EOVERFLOW from above */
1623                         goto out;
1624
1625                 if (*num_found >= sk->nr_items) {
1626                         ret = 1;
1627                         goto out;
1628                 }
1629         }
1630 advance_key:
1631         ret = 0;
1632         test.objectid = sk->max_objectid;
1633         test.type = sk->max_type;
1634         test.offset = sk->max_offset;
1635         if (btrfs_comp_cpu_keys(key, &test) >= 0)
1636                 ret = 1;
1637         else if (key->offset < (u64)-1)
1638                 key->offset++;
1639         else if (key->type < (u8)-1) {
1640                 key->offset = 0;
1641                 key->type++;
1642         } else if (key->objectid < (u64)-1) {
1643                 key->offset = 0;
1644                 key->type = 0;
1645                 key->objectid++;
1646         } else
1647                 ret = 1;
1648 out:
1649         /*
1650          *  0: all items from this leaf copied, continue with next
1651          *  1: * more items can be copied, but unused buffer is too small
1652          *     * all items were found
1653          *     Either way, it will stops the loop which iterates to the next
1654          *     leaf
1655          *  -EOVERFLOW: item was to large for buffer
1656          *  -EFAULT: could not copy extent buffer back to userspace
1657          */
1658         return ret;
1659 }
1660
1661 static noinline int search_ioctl(struct inode *inode,
1662                                  struct btrfs_ioctl_search_key *sk,
1663                                  size_t *buf_size,
1664                                  char __user *ubuf)
1665 {
1666         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
1667         struct btrfs_root *root;
1668         struct btrfs_key key;
1669         struct btrfs_path *path;
1670         int ret;
1671         int num_found = 0;
1672         unsigned long sk_offset = 0;
1673
1674         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
1675                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
1676                 return -EOVERFLOW;
1677         }
1678
1679         path = btrfs_alloc_path();
1680         if (!path)
1681                 return -ENOMEM;
1682
1683         if (sk->tree_id == 0) {
1684                 /* search the root of the inode that was passed */
1685                 root = btrfs_grab_root(BTRFS_I(inode)->root);
1686         } else {
1687                 root = btrfs_get_fs_root(info, sk->tree_id, true);
1688                 if (IS_ERR(root)) {
1689                         btrfs_free_path(path);
1690                         return PTR_ERR(root);
1691                 }
1692         }
1693
1694         key.objectid = sk->min_objectid;
1695         key.type = sk->min_type;
1696         key.offset = sk->min_offset;
1697
1698         while (1) {
1699                 ret = -EFAULT;
1700                 /*
1701                  * Ensure that the whole user buffer is faulted in at sub-page
1702                  * granularity, otherwise the loop may live-lock.
1703                  */
1704                 if (fault_in_subpage_writeable(ubuf + sk_offset,
1705                                                *buf_size - sk_offset))
1706                         break;
1707
1708                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
1709                 if (ret != 0) {
1710                         if (ret > 0)
1711                                 ret = 0;
1712                         goto err;
1713                 }
1714                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
1715                                  &sk_offset, &num_found);
1716                 btrfs_release_path(path);
1717                 if (ret)
1718                         break;
1719
1720         }
1721         if (ret > 0)
1722                 ret = 0;
1723 err:
1724         sk->nr_items = num_found;
1725         btrfs_put_root(root);
1726         btrfs_free_path(path);
1727         return ret;
1728 }
1729
1730 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
1731                                             void __user *argp)
1732 {
1733         struct btrfs_ioctl_search_args __user *uargs = argp;
1734         struct btrfs_ioctl_search_key sk;
1735         int ret;
1736         size_t buf_size;
1737
1738         if (!capable(CAP_SYS_ADMIN))
1739                 return -EPERM;
1740
1741         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
1742                 return -EFAULT;
1743
1744         buf_size = sizeof(uargs->buf);
1745
1746         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
1747
1748         /*
1749          * In the origin implementation an overflow is handled by returning a
1750          * search header with a len of zero, so reset ret.
1751          */
1752         if (ret == -EOVERFLOW)
1753                 ret = 0;
1754
1755         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
1756                 ret = -EFAULT;
1757         return ret;
1758 }
1759
1760 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
1761                                                void __user *argp)
1762 {
1763         struct btrfs_ioctl_search_args_v2 __user *uarg = argp;
1764         struct btrfs_ioctl_search_args_v2 args;
1765         int ret;
1766         size_t buf_size;
1767         const size_t buf_limit = SZ_16M;
1768
1769         if (!capable(CAP_SYS_ADMIN))
1770                 return -EPERM;
1771
1772         /* copy search header and buffer size */
1773         if (copy_from_user(&args, uarg, sizeof(args)))
1774                 return -EFAULT;
1775
1776         buf_size = args.buf_size;
1777
1778         /* limit result size to 16MB */
1779         if (buf_size > buf_limit)
1780                 buf_size = buf_limit;
1781
1782         ret = search_ioctl(inode, &args.key, &buf_size,
1783                            (char __user *)(&uarg->buf[0]));
1784         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
1785                 ret = -EFAULT;
1786         else if (ret == -EOVERFLOW &&
1787                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
1788                 ret = -EFAULT;
1789
1790         return ret;
1791 }
1792
1793 /*
1794  * Search INODE_REFs to identify path name of 'dirid' directory
1795  * in a 'tree_id' tree. and sets path name to 'name'.
1796  */
1797 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1798                                 u64 tree_id, u64 dirid, char *name)
1799 {
1800         struct btrfs_root *root;
1801         struct btrfs_key key;
1802         char *ptr;
1803         int ret = -1;
1804         int slot;
1805         int len;
1806         int total_len = 0;
1807         struct btrfs_inode_ref *iref;
1808         struct extent_buffer *l;
1809         struct btrfs_path *path;
1810
1811         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1812                 name[0]='\0';
1813                 return 0;
1814         }
1815
1816         path = btrfs_alloc_path();
1817         if (!path)
1818                 return -ENOMEM;
1819
1820         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
1821
1822         root = btrfs_get_fs_root(info, tree_id, true);
1823         if (IS_ERR(root)) {
1824                 ret = PTR_ERR(root);
1825                 root = NULL;
1826                 goto out;
1827         }
1828
1829         key.objectid = dirid;
1830         key.type = BTRFS_INODE_REF_KEY;
1831         key.offset = (u64)-1;
1832
1833         while (1) {
1834                 ret = btrfs_search_backwards(root, &key, path);
1835                 if (ret < 0)
1836                         goto out;
1837                 else if (ret > 0) {
1838                         ret = -ENOENT;
1839                         goto out;
1840                 }
1841
1842                 l = path->nodes[0];
1843                 slot = path->slots[0];
1844
1845                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1846                 len = btrfs_inode_ref_name_len(l, iref);
1847                 ptr -= len + 1;
1848                 total_len += len + 1;
1849                 if (ptr < name) {
1850                         ret = -ENAMETOOLONG;
1851                         goto out;
1852                 }
1853
1854                 *(ptr + len) = '/';
1855                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
1856
1857                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1858                         break;
1859
1860                 btrfs_release_path(path);
1861                 key.objectid = key.offset;
1862                 key.offset = (u64)-1;
1863                 dirid = key.objectid;
1864         }
1865         memmove(name, ptr, total_len);
1866         name[total_len] = '\0';
1867         ret = 0;
1868 out:
1869         btrfs_put_root(root);
1870         btrfs_free_path(path);
1871         return ret;
1872 }
1873
1874 static int btrfs_search_path_in_tree_user(struct mnt_idmap *idmap,
1875                                 struct inode *inode,
1876                                 struct btrfs_ioctl_ino_lookup_user_args *args)
1877 {
1878         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1879         struct super_block *sb = inode->i_sb;
1880         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
1881         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
1882         u64 dirid = args->dirid;
1883         unsigned long item_off;
1884         unsigned long item_len;
1885         struct btrfs_inode_ref *iref;
1886         struct btrfs_root_ref *rref;
1887         struct btrfs_root *root = NULL;
1888         struct btrfs_path *path;
1889         struct btrfs_key key, key2;
1890         struct extent_buffer *leaf;
1891         struct inode *temp_inode;
1892         char *ptr;
1893         int slot;
1894         int len;
1895         int total_len = 0;
1896         int ret;
1897
1898         path = btrfs_alloc_path();
1899         if (!path)
1900                 return -ENOMEM;
1901
1902         /*
1903          * If the bottom subvolume does not exist directly under upper_limit,
1904          * construct the path in from the bottom up.
1905          */
1906         if (dirid != upper_limit.objectid) {
1907                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
1908
1909                 root = btrfs_get_fs_root(fs_info, treeid, true);
1910                 if (IS_ERR(root)) {
1911                         ret = PTR_ERR(root);
1912                         goto out;
1913                 }
1914
1915                 key.objectid = dirid;
1916                 key.type = BTRFS_INODE_REF_KEY;
1917                 key.offset = (u64)-1;
1918                 while (1) {
1919                         ret = btrfs_search_backwards(root, &key, path);
1920                         if (ret < 0)
1921                                 goto out_put;
1922                         else if (ret > 0) {
1923                                 ret = -ENOENT;
1924                                 goto out_put;
1925                         }
1926
1927                         leaf = path->nodes[0];
1928                         slot = path->slots[0];
1929
1930                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
1931                         len = btrfs_inode_ref_name_len(leaf, iref);
1932                         ptr -= len + 1;
1933                         total_len += len + 1;
1934                         if (ptr < args->path) {
1935                                 ret = -ENAMETOOLONG;
1936                                 goto out_put;
1937                         }
1938
1939                         *(ptr + len) = '/';
1940                         read_extent_buffer(leaf, ptr,
1941                                         (unsigned long)(iref + 1), len);
1942
1943                         /* Check the read+exec permission of this directory */
1944                         ret = btrfs_previous_item(root, path, dirid,
1945                                                   BTRFS_INODE_ITEM_KEY);
1946                         if (ret < 0) {
1947                                 goto out_put;
1948                         } else if (ret > 0) {
1949                                 ret = -ENOENT;
1950                                 goto out_put;
1951                         }
1952
1953                         leaf = path->nodes[0];
1954                         slot = path->slots[0];
1955                         btrfs_item_key_to_cpu(leaf, &key2, slot);
1956                         if (key2.objectid != dirid) {
1957                                 ret = -ENOENT;
1958                                 goto out_put;
1959                         }
1960
1961                         /*
1962                          * We don't need the path anymore, so release it and
1963                          * avoid deadlocks and lockdep warnings in case
1964                          * btrfs_iget() needs to lookup the inode from its root
1965                          * btree and lock the same leaf.
1966                          */
1967                         btrfs_release_path(path);
1968                         temp_inode = btrfs_iget(sb, key2.objectid, root);
1969                         if (IS_ERR(temp_inode)) {
1970                                 ret = PTR_ERR(temp_inode);
1971                                 goto out_put;
1972                         }
1973                         ret = inode_permission(idmap, temp_inode,
1974                                                MAY_READ | MAY_EXEC);
1975                         iput(temp_inode);
1976                         if (ret) {
1977                                 ret = -EACCES;
1978                                 goto out_put;
1979                         }
1980
1981                         if (key.offset == upper_limit.objectid)
1982                                 break;
1983                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
1984                                 ret = -EACCES;
1985                                 goto out_put;
1986                         }
1987
1988                         key.objectid = key.offset;
1989                         key.offset = (u64)-1;
1990                         dirid = key.objectid;
1991                 }
1992
1993                 memmove(args->path, ptr, total_len);
1994                 args->path[total_len] = '\0';
1995                 btrfs_put_root(root);
1996                 root = NULL;
1997                 btrfs_release_path(path);
1998         }
1999
2000         /* Get the bottom subvolume's name from ROOT_REF */
2001         key.objectid = treeid;
2002         key.type = BTRFS_ROOT_REF_KEY;
2003         key.offset = args->treeid;
2004         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2005         if (ret < 0) {
2006                 goto out;
2007         } else if (ret > 0) {
2008                 ret = -ENOENT;
2009                 goto out;
2010         }
2011
2012         leaf = path->nodes[0];
2013         slot = path->slots[0];
2014         btrfs_item_key_to_cpu(leaf, &key, slot);
2015
2016         item_off = btrfs_item_ptr_offset(leaf, slot);
2017         item_len = btrfs_item_size(leaf, slot);
2018         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2019         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2020         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2021                 ret = -EINVAL;
2022                 goto out;
2023         }
2024
2025         /* Copy subvolume's name */
2026         item_off += sizeof(struct btrfs_root_ref);
2027         item_len -= sizeof(struct btrfs_root_ref);
2028         read_extent_buffer(leaf, args->name, item_off, item_len);
2029         args->name[item_len] = 0;
2030
2031 out_put:
2032         btrfs_put_root(root);
2033 out:
2034         btrfs_free_path(path);
2035         return ret;
2036 }
2037
2038 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2039                                            void __user *argp)
2040 {
2041         struct btrfs_ioctl_ino_lookup_args *args;
2042         int ret = 0;
2043
2044         args = memdup_user(argp, sizeof(*args));
2045         if (IS_ERR(args))
2046                 return PTR_ERR(args);
2047
2048         /*
2049          * Unprivileged query to obtain the containing subvolume root id. The
2050          * path is reset so it's consistent with btrfs_search_path_in_tree.
2051          */
2052         if (args->treeid == 0)
2053                 args->treeid = root->root_key.objectid;
2054
2055         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2056                 args->name[0] = 0;
2057                 goto out;
2058         }
2059
2060         if (!capable(CAP_SYS_ADMIN)) {
2061                 ret = -EPERM;
2062                 goto out;
2063         }
2064
2065         ret = btrfs_search_path_in_tree(root->fs_info,
2066                                         args->treeid, args->objectid,
2067                                         args->name);
2068
2069 out:
2070         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2071                 ret = -EFAULT;
2072
2073         kfree(args);
2074         return ret;
2075 }
2076
2077 /*
2078  * Version of ino_lookup ioctl (unprivileged)
2079  *
2080  * The main differences from ino_lookup ioctl are:
2081  *
2082  *   1. Read + Exec permission will be checked using inode_permission() during
2083  *      path construction. -EACCES will be returned in case of failure.
2084  *   2. Path construction will be stopped at the inode number which corresponds
2085  *      to the fd with which this ioctl is called. If constructed path does not
2086  *      exist under fd's inode, -EACCES will be returned.
2087  *   3. The name of bottom subvolume is also searched and filled.
2088  */
2089 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2090 {
2091         struct btrfs_ioctl_ino_lookup_user_args *args;
2092         struct inode *inode;
2093         int ret;
2094
2095         args = memdup_user(argp, sizeof(*args));
2096         if (IS_ERR(args))
2097                 return PTR_ERR(args);
2098
2099         inode = file_inode(file);
2100
2101         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2102             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2103                 /*
2104                  * The subvolume does not exist under fd with which this is
2105                  * called
2106                  */
2107                 kfree(args);
2108                 return -EACCES;
2109         }
2110
2111         ret = btrfs_search_path_in_tree_user(file_mnt_idmap(file), inode, args);
2112
2113         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2114                 ret = -EFAULT;
2115
2116         kfree(args);
2117         return ret;
2118 }
2119
2120 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2121 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
2122 {
2123         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2124         struct btrfs_fs_info *fs_info;
2125         struct btrfs_root *root;
2126         struct btrfs_path *path;
2127         struct btrfs_key key;
2128         struct btrfs_root_item *root_item;
2129         struct btrfs_root_ref *rref;
2130         struct extent_buffer *leaf;
2131         unsigned long item_off;
2132         unsigned long item_len;
2133         int slot;
2134         int ret = 0;
2135
2136         path = btrfs_alloc_path();
2137         if (!path)
2138                 return -ENOMEM;
2139
2140         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2141         if (!subvol_info) {
2142                 btrfs_free_path(path);
2143                 return -ENOMEM;
2144         }
2145
2146         fs_info = BTRFS_I(inode)->root->fs_info;
2147
2148         /* Get root_item of inode's subvolume */
2149         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2150         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2151         if (IS_ERR(root)) {
2152                 ret = PTR_ERR(root);
2153                 goto out_free;
2154         }
2155         root_item = &root->root_item;
2156
2157         subvol_info->treeid = key.objectid;
2158
2159         subvol_info->generation = btrfs_root_generation(root_item);
2160         subvol_info->flags = btrfs_root_flags(root_item);
2161
2162         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2163         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2164                                                     BTRFS_UUID_SIZE);
2165         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2166                                                     BTRFS_UUID_SIZE);
2167
2168         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2169         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2170         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2171
2172         subvol_info->otransid = btrfs_root_otransid(root_item);
2173         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2174         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2175
2176         subvol_info->stransid = btrfs_root_stransid(root_item);
2177         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2178         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2179
2180         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2181         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2182         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2183
2184         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2185                 /* Search root tree for ROOT_BACKREF of this subvolume */
2186                 key.type = BTRFS_ROOT_BACKREF_KEY;
2187                 key.offset = 0;
2188                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2189                 if (ret < 0) {
2190                         goto out;
2191                 } else if (path->slots[0] >=
2192                            btrfs_header_nritems(path->nodes[0])) {
2193                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2194                         if (ret < 0) {
2195                                 goto out;
2196                         } else if (ret > 0) {
2197                                 ret = -EUCLEAN;
2198                                 goto out;
2199                         }
2200                 }
2201
2202                 leaf = path->nodes[0];
2203                 slot = path->slots[0];
2204                 btrfs_item_key_to_cpu(leaf, &key, slot);
2205                 if (key.objectid == subvol_info->treeid &&
2206                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2207                         subvol_info->parent_id = key.offset;
2208
2209                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2210                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2211
2212                         item_off = btrfs_item_ptr_offset(leaf, slot)
2213                                         + sizeof(struct btrfs_root_ref);
2214                         item_len = btrfs_item_size(leaf, slot)
2215                                         - sizeof(struct btrfs_root_ref);
2216                         read_extent_buffer(leaf, subvol_info->name,
2217                                            item_off, item_len);
2218                 } else {
2219                         ret = -ENOENT;
2220                         goto out;
2221                 }
2222         }
2223
2224         btrfs_free_path(path);
2225         path = NULL;
2226         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2227                 ret = -EFAULT;
2228
2229 out:
2230         btrfs_put_root(root);
2231 out_free:
2232         btrfs_free_path(path);
2233         kfree(subvol_info);
2234         return ret;
2235 }
2236
2237 /*
2238  * Return ROOT_REF information of the subvolume containing this inode
2239  * except the subvolume name.
2240  */
2241 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
2242                                           void __user *argp)
2243 {
2244         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2245         struct btrfs_root_ref *rref;
2246         struct btrfs_path *path;
2247         struct btrfs_key key;
2248         struct extent_buffer *leaf;
2249         u64 objectid;
2250         int slot;
2251         int ret;
2252         u8 found;
2253
2254         path = btrfs_alloc_path();
2255         if (!path)
2256                 return -ENOMEM;
2257
2258         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2259         if (IS_ERR(rootrefs)) {
2260                 btrfs_free_path(path);
2261                 return PTR_ERR(rootrefs);
2262         }
2263
2264         objectid = root->root_key.objectid;
2265         key.objectid = objectid;
2266         key.type = BTRFS_ROOT_REF_KEY;
2267         key.offset = rootrefs->min_treeid;
2268         found = 0;
2269
2270         root = root->fs_info->tree_root;
2271         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2272         if (ret < 0) {
2273                 goto out;
2274         } else if (path->slots[0] >=
2275                    btrfs_header_nritems(path->nodes[0])) {
2276                 ret = btrfs_next_leaf(root, path);
2277                 if (ret < 0) {
2278                         goto out;
2279                 } else if (ret > 0) {
2280                         ret = -EUCLEAN;
2281                         goto out;
2282                 }
2283         }
2284         while (1) {
2285                 leaf = path->nodes[0];
2286                 slot = path->slots[0];
2287
2288                 btrfs_item_key_to_cpu(leaf, &key, slot);
2289                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2290                         ret = 0;
2291                         goto out;
2292                 }
2293
2294                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2295                         ret = -EOVERFLOW;
2296                         goto out;
2297                 }
2298
2299                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2300                 rootrefs->rootref[found].treeid = key.offset;
2301                 rootrefs->rootref[found].dirid =
2302                                   btrfs_root_ref_dirid(leaf, rref);
2303                 found++;
2304
2305                 ret = btrfs_next_item(root, path);
2306                 if (ret < 0) {
2307                         goto out;
2308                 } else if (ret > 0) {
2309                         ret = -EUCLEAN;
2310                         goto out;
2311                 }
2312         }
2313
2314 out:
2315         btrfs_free_path(path);
2316
2317         if (!ret || ret == -EOVERFLOW) {
2318                 rootrefs->num_items = found;
2319                 /* update min_treeid for next search */
2320                 if (found)
2321                         rootrefs->min_treeid =
2322                                 rootrefs->rootref[found - 1].treeid + 1;
2323                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2324                         ret = -EFAULT;
2325         }
2326
2327         kfree(rootrefs);
2328
2329         return ret;
2330 }
2331
2332 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2333                                              void __user *arg,
2334                                              bool destroy_v2)
2335 {
2336         struct dentry *parent = file->f_path.dentry;
2337         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2338         struct dentry *dentry;
2339         struct inode *dir = d_inode(parent);
2340         struct inode *inode;
2341         struct btrfs_root *root = BTRFS_I(dir)->root;
2342         struct btrfs_root *dest = NULL;
2343         struct btrfs_ioctl_vol_args *vol_args = NULL;
2344         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2345         struct mnt_idmap *idmap = file_mnt_idmap(file);
2346         char *subvol_name, *subvol_name_ptr = NULL;
2347         int subvol_namelen;
2348         int err = 0;
2349         bool destroy_parent = false;
2350
2351         /* We don't support snapshots with extent tree v2 yet. */
2352         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2353                 btrfs_err(fs_info,
2354                           "extent tree v2 doesn't support snapshot deletion yet");
2355                 return -EOPNOTSUPP;
2356         }
2357
2358         if (destroy_v2) {
2359                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2360                 if (IS_ERR(vol_args2))
2361                         return PTR_ERR(vol_args2);
2362
2363                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2364                         err = -EOPNOTSUPP;
2365                         goto out;
2366                 }
2367
2368                 /*
2369                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2370                  * name, same as v1 currently does.
2371                  */
2372                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2373                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2374                         subvol_name = vol_args2->name;
2375
2376                         err = mnt_want_write_file(file);
2377                         if (err)
2378                                 goto out;
2379                 } else {
2380                         struct inode *old_dir;
2381
2382                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2383                                 err = -EINVAL;
2384                                 goto out;
2385                         }
2386
2387                         err = mnt_want_write_file(file);
2388                         if (err)
2389                                 goto out;
2390
2391                         dentry = btrfs_get_dentry(fs_info->sb,
2392                                         BTRFS_FIRST_FREE_OBJECTID,
2393                                         vol_args2->subvolid, 0);
2394                         if (IS_ERR(dentry)) {
2395                                 err = PTR_ERR(dentry);
2396                                 goto out_drop_write;
2397                         }
2398
2399                         /*
2400                          * Change the default parent since the subvolume being
2401                          * deleted can be outside of the current mount point.
2402                          */
2403                         parent = btrfs_get_parent(dentry);
2404
2405                         /*
2406                          * At this point dentry->d_name can point to '/' if the
2407                          * subvolume we want to destroy is outsite of the
2408                          * current mount point, so we need to release the
2409                          * current dentry and execute the lookup to return a new
2410                          * one with ->d_name pointing to the
2411                          * <mount point>/subvol_name.
2412                          */
2413                         dput(dentry);
2414                         if (IS_ERR(parent)) {
2415                                 err = PTR_ERR(parent);
2416                                 goto out_drop_write;
2417                         }
2418                         old_dir = dir;
2419                         dir = d_inode(parent);
2420
2421                         /*
2422                          * If v2 was used with SPEC_BY_ID, a new parent was
2423                          * allocated since the subvolume can be outside of the
2424                          * current mount point. Later on we need to release this
2425                          * new parent dentry.
2426                          */
2427                         destroy_parent = true;
2428
2429                         /*
2430                          * On idmapped mounts, deletion via subvolid is
2431                          * restricted to subvolumes that are immediate
2432                          * ancestors of the inode referenced by the file
2433                          * descriptor in the ioctl. Otherwise the idmapping
2434                          * could potentially be abused to delete subvolumes
2435                          * anywhere in the filesystem the user wouldn't be able
2436                          * to delete without an idmapped mount.
2437                          */
2438                         if (old_dir != dir && idmap != &nop_mnt_idmap) {
2439                                 err = -EOPNOTSUPP;
2440                                 goto free_parent;
2441                         }
2442
2443                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2444                                                 fs_info, vol_args2->subvolid);
2445                         if (IS_ERR(subvol_name_ptr)) {
2446                                 err = PTR_ERR(subvol_name_ptr);
2447                                 goto free_parent;
2448                         }
2449                         /* subvol_name_ptr is already nul terminated */
2450                         subvol_name = (char *)kbasename(subvol_name_ptr);
2451                 }
2452         } else {
2453                 vol_args = memdup_user(arg, sizeof(*vol_args));
2454                 if (IS_ERR(vol_args))
2455                         return PTR_ERR(vol_args);
2456
2457                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2458                 subvol_name = vol_args->name;
2459
2460                 err = mnt_want_write_file(file);
2461                 if (err)
2462                         goto out;
2463         }
2464
2465         subvol_namelen = strlen(subvol_name);
2466
2467         if (strchr(subvol_name, '/') ||
2468             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2469                 err = -EINVAL;
2470                 goto free_subvol_name;
2471         }
2472
2473         if (!S_ISDIR(dir->i_mode)) {
2474                 err = -ENOTDIR;
2475                 goto free_subvol_name;
2476         }
2477
2478         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2479         if (err == -EINTR)
2480                 goto free_subvol_name;
2481         dentry = lookup_one(idmap, subvol_name, parent, subvol_namelen);
2482         if (IS_ERR(dentry)) {
2483                 err = PTR_ERR(dentry);
2484                 goto out_unlock_dir;
2485         }
2486
2487         if (d_really_is_negative(dentry)) {
2488                 err = -ENOENT;
2489                 goto out_dput;
2490         }
2491
2492         inode = d_inode(dentry);
2493         dest = BTRFS_I(inode)->root;
2494         if (!capable(CAP_SYS_ADMIN)) {
2495                 /*
2496                  * Regular user.  Only allow this with a special mount
2497                  * option, when the user has write+exec access to the
2498                  * subvol root, and when rmdir(2) would have been
2499                  * allowed.
2500                  *
2501                  * Note that this is _not_ check that the subvol is
2502                  * empty or doesn't contain data that we wouldn't
2503                  * otherwise be able to delete.
2504                  *
2505                  * Users who want to delete empty subvols should try
2506                  * rmdir(2).
2507                  */
2508                 err = -EPERM;
2509                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2510                         goto out_dput;
2511
2512                 /*
2513                  * Do not allow deletion if the parent dir is the same
2514                  * as the dir to be deleted.  That means the ioctl
2515                  * must be called on the dentry referencing the root
2516                  * of the subvol, not a random directory contained
2517                  * within it.
2518                  */
2519                 err = -EINVAL;
2520                 if (root == dest)
2521                         goto out_dput;
2522
2523                 err = inode_permission(idmap, inode, MAY_WRITE | MAY_EXEC);
2524                 if (err)
2525                         goto out_dput;
2526         }
2527
2528         /* check if subvolume may be deleted by a user */
2529         err = btrfs_may_delete(idmap, dir, dentry, 1);
2530         if (err)
2531                 goto out_dput;
2532
2533         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2534                 err = -EINVAL;
2535                 goto out_dput;
2536         }
2537
2538         btrfs_inode_lock(BTRFS_I(inode), 0);
2539         err = btrfs_delete_subvolume(BTRFS_I(dir), dentry);
2540         btrfs_inode_unlock(BTRFS_I(inode), 0);
2541         if (!err)
2542                 d_delete_notify(dir, dentry);
2543
2544 out_dput:
2545         dput(dentry);
2546 out_unlock_dir:
2547         btrfs_inode_unlock(BTRFS_I(dir), 0);
2548 free_subvol_name:
2549         kfree(subvol_name_ptr);
2550 free_parent:
2551         if (destroy_parent)
2552                 dput(parent);
2553 out_drop_write:
2554         mnt_drop_write_file(file);
2555 out:
2556         kfree(vol_args2);
2557         kfree(vol_args);
2558         return err;
2559 }
2560
2561 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2562 {
2563         struct inode *inode = file_inode(file);
2564         struct btrfs_root *root = BTRFS_I(inode)->root;
2565         struct btrfs_ioctl_defrag_range_args range = {0};
2566         int ret;
2567
2568         ret = mnt_want_write_file(file);
2569         if (ret)
2570                 return ret;
2571
2572         if (btrfs_root_readonly(root)) {
2573                 ret = -EROFS;
2574                 goto out;
2575         }
2576
2577         switch (inode->i_mode & S_IFMT) {
2578         case S_IFDIR:
2579                 if (!capable(CAP_SYS_ADMIN)) {
2580                         ret = -EPERM;
2581                         goto out;
2582                 }
2583                 ret = btrfs_defrag_root(root);
2584                 break;
2585         case S_IFREG:
2586                 /*
2587                  * Note that this does not check the file descriptor for write
2588                  * access. This prevents defragmenting executables that are
2589                  * running and allows defrag on files open in read-only mode.
2590                  */
2591                 if (!capable(CAP_SYS_ADMIN) &&
2592                     inode_permission(&nop_mnt_idmap, inode, MAY_WRITE)) {
2593                         ret = -EPERM;
2594                         goto out;
2595                 }
2596
2597                 if (argp) {
2598                         if (copy_from_user(&range, argp, sizeof(range))) {
2599                                 ret = -EFAULT;
2600                                 goto out;
2601                         }
2602                         /* compression requires us to start the IO */
2603                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2604                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
2605                                 range.extent_thresh = (u32)-1;
2606                         }
2607                 } else {
2608                         /* the rest are all set to zero by kzalloc */
2609                         range.len = (u64)-1;
2610                 }
2611                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
2612                                         &range, BTRFS_OLDEST_GENERATION, 0);
2613                 if (ret > 0)
2614                         ret = 0;
2615                 break;
2616         default:
2617                 ret = -EINVAL;
2618         }
2619 out:
2620         mnt_drop_write_file(file);
2621         return ret;
2622 }
2623
2624 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2625 {
2626         struct btrfs_ioctl_vol_args *vol_args;
2627         bool restore_op = false;
2628         int ret;
2629
2630         if (!capable(CAP_SYS_ADMIN))
2631                 return -EPERM;
2632
2633         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2634                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
2635                 return -EINVAL;
2636         }
2637
2638         if (fs_info->fs_devices->temp_fsid) {
2639                 btrfs_err(fs_info,
2640                           "device add not supported on cloned temp-fsid mount");
2641                 return -EINVAL;
2642         }
2643
2644         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
2645                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
2646                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2647
2648                 /*
2649                  * We can do the device add because we have a paused balanced,
2650                  * change the exclusive op type and remember we should bring
2651                  * back the paused balance
2652                  */
2653                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
2654                 btrfs_exclop_start_unlock(fs_info);
2655                 restore_op = true;
2656         }
2657
2658         vol_args = memdup_user(arg, sizeof(*vol_args));
2659         if (IS_ERR(vol_args)) {
2660                 ret = PTR_ERR(vol_args);
2661                 goto out;
2662         }
2663
2664         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2665         ret = btrfs_init_new_device(fs_info, vol_args->name);
2666
2667         if (!ret)
2668                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2669
2670         kfree(vol_args);
2671 out:
2672         if (restore_op)
2673                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
2674         else
2675                 btrfs_exclop_finish(fs_info);
2676         return ret;
2677 }
2678
2679 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2680 {
2681         BTRFS_DEV_LOOKUP_ARGS(args);
2682         struct inode *inode = file_inode(file);
2683         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2684         struct btrfs_ioctl_vol_args_v2 *vol_args;
2685         struct bdev_handle *bdev_handle = NULL;
2686         int ret;
2687         bool cancel = false;
2688
2689         if (!capable(CAP_SYS_ADMIN))
2690                 return -EPERM;
2691
2692         vol_args = memdup_user(arg, sizeof(*vol_args));
2693         if (IS_ERR(vol_args))
2694                 return PTR_ERR(vol_args);
2695
2696         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
2697                 ret = -EOPNOTSUPP;
2698                 goto out;
2699         }
2700
2701         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2702         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2703                 args.devid = vol_args->devid;
2704         } else if (!strcmp("cancel", vol_args->name)) {
2705                 cancel = true;
2706         } else {
2707                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2708                 if (ret)
2709                         goto out;
2710         }
2711
2712         ret = mnt_want_write_file(file);
2713         if (ret)
2714                 goto out;
2715
2716         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2717                                            cancel);
2718         if (ret)
2719                 goto err_drop;
2720
2721         /* Exclusive operation is now claimed */
2722         ret = btrfs_rm_device(fs_info, &args, &bdev_handle);
2723
2724         btrfs_exclop_finish(fs_info);
2725
2726         if (!ret) {
2727                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2728                         btrfs_info(fs_info, "device deleted: id %llu",
2729                                         vol_args->devid);
2730                 else
2731                         btrfs_info(fs_info, "device deleted: %s",
2732                                         vol_args->name);
2733         }
2734 err_drop:
2735         mnt_drop_write_file(file);
2736         if (bdev_handle)
2737                 bdev_release(bdev_handle);
2738 out:
2739         btrfs_put_dev_args_from_path(&args);
2740         kfree(vol_args);
2741         return ret;
2742 }
2743
2744 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2745 {
2746         BTRFS_DEV_LOOKUP_ARGS(args);
2747         struct inode *inode = file_inode(file);
2748         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2749         struct btrfs_ioctl_vol_args *vol_args;
2750         struct bdev_handle *bdev_handle = NULL;
2751         int ret;
2752         bool cancel = false;
2753
2754         if (!capable(CAP_SYS_ADMIN))
2755                 return -EPERM;
2756
2757         vol_args = memdup_user(arg, sizeof(*vol_args));
2758         if (IS_ERR(vol_args))
2759                 return PTR_ERR(vol_args);
2760
2761         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2762         if (!strcmp("cancel", vol_args->name)) {
2763                 cancel = true;
2764         } else {
2765                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2766                 if (ret)
2767                         goto out;
2768         }
2769
2770         ret = mnt_want_write_file(file);
2771         if (ret)
2772                 goto out;
2773
2774         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2775                                            cancel);
2776         if (ret == 0) {
2777                 ret = btrfs_rm_device(fs_info, &args, &bdev_handle);
2778                 if (!ret)
2779                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2780                 btrfs_exclop_finish(fs_info);
2781         }
2782
2783         mnt_drop_write_file(file);
2784         if (bdev_handle)
2785                 bdev_release(bdev_handle);
2786 out:
2787         btrfs_put_dev_args_from_path(&args);
2788         kfree(vol_args);
2789         return ret;
2790 }
2791
2792 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2793                                 void __user *arg)
2794 {
2795         struct btrfs_ioctl_fs_info_args *fi_args;
2796         struct btrfs_device *device;
2797         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2798         u64 flags_in;
2799         int ret = 0;
2800
2801         fi_args = memdup_user(arg, sizeof(*fi_args));
2802         if (IS_ERR(fi_args))
2803                 return PTR_ERR(fi_args);
2804
2805         flags_in = fi_args->flags;
2806         memset(fi_args, 0, sizeof(*fi_args));
2807
2808         rcu_read_lock();
2809         fi_args->num_devices = fs_devices->num_devices;
2810
2811         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2812                 if (device->devid > fi_args->max_id)
2813                         fi_args->max_id = device->devid;
2814         }
2815         rcu_read_unlock();
2816
2817         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
2818         fi_args->nodesize = fs_info->nodesize;
2819         fi_args->sectorsize = fs_info->sectorsize;
2820         fi_args->clone_alignment = fs_info->sectorsize;
2821
2822         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
2823                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
2824                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
2825                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
2826         }
2827
2828         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
2829                 fi_args->generation = btrfs_get_fs_generation(fs_info);
2830                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
2831         }
2832
2833         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
2834                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
2835                        sizeof(fi_args->metadata_uuid));
2836                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
2837         }
2838
2839         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2840                 ret = -EFAULT;
2841
2842         kfree(fi_args);
2843         return ret;
2844 }
2845
2846 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2847                                  void __user *arg)
2848 {
2849         BTRFS_DEV_LOOKUP_ARGS(args);
2850         struct btrfs_ioctl_dev_info_args *di_args;
2851         struct btrfs_device *dev;
2852         int ret = 0;
2853
2854         di_args = memdup_user(arg, sizeof(*di_args));
2855         if (IS_ERR(di_args))
2856                 return PTR_ERR(di_args);
2857
2858         args.devid = di_args->devid;
2859         if (!btrfs_is_empty_uuid(di_args->uuid))
2860                 args.uuid = di_args->uuid;
2861
2862         rcu_read_lock();
2863         dev = btrfs_find_device(fs_info->fs_devices, &args);
2864         if (!dev) {
2865                 ret = -ENODEV;
2866                 goto out;
2867         }
2868
2869         di_args->devid = dev->devid;
2870         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2871         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2872         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2873         memcpy(di_args->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2874         if (dev->name)
2875                 strscpy(di_args->path, btrfs_dev_name(dev), sizeof(di_args->path));
2876         else
2877                 di_args->path[0] = '\0';
2878
2879 out:
2880         rcu_read_unlock();
2881         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2882                 ret = -EFAULT;
2883
2884         kfree(di_args);
2885         return ret;
2886 }
2887
2888 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2889 {
2890         struct inode *inode = file_inode(file);
2891         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2892         struct btrfs_root *root = BTRFS_I(inode)->root;
2893         struct btrfs_root *new_root;
2894         struct btrfs_dir_item *di;
2895         struct btrfs_trans_handle *trans;
2896         struct btrfs_path *path = NULL;
2897         struct btrfs_disk_key disk_key;
2898         struct fscrypt_str name = FSTR_INIT("default", 7);
2899         u64 objectid = 0;
2900         u64 dir_id;
2901         int ret;
2902
2903         if (!capable(CAP_SYS_ADMIN))
2904                 return -EPERM;
2905
2906         ret = mnt_want_write_file(file);
2907         if (ret)
2908                 return ret;
2909
2910         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
2911                 ret = -EFAULT;
2912                 goto out;
2913         }
2914
2915         if (!objectid)
2916                 objectid = BTRFS_FS_TREE_OBJECTID;
2917
2918         new_root = btrfs_get_fs_root(fs_info, objectid, true);
2919         if (IS_ERR(new_root)) {
2920                 ret = PTR_ERR(new_root);
2921                 goto out;
2922         }
2923         if (!is_fstree(new_root->root_key.objectid)) {
2924                 ret = -ENOENT;
2925                 goto out_free;
2926         }
2927
2928         path = btrfs_alloc_path();
2929         if (!path) {
2930                 ret = -ENOMEM;
2931                 goto out_free;
2932         }
2933
2934         trans = btrfs_start_transaction(root, 1);
2935         if (IS_ERR(trans)) {
2936                 ret = PTR_ERR(trans);
2937                 goto out_free;
2938         }
2939
2940         dir_id = btrfs_super_root_dir(fs_info->super_copy);
2941         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
2942                                    dir_id, &name, 1);
2943         if (IS_ERR_OR_NULL(di)) {
2944                 btrfs_release_path(path);
2945                 btrfs_end_transaction(trans);
2946                 btrfs_err(fs_info,
2947                           "Umm, you don't have the default diritem, this isn't going to work");
2948                 ret = -ENOENT;
2949                 goto out_free;
2950         }
2951
2952         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2953         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2954         btrfs_mark_buffer_dirty(trans, path->nodes[0]);
2955         btrfs_release_path(path);
2956
2957         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
2958         btrfs_end_transaction(trans);
2959 out_free:
2960         btrfs_put_root(new_root);
2961         btrfs_free_path(path);
2962 out:
2963         mnt_drop_write_file(file);
2964         return ret;
2965 }
2966
2967 static void get_block_group_info(struct list_head *groups_list,
2968                                  struct btrfs_ioctl_space_info *space)
2969 {
2970         struct btrfs_block_group *block_group;
2971
2972         space->total_bytes = 0;
2973         space->used_bytes = 0;
2974         space->flags = 0;
2975         list_for_each_entry(block_group, groups_list, list) {
2976                 space->flags = block_group->flags;
2977                 space->total_bytes += block_group->length;
2978                 space->used_bytes += block_group->used;
2979         }
2980 }
2981
2982 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
2983                                    void __user *arg)
2984 {
2985         struct btrfs_ioctl_space_args space_args = { 0 };
2986         struct btrfs_ioctl_space_info space;
2987         struct btrfs_ioctl_space_info *dest;
2988         struct btrfs_ioctl_space_info *dest_orig;
2989         struct btrfs_ioctl_space_info __user *user_dest;
2990         struct btrfs_space_info *info;
2991         static const u64 types[] = {
2992                 BTRFS_BLOCK_GROUP_DATA,
2993                 BTRFS_BLOCK_GROUP_SYSTEM,
2994                 BTRFS_BLOCK_GROUP_METADATA,
2995                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
2996         };
2997         int num_types = 4;
2998         int alloc_size;
2999         int ret = 0;
3000         u64 slot_count = 0;
3001         int i, c;
3002
3003         if (copy_from_user(&space_args,
3004                            (struct btrfs_ioctl_space_args __user *)arg,
3005                            sizeof(space_args)))
3006                 return -EFAULT;
3007
3008         for (i = 0; i < num_types; i++) {
3009                 struct btrfs_space_info *tmp;
3010
3011                 info = NULL;
3012                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3013                         if (tmp->flags == types[i]) {
3014                                 info = tmp;
3015                                 break;
3016                         }
3017                 }
3018
3019                 if (!info)
3020                         continue;
3021
3022                 down_read(&info->groups_sem);
3023                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3024                         if (!list_empty(&info->block_groups[c]))
3025                                 slot_count++;
3026                 }
3027                 up_read(&info->groups_sem);
3028         }
3029
3030         /*
3031          * Global block reserve, exported as a space_info
3032          */
3033         slot_count++;
3034
3035         /* space_slots == 0 means they are asking for a count */
3036         if (space_args.space_slots == 0) {
3037                 space_args.total_spaces = slot_count;
3038                 goto out;
3039         }
3040
3041         slot_count = min_t(u64, space_args.space_slots, slot_count);
3042
3043         alloc_size = sizeof(*dest) * slot_count;
3044
3045         /* we generally have at most 6 or so space infos, one for each raid
3046          * level.  So, a whole page should be more than enough for everyone
3047          */
3048         if (alloc_size > PAGE_SIZE)
3049                 return -ENOMEM;
3050
3051         space_args.total_spaces = 0;
3052         dest = kmalloc(alloc_size, GFP_KERNEL);
3053         if (!dest)
3054                 return -ENOMEM;
3055         dest_orig = dest;
3056
3057         /* now we have a buffer to copy into */
3058         for (i = 0; i < num_types; i++) {
3059                 struct btrfs_space_info *tmp;
3060
3061                 if (!slot_count)
3062                         break;
3063
3064                 info = NULL;
3065                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3066                         if (tmp->flags == types[i]) {
3067                                 info = tmp;
3068                                 break;
3069                         }
3070                 }
3071
3072                 if (!info)
3073                         continue;
3074                 down_read(&info->groups_sem);
3075                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3076                         if (!list_empty(&info->block_groups[c])) {
3077                                 get_block_group_info(&info->block_groups[c],
3078                                                      &space);
3079                                 memcpy(dest, &space, sizeof(space));
3080                                 dest++;
3081                                 space_args.total_spaces++;
3082                                 slot_count--;
3083                         }
3084                         if (!slot_count)
3085                                 break;
3086                 }
3087                 up_read(&info->groups_sem);
3088         }
3089
3090         /*
3091          * Add global block reserve
3092          */
3093         if (slot_count) {
3094                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3095
3096                 spin_lock(&block_rsv->lock);
3097                 space.total_bytes = block_rsv->size;
3098                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3099                 spin_unlock(&block_rsv->lock);
3100                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3101                 memcpy(dest, &space, sizeof(space));
3102                 space_args.total_spaces++;
3103         }
3104
3105         user_dest = (struct btrfs_ioctl_space_info __user *)
3106                 (arg + sizeof(struct btrfs_ioctl_space_args));
3107
3108         if (copy_to_user(user_dest, dest_orig, alloc_size))
3109                 ret = -EFAULT;
3110
3111         kfree(dest_orig);
3112 out:
3113         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3114                 ret = -EFAULT;
3115
3116         return ret;
3117 }
3118
3119 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3120                                             void __user *argp)
3121 {
3122         struct btrfs_trans_handle *trans;
3123         u64 transid;
3124
3125         /*
3126          * Start orphan cleanup here for the given root in case it hasn't been
3127          * started already by other means. Errors are handled in the other
3128          * functions during transaction commit.
3129          */
3130         btrfs_orphan_cleanup(root);
3131
3132         trans = btrfs_attach_transaction_barrier(root);
3133         if (IS_ERR(trans)) {
3134                 if (PTR_ERR(trans) != -ENOENT)
3135                         return PTR_ERR(trans);
3136
3137                 /* No running transaction, don't bother */
3138                 transid = btrfs_get_last_trans_committed(root->fs_info);
3139                 goto out;
3140         }
3141         transid = trans->transid;
3142         btrfs_commit_transaction_async(trans);
3143 out:
3144         if (argp)
3145                 if (copy_to_user(argp, &transid, sizeof(transid)))
3146                         return -EFAULT;
3147         return 0;
3148 }
3149
3150 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3151                                            void __user *argp)
3152 {
3153         /* By default wait for the current transaction. */
3154         u64 transid = 0;
3155
3156         if (argp)
3157                 if (copy_from_user(&transid, argp, sizeof(transid)))
3158                         return -EFAULT;
3159
3160         return btrfs_wait_for_commit(fs_info, transid);
3161 }
3162
3163 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3164 {
3165         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3166         struct btrfs_ioctl_scrub_args *sa;
3167         int ret;
3168
3169         if (!capable(CAP_SYS_ADMIN))
3170                 return -EPERM;
3171
3172         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3173                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
3174                 return -EINVAL;
3175         }
3176
3177         sa = memdup_user(arg, sizeof(*sa));
3178         if (IS_ERR(sa))
3179                 return PTR_ERR(sa);
3180
3181         if (sa->flags & ~BTRFS_SCRUB_SUPPORTED_FLAGS) {
3182                 ret = -EOPNOTSUPP;
3183                 goto out;
3184         }
3185
3186         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3187                 ret = mnt_want_write_file(file);
3188                 if (ret)
3189                         goto out;
3190         }
3191
3192         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3193                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3194                               0);
3195
3196         /*
3197          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3198          * error. This is important as it allows user space to know how much
3199          * progress scrub has done. For example, if scrub is canceled we get
3200          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3201          * space. Later user space can inspect the progress from the structure
3202          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3203          * previously (btrfs-progs does this).
3204          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3205          * then return -EFAULT to signal the structure was not copied or it may
3206          * be corrupt and unreliable due to a partial copy.
3207          */
3208         if (copy_to_user(arg, sa, sizeof(*sa)))
3209                 ret = -EFAULT;
3210
3211         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3212                 mnt_drop_write_file(file);
3213 out:
3214         kfree(sa);
3215         return ret;
3216 }
3217
3218 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3219 {
3220         if (!capable(CAP_SYS_ADMIN))
3221                 return -EPERM;
3222
3223         return btrfs_scrub_cancel(fs_info);
3224 }
3225
3226 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3227                                        void __user *arg)
3228 {
3229         struct btrfs_ioctl_scrub_args *sa;
3230         int ret;
3231
3232         if (!capable(CAP_SYS_ADMIN))
3233                 return -EPERM;
3234
3235         sa = memdup_user(arg, sizeof(*sa));
3236         if (IS_ERR(sa))
3237                 return PTR_ERR(sa);
3238
3239         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3240
3241         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3242                 ret = -EFAULT;
3243
3244         kfree(sa);
3245         return ret;
3246 }
3247
3248 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3249                                       void __user *arg)
3250 {
3251         struct btrfs_ioctl_get_dev_stats *sa;
3252         int ret;
3253
3254         sa = memdup_user(arg, sizeof(*sa));
3255         if (IS_ERR(sa))
3256                 return PTR_ERR(sa);
3257
3258         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3259                 kfree(sa);
3260                 return -EPERM;
3261         }
3262
3263         ret = btrfs_get_dev_stats(fs_info, sa);
3264
3265         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3266                 ret = -EFAULT;
3267
3268         kfree(sa);
3269         return ret;
3270 }
3271
3272 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3273                                     void __user *arg)
3274 {
3275         struct btrfs_ioctl_dev_replace_args *p;
3276         int ret;
3277
3278         if (!capable(CAP_SYS_ADMIN))
3279                 return -EPERM;
3280
3281         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3282                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
3283                 return -EINVAL;
3284         }
3285
3286         p = memdup_user(arg, sizeof(*p));
3287         if (IS_ERR(p))
3288                 return PTR_ERR(p);
3289
3290         switch (p->cmd) {
3291         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3292                 if (sb_rdonly(fs_info->sb)) {
3293                         ret = -EROFS;
3294                         goto out;
3295                 }
3296                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3297                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3298                 } else {
3299                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3300                         btrfs_exclop_finish(fs_info);
3301                 }
3302                 break;
3303         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3304                 btrfs_dev_replace_status(fs_info, p);
3305                 ret = 0;
3306                 break;
3307         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3308                 p->result = btrfs_dev_replace_cancel(fs_info);
3309                 ret = 0;
3310                 break;
3311         default:
3312                 ret = -EINVAL;
3313                 break;
3314         }
3315
3316         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3317                 ret = -EFAULT;
3318 out:
3319         kfree(p);
3320         return ret;
3321 }
3322
3323 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3324 {
3325         int ret = 0;
3326         int i;
3327         u64 rel_ptr;
3328         int size;
3329         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3330         struct inode_fs_paths *ipath = NULL;
3331         struct btrfs_path *path;
3332
3333         if (!capable(CAP_DAC_READ_SEARCH))
3334                 return -EPERM;
3335
3336         path = btrfs_alloc_path();
3337         if (!path) {
3338                 ret = -ENOMEM;
3339                 goto out;
3340         }
3341
3342         ipa = memdup_user(arg, sizeof(*ipa));
3343         if (IS_ERR(ipa)) {
3344                 ret = PTR_ERR(ipa);
3345                 ipa = NULL;
3346                 goto out;
3347         }
3348
3349         size = min_t(u32, ipa->size, 4096);
3350         ipath = init_ipath(size, root, path);
3351         if (IS_ERR(ipath)) {
3352                 ret = PTR_ERR(ipath);
3353                 ipath = NULL;
3354                 goto out;
3355         }
3356
3357         ret = paths_from_inode(ipa->inum, ipath);
3358         if (ret < 0)
3359                 goto out;
3360
3361         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3362                 rel_ptr = ipath->fspath->val[i] -
3363                           (u64)(unsigned long)ipath->fspath->val;
3364                 ipath->fspath->val[i] = rel_ptr;
3365         }
3366
3367         btrfs_free_path(path);
3368         path = NULL;
3369         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3370                            ipath->fspath, size);
3371         if (ret) {
3372                 ret = -EFAULT;
3373                 goto out;
3374         }
3375
3376 out:
3377         btrfs_free_path(path);
3378         free_ipath(ipath);
3379         kfree(ipa);
3380
3381         return ret;
3382 }
3383
3384 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3385                                         void __user *arg, int version)
3386 {
3387         int ret = 0;
3388         int size;
3389         struct btrfs_ioctl_logical_ino_args *loi;
3390         struct btrfs_data_container *inodes = NULL;
3391         struct btrfs_path *path = NULL;
3392         bool ignore_offset;
3393
3394         if (!capable(CAP_SYS_ADMIN))
3395                 return -EPERM;
3396
3397         loi = memdup_user(arg, sizeof(*loi));
3398         if (IS_ERR(loi))
3399                 return PTR_ERR(loi);
3400
3401         if (version == 1) {
3402                 ignore_offset = false;
3403                 size = min_t(u32, loi->size, SZ_64K);
3404         } else {
3405                 /* All reserved bits must be 0 for now */
3406                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3407                         ret = -EINVAL;
3408                         goto out_loi;
3409                 }
3410                 /* Only accept flags we have defined so far */
3411                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3412                         ret = -EINVAL;
3413                         goto out_loi;
3414                 }
3415                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3416                 size = min_t(u32, loi->size, SZ_16M);
3417         }
3418
3419         inodes = init_data_container(size);
3420         if (IS_ERR(inodes)) {
3421                 ret = PTR_ERR(inodes);
3422                 goto out_loi;
3423         }
3424
3425         path = btrfs_alloc_path();
3426         if (!path) {
3427                 ret = -ENOMEM;
3428                 goto out;
3429         }
3430         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3431                                           inodes, ignore_offset);
3432         btrfs_free_path(path);
3433         if (ret == -EINVAL)
3434                 ret = -ENOENT;
3435         if (ret < 0)
3436                 goto out;
3437
3438         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3439                            size);
3440         if (ret)
3441                 ret = -EFAULT;
3442
3443 out:
3444         kvfree(inodes);
3445 out_loi:
3446         kfree(loi);
3447
3448         return ret;
3449 }
3450
3451 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3452                                struct btrfs_ioctl_balance_args *bargs)
3453 {
3454         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3455
3456         bargs->flags = bctl->flags;
3457
3458         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3459                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3460         if (atomic_read(&fs_info->balance_pause_req))
3461                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3462         if (atomic_read(&fs_info->balance_cancel_req))
3463                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3464
3465         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3466         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3467         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3468
3469         spin_lock(&fs_info->balance_lock);
3470         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3471         spin_unlock(&fs_info->balance_lock);
3472 }
3473
3474 /*
3475  * Try to acquire fs_info::balance_mutex as well as set BTRFS_EXLCOP_BALANCE as
3476  * required.
3477  *
3478  * @fs_info:       the filesystem
3479  * @excl_acquired: ptr to boolean value which is set to false in case balance
3480  *                 is being resumed
3481  *
3482  * Return 0 on success in which case both fs_info::balance is acquired as well
3483  * as exclusive ops are blocked. In case of failure return an error code.
3484  */
3485 static int btrfs_try_lock_balance(struct btrfs_fs_info *fs_info, bool *excl_acquired)
3486 {
3487         int ret;
3488
3489         /*
3490          * Exclusive operation is locked. Three possibilities:
3491          *   (1) some other op is running
3492          *   (2) balance is running
3493          *   (3) balance is paused -- special case (think resume)
3494          */
3495         while (1) {
3496                 if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
3497                         *excl_acquired = true;
3498                         mutex_lock(&fs_info->balance_mutex);
3499                         return 0;
3500                 }
3501
3502                 mutex_lock(&fs_info->balance_mutex);
3503                 if (fs_info->balance_ctl) {
3504                         /* This is either (2) or (3) */
3505                         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3506                                 /* This is (2) */
3507                                 ret = -EINPROGRESS;
3508                                 goto out_failure;
3509
3510                         } else {
3511                                 mutex_unlock(&fs_info->balance_mutex);
3512                                 /*
3513                                  * Lock released to allow other waiters to
3514                                  * continue, we'll reexamine the status again.
3515                                  */
3516                                 mutex_lock(&fs_info->balance_mutex);
3517
3518                                 if (fs_info->balance_ctl &&
3519                                     !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3520                                         /* This is (3) */
3521                                         *excl_acquired = false;
3522                                         return 0;
3523                                 }
3524                         }
3525                 } else {
3526                         /* This is (1) */
3527                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3528                         goto out_failure;
3529                 }
3530
3531                 mutex_unlock(&fs_info->balance_mutex);
3532         }
3533
3534 out_failure:
3535         mutex_unlock(&fs_info->balance_mutex);
3536         *excl_acquired = false;
3537         return ret;
3538 }
3539
3540 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3541 {
3542         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3543         struct btrfs_fs_info *fs_info = root->fs_info;
3544         struct btrfs_ioctl_balance_args *bargs;
3545         struct btrfs_balance_control *bctl;
3546         bool need_unlock = true;
3547         int ret;
3548
3549         if (!capable(CAP_SYS_ADMIN))
3550                 return -EPERM;
3551
3552         ret = mnt_want_write_file(file);
3553         if (ret)
3554                 return ret;
3555
3556         bargs = memdup_user(arg, sizeof(*bargs));
3557         if (IS_ERR(bargs)) {
3558                 ret = PTR_ERR(bargs);
3559                 bargs = NULL;
3560                 goto out;
3561         }
3562
3563         ret = btrfs_try_lock_balance(fs_info, &need_unlock);
3564         if (ret)
3565                 goto out;
3566
3567         lockdep_assert_held(&fs_info->balance_mutex);
3568
3569         if (bargs->flags & BTRFS_BALANCE_RESUME) {
3570                 if (!fs_info->balance_ctl) {
3571                         ret = -ENOTCONN;
3572                         goto out_unlock;
3573                 }
3574
3575                 bctl = fs_info->balance_ctl;
3576                 spin_lock(&fs_info->balance_lock);
3577                 bctl->flags |= BTRFS_BALANCE_RESUME;
3578                 spin_unlock(&fs_info->balance_lock);
3579                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
3580
3581                 goto do_balance;
3582         }
3583
3584         if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
3585                 ret = -EINVAL;
3586                 goto out_unlock;
3587         }
3588
3589         if (fs_info->balance_ctl) {
3590                 ret = -EINPROGRESS;
3591                 goto out_unlock;
3592         }
3593
3594         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
3595         if (!bctl) {
3596                 ret = -ENOMEM;
3597                 goto out_unlock;
3598         }
3599
3600         memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3601         memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3602         memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3603
3604         bctl->flags = bargs->flags;
3605 do_balance:
3606         /*
3607          * Ownership of bctl and exclusive operation goes to btrfs_balance.
3608          * bctl is freed in reset_balance_state, or, if restriper was paused
3609          * all the way until unmount, in free_fs_info.  The flag should be
3610          * cleared after reset_balance_state.
3611          */
3612         need_unlock = false;
3613
3614         ret = btrfs_balance(fs_info, bctl, bargs);
3615         bctl = NULL;
3616
3617         if (ret == 0 || ret == -ECANCELED) {
3618                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3619                         ret = -EFAULT;
3620         }
3621
3622         kfree(bctl);
3623 out_unlock:
3624         mutex_unlock(&fs_info->balance_mutex);
3625         if (need_unlock)
3626                 btrfs_exclop_finish(fs_info);
3627 out:
3628         mnt_drop_write_file(file);
3629         kfree(bargs);
3630         return ret;
3631 }
3632
3633 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
3634 {
3635         if (!capable(CAP_SYS_ADMIN))
3636                 return -EPERM;
3637
3638         switch (cmd) {
3639         case BTRFS_BALANCE_CTL_PAUSE:
3640                 return btrfs_pause_balance(fs_info);
3641         case BTRFS_BALANCE_CTL_CANCEL:
3642                 return btrfs_cancel_balance(fs_info);
3643         }
3644
3645         return -EINVAL;
3646 }
3647
3648 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
3649                                          void __user *arg)
3650 {
3651         struct btrfs_ioctl_balance_args *bargs;
3652         int ret = 0;
3653
3654         if (!capable(CAP_SYS_ADMIN))
3655                 return -EPERM;
3656
3657         mutex_lock(&fs_info->balance_mutex);
3658         if (!fs_info->balance_ctl) {
3659                 ret = -ENOTCONN;
3660                 goto out;
3661         }
3662
3663         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
3664         if (!bargs) {
3665                 ret = -ENOMEM;
3666                 goto out;
3667         }
3668
3669         btrfs_update_ioctl_balance_args(fs_info, bargs);
3670
3671         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3672                 ret = -EFAULT;
3673
3674         kfree(bargs);
3675 out:
3676         mutex_unlock(&fs_info->balance_mutex);
3677         return ret;
3678 }
3679
3680 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
3681 {
3682         struct inode *inode = file_inode(file);
3683         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3684         struct btrfs_ioctl_quota_ctl_args *sa;
3685         int ret;
3686
3687         if (!capable(CAP_SYS_ADMIN))
3688                 return -EPERM;
3689
3690         ret = mnt_want_write_file(file);
3691         if (ret)
3692                 return ret;
3693
3694         sa = memdup_user(arg, sizeof(*sa));
3695         if (IS_ERR(sa)) {
3696                 ret = PTR_ERR(sa);
3697                 goto drop_write;
3698         }
3699
3700         down_write(&fs_info->subvol_sem);
3701
3702         switch (sa->cmd) {
3703         case BTRFS_QUOTA_CTL_ENABLE:
3704         case BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA:
3705                 ret = btrfs_quota_enable(fs_info, sa);
3706                 break;
3707         case BTRFS_QUOTA_CTL_DISABLE:
3708                 ret = btrfs_quota_disable(fs_info);
3709                 break;
3710         default:
3711                 ret = -EINVAL;
3712                 break;
3713         }
3714
3715         kfree(sa);
3716         up_write(&fs_info->subvol_sem);
3717 drop_write:
3718         mnt_drop_write_file(file);
3719         return ret;
3720 }
3721
3722 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
3723 {
3724         struct inode *inode = file_inode(file);
3725         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3726         struct btrfs_root *root = BTRFS_I(inode)->root;
3727         struct btrfs_ioctl_qgroup_assign_args *sa;
3728         struct btrfs_trans_handle *trans;
3729         int ret;
3730         int err;
3731
3732         if (!capable(CAP_SYS_ADMIN))
3733                 return -EPERM;
3734
3735         ret = mnt_want_write_file(file);
3736         if (ret)
3737                 return ret;
3738
3739         sa = memdup_user(arg, sizeof(*sa));
3740         if (IS_ERR(sa)) {
3741                 ret = PTR_ERR(sa);
3742                 goto drop_write;
3743         }
3744
3745         trans = btrfs_join_transaction(root);
3746         if (IS_ERR(trans)) {
3747                 ret = PTR_ERR(trans);
3748                 goto out;
3749         }
3750
3751         if (sa->assign) {
3752                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
3753         } else {
3754                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
3755         }
3756
3757         /* update qgroup status and info */
3758         mutex_lock(&fs_info->qgroup_ioctl_lock);
3759         err = btrfs_run_qgroups(trans);
3760         mutex_unlock(&fs_info->qgroup_ioctl_lock);
3761         if (err < 0)
3762                 btrfs_handle_fs_error(fs_info, err,
3763                                       "failed to update qgroup status and info");
3764         err = btrfs_end_transaction(trans);
3765         if (err && !ret)
3766                 ret = err;
3767
3768 out:
3769         kfree(sa);
3770 drop_write:
3771         mnt_drop_write_file(file);
3772         return ret;
3773 }
3774
3775 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
3776 {
3777         struct inode *inode = file_inode(file);
3778         struct btrfs_root *root = BTRFS_I(inode)->root;
3779         struct btrfs_ioctl_qgroup_create_args *sa;
3780         struct btrfs_trans_handle *trans;
3781         int ret;
3782         int err;
3783
3784         if (!capable(CAP_SYS_ADMIN))
3785                 return -EPERM;
3786
3787         ret = mnt_want_write_file(file);
3788         if (ret)
3789                 return ret;
3790
3791         sa = memdup_user(arg, sizeof(*sa));
3792         if (IS_ERR(sa)) {
3793                 ret = PTR_ERR(sa);
3794                 goto drop_write;
3795         }
3796
3797         if (!sa->qgroupid) {
3798                 ret = -EINVAL;
3799                 goto out;
3800         }
3801
3802         trans = btrfs_join_transaction(root);
3803         if (IS_ERR(trans)) {
3804                 ret = PTR_ERR(trans);
3805                 goto out;
3806         }
3807
3808         if (sa->create) {
3809                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
3810         } else {
3811                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
3812         }
3813
3814         err = btrfs_end_transaction(trans);
3815         if (err && !ret)
3816                 ret = err;
3817
3818 out:
3819         kfree(sa);
3820 drop_write:
3821         mnt_drop_write_file(file);
3822         return ret;
3823 }
3824
3825 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
3826 {
3827         struct inode *inode = file_inode(file);
3828         struct btrfs_root *root = BTRFS_I(inode)->root;
3829         struct btrfs_ioctl_qgroup_limit_args *sa;
3830         struct btrfs_trans_handle *trans;
3831         int ret;
3832         int err;
3833         u64 qgroupid;
3834
3835         if (!capable(CAP_SYS_ADMIN))
3836                 return -EPERM;
3837
3838         ret = mnt_want_write_file(file);
3839         if (ret)
3840                 return ret;
3841
3842         sa = memdup_user(arg, sizeof(*sa));
3843         if (IS_ERR(sa)) {
3844                 ret = PTR_ERR(sa);
3845                 goto drop_write;
3846         }
3847
3848         trans = btrfs_join_transaction(root);
3849         if (IS_ERR(trans)) {
3850                 ret = PTR_ERR(trans);
3851                 goto out;
3852         }
3853
3854         qgroupid = sa->qgroupid;
3855         if (!qgroupid) {
3856                 /* take the current subvol as qgroup */
3857                 qgroupid = root->root_key.objectid;
3858         }
3859
3860         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
3861
3862         err = btrfs_end_transaction(trans);
3863         if (err && !ret)
3864                 ret = err;
3865
3866 out:
3867         kfree(sa);
3868 drop_write:
3869         mnt_drop_write_file(file);
3870         return ret;
3871 }
3872
3873 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
3874 {
3875         struct inode *inode = file_inode(file);
3876         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3877         struct btrfs_ioctl_quota_rescan_args *qsa;
3878         int ret;
3879
3880         if (!capable(CAP_SYS_ADMIN))
3881                 return -EPERM;
3882
3883         ret = mnt_want_write_file(file);
3884         if (ret)
3885                 return ret;
3886
3887         qsa = memdup_user(arg, sizeof(*qsa));
3888         if (IS_ERR(qsa)) {
3889                 ret = PTR_ERR(qsa);
3890                 goto drop_write;
3891         }
3892
3893         if (qsa->flags) {
3894                 ret = -EINVAL;
3895                 goto out;
3896         }
3897
3898         ret = btrfs_qgroup_rescan(fs_info);
3899
3900 out:
3901         kfree(qsa);
3902 drop_write:
3903         mnt_drop_write_file(file);
3904         return ret;
3905 }
3906
3907 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
3908                                                 void __user *arg)
3909 {
3910         struct btrfs_ioctl_quota_rescan_args qsa = {0};
3911
3912         if (!capable(CAP_SYS_ADMIN))
3913                 return -EPERM;
3914
3915         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3916                 qsa.flags = 1;
3917                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
3918         }
3919
3920         if (copy_to_user(arg, &qsa, sizeof(qsa)))
3921                 return -EFAULT;
3922
3923         return 0;
3924 }
3925
3926 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
3927                                                 void __user *arg)
3928 {
3929         if (!capable(CAP_SYS_ADMIN))
3930                 return -EPERM;
3931
3932         return btrfs_qgroup_wait_for_completion(fs_info, true);
3933 }
3934
3935 static long _btrfs_ioctl_set_received_subvol(struct file *file,
3936                                             struct mnt_idmap *idmap,
3937                                             struct btrfs_ioctl_received_subvol_args *sa)
3938 {
3939         struct inode *inode = file_inode(file);
3940         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3941         struct btrfs_root *root = BTRFS_I(inode)->root;
3942         struct btrfs_root_item *root_item = &root->root_item;
3943         struct btrfs_trans_handle *trans;
3944         struct timespec64 ct = current_time(inode);
3945         int ret = 0;
3946         int received_uuid_changed;
3947
3948         if (!inode_owner_or_capable(idmap, inode))
3949                 return -EPERM;
3950
3951         ret = mnt_want_write_file(file);
3952         if (ret < 0)
3953                 return ret;
3954
3955         down_write(&fs_info->subvol_sem);
3956
3957         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3958                 ret = -EINVAL;
3959                 goto out;
3960         }
3961
3962         if (btrfs_root_readonly(root)) {
3963                 ret = -EROFS;
3964                 goto out;
3965         }
3966
3967         /*
3968          * 1 - root item
3969          * 2 - uuid items (received uuid + subvol uuid)
3970          */
3971         trans = btrfs_start_transaction(root, 3);
3972         if (IS_ERR(trans)) {
3973                 ret = PTR_ERR(trans);
3974                 trans = NULL;
3975                 goto out;
3976         }
3977
3978         sa->rtransid = trans->transid;
3979         sa->rtime.sec = ct.tv_sec;
3980         sa->rtime.nsec = ct.tv_nsec;
3981
3982         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
3983                                        BTRFS_UUID_SIZE);
3984         if (received_uuid_changed &&
3985             !btrfs_is_empty_uuid(root_item->received_uuid)) {
3986                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
3987                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3988                                           root->root_key.objectid);
3989                 if (ret && ret != -ENOENT) {
3990                         btrfs_abort_transaction(trans, ret);
3991                         btrfs_end_transaction(trans);
3992                         goto out;
3993                 }
3994         }
3995         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
3996         btrfs_set_root_stransid(root_item, sa->stransid);
3997         btrfs_set_root_rtransid(root_item, sa->rtransid);
3998         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
3999         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4000         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4001         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4002
4003         ret = btrfs_update_root(trans, fs_info->tree_root,
4004                                 &root->root_key, &root->root_item);
4005         if (ret < 0) {
4006                 btrfs_end_transaction(trans);
4007                 goto out;
4008         }
4009         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4010                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4011                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4012                                           root->root_key.objectid);
4013                 if (ret < 0 && ret != -EEXIST) {
4014                         btrfs_abort_transaction(trans, ret);
4015                         btrfs_end_transaction(trans);
4016                         goto out;
4017                 }
4018         }
4019         ret = btrfs_commit_transaction(trans);
4020 out:
4021         up_write(&fs_info->subvol_sem);
4022         mnt_drop_write_file(file);
4023         return ret;
4024 }
4025
4026 #ifdef CONFIG_64BIT
4027 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4028                                                 void __user *arg)
4029 {
4030         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4031         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4032         int ret = 0;
4033
4034         args32 = memdup_user(arg, sizeof(*args32));
4035         if (IS_ERR(args32))
4036                 return PTR_ERR(args32);
4037
4038         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4039         if (!args64) {
4040                 ret = -ENOMEM;
4041                 goto out;
4042         }
4043
4044         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4045         args64->stransid = args32->stransid;
4046         args64->rtransid = args32->rtransid;
4047         args64->stime.sec = args32->stime.sec;
4048         args64->stime.nsec = args32->stime.nsec;
4049         args64->rtime.sec = args32->rtime.sec;
4050         args64->rtime.nsec = args32->rtime.nsec;
4051         args64->flags = args32->flags;
4052
4053         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), args64);
4054         if (ret)
4055                 goto out;
4056
4057         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4058         args32->stransid = args64->stransid;
4059         args32->rtransid = args64->rtransid;
4060         args32->stime.sec = args64->stime.sec;
4061         args32->stime.nsec = args64->stime.nsec;
4062         args32->rtime.sec = args64->rtime.sec;
4063         args32->rtime.nsec = args64->rtime.nsec;
4064         args32->flags = args64->flags;
4065
4066         ret = copy_to_user(arg, args32, sizeof(*args32));
4067         if (ret)
4068                 ret = -EFAULT;
4069
4070 out:
4071         kfree(args32);
4072         kfree(args64);
4073         return ret;
4074 }
4075 #endif
4076
4077 static long btrfs_ioctl_set_received_subvol(struct file *file,
4078                                             void __user *arg)
4079 {
4080         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4081         int ret = 0;
4082
4083         sa = memdup_user(arg, sizeof(*sa));
4084         if (IS_ERR(sa))
4085                 return PTR_ERR(sa);
4086
4087         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), sa);
4088
4089         if (ret)
4090                 goto out;
4091
4092         ret = copy_to_user(arg, sa, sizeof(*sa));
4093         if (ret)
4094                 ret = -EFAULT;
4095
4096 out:
4097         kfree(sa);
4098         return ret;
4099 }
4100
4101 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4102                                         void __user *arg)
4103 {
4104         size_t len;
4105         int ret;
4106         char label[BTRFS_LABEL_SIZE];
4107
4108         spin_lock(&fs_info->super_lock);
4109         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4110         spin_unlock(&fs_info->super_lock);
4111
4112         len = strnlen(label, BTRFS_LABEL_SIZE);
4113
4114         if (len == BTRFS_LABEL_SIZE) {
4115                 btrfs_warn(fs_info,
4116                            "label is too long, return the first %zu bytes",
4117                            --len);
4118         }
4119
4120         ret = copy_to_user(arg, label, len);
4121
4122         return ret ? -EFAULT : 0;
4123 }
4124
4125 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4126 {
4127         struct inode *inode = file_inode(file);
4128         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4129         struct btrfs_root *root = BTRFS_I(inode)->root;
4130         struct btrfs_super_block *super_block = fs_info->super_copy;
4131         struct btrfs_trans_handle *trans;
4132         char label[BTRFS_LABEL_SIZE];
4133         int ret;
4134
4135         if (!capable(CAP_SYS_ADMIN))
4136                 return -EPERM;
4137
4138         if (copy_from_user(label, arg, sizeof(label)))
4139                 return -EFAULT;
4140
4141         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4142                 btrfs_err(fs_info,
4143                           "unable to set label with more than %d bytes",
4144                           BTRFS_LABEL_SIZE - 1);
4145                 return -EINVAL;
4146         }
4147
4148         ret = mnt_want_write_file(file);
4149         if (ret)
4150                 return ret;
4151
4152         trans = btrfs_start_transaction(root, 0);
4153         if (IS_ERR(trans)) {
4154                 ret = PTR_ERR(trans);
4155                 goto out_unlock;
4156         }
4157
4158         spin_lock(&fs_info->super_lock);
4159         strcpy(super_block->label, label);
4160         spin_unlock(&fs_info->super_lock);
4161         ret = btrfs_commit_transaction(trans);
4162
4163 out_unlock:
4164         mnt_drop_write_file(file);
4165         return ret;
4166 }
4167
4168 #define INIT_FEATURE_FLAGS(suffix) \
4169         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4170           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4171           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4172
4173 int btrfs_ioctl_get_supported_features(void __user *arg)
4174 {
4175         static const struct btrfs_ioctl_feature_flags features[3] = {
4176                 INIT_FEATURE_FLAGS(SUPP),
4177                 INIT_FEATURE_FLAGS(SAFE_SET),
4178                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4179         };
4180
4181         if (copy_to_user(arg, &features, sizeof(features)))
4182                 return -EFAULT;
4183
4184         return 0;
4185 }
4186
4187 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4188                                         void __user *arg)
4189 {
4190         struct btrfs_super_block *super_block = fs_info->super_copy;
4191         struct btrfs_ioctl_feature_flags features;
4192
4193         features.compat_flags = btrfs_super_compat_flags(super_block);
4194         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4195         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4196
4197         if (copy_to_user(arg, &features, sizeof(features)))
4198                 return -EFAULT;
4199
4200         return 0;
4201 }
4202
4203 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4204                               enum btrfs_feature_set set,
4205                               u64 change_mask, u64 flags, u64 supported_flags,
4206                               u64 safe_set, u64 safe_clear)
4207 {
4208         const char *type = btrfs_feature_set_name(set);
4209         char *names;
4210         u64 disallowed, unsupported;
4211         u64 set_mask = flags & change_mask;
4212         u64 clear_mask = ~flags & change_mask;
4213
4214         unsupported = set_mask & ~supported_flags;
4215         if (unsupported) {
4216                 names = btrfs_printable_features(set, unsupported);
4217                 if (names) {
4218                         btrfs_warn(fs_info,
4219                                    "this kernel does not support the %s feature bit%s",
4220                                    names, strchr(names, ',') ? "s" : "");
4221                         kfree(names);
4222                 } else
4223                         btrfs_warn(fs_info,
4224                                    "this kernel does not support %s bits 0x%llx",
4225                                    type, unsupported);
4226                 return -EOPNOTSUPP;
4227         }
4228
4229         disallowed = set_mask & ~safe_set;
4230         if (disallowed) {
4231                 names = btrfs_printable_features(set, disallowed);
4232                 if (names) {
4233                         btrfs_warn(fs_info,
4234                                    "can't set the %s feature bit%s while mounted",
4235                                    names, strchr(names, ',') ? "s" : "");
4236                         kfree(names);
4237                 } else
4238                         btrfs_warn(fs_info,
4239                                    "can't set %s bits 0x%llx while mounted",
4240                                    type, disallowed);
4241                 return -EPERM;
4242         }
4243
4244         disallowed = clear_mask & ~safe_clear;
4245         if (disallowed) {
4246                 names = btrfs_printable_features(set, disallowed);
4247                 if (names) {
4248                         btrfs_warn(fs_info,
4249                                    "can't clear the %s feature bit%s while mounted",
4250                                    names, strchr(names, ',') ? "s" : "");
4251                         kfree(names);
4252                 } else
4253                         btrfs_warn(fs_info,
4254                                    "can't clear %s bits 0x%llx while mounted",
4255                                    type, disallowed);
4256                 return -EPERM;
4257         }
4258
4259         return 0;
4260 }
4261
4262 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4263 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4264                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4265                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4266                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4267
4268 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4269 {
4270         struct inode *inode = file_inode(file);
4271         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4272         struct btrfs_root *root = BTRFS_I(inode)->root;
4273         struct btrfs_super_block *super_block = fs_info->super_copy;
4274         struct btrfs_ioctl_feature_flags flags[2];
4275         struct btrfs_trans_handle *trans;
4276         u64 newflags;
4277         int ret;
4278
4279         if (!capable(CAP_SYS_ADMIN))
4280                 return -EPERM;
4281
4282         if (copy_from_user(flags, arg, sizeof(flags)))
4283                 return -EFAULT;
4284
4285         /* Nothing to do */
4286         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4287             !flags[0].incompat_flags)
4288                 return 0;
4289
4290         ret = check_feature(fs_info, flags[0].compat_flags,
4291                             flags[1].compat_flags, COMPAT);
4292         if (ret)
4293                 return ret;
4294
4295         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4296                             flags[1].compat_ro_flags, COMPAT_RO);
4297         if (ret)
4298                 return ret;
4299
4300         ret = check_feature(fs_info, flags[0].incompat_flags,
4301                             flags[1].incompat_flags, INCOMPAT);
4302         if (ret)
4303                 return ret;
4304
4305         ret = mnt_want_write_file(file);
4306         if (ret)
4307                 return ret;
4308
4309         trans = btrfs_start_transaction(root, 0);
4310         if (IS_ERR(trans)) {
4311                 ret = PTR_ERR(trans);
4312                 goto out_drop_write;
4313         }
4314
4315         spin_lock(&fs_info->super_lock);
4316         newflags = btrfs_super_compat_flags(super_block);
4317         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4318         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4319         btrfs_set_super_compat_flags(super_block, newflags);
4320
4321         newflags = btrfs_super_compat_ro_flags(super_block);
4322         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4323         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4324         btrfs_set_super_compat_ro_flags(super_block, newflags);
4325
4326         newflags = btrfs_super_incompat_flags(super_block);
4327         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4328         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4329         btrfs_set_super_incompat_flags(super_block, newflags);
4330         spin_unlock(&fs_info->super_lock);
4331
4332         ret = btrfs_commit_transaction(trans);
4333 out_drop_write:
4334         mnt_drop_write_file(file);
4335
4336         return ret;
4337 }
4338
4339 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
4340 {
4341         struct btrfs_ioctl_send_args *arg;
4342         int ret;
4343
4344         if (compat) {
4345 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4346                 struct btrfs_ioctl_send_args_32 args32 = { 0 };
4347
4348                 ret = copy_from_user(&args32, argp, sizeof(args32));
4349                 if (ret)
4350                         return -EFAULT;
4351                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4352                 if (!arg)
4353                         return -ENOMEM;
4354                 arg->send_fd = args32.send_fd;
4355                 arg->clone_sources_count = args32.clone_sources_count;
4356                 arg->clone_sources = compat_ptr(args32.clone_sources);
4357                 arg->parent_root = args32.parent_root;
4358                 arg->flags = args32.flags;
4359                 memcpy(arg->reserved, args32.reserved,
4360                        sizeof(args32.reserved));
4361 #else
4362                 return -ENOTTY;
4363 #endif
4364         } else {
4365                 arg = memdup_user(argp, sizeof(*arg));
4366                 if (IS_ERR(arg))
4367                         return PTR_ERR(arg);
4368         }
4369         ret = btrfs_ioctl_send(inode, arg);
4370         kfree(arg);
4371         return ret;
4372 }
4373
4374 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
4375                                     bool compat)
4376 {
4377         struct btrfs_ioctl_encoded_io_args args = { 0 };
4378         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
4379                                              flags);
4380         size_t copy_end;
4381         struct iovec iovstack[UIO_FASTIOV];
4382         struct iovec *iov = iovstack;
4383         struct iov_iter iter;
4384         loff_t pos;
4385         struct kiocb kiocb;
4386         ssize_t ret;
4387
4388         if (!capable(CAP_SYS_ADMIN)) {
4389                 ret = -EPERM;
4390                 goto out_acct;
4391         }
4392
4393         if (compat) {
4394 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4395                 struct btrfs_ioctl_encoded_io_args_32 args32;
4396
4397                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
4398                                        flags);
4399                 if (copy_from_user(&args32, argp, copy_end)) {
4400                         ret = -EFAULT;
4401                         goto out_acct;
4402                 }
4403                 args.iov = compat_ptr(args32.iov);
4404                 args.iovcnt = args32.iovcnt;
4405                 args.offset = args32.offset;
4406                 args.flags = args32.flags;
4407 #else
4408                 return -ENOTTY;
4409 #endif
4410         } else {
4411                 copy_end = copy_end_kernel;
4412                 if (copy_from_user(&args, argp, copy_end)) {
4413                         ret = -EFAULT;
4414                         goto out_acct;
4415                 }
4416         }
4417         if (args.flags != 0) {
4418                 ret = -EINVAL;
4419                 goto out_acct;
4420         }
4421
4422         ret = import_iovec(ITER_DEST, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4423                            &iov, &iter);
4424         if (ret < 0)
4425                 goto out_acct;
4426
4427         if (iov_iter_count(&iter) == 0) {
4428                 ret = 0;
4429                 goto out_iov;
4430         }
4431         pos = args.offset;
4432         ret = rw_verify_area(READ, file, &pos, args.len);
4433         if (ret < 0)
4434                 goto out_iov;
4435
4436         init_sync_kiocb(&kiocb, file);
4437         kiocb.ki_pos = pos;
4438
4439         ret = btrfs_encoded_read(&kiocb, &iter, &args);
4440         if (ret >= 0) {
4441                 fsnotify_access(file);
4442                 if (copy_to_user(argp + copy_end,
4443                                  (char *)&args + copy_end_kernel,
4444                                  sizeof(args) - copy_end_kernel))
4445                         ret = -EFAULT;
4446         }
4447
4448 out_iov:
4449         kfree(iov);
4450 out_acct:
4451         if (ret > 0)
4452                 add_rchar(current, ret);
4453         inc_syscr(current);
4454         return ret;
4455 }
4456
4457 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
4458 {
4459         struct btrfs_ioctl_encoded_io_args args;
4460         struct iovec iovstack[UIO_FASTIOV];
4461         struct iovec *iov = iovstack;
4462         struct iov_iter iter;
4463         loff_t pos;
4464         struct kiocb kiocb;
4465         ssize_t ret;
4466
4467         if (!capable(CAP_SYS_ADMIN)) {
4468                 ret = -EPERM;
4469                 goto out_acct;
4470         }
4471
4472         if (!(file->f_mode & FMODE_WRITE)) {
4473                 ret = -EBADF;
4474                 goto out_acct;
4475         }
4476
4477         if (compat) {
4478 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4479                 struct btrfs_ioctl_encoded_io_args_32 args32;
4480
4481                 if (copy_from_user(&args32, argp, sizeof(args32))) {
4482                         ret = -EFAULT;
4483                         goto out_acct;
4484                 }
4485                 args.iov = compat_ptr(args32.iov);
4486                 args.iovcnt = args32.iovcnt;
4487                 args.offset = args32.offset;
4488                 args.flags = args32.flags;
4489                 args.len = args32.len;
4490                 args.unencoded_len = args32.unencoded_len;
4491                 args.unencoded_offset = args32.unencoded_offset;
4492                 args.compression = args32.compression;
4493                 args.encryption = args32.encryption;
4494                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
4495 #else
4496                 return -ENOTTY;
4497 #endif
4498         } else {
4499                 if (copy_from_user(&args, argp, sizeof(args))) {
4500                         ret = -EFAULT;
4501                         goto out_acct;
4502                 }
4503         }
4504
4505         ret = -EINVAL;
4506         if (args.flags != 0)
4507                 goto out_acct;
4508         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
4509                 goto out_acct;
4510         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
4511             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
4512                 goto out_acct;
4513         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
4514             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
4515                 goto out_acct;
4516         if (args.unencoded_offset > args.unencoded_len)
4517                 goto out_acct;
4518         if (args.len > args.unencoded_len - args.unencoded_offset)
4519                 goto out_acct;
4520
4521         ret = import_iovec(ITER_SOURCE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4522                            &iov, &iter);
4523         if (ret < 0)
4524                 goto out_acct;
4525
4526         file_start_write(file);
4527
4528         if (iov_iter_count(&iter) == 0) {
4529                 ret = 0;
4530                 goto out_end_write;
4531         }
4532         pos = args.offset;
4533         ret = rw_verify_area(WRITE, file, &pos, args.len);
4534         if (ret < 0)
4535                 goto out_end_write;
4536
4537         init_sync_kiocb(&kiocb, file);
4538         ret = kiocb_set_rw_flags(&kiocb, 0);
4539         if (ret)
4540                 goto out_end_write;
4541         kiocb.ki_pos = pos;
4542
4543         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
4544         if (ret > 0)
4545                 fsnotify_modify(file);
4546
4547 out_end_write:
4548         file_end_write(file);
4549         kfree(iov);
4550 out_acct:
4551         if (ret > 0)
4552                 add_wchar(current, ret);
4553         inc_syscw(current);
4554         return ret;
4555 }
4556
4557 long btrfs_ioctl(struct file *file, unsigned int
4558                 cmd, unsigned long arg)
4559 {
4560         struct inode *inode = file_inode(file);
4561         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4562         struct btrfs_root *root = BTRFS_I(inode)->root;
4563         void __user *argp = (void __user *)arg;
4564
4565         switch (cmd) {
4566         case FS_IOC_GETVERSION:
4567                 return btrfs_ioctl_getversion(inode, argp);
4568         case FS_IOC_GETFSLABEL:
4569                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4570         case FS_IOC_SETFSLABEL:
4571                 return btrfs_ioctl_set_fslabel(file, argp);
4572         case FITRIM:
4573                 return btrfs_ioctl_fitrim(fs_info, argp);
4574         case BTRFS_IOC_SNAP_CREATE:
4575                 return btrfs_ioctl_snap_create(file, argp, 0);
4576         case BTRFS_IOC_SNAP_CREATE_V2:
4577                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4578         case BTRFS_IOC_SUBVOL_CREATE:
4579                 return btrfs_ioctl_snap_create(file, argp, 1);
4580         case BTRFS_IOC_SUBVOL_CREATE_V2:
4581                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4582         case BTRFS_IOC_SNAP_DESTROY:
4583                 return btrfs_ioctl_snap_destroy(file, argp, false);
4584         case BTRFS_IOC_SNAP_DESTROY_V2:
4585                 return btrfs_ioctl_snap_destroy(file, argp, true);
4586         case BTRFS_IOC_SUBVOL_GETFLAGS:
4587                 return btrfs_ioctl_subvol_getflags(inode, argp);
4588         case BTRFS_IOC_SUBVOL_SETFLAGS:
4589                 return btrfs_ioctl_subvol_setflags(file, argp);
4590         case BTRFS_IOC_DEFAULT_SUBVOL:
4591                 return btrfs_ioctl_default_subvol(file, argp);
4592         case BTRFS_IOC_DEFRAG:
4593                 return btrfs_ioctl_defrag(file, NULL);
4594         case BTRFS_IOC_DEFRAG_RANGE:
4595                 return btrfs_ioctl_defrag(file, argp);
4596         case BTRFS_IOC_RESIZE:
4597                 return btrfs_ioctl_resize(file, argp);
4598         case BTRFS_IOC_ADD_DEV:
4599                 return btrfs_ioctl_add_dev(fs_info, argp);
4600         case BTRFS_IOC_RM_DEV:
4601                 return btrfs_ioctl_rm_dev(file, argp);
4602         case BTRFS_IOC_RM_DEV_V2:
4603                 return btrfs_ioctl_rm_dev_v2(file, argp);
4604         case BTRFS_IOC_FS_INFO:
4605                 return btrfs_ioctl_fs_info(fs_info, argp);
4606         case BTRFS_IOC_DEV_INFO:
4607                 return btrfs_ioctl_dev_info(fs_info, argp);
4608         case BTRFS_IOC_TREE_SEARCH:
4609                 return btrfs_ioctl_tree_search(inode, argp);
4610         case BTRFS_IOC_TREE_SEARCH_V2:
4611                 return btrfs_ioctl_tree_search_v2(inode, argp);
4612         case BTRFS_IOC_INO_LOOKUP:
4613                 return btrfs_ioctl_ino_lookup(root, argp);
4614         case BTRFS_IOC_INO_PATHS:
4615                 return btrfs_ioctl_ino_to_path(root, argp);
4616         case BTRFS_IOC_LOGICAL_INO:
4617                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4618         case BTRFS_IOC_LOGICAL_INO_V2:
4619                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4620         case BTRFS_IOC_SPACE_INFO:
4621                 return btrfs_ioctl_space_info(fs_info, argp);
4622         case BTRFS_IOC_SYNC: {
4623                 int ret;
4624
4625                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4626                 if (ret)
4627                         return ret;
4628                 ret = btrfs_sync_fs(inode->i_sb, 1);
4629                 /*
4630                  * The transaction thread may want to do more work,
4631                  * namely it pokes the cleaner kthread that will start
4632                  * processing uncleaned subvols.
4633                  */
4634                 wake_up_process(fs_info->transaction_kthread);
4635                 return ret;
4636         }
4637         case BTRFS_IOC_START_SYNC:
4638                 return btrfs_ioctl_start_sync(root, argp);
4639         case BTRFS_IOC_WAIT_SYNC:
4640                 return btrfs_ioctl_wait_sync(fs_info, argp);
4641         case BTRFS_IOC_SCRUB:
4642                 return btrfs_ioctl_scrub(file, argp);
4643         case BTRFS_IOC_SCRUB_CANCEL:
4644                 return btrfs_ioctl_scrub_cancel(fs_info);
4645         case BTRFS_IOC_SCRUB_PROGRESS:
4646                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4647         case BTRFS_IOC_BALANCE_V2:
4648                 return btrfs_ioctl_balance(file, argp);
4649         case BTRFS_IOC_BALANCE_CTL:
4650                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4651         case BTRFS_IOC_BALANCE_PROGRESS:
4652                 return btrfs_ioctl_balance_progress(fs_info, argp);
4653         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4654                 return btrfs_ioctl_set_received_subvol(file, argp);
4655 #ifdef CONFIG_64BIT
4656         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4657                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4658 #endif
4659         case BTRFS_IOC_SEND:
4660                 return _btrfs_ioctl_send(inode, argp, false);
4661 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4662         case BTRFS_IOC_SEND_32:
4663                 return _btrfs_ioctl_send(inode, argp, true);
4664 #endif
4665         case BTRFS_IOC_GET_DEV_STATS:
4666                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4667         case BTRFS_IOC_QUOTA_CTL:
4668                 return btrfs_ioctl_quota_ctl(file, argp);
4669         case BTRFS_IOC_QGROUP_ASSIGN:
4670                 return btrfs_ioctl_qgroup_assign(file, argp);
4671         case BTRFS_IOC_QGROUP_CREATE:
4672                 return btrfs_ioctl_qgroup_create(file, argp);
4673         case BTRFS_IOC_QGROUP_LIMIT:
4674                 return btrfs_ioctl_qgroup_limit(file, argp);
4675         case BTRFS_IOC_QUOTA_RESCAN:
4676                 return btrfs_ioctl_quota_rescan(file, argp);
4677         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4678                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4679         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4680                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4681         case BTRFS_IOC_DEV_REPLACE:
4682                 return btrfs_ioctl_dev_replace(fs_info, argp);
4683         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4684                 return btrfs_ioctl_get_supported_features(argp);
4685         case BTRFS_IOC_GET_FEATURES:
4686                 return btrfs_ioctl_get_features(fs_info, argp);
4687         case BTRFS_IOC_SET_FEATURES:
4688                 return btrfs_ioctl_set_features(file, argp);
4689         case BTRFS_IOC_GET_SUBVOL_INFO:
4690                 return btrfs_ioctl_get_subvol_info(inode, argp);
4691         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4692                 return btrfs_ioctl_get_subvol_rootref(root, argp);
4693         case BTRFS_IOC_INO_LOOKUP_USER:
4694                 return btrfs_ioctl_ino_lookup_user(file, argp);
4695         case FS_IOC_ENABLE_VERITY:
4696                 return fsverity_ioctl_enable(file, (const void __user *)argp);
4697         case FS_IOC_MEASURE_VERITY:
4698                 return fsverity_ioctl_measure(file, argp);
4699         case BTRFS_IOC_ENCODED_READ:
4700                 return btrfs_ioctl_encoded_read(file, argp, false);
4701         case BTRFS_IOC_ENCODED_WRITE:
4702                 return btrfs_ioctl_encoded_write(file, argp, false);
4703 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4704         case BTRFS_IOC_ENCODED_READ_32:
4705                 return btrfs_ioctl_encoded_read(file, argp, true);
4706         case BTRFS_IOC_ENCODED_WRITE_32:
4707                 return btrfs_ioctl_encoded_write(file, argp, true);
4708 #endif
4709         }
4710
4711         return -ENOTTY;
4712 }
4713
4714 #ifdef CONFIG_COMPAT
4715 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4716 {
4717         /*
4718          * These all access 32-bit values anyway so no further
4719          * handling is necessary.
4720          */
4721         switch (cmd) {
4722         case FS_IOC32_GETVERSION:
4723                 cmd = FS_IOC_GETVERSION;
4724                 break;
4725         }
4726
4727         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
4728 }
4729 #endif