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 if (btrfs_ino(BTRFS_I(src_inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1294                         /*
1295                          * Snapshots must be made with the src_inode referring
1296                          * to the subvolume inode, otherwise the permission
1297                          * checking above is useless because we may have
1298                          * permission on a lower directory but not the subvol
1299                          * itself.
1300                          */
1301                         ret = -EINVAL;
1302                 } else {
1303                         ret = btrfs_mksnapshot(&file->f_path, idmap,
1304                                                name, namelen,
1305                                                BTRFS_I(src_inode)->root,
1306                                                readonly, inherit);
1307                 }
1308                 fdput(src);
1309         }
1310 out_drop_write:
1311         mnt_drop_write_file(file);
1312 out:
1313         return ret;
1314 }
1315
1316 static noinline int btrfs_ioctl_snap_create(struct file *file,
1317                                             void __user *arg, int subvol)
1318 {
1319         struct btrfs_ioctl_vol_args *vol_args;
1320         int ret;
1321
1322         if (!S_ISDIR(file_inode(file)->i_mode))
1323                 return -ENOTDIR;
1324
1325         vol_args = memdup_user(arg, sizeof(*vol_args));
1326         if (IS_ERR(vol_args))
1327                 return PTR_ERR(vol_args);
1328         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1329
1330         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1331                                         vol_args->name, vol_args->fd, subvol,
1332                                         false, NULL);
1333
1334         kfree(vol_args);
1335         return ret;
1336 }
1337
1338 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1339                                                void __user *arg, int subvol)
1340 {
1341         struct btrfs_ioctl_vol_args_v2 *vol_args;
1342         int ret;
1343         bool readonly = false;
1344         struct btrfs_qgroup_inherit *inherit = NULL;
1345
1346         if (!S_ISDIR(file_inode(file)->i_mode))
1347                 return -ENOTDIR;
1348
1349         vol_args = memdup_user(arg, sizeof(*vol_args));
1350         if (IS_ERR(vol_args))
1351                 return PTR_ERR(vol_args);
1352         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1353
1354         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1355                 ret = -EOPNOTSUPP;
1356                 goto free_args;
1357         }
1358
1359         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1360                 readonly = true;
1361         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1362                 u64 nums;
1363
1364                 if (vol_args->size < sizeof(*inherit) ||
1365                     vol_args->size > PAGE_SIZE) {
1366                         ret = -EINVAL;
1367                         goto free_args;
1368                 }
1369                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1370                 if (IS_ERR(inherit)) {
1371                         ret = PTR_ERR(inherit);
1372                         goto free_args;
1373                 }
1374
1375                 if (inherit->num_qgroups > PAGE_SIZE ||
1376                     inherit->num_ref_copies > PAGE_SIZE ||
1377                     inherit->num_excl_copies > PAGE_SIZE) {
1378                         ret = -EINVAL;
1379                         goto free_inherit;
1380                 }
1381
1382                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1383                        2 * inherit->num_excl_copies;
1384                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1385                         ret = -EINVAL;
1386                         goto free_inherit;
1387                 }
1388         }
1389
1390         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1391                                         vol_args->name, vol_args->fd, subvol,
1392                                         readonly, inherit);
1393         if (ret)
1394                 goto free_inherit;
1395 free_inherit:
1396         kfree(inherit);
1397 free_args:
1398         kfree(vol_args);
1399         return ret;
1400 }
1401
1402 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
1403                                                 void __user *arg)
1404 {
1405         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1406         struct btrfs_root *root = BTRFS_I(inode)->root;
1407         int ret = 0;
1408         u64 flags = 0;
1409
1410         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1411                 return -EINVAL;
1412
1413         down_read(&fs_info->subvol_sem);
1414         if (btrfs_root_readonly(root))
1415                 flags |= BTRFS_SUBVOL_RDONLY;
1416         up_read(&fs_info->subvol_sem);
1417
1418         if (copy_to_user(arg, &flags, sizeof(flags)))
1419                 ret = -EFAULT;
1420
1421         return ret;
1422 }
1423
1424 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1425                                               void __user *arg)
1426 {
1427         struct inode *inode = file_inode(file);
1428         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1429         struct btrfs_root *root = BTRFS_I(inode)->root;
1430         struct btrfs_trans_handle *trans;
1431         u64 root_flags;
1432         u64 flags;
1433         int ret = 0;
1434
1435         if (!inode_owner_or_capable(file_mnt_idmap(file), inode))
1436                 return -EPERM;
1437
1438         ret = mnt_want_write_file(file);
1439         if (ret)
1440                 goto out;
1441
1442         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1443                 ret = -EINVAL;
1444                 goto out_drop_write;
1445         }
1446
1447         if (copy_from_user(&flags, arg, sizeof(flags))) {
1448                 ret = -EFAULT;
1449                 goto out_drop_write;
1450         }
1451
1452         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1453                 ret = -EOPNOTSUPP;
1454                 goto out_drop_write;
1455         }
1456
1457         down_write(&fs_info->subvol_sem);
1458
1459         /* nothing to do */
1460         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1461                 goto out_drop_sem;
1462
1463         root_flags = btrfs_root_flags(&root->root_item);
1464         if (flags & BTRFS_SUBVOL_RDONLY) {
1465                 btrfs_set_root_flags(&root->root_item,
1466                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1467         } else {
1468                 /*
1469                  * Block RO -> RW transition if this subvolume is involved in
1470                  * send
1471                  */
1472                 spin_lock(&root->root_item_lock);
1473                 if (root->send_in_progress == 0) {
1474                         btrfs_set_root_flags(&root->root_item,
1475                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1476                         spin_unlock(&root->root_item_lock);
1477                 } else {
1478                         spin_unlock(&root->root_item_lock);
1479                         btrfs_warn(fs_info,
1480                                    "Attempt to set subvolume %llu read-write during send",
1481                                    root->root_key.objectid);
1482                         ret = -EPERM;
1483                         goto out_drop_sem;
1484                 }
1485         }
1486
1487         trans = btrfs_start_transaction(root, 1);
1488         if (IS_ERR(trans)) {
1489                 ret = PTR_ERR(trans);
1490                 goto out_reset;
1491         }
1492
1493         ret = btrfs_update_root(trans, fs_info->tree_root,
1494                                 &root->root_key, &root->root_item);
1495         if (ret < 0) {
1496                 btrfs_end_transaction(trans);
1497                 goto out_reset;
1498         }
1499
1500         ret = btrfs_commit_transaction(trans);
1501
1502 out_reset:
1503         if (ret)
1504                 btrfs_set_root_flags(&root->root_item, root_flags);
1505 out_drop_sem:
1506         up_write(&fs_info->subvol_sem);
1507 out_drop_write:
1508         mnt_drop_write_file(file);
1509 out:
1510         return ret;
1511 }
1512
1513 static noinline int key_in_sk(struct btrfs_key *key,
1514                               struct btrfs_ioctl_search_key *sk)
1515 {
1516         struct btrfs_key test;
1517         int ret;
1518
1519         test.objectid = sk->min_objectid;
1520         test.type = sk->min_type;
1521         test.offset = sk->min_offset;
1522
1523         ret = btrfs_comp_cpu_keys(key, &test);
1524         if (ret < 0)
1525                 return 0;
1526
1527         test.objectid = sk->max_objectid;
1528         test.type = sk->max_type;
1529         test.offset = sk->max_offset;
1530
1531         ret = btrfs_comp_cpu_keys(key, &test);
1532         if (ret > 0)
1533                 return 0;
1534         return 1;
1535 }
1536
1537 static noinline int copy_to_sk(struct btrfs_path *path,
1538                                struct btrfs_key *key,
1539                                struct btrfs_ioctl_search_key *sk,
1540                                u64 *buf_size,
1541                                char __user *ubuf,
1542                                unsigned long *sk_offset,
1543                                int *num_found)
1544 {
1545         u64 found_transid;
1546         struct extent_buffer *leaf;
1547         struct btrfs_ioctl_search_header sh;
1548         struct btrfs_key test;
1549         unsigned long item_off;
1550         unsigned long item_len;
1551         int nritems;
1552         int i;
1553         int slot;
1554         int ret = 0;
1555
1556         leaf = path->nodes[0];
1557         slot = path->slots[0];
1558         nritems = btrfs_header_nritems(leaf);
1559
1560         if (btrfs_header_generation(leaf) > sk->max_transid) {
1561                 i = nritems;
1562                 goto advance_key;
1563         }
1564         found_transid = btrfs_header_generation(leaf);
1565
1566         for (i = slot; i < nritems; i++) {
1567                 item_off = btrfs_item_ptr_offset(leaf, i);
1568                 item_len = btrfs_item_size(leaf, i);
1569
1570                 btrfs_item_key_to_cpu(leaf, key, i);
1571                 if (!key_in_sk(key, sk))
1572                         continue;
1573
1574                 if (sizeof(sh) + item_len > *buf_size) {
1575                         if (*num_found) {
1576                                 ret = 1;
1577                                 goto out;
1578                         }
1579
1580                         /*
1581                          * return one empty item back for v1, which does not
1582                          * handle -EOVERFLOW
1583                          */
1584
1585                         *buf_size = sizeof(sh) + item_len;
1586                         item_len = 0;
1587                         ret = -EOVERFLOW;
1588                 }
1589
1590                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1591                         ret = 1;
1592                         goto out;
1593                 }
1594
1595                 sh.objectid = key->objectid;
1596                 sh.offset = key->offset;
1597                 sh.type = key->type;
1598                 sh.len = item_len;
1599                 sh.transid = found_transid;
1600
1601                 /*
1602                  * Copy search result header. If we fault then loop again so we
1603                  * can fault in the pages and -EFAULT there if there's a
1604                  * problem. Otherwise we'll fault and then copy the buffer in
1605                  * properly this next time through
1606                  */
1607                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
1608                         ret = 0;
1609                         goto out;
1610                 }
1611
1612                 *sk_offset += sizeof(sh);
1613
1614                 if (item_len) {
1615                         char __user *up = ubuf + *sk_offset;
1616                         /*
1617                          * Copy the item, same behavior as above, but reset the
1618                          * * sk_offset so we copy the full thing again.
1619                          */
1620                         if (read_extent_buffer_to_user_nofault(leaf, up,
1621                                                 item_off, item_len)) {
1622                                 ret = 0;
1623                                 *sk_offset -= sizeof(sh);
1624                                 goto out;
1625                         }
1626
1627                         *sk_offset += item_len;
1628                 }
1629                 (*num_found)++;
1630
1631                 if (ret) /* -EOVERFLOW from above */
1632                         goto out;
1633
1634                 if (*num_found >= sk->nr_items) {
1635                         ret = 1;
1636                         goto out;
1637                 }
1638         }
1639 advance_key:
1640         ret = 0;
1641         test.objectid = sk->max_objectid;
1642         test.type = sk->max_type;
1643         test.offset = sk->max_offset;
1644         if (btrfs_comp_cpu_keys(key, &test) >= 0)
1645                 ret = 1;
1646         else if (key->offset < (u64)-1)
1647                 key->offset++;
1648         else if (key->type < (u8)-1) {
1649                 key->offset = 0;
1650                 key->type++;
1651         } else if (key->objectid < (u64)-1) {
1652                 key->offset = 0;
1653                 key->type = 0;
1654                 key->objectid++;
1655         } else
1656                 ret = 1;
1657 out:
1658         /*
1659          *  0: all items from this leaf copied, continue with next
1660          *  1: * more items can be copied, but unused buffer is too small
1661          *     * all items were found
1662          *     Either way, it will stops the loop which iterates to the next
1663          *     leaf
1664          *  -EOVERFLOW: item was to large for buffer
1665          *  -EFAULT: could not copy extent buffer back to userspace
1666          */
1667         return ret;
1668 }
1669
1670 static noinline int search_ioctl(struct inode *inode,
1671                                  struct btrfs_ioctl_search_key *sk,
1672                                  u64 *buf_size,
1673                                  char __user *ubuf)
1674 {
1675         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
1676         struct btrfs_root *root;
1677         struct btrfs_key key;
1678         struct btrfs_path *path;
1679         int ret;
1680         int num_found = 0;
1681         unsigned long sk_offset = 0;
1682
1683         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
1684                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
1685                 return -EOVERFLOW;
1686         }
1687
1688         path = btrfs_alloc_path();
1689         if (!path)
1690                 return -ENOMEM;
1691
1692         if (sk->tree_id == 0) {
1693                 /* search the root of the inode that was passed */
1694                 root = btrfs_grab_root(BTRFS_I(inode)->root);
1695         } else {
1696                 root = btrfs_get_fs_root(info, sk->tree_id, true);
1697                 if (IS_ERR(root)) {
1698                         btrfs_free_path(path);
1699                         return PTR_ERR(root);
1700                 }
1701         }
1702
1703         key.objectid = sk->min_objectid;
1704         key.type = sk->min_type;
1705         key.offset = sk->min_offset;
1706
1707         while (1) {
1708                 ret = -EFAULT;
1709                 /*
1710                  * Ensure that the whole user buffer is faulted in at sub-page
1711                  * granularity, otherwise the loop may live-lock.
1712                  */
1713                 if (fault_in_subpage_writeable(ubuf + sk_offset,
1714                                                *buf_size - sk_offset))
1715                         break;
1716
1717                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
1718                 if (ret != 0) {
1719                         if (ret > 0)
1720                                 ret = 0;
1721                         goto err;
1722                 }
1723                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
1724                                  &sk_offset, &num_found);
1725                 btrfs_release_path(path);
1726                 if (ret)
1727                         break;
1728
1729         }
1730         if (ret > 0)
1731                 ret = 0;
1732 err:
1733         sk->nr_items = num_found;
1734         btrfs_put_root(root);
1735         btrfs_free_path(path);
1736         return ret;
1737 }
1738
1739 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
1740                                             void __user *argp)
1741 {
1742         struct btrfs_ioctl_search_args __user *uargs = argp;
1743         struct btrfs_ioctl_search_key sk;
1744         int ret;
1745         u64 buf_size;
1746
1747         if (!capable(CAP_SYS_ADMIN))
1748                 return -EPERM;
1749
1750         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
1751                 return -EFAULT;
1752
1753         buf_size = sizeof(uargs->buf);
1754
1755         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
1756
1757         /*
1758          * In the origin implementation an overflow is handled by returning a
1759          * search header with a len of zero, so reset ret.
1760          */
1761         if (ret == -EOVERFLOW)
1762                 ret = 0;
1763
1764         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
1765                 ret = -EFAULT;
1766         return ret;
1767 }
1768
1769 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
1770                                                void __user *argp)
1771 {
1772         struct btrfs_ioctl_search_args_v2 __user *uarg = argp;
1773         struct btrfs_ioctl_search_args_v2 args;
1774         int ret;
1775         u64 buf_size;
1776         const u64 buf_limit = SZ_16M;
1777
1778         if (!capable(CAP_SYS_ADMIN))
1779                 return -EPERM;
1780
1781         /* copy search header and buffer size */
1782         if (copy_from_user(&args, uarg, sizeof(args)))
1783                 return -EFAULT;
1784
1785         buf_size = args.buf_size;
1786
1787         /* limit result size to 16MB */
1788         if (buf_size > buf_limit)
1789                 buf_size = buf_limit;
1790
1791         ret = search_ioctl(inode, &args.key, &buf_size,
1792                            (char __user *)(&uarg->buf[0]));
1793         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
1794                 ret = -EFAULT;
1795         else if (ret == -EOVERFLOW &&
1796                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
1797                 ret = -EFAULT;
1798
1799         return ret;
1800 }
1801
1802 /*
1803  * Search INODE_REFs to identify path name of 'dirid' directory
1804  * in a 'tree_id' tree. and sets path name to 'name'.
1805  */
1806 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1807                                 u64 tree_id, u64 dirid, char *name)
1808 {
1809         struct btrfs_root *root;
1810         struct btrfs_key key;
1811         char *ptr;
1812         int ret = -1;
1813         int slot;
1814         int len;
1815         int total_len = 0;
1816         struct btrfs_inode_ref *iref;
1817         struct extent_buffer *l;
1818         struct btrfs_path *path;
1819
1820         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1821                 name[0]='\0';
1822                 return 0;
1823         }
1824
1825         path = btrfs_alloc_path();
1826         if (!path)
1827                 return -ENOMEM;
1828
1829         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
1830
1831         root = btrfs_get_fs_root(info, tree_id, true);
1832         if (IS_ERR(root)) {
1833                 ret = PTR_ERR(root);
1834                 root = NULL;
1835                 goto out;
1836         }
1837
1838         key.objectid = dirid;
1839         key.type = BTRFS_INODE_REF_KEY;
1840         key.offset = (u64)-1;
1841
1842         while (1) {
1843                 ret = btrfs_search_backwards(root, &key, path);
1844                 if (ret < 0)
1845                         goto out;
1846                 else if (ret > 0) {
1847                         ret = -ENOENT;
1848                         goto out;
1849                 }
1850
1851                 l = path->nodes[0];
1852                 slot = path->slots[0];
1853
1854                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1855                 len = btrfs_inode_ref_name_len(l, iref);
1856                 ptr -= len + 1;
1857                 total_len += len + 1;
1858                 if (ptr < name) {
1859                         ret = -ENAMETOOLONG;
1860                         goto out;
1861                 }
1862
1863                 *(ptr + len) = '/';
1864                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
1865
1866                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1867                         break;
1868
1869                 btrfs_release_path(path);
1870                 key.objectid = key.offset;
1871                 key.offset = (u64)-1;
1872                 dirid = key.objectid;
1873         }
1874         memmove(name, ptr, total_len);
1875         name[total_len] = '\0';
1876         ret = 0;
1877 out:
1878         btrfs_put_root(root);
1879         btrfs_free_path(path);
1880         return ret;
1881 }
1882
1883 static int btrfs_search_path_in_tree_user(struct mnt_idmap *idmap,
1884                                 struct inode *inode,
1885                                 struct btrfs_ioctl_ino_lookup_user_args *args)
1886 {
1887         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1888         struct super_block *sb = inode->i_sb;
1889         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
1890         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
1891         u64 dirid = args->dirid;
1892         unsigned long item_off;
1893         unsigned long item_len;
1894         struct btrfs_inode_ref *iref;
1895         struct btrfs_root_ref *rref;
1896         struct btrfs_root *root = NULL;
1897         struct btrfs_path *path;
1898         struct btrfs_key key, key2;
1899         struct extent_buffer *leaf;
1900         struct inode *temp_inode;
1901         char *ptr;
1902         int slot;
1903         int len;
1904         int total_len = 0;
1905         int ret;
1906
1907         path = btrfs_alloc_path();
1908         if (!path)
1909                 return -ENOMEM;
1910
1911         /*
1912          * If the bottom subvolume does not exist directly under upper_limit,
1913          * construct the path in from the bottom up.
1914          */
1915         if (dirid != upper_limit.objectid) {
1916                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
1917
1918                 root = btrfs_get_fs_root(fs_info, treeid, true);
1919                 if (IS_ERR(root)) {
1920                         ret = PTR_ERR(root);
1921                         goto out;
1922                 }
1923
1924                 key.objectid = dirid;
1925                 key.type = BTRFS_INODE_REF_KEY;
1926                 key.offset = (u64)-1;
1927                 while (1) {
1928                         ret = btrfs_search_backwards(root, &key, path);
1929                         if (ret < 0)
1930                                 goto out_put;
1931                         else if (ret > 0) {
1932                                 ret = -ENOENT;
1933                                 goto out_put;
1934                         }
1935
1936                         leaf = path->nodes[0];
1937                         slot = path->slots[0];
1938
1939                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
1940                         len = btrfs_inode_ref_name_len(leaf, iref);
1941                         ptr -= len + 1;
1942                         total_len += len + 1;
1943                         if (ptr < args->path) {
1944                                 ret = -ENAMETOOLONG;
1945                                 goto out_put;
1946                         }
1947
1948                         *(ptr + len) = '/';
1949                         read_extent_buffer(leaf, ptr,
1950                                         (unsigned long)(iref + 1), len);
1951
1952                         /* Check the read+exec permission of this directory */
1953                         ret = btrfs_previous_item(root, path, dirid,
1954                                                   BTRFS_INODE_ITEM_KEY);
1955                         if (ret < 0) {
1956                                 goto out_put;
1957                         } else if (ret > 0) {
1958                                 ret = -ENOENT;
1959                                 goto out_put;
1960                         }
1961
1962                         leaf = path->nodes[0];
1963                         slot = path->slots[0];
1964                         btrfs_item_key_to_cpu(leaf, &key2, slot);
1965                         if (key2.objectid != dirid) {
1966                                 ret = -ENOENT;
1967                                 goto out_put;
1968                         }
1969
1970                         /*
1971                          * We don't need the path anymore, so release it and
1972                          * avoid deadlocks and lockdep warnings in case
1973                          * btrfs_iget() needs to lookup the inode from its root
1974                          * btree and lock the same leaf.
1975                          */
1976                         btrfs_release_path(path);
1977                         temp_inode = btrfs_iget(sb, key2.objectid, root);
1978                         if (IS_ERR(temp_inode)) {
1979                                 ret = PTR_ERR(temp_inode);
1980                                 goto out_put;
1981                         }
1982                         ret = inode_permission(idmap, temp_inode,
1983                                                MAY_READ | MAY_EXEC);
1984                         iput(temp_inode);
1985                         if (ret) {
1986                                 ret = -EACCES;
1987                                 goto out_put;
1988                         }
1989
1990                         if (key.offset == upper_limit.objectid)
1991                                 break;
1992                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
1993                                 ret = -EACCES;
1994                                 goto out_put;
1995                         }
1996
1997                         key.objectid = key.offset;
1998                         key.offset = (u64)-1;
1999                         dirid = key.objectid;
2000                 }
2001
2002                 memmove(args->path, ptr, total_len);
2003                 args->path[total_len] = '\0';
2004                 btrfs_put_root(root);
2005                 root = NULL;
2006                 btrfs_release_path(path);
2007         }
2008
2009         /* Get the bottom subvolume's name from ROOT_REF */
2010         key.objectid = treeid;
2011         key.type = BTRFS_ROOT_REF_KEY;
2012         key.offset = args->treeid;
2013         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2014         if (ret < 0) {
2015                 goto out;
2016         } else if (ret > 0) {
2017                 ret = -ENOENT;
2018                 goto out;
2019         }
2020
2021         leaf = path->nodes[0];
2022         slot = path->slots[0];
2023         btrfs_item_key_to_cpu(leaf, &key, slot);
2024
2025         item_off = btrfs_item_ptr_offset(leaf, slot);
2026         item_len = btrfs_item_size(leaf, slot);
2027         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2028         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2029         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2030                 ret = -EINVAL;
2031                 goto out;
2032         }
2033
2034         /* Copy subvolume's name */
2035         item_off += sizeof(struct btrfs_root_ref);
2036         item_len -= sizeof(struct btrfs_root_ref);
2037         read_extent_buffer(leaf, args->name, item_off, item_len);
2038         args->name[item_len] = 0;
2039
2040 out_put:
2041         btrfs_put_root(root);
2042 out:
2043         btrfs_free_path(path);
2044         return ret;
2045 }
2046
2047 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2048                                            void __user *argp)
2049 {
2050         struct btrfs_ioctl_ino_lookup_args *args;
2051         int ret = 0;
2052
2053         args = memdup_user(argp, sizeof(*args));
2054         if (IS_ERR(args))
2055                 return PTR_ERR(args);
2056
2057         /*
2058          * Unprivileged query to obtain the containing subvolume root id. The
2059          * path is reset so it's consistent with btrfs_search_path_in_tree.
2060          */
2061         if (args->treeid == 0)
2062                 args->treeid = root->root_key.objectid;
2063
2064         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2065                 args->name[0] = 0;
2066                 goto out;
2067         }
2068
2069         if (!capable(CAP_SYS_ADMIN)) {
2070                 ret = -EPERM;
2071                 goto out;
2072         }
2073
2074         ret = btrfs_search_path_in_tree(root->fs_info,
2075                                         args->treeid, args->objectid,
2076                                         args->name);
2077
2078 out:
2079         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2080                 ret = -EFAULT;
2081
2082         kfree(args);
2083         return ret;
2084 }
2085
2086 /*
2087  * Version of ino_lookup ioctl (unprivileged)
2088  *
2089  * The main differences from ino_lookup ioctl are:
2090  *
2091  *   1. Read + Exec permission will be checked using inode_permission() during
2092  *      path construction. -EACCES will be returned in case of failure.
2093  *   2. Path construction will be stopped at the inode number which corresponds
2094  *      to the fd with which this ioctl is called. If constructed path does not
2095  *      exist under fd's inode, -EACCES will be returned.
2096  *   3. The name of bottom subvolume is also searched and filled.
2097  */
2098 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2099 {
2100         struct btrfs_ioctl_ino_lookup_user_args *args;
2101         struct inode *inode;
2102         int ret;
2103
2104         args = memdup_user(argp, sizeof(*args));
2105         if (IS_ERR(args))
2106                 return PTR_ERR(args);
2107
2108         inode = file_inode(file);
2109
2110         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2111             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2112                 /*
2113                  * The subvolume does not exist under fd with which this is
2114                  * called
2115                  */
2116                 kfree(args);
2117                 return -EACCES;
2118         }
2119
2120         ret = btrfs_search_path_in_tree_user(file_mnt_idmap(file), inode, args);
2121
2122         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2123                 ret = -EFAULT;
2124
2125         kfree(args);
2126         return ret;
2127 }
2128
2129 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2130 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
2131 {
2132         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2133         struct btrfs_fs_info *fs_info;
2134         struct btrfs_root *root;
2135         struct btrfs_path *path;
2136         struct btrfs_key key;
2137         struct btrfs_root_item *root_item;
2138         struct btrfs_root_ref *rref;
2139         struct extent_buffer *leaf;
2140         unsigned long item_off;
2141         unsigned long item_len;
2142         int slot;
2143         int ret = 0;
2144
2145         path = btrfs_alloc_path();
2146         if (!path)
2147                 return -ENOMEM;
2148
2149         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2150         if (!subvol_info) {
2151                 btrfs_free_path(path);
2152                 return -ENOMEM;
2153         }
2154
2155         fs_info = BTRFS_I(inode)->root->fs_info;
2156
2157         /* Get root_item of inode's subvolume */
2158         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2159         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2160         if (IS_ERR(root)) {
2161                 ret = PTR_ERR(root);
2162                 goto out_free;
2163         }
2164         root_item = &root->root_item;
2165
2166         subvol_info->treeid = key.objectid;
2167
2168         subvol_info->generation = btrfs_root_generation(root_item);
2169         subvol_info->flags = btrfs_root_flags(root_item);
2170
2171         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2172         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2173                                                     BTRFS_UUID_SIZE);
2174         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2175                                                     BTRFS_UUID_SIZE);
2176
2177         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2178         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2179         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2180
2181         subvol_info->otransid = btrfs_root_otransid(root_item);
2182         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2183         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2184
2185         subvol_info->stransid = btrfs_root_stransid(root_item);
2186         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2187         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2188
2189         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2190         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2191         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2192
2193         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2194                 /* Search root tree for ROOT_BACKREF of this subvolume */
2195                 key.type = BTRFS_ROOT_BACKREF_KEY;
2196                 key.offset = 0;
2197                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2198                 if (ret < 0) {
2199                         goto out;
2200                 } else if (path->slots[0] >=
2201                            btrfs_header_nritems(path->nodes[0])) {
2202                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2203                         if (ret < 0) {
2204                                 goto out;
2205                         } else if (ret > 0) {
2206                                 ret = -EUCLEAN;
2207                                 goto out;
2208                         }
2209                 }
2210
2211                 leaf = path->nodes[0];
2212                 slot = path->slots[0];
2213                 btrfs_item_key_to_cpu(leaf, &key, slot);
2214                 if (key.objectid == subvol_info->treeid &&
2215                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2216                         subvol_info->parent_id = key.offset;
2217
2218                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2219                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2220
2221                         item_off = btrfs_item_ptr_offset(leaf, slot)
2222                                         + sizeof(struct btrfs_root_ref);
2223                         item_len = btrfs_item_size(leaf, slot)
2224                                         - sizeof(struct btrfs_root_ref);
2225                         read_extent_buffer(leaf, subvol_info->name,
2226                                            item_off, item_len);
2227                 } else {
2228                         ret = -ENOENT;
2229                         goto out;
2230                 }
2231         }
2232
2233         btrfs_free_path(path);
2234         path = NULL;
2235         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2236                 ret = -EFAULT;
2237
2238 out:
2239         btrfs_put_root(root);
2240 out_free:
2241         btrfs_free_path(path);
2242         kfree(subvol_info);
2243         return ret;
2244 }
2245
2246 /*
2247  * Return ROOT_REF information of the subvolume containing this inode
2248  * except the subvolume name.
2249  */
2250 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
2251                                           void __user *argp)
2252 {
2253         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2254         struct btrfs_root_ref *rref;
2255         struct btrfs_path *path;
2256         struct btrfs_key key;
2257         struct extent_buffer *leaf;
2258         u64 objectid;
2259         int slot;
2260         int ret;
2261         u8 found;
2262
2263         path = btrfs_alloc_path();
2264         if (!path)
2265                 return -ENOMEM;
2266
2267         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2268         if (IS_ERR(rootrefs)) {
2269                 btrfs_free_path(path);
2270                 return PTR_ERR(rootrefs);
2271         }
2272
2273         objectid = root->root_key.objectid;
2274         key.objectid = objectid;
2275         key.type = BTRFS_ROOT_REF_KEY;
2276         key.offset = rootrefs->min_treeid;
2277         found = 0;
2278
2279         root = root->fs_info->tree_root;
2280         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2281         if (ret < 0) {
2282                 goto out;
2283         } else if (path->slots[0] >=
2284                    btrfs_header_nritems(path->nodes[0])) {
2285                 ret = btrfs_next_leaf(root, path);
2286                 if (ret < 0) {
2287                         goto out;
2288                 } else if (ret > 0) {
2289                         ret = -EUCLEAN;
2290                         goto out;
2291                 }
2292         }
2293         while (1) {
2294                 leaf = path->nodes[0];
2295                 slot = path->slots[0];
2296
2297                 btrfs_item_key_to_cpu(leaf, &key, slot);
2298                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2299                         ret = 0;
2300                         goto out;
2301                 }
2302
2303                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2304                         ret = -EOVERFLOW;
2305                         goto out;
2306                 }
2307
2308                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2309                 rootrefs->rootref[found].treeid = key.offset;
2310                 rootrefs->rootref[found].dirid =
2311                                   btrfs_root_ref_dirid(leaf, rref);
2312                 found++;
2313
2314                 ret = btrfs_next_item(root, path);
2315                 if (ret < 0) {
2316                         goto out;
2317                 } else if (ret > 0) {
2318                         ret = -EUCLEAN;
2319                         goto out;
2320                 }
2321         }
2322
2323 out:
2324         btrfs_free_path(path);
2325
2326         if (!ret || ret == -EOVERFLOW) {
2327                 rootrefs->num_items = found;
2328                 /* update min_treeid for next search */
2329                 if (found)
2330                         rootrefs->min_treeid =
2331                                 rootrefs->rootref[found - 1].treeid + 1;
2332                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2333                         ret = -EFAULT;
2334         }
2335
2336         kfree(rootrefs);
2337
2338         return ret;
2339 }
2340
2341 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2342                                              void __user *arg,
2343                                              bool destroy_v2)
2344 {
2345         struct dentry *parent = file->f_path.dentry;
2346         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2347         struct dentry *dentry;
2348         struct inode *dir = d_inode(parent);
2349         struct inode *inode;
2350         struct btrfs_root *root = BTRFS_I(dir)->root;
2351         struct btrfs_root *dest = NULL;
2352         struct btrfs_ioctl_vol_args *vol_args = NULL;
2353         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2354         struct mnt_idmap *idmap = file_mnt_idmap(file);
2355         char *subvol_name, *subvol_name_ptr = NULL;
2356         int subvol_namelen;
2357         int err = 0;
2358         bool destroy_parent = false;
2359
2360         /* We don't support snapshots with extent tree v2 yet. */
2361         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2362                 btrfs_err(fs_info,
2363                           "extent tree v2 doesn't support snapshot deletion yet");
2364                 return -EOPNOTSUPP;
2365         }
2366
2367         if (destroy_v2) {
2368                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2369                 if (IS_ERR(vol_args2))
2370                         return PTR_ERR(vol_args2);
2371
2372                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2373                         err = -EOPNOTSUPP;
2374                         goto out;
2375                 }
2376
2377                 /*
2378                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2379                  * name, same as v1 currently does.
2380                  */
2381                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2382                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2383                         subvol_name = vol_args2->name;
2384
2385                         err = mnt_want_write_file(file);
2386                         if (err)
2387                                 goto out;
2388                 } else {
2389                         struct inode *old_dir;
2390
2391                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2392                                 err = -EINVAL;
2393                                 goto out;
2394                         }
2395
2396                         err = mnt_want_write_file(file);
2397                         if (err)
2398                                 goto out;
2399
2400                         dentry = btrfs_get_dentry(fs_info->sb,
2401                                         BTRFS_FIRST_FREE_OBJECTID,
2402                                         vol_args2->subvolid, 0);
2403                         if (IS_ERR(dentry)) {
2404                                 err = PTR_ERR(dentry);
2405                                 goto out_drop_write;
2406                         }
2407
2408                         /*
2409                          * Change the default parent since the subvolume being
2410                          * deleted can be outside of the current mount point.
2411                          */
2412                         parent = btrfs_get_parent(dentry);
2413
2414                         /*
2415                          * At this point dentry->d_name can point to '/' if the
2416                          * subvolume we want to destroy is outsite of the
2417                          * current mount point, so we need to release the
2418                          * current dentry and execute the lookup to return a new
2419                          * one with ->d_name pointing to the
2420                          * <mount point>/subvol_name.
2421                          */
2422                         dput(dentry);
2423                         if (IS_ERR(parent)) {
2424                                 err = PTR_ERR(parent);
2425                                 goto out_drop_write;
2426                         }
2427                         old_dir = dir;
2428                         dir = d_inode(parent);
2429
2430                         /*
2431                          * If v2 was used with SPEC_BY_ID, a new parent was
2432                          * allocated since the subvolume can be outside of the
2433                          * current mount point. Later on we need to release this
2434                          * new parent dentry.
2435                          */
2436                         destroy_parent = true;
2437
2438                         /*
2439                          * On idmapped mounts, deletion via subvolid is
2440                          * restricted to subvolumes that are immediate
2441                          * ancestors of the inode referenced by the file
2442                          * descriptor in the ioctl. Otherwise the idmapping
2443                          * could potentially be abused to delete subvolumes
2444                          * anywhere in the filesystem the user wouldn't be able
2445                          * to delete without an idmapped mount.
2446                          */
2447                         if (old_dir != dir && idmap != &nop_mnt_idmap) {
2448                                 err = -EOPNOTSUPP;
2449                                 goto free_parent;
2450                         }
2451
2452                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2453                                                 fs_info, vol_args2->subvolid);
2454                         if (IS_ERR(subvol_name_ptr)) {
2455                                 err = PTR_ERR(subvol_name_ptr);
2456                                 goto free_parent;
2457                         }
2458                         /* subvol_name_ptr is already nul terminated */
2459                         subvol_name = (char *)kbasename(subvol_name_ptr);
2460                 }
2461         } else {
2462                 vol_args = memdup_user(arg, sizeof(*vol_args));
2463                 if (IS_ERR(vol_args))
2464                         return PTR_ERR(vol_args);
2465
2466                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2467                 subvol_name = vol_args->name;
2468
2469                 err = mnt_want_write_file(file);
2470                 if (err)
2471                         goto out;
2472         }
2473
2474         subvol_namelen = strlen(subvol_name);
2475
2476         if (strchr(subvol_name, '/') ||
2477             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2478                 err = -EINVAL;
2479                 goto free_subvol_name;
2480         }
2481
2482         if (!S_ISDIR(dir->i_mode)) {
2483                 err = -ENOTDIR;
2484                 goto free_subvol_name;
2485         }
2486
2487         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2488         if (err == -EINTR)
2489                 goto free_subvol_name;
2490         dentry = lookup_one(idmap, subvol_name, parent, subvol_namelen);
2491         if (IS_ERR(dentry)) {
2492                 err = PTR_ERR(dentry);
2493                 goto out_unlock_dir;
2494         }
2495
2496         if (d_really_is_negative(dentry)) {
2497                 err = -ENOENT;
2498                 goto out_dput;
2499         }
2500
2501         inode = d_inode(dentry);
2502         dest = BTRFS_I(inode)->root;
2503         if (!capable(CAP_SYS_ADMIN)) {
2504                 /*
2505                  * Regular user.  Only allow this with a special mount
2506                  * option, when the user has write+exec access to the
2507                  * subvol root, and when rmdir(2) would have been
2508                  * allowed.
2509                  *
2510                  * Note that this is _not_ check that the subvol is
2511                  * empty or doesn't contain data that we wouldn't
2512                  * otherwise be able to delete.
2513                  *
2514                  * Users who want to delete empty subvols should try
2515                  * rmdir(2).
2516                  */
2517                 err = -EPERM;
2518                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2519                         goto out_dput;
2520
2521                 /*
2522                  * Do not allow deletion if the parent dir is the same
2523                  * as the dir to be deleted.  That means the ioctl
2524                  * must be called on the dentry referencing the root
2525                  * of the subvol, not a random directory contained
2526                  * within it.
2527                  */
2528                 err = -EINVAL;
2529                 if (root == dest)
2530                         goto out_dput;
2531
2532                 err = inode_permission(idmap, inode, MAY_WRITE | MAY_EXEC);
2533                 if (err)
2534                         goto out_dput;
2535         }
2536
2537         /* check if subvolume may be deleted by a user */
2538         err = btrfs_may_delete(idmap, dir, dentry, 1);
2539         if (err)
2540                 goto out_dput;
2541
2542         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2543                 err = -EINVAL;
2544                 goto out_dput;
2545         }
2546
2547         btrfs_inode_lock(BTRFS_I(inode), 0);
2548         err = btrfs_delete_subvolume(BTRFS_I(dir), dentry);
2549         btrfs_inode_unlock(BTRFS_I(inode), 0);
2550         if (!err)
2551                 d_delete_notify(dir, dentry);
2552
2553 out_dput:
2554         dput(dentry);
2555 out_unlock_dir:
2556         btrfs_inode_unlock(BTRFS_I(dir), 0);
2557 free_subvol_name:
2558         kfree(subvol_name_ptr);
2559 free_parent:
2560         if (destroy_parent)
2561                 dput(parent);
2562 out_drop_write:
2563         mnt_drop_write_file(file);
2564 out:
2565         kfree(vol_args2);
2566         kfree(vol_args);
2567         return err;
2568 }
2569
2570 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2571 {
2572         struct inode *inode = file_inode(file);
2573         struct btrfs_root *root = BTRFS_I(inode)->root;
2574         struct btrfs_ioctl_defrag_range_args range = {0};
2575         int ret;
2576
2577         ret = mnt_want_write_file(file);
2578         if (ret)
2579                 return ret;
2580
2581         if (btrfs_root_readonly(root)) {
2582                 ret = -EROFS;
2583                 goto out;
2584         }
2585
2586         switch (inode->i_mode & S_IFMT) {
2587         case S_IFDIR:
2588                 if (!capable(CAP_SYS_ADMIN)) {
2589                         ret = -EPERM;
2590                         goto out;
2591                 }
2592                 ret = btrfs_defrag_root(root);
2593                 break;
2594         case S_IFREG:
2595                 /*
2596                  * Note that this does not check the file descriptor for write
2597                  * access. This prevents defragmenting executables that are
2598                  * running and allows defrag on files open in read-only mode.
2599                  */
2600                 if (!capable(CAP_SYS_ADMIN) &&
2601                     inode_permission(&nop_mnt_idmap, inode, MAY_WRITE)) {
2602                         ret = -EPERM;
2603                         goto out;
2604                 }
2605
2606                 if (argp) {
2607                         if (copy_from_user(&range, argp, sizeof(range))) {
2608                                 ret = -EFAULT;
2609                                 goto out;
2610                         }
2611                         /* compression requires us to start the IO */
2612                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2613                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
2614                                 range.extent_thresh = (u32)-1;
2615                         }
2616                 } else {
2617                         /* the rest are all set to zero by kzalloc */
2618                         range.len = (u64)-1;
2619                 }
2620                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
2621                                         &range, BTRFS_OLDEST_GENERATION, 0);
2622                 if (ret > 0)
2623                         ret = 0;
2624                 break;
2625         default:
2626                 ret = -EINVAL;
2627         }
2628 out:
2629         mnt_drop_write_file(file);
2630         return ret;
2631 }
2632
2633 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2634 {
2635         struct btrfs_ioctl_vol_args *vol_args;
2636         bool restore_op = false;
2637         int ret;
2638
2639         if (!capable(CAP_SYS_ADMIN))
2640                 return -EPERM;
2641
2642         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2643                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
2644                 return -EINVAL;
2645         }
2646
2647         if (fs_info->fs_devices->temp_fsid) {
2648                 btrfs_err(fs_info,
2649                           "device add not supported on cloned temp-fsid mount");
2650                 return -EINVAL;
2651         }
2652
2653         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
2654                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
2655                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2656
2657                 /*
2658                  * We can do the device add because we have a paused balanced,
2659                  * change the exclusive op type and remember we should bring
2660                  * back the paused balance
2661                  */
2662                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
2663                 btrfs_exclop_start_unlock(fs_info);
2664                 restore_op = true;
2665         }
2666
2667         vol_args = memdup_user(arg, sizeof(*vol_args));
2668         if (IS_ERR(vol_args)) {
2669                 ret = PTR_ERR(vol_args);
2670                 goto out;
2671         }
2672
2673         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2674         ret = btrfs_init_new_device(fs_info, vol_args->name);
2675
2676         if (!ret)
2677                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2678
2679         kfree(vol_args);
2680 out:
2681         if (restore_op)
2682                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
2683         else
2684                 btrfs_exclop_finish(fs_info);
2685         return ret;
2686 }
2687
2688 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2689 {
2690         BTRFS_DEV_LOOKUP_ARGS(args);
2691         struct inode *inode = file_inode(file);
2692         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2693         struct btrfs_ioctl_vol_args_v2 *vol_args;
2694         struct bdev_handle *bdev_handle = NULL;
2695         int ret;
2696         bool cancel = false;
2697
2698         if (!capable(CAP_SYS_ADMIN))
2699                 return -EPERM;
2700
2701         vol_args = memdup_user(arg, sizeof(*vol_args));
2702         if (IS_ERR(vol_args))
2703                 return PTR_ERR(vol_args);
2704
2705         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
2706                 ret = -EOPNOTSUPP;
2707                 goto out;
2708         }
2709
2710         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2711         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2712                 args.devid = vol_args->devid;
2713         } else if (!strcmp("cancel", vol_args->name)) {
2714                 cancel = true;
2715         } else {
2716                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2717                 if (ret)
2718                         goto out;
2719         }
2720
2721         ret = mnt_want_write_file(file);
2722         if (ret)
2723                 goto out;
2724
2725         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2726                                            cancel);
2727         if (ret)
2728                 goto err_drop;
2729
2730         /* Exclusive operation is now claimed */
2731         ret = btrfs_rm_device(fs_info, &args, &bdev_handle);
2732
2733         btrfs_exclop_finish(fs_info);
2734
2735         if (!ret) {
2736                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2737                         btrfs_info(fs_info, "device deleted: id %llu",
2738                                         vol_args->devid);
2739                 else
2740                         btrfs_info(fs_info, "device deleted: %s",
2741                                         vol_args->name);
2742         }
2743 err_drop:
2744         mnt_drop_write_file(file);
2745         if (bdev_handle)
2746                 bdev_release(bdev_handle);
2747 out:
2748         btrfs_put_dev_args_from_path(&args);
2749         kfree(vol_args);
2750         return ret;
2751 }
2752
2753 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2754 {
2755         BTRFS_DEV_LOOKUP_ARGS(args);
2756         struct inode *inode = file_inode(file);
2757         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2758         struct btrfs_ioctl_vol_args *vol_args;
2759         struct bdev_handle *bdev_handle = NULL;
2760         int ret;
2761         bool cancel = false;
2762
2763         if (!capable(CAP_SYS_ADMIN))
2764                 return -EPERM;
2765
2766         vol_args = memdup_user(arg, sizeof(*vol_args));
2767         if (IS_ERR(vol_args))
2768                 return PTR_ERR(vol_args);
2769
2770         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2771         if (!strcmp("cancel", vol_args->name)) {
2772                 cancel = true;
2773         } else {
2774                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2775                 if (ret)
2776                         goto out;
2777         }
2778
2779         ret = mnt_want_write_file(file);
2780         if (ret)
2781                 goto out;
2782
2783         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2784                                            cancel);
2785         if (ret == 0) {
2786                 ret = btrfs_rm_device(fs_info, &args, &bdev_handle);
2787                 if (!ret)
2788                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2789                 btrfs_exclop_finish(fs_info);
2790         }
2791
2792         mnt_drop_write_file(file);
2793         if (bdev_handle)
2794                 bdev_release(bdev_handle);
2795 out:
2796         btrfs_put_dev_args_from_path(&args);
2797         kfree(vol_args);
2798         return ret;
2799 }
2800
2801 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2802                                 void __user *arg)
2803 {
2804         struct btrfs_ioctl_fs_info_args *fi_args;
2805         struct btrfs_device *device;
2806         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2807         u64 flags_in;
2808         int ret = 0;
2809
2810         fi_args = memdup_user(arg, sizeof(*fi_args));
2811         if (IS_ERR(fi_args))
2812                 return PTR_ERR(fi_args);
2813
2814         flags_in = fi_args->flags;
2815         memset(fi_args, 0, sizeof(*fi_args));
2816
2817         rcu_read_lock();
2818         fi_args->num_devices = fs_devices->num_devices;
2819
2820         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2821                 if (device->devid > fi_args->max_id)
2822                         fi_args->max_id = device->devid;
2823         }
2824         rcu_read_unlock();
2825
2826         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
2827         fi_args->nodesize = fs_info->nodesize;
2828         fi_args->sectorsize = fs_info->sectorsize;
2829         fi_args->clone_alignment = fs_info->sectorsize;
2830
2831         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
2832                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
2833                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
2834                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
2835         }
2836
2837         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
2838                 fi_args->generation = btrfs_get_fs_generation(fs_info);
2839                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
2840         }
2841
2842         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
2843                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
2844                        sizeof(fi_args->metadata_uuid));
2845                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
2846         }
2847
2848         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2849                 ret = -EFAULT;
2850
2851         kfree(fi_args);
2852         return ret;
2853 }
2854
2855 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2856                                  void __user *arg)
2857 {
2858         BTRFS_DEV_LOOKUP_ARGS(args);
2859         struct btrfs_ioctl_dev_info_args *di_args;
2860         struct btrfs_device *dev;
2861         int ret = 0;
2862
2863         di_args = memdup_user(arg, sizeof(*di_args));
2864         if (IS_ERR(di_args))
2865                 return PTR_ERR(di_args);
2866
2867         args.devid = di_args->devid;
2868         if (!btrfs_is_empty_uuid(di_args->uuid))
2869                 args.uuid = di_args->uuid;
2870
2871         rcu_read_lock();
2872         dev = btrfs_find_device(fs_info->fs_devices, &args);
2873         if (!dev) {
2874                 ret = -ENODEV;
2875                 goto out;
2876         }
2877
2878         di_args->devid = dev->devid;
2879         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2880         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2881         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2882         memcpy(di_args->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2883         if (dev->name)
2884                 strscpy(di_args->path, btrfs_dev_name(dev), sizeof(di_args->path));
2885         else
2886                 di_args->path[0] = '\0';
2887
2888 out:
2889         rcu_read_unlock();
2890         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2891                 ret = -EFAULT;
2892
2893         kfree(di_args);
2894         return ret;
2895 }
2896
2897 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2898 {
2899         struct inode *inode = file_inode(file);
2900         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2901         struct btrfs_root *root = BTRFS_I(inode)->root;
2902         struct btrfs_root *new_root;
2903         struct btrfs_dir_item *di;
2904         struct btrfs_trans_handle *trans;
2905         struct btrfs_path *path = NULL;
2906         struct btrfs_disk_key disk_key;
2907         struct fscrypt_str name = FSTR_INIT("default", 7);
2908         u64 objectid = 0;
2909         u64 dir_id;
2910         int ret;
2911
2912         if (!capable(CAP_SYS_ADMIN))
2913                 return -EPERM;
2914
2915         ret = mnt_want_write_file(file);
2916         if (ret)
2917                 return ret;
2918
2919         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
2920                 ret = -EFAULT;
2921                 goto out;
2922         }
2923
2924         if (!objectid)
2925                 objectid = BTRFS_FS_TREE_OBJECTID;
2926
2927         new_root = btrfs_get_fs_root(fs_info, objectid, true);
2928         if (IS_ERR(new_root)) {
2929                 ret = PTR_ERR(new_root);
2930                 goto out;
2931         }
2932         if (!is_fstree(new_root->root_key.objectid)) {
2933                 ret = -ENOENT;
2934                 goto out_free;
2935         }
2936
2937         path = btrfs_alloc_path();
2938         if (!path) {
2939                 ret = -ENOMEM;
2940                 goto out_free;
2941         }
2942
2943         trans = btrfs_start_transaction(root, 1);
2944         if (IS_ERR(trans)) {
2945                 ret = PTR_ERR(trans);
2946                 goto out_free;
2947         }
2948
2949         dir_id = btrfs_super_root_dir(fs_info->super_copy);
2950         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
2951                                    dir_id, &name, 1);
2952         if (IS_ERR_OR_NULL(di)) {
2953                 btrfs_release_path(path);
2954                 btrfs_end_transaction(trans);
2955                 btrfs_err(fs_info,
2956                           "Umm, you don't have the default diritem, this isn't going to work");
2957                 ret = -ENOENT;
2958                 goto out_free;
2959         }
2960
2961         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2962         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2963         btrfs_mark_buffer_dirty(trans, path->nodes[0]);
2964         btrfs_release_path(path);
2965
2966         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
2967         btrfs_end_transaction(trans);
2968 out_free:
2969         btrfs_put_root(new_root);
2970         btrfs_free_path(path);
2971 out:
2972         mnt_drop_write_file(file);
2973         return ret;
2974 }
2975
2976 static void get_block_group_info(struct list_head *groups_list,
2977                                  struct btrfs_ioctl_space_info *space)
2978 {
2979         struct btrfs_block_group *block_group;
2980
2981         space->total_bytes = 0;
2982         space->used_bytes = 0;
2983         space->flags = 0;
2984         list_for_each_entry(block_group, groups_list, list) {
2985                 space->flags = block_group->flags;
2986                 space->total_bytes += block_group->length;
2987                 space->used_bytes += block_group->used;
2988         }
2989 }
2990
2991 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
2992                                    void __user *arg)
2993 {
2994         struct btrfs_ioctl_space_args space_args = { 0 };
2995         struct btrfs_ioctl_space_info space;
2996         struct btrfs_ioctl_space_info *dest;
2997         struct btrfs_ioctl_space_info *dest_orig;
2998         struct btrfs_ioctl_space_info __user *user_dest;
2999         struct btrfs_space_info *info;
3000         static const u64 types[] = {
3001                 BTRFS_BLOCK_GROUP_DATA,
3002                 BTRFS_BLOCK_GROUP_SYSTEM,
3003                 BTRFS_BLOCK_GROUP_METADATA,
3004                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3005         };
3006         int num_types = 4;
3007         int alloc_size;
3008         int ret = 0;
3009         u64 slot_count = 0;
3010         int i, c;
3011
3012         if (copy_from_user(&space_args,
3013                            (struct btrfs_ioctl_space_args __user *)arg,
3014                            sizeof(space_args)))
3015                 return -EFAULT;
3016
3017         for (i = 0; i < num_types; i++) {
3018                 struct btrfs_space_info *tmp;
3019
3020                 info = NULL;
3021                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3022                         if (tmp->flags == types[i]) {
3023                                 info = tmp;
3024                                 break;
3025                         }
3026                 }
3027
3028                 if (!info)
3029                         continue;
3030
3031                 down_read(&info->groups_sem);
3032                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3033                         if (!list_empty(&info->block_groups[c]))
3034                                 slot_count++;
3035                 }
3036                 up_read(&info->groups_sem);
3037         }
3038
3039         /*
3040          * Global block reserve, exported as a space_info
3041          */
3042         slot_count++;
3043
3044         /* space_slots == 0 means they are asking for a count */
3045         if (space_args.space_slots == 0) {
3046                 space_args.total_spaces = slot_count;
3047                 goto out;
3048         }
3049
3050         slot_count = min_t(u64, space_args.space_slots, slot_count);
3051
3052         alloc_size = sizeof(*dest) * slot_count;
3053
3054         /* we generally have at most 6 or so space infos, one for each raid
3055          * level.  So, a whole page should be more than enough for everyone
3056          */
3057         if (alloc_size > PAGE_SIZE)
3058                 return -ENOMEM;
3059
3060         space_args.total_spaces = 0;
3061         dest = kmalloc(alloc_size, GFP_KERNEL);
3062         if (!dest)
3063                 return -ENOMEM;
3064         dest_orig = dest;
3065
3066         /* now we have a buffer to copy into */
3067         for (i = 0; i < num_types; i++) {
3068                 struct btrfs_space_info *tmp;
3069
3070                 if (!slot_count)
3071                         break;
3072
3073                 info = NULL;
3074                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3075                         if (tmp->flags == types[i]) {
3076                                 info = tmp;
3077                                 break;
3078                         }
3079                 }
3080
3081                 if (!info)
3082                         continue;
3083                 down_read(&info->groups_sem);
3084                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3085                         if (!list_empty(&info->block_groups[c])) {
3086                                 get_block_group_info(&info->block_groups[c],
3087                                                      &space);
3088                                 memcpy(dest, &space, sizeof(space));
3089                                 dest++;
3090                                 space_args.total_spaces++;
3091                                 slot_count--;
3092                         }
3093                         if (!slot_count)
3094                                 break;
3095                 }
3096                 up_read(&info->groups_sem);
3097         }
3098
3099         /*
3100          * Add global block reserve
3101          */
3102         if (slot_count) {
3103                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3104
3105                 spin_lock(&block_rsv->lock);
3106                 space.total_bytes = block_rsv->size;
3107                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3108                 spin_unlock(&block_rsv->lock);
3109                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3110                 memcpy(dest, &space, sizeof(space));
3111                 space_args.total_spaces++;
3112         }
3113
3114         user_dest = (struct btrfs_ioctl_space_info __user *)
3115                 (arg + sizeof(struct btrfs_ioctl_space_args));
3116
3117         if (copy_to_user(user_dest, dest_orig, alloc_size))
3118                 ret = -EFAULT;
3119
3120         kfree(dest_orig);
3121 out:
3122         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3123                 ret = -EFAULT;
3124
3125         return ret;
3126 }
3127
3128 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3129                                             void __user *argp)
3130 {
3131         struct btrfs_trans_handle *trans;
3132         u64 transid;
3133
3134         /*
3135          * Start orphan cleanup here for the given root in case it hasn't been
3136          * started already by other means. Errors are handled in the other
3137          * functions during transaction commit.
3138          */
3139         btrfs_orphan_cleanup(root);
3140
3141         trans = btrfs_attach_transaction_barrier(root);
3142         if (IS_ERR(trans)) {
3143                 if (PTR_ERR(trans) != -ENOENT)
3144                         return PTR_ERR(trans);
3145
3146                 /* No running transaction, don't bother */
3147                 transid = btrfs_get_last_trans_committed(root->fs_info);
3148                 goto out;
3149         }
3150         transid = trans->transid;
3151         btrfs_commit_transaction_async(trans);
3152 out:
3153         if (argp)
3154                 if (copy_to_user(argp, &transid, sizeof(transid)))
3155                         return -EFAULT;
3156         return 0;
3157 }
3158
3159 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3160                                            void __user *argp)
3161 {
3162         /* By default wait for the current transaction. */
3163         u64 transid = 0;
3164
3165         if (argp)
3166                 if (copy_from_user(&transid, argp, sizeof(transid)))
3167                         return -EFAULT;
3168
3169         return btrfs_wait_for_commit(fs_info, transid);
3170 }
3171
3172 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3173 {
3174         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3175         struct btrfs_ioctl_scrub_args *sa;
3176         int ret;
3177
3178         if (!capable(CAP_SYS_ADMIN))
3179                 return -EPERM;
3180
3181         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3182                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
3183                 return -EINVAL;
3184         }
3185
3186         sa = memdup_user(arg, sizeof(*sa));
3187         if (IS_ERR(sa))
3188                 return PTR_ERR(sa);
3189
3190         if (sa->flags & ~BTRFS_SCRUB_SUPPORTED_FLAGS) {
3191                 ret = -EOPNOTSUPP;
3192                 goto out;
3193         }
3194
3195         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3196                 ret = mnt_want_write_file(file);
3197                 if (ret)
3198                         goto out;
3199         }
3200
3201         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3202                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3203                               0);
3204
3205         /*
3206          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3207          * error. This is important as it allows user space to know how much
3208          * progress scrub has done. For example, if scrub is canceled we get
3209          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3210          * space. Later user space can inspect the progress from the structure
3211          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3212          * previously (btrfs-progs does this).
3213          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3214          * then return -EFAULT to signal the structure was not copied or it may
3215          * be corrupt and unreliable due to a partial copy.
3216          */
3217         if (copy_to_user(arg, sa, sizeof(*sa)))
3218                 ret = -EFAULT;
3219
3220         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3221                 mnt_drop_write_file(file);
3222 out:
3223         kfree(sa);
3224         return ret;
3225 }
3226
3227 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3228 {
3229         if (!capable(CAP_SYS_ADMIN))
3230                 return -EPERM;
3231
3232         return btrfs_scrub_cancel(fs_info);
3233 }
3234
3235 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3236                                        void __user *arg)
3237 {
3238         struct btrfs_ioctl_scrub_args *sa;
3239         int ret;
3240
3241         if (!capable(CAP_SYS_ADMIN))
3242                 return -EPERM;
3243
3244         sa = memdup_user(arg, sizeof(*sa));
3245         if (IS_ERR(sa))
3246                 return PTR_ERR(sa);
3247
3248         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3249
3250         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3251                 ret = -EFAULT;
3252
3253         kfree(sa);
3254         return ret;
3255 }
3256
3257 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3258                                       void __user *arg)
3259 {
3260         struct btrfs_ioctl_get_dev_stats *sa;
3261         int ret;
3262
3263         sa = memdup_user(arg, sizeof(*sa));
3264         if (IS_ERR(sa))
3265                 return PTR_ERR(sa);
3266
3267         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3268                 kfree(sa);
3269                 return -EPERM;
3270         }
3271
3272         ret = btrfs_get_dev_stats(fs_info, sa);
3273
3274         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3275                 ret = -EFAULT;
3276
3277         kfree(sa);
3278         return ret;
3279 }
3280
3281 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3282                                     void __user *arg)
3283 {
3284         struct btrfs_ioctl_dev_replace_args *p;
3285         int ret;
3286
3287         if (!capable(CAP_SYS_ADMIN))
3288                 return -EPERM;
3289
3290         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3291                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
3292                 return -EINVAL;
3293         }
3294
3295         p = memdup_user(arg, sizeof(*p));
3296         if (IS_ERR(p))
3297                 return PTR_ERR(p);
3298
3299         switch (p->cmd) {
3300         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3301                 if (sb_rdonly(fs_info->sb)) {
3302                         ret = -EROFS;
3303                         goto out;
3304                 }
3305                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3306                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3307                 } else {
3308                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3309                         btrfs_exclop_finish(fs_info);
3310                 }
3311                 break;
3312         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3313                 btrfs_dev_replace_status(fs_info, p);
3314                 ret = 0;
3315                 break;
3316         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3317                 p->result = btrfs_dev_replace_cancel(fs_info);
3318                 ret = 0;
3319                 break;
3320         default:
3321                 ret = -EINVAL;
3322                 break;
3323         }
3324
3325         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3326                 ret = -EFAULT;
3327 out:
3328         kfree(p);
3329         return ret;
3330 }
3331
3332 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3333 {
3334         int ret = 0;
3335         int i;
3336         u64 rel_ptr;
3337         int size;
3338         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3339         struct inode_fs_paths *ipath = NULL;
3340         struct btrfs_path *path;
3341
3342         if (!capable(CAP_DAC_READ_SEARCH))
3343                 return -EPERM;
3344
3345         path = btrfs_alloc_path();
3346         if (!path) {
3347                 ret = -ENOMEM;
3348                 goto out;
3349         }
3350
3351         ipa = memdup_user(arg, sizeof(*ipa));
3352         if (IS_ERR(ipa)) {
3353                 ret = PTR_ERR(ipa);
3354                 ipa = NULL;
3355                 goto out;
3356         }
3357
3358         size = min_t(u32, ipa->size, 4096);
3359         ipath = init_ipath(size, root, path);
3360         if (IS_ERR(ipath)) {
3361                 ret = PTR_ERR(ipath);
3362                 ipath = NULL;
3363                 goto out;
3364         }
3365
3366         ret = paths_from_inode(ipa->inum, ipath);
3367         if (ret < 0)
3368                 goto out;
3369
3370         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3371                 rel_ptr = ipath->fspath->val[i] -
3372                           (u64)(unsigned long)ipath->fspath->val;
3373                 ipath->fspath->val[i] = rel_ptr;
3374         }
3375
3376         btrfs_free_path(path);
3377         path = NULL;
3378         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3379                            ipath->fspath, size);
3380         if (ret) {
3381                 ret = -EFAULT;
3382                 goto out;
3383         }
3384
3385 out:
3386         btrfs_free_path(path);
3387         free_ipath(ipath);
3388         kfree(ipa);
3389
3390         return ret;
3391 }
3392
3393 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3394                                         void __user *arg, int version)
3395 {
3396         int ret = 0;
3397         int size;
3398         struct btrfs_ioctl_logical_ino_args *loi;
3399         struct btrfs_data_container *inodes = NULL;
3400         struct btrfs_path *path = NULL;
3401         bool ignore_offset;
3402
3403         if (!capable(CAP_SYS_ADMIN))
3404                 return -EPERM;
3405
3406         loi = memdup_user(arg, sizeof(*loi));
3407         if (IS_ERR(loi))
3408                 return PTR_ERR(loi);
3409
3410         if (version == 1) {
3411                 ignore_offset = false;
3412                 size = min_t(u32, loi->size, SZ_64K);
3413         } else {
3414                 /* All reserved bits must be 0 for now */
3415                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3416                         ret = -EINVAL;
3417                         goto out_loi;
3418                 }
3419                 /* Only accept flags we have defined so far */
3420                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3421                         ret = -EINVAL;
3422                         goto out_loi;
3423                 }
3424                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3425                 size = min_t(u32, loi->size, SZ_16M);
3426         }
3427
3428         inodes = init_data_container(size);
3429         if (IS_ERR(inodes)) {
3430                 ret = PTR_ERR(inodes);
3431                 goto out_loi;
3432         }
3433
3434         path = btrfs_alloc_path();
3435         if (!path) {
3436                 ret = -ENOMEM;
3437                 goto out;
3438         }
3439         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3440                                           inodes, ignore_offset);
3441         btrfs_free_path(path);
3442         if (ret == -EINVAL)
3443                 ret = -ENOENT;
3444         if (ret < 0)
3445                 goto out;
3446
3447         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3448                            size);
3449         if (ret)
3450                 ret = -EFAULT;
3451
3452 out:
3453         kvfree(inodes);
3454 out_loi:
3455         kfree(loi);
3456
3457         return ret;
3458 }
3459
3460 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3461                                struct btrfs_ioctl_balance_args *bargs)
3462 {
3463         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3464
3465         bargs->flags = bctl->flags;
3466
3467         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3468                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3469         if (atomic_read(&fs_info->balance_pause_req))
3470                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3471         if (atomic_read(&fs_info->balance_cancel_req))
3472                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3473
3474         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3475         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3476         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3477
3478         spin_lock(&fs_info->balance_lock);
3479         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3480         spin_unlock(&fs_info->balance_lock);
3481 }
3482
3483 /*
3484  * Try to acquire fs_info::balance_mutex as well as set BTRFS_EXLCOP_BALANCE as
3485  * required.
3486  *
3487  * @fs_info:       the filesystem
3488  * @excl_acquired: ptr to boolean value which is set to false in case balance
3489  *                 is being resumed
3490  *
3491  * Return 0 on success in which case both fs_info::balance is acquired as well
3492  * as exclusive ops are blocked. In case of failure return an error code.
3493  */
3494 static int btrfs_try_lock_balance(struct btrfs_fs_info *fs_info, bool *excl_acquired)
3495 {
3496         int ret;
3497
3498         /*
3499          * Exclusive operation is locked. Three possibilities:
3500          *   (1) some other op is running
3501          *   (2) balance is running
3502          *   (3) balance is paused -- special case (think resume)
3503          */
3504         while (1) {
3505                 if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
3506                         *excl_acquired = true;
3507                         mutex_lock(&fs_info->balance_mutex);
3508                         return 0;
3509                 }
3510
3511                 mutex_lock(&fs_info->balance_mutex);
3512                 if (fs_info->balance_ctl) {
3513                         /* This is either (2) or (3) */
3514                         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3515                                 /* This is (2) */
3516                                 ret = -EINPROGRESS;
3517                                 goto out_failure;
3518
3519                         } else {
3520                                 mutex_unlock(&fs_info->balance_mutex);
3521                                 /*
3522                                  * Lock released to allow other waiters to
3523                                  * continue, we'll reexamine the status again.
3524                                  */
3525                                 mutex_lock(&fs_info->balance_mutex);
3526
3527                                 if (fs_info->balance_ctl &&
3528                                     !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3529                                         /* This is (3) */
3530                                         *excl_acquired = false;
3531                                         return 0;
3532                                 }
3533                         }
3534                 } else {
3535                         /* This is (1) */
3536                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3537                         goto out_failure;
3538                 }
3539
3540                 mutex_unlock(&fs_info->balance_mutex);
3541         }
3542
3543 out_failure:
3544         mutex_unlock(&fs_info->balance_mutex);
3545         *excl_acquired = false;
3546         return ret;
3547 }
3548
3549 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3550 {
3551         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3552         struct btrfs_fs_info *fs_info = root->fs_info;
3553         struct btrfs_ioctl_balance_args *bargs;
3554         struct btrfs_balance_control *bctl;
3555         bool need_unlock = true;
3556         int ret;
3557
3558         if (!capable(CAP_SYS_ADMIN))
3559                 return -EPERM;
3560
3561         ret = mnt_want_write_file(file);
3562         if (ret)
3563                 return ret;
3564
3565         bargs = memdup_user(arg, sizeof(*bargs));
3566         if (IS_ERR(bargs)) {
3567                 ret = PTR_ERR(bargs);
3568                 bargs = NULL;
3569                 goto out;
3570         }
3571
3572         ret = btrfs_try_lock_balance(fs_info, &need_unlock);
3573         if (ret)
3574                 goto out;
3575
3576         lockdep_assert_held(&fs_info->balance_mutex);
3577
3578         if (bargs->flags & BTRFS_BALANCE_RESUME) {
3579                 if (!fs_info->balance_ctl) {
3580                         ret = -ENOTCONN;
3581                         goto out_unlock;
3582                 }
3583
3584                 bctl = fs_info->balance_ctl;
3585                 spin_lock(&fs_info->balance_lock);
3586                 bctl->flags |= BTRFS_BALANCE_RESUME;
3587                 spin_unlock(&fs_info->balance_lock);
3588                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
3589
3590                 goto do_balance;
3591         }
3592
3593         if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
3594                 ret = -EINVAL;
3595                 goto out_unlock;
3596         }
3597
3598         if (fs_info->balance_ctl) {
3599                 ret = -EINPROGRESS;
3600                 goto out_unlock;
3601         }
3602
3603         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
3604         if (!bctl) {
3605                 ret = -ENOMEM;
3606                 goto out_unlock;
3607         }
3608
3609         memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3610         memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3611         memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3612
3613         bctl->flags = bargs->flags;
3614 do_balance:
3615         /*
3616          * Ownership of bctl and exclusive operation goes to btrfs_balance.
3617          * bctl is freed in reset_balance_state, or, if restriper was paused
3618          * all the way until unmount, in free_fs_info.  The flag should be
3619          * cleared after reset_balance_state.
3620          */
3621         need_unlock = false;
3622
3623         ret = btrfs_balance(fs_info, bctl, bargs);
3624         bctl = NULL;
3625
3626         if (ret == 0 || ret == -ECANCELED) {
3627                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3628                         ret = -EFAULT;
3629         }
3630
3631         kfree(bctl);
3632 out_unlock:
3633         mutex_unlock(&fs_info->balance_mutex);
3634         if (need_unlock)
3635                 btrfs_exclop_finish(fs_info);
3636 out:
3637         mnt_drop_write_file(file);
3638         kfree(bargs);
3639         return ret;
3640 }
3641
3642 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
3643 {
3644         if (!capable(CAP_SYS_ADMIN))
3645                 return -EPERM;
3646
3647         switch (cmd) {
3648         case BTRFS_BALANCE_CTL_PAUSE:
3649                 return btrfs_pause_balance(fs_info);
3650         case BTRFS_BALANCE_CTL_CANCEL:
3651                 return btrfs_cancel_balance(fs_info);
3652         }
3653
3654         return -EINVAL;
3655 }
3656
3657 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
3658                                          void __user *arg)
3659 {
3660         struct btrfs_ioctl_balance_args *bargs;
3661         int ret = 0;
3662
3663         if (!capable(CAP_SYS_ADMIN))
3664                 return -EPERM;
3665
3666         mutex_lock(&fs_info->balance_mutex);
3667         if (!fs_info->balance_ctl) {
3668                 ret = -ENOTCONN;
3669                 goto out;
3670         }
3671
3672         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
3673         if (!bargs) {
3674                 ret = -ENOMEM;
3675                 goto out;
3676         }
3677
3678         btrfs_update_ioctl_balance_args(fs_info, bargs);
3679
3680         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3681                 ret = -EFAULT;
3682
3683         kfree(bargs);
3684 out:
3685         mutex_unlock(&fs_info->balance_mutex);
3686         return ret;
3687 }
3688
3689 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
3690 {
3691         struct inode *inode = file_inode(file);
3692         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3693         struct btrfs_ioctl_quota_ctl_args *sa;
3694         int ret;
3695
3696         if (!capable(CAP_SYS_ADMIN))
3697                 return -EPERM;
3698
3699         ret = mnt_want_write_file(file);
3700         if (ret)
3701                 return ret;
3702
3703         sa = memdup_user(arg, sizeof(*sa));
3704         if (IS_ERR(sa)) {
3705                 ret = PTR_ERR(sa);
3706                 goto drop_write;
3707         }
3708
3709         down_write(&fs_info->subvol_sem);
3710
3711         switch (sa->cmd) {
3712         case BTRFS_QUOTA_CTL_ENABLE:
3713         case BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA:
3714                 ret = btrfs_quota_enable(fs_info, sa);
3715                 break;
3716         case BTRFS_QUOTA_CTL_DISABLE:
3717                 ret = btrfs_quota_disable(fs_info);
3718                 break;
3719         default:
3720                 ret = -EINVAL;
3721                 break;
3722         }
3723
3724         kfree(sa);
3725         up_write(&fs_info->subvol_sem);
3726 drop_write:
3727         mnt_drop_write_file(file);
3728         return ret;
3729 }
3730
3731 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
3732 {
3733         struct inode *inode = file_inode(file);
3734         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3735         struct btrfs_root *root = BTRFS_I(inode)->root;
3736         struct btrfs_ioctl_qgroup_assign_args *sa;
3737         struct btrfs_trans_handle *trans;
3738         int ret;
3739         int err;
3740
3741         if (!capable(CAP_SYS_ADMIN))
3742                 return -EPERM;
3743
3744         ret = mnt_want_write_file(file);
3745         if (ret)
3746                 return ret;
3747
3748         sa = memdup_user(arg, sizeof(*sa));
3749         if (IS_ERR(sa)) {
3750                 ret = PTR_ERR(sa);
3751                 goto drop_write;
3752         }
3753
3754         trans = btrfs_join_transaction(root);
3755         if (IS_ERR(trans)) {
3756                 ret = PTR_ERR(trans);
3757                 goto out;
3758         }
3759
3760         if (sa->assign) {
3761                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
3762         } else {
3763                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
3764         }
3765
3766         /* update qgroup status and info */
3767         mutex_lock(&fs_info->qgroup_ioctl_lock);
3768         err = btrfs_run_qgroups(trans);
3769         mutex_unlock(&fs_info->qgroup_ioctl_lock);
3770         if (err < 0)
3771                 btrfs_handle_fs_error(fs_info, err,
3772                                       "failed to update qgroup status and info");
3773         err = btrfs_end_transaction(trans);
3774         if (err && !ret)
3775                 ret = err;
3776
3777 out:
3778         kfree(sa);
3779 drop_write:
3780         mnt_drop_write_file(file);
3781         return ret;
3782 }
3783
3784 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
3785 {
3786         struct inode *inode = file_inode(file);
3787         struct btrfs_root *root = BTRFS_I(inode)->root;
3788         struct btrfs_ioctl_qgroup_create_args *sa;
3789         struct btrfs_trans_handle *trans;
3790         int ret;
3791         int err;
3792
3793         if (!capable(CAP_SYS_ADMIN))
3794                 return -EPERM;
3795
3796         ret = mnt_want_write_file(file);
3797         if (ret)
3798                 return ret;
3799
3800         sa = memdup_user(arg, sizeof(*sa));
3801         if (IS_ERR(sa)) {
3802                 ret = PTR_ERR(sa);
3803                 goto drop_write;
3804         }
3805
3806         if (!sa->qgroupid) {
3807                 ret = -EINVAL;
3808                 goto out;
3809         }
3810
3811         trans = btrfs_join_transaction(root);
3812         if (IS_ERR(trans)) {
3813                 ret = PTR_ERR(trans);
3814                 goto out;
3815         }
3816
3817         if (sa->create) {
3818                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
3819         } else {
3820                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
3821         }
3822
3823         err = btrfs_end_transaction(trans);
3824         if (err && !ret)
3825                 ret = err;
3826
3827 out:
3828         kfree(sa);
3829 drop_write:
3830         mnt_drop_write_file(file);
3831         return ret;
3832 }
3833
3834 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
3835 {
3836         struct inode *inode = file_inode(file);
3837         struct btrfs_root *root = BTRFS_I(inode)->root;
3838         struct btrfs_ioctl_qgroup_limit_args *sa;
3839         struct btrfs_trans_handle *trans;
3840         int ret;
3841         int err;
3842         u64 qgroupid;
3843
3844         if (!capable(CAP_SYS_ADMIN))
3845                 return -EPERM;
3846
3847         ret = mnt_want_write_file(file);
3848         if (ret)
3849                 return ret;
3850
3851         sa = memdup_user(arg, sizeof(*sa));
3852         if (IS_ERR(sa)) {
3853                 ret = PTR_ERR(sa);
3854                 goto drop_write;
3855         }
3856
3857         trans = btrfs_join_transaction(root);
3858         if (IS_ERR(trans)) {
3859                 ret = PTR_ERR(trans);
3860                 goto out;
3861         }
3862
3863         qgroupid = sa->qgroupid;
3864         if (!qgroupid) {
3865                 /* take the current subvol as qgroup */
3866                 qgroupid = root->root_key.objectid;
3867         }
3868
3869         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
3870
3871         err = btrfs_end_transaction(trans);
3872         if (err && !ret)
3873                 ret = err;
3874
3875 out:
3876         kfree(sa);
3877 drop_write:
3878         mnt_drop_write_file(file);
3879         return ret;
3880 }
3881
3882 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
3883 {
3884         struct inode *inode = file_inode(file);
3885         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3886         struct btrfs_ioctl_quota_rescan_args *qsa;
3887         int ret;
3888
3889         if (!capable(CAP_SYS_ADMIN))
3890                 return -EPERM;
3891
3892         ret = mnt_want_write_file(file);
3893         if (ret)
3894                 return ret;
3895
3896         qsa = memdup_user(arg, sizeof(*qsa));
3897         if (IS_ERR(qsa)) {
3898                 ret = PTR_ERR(qsa);
3899                 goto drop_write;
3900         }
3901
3902         if (qsa->flags) {
3903                 ret = -EINVAL;
3904                 goto out;
3905         }
3906
3907         ret = btrfs_qgroup_rescan(fs_info);
3908
3909 out:
3910         kfree(qsa);
3911 drop_write:
3912         mnt_drop_write_file(file);
3913         return ret;
3914 }
3915
3916 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
3917                                                 void __user *arg)
3918 {
3919         struct btrfs_ioctl_quota_rescan_args qsa = {0};
3920
3921         if (!capable(CAP_SYS_ADMIN))
3922                 return -EPERM;
3923
3924         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3925                 qsa.flags = 1;
3926                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
3927         }
3928
3929         if (copy_to_user(arg, &qsa, sizeof(qsa)))
3930                 return -EFAULT;
3931
3932         return 0;
3933 }
3934
3935 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
3936                                                 void __user *arg)
3937 {
3938         if (!capable(CAP_SYS_ADMIN))
3939                 return -EPERM;
3940
3941         return btrfs_qgroup_wait_for_completion(fs_info, true);
3942 }
3943
3944 static long _btrfs_ioctl_set_received_subvol(struct file *file,
3945                                             struct mnt_idmap *idmap,
3946                                             struct btrfs_ioctl_received_subvol_args *sa)
3947 {
3948         struct inode *inode = file_inode(file);
3949         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3950         struct btrfs_root *root = BTRFS_I(inode)->root;
3951         struct btrfs_root_item *root_item = &root->root_item;
3952         struct btrfs_trans_handle *trans;
3953         struct timespec64 ct = current_time(inode);
3954         int ret = 0;
3955         int received_uuid_changed;
3956
3957         if (!inode_owner_or_capable(idmap, inode))
3958                 return -EPERM;
3959
3960         ret = mnt_want_write_file(file);
3961         if (ret < 0)
3962                 return ret;
3963
3964         down_write(&fs_info->subvol_sem);
3965
3966         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3967                 ret = -EINVAL;
3968                 goto out;
3969         }
3970
3971         if (btrfs_root_readonly(root)) {
3972                 ret = -EROFS;
3973                 goto out;
3974         }
3975
3976         /*
3977          * 1 - root item
3978          * 2 - uuid items (received uuid + subvol uuid)
3979          */
3980         trans = btrfs_start_transaction(root, 3);
3981         if (IS_ERR(trans)) {
3982                 ret = PTR_ERR(trans);
3983                 trans = NULL;
3984                 goto out;
3985         }
3986
3987         sa->rtransid = trans->transid;
3988         sa->rtime.sec = ct.tv_sec;
3989         sa->rtime.nsec = ct.tv_nsec;
3990
3991         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
3992                                        BTRFS_UUID_SIZE);
3993         if (received_uuid_changed &&
3994             !btrfs_is_empty_uuid(root_item->received_uuid)) {
3995                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
3996                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3997                                           root->root_key.objectid);
3998                 if (ret && ret != -ENOENT) {
3999                         btrfs_abort_transaction(trans, ret);
4000                         btrfs_end_transaction(trans);
4001                         goto out;
4002                 }
4003         }
4004         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4005         btrfs_set_root_stransid(root_item, sa->stransid);
4006         btrfs_set_root_rtransid(root_item, sa->rtransid);
4007         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4008         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4009         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4010         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4011
4012         ret = btrfs_update_root(trans, fs_info->tree_root,
4013                                 &root->root_key, &root->root_item);
4014         if (ret < 0) {
4015                 btrfs_end_transaction(trans);
4016                 goto out;
4017         }
4018         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4019                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4020                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4021                                           root->root_key.objectid);
4022                 if (ret < 0 && ret != -EEXIST) {
4023                         btrfs_abort_transaction(trans, ret);
4024                         btrfs_end_transaction(trans);
4025                         goto out;
4026                 }
4027         }
4028         ret = btrfs_commit_transaction(trans);
4029 out:
4030         up_write(&fs_info->subvol_sem);
4031         mnt_drop_write_file(file);
4032         return ret;
4033 }
4034
4035 #ifdef CONFIG_64BIT
4036 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4037                                                 void __user *arg)
4038 {
4039         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4040         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4041         int ret = 0;
4042
4043         args32 = memdup_user(arg, sizeof(*args32));
4044         if (IS_ERR(args32))
4045                 return PTR_ERR(args32);
4046
4047         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4048         if (!args64) {
4049                 ret = -ENOMEM;
4050                 goto out;
4051         }
4052
4053         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4054         args64->stransid = args32->stransid;
4055         args64->rtransid = args32->rtransid;
4056         args64->stime.sec = args32->stime.sec;
4057         args64->stime.nsec = args32->stime.nsec;
4058         args64->rtime.sec = args32->rtime.sec;
4059         args64->rtime.nsec = args32->rtime.nsec;
4060         args64->flags = args32->flags;
4061
4062         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), args64);
4063         if (ret)
4064                 goto out;
4065
4066         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4067         args32->stransid = args64->stransid;
4068         args32->rtransid = args64->rtransid;
4069         args32->stime.sec = args64->stime.sec;
4070         args32->stime.nsec = args64->stime.nsec;
4071         args32->rtime.sec = args64->rtime.sec;
4072         args32->rtime.nsec = args64->rtime.nsec;
4073         args32->flags = args64->flags;
4074
4075         ret = copy_to_user(arg, args32, sizeof(*args32));
4076         if (ret)
4077                 ret = -EFAULT;
4078
4079 out:
4080         kfree(args32);
4081         kfree(args64);
4082         return ret;
4083 }
4084 #endif
4085
4086 static long btrfs_ioctl_set_received_subvol(struct file *file,
4087                                             void __user *arg)
4088 {
4089         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4090         int ret = 0;
4091
4092         sa = memdup_user(arg, sizeof(*sa));
4093         if (IS_ERR(sa))
4094                 return PTR_ERR(sa);
4095
4096         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), sa);
4097
4098         if (ret)
4099                 goto out;
4100
4101         ret = copy_to_user(arg, sa, sizeof(*sa));
4102         if (ret)
4103                 ret = -EFAULT;
4104
4105 out:
4106         kfree(sa);
4107         return ret;
4108 }
4109
4110 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4111                                         void __user *arg)
4112 {
4113         size_t len;
4114         int ret;
4115         char label[BTRFS_LABEL_SIZE];
4116
4117         spin_lock(&fs_info->super_lock);
4118         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4119         spin_unlock(&fs_info->super_lock);
4120
4121         len = strnlen(label, BTRFS_LABEL_SIZE);
4122
4123         if (len == BTRFS_LABEL_SIZE) {
4124                 btrfs_warn(fs_info,
4125                            "label is too long, return the first %zu bytes",
4126                            --len);
4127         }
4128
4129         ret = copy_to_user(arg, label, len);
4130
4131         return ret ? -EFAULT : 0;
4132 }
4133
4134 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4135 {
4136         struct inode *inode = file_inode(file);
4137         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4138         struct btrfs_root *root = BTRFS_I(inode)->root;
4139         struct btrfs_super_block *super_block = fs_info->super_copy;
4140         struct btrfs_trans_handle *trans;
4141         char label[BTRFS_LABEL_SIZE];
4142         int ret;
4143
4144         if (!capable(CAP_SYS_ADMIN))
4145                 return -EPERM;
4146
4147         if (copy_from_user(label, arg, sizeof(label)))
4148                 return -EFAULT;
4149
4150         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4151                 btrfs_err(fs_info,
4152                           "unable to set label with more than %d bytes",
4153                           BTRFS_LABEL_SIZE - 1);
4154                 return -EINVAL;
4155         }
4156
4157         ret = mnt_want_write_file(file);
4158         if (ret)
4159                 return ret;
4160
4161         trans = btrfs_start_transaction(root, 0);
4162         if (IS_ERR(trans)) {
4163                 ret = PTR_ERR(trans);
4164                 goto out_unlock;
4165         }
4166
4167         spin_lock(&fs_info->super_lock);
4168         strcpy(super_block->label, label);
4169         spin_unlock(&fs_info->super_lock);
4170         ret = btrfs_commit_transaction(trans);
4171
4172 out_unlock:
4173         mnt_drop_write_file(file);
4174         return ret;
4175 }
4176
4177 #define INIT_FEATURE_FLAGS(suffix) \
4178         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4179           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4180           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4181
4182 int btrfs_ioctl_get_supported_features(void __user *arg)
4183 {
4184         static const struct btrfs_ioctl_feature_flags features[3] = {
4185                 INIT_FEATURE_FLAGS(SUPP),
4186                 INIT_FEATURE_FLAGS(SAFE_SET),
4187                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4188         };
4189
4190         if (copy_to_user(arg, &features, sizeof(features)))
4191                 return -EFAULT;
4192
4193         return 0;
4194 }
4195
4196 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4197                                         void __user *arg)
4198 {
4199         struct btrfs_super_block *super_block = fs_info->super_copy;
4200         struct btrfs_ioctl_feature_flags features;
4201
4202         features.compat_flags = btrfs_super_compat_flags(super_block);
4203         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4204         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4205
4206         if (copy_to_user(arg, &features, sizeof(features)))
4207                 return -EFAULT;
4208
4209         return 0;
4210 }
4211
4212 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4213                               enum btrfs_feature_set set,
4214                               u64 change_mask, u64 flags, u64 supported_flags,
4215                               u64 safe_set, u64 safe_clear)
4216 {
4217         const char *type = btrfs_feature_set_name(set);
4218         char *names;
4219         u64 disallowed, unsupported;
4220         u64 set_mask = flags & change_mask;
4221         u64 clear_mask = ~flags & change_mask;
4222
4223         unsupported = set_mask & ~supported_flags;
4224         if (unsupported) {
4225                 names = btrfs_printable_features(set, unsupported);
4226                 if (names) {
4227                         btrfs_warn(fs_info,
4228                                    "this kernel does not support the %s feature bit%s",
4229                                    names, strchr(names, ',') ? "s" : "");
4230                         kfree(names);
4231                 } else
4232                         btrfs_warn(fs_info,
4233                                    "this kernel does not support %s bits 0x%llx",
4234                                    type, unsupported);
4235                 return -EOPNOTSUPP;
4236         }
4237
4238         disallowed = set_mask & ~safe_set;
4239         if (disallowed) {
4240                 names = btrfs_printable_features(set, disallowed);
4241                 if (names) {
4242                         btrfs_warn(fs_info,
4243                                    "can't set the %s feature bit%s while mounted",
4244                                    names, strchr(names, ',') ? "s" : "");
4245                         kfree(names);
4246                 } else
4247                         btrfs_warn(fs_info,
4248                                    "can't set %s bits 0x%llx while mounted",
4249                                    type, disallowed);
4250                 return -EPERM;
4251         }
4252
4253         disallowed = clear_mask & ~safe_clear;
4254         if (disallowed) {
4255                 names = btrfs_printable_features(set, disallowed);
4256                 if (names) {
4257                         btrfs_warn(fs_info,
4258                                    "can't clear the %s feature bit%s while mounted",
4259                                    names, strchr(names, ',') ? "s" : "");
4260                         kfree(names);
4261                 } else
4262                         btrfs_warn(fs_info,
4263                                    "can't clear %s bits 0x%llx while mounted",
4264                                    type, disallowed);
4265                 return -EPERM;
4266         }
4267
4268         return 0;
4269 }
4270
4271 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4272 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4273                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4274                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4275                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4276
4277 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4278 {
4279         struct inode *inode = file_inode(file);
4280         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4281         struct btrfs_root *root = BTRFS_I(inode)->root;
4282         struct btrfs_super_block *super_block = fs_info->super_copy;
4283         struct btrfs_ioctl_feature_flags flags[2];
4284         struct btrfs_trans_handle *trans;
4285         u64 newflags;
4286         int ret;
4287
4288         if (!capable(CAP_SYS_ADMIN))
4289                 return -EPERM;
4290
4291         if (copy_from_user(flags, arg, sizeof(flags)))
4292                 return -EFAULT;
4293
4294         /* Nothing to do */
4295         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4296             !flags[0].incompat_flags)
4297                 return 0;
4298
4299         ret = check_feature(fs_info, flags[0].compat_flags,
4300                             flags[1].compat_flags, COMPAT);
4301         if (ret)
4302                 return ret;
4303
4304         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4305                             flags[1].compat_ro_flags, COMPAT_RO);
4306         if (ret)
4307                 return ret;
4308
4309         ret = check_feature(fs_info, flags[0].incompat_flags,
4310                             flags[1].incompat_flags, INCOMPAT);
4311         if (ret)
4312                 return ret;
4313
4314         ret = mnt_want_write_file(file);
4315         if (ret)
4316                 return ret;
4317
4318         trans = btrfs_start_transaction(root, 0);
4319         if (IS_ERR(trans)) {
4320                 ret = PTR_ERR(trans);
4321                 goto out_drop_write;
4322         }
4323
4324         spin_lock(&fs_info->super_lock);
4325         newflags = btrfs_super_compat_flags(super_block);
4326         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4327         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4328         btrfs_set_super_compat_flags(super_block, newflags);
4329
4330         newflags = btrfs_super_compat_ro_flags(super_block);
4331         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4332         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4333         btrfs_set_super_compat_ro_flags(super_block, newflags);
4334
4335         newflags = btrfs_super_incompat_flags(super_block);
4336         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4337         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4338         btrfs_set_super_incompat_flags(super_block, newflags);
4339         spin_unlock(&fs_info->super_lock);
4340
4341         ret = btrfs_commit_transaction(trans);
4342 out_drop_write:
4343         mnt_drop_write_file(file);
4344
4345         return ret;
4346 }
4347
4348 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
4349 {
4350         struct btrfs_ioctl_send_args *arg;
4351         int ret;
4352
4353         if (compat) {
4354 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4355                 struct btrfs_ioctl_send_args_32 args32 = { 0 };
4356
4357                 ret = copy_from_user(&args32, argp, sizeof(args32));
4358                 if (ret)
4359                         return -EFAULT;
4360                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4361                 if (!arg)
4362                         return -ENOMEM;
4363                 arg->send_fd = args32.send_fd;
4364                 arg->clone_sources_count = args32.clone_sources_count;
4365                 arg->clone_sources = compat_ptr(args32.clone_sources);
4366                 arg->parent_root = args32.parent_root;
4367                 arg->flags = args32.flags;
4368                 arg->version = args32.version;
4369                 memcpy(arg->reserved, args32.reserved,
4370                        sizeof(args32.reserved));
4371 #else
4372                 return -ENOTTY;
4373 #endif
4374         } else {
4375                 arg = memdup_user(argp, sizeof(*arg));
4376                 if (IS_ERR(arg))
4377                         return PTR_ERR(arg);
4378         }
4379         ret = btrfs_ioctl_send(inode, arg);
4380         kfree(arg);
4381         return ret;
4382 }
4383
4384 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
4385                                     bool compat)
4386 {
4387         struct btrfs_ioctl_encoded_io_args args = { 0 };
4388         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
4389                                              flags);
4390         size_t copy_end;
4391         struct iovec iovstack[UIO_FASTIOV];
4392         struct iovec *iov = iovstack;
4393         struct iov_iter iter;
4394         loff_t pos;
4395         struct kiocb kiocb;
4396         ssize_t ret;
4397
4398         if (!capable(CAP_SYS_ADMIN)) {
4399                 ret = -EPERM;
4400                 goto out_acct;
4401         }
4402
4403         if (compat) {
4404 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4405                 struct btrfs_ioctl_encoded_io_args_32 args32;
4406
4407                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
4408                                        flags);
4409                 if (copy_from_user(&args32, argp, copy_end)) {
4410                         ret = -EFAULT;
4411                         goto out_acct;
4412                 }
4413                 args.iov = compat_ptr(args32.iov);
4414                 args.iovcnt = args32.iovcnt;
4415                 args.offset = args32.offset;
4416                 args.flags = args32.flags;
4417 #else
4418                 return -ENOTTY;
4419 #endif
4420         } else {
4421                 copy_end = copy_end_kernel;
4422                 if (copy_from_user(&args, argp, copy_end)) {
4423                         ret = -EFAULT;
4424                         goto out_acct;
4425                 }
4426         }
4427         if (args.flags != 0) {
4428                 ret = -EINVAL;
4429                 goto out_acct;
4430         }
4431
4432         ret = import_iovec(ITER_DEST, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4433                            &iov, &iter);
4434         if (ret < 0)
4435                 goto out_acct;
4436
4437         if (iov_iter_count(&iter) == 0) {
4438                 ret = 0;
4439                 goto out_iov;
4440         }
4441         pos = args.offset;
4442         ret = rw_verify_area(READ, file, &pos, args.len);
4443         if (ret < 0)
4444                 goto out_iov;
4445
4446         init_sync_kiocb(&kiocb, file);
4447         kiocb.ki_pos = pos;
4448
4449         ret = btrfs_encoded_read(&kiocb, &iter, &args);
4450         if (ret >= 0) {
4451                 fsnotify_access(file);
4452                 if (copy_to_user(argp + copy_end,
4453                                  (char *)&args + copy_end_kernel,
4454                                  sizeof(args) - copy_end_kernel))
4455                         ret = -EFAULT;
4456         }
4457
4458 out_iov:
4459         kfree(iov);
4460 out_acct:
4461         if (ret > 0)
4462                 add_rchar(current, ret);
4463         inc_syscr(current);
4464         return ret;
4465 }
4466
4467 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
4468 {
4469         struct btrfs_ioctl_encoded_io_args args;
4470         struct iovec iovstack[UIO_FASTIOV];
4471         struct iovec *iov = iovstack;
4472         struct iov_iter iter;
4473         loff_t pos;
4474         struct kiocb kiocb;
4475         ssize_t ret;
4476
4477         if (!capable(CAP_SYS_ADMIN)) {
4478                 ret = -EPERM;
4479                 goto out_acct;
4480         }
4481
4482         if (!(file->f_mode & FMODE_WRITE)) {
4483                 ret = -EBADF;
4484                 goto out_acct;
4485         }
4486
4487         if (compat) {
4488 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4489                 struct btrfs_ioctl_encoded_io_args_32 args32;
4490
4491                 if (copy_from_user(&args32, argp, sizeof(args32))) {
4492                         ret = -EFAULT;
4493                         goto out_acct;
4494                 }
4495                 args.iov = compat_ptr(args32.iov);
4496                 args.iovcnt = args32.iovcnt;
4497                 args.offset = args32.offset;
4498                 args.flags = args32.flags;
4499                 args.len = args32.len;
4500                 args.unencoded_len = args32.unencoded_len;
4501                 args.unencoded_offset = args32.unencoded_offset;
4502                 args.compression = args32.compression;
4503                 args.encryption = args32.encryption;
4504                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
4505 #else
4506                 return -ENOTTY;
4507 #endif
4508         } else {
4509                 if (copy_from_user(&args, argp, sizeof(args))) {
4510                         ret = -EFAULT;
4511                         goto out_acct;
4512                 }
4513         }
4514
4515         ret = -EINVAL;
4516         if (args.flags != 0)
4517                 goto out_acct;
4518         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
4519                 goto out_acct;
4520         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
4521             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
4522                 goto out_acct;
4523         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
4524             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
4525                 goto out_acct;
4526         if (args.unencoded_offset > args.unencoded_len)
4527                 goto out_acct;
4528         if (args.len > args.unencoded_len - args.unencoded_offset)
4529                 goto out_acct;
4530
4531         ret = import_iovec(ITER_SOURCE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4532                            &iov, &iter);
4533         if (ret < 0)
4534                 goto out_acct;
4535
4536         file_start_write(file);
4537
4538         if (iov_iter_count(&iter) == 0) {
4539                 ret = 0;
4540                 goto out_end_write;
4541         }
4542         pos = args.offset;
4543         ret = rw_verify_area(WRITE, file, &pos, args.len);
4544         if (ret < 0)
4545                 goto out_end_write;
4546
4547         init_sync_kiocb(&kiocb, file);
4548         ret = kiocb_set_rw_flags(&kiocb, 0);
4549         if (ret)
4550                 goto out_end_write;
4551         kiocb.ki_pos = pos;
4552
4553         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
4554         if (ret > 0)
4555                 fsnotify_modify(file);
4556
4557 out_end_write:
4558         file_end_write(file);
4559         kfree(iov);
4560 out_acct:
4561         if (ret > 0)
4562                 add_wchar(current, ret);
4563         inc_syscw(current);
4564         return ret;
4565 }
4566
4567 long btrfs_ioctl(struct file *file, unsigned int
4568                 cmd, unsigned long arg)
4569 {
4570         struct inode *inode = file_inode(file);
4571         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4572         struct btrfs_root *root = BTRFS_I(inode)->root;
4573         void __user *argp = (void __user *)arg;
4574
4575         switch (cmd) {
4576         case FS_IOC_GETVERSION:
4577                 return btrfs_ioctl_getversion(inode, argp);
4578         case FS_IOC_GETFSLABEL:
4579                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4580         case FS_IOC_SETFSLABEL:
4581                 return btrfs_ioctl_set_fslabel(file, argp);
4582         case FITRIM:
4583                 return btrfs_ioctl_fitrim(fs_info, argp);
4584         case BTRFS_IOC_SNAP_CREATE:
4585                 return btrfs_ioctl_snap_create(file, argp, 0);
4586         case BTRFS_IOC_SNAP_CREATE_V2:
4587                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4588         case BTRFS_IOC_SUBVOL_CREATE:
4589                 return btrfs_ioctl_snap_create(file, argp, 1);
4590         case BTRFS_IOC_SUBVOL_CREATE_V2:
4591                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4592         case BTRFS_IOC_SNAP_DESTROY:
4593                 return btrfs_ioctl_snap_destroy(file, argp, false);
4594         case BTRFS_IOC_SNAP_DESTROY_V2:
4595                 return btrfs_ioctl_snap_destroy(file, argp, true);
4596         case BTRFS_IOC_SUBVOL_GETFLAGS:
4597                 return btrfs_ioctl_subvol_getflags(inode, argp);
4598         case BTRFS_IOC_SUBVOL_SETFLAGS:
4599                 return btrfs_ioctl_subvol_setflags(file, argp);
4600         case BTRFS_IOC_DEFAULT_SUBVOL:
4601                 return btrfs_ioctl_default_subvol(file, argp);
4602         case BTRFS_IOC_DEFRAG:
4603                 return btrfs_ioctl_defrag(file, NULL);
4604         case BTRFS_IOC_DEFRAG_RANGE:
4605                 return btrfs_ioctl_defrag(file, argp);
4606         case BTRFS_IOC_RESIZE:
4607                 return btrfs_ioctl_resize(file, argp);
4608         case BTRFS_IOC_ADD_DEV:
4609                 return btrfs_ioctl_add_dev(fs_info, argp);
4610         case BTRFS_IOC_RM_DEV:
4611                 return btrfs_ioctl_rm_dev(file, argp);
4612         case BTRFS_IOC_RM_DEV_V2:
4613                 return btrfs_ioctl_rm_dev_v2(file, argp);
4614         case BTRFS_IOC_FS_INFO:
4615                 return btrfs_ioctl_fs_info(fs_info, argp);
4616         case BTRFS_IOC_DEV_INFO:
4617                 return btrfs_ioctl_dev_info(fs_info, argp);
4618         case BTRFS_IOC_TREE_SEARCH:
4619                 return btrfs_ioctl_tree_search(inode, argp);
4620         case BTRFS_IOC_TREE_SEARCH_V2:
4621                 return btrfs_ioctl_tree_search_v2(inode, argp);
4622         case BTRFS_IOC_INO_LOOKUP:
4623                 return btrfs_ioctl_ino_lookup(root, argp);
4624         case BTRFS_IOC_INO_PATHS:
4625                 return btrfs_ioctl_ino_to_path(root, argp);
4626         case BTRFS_IOC_LOGICAL_INO:
4627                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4628         case BTRFS_IOC_LOGICAL_INO_V2:
4629                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4630         case BTRFS_IOC_SPACE_INFO:
4631                 return btrfs_ioctl_space_info(fs_info, argp);
4632         case BTRFS_IOC_SYNC: {
4633                 int ret;
4634
4635                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4636                 if (ret)
4637                         return ret;
4638                 ret = btrfs_sync_fs(inode->i_sb, 1);
4639                 /*
4640                  * The transaction thread may want to do more work,
4641                  * namely it pokes the cleaner kthread that will start
4642                  * processing uncleaned subvols.
4643                  */
4644                 wake_up_process(fs_info->transaction_kthread);
4645                 return ret;
4646         }
4647         case BTRFS_IOC_START_SYNC:
4648                 return btrfs_ioctl_start_sync(root, argp);
4649         case BTRFS_IOC_WAIT_SYNC:
4650                 return btrfs_ioctl_wait_sync(fs_info, argp);
4651         case BTRFS_IOC_SCRUB:
4652                 return btrfs_ioctl_scrub(file, argp);
4653         case BTRFS_IOC_SCRUB_CANCEL:
4654                 return btrfs_ioctl_scrub_cancel(fs_info);
4655         case BTRFS_IOC_SCRUB_PROGRESS:
4656                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4657         case BTRFS_IOC_BALANCE_V2:
4658                 return btrfs_ioctl_balance(file, argp);
4659         case BTRFS_IOC_BALANCE_CTL:
4660                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4661         case BTRFS_IOC_BALANCE_PROGRESS:
4662                 return btrfs_ioctl_balance_progress(fs_info, argp);
4663         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4664                 return btrfs_ioctl_set_received_subvol(file, argp);
4665 #ifdef CONFIG_64BIT
4666         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4667                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4668 #endif
4669         case BTRFS_IOC_SEND:
4670                 return _btrfs_ioctl_send(inode, argp, false);
4671 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4672         case BTRFS_IOC_SEND_32:
4673                 return _btrfs_ioctl_send(inode, argp, true);
4674 #endif
4675         case BTRFS_IOC_GET_DEV_STATS:
4676                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4677         case BTRFS_IOC_QUOTA_CTL:
4678                 return btrfs_ioctl_quota_ctl(file, argp);
4679         case BTRFS_IOC_QGROUP_ASSIGN:
4680                 return btrfs_ioctl_qgroup_assign(file, argp);
4681         case BTRFS_IOC_QGROUP_CREATE:
4682                 return btrfs_ioctl_qgroup_create(file, argp);
4683         case BTRFS_IOC_QGROUP_LIMIT:
4684                 return btrfs_ioctl_qgroup_limit(file, argp);
4685         case BTRFS_IOC_QUOTA_RESCAN:
4686                 return btrfs_ioctl_quota_rescan(file, argp);
4687         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4688                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4689         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4690                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4691         case BTRFS_IOC_DEV_REPLACE:
4692                 return btrfs_ioctl_dev_replace(fs_info, argp);
4693         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4694                 return btrfs_ioctl_get_supported_features(argp);
4695         case BTRFS_IOC_GET_FEATURES:
4696                 return btrfs_ioctl_get_features(fs_info, argp);
4697         case BTRFS_IOC_SET_FEATURES:
4698                 return btrfs_ioctl_set_features(file, argp);
4699         case BTRFS_IOC_GET_SUBVOL_INFO:
4700                 return btrfs_ioctl_get_subvol_info(inode, argp);
4701         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4702                 return btrfs_ioctl_get_subvol_rootref(root, argp);
4703         case BTRFS_IOC_INO_LOOKUP_USER:
4704                 return btrfs_ioctl_ino_lookup_user(file, argp);
4705         case FS_IOC_ENABLE_VERITY:
4706                 return fsverity_ioctl_enable(file, (const void __user *)argp);
4707         case FS_IOC_MEASURE_VERITY:
4708                 return fsverity_ioctl_measure(file, argp);
4709         case BTRFS_IOC_ENCODED_READ:
4710                 return btrfs_ioctl_encoded_read(file, argp, false);
4711         case BTRFS_IOC_ENCODED_WRITE:
4712                 return btrfs_ioctl_encoded_write(file, argp, false);
4713 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4714         case BTRFS_IOC_ENCODED_READ_32:
4715                 return btrfs_ioctl_encoded_read(file, argp, true);
4716         case BTRFS_IOC_ENCODED_WRITE_32:
4717                 return btrfs_ioctl_encoded_write(file, argp, true);
4718 #endif
4719         }
4720
4721         return -ENOTTY;
4722 }
4723
4724 #ifdef CONFIG_COMPAT
4725 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4726 {
4727         /*
4728          * These all access 32-bit values anyway so no further
4729          * handling is necessary.
4730          */
4731         switch (cmd) {
4732         case FS_IOC32_GETVERSION:
4733                 cmd = FS_IOC_GETVERSION;
4734                 break;
4735         }
4736
4737         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
4738 }
4739 #endif