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 = inode_to_fs_info(inode);
 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->sectorsize)
 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 = inode_to_fs_info(dir);
 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         u64 qgroup_reserved = 0;
 607
 608         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 609         if (!root_item)
 610                 return -ENOMEM;
 611
 612         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 613         if (ret)
 614                 goto out_root_item;
 615
 616         /*
 617          * Don't create subvolume whose level is not zero. Or qgroup will be
 618          * screwed up since it assumes subvolume qgroup's level to be 0.
 619          */
 620         if (btrfs_qgroup_level(objectid)) {
 621                 ret = -ENOSPC;
 622                 goto out_root_item;
 623         }
 624
 625         ret = get_anon_bdev(&anon_dev);
 626         if (ret < 0)
 627                 goto out_root_item;
 628
 629         new_inode_args.inode = btrfs_new_subvol_inode(idmap, dir);
 630         if (!new_inode_args.inode) {
 631                 ret = -ENOMEM;
 632                 goto out_anon_dev;
 633         }
 634         ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
 635         if (ret)
 636                 goto out_inode;
 637         trans_num_items += create_subvol_num_items(inherit);
 638
 639         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 640         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 641                                                trans_num_items, false);
 642         if (ret)
 643                 goto out_new_inode_args;
 644         qgroup_reserved = block_rsv.qgroup_rsv_reserved;
 645
 646         trans = btrfs_start_transaction(root, 0);
 647         if (IS_ERR(trans)) {
 648                 ret = PTR_ERR(trans);
 649                 goto out_release_rsv;
 650         }
 651         ret = btrfs_record_root_in_trans(trans, BTRFS_I(dir)->root);
 652         if (ret)
 653                 goto out;
 654         btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
 655         qgroup_reserved = 0;
 656         trans->block_rsv = &block_rsv;
 657         trans->bytes_reserved = block_rsv.size;
 658         /* Tree log can't currently deal with an inode which is a new root. */
 659         btrfs_set_log_full_commit(trans);
 660
 661         ret = btrfs_qgroup_inherit(trans, 0, objectid, root->root_key.objectid, inherit);
 662         if (ret)
 663                 goto out;
 664
 665         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 666                                       0, BTRFS_NESTING_NORMAL);
 667         if (IS_ERR(leaf)) {
 668                 ret = PTR_ERR(leaf);
 669                 goto out;
 670         }
 671
 672         btrfs_mark_buffer_dirty(trans, leaf);
 673
 674         inode_item = &root_item->inode;
 675         btrfs_set_stack_inode_generation(inode_item, 1);
 676         btrfs_set_stack_inode_size(inode_item, 3);
 677         btrfs_set_stack_inode_nlink(inode_item, 1);
 678         btrfs_set_stack_inode_nbytes(inode_item,
 679                                      fs_info->nodesize);
 680         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 681
 682         btrfs_set_root_flags(root_item, 0);
 683         btrfs_set_root_limit(root_item, 0);
 684         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 685
 686         btrfs_set_root_bytenr(root_item, leaf->start);
 687         btrfs_set_root_generation(root_item, trans->transid);
 688         btrfs_set_root_level(root_item, 0);
 689         btrfs_set_root_refs(root_item, 1);
 690         btrfs_set_root_used(root_item, leaf->len);
 691         btrfs_set_root_last_snapshot(root_item, 0);
 692
 693         btrfs_set_root_generation_v2(root_item,
 694                         btrfs_root_generation(root_item));
 695         generate_random_guid(root_item->uuid);
 696         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 697         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 698         root_item->ctime = root_item->otime;
 699         btrfs_set_root_ctransid(root_item, trans->transid);
 700         btrfs_set_root_otransid(root_item, trans->transid);
 701
 702         btrfs_tree_unlock(leaf);
 703
 704         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 705
 706         key.objectid = objectid;
 707         key.offset = 0;
 708         key.type = BTRFS_ROOT_ITEM_KEY;
 709         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 710                                 root_item);
 711         if (ret) {
 712                 /*
 713                  * Since we don't abort the transaction in this case, free the
 714                  * tree block so that we don't leak space and leave the
 715                  * filesystem in an inconsistent state (an extent item in the
 716                  * extent tree with a backreference for a root that does not
 717                  * exists).
 718                  */
 719                 btrfs_tree_lock(leaf);
 720                 btrfs_clear_buffer_dirty(trans, leaf);
 721                 btrfs_tree_unlock(leaf);
 722                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 723                 free_extent_buffer(leaf);
 724                 goto out;
 725         }
 726
 727         free_extent_buffer(leaf);
 728         leaf = NULL;
 729
 730         new_root = btrfs_get_new_fs_root(fs_info, objectid, &anon_dev);
 731         if (IS_ERR(new_root)) {
 732                 ret = PTR_ERR(new_root);
 733                 btrfs_abort_transaction(trans, ret);
 734                 goto out;
 735         }
 736         /* anon_dev is owned by new_root now. */
 737         anon_dev = 0;
 738         BTRFS_I(new_inode_args.inode)->root = new_root;
 739         /* ... and new_root is owned by new_inode_args.inode now. */
 740
 741         ret = btrfs_record_root_in_trans(trans, new_root);
 742         if (ret) {
 743                 btrfs_abort_transaction(trans, ret);
 744                 goto out;
 745         }
 746
 747         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 748                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 749         if (ret) {
 750                 btrfs_abort_transaction(trans, ret);
 751                 goto out;
 752         }
 753
 754         ret = btrfs_create_new_inode(trans, &new_inode_args);
 755         if (ret) {
 756                 btrfs_abort_transaction(trans, ret);
 757                 goto out;
 758         }
 759
 760         d_instantiate_new(dentry, new_inode_args.inode);
 761         new_inode_args.inode = NULL;
 762
 763 out:
 764         trans->block_rsv = NULL;
 765         trans->bytes_reserved = 0;
 766         btrfs_end_transaction(trans);
 767 out_release_rsv:
 768         btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
 769         if (qgroup_reserved)
 770                 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
 771 out_new_inode_args:
 772         btrfs_new_inode_args_destroy(&new_inode_args);
 773 out_inode:
 774         iput(new_inode_args.inode);
 775 out_anon_dev:
 776         if (anon_dev)
 777                 free_anon_bdev(anon_dev);
 778 out_root_item:
 779         kfree(root_item);
 780         return ret;
 781 }
 782
 783 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 784                            struct dentry *dentry, bool readonly,
 785                            struct btrfs_qgroup_inherit *inherit)
 786 {
 787         struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
 788         struct inode *inode;
 789         struct btrfs_pending_snapshot *pending_snapshot;
 790         unsigned int trans_num_items;
 791         struct btrfs_trans_handle *trans;
 792         struct btrfs_block_rsv *block_rsv;
 793         u64 qgroup_reserved = 0;
 794         int ret;
 795
 796         /* We do not support snapshotting right now. */
 797         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 798                 btrfs_warn(fs_info,
 799                            "extent tree v2 doesn't support snapshotting yet");
 800                 return -EOPNOTSUPP;
 801         }
 802
 803         if (btrfs_root_refs(&root->root_item) == 0)
 804                 return -ENOENT;
 805
 806         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 807                 return -EINVAL;
 808
 809         if (atomic_read(&root->nr_swapfiles)) {
 810                 btrfs_warn(fs_info,
 811                            "cannot snapshot subvolume with active swapfile");
 812                 return -ETXTBSY;
 813         }
 814
 815         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 816         if (!pending_snapshot)
 817                 return -ENOMEM;
 818
 819         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 820         if (ret < 0)
 821                 goto free_pending;
 822         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 823                         GFP_KERNEL);
 824         pending_snapshot->path = btrfs_alloc_path();
 825         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 826                 ret = -ENOMEM;
 827                 goto free_pending;
 828         }
 829
 830         block_rsv = &pending_snapshot->block_rsv;
 831         btrfs_init_block_rsv(block_rsv, BTRFS_BLOCK_RSV_TEMP);
 832         /*
 833          * 1 to add dir item
 834          * 1 to add dir index
 835          * 1 to update parent inode item
 836          */
 837         trans_num_items = create_subvol_num_items(inherit) + 3;
 838         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, block_rsv,
 839                                                trans_num_items, false);
 840         if (ret)
 841                 goto free_pending;
 842         qgroup_reserved = block_rsv->qgroup_rsv_reserved;
 843
 844         pending_snapshot->dentry = dentry;
 845         pending_snapshot->root = root;
 846         pending_snapshot->readonly = readonly;
 847         pending_snapshot->dir = dir;
 848         pending_snapshot->inherit = inherit;
 849
 850         trans = btrfs_start_transaction(root, 0);
 851         if (IS_ERR(trans)) {
 852                 ret = PTR_ERR(trans);
 853                 goto fail;
 854         }
 855         ret = btrfs_record_root_in_trans(trans, BTRFS_I(dir)->root);
 856         if (ret) {
 857                 btrfs_end_transaction(trans);
 858                 goto fail;
 859         }
 860         btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
 861         qgroup_reserved = 0;
 862
 863         trans->pending_snapshot = pending_snapshot;
 864
 865         ret = btrfs_commit_transaction(trans);
 866         if (ret)
 867                 goto fail;
 868
 869         ret = pending_snapshot->error;
 870         if (ret)
 871                 goto fail;
 872
 873         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 874         if (ret)
 875                 goto fail;
 876
 877         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 878         if (IS_ERR(inode)) {
 879                 ret = PTR_ERR(inode);
 880                 goto fail;
 881         }
 882
 883         d_instantiate(dentry, inode);
 884         ret = 0;
 885         pending_snapshot->anon_dev = 0;
 886 fail:
 887         /* Prevent double freeing of anon_dev */
 888         if (ret && pending_snapshot->snap)
 889                 pending_snapshot->snap->anon_dev = 0;
 890         btrfs_put_root(pending_snapshot->snap);
 891         btrfs_block_rsv_release(fs_info, block_rsv, (u64)-1, NULL);
 892         if (qgroup_reserved)
 893                 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
 894 free_pending:
 895         if (pending_snapshot->anon_dev)
 896                 free_anon_bdev(pending_snapshot->anon_dev);
 897         kfree(pending_snapshot->root_item);
 898         btrfs_free_path(pending_snapshot->path);
 899         kfree(pending_snapshot);
 900
 901         return ret;
 902 }
 903
 904 /*  copy of may_delete in fs/namei.c()
 905  *      Check whether we can remove a link victim from directory dir, check
 906  *  whether the type of victim is right.
 907  *  1. We can't do it if dir is read-only (done in permission())
 908  *  2. We should have write and exec permissions on dir
 909  *  3. We can't remove anything from append-only dir
 910  *  4. We can't do anything with immutable dir (done in permission())
 911  *  5. If the sticky bit on dir is set we should either
 912  *      a. be owner of dir, or
 913  *      b. be owner of victim, or
 914  *      c. have CAP_FOWNER capability
 915  *  6. If the victim is append-only or immutable we can't do anything with
 916  *     links pointing to it.
 917  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 918  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 919  *  9. We can't remove a root or mountpoint.
 920  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 921  *     nfs_async_unlink().
 922  */
 923
 924 static int btrfs_may_delete(struct mnt_idmap *idmap,
 925                             struct inode *dir, struct dentry *victim, int isdir)
 926 {
 927         int error;
 928
 929         if (d_really_is_negative(victim))
 930                 return -ENOENT;
 931
 932         BUG_ON(d_inode(victim->d_parent) != dir);
 933         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 934
 935         error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
 936         if (error)
 937                 return error;
 938         if (IS_APPEND(dir))
 939                 return -EPERM;
 940         if (check_sticky(idmap, dir, d_inode(victim)) ||
 941             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 942             IS_SWAPFILE(d_inode(victim)))
 943                 return -EPERM;
 944         if (isdir) {
 945                 if (!d_is_dir(victim))
 946                         return -ENOTDIR;
 947                 if (IS_ROOT(victim))
 948                         return -EBUSY;
 949         } else if (d_is_dir(victim))
 950                 return -EISDIR;
 951         if (IS_DEADDIR(dir))
 952                 return -ENOENT;
 953         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 954                 return -EBUSY;
 955         return 0;
 956 }
 957
 958 /* copy of may_create in fs/namei.c() */
 959 static inline int btrfs_may_create(struct mnt_idmap *idmap,
 960                                    struct inode *dir, struct dentry *child)
 961 {
 962         if (d_really_is_positive(child))
 963                 return -EEXIST;
 964         if (IS_DEADDIR(dir))
 965                 return -ENOENT;
 966         if (!fsuidgid_has_mapping(dir->i_sb, idmap))
 967                 return -EOVERFLOW;
 968         return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
 969 }
 970
 971 /*
 972  * Create a new subvolume below @parent.  This is largely modeled after
 973  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 974  * inside this filesystem so it's quite a bit simpler.
 975  */
 976 static noinline int btrfs_mksubvol(const struct path *parent,
 977                                    struct mnt_idmap *idmap,
 978                                    const char *name, int namelen,
 979                                    struct btrfs_root *snap_src,
 980                                    bool readonly,
 981                                    struct btrfs_qgroup_inherit *inherit)
 982 {
 983         struct inode *dir = d_inode(parent->dentry);
 984         struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
 985         struct dentry *dentry;
 986         struct fscrypt_str name_str = FSTR_INIT((char *)name, namelen);
 987         int error;
 988
 989         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 990         if (error == -EINTR)
 991                 return error;
 992
 993         dentry = lookup_one(idmap, name, parent->dentry, namelen);
 994         error = PTR_ERR(dentry);
 995         if (IS_ERR(dentry))
 996                 goto out_unlock;
 997
 998         error = btrfs_may_create(idmap, dir, dentry);
 999         if (error)
1000                 goto out_dput;
1001
1002         /*
1003          * even if this name doesn't exist, we may get hash collisions.
1004          * check for them now when we can safely fail
1005          */
1006         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
1007                                                dir->i_ino, &name_str);
1008         if (error)
1009                 goto out_dput;
1010
1011         down_read(&fs_info->subvol_sem);
1012
1013         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
1014                 goto out_up_read;
1015
1016         if (snap_src)
1017                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
1018         else
1019                 error = create_subvol(idmap, dir, dentry, inherit);
1020
1021         if (!error)
1022                 fsnotify_mkdir(dir, dentry);
1023 out_up_read:
1024         up_read(&fs_info->subvol_sem);
1025 out_dput:
1026         dput(dentry);
1027 out_unlock:
1028         btrfs_inode_unlock(BTRFS_I(dir), 0);
1029         return error;
1030 }
1031
1032 static noinline int btrfs_mksnapshot(const struct path *parent,
1033                                    struct mnt_idmap *idmap,
1034                                    const char *name, int namelen,
1035                                    struct btrfs_root *root,
1036                                    bool readonly,
1037                                    struct btrfs_qgroup_inherit *inherit)
1038 {
1039         int ret;
1040         bool snapshot_force_cow = false;
1041
1042         /*
1043          * Force new buffered writes to reserve space even when NOCOW is
1044          * possible. This is to avoid later writeback (running dealloc) to
1045          * fallback to COW mode and unexpectedly fail with ENOSPC.
1046          */
1047         btrfs_drew_read_lock(&root->snapshot_lock);
1048
1049         ret = btrfs_start_delalloc_snapshot(root, false);
1050         if (ret)
1051                 goto out;
1052
1053         /*
1054          * All previous writes have started writeback in NOCOW mode, so now
1055          * we force future writes to fallback to COW mode during snapshot
1056          * creation.
1057          */
1058         atomic_inc(&root->snapshot_force_cow);
1059         snapshot_force_cow = true;
1060
1061         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1062
1063         ret = btrfs_mksubvol(parent, idmap, name, namelen,
1064                              root, readonly, inherit);
1065 out:
1066         if (snapshot_force_cow)
1067                 atomic_dec(&root->snapshot_force_cow);
1068         btrfs_drew_read_unlock(&root->snapshot_lock);
1069         return ret;
1070 }
1071
1072 /*
1073  * Try to start exclusive operation @type or cancel it if it's running.
1074  *
1075  * Return:
1076  *   0        - normal mode, newly claimed op started
1077  *  >0        - normal mode, something else is running,
1078  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1079  * ECANCELED  - cancel mode, successful cancel
1080  * ENOTCONN   - cancel mode, operation not running anymore
1081  */
1082 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1083                         enum btrfs_exclusive_operation type, bool cancel)
1084 {
1085         if (!cancel) {
1086                 /* Start normal op */
1087                 if (!btrfs_exclop_start(fs_info, type))
1088                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1089                 /* Exclusive operation is now claimed */
1090                 return 0;
1091         }
1092
1093         /* Cancel running op */
1094         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1095                 /*
1096                  * This blocks any exclop finish from setting it to NONE, so we
1097                  * request cancellation. Either it runs and we will wait for it,
1098                  * or it has finished and no waiting will happen.
1099                  */
1100                 atomic_inc(&fs_info->reloc_cancel_req);
1101                 btrfs_exclop_start_unlock(fs_info);
1102
1103                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1104                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1105                                     TASK_INTERRUPTIBLE);
1106
1107                 return -ECANCELED;
1108         }
1109
1110         /* Something else is running or none */
1111         return -ENOTCONN;
1112 }
1113
1114 static noinline int btrfs_ioctl_resize(struct file *file,
1115                                         void __user *arg)
1116 {
1117         BTRFS_DEV_LOOKUP_ARGS(args);
1118         struct inode *inode = file_inode(file);
1119         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1120         u64 new_size;
1121         u64 old_size;
1122         u64 devid = 1;
1123         struct btrfs_root *root = BTRFS_I(inode)->root;
1124         struct btrfs_ioctl_vol_args *vol_args;
1125         struct btrfs_trans_handle *trans;
1126         struct btrfs_device *device = NULL;
1127         char *sizestr;
1128         char *retptr;
1129         char *devstr = NULL;
1130         int ret = 0;
1131         int mod = 0;
1132         bool cancel;
1133
1134         if (!capable(CAP_SYS_ADMIN))
1135                 return -EPERM;
1136
1137         ret = mnt_want_write_file(file);
1138         if (ret)
1139                 return ret;
1140
1141         /*
1142          * Read the arguments before checking exclusivity to be able to
1143          * distinguish regular resize and cancel
1144          */
1145         vol_args = memdup_user(arg, sizeof(*vol_args));
1146         if (IS_ERR(vol_args)) {
1147                 ret = PTR_ERR(vol_args);
1148                 goto out_drop;
1149         }
1150         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1151         sizestr = vol_args->name;
1152         cancel = (strcmp("cancel", sizestr) == 0);
1153         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1154         if (ret)
1155                 goto out_free;
1156         /* Exclusive operation is now claimed */
1157
1158         devstr = strchr(sizestr, ':');
1159         if (devstr) {
1160                 sizestr = devstr + 1;
1161                 *devstr = '\0';
1162                 devstr = vol_args->name;
1163                 ret = kstrtoull(devstr, 10, &devid);
1164                 if (ret)
1165                         goto out_finish;
1166                 if (!devid) {
1167                         ret = -EINVAL;
1168                         goto out_finish;
1169                 }
1170                 btrfs_info(fs_info, "resizing devid %llu", devid);
1171         }
1172
1173         args.devid = devid;
1174         device = btrfs_find_device(fs_info->fs_devices, &args);
1175         if (!device) {
1176                 btrfs_info(fs_info, "resizer unable to find device %llu",
1177                            devid);
1178                 ret = -ENODEV;
1179                 goto out_finish;
1180         }
1181
1182         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1183                 btrfs_info(fs_info,
1184                            "resizer unable to apply on readonly device %llu",
1185                        devid);
1186                 ret = -EPERM;
1187                 goto out_finish;
1188         }
1189
1190         if (!strcmp(sizestr, "max"))
1191                 new_size = bdev_nr_bytes(device->bdev);
1192         else {
1193                 if (sizestr[0] == '-') {
1194                         mod = -1;
1195                         sizestr++;
1196                 } else if (sizestr[0] == '+') {
1197                         mod = 1;
1198                         sizestr++;
1199                 }
1200                 new_size = memparse(sizestr, &retptr);
1201                 if (*retptr != '\0' || new_size == 0) {
1202                         ret = -EINVAL;
1203                         goto out_finish;
1204                 }
1205         }
1206
1207         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1208                 ret = -EPERM;
1209                 goto out_finish;
1210         }
1211
1212         old_size = btrfs_device_get_total_bytes(device);
1213
1214         if (mod < 0) {
1215                 if (new_size > old_size) {
1216                         ret = -EINVAL;
1217                         goto out_finish;
1218                 }
1219                 new_size = old_size - new_size;
1220         } else if (mod > 0) {
1221                 if (new_size > ULLONG_MAX - old_size) {
1222                         ret = -ERANGE;
1223                         goto out_finish;
1224                 }
1225                 new_size = old_size + new_size;
1226         }
1227
1228         if (new_size < SZ_256M) {
1229                 ret = -EINVAL;
1230                 goto out_finish;
1231         }
1232         if (new_size > bdev_nr_bytes(device->bdev)) {
1233                 ret = -EFBIG;
1234                 goto out_finish;
1235         }
1236
1237         new_size = round_down(new_size, fs_info->sectorsize);
1238
1239         if (new_size > old_size) {
1240                 trans = btrfs_start_transaction(root, 0);
1241                 if (IS_ERR(trans)) {
1242                         ret = PTR_ERR(trans);
1243                         goto out_finish;
1244                 }
1245                 ret = btrfs_grow_device(trans, device, new_size);
1246                 btrfs_commit_transaction(trans);
1247         } else if (new_size < old_size) {
1248                 ret = btrfs_shrink_device(device, new_size);
1249         } /* equal, nothing need to do */
1250
1251         if (ret == 0 && new_size != old_size)
1252                 btrfs_info_in_rcu(fs_info,
1253                         "resize device %s (devid %llu) from %llu to %llu",
1254                         btrfs_dev_name(device), device->devid,
1255                         old_size, new_size);
1256 out_finish:
1257         btrfs_exclop_finish(fs_info);
1258 out_free:
1259         kfree(vol_args);
1260 out_drop:
1261         mnt_drop_write_file(file);
1262         return ret;
1263 }
1264
1265 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1266                                 struct mnt_idmap *idmap,
1267                                 const char *name, unsigned long fd, int subvol,
1268                                 bool readonly,
1269                                 struct btrfs_qgroup_inherit *inherit)
1270 {
1271         int namelen;
1272         int ret = 0;
1273
1274         if (!S_ISDIR(file_inode(file)->i_mode))
1275                 return -ENOTDIR;
1276
1277         ret = mnt_want_write_file(file);
1278         if (ret)
1279                 goto out;
1280
1281         namelen = strlen(name);
1282         if (strchr(name, '/')) {
1283                 ret = -EINVAL;
1284                 goto out_drop_write;
1285         }
1286
1287         if (name[0] == '.' &&
1288            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1289                 ret = -EEXIST;
1290                 goto out_drop_write;
1291         }
1292
1293         if (subvol) {
1294                 ret = btrfs_mksubvol(&file->f_path, idmap, name,
1295                                      namelen, NULL, readonly, inherit);
1296         } else {
1297                 struct fd src = fdget(fd);
1298                 struct inode *src_inode;
1299                 if (!src.file) {
1300                         ret = -EINVAL;
1301                         goto out_drop_write;
1302                 }
1303
1304                 src_inode = file_inode(src.file);
1305                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1306                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1307                                    "Snapshot src from another FS");
1308                         ret = -EXDEV;
1309                 } else if (!inode_owner_or_capable(idmap, src_inode)) {
1310                         /*
1311                          * Subvolume creation is not restricted, but snapshots
1312                          * are limited to own subvolumes only
1313                          */
1314                         ret = -EPERM;
1315                 } else if (btrfs_ino(BTRFS_I(src_inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1316                         /*
1317                          * Snapshots must be made with the src_inode referring
1318                          * to the subvolume inode, otherwise the permission
1319                          * checking above is useless because we may have
1320                          * permission on a lower directory but not the subvol
1321                          * itself.
1322                          */
1323                         ret = -EINVAL;
1324                 } else {
1325                         ret = btrfs_mksnapshot(&file->f_path, idmap,
1326                                                name, namelen,
1327                                                BTRFS_I(src_inode)->root,
1328                                                readonly, inherit);
1329                 }
1330                 fdput(src);
1331         }
1332 out_drop_write:
1333         mnt_drop_write_file(file);
1334 out:
1335         return ret;
1336 }
1337
1338 static noinline int btrfs_ioctl_snap_create(struct file *file,
1339                                             void __user *arg, int subvol)
1340 {
1341         struct btrfs_ioctl_vol_args *vol_args;
1342         int ret;
1343
1344         if (!S_ISDIR(file_inode(file)->i_mode))
1345                 return -ENOTDIR;
1346
1347         vol_args = memdup_user(arg, sizeof(*vol_args));
1348         if (IS_ERR(vol_args))
1349                 return PTR_ERR(vol_args);
1350         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1351
1352         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1353                                         vol_args->name, vol_args->fd, subvol,
1354                                         false, NULL);
1355
1356         kfree(vol_args);
1357         return ret;
1358 }
1359
1360 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1361                                                void __user *arg, int subvol)
1362 {
1363         struct btrfs_ioctl_vol_args_v2 *vol_args;
1364         int ret;
1365         bool readonly = false;
1366         struct btrfs_qgroup_inherit *inherit = NULL;
1367
1368         if (!S_ISDIR(file_inode(file)->i_mode))
1369                 return -ENOTDIR;
1370
1371         vol_args = memdup_user(arg, sizeof(*vol_args));
1372         if (IS_ERR(vol_args))
1373                 return PTR_ERR(vol_args);
1374         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1375
1376         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1377                 ret = -EOPNOTSUPP;
1378                 goto free_args;
1379         }
1380
1381         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1382                 readonly = true;
1383         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1384                 struct btrfs_fs_info *fs_info = inode_to_fs_info(file_inode(file));
1385
1386                 if (vol_args->size < sizeof(*inherit) ||
1387                     vol_args->size > PAGE_SIZE) {
1388                         ret = -EINVAL;
1389                         goto free_args;
1390                 }
1391                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1392                 if (IS_ERR(inherit)) {
1393                         ret = PTR_ERR(inherit);
1394                         goto free_args;
1395                 }
1396
1397                 ret = btrfs_qgroup_check_inherit(fs_info, inherit, vol_args->size);
1398                 if (ret < 0)
1399                         goto free_inherit;
1400         }
1401
1402         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1403                                         vol_args->name, vol_args->fd, subvol,
1404                                         readonly, inherit);
1405         if (ret)
1406                 goto free_inherit;
1407 free_inherit:
1408         kfree(inherit);
1409 free_args:
1410         kfree(vol_args);
1411         return ret;
1412 }
1413
1414 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
1415                                                 void __user *arg)
1416 {
1417         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1418         struct btrfs_root *root = BTRFS_I(inode)->root;
1419         int ret = 0;
1420         u64 flags = 0;
1421
1422         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1423                 return -EINVAL;
1424
1425         down_read(&fs_info->subvol_sem);
1426         if (btrfs_root_readonly(root))
1427                 flags |= BTRFS_SUBVOL_RDONLY;
1428         up_read(&fs_info->subvol_sem);
1429
1430         if (copy_to_user(arg, &flags, sizeof(flags)))
1431                 ret = -EFAULT;
1432
1433         return ret;
1434 }
1435
1436 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1437                                               void __user *arg)
1438 {
1439         struct inode *inode = file_inode(file);
1440         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1441         struct btrfs_root *root = BTRFS_I(inode)->root;
1442         struct btrfs_trans_handle *trans;
1443         u64 root_flags;
1444         u64 flags;
1445         int ret = 0;
1446
1447         if (!inode_owner_or_capable(file_mnt_idmap(file), inode))
1448                 return -EPERM;
1449
1450         ret = mnt_want_write_file(file);
1451         if (ret)
1452                 goto out;
1453
1454         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1455                 ret = -EINVAL;
1456                 goto out_drop_write;
1457         }
1458
1459         if (copy_from_user(&flags, arg, sizeof(flags))) {
1460                 ret = -EFAULT;
1461                 goto out_drop_write;
1462         }
1463
1464         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1465                 ret = -EOPNOTSUPP;
1466                 goto out_drop_write;
1467         }
1468
1469         down_write(&fs_info->subvol_sem);
1470
1471         /* nothing to do */
1472         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1473                 goto out_drop_sem;
1474
1475         root_flags = btrfs_root_flags(&root->root_item);
1476         if (flags & BTRFS_SUBVOL_RDONLY) {
1477                 btrfs_set_root_flags(&root->root_item,
1478                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1479         } else {
1480                 /*
1481                  * Block RO -> RW transition if this subvolume is involved in
1482                  * send
1483                  */
1484                 spin_lock(&root->root_item_lock);
1485                 if (root->send_in_progress == 0) {
1486                         btrfs_set_root_flags(&root->root_item,
1487                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1488                         spin_unlock(&root->root_item_lock);
1489                 } else {
1490                         spin_unlock(&root->root_item_lock);
1491                         btrfs_warn(fs_info,
1492                                    "Attempt to set subvolume %llu read-write during send",
1493                                    root->root_key.objectid);
1494                         ret = -EPERM;
1495                         goto out_drop_sem;
1496                 }
1497         }
1498
1499         trans = btrfs_start_transaction(root, 1);
1500         if (IS_ERR(trans)) {
1501                 ret = PTR_ERR(trans);
1502                 goto out_reset;
1503         }
1504
1505         ret = btrfs_update_root(trans, fs_info->tree_root,
1506                                 &root->root_key, &root->root_item);
1507         if (ret < 0) {
1508                 btrfs_end_transaction(trans);
1509                 goto out_reset;
1510         }
1511
1512         ret = btrfs_commit_transaction(trans);
1513
1514 out_reset:
1515         if (ret)
1516                 btrfs_set_root_flags(&root->root_item, root_flags);
1517 out_drop_sem:
1518         up_write(&fs_info->subvol_sem);
1519 out_drop_write:
1520         mnt_drop_write_file(file);
1521 out:
1522         return ret;
1523 }
1524
1525 static noinline int key_in_sk(struct btrfs_key *key,
1526                               struct btrfs_ioctl_search_key *sk)
1527 {
1528         struct btrfs_key test;
1529         int ret;
1530
1531         test.objectid = sk->min_objectid;
1532         test.type = sk->min_type;
1533         test.offset = sk->min_offset;
1534
1535         ret = btrfs_comp_cpu_keys(key, &test);
1536         if (ret < 0)
1537                 return 0;
1538
1539         test.objectid = sk->max_objectid;
1540         test.type = sk->max_type;
1541         test.offset = sk->max_offset;
1542
1543         ret = btrfs_comp_cpu_keys(key, &test);
1544         if (ret > 0)
1545                 return 0;
1546         return 1;
1547 }
1548
1549 static noinline int copy_to_sk(struct btrfs_path *path,
1550                                struct btrfs_key *key,
1551                                struct btrfs_ioctl_search_key *sk,
1552                                u64 *buf_size,
1553                                char __user *ubuf,
1554                                unsigned long *sk_offset,
1555                                int *num_found)
1556 {
1557         u64 found_transid;
1558         struct extent_buffer *leaf;
1559         struct btrfs_ioctl_search_header sh;
1560         struct btrfs_key test;
1561         unsigned long item_off;
1562         unsigned long item_len;
1563         int nritems;
1564         int i;
1565         int slot;
1566         int ret = 0;
1567
1568         leaf = path->nodes[0];
1569         slot = path->slots[0];
1570         nritems = btrfs_header_nritems(leaf);
1571
1572         if (btrfs_header_generation(leaf) > sk->max_transid) {
1573                 i = nritems;
1574                 goto advance_key;
1575         }
1576         found_transid = btrfs_header_generation(leaf);
1577
1578         for (i = slot; i < nritems; i++) {
1579                 item_off = btrfs_item_ptr_offset(leaf, i);
1580                 item_len = btrfs_item_size(leaf, i);
1581
1582                 btrfs_item_key_to_cpu(leaf, key, i);
1583                 if (!key_in_sk(key, sk))
1584                         continue;
1585
1586                 if (sizeof(sh) + item_len > *buf_size) {
1587                         if (*num_found) {
1588                                 ret = 1;
1589                                 goto out;
1590                         }
1591
1592                         /*
1593                          * return one empty item back for v1, which does not
1594                          * handle -EOVERFLOW
1595                          */
1596
1597                         *buf_size = sizeof(sh) + item_len;
1598                         item_len = 0;
1599                         ret = -EOVERFLOW;
1600                 }
1601
1602                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1603                         ret = 1;
1604                         goto out;
1605                 }
1606
1607                 sh.objectid = key->objectid;
1608                 sh.offset = key->offset;
1609                 sh.type = key->type;
1610                 sh.len = item_len;
1611                 sh.transid = found_transid;
1612
1613                 /*
1614                  * Copy search result header. If we fault then loop again so we
1615                  * can fault in the pages and -EFAULT there if there's a
1616                  * problem. Otherwise we'll fault and then copy the buffer in
1617                  * properly this next time through
1618                  */
1619                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
1620                         ret = 0;
1621                         goto out;
1622                 }
1623
1624                 *sk_offset += sizeof(sh);
1625
1626                 if (item_len) {
1627                         char __user *up = ubuf + *sk_offset;
1628                         /*
1629                          * Copy the item, same behavior as above, but reset the
1630                          * * sk_offset so we copy the full thing again.
1631                          */
1632                         if (read_extent_buffer_to_user_nofault(leaf, up,
1633                                                 item_off, item_len)) {
1634                                 ret = 0;
1635                                 *sk_offset -= sizeof(sh);
1636                                 goto out;
1637                         }
1638
1639                         *sk_offset += item_len;
1640                 }
1641                 (*num_found)++;
1642
1643                 if (ret) /* -EOVERFLOW from above */
1644                         goto out;
1645
1646                 if (*num_found >= sk->nr_items) {
1647                         ret = 1;
1648                         goto out;
1649                 }
1650         }
1651 advance_key:
1652         ret = 0;
1653         test.objectid = sk->max_objectid;
1654         test.type = sk->max_type;
1655         test.offset = sk->max_offset;
1656         if (btrfs_comp_cpu_keys(key, &test) >= 0)
1657                 ret = 1;
1658         else if (key->offset < (u64)-1)
1659                 key->offset++;
1660         else if (key->type < (u8)-1) {
1661                 key->offset = 0;
1662                 key->type++;
1663         } else if (key->objectid < (u64)-1) {
1664                 key->offset = 0;
1665                 key->type = 0;
1666                 key->objectid++;
1667         } else
1668                 ret = 1;
1669 out:
1670         /*
1671          *  0: all items from this leaf copied, continue with next
1672          *  1: * more items can be copied, but unused buffer is too small
1673          *     * all items were found
1674          *     Either way, it will stops the loop which iterates to the next
1675          *     leaf
1676          *  -EOVERFLOW: item was to large for buffer
1677          *  -EFAULT: could not copy extent buffer back to userspace
1678          */
1679         return ret;
1680 }
1681
1682 static noinline int search_ioctl(struct inode *inode,
1683                                  struct btrfs_ioctl_search_key *sk,
1684                                  u64 *buf_size,
1685                                  char __user *ubuf)
1686 {
1687         struct btrfs_fs_info *info = inode_to_fs_info(inode);
1688         struct btrfs_root *root;
1689         struct btrfs_key key;
1690         struct btrfs_path *path;
1691         int ret;
1692         int num_found = 0;
1693         unsigned long sk_offset = 0;
1694
1695         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
1696                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
1697                 return -EOVERFLOW;
1698         }
1699
1700         path = btrfs_alloc_path();
1701         if (!path)
1702                 return -ENOMEM;
1703
1704         if (sk->tree_id == 0) {
1705                 /* search the root of the inode that was passed */
1706                 root = btrfs_grab_root(BTRFS_I(inode)->root);
1707         } else {
1708                 root = btrfs_get_fs_root(info, sk->tree_id, true);
1709                 if (IS_ERR(root)) {
1710                         btrfs_free_path(path);
1711                         return PTR_ERR(root);
1712                 }
1713         }
1714
1715         key.objectid = sk->min_objectid;
1716         key.type = sk->min_type;
1717         key.offset = sk->min_offset;
1718
1719         while (1) {
1720                 ret = -EFAULT;
1721                 /*
1722                  * Ensure that the whole user buffer is faulted in at sub-page
1723                  * granularity, otherwise the loop may live-lock.
1724                  */
1725                 if (fault_in_subpage_writeable(ubuf + sk_offset,
1726                                                *buf_size - sk_offset))
1727                         break;
1728
1729                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
1730                 if (ret != 0) {
1731                         if (ret > 0)
1732                                 ret = 0;
1733                         goto err;
1734                 }
1735                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
1736                                  &sk_offset, &num_found);
1737                 btrfs_release_path(path);
1738                 if (ret)
1739                         break;
1740
1741         }
1742         if (ret > 0)
1743                 ret = 0;
1744 err:
1745         sk->nr_items = num_found;
1746         btrfs_put_root(root);
1747         btrfs_free_path(path);
1748         return ret;
1749 }
1750
1751 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
1752                                             void __user *argp)
1753 {
1754         struct btrfs_ioctl_search_args __user *uargs = argp;
1755         struct btrfs_ioctl_search_key sk;
1756         int ret;
1757         u64 buf_size;
1758
1759         if (!capable(CAP_SYS_ADMIN))
1760                 return -EPERM;
1761
1762         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
1763                 return -EFAULT;
1764
1765         buf_size = sizeof(uargs->buf);
1766
1767         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
1768
1769         /*
1770          * In the origin implementation an overflow is handled by returning a
1771          * search header with a len of zero, so reset ret.
1772          */
1773         if (ret == -EOVERFLOW)
1774                 ret = 0;
1775
1776         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
1777                 ret = -EFAULT;
1778         return ret;
1779 }
1780
1781 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
1782                                                void __user *argp)
1783 {
1784         struct btrfs_ioctl_search_args_v2 __user *uarg = argp;
1785         struct btrfs_ioctl_search_args_v2 args;
1786         int ret;
1787         u64 buf_size;
1788         const u64 buf_limit = SZ_16M;
1789
1790         if (!capable(CAP_SYS_ADMIN))
1791                 return -EPERM;
1792
1793         /* copy search header and buffer size */
1794         if (copy_from_user(&args, uarg, sizeof(args)))
1795                 return -EFAULT;
1796
1797         buf_size = args.buf_size;
1798
1799         /* limit result size to 16MB */
1800         if (buf_size > buf_limit)
1801                 buf_size = buf_limit;
1802
1803         ret = search_ioctl(inode, &args.key, &buf_size,
1804                            (char __user *)(&uarg->buf[0]));
1805         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
1806                 ret = -EFAULT;
1807         else if (ret == -EOVERFLOW &&
1808                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
1809                 ret = -EFAULT;
1810
1811         return ret;
1812 }
1813
1814 /*
1815  * Search INODE_REFs to identify path name of 'dirid' directory
1816  * in a 'tree_id' tree. and sets path name to 'name'.
1817  */
1818 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1819                                 u64 tree_id, u64 dirid, char *name)
1820 {
1821         struct btrfs_root *root;
1822         struct btrfs_key key;
1823         char *ptr;
1824         int ret = -1;
1825         int slot;
1826         int len;
1827         int total_len = 0;
1828         struct btrfs_inode_ref *iref;
1829         struct extent_buffer *l;
1830         struct btrfs_path *path;
1831
1832         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1833                 name[0]='\0';
1834                 return 0;
1835         }
1836
1837         path = btrfs_alloc_path();
1838         if (!path)
1839                 return -ENOMEM;
1840
1841         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
1842
1843         root = btrfs_get_fs_root(info, tree_id, true);
1844         if (IS_ERR(root)) {
1845                 ret = PTR_ERR(root);
1846                 root = NULL;
1847                 goto out;
1848         }
1849
1850         key.objectid = dirid;
1851         key.type = BTRFS_INODE_REF_KEY;
1852         key.offset = (u64)-1;
1853
1854         while (1) {
1855                 ret = btrfs_search_backwards(root, &key, path);
1856                 if (ret < 0)
1857                         goto out;
1858                 else if (ret > 0) {
1859                         ret = -ENOENT;
1860                         goto out;
1861                 }
1862
1863                 l = path->nodes[0];
1864                 slot = path->slots[0];
1865
1866                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1867                 len = btrfs_inode_ref_name_len(l, iref);
1868                 ptr -= len + 1;
1869                 total_len += len + 1;
1870                 if (ptr < name) {
1871                         ret = -ENAMETOOLONG;
1872                         goto out;
1873                 }
1874
1875                 *(ptr + len) = '/';
1876                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
1877
1878                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1879                         break;
1880
1881                 btrfs_release_path(path);
1882                 key.objectid = key.offset;
1883                 key.offset = (u64)-1;
1884                 dirid = key.objectid;
1885         }
1886         memmove(name, ptr, total_len);
1887         name[total_len] = '\0';
1888         ret = 0;
1889 out:
1890         btrfs_put_root(root);
1891         btrfs_free_path(path);
1892         return ret;
1893 }
1894
1895 static int btrfs_search_path_in_tree_user(struct mnt_idmap *idmap,
1896                                 struct inode *inode,
1897                                 struct btrfs_ioctl_ino_lookup_user_args *args)
1898 {
1899         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1900         struct super_block *sb = inode->i_sb;
1901         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
1902         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
1903         u64 dirid = args->dirid;
1904         unsigned long item_off;
1905         unsigned long item_len;
1906         struct btrfs_inode_ref *iref;
1907         struct btrfs_root_ref *rref;
1908         struct btrfs_root *root = NULL;
1909         struct btrfs_path *path;
1910         struct btrfs_key key, key2;
1911         struct extent_buffer *leaf;
1912         struct inode *temp_inode;
1913         char *ptr;
1914         int slot;
1915         int len;
1916         int total_len = 0;
1917         int ret;
1918
1919         path = btrfs_alloc_path();
1920         if (!path)
1921                 return -ENOMEM;
1922
1923         /*
1924          * If the bottom subvolume does not exist directly under upper_limit,
1925          * construct the path in from the bottom up.
1926          */
1927         if (dirid != upper_limit.objectid) {
1928                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
1929
1930                 root = btrfs_get_fs_root(fs_info, treeid, true);
1931                 if (IS_ERR(root)) {
1932                         ret = PTR_ERR(root);
1933                         goto out;
1934                 }
1935
1936                 key.objectid = dirid;
1937                 key.type = BTRFS_INODE_REF_KEY;
1938                 key.offset = (u64)-1;
1939                 while (1) {
1940                         ret = btrfs_search_backwards(root, &key, path);
1941                         if (ret < 0)
1942                                 goto out_put;
1943                         else if (ret > 0) {
1944                                 ret = -ENOENT;
1945                                 goto out_put;
1946                         }
1947
1948                         leaf = path->nodes[0];
1949                         slot = path->slots[0];
1950
1951                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
1952                         len = btrfs_inode_ref_name_len(leaf, iref);
1953                         ptr -= len + 1;
1954                         total_len += len + 1;
1955                         if (ptr < args->path) {
1956                                 ret = -ENAMETOOLONG;
1957                                 goto out_put;
1958                         }
1959
1960                         *(ptr + len) = '/';
1961                         read_extent_buffer(leaf, ptr,
1962                                         (unsigned long)(iref + 1), len);
1963
1964                         /* Check the read+exec permission of this directory */
1965                         ret = btrfs_previous_item(root, path, dirid,
1966                                                   BTRFS_INODE_ITEM_KEY);
1967                         if (ret < 0) {
1968                                 goto out_put;
1969                         } else if (ret > 0) {
1970                                 ret = -ENOENT;
1971                                 goto out_put;
1972                         }
1973
1974                         leaf = path->nodes[0];
1975                         slot = path->slots[0];
1976                         btrfs_item_key_to_cpu(leaf, &key2, slot);
1977                         if (key2.objectid != dirid) {
1978                                 ret = -ENOENT;
1979                                 goto out_put;
1980                         }
1981
1982                         /*
1983                          * We don't need the path anymore, so release it and
1984                          * avoid deadlocks and lockdep warnings in case
1985                          * btrfs_iget() needs to lookup the inode from its root
1986                          * btree and lock the same leaf.
1987                          */
1988                         btrfs_release_path(path);
1989                         temp_inode = btrfs_iget(sb, key2.objectid, root);
1990                         if (IS_ERR(temp_inode)) {
1991                                 ret = PTR_ERR(temp_inode);
1992                                 goto out_put;
1993                         }
1994                         ret = inode_permission(idmap, temp_inode,
1995                                                MAY_READ | MAY_EXEC);
1996                         iput(temp_inode);
1997                         if (ret) {
1998                                 ret = -EACCES;
1999                                 goto out_put;
2000                         }
2001
2002                         if (key.offset == upper_limit.objectid)
2003                                 break;
2004                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2005                                 ret = -EACCES;
2006                                 goto out_put;
2007                         }
2008
2009                         key.objectid = key.offset;
2010                         key.offset = (u64)-1;
2011                         dirid = key.objectid;
2012                 }
2013
2014                 memmove(args->path, ptr, total_len);
2015                 args->path[total_len] = '\0';
2016                 btrfs_put_root(root);
2017                 root = NULL;
2018                 btrfs_release_path(path);
2019         }
2020
2021         /* Get the bottom subvolume's name from ROOT_REF */
2022         key.objectid = treeid;
2023         key.type = BTRFS_ROOT_REF_KEY;
2024         key.offset = args->treeid;
2025         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2026         if (ret < 0) {
2027                 goto out;
2028         } else if (ret > 0) {
2029                 ret = -ENOENT;
2030                 goto out;
2031         }
2032
2033         leaf = path->nodes[0];
2034         slot = path->slots[0];
2035         btrfs_item_key_to_cpu(leaf, &key, slot);
2036
2037         item_off = btrfs_item_ptr_offset(leaf, slot);
2038         item_len = btrfs_item_size(leaf, slot);
2039         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2040         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2041         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2042                 ret = -EINVAL;
2043                 goto out;
2044         }
2045
2046         /* Copy subvolume's name */
2047         item_off += sizeof(struct btrfs_root_ref);
2048         item_len -= sizeof(struct btrfs_root_ref);
2049         read_extent_buffer(leaf, args->name, item_off, item_len);
2050         args->name[item_len] = 0;
2051
2052 out_put:
2053         btrfs_put_root(root);
2054 out:
2055         btrfs_free_path(path);
2056         return ret;
2057 }
2058
2059 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2060                                            void __user *argp)
2061 {
2062         struct btrfs_ioctl_ino_lookup_args *args;
2063         int ret = 0;
2064
2065         args = memdup_user(argp, sizeof(*args));
2066         if (IS_ERR(args))
2067                 return PTR_ERR(args);
2068
2069         /*
2070          * Unprivileged query to obtain the containing subvolume root id. The
2071          * path is reset so it's consistent with btrfs_search_path_in_tree.
2072          */
2073         if (args->treeid == 0)
2074                 args->treeid = root->root_key.objectid;
2075
2076         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2077                 args->name[0] = 0;
2078                 goto out;
2079         }
2080
2081         if (!capable(CAP_SYS_ADMIN)) {
2082                 ret = -EPERM;
2083                 goto out;
2084         }
2085
2086         ret = btrfs_search_path_in_tree(root->fs_info,
2087                                         args->treeid, args->objectid,
2088                                         args->name);
2089
2090 out:
2091         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2092                 ret = -EFAULT;
2093
2094         kfree(args);
2095         return ret;
2096 }
2097
2098 /*
2099  * Version of ino_lookup ioctl (unprivileged)
2100  *
2101  * The main differences from ino_lookup ioctl are:
2102  *
2103  *   1. Read + Exec permission will be checked using inode_permission() during
2104  *      path construction. -EACCES will be returned in case of failure.
2105  *   2. Path construction will be stopped at the inode number which corresponds
2106  *      to the fd with which this ioctl is called. If constructed path does not
2107  *      exist under fd's inode, -EACCES will be returned.
2108  *   3. The name of bottom subvolume is also searched and filled.
2109  */
2110 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2111 {
2112         struct btrfs_ioctl_ino_lookup_user_args *args;
2113         struct inode *inode;
2114         int ret;
2115
2116         args = memdup_user(argp, sizeof(*args));
2117         if (IS_ERR(args))
2118                 return PTR_ERR(args);
2119
2120         inode = file_inode(file);
2121
2122         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2123             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2124                 /*
2125                  * The subvolume does not exist under fd with which this is
2126                  * called
2127                  */
2128                 kfree(args);
2129                 return -EACCES;
2130         }
2131
2132         ret = btrfs_search_path_in_tree_user(file_mnt_idmap(file), inode, args);
2133
2134         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2135                 ret = -EFAULT;
2136
2137         kfree(args);
2138         return ret;
2139 }
2140
2141 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2142 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
2143 {
2144         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2145         struct btrfs_fs_info *fs_info;
2146         struct btrfs_root *root;
2147         struct btrfs_path *path;
2148         struct btrfs_key key;
2149         struct btrfs_root_item *root_item;
2150         struct btrfs_root_ref *rref;
2151         struct extent_buffer *leaf;
2152         unsigned long item_off;
2153         unsigned long item_len;
2154         int slot;
2155         int ret = 0;
2156
2157         path = btrfs_alloc_path();
2158         if (!path)
2159                 return -ENOMEM;
2160
2161         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2162         if (!subvol_info) {
2163                 btrfs_free_path(path);
2164                 return -ENOMEM;
2165         }
2166
2167         fs_info = BTRFS_I(inode)->root->fs_info;
2168
2169         /* Get root_item of inode's subvolume */
2170         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2171         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2172         if (IS_ERR(root)) {
2173                 ret = PTR_ERR(root);
2174                 goto out_free;
2175         }
2176         root_item = &root->root_item;
2177
2178         subvol_info->treeid = key.objectid;
2179
2180         subvol_info->generation = btrfs_root_generation(root_item);
2181         subvol_info->flags = btrfs_root_flags(root_item);
2182
2183         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2184         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2185                                                     BTRFS_UUID_SIZE);
2186         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2187                                                     BTRFS_UUID_SIZE);
2188
2189         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2190         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2191         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2192
2193         subvol_info->otransid = btrfs_root_otransid(root_item);
2194         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2195         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2196
2197         subvol_info->stransid = btrfs_root_stransid(root_item);
2198         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2199         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2200
2201         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2202         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2203         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2204
2205         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2206                 /* Search root tree for ROOT_BACKREF of this subvolume */
2207                 key.type = BTRFS_ROOT_BACKREF_KEY;
2208                 key.offset = 0;
2209                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2210                 if (ret < 0) {
2211                         goto out;
2212                 } else if (path->slots[0] >=
2213                            btrfs_header_nritems(path->nodes[0])) {
2214                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2215                         if (ret < 0) {
2216                                 goto out;
2217                         } else if (ret > 0) {
2218                                 ret = -EUCLEAN;
2219                                 goto out;
2220                         }
2221                 }
2222
2223                 leaf = path->nodes[0];
2224                 slot = path->slots[0];
2225                 btrfs_item_key_to_cpu(leaf, &key, slot);
2226                 if (key.objectid == subvol_info->treeid &&
2227                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2228                         subvol_info->parent_id = key.offset;
2229
2230                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2231                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2232
2233                         item_off = btrfs_item_ptr_offset(leaf, slot)
2234                                         + sizeof(struct btrfs_root_ref);
2235                         item_len = btrfs_item_size(leaf, slot)
2236                                         - sizeof(struct btrfs_root_ref);
2237                         read_extent_buffer(leaf, subvol_info->name,
2238                                            item_off, item_len);
2239                 } else {
2240                         ret = -ENOENT;
2241                         goto out;
2242                 }
2243         }
2244
2245         btrfs_free_path(path);
2246         path = NULL;
2247         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2248                 ret = -EFAULT;
2249
2250 out:
2251         btrfs_put_root(root);
2252 out_free:
2253         btrfs_free_path(path);
2254         kfree(subvol_info);
2255         return ret;
2256 }
2257
2258 /*
2259  * Return ROOT_REF information of the subvolume containing this inode
2260  * except the subvolume name.
2261  */
2262 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
2263                                           void __user *argp)
2264 {
2265         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2266         struct btrfs_root_ref *rref;
2267         struct btrfs_path *path;
2268         struct btrfs_key key;
2269         struct extent_buffer *leaf;
2270         u64 objectid;
2271         int slot;
2272         int ret;
2273         u8 found;
2274
2275         path = btrfs_alloc_path();
2276         if (!path)
2277                 return -ENOMEM;
2278
2279         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2280         if (IS_ERR(rootrefs)) {
2281                 btrfs_free_path(path);
2282                 return PTR_ERR(rootrefs);
2283         }
2284
2285         objectid = root->root_key.objectid;
2286         key.objectid = objectid;
2287         key.type = BTRFS_ROOT_REF_KEY;
2288         key.offset = rootrefs->min_treeid;
2289         found = 0;
2290
2291         root = root->fs_info->tree_root;
2292         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2293         if (ret < 0) {
2294                 goto out;
2295         } else if (path->slots[0] >=
2296                    btrfs_header_nritems(path->nodes[0])) {
2297                 ret = btrfs_next_leaf(root, path);
2298                 if (ret < 0) {
2299                         goto out;
2300                 } else if (ret > 0) {
2301                         ret = -EUCLEAN;
2302                         goto out;
2303                 }
2304         }
2305         while (1) {
2306                 leaf = path->nodes[0];
2307                 slot = path->slots[0];
2308
2309                 btrfs_item_key_to_cpu(leaf, &key, slot);
2310                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2311                         ret = 0;
2312                         goto out;
2313                 }
2314
2315                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2316                         ret = -EOVERFLOW;
2317                         goto out;
2318                 }
2319
2320                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2321                 rootrefs->rootref[found].treeid = key.offset;
2322                 rootrefs->rootref[found].dirid =
2323                                   btrfs_root_ref_dirid(leaf, rref);
2324                 found++;
2325
2326                 ret = btrfs_next_item(root, path);
2327                 if (ret < 0) {
2328                         goto out;
2329                 } else if (ret > 0) {
2330                         ret = -EUCLEAN;
2331                         goto out;
2332                 }
2333         }
2334
2335 out:
2336         btrfs_free_path(path);
2337
2338         if (!ret || ret == -EOVERFLOW) {
2339                 rootrefs->num_items = found;
2340                 /* update min_treeid for next search */
2341                 if (found)
2342                         rootrefs->min_treeid =
2343                                 rootrefs->rootref[found - 1].treeid + 1;
2344                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2345                         ret = -EFAULT;
2346         }
2347
2348         kfree(rootrefs);
2349
2350         return ret;
2351 }
2352
2353 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2354                                              void __user *arg,
2355                                              bool destroy_v2)
2356 {
2357         struct dentry *parent = file->f_path.dentry;
2358         struct dentry *dentry;
2359         struct inode *dir = d_inode(parent);
2360         struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
2361         struct inode *inode;
2362         struct btrfs_root *root = BTRFS_I(dir)->root;
2363         struct btrfs_root *dest = NULL;
2364         struct btrfs_ioctl_vol_args *vol_args = NULL;
2365         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2366         struct mnt_idmap *idmap = file_mnt_idmap(file);
2367         char *subvol_name, *subvol_name_ptr = NULL;
2368         int subvol_namelen;
2369         int err = 0;
2370         bool destroy_parent = false;
2371
2372         /* We don't support snapshots with extent tree v2 yet. */
2373         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2374                 btrfs_err(fs_info,
2375                           "extent tree v2 doesn't support snapshot deletion yet");
2376                 return -EOPNOTSUPP;
2377         }
2378
2379         if (destroy_v2) {
2380                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2381                 if (IS_ERR(vol_args2))
2382                         return PTR_ERR(vol_args2);
2383
2384                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2385                         err = -EOPNOTSUPP;
2386                         goto out;
2387                 }
2388
2389                 /*
2390                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2391                  * name, same as v1 currently does.
2392                  */
2393                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2394                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2395                         subvol_name = vol_args2->name;
2396
2397                         err = mnt_want_write_file(file);
2398                         if (err)
2399                                 goto out;
2400                 } else {
2401                         struct inode *old_dir;
2402
2403                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2404                                 err = -EINVAL;
2405                                 goto out;
2406                         }
2407
2408                         err = mnt_want_write_file(file);
2409                         if (err)
2410                                 goto out;
2411
2412                         dentry = btrfs_get_dentry(fs_info->sb,
2413                                         BTRFS_FIRST_FREE_OBJECTID,
2414                                         vol_args2->subvolid, 0);
2415                         if (IS_ERR(dentry)) {
2416                                 err = PTR_ERR(dentry);
2417                                 goto out_drop_write;
2418                         }
2419
2420                         /*
2421                          * Change the default parent since the subvolume being
2422                          * deleted can be outside of the current mount point.
2423                          */
2424                         parent = btrfs_get_parent(dentry);
2425
2426                         /*
2427                          * At this point dentry->d_name can point to '/' if the
2428                          * subvolume we want to destroy is outsite of the
2429                          * current mount point, so we need to release the
2430                          * current dentry and execute the lookup to return a new
2431                          * one with ->d_name pointing to the
2432                          * <mount point>/subvol_name.
2433                          */
2434                         dput(dentry);
2435                         if (IS_ERR(parent)) {
2436                                 err = PTR_ERR(parent);
2437                                 goto out_drop_write;
2438                         }
2439                         old_dir = dir;
2440                         dir = d_inode(parent);
2441
2442                         /*
2443                          * If v2 was used with SPEC_BY_ID, a new parent was
2444                          * allocated since the subvolume can be outside of the
2445                          * current mount point. Later on we need to release this
2446                          * new parent dentry.
2447                          */
2448                         destroy_parent = true;
2449
2450                         /*
2451                          * On idmapped mounts, deletion via subvolid is
2452                          * restricted to subvolumes that are immediate
2453                          * ancestors of the inode referenced by the file
2454                          * descriptor in the ioctl. Otherwise the idmapping
2455                          * could potentially be abused to delete subvolumes
2456                          * anywhere in the filesystem the user wouldn't be able
2457                          * to delete without an idmapped mount.
2458                          */
2459                         if (old_dir != dir && idmap != &nop_mnt_idmap) {
2460                                 err = -EOPNOTSUPP;
2461                                 goto free_parent;
2462                         }
2463
2464                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2465                                                 fs_info, vol_args2->subvolid);
2466                         if (IS_ERR(subvol_name_ptr)) {
2467                                 err = PTR_ERR(subvol_name_ptr);
2468                                 goto free_parent;
2469                         }
2470                         /* subvol_name_ptr is already nul terminated */
2471                         subvol_name = (char *)kbasename(subvol_name_ptr);
2472                 }
2473         } else {
2474                 vol_args = memdup_user(arg, sizeof(*vol_args));
2475                 if (IS_ERR(vol_args))
2476                         return PTR_ERR(vol_args);
2477
2478                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2479                 subvol_name = vol_args->name;
2480
2481                 err = mnt_want_write_file(file);
2482                 if (err)
2483                         goto out;
2484         }
2485
2486         subvol_namelen = strlen(subvol_name);
2487
2488         if (strchr(subvol_name, '/') ||
2489             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2490                 err = -EINVAL;
2491                 goto free_subvol_name;
2492         }
2493
2494         if (!S_ISDIR(dir->i_mode)) {
2495                 err = -ENOTDIR;
2496                 goto free_subvol_name;
2497         }
2498
2499         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2500         if (err == -EINTR)
2501                 goto free_subvol_name;
2502         dentry = lookup_one(idmap, subvol_name, parent, subvol_namelen);
2503         if (IS_ERR(dentry)) {
2504                 err = PTR_ERR(dentry);
2505                 goto out_unlock_dir;
2506         }
2507
2508         if (d_really_is_negative(dentry)) {
2509                 err = -ENOENT;
2510                 goto out_dput;
2511         }
2512
2513         inode = d_inode(dentry);
2514         dest = BTRFS_I(inode)->root;
2515         if (!capable(CAP_SYS_ADMIN)) {
2516                 /*
2517                  * Regular user.  Only allow this with a special mount
2518                  * option, when the user has write+exec access to the
2519                  * subvol root, and when rmdir(2) would have been
2520                  * allowed.
2521                  *
2522                  * Note that this is _not_ check that the subvol is
2523                  * empty or doesn't contain data that we wouldn't
2524                  * otherwise be able to delete.
2525                  *
2526                  * Users who want to delete empty subvols should try
2527                  * rmdir(2).
2528                  */
2529                 err = -EPERM;
2530                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2531                         goto out_dput;
2532
2533                 /*
2534                  * Do not allow deletion if the parent dir is the same
2535                  * as the dir to be deleted.  That means the ioctl
2536                  * must be called on the dentry referencing the root
2537                  * of the subvol, not a random directory contained
2538                  * within it.
2539                  */
2540                 err = -EINVAL;
2541                 if (root == dest)
2542                         goto out_dput;
2543
2544                 err = inode_permission(idmap, inode, MAY_WRITE | MAY_EXEC);
2545                 if (err)
2546                         goto out_dput;
2547         }
2548
2549         /* check if subvolume may be deleted by a user */
2550         err = btrfs_may_delete(idmap, dir, dentry, 1);
2551         if (err)
2552                 goto out_dput;
2553
2554         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2555                 err = -EINVAL;
2556                 goto out_dput;
2557         }
2558
2559         btrfs_inode_lock(BTRFS_I(inode), 0);
2560         err = btrfs_delete_subvolume(BTRFS_I(dir), dentry);
2561         btrfs_inode_unlock(BTRFS_I(inode), 0);
2562         if (!err)
2563                 d_delete_notify(dir, dentry);
2564
2565 out_dput:
2566         dput(dentry);
2567 out_unlock_dir:
2568         btrfs_inode_unlock(BTRFS_I(dir), 0);
2569 free_subvol_name:
2570         kfree(subvol_name_ptr);
2571 free_parent:
2572         if (destroy_parent)
2573                 dput(parent);
2574 out_drop_write:
2575         mnt_drop_write_file(file);
2576 out:
2577         kfree(vol_args2);
2578         kfree(vol_args);
2579         return err;
2580 }
2581
2582 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2583 {
2584         struct inode *inode = file_inode(file);
2585         struct btrfs_root *root = BTRFS_I(inode)->root;
2586         struct btrfs_ioctl_defrag_range_args range = {0};
2587         int ret;
2588
2589         ret = mnt_want_write_file(file);
2590         if (ret)
2591                 return ret;
2592
2593         if (btrfs_root_readonly(root)) {
2594                 ret = -EROFS;
2595                 goto out;
2596         }
2597
2598         switch (inode->i_mode & S_IFMT) {
2599         case S_IFDIR:
2600                 if (!capable(CAP_SYS_ADMIN)) {
2601                         ret = -EPERM;
2602                         goto out;
2603                 }
2604                 ret = btrfs_defrag_root(root);
2605                 break;
2606         case S_IFREG:
2607                 /*
2608                  * Note that this does not check the file descriptor for write
2609                  * access. This prevents defragmenting executables that are
2610                  * running and allows defrag on files open in read-only mode.
2611                  */
2612                 if (!capable(CAP_SYS_ADMIN) &&
2613                     inode_permission(&nop_mnt_idmap, inode, MAY_WRITE)) {
2614                         ret = -EPERM;
2615                         goto out;
2616                 }
2617
2618                 if (argp) {
2619                         if (copy_from_user(&range, argp, sizeof(range))) {
2620                                 ret = -EFAULT;
2621                                 goto out;
2622                         }
2623                         if (range.flags & ~BTRFS_DEFRAG_RANGE_FLAGS_SUPP) {
2624                                 ret = -EOPNOTSUPP;
2625                                 goto out;
2626                         }
2627                         /* compression requires us to start the IO */
2628                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2629                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
2630                                 range.extent_thresh = (u32)-1;
2631                         }
2632                 } else {
2633                         /* the rest are all set to zero by kzalloc */
2634                         range.len = (u64)-1;
2635                 }
2636                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
2637                                         &range, BTRFS_OLDEST_GENERATION, 0);
2638                 if (ret > 0)
2639                         ret = 0;
2640                 break;
2641         default:
2642                 ret = -EINVAL;
2643         }
2644 out:
2645         mnt_drop_write_file(file);
2646         return ret;
2647 }
2648
2649 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2650 {
2651         struct btrfs_ioctl_vol_args *vol_args;
2652         bool restore_op = false;
2653         int ret;
2654
2655         if (!capable(CAP_SYS_ADMIN))
2656                 return -EPERM;
2657
2658         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2659                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
2660                 return -EINVAL;
2661         }
2662
2663         if (fs_info->fs_devices->temp_fsid) {
2664                 btrfs_err(fs_info,
2665                           "device add not supported on cloned temp-fsid mount");
2666                 return -EINVAL;
2667         }
2668
2669         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
2670                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
2671                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2672
2673                 /*
2674                  * We can do the device add because we have a paused balanced,
2675                  * change the exclusive op type and remember we should bring
2676                  * back the paused balance
2677                  */
2678                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
2679                 btrfs_exclop_start_unlock(fs_info);
2680                 restore_op = true;
2681         }
2682
2683         vol_args = memdup_user(arg, sizeof(*vol_args));
2684         if (IS_ERR(vol_args)) {
2685                 ret = PTR_ERR(vol_args);
2686                 goto out;
2687         }
2688
2689         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2690         ret = btrfs_init_new_device(fs_info, vol_args->name);
2691
2692         if (!ret)
2693                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2694
2695         kfree(vol_args);
2696 out:
2697         if (restore_op)
2698                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
2699         else
2700                 btrfs_exclop_finish(fs_info);
2701         return ret;
2702 }
2703
2704 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2705 {
2706         BTRFS_DEV_LOOKUP_ARGS(args);
2707         struct inode *inode = file_inode(file);
2708         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2709         struct btrfs_ioctl_vol_args_v2 *vol_args;
2710         struct bdev_handle *bdev_handle = NULL;
2711         int ret;
2712         bool cancel = false;
2713
2714         if (!capable(CAP_SYS_ADMIN))
2715                 return -EPERM;
2716
2717         vol_args = memdup_user(arg, sizeof(*vol_args));
2718         if (IS_ERR(vol_args))
2719                 return PTR_ERR(vol_args);
2720
2721         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
2722                 ret = -EOPNOTSUPP;
2723                 goto out;
2724         }
2725
2726         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2727         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2728                 args.devid = vol_args->devid;
2729         } else if (!strcmp("cancel", vol_args->name)) {
2730                 cancel = true;
2731         } else {
2732                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2733                 if (ret)
2734                         goto out;
2735         }
2736
2737         ret = mnt_want_write_file(file);
2738         if (ret)
2739                 goto out;
2740
2741         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2742                                            cancel);
2743         if (ret)
2744                 goto err_drop;
2745
2746         /* Exclusive operation is now claimed */
2747         ret = btrfs_rm_device(fs_info, &args, &bdev_handle);
2748
2749         btrfs_exclop_finish(fs_info);
2750
2751         if (!ret) {
2752                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2753                         btrfs_info(fs_info, "device deleted: id %llu",
2754                                         vol_args->devid);
2755                 else
2756                         btrfs_info(fs_info, "device deleted: %s",
2757                                         vol_args->name);
2758         }
2759 err_drop:
2760         mnt_drop_write_file(file);
2761         if (bdev_handle)
2762                 bdev_release(bdev_handle);
2763 out:
2764         btrfs_put_dev_args_from_path(&args);
2765         kfree(vol_args);
2766         return ret;
2767 }
2768
2769 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2770 {
2771         BTRFS_DEV_LOOKUP_ARGS(args);
2772         struct inode *inode = file_inode(file);
2773         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2774         struct btrfs_ioctl_vol_args *vol_args;
2775         struct bdev_handle *bdev_handle = NULL;
2776         int ret;
2777         bool cancel = false;
2778
2779         if (!capable(CAP_SYS_ADMIN))
2780                 return -EPERM;
2781
2782         vol_args = memdup_user(arg, sizeof(*vol_args));
2783         if (IS_ERR(vol_args))
2784                 return PTR_ERR(vol_args);
2785
2786         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2787         if (!strcmp("cancel", vol_args->name)) {
2788                 cancel = true;
2789         } else {
2790                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2791                 if (ret)
2792                         goto out;
2793         }
2794
2795         ret = mnt_want_write_file(file);
2796         if (ret)
2797                 goto out;
2798
2799         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2800                                            cancel);
2801         if (ret == 0) {
2802                 ret = btrfs_rm_device(fs_info, &args, &bdev_handle);
2803                 if (!ret)
2804                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2805                 btrfs_exclop_finish(fs_info);
2806         }
2807
2808         mnt_drop_write_file(file);
2809         if (bdev_handle)
2810                 bdev_release(bdev_handle);
2811 out:
2812         btrfs_put_dev_args_from_path(&args);
2813         kfree(vol_args);
2814         return ret;
2815 }
2816
2817 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2818                                 void __user *arg)
2819 {
2820         struct btrfs_ioctl_fs_info_args *fi_args;
2821         struct btrfs_device *device;
2822         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2823         u64 flags_in;
2824         int ret = 0;
2825
2826         fi_args = memdup_user(arg, sizeof(*fi_args));
2827         if (IS_ERR(fi_args))
2828                 return PTR_ERR(fi_args);
2829
2830         flags_in = fi_args->flags;
2831         memset(fi_args, 0, sizeof(*fi_args));
2832
2833         rcu_read_lock();
2834         fi_args->num_devices = fs_devices->num_devices;
2835
2836         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2837                 if (device->devid > fi_args->max_id)
2838                         fi_args->max_id = device->devid;
2839         }
2840         rcu_read_unlock();
2841
2842         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
2843         fi_args->nodesize = fs_info->nodesize;
2844         fi_args->sectorsize = fs_info->sectorsize;
2845         fi_args->clone_alignment = fs_info->sectorsize;
2846
2847         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
2848                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
2849                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
2850                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
2851         }
2852
2853         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
2854                 fi_args->generation = btrfs_get_fs_generation(fs_info);
2855                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
2856         }
2857
2858         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
2859                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
2860                        sizeof(fi_args->metadata_uuid));
2861                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
2862         }
2863
2864         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2865                 ret = -EFAULT;
2866
2867         kfree(fi_args);
2868         return ret;
2869 }
2870
2871 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2872                                  void __user *arg)
2873 {
2874         BTRFS_DEV_LOOKUP_ARGS(args);
2875         struct btrfs_ioctl_dev_info_args *di_args;
2876         struct btrfs_device *dev;
2877         int ret = 0;
2878
2879         di_args = memdup_user(arg, sizeof(*di_args));
2880         if (IS_ERR(di_args))
2881                 return PTR_ERR(di_args);
2882
2883         args.devid = di_args->devid;
2884         if (!btrfs_is_empty_uuid(di_args->uuid))
2885                 args.uuid = di_args->uuid;
2886
2887         rcu_read_lock();
2888         dev = btrfs_find_device(fs_info->fs_devices, &args);
2889         if (!dev) {
2890                 ret = -ENODEV;
2891                 goto out;
2892         }
2893
2894         di_args->devid = dev->devid;
2895         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2896         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2897         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2898         memcpy(di_args->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2899         if (dev->name)
2900                 strscpy(di_args->path, btrfs_dev_name(dev), sizeof(di_args->path));
2901         else
2902                 di_args->path[0] = '\0';
2903
2904 out:
2905         rcu_read_unlock();
2906         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2907                 ret = -EFAULT;
2908
2909         kfree(di_args);
2910         return ret;
2911 }
2912
2913 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2914 {
2915         struct inode *inode = file_inode(file);
2916         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2917         struct btrfs_root *root = BTRFS_I(inode)->root;
2918         struct btrfs_root *new_root;
2919         struct btrfs_dir_item *di;
2920         struct btrfs_trans_handle *trans;
2921         struct btrfs_path *path = NULL;
2922         struct btrfs_disk_key disk_key;
2923         struct fscrypt_str name = FSTR_INIT("default", 7);
2924         u64 objectid = 0;
2925         u64 dir_id;
2926         int ret;
2927
2928         if (!capable(CAP_SYS_ADMIN))
2929                 return -EPERM;
2930
2931         ret = mnt_want_write_file(file);
2932         if (ret)
2933                 return ret;
2934
2935         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
2936                 ret = -EFAULT;
2937                 goto out;
2938         }
2939
2940         if (!objectid)
2941                 objectid = BTRFS_FS_TREE_OBJECTID;
2942
2943         new_root = btrfs_get_fs_root(fs_info, objectid, true);
2944         if (IS_ERR(new_root)) {
2945                 ret = PTR_ERR(new_root);
2946                 goto out;
2947         }
2948         if (!is_fstree(new_root->root_key.objectid)) {
2949                 ret = -ENOENT;
2950                 goto out_free;
2951         }
2952
2953         path = btrfs_alloc_path();
2954         if (!path) {
2955                 ret = -ENOMEM;
2956                 goto out_free;
2957         }
2958
2959         trans = btrfs_start_transaction(root, 1);
2960         if (IS_ERR(trans)) {
2961                 ret = PTR_ERR(trans);
2962                 goto out_free;
2963         }
2964
2965         dir_id = btrfs_super_root_dir(fs_info->super_copy);
2966         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
2967                                    dir_id, &name, 1);
2968         if (IS_ERR_OR_NULL(di)) {
2969                 btrfs_release_path(path);
2970                 btrfs_end_transaction(trans);
2971                 btrfs_err(fs_info,
2972                           "Umm, you don't have the default diritem, this isn't going to work");
2973                 ret = -ENOENT;
2974                 goto out_free;
2975         }
2976
2977         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2978         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2979         btrfs_mark_buffer_dirty(trans, path->nodes[0]);
2980         btrfs_release_path(path);
2981
2982         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
2983         btrfs_end_transaction(trans);
2984 out_free:
2985         btrfs_put_root(new_root);
2986         btrfs_free_path(path);
2987 out:
2988         mnt_drop_write_file(file);
2989         return ret;
2990 }
2991
2992 static void get_block_group_info(struct list_head *groups_list,
2993                                  struct btrfs_ioctl_space_info *space)
2994 {
2995         struct btrfs_block_group *block_group;
2996
2997         space->total_bytes = 0;
2998         space->used_bytes = 0;
2999         space->flags = 0;
3000         list_for_each_entry(block_group, groups_list, list) {
3001                 space->flags = block_group->flags;
3002                 space->total_bytes += block_group->length;
3003                 space->used_bytes += block_group->used;
3004         }
3005 }
3006
3007 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3008                                    void __user *arg)
3009 {
3010         struct btrfs_ioctl_space_args space_args = { 0 };
3011         struct btrfs_ioctl_space_info space;
3012         struct btrfs_ioctl_space_info *dest;
3013         struct btrfs_ioctl_space_info *dest_orig;
3014         struct btrfs_ioctl_space_info __user *user_dest;
3015         struct btrfs_space_info *info;
3016         static const u64 types[] = {
3017                 BTRFS_BLOCK_GROUP_DATA,
3018                 BTRFS_BLOCK_GROUP_SYSTEM,
3019                 BTRFS_BLOCK_GROUP_METADATA,
3020                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3021         };
3022         int num_types = 4;
3023         int alloc_size;
3024         int ret = 0;
3025         u64 slot_count = 0;
3026         int i, c;
3027
3028         if (copy_from_user(&space_args,
3029                            (struct btrfs_ioctl_space_args __user *)arg,
3030                            sizeof(space_args)))
3031                 return -EFAULT;
3032
3033         for (i = 0; i < num_types; i++) {
3034                 struct btrfs_space_info *tmp;
3035
3036                 info = NULL;
3037                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3038                         if (tmp->flags == types[i]) {
3039                                 info = tmp;
3040                                 break;
3041                         }
3042                 }
3043
3044                 if (!info)
3045                         continue;
3046
3047                 down_read(&info->groups_sem);
3048                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3049                         if (!list_empty(&info->block_groups[c]))
3050                                 slot_count++;
3051                 }
3052                 up_read(&info->groups_sem);
3053         }
3054
3055         /*
3056          * Global block reserve, exported as a space_info
3057          */
3058         slot_count++;
3059
3060         /* space_slots == 0 means they are asking for a count */
3061         if (space_args.space_slots == 0) {
3062                 space_args.total_spaces = slot_count;
3063                 goto out;
3064         }
3065
3066         slot_count = min_t(u64, space_args.space_slots, slot_count);
3067
3068         alloc_size = sizeof(*dest) * slot_count;
3069
3070         /* we generally have at most 6 or so space infos, one for each raid
3071          * level.  So, a whole page should be more than enough for everyone
3072          */
3073         if (alloc_size > PAGE_SIZE)
3074                 return -ENOMEM;
3075
3076         space_args.total_spaces = 0;
3077         dest = kmalloc(alloc_size, GFP_KERNEL);
3078         if (!dest)
3079                 return -ENOMEM;
3080         dest_orig = dest;
3081
3082         /* now we have a buffer to copy into */
3083         for (i = 0; i < num_types; i++) {
3084                 struct btrfs_space_info *tmp;
3085
3086                 if (!slot_count)
3087                         break;
3088
3089                 info = NULL;
3090                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3091                         if (tmp->flags == types[i]) {
3092                                 info = tmp;
3093                                 break;
3094                         }
3095                 }
3096
3097                 if (!info)
3098                         continue;
3099                 down_read(&info->groups_sem);
3100                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3101                         if (!list_empty(&info->block_groups[c])) {
3102                                 get_block_group_info(&info->block_groups[c],
3103                                                      &space);
3104                                 memcpy(dest, &space, sizeof(space));
3105                                 dest++;
3106                                 space_args.total_spaces++;
3107                                 slot_count--;
3108                         }
3109                         if (!slot_count)
3110                                 break;
3111                 }
3112                 up_read(&info->groups_sem);
3113         }
3114
3115         /*
3116          * Add global block reserve
3117          */
3118         if (slot_count) {
3119                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3120
3121                 spin_lock(&block_rsv->lock);
3122                 space.total_bytes = block_rsv->size;
3123                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3124                 spin_unlock(&block_rsv->lock);
3125                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3126                 memcpy(dest, &space, sizeof(space));
3127                 space_args.total_spaces++;
3128         }
3129
3130         user_dest = (struct btrfs_ioctl_space_info __user *)
3131                 (arg + sizeof(struct btrfs_ioctl_space_args));
3132
3133         if (copy_to_user(user_dest, dest_orig, alloc_size))
3134                 ret = -EFAULT;
3135
3136         kfree(dest_orig);
3137 out:
3138         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3139                 ret = -EFAULT;
3140
3141         return ret;
3142 }
3143
3144 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3145                                             void __user *argp)
3146 {
3147         struct btrfs_trans_handle *trans;
3148         u64 transid;
3149
3150         /*
3151          * Start orphan cleanup here for the given root in case it hasn't been
3152          * started already by other means. Errors are handled in the other
3153          * functions during transaction commit.
3154          */
3155         btrfs_orphan_cleanup(root);
3156
3157         trans = btrfs_attach_transaction_barrier(root);
3158         if (IS_ERR(trans)) {
3159                 if (PTR_ERR(trans) != -ENOENT)
3160                         return PTR_ERR(trans);
3161
3162                 /* No running transaction, don't bother */
3163                 transid = btrfs_get_last_trans_committed(root->fs_info);
3164                 goto out;
3165         }
3166         transid = trans->transid;
3167         btrfs_commit_transaction_async(trans);
3168 out:
3169         if (argp)
3170                 if (copy_to_user(argp, &transid, sizeof(transid)))
3171                         return -EFAULT;
3172         return 0;
3173 }
3174
3175 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3176                                            void __user *argp)
3177 {
3178         /* By default wait for the current transaction. */
3179         u64 transid = 0;
3180
3181         if (argp)
3182                 if (copy_from_user(&transid, argp, sizeof(transid)))
3183                         return -EFAULT;
3184
3185         return btrfs_wait_for_commit(fs_info, transid);
3186 }
3187
3188 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3189 {
3190         struct btrfs_fs_info *fs_info = inode_to_fs_info(file_inode(file));
3191         struct btrfs_ioctl_scrub_args *sa;
3192         int ret;
3193
3194         if (!capable(CAP_SYS_ADMIN))
3195                 return -EPERM;
3196
3197         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3198                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
3199                 return -EINVAL;
3200         }
3201
3202         sa = memdup_user(arg, sizeof(*sa));
3203         if (IS_ERR(sa))
3204                 return PTR_ERR(sa);
3205
3206         if (sa->flags & ~BTRFS_SCRUB_SUPPORTED_FLAGS) {
3207                 ret = -EOPNOTSUPP;
3208                 goto out;
3209         }
3210
3211         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3212                 ret = mnt_want_write_file(file);
3213                 if (ret)
3214                         goto out;
3215         }
3216
3217         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3218                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3219                               0);
3220
3221         /*
3222          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3223          * error. This is important as it allows user space to know how much
3224          * progress scrub has done. For example, if scrub is canceled we get
3225          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3226          * space. Later user space can inspect the progress from the structure
3227          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3228          * previously (btrfs-progs does this).
3229          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3230          * then return -EFAULT to signal the structure was not copied or it may
3231          * be corrupt and unreliable due to a partial copy.
3232          */
3233         if (copy_to_user(arg, sa, sizeof(*sa)))
3234                 ret = -EFAULT;
3235
3236         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3237                 mnt_drop_write_file(file);
3238 out:
3239         kfree(sa);
3240         return ret;
3241 }
3242
3243 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3244 {
3245         if (!capable(CAP_SYS_ADMIN))
3246                 return -EPERM;
3247
3248         return btrfs_scrub_cancel(fs_info);
3249 }
3250
3251 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3252                                        void __user *arg)
3253 {
3254         struct btrfs_ioctl_scrub_args *sa;
3255         int ret;
3256
3257         if (!capable(CAP_SYS_ADMIN))
3258                 return -EPERM;
3259
3260         sa = memdup_user(arg, sizeof(*sa));
3261         if (IS_ERR(sa))
3262                 return PTR_ERR(sa);
3263
3264         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3265
3266         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3267                 ret = -EFAULT;
3268
3269         kfree(sa);
3270         return ret;
3271 }
3272
3273 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3274                                       void __user *arg)
3275 {
3276         struct btrfs_ioctl_get_dev_stats *sa;
3277         int ret;
3278
3279         sa = memdup_user(arg, sizeof(*sa));
3280         if (IS_ERR(sa))
3281                 return PTR_ERR(sa);
3282
3283         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3284                 kfree(sa);
3285                 return -EPERM;
3286         }
3287
3288         ret = btrfs_get_dev_stats(fs_info, sa);
3289
3290         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3291                 ret = -EFAULT;
3292
3293         kfree(sa);
3294         return ret;
3295 }
3296
3297 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3298                                     void __user *arg)
3299 {
3300         struct btrfs_ioctl_dev_replace_args *p;
3301         int ret;
3302
3303         if (!capable(CAP_SYS_ADMIN))
3304                 return -EPERM;
3305
3306         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3307                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
3308                 return -EINVAL;
3309         }
3310
3311         p = memdup_user(arg, sizeof(*p));
3312         if (IS_ERR(p))
3313                 return PTR_ERR(p);
3314
3315         switch (p->cmd) {
3316         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3317                 if (sb_rdonly(fs_info->sb)) {
3318                         ret = -EROFS;
3319                         goto out;
3320                 }
3321                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3322                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3323                 } else {
3324                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3325                         btrfs_exclop_finish(fs_info);
3326                 }
3327                 break;
3328         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3329                 btrfs_dev_replace_status(fs_info, p);
3330                 ret = 0;
3331                 break;
3332         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3333                 p->result = btrfs_dev_replace_cancel(fs_info);
3334                 ret = 0;
3335                 break;
3336         default:
3337                 ret = -EINVAL;
3338                 break;
3339         }
3340
3341         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3342                 ret = -EFAULT;
3343 out:
3344         kfree(p);
3345         return ret;
3346 }
3347
3348 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3349 {
3350         int ret = 0;
3351         int i;
3352         u64 rel_ptr;
3353         int size;
3354         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3355         struct inode_fs_paths *ipath = NULL;
3356         struct btrfs_path *path;
3357
3358         if (!capable(CAP_DAC_READ_SEARCH))
3359                 return -EPERM;
3360
3361         path = btrfs_alloc_path();
3362         if (!path) {
3363                 ret = -ENOMEM;
3364                 goto out;
3365         }
3366
3367         ipa = memdup_user(arg, sizeof(*ipa));
3368         if (IS_ERR(ipa)) {
3369                 ret = PTR_ERR(ipa);
3370                 ipa = NULL;
3371                 goto out;
3372         }
3373
3374         size = min_t(u32, ipa->size, 4096);
3375         ipath = init_ipath(size, root, path);
3376         if (IS_ERR(ipath)) {
3377                 ret = PTR_ERR(ipath);
3378                 ipath = NULL;
3379                 goto out;
3380         }
3381
3382         ret = paths_from_inode(ipa->inum, ipath);
3383         if (ret < 0)
3384                 goto out;
3385
3386         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3387                 rel_ptr = ipath->fspath->val[i] -
3388                           (u64)(unsigned long)ipath->fspath->val;
3389                 ipath->fspath->val[i] = rel_ptr;
3390         }
3391
3392         btrfs_free_path(path);
3393         path = NULL;
3394         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3395                            ipath->fspath, size);
3396         if (ret) {
3397                 ret = -EFAULT;
3398                 goto out;
3399         }
3400
3401 out:
3402         btrfs_free_path(path);
3403         free_ipath(ipath);
3404         kfree(ipa);
3405
3406         return ret;
3407 }
3408
3409 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3410                                         void __user *arg, int version)
3411 {
3412         int ret = 0;
3413         int size;
3414         struct btrfs_ioctl_logical_ino_args *loi;
3415         struct btrfs_data_container *inodes = NULL;
3416         struct btrfs_path *path = NULL;
3417         bool ignore_offset;
3418
3419         if (!capable(CAP_SYS_ADMIN))
3420                 return -EPERM;
3421
3422         loi = memdup_user(arg, sizeof(*loi));
3423         if (IS_ERR(loi))
3424                 return PTR_ERR(loi);
3425
3426         if (version == 1) {
3427                 ignore_offset = false;
3428                 size = min_t(u32, loi->size, SZ_64K);
3429         } else {
3430                 /* All reserved bits must be 0 for now */
3431                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3432                         ret = -EINVAL;
3433                         goto out_loi;
3434                 }
3435                 /* Only accept flags we have defined so far */
3436                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3437                         ret = -EINVAL;
3438                         goto out_loi;
3439                 }
3440                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3441                 size = min_t(u32, loi->size, SZ_16M);
3442         }
3443
3444         inodes = init_data_container(size);
3445         if (IS_ERR(inodes)) {
3446                 ret = PTR_ERR(inodes);
3447                 goto out_loi;
3448         }
3449
3450         path = btrfs_alloc_path();
3451         if (!path) {
3452                 ret = -ENOMEM;
3453                 goto out;
3454         }
3455         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3456                                           inodes, ignore_offset);
3457         btrfs_free_path(path);
3458         if (ret == -EINVAL)
3459                 ret = -ENOENT;
3460         if (ret < 0)
3461                 goto out;
3462
3463         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3464                            size);
3465         if (ret)
3466                 ret = -EFAULT;
3467
3468 out:
3469         kvfree(inodes);
3470 out_loi:
3471         kfree(loi);
3472
3473         return ret;
3474 }
3475
3476 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3477                                struct btrfs_ioctl_balance_args *bargs)
3478 {
3479         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3480
3481         bargs->flags = bctl->flags;
3482
3483         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3484                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3485         if (atomic_read(&fs_info->balance_pause_req))
3486                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3487         if (atomic_read(&fs_info->balance_cancel_req))
3488                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3489
3490         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3491         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3492         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3493
3494         spin_lock(&fs_info->balance_lock);
3495         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3496         spin_unlock(&fs_info->balance_lock);
3497 }
3498
3499 /*
3500  * Try to acquire fs_info::balance_mutex as well as set BTRFS_EXLCOP_BALANCE as
3501  * required.
3502  *
3503  * @fs_info:       the filesystem
3504  * @excl_acquired: ptr to boolean value which is set to false in case balance
3505  *                 is being resumed
3506  *
3507  * Return 0 on success in which case both fs_info::balance is acquired as well
3508  * as exclusive ops are blocked. In case of failure return an error code.
3509  */
3510 static int btrfs_try_lock_balance(struct btrfs_fs_info *fs_info, bool *excl_acquired)
3511 {
3512         int ret;
3513
3514         /*
3515          * Exclusive operation is locked. Three possibilities:
3516          *   (1) some other op is running
3517          *   (2) balance is running
3518          *   (3) balance is paused -- special case (think resume)
3519          */
3520         while (1) {
3521                 if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
3522                         *excl_acquired = true;
3523                         mutex_lock(&fs_info->balance_mutex);
3524                         return 0;
3525                 }
3526
3527                 mutex_lock(&fs_info->balance_mutex);
3528                 if (fs_info->balance_ctl) {
3529                         /* This is either (2) or (3) */
3530                         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3531                                 /* This is (2) */
3532                                 ret = -EINPROGRESS;
3533                                 goto out_failure;
3534
3535                         } else {
3536                                 mutex_unlock(&fs_info->balance_mutex);
3537                                 /*
3538                                  * Lock released to allow other waiters to
3539                                  * continue, we'll reexamine the status again.
3540                                  */
3541                                 mutex_lock(&fs_info->balance_mutex);
3542
3543                                 if (fs_info->balance_ctl &&
3544                                     !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3545                                         /* This is (3) */
3546                                         *excl_acquired = false;
3547                                         return 0;
3548                                 }
3549                         }
3550                 } else {
3551                         /* This is (1) */
3552                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3553                         goto out_failure;
3554                 }
3555
3556                 mutex_unlock(&fs_info->balance_mutex);
3557         }
3558
3559 out_failure:
3560         mutex_unlock(&fs_info->balance_mutex);
3561         *excl_acquired = false;
3562         return ret;
3563 }
3564
3565 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3566 {
3567         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3568         struct btrfs_fs_info *fs_info = root->fs_info;
3569         struct btrfs_ioctl_balance_args *bargs;
3570         struct btrfs_balance_control *bctl;
3571         bool need_unlock = true;
3572         int ret;
3573
3574         if (!capable(CAP_SYS_ADMIN))
3575                 return -EPERM;
3576
3577         ret = mnt_want_write_file(file);
3578         if (ret)
3579                 return ret;
3580
3581         bargs = memdup_user(arg, sizeof(*bargs));
3582         if (IS_ERR(bargs)) {
3583                 ret = PTR_ERR(bargs);
3584                 bargs = NULL;
3585                 goto out;
3586         }
3587
3588         ret = btrfs_try_lock_balance(fs_info, &need_unlock);
3589         if (ret)
3590                 goto out;
3591
3592         lockdep_assert_held(&fs_info->balance_mutex);
3593
3594         if (bargs->flags & BTRFS_BALANCE_RESUME) {
3595                 if (!fs_info->balance_ctl) {
3596                         ret = -ENOTCONN;
3597                         goto out_unlock;
3598                 }
3599
3600                 bctl = fs_info->balance_ctl;
3601                 spin_lock(&fs_info->balance_lock);
3602                 bctl->flags |= BTRFS_BALANCE_RESUME;
3603                 spin_unlock(&fs_info->balance_lock);
3604                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
3605
3606                 goto do_balance;
3607         }
3608
3609         if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
3610                 ret = -EINVAL;
3611                 goto out_unlock;
3612         }
3613
3614         if (fs_info->balance_ctl) {
3615                 ret = -EINPROGRESS;
3616                 goto out_unlock;
3617         }
3618
3619         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
3620         if (!bctl) {
3621                 ret = -ENOMEM;
3622                 goto out_unlock;
3623         }
3624
3625         memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3626         memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3627         memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3628
3629         bctl->flags = bargs->flags;
3630 do_balance:
3631         /*
3632          * Ownership of bctl and exclusive operation goes to btrfs_balance.
3633          * bctl is freed in reset_balance_state, or, if restriper was paused
3634          * all the way until unmount, in free_fs_info.  The flag should be
3635          * cleared after reset_balance_state.
3636          */
3637         need_unlock = false;
3638
3639         ret = btrfs_balance(fs_info, bctl, bargs);
3640         bctl = NULL;
3641
3642         if (ret == 0 || ret == -ECANCELED) {
3643                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3644                         ret = -EFAULT;
3645         }
3646
3647         kfree(bctl);
3648 out_unlock:
3649         mutex_unlock(&fs_info->balance_mutex);
3650         if (need_unlock)
3651                 btrfs_exclop_finish(fs_info);
3652 out:
3653         mnt_drop_write_file(file);
3654         kfree(bargs);
3655         return ret;
3656 }
3657
3658 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
3659 {
3660         if (!capable(CAP_SYS_ADMIN))
3661                 return -EPERM;
3662
3663         switch (cmd) {
3664         case BTRFS_BALANCE_CTL_PAUSE:
3665                 return btrfs_pause_balance(fs_info);
3666         case BTRFS_BALANCE_CTL_CANCEL:
3667                 return btrfs_cancel_balance(fs_info);
3668         }
3669
3670         return -EINVAL;
3671 }
3672
3673 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
3674                                          void __user *arg)
3675 {
3676         struct btrfs_ioctl_balance_args *bargs;
3677         int ret = 0;
3678
3679         if (!capable(CAP_SYS_ADMIN))
3680                 return -EPERM;
3681
3682         mutex_lock(&fs_info->balance_mutex);
3683         if (!fs_info->balance_ctl) {
3684                 ret = -ENOTCONN;
3685                 goto out;
3686         }
3687
3688         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
3689         if (!bargs) {
3690                 ret = -ENOMEM;
3691                 goto out;
3692         }
3693
3694         btrfs_update_ioctl_balance_args(fs_info, bargs);
3695
3696         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3697                 ret = -EFAULT;
3698
3699         kfree(bargs);
3700 out:
3701         mutex_unlock(&fs_info->balance_mutex);
3702         return ret;
3703 }
3704
3705 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
3706 {
3707         struct inode *inode = file_inode(file);
3708         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3709         struct btrfs_ioctl_quota_ctl_args *sa;
3710         int ret;
3711
3712         if (!capable(CAP_SYS_ADMIN))
3713                 return -EPERM;
3714
3715         ret = mnt_want_write_file(file);
3716         if (ret)
3717                 return ret;
3718
3719         sa = memdup_user(arg, sizeof(*sa));
3720         if (IS_ERR(sa)) {
3721                 ret = PTR_ERR(sa);
3722                 goto drop_write;
3723         }
3724
3725         down_write(&fs_info->subvol_sem);
3726
3727         switch (sa->cmd) {
3728         case BTRFS_QUOTA_CTL_ENABLE:
3729         case BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA:
3730                 ret = btrfs_quota_enable(fs_info, sa);
3731                 break;
3732         case BTRFS_QUOTA_CTL_DISABLE:
3733                 ret = btrfs_quota_disable(fs_info);
3734                 break;
3735         default:
3736                 ret = -EINVAL;
3737                 break;
3738         }
3739
3740         kfree(sa);
3741         up_write(&fs_info->subvol_sem);
3742 drop_write:
3743         mnt_drop_write_file(file);
3744         return ret;
3745 }
3746
3747 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
3748 {
3749         struct inode *inode = file_inode(file);
3750         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3751         struct btrfs_root *root = BTRFS_I(inode)->root;
3752         struct btrfs_ioctl_qgroup_assign_args *sa;
3753         struct btrfs_trans_handle *trans;
3754         int ret;
3755         int err;
3756
3757         if (!capable(CAP_SYS_ADMIN))
3758                 return -EPERM;
3759
3760         ret = mnt_want_write_file(file);
3761         if (ret)
3762                 return ret;
3763
3764         sa = memdup_user(arg, sizeof(*sa));
3765         if (IS_ERR(sa)) {
3766                 ret = PTR_ERR(sa);
3767                 goto drop_write;
3768         }
3769
3770         trans = btrfs_join_transaction(root);
3771         if (IS_ERR(trans)) {
3772                 ret = PTR_ERR(trans);
3773                 goto out;
3774         }
3775
3776         if (sa->assign) {
3777                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
3778         } else {
3779                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
3780         }
3781
3782         /* update qgroup status and info */
3783         mutex_lock(&fs_info->qgroup_ioctl_lock);
3784         err = btrfs_run_qgroups(trans);
3785         mutex_unlock(&fs_info->qgroup_ioctl_lock);
3786         if (err < 0)
3787                 btrfs_handle_fs_error(fs_info, err,
3788                                       "failed to update qgroup status and info");
3789         err = btrfs_end_transaction(trans);
3790         if (err && !ret)
3791                 ret = err;
3792
3793 out:
3794         kfree(sa);
3795 drop_write:
3796         mnt_drop_write_file(file);
3797         return ret;
3798 }
3799
3800 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
3801 {
3802         struct inode *inode = file_inode(file);
3803         struct btrfs_root *root = BTRFS_I(inode)->root;
3804         struct btrfs_ioctl_qgroup_create_args *sa;
3805         struct btrfs_trans_handle *trans;
3806         int ret;
3807         int err;
3808
3809         if (!capable(CAP_SYS_ADMIN))
3810                 return -EPERM;
3811
3812         ret = mnt_want_write_file(file);
3813         if (ret)
3814                 return ret;
3815
3816         sa = memdup_user(arg, sizeof(*sa));
3817         if (IS_ERR(sa)) {
3818                 ret = PTR_ERR(sa);
3819                 goto drop_write;
3820         }
3821
3822         if (!sa->qgroupid) {
3823                 ret = -EINVAL;
3824                 goto out;
3825         }
3826
3827         if (sa->create && is_fstree(sa->qgroupid)) {
3828                 ret = -EINVAL;
3829                 goto out;
3830         }
3831
3832         trans = btrfs_join_transaction(root);
3833         if (IS_ERR(trans)) {
3834                 ret = PTR_ERR(trans);
3835                 goto out;
3836         }
3837
3838         if (sa->create) {
3839                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
3840         } else {
3841                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
3842         }
3843
3844         err = btrfs_end_transaction(trans);
3845         if (err && !ret)
3846                 ret = err;
3847
3848 out:
3849         kfree(sa);
3850 drop_write:
3851         mnt_drop_write_file(file);
3852         return ret;
3853 }
3854
3855 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
3856 {
3857         struct inode *inode = file_inode(file);
3858         struct btrfs_root *root = BTRFS_I(inode)->root;
3859         struct btrfs_ioctl_qgroup_limit_args *sa;
3860         struct btrfs_trans_handle *trans;
3861         int ret;
3862         int err;
3863         u64 qgroupid;
3864
3865         if (!capable(CAP_SYS_ADMIN))
3866                 return -EPERM;
3867
3868         ret = mnt_want_write_file(file);
3869         if (ret)
3870                 return ret;
3871
3872         sa = memdup_user(arg, sizeof(*sa));
3873         if (IS_ERR(sa)) {
3874                 ret = PTR_ERR(sa);
3875                 goto drop_write;
3876         }
3877
3878         trans = btrfs_join_transaction(root);
3879         if (IS_ERR(trans)) {
3880                 ret = PTR_ERR(trans);
3881                 goto out;
3882         }
3883
3884         qgroupid = sa->qgroupid;
3885         if (!qgroupid) {
3886                 /* take the current subvol as qgroup */
3887                 qgroupid = root->root_key.objectid;
3888         }
3889
3890         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
3891
3892         err = btrfs_end_transaction(trans);
3893         if (err && !ret)
3894                 ret = err;
3895
3896 out:
3897         kfree(sa);
3898 drop_write:
3899         mnt_drop_write_file(file);
3900         return ret;
3901 }
3902
3903 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
3904 {
3905         struct inode *inode = file_inode(file);
3906         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3907         struct btrfs_ioctl_quota_rescan_args *qsa;
3908         int ret;
3909
3910         if (!capable(CAP_SYS_ADMIN))
3911                 return -EPERM;
3912
3913         ret = mnt_want_write_file(file);
3914         if (ret)
3915                 return ret;
3916
3917         qsa = memdup_user(arg, sizeof(*qsa));
3918         if (IS_ERR(qsa)) {
3919                 ret = PTR_ERR(qsa);
3920                 goto drop_write;
3921         }
3922
3923         if (qsa->flags) {
3924                 ret = -EINVAL;
3925                 goto out;
3926         }
3927
3928         ret = btrfs_qgroup_rescan(fs_info);
3929
3930 out:
3931         kfree(qsa);
3932 drop_write:
3933         mnt_drop_write_file(file);
3934         return ret;
3935 }
3936
3937 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
3938                                                 void __user *arg)
3939 {
3940         struct btrfs_ioctl_quota_rescan_args qsa = {0};
3941
3942         if (!capable(CAP_SYS_ADMIN))
3943                 return -EPERM;
3944
3945         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3946                 qsa.flags = 1;
3947                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
3948         }
3949
3950         if (copy_to_user(arg, &qsa, sizeof(qsa)))
3951                 return -EFAULT;
3952
3953         return 0;
3954 }
3955
3956 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
3957                                                 void __user *arg)
3958 {
3959         if (!capable(CAP_SYS_ADMIN))
3960                 return -EPERM;
3961
3962         return btrfs_qgroup_wait_for_completion(fs_info, true);
3963 }
3964
3965 static long _btrfs_ioctl_set_received_subvol(struct file *file,
3966                                             struct mnt_idmap *idmap,
3967                                             struct btrfs_ioctl_received_subvol_args *sa)
3968 {
3969         struct inode *inode = file_inode(file);
3970         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3971         struct btrfs_root *root = BTRFS_I(inode)->root;
3972         struct btrfs_root_item *root_item = &root->root_item;
3973         struct btrfs_trans_handle *trans;
3974         struct timespec64 ct = current_time(inode);
3975         int ret = 0;
3976         int received_uuid_changed;
3977
3978         if (!inode_owner_or_capable(idmap, inode))
3979                 return -EPERM;
3980
3981         ret = mnt_want_write_file(file);
3982         if (ret < 0)
3983                 return ret;
3984
3985         down_write(&fs_info->subvol_sem);
3986
3987         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3988                 ret = -EINVAL;
3989                 goto out;
3990         }
3991
3992         if (btrfs_root_readonly(root)) {
3993                 ret = -EROFS;
3994                 goto out;
3995         }
3996
3997         /*
3998          * 1 - root item
3999          * 2 - uuid items (received uuid + subvol uuid)
4000          */
4001         trans = btrfs_start_transaction(root, 3);
4002         if (IS_ERR(trans)) {
4003                 ret = PTR_ERR(trans);
4004                 trans = NULL;
4005                 goto out;
4006         }
4007
4008         sa->rtransid = trans->transid;
4009         sa->rtime.sec = ct.tv_sec;
4010         sa->rtime.nsec = ct.tv_nsec;
4011
4012         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4013                                        BTRFS_UUID_SIZE);
4014         if (received_uuid_changed &&
4015             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4016                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4017                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4018                                           root->root_key.objectid);
4019                 if (ret && ret != -ENOENT) {
4020                         btrfs_abort_transaction(trans, ret);
4021                         btrfs_end_transaction(trans);
4022                         goto out;
4023                 }
4024         }
4025         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4026         btrfs_set_root_stransid(root_item, sa->stransid);
4027         btrfs_set_root_rtransid(root_item, sa->rtransid);
4028         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4029         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4030         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4031         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4032
4033         ret = btrfs_update_root(trans, fs_info->tree_root,
4034                                 &root->root_key, &root->root_item);
4035         if (ret < 0) {
4036                 btrfs_end_transaction(trans);
4037                 goto out;
4038         }
4039         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4040                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4041                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4042                                           root->root_key.objectid);
4043                 if (ret < 0 && ret != -EEXIST) {
4044                         btrfs_abort_transaction(trans, ret);
4045                         btrfs_end_transaction(trans);
4046                         goto out;
4047                 }
4048         }
4049         ret = btrfs_commit_transaction(trans);
4050 out:
4051         up_write(&fs_info->subvol_sem);
4052         mnt_drop_write_file(file);
4053         return ret;
4054 }
4055
4056 #ifdef CONFIG_64BIT
4057 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4058                                                 void __user *arg)
4059 {
4060         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4061         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4062         int ret = 0;
4063
4064         args32 = memdup_user(arg, sizeof(*args32));
4065         if (IS_ERR(args32))
4066                 return PTR_ERR(args32);
4067
4068         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4069         if (!args64) {
4070                 ret = -ENOMEM;
4071                 goto out;
4072         }
4073
4074         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4075         args64->stransid = args32->stransid;
4076         args64->rtransid = args32->rtransid;
4077         args64->stime.sec = args32->stime.sec;
4078         args64->stime.nsec = args32->stime.nsec;
4079         args64->rtime.sec = args32->rtime.sec;
4080         args64->rtime.nsec = args32->rtime.nsec;
4081         args64->flags = args32->flags;
4082
4083         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), args64);
4084         if (ret)
4085                 goto out;
4086
4087         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4088         args32->stransid = args64->stransid;
4089         args32->rtransid = args64->rtransid;
4090         args32->stime.sec = args64->stime.sec;
4091         args32->stime.nsec = args64->stime.nsec;
4092         args32->rtime.sec = args64->rtime.sec;
4093         args32->rtime.nsec = args64->rtime.nsec;
4094         args32->flags = args64->flags;
4095
4096         ret = copy_to_user(arg, args32, sizeof(*args32));
4097         if (ret)
4098                 ret = -EFAULT;
4099
4100 out:
4101         kfree(args32);
4102         kfree(args64);
4103         return ret;
4104 }
4105 #endif
4106
4107 static long btrfs_ioctl_set_received_subvol(struct file *file,
4108                                             void __user *arg)
4109 {
4110         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4111         int ret = 0;
4112
4113         sa = memdup_user(arg, sizeof(*sa));
4114         if (IS_ERR(sa))
4115                 return PTR_ERR(sa);
4116
4117         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), sa);
4118
4119         if (ret)
4120                 goto out;
4121
4122         ret = copy_to_user(arg, sa, sizeof(*sa));
4123         if (ret)
4124                 ret = -EFAULT;
4125
4126 out:
4127         kfree(sa);
4128         return ret;
4129 }
4130
4131 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4132                                         void __user *arg)
4133 {
4134         size_t len;
4135         int ret;
4136         char label[BTRFS_LABEL_SIZE];
4137
4138         spin_lock(&fs_info->super_lock);
4139         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4140         spin_unlock(&fs_info->super_lock);
4141
4142         len = strnlen(label, BTRFS_LABEL_SIZE);
4143
4144         if (len == BTRFS_LABEL_SIZE) {
4145                 btrfs_warn(fs_info,
4146                            "label is too long, return the first %zu bytes",
4147                            --len);
4148         }
4149
4150         ret = copy_to_user(arg, label, len);
4151
4152         return ret ? -EFAULT : 0;
4153 }
4154
4155 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4156 {
4157         struct inode *inode = file_inode(file);
4158         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4159         struct btrfs_root *root = BTRFS_I(inode)->root;
4160         struct btrfs_super_block *super_block = fs_info->super_copy;
4161         struct btrfs_trans_handle *trans;
4162         char label[BTRFS_LABEL_SIZE];
4163         int ret;
4164
4165         if (!capable(CAP_SYS_ADMIN))
4166                 return -EPERM;
4167
4168         if (copy_from_user(label, arg, sizeof(label)))
4169                 return -EFAULT;
4170
4171         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4172                 btrfs_err(fs_info,
4173                           "unable to set label with more than %d bytes",
4174                           BTRFS_LABEL_SIZE - 1);
4175                 return -EINVAL;
4176         }
4177
4178         ret = mnt_want_write_file(file);
4179         if (ret)
4180                 return ret;
4181
4182         trans = btrfs_start_transaction(root, 0);
4183         if (IS_ERR(trans)) {
4184                 ret = PTR_ERR(trans);
4185                 goto out_unlock;
4186         }
4187
4188         spin_lock(&fs_info->super_lock);
4189         strcpy(super_block->label, label);
4190         spin_unlock(&fs_info->super_lock);
4191         ret = btrfs_commit_transaction(trans);
4192
4193 out_unlock:
4194         mnt_drop_write_file(file);
4195         return ret;
4196 }
4197
4198 #define INIT_FEATURE_FLAGS(suffix) \
4199         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4200           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4201           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4202
4203 int btrfs_ioctl_get_supported_features(void __user *arg)
4204 {
4205         static const struct btrfs_ioctl_feature_flags features[3] = {
4206                 INIT_FEATURE_FLAGS(SUPP),
4207                 INIT_FEATURE_FLAGS(SAFE_SET),
4208                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4209         };
4210
4211         if (copy_to_user(arg, &features, sizeof(features)))
4212                 return -EFAULT;
4213
4214         return 0;
4215 }
4216
4217 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4218                                         void __user *arg)
4219 {
4220         struct btrfs_super_block *super_block = fs_info->super_copy;
4221         struct btrfs_ioctl_feature_flags features;
4222
4223         features.compat_flags = btrfs_super_compat_flags(super_block);
4224         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4225         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4226
4227         if (copy_to_user(arg, &features, sizeof(features)))
4228                 return -EFAULT;
4229
4230         return 0;
4231 }
4232
4233 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4234                               enum btrfs_feature_set set,
4235                               u64 change_mask, u64 flags, u64 supported_flags,
4236                               u64 safe_set, u64 safe_clear)
4237 {
4238         const char *type = btrfs_feature_set_name(set);
4239         char *names;
4240         u64 disallowed, unsupported;
4241         u64 set_mask = flags & change_mask;
4242         u64 clear_mask = ~flags & change_mask;
4243
4244         unsupported = set_mask & ~supported_flags;
4245         if (unsupported) {
4246                 names = btrfs_printable_features(set, unsupported);
4247                 if (names) {
4248                         btrfs_warn(fs_info,
4249                                    "this kernel does not support the %s feature bit%s",
4250                                    names, strchr(names, ',') ? "s" : "");
4251                         kfree(names);
4252                 } else
4253                         btrfs_warn(fs_info,
4254                                    "this kernel does not support %s bits 0x%llx",
4255                                    type, unsupported);
4256                 return -EOPNOTSUPP;
4257         }
4258
4259         disallowed = set_mask & ~safe_set;
4260         if (disallowed) {
4261                 names = btrfs_printable_features(set, disallowed);
4262                 if (names) {
4263                         btrfs_warn(fs_info,
4264                                    "can't set the %s feature bit%s while mounted",
4265                                    names, strchr(names, ',') ? "s" : "");
4266                         kfree(names);
4267                 } else
4268                         btrfs_warn(fs_info,
4269                                    "can't set %s bits 0x%llx while mounted",
4270                                    type, disallowed);
4271                 return -EPERM;
4272         }
4273
4274         disallowed = clear_mask & ~safe_clear;
4275         if (disallowed) {
4276                 names = btrfs_printable_features(set, disallowed);
4277                 if (names) {
4278                         btrfs_warn(fs_info,
4279                                    "can't clear the %s feature bit%s while mounted",
4280                                    names, strchr(names, ',') ? "s" : "");
4281                         kfree(names);
4282                 } else
4283                         btrfs_warn(fs_info,
4284                                    "can't clear %s bits 0x%llx while mounted",
4285                                    type, disallowed);
4286                 return -EPERM;
4287         }
4288
4289         return 0;
4290 }
4291
4292 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4293 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4294                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4295                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4296                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4297
4298 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4299 {
4300         struct inode *inode = file_inode(file);
4301         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4302         struct btrfs_root *root = BTRFS_I(inode)->root;
4303         struct btrfs_super_block *super_block = fs_info->super_copy;
4304         struct btrfs_ioctl_feature_flags flags[2];
4305         struct btrfs_trans_handle *trans;
4306         u64 newflags;
4307         int ret;
4308
4309         if (!capable(CAP_SYS_ADMIN))
4310                 return -EPERM;
4311
4312         if (copy_from_user(flags, arg, sizeof(flags)))
4313                 return -EFAULT;
4314
4315         /* Nothing to do */
4316         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4317             !flags[0].incompat_flags)
4318                 return 0;
4319
4320         ret = check_feature(fs_info, flags[0].compat_flags,
4321                             flags[1].compat_flags, COMPAT);
4322         if (ret)
4323                 return ret;
4324
4325         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4326                             flags[1].compat_ro_flags, COMPAT_RO);
4327         if (ret)
4328                 return ret;
4329
4330         ret = check_feature(fs_info, flags[0].incompat_flags,
4331                             flags[1].incompat_flags, INCOMPAT);
4332         if (ret)
4333                 return ret;
4334
4335         ret = mnt_want_write_file(file);
4336         if (ret)
4337                 return ret;
4338
4339         trans = btrfs_start_transaction(root, 0);
4340         if (IS_ERR(trans)) {
4341                 ret = PTR_ERR(trans);
4342                 goto out_drop_write;
4343         }
4344
4345         spin_lock(&fs_info->super_lock);
4346         newflags = btrfs_super_compat_flags(super_block);
4347         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4348         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4349         btrfs_set_super_compat_flags(super_block, newflags);
4350
4351         newflags = btrfs_super_compat_ro_flags(super_block);
4352         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4353         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4354         btrfs_set_super_compat_ro_flags(super_block, newflags);
4355
4356         newflags = btrfs_super_incompat_flags(super_block);
4357         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4358         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4359         btrfs_set_super_incompat_flags(super_block, newflags);
4360         spin_unlock(&fs_info->super_lock);
4361
4362         ret = btrfs_commit_transaction(trans);
4363 out_drop_write:
4364         mnt_drop_write_file(file);
4365
4366         return ret;
4367 }
4368
4369 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
4370 {
4371         struct btrfs_ioctl_send_args *arg;
4372         int ret;
4373
4374         if (compat) {
4375 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4376                 struct btrfs_ioctl_send_args_32 args32 = { 0 };
4377
4378                 ret = copy_from_user(&args32, argp, sizeof(args32));
4379                 if (ret)
4380                         return -EFAULT;
4381                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4382                 if (!arg)
4383                         return -ENOMEM;
4384                 arg->send_fd = args32.send_fd;
4385                 arg->clone_sources_count = args32.clone_sources_count;
4386                 arg->clone_sources = compat_ptr(args32.clone_sources);
4387                 arg->parent_root = args32.parent_root;
4388                 arg->flags = args32.flags;
4389                 arg->version = args32.version;
4390                 memcpy(arg->reserved, args32.reserved,
4391                        sizeof(args32.reserved));
4392 #else
4393                 return -ENOTTY;
4394 #endif
4395         } else {
4396                 arg = memdup_user(argp, sizeof(*arg));
4397                 if (IS_ERR(arg))
4398                         return PTR_ERR(arg);
4399         }
4400         ret = btrfs_ioctl_send(inode, arg);
4401         kfree(arg);
4402         return ret;
4403 }
4404
4405 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
4406                                     bool compat)
4407 {
4408         struct btrfs_ioctl_encoded_io_args args = { 0 };
4409         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
4410                                              flags);
4411         size_t copy_end;
4412         struct iovec iovstack[UIO_FASTIOV];
4413         struct iovec *iov = iovstack;
4414         struct iov_iter iter;
4415         loff_t pos;
4416         struct kiocb kiocb;
4417         ssize_t ret;
4418
4419         if (!capable(CAP_SYS_ADMIN)) {
4420                 ret = -EPERM;
4421                 goto out_acct;
4422         }
4423
4424         if (compat) {
4425 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4426                 struct btrfs_ioctl_encoded_io_args_32 args32;
4427
4428                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
4429                                        flags);
4430                 if (copy_from_user(&args32, argp, copy_end)) {
4431                         ret = -EFAULT;
4432                         goto out_acct;
4433                 }
4434                 args.iov = compat_ptr(args32.iov);
4435                 args.iovcnt = args32.iovcnt;
4436                 args.offset = args32.offset;
4437                 args.flags = args32.flags;
4438 #else
4439                 return -ENOTTY;
4440 #endif
4441         } else {
4442                 copy_end = copy_end_kernel;
4443                 if (copy_from_user(&args, argp, copy_end)) {
4444                         ret = -EFAULT;
4445                         goto out_acct;
4446                 }
4447         }
4448         if (args.flags != 0) {
4449                 ret = -EINVAL;
4450                 goto out_acct;
4451         }
4452
4453         ret = import_iovec(ITER_DEST, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4454                            &iov, &iter);
4455         if (ret < 0)
4456                 goto out_acct;
4457
4458         if (iov_iter_count(&iter) == 0) {
4459                 ret = 0;
4460                 goto out_iov;
4461         }
4462         pos = args.offset;
4463         ret = rw_verify_area(READ, file, &pos, args.len);
4464         if (ret < 0)
4465                 goto out_iov;
4466
4467         init_sync_kiocb(&kiocb, file);
4468         kiocb.ki_pos = pos;
4469
4470         ret = btrfs_encoded_read(&kiocb, &iter, &args);
4471         if (ret >= 0) {
4472                 fsnotify_access(file);
4473                 if (copy_to_user(argp + copy_end,
4474                                  (char *)&args + copy_end_kernel,
4475                                  sizeof(args) - copy_end_kernel))
4476                         ret = -EFAULT;
4477         }
4478
4479 out_iov:
4480         kfree(iov);
4481 out_acct:
4482         if (ret > 0)
4483                 add_rchar(current, ret);
4484         inc_syscr(current);
4485         return ret;
4486 }
4487
4488 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
4489 {
4490         struct btrfs_ioctl_encoded_io_args args;
4491         struct iovec iovstack[UIO_FASTIOV];
4492         struct iovec *iov = iovstack;
4493         struct iov_iter iter;
4494         loff_t pos;
4495         struct kiocb kiocb;
4496         ssize_t ret;
4497
4498         if (!capable(CAP_SYS_ADMIN)) {
4499                 ret = -EPERM;
4500                 goto out_acct;
4501         }
4502
4503         if (!(file->f_mode & FMODE_WRITE)) {
4504                 ret = -EBADF;
4505                 goto out_acct;
4506         }
4507
4508         if (compat) {
4509 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4510                 struct btrfs_ioctl_encoded_io_args_32 args32;
4511
4512                 if (copy_from_user(&args32, argp, sizeof(args32))) {
4513                         ret = -EFAULT;
4514                         goto out_acct;
4515                 }
4516                 args.iov = compat_ptr(args32.iov);
4517                 args.iovcnt = args32.iovcnt;
4518                 args.offset = args32.offset;
4519                 args.flags = args32.flags;
4520                 args.len = args32.len;
4521                 args.unencoded_len = args32.unencoded_len;
4522                 args.unencoded_offset = args32.unencoded_offset;
4523                 args.compression = args32.compression;
4524                 args.encryption = args32.encryption;
4525                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
4526 #else
4527                 return -ENOTTY;
4528 #endif
4529         } else {
4530                 if (copy_from_user(&args, argp, sizeof(args))) {
4531                         ret = -EFAULT;
4532                         goto out_acct;
4533                 }
4534         }
4535
4536         ret = -EINVAL;
4537         if (args.flags != 0)
4538                 goto out_acct;
4539         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
4540                 goto out_acct;
4541         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
4542             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
4543                 goto out_acct;
4544         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
4545             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
4546                 goto out_acct;
4547         if (args.unencoded_offset > args.unencoded_len)
4548                 goto out_acct;
4549         if (args.len > args.unencoded_len - args.unencoded_offset)
4550                 goto out_acct;
4551
4552         ret = import_iovec(ITER_SOURCE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4553                            &iov, &iter);
4554         if (ret < 0)
4555                 goto out_acct;
4556
4557         if (iov_iter_count(&iter) == 0) {
4558                 ret = 0;
4559                 goto out_iov;
4560         }
4561         pos = args.offset;
4562         ret = rw_verify_area(WRITE, file, &pos, args.len);
4563         if (ret < 0)
4564                 goto out_iov;
4565
4566         init_sync_kiocb(&kiocb, file);
4567         ret = kiocb_set_rw_flags(&kiocb, 0);
4568         if (ret)
4569                 goto out_iov;
4570         kiocb.ki_pos = pos;
4571
4572         file_start_write(file);
4573
4574         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
4575         if (ret > 0)
4576                 fsnotify_modify(file);
4577
4578         file_end_write(file);
4579 out_iov:
4580         kfree(iov);
4581 out_acct:
4582         if (ret > 0)
4583                 add_wchar(current, ret);
4584         inc_syscw(current);
4585         return ret;
4586 }
4587
4588 long btrfs_ioctl(struct file *file, unsigned int
4589                 cmd, unsigned long arg)
4590 {
4591         struct inode *inode = file_inode(file);
4592         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4593         struct btrfs_root *root = BTRFS_I(inode)->root;
4594         void __user *argp = (void __user *)arg;
4595
4596         switch (cmd) {
4597         case FS_IOC_GETVERSION:
4598                 return btrfs_ioctl_getversion(inode, argp);
4599         case FS_IOC_GETFSLABEL:
4600                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4601         case FS_IOC_SETFSLABEL:
4602                 return btrfs_ioctl_set_fslabel(file, argp);
4603         case FITRIM:
4604                 return btrfs_ioctl_fitrim(fs_info, argp);
4605         case BTRFS_IOC_SNAP_CREATE:
4606                 return btrfs_ioctl_snap_create(file, argp, 0);
4607         case BTRFS_IOC_SNAP_CREATE_V2:
4608                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4609         case BTRFS_IOC_SUBVOL_CREATE:
4610                 return btrfs_ioctl_snap_create(file, argp, 1);
4611         case BTRFS_IOC_SUBVOL_CREATE_V2:
4612                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4613         case BTRFS_IOC_SNAP_DESTROY:
4614                 return btrfs_ioctl_snap_destroy(file, argp, false);
4615         case BTRFS_IOC_SNAP_DESTROY_V2:
4616                 return btrfs_ioctl_snap_destroy(file, argp, true);
4617         case BTRFS_IOC_SUBVOL_GETFLAGS:
4618                 return btrfs_ioctl_subvol_getflags(inode, argp);
4619         case BTRFS_IOC_SUBVOL_SETFLAGS:
4620                 return btrfs_ioctl_subvol_setflags(file, argp);
4621         case BTRFS_IOC_DEFAULT_SUBVOL:
4622                 return btrfs_ioctl_default_subvol(file, argp);
4623         case BTRFS_IOC_DEFRAG:
4624                 return btrfs_ioctl_defrag(file, NULL);
4625         case BTRFS_IOC_DEFRAG_RANGE:
4626                 return btrfs_ioctl_defrag(file, argp);
4627         case BTRFS_IOC_RESIZE:
4628                 return btrfs_ioctl_resize(file, argp);
4629         case BTRFS_IOC_ADD_DEV:
4630                 return btrfs_ioctl_add_dev(fs_info, argp);
4631         case BTRFS_IOC_RM_DEV:
4632                 return btrfs_ioctl_rm_dev(file, argp);
4633         case BTRFS_IOC_RM_DEV_V2:
4634                 return btrfs_ioctl_rm_dev_v2(file, argp);
4635         case BTRFS_IOC_FS_INFO:
4636                 return btrfs_ioctl_fs_info(fs_info, argp);
4637         case BTRFS_IOC_DEV_INFO:
4638                 return btrfs_ioctl_dev_info(fs_info, argp);
4639         case BTRFS_IOC_TREE_SEARCH:
4640                 return btrfs_ioctl_tree_search(inode, argp);
4641         case BTRFS_IOC_TREE_SEARCH_V2:
4642                 return btrfs_ioctl_tree_search_v2(inode, argp);
4643         case BTRFS_IOC_INO_LOOKUP:
4644                 return btrfs_ioctl_ino_lookup(root, argp);
4645         case BTRFS_IOC_INO_PATHS:
4646                 return btrfs_ioctl_ino_to_path(root, argp);
4647         case BTRFS_IOC_LOGICAL_INO:
4648                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4649         case BTRFS_IOC_LOGICAL_INO_V2:
4650                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4651         case BTRFS_IOC_SPACE_INFO:
4652                 return btrfs_ioctl_space_info(fs_info, argp);
4653         case BTRFS_IOC_SYNC: {
4654                 int ret;
4655
4656                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4657                 if (ret)
4658                         return ret;
4659                 ret = btrfs_sync_fs(inode->i_sb, 1);
4660                 /*
4661                  * The transaction thread may want to do more work,
4662                  * namely it pokes the cleaner kthread that will start
4663                  * processing uncleaned subvols.
4664                  */
4665                 wake_up_process(fs_info->transaction_kthread);
4666                 return ret;
4667         }
4668         case BTRFS_IOC_START_SYNC:
4669                 return btrfs_ioctl_start_sync(root, argp);
4670         case BTRFS_IOC_WAIT_SYNC:
4671                 return btrfs_ioctl_wait_sync(fs_info, argp);
4672         case BTRFS_IOC_SCRUB:
4673                 return btrfs_ioctl_scrub(file, argp);
4674         case BTRFS_IOC_SCRUB_CANCEL:
4675                 return btrfs_ioctl_scrub_cancel(fs_info);
4676         case BTRFS_IOC_SCRUB_PROGRESS:
4677                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4678         case BTRFS_IOC_BALANCE_V2:
4679                 return btrfs_ioctl_balance(file, argp);
4680         case BTRFS_IOC_BALANCE_CTL:
4681                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4682         case BTRFS_IOC_BALANCE_PROGRESS:
4683                 return btrfs_ioctl_balance_progress(fs_info, argp);
4684         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4685                 return btrfs_ioctl_set_received_subvol(file, argp);
4686 #ifdef CONFIG_64BIT
4687         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4688                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4689 #endif
4690         case BTRFS_IOC_SEND:
4691                 return _btrfs_ioctl_send(inode, argp, false);
4692 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4693         case BTRFS_IOC_SEND_32:
4694                 return _btrfs_ioctl_send(inode, argp, true);
4695 #endif
4696         case BTRFS_IOC_GET_DEV_STATS:
4697                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4698         case BTRFS_IOC_QUOTA_CTL:
4699                 return btrfs_ioctl_quota_ctl(file, argp);
4700         case BTRFS_IOC_QGROUP_ASSIGN:
4701                 return btrfs_ioctl_qgroup_assign(file, argp);
4702         case BTRFS_IOC_QGROUP_CREATE:
4703                 return btrfs_ioctl_qgroup_create(file, argp);
4704         case BTRFS_IOC_QGROUP_LIMIT:
4705                 return btrfs_ioctl_qgroup_limit(file, argp);
4706         case BTRFS_IOC_QUOTA_RESCAN:
4707                 return btrfs_ioctl_quota_rescan(file, argp);
4708         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4709                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4710         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4711                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4712         case BTRFS_IOC_DEV_REPLACE:
4713                 return btrfs_ioctl_dev_replace(fs_info, argp);
4714         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4715                 return btrfs_ioctl_get_supported_features(argp);
4716         case BTRFS_IOC_GET_FEATURES:
4717                 return btrfs_ioctl_get_features(fs_info, argp);
4718         case BTRFS_IOC_SET_FEATURES:
4719                 return btrfs_ioctl_set_features(file, argp);
4720         case BTRFS_IOC_GET_SUBVOL_INFO:
4721                 return btrfs_ioctl_get_subvol_info(inode, argp);
4722         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4723                 return btrfs_ioctl_get_subvol_rootref(root, argp);
4724         case BTRFS_IOC_INO_LOOKUP_USER:
4725                 return btrfs_ioctl_ino_lookup_user(file, argp);
4726         case FS_IOC_ENABLE_VERITY:
4727                 return fsverity_ioctl_enable(file, (const void __user *)argp);
4728         case FS_IOC_MEASURE_VERITY:
4729                 return fsverity_ioctl_measure(file, argp);
4730         case BTRFS_IOC_ENCODED_READ:
4731                 return btrfs_ioctl_encoded_read(file, argp, false);
4732         case BTRFS_IOC_ENCODED_WRITE:
4733                 return btrfs_ioctl_encoded_write(file, argp, false);
4734 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4735         case BTRFS_IOC_ENCODED_READ_32:
4736                 return btrfs_ioctl_encoded_read(file, argp, true);
4737         case BTRFS_IOC_ENCODED_WRITE_32:
4738                 return btrfs_ioctl_encoded_write(file, argp, true);
4739 #endif
4740         }
4741
4742         return -ENOTTY;
4743 }
4744
4745 #ifdef CONFIG_COMPAT
4746 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4747 {
4748         /*
4749          * These all access 32-bit values anyway so no further
4750          * handling is necessary.
4751          */
4752         switch (cmd) {
4753         case FS_IOC32_GETVERSION:
4754                 cmd = FS_IOC_GETVERSION;
4755                 break;
4756         }
4757
4758         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
4759 }
4760 #endif