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 "volumes.h"
  38 #include "locking.h"
  39 #include "backref.h"
  40 #include "send.h"
  41 #include "dev-replace.h"
  42 #include "props.h"
  43 #include "sysfs.h"
  44 #include "qgroup.h"
  45 #include "tree-log.h"
  46 #include "compression.h"
  47 #include "space-info.h"
  48 #include "block-group.h"
  49 #include "fs.h"
  50 #include "accessors.h"
  51 #include "extent-tree.h"
  52 #include "root-tree.h"
  53 #include "defrag.h"
  54 #include "dir-item.h"
  55 #include "uuid-tree.h"
  56 #include "ioctl.h"
  57 #include "file.h"
  58 #include "scrub.h"
  59 #include "super.h"
  60
  61 #ifdef CONFIG_64BIT
  62 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  63  * structures are incorrect, as the timespec structure from userspace
  64  * is 4 bytes too small. We define these alternatives here to teach
  65  * the kernel about the 32-bit struct packing.
  66  */
  67 struct btrfs_ioctl_timespec_32 {
  68         __u64 sec;
  69         __u32 nsec;
  70 } __attribute__ ((__packed__));
  71
  72 struct btrfs_ioctl_received_subvol_args_32 {
  73         char    uuid[BTRFS_UUID_SIZE];  /* in */
  74         __u64   stransid;               /* in */
  75         __u64   rtransid;               /* out */
  76         struct btrfs_ioctl_timespec_32 stime; /* in */
  77         struct btrfs_ioctl_timespec_32 rtime; /* out */
  78         __u64   flags;                  /* in */
  79         __u64   reserved[16];           /* in */
  80 } __attribute__ ((__packed__));
  81
  82 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  83                                 struct btrfs_ioctl_received_subvol_args_32)
  84 #endif
  85
  86 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  87 struct btrfs_ioctl_send_args_32 {
  88         __s64 send_fd;                  /* in */
  89         __u64 clone_sources_count;      /* in */
  90         compat_uptr_t clone_sources;    /* in */
  91         __u64 parent_root;              /* in */
  92         __u64 flags;                    /* in */
  93         __u32 version;                  /* in */
  94         __u8  reserved[28];             /* in */
  95 } __attribute__ ((__packed__));
  96
  97 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  98                                struct btrfs_ioctl_send_args_32)
  99
 100 struct btrfs_ioctl_encoded_io_args_32 {
 101         compat_uptr_t iov;
 102         compat_ulong_t iovcnt;
 103         __s64 offset;
 104         __u64 flags;
 105         __u64 len;
 106         __u64 unencoded_len;
 107         __u64 unencoded_offset;
 108         __u32 compression;
 109         __u32 encryption;
 110         __u8 reserved[64];
 111 };
 112
 113 #define BTRFS_IOC_ENCODED_READ_32 _IOR(BTRFS_IOCTL_MAGIC, 64, \
 114                                        struct btrfs_ioctl_encoded_io_args_32)
 115 #define BTRFS_IOC_ENCODED_WRITE_32 _IOW(BTRFS_IOCTL_MAGIC, 64, \
 116                                         struct btrfs_ioctl_encoded_io_args_32)
 117 #endif
 118
 119 /* Mask out flags that are inappropriate for the given type of inode. */
 120 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
 121                 unsigned int flags)
 122 {
 123         if (S_ISDIR(inode->i_mode))
 124                 return flags;
 125         else if (S_ISREG(inode->i_mode))
 126                 return flags & ~FS_DIRSYNC_FL;
 127         else
 128                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 129 }
 130
 131 /*
 132  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 133  * ioctl.
 134  */
 135 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
 136 {
 137         unsigned int iflags = 0;
 138         u32 flags = binode->flags;
 139         u32 ro_flags = binode->ro_flags;
 140
 141         if (flags & BTRFS_INODE_SYNC)
 142                 iflags |= FS_SYNC_FL;
 143         if (flags & BTRFS_INODE_IMMUTABLE)
 144                 iflags |= FS_IMMUTABLE_FL;
 145         if (flags & BTRFS_INODE_APPEND)
 146                 iflags |= FS_APPEND_FL;
 147         if (flags & BTRFS_INODE_NODUMP)
 148                 iflags |= FS_NODUMP_FL;
 149         if (flags & BTRFS_INODE_NOATIME)
 150                 iflags |= FS_NOATIME_FL;
 151         if (flags & BTRFS_INODE_DIRSYNC)
 152                 iflags |= FS_DIRSYNC_FL;
 153         if (flags & BTRFS_INODE_NODATACOW)
 154                 iflags |= FS_NOCOW_FL;
 155         if (ro_flags & BTRFS_INODE_RO_VERITY)
 156                 iflags |= FS_VERITY_FL;
 157
 158         if (flags & BTRFS_INODE_NOCOMPRESS)
 159                 iflags |= FS_NOCOMP_FL;
 160         else if (flags & BTRFS_INODE_COMPRESS)
 161                 iflags |= FS_COMPR_FL;
 162
 163         return iflags;
 164 }
 165
 166 /*
 167  * Update inode->i_flags based on the btrfs internal flags.
 168  */
 169 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 170 {
 171         struct btrfs_inode *binode = BTRFS_I(inode);
 172         unsigned int new_fl = 0;
 173
 174         if (binode->flags & BTRFS_INODE_SYNC)
 175                 new_fl |= S_SYNC;
 176         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 177                 new_fl |= S_IMMUTABLE;
 178         if (binode->flags & BTRFS_INODE_APPEND)
 179                 new_fl |= S_APPEND;
 180         if (binode->flags & BTRFS_INODE_NOATIME)
 181                 new_fl |= S_NOATIME;
 182         if (binode->flags & BTRFS_INODE_DIRSYNC)
 183                 new_fl |= S_DIRSYNC;
 184         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
 185                 new_fl |= S_VERITY;
 186
 187         set_mask_bits(&inode->i_flags,
 188                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
 189                       S_VERITY, new_fl);
 190 }
 191
 192 /*
 193  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 194  * the old and new flags are not conflicting
 195  */
 196 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 197 {
 198         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 199                       FS_NOATIME_FL | FS_NODUMP_FL | \
 200                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 201                       FS_NOCOMP_FL | FS_COMPR_FL |
 202                       FS_NOCOW_FL))
 203                 return -EOPNOTSUPP;
 204
 205         /* COMPR and NOCOMP on new/old are valid */
 206         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 207                 return -EINVAL;
 208
 209         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 210                 return -EINVAL;
 211
 212         /* NOCOW and compression options are mutually exclusive */
 213         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 214                 return -EINVAL;
 215         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 216                 return -EINVAL;
 217
 218         return 0;
 219 }
 220
 221 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 222                                     unsigned int flags)
 223 {
 224         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 225                 return -EPERM;
 226
 227         return 0;
 228 }
 229
 230 int btrfs_check_ioctl_vol_args_path(const struct btrfs_ioctl_vol_args *vol_args)
 231 {
 232         if (memchr(vol_args->name, 0, sizeof(vol_args->name)) == NULL)
 233                 return -ENAMETOOLONG;
 234         return 0;
 235 }
 236
 237 static int btrfs_check_ioctl_vol_args2_subvol_name(const struct btrfs_ioctl_vol_args_v2 *vol_args2)
 238 {
 239         if (memchr(vol_args2->name, 0, sizeof(vol_args2->name)) == NULL)
 240                 return -ENAMETOOLONG;
 241         return 0;
 242 }
 243
 244 /*
 245  * Set flags/xflags from the internal inode flags. The remaining items of
 246  * fsxattr are zeroed.
 247  */
 248 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 249 {
 250         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
 251
 252         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
 253         return 0;
 254 }
 255
 256 int btrfs_fileattr_set(struct mnt_idmap *idmap,
 257                        struct dentry *dentry, struct fileattr *fa)
 258 {
 259         struct inode *inode = d_inode(dentry);
 260         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
 261         struct btrfs_inode *binode = BTRFS_I(inode);
 262         struct btrfs_root *root = binode->root;
 263         struct btrfs_trans_handle *trans;
 264         unsigned int fsflags, old_fsflags;
 265         int ret;
 266         const char *comp = NULL;
 267         u32 binode_flags;
 268
 269         if (btrfs_root_readonly(root))
 270                 return -EROFS;
 271
 272         if (fileattr_has_fsx(fa))
 273                 return -EOPNOTSUPP;
 274
 275         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
 276         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
 277         ret = check_fsflags(old_fsflags, fsflags);
 278         if (ret)
 279                 return ret;
 280
 281         ret = check_fsflags_compatible(fs_info, fsflags);
 282         if (ret)
 283                 return ret;
 284
 285         binode_flags = binode->flags;
 286         if (fsflags & FS_SYNC_FL)
 287                 binode_flags |= BTRFS_INODE_SYNC;
 288         else
 289                 binode_flags &= ~BTRFS_INODE_SYNC;
 290         if (fsflags & FS_IMMUTABLE_FL)
 291                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 292         else
 293                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 294         if (fsflags & FS_APPEND_FL)
 295                 binode_flags |= BTRFS_INODE_APPEND;
 296         else
 297                 binode_flags &= ~BTRFS_INODE_APPEND;
 298         if (fsflags & FS_NODUMP_FL)
 299                 binode_flags |= BTRFS_INODE_NODUMP;
 300         else
 301                 binode_flags &= ~BTRFS_INODE_NODUMP;
 302         if (fsflags & FS_NOATIME_FL)
 303                 binode_flags |= BTRFS_INODE_NOATIME;
 304         else
 305                 binode_flags &= ~BTRFS_INODE_NOATIME;
 306
 307         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
 308         if (!fa->flags_valid) {
 309                 /* 1 item for the inode */
 310                 trans = btrfs_start_transaction(root, 1);
 311                 if (IS_ERR(trans))
 312                         return PTR_ERR(trans);
 313                 goto update_flags;
 314         }
 315
 316         if (fsflags & FS_DIRSYNC_FL)
 317                 binode_flags |= BTRFS_INODE_DIRSYNC;
 318         else
 319                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 320         if (fsflags & FS_NOCOW_FL) {
 321                 if (S_ISREG(inode->i_mode)) {
 322                         /*
 323                          * It's safe to turn csums off here, no extents exist.
 324                          * Otherwise we want the flag to reflect the real COW
 325                          * status of the file and will not set it.
 326                          */
 327                         if (inode->i_size == 0)
 328                                 binode_flags |= BTRFS_INODE_NODATACOW |
 329                                                 BTRFS_INODE_NODATASUM;
 330                 } else {
 331                         binode_flags |= BTRFS_INODE_NODATACOW;
 332                 }
 333         } else {
 334                 /*
 335                  * Revert back under same assumptions as above
 336                  */
 337                 if (S_ISREG(inode->i_mode)) {
 338                         if (inode->i_size == 0)
 339                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 340                                                   BTRFS_INODE_NODATASUM);
 341                 } else {
 342                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 343                 }
 344         }
 345
 346         /*
 347          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 348          * flag may be changed automatically if compression code won't make
 349          * things smaller.
 350          */
 351         if (fsflags & FS_NOCOMP_FL) {
 352                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 353                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 354         } else if (fsflags & FS_COMPR_FL) {
 355
 356                 if (IS_SWAPFILE(inode))
 357                         return -ETXTBSY;
 358
 359                 binode_flags |= BTRFS_INODE_COMPRESS;
 360                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 361
 362                 comp = btrfs_compress_type2str(fs_info->compress_type);
 363                 if (!comp || comp[0] == 0)
 364                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 365         } else {
 366                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 367         }
 368
 369         /*
 370          * 1 for inode item
 371          * 2 for properties
 372          */
 373         trans = btrfs_start_transaction(root, 3);
 374         if (IS_ERR(trans))
 375                 return PTR_ERR(trans);
 376
 377         if (comp) {
 378                 ret = btrfs_set_prop(trans, BTRFS_I(inode), "btrfs.compression",
 379                                      comp, strlen(comp), 0);
 380                 if (ret) {
 381                         btrfs_abort_transaction(trans, ret);
 382                         goto out_end_trans;
 383                 }
 384         } else {
 385                 ret = btrfs_set_prop(trans, BTRFS_I(inode), "btrfs.compression",
 386                                      NULL, 0, 0);
 387                 if (ret && ret != -ENODATA) {
 388                         btrfs_abort_transaction(trans, ret);
 389                         goto out_end_trans;
 390                 }
 391         }
 392
 393 update_flags:
 394         binode->flags = binode_flags;
 395         btrfs_sync_inode_flags_to_i_flags(inode);
 396         inode_inc_iversion(inode);
 397         inode_set_ctime_current(inode);
 398         ret = btrfs_update_inode(trans, BTRFS_I(inode));
 399
 400  out_end_trans:
 401         btrfs_end_transaction(trans);
 402         return ret;
 403 }
 404
 405 /*
 406  * Start exclusive operation @type, return true on success
 407  */
 408 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 409                         enum btrfs_exclusive_operation type)
 410 {
 411         bool ret = false;
 412
 413         spin_lock(&fs_info->super_lock);
 414         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
 415                 fs_info->exclusive_operation = type;
 416                 ret = true;
 417         }
 418         spin_unlock(&fs_info->super_lock);
 419
 420         return ret;
 421 }
 422
 423 /*
 424  * Conditionally allow to enter the exclusive operation in case it's compatible
 425  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
 426  * btrfs_exclop_finish.
 427  *
 428  * Compatibility:
 429  * - the same type is already running
 430  * - when trying to add a device and balance has been paused
 431  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
 432  *   must check the condition first that would allow none -> @type
 433  */
 434 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
 435                                  enum btrfs_exclusive_operation type)
 436 {
 437         spin_lock(&fs_info->super_lock);
 438         if (fs_info->exclusive_operation == type ||
 439             (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED &&
 440              type == BTRFS_EXCLOP_DEV_ADD))
 441                 return true;
 442
 443         spin_unlock(&fs_info->super_lock);
 444         return false;
 445 }
 446
 447 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
 448 {
 449         spin_unlock(&fs_info->super_lock);
 450 }
 451
 452 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 453 {
 454         spin_lock(&fs_info->super_lock);
 455         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 456         spin_unlock(&fs_info->super_lock);
 457         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 458 }
 459
 460 void btrfs_exclop_balance(struct btrfs_fs_info *fs_info,
 461                           enum btrfs_exclusive_operation op)
 462 {
 463         switch (op) {
 464         case BTRFS_EXCLOP_BALANCE_PAUSED:
 465                 spin_lock(&fs_info->super_lock);
 466                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE ||
 467                        fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD ||
 468                        fs_info->exclusive_operation == BTRFS_EXCLOP_NONE ||
 469                        fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 470                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED;
 471                 spin_unlock(&fs_info->super_lock);
 472                 break;
 473         case BTRFS_EXCLOP_BALANCE:
 474                 spin_lock(&fs_info->super_lock);
 475                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 476                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
 477                 spin_unlock(&fs_info->super_lock);
 478                 break;
 479         default:
 480                 btrfs_warn(fs_info,
 481                         "invalid exclop balance operation %d requested", op);
 482         }
 483 }
 484
 485 static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg)
 486 {
 487         return put_user(inode->i_generation, arg);
 488 }
 489
 490 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 491                                         void __user *arg)
 492 {
 493         struct btrfs_device *device;
 494         struct fstrim_range range;
 495         u64 minlen = ULLONG_MAX;
 496         u64 num_devices = 0;
 497         int ret;
 498
 499         if (!capable(CAP_SYS_ADMIN))
 500                 return -EPERM;
 501
 502         /*
 503          * btrfs_trim_block_group() depends on space cache, which is not
 504          * available in zoned filesystem. So, disallow fitrim on a zoned
 505          * filesystem for now.
 506          */
 507         if (btrfs_is_zoned(fs_info))
 508                 return -EOPNOTSUPP;
 509
 510         /*
 511          * If the fs is mounted with nologreplay, which requires it to be
 512          * mounted in RO mode as well, we can not allow discard on free space
 513          * inside block groups, because log trees refer to extents that are not
 514          * pinned in a block group's free space cache (pinning the extents is
 515          * precisely the first phase of replaying a log tree).
 516          */
 517         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 518                 return -EROFS;
 519
 520         rcu_read_lock();
 521         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 522                                 dev_list) {
 523                 if (!device->bdev || !bdev_max_discard_sectors(device->bdev))
 524                         continue;
 525                 num_devices++;
 526                 minlen = min_t(u64, bdev_discard_granularity(device->bdev),
 527                                     minlen);
 528         }
 529         rcu_read_unlock();
 530
 531         if (!num_devices)
 532                 return -EOPNOTSUPP;
 533         if (copy_from_user(&range, arg, sizeof(range)))
 534                 return -EFAULT;
 535
 536         /*
 537          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 538          * block group is in the logical address space, which can be any
 539          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 540          */
 541         if (range.len < fs_info->sectorsize)
 542                 return -EINVAL;
 543
 544         range.minlen = max(range.minlen, minlen);
 545         ret = btrfs_trim_fs(fs_info, &range);
 546
 547         if (copy_to_user(arg, &range, sizeof(range)))
 548                 return -EFAULT;
 549
 550         return ret;
 551 }
 552
 553 int __pure btrfs_is_empty_uuid(const u8 *uuid)
 554 {
 555         int i;
 556
 557         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 558                 if (uuid[i])
 559                         return 0;
 560         }
 561         return 1;
 562 }
 563
 564 /*
 565  * Calculate the number of transaction items to reserve for creating a subvolume
 566  * or snapshot, not including the inode, directory entries, or parent directory.
 567  */
 568 static unsigned int create_subvol_num_items(struct btrfs_qgroup_inherit *inherit)
 569 {
 570         /*
 571          * 1 to add root block
 572          * 1 to add root item
 573          * 1 to add root ref
 574          * 1 to add root backref
 575          * 1 to add UUID item
 576          * 1 to add qgroup info
 577          * 1 to add qgroup limit
 578          *
 579          * Ideally the last two would only be accounted if qgroups are enabled,
 580          * but that can change between now and the time we would insert them.
 581          */
 582         unsigned int num_items = 7;
 583
 584         if (inherit) {
 585                 /* 2 to add qgroup relations for each inherited qgroup */
 586                 num_items += 2 * inherit->num_qgroups;
 587         }
 588         return num_items;
 589 }
 590
 591 static noinline int create_subvol(struct mnt_idmap *idmap,
 592                                   struct inode *dir, struct dentry *dentry,
 593                                   struct btrfs_qgroup_inherit *inherit)
 594 {
 595         struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
 596         struct btrfs_trans_handle *trans;
 597         struct btrfs_key key;
 598         struct btrfs_root_item *root_item;
 599         struct btrfs_inode_item *inode_item;
 600         struct extent_buffer *leaf;
 601         struct btrfs_root *root = BTRFS_I(dir)->root;
 602         struct btrfs_root *new_root;
 603         struct btrfs_block_rsv block_rsv;
 604         struct timespec64 cur_time = current_time(dir);
 605         struct btrfs_new_inode_args new_inode_args = {
 606                 .dir = dir,
 607                 .dentry = dentry,
 608                 .subvol = true,
 609         };
 610         unsigned int trans_num_items;
 611         int ret;
 612         dev_t anon_dev;
 613         u64 objectid;
 614         u64 qgroup_reserved = 0;
 615
 616         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 617         if (!root_item)
 618                 return -ENOMEM;
 619
 620         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 621         if (ret)
 622                 goto out_root_item;
 623
 624         /*
 625          * Don't create subvolume whose level is not zero. Or qgroup will be
 626          * screwed up since it assumes subvolume qgroup's level to be 0.
 627          */
 628         if (btrfs_qgroup_level(objectid)) {
 629                 ret = -ENOSPC;
 630                 goto out_root_item;
 631         }
 632
 633         ret = get_anon_bdev(&anon_dev);
 634         if (ret < 0)
 635                 goto out_root_item;
 636
 637         new_inode_args.inode = btrfs_new_subvol_inode(idmap, dir);
 638         if (!new_inode_args.inode) {
 639                 ret = -ENOMEM;
 640                 goto out_anon_dev;
 641         }
 642         ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
 643         if (ret)
 644                 goto out_inode;
 645         trans_num_items += create_subvol_num_items(inherit);
 646
 647         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 648         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 649                                                trans_num_items, false);
 650         if (ret)
 651                 goto out_new_inode_args;
 652         qgroup_reserved = block_rsv.qgroup_rsv_reserved;
 653
 654         trans = btrfs_start_transaction(root, 0);
 655         if (IS_ERR(trans)) {
 656                 ret = PTR_ERR(trans);
 657                 goto out_release_rsv;
 658         }
 659         btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
 660         qgroup_reserved = 0;
 661         trans->block_rsv = &block_rsv;
 662         trans->bytes_reserved = block_rsv.size;
 663
 664         ret = btrfs_qgroup_inherit(trans, 0, objectid, btrfs_root_id(root), inherit);
 665         if (ret)
 666                 goto out;
 667
 668         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 669                                       0, BTRFS_NESTING_NORMAL);
 670         if (IS_ERR(leaf)) {
 671                 ret = PTR_ERR(leaf);
 672                 goto out;
 673         }
 674
 675         btrfs_mark_buffer_dirty(trans, leaf);
 676
 677         inode_item = &root_item->inode;
 678         btrfs_set_stack_inode_generation(inode_item, 1);
 679         btrfs_set_stack_inode_size(inode_item, 3);
 680         btrfs_set_stack_inode_nlink(inode_item, 1);
 681         btrfs_set_stack_inode_nbytes(inode_item,
 682                                      fs_info->nodesize);
 683         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 684
 685         btrfs_set_root_flags(root_item, 0);
 686         btrfs_set_root_limit(root_item, 0);
 687         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 688
 689         btrfs_set_root_bytenr(root_item, leaf->start);
 690         btrfs_set_root_generation(root_item, trans->transid);
 691         btrfs_set_root_level(root_item, 0);
 692         btrfs_set_root_refs(root_item, 1);
 693         btrfs_set_root_used(root_item, leaf->len);
 694         btrfs_set_root_last_snapshot(root_item, 0);
 695
 696         btrfs_set_root_generation_v2(root_item,
 697                         btrfs_root_generation(root_item));
 698         generate_random_guid(root_item->uuid);
 699         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 700         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 701         root_item->ctime = root_item->otime;
 702         btrfs_set_root_ctransid(root_item, trans->transid);
 703         btrfs_set_root_otransid(root_item, trans->transid);
 704
 705         btrfs_tree_unlock(leaf);
 706
 707         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 708
 709         key.objectid = objectid;
 710         key.offset = 0;
 711         key.type = BTRFS_ROOT_ITEM_KEY;
 712         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 713                                 root_item);
 714         if (ret) {
 715                 int ret2;
 716
 717                 /*
 718                  * Since we don't abort the transaction in this case, free the
 719                  * tree block so that we don't leak space and leave the
 720                  * filesystem in an inconsistent state (an extent item in the
 721                  * extent tree with a backreference for a root that does not
 722                  * exists).
 723                  */
 724                 btrfs_tree_lock(leaf);
 725                 btrfs_clear_buffer_dirty(trans, leaf);
 726                 btrfs_tree_unlock(leaf);
 727                 ret2 = btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 728                 if (ret2 < 0)
 729                         btrfs_abort_transaction(trans, ret2);
 730                 free_extent_buffer(leaf);
 731                 goto out;
 732         }
 733
 734         free_extent_buffer(leaf);
 735         leaf = NULL;
 736
 737         new_root = btrfs_get_new_fs_root(fs_info, objectid, &anon_dev);
 738         if (IS_ERR(new_root)) {
 739                 ret = PTR_ERR(new_root);
 740                 btrfs_abort_transaction(trans, ret);
 741                 goto out;
 742         }
 743         /* anon_dev is owned by new_root now. */
 744         anon_dev = 0;
 745         BTRFS_I(new_inode_args.inode)->root = new_root;
 746         /* ... and new_root is owned by new_inode_args.inode now. */
 747
 748         ret = btrfs_record_root_in_trans(trans, new_root);
 749         if (ret) {
 750                 btrfs_abort_transaction(trans, ret);
 751                 goto out;
 752         }
 753
 754         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 755                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 756         if (ret) {
 757                 btrfs_abort_transaction(trans, ret);
 758                 goto out;
 759         }
 760
 761         ret = btrfs_create_new_inode(trans, &new_inode_args);
 762         if (ret) {
 763                 btrfs_abort_transaction(trans, ret);
 764                 goto out;
 765         }
 766
 767         btrfs_record_new_subvolume(trans, BTRFS_I(dir));
 768
 769         d_instantiate_new(dentry, new_inode_args.inode);
 770         new_inode_args.inode = NULL;
 771
 772 out:
 773         trans->block_rsv = NULL;
 774         trans->bytes_reserved = 0;
 775         btrfs_end_transaction(trans);
 776 out_release_rsv:
 777         btrfs_block_rsv_release(fs_info, &block_rsv, (u64)-1, NULL);
 778         if (qgroup_reserved)
 779                 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
 780 out_new_inode_args:
 781         btrfs_new_inode_args_destroy(&new_inode_args);
 782 out_inode:
 783         iput(new_inode_args.inode);
 784 out_anon_dev:
 785         if (anon_dev)
 786                 free_anon_bdev(anon_dev);
 787 out_root_item:
 788         kfree(root_item);
 789         return ret;
 790 }
 791
 792 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 793                            struct dentry *dentry, bool readonly,
 794                            struct btrfs_qgroup_inherit *inherit)
 795 {
 796         struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
 797         struct inode *inode;
 798         struct btrfs_pending_snapshot *pending_snapshot;
 799         unsigned int trans_num_items;
 800         struct btrfs_trans_handle *trans;
 801         struct btrfs_block_rsv *block_rsv;
 802         u64 qgroup_reserved = 0;
 803         int ret;
 804
 805         /* We do not support snapshotting right now. */
 806         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 807                 btrfs_warn(fs_info,
 808                            "extent tree v2 doesn't support snapshotting yet");
 809                 return -EOPNOTSUPP;
 810         }
 811
 812         if (btrfs_root_refs(&root->root_item) == 0)
 813                 return -ENOENT;
 814
 815         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 816                 return -EINVAL;
 817
 818         if (atomic_read(&root->nr_swapfiles)) {
 819                 btrfs_warn(fs_info,
 820                            "cannot snapshot subvolume with active swapfile");
 821                 return -ETXTBSY;
 822         }
 823
 824         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 825         if (!pending_snapshot)
 826                 return -ENOMEM;
 827
 828         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 829         if (ret < 0)
 830                 goto free_pending;
 831         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 832                         GFP_KERNEL);
 833         pending_snapshot->path = btrfs_alloc_path();
 834         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 835                 ret = -ENOMEM;
 836                 goto free_pending;
 837         }
 838
 839         block_rsv = &pending_snapshot->block_rsv;
 840         btrfs_init_block_rsv(block_rsv, BTRFS_BLOCK_RSV_TEMP);
 841         /*
 842          * 1 to add dir item
 843          * 1 to add dir index
 844          * 1 to update parent inode item
 845          */
 846         trans_num_items = create_subvol_num_items(inherit) + 3;
 847         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, block_rsv,
 848                                                trans_num_items, false);
 849         if (ret)
 850                 goto free_pending;
 851         qgroup_reserved = block_rsv->qgroup_rsv_reserved;
 852
 853         pending_snapshot->dentry = dentry;
 854         pending_snapshot->root = root;
 855         pending_snapshot->readonly = readonly;
 856         pending_snapshot->dir = BTRFS_I(dir);
 857         pending_snapshot->inherit = inherit;
 858
 859         trans = btrfs_start_transaction(root, 0);
 860         if (IS_ERR(trans)) {
 861                 ret = PTR_ERR(trans);
 862                 goto fail;
 863         }
 864         ret = btrfs_record_root_in_trans(trans, BTRFS_I(dir)->root);
 865         if (ret) {
 866                 btrfs_end_transaction(trans);
 867                 goto fail;
 868         }
 869         btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved);
 870         qgroup_reserved = 0;
 871
 872         trans->pending_snapshot = pending_snapshot;
 873
 874         ret = btrfs_commit_transaction(trans);
 875         if (ret)
 876                 goto fail;
 877
 878         ret = pending_snapshot->error;
 879         if (ret)
 880                 goto fail;
 881
 882         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 883         if (ret)
 884                 goto fail;
 885
 886         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 887         if (IS_ERR(inode)) {
 888                 ret = PTR_ERR(inode);
 889                 goto fail;
 890         }
 891
 892         d_instantiate(dentry, inode);
 893         ret = 0;
 894         pending_snapshot->anon_dev = 0;
 895 fail:
 896         /* Prevent double freeing of anon_dev */
 897         if (ret && pending_snapshot->snap)
 898                 pending_snapshot->snap->anon_dev = 0;
 899         btrfs_put_root(pending_snapshot->snap);
 900         btrfs_block_rsv_release(fs_info, block_rsv, (u64)-1, NULL);
 901         if (qgroup_reserved)
 902                 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved);
 903 free_pending:
 904         if (pending_snapshot->anon_dev)
 905                 free_anon_bdev(pending_snapshot->anon_dev);
 906         kfree(pending_snapshot->root_item);
 907         btrfs_free_path(pending_snapshot->path);
 908         kfree(pending_snapshot);
 909
 910         return ret;
 911 }
 912
 913 /*  copy of may_delete in fs/namei.c()
 914  *      Check whether we can remove a link victim from directory dir, check
 915  *  whether the type of victim is right.
 916  *  1. We can't do it if dir is read-only (done in permission())
 917  *  2. We should have write and exec permissions on dir
 918  *  3. We can't remove anything from append-only dir
 919  *  4. We can't do anything with immutable dir (done in permission())
 920  *  5. If the sticky bit on dir is set we should either
 921  *      a. be owner of dir, or
 922  *      b. be owner of victim, or
 923  *      c. have CAP_FOWNER capability
 924  *  6. If the victim is append-only or immutable we can't do anything with
 925  *     links pointing to it.
 926  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 927  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 928  *  9. We can't remove a root or mountpoint.
 929  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 930  *     nfs_async_unlink().
 931  */
 932
 933 static int btrfs_may_delete(struct mnt_idmap *idmap,
 934                             struct inode *dir, struct dentry *victim, int isdir)
 935 {
 936         int error;
 937
 938         if (d_really_is_negative(victim))
 939                 return -ENOENT;
 940
 941         /* The @victim is not inside @dir. */
 942         if (d_inode(victim->d_parent) != dir)
 943                 return -EINVAL;
 944         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 945
 946         error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
 947         if (error)
 948                 return error;
 949         if (IS_APPEND(dir))
 950                 return -EPERM;
 951         if (check_sticky(idmap, dir, d_inode(victim)) ||
 952             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 953             IS_SWAPFILE(d_inode(victim)))
 954                 return -EPERM;
 955         if (isdir) {
 956                 if (!d_is_dir(victim))
 957                         return -ENOTDIR;
 958                 if (IS_ROOT(victim))
 959                         return -EBUSY;
 960         } else if (d_is_dir(victim))
 961                 return -EISDIR;
 962         if (IS_DEADDIR(dir))
 963                 return -ENOENT;
 964         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 965                 return -EBUSY;
 966         return 0;
 967 }
 968
 969 /* copy of may_create in fs/namei.c() */
 970 static inline int btrfs_may_create(struct mnt_idmap *idmap,
 971                                    struct inode *dir, struct dentry *child)
 972 {
 973         if (d_really_is_positive(child))
 974                 return -EEXIST;
 975         if (IS_DEADDIR(dir))
 976                 return -ENOENT;
 977         if (!fsuidgid_has_mapping(dir->i_sb, idmap))
 978                 return -EOVERFLOW;
 979         return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
 980 }
 981
 982 /*
 983  * Create a new subvolume below @parent.  This is largely modeled after
 984  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 985  * inside this filesystem so it's quite a bit simpler.
 986  */
 987 static noinline int btrfs_mksubvol(const struct path *parent,
 988                                    struct mnt_idmap *idmap,
 989                                    const char *name, int namelen,
 990                                    struct btrfs_root *snap_src,
 991                                    bool readonly,
 992                                    struct btrfs_qgroup_inherit *inherit)
 993 {
 994         struct inode *dir = d_inode(parent->dentry);
 995         struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
 996         struct dentry *dentry;
 997         struct fscrypt_str name_str = FSTR_INIT((char *)name, namelen);
 998         int error;
 999
1000         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
1001         if (error == -EINTR)
1002                 return error;
1003
1004         dentry = lookup_one(idmap, name, parent->dentry, namelen);
1005         error = PTR_ERR(dentry);
1006         if (IS_ERR(dentry))
1007                 goto out_unlock;
1008
1009         error = btrfs_may_create(idmap, dir, dentry);
1010         if (error)
1011                 goto out_dput;
1012
1013         /*
1014          * even if this name doesn't exist, we may get hash collisions.
1015          * check for them now when we can safely fail
1016          */
1017         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
1018                                                dir->i_ino, &name_str);
1019         if (error)
1020                 goto out_dput;
1021
1022         down_read(&fs_info->subvol_sem);
1023
1024         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
1025                 goto out_up_read;
1026
1027         if (snap_src)
1028                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
1029         else
1030                 error = create_subvol(idmap, dir, dentry, inherit);
1031
1032         if (!error)
1033                 fsnotify_mkdir(dir, dentry);
1034 out_up_read:
1035         up_read(&fs_info->subvol_sem);
1036 out_dput:
1037         dput(dentry);
1038 out_unlock:
1039         btrfs_inode_unlock(BTRFS_I(dir), 0);
1040         return error;
1041 }
1042
1043 static noinline int btrfs_mksnapshot(const struct path *parent,
1044                                    struct mnt_idmap *idmap,
1045                                    const char *name, int namelen,
1046                                    struct btrfs_root *root,
1047                                    bool readonly,
1048                                    struct btrfs_qgroup_inherit *inherit)
1049 {
1050         int ret;
1051         bool snapshot_force_cow = false;
1052
1053         /*
1054          * Force new buffered writes to reserve space even when NOCOW is
1055          * possible. This is to avoid later writeback (running dealloc) to
1056          * fallback to COW mode and unexpectedly fail with ENOSPC.
1057          */
1058         btrfs_drew_read_lock(&root->snapshot_lock);
1059
1060         ret = btrfs_start_delalloc_snapshot(root, false);
1061         if (ret)
1062                 goto out;
1063
1064         /*
1065          * All previous writes have started writeback in NOCOW mode, so now
1066          * we force future writes to fallback to COW mode during snapshot
1067          * creation.
1068          */
1069         atomic_inc(&root->snapshot_force_cow);
1070         snapshot_force_cow = true;
1071
1072         btrfs_wait_ordered_extents(root, U64_MAX, NULL);
1073
1074         ret = btrfs_mksubvol(parent, idmap, name, namelen,
1075                              root, readonly, inherit);
1076 out:
1077         if (snapshot_force_cow)
1078                 atomic_dec(&root->snapshot_force_cow);
1079         btrfs_drew_read_unlock(&root->snapshot_lock);
1080         return ret;
1081 }
1082
1083 /*
1084  * Try to start exclusive operation @type or cancel it if it's running.
1085  *
1086  * Return:
1087  *   0        - normal mode, newly claimed op started
1088  *  >0        - normal mode, something else is running,
1089  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1090  * ECANCELED  - cancel mode, successful cancel
1091  * ENOTCONN   - cancel mode, operation not running anymore
1092  */
1093 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1094                         enum btrfs_exclusive_operation type, bool cancel)
1095 {
1096         if (!cancel) {
1097                 /* Start normal op */
1098                 if (!btrfs_exclop_start(fs_info, type))
1099                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1100                 /* Exclusive operation is now claimed */
1101                 return 0;
1102         }
1103
1104         /* Cancel running op */
1105         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1106                 /*
1107                  * This blocks any exclop finish from setting it to NONE, so we
1108                  * request cancellation. Either it runs and we will wait for it,
1109                  * or it has finished and no waiting will happen.
1110                  */
1111                 atomic_inc(&fs_info->reloc_cancel_req);
1112                 btrfs_exclop_start_unlock(fs_info);
1113
1114                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1115                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1116                                     TASK_INTERRUPTIBLE);
1117
1118                 return -ECANCELED;
1119         }
1120
1121         /* Something else is running or none */
1122         return -ENOTCONN;
1123 }
1124
1125 static noinline int btrfs_ioctl_resize(struct file *file,
1126                                         void __user *arg)
1127 {
1128         BTRFS_DEV_LOOKUP_ARGS(args);
1129         struct inode *inode = file_inode(file);
1130         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1131         u64 new_size;
1132         u64 old_size;
1133         u64 devid = 1;
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135         struct btrfs_ioctl_vol_args *vol_args;
1136         struct btrfs_trans_handle *trans;
1137         struct btrfs_device *device = NULL;
1138         char *sizestr;
1139         char *retptr;
1140         char *devstr = NULL;
1141         int ret = 0;
1142         int mod = 0;
1143         bool cancel;
1144
1145         if (!capable(CAP_SYS_ADMIN))
1146                 return -EPERM;
1147
1148         ret = mnt_want_write_file(file);
1149         if (ret)
1150                 return ret;
1151
1152         /*
1153          * Read the arguments before checking exclusivity to be able to
1154          * distinguish regular resize and cancel
1155          */
1156         vol_args = memdup_user(arg, sizeof(*vol_args));
1157         if (IS_ERR(vol_args)) {
1158                 ret = PTR_ERR(vol_args);
1159                 goto out_drop;
1160         }
1161         ret = btrfs_check_ioctl_vol_args_path(vol_args);
1162         if (ret < 0)
1163                 goto out_free;
1164
1165         sizestr = vol_args->name;
1166         cancel = (strcmp("cancel", sizestr) == 0);
1167         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1168         if (ret)
1169                 goto out_free;
1170         /* Exclusive operation is now claimed */
1171
1172         devstr = strchr(sizestr, ':');
1173         if (devstr) {
1174                 sizestr = devstr + 1;
1175                 *devstr = '\0';
1176                 devstr = vol_args->name;
1177                 ret = kstrtoull(devstr, 10, &devid);
1178                 if (ret)
1179                         goto out_finish;
1180                 if (!devid) {
1181                         ret = -EINVAL;
1182                         goto out_finish;
1183                 }
1184                 btrfs_info(fs_info, "resizing devid %llu", devid);
1185         }
1186
1187         args.devid = devid;
1188         device = btrfs_find_device(fs_info->fs_devices, &args);
1189         if (!device) {
1190                 btrfs_info(fs_info, "resizer unable to find device %llu",
1191                            devid);
1192                 ret = -ENODEV;
1193                 goto out_finish;
1194         }
1195
1196         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1197                 btrfs_info(fs_info,
1198                            "resizer unable to apply on readonly device %llu",
1199                        devid);
1200                 ret = -EPERM;
1201                 goto out_finish;
1202         }
1203
1204         if (!strcmp(sizestr, "max"))
1205                 new_size = bdev_nr_bytes(device->bdev);
1206         else {
1207                 if (sizestr[0] == '-') {
1208                         mod = -1;
1209                         sizestr++;
1210                 } else if (sizestr[0] == '+') {
1211                         mod = 1;
1212                         sizestr++;
1213                 }
1214                 new_size = memparse(sizestr, &retptr);
1215                 if (*retptr != '\0' || new_size == 0) {
1216                         ret = -EINVAL;
1217                         goto out_finish;
1218                 }
1219         }
1220
1221         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1222                 ret = -EPERM;
1223                 goto out_finish;
1224         }
1225
1226         old_size = btrfs_device_get_total_bytes(device);
1227
1228         if (mod < 0) {
1229                 if (new_size > old_size) {
1230                         ret = -EINVAL;
1231                         goto out_finish;
1232                 }
1233                 new_size = old_size - new_size;
1234         } else if (mod > 0) {
1235                 if (new_size > ULLONG_MAX - old_size) {
1236                         ret = -ERANGE;
1237                         goto out_finish;
1238                 }
1239                 new_size = old_size + new_size;
1240         }
1241
1242         if (new_size < SZ_256M) {
1243                 ret = -EINVAL;
1244                 goto out_finish;
1245         }
1246         if (new_size > bdev_nr_bytes(device->bdev)) {
1247                 ret = -EFBIG;
1248                 goto out_finish;
1249         }
1250
1251         new_size = round_down(new_size, fs_info->sectorsize);
1252
1253         if (new_size > old_size) {
1254                 trans = btrfs_start_transaction(root, 0);
1255                 if (IS_ERR(trans)) {
1256                         ret = PTR_ERR(trans);
1257                         goto out_finish;
1258                 }
1259                 ret = btrfs_grow_device(trans, device, new_size);
1260                 btrfs_commit_transaction(trans);
1261         } else if (new_size < old_size) {
1262                 ret = btrfs_shrink_device(device, new_size);
1263         } /* equal, nothing need to do */
1264
1265         if (ret == 0 && new_size != old_size)
1266                 btrfs_info_in_rcu(fs_info,
1267                         "resize device %s (devid %llu) from %llu to %llu",
1268                         btrfs_dev_name(device), device->devid,
1269                         old_size, new_size);
1270 out_finish:
1271         btrfs_exclop_finish(fs_info);
1272 out_free:
1273         kfree(vol_args);
1274 out_drop:
1275         mnt_drop_write_file(file);
1276         return ret;
1277 }
1278
1279 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1280                                 struct mnt_idmap *idmap,
1281                                 const char *name, unsigned long fd, int subvol,
1282                                 bool readonly,
1283                                 struct btrfs_qgroup_inherit *inherit)
1284 {
1285         int namelen;
1286         int ret = 0;
1287
1288         if (!S_ISDIR(file_inode(file)->i_mode))
1289                 return -ENOTDIR;
1290
1291         ret = mnt_want_write_file(file);
1292         if (ret)
1293                 goto out;
1294
1295         namelen = strlen(name);
1296         if (strchr(name, '/')) {
1297                 ret = -EINVAL;
1298                 goto out_drop_write;
1299         }
1300
1301         if (name[0] == '.' &&
1302            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1303                 ret = -EEXIST;
1304                 goto out_drop_write;
1305         }
1306
1307         if (subvol) {
1308                 ret = btrfs_mksubvol(&file->f_path, idmap, name,
1309                                      namelen, NULL, readonly, inherit);
1310         } else {
1311                 struct fd src = fdget(fd);
1312                 struct inode *src_inode;
1313                 if (!fd_file(src)) {
1314                         ret = -EINVAL;
1315                         goto out_drop_write;
1316                 }
1317
1318                 src_inode = file_inode(fd_file(src));
1319                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1320                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1321                                    "Snapshot src from another FS");
1322                         ret = -EXDEV;
1323                 } else if (!inode_owner_or_capable(idmap, src_inode)) {
1324                         /*
1325                          * Subvolume creation is not restricted, but snapshots
1326                          * are limited to own subvolumes only
1327                          */
1328                         ret = -EPERM;
1329                 } else if (btrfs_ino(BTRFS_I(src_inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1330                         /*
1331                          * Snapshots must be made with the src_inode referring
1332                          * to the subvolume inode, otherwise the permission
1333                          * checking above is useless because we may have
1334                          * permission on a lower directory but not the subvol
1335                          * itself.
1336                          */
1337                         ret = -EINVAL;
1338                 } else {
1339                         ret = btrfs_mksnapshot(&file->f_path, idmap,
1340                                                name, namelen,
1341                                                BTRFS_I(src_inode)->root,
1342                                                readonly, inherit);
1343                 }
1344                 fdput(src);
1345         }
1346 out_drop_write:
1347         mnt_drop_write_file(file);
1348 out:
1349         return ret;
1350 }
1351
1352 static noinline int btrfs_ioctl_snap_create(struct file *file,
1353                                             void __user *arg, int subvol)
1354 {
1355         struct btrfs_ioctl_vol_args *vol_args;
1356         int ret;
1357
1358         if (!S_ISDIR(file_inode(file)->i_mode))
1359                 return -ENOTDIR;
1360
1361         vol_args = memdup_user(arg, sizeof(*vol_args));
1362         if (IS_ERR(vol_args))
1363                 return PTR_ERR(vol_args);
1364         ret = btrfs_check_ioctl_vol_args_path(vol_args);
1365         if (ret < 0)
1366                 goto out;
1367
1368         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1369                                         vol_args->name, vol_args->fd, subvol,
1370                                         false, NULL);
1371
1372 out:
1373         kfree(vol_args);
1374         return ret;
1375 }
1376
1377 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1378                                                void __user *arg, int subvol)
1379 {
1380         struct btrfs_ioctl_vol_args_v2 *vol_args;
1381         int ret;
1382         bool readonly = false;
1383         struct btrfs_qgroup_inherit *inherit = NULL;
1384
1385         if (!S_ISDIR(file_inode(file)->i_mode))
1386                 return -ENOTDIR;
1387
1388         vol_args = memdup_user(arg, sizeof(*vol_args));
1389         if (IS_ERR(vol_args))
1390                 return PTR_ERR(vol_args);
1391         ret = btrfs_check_ioctl_vol_args2_subvol_name(vol_args);
1392         if (ret < 0)
1393                 goto free_args;
1394
1395         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1396                 ret = -EOPNOTSUPP;
1397                 goto free_args;
1398         }
1399
1400         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1401                 readonly = true;
1402         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1403                 struct btrfs_fs_info *fs_info = inode_to_fs_info(file_inode(file));
1404
1405                 if (vol_args->size < sizeof(*inherit) ||
1406                     vol_args->size > PAGE_SIZE) {
1407                         ret = -EINVAL;
1408                         goto free_args;
1409                 }
1410                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1411                 if (IS_ERR(inherit)) {
1412                         ret = PTR_ERR(inherit);
1413                         goto free_args;
1414                 }
1415
1416                 ret = btrfs_qgroup_check_inherit(fs_info, inherit, vol_args->size);
1417                 if (ret < 0)
1418                         goto free_inherit;
1419         }
1420
1421         ret = __btrfs_ioctl_snap_create(file, file_mnt_idmap(file),
1422                                         vol_args->name, vol_args->fd, subvol,
1423                                         readonly, inherit);
1424         if (ret)
1425                 goto free_inherit;
1426 free_inherit:
1427         kfree(inherit);
1428 free_args:
1429         kfree(vol_args);
1430         return ret;
1431 }
1432
1433 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
1434                                                 void __user *arg)
1435 {
1436         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1437         struct btrfs_root *root = BTRFS_I(inode)->root;
1438         int ret = 0;
1439         u64 flags = 0;
1440
1441         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1442                 return -EINVAL;
1443
1444         down_read(&fs_info->subvol_sem);
1445         if (btrfs_root_readonly(root))
1446                 flags |= BTRFS_SUBVOL_RDONLY;
1447         up_read(&fs_info->subvol_sem);
1448
1449         if (copy_to_user(arg, &flags, sizeof(flags)))
1450                 ret = -EFAULT;
1451
1452         return ret;
1453 }
1454
1455 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1456                                               void __user *arg)
1457 {
1458         struct inode *inode = file_inode(file);
1459         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1460         struct btrfs_root *root = BTRFS_I(inode)->root;
1461         struct btrfs_trans_handle *trans;
1462         u64 root_flags;
1463         u64 flags;
1464         int ret = 0;
1465
1466         if (!inode_owner_or_capable(file_mnt_idmap(file), inode))
1467                 return -EPERM;
1468
1469         ret = mnt_want_write_file(file);
1470         if (ret)
1471                 goto out;
1472
1473         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1474                 ret = -EINVAL;
1475                 goto out_drop_write;
1476         }
1477
1478         if (copy_from_user(&flags, arg, sizeof(flags))) {
1479                 ret = -EFAULT;
1480                 goto out_drop_write;
1481         }
1482
1483         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1484                 ret = -EOPNOTSUPP;
1485                 goto out_drop_write;
1486         }
1487
1488         down_write(&fs_info->subvol_sem);
1489
1490         /* nothing to do */
1491         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1492                 goto out_drop_sem;
1493
1494         root_flags = btrfs_root_flags(&root->root_item);
1495         if (flags & BTRFS_SUBVOL_RDONLY) {
1496                 btrfs_set_root_flags(&root->root_item,
1497                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1498         } else {
1499                 /*
1500                  * Block RO -> RW transition if this subvolume is involved in
1501                  * send
1502                  */
1503                 spin_lock(&root->root_item_lock);
1504                 if (root->send_in_progress == 0) {
1505                         btrfs_set_root_flags(&root->root_item,
1506                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1507                         spin_unlock(&root->root_item_lock);
1508                 } else {
1509                         spin_unlock(&root->root_item_lock);
1510                         btrfs_warn(fs_info,
1511                                    "Attempt to set subvolume %llu read-write during send",
1512                                    btrfs_root_id(root));
1513                         ret = -EPERM;
1514                         goto out_drop_sem;
1515                 }
1516         }
1517
1518         trans = btrfs_start_transaction(root, 1);
1519         if (IS_ERR(trans)) {
1520                 ret = PTR_ERR(trans);
1521                 goto out_reset;
1522         }
1523
1524         ret = btrfs_update_root(trans, fs_info->tree_root,
1525                                 &root->root_key, &root->root_item);
1526         if (ret < 0) {
1527                 btrfs_end_transaction(trans);
1528                 goto out_reset;
1529         }
1530
1531         ret = btrfs_commit_transaction(trans);
1532
1533 out_reset:
1534         if (ret)
1535                 btrfs_set_root_flags(&root->root_item, root_flags);
1536 out_drop_sem:
1537         up_write(&fs_info->subvol_sem);
1538 out_drop_write:
1539         mnt_drop_write_file(file);
1540 out:
1541         return ret;
1542 }
1543
1544 static noinline int key_in_sk(struct btrfs_key *key,
1545                               struct btrfs_ioctl_search_key *sk)
1546 {
1547         struct btrfs_key test;
1548         int ret;
1549
1550         test.objectid = sk->min_objectid;
1551         test.type = sk->min_type;
1552         test.offset = sk->min_offset;
1553
1554         ret = btrfs_comp_cpu_keys(key, &test);
1555         if (ret < 0)
1556                 return 0;
1557
1558         test.objectid = sk->max_objectid;
1559         test.type = sk->max_type;
1560         test.offset = sk->max_offset;
1561
1562         ret = btrfs_comp_cpu_keys(key, &test);
1563         if (ret > 0)
1564                 return 0;
1565         return 1;
1566 }
1567
1568 static noinline int copy_to_sk(struct btrfs_path *path,
1569                                struct btrfs_key *key,
1570                                struct btrfs_ioctl_search_key *sk,
1571                                u64 *buf_size,
1572                                char __user *ubuf,
1573                                unsigned long *sk_offset,
1574                                int *num_found)
1575 {
1576         u64 found_transid;
1577         struct extent_buffer *leaf;
1578         struct btrfs_ioctl_search_header sh;
1579         struct btrfs_key test;
1580         unsigned long item_off;
1581         unsigned long item_len;
1582         int nritems;
1583         int i;
1584         int slot;
1585         int ret = 0;
1586
1587         leaf = path->nodes[0];
1588         slot = path->slots[0];
1589         nritems = btrfs_header_nritems(leaf);
1590
1591         if (btrfs_header_generation(leaf) > sk->max_transid) {
1592                 i = nritems;
1593                 goto advance_key;
1594         }
1595         found_transid = btrfs_header_generation(leaf);
1596
1597         for (i = slot; i < nritems; i++) {
1598                 item_off = btrfs_item_ptr_offset(leaf, i);
1599                 item_len = btrfs_item_size(leaf, i);
1600
1601                 btrfs_item_key_to_cpu(leaf, key, i);
1602                 if (!key_in_sk(key, sk))
1603                         continue;
1604
1605                 if (sizeof(sh) + item_len > *buf_size) {
1606                         if (*num_found) {
1607                                 ret = 1;
1608                                 goto out;
1609                         }
1610
1611                         /*
1612                          * return one empty item back for v1, which does not
1613                          * handle -EOVERFLOW
1614                          */
1615
1616                         *buf_size = sizeof(sh) + item_len;
1617                         item_len = 0;
1618                         ret = -EOVERFLOW;
1619                 }
1620
1621                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
1622                         ret = 1;
1623                         goto out;
1624                 }
1625
1626                 sh.objectid = key->objectid;
1627                 sh.offset = key->offset;
1628                 sh.type = key->type;
1629                 sh.len = item_len;
1630                 sh.transid = found_transid;
1631
1632                 /*
1633                  * Copy search result header. If we fault then loop again so we
1634                  * can fault in the pages and -EFAULT there if there's a
1635                  * problem. Otherwise we'll fault and then copy the buffer in
1636                  * properly this next time through
1637                  */
1638                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
1639                         ret = 0;
1640                         goto out;
1641                 }
1642
1643                 *sk_offset += sizeof(sh);
1644
1645                 if (item_len) {
1646                         char __user *up = ubuf + *sk_offset;
1647                         /*
1648                          * Copy the item, same behavior as above, but reset the
1649                          * * sk_offset so we copy the full thing again.
1650                          */
1651                         if (read_extent_buffer_to_user_nofault(leaf, up,
1652                                                 item_off, item_len)) {
1653                                 ret = 0;
1654                                 *sk_offset -= sizeof(sh);
1655                                 goto out;
1656                         }
1657
1658                         *sk_offset += item_len;
1659                 }
1660                 (*num_found)++;
1661
1662                 if (ret) /* -EOVERFLOW from above */
1663                         goto out;
1664
1665                 if (*num_found >= sk->nr_items) {
1666                         ret = 1;
1667                         goto out;
1668                 }
1669         }
1670 advance_key:
1671         ret = 0;
1672         test.objectid = sk->max_objectid;
1673         test.type = sk->max_type;
1674         test.offset = sk->max_offset;
1675         if (btrfs_comp_cpu_keys(key, &test) >= 0)
1676                 ret = 1;
1677         else if (key->offset < (u64)-1)
1678                 key->offset++;
1679         else if (key->type < (u8)-1) {
1680                 key->offset = 0;
1681                 key->type++;
1682         } else if (key->objectid < (u64)-1) {
1683                 key->offset = 0;
1684                 key->type = 0;
1685                 key->objectid++;
1686         } else
1687                 ret = 1;
1688 out:
1689         /*
1690          *  0: all items from this leaf copied, continue with next
1691          *  1: * more items can be copied, but unused buffer is too small
1692          *     * all items were found
1693          *     Either way, it will stops the loop which iterates to the next
1694          *     leaf
1695          *  -EOVERFLOW: item was to large for buffer
1696          *  -EFAULT: could not copy extent buffer back to userspace
1697          */
1698         return ret;
1699 }
1700
1701 static noinline int search_ioctl(struct inode *inode,
1702                                  struct btrfs_ioctl_search_key *sk,
1703                                  u64 *buf_size,
1704                                  char __user *ubuf)
1705 {
1706         struct btrfs_fs_info *info = inode_to_fs_info(inode);
1707         struct btrfs_root *root;
1708         struct btrfs_key key;
1709         struct btrfs_path *path;
1710         int ret;
1711         int num_found = 0;
1712         unsigned long sk_offset = 0;
1713
1714         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
1715                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
1716                 return -EOVERFLOW;
1717         }
1718
1719         path = btrfs_alloc_path();
1720         if (!path)
1721                 return -ENOMEM;
1722
1723         if (sk->tree_id == 0) {
1724                 /* search the root of the inode that was passed */
1725                 root = btrfs_grab_root(BTRFS_I(inode)->root);
1726         } else {
1727                 root = btrfs_get_fs_root(info, sk->tree_id, true);
1728                 if (IS_ERR(root)) {
1729                         btrfs_free_path(path);
1730                         return PTR_ERR(root);
1731                 }
1732         }
1733
1734         key.objectid = sk->min_objectid;
1735         key.type = sk->min_type;
1736         key.offset = sk->min_offset;
1737
1738         while (1) {
1739                 ret = -EFAULT;
1740                 /*
1741                  * Ensure that the whole user buffer is faulted in at sub-page
1742                  * granularity, otherwise the loop may live-lock.
1743                  */
1744                 if (fault_in_subpage_writeable(ubuf + sk_offset,
1745                                                *buf_size - sk_offset))
1746                         break;
1747
1748                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
1749                 if (ret != 0) {
1750                         if (ret > 0)
1751                                 ret = 0;
1752                         goto err;
1753                 }
1754                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
1755                                  &sk_offset, &num_found);
1756                 btrfs_release_path(path);
1757                 if (ret)
1758                         break;
1759
1760         }
1761         if (ret > 0)
1762                 ret = 0;
1763 err:
1764         sk->nr_items = num_found;
1765         btrfs_put_root(root);
1766         btrfs_free_path(path);
1767         return ret;
1768 }
1769
1770 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
1771                                             void __user *argp)
1772 {
1773         struct btrfs_ioctl_search_args __user *uargs = argp;
1774         struct btrfs_ioctl_search_key sk;
1775         int ret;
1776         u64 buf_size;
1777
1778         if (!capable(CAP_SYS_ADMIN))
1779                 return -EPERM;
1780
1781         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
1782                 return -EFAULT;
1783
1784         buf_size = sizeof(uargs->buf);
1785
1786         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
1787
1788         /*
1789          * In the origin implementation an overflow is handled by returning a
1790          * search header with a len of zero, so reset ret.
1791          */
1792         if (ret == -EOVERFLOW)
1793                 ret = 0;
1794
1795         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
1796                 ret = -EFAULT;
1797         return ret;
1798 }
1799
1800 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
1801                                                void __user *argp)
1802 {
1803         struct btrfs_ioctl_search_args_v2 __user *uarg = argp;
1804         struct btrfs_ioctl_search_args_v2 args;
1805         int ret;
1806         u64 buf_size;
1807         const u64 buf_limit = SZ_16M;
1808
1809         if (!capable(CAP_SYS_ADMIN))
1810                 return -EPERM;
1811
1812         /* copy search header and buffer size */
1813         if (copy_from_user(&args, uarg, sizeof(args)))
1814                 return -EFAULT;
1815
1816         buf_size = args.buf_size;
1817
1818         /* limit result size to 16MB */
1819         if (buf_size > buf_limit)
1820                 buf_size = buf_limit;
1821
1822         ret = search_ioctl(inode, &args.key, &buf_size,
1823                            (char __user *)(&uarg->buf[0]));
1824         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
1825                 ret = -EFAULT;
1826         else if (ret == -EOVERFLOW &&
1827                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
1828                 ret = -EFAULT;
1829
1830         return ret;
1831 }
1832
1833 /*
1834  * Search INODE_REFs to identify path name of 'dirid' directory
1835  * in a 'tree_id' tree. and sets path name to 'name'.
1836  */
1837 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1838                                 u64 tree_id, u64 dirid, char *name)
1839 {
1840         struct btrfs_root *root;
1841         struct btrfs_key key;
1842         char *ptr;
1843         int ret = -1;
1844         int slot;
1845         int len;
1846         int total_len = 0;
1847         struct btrfs_inode_ref *iref;
1848         struct extent_buffer *l;
1849         struct btrfs_path *path;
1850
1851         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1852                 name[0]='\0';
1853                 return 0;
1854         }
1855
1856         path = btrfs_alloc_path();
1857         if (!path)
1858                 return -ENOMEM;
1859
1860         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
1861
1862         root = btrfs_get_fs_root(info, tree_id, true);
1863         if (IS_ERR(root)) {
1864                 ret = PTR_ERR(root);
1865                 root = NULL;
1866                 goto out;
1867         }
1868
1869         key.objectid = dirid;
1870         key.type = BTRFS_INODE_REF_KEY;
1871         key.offset = (u64)-1;
1872
1873         while (1) {
1874                 ret = btrfs_search_backwards(root, &key, path);
1875                 if (ret < 0)
1876                         goto out;
1877                 else if (ret > 0) {
1878                         ret = -ENOENT;
1879                         goto out;
1880                 }
1881
1882                 l = path->nodes[0];
1883                 slot = path->slots[0];
1884
1885                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1886                 len = btrfs_inode_ref_name_len(l, iref);
1887                 ptr -= len + 1;
1888                 total_len += len + 1;
1889                 if (ptr < name) {
1890                         ret = -ENAMETOOLONG;
1891                         goto out;
1892                 }
1893
1894                 *(ptr + len) = '/';
1895                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
1896
1897                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1898                         break;
1899
1900                 btrfs_release_path(path);
1901                 key.objectid = key.offset;
1902                 key.offset = (u64)-1;
1903                 dirid = key.objectid;
1904         }
1905         memmove(name, ptr, total_len);
1906         name[total_len] = '\0';
1907         ret = 0;
1908 out:
1909         btrfs_put_root(root);
1910         btrfs_free_path(path);
1911         return ret;
1912 }
1913
1914 static int btrfs_search_path_in_tree_user(struct mnt_idmap *idmap,
1915                                 struct inode *inode,
1916                                 struct btrfs_ioctl_ino_lookup_user_args *args)
1917 {
1918         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1919         u64 upper_limit = btrfs_ino(BTRFS_I(inode));
1920         u64 treeid = btrfs_root_id(BTRFS_I(inode)->root);
1921         u64 dirid = args->dirid;
1922         unsigned long item_off;
1923         unsigned long item_len;
1924         struct btrfs_inode_ref *iref;
1925         struct btrfs_root_ref *rref;
1926         struct btrfs_root *root = NULL;
1927         struct btrfs_path *path;
1928         struct btrfs_key key, key2;
1929         struct extent_buffer *leaf;
1930         struct inode *temp_inode;
1931         char *ptr;
1932         int slot;
1933         int len;
1934         int total_len = 0;
1935         int ret;
1936
1937         path = btrfs_alloc_path();
1938         if (!path)
1939                 return -ENOMEM;
1940
1941         /*
1942          * If the bottom subvolume does not exist directly under upper_limit,
1943          * construct the path in from the bottom up.
1944          */
1945         if (dirid != upper_limit) {
1946                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
1947
1948                 root = btrfs_get_fs_root(fs_info, treeid, true);
1949                 if (IS_ERR(root)) {
1950                         ret = PTR_ERR(root);
1951                         goto out;
1952                 }
1953
1954                 key.objectid = dirid;
1955                 key.type = BTRFS_INODE_REF_KEY;
1956                 key.offset = (u64)-1;
1957                 while (1) {
1958                         ret = btrfs_search_backwards(root, &key, path);
1959                         if (ret < 0)
1960                                 goto out_put;
1961                         else if (ret > 0) {
1962                                 ret = -ENOENT;
1963                                 goto out_put;
1964                         }
1965
1966                         leaf = path->nodes[0];
1967                         slot = path->slots[0];
1968
1969                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
1970                         len = btrfs_inode_ref_name_len(leaf, iref);
1971                         ptr -= len + 1;
1972                         total_len += len + 1;
1973                         if (ptr < args->path) {
1974                                 ret = -ENAMETOOLONG;
1975                                 goto out_put;
1976                         }
1977
1978                         *(ptr + len) = '/';
1979                         read_extent_buffer(leaf, ptr,
1980                                         (unsigned long)(iref + 1), len);
1981
1982                         /* Check the read+exec permission of this directory */
1983                         ret = btrfs_previous_item(root, path, dirid,
1984                                                   BTRFS_INODE_ITEM_KEY);
1985                         if (ret < 0) {
1986                                 goto out_put;
1987                         } else if (ret > 0) {
1988                                 ret = -ENOENT;
1989                                 goto out_put;
1990                         }
1991
1992                         leaf = path->nodes[0];
1993                         slot = path->slots[0];
1994                         btrfs_item_key_to_cpu(leaf, &key2, slot);
1995                         if (key2.objectid != dirid) {
1996                                 ret = -ENOENT;
1997                                 goto out_put;
1998                         }
1999
2000                         /*
2001                          * We don't need the path anymore, so release it and
2002                          * avoid deadlocks and lockdep warnings in case
2003                          * btrfs_iget() needs to lookup the inode from its root
2004                          * btree and lock the same leaf.
2005                          */
2006                         btrfs_release_path(path);
2007                         temp_inode = btrfs_iget(key2.objectid, root);
2008                         if (IS_ERR(temp_inode)) {
2009                                 ret = PTR_ERR(temp_inode);
2010                                 goto out_put;
2011                         }
2012                         ret = inode_permission(idmap, temp_inode,
2013                                                MAY_READ | MAY_EXEC);
2014                         iput(temp_inode);
2015                         if (ret) {
2016                                 ret = -EACCES;
2017                                 goto out_put;
2018                         }
2019
2020                         if (key.offset == upper_limit)
2021                                 break;
2022                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2023                                 ret = -EACCES;
2024                                 goto out_put;
2025                         }
2026
2027                         key.objectid = key.offset;
2028                         key.offset = (u64)-1;
2029                         dirid = key.objectid;
2030                 }
2031
2032                 memmove(args->path, ptr, total_len);
2033                 args->path[total_len] = '\0';
2034                 btrfs_put_root(root);
2035                 root = NULL;
2036                 btrfs_release_path(path);
2037         }
2038
2039         /* Get the bottom subvolume's name from ROOT_REF */
2040         key.objectid = treeid;
2041         key.type = BTRFS_ROOT_REF_KEY;
2042         key.offset = args->treeid;
2043         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2044         if (ret < 0) {
2045                 goto out;
2046         } else if (ret > 0) {
2047                 ret = -ENOENT;
2048                 goto out;
2049         }
2050
2051         leaf = path->nodes[0];
2052         slot = path->slots[0];
2053         btrfs_item_key_to_cpu(leaf, &key, slot);
2054
2055         item_off = btrfs_item_ptr_offset(leaf, slot);
2056         item_len = btrfs_item_size(leaf, slot);
2057         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2058         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2059         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2060                 ret = -EINVAL;
2061                 goto out;
2062         }
2063
2064         /* Copy subvolume's name */
2065         item_off += sizeof(struct btrfs_root_ref);
2066         item_len -= sizeof(struct btrfs_root_ref);
2067         read_extent_buffer(leaf, args->name, item_off, item_len);
2068         args->name[item_len] = 0;
2069
2070 out_put:
2071         btrfs_put_root(root);
2072 out:
2073         btrfs_free_path(path);
2074         return ret;
2075 }
2076
2077 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2078                                            void __user *argp)
2079 {
2080         struct btrfs_ioctl_ino_lookup_args *args;
2081         int ret = 0;
2082
2083         args = memdup_user(argp, sizeof(*args));
2084         if (IS_ERR(args))
2085                 return PTR_ERR(args);
2086
2087         /*
2088          * Unprivileged query to obtain the containing subvolume root id. The
2089          * path is reset so it's consistent with btrfs_search_path_in_tree.
2090          */
2091         if (args->treeid == 0)
2092                 args->treeid = btrfs_root_id(root);
2093
2094         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2095                 args->name[0] = 0;
2096                 goto out;
2097         }
2098
2099         if (!capable(CAP_SYS_ADMIN)) {
2100                 ret = -EPERM;
2101                 goto out;
2102         }
2103
2104         ret = btrfs_search_path_in_tree(root->fs_info,
2105                                         args->treeid, args->objectid,
2106                                         args->name);
2107
2108 out:
2109         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2110                 ret = -EFAULT;
2111
2112         kfree(args);
2113         return ret;
2114 }
2115
2116 /*
2117  * Version of ino_lookup ioctl (unprivileged)
2118  *
2119  * The main differences from ino_lookup ioctl are:
2120  *
2121  *   1. Read + Exec permission will be checked using inode_permission() during
2122  *      path construction. -EACCES will be returned in case of failure.
2123  *   2. Path construction will be stopped at the inode number which corresponds
2124  *      to the fd with which this ioctl is called. If constructed path does not
2125  *      exist under fd's inode, -EACCES will be returned.
2126  *   3. The name of bottom subvolume is also searched and filled.
2127  */
2128 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2129 {
2130         struct btrfs_ioctl_ino_lookup_user_args *args;
2131         struct inode *inode;
2132         int ret;
2133
2134         args = memdup_user(argp, sizeof(*args));
2135         if (IS_ERR(args))
2136                 return PTR_ERR(args);
2137
2138         inode = file_inode(file);
2139
2140         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2141             btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2142                 /*
2143                  * The subvolume does not exist under fd with which this is
2144                  * called
2145                  */
2146                 kfree(args);
2147                 return -EACCES;
2148         }
2149
2150         ret = btrfs_search_path_in_tree_user(file_mnt_idmap(file), inode, args);
2151
2152         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2153                 ret = -EFAULT;
2154
2155         kfree(args);
2156         return ret;
2157 }
2158
2159 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2160 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
2161 {
2162         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2163         struct btrfs_fs_info *fs_info;
2164         struct btrfs_root *root;
2165         struct btrfs_path *path;
2166         struct btrfs_key key;
2167         struct btrfs_root_item *root_item;
2168         struct btrfs_root_ref *rref;
2169         struct extent_buffer *leaf;
2170         unsigned long item_off;
2171         unsigned long item_len;
2172         int slot;
2173         int ret = 0;
2174
2175         path = btrfs_alloc_path();
2176         if (!path)
2177                 return -ENOMEM;
2178
2179         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2180         if (!subvol_info) {
2181                 btrfs_free_path(path);
2182                 return -ENOMEM;
2183         }
2184
2185         fs_info = BTRFS_I(inode)->root->fs_info;
2186
2187         /* Get root_item of inode's subvolume */
2188         key.objectid = btrfs_root_id(BTRFS_I(inode)->root);
2189         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2190         if (IS_ERR(root)) {
2191                 ret = PTR_ERR(root);
2192                 goto out_free;
2193         }
2194         root_item = &root->root_item;
2195
2196         subvol_info->treeid = key.objectid;
2197
2198         subvol_info->generation = btrfs_root_generation(root_item);
2199         subvol_info->flags = btrfs_root_flags(root_item);
2200
2201         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2202         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2203                                                     BTRFS_UUID_SIZE);
2204         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2205                                                     BTRFS_UUID_SIZE);
2206
2207         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2208         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2209         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2210
2211         subvol_info->otransid = btrfs_root_otransid(root_item);
2212         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2213         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2214
2215         subvol_info->stransid = btrfs_root_stransid(root_item);
2216         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2217         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2218
2219         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2220         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2221         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2222
2223         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2224                 /* Search root tree for ROOT_BACKREF of this subvolume */
2225                 key.type = BTRFS_ROOT_BACKREF_KEY;
2226                 key.offset = 0;
2227                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2228                 if (ret < 0) {
2229                         goto out;
2230                 } else if (path->slots[0] >=
2231                            btrfs_header_nritems(path->nodes[0])) {
2232                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2233                         if (ret < 0) {
2234                                 goto out;
2235                         } else if (ret > 0) {
2236                                 ret = -EUCLEAN;
2237                                 goto out;
2238                         }
2239                 }
2240
2241                 leaf = path->nodes[0];
2242                 slot = path->slots[0];
2243                 btrfs_item_key_to_cpu(leaf, &key, slot);
2244                 if (key.objectid == subvol_info->treeid &&
2245                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2246                         subvol_info->parent_id = key.offset;
2247
2248                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2249                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2250
2251                         item_off = btrfs_item_ptr_offset(leaf, slot)
2252                                         + sizeof(struct btrfs_root_ref);
2253                         item_len = btrfs_item_size(leaf, slot)
2254                                         - sizeof(struct btrfs_root_ref);
2255                         read_extent_buffer(leaf, subvol_info->name,
2256                                            item_off, item_len);
2257                 } else {
2258                         ret = -ENOENT;
2259                         goto out;
2260                 }
2261         }
2262
2263         btrfs_free_path(path);
2264         path = NULL;
2265         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2266                 ret = -EFAULT;
2267
2268 out:
2269         btrfs_put_root(root);
2270 out_free:
2271         btrfs_free_path(path);
2272         kfree(subvol_info);
2273         return ret;
2274 }
2275
2276 /*
2277  * Return ROOT_REF information of the subvolume containing this inode
2278  * except the subvolume name.
2279  */
2280 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
2281                                           void __user *argp)
2282 {
2283         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2284         struct btrfs_root_ref *rref;
2285         struct btrfs_path *path;
2286         struct btrfs_key key;
2287         struct extent_buffer *leaf;
2288         u64 objectid;
2289         int slot;
2290         int ret;
2291         u8 found;
2292
2293         path = btrfs_alloc_path();
2294         if (!path)
2295                 return -ENOMEM;
2296
2297         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2298         if (IS_ERR(rootrefs)) {
2299                 btrfs_free_path(path);
2300                 return PTR_ERR(rootrefs);
2301         }
2302
2303         objectid = btrfs_root_id(root);
2304         key.objectid = objectid;
2305         key.type = BTRFS_ROOT_REF_KEY;
2306         key.offset = rootrefs->min_treeid;
2307         found = 0;
2308
2309         root = root->fs_info->tree_root;
2310         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2311         if (ret < 0) {
2312                 goto out;
2313         } else if (path->slots[0] >=
2314                    btrfs_header_nritems(path->nodes[0])) {
2315                 ret = btrfs_next_leaf(root, path);
2316                 if (ret < 0) {
2317                         goto out;
2318                 } else if (ret > 0) {
2319                         ret = -EUCLEAN;
2320                         goto out;
2321                 }
2322         }
2323         while (1) {
2324                 leaf = path->nodes[0];
2325                 slot = path->slots[0];
2326
2327                 btrfs_item_key_to_cpu(leaf, &key, slot);
2328                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2329                         ret = 0;
2330                         goto out;
2331                 }
2332
2333                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2334                         ret = -EOVERFLOW;
2335                         goto out;
2336                 }
2337
2338                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2339                 rootrefs->rootref[found].treeid = key.offset;
2340                 rootrefs->rootref[found].dirid =
2341                                   btrfs_root_ref_dirid(leaf, rref);
2342                 found++;
2343
2344                 ret = btrfs_next_item(root, path);
2345                 if (ret < 0) {
2346                         goto out;
2347                 } else if (ret > 0) {
2348                         ret = -EUCLEAN;
2349                         goto out;
2350                 }
2351         }
2352
2353 out:
2354         btrfs_free_path(path);
2355
2356         if (!ret || ret == -EOVERFLOW) {
2357                 rootrefs->num_items = found;
2358                 /* update min_treeid for next search */
2359                 if (found)
2360                         rootrefs->min_treeid =
2361                                 rootrefs->rootref[found - 1].treeid + 1;
2362                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2363                         ret = -EFAULT;
2364         }
2365
2366         kfree(rootrefs);
2367
2368         return ret;
2369 }
2370
2371 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2372                                              void __user *arg,
2373                                              bool destroy_v2)
2374 {
2375         struct dentry *parent = file->f_path.dentry;
2376         struct dentry *dentry;
2377         struct inode *dir = d_inode(parent);
2378         struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
2379         struct inode *inode;
2380         struct btrfs_root *root = BTRFS_I(dir)->root;
2381         struct btrfs_root *dest = NULL;
2382         struct btrfs_ioctl_vol_args *vol_args = NULL;
2383         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2384         struct mnt_idmap *idmap = file_mnt_idmap(file);
2385         char *subvol_name, *subvol_name_ptr = NULL;
2386         int subvol_namelen;
2387         int ret = 0;
2388         bool destroy_parent = false;
2389
2390         /* We don't support snapshots with extent tree v2 yet. */
2391         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2392                 btrfs_err(fs_info,
2393                           "extent tree v2 doesn't support snapshot deletion yet");
2394                 return -EOPNOTSUPP;
2395         }
2396
2397         if (destroy_v2) {
2398                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2399                 if (IS_ERR(vol_args2))
2400                         return PTR_ERR(vol_args2);
2401
2402                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2403                         ret = -EOPNOTSUPP;
2404                         goto out;
2405                 }
2406
2407                 /*
2408                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2409                  * name, same as v1 currently does.
2410                  */
2411                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2412                         ret = btrfs_check_ioctl_vol_args2_subvol_name(vol_args2);
2413                         if (ret < 0)
2414                                 goto out;
2415                         subvol_name = vol_args2->name;
2416
2417                         ret = mnt_want_write_file(file);
2418                         if (ret)
2419                                 goto out;
2420                 } else {
2421                         struct inode *old_dir;
2422
2423                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2424                                 ret = -EINVAL;
2425                                 goto out;
2426                         }
2427
2428                         ret = mnt_want_write_file(file);
2429                         if (ret)
2430                                 goto out;
2431
2432                         dentry = btrfs_get_dentry(fs_info->sb,
2433                                         BTRFS_FIRST_FREE_OBJECTID,
2434                                         vol_args2->subvolid, 0);
2435                         if (IS_ERR(dentry)) {
2436                                 ret = PTR_ERR(dentry);
2437                                 goto out_drop_write;
2438                         }
2439
2440                         /*
2441                          * Change the default parent since the subvolume being
2442                          * deleted can be outside of the current mount point.
2443                          */
2444                         parent = btrfs_get_parent(dentry);
2445
2446                         /*
2447                          * At this point dentry->d_name can point to '/' if the
2448                          * subvolume we want to destroy is outsite of the
2449                          * current mount point, so we need to release the
2450                          * current dentry and execute the lookup to return a new
2451                          * one with ->d_name pointing to the
2452                          * <mount point>/subvol_name.
2453                          */
2454                         dput(dentry);
2455                         if (IS_ERR(parent)) {
2456                                 ret = PTR_ERR(parent);
2457                                 goto out_drop_write;
2458                         }
2459                         old_dir = dir;
2460                         dir = d_inode(parent);
2461
2462                         /*
2463                          * If v2 was used with SPEC_BY_ID, a new parent was
2464                          * allocated since the subvolume can be outside of the
2465                          * current mount point. Later on we need to release this
2466                          * new parent dentry.
2467                          */
2468                         destroy_parent = true;
2469
2470                         /*
2471                          * On idmapped mounts, deletion via subvolid is
2472                          * restricted to subvolumes that are immediate
2473                          * ancestors of the inode referenced by the file
2474                          * descriptor in the ioctl. Otherwise the idmapping
2475                          * could potentially be abused to delete subvolumes
2476                          * anywhere in the filesystem the user wouldn't be able
2477                          * to delete without an idmapped mount.
2478                          */
2479                         if (old_dir != dir && idmap != &nop_mnt_idmap) {
2480                                 ret = -EOPNOTSUPP;
2481                                 goto free_parent;
2482                         }
2483
2484                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2485                                                 fs_info, vol_args2->subvolid);
2486                         if (IS_ERR(subvol_name_ptr)) {
2487                                 ret = PTR_ERR(subvol_name_ptr);
2488                                 goto free_parent;
2489                         }
2490                         /* subvol_name_ptr is already nul terminated */
2491                         subvol_name = (char *)kbasename(subvol_name_ptr);
2492                 }
2493         } else {
2494                 vol_args = memdup_user(arg, sizeof(*vol_args));
2495                 if (IS_ERR(vol_args))
2496                         return PTR_ERR(vol_args);
2497
2498                 ret = btrfs_check_ioctl_vol_args_path(vol_args);
2499                 if (ret < 0)
2500                         goto out;
2501
2502                 subvol_name = vol_args->name;
2503
2504                 ret = mnt_want_write_file(file);
2505                 if (ret)
2506                         goto out;
2507         }
2508
2509         subvol_namelen = strlen(subvol_name);
2510
2511         if (strchr(subvol_name, '/') ||
2512             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2513                 ret = -EINVAL;
2514                 goto free_subvol_name;
2515         }
2516
2517         if (!S_ISDIR(dir->i_mode)) {
2518                 ret = -ENOTDIR;
2519                 goto free_subvol_name;
2520         }
2521
2522         ret = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2523         if (ret == -EINTR)
2524                 goto free_subvol_name;
2525         dentry = lookup_one(idmap, subvol_name, parent, subvol_namelen);
2526         if (IS_ERR(dentry)) {
2527                 ret = PTR_ERR(dentry);
2528                 goto out_unlock_dir;
2529         }
2530
2531         if (d_really_is_negative(dentry)) {
2532                 ret = -ENOENT;
2533                 goto out_dput;
2534         }
2535
2536         inode = d_inode(dentry);
2537         dest = BTRFS_I(inode)->root;
2538         if (!capable(CAP_SYS_ADMIN)) {
2539                 /*
2540                  * Regular user.  Only allow this with a special mount
2541                  * option, when the user has write+exec access to the
2542                  * subvol root, and when rmdir(2) would have been
2543                  * allowed.
2544                  *
2545                  * Note that this is _not_ check that the subvol is
2546                  * empty or doesn't contain data that we wouldn't
2547                  * otherwise be able to delete.
2548                  *
2549                  * Users who want to delete empty subvols should try
2550                  * rmdir(2).
2551                  */
2552                 ret = -EPERM;
2553                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2554                         goto out_dput;
2555
2556                 /*
2557                  * Do not allow deletion if the parent dir is the same
2558                  * as the dir to be deleted.  That means the ioctl
2559                  * must be called on the dentry referencing the root
2560                  * of the subvol, not a random directory contained
2561                  * within it.
2562                  */
2563                 ret = -EINVAL;
2564                 if (root == dest)
2565                         goto out_dput;
2566
2567                 ret = inode_permission(idmap, inode, MAY_WRITE | MAY_EXEC);
2568                 if (ret)
2569                         goto out_dput;
2570         }
2571
2572         /* check if subvolume may be deleted by a user */
2573         ret = btrfs_may_delete(idmap, dir, dentry, 1);
2574         if (ret)
2575                 goto out_dput;
2576
2577         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2578                 ret = -EINVAL;
2579                 goto out_dput;
2580         }
2581
2582         btrfs_inode_lock(BTRFS_I(inode), 0);
2583         ret = btrfs_delete_subvolume(BTRFS_I(dir), dentry);
2584         btrfs_inode_unlock(BTRFS_I(inode), 0);
2585         if (!ret)
2586                 d_delete_notify(dir, dentry);
2587
2588 out_dput:
2589         dput(dentry);
2590 out_unlock_dir:
2591         btrfs_inode_unlock(BTRFS_I(dir), 0);
2592 free_subvol_name:
2593         kfree(subvol_name_ptr);
2594 free_parent:
2595         if (destroy_parent)
2596                 dput(parent);
2597 out_drop_write:
2598         mnt_drop_write_file(file);
2599 out:
2600         kfree(vol_args2);
2601         kfree(vol_args);
2602         return ret;
2603 }
2604
2605 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2606 {
2607         struct inode *inode = file_inode(file);
2608         struct btrfs_root *root = BTRFS_I(inode)->root;
2609         struct btrfs_ioctl_defrag_range_args range = {0};
2610         int ret;
2611
2612         ret = mnt_want_write_file(file);
2613         if (ret)
2614                 return ret;
2615
2616         if (btrfs_root_readonly(root)) {
2617                 ret = -EROFS;
2618                 goto out;
2619         }
2620
2621         switch (inode->i_mode & S_IFMT) {
2622         case S_IFDIR:
2623                 if (!capable(CAP_SYS_ADMIN)) {
2624                         ret = -EPERM;
2625                         goto out;
2626                 }
2627                 ret = btrfs_defrag_root(root);
2628                 break;
2629         case S_IFREG:
2630                 /*
2631                  * Note that this does not check the file descriptor for write
2632                  * access. This prevents defragmenting executables that are
2633                  * running and allows defrag on files open in read-only mode.
2634                  */
2635                 if (!capable(CAP_SYS_ADMIN) &&
2636                     inode_permission(&nop_mnt_idmap, inode, MAY_WRITE)) {
2637                         ret = -EPERM;
2638                         goto out;
2639                 }
2640
2641                 if (argp) {
2642                         if (copy_from_user(&range, argp, sizeof(range))) {
2643                                 ret = -EFAULT;
2644                                 goto out;
2645                         }
2646                         if (range.flags & ~BTRFS_DEFRAG_RANGE_FLAGS_SUPP) {
2647                                 ret = -EOPNOTSUPP;
2648                                 goto out;
2649                         }
2650                         /* compression requires us to start the IO */
2651                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2652                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
2653                                 range.extent_thresh = (u32)-1;
2654                         }
2655                 } else {
2656                         /* the rest are all set to zero by kzalloc */
2657                         range.len = (u64)-1;
2658                 }
2659                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
2660                                         &range, BTRFS_OLDEST_GENERATION, 0);
2661                 if (ret > 0)
2662                         ret = 0;
2663                 break;
2664         default:
2665                 ret = -EINVAL;
2666         }
2667 out:
2668         mnt_drop_write_file(file);
2669         return ret;
2670 }
2671
2672 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2673 {
2674         struct btrfs_ioctl_vol_args *vol_args;
2675         bool restore_op = false;
2676         int ret;
2677
2678         if (!capable(CAP_SYS_ADMIN))
2679                 return -EPERM;
2680
2681         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2682                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
2683                 return -EINVAL;
2684         }
2685
2686         if (fs_info->fs_devices->temp_fsid) {
2687                 btrfs_err(fs_info,
2688                           "device add not supported on cloned temp-fsid mount");
2689                 return -EINVAL;
2690         }
2691
2692         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
2693                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
2694                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
2695
2696                 /*
2697                  * We can do the device add because we have a paused balanced,
2698                  * change the exclusive op type and remember we should bring
2699                  * back the paused balance
2700                  */
2701                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
2702                 btrfs_exclop_start_unlock(fs_info);
2703                 restore_op = true;
2704         }
2705
2706         vol_args = memdup_user(arg, sizeof(*vol_args));
2707         if (IS_ERR(vol_args)) {
2708                 ret = PTR_ERR(vol_args);
2709                 goto out;
2710         }
2711
2712         ret = btrfs_check_ioctl_vol_args_path(vol_args);
2713         if (ret < 0)
2714                 goto out_free;
2715
2716         ret = btrfs_init_new_device(fs_info, vol_args->name);
2717
2718         if (!ret)
2719                 btrfs_info(fs_info, "disk added %s", vol_args->name);
2720
2721 out_free:
2722         kfree(vol_args);
2723 out:
2724         if (restore_op)
2725                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
2726         else
2727                 btrfs_exclop_finish(fs_info);
2728         return ret;
2729 }
2730
2731 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
2732 {
2733         BTRFS_DEV_LOOKUP_ARGS(args);
2734         struct inode *inode = file_inode(file);
2735         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2736         struct btrfs_ioctl_vol_args_v2 *vol_args;
2737         struct file *bdev_file = NULL;
2738         int ret;
2739         bool cancel = false;
2740
2741         if (!capable(CAP_SYS_ADMIN))
2742                 return -EPERM;
2743
2744         vol_args = memdup_user(arg, sizeof(*vol_args));
2745         if (IS_ERR(vol_args))
2746                 return PTR_ERR(vol_args);
2747
2748         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
2749                 ret = -EOPNOTSUPP;
2750                 goto out;
2751         }
2752
2753         ret = btrfs_check_ioctl_vol_args2_subvol_name(vol_args);
2754         if (ret < 0)
2755                 goto out;
2756
2757         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
2758                 args.devid = vol_args->devid;
2759         } else if (!strcmp("cancel", vol_args->name)) {
2760                 cancel = true;
2761         } else {
2762                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2763                 if (ret)
2764                         goto out;
2765         }
2766
2767         ret = mnt_want_write_file(file);
2768         if (ret)
2769                 goto out;
2770
2771         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2772                                            cancel);
2773         if (ret)
2774                 goto err_drop;
2775
2776         /* Exclusive operation is now claimed */
2777         ret = btrfs_rm_device(fs_info, &args, &bdev_file);
2778
2779         btrfs_exclop_finish(fs_info);
2780
2781         if (!ret) {
2782                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
2783                         btrfs_info(fs_info, "device deleted: id %llu",
2784                                         vol_args->devid);
2785                 else
2786                         btrfs_info(fs_info, "device deleted: %s",
2787                                         vol_args->name);
2788         }
2789 err_drop:
2790         mnt_drop_write_file(file);
2791         if (bdev_file)
2792                 fput(bdev_file);
2793 out:
2794         btrfs_put_dev_args_from_path(&args);
2795         kfree(vol_args);
2796         return ret;
2797 }
2798
2799 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2800 {
2801         BTRFS_DEV_LOOKUP_ARGS(args);
2802         struct inode *inode = file_inode(file);
2803         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2804         struct btrfs_ioctl_vol_args *vol_args;
2805         struct file *bdev_file = NULL;
2806         int ret;
2807         bool cancel = false;
2808
2809         if (!capable(CAP_SYS_ADMIN))
2810                 return -EPERM;
2811
2812         vol_args = memdup_user(arg, sizeof(*vol_args));
2813         if (IS_ERR(vol_args))
2814                 return PTR_ERR(vol_args);
2815
2816         ret = btrfs_check_ioctl_vol_args_path(vol_args);
2817         if (ret < 0)
2818                 goto out_free;
2819
2820         if (!strcmp("cancel", vol_args->name)) {
2821                 cancel = true;
2822         } else {
2823                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
2824                 if (ret)
2825                         goto out;
2826         }
2827
2828         ret = mnt_want_write_file(file);
2829         if (ret)
2830                 goto out;
2831
2832         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
2833                                            cancel);
2834         if (ret == 0) {
2835                 ret = btrfs_rm_device(fs_info, &args, &bdev_file);
2836                 if (!ret)
2837                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
2838                 btrfs_exclop_finish(fs_info);
2839         }
2840
2841         mnt_drop_write_file(file);
2842         if (bdev_file)
2843                 fput(bdev_file);
2844 out:
2845         btrfs_put_dev_args_from_path(&args);
2846 out_free:
2847         kfree(vol_args);
2848         return ret;
2849 }
2850
2851 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
2852                                 void __user *arg)
2853 {
2854         struct btrfs_ioctl_fs_info_args *fi_args;
2855         struct btrfs_device *device;
2856         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2857         u64 flags_in;
2858         int ret = 0;
2859
2860         fi_args = memdup_user(arg, sizeof(*fi_args));
2861         if (IS_ERR(fi_args))
2862                 return PTR_ERR(fi_args);
2863
2864         flags_in = fi_args->flags;
2865         memset(fi_args, 0, sizeof(*fi_args));
2866
2867         rcu_read_lock();
2868         fi_args->num_devices = fs_devices->num_devices;
2869
2870         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2871                 if (device->devid > fi_args->max_id)
2872                         fi_args->max_id = device->devid;
2873         }
2874         rcu_read_unlock();
2875
2876         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
2877         fi_args->nodesize = fs_info->nodesize;
2878         fi_args->sectorsize = fs_info->sectorsize;
2879         fi_args->clone_alignment = fs_info->sectorsize;
2880
2881         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
2882                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
2883                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
2884                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
2885         }
2886
2887         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
2888                 fi_args->generation = btrfs_get_fs_generation(fs_info);
2889                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
2890         }
2891
2892         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
2893                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
2894                        sizeof(fi_args->metadata_uuid));
2895                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
2896         }
2897
2898         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2899                 ret = -EFAULT;
2900
2901         kfree(fi_args);
2902         return ret;
2903 }
2904
2905 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
2906                                  void __user *arg)
2907 {
2908         BTRFS_DEV_LOOKUP_ARGS(args);
2909         struct btrfs_ioctl_dev_info_args *di_args;
2910         struct btrfs_device *dev;
2911         int ret = 0;
2912
2913         di_args = memdup_user(arg, sizeof(*di_args));
2914         if (IS_ERR(di_args))
2915                 return PTR_ERR(di_args);
2916
2917         args.devid = di_args->devid;
2918         if (!btrfs_is_empty_uuid(di_args->uuid))
2919                 args.uuid = di_args->uuid;
2920
2921         rcu_read_lock();
2922         dev = btrfs_find_device(fs_info->fs_devices, &args);
2923         if (!dev) {
2924                 ret = -ENODEV;
2925                 goto out;
2926         }
2927
2928         di_args->devid = dev->devid;
2929         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
2930         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
2931         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2932         memcpy(di_args->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2933         if (dev->name)
2934                 strscpy(di_args->path, btrfs_dev_name(dev), sizeof(di_args->path));
2935         else
2936                 di_args->path[0] = '\0';
2937
2938 out:
2939         rcu_read_unlock();
2940         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2941                 ret = -EFAULT;
2942
2943         kfree(di_args);
2944         return ret;
2945 }
2946
2947 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2948 {
2949         struct inode *inode = file_inode(file);
2950         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2951         struct btrfs_root *root = BTRFS_I(inode)->root;
2952         struct btrfs_root *new_root;
2953         struct btrfs_dir_item *di;
2954         struct btrfs_trans_handle *trans;
2955         struct btrfs_path *path = NULL;
2956         struct btrfs_disk_key disk_key;
2957         struct fscrypt_str name = FSTR_INIT("default", 7);
2958         u64 objectid = 0;
2959         u64 dir_id;
2960         int ret;
2961
2962         if (!capable(CAP_SYS_ADMIN))
2963                 return -EPERM;
2964
2965         ret = mnt_want_write_file(file);
2966         if (ret)
2967                 return ret;
2968
2969         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
2970                 ret = -EFAULT;
2971                 goto out;
2972         }
2973
2974         if (!objectid)
2975                 objectid = BTRFS_FS_TREE_OBJECTID;
2976
2977         new_root = btrfs_get_fs_root(fs_info, objectid, true);
2978         if (IS_ERR(new_root)) {
2979                 ret = PTR_ERR(new_root);
2980                 goto out;
2981         }
2982         if (!is_fstree(btrfs_root_id(new_root))) {
2983                 ret = -ENOENT;
2984                 goto out_free;
2985         }
2986
2987         path = btrfs_alloc_path();
2988         if (!path) {
2989                 ret = -ENOMEM;
2990                 goto out_free;
2991         }
2992
2993         trans = btrfs_start_transaction(root, 1);
2994         if (IS_ERR(trans)) {
2995                 ret = PTR_ERR(trans);
2996                 goto out_free;
2997         }
2998
2999         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3000         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3001                                    dir_id, &name, 1);
3002         if (IS_ERR_OR_NULL(di)) {
3003                 btrfs_release_path(path);
3004                 btrfs_end_transaction(trans);
3005                 btrfs_err(fs_info,
3006                           "Umm, you don't have the default diritem, this isn't going to work");
3007                 ret = -ENOENT;
3008                 goto out_free;
3009         }
3010
3011         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3012         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3013         btrfs_mark_buffer_dirty(trans, path->nodes[0]);
3014         btrfs_release_path(path);
3015
3016         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3017         btrfs_end_transaction(trans);
3018 out_free:
3019         btrfs_put_root(new_root);
3020         btrfs_free_path(path);
3021 out:
3022         mnt_drop_write_file(file);
3023         return ret;
3024 }
3025
3026 static void get_block_group_info(struct list_head *groups_list,
3027                                  struct btrfs_ioctl_space_info *space)
3028 {
3029         struct btrfs_block_group *block_group;
3030
3031         space->total_bytes = 0;
3032         space->used_bytes = 0;
3033         space->flags = 0;
3034         list_for_each_entry(block_group, groups_list, list) {
3035                 space->flags = block_group->flags;
3036                 space->total_bytes += block_group->length;
3037                 space->used_bytes += block_group->used;
3038         }
3039 }
3040
3041 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3042                                    void __user *arg)
3043 {
3044         struct btrfs_ioctl_space_args space_args = { 0 };
3045         struct btrfs_ioctl_space_info space;
3046         struct btrfs_ioctl_space_info *dest;
3047         struct btrfs_ioctl_space_info *dest_orig;
3048         struct btrfs_ioctl_space_info __user *user_dest;
3049         struct btrfs_space_info *info;
3050         static const u64 types[] = {
3051                 BTRFS_BLOCK_GROUP_DATA,
3052                 BTRFS_BLOCK_GROUP_SYSTEM,
3053                 BTRFS_BLOCK_GROUP_METADATA,
3054                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3055         };
3056         int num_types = 4;
3057         int alloc_size;
3058         int ret = 0;
3059         u64 slot_count = 0;
3060         int i, c;
3061
3062         if (copy_from_user(&space_args,
3063                            (struct btrfs_ioctl_space_args __user *)arg,
3064                            sizeof(space_args)))
3065                 return -EFAULT;
3066
3067         for (i = 0; i < num_types; i++) {
3068                 struct btrfs_space_info *tmp;
3069
3070                 info = NULL;
3071                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3072                         if (tmp->flags == types[i]) {
3073                                 info = tmp;
3074                                 break;
3075                         }
3076                 }
3077
3078                 if (!info)
3079                         continue;
3080
3081                 down_read(&info->groups_sem);
3082                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3083                         if (!list_empty(&info->block_groups[c]))
3084                                 slot_count++;
3085                 }
3086                 up_read(&info->groups_sem);
3087         }
3088
3089         /*
3090          * Global block reserve, exported as a space_info
3091          */
3092         slot_count++;
3093
3094         /* space_slots == 0 means they are asking for a count */
3095         if (space_args.space_slots == 0) {
3096                 space_args.total_spaces = slot_count;
3097                 goto out;
3098         }
3099
3100         slot_count = min_t(u64, space_args.space_slots, slot_count);
3101
3102         alloc_size = sizeof(*dest) * slot_count;
3103
3104         /* we generally have at most 6 or so space infos, one for each raid
3105          * level.  So, a whole page should be more than enough for everyone
3106          */
3107         if (alloc_size > PAGE_SIZE)
3108                 return -ENOMEM;
3109
3110         space_args.total_spaces = 0;
3111         dest = kmalloc(alloc_size, GFP_KERNEL);
3112         if (!dest)
3113                 return -ENOMEM;
3114         dest_orig = dest;
3115
3116         /* now we have a buffer to copy into */
3117         for (i = 0; i < num_types; i++) {
3118                 struct btrfs_space_info *tmp;
3119
3120                 if (!slot_count)
3121                         break;
3122
3123                 info = NULL;
3124                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3125                         if (tmp->flags == types[i]) {
3126                                 info = tmp;
3127                                 break;
3128                         }
3129                 }
3130
3131                 if (!info)
3132                         continue;
3133                 down_read(&info->groups_sem);
3134                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3135                         if (!list_empty(&info->block_groups[c])) {
3136                                 get_block_group_info(&info->block_groups[c],
3137                                                      &space);
3138                                 memcpy(dest, &space, sizeof(space));
3139                                 dest++;
3140                                 space_args.total_spaces++;
3141                                 slot_count--;
3142                         }
3143                         if (!slot_count)
3144                                 break;
3145                 }
3146                 up_read(&info->groups_sem);
3147         }
3148
3149         /*
3150          * Add global block reserve
3151          */
3152         if (slot_count) {
3153                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3154
3155                 spin_lock(&block_rsv->lock);
3156                 space.total_bytes = block_rsv->size;
3157                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3158                 spin_unlock(&block_rsv->lock);
3159                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3160                 memcpy(dest, &space, sizeof(space));
3161                 space_args.total_spaces++;
3162         }
3163
3164         user_dest = (struct btrfs_ioctl_space_info __user *)
3165                 (arg + sizeof(struct btrfs_ioctl_space_args));
3166
3167         if (copy_to_user(user_dest, dest_orig, alloc_size))
3168                 ret = -EFAULT;
3169
3170         kfree(dest_orig);
3171 out:
3172         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3173                 ret = -EFAULT;
3174
3175         return ret;
3176 }
3177
3178 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3179                                             void __user *argp)
3180 {
3181         struct btrfs_trans_handle *trans;
3182         u64 transid;
3183
3184         /*
3185          * Start orphan cleanup here for the given root in case it hasn't been
3186          * started already by other means. Errors are handled in the other
3187          * functions during transaction commit.
3188          */
3189         btrfs_orphan_cleanup(root);
3190
3191         trans = btrfs_attach_transaction_barrier(root);
3192         if (IS_ERR(trans)) {
3193                 if (PTR_ERR(trans) != -ENOENT)
3194                         return PTR_ERR(trans);
3195
3196                 /* No running transaction, don't bother */
3197                 transid = btrfs_get_last_trans_committed(root->fs_info);
3198                 goto out;
3199         }
3200         transid = trans->transid;
3201         btrfs_commit_transaction_async(trans);
3202 out:
3203         if (argp)
3204                 if (copy_to_user(argp, &transid, sizeof(transid)))
3205                         return -EFAULT;
3206         return 0;
3207 }
3208
3209 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3210                                            void __user *argp)
3211 {
3212         /* By default wait for the current transaction. */
3213         u64 transid = 0;
3214
3215         if (argp)
3216                 if (copy_from_user(&transid, argp, sizeof(transid)))
3217                         return -EFAULT;
3218
3219         return btrfs_wait_for_commit(fs_info, transid);
3220 }
3221
3222 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3223 {
3224         struct btrfs_fs_info *fs_info = inode_to_fs_info(file_inode(file));
3225         struct btrfs_ioctl_scrub_args *sa;
3226         int ret;
3227
3228         if (!capable(CAP_SYS_ADMIN))
3229                 return -EPERM;
3230
3231         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3232                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
3233                 return -EINVAL;
3234         }
3235
3236         sa = memdup_user(arg, sizeof(*sa));
3237         if (IS_ERR(sa))
3238                 return PTR_ERR(sa);
3239
3240         if (sa->flags & ~BTRFS_SCRUB_SUPPORTED_FLAGS) {
3241                 ret = -EOPNOTSUPP;
3242                 goto out;
3243         }
3244
3245         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3246                 ret = mnt_want_write_file(file);
3247                 if (ret)
3248                         goto out;
3249         }
3250
3251         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3252                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3253                               0);
3254
3255         /*
3256          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3257          * error. This is important as it allows user space to know how much
3258          * progress scrub has done. For example, if scrub is canceled we get
3259          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3260          * space. Later user space can inspect the progress from the structure
3261          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3262          * previously (btrfs-progs does this).
3263          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3264          * then return -EFAULT to signal the structure was not copied or it may
3265          * be corrupt and unreliable due to a partial copy.
3266          */
3267         if (copy_to_user(arg, sa, sizeof(*sa)))
3268                 ret = -EFAULT;
3269
3270         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3271                 mnt_drop_write_file(file);
3272 out:
3273         kfree(sa);
3274         return ret;
3275 }
3276
3277 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3278 {
3279         if (!capable(CAP_SYS_ADMIN))
3280                 return -EPERM;
3281
3282         return btrfs_scrub_cancel(fs_info);
3283 }
3284
3285 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3286                                        void __user *arg)
3287 {
3288         struct btrfs_ioctl_scrub_args *sa;
3289         int ret;
3290
3291         if (!capable(CAP_SYS_ADMIN))
3292                 return -EPERM;
3293
3294         sa = memdup_user(arg, sizeof(*sa));
3295         if (IS_ERR(sa))
3296                 return PTR_ERR(sa);
3297
3298         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3299
3300         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3301                 ret = -EFAULT;
3302
3303         kfree(sa);
3304         return ret;
3305 }
3306
3307 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3308                                       void __user *arg)
3309 {
3310         struct btrfs_ioctl_get_dev_stats *sa;
3311         int ret;
3312
3313         sa = memdup_user(arg, sizeof(*sa));
3314         if (IS_ERR(sa))
3315                 return PTR_ERR(sa);
3316
3317         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3318                 kfree(sa);
3319                 return -EPERM;
3320         }
3321
3322         ret = btrfs_get_dev_stats(fs_info, sa);
3323
3324         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3325                 ret = -EFAULT;
3326
3327         kfree(sa);
3328         return ret;
3329 }
3330
3331 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3332                                     void __user *arg)
3333 {
3334         struct btrfs_ioctl_dev_replace_args *p;
3335         int ret;
3336
3337         if (!capable(CAP_SYS_ADMIN))
3338                 return -EPERM;
3339
3340         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3341                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
3342                 return -EINVAL;
3343         }
3344
3345         p = memdup_user(arg, sizeof(*p));
3346         if (IS_ERR(p))
3347                 return PTR_ERR(p);
3348
3349         switch (p->cmd) {
3350         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3351                 if (sb_rdonly(fs_info->sb)) {
3352                         ret = -EROFS;
3353                         goto out;
3354                 }
3355                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3356                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3357                 } else {
3358                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3359                         btrfs_exclop_finish(fs_info);
3360                 }
3361                 break;
3362         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3363                 btrfs_dev_replace_status(fs_info, p);
3364                 ret = 0;
3365                 break;
3366         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3367                 p->result = btrfs_dev_replace_cancel(fs_info);
3368                 ret = 0;
3369                 break;
3370         default:
3371                 ret = -EINVAL;
3372                 break;
3373         }
3374
3375         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3376                 ret = -EFAULT;
3377 out:
3378         kfree(p);
3379         return ret;
3380 }
3381
3382 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3383 {
3384         int ret = 0;
3385         int i;
3386         u64 rel_ptr;
3387         int size;
3388         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3389         struct inode_fs_paths *ipath = NULL;
3390         struct btrfs_path *path;
3391
3392         if (!capable(CAP_DAC_READ_SEARCH))
3393                 return -EPERM;
3394
3395         path = btrfs_alloc_path();
3396         if (!path) {
3397                 ret = -ENOMEM;
3398                 goto out;
3399         }
3400
3401         ipa = memdup_user(arg, sizeof(*ipa));
3402         if (IS_ERR(ipa)) {
3403                 ret = PTR_ERR(ipa);
3404                 ipa = NULL;
3405                 goto out;
3406         }
3407
3408         size = min_t(u32, ipa->size, 4096);
3409         ipath = init_ipath(size, root, path);
3410         if (IS_ERR(ipath)) {
3411                 ret = PTR_ERR(ipath);
3412                 ipath = NULL;
3413                 goto out;
3414         }
3415
3416         ret = paths_from_inode(ipa->inum, ipath);
3417         if (ret < 0)
3418                 goto out;
3419
3420         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3421                 rel_ptr = ipath->fspath->val[i] -
3422                           (u64)(unsigned long)ipath->fspath->val;
3423                 ipath->fspath->val[i] = rel_ptr;
3424         }
3425
3426         btrfs_free_path(path);
3427         path = NULL;
3428         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3429                            ipath->fspath, size);
3430         if (ret) {
3431                 ret = -EFAULT;
3432                 goto out;
3433         }
3434
3435 out:
3436         btrfs_free_path(path);
3437         free_ipath(ipath);
3438         kfree(ipa);
3439
3440         return ret;
3441 }
3442
3443 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3444                                         void __user *arg, int version)
3445 {
3446         int ret = 0;
3447         int size;
3448         struct btrfs_ioctl_logical_ino_args *loi;
3449         struct btrfs_data_container *inodes = NULL;
3450         struct btrfs_path *path = NULL;
3451         bool ignore_offset;
3452
3453         if (!capable(CAP_SYS_ADMIN))
3454                 return -EPERM;
3455
3456         loi = memdup_user(arg, sizeof(*loi));
3457         if (IS_ERR(loi))
3458                 return PTR_ERR(loi);
3459
3460         if (version == 1) {
3461                 ignore_offset = false;
3462                 size = min_t(u32, loi->size, SZ_64K);
3463         } else {
3464                 /* All reserved bits must be 0 for now */
3465                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3466                         ret = -EINVAL;
3467                         goto out_loi;
3468                 }
3469                 /* Only accept flags we have defined so far */
3470                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3471                         ret = -EINVAL;
3472                         goto out_loi;
3473                 }
3474                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3475                 size = min_t(u32, loi->size, SZ_16M);
3476         }
3477
3478         inodes = init_data_container(size);
3479         if (IS_ERR(inodes)) {
3480                 ret = PTR_ERR(inodes);
3481                 goto out_loi;
3482         }
3483
3484         path = btrfs_alloc_path();
3485         if (!path) {
3486                 ret = -ENOMEM;
3487                 goto out;
3488         }
3489         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3490                                           inodes, ignore_offset);
3491         btrfs_free_path(path);
3492         if (ret == -EINVAL)
3493                 ret = -ENOENT;
3494         if (ret < 0)
3495                 goto out;
3496
3497         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3498                            size);
3499         if (ret)
3500                 ret = -EFAULT;
3501
3502 out:
3503         kvfree(inodes);
3504 out_loi:
3505         kfree(loi);
3506
3507         return ret;
3508 }
3509
3510 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3511                                struct btrfs_ioctl_balance_args *bargs)
3512 {
3513         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3514
3515         bargs->flags = bctl->flags;
3516
3517         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3518                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3519         if (atomic_read(&fs_info->balance_pause_req))
3520                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3521         if (atomic_read(&fs_info->balance_cancel_req))
3522                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3523
3524         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3525         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3526         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3527
3528         spin_lock(&fs_info->balance_lock);
3529         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3530         spin_unlock(&fs_info->balance_lock);
3531 }
3532
3533 /*
3534  * Try to acquire fs_info::balance_mutex as well as set BTRFS_EXLCOP_BALANCE as
3535  * required.
3536  *
3537  * @fs_info:       the filesystem
3538  * @excl_acquired: ptr to boolean value which is set to false in case balance
3539  *                 is being resumed
3540  *
3541  * Return 0 on success in which case both fs_info::balance is acquired as well
3542  * as exclusive ops are blocked. In case of failure return an error code.
3543  */
3544 static int btrfs_try_lock_balance(struct btrfs_fs_info *fs_info, bool *excl_acquired)
3545 {
3546         int ret;
3547
3548         /*
3549          * Exclusive operation is locked. Three possibilities:
3550          *   (1) some other op is running
3551          *   (2) balance is running
3552          *   (3) balance is paused -- special case (think resume)
3553          */
3554         while (1) {
3555                 if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
3556                         *excl_acquired = true;
3557                         mutex_lock(&fs_info->balance_mutex);
3558                         return 0;
3559                 }
3560
3561                 mutex_lock(&fs_info->balance_mutex);
3562                 if (fs_info->balance_ctl) {
3563                         /* This is either (2) or (3) */
3564                         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3565                                 /* This is (2) */
3566                                 ret = -EINPROGRESS;
3567                                 goto out_failure;
3568
3569                         } else {
3570                                 mutex_unlock(&fs_info->balance_mutex);
3571                                 /*
3572                                  * Lock released to allow other waiters to
3573                                  * continue, we'll reexamine the status again.
3574                                  */
3575                                 mutex_lock(&fs_info->balance_mutex);
3576
3577                                 if (fs_info->balance_ctl &&
3578                                     !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3579                                         /* This is (3) */
3580                                         *excl_acquired = false;
3581                                         return 0;
3582                                 }
3583                         }
3584                 } else {
3585                         /* This is (1) */
3586                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3587                         goto out_failure;
3588                 }
3589
3590                 mutex_unlock(&fs_info->balance_mutex);
3591         }
3592
3593 out_failure:
3594         mutex_unlock(&fs_info->balance_mutex);
3595         *excl_acquired = false;
3596         return ret;
3597 }
3598
3599 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3600 {
3601         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3602         struct btrfs_fs_info *fs_info = root->fs_info;
3603         struct btrfs_ioctl_balance_args *bargs;
3604         struct btrfs_balance_control *bctl;
3605         bool need_unlock = true;
3606         int ret;
3607
3608         if (!capable(CAP_SYS_ADMIN))
3609                 return -EPERM;
3610
3611         ret = mnt_want_write_file(file);
3612         if (ret)
3613                 return ret;
3614
3615         bargs = memdup_user(arg, sizeof(*bargs));
3616         if (IS_ERR(bargs)) {
3617                 ret = PTR_ERR(bargs);
3618                 bargs = NULL;
3619                 goto out;
3620         }
3621
3622         ret = btrfs_try_lock_balance(fs_info, &need_unlock);
3623         if (ret)
3624                 goto out;
3625
3626         lockdep_assert_held(&fs_info->balance_mutex);
3627
3628         if (bargs->flags & BTRFS_BALANCE_RESUME) {
3629                 if (!fs_info->balance_ctl) {
3630                         ret = -ENOTCONN;
3631                         goto out_unlock;
3632                 }
3633
3634                 bctl = fs_info->balance_ctl;
3635                 spin_lock(&fs_info->balance_lock);
3636                 bctl->flags |= BTRFS_BALANCE_RESUME;
3637                 spin_unlock(&fs_info->balance_lock);
3638                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
3639
3640                 goto do_balance;
3641         }
3642
3643         if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
3644                 ret = -EINVAL;
3645                 goto out_unlock;
3646         }
3647
3648         if (fs_info->balance_ctl) {
3649                 ret = -EINPROGRESS;
3650                 goto out_unlock;
3651         }
3652
3653         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
3654         if (!bctl) {
3655                 ret = -ENOMEM;
3656                 goto out_unlock;
3657         }
3658
3659         memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3660         memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3661         memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3662
3663         bctl->flags = bargs->flags;
3664 do_balance:
3665         /*
3666          * Ownership of bctl and exclusive operation goes to btrfs_balance.
3667          * bctl is freed in reset_balance_state, or, if restriper was paused
3668          * all the way until unmount, in free_fs_info.  The flag should be
3669          * cleared after reset_balance_state.
3670          */
3671         need_unlock = false;
3672
3673         ret = btrfs_balance(fs_info, bctl, bargs);
3674         bctl = NULL;
3675
3676         if (ret == 0 || ret == -ECANCELED) {
3677                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3678                         ret = -EFAULT;
3679         }
3680
3681         kfree(bctl);
3682 out_unlock:
3683         mutex_unlock(&fs_info->balance_mutex);
3684         if (need_unlock)
3685                 btrfs_exclop_finish(fs_info);
3686 out:
3687         mnt_drop_write_file(file);
3688         kfree(bargs);
3689         return ret;
3690 }
3691
3692 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
3693 {
3694         if (!capable(CAP_SYS_ADMIN))
3695                 return -EPERM;
3696
3697         switch (cmd) {
3698         case BTRFS_BALANCE_CTL_PAUSE:
3699                 return btrfs_pause_balance(fs_info);
3700         case BTRFS_BALANCE_CTL_CANCEL:
3701                 return btrfs_cancel_balance(fs_info);
3702         }
3703
3704         return -EINVAL;
3705 }
3706
3707 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
3708                                          void __user *arg)
3709 {
3710         struct btrfs_ioctl_balance_args *bargs;
3711         int ret = 0;
3712
3713         if (!capable(CAP_SYS_ADMIN))
3714                 return -EPERM;
3715
3716         mutex_lock(&fs_info->balance_mutex);
3717         if (!fs_info->balance_ctl) {
3718                 ret = -ENOTCONN;
3719                 goto out;
3720         }
3721
3722         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
3723         if (!bargs) {
3724                 ret = -ENOMEM;
3725                 goto out;
3726         }
3727
3728         btrfs_update_ioctl_balance_args(fs_info, bargs);
3729
3730         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3731                 ret = -EFAULT;
3732
3733         kfree(bargs);
3734 out:
3735         mutex_unlock(&fs_info->balance_mutex);
3736         return ret;
3737 }
3738
3739 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
3740 {
3741         struct inode *inode = file_inode(file);
3742         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3743         struct btrfs_ioctl_quota_ctl_args *sa;
3744         int ret;
3745
3746         if (!capable(CAP_SYS_ADMIN))
3747                 return -EPERM;
3748
3749         ret = mnt_want_write_file(file);
3750         if (ret)
3751                 return ret;
3752
3753         sa = memdup_user(arg, sizeof(*sa));
3754         if (IS_ERR(sa)) {
3755                 ret = PTR_ERR(sa);
3756                 goto drop_write;
3757         }
3758
3759         switch (sa->cmd) {
3760         case BTRFS_QUOTA_CTL_ENABLE:
3761         case BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA:
3762                 down_write(&fs_info->subvol_sem);
3763                 ret = btrfs_quota_enable(fs_info, sa);
3764                 up_write(&fs_info->subvol_sem);
3765                 break;
3766         case BTRFS_QUOTA_CTL_DISABLE:
3767                 /*
3768                  * Lock the cleaner mutex to prevent races with concurrent
3769                  * relocation, because relocation may be building backrefs for
3770                  * blocks of the quota root while we are deleting the root. This
3771                  * is like dropping fs roots of deleted snapshots/subvolumes, we
3772                  * need the same protection.
3773                  *
3774                  * This also prevents races between concurrent tasks trying to
3775                  * disable quotas, because we will unlock and relock
3776                  * qgroup_ioctl_lock across BTRFS_FS_QUOTA_ENABLED changes.
3777                  *
3778                  * We take this here because we have the dependency of
3779                  *
3780                  * inode_lock -> subvol_sem
3781                  *
3782                  * because of rename.  With relocation we can prealloc extents,
3783                  * so that makes the dependency chain
3784                  *
3785                  * cleaner_mutex -> inode_lock -> subvol_sem
3786                  *
3787                  * so we must take the cleaner_mutex here before we take the
3788                  * subvol_sem.  The deadlock can't actually happen, but this
3789                  * quiets lockdep.
3790                  */
3791                 mutex_lock(&fs_info->cleaner_mutex);
3792                 down_write(&fs_info->subvol_sem);
3793                 ret = btrfs_quota_disable(fs_info);
3794                 up_write(&fs_info->subvol_sem);
3795                 mutex_unlock(&fs_info->cleaner_mutex);
3796                 break;
3797         default:
3798                 ret = -EINVAL;
3799                 break;
3800         }
3801
3802         kfree(sa);
3803 drop_write:
3804         mnt_drop_write_file(file);
3805         return ret;
3806 }
3807
3808 /*
3809  * Quick check for ioctl handlers if quotas are enabled. Proper locking must be
3810  * done before any operations.
3811  */
3812 static bool qgroup_enabled(struct btrfs_fs_info *fs_info)
3813 {
3814         bool ret = true;
3815
3816         mutex_lock(&fs_info->qgroup_ioctl_lock);
3817         if (!fs_info->quota_root)
3818                 ret = false;
3819         mutex_unlock(&fs_info->qgroup_ioctl_lock);
3820
3821         return ret;
3822 }
3823
3824 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
3825 {
3826         struct inode *inode = file_inode(file);
3827         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3828         struct btrfs_root *root = BTRFS_I(inode)->root;
3829         struct btrfs_ioctl_qgroup_assign_args *sa;
3830         struct btrfs_qgroup_list *prealloc = NULL;
3831         struct btrfs_trans_handle *trans;
3832         int ret;
3833         int err;
3834
3835         if (!capable(CAP_SYS_ADMIN))
3836                 return -EPERM;
3837
3838         if (!qgroup_enabled(root->fs_info))
3839                 return -ENOTCONN;
3840
3841         ret = mnt_want_write_file(file);
3842         if (ret)
3843                 return ret;
3844
3845         sa = memdup_user(arg, sizeof(*sa));
3846         if (IS_ERR(sa)) {
3847                 ret = PTR_ERR(sa);
3848                 goto drop_write;
3849         }
3850
3851         if (sa->assign) {
3852                 prealloc = kzalloc(sizeof(*prealloc), GFP_KERNEL);
3853                 if (!prealloc) {
3854                         ret = -ENOMEM;
3855                         goto drop_write;
3856                 }
3857         }
3858
3859         trans = btrfs_join_transaction(root);
3860         if (IS_ERR(trans)) {
3861                 ret = PTR_ERR(trans);
3862                 goto out;
3863         }
3864
3865         /*
3866          * Prealloc ownership is moved to the relation handler, there it's used
3867          * or freed on error.
3868          */
3869         if (sa->assign) {
3870                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst, prealloc);
3871                 prealloc = NULL;
3872         } else {
3873                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
3874         }
3875
3876         /* update qgroup status and info */
3877         mutex_lock(&fs_info->qgroup_ioctl_lock);
3878         err = btrfs_run_qgroups(trans);
3879         mutex_unlock(&fs_info->qgroup_ioctl_lock);
3880         if (err < 0)
3881                 btrfs_warn(fs_info,
3882                            "qgroup status update failed after %s relation, marked as inconsistent",
3883                            sa->assign ? "adding" : "deleting");
3884         err = btrfs_end_transaction(trans);
3885         if (err && !ret)
3886                 ret = err;
3887
3888 out:
3889         kfree(prealloc);
3890         kfree(sa);
3891 drop_write:
3892         mnt_drop_write_file(file);
3893         return ret;
3894 }
3895
3896 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
3897 {
3898         struct inode *inode = file_inode(file);
3899         struct btrfs_root *root = BTRFS_I(inode)->root;
3900         struct btrfs_ioctl_qgroup_create_args *sa;
3901         struct btrfs_trans_handle *trans;
3902         int ret;
3903         int err;
3904
3905         if (!capable(CAP_SYS_ADMIN))
3906                 return -EPERM;
3907
3908         if (!qgroup_enabled(root->fs_info))
3909                 return -ENOTCONN;
3910
3911         ret = mnt_want_write_file(file);
3912         if (ret)
3913                 return ret;
3914
3915         sa = memdup_user(arg, sizeof(*sa));
3916         if (IS_ERR(sa)) {
3917                 ret = PTR_ERR(sa);
3918                 goto drop_write;
3919         }
3920
3921         if (!sa->qgroupid) {
3922                 ret = -EINVAL;
3923                 goto out;
3924         }
3925
3926         if (sa->create && is_fstree(sa->qgroupid)) {
3927                 ret = -EINVAL;
3928                 goto out;
3929         }
3930
3931         trans = btrfs_join_transaction(root);
3932         if (IS_ERR(trans)) {
3933                 ret = PTR_ERR(trans);
3934                 goto out;
3935         }
3936
3937         if (sa->create) {
3938                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
3939         } else {
3940                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
3941         }
3942
3943         err = btrfs_end_transaction(trans);
3944         if (err && !ret)
3945                 ret = err;
3946
3947 out:
3948         kfree(sa);
3949 drop_write:
3950         mnt_drop_write_file(file);
3951         return ret;
3952 }
3953
3954 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
3955 {
3956         struct inode *inode = file_inode(file);
3957         struct btrfs_root *root = BTRFS_I(inode)->root;
3958         struct btrfs_ioctl_qgroup_limit_args *sa;
3959         struct btrfs_trans_handle *trans;
3960         int ret;
3961         int err;
3962         u64 qgroupid;
3963
3964         if (!capable(CAP_SYS_ADMIN))
3965                 return -EPERM;
3966
3967         if (!qgroup_enabled(root->fs_info))
3968                 return -ENOTCONN;
3969
3970         ret = mnt_want_write_file(file);
3971         if (ret)
3972                 return ret;
3973
3974         sa = memdup_user(arg, sizeof(*sa));
3975         if (IS_ERR(sa)) {
3976                 ret = PTR_ERR(sa);
3977                 goto drop_write;
3978         }
3979
3980         trans = btrfs_join_transaction(root);
3981         if (IS_ERR(trans)) {
3982                 ret = PTR_ERR(trans);
3983                 goto out;
3984         }
3985
3986         qgroupid = sa->qgroupid;
3987         if (!qgroupid) {
3988                 /* take the current subvol as qgroup */
3989                 qgroupid = btrfs_root_id(root);
3990         }
3991
3992         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
3993
3994         err = btrfs_end_transaction(trans);
3995         if (err && !ret)
3996                 ret = err;
3997
3998 out:
3999         kfree(sa);
4000 drop_write:
4001         mnt_drop_write_file(file);
4002         return ret;
4003 }
4004
4005 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4006 {
4007         struct inode *inode = file_inode(file);
4008         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4009         struct btrfs_ioctl_quota_rescan_args *qsa;
4010         int ret;
4011
4012         if (!capable(CAP_SYS_ADMIN))
4013                 return -EPERM;
4014
4015         if (!qgroup_enabled(fs_info))
4016                 return -ENOTCONN;
4017
4018         ret = mnt_want_write_file(file);
4019         if (ret)
4020                 return ret;
4021
4022         qsa = memdup_user(arg, sizeof(*qsa));
4023         if (IS_ERR(qsa)) {
4024                 ret = PTR_ERR(qsa);
4025                 goto drop_write;
4026         }
4027
4028         if (qsa->flags) {
4029                 ret = -EINVAL;
4030                 goto out;
4031         }
4032
4033         ret = btrfs_qgroup_rescan(fs_info);
4034
4035 out:
4036         kfree(qsa);
4037 drop_write:
4038         mnt_drop_write_file(file);
4039         return ret;
4040 }
4041
4042 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4043                                                 void __user *arg)
4044 {
4045         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4046
4047         if (!capable(CAP_SYS_ADMIN))
4048                 return -EPERM;
4049
4050         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4051                 qsa.flags = 1;
4052                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4053         }
4054
4055         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4056                 return -EFAULT;
4057
4058         return 0;
4059 }
4060
4061 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4062                                                 void __user *arg)
4063 {
4064         if (!capable(CAP_SYS_ADMIN))
4065                 return -EPERM;
4066
4067         return btrfs_qgroup_wait_for_completion(fs_info, true);
4068 }
4069
4070 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4071                                             struct mnt_idmap *idmap,
4072                                             struct btrfs_ioctl_received_subvol_args *sa)
4073 {
4074         struct inode *inode = file_inode(file);
4075         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4076         struct btrfs_root *root = BTRFS_I(inode)->root;
4077         struct btrfs_root_item *root_item = &root->root_item;
4078         struct btrfs_trans_handle *trans;
4079         struct timespec64 ct = current_time(inode);
4080         int ret = 0;
4081         int received_uuid_changed;
4082
4083         if (!inode_owner_or_capable(idmap, inode))
4084                 return -EPERM;
4085
4086         ret = mnt_want_write_file(file);
4087         if (ret < 0)
4088                 return ret;
4089
4090         down_write(&fs_info->subvol_sem);
4091
4092         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4093                 ret = -EINVAL;
4094                 goto out;
4095         }
4096
4097         if (btrfs_root_readonly(root)) {
4098                 ret = -EROFS;
4099                 goto out;
4100         }
4101
4102         /*
4103          * 1 - root item
4104          * 2 - uuid items (received uuid + subvol uuid)
4105          */
4106         trans = btrfs_start_transaction(root, 3);
4107         if (IS_ERR(trans)) {
4108                 ret = PTR_ERR(trans);
4109                 trans = NULL;
4110                 goto out;
4111         }
4112
4113         sa->rtransid = trans->transid;
4114         sa->rtime.sec = ct.tv_sec;
4115         sa->rtime.nsec = ct.tv_nsec;
4116
4117         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4118                                        BTRFS_UUID_SIZE);
4119         if (received_uuid_changed &&
4120             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4121                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4122                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4123                                           btrfs_root_id(root));
4124                 if (ret && ret != -ENOENT) {
4125                         btrfs_abort_transaction(trans, ret);
4126                         btrfs_end_transaction(trans);
4127                         goto out;
4128                 }
4129         }
4130         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4131         btrfs_set_root_stransid(root_item, sa->stransid);
4132         btrfs_set_root_rtransid(root_item, sa->rtransid);
4133         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4134         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4135         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4136         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4137
4138         ret = btrfs_update_root(trans, fs_info->tree_root,
4139                                 &root->root_key, &root->root_item);
4140         if (ret < 0) {
4141                 btrfs_end_transaction(trans);
4142                 goto out;
4143         }
4144         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4145                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4146                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4147                                           btrfs_root_id(root));
4148                 if (ret < 0 && ret != -EEXIST) {
4149                         btrfs_abort_transaction(trans, ret);
4150                         btrfs_end_transaction(trans);
4151                         goto out;
4152                 }
4153         }
4154         ret = btrfs_commit_transaction(trans);
4155 out:
4156         up_write(&fs_info->subvol_sem);
4157         mnt_drop_write_file(file);
4158         return ret;
4159 }
4160
4161 #ifdef CONFIG_64BIT
4162 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4163                                                 void __user *arg)
4164 {
4165         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4166         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4167         int ret = 0;
4168
4169         args32 = memdup_user(arg, sizeof(*args32));
4170         if (IS_ERR(args32))
4171                 return PTR_ERR(args32);
4172
4173         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4174         if (!args64) {
4175                 ret = -ENOMEM;
4176                 goto out;
4177         }
4178
4179         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4180         args64->stransid = args32->stransid;
4181         args64->rtransid = args32->rtransid;
4182         args64->stime.sec = args32->stime.sec;
4183         args64->stime.nsec = args32->stime.nsec;
4184         args64->rtime.sec = args32->rtime.sec;
4185         args64->rtime.nsec = args32->rtime.nsec;
4186         args64->flags = args32->flags;
4187
4188         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), args64);
4189         if (ret)
4190                 goto out;
4191
4192         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4193         args32->stransid = args64->stransid;
4194         args32->rtransid = args64->rtransid;
4195         args32->stime.sec = args64->stime.sec;
4196         args32->stime.nsec = args64->stime.nsec;
4197         args32->rtime.sec = args64->rtime.sec;
4198         args32->rtime.nsec = args64->rtime.nsec;
4199         args32->flags = args64->flags;
4200
4201         ret = copy_to_user(arg, args32, sizeof(*args32));
4202         if (ret)
4203                 ret = -EFAULT;
4204
4205 out:
4206         kfree(args32);
4207         kfree(args64);
4208         return ret;
4209 }
4210 #endif
4211
4212 static long btrfs_ioctl_set_received_subvol(struct file *file,
4213                                             void __user *arg)
4214 {
4215         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4216         int ret = 0;
4217
4218         sa = memdup_user(arg, sizeof(*sa));
4219         if (IS_ERR(sa))
4220                 return PTR_ERR(sa);
4221
4222         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_idmap(file), sa);
4223
4224         if (ret)
4225                 goto out;
4226
4227         ret = copy_to_user(arg, sa, sizeof(*sa));
4228         if (ret)
4229                 ret = -EFAULT;
4230
4231 out:
4232         kfree(sa);
4233         return ret;
4234 }
4235
4236 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4237                                         void __user *arg)
4238 {
4239         size_t len;
4240         int ret;
4241         char label[BTRFS_LABEL_SIZE];
4242
4243         spin_lock(&fs_info->super_lock);
4244         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4245         spin_unlock(&fs_info->super_lock);
4246
4247         len = strnlen(label, BTRFS_LABEL_SIZE);
4248
4249         if (len == BTRFS_LABEL_SIZE) {
4250                 btrfs_warn(fs_info,
4251                            "label is too long, return the first %zu bytes",
4252                            --len);
4253         }
4254
4255         ret = copy_to_user(arg, label, len);
4256
4257         return ret ? -EFAULT : 0;
4258 }
4259
4260 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4261 {
4262         struct inode *inode = file_inode(file);
4263         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4264         struct btrfs_root *root = BTRFS_I(inode)->root;
4265         struct btrfs_super_block *super_block = fs_info->super_copy;
4266         struct btrfs_trans_handle *trans;
4267         char label[BTRFS_LABEL_SIZE];
4268         int ret;
4269
4270         if (!capable(CAP_SYS_ADMIN))
4271                 return -EPERM;
4272
4273         if (copy_from_user(label, arg, sizeof(label)))
4274                 return -EFAULT;
4275
4276         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4277                 btrfs_err(fs_info,
4278                           "unable to set label with more than %d bytes",
4279                           BTRFS_LABEL_SIZE - 1);
4280                 return -EINVAL;
4281         }
4282
4283         ret = mnt_want_write_file(file);
4284         if (ret)
4285                 return ret;
4286
4287         trans = btrfs_start_transaction(root, 0);
4288         if (IS_ERR(trans)) {
4289                 ret = PTR_ERR(trans);
4290                 goto out_unlock;
4291         }
4292
4293         spin_lock(&fs_info->super_lock);
4294         strcpy(super_block->label, label);
4295         spin_unlock(&fs_info->super_lock);
4296         ret = btrfs_commit_transaction(trans);
4297
4298 out_unlock:
4299         mnt_drop_write_file(file);
4300         return ret;
4301 }
4302
4303 #define INIT_FEATURE_FLAGS(suffix) \
4304         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4305           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4306           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4307
4308 int btrfs_ioctl_get_supported_features(void __user *arg)
4309 {
4310         static const struct btrfs_ioctl_feature_flags features[3] = {
4311                 INIT_FEATURE_FLAGS(SUPP),
4312                 INIT_FEATURE_FLAGS(SAFE_SET),
4313                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4314         };
4315
4316         if (copy_to_user(arg, &features, sizeof(features)))
4317                 return -EFAULT;
4318
4319         return 0;
4320 }
4321
4322 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4323                                         void __user *arg)
4324 {
4325         struct btrfs_super_block *super_block = fs_info->super_copy;
4326         struct btrfs_ioctl_feature_flags features;
4327
4328         features.compat_flags = btrfs_super_compat_flags(super_block);
4329         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4330         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4331
4332         if (copy_to_user(arg, &features, sizeof(features)))
4333                 return -EFAULT;
4334
4335         return 0;
4336 }
4337
4338 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4339                               enum btrfs_feature_set set,
4340                               u64 change_mask, u64 flags, u64 supported_flags,
4341                               u64 safe_set, u64 safe_clear)
4342 {
4343         const char *type = btrfs_feature_set_name(set);
4344         char *names;
4345         u64 disallowed, unsupported;
4346         u64 set_mask = flags & change_mask;
4347         u64 clear_mask = ~flags & change_mask;
4348
4349         unsupported = set_mask & ~supported_flags;
4350         if (unsupported) {
4351                 names = btrfs_printable_features(set, unsupported);
4352                 if (names) {
4353                         btrfs_warn(fs_info,
4354                                    "this kernel does not support the %s feature bit%s",
4355                                    names, strchr(names, ',') ? "s" : "");
4356                         kfree(names);
4357                 } else
4358                         btrfs_warn(fs_info,
4359                                    "this kernel does not support %s bits 0x%llx",
4360                                    type, unsupported);
4361                 return -EOPNOTSUPP;
4362         }
4363
4364         disallowed = set_mask & ~safe_set;
4365         if (disallowed) {
4366                 names = btrfs_printable_features(set, disallowed);
4367                 if (names) {
4368                         btrfs_warn(fs_info,
4369                                    "can't set the %s feature bit%s while mounted",
4370                                    names, strchr(names, ',') ? "s" : "");
4371                         kfree(names);
4372                 } else
4373                         btrfs_warn(fs_info,
4374                                    "can't set %s bits 0x%llx while mounted",
4375                                    type, disallowed);
4376                 return -EPERM;
4377         }
4378
4379         disallowed = clear_mask & ~safe_clear;
4380         if (disallowed) {
4381                 names = btrfs_printable_features(set, disallowed);
4382                 if (names) {
4383                         btrfs_warn(fs_info,
4384                                    "can't clear the %s feature bit%s while mounted",
4385                                    names, strchr(names, ',') ? "s" : "");
4386                         kfree(names);
4387                 } else
4388                         btrfs_warn(fs_info,
4389                                    "can't clear %s bits 0x%llx while mounted",
4390                                    type, disallowed);
4391                 return -EPERM;
4392         }
4393
4394         return 0;
4395 }
4396
4397 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4398 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4399                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4400                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4401                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4402
4403 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4404 {
4405         struct inode *inode = file_inode(file);
4406         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4407         struct btrfs_root *root = BTRFS_I(inode)->root;
4408         struct btrfs_super_block *super_block = fs_info->super_copy;
4409         struct btrfs_ioctl_feature_flags flags[2];
4410         struct btrfs_trans_handle *trans;
4411         u64 newflags;
4412         int ret;
4413
4414         if (!capable(CAP_SYS_ADMIN))
4415                 return -EPERM;
4416
4417         if (copy_from_user(flags, arg, sizeof(flags)))
4418                 return -EFAULT;
4419
4420         /* Nothing to do */
4421         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4422             !flags[0].incompat_flags)
4423                 return 0;
4424
4425         ret = check_feature(fs_info, flags[0].compat_flags,
4426                             flags[1].compat_flags, COMPAT);
4427         if (ret)
4428                 return ret;
4429
4430         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4431                             flags[1].compat_ro_flags, COMPAT_RO);
4432         if (ret)
4433                 return ret;
4434
4435         ret = check_feature(fs_info, flags[0].incompat_flags,
4436                             flags[1].incompat_flags, INCOMPAT);
4437         if (ret)
4438                 return ret;
4439
4440         ret = mnt_want_write_file(file);
4441         if (ret)
4442                 return ret;
4443
4444         trans = btrfs_start_transaction(root, 0);
4445         if (IS_ERR(trans)) {
4446                 ret = PTR_ERR(trans);
4447                 goto out_drop_write;
4448         }
4449
4450         spin_lock(&fs_info->super_lock);
4451         newflags = btrfs_super_compat_flags(super_block);
4452         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4453         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4454         btrfs_set_super_compat_flags(super_block, newflags);
4455
4456         newflags = btrfs_super_compat_ro_flags(super_block);
4457         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4458         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4459         btrfs_set_super_compat_ro_flags(super_block, newflags);
4460
4461         newflags = btrfs_super_incompat_flags(super_block);
4462         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4463         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4464         btrfs_set_super_incompat_flags(super_block, newflags);
4465         spin_unlock(&fs_info->super_lock);
4466
4467         ret = btrfs_commit_transaction(trans);
4468 out_drop_write:
4469         mnt_drop_write_file(file);
4470
4471         return ret;
4472 }
4473
4474 static int _btrfs_ioctl_send(struct btrfs_inode *inode, void __user *argp, bool compat)
4475 {
4476         struct btrfs_ioctl_send_args *arg;
4477         int ret;
4478
4479         if (compat) {
4480 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4481                 struct btrfs_ioctl_send_args_32 args32 = { 0 };
4482
4483                 ret = copy_from_user(&args32, argp, sizeof(args32));
4484                 if (ret)
4485                         return -EFAULT;
4486                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4487                 if (!arg)
4488                         return -ENOMEM;
4489                 arg->send_fd = args32.send_fd;
4490                 arg->clone_sources_count = args32.clone_sources_count;
4491                 arg->clone_sources = compat_ptr(args32.clone_sources);
4492                 arg->parent_root = args32.parent_root;
4493                 arg->flags = args32.flags;
4494                 arg->version = args32.version;
4495                 memcpy(arg->reserved, args32.reserved,
4496                        sizeof(args32.reserved));
4497 #else
4498                 return -ENOTTY;
4499 #endif
4500         } else {
4501                 arg = memdup_user(argp, sizeof(*arg));
4502                 if (IS_ERR(arg))
4503                         return PTR_ERR(arg);
4504         }
4505         ret = btrfs_ioctl_send(inode, arg);
4506         kfree(arg);
4507         return ret;
4508 }
4509
4510 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
4511                                     bool compat)
4512 {
4513         struct btrfs_ioctl_encoded_io_args args = { 0 };
4514         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
4515                                              flags);
4516         size_t copy_end;
4517         struct iovec iovstack[UIO_FASTIOV];
4518         struct iovec *iov = iovstack;
4519         struct iov_iter iter;
4520         loff_t pos;
4521         struct kiocb kiocb;
4522         ssize_t ret;
4523
4524         if (!capable(CAP_SYS_ADMIN)) {
4525                 ret = -EPERM;
4526                 goto out_acct;
4527         }
4528
4529         if (compat) {
4530 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4531                 struct btrfs_ioctl_encoded_io_args_32 args32;
4532
4533                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
4534                                        flags);
4535                 if (copy_from_user(&args32, argp, copy_end)) {
4536                         ret = -EFAULT;
4537                         goto out_acct;
4538                 }
4539                 args.iov = compat_ptr(args32.iov);
4540                 args.iovcnt = args32.iovcnt;
4541                 args.offset = args32.offset;
4542                 args.flags = args32.flags;
4543 #else
4544                 return -ENOTTY;
4545 #endif
4546         } else {
4547                 copy_end = copy_end_kernel;
4548                 if (copy_from_user(&args, argp, copy_end)) {
4549                         ret = -EFAULT;
4550                         goto out_acct;
4551                 }
4552         }
4553         if (args.flags != 0) {
4554                 ret = -EINVAL;
4555                 goto out_acct;
4556         }
4557
4558         ret = import_iovec(ITER_DEST, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4559                            &iov, &iter);
4560         if (ret < 0)
4561                 goto out_acct;
4562
4563         if (iov_iter_count(&iter) == 0) {
4564                 ret = 0;
4565                 goto out_iov;
4566         }
4567         pos = args.offset;
4568         ret = rw_verify_area(READ, file, &pos, args.len);
4569         if (ret < 0)
4570                 goto out_iov;
4571
4572         init_sync_kiocb(&kiocb, file);
4573         kiocb.ki_pos = pos;
4574
4575         ret = btrfs_encoded_read(&kiocb, &iter, &args);
4576         if (ret >= 0) {
4577                 fsnotify_access(file);
4578                 if (copy_to_user(argp + copy_end,
4579                                  (char *)&args + copy_end_kernel,
4580                                  sizeof(args) - copy_end_kernel))
4581                         ret = -EFAULT;
4582         }
4583
4584 out_iov:
4585         kfree(iov);
4586 out_acct:
4587         if (ret > 0)
4588                 add_rchar(current, ret);
4589         inc_syscr(current);
4590         return ret;
4591 }
4592
4593 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
4594 {
4595         struct btrfs_ioctl_encoded_io_args args;
4596         struct iovec iovstack[UIO_FASTIOV];
4597         struct iovec *iov = iovstack;
4598         struct iov_iter iter;
4599         loff_t pos;
4600         struct kiocb kiocb;
4601         ssize_t ret;
4602
4603         if (!capable(CAP_SYS_ADMIN)) {
4604                 ret = -EPERM;
4605                 goto out_acct;
4606         }
4607
4608         if (!(file->f_mode & FMODE_WRITE)) {
4609                 ret = -EBADF;
4610                 goto out_acct;
4611         }
4612
4613         if (compat) {
4614 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4615                 struct btrfs_ioctl_encoded_io_args_32 args32;
4616
4617                 if (copy_from_user(&args32, argp, sizeof(args32))) {
4618                         ret = -EFAULT;
4619                         goto out_acct;
4620                 }
4621                 args.iov = compat_ptr(args32.iov);
4622                 args.iovcnt = args32.iovcnt;
4623                 args.offset = args32.offset;
4624                 args.flags = args32.flags;
4625                 args.len = args32.len;
4626                 args.unencoded_len = args32.unencoded_len;
4627                 args.unencoded_offset = args32.unencoded_offset;
4628                 args.compression = args32.compression;
4629                 args.encryption = args32.encryption;
4630                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
4631 #else
4632                 return -ENOTTY;
4633 #endif
4634         } else {
4635                 if (copy_from_user(&args, argp, sizeof(args))) {
4636                         ret = -EFAULT;
4637                         goto out_acct;
4638                 }
4639         }
4640
4641         ret = -EINVAL;
4642         if (args.flags != 0)
4643                 goto out_acct;
4644         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
4645                 goto out_acct;
4646         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
4647             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
4648                 goto out_acct;
4649         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
4650             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
4651                 goto out_acct;
4652         if (args.unencoded_offset > args.unencoded_len)
4653                 goto out_acct;
4654         if (args.len > args.unencoded_len - args.unencoded_offset)
4655                 goto out_acct;
4656
4657         ret = import_iovec(ITER_SOURCE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
4658                            &iov, &iter);
4659         if (ret < 0)
4660                 goto out_acct;
4661
4662         if (iov_iter_count(&iter) == 0) {
4663                 ret = 0;
4664                 goto out_iov;
4665         }
4666         pos = args.offset;
4667         ret = rw_verify_area(WRITE, file, &pos, args.len);
4668         if (ret < 0)
4669                 goto out_iov;
4670
4671         init_sync_kiocb(&kiocb, file);
4672         ret = kiocb_set_rw_flags(&kiocb, 0, WRITE);
4673         if (ret)
4674                 goto out_iov;
4675         kiocb.ki_pos = pos;
4676
4677         file_start_write(file);
4678
4679         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
4680         if (ret > 0)
4681                 fsnotify_modify(file);
4682
4683         file_end_write(file);
4684 out_iov:
4685         kfree(iov);
4686 out_acct:
4687         if (ret > 0)
4688                 add_wchar(current, ret);
4689         inc_syscw(current);
4690         return ret;
4691 }
4692
4693 long btrfs_ioctl(struct file *file, unsigned int
4694                 cmd, unsigned long arg)
4695 {
4696         struct inode *inode = file_inode(file);
4697         struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
4698         struct btrfs_root *root = BTRFS_I(inode)->root;
4699         void __user *argp = (void __user *)arg;
4700
4701         switch (cmd) {
4702         case FS_IOC_GETVERSION:
4703                 return btrfs_ioctl_getversion(inode, argp);
4704         case FS_IOC_GETFSLABEL:
4705                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4706         case FS_IOC_SETFSLABEL:
4707                 return btrfs_ioctl_set_fslabel(file, argp);
4708         case FITRIM:
4709                 return btrfs_ioctl_fitrim(fs_info, argp);
4710         case BTRFS_IOC_SNAP_CREATE:
4711                 return btrfs_ioctl_snap_create(file, argp, 0);
4712         case BTRFS_IOC_SNAP_CREATE_V2:
4713                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4714         case BTRFS_IOC_SUBVOL_CREATE:
4715                 return btrfs_ioctl_snap_create(file, argp, 1);
4716         case BTRFS_IOC_SUBVOL_CREATE_V2:
4717                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4718         case BTRFS_IOC_SNAP_DESTROY:
4719                 return btrfs_ioctl_snap_destroy(file, argp, false);
4720         case BTRFS_IOC_SNAP_DESTROY_V2:
4721                 return btrfs_ioctl_snap_destroy(file, argp, true);
4722         case BTRFS_IOC_SUBVOL_GETFLAGS:
4723                 return btrfs_ioctl_subvol_getflags(inode, argp);
4724         case BTRFS_IOC_SUBVOL_SETFLAGS:
4725                 return btrfs_ioctl_subvol_setflags(file, argp);
4726         case BTRFS_IOC_DEFAULT_SUBVOL:
4727                 return btrfs_ioctl_default_subvol(file, argp);
4728         case BTRFS_IOC_DEFRAG:
4729                 return btrfs_ioctl_defrag(file, NULL);
4730         case BTRFS_IOC_DEFRAG_RANGE:
4731                 return btrfs_ioctl_defrag(file, argp);
4732         case BTRFS_IOC_RESIZE:
4733                 return btrfs_ioctl_resize(file, argp);
4734         case BTRFS_IOC_ADD_DEV:
4735                 return btrfs_ioctl_add_dev(fs_info, argp);
4736         case BTRFS_IOC_RM_DEV:
4737                 return btrfs_ioctl_rm_dev(file, argp);
4738         case BTRFS_IOC_RM_DEV_V2:
4739                 return btrfs_ioctl_rm_dev_v2(file, argp);
4740         case BTRFS_IOC_FS_INFO:
4741                 return btrfs_ioctl_fs_info(fs_info, argp);
4742         case BTRFS_IOC_DEV_INFO:
4743                 return btrfs_ioctl_dev_info(fs_info, argp);
4744         case BTRFS_IOC_TREE_SEARCH:
4745                 return btrfs_ioctl_tree_search(inode, argp);
4746         case BTRFS_IOC_TREE_SEARCH_V2:
4747                 return btrfs_ioctl_tree_search_v2(inode, argp);
4748         case BTRFS_IOC_INO_LOOKUP:
4749                 return btrfs_ioctl_ino_lookup(root, argp);
4750         case BTRFS_IOC_INO_PATHS:
4751                 return btrfs_ioctl_ino_to_path(root, argp);
4752         case BTRFS_IOC_LOGICAL_INO:
4753                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4754         case BTRFS_IOC_LOGICAL_INO_V2:
4755                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4756         case BTRFS_IOC_SPACE_INFO:
4757                 return btrfs_ioctl_space_info(fs_info, argp);
4758         case BTRFS_IOC_SYNC: {
4759                 int ret;
4760
4761                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4762                 if (ret)
4763                         return ret;
4764                 ret = btrfs_sync_fs(inode->i_sb, 1);
4765                 /*
4766                  * There may be work for the cleaner kthread to do (subvolume
4767                  * deletion, delayed iputs, defrag inodes, etc), so wake it up.
4768                  */
4769                 wake_up_process(fs_info->cleaner_kthread);
4770                 return ret;
4771         }
4772         case BTRFS_IOC_START_SYNC:
4773                 return btrfs_ioctl_start_sync(root, argp);
4774         case BTRFS_IOC_WAIT_SYNC:
4775                 return btrfs_ioctl_wait_sync(fs_info, argp);
4776         case BTRFS_IOC_SCRUB:
4777                 return btrfs_ioctl_scrub(file, argp);
4778         case BTRFS_IOC_SCRUB_CANCEL:
4779                 return btrfs_ioctl_scrub_cancel(fs_info);
4780         case BTRFS_IOC_SCRUB_PROGRESS:
4781                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4782         case BTRFS_IOC_BALANCE_V2:
4783                 return btrfs_ioctl_balance(file, argp);
4784         case BTRFS_IOC_BALANCE_CTL:
4785                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4786         case BTRFS_IOC_BALANCE_PROGRESS:
4787                 return btrfs_ioctl_balance_progress(fs_info, argp);
4788         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4789                 return btrfs_ioctl_set_received_subvol(file, argp);
4790 #ifdef CONFIG_64BIT
4791         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4792                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4793 #endif
4794         case BTRFS_IOC_SEND:
4795                 return _btrfs_ioctl_send(BTRFS_I(inode), argp, false);
4796 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4797         case BTRFS_IOC_SEND_32:
4798                 return _btrfs_ioctl_send(BTRFS_I(inode), argp, true);
4799 #endif
4800         case BTRFS_IOC_GET_DEV_STATS:
4801                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4802         case BTRFS_IOC_QUOTA_CTL:
4803                 return btrfs_ioctl_quota_ctl(file, argp);
4804         case BTRFS_IOC_QGROUP_ASSIGN:
4805                 return btrfs_ioctl_qgroup_assign(file, argp);
4806         case BTRFS_IOC_QGROUP_CREATE:
4807                 return btrfs_ioctl_qgroup_create(file, argp);
4808         case BTRFS_IOC_QGROUP_LIMIT:
4809                 return btrfs_ioctl_qgroup_limit(file, argp);
4810         case BTRFS_IOC_QUOTA_RESCAN:
4811                 return btrfs_ioctl_quota_rescan(file, argp);
4812         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4813                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4814         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4815                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4816         case BTRFS_IOC_DEV_REPLACE:
4817                 return btrfs_ioctl_dev_replace(fs_info, argp);
4818         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4819                 return btrfs_ioctl_get_supported_features(argp);
4820         case BTRFS_IOC_GET_FEATURES:
4821                 return btrfs_ioctl_get_features(fs_info, argp);
4822         case BTRFS_IOC_SET_FEATURES:
4823                 return btrfs_ioctl_set_features(file, argp);
4824         case BTRFS_IOC_GET_SUBVOL_INFO:
4825                 return btrfs_ioctl_get_subvol_info(inode, argp);
4826         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4827                 return btrfs_ioctl_get_subvol_rootref(root, argp);
4828         case BTRFS_IOC_INO_LOOKUP_USER:
4829                 return btrfs_ioctl_ino_lookup_user(file, argp);
4830         case FS_IOC_ENABLE_VERITY:
4831                 return fsverity_ioctl_enable(file, (const void __user *)argp);
4832         case FS_IOC_MEASURE_VERITY:
4833                 return fsverity_ioctl_measure(file, argp);
4834         case BTRFS_IOC_ENCODED_READ:
4835                 return btrfs_ioctl_encoded_read(file, argp, false);
4836         case BTRFS_IOC_ENCODED_WRITE:
4837                 return btrfs_ioctl_encoded_write(file, argp, false);
4838 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4839         case BTRFS_IOC_ENCODED_READ_32:
4840                 return btrfs_ioctl_encoded_read(file, argp, true);
4841         case BTRFS_IOC_ENCODED_WRITE_32:
4842                 return btrfs_ioctl_encoded_write(file, argp, true);
4843 #endif
4844         }
4845
4846         return -ENOTTY;
4847 }
4848
4849 #ifdef CONFIG_COMPAT
4850 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4851 {
4852         /*
4853          * These all access 32-bit values anyway so no further
4854          * handling is necessary.
4855          */
4856         switch (cmd) {
4857         case FS_IOC32_GETVERSION:
4858                 cmd = FS_IOC_GETVERSION;
4859                 break;
4860         }
4861
4862         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
4863 }
4864 #endif