fs/btrfs/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/file.h>
   9 #include <linux/fs.h>
  10 #include <linux/fsnotify.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/writeback.h>
  19 #include <linux/compat.h>
  20 #include <linux/security.h>
  21 #include <linux/xattr.h>
  22 #include <linux/mm.h>
  23 #include <linux/slab.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/uuid.h>
  26 #include <linux/btrfs.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/iversion.h>
  29 #include <linux/fileattr.h>
  30 #include <linux/fsverity.h>
  31 #include <linux/sched/xacct.h>
  32 #include "ctree.h"
  33 #include "disk-io.h"
  34 #include "export.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "print-tree.h"
  38 #include "volumes.h"
  39 #include "locking.h"
  40 #include "backref.h"
  41 #include "rcu-string.h"
  42 #include "send.h"
  43 #include "dev-replace.h"
  44 #include "props.h"
  45 #include "sysfs.h"
  46 #include "qgroup.h"
  47 #include "tree-log.h"
  48 #include "compression.h"
  49 #include "space-info.h"
  50 #include "delalloc-space.h"
  51 #include "block-group.h"
  52 #include "subpage.h"
  53
  54 #ifdef CONFIG_64BIT
  55 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  56  * structures are incorrect, as the timespec structure from userspace
  57  * is 4 bytes too small. We define these alternatives here to teach
  58  * the kernel about the 32-bit struct packing.
  59  */
  60 struct btrfs_ioctl_timespec_32 {
  61         __u64 sec;
  62         __u32 nsec;
  63 } __attribute__ ((__packed__));
  64
  65 struct btrfs_ioctl_received_subvol_args_32 {
  66         char    uuid[BTRFS_UUID_SIZE];  /* in */
  67         __u64   stransid;               /* in */
  68         __u64   rtransid;               /* out */
  69         struct btrfs_ioctl_timespec_32 stime; /* in */
  70         struct btrfs_ioctl_timespec_32 rtime; /* out */
  71         __u64   flags;                  /* in */
  72         __u64   reserved[16];           /* in */
  73 } __attribute__ ((__packed__));
  74
  75 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  76                                 struct btrfs_ioctl_received_subvol_args_32)
  77 #endif
  78
  79 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  80 struct btrfs_ioctl_send_args_32 {
  81         __s64 send_fd;                  /* in */
  82         __u64 clone_sources_count;      /* in */
  83         compat_uptr_t clone_sources;    /* in */
  84         __u64 parent_root;              /* in */
  85         __u64 flags;                    /* in */
  86         __u32 version;                  /* in */
  87         __u8  reserved[28];             /* in */
  88 } __attribute__ ((__packed__));
  89
  90 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  91                                struct btrfs_ioctl_send_args_32)
  92
  93 struct btrfs_ioctl_encoded_io_args_32 {
  94         compat_uptr_t iov;
  95         compat_ulong_t iovcnt;
  96         __s64 offset;
  97         __u64 flags;
  98         __u64 len;
  99         __u64 unencoded_len;
 100         __u64 unencoded_offset;
 101         __u32 compression;
 102         __u32 encryption;
 103         __u8 reserved[64];
 104 };
 105
 106 #define BTRFS_IOC_ENCODED_READ_32 _IOR(BTRFS_IOCTL_MAGIC, 64, \
 107                                        struct btrfs_ioctl_encoded_io_args_32)
 108 #define BTRFS_IOC_ENCODED_WRITE_32 _IOW(BTRFS_IOCTL_MAGIC, 64, \
 109                                         struct btrfs_ioctl_encoded_io_args_32)
 110 #endif
 111
 112 /* Mask out flags that are inappropriate for the given type of inode. */
 113 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
 114                 unsigned int flags)
 115 {
 116         if (S_ISDIR(inode->i_mode))
 117                 return flags;
 118         else if (S_ISREG(inode->i_mode))
 119                 return flags & ~FS_DIRSYNC_FL;
 120         else
 121                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 122 }
 123
 124 /*
 125  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 126  * ioctl.
 127  */
 128 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
 129 {
 130         unsigned int iflags = 0;
 131         u32 flags = binode->flags;
 132         u32 ro_flags = binode->ro_flags;
 133
 134         if (flags & BTRFS_INODE_SYNC)
 135                 iflags |= FS_SYNC_FL;
 136         if (flags & BTRFS_INODE_IMMUTABLE)
 137                 iflags |= FS_IMMUTABLE_FL;
 138         if (flags & BTRFS_INODE_APPEND)
 139                 iflags |= FS_APPEND_FL;
 140         if (flags & BTRFS_INODE_NODUMP)
 141                 iflags |= FS_NODUMP_FL;
 142         if (flags & BTRFS_INODE_NOATIME)
 143                 iflags |= FS_NOATIME_FL;
 144         if (flags & BTRFS_INODE_DIRSYNC)
 145                 iflags |= FS_DIRSYNC_FL;
 146         if (flags & BTRFS_INODE_NODATACOW)
 147                 iflags |= FS_NOCOW_FL;
 148         if (ro_flags & BTRFS_INODE_RO_VERITY)
 149                 iflags |= FS_VERITY_FL;
 150
 151         if (flags & BTRFS_INODE_NOCOMPRESS)
 152                 iflags |= FS_NOCOMP_FL;
 153         else if (flags & BTRFS_INODE_COMPRESS)
 154                 iflags |= FS_COMPR_FL;
 155
 156         return iflags;
 157 }
 158
 159 /*
 160  * Update inode->i_flags based on the btrfs internal flags.
 161  */
 162 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 163 {
 164         struct btrfs_inode *binode = BTRFS_I(inode);
 165         unsigned int new_fl = 0;
 166
 167         if (binode->flags & BTRFS_INODE_SYNC)
 168                 new_fl |= S_SYNC;
 169         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 170                 new_fl |= S_IMMUTABLE;
 171         if (binode->flags & BTRFS_INODE_APPEND)
 172                 new_fl |= S_APPEND;
 173         if (binode->flags & BTRFS_INODE_NOATIME)
 174                 new_fl |= S_NOATIME;
 175         if (binode->flags & BTRFS_INODE_DIRSYNC)
 176                 new_fl |= S_DIRSYNC;
 177         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
 178                 new_fl |= S_VERITY;
 179
 180         set_mask_bits(&inode->i_flags,
 181                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
 182                       S_VERITY, new_fl);
 183 }
 184
 185 /*
 186  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 187  * the old and new flags are not conflicting
 188  */
 189 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 190 {
 191         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 192                       FS_NOATIME_FL | FS_NODUMP_FL | \
 193                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 194                       FS_NOCOMP_FL | FS_COMPR_FL |
 195                       FS_NOCOW_FL))
 196                 return -EOPNOTSUPP;
 197
 198         /* COMPR and NOCOMP on new/old are valid */
 199         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 200                 return -EINVAL;
 201
 202         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 203                 return -EINVAL;
 204
 205         /* NOCOW and compression options are mutually exclusive */
 206         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 207                 return -EINVAL;
 208         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 209                 return -EINVAL;
 210
 211         return 0;
 212 }
 213
 214 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 215                                     unsigned int flags)
 216 {
 217         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 218                 return -EPERM;
 219
 220         return 0;
 221 }
 222
 223 /*
 224  * Set flags/xflags from the internal inode flags. The remaining items of
 225  * fsxattr are zeroed.
 226  */
 227 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 228 {
 229         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
 230
 231         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
 232         return 0;
 233 }
 234
 235 int btrfs_fileattr_set(struct user_namespace *mnt_userns,
 236                        struct dentry *dentry, struct fileattr *fa)
 237 {
 238         struct inode *inode = d_inode(dentry);
 239         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 240         struct btrfs_inode *binode = BTRFS_I(inode);
 241         struct btrfs_root *root = binode->root;
 242         struct btrfs_trans_handle *trans;
 243         unsigned int fsflags, old_fsflags;
 244         int ret;
 245         const char *comp = NULL;
 246         u32 binode_flags;
 247
 248         if (btrfs_root_readonly(root))
 249                 return -EROFS;
 250
 251         if (fileattr_has_fsx(fa))
 252                 return -EOPNOTSUPP;
 253
 254         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
 255         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
 256         ret = check_fsflags(old_fsflags, fsflags);
 257         if (ret)
 258                 return ret;
 259
 260         ret = check_fsflags_compatible(fs_info, fsflags);
 261         if (ret)
 262                 return ret;
 263
 264         binode_flags = binode->flags;
 265         if (fsflags & FS_SYNC_FL)
 266                 binode_flags |= BTRFS_INODE_SYNC;
 267         else
 268                 binode_flags &= ~BTRFS_INODE_SYNC;
 269         if (fsflags & FS_IMMUTABLE_FL)
 270                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 271         else
 272                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 273         if (fsflags & FS_APPEND_FL)
 274                 binode_flags |= BTRFS_INODE_APPEND;
 275         else
 276                 binode_flags &= ~BTRFS_INODE_APPEND;
 277         if (fsflags & FS_NODUMP_FL)
 278                 binode_flags |= BTRFS_INODE_NODUMP;
 279         else
 280                 binode_flags &= ~BTRFS_INODE_NODUMP;
 281         if (fsflags & FS_NOATIME_FL)
 282                 binode_flags |= BTRFS_INODE_NOATIME;
 283         else
 284                 binode_flags &= ~BTRFS_INODE_NOATIME;
 285
 286         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
 287         if (!fa->flags_valid) {
 288                 /* 1 item for the inode */
 289                 trans = btrfs_start_transaction(root, 1);
 290                 if (IS_ERR(trans))
 291                         return PTR_ERR(trans);
 292                 goto update_flags;
 293         }
 294
 295         if (fsflags & FS_DIRSYNC_FL)
 296                 binode_flags |= BTRFS_INODE_DIRSYNC;
 297         else
 298                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 299         if (fsflags & FS_NOCOW_FL) {
 300                 if (S_ISREG(inode->i_mode)) {
 301                         /*
 302                          * It's safe to turn csums off here, no extents exist.
 303                          * Otherwise we want the flag to reflect the real COW
 304                          * status of the file and will not set it.
 305                          */
 306                         if (inode->i_size == 0)
 307                                 binode_flags |= BTRFS_INODE_NODATACOW |
 308                                                 BTRFS_INODE_NODATASUM;
 309                 } else {
 310                         binode_flags |= BTRFS_INODE_NODATACOW;
 311                 }
 312         } else {
 313                 /*
 314                  * Revert back under same assumptions as above
 315                  */
 316                 if (S_ISREG(inode->i_mode)) {
 317                         if (inode->i_size == 0)
 318                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 319                                                   BTRFS_INODE_NODATASUM);
 320                 } else {
 321                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 322                 }
 323         }
 324
 325         /*
 326          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 327          * flag may be changed automatically if compression code won't make
 328          * things smaller.
 329          */
 330         if (fsflags & FS_NOCOMP_FL) {
 331                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 332                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 333         } else if (fsflags & FS_COMPR_FL) {
 334
 335                 if (IS_SWAPFILE(inode))
 336                         return -ETXTBSY;
 337
 338                 binode_flags |= BTRFS_INODE_COMPRESS;
 339                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 340
 341                 comp = btrfs_compress_type2str(fs_info->compress_type);
 342                 if (!comp || comp[0] == 0)
 343                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 344         } else {
 345                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 346         }
 347
 348         /*
 349          * 1 for inode item
 350          * 2 for properties
 351          */
 352         trans = btrfs_start_transaction(root, 3);
 353         if (IS_ERR(trans))
 354                 return PTR_ERR(trans);
 355
 356         if (comp) {
 357                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 358                                      strlen(comp), 0);
 359                 if (ret) {
 360                         btrfs_abort_transaction(trans, ret);
 361                         goto out_end_trans;
 362                 }
 363         } else {
 364                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 365                                      0, 0);
 366                 if (ret && ret != -ENODATA) {
 367                         btrfs_abort_transaction(trans, ret);
 368                         goto out_end_trans;
 369                 }
 370         }
 371
 372 update_flags:
 373         binode->flags = binode_flags;
 374         btrfs_sync_inode_flags_to_i_flags(inode);
 375         inode_inc_iversion(inode);
 376         inode->i_ctime = current_time(inode);
 377         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 378
 379  out_end_trans:
 380         btrfs_end_transaction(trans);
 381         return ret;
 382 }
 383
 384 /*
 385  * Start exclusive operation @type, return true on success
 386  */
 387 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 388                         enum btrfs_exclusive_operation type)
 389 {
 390         bool ret = false;
 391
 392         spin_lock(&fs_info->super_lock);
 393         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
 394                 fs_info->exclusive_operation = type;
 395                 ret = true;
 396         }
 397         spin_unlock(&fs_info->super_lock);
 398
 399         return ret;
 400 }
 401
 402 /*
 403  * Conditionally allow to enter the exclusive operation in case it's compatible
 404  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
 405  * btrfs_exclop_finish.
 406  *
 407  * Compatibility:
 408  * - the same type is already running
 409  * - when trying to add a device and balance has been paused
 410  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
 411  *   must check the condition first that would allow none -> @type
 412  */
 413 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
 414                                  enum btrfs_exclusive_operation type)
 415 {
 416         spin_lock(&fs_info->super_lock);
 417         if (fs_info->exclusive_operation == type ||
 418             (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED &&
 419              type == BTRFS_EXCLOP_DEV_ADD))
 420                 return true;
 421
 422         spin_unlock(&fs_info->super_lock);
 423         return false;
 424 }
 425
 426 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
 427 {
 428         spin_unlock(&fs_info->super_lock);
 429 }
 430
 431 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 432 {
 433         spin_lock(&fs_info->super_lock);
 434         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 435         spin_unlock(&fs_info->super_lock);
 436         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 437 }
 438
 439 void btrfs_exclop_balance(struct btrfs_fs_info *fs_info,
 440                           enum btrfs_exclusive_operation op)
 441 {
 442         switch (op) {
 443         case BTRFS_EXCLOP_BALANCE_PAUSED:
 444                 spin_lock(&fs_info->super_lock);
 445                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE ||
 446                        fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD);
 447                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED;
 448                 spin_unlock(&fs_info->super_lock);
 449                 break;
 450         case BTRFS_EXCLOP_BALANCE:
 451                 spin_lock(&fs_info->super_lock);
 452                 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
 453                 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
 454                 spin_unlock(&fs_info->super_lock);
 455                 break;
 456         default:
 457                 btrfs_warn(fs_info,
 458                         "invalid exclop balance operation %d requested", op);
 459         }
 460 }
 461
 462 static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg)
 463 {
 464         return put_user(inode->i_generation, arg);
 465 }
 466
 467 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 468                                         void __user *arg)
 469 {
 470         struct btrfs_device *device;
 471         struct fstrim_range range;
 472         u64 minlen = ULLONG_MAX;
 473         u64 num_devices = 0;
 474         int ret;
 475
 476         if (!capable(CAP_SYS_ADMIN))
 477                 return -EPERM;
 478
 479         /*
 480          * btrfs_trim_block_group() depends on space cache, which is not
 481          * available in zoned filesystem. So, disallow fitrim on a zoned
 482          * filesystem for now.
 483          */
 484         if (btrfs_is_zoned(fs_info))
 485                 return -EOPNOTSUPP;
 486
 487         /*
 488          * If the fs is mounted with nologreplay, which requires it to be
 489          * mounted in RO mode as well, we can not allow discard on free space
 490          * inside block groups, because log trees refer to extents that are not
 491          * pinned in a block group's free space cache (pinning the extents is
 492          * precisely the first phase of replaying a log tree).
 493          */
 494         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 495                 return -EROFS;
 496
 497         rcu_read_lock();
 498         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 499                                 dev_list) {
 500                 if (!device->bdev || !bdev_max_discard_sectors(device->bdev))
 501                         continue;
 502                 num_devices++;
 503                 minlen = min_t(u64, bdev_discard_granularity(device->bdev),
 504                                     minlen);
 505         }
 506         rcu_read_unlock();
 507
 508         if (!num_devices)
 509                 return -EOPNOTSUPP;
 510         if (copy_from_user(&range, arg, sizeof(range)))
 511                 return -EFAULT;
 512
 513         /*
 514          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 515          * block group is in the logical address space, which can be any
 516          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 517          */
 518         if (range.len < fs_info->sb->s_blocksize)
 519                 return -EINVAL;
 520
 521         range.minlen = max(range.minlen, minlen);
 522         ret = btrfs_trim_fs(fs_info, &range);
 523         if (ret < 0)
 524                 return ret;
 525
 526         if (copy_to_user(arg, &range, sizeof(range)))
 527                 return -EFAULT;
 528
 529         return 0;
 530 }
 531
 532 int __pure btrfs_is_empty_uuid(u8 *uuid)
 533 {
 534         int i;
 535
 536         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 537                 if (uuid[i])
 538                         return 0;
 539         }
 540         return 1;
 541 }
 542
 543 /*
 544  * Calculate the number of transaction items to reserve for creating a subvolume
 545  * or snapshot, not including the inode, directory entries, or parent directory.
 546  */
 547 static unsigned int create_subvol_num_items(struct btrfs_qgroup_inherit *inherit)
 548 {
 549         /*
 550          * 1 to add root block
 551          * 1 to add root item
 552          * 1 to add root ref
 553          * 1 to add root backref
 554          * 1 to add UUID item
 555          * 1 to add qgroup info
 556          * 1 to add qgroup limit
 557          *
 558          * Ideally the last two would only be accounted if qgroups are enabled,
 559          * but that can change between now and the time we would insert them.
 560          */
 561         unsigned int num_items = 7;
 562
 563         if (inherit) {
 564                 /* 2 to add qgroup relations for each inherited qgroup */
 565                 num_items += 2 * inherit->num_qgroups;
 566         }
 567         return num_items;
 568 }
 569
 570 static noinline int create_subvol(struct user_namespace *mnt_userns,
 571                                   struct inode *dir, struct dentry *dentry,
 572                                   struct btrfs_qgroup_inherit *inherit)
 573 {
 574         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 575         struct btrfs_trans_handle *trans;
 576         struct btrfs_key key;
 577         struct btrfs_root_item *root_item;
 578         struct btrfs_inode_item *inode_item;
 579         struct extent_buffer *leaf;
 580         struct btrfs_root *root = BTRFS_I(dir)->root;
 581         struct btrfs_root *new_root;
 582         struct btrfs_block_rsv block_rsv;
 583         struct timespec64 cur_time = current_time(dir);
 584         struct btrfs_new_inode_args new_inode_args = {
 585                 .dir = dir,
 586                 .dentry = dentry,
 587                 .subvol = true,
 588         };
 589         unsigned int trans_num_items;
 590         int ret;
 591         dev_t anon_dev;
 592         u64 objectid;
 593
 594         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 595         if (!root_item)
 596                 return -ENOMEM;
 597
 598         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 599         if (ret)
 600                 goto out_root_item;
 601
 602         /*
 603          * Don't create subvolume whose level is not zero. Or qgroup will be
 604          * screwed up since it assumes subvolume qgroup's level to be 0.
 605          */
 606         if (btrfs_qgroup_level(objectid)) {
 607                 ret = -ENOSPC;
 608                 goto out_root_item;
 609         }
 610
 611         ret = get_anon_bdev(&anon_dev);
 612         if (ret < 0)
 613                 goto out_root_item;
 614
 615         new_inode_args.inode = btrfs_new_subvol_inode(mnt_userns, dir);
 616         if (!new_inode_args.inode) {
 617                 ret = -ENOMEM;
 618                 goto out_anon_dev;
 619         }
 620         ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
 621         if (ret)
 622                 goto out_inode;
 623         trans_num_items += create_subvol_num_items(inherit);
 624
 625         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 626         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv,
 627                                                trans_num_items, false);
 628         if (ret)
 629                 goto out_new_inode_args;
 630
 631         trans = btrfs_start_transaction(root, 0);
 632         if (IS_ERR(trans)) {
 633                 ret = PTR_ERR(trans);
 634                 btrfs_subvolume_release_metadata(root, &block_rsv);
 635                 goto out_new_inode_args;
 636         }
 637         trans->block_rsv = &block_rsv;
 638         trans->bytes_reserved = block_rsv.size;
 639
 640         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 641         if (ret)
 642                 goto out;
 643
 644         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 645                                       BTRFS_NESTING_NORMAL);
 646         if (IS_ERR(leaf)) {
 647                 ret = PTR_ERR(leaf);
 648                 goto out;
 649         }
 650
 651         btrfs_mark_buffer_dirty(leaf);
 652
 653         inode_item = &root_item->inode;
 654         btrfs_set_stack_inode_generation(inode_item, 1);
 655         btrfs_set_stack_inode_size(inode_item, 3);
 656         btrfs_set_stack_inode_nlink(inode_item, 1);
 657         btrfs_set_stack_inode_nbytes(inode_item,
 658                                      fs_info->nodesize);
 659         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 660
 661         btrfs_set_root_flags(root_item, 0);
 662         btrfs_set_root_limit(root_item, 0);
 663         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 664
 665         btrfs_set_root_bytenr(root_item, leaf->start);
 666         btrfs_set_root_generation(root_item, trans->transid);
 667         btrfs_set_root_level(root_item, 0);
 668         btrfs_set_root_refs(root_item, 1);
 669         btrfs_set_root_used(root_item, leaf->len);
 670         btrfs_set_root_last_snapshot(root_item, 0);
 671
 672         btrfs_set_root_generation_v2(root_item,
 673                         btrfs_root_generation(root_item));
 674         generate_random_guid(root_item->uuid);
 675         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 676         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 677         root_item->ctime = root_item->otime;
 678         btrfs_set_root_ctransid(root_item, trans->transid);
 679         btrfs_set_root_otransid(root_item, trans->transid);
 680
 681         btrfs_tree_unlock(leaf);
 682
 683         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 684
 685         key.objectid = objectid;
 686         key.offset = 0;
 687         key.type = BTRFS_ROOT_ITEM_KEY;
 688         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 689                                 root_item);
 690         if (ret) {
 691                 /*
 692                  * Since we don't abort the transaction in this case, free the
 693                  * tree block so that we don't leak space and leave the
 694                  * filesystem in an inconsistent state (an extent item in the
 695                  * extent tree with a backreference for a root that does not
 696                  * exists).
 697                  */
 698                 btrfs_tree_lock(leaf);
 699                 btrfs_clean_tree_block(leaf);
 700                 btrfs_tree_unlock(leaf);
 701                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 702                 free_extent_buffer(leaf);
 703                 goto out;
 704         }
 705
 706         free_extent_buffer(leaf);
 707         leaf = NULL;
 708
 709         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 710         if (IS_ERR(new_root)) {
 711                 ret = PTR_ERR(new_root);
 712                 btrfs_abort_transaction(trans, ret);
 713                 goto out;
 714         }
 715         /* anon_dev is owned by new_root now. */
 716         anon_dev = 0;
 717         BTRFS_I(new_inode_args.inode)->root = new_root;
 718         /* ... and new_root is owned by new_inode_args.inode now. */
 719
 720         ret = btrfs_record_root_in_trans(trans, new_root);
 721         if (ret) {
 722                 btrfs_abort_transaction(trans, ret);
 723                 goto out;
 724         }
 725
 726         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 727                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 728         if (ret) {
 729                 btrfs_abort_transaction(trans, ret);
 730                 goto out;
 731         }
 732
 733         ret = btrfs_create_new_inode(trans, &new_inode_args);
 734         if (ret) {
 735                 btrfs_abort_transaction(trans, ret);
 736                 goto out;
 737         }
 738
 739         d_instantiate_new(dentry, new_inode_args.inode);
 740         new_inode_args.inode = NULL;
 741
 742 out:
 743         trans->block_rsv = NULL;
 744         trans->bytes_reserved = 0;
 745         btrfs_subvolume_release_metadata(root, &block_rsv);
 746
 747         if (ret)
 748                 btrfs_end_transaction(trans);
 749         else
 750                 ret = btrfs_commit_transaction(trans);
 751 out_new_inode_args:
 752         btrfs_new_inode_args_destroy(&new_inode_args);
 753 out_inode:
 754         iput(new_inode_args.inode);
 755 out_anon_dev:
 756         if (anon_dev)
 757                 free_anon_bdev(anon_dev);
 758 out_root_item:
 759         kfree(root_item);
 760         return ret;
 761 }
 762
 763 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 764                            struct dentry *dentry, bool readonly,
 765                            struct btrfs_qgroup_inherit *inherit)
 766 {
 767         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 768         struct inode *inode;
 769         struct btrfs_pending_snapshot *pending_snapshot;
 770         unsigned int trans_num_items;
 771         struct btrfs_trans_handle *trans;
 772         int ret;
 773
 774         /* We do not support snapshotting right now. */
 775         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 776                 btrfs_warn(fs_info,
 777                            "extent tree v2 doesn't support snapshotting yet");
 778                 return -EOPNOTSUPP;
 779         }
 780
 781         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 782                 return -EINVAL;
 783
 784         if (atomic_read(&root->nr_swapfiles)) {
 785                 btrfs_warn(fs_info,
 786                            "cannot snapshot subvolume with active swapfile");
 787                 return -ETXTBSY;
 788         }
 789
 790         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 791         if (!pending_snapshot)
 792                 return -ENOMEM;
 793
 794         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 795         if (ret < 0)
 796                 goto free_pending;
 797         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 798                         GFP_KERNEL);
 799         pending_snapshot->path = btrfs_alloc_path();
 800         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 801                 ret = -ENOMEM;
 802                 goto free_pending;
 803         }
 804
 805         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 806                              BTRFS_BLOCK_RSV_TEMP);
 807         /*
 808          * 1 to add dir item
 809          * 1 to add dir index
 810          * 1 to update parent inode item
 811          */
 812         trans_num_items = create_subvol_num_items(inherit) + 3;
 813         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 814                                                &pending_snapshot->block_rsv,
 815                                                trans_num_items, false);
 816         if (ret)
 817                 goto free_pending;
 818
 819         pending_snapshot->dentry = dentry;
 820         pending_snapshot->root = root;
 821         pending_snapshot->readonly = readonly;
 822         pending_snapshot->dir = dir;
 823         pending_snapshot->inherit = inherit;
 824
 825         trans = btrfs_start_transaction(root, 0);
 826         if (IS_ERR(trans)) {
 827                 ret = PTR_ERR(trans);
 828                 goto fail;
 829         }
 830
 831         trans->pending_snapshot = pending_snapshot;
 832
 833         ret = btrfs_commit_transaction(trans);
 834         if (ret)
 835                 goto fail;
 836
 837         ret = pending_snapshot->error;
 838         if (ret)
 839                 goto fail;
 840
 841         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 842         if (ret)
 843                 goto fail;
 844
 845         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 846         if (IS_ERR(inode)) {
 847                 ret = PTR_ERR(inode);
 848                 goto fail;
 849         }
 850
 851         d_instantiate(dentry, inode);
 852         ret = 0;
 853         pending_snapshot->anon_dev = 0;
 854 fail:
 855         /* Prevent double freeing of anon_dev */
 856         if (ret && pending_snapshot->snap)
 857                 pending_snapshot->snap->anon_dev = 0;
 858         btrfs_put_root(pending_snapshot->snap);
 859         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 860 free_pending:
 861         if (pending_snapshot->anon_dev)
 862                 free_anon_bdev(pending_snapshot->anon_dev);
 863         kfree(pending_snapshot->root_item);
 864         btrfs_free_path(pending_snapshot->path);
 865         kfree(pending_snapshot);
 866
 867         return ret;
 868 }
 869
 870 /*  copy of may_delete in fs/namei.c()
 871  *      Check whether we can remove a link victim from directory dir, check
 872  *  whether the type of victim is right.
 873  *  1. We can't do it if dir is read-only (done in permission())
 874  *  2. We should have write and exec permissions on dir
 875  *  3. We can't remove anything from append-only dir
 876  *  4. We can't do anything with immutable dir (done in permission())
 877  *  5. If the sticky bit on dir is set we should either
 878  *      a. be owner of dir, or
 879  *      b. be owner of victim, or
 880  *      c. have CAP_FOWNER capability
 881  *  6. If the victim is append-only or immutable we can't do anything with
 882  *     links pointing to it.
 883  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 884  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 885  *  9. We can't remove a root or mountpoint.
 886  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 887  *     nfs_async_unlink().
 888  */
 889
 890 static int btrfs_may_delete(struct user_namespace *mnt_userns,
 891                             struct inode *dir, struct dentry *victim, int isdir)
 892 {
 893         int error;
 894
 895         if (d_really_is_negative(victim))
 896                 return -ENOENT;
 897
 898         BUG_ON(d_inode(victim->d_parent) != dir);
 899         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 900
 901         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 902         if (error)
 903                 return error;
 904         if (IS_APPEND(dir))
 905                 return -EPERM;
 906         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
 907             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 908             IS_SWAPFILE(d_inode(victim)))
 909                 return -EPERM;
 910         if (isdir) {
 911                 if (!d_is_dir(victim))
 912                         return -ENOTDIR;
 913                 if (IS_ROOT(victim))
 914                         return -EBUSY;
 915         } else if (d_is_dir(victim))
 916                 return -EISDIR;
 917         if (IS_DEADDIR(dir))
 918                 return -ENOENT;
 919         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 920                 return -EBUSY;
 921         return 0;
 922 }
 923
 924 /* copy of may_create in fs/namei.c() */
 925 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
 926                                    struct inode *dir, struct dentry *child)
 927 {
 928         if (d_really_is_positive(child))
 929                 return -EEXIST;
 930         if (IS_DEADDIR(dir))
 931                 return -ENOENT;
 932         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
 933                 return -EOVERFLOW;
 934         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 935 }
 936
 937 /*
 938  * Create a new subvolume below @parent.  This is largely modeled after
 939  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 940  * inside this filesystem so it's quite a bit simpler.
 941  */
 942 static noinline int btrfs_mksubvol(const struct path *parent,
 943                                    struct user_namespace *mnt_userns,
 944                                    const char *name, int namelen,
 945                                    struct btrfs_root *snap_src,
 946                                    bool readonly,
 947                                    struct btrfs_qgroup_inherit *inherit)
 948 {
 949         struct inode *dir = d_inode(parent->dentry);
 950         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 951         struct dentry *dentry;
 952         int error;
 953
 954         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 955         if (error == -EINTR)
 956                 return error;
 957
 958         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
 959         error = PTR_ERR(dentry);
 960         if (IS_ERR(dentry))
 961                 goto out_unlock;
 962
 963         error = btrfs_may_create(mnt_userns, dir, dentry);
 964         if (error)
 965                 goto out_dput;
 966
 967         /*
 968          * even if this name doesn't exist, we may get hash collisions.
 969          * check for them now when we can safely fail
 970          */
 971         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 972                                                dir->i_ino, name,
 973                                                namelen);
 974         if (error)
 975                 goto out_dput;
 976
 977         down_read(&fs_info->subvol_sem);
 978
 979         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 980                 goto out_up_read;
 981
 982         if (snap_src)
 983                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 984         else
 985                 error = create_subvol(mnt_userns, dir, dentry, inherit);
 986
 987         if (!error)
 988                 fsnotify_mkdir(dir, dentry);
 989 out_up_read:
 990         up_read(&fs_info->subvol_sem);
 991 out_dput:
 992         dput(dentry);
 993 out_unlock:
 994         btrfs_inode_unlock(dir, 0);
 995         return error;
 996 }
 997
 998 static noinline int btrfs_mksnapshot(const struct path *parent,
 999                                    struct user_namespace *mnt_userns,
1000                                    const char *name, int namelen,
1001                                    struct btrfs_root *root,
1002                                    bool readonly,
1003                                    struct btrfs_qgroup_inherit *inherit)
1004 {
1005         int ret;
1006         bool snapshot_force_cow = false;
1007
1008         /*
1009          * Force new buffered writes to reserve space even when NOCOW is
1010          * possible. This is to avoid later writeback (running dealloc) to
1011          * fallback to COW mode and unexpectedly fail with ENOSPC.
1012          */
1013         btrfs_drew_read_lock(&root->snapshot_lock);
1014
1015         ret = btrfs_start_delalloc_snapshot(root, false);
1016         if (ret)
1017                 goto out;
1018
1019         /*
1020          * All previous writes have started writeback in NOCOW mode, so now
1021          * we force future writes to fallback to COW mode during snapshot
1022          * creation.
1023          */
1024         atomic_inc(&root->snapshot_force_cow);
1025         snapshot_force_cow = true;
1026
1027         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1028
1029         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
1030                              root, readonly, inherit);
1031 out:
1032         if (snapshot_force_cow)
1033                 atomic_dec(&root->snapshot_force_cow);
1034         btrfs_drew_read_unlock(&root->snapshot_lock);
1035         return ret;
1036 }
1037
1038 /*
1039  * Defrag specific helper to get an extent map.
1040  *
1041  * Differences between this and btrfs_get_extent() are:
1042  *
1043  * - No extent_map will be added to inode->extent_tree
1044  *   To reduce memory usage in the long run.
1045  *
1046  * - Extra optimization to skip file extents older than @newer_than
1047  *   By using btrfs_search_forward() we can skip entire file ranges that
1048  *   have extents created in past transactions, because btrfs_search_forward()
1049  *   will not visit leaves and nodes with a generation smaller than given
1050  *   minimal generation threshold (@newer_than).
1051  *
1052  * Return valid em if we find a file extent matching the requirement.
1053  * Return NULL if we can not find a file extent matching the requirement.
1054  *
1055  * Return ERR_PTR() for error.
1056  */
1057 static struct extent_map *defrag_get_extent(struct btrfs_inode *inode,
1058                                             u64 start, u64 newer_than)
1059 {
1060         struct btrfs_root *root = inode->root;
1061         struct btrfs_file_extent_item *fi;
1062         struct btrfs_path path = { 0 };
1063         struct extent_map *em;
1064         struct btrfs_key key;
1065         u64 ino = btrfs_ino(inode);
1066         int ret;
1067
1068         em = alloc_extent_map();
1069         if (!em) {
1070                 ret = -ENOMEM;
1071                 goto err;
1072         }
1073
1074         key.objectid = ino;
1075         key.type = BTRFS_EXTENT_DATA_KEY;
1076         key.offset = start;
1077
1078         if (newer_than) {
1079                 ret = btrfs_search_forward(root, &key, &path, newer_than);
1080                 if (ret < 0)
1081                         goto err;
1082                 /* Can't find anything newer */
1083                 if (ret > 0)
1084                         goto not_found;
1085         } else {
1086                 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1087                 if (ret < 0)
1088                         goto err;
1089         }
1090         if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
1091                 /*
1092                  * If btrfs_search_slot() makes path to point beyond nritems,
1093                  * we should not have an empty leaf, as this inode must at
1094                  * least have its INODE_ITEM.
1095                  */
1096                 ASSERT(btrfs_header_nritems(path.nodes[0]));
1097                 path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1;
1098         }
1099         btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1100         /* Perfect match, no need to go one slot back */
1101         if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY &&
1102             key.offset == start)
1103                 goto iterate;
1104
1105         /* We didn't find a perfect match, needs to go one slot back */
1106         if (path.slots[0] > 0) {
1107                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1108                 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
1109                         path.slots[0]--;
1110         }
1111
1112 iterate:
1113         /* Iterate through the path to find a file extent covering @start */
1114         while (true) {
1115                 u64 extent_end;
1116
1117                 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0]))
1118                         goto next;
1119
1120                 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
1121
1122                 /*
1123                  * We may go one slot back to INODE_REF/XATTR item, then
1124                  * need to go forward until we reach an EXTENT_DATA.
1125                  * But we should still has the correct ino as key.objectid.
1126                  */
1127                 if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY)
1128                         goto next;
1129
1130                 /* It's beyond our target range, definitely not extent found */
1131                 if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY)
1132                         goto not_found;
1133
1134                 /*
1135                  *      |       |<- File extent ->|
1136                  *      \- start
1137                  *
1138                  * This means there is a hole between start and key.offset.
1139                  */
1140                 if (key.offset > start) {
1141                         em->start = start;
1142                         em->orig_start = start;
1143                         em->block_start = EXTENT_MAP_HOLE;
1144                         em->len = key.offset - start;
1145                         break;
1146                 }
1147
1148                 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
1149                                     struct btrfs_file_extent_item);
1150                 extent_end = btrfs_file_extent_end(&path);
1151
1152                 /*
1153                  *      |<- file extent ->|     |
1154                  *                              \- start
1155                  *
1156                  * We haven't reached start, search next slot.
1157                  */
1158                 if (extent_end <= start)
1159                         goto next;
1160
1161                 /* Now this extent covers @start, convert it to em */
1162                 btrfs_extent_item_to_extent_map(inode, &path, fi, false, em);
1163                 break;
1164 next:
1165                 ret = btrfs_next_item(root, &path);
1166                 if (ret < 0)
1167                         goto err;
1168                 if (ret > 0)
1169                         goto not_found;
1170         }
1171         btrfs_release_path(&path);
1172         return em;
1173
1174 not_found:
1175         btrfs_release_path(&path);
1176         free_extent_map(em);
1177         return NULL;
1178
1179 err:
1180         btrfs_release_path(&path);
1181         free_extent_map(em);
1182         return ERR_PTR(ret);
1183 }
1184
1185 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start,
1186                                                u64 newer_than, bool locked)
1187 {
1188         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1189         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1190         struct extent_map *em;
1191         const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize;
1192
1193         /*
1194          * hopefully we have this extent in the tree already, try without
1195          * the full extent lock
1196          */
1197         read_lock(&em_tree->lock);
1198         em = lookup_extent_mapping(em_tree, start, sectorsize);
1199         read_unlock(&em_tree->lock);
1200
1201         /*
1202          * We can get a merged extent, in that case, we need to re-search
1203          * tree to get the original em for defrag.
1204          *
1205          * If @newer_than is 0 or em::generation < newer_than, we can trust
1206          * this em, as either we don't care about the generation, or the
1207          * merged extent map will be rejected anyway.
1208          */
1209         if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) &&
1210             newer_than && em->generation >= newer_than) {
1211                 free_extent_map(em);
1212                 em = NULL;
1213         }
1214
1215         if (!em) {
1216                 struct extent_state *cached = NULL;
1217                 u64 end = start + sectorsize - 1;
1218
1219                 /* get the big lock and read metadata off disk */
1220                 if (!locked)
1221                         lock_extent_bits(io_tree, start, end, &cached);
1222                 em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
1223                 if (!locked)
1224                         unlock_extent_cached(io_tree, start, end, &cached);
1225
1226                 if (IS_ERR(em))
1227                         return NULL;
1228         }
1229
1230         return em;
1231 }
1232
1233 static u32 get_extent_max_capacity(const struct btrfs_fs_info *fs_info,
1234                                    const struct extent_map *em)
1235 {
1236         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1237                 return BTRFS_MAX_COMPRESSED;
1238         return fs_info->max_extent_size;
1239 }
1240
1241 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em,
1242                                      u32 extent_thresh, u64 newer_than, bool locked)
1243 {
1244         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1245         struct extent_map *next;
1246         bool ret = false;
1247
1248         /* this is the last extent */
1249         if (em->start + em->len >= i_size_read(inode))
1250                 return false;
1251
1252         /*
1253          * Here we need to pass @newer_then when checking the next extent, or
1254          * we will hit a case we mark current extent for defrag, but the next
1255          * one will not be a target.
1256          * This will just cause extra IO without really reducing the fragments.
1257          */
1258         next = defrag_lookup_extent(inode, em->start + em->len, newer_than, locked);
1259         /* No more em or hole */
1260         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1261                 goto out;
1262         if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags))
1263                 goto out;
1264         /*
1265          * If the next extent is at its max capacity, defragging current extent
1266          * makes no sense, as the total number of extents won't change.
1267          */
1268         if (next->len >= get_extent_max_capacity(fs_info, em))
1269                 goto out;
1270         /* Skip older extent */
1271         if (next->generation < newer_than)
1272                 goto out;
1273         /* Also check extent size */
1274         if (next->len >= extent_thresh)
1275                 goto out;
1276
1277         ret = true;
1278 out:
1279         free_extent_map(next);
1280         return ret;
1281 }
1282
1283 /*
1284  * Prepare one page to be defragged.
1285  *
1286  * This will ensure:
1287  *
1288  * - Returned page is locked and has been set up properly.
1289  * - No ordered extent exists in the page.
1290  * - The page is uptodate.
1291  *
1292  * NOTE: Caller should also wait for page writeback after the cluster is
1293  * prepared, here we don't do writeback wait for each page.
1294  */
1295 static struct page *defrag_prepare_one_page(struct btrfs_inode *inode,
1296                                             pgoff_t index)
1297 {
1298         struct address_space *mapping = inode->vfs_inode.i_mapping;
1299         gfp_t mask = btrfs_alloc_write_mask(mapping);
1300         u64 page_start = (u64)index << PAGE_SHIFT;
1301         u64 page_end = page_start + PAGE_SIZE - 1;
1302         struct extent_state *cached_state = NULL;
1303         struct page *page;
1304         int ret;
1305
1306 again:
1307         page = find_or_create_page(mapping, index, mask);
1308         if (!page)
1309                 return ERR_PTR(-ENOMEM);
1310
1311         /*
1312          * Since we can defragment files opened read-only, we can encounter
1313          * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We
1314          * can't do I/O using huge pages yet, so return an error for now.
1315          * Filesystem transparent huge pages are typically only used for
1316          * executables that explicitly enable them, so this isn't very
1317          * restrictive.
1318          */
1319         if (PageCompound(page)) {
1320                 unlock_page(page);
1321                 put_page(page);
1322                 return ERR_PTR(-ETXTBSY);
1323         }
1324
1325         ret = set_page_extent_mapped(page);
1326         if (ret < 0) {
1327                 unlock_page(page);
1328                 put_page(page);
1329                 return ERR_PTR(ret);
1330         }
1331
1332         /* Wait for any existing ordered extent in the range */
1333         while (1) {
1334                 struct btrfs_ordered_extent *ordered;
1335
1336                 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
1337                 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
1338                 unlock_extent_cached(&inode->io_tree, page_start, page_end,
1339                                      &cached_state);
1340                 if (!ordered)
1341                         break;
1342
1343                 unlock_page(page);
1344                 btrfs_start_ordered_extent(ordered, 1);
1345                 btrfs_put_ordered_extent(ordered);
1346                 lock_page(page);
1347                 /*
1348                  * We unlocked the page above, so we need check if it was
1349                  * released or not.
1350                  */
1351                 if (page->mapping != mapping || !PagePrivate(page)) {
1352                         unlock_page(page);
1353                         put_page(page);
1354                         goto again;
1355                 }
1356         }
1357
1358         /*
1359          * Now the page range has no ordered extent any more.  Read the page to
1360          * make it uptodate.
1361          */
1362         if (!PageUptodate(page)) {
1363                 btrfs_read_folio(NULL, page_folio(page));
1364                 lock_page(page);
1365                 if (page->mapping != mapping || !PagePrivate(page)) {
1366                         unlock_page(page);
1367                         put_page(page);
1368                         goto again;
1369                 }
1370                 if (!PageUptodate(page)) {
1371                         unlock_page(page);
1372                         put_page(page);
1373                         return ERR_PTR(-EIO);
1374                 }
1375         }
1376         return page;
1377 }
1378
1379 struct defrag_target_range {
1380         struct list_head list;
1381         u64 start;
1382         u64 len;
1383 };
1384
1385 /*
1386  * Collect all valid target extents.
1387  *
1388  * @start:         file offset to lookup
1389  * @len:           length to lookup
1390  * @extent_thresh: file extent size threshold, any extent size >= this value
1391  *                 will be ignored
1392  * @newer_than:    only defrag extents newer than this value
1393  * @do_compress:   whether the defrag is doing compression
1394  *                 if true, @extent_thresh will be ignored and all regular
1395  *                 file extents meeting @newer_than will be targets.
1396  * @locked:        if the range has already held extent lock
1397  * @target_list:   list of targets file extents
1398  */
1399 static int defrag_collect_targets(struct btrfs_inode *inode,
1400                                   u64 start, u64 len, u32 extent_thresh,
1401                                   u64 newer_than, bool do_compress,
1402                                   bool locked, struct list_head *target_list,
1403                                   u64 *last_scanned_ret)
1404 {
1405         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1406         bool last_is_target = false;
1407         u64 cur = start;
1408         int ret = 0;
1409
1410         while (cur < start + len) {
1411                 struct extent_map *em;
1412                 struct defrag_target_range *new;
1413                 bool next_mergeable = true;
1414                 u64 range_len;
1415
1416                 last_is_target = false;
1417                 em = defrag_lookup_extent(&inode->vfs_inode, cur,
1418                                           newer_than, locked);
1419                 if (!em)
1420                         break;
1421
1422                 /*
1423                  * If the file extent is an inlined one, we may still want to
1424                  * defrag it (fallthrough) if it will cause a regular extent.
1425                  * This is for users who want to convert inline extents to
1426                  * regular ones through max_inline= mount option.
1427                  */
1428                 if (em->block_start == EXTENT_MAP_INLINE &&
1429                     em->len <= inode->root->fs_info->max_inline)
1430                         goto next;
1431
1432                 /* Skip hole/delalloc/preallocated extents */
1433                 if (em->block_start == EXTENT_MAP_HOLE ||
1434                     em->block_start == EXTENT_MAP_DELALLOC ||
1435                     test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1436                         goto next;
1437
1438                 /* Skip older extent */
1439                 if (em->generation < newer_than)
1440                         goto next;
1441
1442                 /* This em is under writeback, no need to defrag */
1443                 if (em->generation == (u64)-1)
1444                         goto next;
1445
1446                 /*
1447                  * Our start offset might be in the middle of an existing extent
1448                  * map, so take that into account.
1449                  */
1450                 range_len = em->len - (cur - em->start);
1451                 /*
1452                  * If this range of the extent map is already flagged for delalloc,
1453                  * skip it, because:
1454                  *
1455                  * 1) We could deadlock later, when trying to reserve space for
1456                  *    delalloc, because in case we can't immediately reserve space
1457                  *    the flusher can start delalloc and wait for the respective
1458                  *    ordered extents to complete. The deadlock would happen
1459                  *    because we do the space reservation while holding the range
1460                  *    locked, and starting writeback, or finishing an ordered
1461                  *    extent, requires locking the range;
1462                  *
1463                  * 2) If there's delalloc there, it means there's dirty pages for
1464                  *    which writeback has not started yet (we clean the delalloc
1465                  *    flag when starting writeback and after creating an ordered
1466                  *    extent). If we mark pages in an adjacent range for defrag,
1467                  *    then we will have a larger contiguous range for delalloc,
1468                  *    very likely resulting in a larger extent after writeback is
1469                  *    triggered (except in a case of free space fragmentation).
1470                  */
1471                 if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1,
1472                                    EXTENT_DELALLOC, 0, NULL))
1473                         goto next;
1474
1475                 /*
1476                  * For do_compress case, we want to compress all valid file
1477                  * extents, thus no @extent_thresh or mergeable check.
1478                  */
1479                 if (do_compress)
1480                         goto add;
1481
1482                 /* Skip too large extent */
1483                 if (range_len >= extent_thresh)
1484                         goto next;
1485
1486                 /*
1487                  * Skip extents already at its max capacity, this is mostly for
1488                  * compressed extents, which max cap is only 128K.
1489                  */
1490                 if (em->len >= get_extent_max_capacity(fs_info, em))
1491                         goto next;
1492
1493                 /*
1494                  * Normally there are no more extents after an inline one, thus
1495                  * @next_mergeable will normally be false and not defragged.
1496                  * So if an inline extent passed all above checks, just add it
1497                  * for defrag, and be converted to regular extents.
1498                  */
1499                 if (em->block_start == EXTENT_MAP_INLINE)
1500                         goto add;
1501
1502                 next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em,
1503                                                 extent_thresh, newer_than, locked);
1504                 if (!next_mergeable) {
1505                         struct defrag_target_range *last;
1506
1507                         /* Empty target list, no way to merge with last entry */
1508                         if (list_empty(target_list))
1509                                 goto next;
1510                         last = list_entry(target_list->prev,
1511                                           struct defrag_target_range, list);
1512                         /* Not mergeable with last entry */
1513                         if (last->start + last->len != cur)
1514                                 goto next;
1515
1516                         /* Mergeable, fall through to add it to @target_list. */
1517                 }
1518
1519 add:
1520                 last_is_target = true;
1521                 range_len = min(extent_map_end(em), start + len) - cur;
1522                 /*
1523                  * This one is a good target, check if it can be merged into
1524                  * last range of the target list.
1525                  */
1526                 if (!list_empty(target_list)) {
1527                         struct defrag_target_range *last;
1528
1529                         last = list_entry(target_list->prev,
1530                                           struct defrag_target_range, list);
1531                         ASSERT(last->start + last->len <= cur);
1532                         if (last->start + last->len == cur) {
1533                                 /* Mergeable, enlarge the last entry */
1534                                 last->len += range_len;
1535                                 goto next;
1536                         }
1537                         /* Fall through to allocate a new entry */
1538                 }
1539
1540                 /* Allocate new defrag_target_range */
1541                 new = kmalloc(sizeof(*new), GFP_NOFS);
1542                 if (!new) {
1543                         free_extent_map(em);
1544                         ret = -ENOMEM;
1545                         break;
1546                 }
1547                 new->start = cur;
1548                 new->len = range_len;
1549                 list_add_tail(&new->list, target_list);
1550
1551 next:
1552                 cur = extent_map_end(em);
1553                 free_extent_map(em);
1554         }
1555         if (ret < 0) {
1556                 struct defrag_target_range *entry;
1557                 struct defrag_target_range *tmp;
1558
1559                 list_for_each_entry_safe(entry, tmp, target_list, list) {
1560                         list_del_init(&entry->list);
1561                         kfree(entry);
1562                 }
1563         }
1564         if (!ret && last_scanned_ret) {
1565                 /*
1566                  * If the last extent is not a target, the caller can skip to
1567                  * the end of that extent.
1568                  * Otherwise, we can only go the end of the specified range.
1569                  */
1570                 if (!last_is_target)
1571                         *last_scanned_ret = max(cur, *last_scanned_ret);
1572                 else
1573                         *last_scanned_ret = max(start + len, *last_scanned_ret);
1574         }
1575         return ret;
1576 }
1577
1578 #define CLUSTER_SIZE    (SZ_256K)
1579 static_assert(IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE));
1580
1581 /*
1582  * Defrag one contiguous target range.
1583  *
1584  * @inode:      target inode
1585  * @target:     target range to defrag
1586  * @pages:      locked pages covering the defrag range
1587  * @nr_pages:   number of locked pages
1588  *
1589  * Caller should ensure:
1590  *
1591  * - Pages are prepared
1592  *   Pages should be locked, no ordered extent in the pages range,
1593  *   no writeback.
1594  *
1595  * - Extent bits are locked
1596  */
1597 static int defrag_one_locked_target(struct btrfs_inode *inode,
1598                                     struct defrag_target_range *target,
1599                                     struct page **pages, int nr_pages,
1600                                     struct extent_state **cached_state)
1601 {
1602         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1603         struct extent_changeset *data_reserved = NULL;
1604         const u64 start = target->start;
1605         const u64 len = target->len;
1606         unsigned long last_index = (start + len - 1) >> PAGE_SHIFT;
1607         unsigned long start_index = start >> PAGE_SHIFT;
1608         unsigned long first_index = page_index(pages[0]);
1609         int ret = 0;
1610         int i;
1611
1612         ASSERT(last_index - first_index + 1 <= nr_pages);
1613
1614         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len);
1615         if (ret < 0)
1616                 return ret;
1617         clear_extent_bit(&inode->io_tree, start, start + len - 1,
1618                          EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1619                          EXTENT_DEFRAG, 0, 0, cached_state);
1620         set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state);
1621
1622         /* Update the page status */
1623         for (i = start_index - first_index; i <= last_index - first_index; i++) {
1624                 ClearPageChecked(pages[i]);
1625                 btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len);
1626         }
1627         btrfs_delalloc_release_extents(inode, len);
1628         extent_changeset_free(data_reserved);
1629
1630         return ret;
1631 }
1632
1633 static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len,
1634                             u32 extent_thresh, u64 newer_than, bool do_compress,
1635                             u64 *last_scanned_ret)
1636 {
1637         struct extent_state *cached_state = NULL;
1638         struct defrag_target_range *entry;
1639         struct defrag_target_range *tmp;
1640         LIST_HEAD(target_list);
1641         struct page **pages;
1642         const u32 sectorsize = inode->root->fs_info->sectorsize;
1643         u64 last_index = (start + len - 1) >> PAGE_SHIFT;
1644         u64 start_index = start >> PAGE_SHIFT;
1645         unsigned int nr_pages = last_index - start_index + 1;
1646         int ret = 0;
1647         int i;
1648
1649         ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE);
1650         ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize));
1651
1652         pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
1653         if (!pages)
1654                 return -ENOMEM;
1655
1656         /* Prepare all pages */
1657         for (i = 0; i < nr_pages; i++) {
1658                 pages[i] = defrag_prepare_one_page(inode, start_index + i);
1659                 if (IS_ERR(pages[i])) {
1660                         ret = PTR_ERR(pages[i]);
1661                         pages[i] = NULL;
1662                         goto free_pages;
1663                 }
1664         }
1665         for (i = 0; i < nr_pages; i++)
1666                 wait_on_page_writeback(pages[i]);
1667
1668         /* Lock the pages range */
1669         lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT,
1670                          (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1671                          &cached_state);
1672         /*
1673          * Now we have a consistent view about the extent map, re-check
1674          * which range really needs to be defragged.
1675          *
1676          * And this time we have extent locked already, pass @locked = true
1677          * so that we won't relock the extent range and cause deadlock.
1678          */
1679         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1680                                      newer_than, do_compress, true,
1681                                      &target_list, last_scanned_ret);
1682         if (ret < 0)
1683                 goto unlock_extent;
1684
1685         list_for_each_entry(entry, &target_list, list) {
1686                 ret = defrag_one_locked_target(inode, entry, pages, nr_pages,
1687                                                &cached_state);
1688                 if (ret < 0)
1689                         break;
1690         }
1691
1692         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1693                 list_del_init(&entry->list);
1694                 kfree(entry);
1695         }
1696 unlock_extent:
1697         unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT,
1698                              (last_index << PAGE_SHIFT) + PAGE_SIZE - 1,
1699                              &cached_state);
1700 free_pages:
1701         for (i = 0; i < nr_pages; i++) {
1702                 if (pages[i]) {
1703                         unlock_page(pages[i]);
1704                         put_page(pages[i]);
1705                 }
1706         }
1707         kfree(pages);
1708         return ret;
1709 }
1710
1711 static int defrag_one_cluster(struct btrfs_inode *inode,
1712                               struct file_ra_state *ra,
1713                               u64 start, u32 len, u32 extent_thresh,
1714                               u64 newer_than, bool do_compress,
1715                               unsigned long *sectors_defragged,
1716                               unsigned long max_sectors,
1717                               u64 *last_scanned_ret)
1718 {
1719         const u32 sectorsize = inode->root->fs_info->sectorsize;
1720         struct defrag_target_range *entry;
1721         struct defrag_target_range *tmp;
1722         LIST_HEAD(target_list);
1723         int ret;
1724
1725         ret = defrag_collect_targets(inode, start, len, extent_thresh,
1726                                      newer_than, do_compress, false,
1727                                      &target_list, NULL);
1728         if (ret < 0)
1729                 goto out;
1730
1731         list_for_each_entry(entry, &target_list, list) {
1732                 u32 range_len = entry->len;
1733
1734                 /* Reached or beyond the limit */
1735                 if (max_sectors && *sectors_defragged >= max_sectors) {
1736                         ret = 1;
1737                         break;
1738                 }
1739
1740                 if (max_sectors)
1741                         range_len = min_t(u32, range_len,
1742                                 (max_sectors - *sectors_defragged) * sectorsize);
1743
1744                 /*
1745                  * If defrag_one_range() has updated last_scanned_ret,
1746                  * our range may already be invalid (e.g. hole punched).
1747                  * Skip if our range is before last_scanned_ret, as there is
1748                  * no need to defrag the range anymore.
1749                  */
1750                 if (entry->start + range_len <= *last_scanned_ret)
1751                         continue;
1752
1753                 if (ra)
1754                         page_cache_sync_readahead(inode->vfs_inode.i_mapping,
1755                                 ra, NULL, entry->start >> PAGE_SHIFT,
1756                                 ((entry->start + range_len - 1) >> PAGE_SHIFT) -
1757                                 (entry->start >> PAGE_SHIFT) + 1);
1758                 /*
1759                  * Here we may not defrag any range if holes are punched before
1760                  * we locked the pages.
1761                  * But that's fine, it only affects the @sectors_defragged
1762                  * accounting.
1763                  */
1764                 ret = defrag_one_range(inode, entry->start, range_len,
1765                                        extent_thresh, newer_than, do_compress,
1766                                        last_scanned_ret);
1767                 if (ret < 0)
1768                         break;
1769                 *sectors_defragged += range_len >>
1770                                       inode->root->fs_info->sectorsize_bits;
1771         }
1772 out:
1773         list_for_each_entry_safe(entry, tmp, &target_list, list) {
1774                 list_del_init(&entry->list);
1775                 kfree(entry);
1776         }
1777         if (ret >= 0)
1778                 *last_scanned_ret = max(*last_scanned_ret, start + len);
1779         return ret;
1780 }
1781
1782 /*
1783  * Entry point to file defragmentation.
1784  *
1785  * @inode:         inode to be defragged
1786  * @ra:            readahead state (can be NUL)
1787  * @range:         defrag options including range and flags
1788  * @newer_than:    minimum transid to defrag
1789  * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode
1790  *                 will be defragged.
1791  *
1792  * Return <0 for error.
1793  * Return >=0 for the number of sectors defragged, and range->start will be updated
1794  * to indicate the file offset where next defrag should be started at.
1795  * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without
1796  *  defragging all the range).
1797  */
1798 int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra,
1799                       struct btrfs_ioctl_defrag_range_args *range,
1800                       u64 newer_than, unsigned long max_to_defrag)
1801 {
1802         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1803         unsigned long sectors_defragged = 0;
1804         u64 isize = i_size_read(inode);
1805         u64 cur;
1806         u64 last_byte;
1807         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1808         bool ra_allocated = false;
1809         int compress_type = BTRFS_COMPRESS_ZLIB;
1810         int ret = 0;
1811         u32 extent_thresh = range->extent_thresh;
1812         pgoff_t start_index;
1813
1814         if (isize == 0)
1815                 return 0;
1816
1817         if (range->start >= isize)
1818                 return -EINVAL;
1819
1820         if (do_compress) {
1821                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1822                         return -EINVAL;
1823                 if (range->compress_type)
1824                         compress_type = range->compress_type;
1825         }
1826
1827         if (extent_thresh == 0)
1828                 extent_thresh = SZ_256K;
1829
1830         if (range->start + range->len > range->start) {
1831                 /* Got a specific range */
1832                 last_byte = min(isize, range->start + range->len);
1833         } else {
1834                 /* Defrag until file end */
1835                 last_byte = isize;
1836         }
1837
1838         /* Align the range */
1839         cur = round_down(range->start, fs_info->sectorsize);
1840         last_byte = round_up(last_byte, fs_info->sectorsize) - 1;
1841
1842         /*
1843          * If we were not given a ra, allocate a readahead context. As
1844          * readahead is just an optimization, defrag will work without it so
1845          * we don't error out.
1846          */
1847         if (!ra) {
1848                 ra_allocated = true;
1849                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1850                 if (ra)
1851                         file_ra_state_init(ra, inode->i_mapping);
1852         }
1853
1854         /*
1855          * Make writeback start from the beginning of the range, so that the
1856          * defrag range can be written sequentially.
1857          */
1858         start_index = cur >> PAGE_SHIFT;
1859         if (start_index < inode->i_mapping->writeback_index)
1860                 inode->i_mapping->writeback_index = start_index;
1861
1862         while (cur < last_byte) {
1863                 const unsigned long prev_sectors_defragged = sectors_defragged;
1864                 u64 last_scanned = cur;
1865                 u64 cluster_end;
1866
1867                 if (btrfs_defrag_cancelled(fs_info)) {
1868                         ret = -EAGAIN;
1869                         break;
1870                 }
1871
1872                 /* We want the cluster end at page boundary when possible */
1873                 cluster_end = (((cur >> PAGE_SHIFT) +
1874                                (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1;
1875                 cluster_end = min(cluster_end, last_byte);
1876
1877                 btrfs_inode_lock(inode, 0);
1878                 if (IS_SWAPFILE(inode)) {
1879                         ret = -ETXTBSY;
1880                         btrfs_inode_unlock(inode, 0);
1881                         break;
1882                 }
1883                 if (!(inode->i_sb->s_flags & SB_ACTIVE)) {
1884                         btrfs_inode_unlock(inode, 0);
1885                         break;
1886                 }
1887                 if (do_compress)
1888                         BTRFS_I(inode)->defrag_compress = compress_type;
1889                 ret = defrag_one_cluster(BTRFS_I(inode), ra, cur,
1890                                 cluster_end + 1 - cur, extent_thresh,
1891                                 newer_than, do_compress, &sectors_defragged,
1892                                 max_to_defrag, &last_scanned);
1893
1894                 if (sectors_defragged > prev_sectors_defragged)
1895                         balance_dirty_pages_ratelimited(inode->i_mapping);
1896
1897                 btrfs_inode_unlock(inode, 0);
1898                 if (ret < 0)
1899                         break;
1900                 cur = max(cluster_end + 1, last_scanned);
1901                 if (ret > 0) {
1902                         ret = 0;
1903                         break;
1904                 }
1905                 cond_resched();
1906         }
1907
1908         if (ra_allocated)
1909                 kfree(ra);
1910         /*
1911          * Update range.start for autodefrag, this will indicate where to start
1912          * in next run.
1913          */
1914         range->start = cur;
1915         if (sectors_defragged) {
1916                 /*
1917                  * We have defragged some sectors, for compression case they
1918                  * need to be written back immediately.
1919                  */
1920                 if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) {
1921                         filemap_flush(inode->i_mapping);
1922                         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1923                                      &BTRFS_I(inode)->runtime_flags))
1924                                 filemap_flush(inode->i_mapping);
1925                 }
1926                 if (range->compress_type == BTRFS_COMPRESS_LZO)
1927                         btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1928                 else if (range->compress_type == BTRFS_COMPRESS_ZSTD)
1929                         btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1930                 ret = sectors_defragged;
1931         }
1932         if (do_compress) {
1933                 btrfs_inode_lock(inode, 0);
1934                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1935                 btrfs_inode_unlock(inode, 0);
1936         }
1937         return ret;
1938 }
1939
1940 /*
1941  * Try to start exclusive operation @type or cancel it if it's running.
1942  *
1943  * Return:
1944  *   0        - normal mode, newly claimed op started
1945  *  >0        - normal mode, something else is running,
1946  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1947  * ECANCELED  - cancel mode, successful cancel
1948  * ENOTCONN   - cancel mode, operation not running anymore
1949  */
1950 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1951                         enum btrfs_exclusive_operation type, bool cancel)
1952 {
1953         if (!cancel) {
1954                 /* Start normal op */
1955                 if (!btrfs_exclop_start(fs_info, type))
1956                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1957                 /* Exclusive operation is now claimed */
1958                 return 0;
1959         }
1960
1961         /* Cancel running op */
1962         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1963                 /*
1964                  * This blocks any exclop finish from setting it to NONE, so we
1965                  * request cancellation. Either it runs and we will wait for it,
1966                  * or it has finished and no waiting will happen.
1967                  */
1968                 atomic_inc(&fs_info->reloc_cancel_req);
1969                 btrfs_exclop_start_unlock(fs_info);
1970
1971                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1972                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1973                                     TASK_INTERRUPTIBLE);
1974
1975                 return -ECANCELED;
1976         }
1977
1978         /* Something else is running or none */
1979         return -ENOTCONN;
1980 }
1981
1982 static noinline int btrfs_ioctl_resize(struct file *file,
1983                                         void __user *arg)
1984 {
1985         BTRFS_DEV_LOOKUP_ARGS(args);
1986         struct inode *inode = file_inode(file);
1987         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1988         u64 new_size;
1989         u64 old_size;
1990         u64 devid = 1;
1991         struct btrfs_root *root = BTRFS_I(inode)->root;
1992         struct btrfs_ioctl_vol_args *vol_args;
1993         struct btrfs_trans_handle *trans;
1994         struct btrfs_device *device = NULL;
1995         char *sizestr;
1996         char *retptr;
1997         char *devstr = NULL;
1998         int ret = 0;
1999         int mod = 0;
2000         bool cancel;
2001
2002         if (!capable(CAP_SYS_ADMIN))
2003                 return -EPERM;
2004
2005         ret = mnt_want_write_file(file);
2006         if (ret)
2007                 return ret;
2008
2009         /*
2010          * Read the arguments before checking exclusivity to be able to
2011          * distinguish regular resize and cancel
2012          */
2013         vol_args = memdup_user(arg, sizeof(*vol_args));
2014         if (IS_ERR(vol_args)) {
2015                 ret = PTR_ERR(vol_args);
2016                 goto out_drop;
2017         }
2018         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2019         sizestr = vol_args->name;
2020         cancel = (strcmp("cancel", sizestr) == 0);
2021         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
2022         if (ret)
2023                 goto out_free;
2024         /* Exclusive operation is now claimed */
2025
2026         devstr = strchr(sizestr, ':');
2027         if (devstr) {
2028                 sizestr = devstr + 1;
2029                 *devstr = '\0';
2030                 devstr = vol_args->name;
2031                 ret = kstrtoull(devstr, 10, &devid);
2032                 if (ret)
2033                         goto out_finish;
2034                 if (!devid) {
2035                         ret = -EINVAL;
2036                         goto out_finish;
2037                 }
2038                 btrfs_info(fs_info, "resizing devid %llu", devid);
2039         }
2040
2041         args.devid = devid;
2042         device = btrfs_find_device(fs_info->fs_devices, &args);
2043         if (!device) {
2044                 btrfs_info(fs_info, "resizer unable to find device %llu",
2045                            devid);
2046                 ret = -ENODEV;
2047                 goto out_finish;
2048         }
2049
2050         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2051                 btrfs_info(fs_info,
2052                            "resizer unable to apply on readonly device %llu",
2053                        devid);
2054                 ret = -EPERM;
2055                 goto out_finish;
2056         }
2057
2058         if (!strcmp(sizestr, "max"))
2059                 new_size = bdev_nr_bytes(device->bdev);
2060         else {
2061                 if (sizestr[0] == '-') {
2062                         mod = -1;
2063                         sizestr++;
2064                 } else if (sizestr[0] == '+') {
2065                         mod = 1;
2066                         sizestr++;
2067                 }
2068                 new_size = memparse(sizestr, &retptr);
2069                 if (*retptr != '\0' || new_size == 0) {
2070                         ret = -EINVAL;
2071                         goto out_finish;
2072                 }
2073         }
2074
2075         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2076                 ret = -EPERM;
2077                 goto out_finish;
2078         }
2079
2080         old_size = btrfs_device_get_total_bytes(device);
2081
2082         if (mod < 0) {
2083                 if (new_size > old_size) {
2084                         ret = -EINVAL;
2085                         goto out_finish;
2086                 }
2087                 new_size = old_size - new_size;
2088         } else if (mod > 0) {
2089                 if (new_size > ULLONG_MAX - old_size) {
2090                         ret = -ERANGE;
2091                         goto out_finish;
2092                 }
2093                 new_size = old_size + new_size;
2094         }
2095
2096         if (new_size < SZ_256M) {
2097                 ret = -EINVAL;
2098                 goto out_finish;
2099         }
2100         if (new_size > bdev_nr_bytes(device->bdev)) {
2101                 ret = -EFBIG;
2102                 goto out_finish;
2103         }
2104
2105         new_size = round_down(new_size, fs_info->sectorsize);
2106
2107         if (new_size > old_size) {
2108                 trans = btrfs_start_transaction(root, 0);
2109                 if (IS_ERR(trans)) {
2110                         ret = PTR_ERR(trans);
2111                         goto out_finish;
2112                 }
2113                 ret = btrfs_grow_device(trans, device, new_size);
2114                 btrfs_commit_transaction(trans);
2115         } else if (new_size < old_size) {
2116                 ret = btrfs_shrink_device(device, new_size);
2117         } /* equal, nothing need to do */
2118
2119         if (ret == 0 && new_size != old_size)
2120                 btrfs_info_in_rcu(fs_info,
2121                         "resize device %s (devid %llu) from %llu to %llu",
2122                         rcu_str_deref(device->name), device->devid,
2123                         old_size, new_size);
2124 out_finish:
2125         btrfs_exclop_finish(fs_info);
2126 out_free:
2127         kfree(vol_args);
2128 out_drop:
2129         mnt_drop_write_file(file);
2130         return ret;
2131 }
2132
2133 static noinline int __btrfs_ioctl_snap_create(struct file *file,
2134                                 struct user_namespace *mnt_userns,
2135                                 const char *name, unsigned long fd, int subvol,
2136                                 bool readonly,
2137                                 struct btrfs_qgroup_inherit *inherit)
2138 {
2139         int namelen;
2140         int ret = 0;
2141
2142         if (!S_ISDIR(file_inode(file)->i_mode))
2143                 return -ENOTDIR;
2144
2145         ret = mnt_want_write_file(file);
2146         if (ret)
2147                 goto out;
2148
2149         namelen = strlen(name);
2150         if (strchr(name, '/')) {
2151                 ret = -EINVAL;
2152                 goto out_drop_write;
2153         }
2154
2155         if (name[0] == '.' &&
2156            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
2157                 ret = -EEXIST;
2158                 goto out_drop_write;
2159         }
2160
2161         if (subvol) {
2162                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
2163                                      namelen, NULL, readonly, inherit);
2164         } else {
2165                 struct fd src = fdget(fd);
2166                 struct inode *src_inode;
2167                 if (!src.file) {
2168                         ret = -EINVAL;
2169                         goto out_drop_write;
2170                 }
2171
2172                 src_inode = file_inode(src.file);
2173                 if (src_inode->i_sb != file_inode(file)->i_sb) {
2174                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
2175                                    "Snapshot src from another FS");
2176                         ret = -EXDEV;
2177                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
2178                         /*
2179                          * Subvolume creation is not restricted, but snapshots
2180                          * are limited to own subvolumes only
2181                          */
2182                         ret = -EPERM;
2183                 } else {
2184                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
2185                                                name, namelen,
2186                                                BTRFS_I(src_inode)->root,
2187                                                readonly, inherit);
2188                 }
2189                 fdput(src);
2190         }
2191 out_drop_write:
2192         mnt_drop_write_file(file);
2193 out:
2194         return ret;
2195 }
2196
2197 static noinline int btrfs_ioctl_snap_create(struct file *file,
2198                                             void __user *arg, int subvol)
2199 {
2200         struct btrfs_ioctl_vol_args *vol_args;
2201         int ret;
2202
2203         if (!S_ISDIR(file_inode(file)->i_mode))
2204                 return -ENOTDIR;
2205
2206         vol_args = memdup_user(arg, sizeof(*vol_args));
2207         if (IS_ERR(vol_args))
2208                 return PTR_ERR(vol_args);
2209         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2210
2211         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2212                                         vol_args->name, vol_args->fd, subvol,
2213                                         false, NULL);
2214
2215         kfree(vol_args);
2216         return ret;
2217 }
2218
2219 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
2220                                                void __user *arg, int subvol)
2221 {
2222         struct btrfs_ioctl_vol_args_v2 *vol_args;
2223         int ret;
2224         bool readonly = false;
2225         struct btrfs_qgroup_inherit *inherit = NULL;
2226
2227         if (!S_ISDIR(file_inode(file)->i_mode))
2228                 return -ENOTDIR;
2229
2230         vol_args = memdup_user(arg, sizeof(*vol_args));
2231         if (IS_ERR(vol_args))
2232                 return PTR_ERR(vol_args);
2233         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
2234
2235         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
2236                 ret = -EOPNOTSUPP;
2237                 goto free_args;
2238         }
2239
2240         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
2241                 readonly = true;
2242         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
2243                 u64 nums;
2244
2245                 if (vol_args->size < sizeof(*inherit) ||
2246                     vol_args->size > PAGE_SIZE) {
2247                         ret = -EINVAL;
2248                         goto free_args;
2249                 }
2250                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
2251                 if (IS_ERR(inherit)) {
2252                         ret = PTR_ERR(inherit);
2253                         goto free_args;
2254                 }
2255
2256                 if (inherit->num_qgroups > PAGE_SIZE ||
2257                     inherit->num_ref_copies > PAGE_SIZE ||
2258                     inherit->num_excl_copies > PAGE_SIZE) {
2259                         ret = -EINVAL;
2260                         goto free_inherit;
2261                 }
2262
2263                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2264                        2 * inherit->num_excl_copies;
2265                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
2266                         ret = -EINVAL;
2267                         goto free_inherit;
2268                 }
2269         }
2270
2271         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
2272                                         vol_args->name, vol_args->fd, subvol,
2273                                         readonly, inherit);
2274         if (ret)
2275                 goto free_inherit;
2276 free_inherit:
2277         kfree(inherit);
2278 free_args:
2279         kfree(vol_args);
2280         return ret;
2281 }
2282
2283 static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode,
2284                                                 void __user *arg)
2285 {
2286         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2287         struct btrfs_root *root = BTRFS_I(inode)->root;
2288         int ret = 0;
2289         u64 flags = 0;
2290
2291         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
2292                 return -EINVAL;
2293
2294         down_read(&fs_info->subvol_sem);
2295         if (btrfs_root_readonly(root))
2296                 flags |= BTRFS_SUBVOL_RDONLY;
2297         up_read(&fs_info->subvol_sem);
2298
2299         if (copy_to_user(arg, &flags, sizeof(flags)))
2300                 ret = -EFAULT;
2301
2302         return ret;
2303 }
2304
2305 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2306                                               void __user *arg)
2307 {
2308         struct inode *inode = file_inode(file);
2309         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2310         struct btrfs_root *root = BTRFS_I(inode)->root;
2311         struct btrfs_trans_handle *trans;
2312         u64 root_flags;
2313         u64 flags;
2314         int ret = 0;
2315
2316         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
2317                 return -EPERM;
2318
2319         ret = mnt_want_write_file(file);
2320         if (ret)
2321                 goto out;
2322
2323         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2324                 ret = -EINVAL;
2325                 goto out_drop_write;
2326         }
2327
2328         if (copy_from_user(&flags, arg, sizeof(flags))) {
2329                 ret = -EFAULT;
2330                 goto out_drop_write;
2331         }
2332
2333         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2334                 ret = -EOPNOTSUPP;
2335                 goto out_drop_write;
2336         }
2337
2338         down_write(&fs_info->subvol_sem);
2339
2340         /* nothing to do */
2341         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2342                 goto out_drop_sem;
2343
2344         root_flags = btrfs_root_flags(&root->root_item);
2345         if (flags & BTRFS_SUBVOL_RDONLY) {
2346                 btrfs_set_root_flags(&root->root_item,
2347                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2348         } else {
2349                 /*
2350                  * Block RO -> RW transition if this subvolume is involved in
2351                  * send
2352                  */
2353                 spin_lock(&root->root_item_lock);
2354                 if (root->send_in_progress == 0) {
2355                         btrfs_set_root_flags(&root->root_item,
2356                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2357                         spin_unlock(&root->root_item_lock);
2358                 } else {
2359                         spin_unlock(&root->root_item_lock);
2360                         btrfs_warn(fs_info,
2361                                    "Attempt to set subvolume %llu read-write during send",
2362                                    root->root_key.objectid);
2363                         ret = -EPERM;
2364                         goto out_drop_sem;
2365                 }
2366         }
2367
2368         trans = btrfs_start_transaction(root, 1);
2369         if (IS_ERR(trans)) {
2370                 ret = PTR_ERR(trans);
2371                 goto out_reset;
2372         }
2373
2374         ret = btrfs_update_root(trans, fs_info->tree_root,
2375                                 &root->root_key, &root->root_item);
2376         if (ret < 0) {
2377                 btrfs_end_transaction(trans);
2378                 goto out_reset;
2379         }
2380
2381         ret = btrfs_commit_transaction(trans);
2382
2383 out_reset:
2384         if (ret)
2385                 btrfs_set_root_flags(&root->root_item, root_flags);
2386 out_drop_sem:
2387         up_write(&fs_info->subvol_sem);
2388 out_drop_write:
2389         mnt_drop_write_file(file);
2390 out:
2391         return ret;
2392 }
2393
2394 static noinline int key_in_sk(struct btrfs_key *key,
2395                               struct btrfs_ioctl_search_key *sk)
2396 {
2397         struct btrfs_key test;
2398         int ret;
2399
2400         test.objectid = sk->min_objectid;
2401         test.type = sk->min_type;
2402         test.offset = sk->min_offset;
2403
2404         ret = btrfs_comp_cpu_keys(key, &test);
2405         if (ret < 0)
2406                 return 0;
2407
2408         test.objectid = sk->max_objectid;
2409         test.type = sk->max_type;
2410         test.offset = sk->max_offset;
2411
2412         ret = btrfs_comp_cpu_keys(key, &test);
2413         if (ret > 0)
2414                 return 0;
2415         return 1;
2416 }
2417
2418 static noinline int copy_to_sk(struct btrfs_path *path,
2419                                struct btrfs_key *key,
2420                                struct btrfs_ioctl_search_key *sk,
2421                                size_t *buf_size,
2422                                char __user *ubuf,
2423                                unsigned long *sk_offset,
2424                                int *num_found)
2425 {
2426         u64 found_transid;
2427         struct extent_buffer *leaf;
2428         struct btrfs_ioctl_search_header sh;
2429         struct btrfs_key test;
2430         unsigned long item_off;
2431         unsigned long item_len;
2432         int nritems;
2433         int i;
2434         int slot;
2435         int ret = 0;
2436
2437         leaf = path->nodes[0];
2438         slot = path->slots[0];
2439         nritems = btrfs_header_nritems(leaf);
2440
2441         if (btrfs_header_generation(leaf) > sk->max_transid) {
2442                 i = nritems;
2443                 goto advance_key;
2444         }
2445         found_transid = btrfs_header_generation(leaf);
2446
2447         for (i = slot; i < nritems; i++) {
2448                 item_off = btrfs_item_ptr_offset(leaf, i);
2449                 item_len = btrfs_item_size(leaf, i);
2450
2451                 btrfs_item_key_to_cpu(leaf, key, i);
2452                 if (!key_in_sk(key, sk))
2453                         continue;
2454
2455                 if (sizeof(sh) + item_len > *buf_size) {
2456                         if (*num_found) {
2457                                 ret = 1;
2458                                 goto out;
2459                         }
2460
2461                         /*
2462                          * return one empty item back for v1, which does not
2463                          * handle -EOVERFLOW
2464                          */
2465
2466                         *buf_size = sizeof(sh) + item_len;
2467                         item_len = 0;
2468                         ret = -EOVERFLOW;
2469                 }
2470
2471                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2472                         ret = 1;
2473                         goto out;
2474                 }
2475
2476                 sh.objectid = key->objectid;
2477                 sh.offset = key->offset;
2478                 sh.type = key->type;
2479                 sh.len = item_len;
2480                 sh.transid = found_transid;
2481
2482                 /*
2483                  * Copy search result header. If we fault then loop again so we
2484                  * can fault in the pages and -EFAULT there if there's a
2485                  * problem. Otherwise we'll fault and then copy the buffer in
2486                  * properly this next time through
2487                  */
2488                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2489                         ret = 0;
2490                         goto out;
2491                 }
2492
2493                 *sk_offset += sizeof(sh);
2494
2495                 if (item_len) {
2496                         char __user *up = ubuf + *sk_offset;
2497                         /*
2498                          * Copy the item, same behavior as above, but reset the
2499                          * * sk_offset so we copy the full thing again.
2500                          */
2501                         if (read_extent_buffer_to_user_nofault(leaf, up,
2502                                                 item_off, item_len)) {
2503                                 ret = 0;
2504                                 *sk_offset -= sizeof(sh);
2505                                 goto out;
2506                         }
2507
2508                         *sk_offset += item_len;
2509                 }
2510                 (*num_found)++;
2511
2512                 if (ret) /* -EOVERFLOW from above */
2513                         goto out;
2514
2515                 if (*num_found >= sk->nr_items) {
2516                         ret = 1;
2517                         goto out;
2518                 }
2519         }
2520 advance_key:
2521         ret = 0;
2522         test.objectid = sk->max_objectid;
2523         test.type = sk->max_type;
2524         test.offset = sk->max_offset;
2525         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2526                 ret = 1;
2527         else if (key->offset < (u64)-1)
2528                 key->offset++;
2529         else if (key->type < (u8)-1) {
2530                 key->offset = 0;
2531                 key->type++;
2532         } else if (key->objectid < (u64)-1) {
2533                 key->offset = 0;
2534                 key->type = 0;
2535                 key->objectid++;
2536         } else
2537                 ret = 1;
2538 out:
2539         /*
2540          *  0: all items from this leaf copied, continue with next
2541          *  1: * more items can be copied, but unused buffer is too small
2542          *     * all items were found
2543          *     Either way, it will stops the loop which iterates to the next
2544          *     leaf
2545          *  -EOVERFLOW: item was to large for buffer
2546          *  -EFAULT: could not copy extent buffer back to userspace
2547          */
2548         return ret;
2549 }
2550
2551 static noinline int search_ioctl(struct inode *inode,
2552                                  struct btrfs_ioctl_search_key *sk,
2553                                  size_t *buf_size,
2554                                  char __user *ubuf)
2555 {
2556         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2557         struct btrfs_root *root;
2558         struct btrfs_key key;
2559         struct btrfs_path *path;
2560         int ret;
2561         int num_found = 0;
2562         unsigned long sk_offset = 0;
2563
2564         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2565                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2566                 return -EOVERFLOW;
2567         }
2568
2569         path = btrfs_alloc_path();
2570         if (!path)
2571                 return -ENOMEM;
2572
2573         if (sk->tree_id == 0) {
2574                 /* search the root of the inode that was passed */
2575                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2576         } else {
2577                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2578                 if (IS_ERR(root)) {
2579                         btrfs_free_path(path);
2580                         return PTR_ERR(root);
2581                 }
2582         }
2583
2584         key.objectid = sk->min_objectid;
2585         key.type = sk->min_type;
2586         key.offset = sk->min_offset;
2587
2588         while (1) {
2589                 ret = -EFAULT;
2590                 /*
2591                  * Ensure that the whole user buffer is faulted in at sub-page
2592                  * granularity, otherwise the loop may live-lock.
2593                  */
2594                 if (fault_in_subpage_writeable(ubuf + sk_offset,
2595                                                *buf_size - sk_offset))
2596                         break;
2597
2598                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2599                 if (ret != 0) {
2600                         if (ret > 0)
2601                                 ret = 0;
2602                         goto err;
2603                 }
2604                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2605                                  &sk_offset, &num_found);
2606                 btrfs_release_path(path);
2607                 if (ret)
2608                         break;
2609
2610         }
2611         if (ret > 0)
2612                 ret = 0;
2613 err:
2614         sk->nr_items = num_found;
2615         btrfs_put_root(root);
2616         btrfs_free_path(path);
2617         return ret;
2618 }
2619
2620 static noinline int btrfs_ioctl_tree_search(struct inode *inode,
2621                                             void __user *argp)
2622 {
2623         struct btrfs_ioctl_search_args __user *uargs = argp;
2624         struct btrfs_ioctl_search_key sk;
2625         int ret;
2626         size_t buf_size;
2627
2628         if (!capable(CAP_SYS_ADMIN))
2629                 return -EPERM;
2630
2631         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2632                 return -EFAULT;
2633
2634         buf_size = sizeof(uargs->buf);
2635
2636         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2637
2638         /*
2639          * In the origin implementation an overflow is handled by returning a
2640          * search header with a len of zero, so reset ret.
2641          */
2642         if (ret == -EOVERFLOW)
2643                 ret = 0;
2644
2645         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2646                 ret = -EFAULT;
2647         return ret;
2648 }
2649
2650 static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode,
2651                                                void __user *argp)
2652 {
2653         struct btrfs_ioctl_search_args_v2 __user *uarg = argp;
2654         struct btrfs_ioctl_search_args_v2 args;
2655         int ret;
2656         size_t buf_size;
2657         const size_t buf_limit = SZ_16M;
2658
2659         if (!capable(CAP_SYS_ADMIN))
2660                 return -EPERM;
2661
2662         /* copy search header and buffer size */
2663         if (copy_from_user(&args, uarg, sizeof(args)))
2664                 return -EFAULT;
2665
2666         buf_size = args.buf_size;
2667
2668         /* limit result size to 16MB */
2669         if (buf_size > buf_limit)
2670                 buf_size = buf_limit;
2671
2672         ret = search_ioctl(inode, &args.key, &buf_size,
2673                            (char __user *)(&uarg->buf[0]));
2674         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2675                 ret = -EFAULT;
2676         else if (ret == -EOVERFLOW &&
2677                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2678                 ret = -EFAULT;
2679
2680         return ret;
2681 }
2682
2683 /*
2684  * Search INODE_REFs to identify path name of 'dirid' directory
2685  * in a 'tree_id' tree. and sets path name to 'name'.
2686  */
2687 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2688                                 u64 tree_id, u64 dirid, char *name)
2689 {
2690         struct btrfs_root *root;
2691         struct btrfs_key key;
2692         char *ptr;
2693         int ret = -1;
2694         int slot;
2695         int len;
2696         int total_len = 0;
2697         struct btrfs_inode_ref *iref;
2698         struct extent_buffer *l;
2699         struct btrfs_path *path;
2700
2701         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2702                 name[0]='\0';
2703                 return 0;
2704         }
2705
2706         path = btrfs_alloc_path();
2707         if (!path)
2708                 return -ENOMEM;
2709
2710         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2711
2712         root = btrfs_get_fs_root(info, tree_id, true);
2713         if (IS_ERR(root)) {
2714                 ret = PTR_ERR(root);
2715                 root = NULL;
2716                 goto out;
2717         }
2718
2719         key.objectid = dirid;
2720         key.type = BTRFS_INODE_REF_KEY;
2721         key.offset = (u64)-1;
2722
2723         while (1) {
2724                 ret = btrfs_search_backwards(root, &key, path);
2725                 if (ret < 0)
2726                         goto out;
2727                 else if (ret > 0) {
2728                         ret = -ENOENT;
2729                         goto out;
2730                 }
2731
2732                 l = path->nodes[0];
2733                 slot = path->slots[0];
2734
2735                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2736                 len = btrfs_inode_ref_name_len(l, iref);
2737                 ptr -= len + 1;
2738                 total_len += len + 1;
2739                 if (ptr < name) {
2740                         ret = -ENAMETOOLONG;
2741                         goto out;
2742                 }
2743
2744                 *(ptr + len) = '/';
2745                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2746
2747                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2748                         break;
2749
2750                 btrfs_release_path(path);
2751                 key.objectid = key.offset;
2752                 key.offset = (u64)-1;
2753                 dirid = key.objectid;
2754         }
2755         memmove(name, ptr, total_len);
2756         name[total_len] = '\0';
2757         ret = 0;
2758 out:
2759         btrfs_put_root(root);
2760         btrfs_free_path(path);
2761         return ret;
2762 }
2763
2764 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2765                                 struct inode *inode,
2766                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2767 {
2768         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2769         struct super_block *sb = inode->i_sb;
2770         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2771         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2772         u64 dirid = args->dirid;
2773         unsigned long item_off;
2774         unsigned long item_len;
2775         struct btrfs_inode_ref *iref;
2776         struct btrfs_root_ref *rref;
2777         struct btrfs_root *root = NULL;
2778         struct btrfs_path *path;
2779         struct btrfs_key key, key2;
2780         struct extent_buffer *leaf;
2781         struct inode *temp_inode;
2782         char *ptr;
2783         int slot;
2784         int len;
2785         int total_len = 0;
2786         int ret;
2787
2788         path = btrfs_alloc_path();
2789         if (!path)
2790                 return -ENOMEM;
2791
2792         /*
2793          * If the bottom subvolume does not exist directly under upper_limit,
2794          * construct the path in from the bottom up.
2795          */
2796         if (dirid != upper_limit.objectid) {
2797                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2798
2799                 root = btrfs_get_fs_root(fs_info, treeid, true);
2800                 if (IS_ERR(root)) {
2801                         ret = PTR_ERR(root);
2802                         goto out;
2803                 }
2804
2805                 key.objectid = dirid;
2806                 key.type = BTRFS_INODE_REF_KEY;
2807                 key.offset = (u64)-1;
2808                 while (1) {
2809                         ret = btrfs_search_backwards(root, &key, path);
2810                         if (ret < 0)
2811                                 goto out_put;
2812                         else if (ret > 0) {
2813                                 ret = -ENOENT;
2814                                 goto out_put;
2815                         }
2816
2817                         leaf = path->nodes[0];
2818                         slot = path->slots[0];
2819
2820                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2821                         len = btrfs_inode_ref_name_len(leaf, iref);
2822                         ptr -= len + 1;
2823                         total_len += len + 1;
2824                         if (ptr < args->path) {
2825                                 ret = -ENAMETOOLONG;
2826                                 goto out_put;
2827                         }
2828
2829                         *(ptr + len) = '/';
2830                         read_extent_buffer(leaf, ptr,
2831                                         (unsigned long)(iref + 1), len);
2832
2833                         /* Check the read+exec permission of this directory */
2834                         ret = btrfs_previous_item(root, path, dirid,
2835                                                   BTRFS_INODE_ITEM_KEY);
2836                         if (ret < 0) {
2837                                 goto out_put;
2838                         } else if (ret > 0) {
2839                                 ret = -ENOENT;
2840                                 goto out_put;
2841                         }
2842
2843                         leaf = path->nodes[0];
2844                         slot = path->slots[0];
2845                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2846                         if (key2.objectid != dirid) {
2847                                 ret = -ENOENT;
2848                                 goto out_put;
2849                         }
2850
2851                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2852                         if (IS_ERR(temp_inode)) {
2853                                 ret = PTR_ERR(temp_inode);
2854                                 goto out_put;
2855                         }
2856                         ret = inode_permission(mnt_userns, temp_inode,
2857                                                MAY_READ | MAY_EXEC);
2858                         iput(temp_inode);
2859                         if (ret) {
2860                                 ret = -EACCES;
2861                                 goto out_put;
2862                         }
2863
2864                         if (key.offset == upper_limit.objectid)
2865                                 break;
2866                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2867                                 ret = -EACCES;
2868                                 goto out_put;
2869                         }
2870
2871                         btrfs_release_path(path);
2872                         key.objectid = key.offset;
2873                         key.offset = (u64)-1;
2874                         dirid = key.objectid;
2875                 }
2876
2877                 memmove(args->path, ptr, total_len);
2878                 args->path[total_len] = '\0';
2879                 btrfs_put_root(root);
2880                 root = NULL;
2881                 btrfs_release_path(path);
2882         }
2883
2884         /* Get the bottom subvolume's name from ROOT_REF */
2885         key.objectid = treeid;
2886         key.type = BTRFS_ROOT_REF_KEY;
2887         key.offset = args->treeid;
2888         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2889         if (ret < 0) {
2890                 goto out;
2891         } else if (ret > 0) {
2892                 ret = -ENOENT;
2893                 goto out;
2894         }
2895
2896         leaf = path->nodes[0];
2897         slot = path->slots[0];
2898         btrfs_item_key_to_cpu(leaf, &key, slot);
2899
2900         item_off = btrfs_item_ptr_offset(leaf, slot);
2901         item_len = btrfs_item_size(leaf, slot);
2902         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2903         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2904         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2905                 ret = -EINVAL;
2906                 goto out;
2907         }
2908
2909         /* Copy subvolume's name */
2910         item_off += sizeof(struct btrfs_root_ref);
2911         item_len -= sizeof(struct btrfs_root_ref);
2912         read_extent_buffer(leaf, args->name, item_off, item_len);
2913         args->name[item_len] = 0;
2914
2915 out_put:
2916         btrfs_put_root(root);
2917 out:
2918         btrfs_free_path(path);
2919         return ret;
2920 }
2921
2922 static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root,
2923                                            void __user *argp)
2924 {
2925         struct btrfs_ioctl_ino_lookup_args *args;
2926         int ret = 0;
2927
2928         args = memdup_user(argp, sizeof(*args));
2929         if (IS_ERR(args))
2930                 return PTR_ERR(args);
2931
2932         /*
2933          * Unprivileged query to obtain the containing subvolume root id. The
2934          * path is reset so it's consistent with btrfs_search_path_in_tree.
2935          */
2936         if (args->treeid == 0)
2937                 args->treeid = root->root_key.objectid;
2938
2939         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2940                 args->name[0] = 0;
2941                 goto out;
2942         }
2943
2944         if (!capable(CAP_SYS_ADMIN)) {
2945                 ret = -EPERM;
2946                 goto out;
2947         }
2948
2949         ret = btrfs_search_path_in_tree(root->fs_info,
2950                                         args->treeid, args->objectid,
2951                                         args->name);
2952
2953 out:
2954         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2955                 ret = -EFAULT;
2956
2957         kfree(args);
2958         return ret;
2959 }
2960
2961 /*
2962  * Version of ino_lookup ioctl (unprivileged)
2963  *
2964  * The main differences from ino_lookup ioctl are:
2965  *
2966  *   1. Read + Exec permission will be checked using inode_permission() during
2967  *      path construction. -EACCES will be returned in case of failure.
2968  *   2. Path construction will be stopped at the inode number which corresponds
2969  *      to the fd with which this ioctl is called. If constructed path does not
2970  *      exist under fd's inode, -EACCES will be returned.
2971  *   3. The name of bottom subvolume is also searched and filled.
2972  */
2973 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2974 {
2975         struct btrfs_ioctl_ino_lookup_user_args *args;
2976         struct inode *inode;
2977         int ret;
2978
2979         args = memdup_user(argp, sizeof(*args));
2980         if (IS_ERR(args))
2981                 return PTR_ERR(args);
2982
2983         inode = file_inode(file);
2984
2985         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2986             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2987                 /*
2988                  * The subvolume does not exist under fd with which this is
2989                  * called
2990                  */
2991                 kfree(args);
2992                 return -EACCES;
2993         }
2994
2995         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
2996
2997         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2998                 ret = -EFAULT;
2999
3000         kfree(args);
3001         return ret;
3002 }
3003
3004 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
3005 static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp)
3006 {
3007         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
3008         struct btrfs_fs_info *fs_info;
3009         struct btrfs_root *root;
3010         struct btrfs_path *path;
3011         struct btrfs_key key;
3012         struct btrfs_root_item *root_item;
3013         struct btrfs_root_ref *rref;
3014         struct extent_buffer *leaf;
3015         unsigned long item_off;
3016         unsigned long item_len;
3017         int slot;
3018         int ret = 0;
3019
3020         path = btrfs_alloc_path();
3021         if (!path)
3022                 return -ENOMEM;
3023
3024         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
3025         if (!subvol_info) {
3026                 btrfs_free_path(path);
3027                 return -ENOMEM;
3028         }
3029
3030         fs_info = BTRFS_I(inode)->root->fs_info;
3031
3032         /* Get root_item of inode's subvolume */
3033         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
3034         root = btrfs_get_fs_root(fs_info, key.objectid, true);
3035         if (IS_ERR(root)) {
3036                 ret = PTR_ERR(root);
3037                 goto out_free;
3038         }
3039         root_item = &root->root_item;
3040
3041         subvol_info->treeid = key.objectid;
3042
3043         subvol_info->generation = btrfs_root_generation(root_item);
3044         subvol_info->flags = btrfs_root_flags(root_item);
3045
3046         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
3047         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
3048                                                     BTRFS_UUID_SIZE);
3049         memcpy(subvol_info->received_uuid, root_item->received_uuid,
3050                                                     BTRFS_UUID_SIZE);
3051
3052         subvol_info->ctransid = btrfs_root_ctransid(root_item);
3053         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
3054         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
3055
3056         subvol_info->otransid = btrfs_root_otransid(root_item);
3057         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
3058         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
3059
3060         subvol_info->stransid = btrfs_root_stransid(root_item);
3061         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
3062         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
3063
3064         subvol_info->rtransid = btrfs_root_rtransid(root_item);
3065         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
3066         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
3067
3068         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
3069                 /* Search root tree for ROOT_BACKREF of this subvolume */
3070                 key.type = BTRFS_ROOT_BACKREF_KEY;
3071                 key.offset = 0;
3072                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3073                 if (ret < 0) {
3074                         goto out;
3075                 } else if (path->slots[0] >=
3076                            btrfs_header_nritems(path->nodes[0])) {
3077                         ret = btrfs_next_leaf(fs_info->tree_root, path);
3078                         if (ret < 0) {
3079                                 goto out;
3080                         } else if (ret > 0) {
3081                                 ret = -EUCLEAN;
3082                                 goto out;
3083                         }
3084                 }
3085
3086                 leaf = path->nodes[0];
3087                 slot = path->slots[0];
3088                 btrfs_item_key_to_cpu(leaf, &key, slot);
3089                 if (key.objectid == subvol_info->treeid &&
3090                     key.type == BTRFS_ROOT_BACKREF_KEY) {
3091                         subvol_info->parent_id = key.offset;
3092
3093                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3094                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
3095
3096                         item_off = btrfs_item_ptr_offset(leaf, slot)
3097                                         + sizeof(struct btrfs_root_ref);
3098                         item_len = btrfs_item_size(leaf, slot)
3099                                         - sizeof(struct btrfs_root_ref);
3100                         read_extent_buffer(leaf, subvol_info->name,
3101                                            item_off, item_len);
3102                 } else {
3103                         ret = -ENOENT;
3104                         goto out;
3105                 }
3106         }
3107
3108         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
3109                 ret = -EFAULT;
3110
3111 out:
3112         btrfs_put_root(root);
3113 out_free:
3114         btrfs_free_path(path);
3115         kfree(subvol_info);
3116         return ret;
3117 }
3118
3119 /*
3120  * Return ROOT_REF information of the subvolume containing this inode
3121  * except the subvolume name.
3122  */
3123 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
3124                                           void __user *argp)
3125 {
3126         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
3127         struct btrfs_root_ref *rref;
3128         struct btrfs_path *path;
3129         struct btrfs_key key;
3130         struct extent_buffer *leaf;
3131         u64 objectid;
3132         int slot;
3133         int ret;
3134         u8 found;
3135
3136         path = btrfs_alloc_path();
3137         if (!path)
3138                 return -ENOMEM;
3139
3140         rootrefs = memdup_user(argp, sizeof(*rootrefs));
3141         if (IS_ERR(rootrefs)) {
3142                 btrfs_free_path(path);
3143                 return PTR_ERR(rootrefs);
3144         }
3145
3146         objectid = root->root_key.objectid;
3147         key.objectid = objectid;
3148         key.type = BTRFS_ROOT_REF_KEY;
3149         key.offset = rootrefs->min_treeid;
3150         found = 0;
3151
3152         root = root->fs_info->tree_root;
3153         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3154         if (ret < 0) {
3155                 goto out;
3156         } else if (path->slots[0] >=
3157                    btrfs_header_nritems(path->nodes[0])) {
3158                 ret = btrfs_next_leaf(root, path);
3159                 if (ret < 0) {
3160                         goto out;
3161                 } else if (ret > 0) {
3162                         ret = -EUCLEAN;
3163                         goto out;
3164                 }
3165         }
3166         while (1) {
3167                 leaf = path->nodes[0];
3168                 slot = path->slots[0];
3169
3170                 btrfs_item_key_to_cpu(leaf, &key, slot);
3171                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
3172                         ret = 0;
3173                         goto out;
3174                 }
3175
3176                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
3177                         ret = -EOVERFLOW;
3178                         goto out;
3179                 }
3180
3181                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3182                 rootrefs->rootref[found].treeid = key.offset;
3183                 rootrefs->rootref[found].dirid =
3184                                   btrfs_root_ref_dirid(leaf, rref);
3185                 found++;
3186
3187                 ret = btrfs_next_item(root, path);
3188                 if (ret < 0) {
3189                         goto out;
3190                 } else if (ret > 0) {
3191                         ret = -EUCLEAN;
3192                         goto out;
3193                 }
3194         }
3195
3196 out:
3197         if (!ret || ret == -EOVERFLOW) {
3198                 rootrefs->num_items = found;
3199                 /* update min_treeid for next search */
3200                 if (found)
3201                         rootrefs->min_treeid =
3202                                 rootrefs->rootref[found - 1].treeid + 1;
3203                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
3204                         ret = -EFAULT;
3205         }
3206
3207         kfree(rootrefs);
3208         btrfs_free_path(path);
3209
3210         return ret;
3211 }
3212
3213 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
3214                                              void __user *arg,
3215                                              bool destroy_v2)
3216 {
3217         struct dentry *parent = file->f_path.dentry;
3218         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
3219         struct dentry *dentry;
3220         struct inode *dir = d_inode(parent);
3221         struct inode *inode;
3222         struct btrfs_root *root = BTRFS_I(dir)->root;
3223         struct btrfs_root *dest = NULL;
3224         struct btrfs_ioctl_vol_args *vol_args = NULL;
3225         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
3226         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
3227         char *subvol_name, *subvol_name_ptr = NULL;
3228         int subvol_namelen;
3229         int err = 0;
3230         bool destroy_parent = false;
3231
3232         /* We don't support snapshots with extent tree v2 yet. */
3233         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3234                 btrfs_err(fs_info,
3235                           "extent tree v2 doesn't support snapshot deletion yet");
3236                 return -EOPNOTSUPP;
3237         }
3238
3239         if (destroy_v2) {
3240                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
3241                 if (IS_ERR(vol_args2))
3242                         return PTR_ERR(vol_args2);
3243
3244                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
3245                         err = -EOPNOTSUPP;
3246                         goto out;
3247                 }
3248
3249                 /*
3250                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
3251                  * name, same as v1 currently does.
3252                  */
3253                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
3254                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
3255                         subvol_name = vol_args2->name;
3256
3257                         err = mnt_want_write_file(file);
3258                         if (err)
3259                                 goto out;
3260                 } else {
3261                         struct inode *old_dir;
3262
3263                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
3264                                 err = -EINVAL;
3265                                 goto out;
3266                         }
3267
3268                         err = mnt_want_write_file(file);
3269                         if (err)
3270                                 goto out;
3271
3272                         dentry = btrfs_get_dentry(fs_info->sb,
3273                                         BTRFS_FIRST_FREE_OBJECTID,
3274                                         vol_args2->subvolid, 0, 0);
3275                         if (IS_ERR(dentry)) {
3276                                 err = PTR_ERR(dentry);
3277                                 goto out_drop_write;
3278                         }
3279
3280                         /*
3281                          * Change the default parent since the subvolume being
3282                          * deleted can be outside of the current mount point.
3283                          */
3284                         parent = btrfs_get_parent(dentry);
3285
3286                         /*
3287                          * At this point dentry->d_name can point to '/' if the
3288                          * subvolume we want to destroy is outsite of the
3289                          * current mount point, so we need to release the
3290                          * current dentry and execute the lookup to return a new
3291                          * one with ->d_name pointing to the
3292                          * <mount point>/subvol_name.
3293                          */
3294                         dput(dentry);
3295                         if (IS_ERR(parent)) {
3296                                 err = PTR_ERR(parent);
3297                                 goto out_drop_write;
3298                         }
3299                         old_dir = dir;
3300                         dir = d_inode(parent);
3301
3302                         /*
3303                          * If v2 was used with SPEC_BY_ID, a new parent was
3304                          * allocated since the subvolume can be outside of the
3305                          * current mount point. Later on we need to release this
3306                          * new parent dentry.
3307                          */
3308                         destroy_parent = true;
3309
3310                         /*
3311                          * On idmapped mounts, deletion via subvolid is
3312                          * restricted to subvolumes that are immediate
3313                          * ancestors of the inode referenced by the file
3314                          * descriptor in the ioctl. Otherwise the idmapping
3315                          * could potentially be abused to delete subvolumes
3316                          * anywhere in the filesystem the user wouldn't be able
3317                          * to delete without an idmapped mount.
3318                          */
3319                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3320                                 err = -EOPNOTSUPP;
3321                                 goto free_parent;
3322                         }
3323
3324                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3325                                                 fs_info, vol_args2->subvolid);
3326                         if (IS_ERR(subvol_name_ptr)) {
3327                                 err = PTR_ERR(subvol_name_ptr);
3328                                 goto free_parent;
3329                         }
3330                         /* subvol_name_ptr is already nul terminated */
3331                         subvol_name = (char *)kbasename(subvol_name_ptr);
3332                 }
3333         } else {
3334                 vol_args = memdup_user(arg, sizeof(*vol_args));
3335                 if (IS_ERR(vol_args))
3336                         return PTR_ERR(vol_args);
3337
3338                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3339                 subvol_name = vol_args->name;
3340
3341                 err = mnt_want_write_file(file);
3342                 if (err)
3343                         goto out;
3344         }
3345
3346         subvol_namelen = strlen(subvol_name);
3347
3348         if (strchr(subvol_name, '/') ||
3349             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3350                 err = -EINVAL;
3351                 goto free_subvol_name;
3352         }
3353
3354         if (!S_ISDIR(dir->i_mode)) {
3355                 err = -ENOTDIR;
3356                 goto free_subvol_name;
3357         }
3358
3359         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3360         if (err == -EINTR)
3361                 goto free_subvol_name;
3362         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3363         if (IS_ERR(dentry)) {
3364                 err = PTR_ERR(dentry);
3365                 goto out_unlock_dir;
3366         }
3367
3368         if (d_really_is_negative(dentry)) {
3369                 err = -ENOENT;
3370                 goto out_dput;
3371         }
3372
3373         inode = d_inode(dentry);
3374         dest = BTRFS_I(inode)->root;
3375         if (!capable(CAP_SYS_ADMIN)) {
3376                 /*
3377                  * Regular user.  Only allow this with a special mount
3378                  * option, when the user has write+exec access to the
3379                  * subvol root, and when rmdir(2) would have been
3380                  * allowed.
3381                  *
3382                  * Note that this is _not_ check that the subvol is
3383                  * empty or doesn't contain data that we wouldn't
3384                  * otherwise be able to delete.
3385                  *
3386                  * Users who want to delete empty subvols should try
3387                  * rmdir(2).
3388                  */
3389                 err = -EPERM;
3390                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3391                         goto out_dput;
3392
3393                 /*
3394                  * Do not allow deletion if the parent dir is the same
3395                  * as the dir to be deleted.  That means the ioctl
3396                  * must be called on the dentry referencing the root
3397                  * of the subvol, not a random directory contained
3398                  * within it.
3399                  */
3400                 err = -EINVAL;
3401                 if (root == dest)
3402                         goto out_dput;
3403
3404                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3405                 if (err)
3406                         goto out_dput;
3407         }
3408
3409         /* check if subvolume may be deleted by a user */
3410         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3411         if (err)
3412                 goto out_dput;
3413
3414         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3415                 err = -EINVAL;
3416                 goto out_dput;
3417         }
3418
3419         btrfs_inode_lock(inode, 0);
3420         err = btrfs_delete_subvolume(dir, dentry);
3421         btrfs_inode_unlock(inode, 0);
3422         if (!err)
3423                 d_delete_notify(dir, dentry);
3424
3425 out_dput:
3426         dput(dentry);
3427 out_unlock_dir:
3428         btrfs_inode_unlock(dir, 0);
3429 free_subvol_name:
3430         kfree(subvol_name_ptr);
3431 free_parent:
3432         if (destroy_parent)
3433                 dput(parent);
3434 out_drop_write:
3435         mnt_drop_write_file(file);
3436 out:
3437         kfree(vol_args2);
3438         kfree(vol_args);
3439         return err;
3440 }
3441
3442 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3443 {
3444         struct inode *inode = file_inode(file);
3445         struct btrfs_root *root = BTRFS_I(inode)->root;
3446         struct btrfs_ioctl_defrag_range_args range = {0};
3447         int ret;
3448
3449         ret = mnt_want_write_file(file);
3450         if (ret)
3451                 return ret;
3452
3453         if (btrfs_root_readonly(root)) {
3454                 ret = -EROFS;
3455                 goto out;
3456         }
3457
3458         switch (inode->i_mode & S_IFMT) {
3459         case S_IFDIR:
3460                 if (!capable(CAP_SYS_ADMIN)) {
3461                         ret = -EPERM;
3462                         goto out;
3463                 }
3464                 ret = btrfs_defrag_root(root);
3465                 break;
3466         case S_IFREG:
3467                 /*
3468                  * Note that this does not check the file descriptor for write
3469                  * access. This prevents defragmenting executables that are
3470                  * running and allows defrag on files open in read-only mode.
3471                  */
3472                 if (!capable(CAP_SYS_ADMIN) &&
3473                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3474                         ret = -EPERM;
3475                         goto out;
3476                 }
3477
3478                 if (argp) {
3479                         if (copy_from_user(&range, argp, sizeof(range))) {
3480                                 ret = -EFAULT;
3481                                 goto out;
3482                         }
3483                         /* compression requires us to start the IO */
3484                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3485                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3486                                 range.extent_thresh = (u32)-1;
3487                         }
3488                 } else {
3489                         /* the rest are all set to zero by kzalloc */
3490                         range.len = (u64)-1;
3491                 }
3492                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
3493                                         &range, BTRFS_OLDEST_GENERATION, 0);
3494                 if (ret > 0)
3495                         ret = 0;
3496                 break;
3497         default:
3498                 ret = -EINVAL;
3499         }
3500 out:
3501         mnt_drop_write_file(file);
3502         return ret;
3503 }
3504
3505 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3506 {
3507         struct btrfs_ioctl_vol_args *vol_args;
3508         bool restore_op = false;
3509         int ret;
3510
3511         if (!capable(CAP_SYS_ADMIN))
3512                 return -EPERM;
3513
3514         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3515                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
3516                 return -EINVAL;
3517         }
3518
3519         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
3520                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
3521                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3522
3523                 /*
3524                  * We can do the device add because we have a paused balanced,
3525                  * change the exclusive op type and remember we should bring
3526                  * back the paused balance
3527                  */
3528                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
3529                 btrfs_exclop_start_unlock(fs_info);
3530                 restore_op = true;
3531         }
3532
3533         vol_args = memdup_user(arg, sizeof(*vol_args));
3534         if (IS_ERR(vol_args)) {
3535                 ret = PTR_ERR(vol_args);
3536                 goto out;
3537         }
3538
3539         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3540         ret = btrfs_init_new_device(fs_info, vol_args->name);
3541
3542         if (!ret)
3543                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3544
3545         kfree(vol_args);
3546 out:
3547         if (restore_op)
3548                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
3549         else
3550                 btrfs_exclop_finish(fs_info);
3551         return ret;
3552 }
3553
3554 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3555 {
3556         BTRFS_DEV_LOOKUP_ARGS(args);
3557         struct inode *inode = file_inode(file);
3558         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3559         struct btrfs_ioctl_vol_args_v2 *vol_args;
3560         struct block_device *bdev = NULL;
3561         fmode_t mode;
3562         int ret;
3563         bool cancel = false;
3564
3565         if (!capable(CAP_SYS_ADMIN))
3566                 return -EPERM;
3567
3568         vol_args = memdup_user(arg, sizeof(*vol_args));
3569         if (IS_ERR(vol_args))
3570                 return PTR_ERR(vol_args);
3571
3572         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3573                 ret = -EOPNOTSUPP;
3574                 goto out;
3575         }
3576
3577         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3578         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3579                 args.devid = vol_args->devid;
3580         } else if (!strcmp("cancel", vol_args->name)) {
3581                 cancel = true;
3582         } else {
3583                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3584                 if (ret)
3585                         goto out;
3586         }
3587
3588         ret = mnt_want_write_file(file);
3589         if (ret)
3590                 goto out;
3591
3592         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3593                                            cancel);
3594         if (ret)
3595                 goto err_drop;
3596
3597         /* Exclusive operation is now claimed */
3598         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3599
3600         btrfs_exclop_finish(fs_info);
3601
3602         if (!ret) {
3603                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3604                         btrfs_info(fs_info, "device deleted: id %llu",
3605                                         vol_args->devid);
3606                 else
3607                         btrfs_info(fs_info, "device deleted: %s",
3608                                         vol_args->name);
3609         }
3610 err_drop:
3611         mnt_drop_write_file(file);
3612         if (bdev)
3613                 blkdev_put(bdev, mode);
3614 out:
3615         btrfs_put_dev_args_from_path(&args);
3616         kfree(vol_args);
3617         return ret;
3618 }
3619
3620 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3621 {
3622         BTRFS_DEV_LOOKUP_ARGS(args);
3623         struct inode *inode = file_inode(file);
3624         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3625         struct btrfs_ioctl_vol_args *vol_args;
3626         struct block_device *bdev = NULL;
3627         fmode_t mode;
3628         int ret;
3629         bool cancel = false;
3630
3631         if (!capable(CAP_SYS_ADMIN))
3632                 return -EPERM;
3633
3634         vol_args = memdup_user(arg, sizeof(*vol_args));
3635         if (IS_ERR(vol_args))
3636                 return PTR_ERR(vol_args);
3637
3638         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3639         if (!strcmp("cancel", vol_args->name)) {
3640                 cancel = true;
3641         } else {
3642                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3643                 if (ret)
3644                         goto out;
3645         }
3646
3647         ret = mnt_want_write_file(file);
3648         if (ret)
3649                 goto out;
3650
3651         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3652                                            cancel);
3653         if (ret == 0) {
3654                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3655                 if (!ret)
3656                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3657                 btrfs_exclop_finish(fs_info);
3658         }
3659
3660         mnt_drop_write_file(file);
3661         if (bdev)
3662                 blkdev_put(bdev, mode);
3663 out:
3664         btrfs_put_dev_args_from_path(&args);
3665         kfree(vol_args);
3666         return ret;
3667 }
3668
3669 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3670                                 void __user *arg)
3671 {
3672         struct btrfs_ioctl_fs_info_args *fi_args;
3673         struct btrfs_device *device;
3674         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3675         u64 flags_in;
3676         int ret = 0;
3677
3678         fi_args = memdup_user(arg, sizeof(*fi_args));
3679         if (IS_ERR(fi_args))
3680                 return PTR_ERR(fi_args);
3681
3682         flags_in = fi_args->flags;
3683         memset(fi_args, 0, sizeof(*fi_args));
3684
3685         rcu_read_lock();
3686         fi_args->num_devices = fs_devices->num_devices;
3687
3688         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3689                 if (device->devid > fi_args->max_id)
3690                         fi_args->max_id = device->devid;
3691         }
3692         rcu_read_unlock();
3693
3694         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3695         fi_args->nodesize = fs_info->nodesize;
3696         fi_args->sectorsize = fs_info->sectorsize;
3697         fi_args->clone_alignment = fs_info->sectorsize;
3698
3699         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3700                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3701                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3702                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3703         }
3704
3705         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3706                 fi_args->generation = fs_info->generation;
3707                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3708         }
3709
3710         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3711                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3712                        sizeof(fi_args->metadata_uuid));
3713                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3714         }
3715
3716         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3717                 ret = -EFAULT;
3718
3719         kfree(fi_args);
3720         return ret;
3721 }
3722
3723 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3724                                  void __user *arg)
3725 {
3726         BTRFS_DEV_LOOKUP_ARGS(args);
3727         struct btrfs_ioctl_dev_info_args *di_args;
3728         struct btrfs_device *dev;
3729         int ret = 0;
3730
3731         di_args = memdup_user(arg, sizeof(*di_args));
3732         if (IS_ERR(di_args))
3733                 return PTR_ERR(di_args);
3734
3735         args.devid = di_args->devid;
3736         if (!btrfs_is_empty_uuid(di_args->uuid))
3737                 args.uuid = di_args->uuid;
3738
3739         rcu_read_lock();
3740         dev = btrfs_find_device(fs_info->fs_devices, &args);
3741         if (!dev) {
3742                 ret = -ENODEV;
3743                 goto out;
3744         }
3745
3746         di_args->devid = dev->devid;
3747         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3748         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3749         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3750         if (dev->name) {
3751                 strncpy(di_args->path, rcu_str_deref(dev->name),
3752                                 sizeof(di_args->path) - 1);
3753                 di_args->path[sizeof(di_args->path) - 1] = 0;
3754         } else {
3755                 di_args->path[0] = '\0';
3756         }
3757
3758 out:
3759         rcu_read_unlock();
3760         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3761                 ret = -EFAULT;
3762
3763         kfree(di_args);
3764         return ret;
3765 }
3766
3767 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3768 {
3769         struct inode *inode = file_inode(file);
3770         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3771         struct btrfs_root *root = BTRFS_I(inode)->root;
3772         struct btrfs_root *new_root;
3773         struct btrfs_dir_item *di;
3774         struct btrfs_trans_handle *trans;
3775         struct btrfs_path *path = NULL;
3776         struct btrfs_disk_key disk_key;
3777         u64 objectid = 0;
3778         u64 dir_id;
3779         int ret;
3780
3781         if (!capable(CAP_SYS_ADMIN))
3782                 return -EPERM;
3783
3784         ret = mnt_want_write_file(file);
3785         if (ret)
3786                 return ret;
3787
3788         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3789                 ret = -EFAULT;
3790                 goto out;
3791         }
3792
3793         if (!objectid)
3794                 objectid = BTRFS_FS_TREE_OBJECTID;
3795
3796         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3797         if (IS_ERR(new_root)) {
3798                 ret = PTR_ERR(new_root);
3799                 goto out;
3800         }
3801         if (!is_fstree(new_root->root_key.objectid)) {
3802                 ret = -ENOENT;
3803                 goto out_free;
3804         }
3805
3806         path = btrfs_alloc_path();
3807         if (!path) {
3808                 ret = -ENOMEM;
3809                 goto out_free;
3810         }
3811
3812         trans = btrfs_start_transaction(root, 1);
3813         if (IS_ERR(trans)) {
3814                 ret = PTR_ERR(trans);
3815                 goto out_free;
3816         }
3817
3818         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3819         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3820                                    dir_id, "default", 7, 1);
3821         if (IS_ERR_OR_NULL(di)) {
3822                 btrfs_release_path(path);
3823                 btrfs_end_transaction(trans);
3824                 btrfs_err(fs_info,
3825                           "Umm, you don't have the default diritem, this isn't going to work");
3826                 ret = -ENOENT;
3827                 goto out_free;
3828         }
3829
3830         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3831         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3832         btrfs_mark_buffer_dirty(path->nodes[0]);
3833         btrfs_release_path(path);
3834
3835         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3836         btrfs_end_transaction(trans);
3837 out_free:
3838         btrfs_put_root(new_root);
3839         btrfs_free_path(path);
3840 out:
3841         mnt_drop_write_file(file);
3842         return ret;
3843 }
3844
3845 static void get_block_group_info(struct list_head *groups_list,
3846                                  struct btrfs_ioctl_space_info *space)
3847 {
3848         struct btrfs_block_group *block_group;
3849
3850         space->total_bytes = 0;
3851         space->used_bytes = 0;
3852         space->flags = 0;
3853         list_for_each_entry(block_group, groups_list, list) {
3854                 space->flags = block_group->flags;
3855                 space->total_bytes += block_group->length;
3856                 space->used_bytes += block_group->used;
3857         }
3858 }
3859
3860 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3861                                    void __user *arg)
3862 {
3863         struct btrfs_ioctl_space_args space_args;
3864         struct btrfs_ioctl_space_info space;
3865         struct btrfs_ioctl_space_info *dest;
3866         struct btrfs_ioctl_space_info *dest_orig;
3867         struct btrfs_ioctl_space_info __user *user_dest;
3868         struct btrfs_space_info *info;
3869         static const u64 types[] = {
3870                 BTRFS_BLOCK_GROUP_DATA,
3871                 BTRFS_BLOCK_GROUP_SYSTEM,
3872                 BTRFS_BLOCK_GROUP_METADATA,
3873                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3874         };
3875         int num_types = 4;
3876         int alloc_size;
3877         int ret = 0;
3878         u64 slot_count = 0;
3879         int i, c;
3880
3881         if (copy_from_user(&space_args,
3882                            (struct btrfs_ioctl_space_args __user *)arg,
3883                            sizeof(space_args)))
3884                 return -EFAULT;
3885
3886         for (i = 0; i < num_types; i++) {
3887                 struct btrfs_space_info *tmp;
3888
3889                 info = NULL;
3890                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3891                         if (tmp->flags == types[i]) {
3892                                 info = tmp;
3893                                 break;
3894                         }
3895                 }
3896
3897                 if (!info)
3898                         continue;
3899
3900                 down_read(&info->groups_sem);
3901                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3902                         if (!list_empty(&info->block_groups[c]))
3903                                 slot_count++;
3904                 }
3905                 up_read(&info->groups_sem);
3906         }
3907
3908         /*
3909          * Global block reserve, exported as a space_info
3910          */
3911         slot_count++;
3912
3913         /* space_slots == 0 means they are asking for a count */
3914         if (space_args.space_slots == 0) {
3915                 space_args.total_spaces = slot_count;
3916                 goto out;
3917         }
3918
3919         slot_count = min_t(u64, space_args.space_slots, slot_count);
3920
3921         alloc_size = sizeof(*dest) * slot_count;
3922
3923         /* we generally have at most 6 or so space infos, one for each raid
3924          * level.  So, a whole page should be more than enough for everyone
3925          */
3926         if (alloc_size > PAGE_SIZE)
3927                 return -ENOMEM;
3928
3929         space_args.total_spaces = 0;
3930         dest = kmalloc(alloc_size, GFP_KERNEL);
3931         if (!dest)
3932                 return -ENOMEM;
3933         dest_orig = dest;
3934
3935         /* now we have a buffer to copy into */
3936         for (i = 0; i < num_types; i++) {
3937                 struct btrfs_space_info *tmp;
3938
3939                 if (!slot_count)
3940                         break;
3941
3942                 info = NULL;
3943                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3944                         if (tmp->flags == types[i]) {
3945                                 info = tmp;
3946                                 break;
3947                         }
3948                 }
3949
3950                 if (!info)
3951                         continue;
3952                 down_read(&info->groups_sem);
3953                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3954                         if (!list_empty(&info->block_groups[c])) {
3955                                 get_block_group_info(&info->block_groups[c],
3956                                                      &space);
3957                                 memcpy(dest, &space, sizeof(space));
3958                                 dest++;
3959                                 space_args.total_spaces++;
3960                                 slot_count--;
3961                         }
3962                         if (!slot_count)
3963                                 break;
3964                 }
3965                 up_read(&info->groups_sem);
3966         }
3967
3968         /*
3969          * Add global block reserve
3970          */
3971         if (slot_count) {
3972                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3973
3974                 spin_lock(&block_rsv->lock);
3975                 space.total_bytes = block_rsv->size;
3976                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3977                 spin_unlock(&block_rsv->lock);
3978                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3979                 memcpy(dest, &space, sizeof(space));
3980                 space_args.total_spaces++;
3981         }
3982
3983         user_dest = (struct btrfs_ioctl_space_info __user *)
3984                 (arg + sizeof(struct btrfs_ioctl_space_args));
3985
3986         if (copy_to_user(user_dest, dest_orig, alloc_size))
3987                 ret = -EFAULT;
3988
3989         kfree(dest_orig);
3990 out:
3991         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3992                 ret = -EFAULT;
3993
3994         return ret;
3995 }
3996
3997 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3998                                             void __user *argp)
3999 {
4000         struct btrfs_trans_handle *trans;
4001         u64 transid;
4002
4003         trans = btrfs_attach_transaction_barrier(root);
4004         if (IS_ERR(trans)) {
4005                 if (PTR_ERR(trans) != -ENOENT)
4006                         return PTR_ERR(trans);
4007
4008                 /* No running transaction, don't bother */
4009                 transid = root->fs_info->last_trans_committed;
4010                 goto out;
4011         }
4012         transid = trans->transid;
4013         btrfs_commit_transaction_async(trans);
4014 out:
4015         if (argp)
4016                 if (copy_to_user(argp, &transid, sizeof(transid)))
4017                         return -EFAULT;
4018         return 0;
4019 }
4020
4021 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4022                                            void __user *argp)
4023 {
4024         u64 transid;
4025
4026         if (argp) {
4027                 if (copy_from_user(&transid, argp, sizeof(transid)))
4028                         return -EFAULT;
4029         } else {
4030                 transid = 0;  /* current trans */
4031         }
4032         return btrfs_wait_for_commit(fs_info, transid);
4033 }
4034
4035 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4036 {
4037         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4038         struct btrfs_ioctl_scrub_args *sa;
4039         int ret;
4040
4041         if (!capable(CAP_SYS_ADMIN))
4042                 return -EPERM;
4043
4044         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4045                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
4046                 return -EINVAL;
4047         }
4048
4049         sa = memdup_user(arg, sizeof(*sa));
4050         if (IS_ERR(sa))
4051                 return PTR_ERR(sa);
4052
4053         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4054                 ret = mnt_want_write_file(file);
4055                 if (ret)
4056                         goto out;
4057         }
4058
4059         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4060                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4061                               0);
4062
4063         /*
4064          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4065          * error. This is important as it allows user space to know how much
4066          * progress scrub has done. For example, if scrub is canceled we get
4067          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4068          * space. Later user space can inspect the progress from the structure
4069          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4070          * previously (btrfs-progs does this).
4071          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4072          * then return -EFAULT to signal the structure was not copied or it may
4073          * be corrupt and unreliable due to a partial copy.
4074          */
4075         if (copy_to_user(arg, sa, sizeof(*sa)))
4076                 ret = -EFAULT;
4077
4078         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4079                 mnt_drop_write_file(file);
4080 out:
4081         kfree(sa);
4082         return ret;
4083 }
4084
4085 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4086 {
4087         if (!capable(CAP_SYS_ADMIN))
4088                 return -EPERM;
4089
4090         return btrfs_scrub_cancel(fs_info);
4091 }
4092
4093 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4094                                        void __user *arg)
4095 {
4096         struct btrfs_ioctl_scrub_args *sa;
4097         int ret;
4098
4099         if (!capable(CAP_SYS_ADMIN))
4100                 return -EPERM;
4101
4102         sa = memdup_user(arg, sizeof(*sa));
4103         if (IS_ERR(sa))
4104                 return PTR_ERR(sa);
4105
4106         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4107
4108         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4109                 ret = -EFAULT;
4110
4111         kfree(sa);
4112         return ret;
4113 }
4114
4115 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4116                                       void __user *arg)
4117 {
4118         struct btrfs_ioctl_get_dev_stats *sa;
4119         int ret;
4120
4121         sa = memdup_user(arg, sizeof(*sa));
4122         if (IS_ERR(sa))
4123                 return PTR_ERR(sa);
4124
4125         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4126                 kfree(sa);
4127                 return -EPERM;
4128         }
4129
4130         ret = btrfs_get_dev_stats(fs_info, sa);
4131
4132         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4133                 ret = -EFAULT;
4134
4135         kfree(sa);
4136         return ret;
4137 }
4138
4139 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4140                                     void __user *arg)
4141 {
4142         struct btrfs_ioctl_dev_replace_args *p;
4143         int ret;
4144
4145         if (!capable(CAP_SYS_ADMIN))
4146                 return -EPERM;
4147
4148         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4149                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
4150                 return -EINVAL;
4151         }
4152
4153         p = memdup_user(arg, sizeof(*p));
4154         if (IS_ERR(p))
4155                 return PTR_ERR(p);
4156
4157         switch (p->cmd) {
4158         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4159                 if (sb_rdonly(fs_info->sb)) {
4160                         ret = -EROFS;
4161                         goto out;
4162                 }
4163                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
4164                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4165                 } else {
4166                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4167                         btrfs_exclop_finish(fs_info);
4168                 }
4169                 break;
4170         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4171                 btrfs_dev_replace_status(fs_info, p);
4172                 ret = 0;
4173                 break;
4174         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4175                 p->result = btrfs_dev_replace_cancel(fs_info);
4176                 ret = 0;
4177                 break;
4178         default:
4179                 ret = -EINVAL;
4180                 break;
4181         }
4182
4183         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4184                 ret = -EFAULT;
4185 out:
4186         kfree(p);
4187         return ret;
4188 }
4189
4190 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4191 {
4192         int ret = 0;
4193         int i;
4194         u64 rel_ptr;
4195         int size;
4196         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4197         struct inode_fs_paths *ipath = NULL;
4198         struct btrfs_path *path;
4199
4200         if (!capable(CAP_DAC_READ_SEARCH))
4201                 return -EPERM;
4202
4203         path = btrfs_alloc_path();
4204         if (!path) {
4205                 ret = -ENOMEM;
4206                 goto out;
4207         }
4208
4209         ipa = memdup_user(arg, sizeof(*ipa));
4210         if (IS_ERR(ipa)) {
4211                 ret = PTR_ERR(ipa);
4212                 ipa = NULL;
4213                 goto out;
4214         }
4215
4216         size = min_t(u32, ipa->size, 4096);
4217         ipath = init_ipath(size, root, path);
4218         if (IS_ERR(ipath)) {
4219                 ret = PTR_ERR(ipath);
4220                 ipath = NULL;
4221                 goto out;
4222         }
4223
4224         ret = paths_from_inode(ipa->inum, ipath);
4225         if (ret < 0)
4226                 goto out;
4227
4228         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4229                 rel_ptr = ipath->fspath->val[i] -
4230                           (u64)(unsigned long)ipath->fspath->val;
4231                 ipath->fspath->val[i] = rel_ptr;
4232         }
4233
4234         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4235                            ipath->fspath, size);
4236         if (ret) {
4237                 ret = -EFAULT;
4238                 goto out;
4239         }
4240
4241 out:
4242         btrfs_free_path(path);
4243         free_ipath(ipath);
4244         kfree(ipa);
4245
4246         return ret;
4247 }
4248
4249 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4250                                         void __user *arg, int version)
4251 {
4252         int ret = 0;
4253         int size;
4254         struct btrfs_ioctl_logical_ino_args *loi;
4255         struct btrfs_data_container *inodes = NULL;
4256         struct btrfs_path *path = NULL;
4257         bool ignore_offset;
4258
4259         if (!capable(CAP_SYS_ADMIN))
4260                 return -EPERM;
4261
4262         loi = memdup_user(arg, sizeof(*loi));
4263         if (IS_ERR(loi))
4264                 return PTR_ERR(loi);
4265
4266         if (version == 1) {
4267                 ignore_offset = false;
4268                 size = min_t(u32, loi->size, SZ_64K);
4269         } else {
4270                 /* All reserved bits must be 0 for now */
4271                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4272                         ret = -EINVAL;
4273                         goto out_loi;
4274                 }
4275                 /* Only accept flags we have defined so far */
4276                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4277                         ret = -EINVAL;
4278                         goto out_loi;
4279                 }
4280                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4281                 size = min_t(u32, loi->size, SZ_16M);
4282         }
4283
4284         path = btrfs_alloc_path();
4285         if (!path) {
4286                 ret = -ENOMEM;
4287                 goto out;
4288         }
4289
4290         inodes = init_data_container(size);
4291         if (IS_ERR(inodes)) {
4292                 ret = PTR_ERR(inodes);
4293                 inodes = NULL;
4294                 goto out;
4295         }
4296
4297         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4298                                           inodes, ignore_offset);
4299         if (ret == -EINVAL)
4300                 ret = -ENOENT;
4301         if (ret < 0)
4302                 goto out;
4303
4304         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4305                            size);
4306         if (ret)
4307                 ret = -EFAULT;
4308
4309 out:
4310         btrfs_free_path(path);
4311         kvfree(inodes);
4312 out_loi:
4313         kfree(loi);
4314
4315         return ret;
4316 }
4317
4318 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4319                                struct btrfs_ioctl_balance_args *bargs)
4320 {
4321         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4322
4323         bargs->flags = bctl->flags;
4324
4325         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4326                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4327         if (atomic_read(&fs_info->balance_pause_req))
4328                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4329         if (atomic_read(&fs_info->balance_cancel_req))
4330                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4331
4332         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4333         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4334         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4335
4336         spin_lock(&fs_info->balance_lock);
4337         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4338         spin_unlock(&fs_info->balance_lock);
4339 }
4340
4341 /**
4342  * Try to acquire fs_info::balance_mutex as well as set BTRFS_EXLCOP_BALANCE as
4343  * required.
4344  *
4345  * @fs_info:       the filesystem
4346  * @excl_acquired: ptr to boolean value which is set to false in case balance
4347  *                 is being resumed
4348  *
4349  * Return 0 on success in which case both fs_info::balance is acquired as well
4350  * as exclusive ops are blocked. In case of failure return an error code.
4351  */
4352 static int btrfs_try_lock_balance(struct btrfs_fs_info *fs_info, bool *excl_acquired)
4353 {
4354         int ret;
4355
4356         /*
4357          * Exclusive operation is locked. Three possibilities:
4358          *   (1) some other op is running
4359          *   (2) balance is running
4360          *   (3) balance is paused -- special case (think resume)
4361          */
4362         while (1) {
4363                 if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4364                         *excl_acquired = true;
4365                         mutex_lock(&fs_info->balance_mutex);
4366                         return 0;
4367                 }
4368
4369                 mutex_lock(&fs_info->balance_mutex);
4370                 if (fs_info->balance_ctl) {
4371                         /* This is either (2) or (3) */
4372                         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4373                                 /* This is (2) */
4374                                 ret = -EINPROGRESS;
4375                                 goto out_failure;
4376
4377                         } else {
4378                                 mutex_unlock(&fs_info->balance_mutex);
4379                                 /*
4380                                  * Lock released to allow other waiters to
4381                                  * continue, we'll reexamine the status again.
4382                                  */
4383                                 mutex_lock(&fs_info->balance_mutex);
4384
4385                                 if (fs_info->balance_ctl &&
4386                                     !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4387                                         /* This is (3) */
4388                                         *excl_acquired = false;
4389                                         return 0;
4390                                 }
4391                         }
4392                 } else {
4393                         /* This is (1) */
4394                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4395                         goto out_failure;
4396                 }
4397
4398                 mutex_unlock(&fs_info->balance_mutex);
4399         }
4400
4401 out_failure:
4402         mutex_unlock(&fs_info->balance_mutex);
4403         *excl_acquired = false;
4404         return ret;
4405 }
4406
4407 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4408 {
4409         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4410         struct btrfs_fs_info *fs_info = root->fs_info;
4411         struct btrfs_ioctl_balance_args *bargs;
4412         struct btrfs_balance_control *bctl;
4413         bool need_unlock = true;
4414         int ret;
4415
4416         if (!capable(CAP_SYS_ADMIN))
4417                 return -EPERM;
4418
4419         ret = mnt_want_write_file(file);
4420         if (ret)
4421                 return ret;
4422
4423         bargs = memdup_user(arg, sizeof(*bargs));
4424         if (IS_ERR(bargs)) {
4425                 ret = PTR_ERR(bargs);
4426                 bargs = NULL;
4427                 goto out;
4428         }
4429
4430         ret = btrfs_try_lock_balance(fs_info, &need_unlock);
4431         if (ret)
4432                 goto out;
4433
4434         lockdep_assert_held(&fs_info->balance_mutex);
4435
4436         if (bargs->flags & BTRFS_BALANCE_RESUME) {
4437                 if (!fs_info->balance_ctl) {
4438                         ret = -ENOTCONN;
4439                         goto out_unlock;
4440                 }
4441
4442                 bctl = fs_info->balance_ctl;
4443                 spin_lock(&fs_info->balance_lock);
4444                 bctl->flags |= BTRFS_BALANCE_RESUME;
4445                 spin_unlock(&fs_info->balance_lock);
4446                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
4447
4448                 goto do_balance;
4449         }
4450
4451         if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4452                 ret = -EINVAL;
4453                 goto out_unlock;
4454         }
4455
4456         if (fs_info->balance_ctl) {
4457                 ret = -EINPROGRESS;
4458                 goto out_unlock;
4459         }
4460
4461         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4462         if (!bctl) {
4463                 ret = -ENOMEM;
4464                 goto out_unlock;
4465         }
4466
4467         memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4468         memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4469         memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4470
4471         bctl->flags = bargs->flags;
4472 do_balance:
4473         /*
4474          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4475          * bctl is freed in reset_balance_state, or, if restriper was paused
4476          * all the way until unmount, in free_fs_info.  The flag should be
4477          * cleared after reset_balance_state.
4478          */
4479         need_unlock = false;
4480
4481         ret = btrfs_balance(fs_info, bctl, bargs);
4482         bctl = NULL;
4483
4484         if (ret == 0 || ret == -ECANCELED) {
4485                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4486                         ret = -EFAULT;
4487         }
4488
4489         kfree(bctl);
4490 out_unlock:
4491         mutex_unlock(&fs_info->balance_mutex);
4492         if (need_unlock)
4493                 btrfs_exclop_finish(fs_info);
4494 out:
4495         mnt_drop_write_file(file);
4496         kfree(bargs);
4497         return ret;
4498 }
4499
4500 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4501 {
4502         if (!capable(CAP_SYS_ADMIN))
4503                 return -EPERM;
4504
4505         switch (cmd) {
4506         case BTRFS_BALANCE_CTL_PAUSE:
4507                 return btrfs_pause_balance(fs_info);
4508         case BTRFS_BALANCE_CTL_CANCEL:
4509                 return btrfs_cancel_balance(fs_info);
4510         }
4511
4512         return -EINVAL;
4513 }
4514
4515 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4516                                          void __user *arg)
4517 {
4518         struct btrfs_ioctl_balance_args *bargs;
4519         int ret = 0;
4520
4521         if (!capable(CAP_SYS_ADMIN))
4522                 return -EPERM;
4523
4524         mutex_lock(&fs_info->balance_mutex);
4525         if (!fs_info->balance_ctl) {
4526                 ret = -ENOTCONN;
4527                 goto out;
4528         }
4529
4530         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4531         if (!bargs) {
4532                 ret = -ENOMEM;
4533                 goto out;
4534         }
4535
4536         btrfs_update_ioctl_balance_args(fs_info, bargs);
4537
4538         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4539                 ret = -EFAULT;
4540
4541         kfree(bargs);
4542 out:
4543         mutex_unlock(&fs_info->balance_mutex);
4544         return ret;
4545 }
4546
4547 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4548 {
4549         struct inode *inode = file_inode(file);
4550         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4551         struct btrfs_ioctl_quota_ctl_args *sa;
4552         int ret;
4553
4554         if (!capable(CAP_SYS_ADMIN))
4555                 return -EPERM;
4556
4557         ret = mnt_want_write_file(file);
4558         if (ret)
4559                 return ret;
4560
4561         sa = memdup_user(arg, sizeof(*sa));
4562         if (IS_ERR(sa)) {
4563                 ret = PTR_ERR(sa);
4564                 goto drop_write;
4565         }
4566
4567         down_write(&fs_info->subvol_sem);
4568
4569         switch (sa->cmd) {
4570         case BTRFS_QUOTA_CTL_ENABLE:
4571                 ret = btrfs_quota_enable(fs_info);
4572                 break;
4573         case BTRFS_QUOTA_CTL_DISABLE:
4574                 ret = btrfs_quota_disable(fs_info);
4575                 break;
4576         default:
4577                 ret = -EINVAL;
4578                 break;
4579         }
4580
4581         kfree(sa);
4582         up_write(&fs_info->subvol_sem);
4583 drop_write:
4584         mnt_drop_write_file(file);
4585         return ret;
4586 }
4587
4588 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4589 {
4590         struct inode *inode = file_inode(file);
4591         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4592         struct btrfs_root *root = BTRFS_I(inode)->root;
4593         struct btrfs_ioctl_qgroup_assign_args *sa;
4594         struct btrfs_trans_handle *trans;
4595         int ret;
4596         int err;
4597
4598         if (!capable(CAP_SYS_ADMIN))
4599                 return -EPERM;
4600
4601         ret = mnt_want_write_file(file);
4602         if (ret)
4603                 return ret;
4604
4605         sa = memdup_user(arg, sizeof(*sa));
4606         if (IS_ERR(sa)) {
4607                 ret = PTR_ERR(sa);
4608                 goto drop_write;
4609         }
4610
4611         trans = btrfs_join_transaction(root);
4612         if (IS_ERR(trans)) {
4613                 ret = PTR_ERR(trans);
4614                 goto out;
4615         }
4616
4617         if (sa->assign) {
4618                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4619         } else {
4620                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4621         }
4622
4623         /* update qgroup status and info */
4624         err = btrfs_run_qgroups(trans);
4625         if (err < 0)
4626                 btrfs_handle_fs_error(fs_info, err,
4627                                       "failed to update qgroup status and info");
4628         err = btrfs_end_transaction(trans);
4629         if (err && !ret)
4630                 ret = err;
4631
4632 out:
4633         kfree(sa);
4634 drop_write:
4635         mnt_drop_write_file(file);
4636         return ret;
4637 }
4638
4639 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4640 {
4641         struct inode *inode = file_inode(file);
4642         struct btrfs_root *root = BTRFS_I(inode)->root;
4643         struct btrfs_ioctl_qgroup_create_args *sa;
4644         struct btrfs_trans_handle *trans;
4645         int ret;
4646         int err;
4647
4648         if (!capable(CAP_SYS_ADMIN))
4649                 return -EPERM;
4650
4651         ret = mnt_want_write_file(file);
4652         if (ret)
4653                 return ret;
4654
4655         sa = memdup_user(arg, sizeof(*sa));
4656         if (IS_ERR(sa)) {
4657                 ret = PTR_ERR(sa);
4658                 goto drop_write;
4659         }
4660
4661         if (!sa->qgroupid) {
4662                 ret = -EINVAL;
4663                 goto out;
4664         }
4665
4666         trans = btrfs_join_transaction(root);
4667         if (IS_ERR(trans)) {
4668                 ret = PTR_ERR(trans);
4669                 goto out;
4670         }
4671
4672         if (sa->create) {
4673                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4674         } else {
4675                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4676         }
4677
4678         err = btrfs_end_transaction(trans);
4679         if (err && !ret)
4680                 ret = err;
4681
4682 out:
4683         kfree(sa);
4684 drop_write:
4685         mnt_drop_write_file(file);
4686         return ret;
4687 }
4688
4689 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4690 {
4691         struct inode *inode = file_inode(file);
4692         struct btrfs_root *root = BTRFS_I(inode)->root;
4693         struct btrfs_ioctl_qgroup_limit_args *sa;
4694         struct btrfs_trans_handle *trans;
4695         int ret;
4696         int err;
4697         u64 qgroupid;
4698
4699         if (!capable(CAP_SYS_ADMIN))
4700                 return -EPERM;
4701
4702         ret = mnt_want_write_file(file);
4703         if (ret)
4704                 return ret;
4705
4706         sa = memdup_user(arg, sizeof(*sa));
4707         if (IS_ERR(sa)) {
4708                 ret = PTR_ERR(sa);
4709                 goto drop_write;
4710         }
4711
4712         trans = btrfs_join_transaction(root);
4713         if (IS_ERR(trans)) {
4714                 ret = PTR_ERR(trans);
4715                 goto out;
4716         }
4717
4718         qgroupid = sa->qgroupid;
4719         if (!qgroupid) {
4720                 /* take the current subvol as qgroup */
4721                 qgroupid = root->root_key.objectid;
4722         }
4723
4724         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4725
4726         err = btrfs_end_transaction(trans);
4727         if (err && !ret)
4728                 ret = err;
4729
4730 out:
4731         kfree(sa);
4732 drop_write:
4733         mnt_drop_write_file(file);
4734         return ret;
4735 }
4736
4737 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4738 {
4739         struct inode *inode = file_inode(file);
4740         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4741         struct btrfs_ioctl_quota_rescan_args *qsa;
4742         int ret;
4743
4744         if (!capable(CAP_SYS_ADMIN))
4745                 return -EPERM;
4746
4747         ret = mnt_want_write_file(file);
4748         if (ret)
4749                 return ret;
4750
4751         qsa = memdup_user(arg, sizeof(*qsa));
4752         if (IS_ERR(qsa)) {
4753                 ret = PTR_ERR(qsa);
4754                 goto drop_write;
4755         }
4756
4757         if (qsa->flags) {
4758                 ret = -EINVAL;
4759                 goto out;
4760         }
4761
4762         ret = btrfs_qgroup_rescan(fs_info);
4763
4764 out:
4765         kfree(qsa);
4766 drop_write:
4767         mnt_drop_write_file(file);
4768         return ret;
4769 }
4770
4771 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4772                                                 void __user *arg)
4773 {
4774         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4775
4776         if (!capable(CAP_SYS_ADMIN))
4777                 return -EPERM;
4778
4779         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4780                 qsa.flags = 1;
4781                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4782         }
4783
4784         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4785                 return -EFAULT;
4786
4787         return 0;
4788 }
4789
4790 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4791                                                 void __user *arg)
4792 {
4793         if (!capable(CAP_SYS_ADMIN))
4794                 return -EPERM;
4795
4796         return btrfs_qgroup_wait_for_completion(fs_info, true);
4797 }
4798
4799 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4800                                             struct user_namespace *mnt_userns,
4801                                             struct btrfs_ioctl_received_subvol_args *sa)
4802 {
4803         struct inode *inode = file_inode(file);
4804         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4805         struct btrfs_root *root = BTRFS_I(inode)->root;
4806         struct btrfs_root_item *root_item = &root->root_item;
4807         struct btrfs_trans_handle *trans;
4808         struct timespec64 ct = current_time(inode);
4809         int ret = 0;
4810         int received_uuid_changed;
4811
4812         if (!inode_owner_or_capable(mnt_userns, inode))
4813                 return -EPERM;
4814
4815         ret = mnt_want_write_file(file);
4816         if (ret < 0)
4817                 return ret;
4818
4819         down_write(&fs_info->subvol_sem);
4820
4821         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4822                 ret = -EINVAL;
4823                 goto out;
4824         }
4825
4826         if (btrfs_root_readonly(root)) {
4827                 ret = -EROFS;
4828                 goto out;
4829         }
4830
4831         /*
4832          * 1 - root item
4833          * 2 - uuid items (received uuid + subvol uuid)
4834          */
4835         trans = btrfs_start_transaction(root, 3);
4836         if (IS_ERR(trans)) {
4837                 ret = PTR_ERR(trans);
4838                 trans = NULL;
4839                 goto out;
4840         }
4841
4842         sa->rtransid = trans->transid;
4843         sa->rtime.sec = ct.tv_sec;
4844         sa->rtime.nsec = ct.tv_nsec;
4845
4846         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4847                                        BTRFS_UUID_SIZE);
4848         if (received_uuid_changed &&
4849             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4850                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4851                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4852                                           root->root_key.objectid);
4853                 if (ret && ret != -ENOENT) {
4854                         btrfs_abort_transaction(trans, ret);
4855                         btrfs_end_transaction(trans);
4856                         goto out;
4857                 }
4858         }
4859         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4860         btrfs_set_root_stransid(root_item, sa->stransid);
4861         btrfs_set_root_rtransid(root_item, sa->rtransid);
4862         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4863         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4864         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4865         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4866
4867         ret = btrfs_update_root(trans, fs_info->tree_root,
4868                                 &root->root_key, &root->root_item);
4869         if (ret < 0) {
4870                 btrfs_end_transaction(trans);
4871                 goto out;
4872         }
4873         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4874                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4875                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4876                                           root->root_key.objectid);
4877                 if (ret < 0 && ret != -EEXIST) {
4878                         btrfs_abort_transaction(trans, ret);
4879                         btrfs_end_transaction(trans);
4880                         goto out;
4881                 }
4882         }
4883         ret = btrfs_commit_transaction(trans);
4884 out:
4885         up_write(&fs_info->subvol_sem);
4886         mnt_drop_write_file(file);
4887         return ret;
4888 }
4889
4890 #ifdef CONFIG_64BIT
4891 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4892                                                 void __user *arg)
4893 {
4894         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4895         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4896         int ret = 0;
4897
4898         args32 = memdup_user(arg, sizeof(*args32));
4899         if (IS_ERR(args32))
4900                 return PTR_ERR(args32);
4901
4902         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4903         if (!args64) {
4904                 ret = -ENOMEM;
4905                 goto out;
4906         }
4907
4908         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4909         args64->stransid = args32->stransid;
4910         args64->rtransid = args32->rtransid;
4911         args64->stime.sec = args32->stime.sec;
4912         args64->stime.nsec = args32->stime.nsec;
4913         args64->rtime.sec = args32->rtime.sec;
4914         args64->rtime.nsec = args32->rtime.nsec;
4915         args64->flags = args32->flags;
4916
4917         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4918         if (ret)
4919                 goto out;
4920
4921         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4922         args32->stransid = args64->stransid;
4923         args32->rtransid = args64->rtransid;
4924         args32->stime.sec = args64->stime.sec;
4925         args32->stime.nsec = args64->stime.nsec;
4926         args32->rtime.sec = args64->rtime.sec;
4927         args32->rtime.nsec = args64->rtime.nsec;
4928         args32->flags = args64->flags;
4929
4930         ret = copy_to_user(arg, args32, sizeof(*args32));
4931         if (ret)
4932                 ret = -EFAULT;
4933
4934 out:
4935         kfree(args32);
4936         kfree(args64);
4937         return ret;
4938 }
4939 #endif
4940
4941 static long btrfs_ioctl_set_received_subvol(struct file *file,
4942                                             void __user *arg)
4943 {
4944         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4945         int ret = 0;
4946
4947         sa = memdup_user(arg, sizeof(*sa));
4948         if (IS_ERR(sa))
4949                 return PTR_ERR(sa);
4950
4951         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4952
4953         if (ret)
4954                 goto out;
4955
4956         ret = copy_to_user(arg, sa, sizeof(*sa));
4957         if (ret)
4958                 ret = -EFAULT;
4959
4960 out:
4961         kfree(sa);
4962         return ret;
4963 }
4964
4965 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4966                                         void __user *arg)
4967 {
4968         size_t len;
4969         int ret;
4970         char label[BTRFS_LABEL_SIZE];
4971
4972         spin_lock(&fs_info->super_lock);
4973         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4974         spin_unlock(&fs_info->super_lock);
4975
4976         len = strnlen(label, BTRFS_LABEL_SIZE);
4977
4978         if (len == BTRFS_LABEL_SIZE) {
4979                 btrfs_warn(fs_info,
4980                            "label is too long, return the first %zu bytes",
4981                            --len);
4982         }
4983
4984         ret = copy_to_user(arg, label, len);
4985
4986         return ret ? -EFAULT : 0;
4987 }
4988
4989 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4990 {
4991         struct inode *inode = file_inode(file);
4992         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4993         struct btrfs_root *root = BTRFS_I(inode)->root;
4994         struct btrfs_super_block *super_block = fs_info->super_copy;
4995         struct btrfs_trans_handle *trans;
4996         char label[BTRFS_LABEL_SIZE];
4997         int ret;
4998
4999         if (!capable(CAP_SYS_ADMIN))
5000                 return -EPERM;
5001
5002         if (copy_from_user(label, arg, sizeof(label)))
5003                 return -EFAULT;
5004
5005         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5006                 btrfs_err(fs_info,
5007                           "unable to set label with more than %d bytes",
5008                           BTRFS_LABEL_SIZE - 1);
5009                 return -EINVAL;
5010         }
5011
5012         ret = mnt_want_write_file(file);
5013         if (ret)
5014                 return ret;
5015
5016         trans = btrfs_start_transaction(root, 0);
5017         if (IS_ERR(trans)) {
5018                 ret = PTR_ERR(trans);
5019                 goto out_unlock;
5020         }
5021
5022         spin_lock(&fs_info->super_lock);
5023         strcpy(super_block->label, label);
5024         spin_unlock(&fs_info->super_lock);
5025         ret = btrfs_commit_transaction(trans);
5026
5027 out_unlock:
5028         mnt_drop_write_file(file);
5029         return ret;
5030 }
5031
5032 #define INIT_FEATURE_FLAGS(suffix) \
5033         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5034           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5035           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5036
5037 int btrfs_ioctl_get_supported_features(void __user *arg)
5038 {
5039         static const struct btrfs_ioctl_feature_flags features[3] = {
5040                 INIT_FEATURE_FLAGS(SUPP),
5041                 INIT_FEATURE_FLAGS(SAFE_SET),
5042                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5043         };
5044
5045         if (copy_to_user(arg, &features, sizeof(features)))
5046                 return -EFAULT;
5047
5048         return 0;
5049 }
5050
5051 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5052                                         void __user *arg)
5053 {
5054         struct btrfs_super_block *super_block = fs_info->super_copy;
5055         struct btrfs_ioctl_feature_flags features;
5056
5057         features.compat_flags = btrfs_super_compat_flags(super_block);
5058         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5059         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5060
5061         if (copy_to_user(arg, &features, sizeof(features)))
5062                 return -EFAULT;
5063
5064         return 0;
5065 }
5066
5067 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5068                               enum btrfs_feature_set set,
5069                               u64 change_mask, u64 flags, u64 supported_flags,
5070                               u64 safe_set, u64 safe_clear)
5071 {
5072         const char *type = btrfs_feature_set_name(set);
5073         char *names;
5074         u64 disallowed, unsupported;
5075         u64 set_mask = flags & change_mask;
5076         u64 clear_mask = ~flags & change_mask;
5077
5078         unsupported = set_mask & ~supported_flags;
5079         if (unsupported) {
5080                 names = btrfs_printable_features(set, unsupported);
5081                 if (names) {
5082                         btrfs_warn(fs_info,
5083                                    "this kernel does not support the %s feature bit%s",
5084                                    names, strchr(names, ',') ? "s" : "");
5085                         kfree(names);
5086                 } else
5087                         btrfs_warn(fs_info,
5088                                    "this kernel does not support %s bits 0x%llx",
5089                                    type, unsupported);
5090                 return -EOPNOTSUPP;
5091         }
5092
5093         disallowed = set_mask & ~safe_set;
5094         if (disallowed) {
5095                 names = btrfs_printable_features(set, disallowed);
5096                 if (names) {
5097                         btrfs_warn(fs_info,
5098                                    "can't set the %s feature bit%s while mounted",
5099                                    names, strchr(names, ',') ? "s" : "");
5100                         kfree(names);
5101                 } else
5102                         btrfs_warn(fs_info,
5103                                    "can't set %s bits 0x%llx while mounted",
5104                                    type, disallowed);
5105                 return -EPERM;
5106         }
5107
5108         disallowed = clear_mask & ~safe_clear;
5109         if (disallowed) {
5110                 names = btrfs_printable_features(set, disallowed);
5111                 if (names) {
5112                         btrfs_warn(fs_info,
5113                                    "can't clear the %s feature bit%s while mounted",
5114                                    names, strchr(names, ',') ? "s" : "");
5115                         kfree(names);
5116                 } else
5117                         btrfs_warn(fs_info,
5118                                    "can't clear %s bits 0x%llx while mounted",
5119                                    type, disallowed);
5120                 return -EPERM;
5121         }
5122
5123         return 0;
5124 }
5125
5126 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5127 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5128                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5129                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5130                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5131
5132 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5133 {
5134         struct inode *inode = file_inode(file);
5135         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5136         struct btrfs_root *root = BTRFS_I(inode)->root;
5137         struct btrfs_super_block *super_block = fs_info->super_copy;
5138         struct btrfs_ioctl_feature_flags flags[2];
5139         struct btrfs_trans_handle *trans;
5140         u64 newflags;
5141         int ret;
5142
5143         if (!capable(CAP_SYS_ADMIN))
5144                 return -EPERM;
5145
5146         if (copy_from_user(flags, arg, sizeof(flags)))
5147                 return -EFAULT;
5148
5149         /* Nothing to do */
5150         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5151             !flags[0].incompat_flags)
5152                 return 0;
5153
5154         ret = check_feature(fs_info, flags[0].compat_flags,
5155                             flags[1].compat_flags, COMPAT);
5156         if (ret)
5157                 return ret;
5158
5159         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5160                             flags[1].compat_ro_flags, COMPAT_RO);
5161         if (ret)
5162                 return ret;
5163
5164         ret = check_feature(fs_info, flags[0].incompat_flags,
5165                             flags[1].incompat_flags, INCOMPAT);
5166         if (ret)
5167                 return ret;
5168
5169         ret = mnt_want_write_file(file);
5170         if (ret)
5171                 return ret;
5172
5173         trans = btrfs_start_transaction(root, 0);
5174         if (IS_ERR(trans)) {
5175                 ret = PTR_ERR(trans);
5176                 goto out_drop_write;
5177         }
5178
5179         spin_lock(&fs_info->super_lock);
5180         newflags = btrfs_super_compat_flags(super_block);
5181         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5182         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5183         btrfs_set_super_compat_flags(super_block, newflags);
5184
5185         newflags = btrfs_super_compat_ro_flags(super_block);
5186         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5187         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5188         btrfs_set_super_compat_ro_flags(super_block, newflags);
5189
5190         newflags = btrfs_super_incompat_flags(super_block);
5191         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5192         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5193         btrfs_set_super_incompat_flags(super_block, newflags);
5194         spin_unlock(&fs_info->super_lock);
5195
5196         ret = btrfs_commit_transaction(trans);
5197 out_drop_write:
5198         mnt_drop_write_file(file);
5199
5200         return ret;
5201 }
5202
5203 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
5204 {
5205         struct btrfs_ioctl_send_args *arg;
5206         int ret;
5207
5208         if (compat) {
5209 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5210                 struct btrfs_ioctl_send_args_32 args32;
5211
5212                 ret = copy_from_user(&args32, argp, sizeof(args32));
5213                 if (ret)
5214                         return -EFAULT;
5215                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5216                 if (!arg)
5217                         return -ENOMEM;
5218                 arg->send_fd = args32.send_fd;
5219                 arg->clone_sources_count = args32.clone_sources_count;
5220                 arg->clone_sources = compat_ptr(args32.clone_sources);
5221                 arg->parent_root = args32.parent_root;
5222                 arg->flags = args32.flags;
5223                 memcpy(arg->reserved, args32.reserved,
5224                        sizeof(args32.reserved));
5225 #else
5226                 return -ENOTTY;
5227 #endif
5228         } else {
5229                 arg = memdup_user(argp, sizeof(*arg));
5230                 if (IS_ERR(arg))
5231                         return PTR_ERR(arg);
5232         }
5233         ret = btrfs_ioctl_send(inode, arg);
5234         kfree(arg);
5235         return ret;
5236 }
5237
5238 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
5239                                     bool compat)
5240 {
5241         struct btrfs_ioctl_encoded_io_args args = { 0 };
5242         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
5243                                              flags);
5244         size_t copy_end;
5245         struct iovec iovstack[UIO_FASTIOV];
5246         struct iovec *iov = iovstack;
5247         struct iov_iter iter;
5248         loff_t pos;
5249         struct kiocb kiocb;
5250         ssize_t ret;
5251
5252         if (!capable(CAP_SYS_ADMIN)) {
5253                 ret = -EPERM;
5254                 goto out_acct;
5255         }
5256
5257         if (compat) {
5258 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5259                 struct btrfs_ioctl_encoded_io_args_32 args32;
5260
5261                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
5262                                        flags);
5263                 if (copy_from_user(&args32, argp, copy_end)) {
5264                         ret = -EFAULT;
5265                         goto out_acct;
5266                 }
5267                 args.iov = compat_ptr(args32.iov);
5268                 args.iovcnt = args32.iovcnt;
5269                 args.offset = args32.offset;
5270                 args.flags = args32.flags;
5271 #else
5272                 return -ENOTTY;
5273 #endif
5274         } else {
5275                 copy_end = copy_end_kernel;
5276                 if (copy_from_user(&args, argp, copy_end)) {
5277                         ret = -EFAULT;
5278                         goto out_acct;
5279                 }
5280         }
5281         if (args.flags != 0) {
5282                 ret = -EINVAL;
5283                 goto out_acct;
5284         }
5285
5286         ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5287                            &iov, &iter);
5288         if (ret < 0)
5289                 goto out_acct;
5290
5291         if (iov_iter_count(&iter) == 0) {
5292                 ret = 0;
5293                 goto out_iov;
5294         }
5295         pos = args.offset;
5296         ret = rw_verify_area(READ, file, &pos, args.len);
5297         if (ret < 0)
5298                 goto out_iov;
5299
5300         init_sync_kiocb(&kiocb, file);
5301         kiocb.ki_pos = pos;
5302
5303         ret = btrfs_encoded_read(&kiocb, &iter, &args);
5304         if (ret >= 0) {
5305                 fsnotify_access(file);
5306                 if (copy_to_user(argp + copy_end,
5307                                  (char *)&args + copy_end_kernel,
5308                                  sizeof(args) - copy_end_kernel))
5309                         ret = -EFAULT;
5310         }
5311
5312 out_iov:
5313         kfree(iov);
5314 out_acct:
5315         if (ret > 0)
5316                 add_rchar(current, ret);
5317         inc_syscr(current);
5318         return ret;
5319 }
5320
5321 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
5322 {
5323         struct btrfs_ioctl_encoded_io_args args;
5324         struct iovec iovstack[UIO_FASTIOV];
5325         struct iovec *iov = iovstack;
5326         struct iov_iter iter;
5327         loff_t pos;
5328         struct kiocb kiocb;
5329         ssize_t ret;
5330
5331         if (!capable(CAP_SYS_ADMIN)) {
5332                 ret = -EPERM;
5333                 goto out_acct;
5334         }
5335
5336         if (!(file->f_mode & FMODE_WRITE)) {
5337                 ret = -EBADF;
5338                 goto out_acct;
5339         }
5340
5341         if (compat) {
5342 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5343                 struct btrfs_ioctl_encoded_io_args_32 args32;
5344
5345                 if (copy_from_user(&args32, argp, sizeof(args32))) {
5346                         ret = -EFAULT;
5347                         goto out_acct;
5348                 }
5349                 args.iov = compat_ptr(args32.iov);
5350                 args.iovcnt = args32.iovcnt;
5351                 args.offset = args32.offset;
5352                 args.flags = args32.flags;
5353                 args.len = args32.len;
5354                 args.unencoded_len = args32.unencoded_len;
5355                 args.unencoded_offset = args32.unencoded_offset;
5356                 args.compression = args32.compression;
5357                 args.encryption = args32.encryption;
5358                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
5359 #else
5360                 return -ENOTTY;
5361 #endif
5362         } else {
5363                 if (copy_from_user(&args, argp, sizeof(args))) {
5364                         ret = -EFAULT;
5365                         goto out_acct;
5366                 }
5367         }
5368
5369         ret = -EINVAL;
5370         if (args.flags != 0)
5371                 goto out_acct;
5372         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
5373                 goto out_acct;
5374         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
5375             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
5376                 goto out_acct;
5377         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
5378             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
5379                 goto out_acct;
5380         if (args.unencoded_offset > args.unencoded_len)
5381                 goto out_acct;
5382         if (args.len > args.unencoded_len - args.unencoded_offset)
5383                 goto out_acct;
5384
5385         ret = import_iovec(WRITE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5386                            &iov, &iter);
5387         if (ret < 0)
5388                 goto out_acct;
5389
5390         file_start_write(file);
5391
5392         if (iov_iter_count(&iter) == 0) {
5393                 ret = 0;
5394                 goto out_end_write;
5395         }
5396         pos = args.offset;
5397         ret = rw_verify_area(WRITE, file, &pos, args.len);
5398         if (ret < 0)
5399                 goto out_end_write;
5400
5401         init_sync_kiocb(&kiocb, file);
5402         ret = kiocb_set_rw_flags(&kiocb, 0);
5403         if (ret)
5404                 goto out_end_write;
5405         kiocb.ki_pos = pos;
5406
5407         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
5408         if (ret > 0)
5409                 fsnotify_modify(file);
5410
5411 out_end_write:
5412         file_end_write(file);
5413         kfree(iov);
5414 out_acct:
5415         if (ret > 0)
5416                 add_wchar(current, ret);
5417         inc_syscw(current);
5418         return ret;
5419 }
5420
5421 long btrfs_ioctl(struct file *file, unsigned int
5422                 cmd, unsigned long arg)
5423 {
5424         struct inode *inode = file_inode(file);
5425         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5426         struct btrfs_root *root = BTRFS_I(inode)->root;
5427         void __user *argp = (void __user *)arg;
5428
5429         switch (cmd) {
5430         case FS_IOC_GETVERSION:
5431                 return btrfs_ioctl_getversion(inode, argp);
5432         case FS_IOC_GETFSLABEL:
5433                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5434         case FS_IOC_SETFSLABEL:
5435                 return btrfs_ioctl_set_fslabel(file, argp);
5436         case FITRIM:
5437                 return btrfs_ioctl_fitrim(fs_info, argp);
5438         case BTRFS_IOC_SNAP_CREATE:
5439                 return btrfs_ioctl_snap_create(file, argp, 0);
5440         case BTRFS_IOC_SNAP_CREATE_V2:
5441                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5442         case BTRFS_IOC_SUBVOL_CREATE:
5443                 return btrfs_ioctl_snap_create(file, argp, 1);
5444         case BTRFS_IOC_SUBVOL_CREATE_V2:
5445                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5446         case BTRFS_IOC_SNAP_DESTROY:
5447                 return btrfs_ioctl_snap_destroy(file, argp, false);
5448         case BTRFS_IOC_SNAP_DESTROY_V2:
5449                 return btrfs_ioctl_snap_destroy(file, argp, true);
5450         case BTRFS_IOC_SUBVOL_GETFLAGS:
5451                 return btrfs_ioctl_subvol_getflags(inode, argp);
5452         case BTRFS_IOC_SUBVOL_SETFLAGS:
5453                 return btrfs_ioctl_subvol_setflags(file, argp);
5454         case BTRFS_IOC_DEFAULT_SUBVOL:
5455                 return btrfs_ioctl_default_subvol(file, argp);
5456         case BTRFS_IOC_DEFRAG:
5457                 return btrfs_ioctl_defrag(file, NULL);
5458         case BTRFS_IOC_DEFRAG_RANGE:
5459                 return btrfs_ioctl_defrag(file, argp);
5460         case BTRFS_IOC_RESIZE:
5461                 return btrfs_ioctl_resize(file, argp);
5462         case BTRFS_IOC_ADD_DEV:
5463                 return btrfs_ioctl_add_dev(fs_info, argp);
5464         case BTRFS_IOC_RM_DEV:
5465                 return btrfs_ioctl_rm_dev(file, argp);
5466         case BTRFS_IOC_RM_DEV_V2:
5467                 return btrfs_ioctl_rm_dev_v2(file, argp);
5468         case BTRFS_IOC_FS_INFO:
5469                 return btrfs_ioctl_fs_info(fs_info, argp);
5470         case BTRFS_IOC_DEV_INFO:
5471                 return btrfs_ioctl_dev_info(fs_info, argp);
5472         case BTRFS_IOC_TREE_SEARCH:
5473                 return btrfs_ioctl_tree_search(inode, argp);
5474         case BTRFS_IOC_TREE_SEARCH_V2:
5475                 return btrfs_ioctl_tree_search_v2(inode, argp);
5476         case BTRFS_IOC_INO_LOOKUP:
5477                 return btrfs_ioctl_ino_lookup(root, argp);
5478         case BTRFS_IOC_INO_PATHS:
5479                 return btrfs_ioctl_ino_to_path(root, argp);
5480         case BTRFS_IOC_LOGICAL_INO:
5481                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5482         case BTRFS_IOC_LOGICAL_INO_V2:
5483                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5484         case BTRFS_IOC_SPACE_INFO:
5485                 return btrfs_ioctl_space_info(fs_info, argp);
5486         case BTRFS_IOC_SYNC: {
5487                 int ret;
5488
5489                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
5490                 if (ret)
5491                         return ret;
5492                 ret = btrfs_sync_fs(inode->i_sb, 1);
5493                 /*
5494                  * The transaction thread may want to do more work,
5495                  * namely it pokes the cleaner kthread that will start
5496                  * processing uncleaned subvols.
5497                  */
5498                 wake_up_process(fs_info->transaction_kthread);
5499                 return ret;
5500         }
5501         case BTRFS_IOC_START_SYNC:
5502                 return btrfs_ioctl_start_sync(root, argp);
5503         case BTRFS_IOC_WAIT_SYNC:
5504                 return btrfs_ioctl_wait_sync(fs_info, argp);
5505         case BTRFS_IOC_SCRUB:
5506                 return btrfs_ioctl_scrub(file, argp);
5507         case BTRFS_IOC_SCRUB_CANCEL:
5508                 return btrfs_ioctl_scrub_cancel(fs_info);
5509         case BTRFS_IOC_SCRUB_PROGRESS:
5510                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5511         case BTRFS_IOC_BALANCE_V2:
5512                 return btrfs_ioctl_balance(file, argp);
5513         case BTRFS_IOC_BALANCE_CTL:
5514                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5515         case BTRFS_IOC_BALANCE_PROGRESS:
5516                 return btrfs_ioctl_balance_progress(fs_info, argp);
5517         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5518                 return btrfs_ioctl_set_received_subvol(file, argp);
5519 #ifdef CONFIG_64BIT
5520         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5521                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5522 #endif
5523         case BTRFS_IOC_SEND:
5524                 return _btrfs_ioctl_send(inode, argp, false);
5525 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5526         case BTRFS_IOC_SEND_32:
5527                 return _btrfs_ioctl_send(inode, argp, true);
5528 #endif
5529         case BTRFS_IOC_GET_DEV_STATS:
5530                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5531         case BTRFS_IOC_QUOTA_CTL:
5532                 return btrfs_ioctl_quota_ctl(file, argp);
5533         case BTRFS_IOC_QGROUP_ASSIGN:
5534                 return btrfs_ioctl_qgroup_assign(file, argp);
5535         case BTRFS_IOC_QGROUP_CREATE:
5536                 return btrfs_ioctl_qgroup_create(file, argp);
5537         case BTRFS_IOC_QGROUP_LIMIT:
5538                 return btrfs_ioctl_qgroup_limit(file, argp);
5539         case BTRFS_IOC_QUOTA_RESCAN:
5540                 return btrfs_ioctl_quota_rescan(file, argp);
5541         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5542                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5543         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5544                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5545         case BTRFS_IOC_DEV_REPLACE:
5546                 return btrfs_ioctl_dev_replace(fs_info, argp);
5547         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5548                 return btrfs_ioctl_get_supported_features(argp);
5549         case BTRFS_IOC_GET_FEATURES:
5550                 return btrfs_ioctl_get_features(fs_info, argp);
5551         case BTRFS_IOC_SET_FEATURES:
5552                 return btrfs_ioctl_set_features(file, argp);
5553         case BTRFS_IOC_GET_SUBVOL_INFO:
5554                 return btrfs_ioctl_get_subvol_info(inode, argp);
5555         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5556                 return btrfs_ioctl_get_subvol_rootref(root, argp);
5557         case BTRFS_IOC_INO_LOOKUP_USER:
5558                 return btrfs_ioctl_ino_lookup_user(file, argp);
5559         case FS_IOC_ENABLE_VERITY:
5560                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5561         case FS_IOC_MEASURE_VERITY:
5562                 return fsverity_ioctl_measure(file, argp);
5563         case BTRFS_IOC_ENCODED_READ:
5564                 return btrfs_ioctl_encoded_read(file, argp, false);
5565         case BTRFS_IOC_ENCODED_WRITE:
5566                 return btrfs_ioctl_encoded_write(file, argp, false);
5567 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5568         case BTRFS_IOC_ENCODED_READ_32:
5569                 return btrfs_ioctl_encoded_read(file, argp, true);
5570         case BTRFS_IOC_ENCODED_WRITE_32:
5571                 return btrfs_ioctl_encoded_write(file, argp, true);
5572 #endif
5573         }
5574
5575         return -ENOTTY;
5576 }
5577
5578 #ifdef CONFIG_COMPAT
5579 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5580 {
5581         /*
5582          * These all access 32-bit values anyway so no further
5583          * handling is necessary.
5584          */
5585         switch (cmd) {
5586         case FS_IOC32_GETVERSION:
5587                 cmd = FS_IOC_GETVERSION;
5588                 break;
5589         }
5590
5591         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5592 }
5593 #endif