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(io_tree, start, end, &cached);
1222                 em = defrag_get_extent(BTRFS_I(inode), start, newer_than);
1223                 if (!locked)
1224                         unlock_extent(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(&inode->io_tree, page_start, page_end, &cached_state);
1337                 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
1338                 unlock_extent(&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, 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(&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(&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         btrfs_free_path(path);
3109         path = NULL;
3110         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
3111                 ret = -EFAULT;
3112
3113 out:
3114         btrfs_put_root(root);
3115 out_free:
3116         btrfs_free_path(path);
3117         kfree(subvol_info);
3118         return ret;
3119 }
3120
3121 /*
3122  * Return ROOT_REF information of the subvolume containing this inode
3123  * except the subvolume name.
3124  */
3125 static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root,
3126                                           void __user *argp)
3127 {
3128         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
3129         struct btrfs_root_ref *rref;
3130         struct btrfs_path *path;
3131         struct btrfs_key key;
3132         struct extent_buffer *leaf;
3133         u64 objectid;
3134         int slot;
3135         int ret;
3136         u8 found;
3137
3138         path = btrfs_alloc_path();
3139         if (!path)
3140                 return -ENOMEM;
3141
3142         rootrefs = memdup_user(argp, sizeof(*rootrefs));
3143         if (IS_ERR(rootrefs)) {
3144                 btrfs_free_path(path);
3145                 return PTR_ERR(rootrefs);
3146         }
3147
3148         objectid = root->root_key.objectid;
3149         key.objectid = objectid;
3150         key.type = BTRFS_ROOT_REF_KEY;
3151         key.offset = rootrefs->min_treeid;
3152         found = 0;
3153
3154         root = root->fs_info->tree_root;
3155         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3156         if (ret < 0) {
3157                 goto out;
3158         } else if (path->slots[0] >=
3159                    btrfs_header_nritems(path->nodes[0])) {
3160                 ret = btrfs_next_leaf(root, path);
3161                 if (ret < 0) {
3162                         goto out;
3163                 } else if (ret > 0) {
3164                         ret = -EUCLEAN;
3165                         goto out;
3166                 }
3167         }
3168         while (1) {
3169                 leaf = path->nodes[0];
3170                 slot = path->slots[0];
3171
3172                 btrfs_item_key_to_cpu(leaf, &key, slot);
3173                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
3174                         ret = 0;
3175                         goto out;
3176                 }
3177
3178                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
3179                         ret = -EOVERFLOW;
3180                         goto out;
3181                 }
3182
3183                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
3184                 rootrefs->rootref[found].treeid = key.offset;
3185                 rootrefs->rootref[found].dirid =
3186                                   btrfs_root_ref_dirid(leaf, rref);
3187                 found++;
3188
3189                 ret = btrfs_next_item(root, path);
3190                 if (ret < 0) {
3191                         goto out;
3192                 } else if (ret > 0) {
3193                         ret = -EUCLEAN;
3194                         goto out;
3195                 }
3196         }
3197
3198 out:
3199         btrfs_free_path(path);
3200
3201         if (!ret || ret == -EOVERFLOW) {
3202                 rootrefs->num_items = found;
3203                 /* update min_treeid for next search */
3204                 if (found)
3205                         rootrefs->min_treeid =
3206                                 rootrefs->rootref[found - 1].treeid + 1;
3207                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
3208                         ret = -EFAULT;
3209         }
3210
3211         kfree(rootrefs);
3212
3213         return ret;
3214 }
3215
3216 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
3217                                              void __user *arg,
3218                                              bool destroy_v2)
3219 {
3220         struct dentry *parent = file->f_path.dentry;
3221         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
3222         struct dentry *dentry;
3223         struct inode *dir = d_inode(parent);
3224         struct inode *inode;
3225         struct btrfs_root *root = BTRFS_I(dir)->root;
3226         struct btrfs_root *dest = NULL;
3227         struct btrfs_ioctl_vol_args *vol_args = NULL;
3228         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
3229         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
3230         char *subvol_name, *subvol_name_ptr = NULL;
3231         int subvol_namelen;
3232         int err = 0;
3233         bool destroy_parent = false;
3234
3235         /* We don't support snapshots with extent tree v2 yet. */
3236         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3237                 btrfs_err(fs_info,
3238                           "extent tree v2 doesn't support snapshot deletion yet");
3239                 return -EOPNOTSUPP;
3240         }
3241
3242         if (destroy_v2) {
3243                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
3244                 if (IS_ERR(vol_args2))
3245                         return PTR_ERR(vol_args2);
3246
3247                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
3248                         err = -EOPNOTSUPP;
3249                         goto out;
3250                 }
3251
3252                 /*
3253                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
3254                  * name, same as v1 currently does.
3255                  */
3256                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
3257                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
3258                         subvol_name = vol_args2->name;
3259
3260                         err = mnt_want_write_file(file);
3261                         if (err)
3262                                 goto out;
3263                 } else {
3264                         struct inode *old_dir;
3265
3266                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
3267                                 err = -EINVAL;
3268                                 goto out;
3269                         }
3270
3271                         err = mnt_want_write_file(file);
3272                         if (err)
3273                                 goto out;
3274
3275                         dentry = btrfs_get_dentry(fs_info->sb,
3276                                         BTRFS_FIRST_FREE_OBJECTID,
3277                                         vol_args2->subvolid, 0, 0);
3278                         if (IS_ERR(dentry)) {
3279                                 err = PTR_ERR(dentry);
3280                                 goto out_drop_write;
3281                         }
3282
3283                         /*
3284                          * Change the default parent since the subvolume being
3285                          * deleted can be outside of the current mount point.
3286                          */
3287                         parent = btrfs_get_parent(dentry);
3288
3289                         /*
3290                          * At this point dentry->d_name can point to '/' if the
3291                          * subvolume we want to destroy is outsite of the
3292                          * current mount point, so we need to release the
3293                          * current dentry and execute the lookup to return a new
3294                          * one with ->d_name pointing to the
3295                          * <mount point>/subvol_name.
3296                          */
3297                         dput(dentry);
3298                         if (IS_ERR(parent)) {
3299                                 err = PTR_ERR(parent);
3300                                 goto out_drop_write;
3301                         }
3302                         old_dir = dir;
3303                         dir = d_inode(parent);
3304
3305                         /*
3306                          * If v2 was used with SPEC_BY_ID, a new parent was
3307                          * allocated since the subvolume can be outside of the
3308                          * current mount point. Later on we need to release this
3309                          * new parent dentry.
3310                          */
3311                         destroy_parent = true;
3312
3313                         /*
3314                          * On idmapped mounts, deletion via subvolid is
3315                          * restricted to subvolumes that are immediate
3316                          * ancestors of the inode referenced by the file
3317                          * descriptor in the ioctl. Otherwise the idmapping
3318                          * could potentially be abused to delete subvolumes
3319                          * anywhere in the filesystem the user wouldn't be able
3320                          * to delete without an idmapped mount.
3321                          */
3322                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3323                                 err = -EOPNOTSUPP;
3324                                 goto free_parent;
3325                         }
3326
3327                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3328                                                 fs_info, vol_args2->subvolid);
3329                         if (IS_ERR(subvol_name_ptr)) {
3330                                 err = PTR_ERR(subvol_name_ptr);
3331                                 goto free_parent;
3332                         }
3333                         /* subvol_name_ptr is already nul terminated */
3334                         subvol_name = (char *)kbasename(subvol_name_ptr);
3335                 }
3336         } else {
3337                 vol_args = memdup_user(arg, sizeof(*vol_args));
3338                 if (IS_ERR(vol_args))
3339                         return PTR_ERR(vol_args);
3340
3341                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3342                 subvol_name = vol_args->name;
3343
3344                 err = mnt_want_write_file(file);
3345                 if (err)
3346                         goto out;
3347         }
3348
3349         subvol_namelen = strlen(subvol_name);
3350
3351         if (strchr(subvol_name, '/') ||
3352             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3353                 err = -EINVAL;
3354                 goto free_subvol_name;
3355         }
3356
3357         if (!S_ISDIR(dir->i_mode)) {
3358                 err = -ENOTDIR;
3359                 goto free_subvol_name;
3360         }
3361
3362         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3363         if (err == -EINTR)
3364                 goto free_subvol_name;
3365         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3366         if (IS_ERR(dentry)) {
3367                 err = PTR_ERR(dentry);
3368                 goto out_unlock_dir;
3369         }
3370
3371         if (d_really_is_negative(dentry)) {
3372                 err = -ENOENT;
3373                 goto out_dput;
3374         }
3375
3376         inode = d_inode(dentry);
3377         dest = BTRFS_I(inode)->root;
3378         if (!capable(CAP_SYS_ADMIN)) {
3379                 /*
3380                  * Regular user.  Only allow this with a special mount
3381                  * option, when the user has write+exec access to the
3382                  * subvol root, and when rmdir(2) would have been
3383                  * allowed.
3384                  *
3385                  * Note that this is _not_ check that the subvol is
3386                  * empty or doesn't contain data that we wouldn't
3387                  * otherwise be able to delete.
3388                  *
3389                  * Users who want to delete empty subvols should try
3390                  * rmdir(2).
3391                  */
3392                 err = -EPERM;
3393                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3394                         goto out_dput;
3395
3396                 /*
3397                  * Do not allow deletion if the parent dir is the same
3398                  * as the dir to be deleted.  That means the ioctl
3399                  * must be called on the dentry referencing the root
3400                  * of the subvol, not a random directory contained
3401                  * within it.
3402                  */
3403                 err = -EINVAL;
3404                 if (root == dest)
3405                         goto out_dput;
3406
3407                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3408                 if (err)
3409                         goto out_dput;
3410         }
3411
3412         /* check if subvolume may be deleted by a user */
3413         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3414         if (err)
3415                 goto out_dput;
3416
3417         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3418                 err = -EINVAL;
3419                 goto out_dput;
3420         }
3421
3422         btrfs_inode_lock(inode, 0);
3423         err = btrfs_delete_subvolume(dir, dentry);
3424         btrfs_inode_unlock(inode, 0);
3425         if (!err)
3426                 d_delete_notify(dir, dentry);
3427
3428 out_dput:
3429         dput(dentry);
3430 out_unlock_dir:
3431         btrfs_inode_unlock(dir, 0);
3432 free_subvol_name:
3433         kfree(subvol_name_ptr);
3434 free_parent:
3435         if (destroy_parent)
3436                 dput(parent);
3437 out_drop_write:
3438         mnt_drop_write_file(file);
3439 out:
3440         kfree(vol_args2);
3441         kfree(vol_args);
3442         return err;
3443 }
3444
3445 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3446 {
3447         struct inode *inode = file_inode(file);
3448         struct btrfs_root *root = BTRFS_I(inode)->root;
3449         struct btrfs_ioctl_defrag_range_args range = {0};
3450         int ret;
3451
3452         ret = mnt_want_write_file(file);
3453         if (ret)
3454                 return ret;
3455
3456         if (btrfs_root_readonly(root)) {
3457                 ret = -EROFS;
3458                 goto out;
3459         }
3460
3461         switch (inode->i_mode & S_IFMT) {
3462         case S_IFDIR:
3463                 if (!capable(CAP_SYS_ADMIN)) {
3464                         ret = -EPERM;
3465                         goto out;
3466                 }
3467                 ret = btrfs_defrag_root(root);
3468                 break;
3469         case S_IFREG:
3470                 /*
3471                  * Note that this does not check the file descriptor for write
3472                  * access. This prevents defragmenting executables that are
3473                  * running and allows defrag on files open in read-only mode.
3474                  */
3475                 if (!capable(CAP_SYS_ADMIN) &&
3476                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3477                         ret = -EPERM;
3478                         goto out;
3479                 }
3480
3481                 if (argp) {
3482                         if (copy_from_user(&range, argp, sizeof(range))) {
3483                                 ret = -EFAULT;
3484                                 goto out;
3485                         }
3486                         /* compression requires us to start the IO */
3487                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3488                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3489                                 range.extent_thresh = (u32)-1;
3490                         }
3491                 } else {
3492                         /* the rest are all set to zero by kzalloc */
3493                         range.len = (u64)-1;
3494                 }
3495                 ret = btrfs_defrag_file(file_inode(file), &file->f_ra,
3496                                         &range, BTRFS_OLDEST_GENERATION, 0);
3497                 if (ret > 0)
3498                         ret = 0;
3499                 break;
3500         default:
3501                 ret = -EINVAL;
3502         }
3503 out:
3504         mnt_drop_write_file(file);
3505         return ret;
3506 }
3507
3508 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3509 {
3510         struct btrfs_ioctl_vol_args *vol_args;
3511         bool restore_op = false;
3512         int ret;
3513
3514         if (!capable(CAP_SYS_ADMIN))
3515                 return -EPERM;
3516
3517         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3518                 btrfs_err(fs_info, "device add not supported on extent tree v2 yet");
3519                 return -EINVAL;
3520         }
3521
3522         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) {
3523                 if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD))
3524                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3525
3526                 /*
3527                  * We can do the device add because we have a paused balanced,
3528                  * change the exclusive op type and remember we should bring
3529                  * back the paused balance
3530                  */
3531                 fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD;
3532                 btrfs_exclop_start_unlock(fs_info);
3533                 restore_op = true;
3534         }
3535
3536         vol_args = memdup_user(arg, sizeof(*vol_args));
3537         if (IS_ERR(vol_args)) {
3538                 ret = PTR_ERR(vol_args);
3539                 goto out;
3540         }
3541
3542         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3543         ret = btrfs_init_new_device(fs_info, vol_args->name);
3544
3545         if (!ret)
3546                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3547
3548         kfree(vol_args);
3549 out:
3550         if (restore_op)
3551                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
3552         else
3553                 btrfs_exclop_finish(fs_info);
3554         return ret;
3555 }
3556
3557 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3558 {
3559         BTRFS_DEV_LOOKUP_ARGS(args);
3560         struct inode *inode = file_inode(file);
3561         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3562         struct btrfs_ioctl_vol_args_v2 *vol_args;
3563         struct block_device *bdev = NULL;
3564         fmode_t mode;
3565         int ret;
3566         bool cancel = false;
3567
3568         if (!capable(CAP_SYS_ADMIN))
3569                 return -EPERM;
3570
3571         vol_args = memdup_user(arg, sizeof(*vol_args));
3572         if (IS_ERR(vol_args))
3573                 return PTR_ERR(vol_args);
3574
3575         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3576                 ret = -EOPNOTSUPP;
3577                 goto out;
3578         }
3579
3580         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3581         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3582                 args.devid = vol_args->devid;
3583         } else if (!strcmp("cancel", vol_args->name)) {
3584                 cancel = true;
3585         } else {
3586                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3587                 if (ret)
3588                         goto out;
3589         }
3590
3591         ret = mnt_want_write_file(file);
3592         if (ret)
3593                 goto out;
3594
3595         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3596                                            cancel);
3597         if (ret)
3598                 goto err_drop;
3599
3600         /* Exclusive operation is now claimed */
3601         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3602
3603         btrfs_exclop_finish(fs_info);
3604
3605         if (!ret) {
3606                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3607                         btrfs_info(fs_info, "device deleted: id %llu",
3608                                         vol_args->devid);
3609                 else
3610                         btrfs_info(fs_info, "device deleted: %s",
3611                                         vol_args->name);
3612         }
3613 err_drop:
3614         mnt_drop_write_file(file);
3615         if (bdev)
3616                 blkdev_put(bdev, mode);
3617 out:
3618         btrfs_put_dev_args_from_path(&args);
3619         kfree(vol_args);
3620         return ret;
3621 }
3622
3623 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3624 {
3625         BTRFS_DEV_LOOKUP_ARGS(args);
3626         struct inode *inode = file_inode(file);
3627         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3628         struct btrfs_ioctl_vol_args *vol_args;
3629         struct block_device *bdev = NULL;
3630         fmode_t mode;
3631         int ret;
3632         bool cancel = false;
3633
3634         if (!capable(CAP_SYS_ADMIN))
3635                 return -EPERM;
3636
3637         vol_args = memdup_user(arg, sizeof(*vol_args));
3638         if (IS_ERR(vol_args))
3639                 return PTR_ERR(vol_args);
3640
3641         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3642         if (!strcmp("cancel", vol_args->name)) {
3643                 cancel = true;
3644         } else {
3645                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3646                 if (ret)
3647                         goto out;
3648         }
3649
3650         ret = mnt_want_write_file(file);
3651         if (ret)
3652                 goto out;
3653
3654         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3655                                            cancel);
3656         if (ret == 0) {
3657                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3658                 if (!ret)
3659                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3660                 btrfs_exclop_finish(fs_info);
3661         }
3662
3663         mnt_drop_write_file(file);
3664         if (bdev)
3665                 blkdev_put(bdev, mode);
3666 out:
3667         btrfs_put_dev_args_from_path(&args);
3668         kfree(vol_args);
3669         return ret;
3670 }
3671
3672 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3673                                 void __user *arg)
3674 {
3675         struct btrfs_ioctl_fs_info_args *fi_args;
3676         struct btrfs_device *device;
3677         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3678         u64 flags_in;
3679         int ret = 0;
3680
3681         fi_args = memdup_user(arg, sizeof(*fi_args));
3682         if (IS_ERR(fi_args))
3683                 return PTR_ERR(fi_args);
3684
3685         flags_in = fi_args->flags;
3686         memset(fi_args, 0, sizeof(*fi_args));
3687
3688         rcu_read_lock();
3689         fi_args->num_devices = fs_devices->num_devices;
3690
3691         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3692                 if (device->devid > fi_args->max_id)
3693                         fi_args->max_id = device->devid;
3694         }
3695         rcu_read_unlock();
3696
3697         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3698         fi_args->nodesize = fs_info->nodesize;
3699         fi_args->sectorsize = fs_info->sectorsize;
3700         fi_args->clone_alignment = fs_info->sectorsize;
3701
3702         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3703                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3704                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3705                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3706         }
3707
3708         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3709                 fi_args->generation = fs_info->generation;
3710                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3711         }
3712
3713         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3714                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3715                        sizeof(fi_args->metadata_uuid));
3716                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3717         }
3718
3719         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3720                 ret = -EFAULT;
3721
3722         kfree(fi_args);
3723         return ret;
3724 }
3725
3726 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3727                                  void __user *arg)
3728 {
3729         BTRFS_DEV_LOOKUP_ARGS(args);
3730         struct btrfs_ioctl_dev_info_args *di_args;
3731         struct btrfs_device *dev;
3732         int ret = 0;
3733
3734         di_args = memdup_user(arg, sizeof(*di_args));
3735         if (IS_ERR(di_args))
3736                 return PTR_ERR(di_args);
3737
3738         args.devid = di_args->devid;
3739         if (!btrfs_is_empty_uuid(di_args->uuid))
3740                 args.uuid = di_args->uuid;
3741
3742         rcu_read_lock();
3743         dev = btrfs_find_device(fs_info->fs_devices, &args);
3744         if (!dev) {
3745                 ret = -ENODEV;
3746                 goto out;
3747         }
3748
3749         di_args->devid = dev->devid;
3750         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3751         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3752         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3753         if (dev->name) {
3754                 strncpy(di_args->path, rcu_str_deref(dev->name),
3755                                 sizeof(di_args->path) - 1);
3756                 di_args->path[sizeof(di_args->path) - 1] = 0;
3757         } else {
3758                 di_args->path[0] = '\0';
3759         }
3760
3761 out:
3762         rcu_read_unlock();
3763         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3764                 ret = -EFAULT;
3765
3766         kfree(di_args);
3767         return ret;
3768 }
3769
3770 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3771 {
3772         struct inode *inode = file_inode(file);
3773         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3774         struct btrfs_root *root = BTRFS_I(inode)->root;
3775         struct btrfs_root *new_root;
3776         struct btrfs_dir_item *di;
3777         struct btrfs_trans_handle *trans;
3778         struct btrfs_path *path = NULL;
3779         struct btrfs_disk_key disk_key;
3780         u64 objectid = 0;
3781         u64 dir_id;
3782         int ret;
3783
3784         if (!capable(CAP_SYS_ADMIN))
3785                 return -EPERM;
3786
3787         ret = mnt_want_write_file(file);
3788         if (ret)
3789                 return ret;
3790
3791         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3792                 ret = -EFAULT;
3793                 goto out;
3794         }
3795
3796         if (!objectid)
3797                 objectid = BTRFS_FS_TREE_OBJECTID;
3798
3799         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3800         if (IS_ERR(new_root)) {
3801                 ret = PTR_ERR(new_root);
3802                 goto out;
3803         }
3804         if (!is_fstree(new_root->root_key.objectid)) {
3805                 ret = -ENOENT;
3806                 goto out_free;
3807         }
3808
3809         path = btrfs_alloc_path();
3810         if (!path) {
3811                 ret = -ENOMEM;
3812                 goto out_free;
3813         }
3814
3815         trans = btrfs_start_transaction(root, 1);
3816         if (IS_ERR(trans)) {
3817                 ret = PTR_ERR(trans);
3818                 goto out_free;
3819         }
3820
3821         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3822         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3823                                    dir_id, "default", 7, 1);
3824         if (IS_ERR_OR_NULL(di)) {
3825                 btrfs_release_path(path);
3826                 btrfs_end_transaction(trans);
3827                 btrfs_err(fs_info,
3828                           "Umm, you don't have the default diritem, this isn't going to work");
3829                 ret = -ENOENT;
3830                 goto out_free;
3831         }
3832
3833         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3834         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3835         btrfs_mark_buffer_dirty(path->nodes[0]);
3836         btrfs_release_path(path);
3837
3838         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3839         btrfs_end_transaction(trans);
3840 out_free:
3841         btrfs_put_root(new_root);
3842         btrfs_free_path(path);
3843 out:
3844         mnt_drop_write_file(file);
3845         return ret;
3846 }
3847
3848 static void get_block_group_info(struct list_head *groups_list,
3849                                  struct btrfs_ioctl_space_info *space)
3850 {
3851         struct btrfs_block_group *block_group;
3852
3853         space->total_bytes = 0;
3854         space->used_bytes = 0;
3855         space->flags = 0;
3856         list_for_each_entry(block_group, groups_list, list) {
3857                 space->flags = block_group->flags;
3858                 space->total_bytes += block_group->length;
3859                 space->used_bytes += block_group->used;
3860         }
3861 }
3862
3863 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3864                                    void __user *arg)
3865 {
3866         struct btrfs_ioctl_space_args space_args;
3867         struct btrfs_ioctl_space_info space;
3868         struct btrfs_ioctl_space_info *dest;
3869         struct btrfs_ioctl_space_info *dest_orig;
3870         struct btrfs_ioctl_space_info __user *user_dest;
3871         struct btrfs_space_info *info;
3872         static const u64 types[] = {
3873                 BTRFS_BLOCK_GROUP_DATA,
3874                 BTRFS_BLOCK_GROUP_SYSTEM,
3875                 BTRFS_BLOCK_GROUP_METADATA,
3876                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3877         };
3878         int num_types = 4;
3879         int alloc_size;
3880         int ret = 0;
3881         u64 slot_count = 0;
3882         int i, c;
3883
3884         if (copy_from_user(&space_args,
3885                            (struct btrfs_ioctl_space_args __user *)arg,
3886                            sizeof(space_args)))
3887                 return -EFAULT;
3888
3889         for (i = 0; i < num_types; i++) {
3890                 struct btrfs_space_info *tmp;
3891
3892                 info = NULL;
3893                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3894                         if (tmp->flags == types[i]) {
3895                                 info = tmp;
3896                                 break;
3897                         }
3898                 }
3899
3900                 if (!info)
3901                         continue;
3902
3903                 down_read(&info->groups_sem);
3904                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3905                         if (!list_empty(&info->block_groups[c]))
3906                                 slot_count++;
3907                 }
3908                 up_read(&info->groups_sem);
3909         }
3910
3911         /*
3912          * Global block reserve, exported as a space_info
3913          */
3914         slot_count++;
3915
3916         /* space_slots == 0 means they are asking for a count */
3917         if (space_args.space_slots == 0) {
3918                 space_args.total_spaces = slot_count;
3919                 goto out;
3920         }
3921
3922         slot_count = min_t(u64, space_args.space_slots, slot_count);
3923
3924         alloc_size = sizeof(*dest) * slot_count;
3925
3926         /* we generally have at most 6 or so space infos, one for each raid
3927          * level.  So, a whole page should be more than enough for everyone
3928          */
3929         if (alloc_size > PAGE_SIZE)
3930                 return -ENOMEM;
3931
3932         space_args.total_spaces = 0;
3933         dest = kmalloc(alloc_size, GFP_KERNEL);
3934         if (!dest)
3935                 return -ENOMEM;
3936         dest_orig = dest;
3937
3938         /* now we have a buffer to copy into */
3939         for (i = 0; i < num_types; i++) {
3940                 struct btrfs_space_info *tmp;
3941
3942                 if (!slot_count)
3943                         break;
3944
3945                 info = NULL;
3946                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3947                         if (tmp->flags == types[i]) {
3948                                 info = tmp;
3949                                 break;
3950                         }
3951                 }
3952
3953                 if (!info)
3954                         continue;
3955                 down_read(&info->groups_sem);
3956                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3957                         if (!list_empty(&info->block_groups[c])) {
3958                                 get_block_group_info(&info->block_groups[c],
3959                                                      &space);
3960                                 memcpy(dest, &space, sizeof(space));
3961                                 dest++;
3962                                 space_args.total_spaces++;
3963                                 slot_count--;
3964                         }
3965                         if (!slot_count)
3966                                 break;
3967                 }
3968                 up_read(&info->groups_sem);
3969         }
3970
3971         /*
3972          * Add global block reserve
3973          */
3974         if (slot_count) {
3975                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3976
3977                 spin_lock(&block_rsv->lock);
3978                 space.total_bytes = block_rsv->size;
3979                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3980                 spin_unlock(&block_rsv->lock);
3981                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3982                 memcpy(dest, &space, sizeof(space));
3983                 space_args.total_spaces++;
3984         }
3985
3986         user_dest = (struct btrfs_ioctl_space_info __user *)
3987                 (arg + sizeof(struct btrfs_ioctl_space_args));
3988
3989         if (copy_to_user(user_dest, dest_orig, alloc_size))
3990                 ret = -EFAULT;
3991
3992         kfree(dest_orig);
3993 out:
3994         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3995                 ret = -EFAULT;
3996
3997         return ret;
3998 }
3999
4000 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4001                                             void __user *argp)
4002 {
4003         struct btrfs_trans_handle *trans;
4004         u64 transid;
4005
4006         trans = btrfs_attach_transaction_barrier(root);
4007         if (IS_ERR(trans)) {
4008                 if (PTR_ERR(trans) != -ENOENT)
4009                         return PTR_ERR(trans);
4010
4011                 /* No running transaction, don't bother */
4012                 transid = root->fs_info->last_trans_committed;
4013                 goto out;
4014         }
4015         transid = trans->transid;
4016         btrfs_commit_transaction_async(trans);
4017 out:
4018         if (argp)
4019                 if (copy_to_user(argp, &transid, sizeof(transid)))
4020                         return -EFAULT;
4021         return 0;
4022 }
4023
4024 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4025                                            void __user *argp)
4026 {
4027         u64 transid;
4028
4029         if (argp) {
4030                 if (copy_from_user(&transid, argp, sizeof(transid)))
4031                         return -EFAULT;
4032         } else {
4033                 transid = 0;  /* current trans */
4034         }
4035         return btrfs_wait_for_commit(fs_info, transid);
4036 }
4037
4038 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4039 {
4040         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4041         struct btrfs_ioctl_scrub_args *sa;
4042         int ret;
4043
4044         if (!capable(CAP_SYS_ADMIN))
4045                 return -EPERM;
4046
4047         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4048                 btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet");
4049                 return -EINVAL;
4050         }
4051
4052         sa = memdup_user(arg, sizeof(*sa));
4053         if (IS_ERR(sa))
4054                 return PTR_ERR(sa);
4055
4056         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4057                 ret = mnt_want_write_file(file);
4058                 if (ret)
4059                         goto out;
4060         }
4061
4062         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4063                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4064                               0);
4065
4066         /*
4067          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4068          * error. This is important as it allows user space to know how much
4069          * progress scrub has done. For example, if scrub is canceled we get
4070          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4071          * space. Later user space can inspect the progress from the structure
4072          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4073          * previously (btrfs-progs does this).
4074          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4075          * then return -EFAULT to signal the structure was not copied or it may
4076          * be corrupt and unreliable due to a partial copy.
4077          */
4078         if (copy_to_user(arg, sa, sizeof(*sa)))
4079                 ret = -EFAULT;
4080
4081         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4082                 mnt_drop_write_file(file);
4083 out:
4084         kfree(sa);
4085         return ret;
4086 }
4087
4088 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4089 {
4090         if (!capable(CAP_SYS_ADMIN))
4091                 return -EPERM;
4092
4093         return btrfs_scrub_cancel(fs_info);
4094 }
4095
4096 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4097                                        void __user *arg)
4098 {
4099         struct btrfs_ioctl_scrub_args *sa;
4100         int ret;
4101
4102         if (!capable(CAP_SYS_ADMIN))
4103                 return -EPERM;
4104
4105         sa = memdup_user(arg, sizeof(*sa));
4106         if (IS_ERR(sa))
4107                 return PTR_ERR(sa);
4108
4109         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4110
4111         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4112                 ret = -EFAULT;
4113
4114         kfree(sa);
4115         return ret;
4116 }
4117
4118 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4119                                       void __user *arg)
4120 {
4121         struct btrfs_ioctl_get_dev_stats *sa;
4122         int ret;
4123
4124         sa = memdup_user(arg, sizeof(*sa));
4125         if (IS_ERR(sa))
4126                 return PTR_ERR(sa);
4127
4128         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4129                 kfree(sa);
4130                 return -EPERM;
4131         }
4132
4133         ret = btrfs_get_dev_stats(fs_info, sa);
4134
4135         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4136                 ret = -EFAULT;
4137
4138         kfree(sa);
4139         return ret;
4140 }
4141
4142 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4143                                     void __user *arg)
4144 {
4145         struct btrfs_ioctl_dev_replace_args *p;
4146         int ret;
4147
4148         if (!capable(CAP_SYS_ADMIN))
4149                 return -EPERM;
4150
4151         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
4152                 btrfs_err(fs_info, "device replace not supported on extent tree v2 yet");
4153                 return -EINVAL;
4154         }
4155
4156         p = memdup_user(arg, sizeof(*p));
4157         if (IS_ERR(p))
4158                 return PTR_ERR(p);
4159
4160         switch (p->cmd) {
4161         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4162                 if (sb_rdonly(fs_info->sb)) {
4163                         ret = -EROFS;
4164                         goto out;
4165                 }
4166                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
4167                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4168                 } else {
4169                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4170                         btrfs_exclop_finish(fs_info);
4171                 }
4172                 break;
4173         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4174                 btrfs_dev_replace_status(fs_info, p);
4175                 ret = 0;
4176                 break;
4177         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4178                 p->result = btrfs_dev_replace_cancel(fs_info);
4179                 ret = 0;
4180                 break;
4181         default:
4182                 ret = -EINVAL;
4183                 break;
4184         }
4185
4186         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4187                 ret = -EFAULT;
4188 out:
4189         kfree(p);
4190         return ret;
4191 }
4192
4193 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4194 {
4195         int ret = 0;
4196         int i;
4197         u64 rel_ptr;
4198         int size;
4199         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4200         struct inode_fs_paths *ipath = NULL;
4201         struct btrfs_path *path;
4202
4203         if (!capable(CAP_DAC_READ_SEARCH))
4204                 return -EPERM;
4205
4206         path = btrfs_alloc_path();
4207         if (!path) {
4208                 ret = -ENOMEM;
4209                 goto out;
4210         }
4211
4212         ipa = memdup_user(arg, sizeof(*ipa));
4213         if (IS_ERR(ipa)) {
4214                 ret = PTR_ERR(ipa);
4215                 ipa = NULL;
4216                 goto out;
4217         }
4218
4219         size = min_t(u32, ipa->size, 4096);
4220         ipath = init_ipath(size, root, path);
4221         if (IS_ERR(ipath)) {
4222                 ret = PTR_ERR(ipath);
4223                 ipath = NULL;
4224                 goto out;
4225         }
4226
4227         ret = paths_from_inode(ipa->inum, ipath);
4228         if (ret < 0)
4229                 goto out;
4230
4231         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4232                 rel_ptr = ipath->fspath->val[i] -
4233                           (u64)(unsigned long)ipath->fspath->val;
4234                 ipath->fspath->val[i] = rel_ptr;
4235         }
4236
4237         btrfs_free_path(path);
4238         path = NULL;
4239         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4240                            ipath->fspath, size);
4241         if (ret) {
4242                 ret = -EFAULT;
4243                 goto out;
4244         }
4245
4246 out:
4247         btrfs_free_path(path);
4248         free_ipath(ipath);
4249         kfree(ipa);
4250
4251         return ret;
4252 }
4253
4254 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4255                                         void __user *arg, int version)
4256 {
4257         int ret = 0;
4258         int size;
4259         struct btrfs_ioctl_logical_ino_args *loi;
4260         struct btrfs_data_container *inodes = NULL;
4261         struct btrfs_path *path = NULL;
4262         bool ignore_offset;
4263
4264         if (!capable(CAP_SYS_ADMIN))
4265                 return -EPERM;
4266
4267         loi = memdup_user(arg, sizeof(*loi));
4268         if (IS_ERR(loi))
4269                 return PTR_ERR(loi);
4270
4271         if (version == 1) {
4272                 ignore_offset = false;
4273                 size = min_t(u32, loi->size, SZ_64K);
4274         } else {
4275                 /* All reserved bits must be 0 for now */
4276                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4277                         ret = -EINVAL;
4278                         goto out_loi;
4279                 }
4280                 /* Only accept flags we have defined so far */
4281                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4282                         ret = -EINVAL;
4283                         goto out_loi;
4284                 }
4285                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4286                 size = min_t(u32, loi->size, SZ_16M);
4287         }
4288
4289         inodes = init_data_container(size);
4290         if (IS_ERR(inodes)) {
4291                 ret = PTR_ERR(inodes);
4292                 goto out_loi;
4293         }
4294
4295         path = btrfs_alloc_path();
4296         if (!path) {
4297                 ret = -ENOMEM;
4298                 goto out;
4299         }
4300         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4301                                           inodes, ignore_offset);
4302         btrfs_free_path(path);
4303         if (ret == -EINVAL)
4304                 ret = -ENOENT;
4305         if (ret < 0)
4306                 goto out;
4307
4308         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4309                            size);
4310         if (ret)
4311                 ret = -EFAULT;
4312
4313 out:
4314         kvfree(inodes);
4315 out_loi:
4316         kfree(loi);
4317
4318         return ret;
4319 }
4320
4321 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4322                                struct btrfs_ioctl_balance_args *bargs)
4323 {
4324         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4325
4326         bargs->flags = bctl->flags;
4327
4328         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4329                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4330         if (atomic_read(&fs_info->balance_pause_req))
4331                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4332         if (atomic_read(&fs_info->balance_cancel_req))
4333                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4334
4335         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4336         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4337         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4338
4339         spin_lock(&fs_info->balance_lock);
4340         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4341         spin_unlock(&fs_info->balance_lock);
4342 }
4343
4344 /**
4345  * Try to acquire fs_info::balance_mutex as well as set BTRFS_EXLCOP_BALANCE as
4346  * required.
4347  *
4348  * @fs_info:       the filesystem
4349  * @excl_acquired: ptr to boolean value which is set to false in case balance
4350  *                 is being resumed
4351  *
4352  * Return 0 on success in which case both fs_info::balance is acquired as well
4353  * as exclusive ops are blocked. In case of failure return an error code.
4354  */
4355 static int btrfs_try_lock_balance(struct btrfs_fs_info *fs_info, bool *excl_acquired)
4356 {
4357         int ret;
4358
4359         /*
4360          * Exclusive operation is locked. Three possibilities:
4361          *   (1) some other op is running
4362          *   (2) balance is running
4363          *   (3) balance is paused -- special case (think resume)
4364          */
4365         while (1) {
4366                 if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4367                         *excl_acquired = true;
4368                         mutex_lock(&fs_info->balance_mutex);
4369                         return 0;
4370                 }
4371
4372                 mutex_lock(&fs_info->balance_mutex);
4373                 if (fs_info->balance_ctl) {
4374                         /* This is either (2) or (3) */
4375                         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4376                                 /* This is (2) */
4377                                 ret = -EINPROGRESS;
4378                                 goto out_failure;
4379
4380                         } else {
4381                                 mutex_unlock(&fs_info->balance_mutex);
4382                                 /*
4383                                  * Lock released to allow other waiters to
4384                                  * continue, we'll reexamine the status again.
4385                                  */
4386                                 mutex_lock(&fs_info->balance_mutex);
4387
4388                                 if (fs_info->balance_ctl &&
4389                                     !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4390                                         /* This is (3) */
4391                                         *excl_acquired = false;
4392                                         return 0;
4393                                 }
4394                         }
4395                 } else {
4396                         /* This is (1) */
4397                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4398                         goto out_failure;
4399                 }
4400
4401                 mutex_unlock(&fs_info->balance_mutex);
4402         }
4403
4404 out_failure:
4405         mutex_unlock(&fs_info->balance_mutex);
4406         *excl_acquired = false;
4407         return ret;
4408 }
4409
4410 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4411 {
4412         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4413         struct btrfs_fs_info *fs_info = root->fs_info;
4414         struct btrfs_ioctl_balance_args *bargs;
4415         struct btrfs_balance_control *bctl;
4416         bool need_unlock = true;
4417         int ret;
4418
4419         if (!capable(CAP_SYS_ADMIN))
4420                 return -EPERM;
4421
4422         ret = mnt_want_write_file(file);
4423         if (ret)
4424                 return ret;
4425
4426         bargs = memdup_user(arg, sizeof(*bargs));
4427         if (IS_ERR(bargs)) {
4428                 ret = PTR_ERR(bargs);
4429                 bargs = NULL;
4430                 goto out;
4431         }
4432
4433         ret = btrfs_try_lock_balance(fs_info, &need_unlock);
4434         if (ret)
4435                 goto out;
4436
4437         lockdep_assert_held(&fs_info->balance_mutex);
4438
4439         if (bargs->flags & BTRFS_BALANCE_RESUME) {
4440                 if (!fs_info->balance_ctl) {
4441                         ret = -ENOTCONN;
4442                         goto out_unlock;
4443                 }
4444
4445                 bctl = fs_info->balance_ctl;
4446                 spin_lock(&fs_info->balance_lock);
4447                 bctl->flags |= BTRFS_BALANCE_RESUME;
4448                 spin_unlock(&fs_info->balance_lock);
4449                 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE);
4450
4451                 goto do_balance;
4452         }
4453
4454         if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4455                 ret = -EINVAL;
4456                 goto out_unlock;
4457         }
4458
4459         if (fs_info->balance_ctl) {
4460                 ret = -EINPROGRESS;
4461                 goto out_unlock;
4462         }
4463
4464         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4465         if (!bctl) {
4466                 ret = -ENOMEM;
4467                 goto out_unlock;
4468         }
4469
4470         memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4471         memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4472         memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4473
4474         bctl->flags = bargs->flags;
4475 do_balance:
4476         /*
4477          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4478          * bctl is freed in reset_balance_state, or, if restriper was paused
4479          * all the way until unmount, in free_fs_info.  The flag should be
4480          * cleared after reset_balance_state.
4481          */
4482         need_unlock = false;
4483
4484         ret = btrfs_balance(fs_info, bctl, bargs);
4485         bctl = NULL;
4486
4487         if (ret == 0 || ret == -ECANCELED) {
4488                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4489                         ret = -EFAULT;
4490         }
4491
4492         kfree(bctl);
4493 out_unlock:
4494         mutex_unlock(&fs_info->balance_mutex);
4495         if (need_unlock)
4496                 btrfs_exclop_finish(fs_info);
4497 out:
4498         mnt_drop_write_file(file);
4499         kfree(bargs);
4500         return ret;
4501 }
4502
4503 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4504 {
4505         if (!capable(CAP_SYS_ADMIN))
4506                 return -EPERM;
4507
4508         switch (cmd) {
4509         case BTRFS_BALANCE_CTL_PAUSE:
4510                 return btrfs_pause_balance(fs_info);
4511         case BTRFS_BALANCE_CTL_CANCEL:
4512                 return btrfs_cancel_balance(fs_info);
4513         }
4514
4515         return -EINVAL;
4516 }
4517
4518 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4519                                          void __user *arg)
4520 {
4521         struct btrfs_ioctl_balance_args *bargs;
4522         int ret = 0;
4523
4524         if (!capable(CAP_SYS_ADMIN))
4525                 return -EPERM;
4526
4527         mutex_lock(&fs_info->balance_mutex);
4528         if (!fs_info->balance_ctl) {
4529                 ret = -ENOTCONN;
4530                 goto out;
4531         }
4532
4533         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4534         if (!bargs) {
4535                 ret = -ENOMEM;
4536                 goto out;
4537         }
4538
4539         btrfs_update_ioctl_balance_args(fs_info, bargs);
4540
4541         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4542                 ret = -EFAULT;
4543
4544         kfree(bargs);
4545 out:
4546         mutex_unlock(&fs_info->balance_mutex);
4547         return ret;
4548 }
4549
4550 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4551 {
4552         struct inode *inode = file_inode(file);
4553         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4554         struct btrfs_ioctl_quota_ctl_args *sa;
4555         int ret;
4556
4557         if (!capable(CAP_SYS_ADMIN))
4558                 return -EPERM;
4559
4560         ret = mnt_want_write_file(file);
4561         if (ret)
4562                 return ret;
4563
4564         sa = memdup_user(arg, sizeof(*sa));
4565         if (IS_ERR(sa)) {
4566                 ret = PTR_ERR(sa);
4567                 goto drop_write;
4568         }
4569
4570         down_write(&fs_info->subvol_sem);
4571
4572         switch (sa->cmd) {
4573         case BTRFS_QUOTA_CTL_ENABLE:
4574                 ret = btrfs_quota_enable(fs_info);
4575                 break;
4576         case BTRFS_QUOTA_CTL_DISABLE:
4577                 ret = btrfs_quota_disable(fs_info);
4578                 break;
4579         default:
4580                 ret = -EINVAL;
4581                 break;
4582         }
4583
4584         kfree(sa);
4585         up_write(&fs_info->subvol_sem);
4586 drop_write:
4587         mnt_drop_write_file(file);
4588         return ret;
4589 }
4590
4591 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4592 {
4593         struct inode *inode = file_inode(file);
4594         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4595         struct btrfs_root *root = BTRFS_I(inode)->root;
4596         struct btrfs_ioctl_qgroup_assign_args *sa;
4597         struct btrfs_trans_handle *trans;
4598         int ret;
4599         int err;
4600
4601         if (!capable(CAP_SYS_ADMIN))
4602                 return -EPERM;
4603
4604         ret = mnt_want_write_file(file);
4605         if (ret)
4606                 return ret;
4607
4608         sa = memdup_user(arg, sizeof(*sa));
4609         if (IS_ERR(sa)) {
4610                 ret = PTR_ERR(sa);
4611                 goto drop_write;
4612         }
4613
4614         trans = btrfs_join_transaction(root);
4615         if (IS_ERR(trans)) {
4616                 ret = PTR_ERR(trans);
4617                 goto out;
4618         }
4619
4620         if (sa->assign) {
4621                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4622         } else {
4623                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4624         }
4625
4626         /* update qgroup status and info */
4627         err = btrfs_run_qgroups(trans);
4628         if (err < 0)
4629                 btrfs_handle_fs_error(fs_info, err,
4630                                       "failed to update qgroup status and info");
4631         err = btrfs_end_transaction(trans);
4632         if (err && !ret)
4633                 ret = err;
4634
4635 out:
4636         kfree(sa);
4637 drop_write:
4638         mnt_drop_write_file(file);
4639         return ret;
4640 }
4641
4642 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4643 {
4644         struct inode *inode = file_inode(file);
4645         struct btrfs_root *root = BTRFS_I(inode)->root;
4646         struct btrfs_ioctl_qgroup_create_args *sa;
4647         struct btrfs_trans_handle *trans;
4648         int ret;
4649         int err;
4650
4651         if (!capable(CAP_SYS_ADMIN))
4652                 return -EPERM;
4653
4654         ret = mnt_want_write_file(file);
4655         if (ret)
4656                 return ret;
4657
4658         sa = memdup_user(arg, sizeof(*sa));
4659         if (IS_ERR(sa)) {
4660                 ret = PTR_ERR(sa);
4661                 goto drop_write;
4662         }
4663
4664         if (!sa->qgroupid) {
4665                 ret = -EINVAL;
4666                 goto out;
4667         }
4668
4669         trans = btrfs_join_transaction(root);
4670         if (IS_ERR(trans)) {
4671                 ret = PTR_ERR(trans);
4672                 goto out;
4673         }
4674
4675         if (sa->create) {
4676                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4677         } else {
4678                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4679         }
4680
4681         err = btrfs_end_transaction(trans);
4682         if (err && !ret)
4683                 ret = err;
4684
4685 out:
4686         kfree(sa);
4687 drop_write:
4688         mnt_drop_write_file(file);
4689         return ret;
4690 }
4691
4692 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4693 {
4694         struct inode *inode = file_inode(file);
4695         struct btrfs_root *root = BTRFS_I(inode)->root;
4696         struct btrfs_ioctl_qgroup_limit_args *sa;
4697         struct btrfs_trans_handle *trans;
4698         int ret;
4699         int err;
4700         u64 qgroupid;
4701
4702         if (!capable(CAP_SYS_ADMIN))
4703                 return -EPERM;
4704
4705         ret = mnt_want_write_file(file);
4706         if (ret)
4707                 return ret;
4708
4709         sa = memdup_user(arg, sizeof(*sa));
4710         if (IS_ERR(sa)) {
4711                 ret = PTR_ERR(sa);
4712                 goto drop_write;
4713         }
4714
4715         trans = btrfs_join_transaction(root);
4716         if (IS_ERR(trans)) {
4717                 ret = PTR_ERR(trans);
4718                 goto out;
4719         }
4720
4721         qgroupid = sa->qgroupid;
4722         if (!qgroupid) {
4723                 /* take the current subvol as qgroup */
4724                 qgroupid = root->root_key.objectid;
4725         }
4726
4727         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4728
4729         err = btrfs_end_transaction(trans);
4730         if (err && !ret)
4731                 ret = err;
4732
4733 out:
4734         kfree(sa);
4735 drop_write:
4736         mnt_drop_write_file(file);
4737         return ret;
4738 }
4739
4740 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4741 {
4742         struct inode *inode = file_inode(file);
4743         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4744         struct btrfs_ioctl_quota_rescan_args *qsa;
4745         int ret;
4746
4747         if (!capable(CAP_SYS_ADMIN))
4748                 return -EPERM;
4749
4750         ret = mnt_want_write_file(file);
4751         if (ret)
4752                 return ret;
4753
4754         qsa = memdup_user(arg, sizeof(*qsa));
4755         if (IS_ERR(qsa)) {
4756                 ret = PTR_ERR(qsa);
4757                 goto drop_write;
4758         }
4759
4760         if (qsa->flags) {
4761                 ret = -EINVAL;
4762                 goto out;
4763         }
4764
4765         ret = btrfs_qgroup_rescan(fs_info);
4766
4767 out:
4768         kfree(qsa);
4769 drop_write:
4770         mnt_drop_write_file(file);
4771         return ret;
4772 }
4773
4774 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4775                                                 void __user *arg)
4776 {
4777         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4778
4779         if (!capable(CAP_SYS_ADMIN))
4780                 return -EPERM;
4781
4782         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4783                 qsa.flags = 1;
4784                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4785         }
4786
4787         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4788                 return -EFAULT;
4789
4790         return 0;
4791 }
4792
4793 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4794                                                 void __user *arg)
4795 {
4796         if (!capable(CAP_SYS_ADMIN))
4797                 return -EPERM;
4798
4799         return btrfs_qgroup_wait_for_completion(fs_info, true);
4800 }
4801
4802 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4803                                             struct user_namespace *mnt_userns,
4804                                             struct btrfs_ioctl_received_subvol_args *sa)
4805 {
4806         struct inode *inode = file_inode(file);
4807         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4808         struct btrfs_root *root = BTRFS_I(inode)->root;
4809         struct btrfs_root_item *root_item = &root->root_item;
4810         struct btrfs_trans_handle *trans;
4811         struct timespec64 ct = current_time(inode);
4812         int ret = 0;
4813         int received_uuid_changed;
4814
4815         if (!inode_owner_or_capable(mnt_userns, inode))
4816                 return -EPERM;
4817
4818         ret = mnt_want_write_file(file);
4819         if (ret < 0)
4820                 return ret;
4821
4822         down_write(&fs_info->subvol_sem);
4823
4824         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4825                 ret = -EINVAL;
4826                 goto out;
4827         }
4828
4829         if (btrfs_root_readonly(root)) {
4830                 ret = -EROFS;
4831                 goto out;
4832         }
4833
4834         /*
4835          * 1 - root item
4836          * 2 - uuid items (received uuid + subvol uuid)
4837          */
4838         trans = btrfs_start_transaction(root, 3);
4839         if (IS_ERR(trans)) {
4840                 ret = PTR_ERR(trans);
4841                 trans = NULL;
4842                 goto out;
4843         }
4844
4845         sa->rtransid = trans->transid;
4846         sa->rtime.sec = ct.tv_sec;
4847         sa->rtime.nsec = ct.tv_nsec;
4848
4849         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4850                                        BTRFS_UUID_SIZE);
4851         if (received_uuid_changed &&
4852             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4853                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4854                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4855                                           root->root_key.objectid);
4856                 if (ret && ret != -ENOENT) {
4857                         btrfs_abort_transaction(trans, ret);
4858                         btrfs_end_transaction(trans);
4859                         goto out;
4860                 }
4861         }
4862         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4863         btrfs_set_root_stransid(root_item, sa->stransid);
4864         btrfs_set_root_rtransid(root_item, sa->rtransid);
4865         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4866         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4867         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4868         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4869
4870         ret = btrfs_update_root(trans, fs_info->tree_root,
4871                                 &root->root_key, &root->root_item);
4872         if (ret < 0) {
4873                 btrfs_end_transaction(trans);
4874                 goto out;
4875         }
4876         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4877                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4878                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4879                                           root->root_key.objectid);
4880                 if (ret < 0 && ret != -EEXIST) {
4881                         btrfs_abort_transaction(trans, ret);
4882                         btrfs_end_transaction(trans);
4883                         goto out;
4884                 }
4885         }
4886         ret = btrfs_commit_transaction(trans);
4887 out:
4888         up_write(&fs_info->subvol_sem);
4889         mnt_drop_write_file(file);
4890         return ret;
4891 }
4892
4893 #ifdef CONFIG_64BIT
4894 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4895                                                 void __user *arg)
4896 {
4897         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4898         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4899         int ret = 0;
4900
4901         args32 = memdup_user(arg, sizeof(*args32));
4902         if (IS_ERR(args32))
4903                 return PTR_ERR(args32);
4904
4905         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4906         if (!args64) {
4907                 ret = -ENOMEM;
4908                 goto out;
4909         }
4910
4911         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4912         args64->stransid = args32->stransid;
4913         args64->rtransid = args32->rtransid;
4914         args64->stime.sec = args32->stime.sec;
4915         args64->stime.nsec = args32->stime.nsec;
4916         args64->rtime.sec = args32->rtime.sec;
4917         args64->rtime.nsec = args32->rtime.nsec;
4918         args64->flags = args32->flags;
4919
4920         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4921         if (ret)
4922                 goto out;
4923
4924         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4925         args32->stransid = args64->stransid;
4926         args32->rtransid = args64->rtransid;
4927         args32->stime.sec = args64->stime.sec;
4928         args32->stime.nsec = args64->stime.nsec;
4929         args32->rtime.sec = args64->rtime.sec;
4930         args32->rtime.nsec = args64->rtime.nsec;
4931         args32->flags = args64->flags;
4932
4933         ret = copy_to_user(arg, args32, sizeof(*args32));
4934         if (ret)
4935                 ret = -EFAULT;
4936
4937 out:
4938         kfree(args32);
4939         kfree(args64);
4940         return ret;
4941 }
4942 #endif
4943
4944 static long btrfs_ioctl_set_received_subvol(struct file *file,
4945                                             void __user *arg)
4946 {
4947         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4948         int ret = 0;
4949
4950         sa = memdup_user(arg, sizeof(*sa));
4951         if (IS_ERR(sa))
4952                 return PTR_ERR(sa);
4953
4954         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4955
4956         if (ret)
4957                 goto out;
4958
4959         ret = copy_to_user(arg, sa, sizeof(*sa));
4960         if (ret)
4961                 ret = -EFAULT;
4962
4963 out:
4964         kfree(sa);
4965         return ret;
4966 }
4967
4968 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4969                                         void __user *arg)
4970 {
4971         size_t len;
4972         int ret;
4973         char label[BTRFS_LABEL_SIZE];
4974
4975         spin_lock(&fs_info->super_lock);
4976         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4977         spin_unlock(&fs_info->super_lock);
4978
4979         len = strnlen(label, BTRFS_LABEL_SIZE);
4980
4981         if (len == BTRFS_LABEL_SIZE) {
4982                 btrfs_warn(fs_info,
4983                            "label is too long, return the first %zu bytes",
4984                            --len);
4985         }
4986
4987         ret = copy_to_user(arg, label, len);
4988
4989         return ret ? -EFAULT : 0;
4990 }
4991
4992 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4993 {
4994         struct inode *inode = file_inode(file);
4995         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4996         struct btrfs_root *root = BTRFS_I(inode)->root;
4997         struct btrfs_super_block *super_block = fs_info->super_copy;
4998         struct btrfs_trans_handle *trans;
4999         char label[BTRFS_LABEL_SIZE];
5000         int ret;
5001
5002         if (!capable(CAP_SYS_ADMIN))
5003                 return -EPERM;
5004
5005         if (copy_from_user(label, arg, sizeof(label)))
5006                 return -EFAULT;
5007
5008         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5009                 btrfs_err(fs_info,
5010                           "unable to set label with more than %d bytes",
5011                           BTRFS_LABEL_SIZE - 1);
5012                 return -EINVAL;
5013         }
5014
5015         ret = mnt_want_write_file(file);
5016         if (ret)
5017                 return ret;
5018
5019         trans = btrfs_start_transaction(root, 0);
5020         if (IS_ERR(trans)) {
5021                 ret = PTR_ERR(trans);
5022                 goto out_unlock;
5023         }
5024
5025         spin_lock(&fs_info->super_lock);
5026         strcpy(super_block->label, label);
5027         spin_unlock(&fs_info->super_lock);
5028         ret = btrfs_commit_transaction(trans);
5029
5030 out_unlock:
5031         mnt_drop_write_file(file);
5032         return ret;
5033 }
5034
5035 #define INIT_FEATURE_FLAGS(suffix) \
5036         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5037           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5038           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5039
5040 int btrfs_ioctl_get_supported_features(void __user *arg)
5041 {
5042         static const struct btrfs_ioctl_feature_flags features[3] = {
5043                 INIT_FEATURE_FLAGS(SUPP),
5044                 INIT_FEATURE_FLAGS(SAFE_SET),
5045                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5046         };
5047
5048         if (copy_to_user(arg, &features, sizeof(features)))
5049                 return -EFAULT;
5050
5051         return 0;
5052 }
5053
5054 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5055                                         void __user *arg)
5056 {
5057         struct btrfs_super_block *super_block = fs_info->super_copy;
5058         struct btrfs_ioctl_feature_flags features;
5059
5060         features.compat_flags = btrfs_super_compat_flags(super_block);
5061         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5062         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5063
5064         if (copy_to_user(arg, &features, sizeof(features)))
5065                 return -EFAULT;
5066
5067         return 0;
5068 }
5069
5070 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5071                               enum btrfs_feature_set set,
5072                               u64 change_mask, u64 flags, u64 supported_flags,
5073                               u64 safe_set, u64 safe_clear)
5074 {
5075         const char *type = btrfs_feature_set_name(set);
5076         char *names;
5077         u64 disallowed, unsupported;
5078         u64 set_mask = flags & change_mask;
5079         u64 clear_mask = ~flags & change_mask;
5080
5081         unsupported = set_mask & ~supported_flags;
5082         if (unsupported) {
5083                 names = btrfs_printable_features(set, unsupported);
5084                 if (names) {
5085                         btrfs_warn(fs_info,
5086                                    "this kernel does not support the %s feature bit%s",
5087                                    names, strchr(names, ',') ? "s" : "");
5088                         kfree(names);
5089                 } else
5090                         btrfs_warn(fs_info,
5091                                    "this kernel does not support %s bits 0x%llx",
5092                                    type, unsupported);
5093                 return -EOPNOTSUPP;
5094         }
5095
5096         disallowed = set_mask & ~safe_set;
5097         if (disallowed) {
5098                 names = btrfs_printable_features(set, disallowed);
5099                 if (names) {
5100                         btrfs_warn(fs_info,
5101                                    "can't set the %s feature bit%s while mounted",
5102                                    names, strchr(names, ',') ? "s" : "");
5103                         kfree(names);
5104                 } else
5105                         btrfs_warn(fs_info,
5106                                    "can't set %s bits 0x%llx while mounted",
5107                                    type, disallowed);
5108                 return -EPERM;
5109         }
5110
5111         disallowed = clear_mask & ~safe_clear;
5112         if (disallowed) {
5113                 names = btrfs_printable_features(set, disallowed);
5114                 if (names) {
5115                         btrfs_warn(fs_info,
5116                                    "can't clear the %s feature bit%s while mounted",
5117                                    names, strchr(names, ',') ? "s" : "");
5118                         kfree(names);
5119                 } else
5120                         btrfs_warn(fs_info,
5121                                    "can't clear %s bits 0x%llx while mounted",
5122                                    type, disallowed);
5123                 return -EPERM;
5124         }
5125
5126         return 0;
5127 }
5128
5129 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5130 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5131                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5132                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5133                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5134
5135 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5136 {
5137         struct inode *inode = file_inode(file);
5138         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5139         struct btrfs_root *root = BTRFS_I(inode)->root;
5140         struct btrfs_super_block *super_block = fs_info->super_copy;
5141         struct btrfs_ioctl_feature_flags flags[2];
5142         struct btrfs_trans_handle *trans;
5143         u64 newflags;
5144         int ret;
5145
5146         if (!capable(CAP_SYS_ADMIN))
5147                 return -EPERM;
5148
5149         if (copy_from_user(flags, arg, sizeof(flags)))
5150                 return -EFAULT;
5151
5152         /* Nothing to do */
5153         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5154             !flags[0].incompat_flags)
5155                 return 0;
5156
5157         ret = check_feature(fs_info, flags[0].compat_flags,
5158                             flags[1].compat_flags, COMPAT);
5159         if (ret)
5160                 return ret;
5161
5162         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5163                             flags[1].compat_ro_flags, COMPAT_RO);
5164         if (ret)
5165                 return ret;
5166
5167         ret = check_feature(fs_info, flags[0].incompat_flags,
5168                             flags[1].incompat_flags, INCOMPAT);
5169         if (ret)
5170                 return ret;
5171
5172         ret = mnt_want_write_file(file);
5173         if (ret)
5174                 return ret;
5175
5176         trans = btrfs_start_transaction(root, 0);
5177         if (IS_ERR(trans)) {
5178                 ret = PTR_ERR(trans);
5179                 goto out_drop_write;
5180         }
5181
5182         spin_lock(&fs_info->super_lock);
5183         newflags = btrfs_super_compat_flags(super_block);
5184         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5185         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5186         btrfs_set_super_compat_flags(super_block, newflags);
5187
5188         newflags = btrfs_super_compat_ro_flags(super_block);
5189         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5190         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5191         btrfs_set_super_compat_ro_flags(super_block, newflags);
5192
5193         newflags = btrfs_super_incompat_flags(super_block);
5194         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5195         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5196         btrfs_set_super_incompat_flags(super_block, newflags);
5197         spin_unlock(&fs_info->super_lock);
5198
5199         ret = btrfs_commit_transaction(trans);
5200 out_drop_write:
5201         mnt_drop_write_file(file);
5202
5203         return ret;
5204 }
5205
5206 static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat)
5207 {
5208         struct btrfs_ioctl_send_args *arg;
5209         int ret;
5210
5211         if (compat) {
5212 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5213                 struct btrfs_ioctl_send_args_32 args32;
5214
5215                 ret = copy_from_user(&args32, argp, sizeof(args32));
5216                 if (ret)
5217                         return -EFAULT;
5218                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5219                 if (!arg)
5220                         return -ENOMEM;
5221                 arg->send_fd = args32.send_fd;
5222                 arg->clone_sources_count = args32.clone_sources_count;
5223                 arg->clone_sources = compat_ptr(args32.clone_sources);
5224                 arg->parent_root = args32.parent_root;
5225                 arg->flags = args32.flags;
5226                 memcpy(arg->reserved, args32.reserved,
5227                        sizeof(args32.reserved));
5228 #else
5229                 return -ENOTTY;
5230 #endif
5231         } else {
5232                 arg = memdup_user(argp, sizeof(*arg));
5233                 if (IS_ERR(arg))
5234                         return PTR_ERR(arg);
5235         }
5236         ret = btrfs_ioctl_send(inode, arg);
5237         kfree(arg);
5238         return ret;
5239 }
5240
5241 static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp,
5242                                     bool compat)
5243 {
5244         struct btrfs_ioctl_encoded_io_args args = { 0 };
5245         size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
5246                                              flags);
5247         size_t copy_end;
5248         struct iovec iovstack[UIO_FASTIOV];
5249         struct iovec *iov = iovstack;
5250         struct iov_iter iter;
5251         loff_t pos;
5252         struct kiocb kiocb;
5253         ssize_t ret;
5254
5255         if (!capable(CAP_SYS_ADMIN)) {
5256                 ret = -EPERM;
5257                 goto out_acct;
5258         }
5259
5260         if (compat) {
5261 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5262                 struct btrfs_ioctl_encoded_io_args_32 args32;
5263
5264                 copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
5265                                        flags);
5266                 if (copy_from_user(&args32, argp, copy_end)) {
5267                         ret = -EFAULT;
5268                         goto out_acct;
5269                 }
5270                 args.iov = compat_ptr(args32.iov);
5271                 args.iovcnt = args32.iovcnt;
5272                 args.offset = args32.offset;
5273                 args.flags = args32.flags;
5274 #else
5275                 return -ENOTTY;
5276 #endif
5277         } else {
5278                 copy_end = copy_end_kernel;
5279                 if (copy_from_user(&args, argp, copy_end)) {
5280                         ret = -EFAULT;
5281                         goto out_acct;
5282                 }
5283         }
5284         if (args.flags != 0) {
5285                 ret = -EINVAL;
5286                 goto out_acct;
5287         }
5288
5289         ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5290                            &iov, &iter);
5291         if (ret < 0)
5292                 goto out_acct;
5293
5294         if (iov_iter_count(&iter) == 0) {
5295                 ret = 0;
5296                 goto out_iov;
5297         }
5298         pos = args.offset;
5299         ret = rw_verify_area(READ, file, &pos, args.len);
5300         if (ret < 0)
5301                 goto out_iov;
5302
5303         init_sync_kiocb(&kiocb, file);
5304         kiocb.ki_pos = pos;
5305
5306         ret = btrfs_encoded_read(&kiocb, &iter, &args);
5307         if (ret >= 0) {
5308                 fsnotify_access(file);
5309                 if (copy_to_user(argp + copy_end,
5310                                  (char *)&args + copy_end_kernel,
5311                                  sizeof(args) - copy_end_kernel))
5312                         ret = -EFAULT;
5313         }
5314
5315 out_iov:
5316         kfree(iov);
5317 out_acct:
5318         if (ret > 0)
5319                 add_rchar(current, ret);
5320         inc_syscr(current);
5321         return ret;
5322 }
5323
5324 static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat)
5325 {
5326         struct btrfs_ioctl_encoded_io_args args;
5327         struct iovec iovstack[UIO_FASTIOV];
5328         struct iovec *iov = iovstack;
5329         struct iov_iter iter;
5330         loff_t pos;
5331         struct kiocb kiocb;
5332         ssize_t ret;
5333
5334         if (!capable(CAP_SYS_ADMIN)) {
5335                 ret = -EPERM;
5336                 goto out_acct;
5337         }
5338
5339         if (!(file->f_mode & FMODE_WRITE)) {
5340                 ret = -EBADF;
5341                 goto out_acct;
5342         }
5343
5344         if (compat) {
5345 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5346                 struct btrfs_ioctl_encoded_io_args_32 args32;
5347
5348                 if (copy_from_user(&args32, argp, sizeof(args32))) {
5349                         ret = -EFAULT;
5350                         goto out_acct;
5351                 }
5352                 args.iov = compat_ptr(args32.iov);
5353                 args.iovcnt = args32.iovcnt;
5354                 args.offset = args32.offset;
5355                 args.flags = args32.flags;
5356                 args.len = args32.len;
5357                 args.unencoded_len = args32.unencoded_len;
5358                 args.unencoded_offset = args32.unencoded_offset;
5359                 args.compression = args32.compression;
5360                 args.encryption = args32.encryption;
5361                 memcpy(args.reserved, args32.reserved, sizeof(args.reserved));
5362 #else
5363                 return -ENOTTY;
5364 #endif
5365         } else {
5366                 if (copy_from_user(&args, argp, sizeof(args))) {
5367                         ret = -EFAULT;
5368                         goto out_acct;
5369                 }
5370         }
5371
5372         ret = -EINVAL;
5373         if (args.flags != 0)
5374                 goto out_acct;
5375         if (memchr_inv(args.reserved, 0, sizeof(args.reserved)))
5376                 goto out_acct;
5377         if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE &&
5378             args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE)
5379                 goto out_acct;
5380         if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES ||
5381             args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES)
5382                 goto out_acct;
5383         if (args.unencoded_offset > args.unencoded_len)
5384                 goto out_acct;
5385         if (args.len > args.unencoded_len - args.unencoded_offset)
5386                 goto out_acct;
5387
5388         ret = import_iovec(WRITE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
5389                            &iov, &iter);
5390         if (ret < 0)
5391                 goto out_acct;
5392
5393         file_start_write(file);
5394
5395         if (iov_iter_count(&iter) == 0) {
5396                 ret = 0;
5397                 goto out_end_write;
5398         }
5399         pos = args.offset;
5400         ret = rw_verify_area(WRITE, file, &pos, args.len);
5401         if (ret < 0)
5402                 goto out_end_write;
5403
5404         init_sync_kiocb(&kiocb, file);
5405         ret = kiocb_set_rw_flags(&kiocb, 0);
5406         if (ret)
5407                 goto out_end_write;
5408         kiocb.ki_pos = pos;
5409
5410         ret = btrfs_do_write_iter(&kiocb, &iter, &args);
5411         if (ret > 0)
5412                 fsnotify_modify(file);
5413
5414 out_end_write:
5415         file_end_write(file);
5416         kfree(iov);
5417 out_acct:
5418         if (ret > 0)
5419                 add_wchar(current, ret);
5420         inc_syscw(current);
5421         return ret;
5422 }
5423
5424 long btrfs_ioctl(struct file *file, unsigned int
5425                 cmd, unsigned long arg)
5426 {
5427         struct inode *inode = file_inode(file);
5428         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5429         struct btrfs_root *root = BTRFS_I(inode)->root;
5430         void __user *argp = (void __user *)arg;
5431
5432         switch (cmd) {
5433         case FS_IOC_GETVERSION:
5434                 return btrfs_ioctl_getversion(inode, argp);
5435         case FS_IOC_GETFSLABEL:
5436                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5437         case FS_IOC_SETFSLABEL:
5438                 return btrfs_ioctl_set_fslabel(file, argp);
5439         case FITRIM:
5440                 return btrfs_ioctl_fitrim(fs_info, argp);
5441         case BTRFS_IOC_SNAP_CREATE:
5442                 return btrfs_ioctl_snap_create(file, argp, 0);
5443         case BTRFS_IOC_SNAP_CREATE_V2:
5444                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5445         case BTRFS_IOC_SUBVOL_CREATE:
5446                 return btrfs_ioctl_snap_create(file, argp, 1);
5447         case BTRFS_IOC_SUBVOL_CREATE_V2:
5448                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5449         case BTRFS_IOC_SNAP_DESTROY:
5450                 return btrfs_ioctl_snap_destroy(file, argp, false);
5451         case BTRFS_IOC_SNAP_DESTROY_V2:
5452                 return btrfs_ioctl_snap_destroy(file, argp, true);
5453         case BTRFS_IOC_SUBVOL_GETFLAGS:
5454                 return btrfs_ioctl_subvol_getflags(inode, argp);
5455         case BTRFS_IOC_SUBVOL_SETFLAGS:
5456                 return btrfs_ioctl_subvol_setflags(file, argp);
5457         case BTRFS_IOC_DEFAULT_SUBVOL:
5458                 return btrfs_ioctl_default_subvol(file, argp);
5459         case BTRFS_IOC_DEFRAG:
5460                 return btrfs_ioctl_defrag(file, NULL);
5461         case BTRFS_IOC_DEFRAG_RANGE:
5462                 return btrfs_ioctl_defrag(file, argp);
5463         case BTRFS_IOC_RESIZE:
5464                 return btrfs_ioctl_resize(file, argp);
5465         case BTRFS_IOC_ADD_DEV:
5466                 return btrfs_ioctl_add_dev(fs_info, argp);
5467         case BTRFS_IOC_RM_DEV:
5468                 return btrfs_ioctl_rm_dev(file, argp);
5469         case BTRFS_IOC_RM_DEV_V2:
5470                 return btrfs_ioctl_rm_dev_v2(file, argp);
5471         case BTRFS_IOC_FS_INFO:
5472                 return btrfs_ioctl_fs_info(fs_info, argp);
5473         case BTRFS_IOC_DEV_INFO:
5474                 return btrfs_ioctl_dev_info(fs_info, argp);
5475         case BTRFS_IOC_TREE_SEARCH:
5476                 return btrfs_ioctl_tree_search(inode, argp);
5477         case BTRFS_IOC_TREE_SEARCH_V2:
5478                 return btrfs_ioctl_tree_search_v2(inode, argp);
5479         case BTRFS_IOC_INO_LOOKUP:
5480                 return btrfs_ioctl_ino_lookup(root, argp);
5481         case BTRFS_IOC_INO_PATHS:
5482                 return btrfs_ioctl_ino_to_path(root, argp);
5483         case BTRFS_IOC_LOGICAL_INO:
5484                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5485         case BTRFS_IOC_LOGICAL_INO_V2:
5486                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5487         case BTRFS_IOC_SPACE_INFO:
5488                 return btrfs_ioctl_space_info(fs_info, argp);
5489         case BTRFS_IOC_SYNC: {
5490                 int ret;
5491
5492                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
5493                 if (ret)
5494                         return ret;
5495                 ret = btrfs_sync_fs(inode->i_sb, 1);
5496                 /*
5497                  * The transaction thread may want to do more work,
5498                  * namely it pokes the cleaner kthread that will start
5499                  * processing uncleaned subvols.
5500                  */
5501                 wake_up_process(fs_info->transaction_kthread);
5502                 return ret;
5503         }
5504         case BTRFS_IOC_START_SYNC:
5505                 return btrfs_ioctl_start_sync(root, argp);
5506         case BTRFS_IOC_WAIT_SYNC:
5507                 return btrfs_ioctl_wait_sync(fs_info, argp);
5508         case BTRFS_IOC_SCRUB:
5509                 return btrfs_ioctl_scrub(file, argp);
5510         case BTRFS_IOC_SCRUB_CANCEL:
5511                 return btrfs_ioctl_scrub_cancel(fs_info);
5512         case BTRFS_IOC_SCRUB_PROGRESS:
5513                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5514         case BTRFS_IOC_BALANCE_V2:
5515                 return btrfs_ioctl_balance(file, argp);
5516         case BTRFS_IOC_BALANCE_CTL:
5517                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5518         case BTRFS_IOC_BALANCE_PROGRESS:
5519                 return btrfs_ioctl_balance_progress(fs_info, argp);
5520         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5521                 return btrfs_ioctl_set_received_subvol(file, argp);
5522 #ifdef CONFIG_64BIT
5523         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5524                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5525 #endif
5526         case BTRFS_IOC_SEND:
5527                 return _btrfs_ioctl_send(inode, argp, false);
5528 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5529         case BTRFS_IOC_SEND_32:
5530                 return _btrfs_ioctl_send(inode, argp, true);
5531 #endif
5532         case BTRFS_IOC_GET_DEV_STATS:
5533                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5534         case BTRFS_IOC_QUOTA_CTL:
5535                 return btrfs_ioctl_quota_ctl(file, argp);
5536         case BTRFS_IOC_QGROUP_ASSIGN:
5537                 return btrfs_ioctl_qgroup_assign(file, argp);
5538         case BTRFS_IOC_QGROUP_CREATE:
5539                 return btrfs_ioctl_qgroup_create(file, argp);
5540         case BTRFS_IOC_QGROUP_LIMIT:
5541                 return btrfs_ioctl_qgroup_limit(file, argp);
5542         case BTRFS_IOC_QUOTA_RESCAN:
5543                 return btrfs_ioctl_quota_rescan(file, argp);
5544         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5545                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5546         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5547                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5548         case BTRFS_IOC_DEV_REPLACE:
5549                 return btrfs_ioctl_dev_replace(fs_info, argp);
5550         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5551                 return btrfs_ioctl_get_supported_features(argp);
5552         case BTRFS_IOC_GET_FEATURES:
5553                 return btrfs_ioctl_get_features(fs_info, argp);
5554         case BTRFS_IOC_SET_FEATURES:
5555                 return btrfs_ioctl_set_features(file, argp);
5556         case BTRFS_IOC_GET_SUBVOL_INFO:
5557                 return btrfs_ioctl_get_subvol_info(inode, argp);
5558         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5559                 return btrfs_ioctl_get_subvol_rootref(root, argp);
5560         case BTRFS_IOC_INO_LOOKUP_USER:
5561                 return btrfs_ioctl_ino_lookup_user(file, argp);
5562         case FS_IOC_ENABLE_VERITY:
5563                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5564         case FS_IOC_MEASURE_VERITY:
5565                 return fsverity_ioctl_measure(file, argp);
5566         case BTRFS_IOC_ENCODED_READ:
5567                 return btrfs_ioctl_encoded_read(file, argp, false);
5568         case BTRFS_IOC_ENCODED_WRITE:
5569                 return btrfs_ioctl_encoded_write(file, argp, false);
5570 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5571         case BTRFS_IOC_ENCODED_READ_32:
5572                 return btrfs_ioctl_encoded_read(file, argp, true);
5573         case BTRFS_IOC_ENCODED_WRITE_32:
5574                 return btrfs_ioctl_encoded_write(file, argp, true);
5575 #endif
5576         }
5577
5578         return -ENOTTY;
5579 }
5580
5581 #ifdef CONFIG_COMPAT
5582 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5583 {
5584         /*
5585          * These all access 32-bit values anyway so no further
5586          * handling is necessary.
5587          */
5588         switch (cmd) {
5589         case FS_IOC32_GETVERSION:
5590                 cmd = FS_IOC_GETVERSION;
5591                 break;
5592         }
5593
5594         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5595 }
5596 #endif