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 "ctree.h"
  32 #include "disk-io.h"
  33 #include "export.h"
  34 #include "transaction.h"
  35 #include "btrfs_inode.h"
  36 #include "print-tree.h"
  37 #include "volumes.h"
  38 #include "locking.h"
  39 #include "backref.h"
  40 #include "rcu-string.h"
  41 #include "send.h"
  42 #include "dev-replace.h"
  43 #include "props.h"
  44 #include "sysfs.h"
  45 #include "qgroup.h"
  46 #include "tree-log.h"
  47 #include "compression.h"
  48 #include "space-info.h"
  49 #include "delalloc-space.h"
  50 #include "block-group.h"
  51
  52 #ifdef CONFIG_64BIT
  53 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  54  * structures are incorrect, as the timespec structure from userspace
  55  * is 4 bytes too small. We define these alternatives here to teach
  56  * the kernel about the 32-bit struct packing.
  57  */
  58 struct btrfs_ioctl_timespec_32 {
  59         __u64 sec;
  60         __u32 nsec;
  61 } __attribute__ ((__packed__));
  62
  63 struct btrfs_ioctl_received_subvol_args_32 {
  64         char    uuid[BTRFS_UUID_SIZE];  /* in */
  65         __u64   stransid;               /* in */
  66         __u64   rtransid;               /* out */
  67         struct btrfs_ioctl_timespec_32 stime; /* in */
  68         struct btrfs_ioctl_timespec_32 rtime; /* out */
  69         __u64   flags;                  /* in */
  70         __u64   reserved[16];           /* in */
  71 } __attribute__ ((__packed__));
  72
  73 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  74                                 struct btrfs_ioctl_received_subvol_args_32)
  75 #endif
  76
  77 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  78 struct btrfs_ioctl_send_args_32 {
  79         __s64 send_fd;                  /* in */
  80         __u64 clone_sources_count;      /* in */
  81         compat_uptr_t clone_sources;    /* in */
  82         __u64 parent_root;              /* in */
  83         __u64 flags;                    /* in */
  84         __u64 reserved[4];              /* in */
  85 } __attribute__ ((__packed__));
  86
  87 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  88                                struct btrfs_ioctl_send_args_32)
  89 #endif
  90
  91 /* Mask out flags that are inappropriate for the given type of inode. */
  92 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  93                 unsigned int flags)
  94 {
  95         if (S_ISDIR(inode->i_mode))
  96                 return flags;
  97         else if (S_ISREG(inode->i_mode))
  98                 return flags & ~FS_DIRSYNC_FL;
  99         else
 100                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 101 }
 102
 103 /*
 104  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 105  * ioctl.
 106  */
 107 static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode)
 108 {
 109         unsigned int iflags = 0;
 110         u32 flags = binode->flags;
 111         u32 ro_flags = binode->ro_flags;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127         if (ro_flags & BTRFS_INODE_RO_VERITY)
 128                 iflags |= FS_VERITY_FL;
 129
 130         if (flags & BTRFS_INODE_NOCOMPRESS)
 131                 iflags |= FS_NOCOMP_FL;
 132         else if (flags & BTRFS_INODE_COMPRESS)
 133                 iflags |= FS_COMPR_FL;
 134
 135         return iflags;
 136 }
 137
 138 /*
 139  * Update inode->i_flags based on the btrfs internal flags.
 140  */
 141 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 142 {
 143         struct btrfs_inode *binode = BTRFS_I(inode);
 144         unsigned int new_fl = 0;
 145
 146         if (binode->flags & BTRFS_INODE_SYNC)
 147                 new_fl |= S_SYNC;
 148         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 149                 new_fl |= S_IMMUTABLE;
 150         if (binode->flags & BTRFS_INODE_APPEND)
 151                 new_fl |= S_APPEND;
 152         if (binode->flags & BTRFS_INODE_NOATIME)
 153                 new_fl |= S_NOATIME;
 154         if (binode->flags & BTRFS_INODE_DIRSYNC)
 155                 new_fl |= S_DIRSYNC;
 156         if (binode->ro_flags & BTRFS_INODE_RO_VERITY)
 157                 new_fl |= S_VERITY;
 158
 159         set_mask_bits(&inode->i_flags,
 160                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC |
 161                       S_VERITY, new_fl);
 162 }
 163
 164 /*
 165  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 166  * the old and new flags are not conflicting
 167  */
 168 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 169 {
 170         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 171                       FS_NOATIME_FL | FS_NODUMP_FL | \
 172                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 173                       FS_NOCOMP_FL | FS_COMPR_FL |
 174                       FS_NOCOW_FL))
 175                 return -EOPNOTSUPP;
 176
 177         /* COMPR and NOCOMP on new/old are valid */
 178         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 179                 return -EINVAL;
 180
 181         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 182                 return -EINVAL;
 183
 184         /* NOCOW and compression options are mutually exclusive */
 185         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 186                 return -EINVAL;
 187         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 188                 return -EINVAL;
 189
 190         return 0;
 191 }
 192
 193 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 194                                     unsigned int flags)
 195 {
 196         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 197                 return -EPERM;
 198
 199         return 0;
 200 }
 201
 202 /*
 203  * Set flags/xflags from the internal inode flags. The remaining items of
 204  * fsxattr are zeroed.
 205  */
 206 int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 207 {
 208         struct btrfs_inode *binode = BTRFS_I(d_inode(dentry));
 209
 210         fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode));
 211         return 0;
 212 }
 213
 214 int btrfs_fileattr_set(struct user_namespace *mnt_userns,
 215                        struct dentry *dentry, struct fileattr *fa)
 216 {
 217         struct inode *inode = d_inode(dentry);
 218         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 219         struct btrfs_inode *binode = BTRFS_I(inode);
 220         struct btrfs_root *root = binode->root;
 221         struct btrfs_trans_handle *trans;
 222         unsigned int fsflags, old_fsflags;
 223         int ret;
 224         const char *comp = NULL;
 225         u32 binode_flags;
 226
 227         if (btrfs_root_readonly(root))
 228                 return -EROFS;
 229
 230         if (fileattr_has_fsx(fa))
 231                 return -EOPNOTSUPP;
 232
 233         fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags);
 234         old_fsflags = btrfs_inode_flags_to_fsflags(binode);
 235         ret = check_fsflags(old_fsflags, fsflags);
 236         if (ret)
 237                 return ret;
 238
 239         ret = check_fsflags_compatible(fs_info, fsflags);
 240         if (ret)
 241                 return ret;
 242
 243         binode_flags = binode->flags;
 244         if (fsflags & FS_SYNC_FL)
 245                 binode_flags |= BTRFS_INODE_SYNC;
 246         else
 247                 binode_flags &= ~BTRFS_INODE_SYNC;
 248         if (fsflags & FS_IMMUTABLE_FL)
 249                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 250         else
 251                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 252         if (fsflags & FS_APPEND_FL)
 253                 binode_flags |= BTRFS_INODE_APPEND;
 254         else
 255                 binode_flags &= ~BTRFS_INODE_APPEND;
 256         if (fsflags & FS_NODUMP_FL)
 257                 binode_flags |= BTRFS_INODE_NODUMP;
 258         else
 259                 binode_flags &= ~BTRFS_INODE_NODUMP;
 260         if (fsflags & FS_NOATIME_FL)
 261                 binode_flags |= BTRFS_INODE_NOATIME;
 262         else
 263                 binode_flags &= ~BTRFS_INODE_NOATIME;
 264
 265         /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */
 266         if (!fa->flags_valid) {
 267                 /* 1 item for the inode */
 268                 trans = btrfs_start_transaction(root, 1);
 269                 if (IS_ERR(trans))
 270                         return PTR_ERR(trans);
 271                 goto update_flags;
 272         }
 273
 274         if (fsflags & FS_DIRSYNC_FL)
 275                 binode_flags |= BTRFS_INODE_DIRSYNC;
 276         else
 277                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 278         if (fsflags & FS_NOCOW_FL) {
 279                 if (S_ISREG(inode->i_mode)) {
 280                         /*
 281                          * It's safe to turn csums off here, no extents exist.
 282                          * Otherwise we want the flag to reflect the real COW
 283                          * status of the file and will not set it.
 284                          */
 285                         if (inode->i_size == 0)
 286                                 binode_flags |= BTRFS_INODE_NODATACOW |
 287                                                 BTRFS_INODE_NODATASUM;
 288                 } else {
 289                         binode_flags |= BTRFS_INODE_NODATACOW;
 290                 }
 291         } else {
 292                 /*
 293                  * Revert back under same assumptions as above
 294                  */
 295                 if (S_ISREG(inode->i_mode)) {
 296                         if (inode->i_size == 0)
 297                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 298                                                   BTRFS_INODE_NODATASUM);
 299                 } else {
 300                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 301                 }
 302         }
 303
 304         /*
 305          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 306          * flag may be changed automatically if compression code won't make
 307          * things smaller.
 308          */
 309         if (fsflags & FS_NOCOMP_FL) {
 310                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 311                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 312         } else if (fsflags & FS_COMPR_FL) {
 313
 314                 if (IS_SWAPFILE(inode))
 315                         return -ETXTBSY;
 316
 317                 binode_flags |= BTRFS_INODE_COMPRESS;
 318                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 319
 320                 comp = btrfs_compress_type2str(fs_info->compress_type);
 321                 if (!comp || comp[0] == 0)
 322                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 323         } else {
 324                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 325         }
 326
 327         /*
 328          * 1 for inode item
 329          * 2 for properties
 330          */
 331         trans = btrfs_start_transaction(root, 3);
 332         if (IS_ERR(trans))
 333                 return PTR_ERR(trans);
 334
 335         if (comp) {
 336                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 337                                      strlen(comp), 0);
 338                 if (ret) {
 339                         btrfs_abort_transaction(trans, ret);
 340                         goto out_end_trans;
 341                 }
 342         } else {
 343                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 344                                      0, 0);
 345                 if (ret && ret != -ENODATA) {
 346                         btrfs_abort_transaction(trans, ret);
 347                         goto out_end_trans;
 348                 }
 349         }
 350
 351 update_flags:
 352         binode->flags = binode_flags;
 353         btrfs_sync_inode_flags_to_i_flags(inode);
 354         inode_inc_iversion(inode);
 355         inode->i_ctime = current_time(inode);
 356         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 357
 358  out_end_trans:
 359         btrfs_end_transaction(trans);
 360         return ret;
 361 }
 362
 363 /*
 364  * Start exclusive operation @type, return true on success
 365  */
 366 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 367                         enum btrfs_exclusive_operation type)
 368 {
 369         bool ret = false;
 370
 371         spin_lock(&fs_info->super_lock);
 372         if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) {
 373                 fs_info->exclusive_operation = type;
 374                 ret = true;
 375         }
 376         spin_unlock(&fs_info->super_lock);
 377
 378         return ret;
 379 }
 380
 381 /*
 382  * Conditionally allow to enter the exclusive operation in case it's compatible
 383  * with the running one.  This must be paired with btrfs_exclop_start_unlock and
 384  * btrfs_exclop_finish.
 385  *
 386  * Compatibility:
 387  * - the same type is already running
 388  * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller
 389  *   must check the condition first that would allow none -> @type
 390  */
 391 bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info,
 392                                  enum btrfs_exclusive_operation type)
 393 {
 394         spin_lock(&fs_info->super_lock);
 395         if (fs_info->exclusive_operation == type)
 396                 return true;
 397
 398         spin_unlock(&fs_info->super_lock);
 399         return false;
 400 }
 401
 402 void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info)
 403 {
 404         spin_unlock(&fs_info->super_lock);
 405 }
 406
 407 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 408 {
 409         spin_lock(&fs_info->super_lock);
 410         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 411         spin_unlock(&fs_info->super_lock);
 412         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 413 }
 414
 415 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 416 {
 417         struct inode *inode = file_inode(file);
 418
 419         return put_user(inode->i_generation, arg);
 420 }
 421
 422 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 423                                         void __user *arg)
 424 {
 425         struct btrfs_device *device;
 426         struct request_queue *q;
 427         struct fstrim_range range;
 428         u64 minlen = ULLONG_MAX;
 429         u64 num_devices = 0;
 430         int ret;
 431
 432         if (!capable(CAP_SYS_ADMIN))
 433                 return -EPERM;
 434
 435         /*
 436          * btrfs_trim_block_group() depends on space cache, which is not
 437          * available in zoned filesystem. So, disallow fitrim on a zoned
 438          * filesystem for now.
 439          */
 440         if (btrfs_is_zoned(fs_info))
 441                 return -EOPNOTSUPP;
 442
 443         /*
 444          * If the fs is mounted with nologreplay, which requires it to be
 445          * mounted in RO mode as well, we can not allow discard on free space
 446          * inside block groups, because log trees refer to extents that are not
 447          * pinned in a block group's free space cache (pinning the extents is
 448          * precisely the first phase of replaying a log tree).
 449          */
 450         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 451                 return -EROFS;
 452
 453         rcu_read_lock();
 454         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 455                                 dev_list) {
 456                 if (!device->bdev)
 457                         continue;
 458                 q = bdev_get_queue(device->bdev);
 459                 if (blk_queue_discard(q)) {
 460                         num_devices++;
 461                         minlen = min_t(u64, q->limits.discard_granularity,
 462                                      minlen);
 463                 }
 464         }
 465         rcu_read_unlock();
 466
 467         if (!num_devices)
 468                 return -EOPNOTSUPP;
 469         if (copy_from_user(&range, arg, sizeof(range)))
 470                 return -EFAULT;
 471
 472         /*
 473          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 474          * block group is in the logical address space, which can be any
 475          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 476          */
 477         if (range.len < fs_info->sb->s_blocksize)
 478                 return -EINVAL;
 479
 480         range.minlen = max(range.minlen, minlen);
 481         ret = btrfs_trim_fs(fs_info, &range);
 482         if (ret < 0)
 483                 return ret;
 484
 485         if (copy_to_user(arg, &range, sizeof(range)))
 486                 return -EFAULT;
 487
 488         return 0;
 489 }
 490
 491 int __pure btrfs_is_empty_uuid(u8 *uuid)
 492 {
 493         int i;
 494
 495         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 496                 if (uuid[i])
 497                         return 0;
 498         }
 499         return 1;
 500 }
 501
 502 static noinline int create_subvol(struct user_namespace *mnt_userns,
 503                                   struct inode *dir, struct dentry *dentry,
 504                                   const char *name, int namelen,
 505                                   struct btrfs_qgroup_inherit *inherit)
 506 {
 507         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 508         struct btrfs_trans_handle *trans;
 509         struct btrfs_key key;
 510         struct btrfs_root_item *root_item;
 511         struct btrfs_inode_item *inode_item;
 512         struct extent_buffer *leaf;
 513         struct btrfs_root *root = BTRFS_I(dir)->root;
 514         struct btrfs_root *new_root;
 515         struct btrfs_block_rsv block_rsv;
 516         struct timespec64 cur_time = current_time(dir);
 517         struct inode *inode;
 518         int ret;
 519         int err;
 520         dev_t anon_dev = 0;
 521         u64 objectid;
 522         u64 index = 0;
 523
 524         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 525         if (!root_item)
 526                 return -ENOMEM;
 527
 528         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 529         if (ret)
 530                 goto fail_free;
 531
 532         ret = get_anon_bdev(&anon_dev);
 533         if (ret < 0)
 534                 goto fail_free;
 535
 536         /*
 537          * Don't create subvolume whose level is not zero. Or qgroup will be
 538          * screwed up since it assumes subvolume qgroup's level to be 0.
 539          */
 540         if (btrfs_qgroup_level(objectid)) {
 541                 ret = -ENOSPC;
 542                 goto fail_free;
 543         }
 544
 545         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 546         /*
 547          * The same as the snapshot creation, please see the comment
 548          * of create_snapshot().
 549          */
 550         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 551         if (ret)
 552                 goto fail_free;
 553
 554         trans = btrfs_start_transaction(root, 0);
 555         if (IS_ERR(trans)) {
 556                 ret = PTR_ERR(trans);
 557                 btrfs_subvolume_release_metadata(root, &block_rsv);
 558                 goto fail_free;
 559         }
 560         trans->block_rsv = &block_rsv;
 561         trans->bytes_reserved = block_rsv.size;
 562
 563         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 564         if (ret)
 565                 goto fail;
 566
 567         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 568                                       BTRFS_NESTING_NORMAL);
 569         if (IS_ERR(leaf)) {
 570                 ret = PTR_ERR(leaf);
 571                 goto fail;
 572         }
 573
 574         btrfs_mark_buffer_dirty(leaf);
 575
 576         inode_item = &root_item->inode;
 577         btrfs_set_stack_inode_generation(inode_item, 1);
 578         btrfs_set_stack_inode_size(inode_item, 3);
 579         btrfs_set_stack_inode_nlink(inode_item, 1);
 580         btrfs_set_stack_inode_nbytes(inode_item,
 581                                      fs_info->nodesize);
 582         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 583
 584         btrfs_set_root_flags(root_item, 0);
 585         btrfs_set_root_limit(root_item, 0);
 586         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 587
 588         btrfs_set_root_bytenr(root_item, leaf->start);
 589         btrfs_set_root_generation(root_item, trans->transid);
 590         btrfs_set_root_level(root_item, 0);
 591         btrfs_set_root_refs(root_item, 1);
 592         btrfs_set_root_used(root_item, leaf->len);
 593         btrfs_set_root_last_snapshot(root_item, 0);
 594
 595         btrfs_set_root_generation_v2(root_item,
 596                         btrfs_root_generation(root_item));
 597         generate_random_guid(root_item->uuid);
 598         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 599         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 600         root_item->ctime = root_item->otime;
 601         btrfs_set_root_ctransid(root_item, trans->transid);
 602         btrfs_set_root_otransid(root_item, trans->transid);
 603
 604         btrfs_tree_unlock(leaf);
 605
 606         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 607
 608         key.objectid = objectid;
 609         key.offset = 0;
 610         key.type = BTRFS_ROOT_ITEM_KEY;
 611         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 612                                 root_item);
 613         if (ret) {
 614                 /*
 615                  * Since we don't abort the transaction in this case, free the
 616                  * tree block so that we don't leak space and leave the
 617                  * filesystem in an inconsistent state (an extent item in the
 618                  * extent tree with a backreference for a root that does not
 619                  * exists).
 620                  */
 621                 btrfs_tree_lock(leaf);
 622                 btrfs_clean_tree_block(leaf);
 623                 btrfs_tree_unlock(leaf);
 624                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 625                 free_extent_buffer(leaf);
 626                 goto fail;
 627         }
 628
 629         free_extent_buffer(leaf);
 630         leaf = NULL;
 631
 632         key.offset = (u64)-1;
 633         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 634         if (IS_ERR(new_root)) {
 635                 free_anon_bdev(anon_dev);
 636                 ret = PTR_ERR(new_root);
 637                 btrfs_abort_transaction(trans, ret);
 638                 goto fail;
 639         }
 640         /* Freeing will be done in btrfs_put_root() of new_root */
 641         anon_dev = 0;
 642
 643         ret = btrfs_record_root_in_trans(trans, new_root);
 644         if (ret) {
 645                 btrfs_put_root(new_root);
 646                 btrfs_abort_transaction(trans, ret);
 647                 goto fail;
 648         }
 649
 650         ret = btrfs_create_subvol_root(trans, new_root, root, mnt_userns);
 651         btrfs_put_root(new_root);
 652         if (ret) {
 653                 /* We potentially lose an unused inode item here */
 654                 btrfs_abort_transaction(trans, ret);
 655                 goto fail;
 656         }
 657
 658         /*
 659          * insert the directory item
 660          */
 661         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 662         if (ret) {
 663                 btrfs_abort_transaction(trans, ret);
 664                 goto fail;
 665         }
 666
 667         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 668                                     BTRFS_FT_DIR, index);
 669         if (ret) {
 670                 btrfs_abort_transaction(trans, ret);
 671                 goto fail;
 672         }
 673
 674         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 675         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 676         if (ret) {
 677                 btrfs_abort_transaction(trans, ret);
 678                 goto fail;
 679         }
 680
 681         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 682                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 683         if (ret) {
 684                 btrfs_abort_transaction(trans, ret);
 685                 goto fail;
 686         }
 687
 688         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 689                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 690         if (ret)
 691                 btrfs_abort_transaction(trans, ret);
 692
 693 fail:
 694         kfree(root_item);
 695         trans->block_rsv = NULL;
 696         trans->bytes_reserved = 0;
 697         btrfs_subvolume_release_metadata(root, &block_rsv);
 698
 699         err = btrfs_commit_transaction(trans);
 700         if (err && !ret)
 701                 ret = err;
 702
 703         if (!ret) {
 704                 inode = btrfs_lookup_dentry(dir, dentry);
 705                 if (IS_ERR(inode))
 706                         return PTR_ERR(inode);
 707                 d_instantiate(dentry, inode);
 708         }
 709         return ret;
 710
 711 fail_free:
 712         if (anon_dev)
 713                 free_anon_bdev(anon_dev);
 714         kfree(root_item);
 715         return ret;
 716 }
 717
 718 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 719                            struct dentry *dentry, bool readonly,
 720                            struct btrfs_qgroup_inherit *inherit)
 721 {
 722         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 723         struct inode *inode;
 724         struct btrfs_pending_snapshot *pending_snapshot;
 725         struct btrfs_trans_handle *trans;
 726         int ret;
 727
 728         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 729                 return -EINVAL;
 730
 731         if (atomic_read(&root->nr_swapfiles)) {
 732                 btrfs_warn(fs_info,
 733                            "cannot snapshot subvolume with active swapfile");
 734                 return -ETXTBSY;
 735         }
 736
 737         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 738         if (!pending_snapshot)
 739                 return -ENOMEM;
 740
 741         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 742         if (ret < 0)
 743                 goto free_pending;
 744         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 745                         GFP_KERNEL);
 746         pending_snapshot->path = btrfs_alloc_path();
 747         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 748                 ret = -ENOMEM;
 749                 goto free_pending;
 750         }
 751
 752         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 753                              BTRFS_BLOCK_RSV_TEMP);
 754         /*
 755          * 1 - parent dir inode
 756          * 2 - dir entries
 757          * 1 - root item
 758          * 2 - root ref/backref
 759          * 1 - root of snapshot
 760          * 1 - UUID item
 761          */
 762         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 763                                         &pending_snapshot->block_rsv, 8,
 764                                         false);
 765         if (ret)
 766                 goto free_pending;
 767
 768         pending_snapshot->dentry = dentry;
 769         pending_snapshot->root = root;
 770         pending_snapshot->readonly = readonly;
 771         pending_snapshot->dir = dir;
 772         pending_snapshot->inherit = inherit;
 773
 774         trans = btrfs_start_transaction(root, 0);
 775         if (IS_ERR(trans)) {
 776                 ret = PTR_ERR(trans);
 777                 goto fail;
 778         }
 779
 780         trans->pending_snapshot = pending_snapshot;
 781
 782         ret = btrfs_commit_transaction(trans);
 783         if (ret)
 784                 goto fail;
 785
 786         ret = pending_snapshot->error;
 787         if (ret)
 788                 goto fail;
 789
 790         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 791         if (ret)
 792                 goto fail;
 793
 794         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 795         if (IS_ERR(inode)) {
 796                 ret = PTR_ERR(inode);
 797                 goto fail;
 798         }
 799
 800         d_instantiate(dentry, inode);
 801         ret = 0;
 802         pending_snapshot->anon_dev = 0;
 803 fail:
 804         /* Prevent double freeing of anon_dev */
 805         if (ret && pending_snapshot->snap)
 806                 pending_snapshot->snap->anon_dev = 0;
 807         btrfs_put_root(pending_snapshot->snap);
 808         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 809 free_pending:
 810         if (pending_snapshot->anon_dev)
 811                 free_anon_bdev(pending_snapshot->anon_dev);
 812         kfree(pending_snapshot->root_item);
 813         btrfs_free_path(pending_snapshot->path);
 814         kfree(pending_snapshot);
 815
 816         return ret;
 817 }
 818
 819 /*  copy of may_delete in fs/namei.c()
 820  *      Check whether we can remove a link victim from directory dir, check
 821  *  whether the type of victim is right.
 822  *  1. We can't do it if dir is read-only (done in permission())
 823  *  2. We should have write and exec permissions on dir
 824  *  3. We can't remove anything from append-only dir
 825  *  4. We can't do anything with immutable dir (done in permission())
 826  *  5. If the sticky bit on dir is set we should either
 827  *      a. be owner of dir, or
 828  *      b. be owner of victim, or
 829  *      c. have CAP_FOWNER capability
 830  *  6. If the victim is append-only or immutable we can't do anything with
 831  *     links pointing to it.
 832  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 833  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 834  *  9. We can't remove a root or mountpoint.
 835  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 836  *     nfs_async_unlink().
 837  */
 838
 839 static int btrfs_may_delete(struct user_namespace *mnt_userns,
 840                             struct inode *dir, struct dentry *victim, int isdir)
 841 {
 842         int error;
 843
 844         if (d_really_is_negative(victim))
 845                 return -ENOENT;
 846
 847         BUG_ON(d_inode(victim->d_parent) != dir);
 848         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 849
 850         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 851         if (error)
 852                 return error;
 853         if (IS_APPEND(dir))
 854                 return -EPERM;
 855         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
 856             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 857             IS_SWAPFILE(d_inode(victim)))
 858                 return -EPERM;
 859         if (isdir) {
 860                 if (!d_is_dir(victim))
 861                         return -ENOTDIR;
 862                 if (IS_ROOT(victim))
 863                         return -EBUSY;
 864         } else if (d_is_dir(victim))
 865                 return -EISDIR;
 866         if (IS_DEADDIR(dir))
 867                 return -ENOENT;
 868         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 869                 return -EBUSY;
 870         return 0;
 871 }
 872
 873 /* copy of may_create in fs/namei.c() */
 874 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
 875                                    struct inode *dir, struct dentry *child)
 876 {
 877         if (d_really_is_positive(child))
 878                 return -EEXIST;
 879         if (IS_DEADDIR(dir))
 880                 return -ENOENT;
 881         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
 882                 return -EOVERFLOW;
 883         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 884 }
 885
 886 /*
 887  * Create a new subvolume below @parent.  This is largely modeled after
 888  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 889  * inside this filesystem so it's quite a bit simpler.
 890  */
 891 static noinline int btrfs_mksubvol(const struct path *parent,
 892                                    struct user_namespace *mnt_userns,
 893                                    const char *name, int namelen,
 894                                    struct btrfs_root *snap_src,
 895                                    bool readonly,
 896                                    struct btrfs_qgroup_inherit *inherit)
 897 {
 898         struct inode *dir = d_inode(parent->dentry);
 899         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 900         struct dentry *dentry;
 901         int error;
 902
 903         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 904         if (error == -EINTR)
 905                 return error;
 906
 907         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
 908         error = PTR_ERR(dentry);
 909         if (IS_ERR(dentry))
 910                 goto out_unlock;
 911
 912         error = btrfs_may_create(mnt_userns, dir, dentry);
 913         if (error)
 914                 goto out_dput;
 915
 916         /*
 917          * even if this name doesn't exist, we may get hash collisions.
 918          * check for them now when we can safely fail
 919          */
 920         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 921                                                dir->i_ino, name,
 922                                                namelen);
 923         if (error)
 924                 goto out_dput;
 925
 926         down_read(&fs_info->subvol_sem);
 927
 928         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 929                 goto out_up_read;
 930
 931         if (snap_src)
 932                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 933         else
 934                 error = create_subvol(mnt_userns, dir, dentry, name, namelen, inherit);
 935
 936         if (!error)
 937                 fsnotify_mkdir(dir, dentry);
 938 out_up_read:
 939         up_read(&fs_info->subvol_sem);
 940 out_dput:
 941         dput(dentry);
 942 out_unlock:
 943         btrfs_inode_unlock(dir, 0);
 944         return error;
 945 }
 946
 947 static noinline int btrfs_mksnapshot(const struct path *parent,
 948                                    struct user_namespace *mnt_userns,
 949                                    const char *name, int namelen,
 950                                    struct btrfs_root *root,
 951                                    bool readonly,
 952                                    struct btrfs_qgroup_inherit *inherit)
 953 {
 954         int ret;
 955         bool snapshot_force_cow = false;
 956
 957         /*
 958          * Force new buffered writes to reserve space even when NOCOW is
 959          * possible. This is to avoid later writeback (running dealloc) to
 960          * fallback to COW mode and unexpectedly fail with ENOSPC.
 961          */
 962         btrfs_drew_read_lock(&root->snapshot_lock);
 963
 964         ret = btrfs_start_delalloc_snapshot(root, false);
 965         if (ret)
 966                 goto out;
 967
 968         /*
 969          * All previous writes have started writeback in NOCOW mode, so now
 970          * we force future writes to fallback to COW mode during snapshot
 971          * creation.
 972          */
 973         atomic_inc(&root->snapshot_force_cow);
 974         snapshot_force_cow = true;
 975
 976         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 977
 978         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
 979                              root, readonly, inherit);
 980 out:
 981         if (snapshot_force_cow)
 982                 atomic_dec(&root->snapshot_force_cow);
 983         btrfs_drew_read_unlock(&root->snapshot_lock);
 984         return ret;
 985 }
 986
 987 /*
 988  * When we're defragging a range, we don't want to kick it off again
 989  * if it is really just waiting for delalloc to send it down.
 990  * If we find a nice big extent or delalloc range for the bytes in the
 991  * file you want to defrag, we return 0 to let you know to skip this
 992  * part of the file
 993  */
 994 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 995 {
 996         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 997         struct extent_map *em = NULL;
 998         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 999         u64 end;
1000
1001         read_lock(&em_tree->lock);
1002         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1003         read_unlock(&em_tree->lock);
1004
1005         if (em) {
1006                 end = extent_map_end(em);
1007                 free_extent_map(em);
1008                 if (end - offset > thresh)
1009                         return 0;
1010         }
1011         /* if we already have a nice delalloc here, just stop */
1012         thresh /= 2;
1013         end = count_range_bits(io_tree, &offset, offset + thresh,
1014                                thresh, EXTENT_DELALLOC, 1);
1015         if (end >= thresh)
1016                 return 0;
1017         return 1;
1018 }
1019
1020 /*
1021  * helper function to walk through a file and find extents
1022  * newer than a specific transid, and smaller than thresh.
1023  *
1024  * This is used by the defragging code to find new and small
1025  * extents
1026  */
1027 static int find_new_extents(struct btrfs_root *root,
1028                             struct inode *inode, u64 newer_than,
1029                             u64 *off, u32 thresh)
1030 {
1031         struct btrfs_path *path;
1032         struct btrfs_key min_key;
1033         struct extent_buffer *leaf;
1034         struct btrfs_file_extent_item *extent;
1035         int type;
1036         int ret;
1037         u64 ino = btrfs_ino(BTRFS_I(inode));
1038
1039         path = btrfs_alloc_path();
1040         if (!path)
1041                 return -ENOMEM;
1042
1043         min_key.objectid = ino;
1044         min_key.type = BTRFS_EXTENT_DATA_KEY;
1045         min_key.offset = *off;
1046
1047         while (1) {
1048                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1049                 if (ret != 0)
1050                         goto none;
1051 process_slot:
1052                 if (min_key.objectid != ino)
1053                         goto none;
1054                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1055                         goto none;
1056
1057                 leaf = path->nodes[0];
1058                 extent = btrfs_item_ptr(leaf, path->slots[0],
1059                                         struct btrfs_file_extent_item);
1060
1061                 type = btrfs_file_extent_type(leaf, extent);
1062                 if (type == BTRFS_FILE_EXTENT_REG &&
1063                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1064                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1065                         *off = min_key.offset;
1066                         btrfs_free_path(path);
1067                         return 0;
1068                 }
1069
1070                 path->slots[0]++;
1071                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1072                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1073                         goto process_slot;
1074                 }
1075
1076                 if (min_key.offset == (u64)-1)
1077                         goto none;
1078
1079                 min_key.offset++;
1080                 btrfs_release_path(path);
1081         }
1082 none:
1083         btrfs_free_path(path);
1084         return -ENOENT;
1085 }
1086
1087 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1088 {
1089         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1090         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1091         struct extent_map *em;
1092         u64 len = PAGE_SIZE;
1093
1094         /*
1095          * hopefully we have this extent in the tree already, try without
1096          * the full extent lock
1097          */
1098         read_lock(&em_tree->lock);
1099         em = lookup_extent_mapping(em_tree, start, len);
1100         read_unlock(&em_tree->lock);
1101
1102         if (!em) {
1103                 struct extent_state *cached = NULL;
1104                 u64 end = start + len - 1;
1105
1106                 /* get the big lock and read metadata off disk */
1107                 lock_extent_bits(io_tree, start, end, &cached);
1108                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1109                 unlock_extent_cached(io_tree, start, end, &cached);
1110
1111                 if (IS_ERR(em))
1112                         return NULL;
1113         }
1114
1115         return em;
1116 }
1117
1118 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1119 {
1120         struct extent_map *next;
1121         bool ret = true;
1122
1123         /* this is the last extent */
1124         if (em->start + em->len >= i_size_read(inode))
1125                 return false;
1126
1127         next = defrag_lookup_extent(inode, em->start + em->len);
1128         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1129                 ret = false;
1130         else if ((em->block_start + em->block_len == next->block_start) &&
1131                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1132                 ret = false;
1133
1134         free_extent_map(next);
1135         return ret;
1136 }
1137
1138 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1139                                u64 *last_len, u64 *skip, u64 *defrag_end,
1140                                int compress)
1141 {
1142         struct extent_map *em;
1143         int ret = 1;
1144         bool next_mergeable = true;
1145         bool prev_mergeable = true;
1146
1147         /*
1148          * make sure that once we start defragging an extent, we keep on
1149          * defragging it
1150          */
1151         if (start < *defrag_end)
1152                 return 1;
1153
1154         *skip = 0;
1155
1156         em = defrag_lookup_extent(inode, start);
1157         if (!em)
1158                 return 0;
1159
1160         /* this will cover holes, and inline extents */
1161         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1162                 ret = 0;
1163                 goto out;
1164         }
1165
1166         if (!*defrag_end)
1167                 prev_mergeable = false;
1168
1169         next_mergeable = defrag_check_next_extent(inode, em);
1170         /*
1171          * we hit a real extent, if it is big or the next extent is not a
1172          * real extent, don't bother defragging it
1173          */
1174         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1175             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1176                 ret = 0;
1177 out:
1178         /*
1179          * last_len ends up being a counter of how many bytes we've defragged.
1180          * every time we choose not to defrag an extent, we reset *last_len
1181          * so that the next tiny extent will force a defrag.
1182          *
1183          * The end result of this is that tiny extents before a single big
1184          * extent will force at least part of that big extent to be defragged.
1185          */
1186         if (ret) {
1187                 *defrag_end = extent_map_end(em);
1188         } else {
1189                 *last_len = 0;
1190                 *skip = extent_map_end(em);
1191                 *defrag_end = 0;
1192         }
1193
1194         free_extent_map(em);
1195         return ret;
1196 }
1197
1198 /*
1199  * it doesn't do much good to defrag one or two pages
1200  * at a time.  This pulls in a nice chunk of pages
1201  * to COW and defrag.
1202  *
1203  * It also makes sure the delalloc code has enough
1204  * dirty data to avoid making new small extents as part
1205  * of the defrag
1206  *
1207  * It's a good idea to start RA on this range
1208  * before calling this.
1209  */
1210 static int cluster_pages_for_defrag(struct inode *inode,
1211                                     struct page **pages,
1212                                     unsigned long start_index,
1213                                     unsigned long num_pages)
1214 {
1215         unsigned long file_end;
1216         u64 isize = i_size_read(inode);
1217         u64 page_start;
1218         u64 page_end;
1219         u64 page_cnt;
1220         u64 start = (u64)start_index << PAGE_SHIFT;
1221         u64 search_start;
1222         int ret;
1223         int i;
1224         int i_done;
1225         struct btrfs_ordered_extent *ordered;
1226         struct extent_state *cached_state = NULL;
1227         struct extent_io_tree *tree;
1228         struct extent_changeset *data_reserved = NULL;
1229         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1230
1231         file_end = (isize - 1) >> PAGE_SHIFT;
1232         if (!isize || start_index > file_end)
1233                 return 0;
1234
1235         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1236
1237         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1238                         start, page_cnt << PAGE_SHIFT);
1239         if (ret)
1240                 return ret;
1241         i_done = 0;
1242         tree = &BTRFS_I(inode)->io_tree;
1243
1244         /* step one, lock all the pages */
1245         for (i = 0; i < page_cnt; i++) {
1246                 struct page *page;
1247 again:
1248                 page = find_or_create_page(inode->i_mapping,
1249                                            start_index + i, mask);
1250                 if (!page)
1251                         break;
1252
1253                 ret = set_page_extent_mapped(page);
1254                 if (ret < 0) {
1255                         unlock_page(page);
1256                         put_page(page);
1257                         break;
1258                 }
1259
1260                 page_start = page_offset(page);
1261                 page_end = page_start + PAGE_SIZE - 1;
1262                 while (1) {
1263                         lock_extent_bits(tree, page_start, page_end,
1264                                          &cached_state);
1265                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1266                                                               page_start);
1267                         unlock_extent_cached(tree, page_start, page_end,
1268                                              &cached_state);
1269                         if (!ordered)
1270                                 break;
1271
1272                         unlock_page(page);
1273                         btrfs_start_ordered_extent(ordered, 1);
1274                         btrfs_put_ordered_extent(ordered);
1275                         lock_page(page);
1276                         /*
1277                          * we unlocked the page above, so we need check if
1278                          * it was released or not.
1279                          */
1280                         if (page->mapping != inode->i_mapping) {
1281                                 unlock_page(page);
1282                                 put_page(page);
1283                                 goto again;
1284                         }
1285                 }
1286
1287                 if (!PageUptodate(page)) {
1288                         btrfs_readpage(NULL, page);
1289                         lock_page(page);
1290                         if (!PageUptodate(page)) {
1291                                 unlock_page(page);
1292                                 put_page(page);
1293                                 ret = -EIO;
1294                                 break;
1295                         }
1296                 }
1297
1298                 if (page->mapping != inode->i_mapping) {
1299                         unlock_page(page);
1300                         put_page(page);
1301                         goto again;
1302                 }
1303
1304                 pages[i] = page;
1305                 i_done++;
1306         }
1307         if (!i_done || ret)
1308                 goto out;
1309
1310         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1311                 goto out;
1312
1313         /*
1314          * so now we have a nice long stream of locked
1315          * and up to date pages, lets wait on them
1316          */
1317         for (i = 0; i < i_done; i++)
1318                 wait_on_page_writeback(pages[i]);
1319
1320         page_start = page_offset(pages[0]);
1321         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1322
1323         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1324                          page_start, page_end - 1, &cached_state);
1325
1326         /*
1327          * When defragmenting we skip ranges that have holes or inline extents,
1328          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1329          * space. At btrfs_defrag_file(), we check if a range should be defragged
1330          * before locking the inode and then, if it should, we trigger a sync
1331          * page cache readahead - we lock the inode only after that to avoid
1332          * blocking for too long other tasks that possibly want to operate on
1333          * other file ranges. But before we were able to get the inode lock,
1334          * some other task may have punched a hole in the range, or we may have
1335          * now an inline extent, in which case we should not defrag. So check
1336          * for that here, where we have the inode and the range locked, and bail
1337          * out if that happened.
1338          */
1339         search_start = page_start;
1340         while (search_start < page_end) {
1341                 struct extent_map *em;
1342
1343                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1344                                       page_end - search_start);
1345                 if (IS_ERR(em)) {
1346                         ret = PTR_ERR(em);
1347                         goto out_unlock_range;
1348                 }
1349                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1350                         free_extent_map(em);
1351                         /* Ok, 0 means we did not defrag anything */
1352                         ret = 0;
1353                         goto out_unlock_range;
1354                 }
1355                 search_start = extent_map_end(em);
1356                 free_extent_map(em);
1357         }
1358
1359         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1360                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1361                           EXTENT_DEFRAG, 0, 0, &cached_state);
1362
1363         if (i_done != page_cnt) {
1364                 spin_lock(&BTRFS_I(inode)->lock);
1365                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1366                 spin_unlock(&BTRFS_I(inode)->lock);
1367                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1368                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1369         }
1370
1371
1372         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1373                           &cached_state);
1374
1375         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1376                              page_start, page_end - 1, &cached_state);
1377
1378         for (i = 0; i < i_done; i++) {
1379                 clear_page_dirty_for_io(pages[i]);
1380                 ClearPageChecked(pages[i]);
1381                 set_page_dirty(pages[i]);
1382                 unlock_page(pages[i]);
1383                 put_page(pages[i]);
1384         }
1385         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1386         extent_changeset_free(data_reserved);
1387         return i_done;
1388
1389 out_unlock_range:
1390         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1391                              page_start, page_end - 1, &cached_state);
1392 out:
1393         for (i = 0; i < i_done; i++) {
1394                 unlock_page(pages[i]);
1395                 put_page(pages[i]);
1396         }
1397         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1398                         start, page_cnt << PAGE_SHIFT, true);
1399         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1400         extent_changeset_free(data_reserved);
1401         return ret;
1402
1403 }
1404
1405 int btrfs_defrag_file(struct inode *inode, struct file *file,
1406                       struct btrfs_ioctl_defrag_range_args *range,
1407                       u64 newer_than, unsigned long max_to_defrag)
1408 {
1409         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1410         struct btrfs_root *root = BTRFS_I(inode)->root;
1411         struct file_ra_state *ra = NULL;
1412         unsigned long last_index;
1413         u64 isize = i_size_read(inode);
1414         u64 last_len = 0;
1415         u64 skip = 0;
1416         u64 defrag_end = 0;
1417         u64 newer_off = range->start;
1418         unsigned long i;
1419         unsigned long ra_index = 0;
1420         int ret;
1421         int defrag_count = 0;
1422         int compress_type = BTRFS_COMPRESS_ZLIB;
1423         u32 extent_thresh = range->extent_thresh;
1424         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1425         unsigned long cluster = max_cluster;
1426         u64 new_align = ~((u64)SZ_128K - 1);
1427         struct page **pages = NULL;
1428         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1429
1430         if (isize == 0)
1431                 return 0;
1432
1433         if (range->start >= isize)
1434                 return -EINVAL;
1435
1436         if (do_compress) {
1437                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1438                         return -EINVAL;
1439                 if (range->compress_type)
1440                         compress_type = range->compress_type;
1441         }
1442
1443         if (extent_thresh == 0)
1444                 extent_thresh = SZ_256K;
1445
1446         /*
1447          * If we were not given a file, allocate a readahead context. As
1448          * readahead is just an optimization, defrag will work without it so
1449          * we don't error out.
1450          */
1451         if (!file) {
1452                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1453                 if (ra)
1454                         file_ra_state_init(ra, inode->i_mapping);
1455         } else {
1456                 ra = &file->f_ra;
1457         }
1458
1459         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1460         if (!pages) {
1461                 ret = -ENOMEM;
1462                 goto out_ra;
1463         }
1464
1465         /* find the last page to defrag */
1466         if (range->start + range->len > range->start) {
1467                 last_index = min_t(u64, isize - 1,
1468                          range->start + range->len - 1) >> PAGE_SHIFT;
1469         } else {
1470                 last_index = (isize - 1) >> PAGE_SHIFT;
1471         }
1472
1473         if (newer_than) {
1474                 ret = find_new_extents(root, inode, newer_than,
1475                                        &newer_off, SZ_64K);
1476                 if (!ret) {
1477                         range->start = newer_off;
1478                         /*
1479                          * we always align our defrag to help keep
1480                          * the extents in the file evenly spaced
1481                          */
1482                         i = (newer_off & new_align) >> PAGE_SHIFT;
1483                 } else
1484                         goto out_ra;
1485         } else {
1486                 i = range->start >> PAGE_SHIFT;
1487         }
1488         if (!max_to_defrag)
1489                 max_to_defrag = last_index - i + 1;
1490
1491         /*
1492          * make writeback starts from i, so the defrag range can be
1493          * written sequentially.
1494          */
1495         if (i < inode->i_mapping->writeback_index)
1496                 inode->i_mapping->writeback_index = i;
1497
1498         while (i <= last_index && defrag_count < max_to_defrag &&
1499                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1500                 /*
1501                  * make sure we stop running if someone unmounts
1502                  * the FS
1503                  */
1504                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1505                         break;
1506
1507                 if (btrfs_defrag_cancelled(fs_info)) {
1508                         btrfs_debug(fs_info, "defrag_file cancelled");
1509                         ret = -EAGAIN;
1510                         goto error;
1511                 }
1512
1513                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1514                                          extent_thresh, &last_len, &skip,
1515                                          &defrag_end, do_compress)){
1516                         unsigned long next;
1517                         /*
1518                          * the should_defrag function tells us how much to skip
1519                          * bump our counter by the suggested amount
1520                          */
1521                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1522                         i = max(i + 1, next);
1523                         continue;
1524                 }
1525
1526                 if (!newer_than) {
1527                         cluster = (PAGE_ALIGN(defrag_end) >>
1528                                    PAGE_SHIFT) - i;
1529                         cluster = min(cluster, max_cluster);
1530                 } else {
1531                         cluster = max_cluster;
1532                 }
1533
1534                 if (i + cluster > ra_index) {
1535                         ra_index = max(i, ra_index);
1536                         if (ra)
1537                                 page_cache_sync_readahead(inode->i_mapping, ra,
1538                                                 file, ra_index, cluster);
1539                         ra_index += cluster;
1540                 }
1541
1542                 btrfs_inode_lock(inode, 0);
1543                 if (IS_SWAPFILE(inode)) {
1544                         ret = -ETXTBSY;
1545                 } else {
1546                         if (do_compress)
1547                                 BTRFS_I(inode)->defrag_compress = compress_type;
1548                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1549                 }
1550                 if (ret < 0) {
1551                         btrfs_inode_unlock(inode, 0);
1552                         goto out_ra;
1553                 }
1554
1555                 defrag_count += ret;
1556                 balance_dirty_pages_ratelimited(inode->i_mapping);
1557                 btrfs_inode_unlock(inode, 0);
1558
1559                 if (newer_than) {
1560                         if (newer_off == (u64)-1)
1561                                 break;
1562
1563                         if (ret > 0)
1564                                 i += ret;
1565
1566                         newer_off = max(newer_off + 1,
1567                                         (u64)i << PAGE_SHIFT);
1568
1569                         ret = find_new_extents(root, inode, newer_than,
1570                                                &newer_off, SZ_64K);
1571                         if (!ret) {
1572                                 range->start = newer_off;
1573                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1574                         } else {
1575                                 break;
1576                         }
1577                 } else {
1578                         if (ret > 0) {
1579                                 i += ret;
1580                                 last_len += ret << PAGE_SHIFT;
1581                         } else {
1582                                 i++;
1583                                 last_len = 0;
1584                         }
1585                 }
1586         }
1587
1588         ret = defrag_count;
1589 error:
1590         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1591                 filemap_flush(inode->i_mapping);
1592                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1593                              &BTRFS_I(inode)->runtime_flags))
1594                         filemap_flush(inode->i_mapping);
1595         }
1596
1597         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1598                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1599         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1600                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1601         }
1602
1603 out_ra:
1604         if (do_compress) {
1605                 btrfs_inode_lock(inode, 0);
1606                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1607                 btrfs_inode_unlock(inode, 0);
1608         }
1609         if (!file)
1610                 kfree(ra);
1611         kfree(pages);
1612         return ret;
1613 }
1614
1615 /*
1616  * Try to start exclusive operation @type or cancel it if it's running.
1617  *
1618  * Return:
1619  *   0        - normal mode, newly claimed op started
1620  *  >0        - normal mode, something else is running,
1621  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1622  * ECANCELED  - cancel mode, successful cancel
1623  * ENOTCONN   - cancel mode, operation not running anymore
1624  */
1625 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1626                         enum btrfs_exclusive_operation type, bool cancel)
1627 {
1628         if (!cancel) {
1629                 /* Start normal op */
1630                 if (!btrfs_exclop_start(fs_info, type))
1631                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1632                 /* Exclusive operation is now claimed */
1633                 return 0;
1634         }
1635
1636         /* Cancel running op */
1637         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1638                 /*
1639                  * This blocks any exclop finish from setting it to NONE, so we
1640                  * request cancellation. Either it runs and we will wait for it,
1641                  * or it has finished and no waiting will happen.
1642                  */
1643                 atomic_inc(&fs_info->reloc_cancel_req);
1644                 btrfs_exclop_start_unlock(fs_info);
1645
1646                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1647                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1648                                     TASK_INTERRUPTIBLE);
1649
1650                 return -ECANCELED;
1651         }
1652
1653         /* Something else is running or none */
1654         return -ENOTCONN;
1655 }
1656
1657 static noinline int btrfs_ioctl_resize(struct file *file,
1658                                         void __user *arg)
1659 {
1660         BTRFS_DEV_LOOKUP_ARGS(args);
1661         struct inode *inode = file_inode(file);
1662         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1663         u64 new_size;
1664         u64 old_size;
1665         u64 devid = 1;
1666         struct btrfs_root *root = BTRFS_I(inode)->root;
1667         struct btrfs_ioctl_vol_args *vol_args;
1668         struct btrfs_trans_handle *trans;
1669         struct btrfs_device *device = NULL;
1670         char *sizestr;
1671         char *retptr;
1672         char *devstr = NULL;
1673         int ret = 0;
1674         int mod = 0;
1675         bool cancel;
1676
1677         if (!capable(CAP_SYS_ADMIN))
1678                 return -EPERM;
1679
1680         ret = mnt_want_write_file(file);
1681         if (ret)
1682                 return ret;
1683
1684         /*
1685          * Read the arguments before checking exclusivity to be able to
1686          * distinguish regular resize and cancel
1687          */
1688         vol_args = memdup_user(arg, sizeof(*vol_args));
1689         if (IS_ERR(vol_args)) {
1690                 ret = PTR_ERR(vol_args);
1691                 goto out_drop;
1692         }
1693         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1694         sizestr = vol_args->name;
1695         cancel = (strcmp("cancel", sizestr) == 0);
1696         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1697         if (ret)
1698                 goto out_free;
1699         /* Exclusive operation is now claimed */
1700
1701         devstr = strchr(sizestr, ':');
1702         if (devstr) {
1703                 sizestr = devstr + 1;
1704                 *devstr = '\0';
1705                 devstr = vol_args->name;
1706                 ret = kstrtoull(devstr, 10, &devid);
1707                 if (ret)
1708                         goto out_finish;
1709                 if (!devid) {
1710                         ret = -EINVAL;
1711                         goto out_finish;
1712                 }
1713                 btrfs_info(fs_info, "resizing devid %llu", devid);
1714         }
1715
1716         args.devid = devid;
1717         device = btrfs_find_device(fs_info->fs_devices, &args);
1718         if (!device) {
1719                 btrfs_info(fs_info, "resizer unable to find device %llu",
1720                            devid);
1721                 ret = -ENODEV;
1722                 goto out_finish;
1723         }
1724
1725         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1726                 btrfs_info(fs_info,
1727                            "resizer unable to apply on readonly device %llu",
1728                        devid);
1729                 ret = -EPERM;
1730                 goto out_finish;
1731         }
1732
1733         if (!strcmp(sizestr, "max"))
1734                 new_size = device->bdev->bd_inode->i_size;
1735         else {
1736                 if (sizestr[0] == '-') {
1737                         mod = -1;
1738                         sizestr++;
1739                 } else if (sizestr[0] == '+') {
1740                         mod = 1;
1741                         sizestr++;
1742                 }
1743                 new_size = memparse(sizestr, &retptr);
1744                 if (*retptr != '\0' || new_size == 0) {
1745                         ret = -EINVAL;
1746                         goto out_finish;
1747                 }
1748         }
1749
1750         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1751                 ret = -EPERM;
1752                 goto out_finish;
1753         }
1754
1755         old_size = btrfs_device_get_total_bytes(device);
1756
1757         if (mod < 0) {
1758                 if (new_size > old_size) {
1759                         ret = -EINVAL;
1760                         goto out_finish;
1761                 }
1762                 new_size = old_size - new_size;
1763         } else if (mod > 0) {
1764                 if (new_size > ULLONG_MAX - old_size) {
1765                         ret = -ERANGE;
1766                         goto out_finish;
1767                 }
1768                 new_size = old_size + new_size;
1769         }
1770
1771         if (new_size < SZ_256M) {
1772                 ret = -EINVAL;
1773                 goto out_finish;
1774         }
1775         if (new_size > device->bdev->bd_inode->i_size) {
1776                 ret = -EFBIG;
1777                 goto out_finish;
1778         }
1779
1780         new_size = round_down(new_size, fs_info->sectorsize);
1781
1782         if (new_size > old_size) {
1783                 trans = btrfs_start_transaction(root, 0);
1784                 if (IS_ERR(trans)) {
1785                         ret = PTR_ERR(trans);
1786                         goto out_finish;
1787                 }
1788                 ret = btrfs_grow_device(trans, device, new_size);
1789                 btrfs_commit_transaction(trans);
1790         } else if (new_size < old_size) {
1791                 ret = btrfs_shrink_device(device, new_size);
1792         } /* equal, nothing need to do */
1793
1794         if (ret == 0 && new_size != old_size)
1795                 btrfs_info_in_rcu(fs_info,
1796                         "resize device %s (devid %llu) from %llu to %llu",
1797                         rcu_str_deref(device->name), device->devid,
1798                         old_size, new_size);
1799 out_finish:
1800         btrfs_exclop_finish(fs_info);
1801 out_free:
1802         kfree(vol_args);
1803 out_drop:
1804         mnt_drop_write_file(file);
1805         return ret;
1806 }
1807
1808 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1809                                 struct user_namespace *mnt_userns,
1810                                 const char *name, unsigned long fd, int subvol,
1811                                 bool readonly,
1812                                 struct btrfs_qgroup_inherit *inherit)
1813 {
1814         int namelen;
1815         int ret = 0;
1816
1817         if (!S_ISDIR(file_inode(file)->i_mode))
1818                 return -ENOTDIR;
1819
1820         ret = mnt_want_write_file(file);
1821         if (ret)
1822                 goto out;
1823
1824         namelen = strlen(name);
1825         if (strchr(name, '/')) {
1826                 ret = -EINVAL;
1827                 goto out_drop_write;
1828         }
1829
1830         if (name[0] == '.' &&
1831            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1832                 ret = -EEXIST;
1833                 goto out_drop_write;
1834         }
1835
1836         if (subvol) {
1837                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
1838                                      namelen, NULL, readonly, inherit);
1839         } else {
1840                 struct fd src = fdget(fd);
1841                 struct inode *src_inode;
1842                 if (!src.file) {
1843                         ret = -EINVAL;
1844                         goto out_drop_write;
1845                 }
1846
1847                 src_inode = file_inode(src.file);
1848                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1849                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1850                                    "Snapshot src from another FS");
1851                         ret = -EXDEV;
1852                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
1853                         /*
1854                          * Subvolume creation is not restricted, but snapshots
1855                          * are limited to own subvolumes only
1856                          */
1857                         ret = -EPERM;
1858                 } else {
1859                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
1860                                                name, namelen,
1861                                                BTRFS_I(src_inode)->root,
1862                                                readonly, inherit);
1863                 }
1864                 fdput(src);
1865         }
1866 out_drop_write:
1867         mnt_drop_write_file(file);
1868 out:
1869         return ret;
1870 }
1871
1872 static noinline int btrfs_ioctl_snap_create(struct file *file,
1873                                             void __user *arg, int subvol)
1874 {
1875         struct btrfs_ioctl_vol_args *vol_args;
1876         int ret;
1877
1878         if (!S_ISDIR(file_inode(file)->i_mode))
1879                 return -ENOTDIR;
1880
1881         vol_args = memdup_user(arg, sizeof(*vol_args));
1882         if (IS_ERR(vol_args))
1883                 return PTR_ERR(vol_args);
1884         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1885
1886         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
1887                                         vol_args->name, vol_args->fd, subvol,
1888                                         false, NULL);
1889
1890         kfree(vol_args);
1891         return ret;
1892 }
1893
1894 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1895                                                void __user *arg, int subvol)
1896 {
1897         struct btrfs_ioctl_vol_args_v2 *vol_args;
1898         int ret;
1899         bool readonly = false;
1900         struct btrfs_qgroup_inherit *inherit = NULL;
1901
1902         if (!S_ISDIR(file_inode(file)->i_mode))
1903                 return -ENOTDIR;
1904
1905         vol_args = memdup_user(arg, sizeof(*vol_args));
1906         if (IS_ERR(vol_args))
1907                 return PTR_ERR(vol_args);
1908         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1909
1910         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1911                 ret = -EOPNOTSUPP;
1912                 goto free_args;
1913         }
1914
1915         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1916                 readonly = true;
1917         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1918                 u64 nums;
1919
1920                 if (vol_args->size < sizeof(*inherit) ||
1921                     vol_args->size > PAGE_SIZE) {
1922                         ret = -EINVAL;
1923                         goto free_args;
1924                 }
1925                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1926                 if (IS_ERR(inherit)) {
1927                         ret = PTR_ERR(inherit);
1928                         goto free_args;
1929                 }
1930
1931                 if (inherit->num_qgroups > PAGE_SIZE ||
1932                     inherit->num_ref_copies > PAGE_SIZE ||
1933                     inherit->num_excl_copies > PAGE_SIZE) {
1934                         ret = -EINVAL;
1935                         goto free_inherit;
1936                 }
1937
1938                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1939                        2 * inherit->num_excl_copies;
1940                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1941                         ret = -EINVAL;
1942                         goto free_inherit;
1943                 }
1944         }
1945
1946         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
1947                                         vol_args->name, vol_args->fd, subvol,
1948                                         readonly, inherit);
1949         if (ret)
1950                 goto free_inherit;
1951 free_inherit:
1952         kfree(inherit);
1953 free_args:
1954         kfree(vol_args);
1955         return ret;
1956 }
1957
1958 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1959                                                 void __user *arg)
1960 {
1961         struct inode *inode = file_inode(file);
1962         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1963         struct btrfs_root *root = BTRFS_I(inode)->root;
1964         int ret = 0;
1965         u64 flags = 0;
1966
1967         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1968                 return -EINVAL;
1969
1970         down_read(&fs_info->subvol_sem);
1971         if (btrfs_root_readonly(root))
1972                 flags |= BTRFS_SUBVOL_RDONLY;
1973         up_read(&fs_info->subvol_sem);
1974
1975         if (copy_to_user(arg, &flags, sizeof(flags)))
1976                 ret = -EFAULT;
1977
1978         return ret;
1979 }
1980
1981 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1982                                               void __user *arg)
1983 {
1984         struct inode *inode = file_inode(file);
1985         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1986         struct btrfs_root *root = BTRFS_I(inode)->root;
1987         struct btrfs_trans_handle *trans;
1988         u64 root_flags;
1989         u64 flags;
1990         int ret = 0;
1991
1992         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
1993                 return -EPERM;
1994
1995         ret = mnt_want_write_file(file);
1996         if (ret)
1997                 goto out;
1998
1999         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2000                 ret = -EINVAL;
2001                 goto out_drop_write;
2002         }
2003
2004         if (copy_from_user(&flags, arg, sizeof(flags))) {
2005                 ret = -EFAULT;
2006                 goto out_drop_write;
2007         }
2008
2009         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2010                 ret = -EOPNOTSUPP;
2011                 goto out_drop_write;
2012         }
2013
2014         down_write(&fs_info->subvol_sem);
2015
2016         /* nothing to do */
2017         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2018                 goto out_drop_sem;
2019
2020         root_flags = btrfs_root_flags(&root->root_item);
2021         if (flags & BTRFS_SUBVOL_RDONLY) {
2022                 btrfs_set_root_flags(&root->root_item,
2023                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2024         } else {
2025                 /*
2026                  * Block RO -> RW transition if this subvolume is involved in
2027                  * send
2028                  */
2029                 spin_lock(&root->root_item_lock);
2030                 if (root->send_in_progress == 0) {
2031                         btrfs_set_root_flags(&root->root_item,
2032                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2033                         spin_unlock(&root->root_item_lock);
2034                 } else {
2035                         spin_unlock(&root->root_item_lock);
2036                         btrfs_warn(fs_info,
2037                                    "Attempt to set subvolume %llu read-write during send",
2038                                    root->root_key.objectid);
2039                         ret = -EPERM;
2040                         goto out_drop_sem;
2041                 }
2042         }
2043
2044         trans = btrfs_start_transaction(root, 1);
2045         if (IS_ERR(trans)) {
2046                 ret = PTR_ERR(trans);
2047                 goto out_reset;
2048         }
2049
2050         ret = btrfs_update_root(trans, fs_info->tree_root,
2051                                 &root->root_key, &root->root_item);
2052         if (ret < 0) {
2053                 btrfs_end_transaction(trans);
2054                 goto out_reset;
2055         }
2056
2057         ret = btrfs_commit_transaction(trans);
2058
2059 out_reset:
2060         if (ret)
2061                 btrfs_set_root_flags(&root->root_item, root_flags);
2062 out_drop_sem:
2063         up_write(&fs_info->subvol_sem);
2064 out_drop_write:
2065         mnt_drop_write_file(file);
2066 out:
2067         return ret;
2068 }
2069
2070 static noinline int key_in_sk(struct btrfs_key *key,
2071                               struct btrfs_ioctl_search_key *sk)
2072 {
2073         struct btrfs_key test;
2074         int ret;
2075
2076         test.objectid = sk->min_objectid;
2077         test.type = sk->min_type;
2078         test.offset = sk->min_offset;
2079
2080         ret = btrfs_comp_cpu_keys(key, &test);
2081         if (ret < 0)
2082                 return 0;
2083
2084         test.objectid = sk->max_objectid;
2085         test.type = sk->max_type;
2086         test.offset = sk->max_offset;
2087
2088         ret = btrfs_comp_cpu_keys(key, &test);
2089         if (ret > 0)
2090                 return 0;
2091         return 1;
2092 }
2093
2094 static noinline int copy_to_sk(struct btrfs_path *path,
2095                                struct btrfs_key *key,
2096                                struct btrfs_ioctl_search_key *sk,
2097                                size_t *buf_size,
2098                                char __user *ubuf,
2099                                unsigned long *sk_offset,
2100                                int *num_found)
2101 {
2102         u64 found_transid;
2103         struct extent_buffer *leaf;
2104         struct btrfs_ioctl_search_header sh;
2105         struct btrfs_key test;
2106         unsigned long item_off;
2107         unsigned long item_len;
2108         int nritems;
2109         int i;
2110         int slot;
2111         int ret = 0;
2112
2113         leaf = path->nodes[0];
2114         slot = path->slots[0];
2115         nritems = btrfs_header_nritems(leaf);
2116
2117         if (btrfs_header_generation(leaf) > sk->max_transid) {
2118                 i = nritems;
2119                 goto advance_key;
2120         }
2121         found_transid = btrfs_header_generation(leaf);
2122
2123         for (i = slot; i < nritems; i++) {
2124                 item_off = btrfs_item_ptr_offset(leaf, i);
2125                 item_len = btrfs_item_size_nr(leaf, i);
2126
2127                 btrfs_item_key_to_cpu(leaf, key, i);
2128                 if (!key_in_sk(key, sk))
2129                         continue;
2130
2131                 if (sizeof(sh) + item_len > *buf_size) {
2132                         if (*num_found) {
2133                                 ret = 1;
2134                                 goto out;
2135                         }
2136
2137                         /*
2138                          * return one empty item back for v1, which does not
2139                          * handle -EOVERFLOW
2140                          */
2141
2142                         *buf_size = sizeof(sh) + item_len;
2143                         item_len = 0;
2144                         ret = -EOVERFLOW;
2145                 }
2146
2147                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2148                         ret = 1;
2149                         goto out;
2150                 }
2151
2152                 sh.objectid = key->objectid;
2153                 sh.offset = key->offset;
2154                 sh.type = key->type;
2155                 sh.len = item_len;
2156                 sh.transid = found_transid;
2157
2158                 /*
2159                  * Copy search result header. If we fault then loop again so we
2160                  * can fault in the pages and -EFAULT there if there's a
2161                  * problem. Otherwise we'll fault and then copy the buffer in
2162                  * properly this next time through
2163                  */
2164                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2165                         ret = 0;
2166                         goto out;
2167                 }
2168
2169                 *sk_offset += sizeof(sh);
2170
2171                 if (item_len) {
2172                         char __user *up = ubuf + *sk_offset;
2173                         /*
2174                          * Copy the item, same behavior as above, but reset the
2175                          * * sk_offset so we copy the full thing again.
2176                          */
2177                         if (read_extent_buffer_to_user_nofault(leaf, up,
2178                                                 item_off, item_len)) {
2179                                 ret = 0;
2180                                 *sk_offset -= sizeof(sh);
2181                                 goto out;
2182                         }
2183
2184                         *sk_offset += item_len;
2185                 }
2186                 (*num_found)++;
2187
2188                 if (ret) /* -EOVERFLOW from above */
2189                         goto out;
2190
2191                 if (*num_found >= sk->nr_items) {
2192                         ret = 1;
2193                         goto out;
2194                 }
2195         }
2196 advance_key:
2197         ret = 0;
2198         test.objectid = sk->max_objectid;
2199         test.type = sk->max_type;
2200         test.offset = sk->max_offset;
2201         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2202                 ret = 1;
2203         else if (key->offset < (u64)-1)
2204                 key->offset++;
2205         else if (key->type < (u8)-1) {
2206                 key->offset = 0;
2207                 key->type++;
2208         } else if (key->objectid < (u64)-1) {
2209                 key->offset = 0;
2210                 key->type = 0;
2211                 key->objectid++;
2212         } else
2213                 ret = 1;
2214 out:
2215         /*
2216          *  0: all items from this leaf copied, continue with next
2217          *  1: * more items can be copied, but unused buffer is too small
2218          *     * all items were found
2219          *     Either way, it will stops the loop which iterates to the next
2220          *     leaf
2221          *  -EOVERFLOW: item was to large for buffer
2222          *  -EFAULT: could not copy extent buffer back to userspace
2223          */
2224         return ret;
2225 }
2226
2227 static noinline int search_ioctl(struct inode *inode,
2228                                  struct btrfs_ioctl_search_key *sk,
2229                                  size_t *buf_size,
2230                                  char __user *ubuf)
2231 {
2232         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2233         struct btrfs_root *root;
2234         struct btrfs_key key;
2235         struct btrfs_path *path;
2236         int ret;
2237         int num_found = 0;
2238         unsigned long sk_offset = 0;
2239
2240         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2241                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2242                 return -EOVERFLOW;
2243         }
2244
2245         path = btrfs_alloc_path();
2246         if (!path)
2247                 return -ENOMEM;
2248
2249         if (sk->tree_id == 0) {
2250                 /* search the root of the inode that was passed */
2251                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2252         } else {
2253                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2254                 if (IS_ERR(root)) {
2255                         btrfs_free_path(path);
2256                         return PTR_ERR(root);
2257                 }
2258         }
2259
2260         key.objectid = sk->min_objectid;
2261         key.type = sk->min_type;
2262         key.offset = sk->min_offset;
2263
2264         while (1) {
2265                 ret = -EFAULT;
2266                 if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset))
2267                         break;
2268
2269                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2270                 if (ret != 0) {
2271                         if (ret > 0)
2272                                 ret = 0;
2273                         goto err;
2274                 }
2275                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2276                                  &sk_offset, &num_found);
2277                 btrfs_release_path(path);
2278                 if (ret)
2279                         break;
2280
2281         }
2282         if (ret > 0)
2283                 ret = 0;
2284 err:
2285         sk->nr_items = num_found;
2286         btrfs_put_root(root);
2287         btrfs_free_path(path);
2288         return ret;
2289 }
2290
2291 static noinline int btrfs_ioctl_tree_search(struct file *file,
2292                                            void __user *argp)
2293 {
2294         struct btrfs_ioctl_search_args __user *uargs;
2295         struct btrfs_ioctl_search_key sk;
2296         struct inode *inode;
2297         int ret;
2298         size_t buf_size;
2299
2300         if (!capable(CAP_SYS_ADMIN))
2301                 return -EPERM;
2302
2303         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2304
2305         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2306                 return -EFAULT;
2307
2308         buf_size = sizeof(uargs->buf);
2309
2310         inode = file_inode(file);
2311         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2312
2313         /*
2314          * In the origin implementation an overflow is handled by returning a
2315          * search header with a len of zero, so reset ret.
2316          */
2317         if (ret == -EOVERFLOW)
2318                 ret = 0;
2319
2320         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2321                 ret = -EFAULT;
2322         return ret;
2323 }
2324
2325 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2326                                                void __user *argp)
2327 {
2328         struct btrfs_ioctl_search_args_v2 __user *uarg;
2329         struct btrfs_ioctl_search_args_v2 args;
2330         struct inode *inode;
2331         int ret;
2332         size_t buf_size;
2333         const size_t buf_limit = SZ_16M;
2334
2335         if (!capable(CAP_SYS_ADMIN))
2336                 return -EPERM;
2337
2338         /* copy search header and buffer size */
2339         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2340         if (copy_from_user(&args, uarg, sizeof(args)))
2341                 return -EFAULT;
2342
2343         buf_size = args.buf_size;
2344
2345         /* limit result size to 16MB */
2346         if (buf_size > buf_limit)
2347                 buf_size = buf_limit;
2348
2349         inode = file_inode(file);
2350         ret = search_ioctl(inode, &args.key, &buf_size,
2351                            (char __user *)(&uarg->buf[0]));
2352         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2353                 ret = -EFAULT;
2354         else if (ret == -EOVERFLOW &&
2355                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2356                 ret = -EFAULT;
2357
2358         return ret;
2359 }
2360
2361 /*
2362  * Search INODE_REFs to identify path name of 'dirid' directory
2363  * in a 'tree_id' tree. and sets path name to 'name'.
2364  */
2365 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2366                                 u64 tree_id, u64 dirid, char *name)
2367 {
2368         struct btrfs_root *root;
2369         struct btrfs_key key;
2370         char *ptr;
2371         int ret = -1;
2372         int slot;
2373         int len;
2374         int total_len = 0;
2375         struct btrfs_inode_ref *iref;
2376         struct extent_buffer *l;
2377         struct btrfs_path *path;
2378
2379         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2380                 name[0]='\0';
2381                 return 0;
2382         }
2383
2384         path = btrfs_alloc_path();
2385         if (!path)
2386                 return -ENOMEM;
2387
2388         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2389
2390         root = btrfs_get_fs_root(info, tree_id, true);
2391         if (IS_ERR(root)) {
2392                 ret = PTR_ERR(root);
2393                 root = NULL;
2394                 goto out;
2395         }
2396
2397         key.objectid = dirid;
2398         key.type = BTRFS_INODE_REF_KEY;
2399         key.offset = (u64)-1;
2400
2401         while (1) {
2402                 ret = btrfs_search_backwards(root, &key, path);
2403                 if (ret < 0)
2404                         goto out;
2405                 else if (ret > 0) {
2406                         ret = -ENOENT;
2407                         goto out;
2408                 }
2409
2410                 l = path->nodes[0];
2411                 slot = path->slots[0];
2412
2413                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2414                 len = btrfs_inode_ref_name_len(l, iref);
2415                 ptr -= len + 1;
2416                 total_len += len + 1;
2417                 if (ptr < name) {
2418                         ret = -ENAMETOOLONG;
2419                         goto out;
2420                 }
2421
2422                 *(ptr + len) = '/';
2423                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2424
2425                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2426                         break;
2427
2428                 btrfs_release_path(path);
2429                 key.objectid = key.offset;
2430                 key.offset = (u64)-1;
2431                 dirid = key.objectid;
2432         }
2433         memmove(name, ptr, total_len);
2434         name[total_len] = '\0';
2435         ret = 0;
2436 out:
2437         btrfs_put_root(root);
2438         btrfs_free_path(path);
2439         return ret;
2440 }
2441
2442 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2443                                 struct inode *inode,
2444                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2445 {
2446         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2447         struct super_block *sb = inode->i_sb;
2448         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2449         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2450         u64 dirid = args->dirid;
2451         unsigned long item_off;
2452         unsigned long item_len;
2453         struct btrfs_inode_ref *iref;
2454         struct btrfs_root_ref *rref;
2455         struct btrfs_root *root = NULL;
2456         struct btrfs_path *path;
2457         struct btrfs_key key, key2;
2458         struct extent_buffer *leaf;
2459         struct inode *temp_inode;
2460         char *ptr;
2461         int slot;
2462         int len;
2463         int total_len = 0;
2464         int ret;
2465
2466         path = btrfs_alloc_path();
2467         if (!path)
2468                 return -ENOMEM;
2469
2470         /*
2471          * If the bottom subvolume does not exist directly under upper_limit,
2472          * construct the path in from the bottom up.
2473          */
2474         if (dirid != upper_limit.objectid) {
2475                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2476
2477                 root = btrfs_get_fs_root(fs_info, treeid, true);
2478                 if (IS_ERR(root)) {
2479                         ret = PTR_ERR(root);
2480                         goto out;
2481                 }
2482
2483                 key.objectid = dirid;
2484                 key.type = BTRFS_INODE_REF_KEY;
2485                 key.offset = (u64)-1;
2486                 while (1) {
2487                         ret = btrfs_search_backwards(root, &key, path);
2488                         if (ret < 0)
2489                                 goto out_put;
2490                         else if (ret > 0) {
2491                                 ret = -ENOENT;
2492                                 goto out_put;
2493                         }
2494
2495                         leaf = path->nodes[0];
2496                         slot = path->slots[0];
2497
2498                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2499                         len = btrfs_inode_ref_name_len(leaf, iref);
2500                         ptr -= len + 1;
2501                         total_len += len + 1;
2502                         if (ptr < args->path) {
2503                                 ret = -ENAMETOOLONG;
2504                                 goto out_put;
2505                         }
2506
2507                         *(ptr + len) = '/';
2508                         read_extent_buffer(leaf, ptr,
2509                                         (unsigned long)(iref + 1), len);
2510
2511                         /* Check the read+exec permission of this directory */
2512                         ret = btrfs_previous_item(root, path, dirid,
2513                                                   BTRFS_INODE_ITEM_KEY);
2514                         if (ret < 0) {
2515                                 goto out_put;
2516                         } else if (ret > 0) {
2517                                 ret = -ENOENT;
2518                                 goto out_put;
2519                         }
2520
2521                         leaf = path->nodes[0];
2522                         slot = path->slots[0];
2523                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2524                         if (key2.objectid != dirid) {
2525                                 ret = -ENOENT;
2526                                 goto out_put;
2527                         }
2528
2529                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2530                         if (IS_ERR(temp_inode)) {
2531                                 ret = PTR_ERR(temp_inode);
2532                                 goto out_put;
2533                         }
2534                         ret = inode_permission(mnt_userns, temp_inode,
2535                                                MAY_READ | MAY_EXEC);
2536                         iput(temp_inode);
2537                         if (ret) {
2538                                 ret = -EACCES;
2539                                 goto out_put;
2540                         }
2541
2542                         if (key.offset == upper_limit.objectid)
2543                                 break;
2544                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2545                                 ret = -EACCES;
2546                                 goto out_put;
2547                         }
2548
2549                         btrfs_release_path(path);
2550                         key.objectid = key.offset;
2551                         key.offset = (u64)-1;
2552                         dirid = key.objectid;
2553                 }
2554
2555                 memmove(args->path, ptr, total_len);
2556                 args->path[total_len] = '\0';
2557                 btrfs_put_root(root);
2558                 root = NULL;
2559                 btrfs_release_path(path);
2560         }
2561
2562         /* Get the bottom subvolume's name from ROOT_REF */
2563         key.objectid = treeid;
2564         key.type = BTRFS_ROOT_REF_KEY;
2565         key.offset = args->treeid;
2566         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2567         if (ret < 0) {
2568                 goto out;
2569         } else if (ret > 0) {
2570                 ret = -ENOENT;
2571                 goto out;
2572         }
2573
2574         leaf = path->nodes[0];
2575         slot = path->slots[0];
2576         btrfs_item_key_to_cpu(leaf, &key, slot);
2577
2578         item_off = btrfs_item_ptr_offset(leaf, slot);
2579         item_len = btrfs_item_size_nr(leaf, slot);
2580         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2581         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2582         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2583                 ret = -EINVAL;
2584                 goto out;
2585         }
2586
2587         /* Copy subvolume's name */
2588         item_off += sizeof(struct btrfs_root_ref);
2589         item_len -= sizeof(struct btrfs_root_ref);
2590         read_extent_buffer(leaf, args->name, item_off, item_len);
2591         args->name[item_len] = 0;
2592
2593 out_put:
2594         btrfs_put_root(root);
2595 out:
2596         btrfs_free_path(path);
2597         return ret;
2598 }
2599
2600 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2601                                            void __user *argp)
2602 {
2603         struct btrfs_ioctl_ino_lookup_args *args;
2604         struct inode *inode;
2605         int ret = 0;
2606
2607         args = memdup_user(argp, sizeof(*args));
2608         if (IS_ERR(args))
2609                 return PTR_ERR(args);
2610
2611         inode = file_inode(file);
2612
2613         /*
2614          * Unprivileged query to obtain the containing subvolume root id. The
2615          * path is reset so it's consistent with btrfs_search_path_in_tree.
2616          */
2617         if (args->treeid == 0)
2618                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2619
2620         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2621                 args->name[0] = 0;
2622                 goto out;
2623         }
2624
2625         if (!capable(CAP_SYS_ADMIN)) {
2626                 ret = -EPERM;
2627                 goto out;
2628         }
2629
2630         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2631                                         args->treeid, args->objectid,
2632                                         args->name);
2633
2634 out:
2635         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2636                 ret = -EFAULT;
2637
2638         kfree(args);
2639         return ret;
2640 }
2641
2642 /*
2643  * Version of ino_lookup ioctl (unprivileged)
2644  *
2645  * The main differences from ino_lookup ioctl are:
2646  *
2647  *   1. Read + Exec permission will be checked using inode_permission() during
2648  *      path construction. -EACCES will be returned in case of failure.
2649  *   2. Path construction will be stopped at the inode number which corresponds
2650  *      to the fd with which this ioctl is called. If constructed path does not
2651  *      exist under fd's inode, -EACCES will be returned.
2652  *   3. The name of bottom subvolume is also searched and filled.
2653  */
2654 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2655 {
2656         struct btrfs_ioctl_ino_lookup_user_args *args;
2657         struct inode *inode;
2658         int ret;
2659
2660         args = memdup_user(argp, sizeof(*args));
2661         if (IS_ERR(args))
2662                 return PTR_ERR(args);
2663
2664         inode = file_inode(file);
2665
2666         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2667             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2668                 /*
2669                  * The subvolume does not exist under fd with which this is
2670                  * called
2671                  */
2672                 kfree(args);
2673                 return -EACCES;
2674         }
2675
2676         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
2677
2678         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2679                 ret = -EFAULT;
2680
2681         kfree(args);
2682         return ret;
2683 }
2684
2685 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2686 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2687 {
2688         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2689         struct btrfs_fs_info *fs_info;
2690         struct btrfs_root *root;
2691         struct btrfs_path *path;
2692         struct btrfs_key key;
2693         struct btrfs_root_item *root_item;
2694         struct btrfs_root_ref *rref;
2695         struct extent_buffer *leaf;
2696         unsigned long item_off;
2697         unsigned long item_len;
2698         struct inode *inode;
2699         int slot;
2700         int ret = 0;
2701
2702         path = btrfs_alloc_path();
2703         if (!path)
2704                 return -ENOMEM;
2705
2706         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2707         if (!subvol_info) {
2708                 btrfs_free_path(path);
2709                 return -ENOMEM;
2710         }
2711
2712         inode = file_inode(file);
2713         fs_info = BTRFS_I(inode)->root->fs_info;
2714
2715         /* Get root_item of inode's subvolume */
2716         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2717         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2718         if (IS_ERR(root)) {
2719                 ret = PTR_ERR(root);
2720                 goto out_free;
2721         }
2722         root_item = &root->root_item;
2723
2724         subvol_info->treeid = key.objectid;
2725
2726         subvol_info->generation = btrfs_root_generation(root_item);
2727         subvol_info->flags = btrfs_root_flags(root_item);
2728
2729         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2730         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2731                                                     BTRFS_UUID_SIZE);
2732         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2733                                                     BTRFS_UUID_SIZE);
2734
2735         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2736         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2737         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2738
2739         subvol_info->otransid = btrfs_root_otransid(root_item);
2740         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2741         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2742
2743         subvol_info->stransid = btrfs_root_stransid(root_item);
2744         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2745         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2746
2747         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2748         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2749         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2750
2751         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2752                 /* Search root tree for ROOT_BACKREF of this subvolume */
2753                 key.type = BTRFS_ROOT_BACKREF_KEY;
2754                 key.offset = 0;
2755                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2756                 if (ret < 0) {
2757                         goto out;
2758                 } else if (path->slots[0] >=
2759                            btrfs_header_nritems(path->nodes[0])) {
2760                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2761                         if (ret < 0) {
2762                                 goto out;
2763                         } else if (ret > 0) {
2764                                 ret = -EUCLEAN;
2765                                 goto out;
2766                         }
2767                 }
2768
2769                 leaf = path->nodes[0];
2770                 slot = path->slots[0];
2771                 btrfs_item_key_to_cpu(leaf, &key, slot);
2772                 if (key.objectid == subvol_info->treeid &&
2773                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2774                         subvol_info->parent_id = key.offset;
2775
2776                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2777                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2778
2779                         item_off = btrfs_item_ptr_offset(leaf, slot)
2780                                         + sizeof(struct btrfs_root_ref);
2781                         item_len = btrfs_item_size_nr(leaf, slot)
2782                                         - sizeof(struct btrfs_root_ref);
2783                         read_extent_buffer(leaf, subvol_info->name,
2784                                            item_off, item_len);
2785                 } else {
2786                         ret = -ENOENT;
2787                         goto out;
2788                 }
2789         }
2790
2791         btrfs_free_path(path);
2792         path = NULL;
2793         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2794                 ret = -EFAULT;
2795
2796 out:
2797         btrfs_put_root(root);
2798 out_free:
2799         btrfs_free_path(path);
2800         kfree(subvol_info);
2801         return ret;
2802 }
2803
2804 /*
2805  * Return ROOT_REF information of the subvolume containing this inode
2806  * except the subvolume name.
2807  */
2808 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2809 {
2810         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2811         struct btrfs_root_ref *rref;
2812         struct btrfs_root *root;
2813         struct btrfs_path *path;
2814         struct btrfs_key key;
2815         struct extent_buffer *leaf;
2816         struct inode *inode;
2817         u64 objectid;
2818         int slot;
2819         int ret;
2820         u8 found;
2821
2822         path = btrfs_alloc_path();
2823         if (!path)
2824                 return -ENOMEM;
2825
2826         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2827         if (IS_ERR(rootrefs)) {
2828                 btrfs_free_path(path);
2829                 return PTR_ERR(rootrefs);
2830         }
2831
2832         inode = file_inode(file);
2833         root = BTRFS_I(inode)->root->fs_info->tree_root;
2834         objectid = BTRFS_I(inode)->root->root_key.objectid;
2835
2836         key.objectid = objectid;
2837         key.type = BTRFS_ROOT_REF_KEY;
2838         key.offset = rootrefs->min_treeid;
2839         found = 0;
2840
2841         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2842         if (ret < 0) {
2843                 goto out;
2844         } else if (path->slots[0] >=
2845                    btrfs_header_nritems(path->nodes[0])) {
2846                 ret = btrfs_next_leaf(root, path);
2847                 if (ret < 0) {
2848                         goto out;
2849                 } else if (ret > 0) {
2850                         ret = -EUCLEAN;
2851                         goto out;
2852                 }
2853         }
2854         while (1) {
2855                 leaf = path->nodes[0];
2856                 slot = path->slots[0];
2857
2858                 btrfs_item_key_to_cpu(leaf, &key, slot);
2859                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2860                         ret = 0;
2861                         goto out;
2862                 }
2863
2864                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2865                         ret = -EOVERFLOW;
2866                         goto out;
2867                 }
2868
2869                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2870                 rootrefs->rootref[found].treeid = key.offset;
2871                 rootrefs->rootref[found].dirid =
2872                                   btrfs_root_ref_dirid(leaf, rref);
2873                 found++;
2874
2875                 ret = btrfs_next_item(root, path);
2876                 if (ret < 0) {
2877                         goto out;
2878                 } else if (ret > 0) {
2879                         ret = -EUCLEAN;
2880                         goto out;
2881                 }
2882         }
2883
2884 out:
2885         btrfs_free_path(path);
2886
2887         if (!ret || ret == -EOVERFLOW) {
2888                 rootrefs->num_items = found;
2889                 /* update min_treeid for next search */
2890                 if (found)
2891                         rootrefs->min_treeid =
2892                                 rootrefs->rootref[found - 1].treeid + 1;
2893                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2894                         ret = -EFAULT;
2895         }
2896
2897         kfree(rootrefs);
2898
2899         return ret;
2900 }
2901
2902 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2903                                              void __user *arg,
2904                                              bool destroy_v2)
2905 {
2906         struct dentry *parent = file->f_path.dentry;
2907         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2908         struct dentry *dentry;
2909         struct inode *dir = d_inode(parent);
2910         struct inode *inode;
2911         struct btrfs_root *root = BTRFS_I(dir)->root;
2912         struct btrfs_root *dest = NULL;
2913         struct btrfs_ioctl_vol_args *vol_args = NULL;
2914         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2915         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
2916         char *subvol_name, *subvol_name_ptr = NULL;
2917         int subvol_namelen;
2918         int err = 0;
2919         bool destroy_parent = false;
2920
2921         if (destroy_v2) {
2922                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2923                 if (IS_ERR(vol_args2))
2924                         return PTR_ERR(vol_args2);
2925
2926                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2927                         err = -EOPNOTSUPP;
2928                         goto out;
2929                 }
2930
2931                 /*
2932                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2933                  * name, same as v1 currently does.
2934                  */
2935                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2936                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2937                         subvol_name = vol_args2->name;
2938
2939                         err = mnt_want_write_file(file);
2940                         if (err)
2941                                 goto out;
2942                 } else {
2943                         struct inode *old_dir;
2944
2945                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2946                                 err = -EINVAL;
2947                                 goto out;
2948                         }
2949
2950                         err = mnt_want_write_file(file);
2951                         if (err)
2952                                 goto out;
2953
2954                         dentry = btrfs_get_dentry(fs_info->sb,
2955                                         BTRFS_FIRST_FREE_OBJECTID,
2956                                         vol_args2->subvolid, 0, 0);
2957                         if (IS_ERR(dentry)) {
2958                                 err = PTR_ERR(dentry);
2959                                 goto out_drop_write;
2960                         }
2961
2962                         /*
2963                          * Change the default parent since the subvolume being
2964                          * deleted can be outside of the current mount point.
2965                          */
2966                         parent = btrfs_get_parent(dentry);
2967
2968                         /*
2969                          * At this point dentry->d_name can point to '/' if the
2970                          * subvolume we want to destroy is outsite of the
2971                          * current mount point, so we need to release the
2972                          * current dentry and execute the lookup to return a new
2973                          * one with ->d_name pointing to the
2974                          * <mount point>/subvol_name.
2975                          */
2976                         dput(dentry);
2977                         if (IS_ERR(parent)) {
2978                                 err = PTR_ERR(parent);
2979                                 goto out_drop_write;
2980                         }
2981                         old_dir = dir;
2982                         dir = d_inode(parent);
2983
2984                         /*
2985                          * If v2 was used with SPEC_BY_ID, a new parent was
2986                          * allocated since the subvolume can be outside of the
2987                          * current mount point. Later on we need to release this
2988                          * new parent dentry.
2989                          */
2990                         destroy_parent = true;
2991
2992                         /*
2993                          * On idmapped mounts, deletion via subvolid is
2994                          * restricted to subvolumes that are immediate
2995                          * ancestors of the inode referenced by the file
2996                          * descriptor in the ioctl. Otherwise the idmapping
2997                          * could potentially be abused to delete subvolumes
2998                          * anywhere in the filesystem the user wouldn't be able
2999                          * to delete without an idmapped mount.
3000                          */
3001                         if (old_dir != dir && mnt_userns != &init_user_ns) {
3002                                 err = -EOPNOTSUPP;
3003                                 goto free_parent;
3004                         }
3005
3006                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3007                                                 fs_info, vol_args2->subvolid);
3008                         if (IS_ERR(subvol_name_ptr)) {
3009                                 err = PTR_ERR(subvol_name_ptr);
3010                                 goto free_parent;
3011                         }
3012                         /* subvol_name_ptr is already nul terminated */
3013                         subvol_name = (char *)kbasename(subvol_name_ptr);
3014                 }
3015         } else {
3016                 vol_args = memdup_user(arg, sizeof(*vol_args));
3017                 if (IS_ERR(vol_args))
3018                         return PTR_ERR(vol_args);
3019
3020                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3021                 subvol_name = vol_args->name;
3022
3023                 err = mnt_want_write_file(file);
3024                 if (err)
3025                         goto out;
3026         }
3027
3028         subvol_namelen = strlen(subvol_name);
3029
3030         if (strchr(subvol_name, '/') ||
3031             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3032                 err = -EINVAL;
3033                 goto free_subvol_name;
3034         }
3035
3036         if (!S_ISDIR(dir->i_mode)) {
3037                 err = -ENOTDIR;
3038                 goto free_subvol_name;
3039         }
3040
3041         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3042         if (err == -EINTR)
3043                 goto free_subvol_name;
3044         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3045         if (IS_ERR(dentry)) {
3046                 err = PTR_ERR(dentry);
3047                 goto out_unlock_dir;
3048         }
3049
3050         if (d_really_is_negative(dentry)) {
3051                 err = -ENOENT;
3052                 goto out_dput;
3053         }
3054
3055         inode = d_inode(dentry);
3056         dest = BTRFS_I(inode)->root;
3057         if (!capable(CAP_SYS_ADMIN)) {
3058                 /*
3059                  * Regular user.  Only allow this with a special mount
3060                  * option, when the user has write+exec access to the
3061                  * subvol root, and when rmdir(2) would have been
3062                  * allowed.
3063                  *
3064                  * Note that this is _not_ check that the subvol is
3065                  * empty or doesn't contain data that we wouldn't
3066                  * otherwise be able to delete.
3067                  *
3068                  * Users who want to delete empty subvols should try
3069                  * rmdir(2).
3070                  */
3071                 err = -EPERM;
3072                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3073                         goto out_dput;
3074
3075                 /*
3076                  * Do not allow deletion if the parent dir is the same
3077                  * as the dir to be deleted.  That means the ioctl
3078                  * must be called on the dentry referencing the root
3079                  * of the subvol, not a random directory contained
3080                  * within it.
3081                  */
3082                 err = -EINVAL;
3083                 if (root == dest)
3084                         goto out_dput;
3085
3086                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3087                 if (err)
3088                         goto out_dput;
3089         }
3090
3091         /* check if subvolume may be deleted by a user */
3092         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3093         if (err)
3094                 goto out_dput;
3095
3096         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3097                 err = -EINVAL;
3098                 goto out_dput;
3099         }
3100
3101         btrfs_inode_lock(inode, 0);
3102         err = btrfs_delete_subvolume(dir, dentry);
3103         btrfs_inode_unlock(inode, 0);
3104         if (!err)
3105                 d_delete_notify(dir, dentry);
3106
3107 out_dput:
3108         dput(dentry);
3109 out_unlock_dir:
3110         btrfs_inode_unlock(dir, 0);
3111 free_subvol_name:
3112         kfree(subvol_name_ptr);
3113 free_parent:
3114         if (destroy_parent)
3115                 dput(parent);
3116 out_drop_write:
3117         mnt_drop_write_file(file);
3118 out:
3119         kfree(vol_args2);
3120         kfree(vol_args);
3121         return err;
3122 }
3123
3124 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3125 {
3126         struct inode *inode = file_inode(file);
3127         struct btrfs_root *root = BTRFS_I(inode)->root;
3128         struct btrfs_ioctl_defrag_range_args range = {0};
3129         int ret;
3130
3131         ret = mnt_want_write_file(file);
3132         if (ret)
3133                 return ret;
3134
3135         if (btrfs_root_readonly(root)) {
3136                 ret = -EROFS;
3137                 goto out;
3138         }
3139
3140         /* Subpage defrag will be supported in later commits */
3141         if (root->fs_info->sectorsize < PAGE_SIZE) {
3142                 ret = -ENOTTY;
3143                 goto out;
3144         }
3145
3146         switch (inode->i_mode & S_IFMT) {
3147         case S_IFDIR:
3148                 if (!capable(CAP_SYS_ADMIN)) {
3149                         ret = -EPERM;
3150                         goto out;
3151                 }
3152                 ret = btrfs_defrag_root(root);
3153                 break;
3154         case S_IFREG:
3155                 /*
3156                  * Note that this does not check the file descriptor for write
3157                  * access. This prevents defragmenting executables that are
3158                  * running and allows defrag on files open in read-only mode.
3159                  */
3160                 if (!capable(CAP_SYS_ADMIN) &&
3161                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3162                         ret = -EPERM;
3163                         goto out;
3164                 }
3165
3166                 if (argp) {
3167                         if (copy_from_user(&range, argp, sizeof(range))) {
3168                                 ret = -EFAULT;
3169                                 goto out;
3170                         }
3171                         /* compression requires us to start the IO */
3172                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3173                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3174                                 range.extent_thresh = (u32)-1;
3175                         }
3176                 } else {
3177                         /* the rest are all set to zero by kzalloc */
3178                         range.len = (u64)-1;
3179                 }
3180                 ret = btrfs_defrag_file(file_inode(file), file,
3181                                         &range, BTRFS_OLDEST_GENERATION, 0);
3182                 if (ret > 0)
3183                         ret = 0;
3184                 break;
3185         default:
3186                 ret = -EINVAL;
3187         }
3188 out:
3189         mnt_drop_write_file(file);
3190         return ret;
3191 }
3192
3193 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3194 {
3195         struct btrfs_ioctl_vol_args *vol_args;
3196         int ret;
3197
3198         if (!capable(CAP_SYS_ADMIN))
3199                 return -EPERM;
3200
3201         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3202                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3203
3204         vol_args = memdup_user(arg, sizeof(*vol_args));
3205         if (IS_ERR(vol_args)) {
3206                 ret = PTR_ERR(vol_args);
3207                 goto out;
3208         }
3209
3210         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3211         ret = btrfs_init_new_device(fs_info, vol_args->name);
3212
3213         if (!ret)
3214                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3215
3216         kfree(vol_args);
3217 out:
3218         btrfs_exclop_finish(fs_info);
3219         return ret;
3220 }
3221
3222 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3223 {
3224         BTRFS_DEV_LOOKUP_ARGS(args);
3225         struct inode *inode = file_inode(file);
3226         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3227         struct btrfs_ioctl_vol_args_v2 *vol_args;
3228         struct block_device *bdev = NULL;
3229         fmode_t mode;
3230         int ret;
3231         bool cancel = false;
3232
3233         if (!capable(CAP_SYS_ADMIN))
3234                 return -EPERM;
3235
3236         vol_args = memdup_user(arg, sizeof(*vol_args));
3237         if (IS_ERR(vol_args))
3238                 return PTR_ERR(vol_args);
3239
3240         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3241                 ret = -EOPNOTSUPP;
3242                 goto out;
3243         }
3244
3245         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3246         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3247                 args.devid = vol_args->devid;
3248         } else if (!strcmp("cancel", vol_args->name)) {
3249                 cancel = true;
3250         } else {
3251                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3252                 if (ret)
3253                         goto out;
3254         }
3255
3256         ret = mnt_want_write_file(file);
3257         if (ret)
3258                 goto out;
3259
3260         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3261                                            cancel);
3262         if (ret)
3263                 goto err_drop;
3264
3265         /* Exclusive operation is now claimed */
3266         ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3267
3268         btrfs_exclop_finish(fs_info);
3269
3270         if (!ret) {
3271                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3272                         btrfs_info(fs_info, "device deleted: id %llu",
3273                                         vol_args->devid);
3274                 else
3275                         btrfs_info(fs_info, "device deleted: %s",
3276                                         vol_args->name);
3277         }
3278 err_drop:
3279         mnt_drop_write_file(file);
3280         if (bdev)
3281                 blkdev_put(bdev, mode);
3282 out:
3283         btrfs_put_dev_args_from_path(&args);
3284         kfree(vol_args);
3285         return ret;
3286 }
3287
3288 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3289 {
3290         BTRFS_DEV_LOOKUP_ARGS(args);
3291         struct inode *inode = file_inode(file);
3292         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3293         struct btrfs_ioctl_vol_args *vol_args;
3294         struct block_device *bdev = NULL;
3295         fmode_t mode;
3296         int ret;
3297         bool cancel = false;
3298
3299         if (!capable(CAP_SYS_ADMIN))
3300                 return -EPERM;
3301
3302         vol_args = memdup_user(arg, sizeof(*vol_args));
3303         if (IS_ERR(vol_args))
3304                 return PTR_ERR(vol_args);
3305
3306         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3307         if (!strcmp("cancel", vol_args->name)) {
3308                 cancel = true;
3309         } else {
3310                 ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name);
3311                 if (ret)
3312                         goto out;
3313         }
3314
3315         ret = mnt_want_write_file(file);
3316         if (ret)
3317                 goto out;
3318
3319         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3320                                            cancel);
3321         if (ret == 0) {
3322                 ret = btrfs_rm_device(fs_info, &args, &bdev, &mode);
3323                 if (!ret)
3324                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3325                 btrfs_exclop_finish(fs_info);
3326         }
3327
3328         mnt_drop_write_file(file);
3329         if (bdev)
3330                 blkdev_put(bdev, mode);
3331 out:
3332         btrfs_put_dev_args_from_path(&args);
3333         kfree(vol_args);
3334         return ret;
3335 }
3336
3337 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3338                                 void __user *arg)
3339 {
3340         struct btrfs_ioctl_fs_info_args *fi_args;
3341         struct btrfs_device *device;
3342         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3343         u64 flags_in;
3344         int ret = 0;
3345
3346         fi_args = memdup_user(arg, sizeof(*fi_args));
3347         if (IS_ERR(fi_args))
3348                 return PTR_ERR(fi_args);
3349
3350         flags_in = fi_args->flags;
3351         memset(fi_args, 0, sizeof(*fi_args));
3352
3353         rcu_read_lock();
3354         fi_args->num_devices = fs_devices->num_devices;
3355
3356         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3357                 if (device->devid > fi_args->max_id)
3358                         fi_args->max_id = device->devid;
3359         }
3360         rcu_read_unlock();
3361
3362         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3363         fi_args->nodesize = fs_info->nodesize;
3364         fi_args->sectorsize = fs_info->sectorsize;
3365         fi_args->clone_alignment = fs_info->sectorsize;
3366
3367         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3368                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3369                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3370                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3371         }
3372
3373         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3374                 fi_args->generation = fs_info->generation;
3375                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3376         }
3377
3378         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3379                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3380                        sizeof(fi_args->metadata_uuid));
3381                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3382         }
3383
3384         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3385                 ret = -EFAULT;
3386
3387         kfree(fi_args);
3388         return ret;
3389 }
3390
3391 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3392                                  void __user *arg)
3393 {
3394         BTRFS_DEV_LOOKUP_ARGS(args);
3395         struct btrfs_ioctl_dev_info_args *di_args;
3396         struct btrfs_device *dev;
3397         int ret = 0;
3398
3399         di_args = memdup_user(arg, sizeof(*di_args));
3400         if (IS_ERR(di_args))
3401                 return PTR_ERR(di_args);
3402
3403         args.devid = di_args->devid;
3404         if (!btrfs_is_empty_uuid(di_args->uuid))
3405                 args.uuid = di_args->uuid;
3406
3407         rcu_read_lock();
3408         dev = btrfs_find_device(fs_info->fs_devices, &args);
3409         if (!dev) {
3410                 ret = -ENODEV;
3411                 goto out;
3412         }
3413
3414         di_args->devid = dev->devid;
3415         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3416         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3417         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3418         if (dev->name)
3419                 strscpy(di_args->path, rcu_str_deref(dev->name), sizeof(di_args->path));
3420         else
3421                 di_args->path[0] = '\0';
3422
3423 out:
3424         rcu_read_unlock();
3425         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3426                 ret = -EFAULT;
3427
3428         kfree(di_args);
3429         return ret;
3430 }
3431
3432 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3433 {
3434         struct inode *inode = file_inode(file);
3435         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3436         struct btrfs_root *root = BTRFS_I(inode)->root;
3437         struct btrfs_root *new_root;
3438         struct btrfs_dir_item *di;
3439         struct btrfs_trans_handle *trans;
3440         struct btrfs_path *path = NULL;
3441         struct btrfs_disk_key disk_key;
3442         u64 objectid = 0;
3443         u64 dir_id;
3444         int ret;
3445
3446         if (!capable(CAP_SYS_ADMIN))
3447                 return -EPERM;
3448
3449         ret = mnt_want_write_file(file);
3450         if (ret)
3451                 return ret;
3452
3453         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3454                 ret = -EFAULT;
3455                 goto out;
3456         }
3457
3458         if (!objectid)
3459                 objectid = BTRFS_FS_TREE_OBJECTID;
3460
3461         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3462         if (IS_ERR(new_root)) {
3463                 ret = PTR_ERR(new_root);
3464                 goto out;
3465         }
3466         if (!is_fstree(new_root->root_key.objectid)) {
3467                 ret = -ENOENT;
3468                 goto out_free;
3469         }
3470
3471         path = btrfs_alloc_path();
3472         if (!path) {
3473                 ret = -ENOMEM;
3474                 goto out_free;
3475         }
3476
3477         trans = btrfs_start_transaction(root, 1);
3478         if (IS_ERR(trans)) {
3479                 ret = PTR_ERR(trans);
3480                 goto out_free;
3481         }
3482
3483         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3484         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3485                                    dir_id, "default", 7, 1);
3486         if (IS_ERR_OR_NULL(di)) {
3487                 btrfs_release_path(path);
3488                 btrfs_end_transaction(trans);
3489                 btrfs_err(fs_info,
3490                           "Umm, you don't have the default diritem, this isn't going to work");
3491                 ret = -ENOENT;
3492                 goto out_free;
3493         }
3494
3495         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3496         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3497         btrfs_mark_buffer_dirty(path->nodes[0]);
3498         btrfs_release_path(path);
3499
3500         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3501         btrfs_end_transaction(trans);
3502 out_free:
3503         btrfs_put_root(new_root);
3504         btrfs_free_path(path);
3505 out:
3506         mnt_drop_write_file(file);
3507         return ret;
3508 }
3509
3510 static void get_block_group_info(struct list_head *groups_list,
3511                                  struct btrfs_ioctl_space_info *space)
3512 {
3513         struct btrfs_block_group *block_group;
3514
3515         space->total_bytes = 0;
3516         space->used_bytes = 0;
3517         space->flags = 0;
3518         list_for_each_entry(block_group, groups_list, list) {
3519                 space->flags = block_group->flags;
3520                 space->total_bytes += block_group->length;
3521                 space->used_bytes += block_group->used;
3522         }
3523 }
3524
3525 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3526                                    void __user *arg)
3527 {
3528         struct btrfs_ioctl_space_args space_args;
3529         struct btrfs_ioctl_space_info space;
3530         struct btrfs_ioctl_space_info *dest;
3531         struct btrfs_ioctl_space_info *dest_orig;
3532         struct btrfs_ioctl_space_info __user *user_dest;
3533         struct btrfs_space_info *info;
3534         static const u64 types[] = {
3535                 BTRFS_BLOCK_GROUP_DATA,
3536                 BTRFS_BLOCK_GROUP_SYSTEM,
3537                 BTRFS_BLOCK_GROUP_METADATA,
3538                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3539         };
3540         int num_types = 4;
3541         int alloc_size;
3542         int ret = 0;
3543         u64 slot_count = 0;
3544         int i, c;
3545
3546         if (copy_from_user(&space_args,
3547                            (struct btrfs_ioctl_space_args __user *)arg,
3548                            sizeof(space_args)))
3549                 return -EFAULT;
3550
3551         for (i = 0; i < num_types; i++) {
3552                 struct btrfs_space_info *tmp;
3553
3554                 info = NULL;
3555                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3556                         if (tmp->flags == types[i]) {
3557                                 info = tmp;
3558                                 break;
3559                         }
3560                 }
3561
3562                 if (!info)
3563                         continue;
3564
3565                 down_read(&info->groups_sem);
3566                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3567                         if (!list_empty(&info->block_groups[c]))
3568                                 slot_count++;
3569                 }
3570                 up_read(&info->groups_sem);
3571         }
3572
3573         /*
3574          * Global block reserve, exported as a space_info
3575          */
3576         slot_count++;
3577
3578         /* space_slots == 0 means they are asking for a count */
3579         if (space_args.space_slots == 0) {
3580                 space_args.total_spaces = slot_count;
3581                 goto out;
3582         }
3583
3584         slot_count = min_t(u64, space_args.space_slots, slot_count);
3585
3586         alloc_size = sizeof(*dest) * slot_count;
3587
3588         /* we generally have at most 6 or so space infos, one for each raid
3589          * level.  So, a whole page should be more than enough for everyone
3590          */
3591         if (alloc_size > PAGE_SIZE)
3592                 return -ENOMEM;
3593
3594         space_args.total_spaces = 0;
3595         dest = kmalloc(alloc_size, GFP_KERNEL);
3596         if (!dest)
3597                 return -ENOMEM;
3598         dest_orig = dest;
3599
3600         /* now we have a buffer to copy into */
3601         for (i = 0; i < num_types; i++) {
3602                 struct btrfs_space_info *tmp;
3603
3604                 if (!slot_count)
3605                         break;
3606
3607                 info = NULL;
3608                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3609                         if (tmp->flags == types[i]) {
3610                                 info = tmp;
3611                                 break;
3612                         }
3613                 }
3614
3615                 if (!info)
3616                         continue;
3617                 down_read(&info->groups_sem);
3618                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3619                         if (!list_empty(&info->block_groups[c])) {
3620                                 get_block_group_info(&info->block_groups[c],
3621                                                      &space);
3622                                 memcpy(dest, &space, sizeof(space));
3623                                 dest++;
3624                                 space_args.total_spaces++;
3625                                 slot_count--;
3626                         }
3627                         if (!slot_count)
3628                                 break;
3629                 }
3630                 up_read(&info->groups_sem);
3631         }
3632
3633         /*
3634          * Add global block reserve
3635          */
3636         if (slot_count) {
3637                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3638
3639                 spin_lock(&block_rsv->lock);
3640                 space.total_bytes = block_rsv->size;
3641                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3642                 spin_unlock(&block_rsv->lock);
3643                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3644                 memcpy(dest, &space, sizeof(space));
3645                 space_args.total_spaces++;
3646         }
3647
3648         user_dest = (struct btrfs_ioctl_space_info __user *)
3649                 (arg + sizeof(struct btrfs_ioctl_space_args));
3650
3651         if (copy_to_user(user_dest, dest_orig, alloc_size))
3652                 ret = -EFAULT;
3653
3654         kfree(dest_orig);
3655 out:
3656         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3657                 ret = -EFAULT;
3658
3659         return ret;
3660 }
3661
3662 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3663                                             void __user *argp)
3664 {
3665         struct btrfs_trans_handle *trans;
3666         u64 transid;
3667         int ret;
3668
3669         trans = btrfs_attach_transaction_barrier(root);
3670         if (IS_ERR(trans)) {
3671                 if (PTR_ERR(trans) != -ENOENT)
3672                         return PTR_ERR(trans);
3673
3674                 /* No running transaction, don't bother */
3675                 transid = root->fs_info->last_trans_committed;
3676                 goto out;
3677         }
3678         transid = trans->transid;
3679         ret = btrfs_commit_transaction_async(trans);
3680         if (ret) {
3681                 btrfs_end_transaction(trans);
3682                 return ret;
3683         }
3684 out:
3685         if (argp)
3686                 if (copy_to_user(argp, &transid, sizeof(transid)))
3687                         return -EFAULT;
3688         return 0;
3689 }
3690
3691 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3692                                            void __user *argp)
3693 {
3694         u64 transid;
3695
3696         if (argp) {
3697                 if (copy_from_user(&transid, argp, sizeof(transid)))
3698                         return -EFAULT;
3699         } else {
3700                 transid = 0;  /* current trans */
3701         }
3702         return btrfs_wait_for_commit(fs_info, transid);
3703 }
3704
3705 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3706 {
3707         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3708         struct btrfs_ioctl_scrub_args *sa;
3709         int ret;
3710
3711         if (!capable(CAP_SYS_ADMIN))
3712                 return -EPERM;
3713
3714         sa = memdup_user(arg, sizeof(*sa));
3715         if (IS_ERR(sa))
3716                 return PTR_ERR(sa);
3717
3718         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3719                 ret = mnt_want_write_file(file);
3720                 if (ret)
3721                         goto out;
3722         }
3723
3724         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3725                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3726                               0);
3727
3728         /*
3729          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3730          * error. This is important as it allows user space to know how much
3731          * progress scrub has done. For example, if scrub is canceled we get
3732          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3733          * space. Later user space can inspect the progress from the structure
3734          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3735          * previously (btrfs-progs does this).
3736          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3737          * then return -EFAULT to signal the structure was not copied or it may
3738          * be corrupt and unreliable due to a partial copy.
3739          */
3740         if (copy_to_user(arg, sa, sizeof(*sa)))
3741                 ret = -EFAULT;
3742
3743         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3744                 mnt_drop_write_file(file);
3745 out:
3746         kfree(sa);
3747         return ret;
3748 }
3749
3750 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3751 {
3752         if (!capable(CAP_SYS_ADMIN))
3753                 return -EPERM;
3754
3755         return btrfs_scrub_cancel(fs_info);
3756 }
3757
3758 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3759                                        void __user *arg)
3760 {
3761         struct btrfs_ioctl_scrub_args *sa;
3762         int ret;
3763
3764         if (!capable(CAP_SYS_ADMIN))
3765                 return -EPERM;
3766
3767         sa = memdup_user(arg, sizeof(*sa));
3768         if (IS_ERR(sa))
3769                 return PTR_ERR(sa);
3770
3771         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3772
3773         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3774                 ret = -EFAULT;
3775
3776         kfree(sa);
3777         return ret;
3778 }
3779
3780 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3781                                       void __user *arg)
3782 {
3783         struct btrfs_ioctl_get_dev_stats *sa;
3784         int ret;
3785
3786         sa = memdup_user(arg, sizeof(*sa));
3787         if (IS_ERR(sa))
3788                 return PTR_ERR(sa);
3789
3790         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3791                 kfree(sa);
3792                 return -EPERM;
3793         }
3794
3795         ret = btrfs_get_dev_stats(fs_info, sa);
3796
3797         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3798                 ret = -EFAULT;
3799
3800         kfree(sa);
3801         return ret;
3802 }
3803
3804 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3805                                     void __user *arg)
3806 {
3807         struct btrfs_ioctl_dev_replace_args *p;
3808         int ret;
3809
3810         if (!capable(CAP_SYS_ADMIN))
3811                 return -EPERM;
3812
3813         p = memdup_user(arg, sizeof(*p));
3814         if (IS_ERR(p))
3815                 return PTR_ERR(p);
3816
3817         switch (p->cmd) {
3818         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3819                 if (sb_rdonly(fs_info->sb)) {
3820                         ret = -EROFS;
3821                         goto out;
3822                 }
3823                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3824                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3825                 } else {
3826                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3827                         btrfs_exclop_finish(fs_info);
3828                 }
3829                 break;
3830         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3831                 btrfs_dev_replace_status(fs_info, p);
3832                 ret = 0;
3833                 break;
3834         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3835                 p->result = btrfs_dev_replace_cancel(fs_info);
3836                 ret = 0;
3837                 break;
3838         default:
3839                 ret = -EINVAL;
3840                 break;
3841         }
3842
3843         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3844                 ret = -EFAULT;
3845 out:
3846         kfree(p);
3847         return ret;
3848 }
3849
3850 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3851 {
3852         int ret = 0;
3853         int i;
3854         u64 rel_ptr;
3855         int size;
3856         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3857         struct inode_fs_paths *ipath = NULL;
3858         struct btrfs_path *path;
3859
3860         if (!capable(CAP_DAC_READ_SEARCH))
3861                 return -EPERM;
3862
3863         path = btrfs_alloc_path();
3864         if (!path) {
3865                 ret = -ENOMEM;
3866                 goto out;
3867         }
3868
3869         ipa = memdup_user(arg, sizeof(*ipa));
3870         if (IS_ERR(ipa)) {
3871                 ret = PTR_ERR(ipa);
3872                 ipa = NULL;
3873                 goto out;
3874         }
3875
3876         size = min_t(u32, ipa->size, 4096);
3877         ipath = init_ipath(size, root, path);
3878         if (IS_ERR(ipath)) {
3879                 ret = PTR_ERR(ipath);
3880                 ipath = NULL;
3881                 goto out;
3882         }
3883
3884         ret = paths_from_inode(ipa->inum, ipath);
3885         if (ret < 0)
3886                 goto out;
3887
3888         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3889                 rel_ptr = ipath->fspath->val[i] -
3890                           (u64)(unsigned long)ipath->fspath->val;
3891                 ipath->fspath->val[i] = rel_ptr;
3892         }
3893
3894         btrfs_free_path(path);
3895         path = NULL;
3896         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3897                            ipath->fspath, size);
3898         if (ret) {
3899                 ret = -EFAULT;
3900                 goto out;
3901         }
3902
3903 out:
3904         btrfs_free_path(path);
3905         free_ipath(ipath);
3906         kfree(ipa);
3907
3908         return ret;
3909 }
3910
3911 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3912                                         void __user *arg, int version)
3913 {
3914         int ret = 0;
3915         int size;
3916         struct btrfs_ioctl_logical_ino_args *loi;
3917         struct btrfs_data_container *inodes = NULL;
3918         struct btrfs_path *path = NULL;
3919         bool ignore_offset;
3920
3921         if (!capable(CAP_SYS_ADMIN))
3922                 return -EPERM;
3923
3924         loi = memdup_user(arg, sizeof(*loi));
3925         if (IS_ERR(loi))
3926                 return PTR_ERR(loi);
3927
3928         if (version == 1) {
3929                 ignore_offset = false;
3930                 size = min_t(u32, loi->size, SZ_64K);
3931         } else {
3932                 /* All reserved bits must be 0 for now */
3933                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3934                         ret = -EINVAL;
3935                         goto out_loi;
3936                 }
3937                 /* Only accept flags we have defined so far */
3938                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3939                         ret = -EINVAL;
3940                         goto out_loi;
3941                 }
3942                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3943                 size = min_t(u32, loi->size, SZ_16M);
3944         }
3945
3946         inodes = init_data_container(size);
3947         if (IS_ERR(inodes)) {
3948                 ret = PTR_ERR(inodes);
3949                 goto out_loi;
3950         }
3951
3952         path = btrfs_alloc_path();
3953         if (!path) {
3954                 ret = -ENOMEM;
3955                 goto out;
3956         }
3957         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3958                                           inodes, ignore_offset);
3959         btrfs_free_path(path);
3960         if (ret == -EINVAL)
3961                 ret = -ENOENT;
3962         if (ret < 0)
3963                 goto out;
3964
3965         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3966                            size);
3967         if (ret)
3968                 ret = -EFAULT;
3969
3970 out:
3971         kvfree(inodes);
3972 out_loi:
3973         kfree(loi);
3974
3975         return ret;
3976 }
3977
3978 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3979                                struct btrfs_ioctl_balance_args *bargs)
3980 {
3981         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3982
3983         bargs->flags = bctl->flags;
3984
3985         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3986                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3987         if (atomic_read(&fs_info->balance_pause_req))
3988                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3989         if (atomic_read(&fs_info->balance_cancel_req))
3990                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3991
3992         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3993         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3994         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3995
3996         spin_lock(&fs_info->balance_lock);
3997         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3998         spin_unlock(&fs_info->balance_lock);
3999 }
4000
4001 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4002 {
4003         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4004         struct btrfs_fs_info *fs_info = root->fs_info;
4005         struct btrfs_ioctl_balance_args *bargs;
4006         struct btrfs_balance_control *bctl;
4007         bool need_unlock; /* for mut. excl. ops lock */
4008         int ret;
4009
4010         if (!capable(CAP_SYS_ADMIN))
4011                 return -EPERM;
4012
4013         ret = mnt_want_write_file(file);
4014         if (ret)
4015                 return ret;
4016
4017 again:
4018         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4019                 mutex_lock(&fs_info->balance_mutex);
4020                 need_unlock = true;
4021                 goto locked;
4022         }
4023
4024         /*
4025          * mut. excl. ops lock is locked.  Three possibilities:
4026          *   (1) some other op is running
4027          *   (2) balance is running
4028          *   (3) balance is paused -- special case (think resume)
4029          */
4030         mutex_lock(&fs_info->balance_mutex);
4031         if (fs_info->balance_ctl) {
4032                 /* this is either (2) or (3) */
4033                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4034                         mutex_unlock(&fs_info->balance_mutex);
4035                         /*
4036                          * Lock released to allow other waiters to continue,
4037                          * we'll reexamine the status again.
4038                          */
4039                         mutex_lock(&fs_info->balance_mutex);
4040
4041                         if (fs_info->balance_ctl &&
4042                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4043                                 /* this is (3) */
4044                                 need_unlock = false;
4045                                 goto locked;
4046                         }
4047
4048                         mutex_unlock(&fs_info->balance_mutex);
4049                         goto again;
4050                 } else {
4051                         /* this is (2) */
4052                         mutex_unlock(&fs_info->balance_mutex);
4053                         ret = -EINPROGRESS;
4054                         goto out;
4055                 }
4056         } else {
4057                 /* this is (1) */
4058                 mutex_unlock(&fs_info->balance_mutex);
4059                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4060                 goto out;
4061         }
4062
4063 locked:
4064
4065         if (arg) {
4066                 bargs = memdup_user(arg, sizeof(*bargs));
4067                 if (IS_ERR(bargs)) {
4068                         ret = PTR_ERR(bargs);
4069                         goto out_unlock;
4070                 }
4071
4072                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4073                         if (!fs_info->balance_ctl) {
4074                                 ret = -ENOTCONN;
4075                                 goto out_bargs;
4076                         }
4077
4078                         bctl = fs_info->balance_ctl;
4079                         spin_lock(&fs_info->balance_lock);
4080                         bctl->flags |= BTRFS_BALANCE_RESUME;
4081                         spin_unlock(&fs_info->balance_lock);
4082
4083                         goto do_balance;
4084                 }
4085         } else {
4086                 bargs = NULL;
4087         }
4088
4089         if (fs_info->balance_ctl) {
4090                 ret = -EINPROGRESS;
4091                 goto out_bargs;
4092         }
4093
4094         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4095         if (!bctl) {
4096                 ret = -ENOMEM;
4097                 goto out_bargs;
4098         }
4099
4100         if (arg) {
4101                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4102                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4103                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4104
4105                 bctl->flags = bargs->flags;
4106         } else {
4107                 /* balance everything - no filters */
4108                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4109         }
4110
4111         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4112                 ret = -EINVAL;
4113                 goto out_bctl;
4114         }
4115
4116 do_balance:
4117         /*
4118          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4119          * bctl is freed in reset_balance_state, or, if restriper was paused
4120          * all the way until unmount, in free_fs_info.  The flag should be
4121          * cleared after reset_balance_state.
4122          */
4123         need_unlock = false;
4124
4125         ret = btrfs_balance(fs_info, bctl, bargs);
4126         bctl = NULL;
4127
4128         if ((ret == 0 || ret == -ECANCELED) && arg) {
4129                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4130                         ret = -EFAULT;
4131         }
4132
4133 out_bctl:
4134         kfree(bctl);
4135 out_bargs:
4136         kfree(bargs);
4137 out_unlock:
4138         mutex_unlock(&fs_info->balance_mutex);
4139         if (need_unlock)
4140                 btrfs_exclop_finish(fs_info);
4141 out:
4142         mnt_drop_write_file(file);
4143         return ret;
4144 }
4145
4146 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4147 {
4148         if (!capable(CAP_SYS_ADMIN))
4149                 return -EPERM;
4150
4151         switch (cmd) {
4152         case BTRFS_BALANCE_CTL_PAUSE:
4153                 return btrfs_pause_balance(fs_info);
4154         case BTRFS_BALANCE_CTL_CANCEL:
4155                 return btrfs_cancel_balance(fs_info);
4156         }
4157
4158         return -EINVAL;
4159 }
4160
4161 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4162                                          void __user *arg)
4163 {
4164         struct btrfs_ioctl_balance_args *bargs;
4165         int ret = 0;
4166
4167         if (!capable(CAP_SYS_ADMIN))
4168                 return -EPERM;
4169
4170         mutex_lock(&fs_info->balance_mutex);
4171         if (!fs_info->balance_ctl) {
4172                 ret = -ENOTCONN;
4173                 goto out;
4174         }
4175
4176         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4177         if (!bargs) {
4178                 ret = -ENOMEM;
4179                 goto out;
4180         }
4181
4182         btrfs_update_ioctl_balance_args(fs_info, bargs);
4183
4184         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4185                 ret = -EFAULT;
4186
4187         kfree(bargs);
4188 out:
4189         mutex_unlock(&fs_info->balance_mutex);
4190         return ret;
4191 }
4192
4193 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4194 {
4195         struct inode *inode = file_inode(file);
4196         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4197         struct btrfs_ioctl_quota_ctl_args *sa;
4198         int ret;
4199
4200         if (!capable(CAP_SYS_ADMIN))
4201                 return -EPERM;
4202
4203         ret = mnt_want_write_file(file);
4204         if (ret)
4205                 return ret;
4206
4207         sa = memdup_user(arg, sizeof(*sa));
4208         if (IS_ERR(sa)) {
4209                 ret = PTR_ERR(sa);
4210                 goto drop_write;
4211         }
4212
4213         down_write(&fs_info->subvol_sem);
4214
4215         switch (sa->cmd) {
4216         case BTRFS_QUOTA_CTL_ENABLE:
4217                 ret = btrfs_quota_enable(fs_info);
4218                 break;
4219         case BTRFS_QUOTA_CTL_DISABLE:
4220                 ret = btrfs_quota_disable(fs_info);
4221                 break;
4222         default:
4223                 ret = -EINVAL;
4224                 break;
4225         }
4226
4227         kfree(sa);
4228         up_write(&fs_info->subvol_sem);
4229 drop_write:
4230         mnt_drop_write_file(file);
4231         return ret;
4232 }
4233
4234 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4235 {
4236         struct inode *inode = file_inode(file);
4237         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4238         struct btrfs_root *root = BTRFS_I(inode)->root;
4239         struct btrfs_ioctl_qgroup_assign_args *sa;
4240         struct btrfs_trans_handle *trans;
4241         int ret;
4242         int err;
4243
4244         if (!capable(CAP_SYS_ADMIN))
4245                 return -EPERM;
4246
4247         ret = mnt_want_write_file(file);
4248         if (ret)
4249                 return ret;
4250
4251         sa = memdup_user(arg, sizeof(*sa));
4252         if (IS_ERR(sa)) {
4253                 ret = PTR_ERR(sa);
4254                 goto drop_write;
4255         }
4256
4257         trans = btrfs_join_transaction(root);
4258         if (IS_ERR(trans)) {
4259                 ret = PTR_ERR(trans);
4260                 goto out;
4261         }
4262
4263         if (sa->assign) {
4264                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4265         } else {
4266                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4267         }
4268
4269         /* update qgroup status and info */
4270         err = btrfs_run_qgroups(trans);
4271         if (err < 0)
4272                 btrfs_handle_fs_error(fs_info, err,
4273                                       "failed to update qgroup status and info");
4274         err = btrfs_end_transaction(trans);
4275         if (err && !ret)
4276                 ret = err;
4277
4278 out:
4279         kfree(sa);
4280 drop_write:
4281         mnt_drop_write_file(file);
4282         return ret;
4283 }
4284
4285 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4286 {
4287         struct inode *inode = file_inode(file);
4288         struct btrfs_root *root = BTRFS_I(inode)->root;
4289         struct btrfs_ioctl_qgroup_create_args *sa;
4290         struct btrfs_trans_handle *trans;
4291         int ret;
4292         int err;
4293
4294         if (!capable(CAP_SYS_ADMIN))
4295                 return -EPERM;
4296
4297         ret = mnt_want_write_file(file);
4298         if (ret)
4299                 return ret;
4300
4301         sa = memdup_user(arg, sizeof(*sa));
4302         if (IS_ERR(sa)) {
4303                 ret = PTR_ERR(sa);
4304                 goto drop_write;
4305         }
4306
4307         if (!sa->qgroupid) {
4308                 ret = -EINVAL;
4309                 goto out;
4310         }
4311
4312         trans = btrfs_join_transaction(root);
4313         if (IS_ERR(trans)) {
4314                 ret = PTR_ERR(trans);
4315                 goto out;
4316         }
4317
4318         if (sa->create) {
4319                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4320         } else {
4321                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4322         }
4323
4324         err = btrfs_end_transaction(trans);
4325         if (err && !ret)
4326                 ret = err;
4327
4328 out:
4329         kfree(sa);
4330 drop_write:
4331         mnt_drop_write_file(file);
4332         return ret;
4333 }
4334
4335 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4336 {
4337         struct inode *inode = file_inode(file);
4338         struct btrfs_root *root = BTRFS_I(inode)->root;
4339         struct btrfs_ioctl_qgroup_limit_args *sa;
4340         struct btrfs_trans_handle *trans;
4341         int ret;
4342         int err;
4343         u64 qgroupid;
4344
4345         if (!capable(CAP_SYS_ADMIN))
4346                 return -EPERM;
4347
4348         ret = mnt_want_write_file(file);
4349         if (ret)
4350                 return ret;
4351
4352         sa = memdup_user(arg, sizeof(*sa));
4353         if (IS_ERR(sa)) {
4354                 ret = PTR_ERR(sa);
4355                 goto drop_write;
4356         }
4357
4358         trans = btrfs_join_transaction(root);
4359         if (IS_ERR(trans)) {
4360                 ret = PTR_ERR(trans);
4361                 goto out;
4362         }
4363
4364         qgroupid = sa->qgroupid;
4365         if (!qgroupid) {
4366                 /* take the current subvol as qgroup */
4367                 qgroupid = root->root_key.objectid;
4368         }
4369
4370         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4371
4372         err = btrfs_end_transaction(trans);
4373         if (err && !ret)
4374                 ret = err;
4375
4376 out:
4377         kfree(sa);
4378 drop_write:
4379         mnt_drop_write_file(file);
4380         return ret;
4381 }
4382
4383 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4384 {
4385         struct inode *inode = file_inode(file);
4386         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4387         struct btrfs_ioctl_quota_rescan_args *qsa;
4388         int ret;
4389
4390         if (!capable(CAP_SYS_ADMIN))
4391                 return -EPERM;
4392
4393         ret = mnt_want_write_file(file);
4394         if (ret)
4395                 return ret;
4396
4397         qsa = memdup_user(arg, sizeof(*qsa));
4398         if (IS_ERR(qsa)) {
4399                 ret = PTR_ERR(qsa);
4400                 goto drop_write;
4401         }
4402
4403         if (qsa->flags) {
4404                 ret = -EINVAL;
4405                 goto out;
4406         }
4407
4408         ret = btrfs_qgroup_rescan(fs_info);
4409
4410 out:
4411         kfree(qsa);
4412 drop_write:
4413         mnt_drop_write_file(file);
4414         return ret;
4415 }
4416
4417 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4418                                                 void __user *arg)
4419 {
4420         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4421         int ret = 0;
4422
4423         if (!capable(CAP_SYS_ADMIN))
4424                 return -EPERM;
4425
4426         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4427                 qsa.flags = 1;
4428                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4429         }
4430
4431         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4432                 ret = -EFAULT;
4433
4434         return ret;
4435 }
4436
4437 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4438                                                 void __user *arg)
4439 {
4440         if (!capable(CAP_SYS_ADMIN))
4441                 return -EPERM;
4442
4443         return btrfs_qgroup_wait_for_completion(fs_info, true);
4444 }
4445
4446 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4447                                             struct user_namespace *mnt_userns,
4448                                             struct btrfs_ioctl_received_subvol_args *sa)
4449 {
4450         struct inode *inode = file_inode(file);
4451         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4452         struct btrfs_root *root = BTRFS_I(inode)->root;
4453         struct btrfs_root_item *root_item = &root->root_item;
4454         struct btrfs_trans_handle *trans;
4455         struct timespec64 ct = current_time(inode);
4456         int ret = 0;
4457         int received_uuid_changed;
4458
4459         if (!inode_owner_or_capable(mnt_userns, inode))
4460                 return -EPERM;
4461
4462         ret = mnt_want_write_file(file);
4463         if (ret < 0)
4464                 return ret;
4465
4466         down_write(&fs_info->subvol_sem);
4467
4468         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4469                 ret = -EINVAL;
4470                 goto out;
4471         }
4472
4473         if (btrfs_root_readonly(root)) {
4474                 ret = -EROFS;
4475                 goto out;
4476         }
4477
4478         /*
4479          * 1 - root item
4480          * 2 - uuid items (received uuid + subvol uuid)
4481          */
4482         trans = btrfs_start_transaction(root, 3);
4483         if (IS_ERR(trans)) {
4484                 ret = PTR_ERR(trans);
4485                 trans = NULL;
4486                 goto out;
4487         }
4488
4489         sa->rtransid = trans->transid;
4490         sa->rtime.sec = ct.tv_sec;
4491         sa->rtime.nsec = ct.tv_nsec;
4492
4493         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4494                                        BTRFS_UUID_SIZE);
4495         if (received_uuid_changed &&
4496             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4497                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4498                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4499                                           root->root_key.objectid);
4500                 if (ret && ret != -ENOENT) {
4501                         btrfs_abort_transaction(trans, ret);
4502                         btrfs_end_transaction(trans);
4503                         goto out;
4504                 }
4505         }
4506         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4507         btrfs_set_root_stransid(root_item, sa->stransid);
4508         btrfs_set_root_rtransid(root_item, sa->rtransid);
4509         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4510         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4511         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4512         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4513
4514         ret = btrfs_update_root(trans, fs_info->tree_root,
4515                                 &root->root_key, &root->root_item);
4516         if (ret < 0) {
4517                 btrfs_end_transaction(trans);
4518                 goto out;
4519         }
4520         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4521                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4522                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4523                                           root->root_key.objectid);
4524                 if (ret < 0 && ret != -EEXIST) {
4525                         btrfs_abort_transaction(trans, ret);
4526                         btrfs_end_transaction(trans);
4527                         goto out;
4528                 }
4529         }
4530         ret = btrfs_commit_transaction(trans);
4531 out:
4532         up_write(&fs_info->subvol_sem);
4533         mnt_drop_write_file(file);
4534         return ret;
4535 }
4536
4537 #ifdef CONFIG_64BIT
4538 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4539                                                 void __user *arg)
4540 {
4541         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4542         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4543         int ret = 0;
4544
4545         args32 = memdup_user(arg, sizeof(*args32));
4546         if (IS_ERR(args32))
4547                 return PTR_ERR(args32);
4548
4549         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4550         if (!args64) {
4551                 ret = -ENOMEM;
4552                 goto out;
4553         }
4554
4555         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4556         args64->stransid = args32->stransid;
4557         args64->rtransid = args32->rtransid;
4558         args64->stime.sec = args32->stime.sec;
4559         args64->stime.nsec = args32->stime.nsec;
4560         args64->rtime.sec = args32->rtime.sec;
4561         args64->rtime.nsec = args32->rtime.nsec;
4562         args64->flags = args32->flags;
4563
4564         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4565         if (ret)
4566                 goto out;
4567
4568         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4569         args32->stransid = args64->stransid;
4570         args32->rtransid = args64->rtransid;
4571         args32->stime.sec = args64->stime.sec;
4572         args32->stime.nsec = args64->stime.nsec;
4573         args32->rtime.sec = args64->rtime.sec;
4574         args32->rtime.nsec = args64->rtime.nsec;
4575         args32->flags = args64->flags;
4576
4577         ret = copy_to_user(arg, args32, sizeof(*args32));
4578         if (ret)
4579                 ret = -EFAULT;
4580
4581 out:
4582         kfree(args32);
4583         kfree(args64);
4584         return ret;
4585 }
4586 #endif
4587
4588 static long btrfs_ioctl_set_received_subvol(struct file *file,
4589                                             void __user *arg)
4590 {
4591         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4592         int ret = 0;
4593
4594         sa = memdup_user(arg, sizeof(*sa));
4595         if (IS_ERR(sa))
4596                 return PTR_ERR(sa);
4597
4598         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4599
4600         if (ret)
4601                 goto out;
4602
4603         ret = copy_to_user(arg, sa, sizeof(*sa));
4604         if (ret)
4605                 ret = -EFAULT;
4606
4607 out:
4608         kfree(sa);
4609         return ret;
4610 }
4611
4612 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4613                                         void __user *arg)
4614 {
4615         size_t len;
4616         int ret;
4617         char label[BTRFS_LABEL_SIZE];
4618
4619         spin_lock(&fs_info->super_lock);
4620         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4621         spin_unlock(&fs_info->super_lock);
4622
4623         len = strnlen(label, BTRFS_LABEL_SIZE);
4624
4625         if (len == BTRFS_LABEL_SIZE) {
4626                 btrfs_warn(fs_info,
4627                            "label is too long, return the first %zu bytes",
4628                            --len);
4629         }
4630
4631         ret = copy_to_user(arg, label, len);
4632
4633         return ret ? -EFAULT : 0;
4634 }
4635
4636 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4637 {
4638         struct inode *inode = file_inode(file);
4639         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4640         struct btrfs_root *root = BTRFS_I(inode)->root;
4641         struct btrfs_super_block *super_block = fs_info->super_copy;
4642         struct btrfs_trans_handle *trans;
4643         char label[BTRFS_LABEL_SIZE];
4644         int ret;
4645
4646         if (!capable(CAP_SYS_ADMIN))
4647                 return -EPERM;
4648
4649         if (copy_from_user(label, arg, sizeof(label)))
4650                 return -EFAULT;
4651
4652         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4653                 btrfs_err(fs_info,
4654                           "unable to set label with more than %d bytes",
4655                           BTRFS_LABEL_SIZE - 1);
4656                 return -EINVAL;
4657         }
4658
4659         ret = mnt_want_write_file(file);
4660         if (ret)
4661                 return ret;
4662
4663         trans = btrfs_start_transaction(root, 0);
4664         if (IS_ERR(trans)) {
4665                 ret = PTR_ERR(trans);
4666                 goto out_unlock;
4667         }
4668
4669         spin_lock(&fs_info->super_lock);
4670         strcpy(super_block->label, label);
4671         spin_unlock(&fs_info->super_lock);
4672         ret = btrfs_commit_transaction(trans);
4673
4674 out_unlock:
4675         mnt_drop_write_file(file);
4676         return ret;
4677 }
4678
4679 #define INIT_FEATURE_FLAGS(suffix) \
4680         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4681           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4682           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4683
4684 int btrfs_ioctl_get_supported_features(void __user *arg)
4685 {
4686         static const struct btrfs_ioctl_feature_flags features[3] = {
4687                 INIT_FEATURE_FLAGS(SUPP),
4688                 INIT_FEATURE_FLAGS(SAFE_SET),
4689                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4690         };
4691
4692         if (copy_to_user(arg, &features, sizeof(features)))
4693                 return -EFAULT;
4694
4695         return 0;
4696 }
4697
4698 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4699                                         void __user *arg)
4700 {
4701         struct btrfs_super_block *super_block = fs_info->super_copy;
4702         struct btrfs_ioctl_feature_flags features;
4703
4704         features.compat_flags = btrfs_super_compat_flags(super_block);
4705         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4706         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4707
4708         if (copy_to_user(arg, &features, sizeof(features)))
4709                 return -EFAULT;
4710
4711         return 0;
4712 }
4713
4714 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4715                               enum btrfs_feature_set set,
4716                               u64 change_mask, u64 flags, u64 supported_flags,
4717                               u64 safe_set, u64 safe_clear)
4718 {
4719         const char *type = btrfs_feature_set_name(set);
4720         char *names;
4721         u64 disallowed, unsupported;
4722         u64 set_mask = flags & change_mask;
4723         u64 clear_mask = ~flags & change_mask;
4724
4725         unsupported = set_mask & ~supported_flags;
4726         if (unsupported) {
4727                 names = btrfs_printable_features(set, unsupported);
4728                 if (names) {
4729                         btrfs_warn(fs_info,
4730                                    "this kernel does not support the %s feature bit%s",
4731                                    names, strchr(names, ',') ? "s" : "");
4732                         kfree(names);
4733                 } else
4734                         btrfs_warn(fs_info,
4735                                    "this kernel does not support %s bits 0x%llx",
4736                                    type, unsupported);
4737                 return -EOPNOTSUPP;
4738         }
4739
4740         disallowed = set_mask & ~safe_set;
4741         if (disallowed) {
4742                 names = btrfs_printable_features(set, disallowed);
4743                 if (names) {
4744                         btrfs_warn(fs_info,
4745                                    "can't set the %s feature bit%s while mounted",
4746                                    names, strchr(names, ',') ? "s" : "");
4747                         kfree(names);
4748                 } else
4749                         btrfs_warn(fs_info,
4750                                    "can't set %s bits 0x%llx while mounted",
4751                                    type, disallowed);
4752                 return -EPERM;
4753         }
4754
4755         disallowed = clear_mask & ~safe_clear;
4756         if (disallowed) {
4757                 names = btrfs_printable_features(set, disallowed);
4758                 if (names) {
4759                         btrfs_warn(fs_info,
4760                                    "can't clear the %s feature bit%s while mounted",
4761                                    names, strchr(names, ',') ? "s" : "");
4762                         kfree(names);
4763                 } else
4764                         btrfs_warn(fs_info,
4765                                    "can't clear %s bits 0x%llx while mounted",
4766                                    type, disallowed);
4767                 return -EPERM;
4768         }
4769
4770         return 0;
4771 }
4772
4773 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4774 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4775                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4776                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4777                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4778
4779 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4780 {
4781         struct inode *inode = file_inode(file);
4782         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4783         struct btrfs_root *root = BTRFS_I(inode)->root;
4784         struct btrfs_super_block *super_block = fs_info->super_copy;
4785         struct btrfs_ioctl_feature_flags flags[2];
4786         struct btrfs_trans_handle *trans;
4787         u64 newflags;
4788         int ret;
4789
4790         if (!capable(CAP_SYS_ADMIN))
4791                 return -EPERM;
4792
4793         if (copy_from_user(flags, arg, sizeof(flags)))
4794                 return -EFAULT;
4795
4796         /* Nothing to do */
4797         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4798             !flags[0].incompat_flags)
4799                 return 0;
4800
4801         ret = check_feature(fs_info, flags[0].compat_flags,
4802                             flags[1].compat_flags, COMPAT);
4803         if (ret)
4804                 return ret;
4805
4806         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4807                             flags[1].compat_ro_flags, COMPAT_RO);
4808         if (ret)
4809                 return ret;
4810
4811         ret = check_feature(fs_info, flags[0].incompat_flags,
4812                             flags[1].incompat_flags, INCOMPAT);
4813         if (ret)
4814                 return ret;
4815
4816         ret = mnt_want_write_file(file);
4817         if (ret)
4818                 return ret;
4819
4820         trans = btrfs_start_transaction(root, 0);
4821         if (IS_ERR(trans)) {
4822                 ret = PTR_ERR(trans);
4823                 goto out_drop_write;
4824         }
4825
4826         spin_lock(&fs_info->super_lock);
4827         newflags = btrfs_super_compat_flags(super_block);
4828         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4829         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4830         btrfs_set_super_compat_flags(super_block, newflags);
4831
4832         newflags = btrfs_super_compat_ro_flags(super_block);
4833         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4834         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4835         btrfs_set_super_compat_ro_flags(super_block, newflags);
4836
4837         newflags = btrfs_super_incompat_flags(super_block);
4838         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4839         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4840         btrfs_set_super_incompat_flags(super_block, newflags);
4841         spin_unlock(&fs_info->super_lock);
4842
4843         ret = btrfs_commit_transaction(trans);
4844 out_drop_write:
4845         mnt_drop_write_file(file);
4846
4847         return ret;
4848 }
4849
4850 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4851 {
4852         struct btrfs_ioctl_send_args *arg;
4853         int ret;
4854
4855         if (compat) {
4856 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4857                 struct btrfs_ioctl_send_args_32 args32;
4858
4859                 ret = copy_from_user(&args32, argp, sizeof(args32));
4860                 if (ret)
4861                         return -EFAULT;
4862                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4863                 if (!arg)
4864                         return -ENOMEM;
4865                 arg->send_fd = args32.send_fd;
4866                 arg->clone_sources_count = args32.clone_sources_count;
4867                 arg->clone_sources = compat_ptr(args32.clone_sources);
4868                 arg->parent_root = args32.parent_root;
4869                 arg->flags = args32.flags;
4870                 memcpy(arg->reserved, args32.reserved,
4871                        sizeof(args32.reserved));
4872 #else
4873                 return -ENOTTY;
4874 #endif
4875         } else {
4876                 arg = memdup_user(argp, sizeof(*arg));
4877                 if (IS_ERR(arg))
4878                         return PTR_ERR(arg);
4879         }
4880         ret = btrfs_ioctl_send(file, arg);
4881         kfree(arg);
4882         return ret;
4883 }
4884
4885 long btrfs_ioctl(struct file *file, unsigned int
4886                 cmd, unsigned long arg)
4887 {
4888         struct inode *inode = file_inode(file);
4889         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4890         struct btrfs_root *root = BTRFS_I(inode)->root;
4891         void __user *argp = (void __user *)arg;
4892
4893         switch (cmd) {
4894         case FS_IOC_GETVERSION:
4895                 return btrfs_ioctl_getversion(file, argp);
4896         case FS_IOC_GETFSLABEL:
4897                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4898         case FS_IOC_SETFSLABEL:
4899                 return btrfs_ioctl_set_fslabel(file, argp);
4900         case FITRIM:
4901                 return btrfs_ioctl_fitrim(fs_info, argp);
4902         case BTRFS_IOC_SNAP_CREATE:
4903                 return btrfs_ioctl_snap_create(file, argp, 0);
4904         case BTRFS_IOC_SNAP_CREATE_V2:
4905                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4906         case BTRFS_IOC_SUBVOL_CREATE:
4907                 return btrfs_ioctl_snap_create(file, argp, 1);
4908         case BTRFS_IOC_SUBVOL_CREATE_V2:
4909                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4910         case BTRFS_IOC_SNAP_DESTROY:
4911                 return btrfs_ioctl_snap_destroy(file, argp, false);
4912         case BTRFS_IOC_SNAP_DESTROY_V2:
4913                 return btrfs_ioctl_snap_destroy(file, argp, true);
4914         case BTRFS_IOC_SUBVOL_GETFLAGS:
4915                 return btrfs_ioctl_subvol_getflags(file, argp);
4916         case BTRFS_IOC_SUBVOL_SETFLAGS:
4917                 return btrfs_ioctl_subvol_setflags(file, argp);
4918         case BTRFS_IOC_DEFAULT_SUBVOL:
4919                 return btrfs_ioctl_default_subvol(file, argp);
4920         case BTRFS_IOC_DEFRAG:
4921                 return btrfs_ioctl_defrag(file, NULL);
4922         case BTRFS_IOC_DEFRAG_RANGE:
4923                 return btrfs_ioctl_defrag(file, argp);
4924         case BTRFS_IOC_RESIZE:
4925                 return btrfs_ioctl_resize(file, argp);
4926         case BTRFS_IOC_ADD_DEV:
4927                 return btrfs_ioctl_add_dev(fs_info, argp);
4928         case BTRFS_IOC_RM_DEV:
4929                 return btrfs_ioctl_rm_dev(file, argp);
4930         case BTRFS_IOC_RM_DEV_V2:
4931                 return btrfs_ioctl_rm_dev_v2(file, argp);
4932         case BTRFS_IOC_FS_INFO:
4933                 return btrfs_ioctl_fs_info(fs_info, argp);
4934         case BTRFS_IOC_DEV_INFO:
4935                 return btrfs_ioctl_dev_info(fs_info, argp);
4936         case BTRFS_IOC_BALANCE:
4937                 return btrfs_ioctl_balance(file, NULL);
4938         case BTRFS_IOC_TREE_SEARCH:
4939                 return btrfs_ioctl_tree_search(file, argp);
4940         case BTRFS_IOC_TREE_SEARCH_V2:
4941                 return btrfs_ioctl_tree_search_v2(file, argp);
4942         case BTRFS_IOC_INO_LOOKUP:
4943                 return btrfs_ioctl_ino_lookup(file, argp);
4944         case BTRFS_IOC_INO_PATHS:
4945                 return btrfs_ioctl_ino_to_path(root, argp);
4946         case BTRFS_IOC_LOGICAL_INO:
4947                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4948         case BTRFS_IOC_LOGICAL_INO_V2:
4949                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4950         case BTRFS_IOC_SPACE_INFO:
4951                 return btrfs_ioctl_space_info(fs_info, argp);
4952         case BTRFS_IOC_SYNC: {
4953                 int ret;
4954
4955                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4956                 if (ret)
4957                         return ret;
4958                 ret = btrfs_sync_fs(inode->i_sb, 1);
4959                 /*
4960                  * The transaction thread may want to do more work,
4961                  * namely it pokes the cleaner kthread that will start
4962                  * processing uncleaned subvols.
4963                  */
4964                 wake_up_process(fs_info->transaction_kthread);
4965                 return ret;
4966         }
4967         case BTRFS_IOC_START_SYNC:
4968                 return btrfs_ioctl_start_sync(root, argp);
4969         case BTRFS_IOC_WAIT_SYNC:
4970                 return btrfs_ioctl_wait_sync(fs_info, argp);
4971         case BTRFS_IOC_SCRUB:
4972                 return btrfs_ioctl_scrub(file, argp);
4973         case BTRFS_IOC_SCRUB_CANCEL:
4974                 return btrfs_ioctl_scrub_cancel(fs_info);
4975         case BTRFS_IOC_SCRUB_PROGRESS:
4976                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4977         case BTRFS_IOC_BALANCE_V2:
4978                 return btrfs_ioctl_balance(file, argp);
4979         case BTRFS_IOC_BALANCE_CTL:
4980                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4981         case BTRFS_IOC_BALANCE_PROGRESS:
4982                 return btrfs_ioctl_balance_progress(fs_info, argp);
4983         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4984                 return btrfs_ioctl_set_received_subvol(file, argp);
4985 #ifdef CONFIG_64BIT
4986         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4987                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4988 #endif
4989         case BTRFS_IOC_SEND:
4990                 return _btrfs_ioctl_send(file, argp, false);
4991 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4992         case BTRFS_IOC_SEND_32:
4993                 return _btrfs_ioctl_send(file, argp, true);
4994 #endif
4995         case BTRFS_IOC_GET_DEV_STATS:
4996                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4997         case BTRFS_IOC_QUOTA_CTL:
4998                 return btrfs_ioctl_quota_ctl(file, argp);
4999         case BTRFS_IOC_QGROUP_ASSIGN:
5000                 return btrfs_ioctl_qgroup_assign(file, argp);
5001         case BTRFS_IOC_QGROUP_CREATE:
5002                 return btrfs_ioctl_qgroup_create(file, argp);
5003         case BTRFS_IOC_QGROUP_LIMIT:
5004                 return btrfs_ioctl_qgroup_limit(file, argp);
5005         case BTRFS_IOC_QUOTA_RESCAN:
5006                 return btrfs_ioctl_quota_rescan(file, argp);
5007         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5008                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5009         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5010                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5011         case BTRFS_IOC_DEV_REPLACE:
5012                 return btrfs_ioctl_dev_replace(fs_info, argp);
5013         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5014                 return btrfs_ioctl_get_supported_features(argp);
5015         case BTRFS_IOC_GET_FEATURES:
5016                 return btrfs_ioctl_get_features(fs_info, argp);
5017         case BTRFS_IOC_SET_FEATURES:
5018                 return btrfs_ioctl_set_features(file, argp);
5019         case BTRFS_IOC_GET_SUBVOL_INFO:
5020                 return btrfs_ioctl_get_subvol_info(file, argp);
5021         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5022                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5023         case BTRFS_IOC_INO_LOOKUP_USER:
5024                 return btrfs_ioctl_ino_lookup_user(file, argp);
5025         case FS_IOC_ENABLE_VERITY:
5026                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5027         case FS_IOC_MEASURE_VERITY:
5028                 return fsverity_ioctl_measure(file, argp);
5029         }
5030
5031         return -ENOTTY;
5032 }
5033
5034 #ifdef CONFIG_COMPAT
5035 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5036 {
5037         /*
5038          * These all access 32-bit values anyway so no further
5039          * handling is necessary.
5040          */
5041         switch (cmd) {
5042         case FS_IOC32_GETVERSION:
5043                 cmd = FS_IOC_GETVERSION;
5044                 break;
5045         }
5046
5047         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5048 }
5049 #endif