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). Also no need to have the tree block locked since it
 620                  * is not in any tree at this point, so no other task can find
 621                  * it and use it.
 622                  */
 623                 btrfs_free_tree_block(trans, objectid, leaf, 0, 1);
 624                 free_extent_buffer(leaf);
 625                 goto fail;
 626         }
 627
 628         free_extent_buffer(leaf);
 629         leaf = NULL;
 630
 631         key.offset = (u64)-1;
 632         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 633         if (IS_ERR(new_root)) {
 634                 free_anon_bdev(anon_dev);
 635                 ret = PTR_ERR(new_root);
 636                 btrfs_abort_transaction(trans, ret);
 637                 goto fail;
 638         }
 639         /* Freeing will be done in btrfs_put_root() of new_root */
 640         anon_dev = 0;
 641
 642         ret = btrfs_record_root_in_trans(trans, new_root);
 643         if (ret) {
 644                 btrfs_put_root(new_root);
 645                 btrfs_abort_transaction(trans, ret);
 646                 goto fail;
 647         }
 648
 649         ret = btrfs_create_subvol_root(trans, new_root, root, mnt_userns);
 650         btrfs_put_root(new_root);
 651         if (ret) {
 652                 /* We potentially lose an unused inode item here */
 653                 btrfs_abort_transaction(trans, ret);
 654                 goto fail;
 655         }
 656
 657         /*
 658          * insert the directory item
 659          */
 660         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 661         if (ret) {
 662                 btrfs_abort_transaction(trans, ret);
 663                 goto fail;
 664         }
 665
 666         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 667                                     BTRFS_FT_DIR, index);
 668         if (ret) {
 669                 btrfs_abort_transaction(trans, ret);
 670                 goto fail;
 671         }
 672
 673         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 674         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 675         if (ret) {
 676                 btrfs_abort_transaction(trans, ret);
 677                 goto fail;
 678         }
 679
 680         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 681                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 682         if (ret) {
 683                 btrfs_abort_transaction(trans, ret);
 684                 goto fail;
 685         }
 686
 687         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 688                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 689         if (ret)
 690                 btrfs_abort_transaction(trans, ret);
 691
 692 fail:
 693         kfree(root_item);
 694         trans->block_rsv = NULL;
 695         trans->bytes_reserved = 0;
 696         btrfs_subvolume_release_metadata(root, &block_rsv);
 697
 698         err = btrfs_commit_transaction(trans);
 699         if (err && !ret)
 700                 ret = err;
 701
 702         if (!ret) {
 703                 inode = btrfs_lookup_dentry(dir, dentry);
 704                 if (IS_ERR(inode))
 705                         return PTR_ERR(inode);
 706                 d_instantiate(dentry, inode);
 707         }
 708         return ret;
 709
 710 fail_free:
 711         if (anon_dev)
 712                 free_anon_bdev(anon_dev);
 713         kfree(root_item);
 714         return ret;
 715 }
 716
 717 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 718                            struct dentry *dentry, bool readonly,
 719                            struct btrfs_qgroup_inherit *inherit)
 720 {
 721         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 722         struct inode *inode;
 723         struct btrfs_pending_snapshot *pending_snapshot;
 724         struct btrfs_trans_handle *trans;
 725         int ret;
 726
 727         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 728                 return -EINVAL;
 729
 730         if (atomic_read(&root->nr_swapfiles)) {
 731                 btrfs_warn(fs_info,
 732                            "cannot snapshot subvolume with active swapfile");
 733                 return -ETXTBSY;
 734         }
 735
 736         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 737         if (!pending_snapshot)
 738                 return -ENOMEM;
 739
 740         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 741         if (ret < 0)
 742                 goto free_pending;
 743         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 744                         GFP_KERNEL);
 745         pending_snapshot->path = btrfs_alloc_path();
 746         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 747                 ret = -ENOMEM;
 748                 goto free_pending;
 749         }
 750
 751         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 752                              BTRFS_BLOCK_RSV_TEMP);
 753         /*
 754          * 1 - parent dir inode
 755          * 2 - dir entries
 756          * 1 - root item
 757          * 2 - root ref/backref
 758          * 1 - root of snapshot
 759          * 1 - UUID item
 760          */
 761         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 762                                         &pending_snapshot->block_rsv, 8,
 763                                         false);
 764         if (ret)
 765                 goto free_pending;
 766
 767         pending_snapshot->dentry = dentry;
 768         pending_snapshot->root = root;
 769         pending_snapshot->readonly = readonly;
 770         pending_snapshot->dir = dir;
 771         pending_snapshot->inherit = inherit;
 772
 773         trans = btrfs_start_transaction(root, 0);
 774         if (IS_ERR(trans)) {
 775                 ret = PTR_ERR(trans);
 776                 goto fail;
 777         }
 778
 779         trans->pending_snapshot = pending_snapshot;
 780
 781         ret = btrfs_commit_transaction(trans);
 782         if (ret)
 783                 goto fail;
 784
 785         ret = pending_snapshot->error;
 786         if (ret)
 787                 goto fail;
 788
 789         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 790         if (ret)
 791                 goto fail;
 792
 793         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 794         if (IS_ERR(inode)) {
 795                 ret = PTR_ERR(inode);
 796                 goto fail;
 797         }
 798
 799         d_instantiate(dentry, inode);
 800         ret = 0;
 801         pending_snapshot->anon_dev = 0;
 802 fail:
 803         /* Prevent double freeing of anon_dev */
 804         if (ret && pending_snapshot->snap)
 805                 pending_snapshot->snap->anon_dev = 0;
 806         btrfs_put_root(pending_snapshot->snap);
 807         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 808 free_pending:
 809         if (pending_snapshot->anon_dev)
 810                 free_anon_bdev(pending_snapshot->anon_dev);
 811         kfree(pending_snapshot->root_item);
 812         btrfs_free_path(pending_snapshot->path);
 813         kfree(pending_snapshot);
 814
 815         return ret;
 816 }
 817
 818 /*  copy of may_delete in fs/namei.c()
 819  *      Check whether we can remove a link victim from directory dir, check
 820  *  whether the type of victim is right.
 821  *  1. We can't do it if dir is read-only (done in permission())
 822  *  2. We should have write and exec permissions on dir
 823  *  3. We can't remove anything from append-only dir
 824  *  4. We can't do anything with immutable dir (done in permission())
 825  *  5. If the sticky bit on dir is set we should either
 826  *      a. be owner of dir, or
 827  *      b. be owner of victim, or
 828  *      c. have CAP_FOWNER capability
 829  *  6. If the victim is append-only or immutable we can't do anything with
 830  *     links pointing to it.
 831  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 832  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 833  *  9. We can't remove a root or mountpoint.
 834  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 835  *     nfs_async_unlink().
 836  */
 837
 838 static int btrfs_may_delete(struct user_namespace *mnt_userns,
 839                             struct inode *dir, struct dentry *victim, int isdir)
 840 {
 841         int error;
 842
 843         if (d_really_is_negative(victim))
 844                 return -ENOENT;
 845
 846         BUG_ON(d_inode(victim->d_parent) != dir);
 847         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 848
 849         error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 850         if (error)
 851                 return error;
 852         if (IS_APPEND(dir))
 853                 return -EPERM;
 854         if (check_sticky(mnt_userns, dir, d_inode(victim)) ||
 855             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 856             IS_SWAPFILE(d_inode(victim)))
 857                 return -EPERM;
 858         if (isdir) {
 859                 if (!d_is_dir(victim))
 860                         return -ENOTDIR;
 861                 if (IS_ROOT(victim))
 862                         return -EBUSY;
 863         } else if (d_is_dir(victim))
 864                 return -EISDIR;
 865         if (IS_DEADDIR(dir))
 866                 return -ENOENT;
 867         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 868                 return -EBUSY;
 869         return 0;
 870 }
 871
 872 /* copy of may_create in fs/namei.c() */
 873 static inline int btrfs_may_create(struct user_namespace *mnt_userns,
 874                                    struct inode *dir, struct dentry *child)
 875 {
 876         if (d_really_is_positive(child))
 877                 return -EEXIST;
 878         if (IS_DEADDIR(dir))
 879                 return -ENOENT;
 880         if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
 881                 return -EOVERFLOW;
 882         return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
 883 }
 884
 885 /*
 886  * Create a new subvolume below @parent.  This is largely modeled after
 887  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 888  * inside this filesystem so it's quite a bit simpler.
 889  */
 890 static noinline int btrfs_mksubvol(const struct path *parent,
 891                                    struct user_namespace *mnt_userns,
 892                                    const char *name, int namelen,
 893                                    struct btrfs_root *snap_src,
 894                                    bool readonly,
 895                                    struct btrfs_qgroup_inherit *inherit)
 896 {
 897         struct inode *dir = d_inode(parent->dentry);
 898         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 899         struct dentry *dentry;
 900         int error;
 901
 902         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 903         if (error == -EINTR)
 904                 return error;
 905
 906         dentry = lookup_one(mnt_userns, name, parent->dentry, namelen);
 907         error = PTR_ERR(dentry);
 908         if (IS_ERR(dentry))
 909                 goto out_unlock;
 910
 911         error = btrfs_may_create(mnt_userns, dir, dentry);
 912         if (error)
 913                 goto out_dput;
 914
 915         /*
 916          * even if this name doesn't exist, we may get hash collisions.
 917          * check for them now when we can safely fail
 918          */
 919         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 920                                                dir->i_ino, name,
 921                                                namelen);
 922         if (error)
 923                 goto out_dput;
 924
 925         down_read(&fs_info->subvol_sem);
 926
 927         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 928                 goto out_up_read;
 929
 930         if (snap_src)
 931                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 932         else
 933                 error = create_subvol(mnt_userns, dir, dentry, name, namelen, inherit);
 934
 935         if (!error)
 936                 fsnotify_mkdir(dir, dentry);
 937 out_up_read:
 938         up_read(&fs_info->subvol_sem);
 939 out_dput:
 940         dput(dentry);
 941 out_unlock:
 942         btrfs_inode_unlock(dir, 0);
 943         return error;
 944 }
 945
 946 static noinline int btrfs_mksnapshot(const struct path *parent,
 947                                    struct user_namespace *mnt_userns,
 948                                    const char *name, int namelen,
 949                                    struct btrfs_root *root,
 950                                    bool readonly,
 951                                    struct btrfs_qgroup_inherit *inherit)
 952 {
 953         int ret;
 954         bool snapshot_force_cow = false;
 955
 956         /*
 957          * Force new buffered writes to reserve space even when NOCOW is
 958          * possible. This is to avoid later writeback (running dealloc) to
 959          * fallback to COW mode and unexpectedly fail with ENOSPC.
 960          */
 961         btrfs_drew_read_lock(&root->snapshot_lock);
 962
 963         ret = btrfs_start_delalloc_snapshot(root, false);
 964         if (ret)
 965                 goto out;
 966
 967         /*
 968          * All previous writes have started writeback in NOCOW mode, so now
 969          * we force future writes to fallback to COW mode during snapshot
 970          * creation.
 971          */
 972         atomic_inc(&root->snapshot_force_cow);
 973         snapshot_force_cow = true;
 974
 975         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 976
 977         ret = btrfs_mksubvol(parent, mnt_userns, name, namelen,
 978                              root, readonly, inherit);
 979 out:
 980         if (snapshot_force_cow)
 981                 atomic_dec(&root->snapshot_force_cow);
 982         btrfs_drew_read_unlock(&root->snapshot_lock);
 983         return ret;
 984 }
 985
 986 /*
 987  * When we're defragging a range, we don't want to kick it off again
 988  * if it is really just waiting for delalloc to send it down.
 989  * If we find a nice big extent or delalloc range for the bytes in the
 990  * file you want to defrag, we return 0 to let you know to skip this
 991  * part of the file
 992  */
 993 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 994 {
 995         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 996         struct extent_map *em = NULL;
 997         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 998         u64 end;
 999
1000         read_lock(&em_tree->lock);
1001         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1002         read_unlock(&em_tree->lock);
1003
1004         if (em) {
1005                 end = extent_map_end(em);
1006                 free_extent_map(em);
1007                 if (end - offset > thresh)
1008                         return 0;
1009         }
1010         /* if we already have a nice delalloc here, just stop */
1011         thresh /= 2;
1012         end = count_range_bits(io_tree, &offset, offset + thresh,
1013                                thresh, EXTENT_DELALLOC, 1);
1014         if (end >= thresh)
1015                 return 0;
1016         return 1;
1017 }
1018
1019 /*
1020  * helper function to walk through a file and find extents
1021  * newer than a specific transid, and smaller than thresh.
1022  *
1023  * This is used by the defragging code to find new and small
1024  * extents
1025  */
1026 static int find_new_extents(struct btrfs_root *root,
1027                             struct inode *inode, u64 newer_than,
1028                             u64 *off, u32 thresh)
1029 {
1030         struct btrfs_path *path;
1031         struct btrfs_key min_key;
1032         struct extent_buffer *leaf;
1033         struct btrfs_file_extent_item *extent;
1034         int type;
1035         int ret;
1036         u64 ino = btrfs_ino(BTRFS_I(inode));
1037
1038         path = btrfs_alloc_path();
1039         if (!path)
1040                 return -ENOMEM;
1041
1042         min_key.objectid = ino;
1043         min_key.type = BTRFS_EXTENT_DATA_KEY;
1044         min_key.offset = *off;
1045
1046         while (1) {
1047                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1048                 if (ret != 0)
1049                         goto none;
1050 process_slot:
1051                 if (min_key.objectid != ino)
1052                         goto none;
1053                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1054                         goto none;
1055
1056                 leaf = path->nodes[0];
1057                 extent = btrfs_item_ptr(leaf, path->slots[0],
1058                                         struct btrfs_file_extent_item);
1059
1060                 type = btrfs_file_extent_type(leaf, extent);
1061                 if (type == BTRFS_FILE_EXTENT_REG &&
1062                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1063                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1064                         *off = min_key.offset;
1065                         btrfs_free_path(path);
1066                         return 0;
1067                 }
1068
1069                 path->slots[0]++;
1070                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1071                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1072                         goto process_slot;
1073                 }
1074
1075                 if (min_key.offset == (u64)-1)
1076                         goto none;
1077
1078                 min_key.offset++;
1079                 btrfs_release_path(path);
1080         }
1081 none:
1082         btrfs_free_path(path);
1083         return -ENOENT;
1084 }
1085
1086 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1087 {
1088         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1089         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1090         struct extent_map *em;
1091         u64 len = PAGE_SIZE;
1092
1093         /*
1094          * hopefully we have this extent in the tree already, try without
1095          * the full extent lock
1096          */
1097         read_lock(&em_tree->lock);
1098         em = lookup_extent_mapping(em_tree, start, len);
1099         read_unlock(&em_tree->lock);
1100
1101         if (!em) {
1102                 struct extent_state *cached = NULL;
1103                 u64 end = start + len - 1;
1104
1105                 /* get the big lock and read metadata off disk */
1106                 lock_extent_bits(io_tree, start, end, &cached);
1107                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1108                 unlock_extent_cached(io_tree, start, end, &cached);
1109
1110                 if (IS_ERR(em))
1111                         return NULL;
1112         }
1113
1114         return em;
1115 }
1116
1117 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1118 {
1119         struct extent_map *next;
1120         bool ret = true;
1121
1122         /* this is the last extent */
1123         if (em->start + em->len >= i_size_read(inode))
1124                 return false;
1125
1126         next = defrag_lookup_extent(inode, em->start + em->len);
1127         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1128                 ret = false;
1129         else if ((em->block_start + em->block_len == next->block_start) &&
1130                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1131                 ret = false;
1132
1133         free_extent_map(next);
1134         return ret;
1135 }
1136
1137 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1138                                u64 *last_len, u64 *skip, u64 *defrag_end,
1139                                int compress)
1140 {
1141         struct extent_map *em;
1142         int ret = 1;
1143         bool next_mergeable = true;
1144         bool prev_mergeable = true;
1145
1146         /*
1147          * make sure that once we start defragging an extent, we keep on
1148          * defragging it
1149          */
1150         if (start < *defrag_end)
1151                 return 1;
1152
1153         *skip = 0;
1154
1155         em = defrag_lookup_extent(inode, start);
1156         if (!em)
1157                 return 0;
1158
1159         /* this will cover holes, and inline extents */
1160         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1161                 ret = 0;
1162                 goto out;
1163         }
1164
1165         if (!*defrag_end)
1166                 prev_mergeable = false;
1167
1168         next_mergeable = defrag_check_next_extent(inode, em);
1169         /*
1170          * we hit a real extent, if it is big or the next extent is not a
1171          * real extent, don't bother defragging it
1172          */
1173         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1174             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1175                 ret = 0;
1176 out:
1177         /*
1178          * last_len ends up being a counter of how many bytes we've defragged.
1179          * every time we choose not to defrag an extent, we reset *last_len
1180          * so that the next tiny extent will force a defrag.
1181          *
1182          * The end result of this is that tiny extents before a single big
1183          * extent will force at least part of that big extent to be defragged.
1184          */
1185         if (ret) {
1186                 *defrag_end = extent_map_end(em);
1187         } else {
1188                 *last_len = 0;
1189                 *skip = extent_map_end(em);
1190                 *defrag_end = 0;
1191         }
1192
1193         free_extent_map(em);
1194         return ret;
1195 }
1196
1197 /*
1198  * it doesn't do much good to defrag one or two pages
1199  * at a time.  This pulls in a nice chunk of pages
1200  * to COW and defrag.
1201  *
1202  * It also makes sure the delalloc code has enough
1203  * dirty data to avoid making new small extents as part
1204  * of the defrag
1205  *
1206  * It's a good idea to start RA on this range
1207  * before calling this.
1208  */
1209 static int cluster_pages_for_defrag(struct inode *inode,
1210                                     struct page **pages,
1211                                     unsigned long start_index,
1212                                     unsigned long num_pages)
1213 {
1214         unsigned long file_end;
1215         u64 isize = i_size_read(inode);
1216         u64 page_start;
1217         u64 page_end;
1218         u64 page_cnt;
1219         u64 start = (u64)start_index << PAGE_SHIFT;
1220         u64 search_start;
1221         int ret;
1222         int i;
1223         int i_done;
1224         struct btrfs_ordered_extent *ordered;
1225         struct extent_state *cached_state = NULL;
1226         struct extent_io_tree *tree;
1227         struct extent_changeset *data_reserved = NULL;
1228         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1229
1230         file_end = (isize - 1) >> PAGE_SHIFT;
1231         if (!isize || start_index > file_end)
1232                 return 0;
1233
1234         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1235
1236         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1237                         start, page_cnt << PAGE_SHIFT);
1238         if (ret)
1239                 return ret;
1240         i_done = 0;
1241         tree = &BTRFS_I(inode)->io_tree;
1242
1243         /* step one, lock all the pages */
1244         for (i = 0; i < page_cnt; i++) {
1245                 struct page *page;
1246 again:
1247                 page = find_or_create_page(inode->i_mapping,
1248                                            start_index + i, mask);
1249                 if (!page)
1250                         break;
1251
1252                 ret = set_page_extent_mapped(page);
1253                 if (ret < 0) {
1254                         unlock_page(page);
1255                         put_page(page);
1256                         break;
1257                 }
1258
1259                 page_start = page_offset(page);
1260                 page_end = page_start + PAGE_SIZE - 1;
1261                 while (1) {
1262                         lock_extent_bits(tree, page_start, page_end,
1263                                          &cached_state);
1264                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1265                                                               page_start);
1266                         unlock_extent_cached(tree, page_start, page_end,
1267                                              &cached_state);
1268                         if (!ordered)
1269                                 break;
1270
1271                         unlock_page(page);
1272                         btrfs_start_ordered_extent(ordered, 1);
1273                         btrfs_put_ordered_extent(ordered);
1274                         lock_page(page);
1275                         /*
1276                          * we unlocked the page above, so we need check if
1277                          * it was released or not.
1278                          */
1279                         if (page->mapping != inode->i_mapping) {
1280                                 unlock_page(page);
1281                                 put_page(page);
1282                                 goto again;
1283                         }
1284                 }
1285
1286                 if (!PageUptodate(page)) {
1287                         btrfs_readpage(NULL, page);
1288                         lock_page(page);
1289                         if (!PageUptodate(page)) {
1290                                 unlock_page(page);
1291                                 put_page(page);
1292                                 ret = -EIO;
1293                                 break;
1294                         }
1295                 }
1296
1297                 if (page->mapping != inode->i_mapping) {
1298                         unlock_page(page);
1299                         put_page(page);
1300                         goto again;
1301                 }
1302
1303                 pages[i] = page;
1304                 i_done++;
1305         }
1306         if (!i_done || ret)
1307                 goto out;
1308
1309         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1310                 goto out;
1311
1312         /*
1313          * so now we have a nice long stream of locked
1314          * and up to date pages, lets wait on them
1315          */
1316         for (i = 0; i < i_done; i++)
1317                 wait_on_page_writeback(pages[i]);
1318
1319         page_start = page_offset(pages[0]);
1320         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1321
1322         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1323                          page_start, page_end - 1, &cached_state);
1324
1325         /*
1326          * When defragmenting we skip ranges that have holes or inline extents,
1327          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1328          * space. At btrfs_defrag_file(), we check if a range should be defragged
1329          * before locking the inode and then, if it should, we trigger a sync
1330          * page cache readahead - we lock the inode only after that to avoid
1331          * blocking for too long other tasks that possibly want to operate on
1332          * other file ranges. But before we were able to get the inode lock,
1333          * some other task may have punched a hole in the range, or we may have
1334          * now an inline extent, in which case we should not defrag. So check
1335          * for that here, where we have the inode and the range locked, and bail
1336          * out if that happened.
1337          */
1338         search_start = page_start;
1339         while (search_start < page_end) {
1340                 struct extent_map *em;
1341
1342                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1343                                       page_end - search_start);
1344                 if (IS_ERR(em)) {
1345                         ret = PTR_ERR(em);
1346                         goto out_unlock_range;
1347                 }
1348                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1349                         free_extent_map(em);
1350                         /* Ok, 0 means we did not defrag anything */
1351                         ret = 0;
1352                         goto out_unlock_range;
1353                 }
1354                 search_start = extent_map_end(em);
1355                 free_extent_map(em);
1356         }
1357
1358         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1359                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1360                           EXTENT_DEFRAG, 0, 0, &cached_state);
1361
1362         if (i_done != page_cnt) {
1363                 spin_lock(&BTRFS_I(inode)->lock);
1364                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1365                 spin_unlock(&BTRFS_I(inode)->lock);
1366                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1367                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1368         }
1369
1370
1371         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1372                           &cached_state);
1373
1374         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1375                              page_start, page_end - 1, &cached_state);
1376
1377         for (i = 0; i < i_done; i++) {
1378                 clear_page_dirty_for_io(pages[i]);
1379                 ClearPageChecked(pages[i]);
1380                 set_page_dirty(pages[i]);
1381                 unlock_page(pages[i]);
1382                 put_page(pages[i]);
1383         }
1384         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1385         extent_changeset_free(data_reserved);
1386         return i_done;
1387
1388 out_unlock_range:
1389         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1390                              page_start, page_end - 1, &cached_state);
1391 out:
1392         for (i = 0; i < i_done; i++) {
1393                 unlock_page(pages[i]);
1394                 put_page(pages[i]);
1395         }
1396         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1397                         start, page_cnt << PAGE_SHIFT, true);
1398         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1399         extent_changeset_free(data_reserved);
1400         return ret;
1401
1402 }
1403
1404 int btrfs_defrag_file(struct inode *inode, struct file *file,
1405                       struct btrfs_ioctl_defrag_range_args *range,
1406                       u64 newer_than, unsigned long max_to_defrag)
1407 {
1408         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410         struct file_ra_state *ra = NULL;
1411         unsigned long last_index;
1412         u64 isize = i_size_read(inode);
1413         u64 last_len = 0;
1414         u64 skip = 0;
1415         u64 defrag_end = 0;
1416         u64 newer_off = range->start;
1417         unsigned long i;
1418         unsigned long ra_index = 0;
1419         int ret;
1420         int defrag_count = 0;
1421         int compress_type = BTRFS_COMPRESS_ZLIB;
1422         u32 extent_thresh = range->extent_thresh;
1423         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1424         unsigned long cluster = max_cluster;
1425         u64 new_align = ~((u64)SZ_128K - 1);
1426         struct page **pages = NULL;
1427         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1428
1429         if (isize == 0)
1430                 return 0;
1431
1432         if (range->start >= isize)
1433                 return -EINVAL;
1434
1435         if (do_compress) {
1436                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1437                         return -EINVAL;
1438                 if (range->compress_type)
1439                         compress_type = range->compress_type;
1440         }
1441
1442         if (extent_thresh == 0)
1443                 extent_thresh = SZ_256K;
1444
1445         /*
1446          * If we were not given a file, allocate a readahead context. As
1447          * readahead is just an optimization, defrag will work without it so
1448          * we don't error out.
1449          */
1450         if (!file) {
1451                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1452                 if (ra)
1453                         file_ra_state_init(ra, inode->i_mapping);
1454         } else {
1455                 ra = &file->f_ra;
1456         }
1457
1458         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1459         if (!pages) {
1460                 ret = -ENOMEM;
1461                 goto out_ra;
1462         }
1463
1464         /* find the last page to defrag */
1465         if (range->start + range->len > range->start) {
1466                 last_index = min_t(u64, isize - 1,
1467                          range->start + range->len - 1) >> PAGE_SHIFT;
1468         } else {
1469                 last_index = (isize - 1) >> PAGE_SHIFT;
1470         }
1471
1472         if (newer_than) {
1473                 ret = find_new_extents(root, inode, newer_than,
1474                                        &newer_off, SZ_64K);
1475                 if (!ret) {
1476                         range->start = newer_off;
1477                         /*
1478                          * we always align our defrag to help keep
1479                          * the extents in the file evenly spaced
1480                          */
1481                         i = (newer_off & new_align) >> PAGE_SHIFT;
1482                 } else
1483                         goto out_ra;
1484         } else {
1485                 i = range->start >> PAGE_SHIFT;
1486         }
1487         if (!max_to_defrag)
1488                 max_to_defrag = last_index - i + 1;
1489
1490         /*
1491          * make writeback starts from i, so the defrag range can be
1492          * written sequentially.
1493          */
1494         if (i < inode->i_mapping->writeback_index)
1495                 inode->i_mapping->writeback_index = i;
1496
1497         while (i <= last_index && defrag_count < max_to_defrag &&
1498                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1499                 /*
1500                  * make sure we stop running if someone unmounts
1501                  * the FS
1502                  */
1503                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1504                         break;
1505
1506                 if (btrfs_defrag_cancelled(fs_info)) {
1507                         btrfs_debug(fs_info, "defrag_file cancelled");
1508                         ret = -EAGAIN;
1509                         goto error;
1510                 }
1511
1512                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1513                                          extent_thresh, &last_len, &skip,
1514                                          &defrag_end, do_compress)){
1515                         unsigned long next;
1516                         /*
1517                          * the should_defrag function tells us how much to skip
1518                          * bump our counter by the suggested amount
1519                          */
1520                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1521                         i = max(i + 1, next);
1522                         continue;
1523                 }
1524
1525                 if (!newer_than) {
1526                         cluster = (PAGE_ALIGN(defrag_end) >>
1527                                    PAGE_SHIFT) - i;
1528                         cluster = min(cluster, max_cluster);
1529                 } else {
1530                         cluster = max_cluster;
1531                 }
1532
1533                 if (i + cluster > ra_index) {
1534                         ra_index = max(i, ra_index);
1535                         if (ra)
1536                                 page_cache_sync_readahead(inode->i_mapping, ra,
1537                                                 file, ra_index, cluster);
1538                         ra_index += cluster;
1539                 }
1540
1541                 btrfs_inode_lock(inode, 0);
1542                 if (IS_SWAPFILE(inode)) {
1543                         ret = -ETXTBSY;
1544                 } else {
1545                         if (do_compress)
1546                                 BTRFS_I(inode)->defrag_compress = compress_type;
1547                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1548                 }
1549                 if (ret < 0) {
1550                         btrfs_inode_unlock(inode, 0);
1551                         goto out_ra;
1552                 }
1553
1554                 defrag_count += ret;
1555                 balance_dirty_pages_ratelimited(inode->i_mapping);
1556                 btrfs_inode_unlock(inode, 0);
1557
1558                 if (newer_than) {
1559                         if (newer_off == (u64)-1)
1560                                 break;
1561
1562                         if (ret > 0)
1563                                 i += ret;
1564
1565                         newer_off = max(newer_off + 1,
1566                                         (u64)i << PAGE_SHIFT);
1567
1568                         ret = find_new_extents(root, inode, newer_than,
1569                                                &newer_off, SZ_64K);
1570                         if (!ret) {
1571                                 range->start = newer_off;
1572                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1573                         } else {
1574                                 break;
1575                         }
1576                 } else {
1577                         if (ret > 0) {
1578                                 i += ret;
1579                                 last_len += ret << PAGE_SHIFT;
1580                         } else {
1581                                 i++;
1582                                 last_len = 0;
1583                         }
1584                 }
1585         }
1586
1587         ret = defrag_count;
1588 error:
1589         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1590                 filemap_flush(inode->i_mapping);
1591                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1592                              &BTRFS_I(inode)->runtime_flags))
1593                         filemap_flush(inode->i_mapping);
1594         }
1595
1596         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1597                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1598         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1599                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1600         }
1601
1602 out_ra:
1603         if (do_compress) {
1604                 btrfs_inode_lock(inode, 0);
1605                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1606                 btrfs_inode_unlock(inode, 0);
1607         }
1608         if (!file)
1609                 kfree(ra);
1610         kfree(pages);
1611         return ret;
1612 }
1613
1614 /*
1615  * Try to start exclusive operation @type or cancel it if it's running.
1616  *
1617  * Return:
1618  *   0        - normal mode, newly claimed op started
1619  *  >0        - normal mode, something else is running,
1620  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1621  * ECANCELED  - cancel mode, successful cancel
1622  * ENOTCONN   - cancel mode, operation not running anymore
1623  */
1624 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1625                         enum btrfs_exclusive_operation type, bool cancel)
1626 {
1627         if (!cancel) {
1628                 /* Start normal op */
1629                 if (!btrfs_exclop_start(fs_info, type))
1630                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1631                 /* Exclusive operation is now claimed */
1632                 return 0;
1633         }
1634
1635         /* Cancel running op */
1636         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1637                 /*
1638                  * This blocks any exclop finish from setting it to NONE, so we
1639                  * request cancellation. Either it runs and we will wait for it,
1640                  * or it has finished and no waiting will happen.
1641                  */
1642                 atomic_inc(&fs_info->reloc_cancel_req);
1643                 btrfs_exclop_start_unlock(fs_info);
1644
1645                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1646                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1647                                     TASK_INTERRUPTIBLE);
1648
1649                 return -ECANCELED;
1650         }
1651
1652         /* Something else is running or none */
1653         return -ENOTCONN;
1654 }
1655
1656 static noinline int btrfs_ioctl_resize(struct file *file,
1657                                         void __user *arg)
1658 {
1659         struct inode *inode = file_inode(file);
1660         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1661         u64 new_size;
1662         u64 old_size;
1663         u64 devid = 1;
1664         struct btrfs_root *root = BTRFS_I(inode)->root;
1665         struct btrfs_ioctl_vol_args *vol_args;
1666         struct btrfs_trans_handle *trans;
1667         struct btrfs_device *device = NULL;
1668         char *sizestr;
1669         char *retptr;
1670         char *devstr = NULL;
1671         int ret = 0;
1672         int mod = 0;
1673         bool cancel;
1674
1675         if (!capable(CAP_SYS_ADMIN))
1676                 return -EPERM;
1677
1678         ret = mnt_want_write_file(file);
1679         if (ret)
1680                 return ret;
1681
1682         /*
1683          * Read the arguments before checking exclusivity to be able to
1684          * distinguish regular resize and cancel
1685          */
1686         vol_args = memdup_user(arg, sizeof(*vol_args));
1687         if (IS_ERR(vol_args)) {
1688                 ret = PTR_ERR(vol_args);
1689                 goto out_drop;
1690         }
1691         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1692         sizestr = vol_args->name;
1693         cancel = (strcmp("cancel", sizestr) == 0);
1694         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1695         if (ret)
1696                 goto out_free;
1697         /* Exclusive operation is now claimed */
1698
1699         devstr = strchr(sizestr, ':');
1700         if (devstr) {
1701                 sizestr = devstr + 1;
1702                 *devstr = '\0';
1703                 devstr = vol_args->name;
1704                 ret = kstrtoull(devstr, 10, &devid);
1705                 if (ret)
1706                         goto out_finish;
1707                 if (!devid) {
1708                         ret = -EINVAL;
1709                         goto out_finish;
1710                 }
1711                 btrfs_info(fs_info, "resizing devid %llu", devid);
1712         }
1713
1714         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
1715         if (!device) {
1716                 btrfs_info(fs_info, "resizer unable to find device %llu",
1717                            devid);
1718                 ret = -ENODEV;
1719                 goto out_finish;
1720         }
1721
1722         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1723                 btrfs_info(fs_info,
1724                            "resizer unable to apply on readonly device %llu",
1725                        devid);
1726                 ret = -EPERM;
1727                 goto out_finish;
1728         }
1729
1730         if (!strcmp(sizestr, "max"))
1731                 new_size = device->bdev->bd_inode->i_size;
1732         else {
1733                 if (sizestr[0] == '-') {
1734                         mod = -1;
1735                         sizestr++;
1736                 } else if (sizestr[0] == '+') {
1737                         mod = 1;
1738                         sizestr++;
1739                 }
1740                 new_size = memparse(sizestr, &retptr);
1741                 if (*retptr != '\0' || new_size == 0) {
1742                         ret = -EINVAL;
1743                         goto out_finish;
1744                 }
1745         }
1746
1747         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1748                 ret = -EPERM;
1749                 goto out_finish;
1750         }
1751
1752         old_size = btrfs_device_get_total_bytes(device);
1753
1754         if (mod < 0) {
1755                 if (new_size > old_size) {
1756                         ret = -EINVAL;
1757                         goto out_finish;
1758                 }
1759                 new_size = old_size - new_size;
1760         } else if (mod > 0) {
1761                 if (new_size > ULLONG_MAX - old_size) {
1762                         ret = -ERANGE;
1763                         goto out_finish;
1764                 }
1765                 new_size = old_size + new_size;
1766         }
1767
1768         if (new_size < SZ_256M) {
1769                 ret = -EINVAL;
1770                 goto out_finish;
1771         }
1772         if (new_size > device->bdev->bd_inode->i_size) {
1773                 ret = -EFBIG;
1774                 goto out_finish;
1775         }
1776
1777         new_size = round_down(new_size, fs_info->sectorsize);
1778
1779         if (new_size > old_size) {
1780                 trans = btrfs_start_transaction(root, 0);
1781                 if (IS_ERR(trans)) {
1782                         ret = PTR_ERR(trans);
1783                         goto out_finish;
1784                 }
1785                 ret = btrfs_grow_device(trans, device, new_size);
1786                 btrfs_commit_transaction(trans);
1787         } else if (new_size < old_size) {
1788                 ret = btrfs_shrink_device(device, new_size);
1789         } /* equal, nothing need to do */
1790
1791         if (ret == 0 && new_size != old_size)
1792                 btrfs_info_in_rcu(fs_info,
1793                         "resize device %s (devid %llu) from %llu to %llu",
1794                         rcu_str_deref(device->name), device->devid,
1795                         old_size, new_size);
1796 out_finish:
1797         btrfs_exclop_finish(fs_info);
1798 out_free:
1799         kfree(vol_args);
1800 out_drop:
1801         mnt_drop_write_file(file);
1802         return ret;
1803 }
1804
1805 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1806                                 struct user_namespace *mnt_userns,
1807                                 const char *name, unsigned long fd, int subvol,
1808                                 bool readonly,
1809                                 struct btrfs_qgroup_inherit *inherit)
1810 {
1811         int namelen;
1812         int ret = 0;
1813
1814         if (!S_ISDIR(file_inode(file)->i_mode))
1815                 return -ENOTDIR;
1816
1817         ret = mnt_want_write_file(file);
1818         if (ret)
1819                 goto out;
1820
1821         namelen = strlen(name);
1822         if (strchr(name, '/')) {
1823                 ret = -EINVAL;
1824                 goto out_drop_write;
1825         }
1826
1827         if (name[0] == '.' &&
1828            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1829                 ret = -EEXIST;
1830                 goto out_drop_write;
1831         }
1832
1833         if (subvol) {
1834                 ret = btrfs_mksubvol(&file->f_path, mnt_userns, name,
1835                                      namelen, NULL, readonly, inherit);
1836         } else {
1837                 struct fd src = fdget(fd);
1838                 struct inode *src_inode;
1839                 if (!src.file) {
1840                         ret = -EINVAL;
1841                         goto out_drop_write;
1842                 }
1843
1844                 src_inode = file_inode(src.file);
1845                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1846                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1847                                    "Snapshot src from another FS");
1848                         ret = -EXDEV;
1849                 } else if (!inode_owner_or_capable(mnt_userns, src_inode)) {
1850                         /*
1851                          * Subvolume creation is not restricted, but snapshots
1852                          * are limited to own subvolumes only
1853                          */
1854                         ret = -EPERM;
1855                 } else {
1856                         ret = btrfs_mksnapshot(&file->f_path, mnt_userns,
1857                                                name, namelen,
1858                                                BTRFS_I(src_inode)->root,
1859                                                readonly, inherit);
1860                 }
1861                 fdput(src);
1862         }
1863 out_drop_write:
1864         mnt_drop_write_file(file);
1865 out:
1866         return ret;
1867 }
1868
1869 static noinline int btrfs_ioctl_snap_create(struct file *file,
1870                                             void __user *arg, int subvol)
1871 {
1872         struct btrfs_ioctl_vol_args *vol_args;
1873         int ret;
1874
1875         if (!S_ISDIR(file_inode(file)->i_mode))
1876                 return -ENOTDIR;
1877
1878         vol_args = memdup_user(arg, sizeof(*vol_args));
1879         if (IS_ERR(vol_args))
1880                 return PTR_ERR(vol_args);
1881         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1882
1883         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
1884                                         vol_args->name, vol_args->fd, subvol,
1885                                         false, NULL);
1886
1887         kfree(vol_args);
1888         return ret;
1889 }
1890
1891 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1892                                                void __user *arg, int subvol)
1893 {
1894         struct btrfs_ioctl_vol_args_v2 *vol_args;
1895         int ret;
1896         bool readonly = false;
1897         struct btrfs_qgroup_inherit *inherit = NULL;
1898
1899         if (!S_ISDIR(file_inode(file)->i_mode))
1900                 return -ENOTDIR;
1901
1902         vol_args = memdup_user(arg, sizeof(*vol_args));
1903         if (IS_ERR(vol_args))
1904                 return PTR_ERR(vol_args);
1905         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1906
1907         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1908                 ret = -EOPNOTSUPP;
1909                 goto free_args;
1910         }
1911
1912         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1913                 readonly = true;
1914         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1915                 u64 nums;
1916
1917                 if (vol_args->size < sizeof(*inherit) ||
1918                     vol_args->size > PAGE_SIZE) {
1919                         ret = -EINVAL;
1920                         goto free_args;
1921                 }
1922                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1923                 if (IS_ERR(inherit)) {
1924                         ret = PTR_ERR(inherit);
1925                         goto free_args;
1926                 }
1927
1928                 if (inherit->num_qgroups > PAGE_SIZE ||
1929                     inherit->num_ref_copies > PAGE_SIZE ||
1930                     inherit->num_excl_copies > PAGE_SIZE) {
1931                         ret = -EINVAL;
1932                         goto free_inherit;
1933                 }
1934
1935                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1936                        2 * inherit->num_excl_copies;
1937                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1938                         ret = -EINVAL;
1939                         goto free_inherit;
1940                 }
1941         }
1942
1943         ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file),
1944                                         vol_args->name, vol_args->fd, subvol,
1945                                         readonly, inherit);
1946         if (ret)
1947                 goto free_inherit;
1948 free_inherit:
1949         kfree(inherit);
1950 free_args:
1951         kfree(vol_args);
1952         return ret;
1953 }
1954
1955 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1956                                                 void __user *arg)
1957 {
1958         struct inode *inode = file_inode(file);
1959         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1960         struct btrfs_root *root = BTRFS_I(inode)->root;
1961         int ret = 0;
1962         u64 flags = 0;
1963
1964         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1965                 return -EINVAL;
1966
1967         down_read(&fs_info->subvol_sem);
1968         if (btrfs_root_readonly(root))
1969                 flags |= BTRFS_SUBVOL_RDONLY;
1970         up_read(&fs_info->subvol_sem);
1971
1972         if (copy_to_user(arg, &flags, sizeof(flags)))
1973                 ret = -EFAULT;
1974
1975         return ret;
1976 }
1977
1978 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1979                                               void __user *arg)
1980 {
1981         struct inode *inode = file_inode(file);
1982         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1983         struct btrfs_root *root = BTRFS_I(inode)->root;
1984         struct btrfs_trans_handle *trans;
1985         u64 root_flags;
1986         u64 flags;
1987         int ret = 0;
1988
1989         if (!inode_owner_or_capable(file_mnt_user_ns(file), inode))
1990                 return -EPERM;
1991
1992         ret = mnt_want_write_file(file);
1993         if (ret)
1994                 goto out;
1995
1996         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1997                 ret = -EINVAL;
1998                 goto out_drop_write;
1999         }
2000
2001         if (copy_from_user(&flags, arg, sizeof(flags))) {
2002                 ret = -EFAULT;
2003                 goto out_drop_write;
2004         }
2005
2006         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2007                 ret = -EOPNOTSUPP;
2008                 goto out_drop_write;
2009         }
2010
2011         down_write(&fs_info->subvol_sem);
2012
2013         /* nothing to do */
2014         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2015                 goto out_drop_sem;
2016
2017         root_flags = btrfs_root_flags(&root->root_item);
2018         if (flags & BTRFS_SUBVOL_RDONLY) {
2019                 btrfs_set_root_flags(&root->root_item,
2020                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2021         } else {
2022                 /*
2023                  * Block RO -> RW transition if this subvolume is involved in
2024                  * send
2025                  */
2026                 spin_lock(&root->root_item_lock);
2027                 if (root->send_in_progress == 0) {
2028                         btrfs_set_root_flags(&root->root_item,
2029                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2030                         spin_unlock(&root->root_item_lock);
2031                 } else {
2032                         spin_unlock(&root->root_item_lock);
2033                         btrfs_warn(fs_info,
2034                                    "Attempt to set subvolume %llu read-write during send",
2035                                    root->root_key.objectid);
2036                         ret = -EPERM;
2037                         goto out_drop_sem;
2038                 }
2039         }
2040
2041         trans = btrfs_start_transaction(root, 1);
2042         if (IS_ERR(trans)) {
2043                 ret = PTR_ERR(trans);
2044                 goto out_reset;
2045         }
2046
2047         ret = btrfs_update_root(trans, fs_info->tree_root,
2048                                 &root->root_key, &root->root_item);
2049         if (ret < 0) {
2050                 btrfs_end_transaction(trans);
2051                 goto out_reset;
2052         }
2053
2054         ret = btrfs_commit_transaction(trans);
2055
2056 out_reset:
2057         if (ret)
2058                 btrfs_set_root_flags(&root->root_item, root_flags);
2059 out_drop_sem:
2060         up_write(&fs_info->subvol_sem);
2061 out_drop_write:
2062         mnt_drop_write_file(file);
2063 out:
2064         return ret;
2065 }
2066
2067 static noinline int key_in_sk(struct btrfs_key *key,
2068                               struct btrfs_ioctl_search_key *sk)
2069 {
2070         struct btrfs_key test;
2071         int ret;
2072
2073         test.objectid = sk->min_objectid;
2074         test.type = sk->min_type;
2075         test.offset = sk->min_offset;
2076
2077         ret = btrfs_comp_cpu_keys(key, &test);
2078         if (ret < 0)
2079                 return 0;
2080
2081         test.objectid = sk->max_objectid;
2082         test.type = sk->max_type;
2083         test.offset = sk->max_offset;
2084
2085         ret = btrfs_comp_cpu_keys(key, &test);
2086         if (ret > 0)
2087                 return 0;
2088         return 1;
2089 }
2090
2091 static noinline int copy_to_sk(struct btrfs_path *path,
2092                                struct btrfs_key *key,
2093                                struct btrfs_ioctl_search_key *sk,
2094                                size_t *buf_size,
2095                                char __user *ubuf,
2096                                unsigned long *sk_offset,
2097                                int *num_found)
2098 {
2099         u64 found_transid;
2100         struct extent_buffer *leaf;
2101         struct btrfs_ioctl_search_header sh;
2102         struct btrfs_key test;
2103         unsigned long item_off;
2104         unsigned long item_len;
2105         int nritems;
2106         int i;
2107         int slot;
2108         int ret = 0;
2109
2110         leaf = path->nodes[0];
2111         slot = path->slots[0];
2112         nritems = btrfs_header_nritems(leaf);
2113
2114         if (btrfs_header_generation(leaf) > sk->max_transid) {
2115                 i = nritems;
2116                 goto advance_key;
2117         }
2118         found_transid = btrfs_header_generation(leaf);
2119
2120         for (i = slot; i < nritems; i++) {
2121                 item_off = btrfs_item_ptr_offset(leaf, i);
2122                 item_len = btrfs_item_size_nr(leaf, i);
2123
2124                 btrfs_item_key_to_cpu(leaf, key, i);
2125                 if (!key_in_sk(key, sk))
2126                         continue;
2127
2128                 if (sizeof(sh) + item_len > *buf_size) {
2129                         if (*num_found) {
2130                                 ret = 1;
2131                                 goto out;
2132                         }
2133
2134                         /*
2135                          * return one empty item back for v1, which does not
2136                          * handle -EOVERFLOW
2137                          */
2138
2139                         *buf_size = sizeof(sh) + item_len;
2140                         item_len = 0;
2141                         ret = -EOVERFLOW;
2142                 }
2143
2144                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2145                         ret = 1;
2146                         goto out;
2147                 }
2148
2149                 sh.objectid = key->objectid;
2150                 sh.offset = key->offset;
2151                 sh.type = key->type;
2152                 sh.len = item_len;
2153                 sh.transid = found_transid;
2154
2155                 /*
2156                  * Copy search result header. If we fault then loop again so we
2157                  * can fault in the pages and -EFAULT there if there's a
2158                  * problem. Otherwise we'll fault and then copy the buffer in
2159                  * properly this next time through
2160                  */
2161                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2162                         ret = 0;
2163                         goto out;
2164                 }
2165
2166                 *sk_offset += sizeof(sh);
2167
2168                 if (item_len) {
2169                         char __user *up = ubuf + *sk_offset;
2170                         /*
2171                          * Copy the item, same behavior as above, but reset the
2172                          * * sk_offset so we copy the full thing again.
2173                          */
2174                         if (read_extent_buffer_to_user_nofault(leaf, up,
2175                                                 item_off, item_len)) {
2176                                 ret = 0;
2177                                 *sk_offset -= sizeof(sh);
2178                                 goto out;
2179                         }
2180
2181                         *sk_offset += item_len;
2182                 }
2183                 (*num_found)++;
2184
2185                 if (ret) /* -EOVERFLOW from above */
2186                         goto out;
2187
2188                 if (*num_found >= sk->nr_items) {
2189                         ret = 1;
2190                         goto out;
2191                 }
2192         }
2193 advance_key:
2194         ret = 0;
2195         test.objectid = sk->max_objectid;
2196         test.type = sk->max_type;
2197         test.offset = sk->max_offset;
2198         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2199                 ret = 1;
2200         else if (key->offset < (u64)-1)
2201                 key->offset++;
2202         else if (key->type < (u8)-1) {
2203                 key->offset = 0;
2204                 key->type++;
2205         } else if (key->objectid < (u64)-1) {
2206                 key->offset = 0;
2207                 key->type = 0;
2208                 key->objectid++;
2209         } else
2210                 ret = 1;
2211 out:
2212         /*
2213          *  0: all items from this leaf copied, continue with next
2214          *  1: * more items can be copied, but unused buffer is too small
2215          *     * all items were found
2216          *     Either way, it will stops the loop which iterates to the next
2217          *     leaf
2218          *  -EOVERFLOW: item was to large for buffer
2219          *  -EFAULT: could not copy extent buffer back to userspace
2220          */
2221         return ret;
2222 }
2223
2224 static noinline int search_ioctl(struct inode *inode,
2225                                  struct btrfs_ioctl_search_key *sk,
2226                                  size_t *buf_size,
2227                                  char __user *ubuf)
2228 {
2229         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2230         struct btrfs_root *root;
2231         struct btrfs_key key;
2232         struct btrfs_path *path;
2233         int ret;
2234         int num_found = 0;
2235         unsigned long sk_offset = 0;
2236
2237         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2238                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2239                 return -EOVERFLOW;
2240         }
2241
2242         path = btrfs_alloc_path();
2243         if (!path)
2244                 return -ENOMEM;
2245
2246         if (sk->tree_id == 0) {
2247                 /* search the root of the inode that was passed */
2248                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2249         } else {
2250                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2251                 if (IS_ERR(root)) {
2252                         btrfs_free_path(path);
2253                         return PTR_ERR(root);
2254                 }
2255         }
2256
2257         key.objectid = sk->min_objectid;
2258         key.type = sk->min_type;
2259         key.offset = sk->min_offset;
2260
2261         while (1) {
2262                 ret = -EFAULT;
2263                 if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset))
2264                         break;
2265
2266                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2267                 if (ret != 0) {
2268                         if (ret > 0)
2269                                 ret = 0;
2270                         goto err;
2271                 }
2272                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2273                                  &sk_offset, &num_found);
2274                 btrfs_release_path(path);
2275                 if (ret)
2276                         break;
2277
2278         }
2279         if (ret > 0)
2280                 ret = 0;
2281 err:
2282         sk->nr_items = num_found;
2283         btrfs_put_root(root);
2284         btrfs_free_path(path);
2285         return ret;
2286 }
2287
2288 static noinline int btrfs_ioctl_tree_search(struct file *file,
2289                                            void __user *argp)
2290 {
2291         struct btrfs_ioctl_search_args __user *uargs;
2292         struct btrfs_ioctl_search_key sk;
2293         struct inode *inode;
2294         int ret;
2295         size_t buf_size;
2296
2297         if (!capable(CAP_SYS_ADMIN))
2298                 return -EPERM;
2299
2300         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2301
2302         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2303                 return -EFAULT;
2304
2305         buf_size = sizeof(uargs->buf);
2306
2307         inode = file_inode(file);
2308         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2309
2310         /*
2311          * In the origin implementation an overflow is handled by returning a
2312          * search header with a len of zero, so reset ret.
2313          */
2314         if (ret == -EOVERFLOW)
2315                 ret = 0;
2316
2317         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2318                 ret = -EFAULT;
2319         return ret;
2320 }
2321
2322 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2323                                                void __user *argp)
2324 {
2325         struct btrfs_ioctl_search_args_v2 __user *uarg;
2326         struct btrfs_ioctl_search_args_v2 args;
2327         struct inode *inode;
2328         int ret;
2329         size_t buf_size;
2330         const size_t buf_limit = SZ_16M;
2331
2332         if (!capable(CAP_SYS_ADMIN))
2333                 return -EPERM;
2334
2335         /* copy search header and buffer size */
2336         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2337         if (copy_from_user(&args, uarg, sizeof(args)))
2338                 return -EFAULT;
2339
2340         buf_size = args.buf_size;
2341
2342         /* limit result size to 16MB */
2343         if (buf_size > buf_limit)
2344                 buf_size = buf_limit;
2345
2346         inode = file_inode(file);
2347         ret = search_ioctl(inode, &args.key, &buf_size,
2348                            (char __user *)(&uarg->buf[0]));
2349         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2350                 ret = -EFAULT;
2351         else if (ret == -EOVERFLOW &&
2352                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2353                 ret = -EFAULT;
2354
2355         return ret;
2356 }
2357
2358 /*
2359  * Search INODE_REFs to identify path name of 'dirid' directory
2360  * in a 'tree_id' tree. and sets path name to 'name'.
2361  */
2362 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2363                                 u64 tree_id, u64 dirid, char *name)
2364 {
2365         struct btrfs_root *root;
2366         struct btrfs_key key;
2367         char *ptr;
2368         int ret = -1;
2369         int slot;
2370         int len;
2371         int total_len = 0;
2372         struct btrfs_inode_ref *iref;
2373         struct extent_buffer *l;
2374         struct btrfs_path *path;
2375
2376         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2377                 name[0]='\0';
2378                 return 0;
2379         }
2380
2381         path = btrfs_alloc_path();
2382         if (!path)
2383                 return -ENOMEM;
2384
2385         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2386
2387         root = btrfs_get_fs_root(info, tree_id, true);
2388         if (IS_ERR(root)) {
2389                 ret = PTR_ERR(root);
2390                 root = NULL;
2391                 goto out;
2392         }
2393
2394         key.objectid = dirid;
2395         key.type = BTRFS_INODE_REF_KEY;
2396         key.offset = (u64)-1;
2397
2398         while (1) {
2399                 ret = btrfs_search_backwards(root, &key, path);
2400                 if (ret < 0)
2401                         goto out;
2402                 else if (ret > 0) {
2403                         ret = -ENOENT;
2404                         goto out;
2405                 }
2406
2407                 l = path->nodes[0];
2408                 slot = path->slots[0];
2409
2410                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2411                 len = btrfs_inode_ref_name_len(l, iref);
2412                 ptr -= len + 1;
2413                 total_len += len + 1;
2414                 if (ptr < name) {
2415                         ret = -ENAMETOOLONG;
2416                         goto out;
2417                 }
2418
2419                 *(ptr + len) = '/';
2420                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2421
2422                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2423                         break;
2424
2425                 btrfs_release_path(path);
2426                 key.objectid = key.offset;
2427                 key.offset = (u64)-1;
2428                 dirid = key.objectid;
2429         }
2430         memmove(name, ptr, total_len);
2431         name[total_len] = '\0';
2432         ret = 0;
2433 out:
2434         btrfs_put_root(root);
2435         btrfs_free_path(path);
2436         return ret;
2437 }
2438
2439 static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns,
2440                                 struct inode *inode,
2441                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2442 {
2443         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2444         struct super_block *sb = inode->i_sb;
2445         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2446         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2447         u64 dirid = args->dirid;
2448         unsigned long item_off;
2449         unsigned long item_len;
2450         struct btrfs_inode_ref *iref;
2451         struct btrfs_root_ref *rref;
2452         struct btrfs_root *root = NULL;
2453         struct btrfs_path *path;
2454         struct btrfs_key key, key2;
2455         struct extent_buffer *leaf;
2456         struct inode *temp_inode;
2457         char *ptr;
2458         int slot;
2459         int len;
2460         int total_len = 0;
2461         int ret;
2462
2463         path = btrfs_alloc_path();
2464         if (!path)
2465                 return -ENOMEM;
2466
2467         /*
2468          * If the bottom subvolume does not exist directly under upper_limit,
2469          * construct the path in from the bottom up.
2470          */
2471         if (dirid != upper_limit.objectid) {
2472                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2473
2474                 root = btrfs_get_fs_root(fs_info, treeid, true);
2475                 if (IS_ERR(root)) {
2476                         ret = PTR_ERR(root);
2477                         goto out;
2478                 }
2479
2480                 key.objectid = dirid;
2481                 key.type = BTRFS_INODE_REF_KEY;
2482                 key.offset = (u64)-1;
2483                 while (1) {
2484                         ret = btrfs_search_backwards(root, &key, path);
2485                         if (ret < 0)
2486                                 goto out_put;
2487                         else if (ret > 0) {
2488                                 ret = -ENOENT;
2489                                 goto out_put;
2490                         }
2491
2492                         leaf = path->nodes[0];
2493                         slot = path->slots[0];
2494
2495                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2496                         len = btrfs_inode_ref_name_len(leaf, iref);
2497                         ptr -= len + 1;
2498                         total_len += len + 1;
2499                         if (ptr < args->path) {
2500                                 ret = -ENAMETOOLONG;
2501                                 goto out_put;
2502                         }
2503
2504                         *(ptr + len) = '/';
2505                         read_extent_buffer(leaf, ptr,
2506                                         (unsigned long)(iref + 1), len);
2507
2508                         /* Check the read+exec permission of this directory */
2509                         ret = btrfs_previous_item(root, path, dirid,
2510                                                   BTRFS_INODE_ITEM_KEY);
2511                         if (ret < 0) {
2512                                 goto out_put;
2513                         } else if (ret > 0) {
2514                                 ret = -ENOENT;
2515                                 goto out_put;
2516                         }
2517
2518                         leaf = path->nodes[0];
2519                         slot = path->slots[0];
2520                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2521                         if (key2.objectid != dirid) {
2522                                 ret = -ENOENT;
2523                                 goto out_put;
2524                         }
2525
2526                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2527                         if (IS_ERR(temp_inode)) {
2528                                 ret = PTR_ERR(temp_inode);
2529                                 goto out_put;
2530                         }
2531                         ret = inode_permission(mnt_userns, temp_inode,
2532                                                MAY_READ | MAY_EXEC);
2533                         iput(temp_inode);
2534                         if (ret) {
2535                                 ret = -EACCES;
2536                                 goto out_put;
2537                         }
2538
2539                         if (key.offset == upper_limit.objectid)
2540                                 break;
2541                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2542                                 ret = -EACCES;
2543                                 goto out_put;
2544                         }
2545
2546                         btrfs_release_path(path);
2547                         key.objectid = key.offset;
2548                         key.offset = (u64)-1;
2549                         dirid = key.objectid;
2550                 }
2551
2552                 memmove(args->path, ptr, total_len);
2553                 args->path[total_len] = '\0';
2554                 btrfs_put_root(root);
2555                 root = NULL;
2556                 btrfs_release_path(path);
2557         }
2558
2559         /* Get the bottom subvolume's name from ROOT_REF */
2560         key.objectid = treeid;
2561         key.type = BTRFS_ROOT_REF_KEY;
2562         key.offset = args->treeid;
2563         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2564         if (ret < 0) {
2565                 goto out;
2566         } else if (ret > 0) {
2567                 ret = -ENOENT;
2568                 goto out;
2569         }
2570
2571         leaf = path->nodes[0];
2572         slot = path->slots[0];
2573         btrfs_item_key_to_cpu(leaf, &key, slot);
2574
2575         item_off = btrfs_item_ptr_offset(leaf, slot);
2576         item_len = btrfs_item_size_nr(leaf, slot);
2577         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2578         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2579         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2580                 ret = -EINVAL;
2581                 goto out;
2582         }
2583
2584         /* Copy subvolume's name */
2585         item_off += sizeof(struct btrfs_root_ref);
2586         item_len -= sizeof(struct btrfs_root_ref);
2587         read_extent_buffer(leaf, args->name, item_off, item_len);
2588         args->name[item_len] = 0;
2589
2590 out_put:
2591         btrfs_put_root(root);
2592 out:
2593         btrfs_free_path(path);
2594         return ret;
2595 }
2596
2597 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2598                                            void __user *argp)
2599 {
2600         struct btrfs_ioctl_ino_lookup_args *args;
2601         struct inode *inode;
2602         int ret = 0;
2603
2604         args = memdup_user(argp, sizeof(*args));
2605         if (IS_ERR(args))
2606                 return PTR_ERR(args);
2607
2608         inode = file_inode(file);
2609
2610         /*
2611          * Unprivileged query to obtain the containing subvolume root id. The
2612          * path is reset so it's consistent with btrfs_search_path_in_tree.
2613          */
2614         if (args->treeid == 0)
2615                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2616
2617         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2618                 args->name[0] = 0;
2619                 goto out;
2620         }
2621
2622         if (!capable(CAP_SYS_ADMIN)) {
2623                 ret = -EPERM;
2624                 goto out;
2625         }
2626
2627         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2628                                         args->treeid, args->objectid,
2629                                         args->name);
2630
2631 out:
2632         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2633                 ret = -EFAULT;
2634
2635         kfree(args);
2636         return ret;
2637 }
2638
2639 /*
2640  * Version of ino_lookup ioctl (unprivileged)
2641  *
2642  * The main differences from ino_lookup ioctl are:
2643  *
2644  *   1. Read + Exec permission will be checked using inode_permission() during
2645  *      path construction. -EACCES will be returned in case of failure.
2646  *   2. Path construction will be stopped at the inode number which corresponds
2647  *      to the fd with which this ioctl is called. If constructed path does not
2648  *      exist under fd's inode, -EACCES will be returned.
2649  *   3. The name of bottom subvolume is also searched and filled.
2650  */
2651 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2652 {
2653         struct btrfs_ioctl_ino_lookup_user_args *args;
2654         struct inode *inode;
2655         int ret;
2656
2657         args = memdup_user(argp, sizeof(*args));
2658         if (IS_ERR(args))
2659                 return PTR_ERR(args);
2660
2661         inode = file_inode(file);
2662
2663         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2664             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2665                 /*
2666                  * The subvolume does not exist under fd with which this is
2667                  * called
2668                  */
2669                 kfree(args);
2670                 return -EACCES;
2671         }
2672
2673         ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args);
2674
2675         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2676                 ret = -EFAULT;
2677
2678         kfree(args);
2679         return ret;
2680 }
2681
2682 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2683 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2684 {
2685         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2686         struct btrfs_fs_info *fs_info;
2687         struct btrfs_root *root;
2688         struct btrfs_path *path;
2689         struct btrfs_key key;
2690         struct btrfs_root_item *root_item;
2691         struct btrfs_root_ref *rref;
2692         struct extent_buffer *leaf;
2693         unsigned long item_off;
2694         unsigned long item_len;
2695         struct inode *inode;
2696         int slot;
2697         int ret = 0;
2698
2699         path = btrfs_alloc_path();
2700         if (!path)
2701                 return -ENOMEM;
2702
2703         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2704         if (!subvol_info) {
2705                 btrfs_free_path(path);
2706                 return -ENOMEM;
2707         }
2708
2709         inode = file_inode(file);
2710         fs_info = BTRFS_I(inode)->root->fs_info;
2711
2712         /* Get root_item of inode's subvolume */
2713         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2714         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2715         if (IS_ERR(root)) {
2716                 ret = PTR_ERR(root);
2717                 goto out_free;
2718         }
2719         root_item = &root->root_item;
2720
2721         subvol_info->treeid = key.objectid;
2722
2723         subvol_info->generation = btrfs_root_generation(root_item);
2724         subvol_info->flags = btrfs_root_flags(root_item);
2725
2726         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2727         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2728                                                     BTRFS_UUID_SIZE);
2729         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2730                                                     BTRFS_UUID_SIZE);
2731
2732         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2733         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2734         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2735
2736         subvol_info->otransid = btrfs_root_otransid(root_item);
2737         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2738         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2739
2740         subvol_info->stransid = btrfs_root_stransid(root_item);
2741         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2742         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2743
2744         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2745         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2746         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2747
2748         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2749                 /* Search root tree for ROOT_BACKREF of this subvolume */
2750                 key.type = BTRFS_ROOT_BACKREF_KEY;
2751                 key.offset = 0;
2752                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2753                 if (ret < 0) {
2754                         goto out;
2755                 } else if (path->slots[0] >=
2756                            btrfs_header_nritems(path->nodes[0])) {
2757                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2758                         if (ret < 0) {
2759                                 goto out;
2760                         } else if (ret > 0) {
2761                                 ret = -EUCLEAN;
2762                                 goto out;
2763                         }
2764                 }
2765
2766                 leaf = path->nodes[0];
2767                 slot = path->slots[0];
2768                 btrfs_item_key_to_cpu(leaf, &key, slot);
2769                 if (key.objectid == subvol_info->treeid &&
2770                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2771                         subvol_info->parent_id = key.offset;
2772
2773                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2774                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2775
2776                         item_off = btrfs_item_ptr_offset(leaf, slot)
2777                                         + sizeof(struct btrfs_root_ref);
2778                         item_len = btrfs_item_size_nr(leaf, slot)
2779                                         - sizeof(struct btrfs_root_ref);
2780                         read_extent_buffer(leaf, subvol_info->name,
2781                                            item_off, item_len);
2782                 } else {
2783                         ret = -ENOENT;
2784                         goto out;
2785                 }
2786         }
2787
2788         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2789                 ret = -EFAULT;
2790
2791 out:
2792         btrfs_put_root(root);
2793 out_free:
2794         btrfs_free_path(path);
2795         kfree(subvol_info);
2796         return ret;
2797 }
2798
2799 /*
2800  * Return ROOT_REF information of the subvolume containing this inode
2801  * except the subvolume name.
2802  */
2803 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2804 {
2805         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2806         struct btrfs_root_ref *rref;
2807         struct btrfs_root *root;
2808         struct btrfs_path *path;
2809         struct btrfs_key key;
2810         struct extent_buffer *leaf;
2811         struct inode *inode;
2812         u64 objectid;
2813         int slot;
2814         int ret;
2815         u8 found;
2816
2817         path = btrfs_alloc_path();
2818         if (!path)
2819                 return -ENOMEM;
2820
2821         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2822         if (IS_ERR(rootrefs)) {
2823                 btrfs_free_path(path);
2824                 return PTR_ERR(rootrefs);
2825         }
2826
2827         inode = file_inode(file);
2828         root = BTRFS_I(inode)->root->fs_info->tree_root;
2829         objectid = BTRFS_I(inode)->root->root_key.objectid;
2830
2831         key.objectid = objectid;
2832         key.type = BTRFS_ROOT_REF_KEY;
2833         key.offset = rootrefs->min_treeid;
2834         found = 0;
2835
2836         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2837         if (ret < 0) {
2838                 goto out;
2839         } else if (path->slots[0] >=
2840                    btrfs_header_nritems(path->nodes[0])) {
2841                 ret = btrfs_next_leaf(root, path);
2842                 if (ret < 0) {
2843                         goto out;
2844                 } else if (ret > 0) {
2845                         ret = -EUCLEAN;
2846                         goto out;
2847                 }
2848         }
2849         while (1) {
2850                 leaf = path->nodes[0];
2851                 slot = path->slots[0];
2852
2853                 btrfs_item_key_to_cpu(leaf, &key, slot);
2854                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2855                         ret = 0;
2856                         goto out;
2857                 }
2858
2859                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2860                         ret = -EOVERFLOW;
2861                         goto out;
2862                 }
2863
2864                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2865                 rootrefs->rootref[found].treeid = key.offset;
2866                 rootrefs->rootref[found].dirid =
2867                                   btrfs_root_ref_dirid(leaf, rref);
2868                 found++;
2869
2870                 ret = btrfs_next_item(root, path);
2871                 if (ret < 0) {
2872                         goto out;
2873                 } else if (ret > 0) {
2874                         ret = -EUCLEAN;
2875                         goto out;
2876                 }
2877         }
2878
2879 out:
2880         if (!ret || ret == -EOVERFLOW) {
2881                 rootrefs->num_items = found;
2882                 /* update min_treeid for next search */
2883                 if (found)
2884                         rootrefs->min_treeid =
2885                                 rootrefs->rootref[found - 1].treeid + 1;
2886                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2887                         ret = -EFAULT;
2888         }
2889
2890         kfree(rootrefs);
2891         btrfs_free_path(path);
2892
2893         return ret;
2894 }
2895
2896 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2897                                              void __user *arg,
2898                                              bool destroy_v2)
2899 {
2900         struct dentry *parent = file->f_path.dentry;
2901         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2902         struct dentry *dentry;
2903         struct inode *dir = d_inode(parent);
2904         struct inode *inode;
2905         struct btrfs_root *root = BTRFS_I(dir)->root;
2906         struct btrfs_root *dest = NULL;
2907         struct btrfs_ioctl_vol_args *vol_args = NULL;
2908         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2909         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
2910         char *subvol_name, *subvol_name_ptr = NULL;
2911         int subvol_namelen;
2912         int err = 0;
2913         bool destroy_parent = false;
2914
2915         if (destroy_v2) {
2916                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2917                 if (IS_ERR(vol_args2))
2918                         return PTR_ERR(vol_args2);
2919
2920                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2921                         err = -EOPNOTSUPP;
2922                         goto out;
2923                 }
2924
2925                 /*
2926                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2927                  * name, same as v1 currently does.
2928                  */
2929                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2930                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2931                         subvol_name = vol_args2->name;
2932
2933                         err = mnt_want_write_file(file);
2934                         if (err)
2935                                 goto out;
2936                 } else {
2937                         struct inode *old_dir;
2938
2939                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2940                                 err = -EINVAL;
2941                                 goto out;
2942                         }
2943
2944                         err = mnt_want_write_file(file);
2945                         if (err)
2946                                 goto out;
2947
2948                         dentry = btrfs_get_dentry(fs_info->sb,
2949                                         BTRFS_FIRST_FREE_OBJECTID,
2950                                         vol_args2->subvolid, 0, 0);
2951                         if (IS_ERR(dentry)) {
2952                                 err = PTR_ERR(dentry);
2953                                 goto out_drop_write;
2954                         }
2955
2956                         /*
2957                          * Change the default parent since the subvolume being
2958                          * deleted can be outside of the current mount point.
2959                          */
2960                         parent = btrfs_get_parent(dentry);
2961
2962                         /*
2963                          * At this point dentry->d_name can point to '/' if the
2964                          * subvolume we want to destroy is outsite of the
2965                          * current mount point, so we need to release the
2966                          * current dentry and execute the lookup to return a new
2967                          * one with ->d_name pointing to the
2968                          * <mount point>/subvol_name.
2969                          */
2970                         dput(dentry);
2971                         if (IS_ERR(parent)) {
2972                                 err = PTR_ERR(parent);
2973                                 goto out_drop_write;
2974                         }
2975                         old_dir = dir;
2976                         dir = d_inode(parent);
2977
2978                         /*
2979                          * If v2 was used with SPEC_BY_ID, a new parent was
2980                          * allocated since the subvolume can be outside of the
2981                          * current mount point. Later on we need to release this
2982                          * new parent dentry.
2983                          */
2984                         destroy_parent = true;
2985
2986                         /*
2987                          * On idmapped mounts, deletion via subvolid is
2988                          * restricted to subvolumes that are immediate
2989                          * ancestors of the inode referenced by the file
2990                          * descriptor in the ioctl. Otherwise the idmapping
2991                          * could potentially be abused to delete subvolumes
2992                          * anywhere in the filesystem the user wouldn't be able
2993                          * to delete without an idmapped mount.
2994                          */
2995                         if (old_dir != dir && mnt_userns != &init_user_ns) {
2996                                 err = -EOPNOTSUPP;
2997                                 goto free_parent;
2998                         }
2999
3000                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3001                                                 fs_info, vol_args2->subvolid);
3002                         if (IS_ERR(subvol_name_ptr)) {
3003                                 err = PTR_ERR(subvol_name_ptr);
3004                                 goto free_parent;
3005                         }
3006                         /* subvol_name_ptr is already nul terminated */
3007                         subvol_name = (char *)kbasename(subvol_name_ptr);
3008                 }
3009         } else {
3010                 vol_args = memdup_user(arg, sizeof(*vol_args));
3011                 if (IS_ERR(vol_args))
3012                         return PTR_ERR(vol_args);
3013
3014                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3015                 subvol_name = vol_args->name;
3016
3017                 err = mnt_want_write_file(file);
3018                 if (err)
3019                         goto out;
3020         }
3021
3022         subvol_namelen = strlen(subvol_name);
3023
3024         if (strchr(subvol_name, '/') ||
3025             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3026                 err = -EINVAL;
3027                 goto free_subvol_name;
3028         }
3029
3030         if (!S_ISDIR(dir->i_mode)) {
3031                 err = -ENOTDIR;
3032                 goto free_subvol_name;
3033         }
3034
3035         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3036         if (err == -EINTR)
3037                 goto free_subvol_name;
3038         dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen);
3039         if (IS_ERR(dentry)) {
3040                 err = PTR_ERR(dentry);
3041                 goto out_unlock_dir;
3042         }
3043
3044         if (d_really_is_negative(dentry)) {
3045                 err = -ENOENT;
3046                 goto out_dput;
3047         }
3048
3049         inode = d_inode(dentry);
3050         dest = BTRFS_I(inode)->root;
3051         if (!capable(CAP_SYS_ADMIN)) {
3052                 /*
3053                  * Regular user.  Only allow this with a special mount
3054                  * option, when the user has write+exec access to the
3055                  * subvol root, and when rmdir(2) would have been
3056                  * allowed.
3057                  *
3058                  * Note that this is _not_ check that the subvol is
3059                  * empty or doesn't contain data that we wouldn't
3060                  * otherwise be able to delete.
3061                  *
3062                  * Users who want to delete empty subvols should try
3063                  * rmdir(2).
3064                  */
3065                 err = -EPERM;
3066                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3067                         goto out_dput;
3068
3069                 /*
3070                  * Do not allow deletion if the parent dir is the same
3071                  * as the dir to be deleted.  That means the ioctl
3072                  * must be called on the dentry referencing the root
3073                  * of the subvol, not a random directory contained
3074                  * within it.
3075                  */
3076                 err = -EINVAL;
3077                 if (root == dest)
3078                         goto out_dput;
3079
3080                 err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC);
3081                 if (err)
3082                         goto out_dput;
3083         }
3084
3085         /* check if subvolume may be deleted by a user */
3086         err = btrfs_may_delete(mnt_userns, dir, dentry, 1);
3087         if (err)
3088                 goto out_dput;
3089
3090         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3091                 err = -EINVAL;
3092                 goto out_dput;
3093         }
3094
3095         btrfs_inode_lock(inode, 0);
3096         err = btrfs_delete_subvolume(dir, dentry);
3097         btrfs_inode_unlock(inode, 0);
3098         if (!err)
3099                 d_delete_notify(dir, dentry);
3100
3101 out_dput:
3102         dput(dentry);
3103 out_unlock_dir:
3104         btrfs_inode_unlock(dir, 0);
3105 free_subvol_name:
3106         kfree(subvol_name_ptr);
3107 free_parent:
3108         if (destroy_parent)
3109                 dput(parent);
3110 out_drop_write:
3111         mnt_drop_write_file(file);
3112 out:
3113         kfree(vol_args2);
3114         kfree(vol_args);
3115         return err;
3116 }
3117
3118 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3119 {
3120         struct inode *inode = file_inode(file);
3121         struct btrfs_root *root = BTRFS_I(inode)->root;
3122         struct btrfs_ioctl_defrag_range_args range = {0};
3123         int ret;
3124
3125         ret = mnt_want_write_file(file);
3126         if (ret)
3127                 return ret;
3128
3129         if (btrfs_root_readonly(root)) {
3130                 ret = -EROFS;
3131                 goto out;
3132         }
3133
3134         /* Subpage defrag will be supported in later commits */
3135         if (root->fs_info->sectorsize < PAGE_SIZE) {
3136                 ret = -ENOTTY;
3137                 goto out;
3138         }
3139
3140         switch (inode->i_mode & S_IFMT) {
3141         case S_IFDIR:
3142                 if (!capable(CAP_SYS_ADMIN)) {
3143                         ret = -EPERM;
3144                         goto out;
3145                 }
3146                 ret = btrfs_defrag_root(root);
3147                 break;
3148         case S_IFREG:
3149                 /*
3150                  * Note that this does not check the file descriptor for write
3151                  * access. This prevents defragmenting executables that are
3152                  * running and allows defrag on files open in read-only mode.
3153                  */
3154                 if (!capable(CAP_SYS_ADMIN) &&
3155                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3156                         ret = -EPERM;
3157                         goto out;
3158                 }
3159
3160                 if (argp) {
3161                         if (copy_from_user(&range, argp, sizeof(range))) {
3162                                 ret = -EFAULT;
3163                                 goto out;
3164                         }
3165                         /* compression requires us to start the IO */
3166                         if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3167                                 range.flags |= BTRFS_DEFRAG_RANGE_START_IO;
3168                                 range.extent_thresh = (u32)-1;
3169                         }
3170                 } else {
3171                         /* the rest are all set to zero by kzalloc */
3172                         range.len = (u64)-1;
3173                 }
3174                 ret = btrfs_defrag_file(file_inode(file), file,
3175                                         &range, BTRFS_OLDEST_GENERATION, 0);
3176                 if (ret > 0)
3177                         ret = 0;
3178                 break;
3179         default:
3180                 ret = -EINVAL;
3181         }
3182 out:
3183         mnt_drop_write_file(file);
3184         return ret;
3185 }
3186
3187 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3188 {
3189         struct btrfs_ioctl_vol_args *vol_args;
3190         int ret;
3191
3192         if (!capable(CAP_SYS_ADMIN))
3193                 return -EPERM;
3194
3195         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3196                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3197
3198         vol_args = memdup_user(arg, sizeof(*vol_args));
3199         if (IS_ERR(vol_args)) {
3200                 ret = PTR_ERR(vol_args);
3201                 goto out;
3202         }
3203
3204         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3205         ret = btrfs_init_new_device(fs_info, vol_args->name);
3206
3207         if (!ret)
3208                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3209
3210         kfree(vol_args);
3211 out:
3212         btrfs_exclop_finish(fs_info);
3213         return ret;
3214 }
3215
3216 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3217 {
3218         struct inode *inode = file_inode(file);
3219         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3220         struct btrfs_ioctl_vol_args_v2 *vol_args;
3221         struct block_device *bdev = NULL;
3222         fmode_t mode;
3223         int ret;
3224         bool cancel = false;
3225
3226         if (!capable(CAP_SYS_ADMIN))
3227                 return -EPERM;
3228
3229         ret = mnt_want_write_file(file);
3230         if (ret)
3231                 return ret;
3232
3233         vol_args = memdup_user(arg, sizeof(*vol_args));
3234         if (IS_ERR(vol_args)) {
3235                 ret = PTR_ERR(vol_args);
3236                 goto err_drop;
3237         }
3238
3239         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3240                 ret = -EOPNOTSUPP;
3241                 goto out;
3242         }
3243         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3244         if (!(vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) &&
3245             strcmp("cancel", vol_args->name) == 0)
3246                 cancel = true;
3247
3248         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3249                                            cancel);
3250         if (ret)
3251                 goto out;
3252         /* Exclusive operation is now claimed */
3253
3254         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3255                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid, &bdev, &mode);
3256         else
3257                 ret = btrfs_rm_device(fs_info, vol_args->name, 0, &bdev, &mode);
3258
3259         btrfs_exclop_finish(fs_info);
3260
3261         if (!ret) {
3262                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3263                         btrfs_info(fs_info, "device deleted: id %llu",
3264                                         vol_args->devid);
3265                 else
3266                         btrfs_info(fs_info, "device deleted: %s",
3267                                         vol_args->name);
3268         }
3269 out:
3270         kfree(vol_args);
3271 err_drop:
3272         mnt_drop_write_file(file);
3273         if (bdev)
3274                 blkdev_put(bdev, mode);
3275         return ret;
3276 }
3277
3278 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3279 {
3280         struct inode *inode = file_inode(file);
3281         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3282         struct btrfs_ioctl_vol_args *vol_args;
3283         struct block_device *bdev = NULL;
3284         fmode_t mode;
3285         int ret;
3286         bool cancel;
3287
3288         if (!capable(CAP_SYS_ADMIN))
3289                 return -EPERM;
3290
3291         ret = mnt_want_write_file(file);
3292         if (ret)
3293                 return ret;
3294
3295         vol_args = memdup_user(arg, sizeof(*vol_args));
3296         if (IS_ERR(vol_args)) {
3297                 ret = PTR_ERR(vol_args);
3298                 goto out_drop_write;
3299         }
3300         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3301         cancel = (strcmp("cancel", vol_args->name) == 0);
3302
3303         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3304                                            cancel);
3305         if (ret == 0) {
3306                 ret = btrfs_rm_device(fs_info, vol_args->name, 0, &bdev, &mode);
3307                 if (!ret)
3308                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3309                 btrfs_exclop_finish(fs_info);
3310         }
3311
3312         kfree(vol_args);
3313 out_drop_write:
3314         mnt_drop_write_file(file);
3315         if (bdev)
3316                 blkdev_put(bdev, mode);
3317         return ret;
3318 }
3319
3320 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3321                                 void __user *arg)
3322 {
3323         struct btrfs_ioctl_fs_info_args *fi_args;
3324         struct btrfs_device *device;
3325         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3326         u64 flags_in;
3327         int ret = 0;
3328
3329         fi_args = memdup_user(arg, sizeof(*fi_args));
3330         if (IS_ERR(fi_args))
3331                 return PTR_ERR(fi_args);
3332
3333         flags_in = fi_args->flags;
3334         memset(fi_args, 0, sizeof(*fi_args));
3335
3336         rcu_read_lock();
3337         fi_args->num_devices = fs_devices->num_devices;
3338
3339         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3340                 if (device->devid > fi_args->max_id)
3341                         fi_args->max_id = device->devid;
3342         }
3343         rcu_read_unlock();
3344
3345         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3346         fi_args->nodesize = fs_info->nodesize;
3347         fi_args->sectorsize = fs_info->sectorsize;
3348         fi_args->clone_alignment = fs_info->sectorsize;
3349
3350         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3351                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3352                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3353                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3354         }
3355
3356         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3357                 fi_args->generation = fs_info->generation;
3358                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3359         }
3360
3361         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3362                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3363                        sizeof(fi_args->metadata_uuid));
3364                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3365         }
3366
3367         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3368                 ret = -EFAULT;
3369
3370         kfree(fi_args);
3371         return ret;
3372 }
3373
3374 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3375                                  void __user *arg)
3376 {
3377         struct btrfs_ioctl_dev_info_args *di_args;
3378         struct btrfs_device *dev;
3379         int ret = 0;
3380         char *s_uuid = NULL;
3381
3382         di_args = memdup_user(arg, sizeof(*di_args));
3383         if (IS_ERR(di_args))
3384                 return PTR_ERR(di_args);
3385
3386         if (!btrfs_is_empty_uuid(di_args->uuid))
3387                 s_uuid = di_args->uuid;
3388
3389         rcu_read_lock();
3390         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3391                                 NULL);
3392
3393         if (!dev) {
3394                 ret = -ENODEV;
3395                 goto out;
3396         }
3397
3398         di_args->devid = dev->devid;
3399         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3400         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3401         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3402         if (dev->name) {
3403                 strncpy(di_args->path, rcu_str_deref(dev->name),
3404                                 sizeof(di_args->path) - 1);
3405                 di_args->path[sizeof(di_args->path) - 1] = 0;
3406         } else {
3407                 di_args->path[0] = '\0';
3408         }
3409
3410 out:
3411         rcu_read_unlock();
3412         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3413                 ret = -EFAULT;
3414
3415         kfree(di_args);
3416         return ret;
3417 }
3418
3419 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3420 {
3421         struct inode *inode = file_inode(file);
3422         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3423         struct btrfs_root *root = BTRFS_I(inode)->root;
3424         struct btrfs_root *new_root;
3425         struct btrfs_dir_item *di;
3426         struct btrfs_trans_handle *trans;
3427         struct btrfs_path *path = NULL;
3428         struct btrfs_disk_key disk_key;
3429         u64 objectid = 0;
3430         u64 dir_id;
3431         int ret;
3432
3433         if (!capable(CAP_SYS_ADMIN))
3434                 return -EPERM;
3435
3436         ret = mnt_want_write_file(file);
3437         if (ret)
3438                 return ret;
3439
3440         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3441                 ret = -EFAULT;
3442                 goto out;
3443         }
3444
3445         if (!objectid)
3446                 objectid = BTRFS_FS_TREE_OBJECTID;
3447
3448         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3449         if (IS_ERR(new_root)) {
3450                 ret = PTR_ERR(new_root);
3451                 goto out;
3452         }
3453         if (!is_fstree(new_root->root_key.objectid)) {
3454                 ret = -ENOENT;
3455                 goto out_free;
3456         }
3457
3458         path = btrfs_alloc_path();
3459         if (!path) {
3460                 ret = -ENOMEM;
3461                 goto out_free;
3462         }
3463
3464         trans = btrfs_start_transaction(root, 1);
3465         if (IS_ERR(trans)) {
3466                 ret = PTR_ERR(trans);
3467                 goto out_free;
3468         }
3469
3470         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3471         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3472                                    dir_id, "default", 7, 1);
3473         if (IS_ERR_OR_NULL(di)) {
3474                 btrfs_release_path(path);
3475                 btrfs_end_transaction(trans);
3476                 btrfs_err(fs_info,
3477                           "Umm, you don't have the default diritem, this isn't going to work");
3478                 ret = -ENOENT;
3479                 goto out_free;
3480         }
3481
3482         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3483         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3484         btrfs_mark_buffer_dirty(path->nodes[0]);
3485         btrfs_release_path(path);
3486
3487         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3488         btrfs_end_transaction(trans);
3489 out_free:
3490         btrfs_put_root(new_root);
3491         btrfs_free_path(path);
3492 out:
3493         mnt_drop_write_file(file);
3494         return ret;
3495 }
3496
3497 static void get_block_group_info(struct list_head *groups_list,
3498                                  struct btrfs_ioctl_space_info *space)
3499 {
3500         struct btrfs_block_group *block_group;
3501
3502         space->total_bytes = 0;
3503         space->used_bytes = 0;
3504         space->flags = 0;
3505         list_for_each_entry(block_group, groups_list, list) {
3506                 space->flags = block_group->flags;
3507                 space->total_bytes += block_group->length;
3508                 space->used_bytes += block_group->used;
3509         }
3510 }
3511
3512 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3513                                    void __user *arg)
3514 {
3515         struct btrfs_ioctl_space_args space_args;
3516         struct btrfs_ioctl_space_info space;
3517         struct btrfs_ioctl_space_info *dest;
3518         struct btrfs_ioctl_space_info *dest_orig;
3519         struct btrfs_ioctl_space_info __user *user_dest;
3520         struct btrfs_space_info *info;
3521         static const u64 types[] = {
3522                 BTRFS_BLOCK_GROUP_DATA,
3523                 BTRFS_BLOCK_GROUP_SYSTEM,
3524                 BTRFS_BLOCK_GROUP_METADATA,
3525                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3526         };
3527         int num_types = 4;
3528         int alloc_size;
3529         int ret = 0;
3530         u64 slot_count = 0;
3531         int i, c;
3532
3533         if (copy_from_user(&space_args,
3534                            (struct btrfs_ioctl_space_args __user *)arg,
3535                            sizeof(space_args)))
3536                 return -EFAULT;
3537
3538         for (i = 0; i < num_types; i++) {
3539                 struct btrfs_space_info *tmp;
3540
3541                 info = NULL;
3542                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3543                         if (tmp->flags == types[i]) {
3544                                 info = tmp;
3545                                 break;
3546                         }
3547                 }
3548
3549                 if (!info)
3550                         continue;
3551
3552                 down_read(&info->groups_sem);
3553                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3554                         if (!list_empty(&info->block_groups[c]))
3555                                 slot_count++;
3556                 }
3557                 up_read(&info->groups_sem);
3558         }
3559
3560         /*
3561          * Global block reserve, exported as a space_info
3562          */
3563         slot_count++;
3564
3565         /* space_slots == 0 means they are asking for a count */
3566         if (space_args.space_slots == 0) {
3567                 space_args.total_spaces = slot_count;
3568                 goto out;
3569         }
3570
3571         slot_count = min_t(u64, space_args.space_slots, slot_count);
3572
3573         alloc_size = sizeof(*dest) * slot_count;
3574
3575         /* we generally have at most 6 or so space infos, one for each raid
3576          * level.  So, a whole page should be more than enough for everyone
3577          */
3578         if (alloc_size > PAGE_SIZE)
3579                 return -ENOMEM;
3580
3581         space_args.total_spaces = 0;
3582         dest = kmalloc(alloc_size, GFP_KERNEL);
3583         if (!dest)
3584                 return -ENOMEM;
3585         dest_orig = dest;
3586
3587         /* now we have a buffer to copy into */
3588         for (i = 0; i < num_types; i++) {
3589                 struct btrfs_space_info *tmp;
3590
3591                 if (!slot_count)
3592                         break;
3593
3594                 info = NULL;
3595                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3596                         if (tmp->flags == types[i]) {
3597                                 info = tmp;
3598                                 break;
3599                         }
3600                 }
3601
3602                 if (!info)
3603                         continue;
3604                 down_read(&info->groups_sem);
3605                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3606                         if (!list_empty(&info->block_groups[c])) {
3607                                 get_block_group_info(&info->block_groups[c],
3608                                                      &space);
3609                                 memcpy(dest, &space, sizeof(space));
3610                                 dest++;
3611                                 space_args.total_spaces++;
3612                                 slot_count--;
3613                         }
3614                         if (!slot_count)
3615                                 break;
3616                 }
3617                 up_read(&info->groups_sem);
3618         }
3619
3620         /*
3621          * Add global block reserve
3622          */
3623         if (slot_count) {
3624                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3625
3626                 spin_lock(&block_rsv->lock);
3627                 space.total_bytes = block_rsv->size;
3628                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3629                 spin_unlock(&block_rsv->lock);
3630                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3631                 memcpy(dest, &space, sizeof(space));
3632                 space_args.total_spaces++;
3633         }
3634
3635         user_dest = (struct btrfs_ioctl_space_info __user *)
3636                 (arg + sizeof(struct btrfs_ioctl_space_args));
3637
3638         if (copy_to_user(user_dest, dest_orig, alloc_size))
3639                 ret = -EFAULT;
3640
3641         kfree(dest_orig);
3642 out:
3643         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3644                 ret = -EFAULT;
3645
3646         return ret;
3647 }
3648
3649 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3650                                             void __user *argp)
3651 {
3652         struct btrfs_trans_handle *trans;
3653         u64 transid;
3654         int ret;
3655
3656         trans = btrfs_attach_transaction_barrier(root);
3657         if (IS_ERR(trans)) {
3658                 if (PTR_ERR(trans) != -ENOENT)
3659                         return PTR_ERR(trans);
3660
3661                 /* No running transaction, don't bother */
3662                 transid = root->fs_info->last_trans_committed;
3663                 goto out;
3664         }
3665         transid = trans->transid;
3666         ret = btrfs_commit_transaction_async(trans);
3667         if (ret) {
3668                 btrfs_end_transaction(trans);
3669                 return ret;
3670         }
3671 out:
3672         if (argp)
3673                 if (copy_to_user(argp, &transid, sizeof(transid)))
3674                         return -EFAULT;
3675         return 0;
3676 }
3677
3678 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3679                                            void __user *argp)
3680 {
3681         u64 transid;
3682
3683         if (argp) {
3684                 if (copy_from_user(&transid, argp, sizeof(transid)))
3685                         return -EFAULT;
3686         } else {
3687                 transid = 0;  /* current trans */
3688         }
3689         return btrfs_wait_for_commit(fs_info, transid);
3690 }
3691
3692 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3693 {
3694         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3695         struct btrfs_ioctl_scrub_args *sa;
3696         int ret;
3697
3698         if (!capable(CAP_SYS_ADMIN))
3699                 return -EPERM;
3700
3701         sa = memdup_user(arg, sizeof(*sa));
3702         if (IS_ERR(sa))
3703                 return PTR_ERR(sa);
3704
3705         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3706                 ret = mnt_want_write_file(file);
3707                 if (ret)
3708                         goto out;
3709         }
3710
3711         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3712                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3713                               0);
3714
3715         /*
3716          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3717          * error. This is important as it allows user space to know how much
3718          * progress scrub has done. For example, if scrub is canceled we get
3719          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3720          * space. Later user space can inspect the progress from the structure
3721          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3722          * previously (btrfs-progs does this).
3723          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3724          * then return -EFAULT to signal the structure was not copied or it may
3725          * be corrupt and unreliable due to a partial copy.
3726          */
3727         if (copy_to_user(arg, sa, sizeof(*sa)))
3728                 ret = -EFAULT;
3729
3730         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3731                 mnt_drop_write_file(file);
3732 out:
3733         kfree(sa);
3734         return ret;
3735 }
3736
3737 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3738 {
3739         if (!capable(CAP_SYS_ADMIN))
3740                 return -EPERM;
3741
3742         return btrfs_scrub_cancel(fs_info);
3743 }
3744
3745 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3746                                        void __user *arg)
3747 {
3748         struct btrfs_ioctl_scrub_args *sa;
3749         int ret;
3750
3751         if (!capable(CAP_SYS_ADMIN))
3752                 return -EPERM;
3753
3754         sa = memdup_user(arg, sizeof(*sa));
3755         if (IS_ERR(sa))
3756                 return PTR_ERR(sa);
3757
3758         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3759
3760         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3761                 ret = -EFAULT;
3762
3763         kfree(sa);
3764         return ret;
3765 }
3766
3767 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3768                                       void __user *arg)
3769 {
3770         struct btrfs_ioctl_get_dev_stats *sa;
3771         int ret;
3772
3773         sa = memdup_user(arg, sizeof(*sa));
3774         if (IS_ERR(sa))
3775                 return PTR_ERR(sa);
3776
3777         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3778                 kfree(sa);
3779                 return -EPERM;
3780         }
3781
3782         ret = btrfs_get_dev_stats(fs_info, sa);
3783
3784         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3785                 ret = -EFAULT;
3786
3787         kfree(sa);
3788         return ret;
3789 }
3790
3791 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3792                                     void __user *arg)
3793 {
3794         struct btrfs_ioctl_dev_replace_args *p;
3795         int ret;
3796
3797         if (!capable(CAP_SYS_ADMIN))
3798                 return -EPERM;
3799
3800         p = memdup_user(arg, sizeof(*p));
3801         if (IS_ERR(p))
3802                 return PTR_ERR(p);
3803
3804         switch (p->cmd) {
3805         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3806                 if (sb_rdonly(fs_info->sb)) {
3807                         ret = -EROFS;
3808                         goto out;
3809                 }
3810                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3811                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3812                 } else {
3813                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3814                         btrfs_exclop_finish(fs_info);
3815                 }
3816                 break;
3817         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3818                 btrfs_dev_replace_status(fs_info, p);
3819                 ret = 0;
3820                 break;
3821         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3822                 p->result = btrfs_dev_replace_cancel(fs_info);
3823                 ret = 0;
3824                 break;
3825         default:
3826                 ret = -EINVAL;
3827                 break;
3828         }
3829
3830         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3831                 ret = -EFAULT;
3832 out:
3833         kfree(p);
3834         return ret;
3835 }
3836
3837 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3838 {
3839         int ret = 0;
3840         int i;
3841         u64 rel_ptr;
3842         int size;
3843         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3844         struct inode_fs_paths *ipath = NULL;
3845         struct btrfs_path *path;
3846
3847         if (!capable(CAP_DAC_READ_SEARCH))
3848                 return -EPERM;
3849
3850         path = btrfs_alloc_path();
3851         if (!path) {
3852                 ret = -ENOMEM;
3853                 goto out;
3854         }
3855
3856         ipa = memdup_user(arg, sizeof(*ipa));
3857         if (IS_ERR(ipa)) {
3858                 ret = PTR_ERR(ipa);
3859                 ipa = NULL;
3860                 goto out;
3861         }
3862
3863         size = min_t(u32, ipa->size, 4096);
3864         ipath = init_ipath(size, root, path);
3865         if (IS_ERR(ipath)) {
3866                 ret = PTR_ERR(ipath);
3867                 ipath = NULL;
3868                 goto out;
3869         }
3870
3871         ret = paths_from_inode(ipa->inum, ipath);
3872         if (ret < 0)
3873                 goto out;
3874
3875         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3876                 rel_ptr = ipath->fspath->val[i] -
3877                           (u64)(unsigned long)ipath->fspath->val;
3878                 ipath->fspath->val[i] = rel_ptr;
3879         }
3880
3881         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3882                            ipath->fspath, size);
3883         if (ret) {
3884                 ret = -EFAULT;
3885                 goto out;
3886         }
3887
3888 out:
3889         btrfs_free_path(path);
3890         free_ipath(ipath);
3891         kfree(ipa);
3892
3893         return ret;
3894 }
3895
3896 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3897 {
3898         struct btrfs_data_container *inodes = ctx;
3899         const size_t c = 3 * sizeof(u64);
3900
3901         if (inodes->bytes_left >= c) {
3902                 inodes->bytes_left -= c;
3903                 inodes->val[inodes->elem_cnt] = inum;
3904                 inodes->val[inodes->elem_cnt + 1] = offset;
3905                 inodes->val[inodes->elem_cnt + 2] = root;
3906                 inodes->elem_cnt += 3;
3907         } else {
3908                 inodes->bytes_missing += c - inodes->bytes_left;
3909                 inodes->bytes_left = 0;
3910                 inodes->elem_missed += 3;
3911         }
3912
3913         return 0;
3914 }
3915
3916 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3917                                         void __user *arg, int version)
3918 {
3919         int ret = 0;
3920         int size;
3921         struct btrfs_ioctl_logical_ino_args *loi;
3922         struct btrfs_data_container *inodes = NULL;
3923         struct btrfs_path *path = NULL;
3924         bool ignore_offset;
3925
3926         if (!capable(CAP_SYS_ADMIN))
3927                 return -EPERM;
3928
3929         loi = memdup_user(arg, sizeof(*loi));
3930         if (IS_ERR(loi))
3931                 return PTR_ERR(loi);
3932
3933         if (version == 1) {
3934                 ignore_offset = false;
3935                 size = min_t(u32, loi->size, SZ_64K);
3936         } else {
3937                 /* All reserved bits must be 0 for now */
3938                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3939                         ret = -EINVAL;
3940                         goto out_loi;
3941                 }
3942                 /* Only accept flags we have defined so far */
3943                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3944                         ret = -EINVAL;
3945                         goto out_loi;
3946                 }
3947                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3948                 size = min_t(u32, loi->size, SZ_16M);
3949         }
3950
3951         path = btrfs_alloc_path();
3952         if (!path) {
3953                 ret = -ENOMEM;
3954                 goto out;
3955         }
3956
3957         inodes = init_data_container(size);
3958         if (IS_ERR(inodes)) {
3959                 ret = PTR_ERR(inodes);
3960                 inodes = NULL;
3961                 goto out;
3962         }
3963
3964         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3965                                           build_ino_list, inodes, ignore_offset);
3966         if (ret == -EINVAL)
3967                 ret = -ENOENT;
3968         if (ret < 0)
3969                 goto out;
3970
3971         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3972                            size);
3973         if (ret)
3974                 ret = -EFAULT;
3975
3976 out:
3977         btrfs_free_path(path);
3978         kvfree(inodes);
3979 out_loi:
3980         kfree(loi);
3981
3982         return ret;
3983 }
3984
3985 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3986                                struct btrfs_ioctl_balance_args *bargs)
3987 {
3988         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3989
3990         bargs->flags = bctl->flags;
3991
3992         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3993                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3994         if (atomic_read(&fs_info->balance_pause_req))
3995                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3996         if (atomic_read(&fs_info->balance_cancel_req))
3997                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3998
3999         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4000         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4001         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4002
4003         spin_lock(&fs_info->balance_lock);
4004         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4005         spin_unlock(&fs_info->balance_lock);
4006 }
4007
4008 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4009 {
4010         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4011         struct btrfs_fs_info *fs_info = root->fs_info;
4012         struct btrfs_ioctl_balance_args *bargs;
4013         struct btrfs_balance_control *bctl;
4014         bool need_unlock; /* for mut. excl. ops lock */
4015         int ret;
4016
4017         if (!capable(CAP_SYS_ADMIN))
4018                 return -EPERM;
4019
4020         ret = mnt_want_write_file(file);
4021         if (ret)
4022                 return ret;
4023
4024 again:
4025         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4026                 mutex_lock(&fs_info->balance_mutex);
4027                 need_unlock = true;
4028                 goto locked;
4029         }
4030
4031         /*
4032          * mut. excl. ops lock is locked.  Three possibilities:
4033          *   (1) some other op is running
4034          *   (2) balance is running
4035          *   (3) balance is paused -- special case (think resume)
4036          */
4037         mutex_lock(&fs_info->balance_mutex);
4038         if (fs_info->balance_ctl) {
4039                 /* this is either (2) or (3) */
4040                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4041                         mutex_unlock(&fs_info->balance_mutex);
4042                         /*
4043                          * Lock released to allow other waiters to continue,
4044                          * we'll reexamine the status again.
4045                          */
4046                         mutex_lock(&fs_info->balance_mutex);
4047
4048                         if (fs_info->balance_ctl &&
4049                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4050                                 /* this is (3) */
4051                                 need_unlock = false;
4052                                 goto locked;
4053                         }
4054
4055                         mutex_unlock(&fs_info->balance_mutex);
4056                         goto again;
4057                 } else {
4058                         /* this is (2) */
4059                         mutex_unlock(&fs_info->balance_mutex);
4060                         ret = -EINPROGRESS;
4061                         goto out;
4062                 }
4063         } else {
4064                 /* this is (1) */
4065                 mutex_unlock(&fs_info->balance_mutex);
4066                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4067                 goto out;
4068         }
4069
4070 locked:
4071
4072         if (arg) {
4073                 bargs = memdup_user(arg, sizeof(*bargs));
4074                 if (IS_ERR(bargs)) {
4075                         ret = PTR_ERR(bargs);
4076                         goto out_unlock;
4077                 }
4078
4079                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4080                         if (!fs_info->balance_ctl) {
4081                                 ret = -ENOTCONN;
4082                                 goto out_bargs;
4083                         }
4084
4085                         bctl = fs_info->balance_ctl;
4086                         spin_lock(&fs_info->balance_lock);
4087                         bctl->flags |= BTRFS_BALANCE_RESUME;
4088                         spin_unlock(&fs_info->balance_lock);
4089
4090                         goto do_balance;
4091                 }
4092         } else {
4093                 bargs = NULL;
4094         }
4095
4096         if (fs_info->balance_ctl) {
4097                 ret = -EINPROGRESS;
4098                 goto out_bargs;
4099         }
4100
4101         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4102         if (!bctl) {
4103                 ret = -ENOMEM;
4104                 goto out_bargs;
4105         }
4106
4107         if (arg) {
4108                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4109                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4110                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4111
4112                 bctl->flags = bargs->flags;
4113         } else {
4114                 /* balance everything - no filters */
4115                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4116         }
4117
4118         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4119                 ret = -EINVAL;
4120                 goto out_bctl;
4121         }
4122
4123 do_balance:
4124         /*
4125          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4126          * bctl is freed in reset_balance_state, or, if restriper was paused
4127          * all the way until unmount, in free_fs_info.  The flag should be
4128          * cleared after reset_balance_state.
4129          */
4130         need_unlock = false;
4131
4132         ret = btrfs_balance(fs_info, bctl, bargs);
4133         bctl = NULL;
4134
4135         if ((ret == 0 || ret == -ECANCELED) && arg) {
4136                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4137                         ret = -EFAULT;
4138         }
4139
4140 out_bctl:
4141         kfree(bctl);
4142 out_bargs:
4143         kfree(bargs);
4144 out_unlock:
4145         mutex_unlock(&fs_info->balance_mutex);
4146         if (need_unlock)
4147                 btrfs_exclop_finish(fs_info);
4148 out:
4149         mnt_drop_write_file(file);
4150         return ret;
4151 }
4152
4153 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4154 {
4155         if (!capable(CAP_SYS_ADMIN))
4156                 return -EPERM;
4157
4158         switch (cmd) {
4159         case BTRFS_BALANCE_CTL_PAUSE:
4160                 return btrfs_pause_balance(fs_info);
4161         case BTRFS_BALANCE_CTL_CANCEL:
4162                 return btrfs_cancel_balance(fs_info);
4163         }
4164
4165         return -EINVAL;
4166 }
4167
4168 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4169                                          void __user *arg)
4170 {
4171         struct btrfs_ioctl_balance_args *bargs;
4172         int ret = 0;
4173
4174         if (!capable(CAP_SYS_ADMIN))
4175                 return -EPERM;
4176
4177         mutex_lock(&fs_info->balance_mutex);
4178         if (!fs_info->balance_ctl) {
4179                 ret = -ENOTCONN;
4180                 goto out;
4181         }
4182
4183         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4184         if (!bargs) {
4185                 ret = -ENOMEM;
4186                 goto out;
4187         }
4188
4189         btrfs_update_ioctl_balance_args(fs_info, bargs);
4190
4191         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4192                 ret = -EFAULT;
4193
4194         kfree(bargs);
4195 out:
4196         mutex_unlock(&fs_info->balance_mutex);
4197         return ret;
4198 }
4199
4200 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4201 {
4202         struct inode *inode = file_inode(file);
4203         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4204         struct btrfs_ioctl_quota_ctl_args *sa;
4205         int ret;
4206
4207         if (!capable(CAP_SYS_ADMIN))
4208                 return -EPERM;
4209
4210         ret = mnt_want_write_file(file);
4211         if (ret)
4212                 return ret;
4213
4214         sa = memdup_user(arg, sizeof(*sa));
4215         if (IS_ERR(sa)) {
4216                 ret = PTR_ERR(sa);
4217                 goto drop_write;
4218         }
4219
4220         down_write(&fs_info->subvol_sem);
4221
4222         switch (sa->cmd) {
4223         case BTRFS_QUOTA_CTL_ENABLE:
4224                 ret = btrfs_quota_enable(fs_info);
4225                 break;
4226         case BTRFS_QUOTA_CTL_DISABLE:
4227                 ret = btrfs_quota_disable(fs_info);
4228                 break;
4229         default:
4230                 ret = -EINVAL;
4231                 break;
4232         }
4233
4234         kfree(sa);
4235         up_write(&fs_info->subvol_sem);
4236 drop_write:
4237         mnt_drop_write_file(file);
4238         return ret;
4239 }
4240
4241 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4242 {
4243         struct inode *inode = file_inode(file);
4244         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4245         struct btrfs_root *root = BTRFS_I(inode)->root;
4246         struct btrfs_ioctl_qgroup_assign_args *sa;
4247         struct btrfs_trans_handle *trans;
4248         int ret;
4249         int err;
4250
4251         if (!capable(CAP_SYS_ADMIN))
4252                 return -EPERM;
4253
4254         ret = mnt_want_write_file(file);
4255         if (ret)
4256                 return ret;
4257
4258         sa = memdup_user(arg, sizeof(*sa));
4259         if (IS_ERR(sa)) {
4260                 ret = PTR_ERR(sa);
4261                 goto drop_write;
4262         }
4263
4264         trans = btrfs_join_transaction(root);
4265         if (IS_ERR(trans)) {
4266                 ret = PTR_ERR(trans);
4267                 goto out;
4268         }
4269
4270         if (sa->assign) {
4271                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4272         } else {
4273                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4274         }
4275
4276         /* update qgroup status and info */
4277         err = btrfs_run_qgroups(trans);
4278         if (err < 0)
4279                 btrfs_handle_fs_error(fs_info, err,
4280                                       "failed to update qgroup status and info");
4281         err = btrfs_end_transaction(trans);
4282         if (err && !ret)
4283                 ret = err;
4284
4285 out:
4286         kfree(sa);
4287 drop_write:
4288         mnt_drop_write_file(file);
4289         return ret;
4290 }
4291
4292 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4293 {
4294         struct inode *inode = file_inode(file);
4295         struct btrfs_root *root = BTRFS_I(inode)->root;
4296         struct btrfs_ioctl_qgroup_create_args *sa;
4297         struct btrfs_trans_handle *trans;
4298         int ret;
4299         int err;
4300
4301         if (!capable(CAP_SYS_ADMIN))
4302                 return -EPERM;
4303
4304         ret = mnt_want_write_file(file);
4305         if (ret)
4306                 return ret;
4307
4308         sa = memdup_user(arg, sizeof(*sa));
4309         if (IS_ERR(sa)) {
4310                 ret = PTR_ERR(sa);
4311                 goto drop_write;
4312         }
4313
4314         if (!sa->qgroupid) {
4315                 ret = -EINVAL;
4316                 goto out;
4317         }
4318
4319         trans = btrfs_join_transaction(root);
4320         if (IS_ERR(trans)) {
4321                 ret = PTR_ERR(trans);
4322                 goto out;
4323         }
4324
4325         if (sa->create) {
4326                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4327         } else {
4328                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4329         }
4330
4331         err = btrfs_end_transaction(trans);
4332         if (err && !ret)
4333                 ret = err;
4334
4335 out:
4336         kfree(sa);
4337 drop_write:
4338         mnt_drop_write_file(file);
4339         return ret;
4340 }
4341
4342 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4343 {
4344         struct inode *inode = file_inode(file);
4345         struct btrfs_root *root = BTRFS_I(inode)->root;
4346         struct btrfs_ioctl_qgroup_limit_args *sa;
4347         struct btrfs_trans_handle *trans;
4348         int ret;
4349         int err;
4350         u64 qgroupid;
4351
4352         if (!capable(CAP_SYS_ADMIN))
4353                 return -EPERM;
4354
4355         ret = mnt_want_write_file(file);
4356         if (ret)
4357                 return ret;
4358
4359         sa = memdup_user(arg, sizeof(*sa));
4360         if (IS_ERR(sa)) {
4361                 ret = PTR_ERR(sa);
4362                 goto drop_write;
4363         }
4364
4365         trans = btrfs_join_transaction(root);
4366         if (IS_ERR(trans)) {
4367                 ret = PTR_ERR(trans);
4368                 goto out;
4369         }
4370
4371         qgroupid = sa->qgroupid;
4372         if (!qgroupid) {
4373                 /* take the current subvol as qgroup */
4374                 qgroupid = root->root_key.objectid;
4375         }
4376
4377         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4378
4379         err = btrfs_end_transaction(trans);
4380         if (err && !ret)
4381                 ret = err;
4382
4383 out:
4384         kfree(sa);
4385 drop_write:
4386         mnt_drop_write_file(file);
4387         return ret;
4388 }
4389
4390 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4391 {
4392         struct inode *inode = file_inode(file);
4393         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4394         struct btrfs_ioctl_quota_rescan_args *qsa;
4395         int ret;
4396
4397         if (!capable(CAP_SYS_ADMIN))
4398                 return -EPERM;
4399
4400         ret = mnt_want_write_file(file);
4401         if (ret)
4402                 return ret;
4403
4404         qsa = memdup_user(arg, sizeof(*qsa));
4405         if (IS_ERR(qsa)) {
4406                 ret = PTR_ERR(qsa);
4407                 goto drop_write;
4408         }
4409
4410         if (qsa->flags) {
4411                 ret = -EINVAL;
4412                 goto out;
4413         }
4414
4415         ret = btrfs_qgroup_rescan(fs_info);
4416
4417 out:
4418         kfree(qsa);
4419 drop_write:
4420         mnt_drop_write_file(file);
4421         return ret;
4422 }
4423
4424 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4425                                                 void __user *arg)
4426 {
4427         struct btrfs_ioctl_quota_rescan_args qsa = {0};
4428         int ret = 0;
4429
4430         if (!capable(CAP_SYS_ADMIN))
4431                 return -EPERM;
4432
4433         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4434                 qsa.flags = 1;
4435                 qsa.progress = fs_info->qgroup_rescan_progress.objectid;
4436         }
4437
4438         if (copy_to_user(arg, &qsa, sizeof(qsa)))
4439                 ret = -EFAULT;
4440
4441         return ret;
4442 }
4443
4444 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4445                                                 void __user *arg)
4446 {
4447         if (!capable(CAP_SYS_ADMIN))
4448                 return -EPERM;
4449
4450         return btrfs_qgroup_wait_for_completion(fs_info, true);
4451 }
4452
4453 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4454                                             struct user_namespace *mnt_userns,
4455                                             struct btrfs_ioctl_received_subvol_args *sa)
4456 {
4457         struct inode *inode = file_inode(file);
4458         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4459         struct btrfs_root *root = BTRFS_I(inode)->root;
4460         struct btrfs_root_item *root_item = &root->root_item;
4461         struct btrfs_trans_handle *trans;
4462         struct timespec64 ct = current_time(inode);
4463         int ret = 0;
4464         int received_uuid_changed;
4465
4466         if (!inode_owner_or_capable(mnt_userns, inode))
4467                 return -EPERM;
4468
4469         ret = mnt_want_write_file(file);
4470         if (ret < 0)
4471                 return ret;
4472
4473         down_write(&fs_info->subvol_sem);
4474
4475         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4476                 ret = -EINVAL;
4477                 goto out;
4478         }
4479
4480         if (btrfs_root_readonly(root)) {
4481                 ret = -EROFS;
4482                 goto out;
4483         }
4484
4485         /*
4486          * 1 - root item
4487          * 2 - uuid items (received uuid + subvol uuid)
4488          */
4489         trans = btrfs_start_transaction(root, 3);
4490         if (IS_ERR(trans)) {
4491                 ret = PTR_ERR(trans);
4492                 trans = NULL;
4493                 goto out;
4494         }
4495
4496         sa->rtransid = trans->transid;
4497         sa->rtime.sec = ct.tv_sec;
4498         sa->rtime.nsec = ct.tv_nsec;
4499
4500         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4501                                        BTRFS_UUID_SIZE);
4502         if (received_uuid_changed &&
4503             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4504                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4505                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4506                                           root->root_key.objectid);
4507                 if (ret && ret != -ENOENT) {
4508                         btrfs_abort_transaction(trans, ret);
4509                         btrfs_end_transaction(trans);
4510                         goto out;
4511                 }
4512         }
4513         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4514         btrfs_set_root_stransid(root_item, sa->stransid);
4515         btrfs_set_root_rtransid(root_item, sa->rtransid);
4516         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4517         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4518         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4519         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4520
4521         ret = btrfs_update_root(trans, fs_info->tree_root,
4522                                 &root->root_key, &root->root_item);
4523         if (ret < 0) {
4524                 btrfs_end_transaction(trans);
4525                 goto out;
4526         }
4527         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4528                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4529                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4530                                           root->root_key.objectid);
4531                 if (ret < 0 && ret != -EEXIST) {
4532                         btrfs_abort_transaction(trans, ret);
4533                         btrfs_end_transaction(trans);
4534                         goto out;
4535                 }
4536         }
4537         ret = btrfs_commit_transaction(trans);
4538 out:
4539         up_write(&fs_info->subvol_sem);
4540         mnt_drop_write_file(file);
4541         return ret;
4542 }
4543
4544 #ifdef CONFIG_64BIT
4545 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4546                                                 void __user *arg)
4547 {
4548         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4549         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4550         int ret = 0;
4551
4552         args32 = memdup_user(arg, sizeof(*args32));
4553         if (IS_ERR(args32))
4554                 return PTR_ERR(args32);
4555
4556         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4557         if (!args64) {
4558                 ret = -ENOMEM;
4559                 goto out;
4560         }
4561
4562         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4563         args64->stransid = args32->stransid;
4564         args64->rtransid = args32->rtransid;
4565         args64->stime.sec = args32->stime.sec;
4566         args64->stime.nsec = args32->stime.nsec;
4567         args64->rtime.sec = args32->rtime.sec;
4568         args64->rtime.nsec = args32->rtime.nsec;
4569         args64->flags = args32->flags;
4570
4571         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64);
4572         if (ret)
4573                 goto out;
4574
4575         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4576         args32->stransid = args64->stransid;
4577         args32->rtransid = args64->rtransid;
4578         args32->stime.sec = args64->stime.sec;
4579         args32->stime.nsec = args64->stime.nsec;
4580         args32->rtime.sec = args64->rtime.sec;
4581         args32->rtime.nsec = args64->rtime.nsec;
4582         args32->flags = args64->flags;
4583
4584         ret = copy_to_user(arg, args32, sizeof(*args32));
4585         if (ret)
4586                 ret = -EFAULT;
4587
4588 out:
4589         kfree(args32);
4590         kfree(args64);
4591         return ret;
4592 }
4593 #endif
4594
4595 static long btrfs_ioctl_set_received_subvol(struct file *file,
4596                                             void __user *arg)
4597 {
4598         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4599         int ret = 0;
4600
4601         sa = memdup_user(arg, sizeof(*sa));
4602         if (IS_ERR(sa))
4603                 return PTR_ERR(sa);
4604
4605         ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa);
4606
4607         if (ret)
4608                 goto out;
4609
4610         ret = copy_to_user(arg, sa, sizeof(*sa));
4611         if (ret)
4612                 ret = -EFAULT;
4613
4614 out:
4615         kfree(sa);
4616         return ret;
4617 }
4618
4619 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4620                                         void __user *arg)
4621 {
4622         size_t len;
4623         int ret;
4624         char label[BTRFS_LABEL_SIZE];
4625
4626         spin_lock(&fs_info->super_lock);
4627         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4628         spin_unlock(&fs_info->super_lock);
4629
4630         len = strnlen(label, BTRFS_LABEL_SIZE);
4631
4632         if (len == BTRFS_LABEL_SIZE) {
4633                 btrfs_warn(fs_info,
4634                            "label is too long, return the first %zu bytes",
4635                            --len);
4636         }
4637
4638         ret = copy_to_user(arg, label, len);
4639
4640         return ret ? -EFAULT : 0;
4641 }
4642
4643 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4644 {
4645         struct inode *inode = file_inode(file);
4646         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4647         struct btrfs_root *root = BTRFS_I(inode)->root;
4648         struct btrfs_super_block *super_block = fs_info->super_copy;
4649         struct btrfs_trans_handle *trans;
4650         char label[BTRFS_LABEL_SIZE];
4651         int ret;
4652
4653         if (!capable(CAP_SYS_ADMIN))
4654                 return -EPERM;
4655
4656         if (copy_from_user(label, arg, sizeof(label)))
4657                 return -EFAULT;
4658
4659         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4660                 btrfs_err(fs_info,
4661                           "unable to set label with more than %d bytes",
4662                           BTRFS_LABEL_SIZE - 1);
4663                 return -EINVAL;
4664         }
4665
4666         ret = mnt_want_write_file(file);
4667         if (ret)
4668                 return ret;
4669
4670         trans = btrfs_start_transaction(root, 0);
4671         if (IS_ERR(trans)) {
4672                 ret = PTR_ERR(trans);
4673                 goto out_unlock;
4674         }
4675
4676         spin_lock(&fs_info->super_lock);
4677         strcpy(super_block->label, label);
4678         spin_unlock(&fs_info->super_lock);
4679         ret = btrfs_commit_transaction(trans);
4680
4681 out_unlock:
4682         mnt_drop_write_file(file);
4683         return ret;
4684 }
4685
4686 #define INIT_FEATURE_FLAGS(suffix) \
4687         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4688           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4689           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4690
4691 int btrfs_ioctl_get_supported_features(void __user *arg)
4692 {
4693         static const struct btrfs_ioctl_feature_flags features[3] = {
4694                 INIT_FEATURE_FLAGS(SUPP),
4695                 INIT_FEATURE_FLAGS(SAFE_SET),
4696                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4697         };
4698
4699         if (copy_to_user(arg, &features, sizeof(features)))
4700                 return -EFAULT;
4701
4702         return 0;
4703 }
4704
4705 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4706                                         void __user *arg)
4707 {
4708         struct btrfs_super_block *super_block = fs_info->super_copy;
4709         struct btrfs_ioctl_feature_flags features;
4710
4711         features.compat_flags = btrfs_super_compat_flags(super_block);
4712         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4713         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4714
4715         if (copy_to_user(arg, &features, sizeof(features)))
4716                 return -EFAULT;
4717
4718         return 0;
4719 }
4720
4721 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4722                               enum btrfs_feature_set set,
4723                               u64 change_mask, u64 flags, u64 supported_flags,
4724                               u64 safe_set, u64 safe_clear)
4725 {
4726         const char *type = btrfs_feature_set_name(set);
4727         char *names;
4728         u64 disallowed, unsupported;
4729         u64 set_mask = flags & change_mask;
4730         u64 clear_mask = ~flags & change_mask;
4731
4732         unsupported = set_mask & ~supported_flags;
4733         if (unsupported) {
4734                 names = btrfs_printable_features(set, unsupported);
4735                 if (names) {
4736                         btrfs_warn(fs_info,
4737                                    "this kernel does not support the %s feature bit%s",
4738                                    names, strchr(names, ',') ? "s" : "");
4739                         kfree(names);
4740                 } else
4741                         btrfs_warn(fs_info,
4742                                    "this kernel does not support %s bits 0x%llx",
4743                                    type, unsupported);
4744                 return -EOPNOTSUPP;
4745         }
4746
4747         disallowed = set_mask & ~safe_set;
4748         if (disallowed) {
4749                 names = btrfs_printable_features(set, disallowed);
4750                 if (names) {
4751                         btrfs_warn(fs_info,
4752                                    "can't set the %s feature bit%s while mounted",
4753                                    names, strchr(names, ',') ? "s" : "");
4754                         kfree(names);
4755                 } else
4756                         btrfs_warn(fs_info,
4757                                    "can't set %s bits 0x%llx while mounted",
4758                                    type, disallowed);
4759                 return -EPERM;
4760         }
4761
4762         disallowed = clear_mask & ~safe_clear;
4763         if (disallowed) {
4764                 names = btrfs_printable_features(set, disallowed);
4765                 if (names) {
4766                         btrfs_warn(fs_info,
4767                                    "can't clear the %s feature bit%s while mounted",
4768                                    names, strchr(names, ',') ? "s" : "");
4769                         kfree(names);
4770                 } else
4771                         btrfs_warn(fs_info,
4772                                    "can't clear %s bits 0x%llx while mounted",
4773                                    type, disallowed);
4774                 return -EPERM;
4775         }
4776
4777         return 0;
4778 }
4779
4780 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4781 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4782                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4783                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4784                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4785
4786 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4787 {
4788         struct inode *inode = file_inode(file);
4789         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4790         struct btrfs_root *root = BTRFS_I(inode)->root;
4791         struct btrfs_super_block *super_block = fs_info->super_copy;
4792         struct btrfs_ioctl_feature_flags flags[2];
4793         struct btrfs_trans_handle *trans;
4794         u64 newflags;
4795         int ret;
4796
4797         if (!capable(CAP_SYS_ADMIN))
4798                 return -EPERM;
4799
4800         if (copy_from_user(flags, arg, sizeof(flags)))
4801                 return -EFAULT;
4802
4803         /* Nothing to do */
4804         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4805             !flags[0].incompat_flags)
4806                 return 0;
4807
4808         ret = check_feature(fs_info, flags[0].compat_flags,
4809                             flags[1].compat_flags, COMPAT);
4810         if (ret)
4811                 return ret;
4812
4813         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4814                             flags[1].compat_ro_flags, COMPAT_RO);
4815         if (ret)
4816                 return ret;
4817
4818         ret = check_feature(fs_info, flags[0].incompat_flags,
4819                             flags[1].incompat_flags, INCOMPAT);
4820         if (ret)
4821                 return ret;
4822
4823         ret = mnt_want_write_file(file);
4824         if (ret)
4825                 return ret;
4826
4827         trans = btrfs_start_transaction(root, 0);
4828         if (IS_ERR(trans)) {
4829                 ret = PTR_ERR(trans);
4830                 goto out_drop_write;
4831         }
4832
4833         spin_lock(&fs_info->super_lock);
4834         newflags = btrfs_super_compat_flags(super_block);
4835         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4836         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4837         btrfs_set_super_compat_flags(super_block, newflags);
4838
4839         newflags = btrfs_super_compat_ro_flags(super_block);
4840         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4841         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4842         btrfs_set_super_compat_ro_flags(super_block, newflags);
4843
4844         newflags = btrfs_super_incompat_flags(super_block);
4845         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4846         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4847         btrfs_set_super_incompat_flags(super_block, newflags);
4848         spin_unlock(&fs_info->super_lock);
4849
4850         ret = btrfs_commit_transaction(trans);
4851 out_drop_write:
4852         mnt_drop_write_file(file);
4853
4854         return ret;
4855 }
4856
4857 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4858 {
4859         struct btrfs_ioctl_send_args *arg;
4860         int ret;
4861
4862         if (compat) {
4863 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4864                 struct btrfs_ioctl_send_args_32 args32;
4865
4866                 ret = copy_from_user(&args32, argp, sizeof(args32));
4867                 if (ret)
4868                         return -EFAULT;
4869                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4870                 if (!arg)
4871                         return -ENOMEM;
4872                 arg->send_fd = args32.send_fd;
4873                 arg->clone_sources_count = args32.clone_sources_count;
4874                 arg->clone_sources = compat_ptr(args32.clone_sources);
4875                 arg->parent_root = args32.parent_root;
4876                 arg->flags = args32.flags;
4877                 memcpy(arg->reserved, args32.reserved,
4878                        sizeof(args32.reserved));
4879 #else
4880                 return -ENOTTY;
4881 #endif
4882         } else {
4883                 arg = memdup_user(argp, sizeof(*arg));
4884                 if (IS_ERR(arg))
4885                         return PTR_ERR(arg);
4886         }
4887         ret = btrfs_ioctl_send(file, arg);
4888         kfree(arg);
4889         return ret;
4890 }
4891
4892 long btrfs_ioctl(struct file *file, unsigned int
4893                 cmd, unsigned long arg)
4894 {
4895         struct inode *inode = file_inode(file);
4896         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4897         struct btrfs_root *root = BTRFS_I(inode)->root;
4898         void __user *argp = (void __user *)arg;
4899
4900         switch (cmd) {
4901         case FS_IOC_GETVERSION:
4902                 return btrfs_ioctl_getversion(file, argp);
4903         case FS_IOC_GETFSLABEL:
4904                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4905         case FS_IOC_SETFSLABEL:
4906                 return btrfs_ioctl_set_fslabel(file, argp);
4907         case FITRIM:
4908                 return btrfs_ioctl_fitrim(fs_info, argp);
4909         case BTRFS_IOC_SNAP_CREATE:
4910                 return btrfs_ioctl_snap_create(file, argp, 0);
4911         case BTRFS_IOC_SNAP_CREATE_V2:
4912                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4913         case BTRFS_IOC_SUBVOL_CREATE:
4914                 return btrfs_ioctl_snap_create(file, argp, 1);
4915         case BTRFS_IOC_SUBVOL_CREATE_V2:
4916                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4917         case BTRFS_IOC_SNAP_DESTROY:
4918                 return btrfs_ioctl_snap_destroy(file, argp, false);
4919         case BTRFS_IOC_SNAP_DESTROY_V2:
4920                 return btrfs_ioctl_snap_destroy(file, argp, true);
4921         case BTRFS_IOC_SUBVOL_GETFLAGS:
4922                 return btrfs_ioctl_subvol_getflags(file, argp);
4923         case BTRFS_IOC_SUBVOL_SETFLAGS:
4924                 return btrfs_ioctl_subvol_setflags(file, argp);
4925         case BTRFS_IOC_DEFAULT_SUBVOL:
4926                 return btrfs_ioctl_default_subvol(file, argp);
4927         case BTRFS_IOC_DEFRAG:
4928                 return btrfs_ioctl_defrag(file, NULL);
4929         case BTRFS_IOC_DEFRAG_RANGE:
4930                 return btrfs_ioctl_defrag(file, argp);
4931         case BTRFS_IOC_RESIZE:
4932                 return btrfs_ioctl_resize(file, argp);
4933         case BTRFS_IOC_ADD_DEV:
4934                 return btrfs_ioctl_add_dev(fs_info, argp);
4935         case BTRFS_IOC_RM_DEV:
4936                 return btrfs_ioctl_rm_dev(file, argp);
4937         case BTRFS_IOC_RM_DEV_V2:
4938                 return btrfs_ioctl_rm_dev_v2(file, argp);
4939         case BTRFS_IOC_FS_INFO:
4940                 return btrfs_ioctl_fs_info(fs_info, argp);
4941         case BTRFS_IOC_DEV_INFO:
4942                 return btrfs_ioctl_dev_info(fs_info, argp);
4943         case BTRFS_IOC_BALANCE:
4944                 return btrfs_ioctl_balance(file, NULL);
4945         case BTRFS_IOC_TREE_SEARCH:
4946                 return btrfs_ioctl_tree_search(file, argp);
4947         case BTRFS_IOC_TREE_SEARCH_V2:
4948                 return btrfs_ioctl_tree_search_v2(file, argp);
4949         case BTRFS_IOC_INO_LOOKUP:
4950                 return btrfs_ioctl_ino_lookup(file, argp);
4951         case BTRFS_IOC_INO_PATHS:
4952                 return btrfs_ioctl_ino_to_path(root, argp);
4953         case BTRFS_IOC_LOGICAL_INO:
4954                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4955         case BTRFS_IOC_LOGICAL_INO_V2:
4956                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4957         case BTRFS_IOC_SPACE_INFO:
4958                 return btrfs_ioctl_space_info(fs_info, argp);
4959         case BTRFS_IOC_SYNC: {
4960                 int ret;
4961
4962                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4963                 if (ret)
4964                         return ret;
4965                 ret = btrfs_sync_fs(inode->i_sb, 1);
4966                 /*
4967                  * The transaction thread may want to do more work,
4968                  * namely it pokes the cleaner kthread that will start
4969                  * processing uncleaned subvols.
4970                  */
4971                 wake_up_process(fs_info->transaction_kthread);
4972                 return ret;
4973         }
4974         case BTRFS_IOC_START_SYNC:
4975                 return btrfs_ioctl_start_sync(root, argp);
4976         case BTRFS_IOC_WAIT_SYNC:
4977                 return btrfs_ioctl_wait_sync(fs_info, argp);
4978         case BTRFS_IOC_SCRUB:
4979                 return btrfs_ioctl_scrub(file, argp);
4980         case BTRFS_IOC_SCRUB_CANCEL:
4981                 return btrfs_ioctl_scrub_cancel(fs_info);
4982         case BTRFS_IOC_SCRUB_PROGRESS:
4983                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4984         case BTRFS_IOC_BALANCE_V2:
4985                 return btrfs_ioctl_balance(file, argp);
4986         case BTRFS_IOC_BALANCE_CTL:
4987                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4988         case BTRFS_IOC_BALANCE_PROGRESS:
4989                 return btrfs_ioctl_balance_progress(fs_info, argp);
4990         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4991                 return btrfs_ioctl_set_received_subvol(file, argp);
4992 #ifdef CONFIG_64BIT
4993         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4994                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4995 #endif
4996         case BTRFS_IOC_SEND:
4997                 return _btrfs_ioctl_send(file, argp, false);
4998 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4999         case BTRFS_IOC_SEND_32:
5000                 return _btrfs_ioctl_send(file, argp, true);
5001 #endif
5002         case BTRFS_IOC_GET_DEV_STATS:
5003                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5004         case BTRFS_IOC_QUOTA_CTL:
5005                 return btrfs_ioctl_quota_ctl(file, argp);
5006         case BTRFS_IOC_QGROUP_ASSIGN:
5007                 return btrfs_ioctl_qgroup_assign(file, argp);
5008         case BTRFS_IOC_QGROUP_CREATE:
5009                 return btrfs_ioctl_qgroup_create(file, argp);
5010         case BTRFS_IOC_QGROUP_LIMIT:
5011                 return btrfs_ioctl_qgroup_limit(file, argp);
5012         case BTRFS_IOC_QUOTA_RESCAN:
5013                 return btrfs_ioctl_quota_rescan(file, argp);
5014         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5015                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5016         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5017                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5018         case BTRFS_IOC_DEV_REPLACE:
5019                 return btrfs_ioctl_dev_replace(fs_info, argp);
5020         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5021                 return btrfs_ioctl_get_supported_features(argp);
5022         case BTRFS_IOC_GET_FEATURES:
5023                 return btrfs_ioctl_get_features(fs_info, argp);
5024         case BTRFS_IOC_SET_FEATURES:
5025                 return btrfs_ioctl_set_features(file, argp);
5026         case BTRFS_IOC_GET_SUBVOL_INFO:
5027                 return btrfs_ioctl_get_subvol_info(file, argp);
5028         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5029                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5030         case BTRFS_IOC_INO_LOOKUP_USER:
5031                 return btrfs_ioctl_ino_lookup_user(file, argp);
5032         case FS_IOC_ENABLE_VERITY:
5033                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5034         case FS_IOC_MEASURE_VERITY:
5035                 return fsverity_ioctl_measure(file, argp);
5036         }
5037
5038         return -ENOTTY;
5039 }
5040
5041 #ifdef CONFIG_COMPAT
5042 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5043 {
5044         /*
5045          * These all access 32-bit values anyway so no further
5046          * handling is necessary.
5047          */
5048         switch (cmd) {
5049         case FS_IOC32_GETVERSION:
5050                 cmd = FS_IOC_GETVERSION;
5051                 break;
5052         }
5053
5054         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5055 }
5056 #endif