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