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 inode *dir,
 503                                   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);
 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 inode *dir, struct dentry *victim, int isdir)
 841 {
 842         int error;
 843
 844         if (d_really_is_negative(victim))
 845                 return -ENOENT;
 846
 847         BUG_ON(d_inode(victim->d_parent) != dir);
 848         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 849
 850         error = inode_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
 851         if (error)
 852                 return error;
 853         if (IS_APPEND(dir))
 854                 return -EPERM;
 855         if (check_sticky(&init_user_ns, dir, d_inode(victim)) ||
 856             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 857             IS_SWAPFILE(d_inode(victim)))
 858                 return -EPERM;
 859         if (isdir) {
 860                 if (!d_is_dir(victim))
 861                         return -ENOTDIR;
 862                 if (IS_ROOT(victim))
 863                         return -EBUSY;
 864         } else if (d_is_dir(victim))
 865                 return -EISDIR;
 866         if (IS_DEADDIR(dir))
 867                 return -ENOENT;
 868         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 869                 return -EBUSY;
 870         return 0;
 871 }
 872
 873 /* copy of may_create in fs/namei.c() */
 874 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 875 {
 876         if (d_really_is_positive(child))
 877                 return -EEXIST;
 878         if (IS_DEADDIR(dir))
 879                 return -ENOENT;
 880         return inode_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
 881 }
 882
 883 /*
 884  * Create a new subvolume below @parent.  This is largely modeled after
 885  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 886  * inside this filesystem so it's quite a bit simpler.
 887  */
 888 static noinline int btrfs_mksubvol(const struct path *parent,
 889                                    const char *name, int namelen,
 890                                    struct btrfs_root *snap_src,
 891                                    bool readonly,
 892                                    struct btrfs_qgroup_inherit *inherit)
 893 {
 894         struct inode *dir = d_inode(parent->dentry);
 895         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 896         struct dentry *dentry;
 897         int error;
 898
 899         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 900         if (error == -EINTR)
 901                 return error;
 902
 903         dentry = lookup_one_len(name, parent->dentry, namelen);
 904         error = PTR_ERR(dentry);
 905         if (IS_ERR(dentry))
 906                 goto out_unlock;
 907
 908         error = btrfs_may_create(dir, dentry);
 909         if (error)
 910                 goto out_dput;
 911
 912         /*
 913          * even if this name doesn't exist, we may get hash collisions.
 914          * check for them now when we can safely fail
 915          */
 916         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 917                                                dir->i_ino, name,
 918                                                namelen);
 919         if (error)
 920                 goto out_dput;
 921
 922         down_read(&fs_info->subvol_sem);
 923
 924         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 925                 goto out_up_read;
 926
 927         if (snap_src)
 928                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 929         else
 930                 error = create_subvol(dir, dentry, name, namelen, inherit);
 931
 932         if (!error)
 933                 fsnotify_mkdir(dir, dentry);
 934 out_up_read:
 935         up_read(&fs_info->subvol_sem);
 936 out_dput:
 937         dput(dentry);
 938 out_unlock:
 939         btrfs_inode_unlock(dir, 0);
 940         return error;
 941 }
 942
 943 static noinline int btrfs_mksnapshot(const struct path *parent,
 944                                    const char *name, int namelen,
 945                                    struct btrfs_root *root,
 946                                    bool readonly,
 947                                    struct btrfs_qgroup_inherit *inherit)
 948 {
 949         int ret;
 950         bool snapshot_force_cow = false;
 951
 952         /*
 953          * Force new buffered writes to reserve space even when NOCOW is
 954          * possible. This is to avoid later writeback (running dealloc) to
 955          * fallback to COW mode and unexpectedly fail with ENOSPC.
 956          */
 957         btrfs_drew_read_lock(&root->snapshot_lock);
 958
 959         ret = btrfs_start_delalloc_snapshot(root, false);
 960         if (ret)
 961                 goto out;
 962
 963         /*
 964          * All previous writes have started writeback in NOCOW mode, so now
 965          * we force future writes to fallback to COW mode during snapshot
 966          * creation.
 967          */
 968         atomic_inc(&root->snapshot_force_cow);
 969         snapshot_force_cow = true;
 970
 971         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 972
 973         ret = btrfs_mksubvol(parent, name, namelen,
 974                              root, readonly, inherit);
 975 out:
 976         if (snapshot_force_cow)
 977                 atomic_dec(&root->snapshot_force_cow);
 978         btrfs_drew_read_unlock(&root->snapshot_lock);
 979         return ret;
 980 }
 981
 982 /*
 983  * When we're defragging a range, we don't want to kick it off again
 984  * if it is really just waiting for delalloc to send it down.
 985  * If we find a nice big extent or delalloc range for the bytes in the
 986  * file you want to defrag, we return 0 to let you know to skip this
 987  * part of the file
 988  */
 989 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 990 {
 991         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 992         struct extent_map *em = NULL;
 993         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 994         u64 end;
 995
 996         read_lock(&em_tree->lock);
 997         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
 998         read_unlock(&em_tree->lock);
 999
1000         if (em) {
1001                 end = extent_map_end(em);
1002                 free_extent_map(em);
1003                 if (end - offset > thresh)
1004                         return 0;
1005         }
1006         /* if we already have a nice delalloc here, just stop */
1007         thresh /= 2;
1008         end = count_range_bits(io_tree, &offset, offset + thresh,
1009                                thresh, EXTENT_DELALLOC, 1);
1010         if (end >= thresh)
1011                 return 0;
1012         return 1;
1013 }
1014
1015 /*
1016  * helper function to walk through a file and find extents
1017  * newer than a specific transid, and smaller than thresh.
1018  *
1019  * This is used by the defragging code to find new and small
1020  * extents
1021  */
1022 static int find_new_extents(struct btrfs_root *root,
1023                             struct inode *inode, u64 newer_than,
1024                             u64 *off, u32 thresh)
1025 {
1026         struct btrfs_path *path;
1027         struct btrfs_key min_key;
1028         struct extent_buffer *leaf;
1029         struct btrfs_file_extent_item *extent;
1030         int type;
1031         int ret;
1032         u64 ino = btrfs_ino(BTRFS_I(inode));
1033
1034         path = btrfs_alloc_path();
1035         if (!path)
1036                 return -ENOMEM;
1037
1038         min_key.objectid = ino;
1039         min_key.type = BTRFS_EXTENT_DATA_KEY;
1040         min_key.offset = *off;
1041
1042         while (1) {
1043                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1044                 if (ret != 0)
1045                         goto none;
1046 process_slot:
1047                 if (min_key.objectid != ino)
1048                         goto none;
1049                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1050                         goto none;
1051
1052                 leaf = path->nodes[0];
1053                 extent = btrfs_item_ptr(leaf, path->slots[0],
1054                                         struct btrfs_file_extent_item);
1055
1056                 type = btrfs_file_extent_type(leaf, extent);
1057                 if (type == BTRFS_FILE_EXTENT_REG &&
1058                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1059                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1060                         *off = min_key.offset;
1061                         btrfs_free_path(path);
1062                         return 0;
1063                 }
1064
1065                 path->slots[0]++;
1066                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1067                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1068                         goto process_slot;
1069                 }
1070
1071                 if (min_key.offset == (u64)-1)
1072                         goto none;
1073
1074                 min_key.offset++;
1075                 btrfs_release_path(path);
1076         }
1077 none:
1078         btrfs_free_path(path);
1079         return -ENOENT;
1080 }
1081
1082 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1083 {
1084         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1085         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1086         struct extent_map *em;
1087         u64 len = PAGE_SIZE;
1088
1089         /*
1090          * hopefully we have this extent in the tree already, try without
1091          * the full extent lock
1092          */
1093         read_lock(&em_tree->lock);
1094         em = lookup_extent_mapping(em_tree, start, len);
1095         read_unlock(&em_tree->lock);
1096
1097         if (!em) {
1098                 struct extent_state *cached = NULL;
1099                 u64 end = start + len - 1;
1100
1101                 /* get the big lock and read metadata off disk */
1102                 lock_extent_bits(io_tree, start, end, &cached);
1103                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1104                 unlock_extent_cached(io_tree, start, end, &cached);
1105
1106                 if (IS_ERR(em))
1107                         return NULL;
1108         }
1109
1110         return em;
1111 }
1112
1113 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1114 {
1115         struct extent_map *next;
1116         bool ret = true;
1117
1118         /* this is the last extent */
1119         if (em->start + em->len >= i_size_read(inode))
1120                 return false;
1121
1122         next = defrag_lookup_extent(inode, em->start + em->len);
1123         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1124                 ret = false;
1125         else if ((em->block_start + em->block_len == next->block_start) &&
1126                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1127                 ret = false;
1128
1129         free_extent_map(next);
1130         return ret;
1131 }
1132
1133 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1134                                u64 *last_len, u64 *skip, u64 *defrag_end,
1135                                int compress)
1136 {
1137         struct extent_map *em;
1138         int ret = 1;
1139         bool next_mergeable = true;
1140         bool prev_mergeable = true;
1141
1142         /*
1143          * make sure that once we start defragging an extent, we keep on
1144          * defragging it
1145          */
1146         if (start < *defrag_end)
1147                 return 1;
1148
1149         *skip = 0;
1150
1151         em = defrag_lookup_extent(inode, start);
1152         if (!em)
1153                 return 0;
1154
1155         /* this will cover holes, and inline extents */
1156         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1157                 ret = 0;
1158                 goto out;
1159         }
1160
1161         if (!*defrag_end)
1162                 prev_mergeable = false;
1163
1164         next_mergeable = defrag_check_next_extent(inode, em);
1165         /*
1166          * we hit a real extent, if it is big or the next extent is not a
1167          * real extent, don't bother defragging it
1168          */
1169         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1170             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1171                 ret = 0;
1172 out:
1173         /*
1174          * last_len ends up being a counter of how many bytes we've defragged.
1175          * every time we choose not to defrag an extent, we reset *last_len
1176          * so that the next tiny extent will force a defrag.
1177          *
1178          * The end result of this is that tiny extents before a single big
1179          * extent will force at least part of that big extent to be defragged.
1180          */
1181         if (ret) {
1182                 *defrag_end = extent_map_end(em);
1183         } else {
1184                 *last_len = 0;
1185                 *skip = extent_map_end(em);
1186                 *defrag_end = 0;
1187         }
1188
1189         free_extent_map(em);
1190         return ret;
1191 }
1192
1193 /*
1194  * it doesn't do much good to defrag one or two pages
1195  * at a time.  This pulls in a nice chunk of pages
1196  * to COW and defrag.
1197  *
1198  * It also makes sure the delalloc code has enough
1199  * dirty data to avoid making new small extents as part
1200  * of the defrag
1201  *
1202  * It's a good idea to start RA on this range
1203  * before calling this.
1204  */
1205 static int cluster_pages_for_defrag(struct inode *inode,
1206                                     struct page **pages,
1207                                     unsigned long start_index,
1208                                     unsigned long num_pages)
1209 {
1210         unsigned long file_end;
1211         u64 isize = i_size_read(inode);
1212         u64 page_start;
1213         u64 page_end;
1214         u64 page_cnt;
1215         u64 start = (u64)start_index << PAGE_SHIFT;
1216         u64 search_start;
1217         int ret;
1218         int i;
1219         int i_done;
1220         struct btrfs_ordered_extent *ordered;
1221         struct extent_state *cached_state = NULL;
1222         struct extent_io_tree *tree;
1223         struct extent_changeset *data_reserved = NULL;
1224         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1225
1226         file_end = (isize - 1) >> PAGE_SHIFT;
1227         if (!isize || start_index > file_end)
1228                 return 0;
1229
1230         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1231
1232         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1233                         start, page_cnt << PAGE_SHIFT);
1234         if (ret)
1235                 return ret;
1236         i_done = 0;
1237         tree = &BTRFS_I(inode)->io_tree;
1238
1239         /* step one, lock all the pages */
1240         for (i = 0; i < page_cnt; i++) {
1241                 struct page *page;
1242 again:
1243                 page = find_or_create_page(inode->i_mapping,
1244                                            start_index + i, mask);
1245                 if (!page)
1246                         break;
1247
1248                 ret = set_page_extent_mapped(page);
1249                 if (ret < 0) {
1250                         unlock_page(page);
1251                         put_page(page);
1252                         break;
1253                 }
1254
1255                 page_start = page_offset(page);
1256                 page_end = page_start + PAGE_SIZE - 1;
1257                 while (1) {
1258                         lock_extent_bits(tree, page_start, page_end,
1259                                          &cached_state);
1260                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1261                                                               page_start);
1262                         unlock_extent_cached(tree, page_start, page_end,
1263                                              &cached_state);
1264                         if (!ordered)
1265                                 break;
1266
1267                         unlock_page(page);
1268                         btrfs_start_ordered_extent(ordered, 1);
1269                         btrfs_put_ordered_extent(ordered);
1270                         lock_page(page);
1271                         /*
1272                          * we unlocked the page above, so we need check if
1273                          * it was released or not.
1274                          */
1275                         if (page->mapping != inode->i_mapping) {
1276                                 unlock_page(page);
1277                                 put_page(page);
1278                                 goto again;
1279                         }
1280                 }
1281
1282                 if (!PageUptodate(page)) {
1283                         btrfs_readpage(NULL, page);
1284                         lock_page(page);
1285                         if (!PageUptodate(page)) {
1286                                 unlock_page(page);
1287                                 put_page(page);
1288                                 ret = -EIO;
1289                                 break;
1290                         }
1291                 }
1292
1293                 if (page->mapping != inode->i_mapping) {
1294                         unlock_page(page);
1295                         put_page(page);
1296                         goto again;
1297                 }
1298
1299                 pages[i] = page;
1300                 i_done++;
1301         }
1302         if (!i_done || ret)
1303                 goto out;
1304
1305         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1306                 goto out;
1307
1308         /*
1309          * so now we have a nice long stream of locked
1310          * and up to date pages, lets wait on them
1311          */
1312         for (i = 0; i < i_done; i++)
1313                 wait_on_page_writeback(pages[i]);
1314
1315         page_start = page_offset(pages[0]);
1316         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1317
1318         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1319                          page_start, page_end - 1, &cached_state);
1320
1321         /*
1322          * When defragmenting we skip ranges that have holes or inline extents,
1323          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1324          * space. At btrfs_defrag_file(), we check if a range should be defragged
1325          * before locking the inode and then, if it should, we trigger a sync
1326          * page cache readahead - we lock the inode only after that to avoid
1327          * blocking for too long other tasks that possibly want to operate on
1328          * other file ranges. But before we were able to get the inode lock,
1329          * some other task may have punched a hole in the range, or we may have
1330          * now an inline extent, in which case we should not defrag. So check
1331          * for that here, where we have the inode and the range locked, and bail
1332          * out if that happened.
1333          */
1334         search_start = page_start;
1335         while (search_start < page_end) {
1336                 struct extent_map *em;
1337
1338                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1339                                       page_end - search_start);
1340                 if (IS_ERR(em)) {
1341                         ret = PTR_ERR(em);
1342                         goto out_unlock_range;
1343                 }
1344                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1345                         free_extent_map(em);
1346                         /* Ok, 0 means we did not defrag anything */
1347                         ret = 0;
1348                         goto out_unlock_range;
1349                 }
1350                 search_start = extent_map_end(em);
1351                 free_extent_map(em);
1352         }
1353
1354         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1355                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1356                           EXTENT_DEFRAG, 0, 0, &cached_state);
1357
1358         if (i_done != page_cnt) {
1359                 spin_lock(&BTRFS_I(inode)->lock);
1360                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1361                 spin_unlock(&BTRFS_I(inode)->lock);
1362                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1363                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1364         }
1365
1366
1367         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1368                           &cached_state);
1369
1370         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1371                              page_start, page_end - 1, &cached_state);
1372
1373         for (i = 0; i < i_done; i++) {
1374                 clear_page_dirty_for_io(pages[i]);
1375                 ClearPageChecked(pages[i]);
1376                 set_page_dirty(pages[i]);
1377                 unlock_page(pages[i]);
1378                 put_page(pages[i]);
1379         }
1380         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1381         extent_changeset_free(data_reserved);
1382         return i_done;
1383
1384 out_unlock_range:
1385         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1386                              page_start, page_end - 1, &cached_state);
1387 out:
1388         for (i = 0; i < i_done; i++) {
1389                 unlock_page(pages[i]);
1390                 put_page(pages[i]);
1391         }
1392         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1393                         start, page_cnt << PAGE_SHIFT, true);
1394         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1395         extent_changeset_free(data_reserved);
1396         return ret;
1397
1398 }
1399
1400 int btrfs_defrag_file(struct inode *inode, struct file *file,
1401                       struct btrfs_ioctl_defrag_range_args *range,
1402                       u64 newer_than, unsigned long max_to_defrag)
1403 {
1404         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1405         struct btrfs_root *root = BTRFS_I(inode)->root;
1406         struct file_ra_state *ra = NULL;
1407         unsigned long last_index;
1408         u64 isize = i_size_read(inode);
1409         u64 last_len = 0;
1410         u64 skip = 0;
1411         u64 defrag_end = 0;
1412         u64 newer_off = range->start;
1413         unsigned long i;
1414         unsigned long ra_index = 0;
1415         int ret;
1416         int defrag_count = 0;
1417         int compress_type = BTRFS_COMPRESS_ZLIB;
1418         u32 extent_thresh = range->extent_thresh;
1419         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1420         unsigned long cluster = max_cluster;
1421         u64 new_align = ~((u64)SZ_128K - 1);
1422         struct page **pages = NULL;
1423         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1424
1425         if (isize == 0)
1426                 return 0;
1427
1428         if (range->start >= isize)
1429                 return -EINVAL;
1430
1431         if (do_compress) {
1432                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1433                         return -EINVAL;
1434                 if (range->compress_type)
1435                         compress_type = range->compress_type;
1436         }
1437
1438         if (extent_thresh == 0)
1439                 extent_thresh = SZ_256K;
1440
1441         /*
1442          * If we were not given a file, allocate a readahead context. As
1443          * readahead is just an optimization, defrag will work without it so
1444          * we don't error out.
1445          */
1446         if (!file) {
1447                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1448                 if (ra)
1449                         file_ra_state_init(ra, inode->i_mapping);
1450         } else {
1451                 ra = &file->f_ra;
1452         }
1453
1454         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1455         if (!pages) {
1456                 ret = -ENOMEM;
1457                 goto out_ra;
1458         }
1459
1460         /* find the last page to defrag */
1461         if (range->start + range->len > range->start) {
1462                 last_index = min_t(u64, isize - 1,
1463                          range->start + range->len - 1) >> PAGE_SHIFT;
1464         } else {
1465                 last_index = (isize - 1) >> PAGE_SHIFT;
1466         }
1467
1468         if (newer_than) {
1469                 ret = find_new_extents(root, inode, newer_than,
1470                                        &newer_off, SZ_64K);
1471                 if (!ret) {
1472                         range->start = newer_off;
1473                         /*
1474                          * we always align our defrag to help keep
1475                          * the extents in the file evenly spaced
1476                          */
1477                         i = (newer_off & new_align) >> PAGE_SHIFT;
1478                 } else
1479                         goto out_ra;
1480         } else {
1481                 i = range->start >> PAGE_SHIFT;
1482         }
1483         if (!max_to_defrag)
1484                 max_to_defrag = last_index - i + 1;
1485
1486         /*
1487          * make writeback starts from i, so the defrag range can be
1488          * written sequentially.
1489          */
1490         if (i < inode->i_mapping->writeback_index)
1491                 inode->i_mapping->writeback_index = i;
1492
1493         while (i <= last_index && defrag_count < max_to_defrag &&
1494                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1495                 /*
1496                  * make sure we stop running if someone unmounts
1497                  * the FS
1498                  */
1499                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1500                         break;
1501
1502                 if (btrfs_defrag_cancelled(fs_info)) {
1503                         btrfs_debug(fs_info, "defrag_file cancelled");
1504                         ret = -EAGAIN;
1505                         goto error;
1506                 }
1507
1508                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1509                                          extent_thresh, &last_len, &skip,
1510                                          &defrag_end, do_compress)){
1511                         unsigned long next;
1512                         /*
1513                          * the should_defrag function tells us how much to skip
1514                          * bump our counter by the suggested amount
1515                          */
1516                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1517                         i = max(i + 1, next);
1518                         continue;
1519                 }
1520
1521                 if (!newer_than) {
1522                         cluster = (PAGE_ALIGN(defrag_end) >>
1523                                    PAGE_SHIFT) - i;
1524                         cluster = min(cluster, max_cluster);
1525                 } else {
1526                         cluster = max_cluster;
1527                 }
1528
1529                 if (i + cluster > ra_index) {
1530                         ra_index = max(i, ra_index);
1531                         if (ra)
1532                                 page_cache_sync_readahead(inode->i_mapping, ra,
1533                                                 file, ra_index, cluster);
1534                         ra_index += cluster;
1535                 }
1536
1537                 btrfs_inode_lock(inode, 0);
1538                 if (IS_SWAPFILE(inode)) {
1539                         ret = -ETXTBSY;
1540                 } else {
1541                         if (do_compress)
1542                                 BTRFS_I(inode)->defrag_compress = compress_type;
1543                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1544                 }
1545                 if (ret < 0) {
1546                         btrfs_inode_unlock(inode, 0);
1547                         goto out_ra;
1548                 }
1549
1550                 defrag_count += ret;
1551                 balance_dirty_pages_ratelimited(inode->i_mapping);
1552                 btrfs_inode_unlock(inode, 0);
1553
1554                 if (newer_than) {
1555                         if (newer_off == (u64)-1)
1556                                 break;
1557
1558                         if (ret > 0)
1559                                 i += ret;
1560
1561                         newer_off = max(newer_off + 1,
1562                                         (u64)i << PAGE_SHIFT);
1563
1564                         ret = find_new_extents(root, inode, newer_than,
1565                                                &newer_off, SZ_64K);
1566                         if (!ret) {
1567                                 range->start = newer_off;
1568                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1569                         } else {
1570                                 break;
1571                         }
1572                 } else {
1573                         if (ret > 0) {
1574                                 i += ret;
1575                                 last_len += ret << PAGE_SHIFT;
1576                         } else {
1577                                 i++;
1578                                 last_len = 0;
1579                         }
1580                 }
1581         }
1582
1583         ret = defrag_count;
1584 error:
1585         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1586                 filemap_flush(inode->i_mapping);
1587                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1588                              &BTRFS_I(inode)->runtime_flags))
1589                         filemap_flush(inode->i_mapping);
1590         }
1591
1592         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1593                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1594         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1595                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1596         }
1597
1598 out_ra:
1599         if (do_compress) {
1600                 btrfs_inode_lock(inode, 0);
1601                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1602                 btrfs_inode_unlock(inode, 0);
1603         }
1604         if (!file)
1605                 kfree(ra);
1606         kfree(pages);
1607         return ret;
1608 }
1609
1610 /*
1611  * Try to start exclusive operation @type or cancel it if it's running.
1612  *
1613  * Return:
1614  *   0        - normal mode, newly claimed op started
1615  *  >0        - normal mode, something else is running,
1616  *              return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space
1617  * ECANCELED  - cancel mode, successful cancel
1618  * ENOTCONN   - cancel mode, operation not running anymore
1619  */
1620 static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info,
1621                         enum btrfs_exclusive_operation type, bool cancel)
1622 {
1623         if (!cancel) {
1624                 /* Start normal op */
1625                 if (!btrfs_exclop_start(fs_info, type))
1626                         return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1627                 /* Exclusive operation is now claimed */
1628                 return 0;
1629         }
1630
1631         /* Cancel running op */
1632         if (btrfs_exclop_start_try_lock(fs_info, type)) {
1633                 /*
1634                  * This blocks any exclop finish from setting it to NONE, so we
1635                  * request cancellation. Either it runs and we will wait for it,
1636                  * or it has finished and no waiting will happen.
1637                  */
1638                 atomic_inc(&fs_info->reloc_cancel_req);
1639                 btrfs_exclop_start_unlock(fs_info);
1640
1641                 if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags))
1642                         wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING,
1643                                     TASK_INTERRUPTIBLE);
1644
1645                 return -ECANCELED;
1646         }
1647
1648         /* Something else is running or none */
1649         return -ENOTCONN;
1650 }
1651
1652 static noinline int btrfs_ioctl_resize(struct file *file,
1653                                         void __user *arg)
1654 {
1655         struct inode *inode = file_inode(file);
1656         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1657         u64 new_size;
1658         u64 old_size;
1659         u64 devid = 1;
1660         struct btrfs_root *root = BTRFS_I(inode)->root;
1661         struct btrfs_ioctl_vol_args *vol_args;
1662         struct btrfs_trans_handle *trans;
1663         struct btrfs_device *device = NULL;
1664         char *sizestr;
1665         char *retptr;
1666         char *devstr = NULL;
1667         int ret = 0;
1668         int mod = 0;
1669         bool cancel;
1670
1671         if (!capable(CAP_SYS_ADMIN))
1672                 return -EPERM;
1673
1674         ret = mnt_want_write_file(file);
1675         if (ret)
1676                 return ret;
1677
1678         /*
1679          * Read the arguments before checking exclusivity to be able to
1680          * distinguish regular resize and cancel
1681          */
1682         vol_args = memdup_user(arg, sizeof(*vol_args));
1683         if (IS_ERR(vol_args)) {
1684                 ret = PTR_ERR(vol_args);
1685                 goto out_drop;
1686         }
1687         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1688         sizestr = vol_args->name;
1689         cancel = (strcmp("cancel", sizestr) == 0);
1690         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel);
1691         if (ret)
1692                 goto out_free;
1693         /* Exclusive operation is now claimed */
1694
1695         devstr = strchr(sizestr, ':');
1696         if (devstr) {
1697                 sizestr = devstr + 1;
1698                 *devstr = '\0';
1699                 devstr = vol_args->name;
1700                 ret = kstrtoull(devstr, 10, &devid);
1701                 if (ret)
1702                         goto out_finish;
1703                 if (!devid) {
1704                         ret = -EINVAL;
1705                         goto out_finish;
1706                 }
1707                 btrfs_info(fs_info, "resizing devid %llu", devid);
1708         }
1709
1710         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
1711         if (!device) {
1712                 btrfs_info(fs_info, "resizer unable to find device %llu",
1713                            devid);
1714                 ret = -ENODEV;
1715                 goto out_finish;
1716         }
1717
1718         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1719                 btrfs_info(fs_info,
1720                            "resizer unable to apply on readonly device %llu",
1721                        devid);
1722                 ret = -EPERM;
1723                 goto out_finish;
1724         }
1725
1726         if (!strcmp(sizestr, "max"))
1727                 new_size = device->bdev->bd_inode->i_size;
1728         else {
1729                 if (sizestr[0] == '-') {
1730                         mod = -1;
1731                         sizestr++;
1732                 } else if (sizestr[0] == '+') {
1733                         mod = 1;
1734                         sizestr++;
1735                 }
1736                 new_size = memparse(sizestr, &retptr);
1737                 if (*retptr != '\0' || new_size == 0) {
1738                         ret = -EINVAL;
1739                         goto out_finish;
1740                 }
1741         }
1742
1743         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1744                 ret = -EPERM;
1745                 goto out_finish;
1746         }
1747
1748         old_size = btrfs_device_get_total_bytes(device);
1749
1750         if (mod < 0) {
1751                 if (new_size > old_size) {
1752                         ret = -EINVAL;
1753                         goto out_finish;
1754                 }
1755                 new_size = old_size - new_size;
1756         } else if (mod > 0) {
1757                 if (new_size > ULLONG_MAX - old_size) {
1758                         ret = -ERANGE;
1759                         goto out_finish;
1760                 }
1761                 new_size = old_size + new_size;
1762         }
1763
1764         if (new_size < SZ_256M) {
1765                 ret = -EINVAL;
1766                 goto out_finish;
1767         }
1768         if (new_size > device->bdev->bd_inode->i_size) {
1769                 ret = -EFBIG;
1770                 goto out_finish;
1771         }
1772
1773         new_size = round_down(new_size, fs_info->sectorsize);
1774
1775         if (new_size > old_size) {
1776                 trans = btrfs_start_transaction(root, 0);
1777                 if (IS_ERR(trans)) {
1778                         ret = PTR_ERR(trans);
1779                         goto out_finish;
1780                 }
1781                 ret = btrfs_grow_device(trans, device, new_size);
1782                 btrfs_commit_transaction(trans);
1783         } else if (new_size < old_size) {
1784                 ret = btrfs_shrink_device(device, new_size);
1785         } /* equal, nothing need to do */
1786
1787         if (ret == 0 && new_size != old_size)
1788                 btrfs_info_in_rcu(fs_info,
1789                         "resize device %s (devid %llu) from %llu to %llu",
1790                         rcu_str_deref(device->name), device->devid,
1791                         old_size, new_size);
1792 out_finish:
1793         btrfs_exclop_finish(fs_info);
1794 out_free:
1795         kfree(vol_args);
1796 out_drop:
1797         mnt_drop_write_file(file);
1798         return ret;
1799 }
1800
1801 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1802                                 const char *name, unsigned long fd, int subvol,
1803                                 bool readonly,
1804                                 struct btrfs_qgroup_inherit *inherit)
1805 {
1806         int namelen;
1807         int ret = 0;
1808
1809         if (!S_ISDIR(file_inode(file)->i_mode))
1810                 return -ENOTDIR;
1811
1812         ret = mnt_want_write_file(file);
1813         if (ret)
1814                 goto out;
1815
1816         namelen = strlen(name);
1817         if (strchr(name, '/')) {
1818                 ret = -EINVAL;
1819                 goto out_drop_write;
1820         }
1821
1822         if (name[0] == '.' &&
1823            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1824                 ret = -EEXIST;
1825                 goto out_drop_write;
1826         }
1827
1828         if (subvol) {
1829                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1830                                      NULL, readonly, inherit);
1831         } else {
1832                 struct fd src = fdget(fd);
1833                 struct inode *src_inode;
1834                 if (!src.file) {
1835                         ret = -EINVAL;
1836                         goto out_drop_write;
1837                 }
1838
1839                 src_inode = file_inode(src.file);
1840                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1841                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1842                                    "Snapshot src from another FS");
1843                         ret = -EXDEV;
1844                 } else if (!inode_owner_or_capable(&init_user_ns, src_inode)) {
1845                         /*
1846                          * Subvolume creation is not restricted, but snapshots
1847                          * are limited to own subvolumes only
1848                          */
1849                         ret = -EPERM;
1850                 } else {
1851                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1852                                              BTRFS_I(src_inode)->root,
1853                                              readonly, inherit);
1854                 }
1855                 fdput(src);
1856         }
1857 out_drop_write:
1858         mnt_drop_write_file(file);
1859 out:
1860         return ret;
1861 }
1862
1863 static noinline int btrfs_ioctl_snap_create(struct file *file,
1864                                             void __user *arg, int subvol)
1865 {
1866         struct btrfs_ioctl_vol_args *vol_args;
1867         int ret;
1868
1869         if (!S_ISDIR(file_inode(file)->i_mode))
1870                 return -ENOTDIR;
1871
1872         vol_args = memdup_user(arg, sizeof(*vol_args));
1873         if (IS_ERR(vol_args))
1874                 return PTR_ERR(vol_args);
1875         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1876
1877         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1878                                         subvol, false, NULL);
1879
1880         kfree(vol_args);
1881         return ret;
1882 }
1883
1884 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1885                                                void __user *arg, int subvol)
1886 {
1887         struct btrfs_ioctl_vol_args_v2 *vol_args;
1888         int ret;
1889         bool readonly = false;
1890         struct btrfs_qgroup_inherit *inherit = NULL;
1891
1892         if (!S_ISDIR(file_inode(file)->i_mode))
1893                 return -ENOTDIR;
1894
1895         vol_args = memdup_user(arg, sizeof(*vol_args));
1896         if (IS_ERR(vol_args))
1897                 return PTR_ERR(vol_args);
1898         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1899
1900         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1901                 ret = -EOPNOTSUPP;
1902                 goto free_args;
1903         }
1904
1905         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1906                 readonly = true;
1907         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1908                 u64 nums;
1909
1910                 if (vol_args->size < sizeof(*inherit) ||
1911                     vol_args->size > PAGE_SIZE) {
1912                         ret = -EINVAL;
1913                         goto free_args;
1914                 }
1915                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1916                 if (IS_ERR(inherit)) {
1917                         ret = PTR_ERR(inherit);
1918                         goto free_args;
1919                 }
1920
1921                 if (inherit->num_qgroups > PAGE_SIZE ||
1922                     inherit->num_ref_copies > PAGE_SIZE ||
1923                     inherit->num_excl_copies > PAGE_SIZE) {
1924                         ret = -EINVAL;
1925                         goto free_inherit;
1926                 }
1927
1928                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1929                        2 * inherit->num_excl_copies;
1930                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1931                         ret = -EINVAL;
1932                         goto free_inherit;
1933                 }
1934         }
1935
1936         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1937                                         subvol, readonly, inherit);
1938         if (ret)
1939                 goto free_inherit;
1940 free_inherit:
1941         kfree(inherit);
1942 free_args:
1943         kfree(vol_args);
1944         return ret;
1945 }
1946
1947 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1948                                                 void __user *arg)
1949 {
1950         struct inode *inode = file_inode(file);
1951         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1952         struct btrfs_root *root = BTRFS_I(inode)->root;
1953         int ret = 0;
1954         u64 flags = 0;
1955
1956         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1957                 return -EINVAL;
1958
1959         down_read(&fs_info->subvol_sem);
1960         if (btrfs_root_readonly(root))
1961                 flags |= BTRFS_SUBVOL_RDONLY;
1962         up_read(&fs_info->subvol_sem);
1963
1964         if (copy_to_user(arg, &flags, sizeof(flags)))
1965                 ret = -EFAULT;
1966
1967         return ret;
1968 }
1969
1970 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1971                                               void __user *arg)
1972 {
1973         struct inode *inode = file_inode(file);
1974         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1975         struct btrfs_root *root = BTRFS_I(inode)->root;
1976         struct btrfs_trans_handle *trans;
1977         u64 root_flags;
1978         u64 flags;
1979         int ret = 0;
1980
1981         if (!inode_owner_or_capable(&init_user_ns, inode))
1982                 return -EPERM;
1983
1984         ret = mnt_want_write_file(file);
1985         if (ret)
1986                 goto out;
1987
1988         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1989                 ret = -EINVAL;
1990                 goto out_drop_write;
1991         }
1992
1993         if (copy_from_user(&flags, arg, sizeof(flags))) {
1994                 ret = -EFAULT;
1995                 goto out_drop_write;
1996         }
1997
1998         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1999                 ret = -EOPNOTSUPP;
2000                 goto out_drop_write;
2001         }
2002
2003         down_write(&fs_info->subvol_sem);
2004
2005         /* nothing to do */
2006         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2007                 goto out_drop_sem;
2008
2009         root_flags = btrfs_root_flags(&root->root_item);
2010         if (flags & BTRFS_SUBVOL_RDONLY) {
2011                 btrfs_set_root_flags(&root->root_item,
2012                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2013         } else {
2014                 /*
2015                  * Block RO -> RW transition if this subvolume is involved in
2016                  * send
2017                  */
2018                 spin_lock(&root->root_item_lock);
2019                 if (root->send_in_progress == 0) {
2020                         btrfs_set_root_flags(&root->root_item,
2021                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2022                         spin_unlock(&root->root_item_lock);
2023                 } else {
2024                         spin_unlock(&root->root_item_lock);
2025                         btrfs_warn(fs_info,
2026                                    "Attempt to set subvolume %llu read-write during send",
2027                                    root->root_key.objectid);
2028                         ret = -EPERM;
2029                         goto out_drop_sem;
2030                 }
2031         }
2032
2033         trans = btrfs_start_transaction(root, 1);
2034         if (IS_ERR(trans)) {
2035                 ret = PTR_ERR(trans);
2036                 goto out_reset;
2037         }
2038
2039         ret = btrfs_update_root(trans, fs_info->tree_root,
2040                                 &root->root_key, &root->root_item);
2041         if (ret < 0) {
2042                 btrfs_end_transaction(trans);
2043                 goto out_reset;
2044         }
2045
2046         ret = btrfs_commit_transaction(trans);
2047
2048 out_reset:
2049         if (ret)
2050                 btrfs_set_root_flags(&root->root_item, root_flags);
2051 out_drop_sem:
2052         up_write(&fs_info->subvol_sem);
2053 out_drop_write:
2054         mnt_drop_write_file(file);
2055 out:
2056         return ret;
2057 }
2058
2059 static noinline int key_in_sk(struct btrfs_key *key,
2060                               struct btrfs_ioctl_search_key *sk)
2061 {
2062         struct btrfs_key test;
2063         int ret;
2064
2065         test.objectid = sk->min_objectid;
2066         test.type = sk->min_type;
2067         test.offset = sk->min_offset;
2068
2069         ret = btrfs_comp_cpu_keys(key, &test);
2070         if (ret < 0)
2071                 return 0;
2072
2073         test.objectid = sk->max_objectid;
2074         test.type = sk->max_type;
2075         test.offset = sk->max_offset;
2076
2077         ret = btrfs_comp_cpu_keys(key, &test);
2078         if (ret > 0)
2079                 return 0;
2080         return 1;
2081 }
2082
2083 static noinline int copy_to_sk(struct btrfs_path *path,
2084                                struct btrfs_key *key,
2085                                struct btrfs_ioctl_search_key *sk,
2086                                size_t *buf_size,
2087                                char __user *ubuf,
2088                                unsigned long *sk_offset,
2089                                int *num_found)
2090 {
2091         u64 found_transid;
2092         struct extent_buffer *leaf;
2093         struct btrfs_ioctl_search_header sh;
2094         struct btrfs_key test;
2095         unsigned long item_off;
2096         unsigned long item_len;
2097         int nritems;
2098         int i;
2099         int slot;
2100         int ret = 0;
2101
2102         leaf = path->nodes[0];
2103         slot = path->slots[0];
2104         nritems = btrfs_header_nritems(leaf);
2105
2106         if (btrfs_header_generation(leaf) > sk->max_transid) {
2107                 i = nritems;
2108                 goto advance_key;
2109         }
2110         found_transid = btrfs_header_generation(leaf);
2111
2112         for (i = slot; i < nritems; i++) {
2113                 item_off = btrfs_item_ptr_offset(leaf, i);
2114                 item_len = btrfs_item_size_nr(leaf, i);
2115
2116                 btrfs_item_key_to_cpu(leaf, key, i);
2117                 if (!key_in_sk(key, sk))
2118                         continue;
2119
2120                 if (sizeof(sh) + item_len > *buf_size) {
2121                         if (*num_found) {
2122                                 ret = 1;
2123                                 goto out;
2124                         }
2125
2126                         /*
2127                          * return one empty item back for v1, which does not
2128                          * handle -EOVERFLOW
2129                          */
2130
2131                         *buf_size = sizeof(sh) + item_len;
2132                         item_len = 0;
2133                         ret = -EOVERFLOW;
2134                 }
2135
2136                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2137                         ret = 1;
2138                         goto out;
2139                 }
2140
2141                 sh.objectid = key->objectid;
2142                 sh.offset = key->offset;
2143                 sh.type = key->type;
2144                 sh.len = item_len;
2145                 sh.transid = found_transid;
2146
2147                 /*
2148                  * Copy search result header. If we fault then loop again so we
2149                  * can fault in the pages and -EFAULT there if there's a
2150                  * problem. Otherwise we'll fault and then copy the buffer in
2151                  * properly this next time through
2152                  */
2153                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2154                         ret = 0;
2155                         goto out;
2156                 }
2157
2158                 *sk_offset += sizeof(sh);
2159
2160                 if (item_len) {
2161                         char __user *up = ubuf + *sk_offset;
2162                         /*
2163                          * Copy the item, same behavior as above, but reset the
2164                          * * sk_offset so we copy the full thing again.
2165                          */
2166                         if (read_extent_buffer_to_user_nofault(leaf, up,
2167                                                 item_off, item_len)) {
2168                                 ret = 0;
2169                                 *sk_offset -= sizeof(sh);
2170                                 goto out;
2171                         }
2172
2173                         *sk_offset += item_len;
2174                 }
2175                 (*num_found)++;
2176
2177                 if (ret) /* -EOVERFLOW from above */
2178                         goto out;
2179
2180                 if (*num_found >= sk->nr_items) {
2181                         ret = 1;
2182                         goto out;
2183                 }
2184         }
2185 advance_key:
2186         ret = 0;
2187         test.objectid = sk->max_objectid;
2188         test.type = sk->max_type;
2189         test.offset = sk->max_offset;
2190         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2191                 ret = 1;
2192         else if (key->offset < (u64)-1)
2193                 key->offset++;
2194         else if (key->type < (u8)-1) {
2195                 key->offset = 0;
2196                 key->type++;
2197         } else if (key->objectid < (u64)-1) {
2198                 key->offset = 0;
2199                 key->type = 0;
2200                 key->objectid++;
2201         } else
2202                 ret = 1;
2203 out:
2204         /*
2205          *  0: all items from this leaf copied, continue with next
2206          *  1: * more items can be copied, but unused buffer is too small
2207          *     * all items were found
2208          *     Either way, it will stops the loop which iterates to the next
2209          *     leaf
2210          *  -EOVERFLOW: item was to large for buffer
2211          *  -EFAULT: could not copy extent buffer back to userspace
2212          */
2213         return ret;
2214 }
2215
2216 static noinline int search_ioctl(struct inode *inode,
2217                                  struct btrfs_ioctl_search_key *sk,
2218                                  size_t *buf_size,
2219                                  char __user *ubuf)
2220 {
2221         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2222         struct btrfs_root *root;
2223         struct btrfs_key key;
2224         struct btrfs_path *path;
2225         int ret;
2226         int num_found = 0;
2227         unsigned long sk_offset = 0;
2228
2229         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2230                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2231                 return -EOVERFLOW;
2232         }
2233
2234         path = btrfs_alloc_path();
2235         if (!path)
2236                 return -ENOMEM;
2237
2238         if (sk->tree_id == 0) {
2239                 /* search the root of the inode that was passed */
2240                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2241         } else {
2242                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2243                 if (IS_ERR(root)) {
2244                         btrfs_free_path(path);
2245                         return PTR_ERR(root);
2246                 }
2247         }
2248
2249         key.objectid = sk->min_objectid;
2250         key.type = sk->min_type;
2251         key.offset = sk->min_offset;
2252
2253         while (1) {
2254                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2255                                                *buf_size - sk_offset);
2256                 if (ret)
2257                         break;
2258
2259                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2260                 if (ret != 0) {
2261                         if (ret > 0)
2262                                 ret = 0;
2263                         goto err;
2264                 }
2265                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2266                                  &sk_offset, &num_found);
2267                 btrfs_release_path(path);
2268                 if (ret)
2269                         break;
2270
2271         }
2272         if (ret > 0)
2273                 ret = 0;
2274 err:
2275         sk->nr_items = num_found;
2276         btrfs_put_root(root);
2277         btrfs_free_path(path);
2278         return ret;
2279 }
2280
2281 static noinline int btrfs_ioctl_tree_search(struct file *file,
2282                                            void __user *argp)
2283 {
2284         struct btrfs_ioctl_search_args __user *uargs;
2285         struct btrfs_ioctl_search_key sk;
2286         struct inode *inode;
2287         int ret;
2288         size_t buf_size;
2289
2290         if (!capable(CAP_SYS_ADMIN))
2291                 return -EPERM;
2292
2293         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2294
2295         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2296                 return -EFAULT;
2297
2298         buf_size = sizeof(uargs->buf);
2299
2300         inode = file_inode(file);
2301         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2302
2303         /*
2304          * In the origin implementation an overflow is handled by returning a
2305          * search header with a len of zero, so reset ret.
2306          */
2307         if (ret == -EOVERFLOW)
2308                 ret = 0;
2309
2310         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2311                 ret = -EFAULT;
2312         return ret;
2313 }
2314
2315 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2316                                                void __user *argp)
2317 {
2318         struct btrfs_ioctl_search_args_v2 __user *uarg;
2319         struct btrfs_ioctl_search_args_v2 args;
2320         struct inode *inode;
2321         int ret;
2322         size_t buf_size;
2323         const size_t buf_limit = SZ_16M;
2324
2325         if (!capable(CAP_SYS_ADMIN))
2326                 return -EPERM;
2327
2328         /* copy search header and buffer size */
2329         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2330         if (copy_from_user(&args, uarg, sizeof(args)))
2331                 return -EFAULT;
2332
2333         buf_size = args.buf_size;
2334
2335         /* limit result size to 16MB */
2336         if (buf_size > buf_limit)
2337                 buf_size = buf_limit;
2338
2339         inode = file_inode(file);
2340         ret = search_ioctl(inode, &args.key, &buf_size,
2341                            (char __user *)(&uarg->buf[0]));
2342         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2343                 ret = -EFAULT;
2344         else if (ret == -EOVERFLOW &&
2345                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2346                 ret = -EFAULT;
2347
2348         return ret;
2349 }
2350
2351 /*
2352  * Search INODE_REFs to identify path name of 'dirid' directory
2353  * in a 'tree_id' tree. and sets path name to 'name'.
2354  */
2355 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2356                                 u64 tree_id, u64 dirid, char *name)
2357 {
2358         struct btrfs_root *root;
2359         struct btrfs_key key;
2360         char *ptr;
2361         int ret = -1;
2362         int slot;
2363         int len;
2364         int total_len = 0;
2365         struct btrfs_inode_ref *iref;
2366         struct extent_buffer *l;
2367         struct btrfs_path *path;
2368
2369         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2370                 name[0]='\0';
2371                 return 0;
2372         }
2373
2374         path = btrfs_alloc_path();
2375         if (!path)
2376                 return -ENOMEM;
2377
2378         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2379
2380         root = btrfs_get_fs_root(info, tree_id, true);
2381         if (IS_ERR(root)) {
2382                 ret = PTR_ERR(root);
2383                 root = NULL;
2384                 goto out;
2385         }
2386
2387         key.objectid = dirid;
2388         key.type = BTRFS_INODE_REF_KEY;
2389         key.offset = (u64)-1;
2390
2391         while (1) {
2392                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2393                 if (ret < 0)
2394                         goto out;
2395                 else if (ret > 0) {
2396                         ret = btrfs_previous_item(root, path, dirid,
2397                                                   BTRFS_INODE_REF_KEY);
2398                         if (ret < 0)
2399                                 goto out;
2400                         else if (ret > 0) {
2401                                 ret = -ENOENT;
2402                                 goto out;
2403                         }
2404                 }
2405
2406                 l = path->nodes[0];
2407                 slot = path->slots[0];
2408                 btrfs_item_key_to_cpu(l, &key, slot);
2409
2410                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2411                 len = btrfs_inode_ref_name_len(l, iref);
2412                 ptr -= len + 1;
2413                 total_len += len + 1;
2414                 if (ptr < name) {
2415                         ret = -ENAMETOOLONG;
2416                         goto out;
2417                 }
2418
2419                 *(ptr + len) = '/';
2420                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2421
2422                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2423                         break;
2424
2425                 btrfs_release_path(path);
2426                 key.objectid = key.offset;
2427                 key.offset = (u64)-1;
2428                 dirid = key.objectid;
2429         }
2430         memmove(name, ptr, total_len);
2431         name[total_len] = '\0';
2432         ret = 0;
2433 out:
2434         btrfs_put_root(root);
2435         btrfs_free_path(path);
2436         return ret;
2437 }
2438
2439 static int btrfs_search_path_in_tree_user(struct inode *inode,
2440                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2441 {
2442         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2443         struct super_block *sb = inode->i_sb;
2444         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2445         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2446         u64 dirid = args->dirid;
2447         unsigned long item_off;
2448         unsigned long item_len;
2449         struct btrfs_inode_ref *iref;
2450         struct btrfs_root_ref *rref;
2451         struct btrfs_root *root = NULL;
2452         struct btrfs_path *path;
2453         struct btrfs_key key, key2;
2454         struct extent_buffer *leaf;
2455         struct inode *temp_inode;
2456         char *ptr;
2457         int slot;
2458         int len;
2459         int total_len = 0;
2460         int ret;
2461
2462         path = btrfs_alloc_path();
2463         if (!path)
2464                 return -ENOMEM;
2465
2466         /*
2467          * If the bottom subvolume does not exist directly under upper_limit,
2468          * construct the path in from the bottom up.
2469          */
2470         if (dirid != upper_limit.objectid) {
2471                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2472
2473                 root = btrfs_get_fs_root(fs_info, treeid, true);
2474                 if (IS_ERR(root)) {
2475                         ret = PTR_ERR(root);
2476                         goto out;
2477                 }
2478
2479                 key.objectid = dirid;
2480                 key.type = BTRFS_INODE_REF_KEY;
2481                 key.offset = (u64)-1;
2482                 while (1) {
2483                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2484                         if (ret < 0) {
2485                                 goto out_put;
2486                         } else if (ret > 0) {
2487                                 ret = btrfs_previous_item(root, path, dirid,
2488                                                           BTRFS_INODE_REF_KEY);
2489                                 if (ret < 0) {
2490                                         goto out_put;
2491                                 } else if (ret > 0) {
2492                                         ret = -ENOENT;
2493                                         goto out_put;
2494                                 }
2495                         }
2496
2497                         leaf = path->nodes[0];
2498                         slot = path->slots[0];
2499                         btrfs_item_key_to_cpu(leaf, &key, slot);
2500
2501                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2502                         len = btrfs_inode_ref_name_len(leaf, iref);
2503                         ptr -= len + 1;
2504                         total_len += len + 1;
2505                         if (ptr < args->path) {
2506                                 ret = -ENAMETOOLONG;
2507                                 goto out_put;
2508                         }
2509
2510                         *(ptr + len) = '/';
2511                         read_extent_buffer(leaf, ptr,
2512                                         (unsigned long)(iref + 1), len);
2513
2514                         /* Check the read+exec permission of this directory */
2515                         ret = btrfs_previous_item(root, path, dirid,
2516                                                   BTRFS_INODE_ITEM_KEY);
2517                         if (ret < 0) {
2518                                 goto out_put;
2519                         } else if (ret > 0) {
2520                                 ret = -ENOENT;
2521                                 goto out_put;
2522                         }
2523
2524                         leaf = path->nodes[0];
2525                         slot = path->slots[0];
2526                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2527                         if (key2.objectid != dirid) {
2528                                 ret = -ENOENT;
2529                                 goto out_put;
2530                         }
2531
2532                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2533                         if (IS_ERR(temp_inode)) {
2534                                 ret = PTR_ERR(temp_inode);
2535                                 goto out_put;
2536                         }
2537                         ret = inode_permission(&init_user_ns, temp_inode,
2538                                                MAY_READ | MAY_EXEC);
2539                         iput(temp_inode);
2540                         if (ret) {
2541                                 ret = -EACCES;
2542                                 goto out_put;
2543                         }
2544
2545                         if (key.offset == upper_limit.objectid)
2546                                 break;
2547                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2548                                 ret = -EACCES;
2549                                 goto out_put;
2550                         }
2551
2552                         btrfs_release_path(path);
2553                         key.objectid = key.offset;
2554                         key.offset = (u64)-1;
2555                         dirid = key.objectid;
2556                 }
2557
2558                 memmove(args->path, ptr, total_len);
2559                 args->path[total_len] = '\0';
2560                 btrfs_put_root(root);
2561                 root = NULL;
2562                 btrfs_release_path(path);
2563         }
2564
2565         /* Get the bottom subvolume's name from ROOT_REF */
2566         key.objectid = treeid;
2567         key.type = BTRFS_ROOT_REF_KEY;
2568         key.offset = args->treeid;
2569         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2570         if (ret < 0) {
2571                 goto out;
2572         } else if (ret > 0) {
2573                 ret = -ENOENT;
2574                 goto out;
2575         }
2576
2577         leaf = path->nodes[0];
2578         slot = path->slots[0];
2579         btrfs_item_key_to_cpu(leaf, &key, slot);
2580
2581         item_off = btrfs_item_ptr_offset(leaf, slot);
2582         item_len = btrfs_item_size_nr(leaf, slot);
2583         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2584         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2585         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2586                 ret = -EINVAL;
2587                 goto out;
2588         }
2589
2590         /* Copy subvolume's name */
2591         item_off += sizeof(struct btrfs_root_ref);
2592         item_len -= sizeof(struct btrfs_root_ref);
2593         read_extent_buffer(leaf, args->name, item_off, item_len);
2594         args->name[item_len] = 0;
2595
2596 out_put:
2597         btrfs_put_root(root);
2598 out:
2599         btrfs_free_path(path);
2600         return ret;
2601 }
2602
2603 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2604                                            void __user *argp)
2605 {
2606         struct btrfs_ioctl_ino_lookup_args *args;
2607         struct inode *inode;
2608         int ret = 0;
2609
2610         args = memdup_user(argp, sizeof(*args));
2611         if (IS_ERR(args))
2612                 return PTR_ERR(args);
2613
2614         inode = file_inode(file);
2615
2616         /*
2617          * Unprivileged query to obtain the containing subvolume root id. The
2618          * path is reset so it's consistent with btrfs_search_path_in_tree.
2619          */
2620         if (args->treeid == 0)
2621                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2622
2623         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2624                 args->name[0] = 0;
2625                 goto out;
2626         }
2627
2628         if (!capable(CAP_SYS_ADMIN)) {
2629                 ret = -EPERM;
2630                 goto out;
2631         }
2632
2633         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2634                                         args->treeid, args->objectid,
2635                                         args->name);
2636
2637 out:
2638         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2639                 ret = -EFAULT;
2640
2641         kfree(args);
2642         return ret;
2643 }
2644
2645 /*
2646  * Version of ino_lookup ioctl (unprivileged)
2647  *
2648  * The main differences from ino_lookup ioctl are:
2649  *
2650  *   1. Read + Exec permission will be checked using inode_permission() during
2651  *      path construction. -EACCES will be returned in case of failure.
2652  *   2. Path construction will be stopped at the inode number which corresponds
2653  *      to the fd with which this ioctl is called. If constructed path does not
2654  *      exist under fd's inode, -EACCES will be returned.
2655  *   3. The name of bottom subvolume is also searched and filled.
2656  */
2657 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2658 {
2659         struct btrfs_ioctl_ino_lookup_user_args *args;
2660         struct inode *inode;
2661         int ret;
2662
2663         args = memdup_user(argp, sizeof(*args));
2664         if (IS_ERR(args))
2665                 return PTR_ERR(args);
2666
2667         inode = file_inode(file);
2668
2669         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2670             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2671                 /*
2672                  * The subvolume does not exist under fd with which this is
2673                  * called
2674                  */
2675                 kfree(args);
2676                 return -EACCES;
2677         }
2678
2679         ret = btrfs_search_path_in_tree_user(inode, args);
2680
2681         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2682                 ret = -EFAULT;
2683
2684         kfree(args);
2685         return ret;
2686 }
2687
2688 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2689 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2690 {
2691         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2692         struct btrfs_fs_info *fs_info;
2693         struct btrfs_root *root;
2694         struct btrfs_path *path;
2695         struct btrfs_key key;
2696         struct btrfs_root_item *root_item;
2697         struct btrfs_root_ref *rref;
2698         struct extent_buffer *leaf;
2699         unsigned long item_off;
2700         unsigned long item_len;
2701         struct inode *inode;
2702         int slot;
2703         int ret = 0;
2704
2705         path = btrfs_alloc_path();
2706         if (!path)
2707                 return -ENOMEM;
2708
2709         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2710         if (!subvol_info) {
2711                 btrfs_free_path(path);
2712                 return -ENOMEM;
2713         }
2714
2715         inode = file_inode(file);
2716         fs_info = BTRFS_I(inode)->root->fs_info;
2717
2718         /* Get root_item of inode's subvolume */
2719         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2720         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2721         if (IS_ERR(root)) {
2722                 ret = PTR_ERR(root);
2723                 goto out_free;
2724         }
2725         root_item = &root->root_item;
2726
2727         subvol_info->treeid = key.objectid;
2728
2729         subvol_info->generation = btrfs_root_generation(root_item);
2730         subvol_info->flags = btrfs_root_flags(root_item);
2731
2732         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2733         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2734                                                     BTRFS_UUID_SIZE);
2735         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2736                                                     BTRFS_UUID_SIZE);
2737
2738         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2739         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2740         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2741
2742         subvol_info->otransid = btrfs_root_otransid(root_item);
2743         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2744         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2745
2746         subvol_info->stransid = btrfs_root_stransid(root_item);
2747         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2748         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2749
2750         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2751         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2752         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2753
2754         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2755                 /* Search root tree for ROOT_BACKREF of this subvolume */
2756                 key.type = BTRFS_ROOT_BACKREF_KEY;
2757                 key.offset = 0;
2758                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2759                 if (ret < 0) {
2760                         goto out;
2761                 } else if (path->slots[0] >=
2762                            btrfs_header_nritems(path->nodes[0])) {
2763                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2764                         if (ret < 0) {
2765                                 goto out;
2766                         } else if (ret > 0) {
2767                                 ret = -EUCLEAN;
2768                                 goto out;
2769                         }
2770                 }
2771
2772                 leaf = path->nodes[0];
2773                 slot = path->slots[0];
2774                 btrfs_item_key_to_cpu(leaf, &key, slot);
2775                 if (key.objectid == subvol_info->treeid &&
2776                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2777                         subvol_info->parent_id = key.offset;
2778
2779                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2780                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2781
2782                         item_off = btrfs_item_ptr_offset(leaf, slot)
2783                                         + sizeof(struct btrfs_root_ref);
2784                         item_len = btrfs_item_size_nr(leaf, slot)
2785                                         - sizeof(struct btrfs_root_ref);
2786                         read_extent_buffer(leaf, subvol_info->name,
2787                                            item_off, item_len);
2788                 } else {
2789                         ret = -ENOENT;
2790                         goto out;
2791                 }
2792         }
2793
2794         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2795                 ret = -EFAULT;
2796
2797 out:
2798         btrfs_put_root(root);
2799 out_free:
2800         btrfs_free_path(path);
2801         kfree(subvol_info);
2802         return ret;
2803 }
2804
2805 /*
2806  * Return ROOT_REF information of the subvolume containing this inode
2807  * except the subvolume name.
2808  */
2809 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2810 {
2811         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2812         struct btrfs_root_ref *rref;
2813         struct btrfs_root *root;
2814         struct btrfs_path *path;
2815         struct btrfs_key key;
2816         struct extent_buffer *leaf;
2817         struct inode *inode;
2818         u64 objectid;
2819         int slot;
2820         int ret;
2821         u8 found;
2822
2823         path = btrfs_alloc_path();
2824         if (!path)
2825                 return -ENOMEM;
2826
2827         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2828         if (IS_ERR(rootrefs)) {
2829                 btrfs_free_path(path);
2830                 return PTR_ERR(rootrefs);
2831         }
2832
2833         inode = file_inode(file);
2834         root = BTRFS_I(inode)->root->fs_info->tree_root;
2835         objectid = BTRFS_I(inode)->root->root_key.objectid;
2836
2837         key.objectid = objectid;
2838         key.type = BTRFS_ROOT_REF_KEY;
2839         key.offset = rootrefs->min_treeid;
2840         found = 0;
2841
2842         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2843         if (ret < 0) {
2844                 goto out;
2845         } else if (path->slots[0] >=
2846                    btrfs_header_nritems(path->nodes[0])) {
2847                 ret = btrfs_next_leaf(root, path);
2848                 if (ret < 0) {
2849                         goto out;
2850                 } else if (ret > 0) {
2851                         ret = -EUCLEAN;
2852                         goto out;
2853                 }
2854         }
2855         while (1) {
2856                 leaf = path->nodes[0];
2857                 slot = path->slots[0];
2858
2859                 btrfs_item_key_to_cpu(leaf, &key, slot);
2860                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2861                         ret = 0;
2862                         goto out;
2863                 }
2864
2865                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2866                         ret = -EOVERFLOW;
2867                         goto out;
2868                 }
2869
2870                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2871                 rootrefs->rootref[found].treeid = key.offset;
2872                 rootrefs->rootref[found].dirid =
2873                                   btrfs_root_ref_dirid(leaf, rref);
2874                 found++;
2875
2876                 ret = btrfs_next_item(root, path);
2877                 if (ret < 0) {
2878                         goto out;
2879                 } else if (ret > 0) {
2880                         ret = -EUCLEAN;
2881                         goto out;
2882                 }
2883         }
2884
2885 out:
2886         if (!ret || ret == -EOVERFLOW) {
2887                 rootrefs->num_items = found;
2888                 /* update min_treeid for next search */
2889                 if (found)
2890                         rootrefs->min_treeid =
2891                                 rootrefs->rootref[found - 1].treeid + 1;
2892                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2893                         ret = -EFAULT;
2894         }
2895
2896         kfree(rootrefs);
2897         btrfs_free_path(path);
2898
2899         return ret;
2900 }
2901
2902 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2903                                              void __user *arg,
2904                                              bool destroy_v2)
2905 {
2906         struct dentry *parent = file->f_path.dentry;
2907         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2908         struct dentry *dentry;
2909         struct inode *dir = d_inode(parent);
2910         struct inode *inode;
2911         struct btrfs_root *root = BTRFS_I(dir)->root;
2912         struct btrfs_root *dest = NULL;
2913         struct btrfs_ioctl_vol_args *vol_args = NULL;
2914         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2915         char *subvol_name, *subvol_name_ptr = NULL;
2916         int subvol_namelen;
2917         int err = 0;
2918         bool destroy_parent = false;
2919
2920         if (destroy_v2) {
2921                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2922                 if (IS_ERR(vol_args2))
2923                         return PTR_ERR(vol_args2);
2924
2925                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2926                         err = -EOPNOTSUPP;
2927                         goto out;
2928                 }
2929
2930                 /*
2931                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2932                  * name, same as v1 currently does.
2933                  */
2934                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2935                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2936                         subvol_name = vol_args2->name;
2937
2938                         err = mnt_want_write_file(file);
2939                         if (err)
2940                                 goto out;
2941                 } else {
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                         dir = d_inode(parent);
2979
2980                         /*
2981                          * If v2 was used with SPEC_BY_ID, a new parent was
2982                          * allocated since the subvolume can be outside of the
2983                          * current mount point. Later on we need to release this
2984                          * new parent dentry.
2985                          */
2986                         destroy_parent = true;
2987
2988                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2989                                                 fs_info, vol_args2->subvolid);
2990                         if (IS_ERR(subvol_name_ptr)) {
2991                                 err = PTR_ERR(subvol_name_ptr);
2992                                 goto free_parent;
2993                         }
2994                         /* subvol_name_ptr is already nul terminated */
2995                         subvol_name = (char *)kbasename(subvol_name_ptr);
2996                 }
2997         } else {
2998                 vol_args = memdup_user(arg, sizeof(*vol_args));
2999                 if (IS_ERR(vol_args))
3000                         return PTR_ERR(vol_args);
3001
3002                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3003                 subvol_name = vol_args->name;
3004
3005                 err = mnt_want_write_file(file);
3006                 if (err)
3007                         goto out;
3008         }
3009
3010         subvol_namelen = strlen(subvol_name);
3011
3012         if (strchr(subvol_name, '/') ||
3013             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3014                 err = -EINVAL;
3015                 goto free_subvol_name;
3016         }
3017
3018         if (!S_ISDIR(dir->i_mode)) {
3019                 err = -ENOTDIR;
3020                 goto free_subvol_name;
3021         }
3022
3023         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3024         if (err == -EINTR)
3025                 goto free_subvol_name;
3026         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
3027         if (IS_ERR(dentry)) {
3028                 err = PTR_ERR(dentry);
3029                 goto out_unlock_dir;
3030         }
3031
3032         if (d_really_is_negative(dentry)) {
3033                 err = -ENOENT;
3034                 goto out_dput;
3035         }
3036
3037         inode = d_inode(dentry);
3038         dest = BTRFS_I(inode)->root;
3039         if (!capable(CAP_SYS_ADMIN)) {
3040                 /*
3041                  * Regular user.  Only allow this with a special mount
3042                  * option, when the user has write+exec access to the
3043                  * subvol root, and when rmdir(2) would have been
3044                  * allowed.
3045                  *
3046                  * Note that this is _not_ check that the subvol is
3047                  * empty or doesn't contain data that we wouldn't
3048                  * otherwise be able to delete.
3049                  *
3050                  * Users who want to delete empty subvols should try
3051                  * rmdir(2).
3052                  */
3053                 err = -EPERM;
3054                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3055                         goto out_dput;
3056
3057                 /*
3058                  * Do not allow deletion if the parent dir is the same
3059                  * as the dir to be deleted.  That means the ioctl
3060                  * must be called on the dentry referencing the root
3061                  * of the subvol, not a random directory contained
3062                  * within it.
3063                  */
3064                 err = -EINVAL;
3065                 if (root == dest)
3066                         goto out_dput;
3067
3068                 err = inode_permission(&init_user_ns, inode,
3069                                        MAY_WRITE | MAY_EXEC);
3070                 if (err)
3071                         goto out_dput;
3072         }
3073
3074         /* check if subvolume may be deleted by a user */
3075         err = btrfs_may_delete(dir, dentry, 1);
3076         if (err)
3077                 goto out_dput;
3078
3079         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3080                 err = -EINVAL;
3081                 goto out_dput;
3082         }
3083
3084         btrfs_inode_lock(inode, 0);
3085         err = btrfs_delete_subvolume(dir, dentry);
3086         btrfs_inode_unlock(inode, 0);
3087         if (!err) {
3088                 fsnotify_rmdir(dir, dentry);
3089                 d_delete(dentry);
3090         }
3091
3092 out_dput:
3093         dput(dentry);
3094 out_unlock_dir:
3095         btrfs_inode_unlock(dir, 0);
3096 free_subvol_name:
3097         kfree(subvol_name_ptr);
3098 free_parent:
3099         if (destroy_parent)
3100                 dput(parent);
3101 out_drop_write:
3102         mnt_drop_write_file(file);
3103 out:
3104         kfree(vol_args2);
3105         kfree(vol_args);
3106         return err;
3107 }
3108
3109 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3110 {
3111         struct inode *inode = file_inode(file);
3112         struct btrfs_root *root = BTRFS_I(inode)->root;
3113         struct btrfs_ioctl_defrag_range_args *range;
3114         int ret;
3115
3116         ret = mnt_want_write_file(file);
3117         if (ret)
3118                 return ret;
3119
3120         if (btrfs_root_readonly(root)) {
3121                 ret = -EROFS;
3122                 goto out;
3123         }
3124
3125         /* Subpage defrag will be supported in later commits */
3126         if (root->fs_info->sectorsize < PAGE_SIZE) {
3127                 ret = -ENOTTY;
3128                 goto out;
3129         }
3130
3131         switch (inode->i_mode & S_IFMT) {
3132         case S_IFDIR:
3133                 if (!capable(CAP_SYS_ADMIN)) {
3134                         ret = -EPERM;
3135                         goto out;
3136                 }
3137                 ret = btrfs_defrag_root(root);
3138                 break;
3139         case S_IFREG:
3140                 /*
3141                  * Note that this does not check the file descriptor for write
3142                  * access. This prevents defragmenting executables that are
3143                  * running and allows defrag on files open in read-only mode.
3144                  */
3145                 if (!capable(CAP_SYS_ADMIN) &&
3146                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3147                         ret = -EPERM;
3148                         goto out;
3149                 }
3150
3151                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3152                 if (!range) {
3153                         ret = -ENOMEM;
3154                         goto out;
3155                 }
3156
3157                 if (argp) {
3158                         if (copy_from_user(range, argp,
3159                                            sizeof(*range))) {
3160                                 ret = -EFAULT;
3161                                 kfree(range);
3162                                 goto out;
3163                         }
3164                         /* compression requires us to start the IO */
3165                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3166                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3167                                 range->extent_thresh = (u32)-1;
3168                         }
3169                 } else {
3170                         /* the rest are all set to zero by kzalloc */
3171                         range->len = (u64)-1;
3172                 }
3173                 ret = btrfs_defrag_file(file_inode(file), file,
3174                                         range, BTRFS_OLDEST_GENERATION, 0);
3175                 if (ret > 0)
3176                         ret = 0;
3177                 kfree(range);
3178                 break;
3179         default:
3180                 ret = -EINVAL;
3181         }
3182 out:
3183         mnt_drop_write_file(file);
3184         return ret;
3185 }
3186
3187 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3188 {
3189         struct btrfs_ioctl_vol_args *vol_args;
3190         int ret;
3191
3192         if (!capable(CAP_SYS_ADMIN))
3193                 return -EPERM;
3194
3195         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3196                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3197
3198         vol_args = memdup_user(arg, sizeof(*vol_args));
3199         if (IS_ERR(vol_args)) {
3200                 ret = PTR_ERR(vol_args);
3201                 goto out;
3202         }
3203
3204         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3205         ret = btrfs_init_new_device(fs_info, vol_args->name);
3206
3207         if (!ret)
3208                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3209
3210         kfree(vol_args);
3211 out:
3212         btrfs_exclop_finish(fs_info);
3213         return ret;
3214 }
3215
3216 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3217 {
3218         struct inode *inode = file_inode(file);
3219         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3220         struct btrfs_ioctl_vol_args_v2 *vol_args;
3221         int ret;
3222         bool cancel = false;
3223
3224         if (!capable(CAP_SYS_ADMIN))
3225                 return -EPERM;
3226
3227         ret = mnt_want_write_file(file);
3228         if (ret)
3229                 return ret;
3230
3231         vol_args = memdup_user(arg, sizeof(*vol_args));
3232         if (IS_ERR(vol_args)) {
3233                 ret = PTR_ERR(vol_args);
3234                 goto err_drop;
3235         }
3236
3237         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3238                 ret = -EOPNOTSUPP;
3239                 goto out;
3240         }
3241         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3242         if (!(vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) &&
3243             strcmp("cancel", vol_args->name) == 0)
3244                 cancel = true;
3245
3246         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3247                                            cancel);
3248         if (ret)
3249                 goto out;
3250         /* Exclusive operation is now claimed */
3251
3252         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3253                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3254         else
3255                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3256
3257         btrfs_exclop_finish(fs_info);
3258
3259         if (!ret) {
3260                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3261                         btrfs_info(fs_info, "device deleted: id %llu",
3262                                         vol_args->devid);
3263                 else
3264                         btrfs_info(fs_info, "device deleted: %s",
3265                                         vol_args->name);
3266         }
3267 out:
3268         kfree(vol_args);
3269 err_drop:
3270         mnt_drop_write_file(file);
3271         return ret;
3272 }
3273
3274 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3275 {
3276         struct inode *inode = file_inode(file);
3277         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3278         struct btrfs_ioctl_vol_args *vol_args;
3279         int ret;
3280         bool cancel;
3281
3282         if (!capable(CAP_SYS_ADMIN))
3283                 return -EPERM;
3284
3285         ret = mnt_want_write_file(file);
3286         if (ret)
3287                 return ret;
3288
3289         vol_args = memdup_user(arg, sizeof(*vol_args));
3290         if (IS_ERR(vol_args)) {
3291                 ret = PTR_ERR(vol_args);
3292                 goto out_drop_write;
3293         }
3294         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3295         cancel = (strcmp("cancel", vol_args->name) == 0);
3296
3297         ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE,
3298                                            cancel);
3299         if (ret == 0) {
3300                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3301                 if (!ret)
3302                         btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3303                 btrfs_exclop_finish(fs_info);
3304         }
3305
3306         kfree(vol_args);
3307 out_drop_write:
3308         mnt_drop_write_file(file);
3309
3310         return ret;
3311 }
3312
3313 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3314                                 void __user *arg)
3315 {
3316         struct btrfs_ioctl_fs_info_args *fi_args;
3317         struct btrfs_device *device;
3318         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3319         u64 flags_in;
3320         int ret = 0;
3321
3322         fi_args = memdup_user(arg, sizeof(*fi_args));
3323         if (IS_ERR(fi_args))
3324                 return PTR_ERR(fi_args);
3325
3326         flags_in = fi_args->flags;
3327         memset(fi_args, 0, sizeof(*fi_args));
3328
3329         rcu_read_lock();
3330         fi_args->num_devices = fs_devices->num_devices;
3331
3332         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3333                 if (device->devid > fi_args->max_id)
3334                         fi_args->max_id = device->devid;
3335         }
3336         rcu_read_unlock();
3337
3338         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3339         fi_args->nodesize = fs_info->nodesize;
3340         fi_args->sectorsize = fs_info->sectorsize;
3341         fi_args->clone_alignment = fs_info->sectorsize;
3342
3343         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3344                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3345                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3346                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3347         }
3348
3349         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3350                 fi_args->generation = fs_info->generation;
3351                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3352         }
3353
3354         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3355                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3356                        sizeof(fi_args->metadata_uuid));
3357                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3358         }
3359
3360         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3361                 ret = -EFAULT;
3362
3363         kfree(fi_args);
3364         return ret;
3365 }
3366
3367 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3368                                  void __user *arg)
3369 {
3370         struct btrfs_ioctl_dev_info_args *di_args;
3371         struct btrfs_device *dev;
3372         int ret = 0;
3373         char *s_uuid = NULL;
3374
3375         di_args = memdup_user(arg, sizeof(*di_args));
3376         if (IS_ERR(di_args))
3377                 return PTR_ERR(di_args);
3378
3379         if (!btrfs_is_empty_uuid(di_args->uuid))
3380                 s_uuid = di_args->uuid;
3381
3382         rcu_read_lock();
3383         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3384                                 NULL);
3385
3386         if (!dev) {
3387                 ret = -ENODEV;
3388                 goto out;
3389         }
3390
3391         di_args->devid = dev->devid;
3392         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3393         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3394         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3395         if (dev->name) {
3396                 strncpy(di_args->path, rcu_str_deref(dev->name),
3397                                 sizeof(di_args->path) - 1);
3398                 di_args->path[sizeof(di_args->path) - 1] = 0;
3399         } else {
3400                 di_args->path[0] = '\0';
3401         }
3402
3403 out:
3404         rcu_read_unlock();
3405         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3406                 ret = -EFAULT;
3407
3408         kfree(di_args);
3409         return ret;
3410 }
3411
3412 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3413 {
3414         struct inode *inode = file_inode(file);
3415         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3416         struct btrfs_root *root = BTRFS_I(inode)->root;
3417         struct btrfs_root *new_root;
3418         struct btrfs_dir_item *di;
3419         struct btrfs_trans_handle *trans;
3420         struct btrfs_path *path = NULL;
3421         struct btrfs_disk_key disk_key;
3422         u64 objectid = 0;
3423         u64 dir_id;
3424         int ret;
3425
3426         if (!capable(CAP_SYS_ADMIN))
3427                 return -EPERM;
3428
3429         ret = mnt_want_write_file(file);
3430         if (ret)
3431                 return ret;
3432
3433         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3434                 ret = -EFAULT;
3435                 goto out;
3436         }
3437
3438         if (!objectid)
3439                 objectid = BTRFS_FS_TREE_OBJECTID;
3440
3441         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3442         if (IS_ERR(new_root)) {
3443                 ret = PTR_ERR(new_root);
3444                 goto out;
3445         }
3446         if (!is_fstree(new_root->root_key.objectid)) {
3447                 ret = -ENOENT;
3448                 goto out_free;
3449         }
3450
3451         path = btrfs_alloc_path();
3452         if (!path) {
3453                 ret = -ENOMEM;
3454                 goto out_free;
3455         }
3456
3457         trans = btrfs_start_transaction(root, 1);
3458         if (IS_ERR(trans)) {
3459                 ret = PTR_ERR(trans);
3460                 goto out_free;
3461         }
3462
3463         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3464         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3465                                    dir_id, "default", 7, 1);
3466         if (IS_ERR_OR_NULL(di)) {
3467                 btrfs_release_path(path);
3468                 btrfs_end_transaction(trans);
3469                 btrfs_err(fs_info,
3470                           "Umm, you don't have the default diritem, this isn't going to work");
3471                 ret = -ENOENT;
3472                 goto out_free;
3473         }
3474
3475         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3476         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3477         btrfs_mark_buffer_dirty(path->nodes[0]);
3478         btrfs_release_path(path);
3479
3480         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3481         btrfs_end_transaction(trans);
3482 out_free:
3483         btrfs_put_root(new_root);
3484         btrfs_free_path(path);
3485 out:
3486         mnt_drop_write_file(file);
3487         return ret;
3488 }
3489
3490 static void get_block_group_info(struct list_head *groups_list,
3491                                  struct btrfs_ioctl_space_info *space)
3492 {
3493         struct btrfs_block_group *block_group;
3494
3495         space->total_bytes = 0;
3496         space->used_bytes = 0;
3497         space->flags = 0;
3498         list_for_each_entry(block_group, groups_list, list) {
3499                 space->flags = block_group->flags;
3500                 space->total_bytes += block_group->length;
3501                 space->used_bytes += block_group->used;
3502         }
3503 }
3504
3505 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3506                                    void __user *arg)
3507 {
3508         struct btrfs_ioctl_space_args space_args;
3509         struct btrfs_ioctl_space_info space;
3510         struct btrfs_ioctl_space_info *dest;
3511         struct btrfs_ioctl_space_info *dest_orig;
3512         struct btrfs_ioctl_space_info __user *user_dest;
3513         struct btrfs_space_info *info;
3514         static const u64 types[] = {
3515                 BTRFS_BLOCK_GROUP_DATA,
3516                 BTRFS_BLOCK_GROUP_SYSTEM,
3517                 BTRFS_BLOCK_GROUP_METADATA,
3518                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3519         };
3520         int num_types = 4;
3521         int alloc_size;
3522         int ret = 0;
3523         u64 slot_count = 0;
3524         int i, c;
3525
3526         if (copy_from_user(&space_args,
3527                            (struct btrfs_ioctl_space_args __user *)arg,
3528                            sizeof(space_args)))
3529                 return -EFAULT;
3530
3531         for (i = 0; i < num_types; i++) {
3532                 struct btrfs_space_info *tmp;
3533
3534                 info = NULL;
3535                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3536                         if (tmp->flags == types[i]) {
3537                                 info = tmp;
3538                                 break;
3539                         }
3540                 }
3541
3542                 if (!info)
3543                         continue;
3544
3545                 down_read(&info->groups_sem);
3546                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3547                         if (!list_empty(&info->block_groups[c]))
3548                                 slot_count++;
3549                 }
3550                 up_read(&info->groups_sem);
3551         }
3552
3553         /*
3554          * Global block reserve, exported as a space_info
3555          */
3556         slot_count++;
3557
3558         /* space_slots == 0 means they are asking for a count */
3559         if (space_args.space_slots == 0) {
3560                 space_args.total_spaces = slot_count;
3561                 goto out;
3562         }
3563
3564         slot_count = min_t(u64, space_args.space_slots, slot_count);
3565
3566         alloc_size = sizeof(*dest) * slot_count;
3567
3568         /* we generally have at most 6 or so space infos, one for each raid
3569          * level.  So, a whole page should be more than enough for everyone
3570          */
3571         if (alloc_size > PAGE_SIZE)
3572                 return -ENOMEM;
3573
3574         space_args.total_spaces = 0;
3575         dest = kmalloc(alloc_size, GFP_KERNEL);
3576         if (!dest)
3577                 return -ENOMEM;
3578         dest_orig = dest;
3579
3580         /* now we have a buffer to copy into */
3581         for (i = 0; i < num_types; i++) {
3582                 struct btrfs_space_info *tmp;
3583
3584                 if (!slot_count)
3585                         break;
3586
3587                 info = NULL;
3588                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3589                         if (tmp->flags == types[i]) {
3590                                 info = tmp;
3591                                 break;
3592                         }
3593                 }
3594
3595                 if (!info)
3596                         continue;
3597                 down_read(&info->groups_sem);
3598                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3599                         if (!list_empty(&info->block_groups[c])) {
3600                                 get_block_group_info(&info->block_groups[c],
3601                                                      &space);
3602                                 memcpy(dest, &space, sizeof(space));
3603                                 dest++;
3604                                 space_args.total_spaces++;
3605                                 slot_count--;
3606                         }
3607                         if (!slot_count)
3608                                 break;
3609                 }
3610                 up_read(&info->groups_sem);
3611         }
3612
3613         /*
3614          * Add global block reserve
3615          */
3616         if (slot_count) {
3617                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3618
3619                 spin_lock(&block_rsv->lock);
3620                 space.total_bytes = block_rsv->size;
3621                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3622                 spin_unlock(&block_rsv->lock);
3623                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3624                 memcpy(dest, &space, sizeof(space));
3625                 space_args.total_spaces++;
3626         }
3627
3628         user_dest = (struct btrfs_ioctl_space_info __user *)
3629                 (arg + sizeof(struct btrfs_ioctl_space_args));
3630
3631         if (copy_to_user(user_dest, dest_orig, alloc_size))
3632                 ret = -EFAULT;
3633
3634         kfree(dest_orig);
3635 out:
3636         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3637                 ret = -EFAULT;
3638
3639         return ret;
3640 }
3641
3642 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3643                                             void __user *argp)
3644 {
3645         struct btrfs_trans_handle *trans;
3646         u64 transid;
3647         int ret;
3648
3649         trans = btrfs_attach_transaction_barrier(root);
3650         if (IS_ERR(trans)) {
3651                 if (PTR_ERR(trans) != -ENOENT)
3652                         return PTR_ERR(trans);
3653
3654                 /* No running transaction, don't bother */
3655                 transid = root->fs_info->last_trans_committed;
3656                 goto out;
3657         }
3658         transid = trans->transid;
3659         ret = btrfs_commit_transaction_async(trans);
3660         if (ret) {
3661                 btrfs_end_transaction(trans);
3662                 return ret;
3663         }
3664 out:
3665         if (argp)
3666                 if (copy_to_user(argp, &transid, sizeof(transid)))
3667                         return -EFAULT;
3668         return 0;
3669 }
3670
3671 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3672                                            void __user *argp)
3673 {
3674         u64 transid;
3675
3676         if (argp) {
3677                 if (copy_from_user(&transid, argp, sizeof(transid)))
3678                         return -EFAULT;
3679         } else {
3680                 transid = 0;  /* current trans */
3681         }
3682         return btrfs_wait_for_commit(fs_info, transid);
3683 }
3684
3685 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3686 {
3687         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3688         struct btrfs_ioctl_scrub_args *sa;
3689         int ret;
3690
3691         if (!capable(CAP_SYS_ADMIN))
3692                 return -EPERM;
3693
3694         sa = memdup_user(arg, sizeof(*sa));
3695         if (IS_ERR(sa))
3696                 return PTR_ERR(sa);
3697
3698         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3699                 ret = mnt_want_write_file(file);
3700                 if (ret)
3701                         goto out;
3702         }
3703
3704         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3705                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3706                               0);
3707
3708         /*
3709          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3710          * error. This is important as it allows user space to know how much
3711          * progress scrub has done. For example, if scrub is canceled we get
3712          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3713          * space. Later user space can inspect the progress from the structure
3714          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3715          * previously (btrfs-progs does this).
3716          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3717          * then return -EFAULT to signal the structure was not copied or it may
3718          * be corrupt and unreliable due to a partial copy.
3719          */
3720         if (copy_to_user(arg, sa, sizeof(*sa)))
3721                 ret = -EFAULT;
3722
3723         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3724                 mnt_drop_write_file(file);
3725 out:
3726         kfree(sa);
3727         return ret;
3728 }
3729
3730 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3731 {
3732         if (!capable(CAP_SYS_ADMIN))
3733                 return -EPERM;
3734
3735         return btrfs_scrub_cancel(fs_info);
3736 }
3737
3738 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3739                                        void __user *arg)
3740 {
3741         struct btrfs_ioctl_scrub_args *sa;
3742         int ret;
3743
3744         if (!capable(CAP_SYS_ADMIN))
3745                 return -EPERM;
3746
3747         sa = memdup_user(arg, sizeof(*sa));
3748         if (IS_ERR(sa))
3749                 return PTR_ERR(sa);
3750
3751         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3752
3753         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3754                 ret = -EFAULT;
3755
3756         kfree(sa);
3757         return ret;
3758 }
3759
3760 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3761                                       void __user *arg)
3762 {
3763         struct btrfs_ioctl_get_dev_stats *sa;
3764         int ret;
3765
3766         sa = memdup_user(arg, sizeof(*sa));
3767         if (IS_ERR(sa))
3768                 return PTR_ERR(sa);
3769
3770         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3771                 kfree(sa);
3772                 return -EPERM;
3773         }
3774
3775         ret = btrfs_get_dev_stats(fs_info, sa);
3776
3777         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3778                 ret = -EFAULT;
3779
3780         kfree(sa);
3781         return ret;
3782 }
3783
3784 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3785                                     void __user *arg)
3786 {
3787         struct btrfs_ioctl_dev_replace_args *p;
3788         int ret;
3789
3790         if (!capable(CAP_SYS_ADMIN))
3791                 return -EPERM;
3792
3793         p = memdup_user(arg, sizeof(*p));
3794         if (IS_ERR(p))
3795                 return PTR_ERR(p);
3796
3797         switch (p->cmd) {
3798         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3799                 if (sb_rdonly(fs_info->sb)) {
3800                         ret = -EROFS;
3801                         goto out;
3802                 }
3803                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3804                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3805                 } else {
3806                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3807                         btrfs_exclop_finish(fs_info);
3808                 }
3809                 break;
3810         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3811                 btrfs_dev_replace_status(fs_info, p);
3812                 ret = 0;
3813                 break;
3814         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3815                 p->result = btrfs_dev_replace_cancel(fs_info);
3816                 ret = 0;
3817                 break;
3818         default:
3819                 ret = -EINVAL;
3820                 break;
3821         }
3822
3823         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3824                 ret = -EFAULT;
3825 out:
3826         kfree(p);
3827         return ret;
3828 }
3829
3830 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3831 {
3832         int ret = 0;
3833         int i;
3834         u64 rel_ptr;
3835         int size;
3836         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3837         struct inode_fs_paths *ipath = NULL;
3838         struct btrfs_path *path;
3839
3840         if (!capable(CAP_DAC_READ_SEARCH))
3841                 return -EPERM;
3842
3843         path = btrfs_alloc_path();
3844         if (!path) {
3845                 ret = -ENOMEM;
3846                 goto out;
3847         }
3848
3849         ipa = memdup_user(arg, sizeof(*ipa));
3850         if (IS_ERR(ipa)) {
3851                 ret = PTR_ERR(ipa);
3852                 ipa = NULL;
3853                 goto out;
3854         }
3855
3856         size = min_t(u32, ipa->size, 4096);
3857         ipath = init_ipath(size, root, path);
3858         if (IS_ERR(ipath)) {
3859                 ret = PTR_ERR(ipath);
3860                 ipath = NULL;
3861                 goto out;
3862         }
3863
3864         ret = paths_from_inode(ipa->inum, ipath);
3865         if (ret < 0)
3866                 goto out;
3867
3868         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3869                 rel_ptr = ipath->fspath->val[i] -
3870                           (u64)(unsigned long)ipath->fspath->val;
3871                 ipath->fspath->val[i] = rel_ptr;
3872         }
3873
3874         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3875                            ipath->fspath, size);
3876         if (ret) {
3877                 ret = -EFAULT;
3878                 goto out;
3879         }
3880
3881 out:
3882         btrfs_free_path(path);
3883         free_ipath(ipath);
3884         kfree(ipa);
3885
3886         return ret;
3887 }
3888
3889 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3890 {
3891         struct btrfs_data_container *inodes = ctx;
3892         const size_t c = 3 * sizeof(u64);
3893
3894         if (inodes->bytes_left >= c) {
3895                 inodes->bytes_left -= c;
3896                 inodes->val[inodes->elem_cnt] = inum;
3897                 inodes->val[inodes->elem_cnt + 1] = offset;
3898                 inodes->val[inodes->elem_cnt + 2] = root;
3899                 inodes->elem_cnt += 3;
3900         } else {
3901                 inodes->bytes_missing += c - inodes->bytes_left;
3902                 inodes->bytes_left = 0;
3903                 inodes->elem_missed += 3;
3904         }
3905
3906         return 0;
3907 }
3908
3909 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3910                                         void __user *arg, int version)
3911 {
3912         int ret = 0;
3913         int size;
3914         struct btrfs_ioctl_logical_ino_args *loi;
3915         struct btrfs_data_container *inodes = NULL;
3916         struct btrfs_path *path = NULL;
3917         bool ignore_offset;
3918
3919         if (!capable(CAP_SYS_ADMIN))
3920                 return -EPERM;
3921
3922         loi = memdup_user(arg, sizeof(*loi));
3923         if (IS_ERR(loi))
3924                 return PTR_ERR(loi);
3925
3926         if (version == 1) {
3927                 ignore_offset = false;
3928                 size = min_t(u32, loi->size, SZ_64K);
3929         } else {
3930                 /* All reserved bits must be 0 for now */
3931                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3932                         ret = -EINVAL;
3933                         goto out_loi;
3934                 }
3935                 /* Only accept flags we have defined so far */
3936                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3937                         ret = -EINVAL;
3938                         goto out_loi;
3939                 }
3940                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3941                 size = min_t(u32, loi->size, SZ_16M);
3942         }
3943
3944         path = btrfs_alloc_path();
3945         if (!path) {
3946                 ret = -ENOMEM;
3947                 goto out;
3948         }
3949
3950         inodes = init_data_container(size);
3951         if (IS_ERR(inodes)) {
3952                 ret = PTR_ERR(inodes);
3953                 inodes = NULL;
3954                 goto out;
3955         }
3956
3957         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3958                                           build_ino_list, inodes, ignore_offset);
3959         if (ret == -EINVAL)
3960                 ret = -ENOENT;
3961         if (ret < 0)
3962                 goto out;
3963
3964         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3965                            size);
3966         if (ret)
3967                 ret = -EFAULT;
3968
3969 out:
3970         btrfs_free_path(path);
3971         kvfree(inodes);
3972 out_loi:
3973         kfree(loi);
3974
3975         return ret;
3976 }
3977
3978 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3979                                struct btrfs_ioctl_balance_args *bargs)
3980 {
3981         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3982
3983         bargs->flags = bctl->flags;
3984
3985         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3986                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3987         if (atomic_read(&fs_info->balance_pause_req))
3988                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3989         if (atomic_read(&fs_info->balance_cancel_req))
3990                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3991
3992         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3993         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3994         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3995
3996         spin_lock(&fs_info->balance_lock);
3997         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3998         spin_unlock(&fs_info->balance_lock);
3999 }
4000
4001 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4002 {
4003         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4004         struct btrfs_fs_info *fs_info = root->fs_info;
4005         struct btrfs_ioctl_balance_args *bargs;
4006         struct btrfs_balance_control *bctl;
4007         bool need_unlock; /* for mut. excl. ops lock */
4008         int ret;
4009
4010         if (!capable(CAP_SYS_ADMIN))
4011                 return -EPERM;
4012
4013         ret = mnt_want_write_file(file);
4014         if (ret)
4015                 return ret;
4016
4017 again:
4018         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4019                 mutex_lock(&fs_info->balance_mutex);
4020                 need_unlock = true;
4021                 goto locked;
4022         }
4023
4024         /*
4025          * mut. excl. ops lock is locked.  Three possibilities:
4026          *   (1) some other op is running
4027          *   (2) balance is running
4028          *   (3) balance is paused -- special case (think resume)
4029          */
4030         mutex_lock(&fs_info->balance_mutex);
4031         if (fs_info->balance_ctl) {
4032                 /* this is either (2) or (3) */
4033                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4034                         mutex_unlock(&fs_info->balance_mutex);
4035                         /*
4036                          * Lock released to allow other waiters to continue,
4037                          * we'll reexamine the status again.
4038                          */
4039                         mutex_lock(&fs_info->balance_mutex);
4040
4041                         if (fs_info->balance_ctl &&
4042                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4043                                 /* this is (3) */
4044                                 need_unlock = false;
4045                                 goto locked;
4046                         }
4047
4048                         mutex_unlock(&fs_info->balance_mutex);
4049                         goto again;
4050                 } else {
4051                         /* this is (2) */
4052                         mutex_unlock(&fs_info->balance_mutex);
4053                         ret = -EINPROGRESS;
4054                         goto out;
4055                 }
4056         } else {
4057                 /* this is (1) */
4058                 mutex_unlock(&fs_info->balance_mutex);
4059                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4060                 goto out;
4061         }
4062
4063 locked:
4064
4065         if (arg) {
4066                 bargs = memdup_user(arg, sizeof(*bargs));
4067                 if (IS_ERR(bargs)) {
4068                         ret = PTR_ERR(bargs);
4069                         goto out_unlock;
4070                 }
4071
4072                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4073                         if (!fs_info->balance_ctl) {
4074                                 ret = -ENOTCONN;
4075                                 goto out_bargs;
4076                         }
4077
4078                         bctl = fs_info->balance_ctl;
4079                         spin_lock(&fs_info->balance_lock);
4080                         bctl->flags |= BTRFS_BALANCE_RESUME;
4081                         spin_unlock(&fs_info->balance_lock);
4082
4083                         goto do_balance;
4084                 }
4085         } else {
4086                 bargs = NULL;
4087         }
4088
4089         if (fs_info->balance_ctl) {
4090                 ret = -EINPROGRESS;
4091                 goto out_bargs;
4092         }
4093
4094         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4095         if (!bctl) {
4096                 ret = -ENOMEM;
4097                 goto out_bargs;
4098         }
4099
4100         if (arg) {
4101                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4102                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4103                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4104
4105                 bctl->flags = bargs->flags;
4106         } else {
4107                 /* balance everything - no filters */
4108                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4109         }
4110
4111         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4112                 ret = -EINVAL;
4113                 goto out_bctl;
4114         }
4115
4116 do_balance:
4117         /*
4118          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4119          * bctl is freed in reset_balance_state, or, if restriper was paused
4120          * all the way until unmount, in free_fs_info.  The flag should be
4121          * cleared after reset_balance_state.
4122          */
4123         need_unlock = false;
4124
4125         ret = btrfs_balance(fs_info, bctl, bargs);
4126         bctl = NULL;
4127
4128         if ((ret == 0 || ret == -ECANCELED) && arg) {
4129                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4130                         ret = -EFAULT;
4131         }
4132
4133 out_bctl:
4134         kfree(bctl);
4135 out_bargs:
4136         kfree(bargs);
4137 out_unlock:
4138         mutex_unlock(&fs_info->balance_mutex);
4139         if (need_unlock)
4140                 btrfs_exclop_finish(fs_info);
4141 out:
4142         mnt_drop_write_file(file);
4143         return ret;
4144 }
4145
4146 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4147 {
4148         if (!capable(CAP_SYS_ADMIN))
4149                 return -EPERM;
4150
4151         switch (cmd) {
4152         case BTRFS_BALANCE_CTL_PAUSE:
4153                 return btrfs_pause_balance(fs_info);
4154         case BTRFS_BALANCE_CTL_CANCEL:
4155                 return btrfs_cancel_balance(fs_info);
4156         }
4157
4158         return -EINVAL;
4159 }
4160
4161 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4162                                          void __user *arg)
4163 {
4164         struct btrfs_ioctl_balance_args *bargs;
4165         int ret = 0;
4166
4167         if (!capable(CAP_SYS_ADMIN))
4168                 return -EPERM;
4169
4170         mutex_lock(&fs_info->balance_mutex);
4171         if (!fs_info->balance_ctl) {
4172                 ret = -ENOTCONN;
4173                 goto out;
4174         }
4175
4176         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4177         if (!bargs) {
4178                 ret = -ENOMEM;
4179                 goto out;
4180         }
4181
4182         btrfs_update_ioctl_balance_args(fs_info, bargs);
4183
4184         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4185                 ret = -EFAULT;
4186
4187         kfree(bargs);
4188 out:
4189         mutex_unlock(&fs_info->balance_mutex);
4190         return ret;
4191 }
4192
4193 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4194 {
4195         struct inode *inode = file_inode(file);
4196         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4197         struct btrfs_ioctl_quota_ctl_args *sa;
4198         int ret;
4199
4200         if (!capable(CAP_SYS_ADMIN))
4201                 return -EPERM;
4202
4203         ret = mnt_want_write_file(file);
4204         if (ret)
4205                 return ret;
4206
4207         sa = memdup_user(arg, sizeof(*sa));
4208         if (IS_ERR(sa)) {
4209                 ret = PTR_ERR(sa);
4210                 goto drop_write;
4211         }
4212
4213         down_write(&fs_info->subvol_sem);
4214
4215         switch (sa->cmd) {
4216         case BTRFS_QUOTA_CTL_ENABLE:
4217                 ret = btrfs_quota_enable(fs_info);
4218                 break;
4219         case BTRFS_QUOTA_CTL_DISABLE:
4220                 ret = btrfs_quota_disable(fs_info);
4221                 break;
4222         default:
4223                 ret = -EINVAL;
4224                 break;
4225         }
4226
4227         kfree(sa);
4228         up_write(&fs_info->subvol_sem);
4229 drop_write:
4230         mnt_drop_write_file(file);
4231         return ret;
4232 }
4233
4234 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4235 {
4236         struct inode *inode = file_inode(file);
4237         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4238         struct btrfs_root *root = BTRFS_I(inode)->root;
4239         struct btrfs_ioctl_qgroup_assign_args *sa;
4240         struct btrfs_trans_handle *trans;
4241         int ret;
4242         int err;
4243
4244         if (!capable(CAP_SYS_ADMIN))
4245                 return -EPERM;
4246
4247         ret = mnt_want_write_file(file);
4248         if (ret)
4249                 return ret;
4250
4251         sa = memdup_user(arg, sizeof(*sa));
4252         if (IS_ERR(sa)) {
4253                 ret = PTR_ERR(sa);
4254                 goto drop_write;
4255         }
4256
4257         trans = btrfs_join_transaction(root);
4258         if (IS_ERR(trans)) {
4259                 ret = PTR_ERR(trans);
4260                 goto out;
4261         }
4262
4263         if (sa->assign) {
4264                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4265         } else {
4266                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4267         }
4268
4269         /* update qgroup status and info */
4270         err = btrfs_run_qgroups(trans);
4271         if (err < 0)
4272                 btrfs_handle_fs_error(fs_info, err,
4273                                       "failed to update qgroup status and info");
4274         err = btrfs_end_transaction(trans);
4275         if (err && !ret)
4276                 ret = err;
4277
4278 out:
4279         kfree(sa);
4280 drop_write:
4281         mnt_drop_write_file(file);
4282         return ret;
4283 }
4284
4285 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4286 {
4287         struct inode *inode = file_inode(file);
4288         struct btrfs_root *root = BTRFS_I(inode)->root;
4289         struct btrfs_ioctl_qgroup_create_args *sa;
4290         struct btrfs_trans_handle *trans;
4291         int ret;
4292         int err;
4293
4294         if (!capable(CAP_SYS_ADMIN))
4295                 return -EPERM;
4296
4297         ret = mnt_want_write_file(file);
4298         if (ret)
4299                 return ret;
4300
4301         sa = memdup_user(arg, sizeof(*sa));
4302         if (IS_ERR(sa)) {
4303                 ret = PTR_ERR(sa);
4304                 goto drop_write;
4305         }
4306
4307         if (!sa->qgroupid) {
4308                 ret = -EINVAL;
4309                 goto out;
4310         }
4311
4312         trans = btrfs_join_transaction(root);
4313         if (IS_ERR(trans)) {
4314                 ret = PTR_ERR(trans);
4315                 goto out;
4316         }
4317
4318         if (sa->create) {
4319                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4320         } else {
4321                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4322         }
4323
4324         err = btrfs_end_transaction(trans);
4325         if (err && !ret)
4326                 ret = err;
4327
4328 out:
4329         kfree(sa);
4330 drop_write:
4331         mnt_drop_write_file(file);
4332         return ret;
4333 }
4334
4335 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4336 {
4337         struct inode *inode = file_inode(file);
4338         struct btrfs_root *root = BTRFS_I(inode)->root;
4339         struct btrfs_ioctl_qgroup_limit_args *sa;
4340         struct btrfs_trans_handle *trans;
4341         int ret;
4342         int err;
4343         u64 qgroupid;
4344
4345         if (!capable(CAP_SYS_ADMIN))
4346                 return -EPERM;
4347
4348         ret = mnt_want_write_file(file);
4349         if (ret)
4350                 return ret;
4351
4352         sa = memdup_user(arg, sizeof(*sa));
4353         if (IS_ERR(sa)) {
4354                 ret = PTR_ERR(sa);
4355                 goto drop_write;
4356         }
4357
4358         trans = btrfs_join_transaction(root);
4359         if (IS_ERR(trans)) {
4360                 ret = PTR_ERR(trans);
4361                 goto out;
4362         }
4363
4364         qgroupid = sa->qgroupid;
4365         if (!qgroupid) {
4366                 /* take the current subvol as qgroup */
4367                 qgroupid = root->root_key.objectid;
4368         }
4369
4370         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4371
4372         err = btrfs_end_transaction(trans);
4373         if (err && !ret)
4374                 ret = err;
4375
4376 out:
4377         kfree(sa);
4378 drop_write:
4379         mnt_drop_write_file(file);
4380         return ret;
4381 }
4382
4383 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4384 {
4385         struct inode *inode = file_inode(file);
4386         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4387         struct btrfs_ioctl_quota_rescan_args *qsa;
4388         int ret;
4389
4390         if (!capable(CAP_SYS_ADMIN))
4391                 return -EPERM;
4392
4393         ret = mnt_want_write_file(file);
4394         if (ret)
4395                 return ret;
4396
4397         qsa = memdup_user(arg, sizeof(*qsa));
4398         if (IS_ERR(qsa)) {
4399                 ret = PTR_ERR(qsa);
4400                 goto drop_write;
4401         }
4402
4403         if (qsa->flags) {
4404                 ret = -EINVAL;
4405                 goto out;
4406         }
4407
4408         ret = btrfs_qgroup_rescan(fs_info);
4409
4410 out:
4411         kfree(qsa);
4412 drop_write:
4413         mnt_drop_write_file(file);
4414         return ret;
4415 }
4416
4417 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4418                                                 void __user *arg)
4419 {
4420         struct btrfs_ioctl_quota_rescan_args *qsa;
4421         int ret = 0;
4422
4423         if (!capable(CAP_SYS_ADMIN))
4424                 return -EPERM;
4425
4426         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4427         if (!qsa)
4428                 return -ENOMEM;
4429
4430         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4431                 qsa->flags = 1;
4432                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4433         }
4434
4435         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4436                 ret = -EFAULT;
4437
4438         kfree(qsa);
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 btrfs_ioctl_received_subvol_args *sa)
4453 {
4454         struct inode *inode = file_inode(file);
4455         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4456         struct btrfs_root *root = BTRFS_I(inode)->root;
4457         struct btrfs_root_item *root_item = &root->root_item;
4458         struct btrfs_trans_handle *trans;
4459         struct timespec64 ct = current_time(inode);
4460         int ret = 0;
4461         int received_uuid_changed;
4462
4463         if (!inode_owner_or_capable(&init_user_ns, inode))
4464                 return -EPERM;
4465
4466         ret = mnt_want_write_file(file);
4467         if (ret < 0)
4468                 return ret;
4469
4470         down_write(&fs_info->subvol_sem);
4471
4472         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4473                 ret = -EINVAL;
4474                 goto out;
4475         }
4476
4477         if (btrfs_root_readonly(root)) {
4478                 ret = -EROFS;
4479                 goto out;
4480         }
4481
4482         /*
4483          * 1 - root item
4484          * 2 - uuid items (received uuid + subvol uuid)
4485          */
4486         trans = btrfs_start_transaction(root, 3);
4487         if (IS_ERR(trans)) {
4488                 ret = PTR_ERR(trans);
4489                 trans = NULL;
4490                 goto out;
4491         }
4492
4493         sa->rtransid = trans->transid;
4494         sa->rtime.sec = ct.tv_sec;
4495         sa->rtime.nsec = ct.tv_nsec;
4496
4497         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4498                                        BTRFS_UUID_SIZE);
4499         if (received_uuid_changed &&
4500             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4501                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4502                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4503                                           root->root_key.objectid);
4504                 if (ret && ret != -ENOENT) {
4505                         btrfs_abort_transaction(trans, ret);
4506                         btrfs_end_transaction(trans);
4507                         goto out;
4508                 }
4509         }
4510         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4511         btrfs_set_root_stransid(root_item, sa->stransid);
4512         btrfs_set_root_rtransid(root_item, sa->rtransid);
4513         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4514         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4515         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4516         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4517
4518         ret = btrfs_update_root(trans, fs_info->tree_root,
4519                                 &root->root_key, &root->root_item);
4520         if (ret < 0) {
4521                 btrfs_end_transaction(trans);
4522                 goto out;
4523         }
4524         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4525                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4526                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4527                                           root->root_key.objectid);
4528                 if (ret < 0 && ret != -EEXIST) {
4529                         btrfs_abort_transaction(trans, ret);
4530                         btrfs_end_transaction(trans);
4531                         goto out;
4532                 }
4533         }
4534         ret = btrfs_commit_transaction(trans);
4535 out:
4536         up_write(&fs_info->subvol_sem);
4537         mnt_drop_write_file(file);
4538         return ret;
4539 }
4540
4541 #ifdef CONFIG_64BIT
4542 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4543                                                 void __user *arg)
4544 {
4545         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4546         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4547         int ret = 0;
4548
4549         args32 = memdup_user(arg, sizeof(*args32));
4550         if (IS_ERR(args32))
4551                 return PTR_ERR(args32);
4552
4553         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4554         if (!args64) {
4555                 ret = -ENOMEM;
4556                 goto out;
4557         }
4558
4559         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4560         args64->stransid = args32->stransid;
4561         args64->rtransid = args32->rtransid;
4562         args64->stime.sec = args32->stime.sec;
4563         args64->stime.nsec = args32->stime.nsec;
4564         args64->rtime.sec = args32->rtime.sec;
4565         args64->rtime.nsec = args32->rtime.nsec;
4566         args64->flags = args32->flags;
4567
4568         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4569         if (ret)
4570                 goto out;
4571
4572         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4573         args32->stransid = args64->stransid;
4574         args32->rtransid = args64->rtransid;
4575         args32->stime.sec = args64->stime.sec;
4576         args32->stime.nsec = args64->stime.nsec;
4577         args32->rtime.sec = args64->rtime.sec;
4578         args32->rtime.nsec = args64->rtime.nsec;
4579         args32->flags = args64->flags;
4580
4581         ret = copy_to_user(arg, args32, sizeof(*args32));
4582         if (ret)
4583                 ret = -EFAULT;
4584
4585 out:
4586         kfree(args32);
4587         kfree(args64);
4588         return ret;
4589 }
4590 #endif
4591
4592 static long btrfs_ioctl_set_received_subvol(struct file *file,
4593                                             void __user *arg)
4594 {
4595         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4596         int ret = 0;
4597
4598         sa = memdup_user(arg, sizeof(*sa));
4599         if (IS_ERR(sa))
4600                 return PTR_ERR(sa);
4601
4602         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4603
4604         if (ret)
4605                 goto out;
4606
4607         ret = copy_to_user(arg, sa, sizeof(*sa));
4608         if (ret)
4609                 ret = -EFAULT;
4610
4611 out:
4612         kfree(sa);
4613         return ret;
4614 }
4615
4616 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4617                                         void __user *arg)
4618 {
4619         size_t len;
4620         int ret;
4621         char label[BTRFS_LABEL_SIZE];
4622
4623         spin_lock(&fs_info->super_lock);
4624         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4625         spin_unlock(&fs_info->super_lock);
4626
4627         len = strnlen(label, BTRFS_LABEL_SIZE);
4628
4629         if (len == BTRFS_LABEL_SIZE) {
4630                 btrfs_warn(fs_info,
4631                            "label is too long, return the first %zu bytes",
4632                            --len);
4633         }
4634
4635         ret = copy_to_user(arg, label, len);
4636
4637         return ret ? -EFAULT : 0;
4638 }
4639
4640 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4641 {
4642         struct inode *inode = file_inode(file);
4643         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4644         struct btrfs_root *root = BTRFS_I(inode)->root;
4645         struct btrfs_super_block *super_block = fs_info->super_copy;
4646         struct btrfs_trans_handle *trans;
4647         char label[BTRFS_LABEL_SIZE];
4648         int ret;
4649
4650         if (!capable(CAP_SYS_ADMIN))
4651                 return -EPERM;
4652
4653         if (copy_from_user(label, arg, sizeof(label)))
4654                 return -EFAULT;
4655
4656         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4657                 btrfs_err(fs_info,
4658                           "unable to set label with more than %d bytes",
4659                           BTRFS_LABEL_SIZE - 1);
4660                 return -EINVAL;
4661         }
4662
4663         ret = mnt_want_write_file(file);
4664         if (ret)
4665                 return ret;
4666
4667         trans = btrfs_start_transaction(root, 0);
4668         if (IS_ERR(trans)) {
4669                 ret = PTR_ERR(trans);
4670                 goto out_unlock;
4671         }
4672
4673         spin_lock(&fs_info->super_lock);
4674         strcpy(super_block->label, label);
4675         spin_unlock(&fs_info->super_lock);
4676         ret = btrfs_commit_transaction(trans);
4677
4678 out_unlock:
4679         mnt_drop_write_file(file);
4680         return ret;
4681 }
4682
4683 #define INIT_FEATURE_FLAGS(suffix) \
4684         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4685           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4686           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4687
4688 int btrfs_ioctl_get_supported_features(void __user *arg)
4689 {
4690         static const struct btrfs_ioctl_feature_flags features[3] = {
4691                 INIT_FEATURE_FLAGS(SUPP),
4692                 INIT_FEATURE_FLAGS(SAFE_SET),
4693                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4694         };
4695
4696         if (copy_to_user(arg, &features, sizeof(features)))
4697                 return -EFAULT;
4698
4699         return 0;
4700 }
4701
4702 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4703                                         void __user *arg)
4704 {
4705         struct btrfs_super_block *super_block = fs_info->super_copy;
4706         struct btrfs_ioctl_feature_flags features;
4707
4708         features.compat_flags = btrfs_super_compat_flags(super_block);
4709         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4710         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4711
4712         if (copy_to_user(arg, &features, sizeof(features)))
4713                 return -EFAULT;
4714
4715         return 0;
4716 }
4717
4718 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4719                               enum btrfs_feature_set set,
4720                               u64 change_mask, u64 flags, u64 supported_flags,
4721                               u64 safe_set, u64 safe_clear)
4722 {
4723         const char *type = btrfs_feature_set_name(set);
4724         char *names;
4725         u64 disallowed, unsupported;
4726         u64 set_mask = flags & change_mask;
4727         u64 clear_mask = ~flags & change_mask;
4728
4729         unsupported = set_mask & ~supported_flags;
4730         if (unsupported) {
4731                 names = btrfs_printable_features(set, unsupported);
4732                 if (names) {
4733                         btrfs_warn(fs_info,
4734                                    "this kernel does not support the %s feature bit%s",
4735                                    names, strchr(names, ',') ? "s" : "");
4736                         kfree(names);
4737                 } else
4738                         btrfs_warn(fs_info,
4739                                    "this kernel does not support %s bits 0x%llx",
4740                                    type, unsupported);
4741                 return -EOPNOTSUPP;
4742         }
4743
4744         disallowed = set_mask & ~safe_set;
4745         if (disallowed) {
4746                 names = btrfs_printable_features(set, disallowed);
4747                 if (names) {
4748                         btrfs_warn(fs_info,
4749                                    "can't set the %s feature bit%s while mounted",
4750                                    names, strchr(names, ',') ? "s" : "");
4751                         kfree(names);
4752                 } else
4753                         btrfs_warn(fs_info,
4754                                    "can't set %s bits 0x%llx while mounted",
4755                                    type, disallowed);
4756                 return -EPERM;
4757         }
4758
4759         disallowed = clear_mask & ~safe_clear;
4760         if (disallowed) {
4761                 names = btrfs_printable_features(set, disallowed);
4762                 if (names) {
4763                         btrfs_warn(fs_info,
4764                                    "can't clear the %s feature bit%s while mounted",
4765                                    names, strchr(names, ',') ? "s" : "");
4766                         kfree(names);
4767                 } else
4768                         btrfs_warn(fs_info,
4769                                    "can't clear %s bits 0x%llx while mounted",
4770                                    type, disallowed);
4771                 return -EPERM;
4772         }
4773
4774         return 0;
4775 }
4776
4777 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4778 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4779                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4780                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4781                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4782
4783 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4784 {
4785         struct inode *inode = file_inode(file);
4786         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4787         struct btrfs_root *root = BTRFS_I(inode)->root;
4788         struct btrfs_super_block *super_block = fs_info->super_copy;
4789         struct btrfs_ioctl_feature_flags flags[2];
4790         struct btrfs_trans_handle *trans;
4791         u64 newflags;
4792         int ret;
4793
4794         if (!capable(CAP_SYS_ADMIN))
4795                 return -EPERM;
4796
4797         if (copy_from_user(flags, arg, sizeof(flags)))
4798                 return -EFAULT;
4799
4800         /* Nothing to do */
4801         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4802             !flags[0].incompat_flags)
4803                 return 0;
4804
4805         ret = check_feature(fs_info, flags[0].compat_flags,
4806                             flags[1].compat_flags, COMPAT);
4807         if (ret)
4808                 return ret;
4809
4810         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4811                             flags[1].compat_ro_flags, COMPAT_RO);
4812         if (ret)
4813                 return ret;
4814
4815         ret = check_feature(fs_info, flags[0].incompat_flags,
4816                             flags[1].incompat_flags, INCOMPAT);
4817         if (ret)
4818                 return ret;
4819
4820         ret = mnt_want_write_file(file);
4821         if (ret)
4822                 return ret;
4823
4824         trans = btrfs_start_transaction(root, 0);
4825         if (IS_ERR(trans)) {
4826                 ret = PTR_ERR(trans);
4827                 goto out_drop_write;
4828         }
4829
4830         spin_lock(&fs_info->super_lock);
4831         newflags = btrfs_super_compat_flags(super_block);
4832         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4833         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4834         btrfs_set_super_compat_flags(super_block, newflags);
4835
4836         newflags = btrfs_super_compat_ro_flags(super_block);
4837         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4838         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4839         btrfs_set_super_compat_ro_flags(super_block, newflags);
4840
4841         newflags = btrfs_super_incompat_flags(super_block);
4842         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4843         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4844         btrfs_set_super_incompat_flags(super_block, newflags);
4845         spin_unlock(&fs_info->super_lock);
4846
4847         ret = btrfs_commit_transaction(trans);
4848 out_drop_write:
4849         mnt_drop_write_file(file);
4850
4851         return ret;
4852 }
4853
4854 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4855 {
4856         struct btrfs_ioctl_send_args *arg;
4857         int ret;
4858
4859         if (compat) {
4860 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4861                 struct btrfs_ioctl_send_args_32 args32;
4862
4863                 ret = copy_from_user(&args32, argp, sizeof(args32));
4864                 if (ret)
4865                         return -EFAULT;
4866                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4867                 if (!arg)
4868                         return -ENOMEM;
4869                 arg->send_fd = args32.send_fd;
4870                 arg->clone_sources_count = args32.clone_sources_count;
4871                 arg->clone_sources = compat_ptr(args32.clone_sources);
4872                 arg->parent_root = args32.parent_root;
4873                 arg->flags = args32.flags;
4874                 memcpy(arg->reserved, args32.reserved,
4875                        sizeof(args32.reserved));
4876 #else
4877                 return -ENOTTY;
4878 #endif
4879         } else {
4880                 arg = memdup_user(argp, sizeof(*arg));
4881                 if (IS_ERR(arg))
4882                         return PTR_ERR(arg);
4883         }
4884         ret = btrfs_ioctl_send(file, arg);
4885         kfree(arg);
4886         return ret;
4887 }
4888
4889 long btrfs_ioctl(struct file *file, unsigned int
4890                 cmd, unsigned long arg)
4891 {
4892         struct inode *inode = file_inode(file);
4893         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4894         struct btrfs_root *root = BTRFS_I(inode)->root;
4895         void __user *argp = (void __user *)arg;
4896
4897         switch (cmd) {
4898         case FS_IOC_GETVERSION:
4899                 return btrfs_ioctl_getversion(file, argp);
4900         case FS_IOC_GETFSLABEL:
4901                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4902         case FS_IOC_SETFSLABEL:
4903                 return btrfs_ioctl_set_fslabel(file, argp);
4904         case FITRIM:
4905                 return btrfs_ioctl_fitrim(fs_info, argp);
4906         case BTRFS_IOC_SNAP_CREATE:
4907                 return btrfs_ioctl_snap_create(file, argp, 0);
4908         case BTRFS_IOC_SNAP_CREATE_V2:
4909                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4910         case BTRFS_IOC_SUBVOL_CREATE:
4911                 return btrfs_ioctl_snap_create(file, argp, 1);
4912         case BTRFS_IOC_SUBVOL_CREATE_V2:
4913                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4914         case BTRFS_IOC_SNAP_DESTROY:
4915                 return btrfs_ioctl_snap_destroy(file, argp, false);
4916         case BTRFS_IOC_SNAP_DESTROY_V2:
4917                 return btrfs_ioctl_snap_destroy(file, argp, true);
4918         case BTRFS_IOC_SUBVOL_GETFLAGS:
4919                 return btrfs_ioctl_subvol_getflags(file, argp);
4920         case BTRFS_IOC_SUBVOL_SETFLAGS:
4921                 return btrfs_ioctl_subvol_setflags(file, argp);
4922         case BTRFS_IOC_DEFAULT_SUBVOL:
4923                 return btrfs_ioctl_default_subvol(file, argp);
4924         case BTRFS_IOC_DEFRAG:
4925                 return btrfs_ioctl_defrag(file, NULL);
4926         case BTRFS_IOC_DEFRAG_RANGE:
4927                 return btrfs_ioctl_defrag(file, argp);
4928         case BTRFS_IOC_RESIZE:
4929                 return btrfs_ioctl_resize(file, argp);
4930         case BTRFS_IOC_ADD_DEV:
4931                 return btrfs_ioctl_add_dev(fs_info, argp);
4932         case BTRFS_IOC_RM_DEV:
4933                 return btrfs_ioctl_rm_dev(file, argp);
4934         case BTRFS_IOC_RM_DEV_V2:
4935                 return btrfs_ioctl_rm_dev_v2(file, argp);
4936         case BTRFS_IOC_FS_INFO:
4937                 return btrfs_ioctl_fs_info(fs_info, argp);
4938         case BTRFS_IOC_DEV_INFO:
4939                 return btrfs_ioctl_dev_info(fs_info, argp);
4940         case BTRFS_IOC_BALANCE:
4941                 return btrfs_ioctl_balance(file, NULL);
4942         case BTRFS_IOC_TREE_SEARCH:
4943                 return btrfs_ioctl_tree_search(file, argp);
4944         case BTRFS_IOC_TREE_SEARCH_V2:
4945                 return btrfs_ioctl_tree_search_v2(file, argp);
4946         case BTRFS_IOC_INO_LOOKUP:
4947                 return btrfs_ioctl_ino_lookup(file, argp);
4948         case BTRFS_IOC_INO_PATHS:
4949                 return btrfs_ioctl_ino_to_path(root, argp);
4950         case BTRFS_IOC_LOGICAL_INO:
4951                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4952         case BTRFS_IOC_LOGICAL_INO_V2:
4953                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4954         case BTRFS_IOC_SPACE_INFO:
4955                 return btrfs_ioctl_space_info(fs_info, argp);
4956         case BTRFS_IOC_SYNC: {
4957                 int ret;
4958
4959                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4960                 if (ret)
4961                         return ret;
4962                 ret = btrfs_sync_fs(inode->i_sb, 1);
4963                 /*
4964                  * The transaction thread may want to do more work,
4965                  * namely it pokes the cleaner kthread that will start
4966                  * processing uncleaned subvols.
4967                  */
4968                 wake_up_process(fs_info->transaction_kthread);
4969                 return ret;
4970         }
4971         case BTRFS_IOC_START_SYNC:
4972                 return btrfs_ioctl_start_sync(root, argp);
4973         case BTRFS_IOC_WAIT_SYNC:
4974                 return btrfs_ioctl_wait_sync(fs_info, argp);
4975         case BTRFS_IOC_SCRUB:
4976                 return btrfs_ioctl_scrub(file, argp);
4977         case BTRFS_IOC_SCRUB_CANCEL:
4978                 return btrfs_ioctl_scrub_cancel(fs_info);
4979         case BTRFS_IOC_SCRUB_PROGRESS:
4980                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4981         case BTRFS_IOC_BALANCE_V2:
4982                 return btrfs_ioctl_balance(file, argp);
4983         case BTRFS_IOC_BALANCE_CTL:
4984                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4985         case BTRFS_IOC_BALANCE_PROGRESS:
4986                 return btrfs_ioctl_balance_progress(fs_info, argp);
4987         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4988                 return btrfs_ioctl_set_received_subvol(file, argp);
4989 #ifdef CONFIG_64BIT
4990         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4991                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4992 #endif
4993         case BTRFS_IOC_SEND:
4994                 return _btrfs_ioctl_send(file, argp, false);
4995 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4996         case BTRFS_IOC_SEND_32:
4997                 return _btrfs_ioctl_send(file, argp, true);
4998 #endif
4999         case BTRFS_IOC_GET_DEV_STATS:
5000                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5001         case BTRFS_IOC_QUOTA_CTL:
5002                 return btrfs_ioctl_quota_ctl(file, argp);
5003         case BTRFS_IOC_QGROUP_ASSIGN:
5004                 return btrfs_ioctl_qgroup_assign(file, argp);
5005         case BTRFS_IOC_QGROUP_CREATE:
5006                 return btrfs_ioctl_qgroup_create(file, argp);
5007         case BTRFS_IOC_QGROUP_LIMIT:
5008                 return btrfs_ioctl_qgroup_limit(file, argp);
5009         case BTRFS_IOC_QUOTA_RESCAN:
5010                 return btrfs_ioctl_quota_rescan(file, argp);
5011         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5012                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5013         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5014                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5015         case BTRFS_IOC_DEV_REPLACE:
5016                 return btrfs_ioctl_dev_replace(fs_info, argp);
5017         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5018                 return btrfs_ioctl_get_supported_features(argp);
5019         case BTRFS_IOC_GET_FEATURES:
5020                 return btrfs_ioctl_get_features(fs_info, argp);
5021         case BTRFS_IOC_SET_FEATURES:
5022                 return btrfs_ioctl_set_features(file, argp);
5023         case BTRFS_IOC_GET_SUBVOL_INFO:
5024                 return btrfs_ioctl_get_subvol_info(file, argp);
5025         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5026                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5027         case BTRFS_IOC_INO_LOOKUP_USER:
5028                 return btrfs_ioctl_ino_lookup_user(file, argp);
5029         case FS_IOC_ENABLE_VERITY:
5030                 return fsverity_ioctl_enable(file, (const void __user *)argp);
5031         case FS_IOC_MEASURE_VERITY:
5032                 return fsverity_ioctl_measure(file, argp);
5033         }
5034
5035         return -ENOTTY;
5036 }
5037
5038 #ifdef CONFIG_COMPAT
5039 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5040 {
5041         /*
5042          * These all access 32-bit values anyway so no further
5043          * handling is necessary.
5044          */
5045         switch (cmd) {
5046         case FS_IOC32_GETVERSION:
5047                 cmd = FS_IOC_GETVERSION;
5048                 break;
5049         }
5050
5051         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5052 }
5053 #endif