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 "ctree.h"
  30 #include "disk-io.h"
  31 #include "export.h"
  32 #include "transaction.h"
  33 #include "btrfs_inode.h"
  34 #include "print-tree.h"
  35 #include "volumes.h"
  36 #include "locking.h"
  37 #include "inode-map.h"
  38 #include "backref.h"
  39 #include "rcu-string.h"
  40 #include "send.h"
  41 #include "dev-replace.h"
  42 #include "props.h"
  43 #include "sysfs.h"
  44 #include "qgroup.h"
  45 #include "tree-log.h"
  46 #include "compression.h"
  47 #include "space-info.h"
  48 #include "delalloc-space.h"
  49 #include "block-group.h"
  50
  51 #ifdef CONFIG_64BIT
  52 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  53  * structures are incorrect, as the timespec structure from userspace
  54  * is 4 bytes too small. We define these alternatives here to teach
  55  * the kernel about the 32-bit struct packing.
  56  */
  57 struct btrfs_ioctl_timespec_32 {
  58         __u64 sec;
  59         __u32 nsec;
  60 } __attribute__ ((__packed__));
  61
  62 struct btrfs_ioctl_received_subvol_args_32 {
  63         char    uuid[BTRFS_UUID_SIZE];  /* in */
  64         __u64   stransid;               /* in */
  65         __u64   rtransid;               /* out */
  66         struct btrfs_ioctl_timespec_32 stime; /* in */
  67         struct btrfs_ioctl_timespec_32 rtime; /* out */
  68         __u64   flags;                  /* in */
  69         __u64   reserved[16];           /* in */
  70 } __attribute__ ((__packed__));
  71
  72 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  73                                 struct btrfs_ioctl_received_subvol_args_32)
  74 #endif
  75
  76 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  77 struct btrfs_ioctl_send_args_32 {
  78         __s64 send_fd;                  /* in */
  79         __u64 clone_sources_count;      /* in */
  80         compat_uptr_t clone_sources;    /* in */
  81         __u64 parent_root;              /* in */
  82         __u64 flags;                    /* in */
  83         __u64 reserved[4];              /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  87                                struct btrfs_ioctl_send_args_32)
  88 #endif
  89
  90 /* Mask out flags that are inappropriate for the given type of inode. */
  91 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  92                 unsigned int flags)
  93 {
  94         if (S_ISDIR(inode->i_mode))
  95                 return flags;
  96         else if (S_ISREG(inode->i_mode))
  97                 return flags & ~FS_DIRSYNC_FL;
  98         else
  99                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 100 }
 101
 102 /*
 103  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 104  * ioctl.
 105  */
 106 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if (flags & BTRFS_INODE_NOCOMPRESS)
 126                 iflags |= FS_NOCOMP_FL;
 127         else if (flags & BTRFS_INODE_COMPRESS)
 128                 iflags |= FS_COMPR_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 137 {
 138         struct btrfs_inode *binode = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (binode->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (binode->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (binode->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (binode->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 158 {
 159         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 160         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 161
 162         if (copy_to_user(arg, &flags, sizeof(flags)))
 163                 return -EFAULT;
 164         return 0;
 165 }
 166
 167 /*
 168  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 169  * the old and new flags are not conflicting
 170  */
 171 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 172 {
 173         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 174                       FS_NOATIME_FL | FS_NODUMP_FL | \
 175                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 176                       FS_NOCOMP_FL | FS_COMPR_FL |
 177                       FS_NOCOW_FL))
 178                 return -EOPNOTSUPP;
 179
 180         /* COMPR and NOCOMP on new/old are valid */
 181         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 182                 return -EINVAL;
 183
 184         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 185                 return -EINVAL;
 186
 187         /* NOCOW and compression options are mutually exclusive */
 188         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 189                 return -EINVAL;
 190         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 191                 return -EINVAL;
 192
 193         return 0;
 194 }
 195
 196 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 197 {
 198         struct inode *inode = file_inode(file);
 199         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 200         struct btrfs_inode *binode = BTRFS_I(inode);
 201         struct btrfs_root *root = binode->root;
 202         struct btrfs_trans_handle *trans;
 203         unsigned int fsflags, old_fsflags;
 204         int ret;
 205         const char *comp = NULL;
 206         u32 binode_flags;
 207
 208         if (!inode_owner_or_capable(inode))
 209                 return -EPERM;
 210
 211         if (btrfs_root_readonly(root))
 212                 return -EROFS;
 213
 214         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 215                 return -EFAULT;
 216
 217         ret = mnt_want_write_file(file);
 218         if (ret)
 219                 return ret;
 220
 221         inode_lock(inode);
 222         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 223         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 224
 225         ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 226         if (ret)
 227                 goto out_unlock;
 228
 229         ret = check_fsflags(old_fsflags, fsflags);
 230         if (ret)
 231                 goto out_unlock;
 232
 233         binode_flags = binode->flags;
 234         if (fsflags & FS_SYNC_FL)
 235                 binode_flags |= BTRFS_INODE_SYNC;
 236         else
 237                 binode_flags &= ~BTRFS_INODE_SYNC;
 238         if (fsflags & FS_IMMUTABLE_FL)
 239                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 240         else
 241                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 242         if (fsflags & FS_APPEND_FL)
 243                 binode_flags |= BTRFS_INODE_APPEND;
 244         else
 245                 binode_flags &= ~BTRFS_INODE_APPEND;
 246         if (fsflags & FS_NODUMP_FL)
 247                 binode_flags |= BTRFS_INODE_NODUMP;
 248         else
 249                 binode_flags &= ~BTRFS_INODE_NODUMP;
 250         if (fsflags & FS_NOATIME_FL)
 251                 binode_flags |= BTRFS_INODE_NOATIME;
 252         else
 253                 binode_flags &= ~BTRFS_INODE_NOATIME;
 254         if (fsflags & FS_DIRSYNC_FL)
 255                 binode_flags |= BTRFS_INODE_DIRSYNC;
 256         else
 257                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 258         if (fsflags & FS_NOCOW_FL) {
 259                 if (S_ISREG(inode->i_mode)) {
 260                         /*
 261                          * It's safe to turn csums off here, no extents exist.
 262                          * Otherwise we want the flag to reflect the real COW
 263                          * status of the file and will not set it.
 264                          */
 265                         if (inode->i_size == 0)
 266                                 binode_flags |= BTRFS_INODE_NODATACOW |
 267                                                 BTRFS_INODE_NODATASUM;
 268                 } else {
 269                         binode_flags |= BTRFS_INODE_NODATACOW;
 270                 }
 271         } else {
 272                 /*
 273                  * Revert back under same assumptions as above
 274                  */
 275                 if (S_ISREG(inode->i_mode)) {
 276                         if (inode->i_size == 0)
 277                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 278                                                   BTRFS_INODE_NODATASUM);
 279                 } else {
 280                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 281                 }
 282         }
 283
 284         /*
 285          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 286          * flag may be changed automatically if compression code won't make
 287          * things smaller.
 288          */
 289         if (fsflags & FS_NOCOMP_FL) {
 290                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 291                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 292         } else if (fsflags & FS_COMPR_FL) {
 293
 294                 if (IS_SWAPFILE(inode)) {
 295                         ret = -ETXTBSY;
 296                         goto out_unlock;
 297                 }
 298
 299                 binode_flags |= BTRFS_INODE_COMPRESS;
 300                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 301
 302                 comp = btrfs_compress_type2str(fs_info->compress_type);
 303                 if (!comp || comp[0] == 0)
 304                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 305         } else {
 306                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 307         }
 308
 309         /*
 310          * 1 for inode item
 311          * 2 for properties
 312          */
 313         trans = btrfs_start_transaction(root, 3);
 314         if (IS_ERR(trans)) {
 315                 ret = PTR_ERR(trans);
 316                 goto out_unlock;
 317         }
 318
 319         if (comp) {
 320                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 321                                      strlen(comp), 0);
 322                 if (ret) {
 323                         btrfs_abort_transaction(trans, ret);
 324                         goto out_end_trans;
 325                 }
 326         } else {
 327                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 328                                      0, 0);
 329                 if (ret && ret != -ENODATA) {
 330                         btrfs_abort_transaction(trans, ret);
 331                         goto out_end_trans;
 332                 }
 333         }
 334
 335         binode->flags = binode_flags;
 336         btrfs_sync_inode_flags_to_i_flags(inode);
 337         inode_inc_iversion(inode);
 338         inode->i_ctime = current_time(inode);
 339         ret = btrfs_update_inode(trans, root, inode);
 340
 341  out_end_trans:
 342         btrfs_end_transaction(trans);
 343  out_unlock:
 344         inode_unlock(inode);
 345         mnt_drop_write_file(file);
 346         return ret;
 347 }
 348
 349 /*
 350  * Translate btrfs internal inode flags to xflags as expected by the
 351  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 352  * silently dropped.
 353  */
 354 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 355 {
 356         unsigned int xflags = 0;
 357
 358         if (flags & BTRFS_INODE_APPEND)
 359                 xflags |= FS_XFLAG_APPEND;
 360         if (flags & BTRFS_INODE_IMMUTABLE)
 361                 xflags |= FS_XFLAG_IMMUTABLE;
 362         if (flags & BTRFS_INODE_NOATIME)
 363                 xflags |= FS_XFLAG_NOATIME;
 364         if (flags & BTRFS_INODE_NODUMP)
 365                 xflags |= FS_XFLAG_NODUMP;
 366         if (flags & BTRFS_INODE_SYNC)
 367                 xflags |= FS_XFLAG_SYNC;
 368
 369         return xflags;
 370 }
 371
 372 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 373 static int check_xflags(unsigned int flags)
 374 {
 375         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 376                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 377                 return -EOPNOTSUPP;
 378         return 0;
 379 }
 380
 381 /*
 382  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 383  * are zeroed.
 384  */
 385 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 386 {
 387         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 388         struct fsxattr fa;
 389
 390         simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 391         if (copy_to_user(arg, &fa, sizeof(fa)))
 392                 return -EFAULT;
 393
 394         return 0;
 395 }
 396
 397 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 398 {
 399         struct inode *inode = file_inode(file);
 400         struct btrfs_inode *binode = BTRFS_I(inode);
 401         struct btrfs_root *root = binode->root;
 402         struct btrfs_trans_handle *trans;
 403         struct fsxattr fa, old_fa;
 404         unsigned old_flags;
 405         unsigned old_i_flags;
 406         int ret = 0;
 407
 408         if (!inode_owner_or_capable(inode))
 409                 return -EPERM;
 410
 411         if (btrfs_root_readonly(root))
 412                 return -EROFS;
 413
 414         if (copy_from_user(&fa, arg, sizeof(fa)))
 415                 return -EFAULT;
 416
 417         ret = check_xflags(fa.fsx_xflags);
 418         if (ret)
 419                 return ret;
 420
 421         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 422                 return -EOPNOTSUPP;
 423
 424         ret = mnt_want_write_file(file);
 425         if (ret)
 426                 return ret;
 427
 428         inode_lock(inode);
 429
 430         old_flags = binode->flags;
 431         old_i_flags = inode->i_flags;
 432
 433         simple_fill_fsxattr(&old_fa,
 434                             btrfs_inode_flags_to_xflags(binode->flags));
 435         ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 436         if (ret)
 437                 goto out_unlock;
 438
 439         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 440                 binode->flags |= BTRFS_INODE_SYNC;
 441         else
 442                 binode->flags &= ~BTRFS_INODE_SYNC;
 443         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 444                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 445         else
 446                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 447         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 448                 binode->flags |= BTRFS_INODE_APPEND;
 449         else
 450                 binode->flags &= ~BTRFS_INODE_APPEND;
 451         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 452                 binode->flags |= BTRFS_INODE_NODUMP;
 453         else
 454                 binode->flags &= ~BTRFS_INODE_NODUMP;
 455         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 456                 binode->flags |= BTRFS_INODE_NOATIME;
 457         else
 458                 binode->flags &= ~BTRFS_INODE_NOATIME;
 459
 460         /* 1 item for the inode */
 461         trans = btrfs_start_transaction(root, 1);
 462         if (IS_ERR(trans)) {
 463                 ret = PTR_ERR(trans);
 464                 goto out_unlock;
 465         }
 466
 467         btrfs_sync_inode_flags_to_i_flags(inode);
 468         inode_inc_iversion(inode);
 469         inode->i_ctime = current_time(inode);
 470         ret = btrfs_update_inode(trans, root, inode);
 471
 472         btrfs_end_transaction(trans);
 473
 474 out_unlock:
 475         if (ret) {
 476                 binode->flags = old_flags;
 477                 inode->i_flags = old_i_flags;
 478         }
 479
 480         inode_unlock(inode);
 481         mnt_drop_write_file(file);
 482
 483         return ret;
 484 }
 485
 486 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 487 {
 488         struct inode *inode = file_inode(file);
 489
 490         return put_user(inode->i_generation, arg);
 491 }
 492
 493 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 494                                         void __user *arg)
 495 {
 496         struct btrfs_device *device;
 497         struct request_queue *q;
 498         struct fstrim_range range;
 499         u64 minlen = ULLONG_MAX;
 500         u64 num_devices = 0;
 501         int ret;
 502
 503         if (!capable(CAP_SYS_ADMIN))
 504                 return -EPERM;
 505
 506         /*
 507          * If the fs is mounted with nologreplay, which requires it to be
 508          * mounted in RO mode as well, we can not allow discard on free space
 509          * inside block groups, because log trees refer to extents that are not
 510          * pinned in a block group's free space cache (pinning the extents is
 511          * precisely the first phase of replaying a log tree).
 512          */
 513         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 514                 return -EROFS;
 515
 516         rcu_read_lock();
 517         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 518                                 dev_list) {
 519                 if (!device->bdev)
 520                         continue;
 521                 q = bdev_get_queue(device->bdev);
 522                 if (blk_queue_discard(q)) {
 523                         num_devices++;
 524                         minlen = min_t(u64, q->limits.discard_granularity,
 525                                      minlen);
 526                 }
 527         }
 528         rcu_read_unlock();
 529
 530         if (!num_devices)
 531                 return -EOPNOTSUPP;
 532         if (copy_from_user(&range, arg, sizeof(range)))
 533                 return -EFAULT;
 534
 535         /*
 536          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 537          * block group is in the logical address space, which can be any
 538          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 539          */
 540         if (range.len < fs_info->sb->s_blocksize)
 541                 return -EINVAL;
 542
 543         range.minlen = max(range.minlen, minlen);
 544         ret = btrfs_trim_fs(fs_info, &range);
 545         if (ret < 0)
 546                 return ret;
 547
 548         if (copy_to_user(arg, &range, sizeof(range)))
 549                 return -EFAULT;
 550
 551         return 0;
 552 }
 553
 554 int __pure btrfs_is_empty_uuid(u8 *uuid)
 555 {
 556         int i;
 557
 558         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 559                 if (uuid[i])
 560                         return 0;
 561         }
 562         return 1;
 563 }
 564
 565 static noinline int create_subvol(struct inode *dir,
 566                                   struct dentry *dentry,
 567                                   const char *name, int namelen,
 568                                   struct btrfs_qgroup_inherit *inherit)
 569 {
 570         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 571         struct btrfs_trans_handle *trans;
 572         struct btrfs_key key;
 573         struct btrfs_root_item *root_item;
 574         struct btrfs_inode_item *inode_item;
 575         struct extent_buffer *leaf;
 576         struct btrfs_root *root = BTRFS_I(dir)->root;
 577         struct btrfs_root *new_root;
 578         struct btrfs_block_rsv block_rsv;
 579         struct timespec64 cur_time = current_time(dir);
 580         struct inode *inode;
 581         int ret;
 582         int err;
 583         dev_t anon_dev = 0;
 584         u64 objectid;
 585         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 586         u64 index = 0;
 587
 588         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 589         if (!root_item)
 590                 return -ENOMEM;
 591
 592         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 593         if (ret)
 594                 goto fail_free;
 595
 596         ret = get_anon_bdev(&anon_dev);
 597         if (ret < 0)
 598                 goto fail_free;
 599
 600         /*
 601          * Don't create subvolume whose level is not zero. Or qgroup will be
 602          * screwed up since it assumes subvolume qgroup's level to be 0.
 603          */
 604         if (btrfs_qgroup_level(objectid)) {
 605                 ret = -ENOSPC;
 606                 goto fail_free;
 607         }
 608
 609         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 610         /*
 611          * The same as the snapshot creation, please see the comment
 612          * of create_snapshot().
 613          */
 614         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 615         if (ret)
 616                 goto fail_free;
 617
 618         trans = btrfs_start_transaction(root, 0);
 619         if (IS_ERR(trans)) {
 620                 ret = PTR_ERR(trans);
 621                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 622                 goto fail_free;
 623         }
 624         trans->block_rsv = &block_rsv;
 625         trans->bytes_reserved = block_rsv.size;
 626
 627         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 628         if (ret)
 629                 goto fail;
 630
 631         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 632         if (IS_ERR(leaf)) {
 633                 ret = PTR_ERR(leaf);
 634                 goto fail;
 635         }
 636
 637         btrfs_mark_buffer_dirty(leaf);
 638
 639         inode_item = &root_item->inode;
 640         btrfs_set_stack_inode_generation(inode_item, 1);
 641         btrfs_set_stack_inode_size(inode_item, 3);
 642         btrfs_set_stack_inode_nlink(inode_item, 1);
 643         btrfs_set_stack_inode_nbytes(inode_item,
 644                                      fs_info->nodesize);
 645         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 646
 647         btrfs_set_root_flags(root_item, 0);
 648         btrfs_set_root_limit(root_item, 0);
 649         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 650
 651         btrfs_set_root_bytenr(root_item, leaf->start);
 652         btrfs_set_root_generation(root_item, trans->transid);
 653         btrfs_set_root_level(root_item, 0);
 654         btrfs_set_root_refs(root_item, 1);
 655         btrfs_set_root_used(root_item, leaf->len);
 656         btrfs_set_root_last_snapshot(root_item, 0);
 657
 658         btrfs_set_root_generation_v2(root_item,
 659                         btrfs_root_generation(root_item));
 660         generate_random_guid(root_item->uuid);
 661         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 662         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 663         root_item->ctime = root_item->otime;
 664         btrfs_set_root_ctransid(root_item, trans->transid);
 665         btrfs_set_root_otransid(root_item, trans->transid);
 666
 667         btrfs_tree_unlock(leaf);
 668         free_extent_buffer(leaf);
 669         leaf = NULL;
 670
 671         btrfs_set_root_dirid(root_item, new_dirid);
 672
 673         key.objectid = objectid;
 674         key.offset = 0;
 675         key.type = BTRFS_ROOT_ITEM_KEY;
 676         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 677                                 root_item);
 678         if (ret)
 679                 goto fail;
 680
 681         key.offset = (u64)-1;
 682         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 683         if (IS_ERR(new_root)) {
 684                 free_anon_bdev(anon_dev);
 685                 ret = PTR_ERR(new_root);
 686                 btrfs_abort_transaction(trans, ret);
 687                 goto fail;
 688         }
 689         /* Freeing will be done in btrfs_put_root() of new_root */
 690         anon_dev = 0;
 691
 692         btrfs_record_root_in_trans(trans, new_root);
 693
 694         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 695         btrfs_put_root(new_root);
 696         if (ret) {
 697                 /* We potentially lose an unused inode item here */
 698                 btrfs_abort_transaction(trans, ret);
 699                 goto fail;
 700         }
 701
 702         mutex_lock(&new_root->objectid_mutex);
 703         new_root->highest_objectid = new_dirid;
 704         mutex_unlock(&new_root->objectid_mutex);
 705
 706         /*
 707          * insert the directory item
 708          */
 709         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 710         if (ret) {
 711                 btrfs_abort_transaction(trans, ret);
 712                 goto fail;
 713         }
 714
 715         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 716                                     BTRFS_FT_DIR, index);
 717         if (ret) {
 718                 btrfs_abort_transaction(trans, ret);
 719                 goto fail;
 720         }
 721
 722         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 723         ret = btrfs_update_inode(trans, root, dir);
 724         if (ret) {
 725                 btrfs_abort_transaction(trans, ret);
 726                 goto fail;
 727         }
 728
 729         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 730                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 731         if (ret) {
 732                 btrfs_abort_transaction(trans, ret);
 733                 goto fail;
 734         }
 735
 736         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 737                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 738         if (ret)
 739                 btrfs_abort_transaction(trans, ret);
 740
 741 fail:
 742         kfree(root_item);
 743         trans->block_rsv = NULL;
 744         trans->bytes_reserved = 0;
 745         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 746
 747         err = btrfs_commit_transaction(trans);
 748         if (err && !ret)
 749                 ret = err;
 750
 751         if (!ret) {
 752                 inode = btrfs_lookup_dentry(dir, dentry);
 753                 if (IS_ERR(inode))
 754                         return PTR_ERR(inode);
 755                 d_instantiate(dentry, inode);
 756         }
 757         return ret;
 758
 759 fail_free:
 760         if (anon_dev)
 761                 free_anon_bdev(anon_dev);
 762         kfree(root_item);
 763         return ret;
 764 }
 765
 766 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 767                            struct dentry *dentry, bool readonly,
 768                            struct btrfs_qgroup_inherit *inherit)
 769 {
 770         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 771         struct inode *inode;
 772         struct btrfs_pending_snapshot *pending_snapshot;
 773         struct btrfs_trans_handle *trans;
 774         int ret;
 775
 776         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 777                 return -EINVAL;
 778
 779         if (atomic_read(&root->nr_swapfiles)) {
 780                 btrfs_warn(fs_info,
 781                            "cannot snapshot subvolume with active swapfile");
 782                 return -ETXTBSY;
 783         }
 784
 785         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 786         if (!pending_snapshot)
 787                 return -ENOMEM;
 788
 789         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 790         if (ret < 0)
 791                 goto free_pending;
 792         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 793                         GFP_KERNEL);
 794         pending_snapshot->path = btrfs_alloc_path();
 795         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 796                 ret = -ENOMEM;
 797                 goto free_pending;
 798         }
 799
 800         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 801                              BTRFS_BLOCK_RSV_TEMP);
 802         /*
 803          * 1 - parent dir inode
 804          * 2 - dir entries
 805          * 1 - root item
 806          * 2 - root ref/backref
 807          * 1 - root of snapshot
 808          * 1 - UUID item
 809          */
 810         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 811                                         &pending_snapshot->block_rsv, 8,
 812                                         false);
 813         if (ret)
 814                 goto free_pending;
 815
 816         pending_snapshot->dentry = dentry;
 817         pending_snapshot->root = root;
 818         pending_snapshot->readonly = readonly;
 819         pending_snapshot->dir = dir;
 820         pending_snapshot->inherit = inherit;
 821
 822         trans = btrfs_start_transaction(root, 0);
 823         if (IS_ERR(trans)) {
 824                 ret = PTR_ERR(trans);
 825                 goto fail;
 826         }
 827
 828         spin_lock(&fs_info->trans_lock);
 829         list_add(&pending_snapshot->list,
 830                  &trans->transaction->pending_snapshots);
 831         spin_unlock(&fs_info->trans_lock);
 832
 833         ret = btrfs_commit_transaction(trans);
 834         if (ret)
 835                 goto fail;
 836
 837         ret = pending_snapshot->error;
 838         if (ret)
 839                 goto fail;
 840
 841         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 842         if (ret)
 843                 goto fail;
 844
 845         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 846         if (IS_ERR(inode)) {
 847                 ret = PTR_ERR(inode);
 848                 goto fail;
 849         }
 850
 851         d_instantiate(dentry, inode);
 852         ret = 0;
 853         pending_snapshot->anon_dev = 0;
 854 fail:
 855         /* Prevent double freeing of anon_dev */
 856         if (ret && pending_snapshot->snap)
 857                 pending_snapshot->snap->anon_dev = 0;
 858         btrfs_put_root(pending_snapshot->snap);
 859         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 860 free_pending:
 861         if (pending_snapshot->anon_dev)
 862                 free_anon_bdev(pending_snapshot->anon_dev);
 863         kfree(pending_snapshot->root_item);
 864         btrfs_free_path(pending_snapshot->path);
 865         kfree(pending_snapshot);
 866
 867         return ret;
 868 }
 869
 870 /*  copy of may_delete in fs/namei.c()
 871  *      Check whether we can remove a link victim from directory dir, check
 872  *  whether the type of victim is right.
 873  *  1. We can't do it if dir is read-only (done in permission())
 874  *  2. We should have write and exec permissions on dir
 875  *  3. We can't remove anything from append-only dir
 876  *  4. We can't do anything with immutable dir (done in permission())
 877  *  5. If the sticky bit on dir is set we should either
 878  *      a. be owner of dir, or
 879  *      b. be owner of victim, or
 880  *      c. have CAP_FOWNER capability
 881  *  6. If the victim is append-only or immutable we can't do anything with
 882  *     links pointing to it.
 883  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 884  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 885  *  9. We can't remove a root or mountpoint.
 886  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 887  *     nfs_async_unlink().
 888  */
 889
 890 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 891 {
 892         int error;
 893
 894         if (d_really_is_negative(victim))
 895                 return -ENOENT;
 896
 897         BUG_ON(d_inode(victim->d_parent) != dir);
 898         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 899
 900         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 901         if (error)
 902                 return error;
 903         if (IS_APPEND(dir))
 904                 return -EPERM;
 905         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 906             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 907                 return -EPERM;
 908         if (isdir) {
 909                 if (!d_is_dir(victim))
 910                         return -ENOTDIR;
 911                 if (IS_ROOT(victim))
 912                         return -EBUSY;
 913         } else if (d_is_dir(victim))
 914                 return -EISDIR;
 915         if (IS_DEADDIR(dir))
 916                 return -ENOENT;
 917         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 918                 return -EBUSY;
 919         return 0;
 920 }
 921
 922 /* copy of may_create in fs/namei.c() */
 923 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 924 {
 925         if (d_really_is_positive(child))
 926                 return -EEXIST;
 927         if (IS_DEADDIR(dir))
 928                 return -ENOENT;
 929         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 930 }
 931
 932 /*
 933  * Create a new subvolume below @parent.  This is largely modeled after
 934  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 935  * inside this filesystem so it's quite a bit simpler.
 936  */
 937 static noinline int btrfs_mksubvol(const struct path *parent,
 938                                    const char *name, int namelen,
 939                                    struct btrfs_root *snap_src,
 940                                    bool readonly,
 941                                    struct btrfs_qgroup_inherit *inherit)
 942 {
 943         struct inode *dir = d_inode(parent->dentry);
 944         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 945         struct dentry *dentry;
 946         int error;
 947
 948         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 949         if (error == -EINTR)
 950                 return error;
 951
 952         dentry = lookup_one_len(name, parent->dentry, namelen);
 953         error = PTR_ERR(dentry);
 954         if (IS_ERR(dentry))
 955                 goto out_unlock;
 956
 957         error = btrfs_may_create(dir, dentry);
 958         if (error)
 959                 goto out_dput;
 960
 961         /*
 962          * even if this name doesn't exist, we may get hash collisions.
 963          * check for them now when we can safely fail
 964          */
 965         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 966                                                dir->i_ino, name,
 967                                                namelen);
 968         if (error)
 969                 goto out_dput;
 970
 971         down_read(&fs_info->subvol_sem);
 972
 973         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 974                 goto out_up_read;
 975
 976         if (snap_src)
 977                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 978         else
 979                 error = create_subvol(dir, dentry, name, namelen, inherit);
 980
 981         if (!error)
 982                 fsnotify_mkdir(dir, dentry);
 983 out_up_read:
 984         up_read(&fs_info->subvol_sem);
 985 out_dput:
 986         dput(dentry);
 987 out_unlock:
 988         inode_unlock(dir);
 989         return error;
 990 }
 991
 992 static noinline int btrfs_mksnapshot(const struct path *parent,
 993                                    const char *name, int namelen,
 994                                    struct btrfs_root *root,
 995                                    bool readonly,
 996                                    struct btrfs_qgroup_inherit *inherit)
 997 {
 998         int ret;
 999         bool snapshot_force_cow = false;
1000
1001         /*
1002          * Force new buffered writes to reserve space even when NOCOW is
1003          * possible. This is to avoid later writeback (running dealloc) to
1004          * fallback to COW mode and unexpectedly fail with ENOSPC.
1005          */
1006         btrfs_drew_read_lock(&root->snapshot_lock);
1007
1008         ret = btrfs_start_delalloc_snapshot(root);
1009         if (ret)
1010                 goto out;
1011
1012         /*
1013          * All previous writes have started writeback in NOCOW mode, so now
1014          * we force future writes to fallback to COW mode during snapshot
1015          * creation.
1016          */
1017         atomic_inc(&root->snapshot_force_cow);
1018         snapshot_force_cow = true;
1019
1020         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1021
1022         ret = btrfs_mksubvol(parent, name, namelen,
1023                              root, readonly, inherit);
1024 out:
1025         if (snapshot_force_cow)
1026                 atomic_dec(&root->snapshot_force_cow);
1027         btrfs_drew_read_unlock(&root->snapshot_lock);
1028         return ret;
1029 }
1030
1031 /*
1032  * When we're defragging a range, we don't want to kick it off again
1033  * if it is really just waiting for delalloc to send it down.
1034  * If we find a nice big extent or delalloc range for the bytes in the
1035  * file you want to defrag, we return 0 to let you know to skip this
1036  * part of the file
1037  */
1038 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1039 {
1040         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1041         struct extent_map *em = NULL;
1042         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1043         u64 end;
1044
1045         read_lock(&em_tree->lock);
1046         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1047         read_unlock(&em_tree->lock);
1048
1049         if (em) {
1050                 end = extent_map_end(em);
1051                 free_extent_map(em);
1052                 if (end - offset > thresh)
1053                         return 0;
1054         }
1055         /* if we already have a nice delalloc here, just stop */
1056         thresh /= 2;
1057         end = count_range_bits(io_tree, &offset, offset + thresh,
1058                                thresh, EXTENT_DELALLOC, 1);
1059         if (end >= thresh)
1060                 return 0;
1061         return 1;
1062 }
1063
1064 /*
1065  * helper function to walk through a file and find extents
1066  * newer than a specific transid, and smaller than thresh.
1067  *
1068  * This is used by the defragging code to find new and small
1069  * extents
1070  */
1071 static int find_new_extents(struct btrfs_root *root,
1072                             struct inode *inode, u64 newer_than,
1073                             u64 *off, u32 thresh)
1074 {
1075         struct btrfs_path *path;
1076         struct btrfs_key min_key;
1077         struct extent_buffer *leaf;
1078         struct btrfs_file_extent_item *extent;
1079         int type;
1080         int ret;
1081         u64 ino = btrfs_ino(BTRFS_I(inode));
1082
1083         path = btrfs_alloc_path();
1084         if (!path)
1085                 return -ENOMEM;
1086
1087         min_key.objectid = ino;
1088         min_key.type = BTRFS_EXTENT_DATA_KEY;
1089         min_key.offset = *off;
1090
1091         while (1) {
1092                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1093                 if (ret != 0)
1094                         goto none;
1095 process_slot:
1096                 if (min_key.objectid != ino)
1097                         goto none;
1098                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1099                         goto none;
1100
1101                 leaf = path->nodes[0];
1102                 extent = btrfs_item_ptr(leaf, path->slots[0],
1103                                         struct btrfs_file_extent_item);
1104
1105                 type = btrfs_file_extent_type(leaf, extent);
1106                 if (type == BTRFS_FILE_EXTENT_REG &&
1107                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1108                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1109                         *off = min_key.offset;
1110                         btrfs_free_path(path);
1111                         return 0;
1112                 }
1113
1114                 path->slots[0]++;
1115                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1116                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1117                         goto process_slot;
1118                 }
1119
1120                 if (min_key.offset == (u64)-1)
1121                         goto none;
1122
1123                 min_key.offset++;
1124                 btrfs_release_path(path);
1125         }
1126 none:
1127         btrfs_free_path(path);
1128         return -ENOENT;
1129 }
1130
1131 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1132 {
1133         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1134         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1135         struct extent_map *em;
1136         u64 len = PAGE_SIZE;
1137
1138         /*
1139          * hopefully we have this extent in the tree already, try without
1140          * the full extent lock
1141          */
1142         read_lock(&em_tree->lock);
1143         em = lookup_extent_mapping(em_tree, start, len);
1144         read_unlock(&em_tree->lock);
1145
1146         if (!em) {
1147                 struct extent_state *cached = NULL;
1148                 u64 end = start + len - 1;
1149
1150                 /* get the big lock and read metadata off disk */
1151                 lock_extent_bits(io_tree, start, end, &cached);
1152                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1153                 unlock_extent_cached(io_tree, start, end, &cached);
1154
1155                 if (IS_ERR(em))
1156                         return NULL;
1157         }
1158
1159         return em;
1160 }
1161
1162 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1163 {
1164         struct extent_map *next;
1165         bool ret = true;
1166
1167         /* this is the last extent */
1168         if (em->start + em->len >= i_size_read(inode))
1169                 return false;
1170
1171         next = defrag_lookup_extent(inode, em->start + em->len);
1172         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1173                 ret = false;
1174         else if ((em->block_start + em->block_len == next->block_start) &&
1175                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1176                 ret = false;
1177
1178         free_extent_map(next);
1179         return ret;
1180 }
1181
1182 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1183                                u64 *last_len, u64 *skip, u64 *defrag_end,
1184                                int compress)
1185 {
1186         struct extent_map *em;
1187         int ret = 1;
1188         bool next_mergeable = true;
1189         bool prev_mergeable = true;
1190
1191         /*
1192          * make sure that once we start defragging an extent, we keep on
1193          * defragging it
1194          */
1195         if (start < *defrag_end)
1196                 return 1;
1197
1198         *skip = 0;
1199
1200         em = defrag_lookup_extent(inode, start);
1201         if (!em)
1202                 return 0;
1203
1204         /* this will cover holes, and inline extents */
1205         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1206                 ret = 0;
1207                 goto out;
1208         }
1209
1210         if (!*defrag_end)
1211                 prev_mergeable = false;
1212
1213         next_mergeable = defrag_check_next_extent(inode, em);
1214         /*
1215          * we hit a real extent, if it is big or the next extent is not a
1216          * real extent, don't bother defragging it
1217          */
1218         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1219             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1220                 ret = 0;
1221 out:
1222         /*
1223          * last_len ends up being a counter of how many bytes we've defragged.
1224          * every time we choose not to defrag an extent, we reset *last_len
1225          * so that the next tiny extent will force a defrag.
1226          *
1227          * The end result of this is that tiny extents before a single big
1228          * extent will force at least part of that big extent to be defragged.
1229          */
1230         if (ret) {
1231                 *defrag_end = extent_map_end(em);
1232         } else {
1233                 *last_len = 0;
1234                 *skip = extent_map_end(em);
1235                 *defrag_end = 0;
1236         }
1237
1238         free_extent_map(em);
1239         return ret;
1240 }
1241
1242 /*
1243  * it doesn't do much good to defrag one or two pages
1244  * at a time.  This pulls in a nice chunk of pages
1245  * to COW and defrag.
1246  *
1247  * It also makes sure the delalloc code has enough
1248  * dirty data to avoid making new small extents as part
1249  * of the defrag
1250  *
1251  * It's a good idea to start RA on this range
1252  * before calling this.
1253  */
1254 static int cluster_pages_for_defrag(struct inode *inode,
1255                                     struct page **pages,
1256                                     unsigned long start_index,
1257                                     unsigned long num_pages)
1258 {
1259         unsigned long file_end;
1260         u64 isize = i_size_read(inode);
1261         u64 page_start;
1262         u64 page_end;
1263         u64 page_cnt;
1264         int ret;
1265         int i;
1266         int i_done;
1267         struct btrfs_ordered_extent *ordered;
1268         struct extent_state *cached_state = NULL;
1269         struct extent_io_tree *tree;
1270         struct extent_changeset *data_reserved = NULL;
1271         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1272
1273         file_end = (isize - 1) >> PAGE_SHIFT;
1274         if (!isize || start_index > file_end)
1275                 return 0;
1276
1277         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1278
1279         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1280                         start_index << PAGE_SHIFT,
1281                         page_cnt << PAGE_SHIFT);
1282         if (ret)
1283                 return ret;
1284         i_done = 0;
1285         tree = &BTRFS_I(inode)->io_tree;
1286
1287         /* step one, lock all the pages */
1288         for (i = 0; i < page_cnt; i++) {
1289                 struct page *page;
1290 again:
1291                 page = find_or_create_page(inode->i_mapping,
1292                                            start_index + i, mask);
1293                 if (!page)
1294                         break;
1295
1296                 page_start = page_offset(page);
1297                 page_end = page_start + PAGE_SIZE - 1;
1298                 while (1) {
1299                         lock_extent_bits(tree, page_start, page_end,
1300                                          &cached_state);
1301                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1302                                                               page_start);
1303                         unlock_extent_cached(tree, page_start, page_end,
1304                                              &cached_state);
1305                         if (!ordered)
1306                                 break;
1307
1308                         unlock_page(page);
1309                         btrfs_start_ordered_extent(inode, ordered, 1);
1310                         btrfs_put_ordered_extent(ordered);
1311                         lock_page(page);
1312                         /*
1313                          * we unlocked the page above, so we need check if
1314                          * it was released or not.
1315                          */
1316                         if (page->mapping != inode->i_mapping) {
1317                                 unlock_page(page);
1318                                 put_page(page);
1319                                 goto again;
1320                         }
1321                 }
1322
1323                 if (!PageUptodate(page)) {
1324                         btrfs_readpage(NULL, page);
1325                         lock_page(page);
1326                         if (!PageUptodate(page)) {
1327                                 unlock_page(page);
1328                                 put_page(page);
1329                                 ret = -EIO;
1330                                 break;
1331                         }
1332                 }
1333
1334                 if (page->mapping != inode->i_mapping) {
1335                         unlock_page(page);
1336                         put_page(page);
1337                         goto again;
1338                 }
1339
1340                 pages[i] = page;
1341                 i_done++;
1342         }
1343         if (!i_done || ret)
1344                 goto out;
1345
1346         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1347                 goto out;
1348
1349         /*
1350          * so now we have a nice long stream of locked
1351          * and up to date pages, lets wait on them
1352          */
1353         for (i = 0; i < i_done; i++)
1354                 wait_on_page_writeback(pages[i]);
1355
1356         page_start = page_offset(pages[0]);
1357         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1358
1359         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1360                          page_start, page_end - 1, &cached_state);
1361         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1362                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1363                           EXTENT_DEFRAG, 0, 0, &cached_state);
1364
1365         if (i_done != page_cnt) {
1366                 spin_lock(&BTRFS_I(inode)->lock);
1367                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1368                 spin_unlock(&BTRFS_I(inode)->lock);
1369                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1370                                 start_index << PAGE_SHIFT,
1371                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1372         }
1373
1374
1375         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1376                           &cached_state);
1377
1378         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1379                              page_start, page_end - 1, &cached_state);
1380
1381         for (i = 0; i < i_done; i++) {
1382                 clear_page_dirty_for_io(pages[i]);
1383                 ClearPageChecked(pages[i]);
1384                 set_page_extent_mapped(pages[i]);
1385                 set_page_dirty(pages[i]);
1386                 unlock_page(pages[i]);
1387                 put_page(pages[i]);
1388         }
1389         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1390         extent_changeset_free(data_reserved);
1391         return i_done;
1392 out:
1393         for (i = 0; i < i_done; i++) {
1394                 unlock_page(pages[i]);
1395                 put_page(pages[i]);
1396         }
1397         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1398                         start_index << PAGE_SHIFT,
1399                         page_cnt << PAGE_SHIFT, true);
1400         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1401         extent_changeset_free(data_reserved);
1402         return ret;
1403
1404 }
1405
1406 int btrfs_defrag_file(struct inode *inode, struct file *file,
1407                       struct btrfs_ioctl_defrag_range_args *range,
1408                       u64 newer_than, unsigned long max_to_defrag)
1409 {
1410         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1411         struct btrfs_root *root = BTRFS_I(inode)->root;
1412         struct file_ra_state *ra = NULL;
1413         unsigned long last_index;
1414         u64 isize = i_size_read(inode);
1415         u64 last_len = 0;
1416         u64 skip = 0;
1417         u64 defrag_end = 0;
1418         u64 newer_off = range->start;
1419         unsigned long i;
1420         unsigned long ra_index = 0;
1421         int ret;
1422         int defrag_count = 0;
1423         int compress_type = BTRFS_COMPRESS_ZLIB;
1424         u32 extent_thresh = range->extent_thresh;
1425         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1426         unsigned long cluster = max_cluster;
1427         u64 new_align = ~((u64)SZ_128K - 1);
1428         struct page **pages = NULL;
1429         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1430
1431         if (isize == 0)
1432                 return 0;
1433
1434         if (range->start >= isize)
1435                 return -EINVAL;
1436
1437         if (do_compress) {
1438                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1439                         return -EINVAL;
1440                 if (range->compress_type)
1441                         compress_type = range->compress_type;
1442         }
1443
1444         if (extent_thresh == 0)
1445                 extent_thresh = SZ_256K;
1446
1447         /*
1448          * If we were not given a file, allocate a readahead context. As
1449          * readahead is just an optimization, defrag will work without it so
1450          * we don't error out.
1451          */
1452         if (!file) {
1453                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1454                 if (ra)
1455                         file_ra_state_init(ra, inode->i_mapping);
1456         } else {
1457                 ra = &file->f_ra;
1458         }
1459
1460         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1461         if (!pages) {
1462                 ret = -ENOMEM;
1463                 goto out_ra;
1464         }
1465
1466         /* find the last page to defrag */
1467         if (range->start + range->len > range->start) {
1468                 last_index = min_t(u64, isize - 1,
1469                          range->start + range->len - 1) >> PAGE_SHIFT;
1470         } else {
1471                 last_index = (isize - 1) >> PAGE_SHIFT;
1472         }
1473
1474         if (newer_than) {
1475                 ret = find_new_extents(root, inode, newer_than,
1476                                        &newer_off, SZ_64K);
1477                 if (!ret) {
1478                         range->start = newer_off;
1479                         /*
1480                          * we always align our defrag to help keep
1481                          * the extents in the file evenly spaced
1482                          */
1483                         i = (newer_off & new_align) >> PAGE_SHIFT;
1484                 } else
1485                         goto out_ra;
1486         } else {
1487                 i = range->start >> PAGE_SHIFT;
1488         }
1489         if (!max_to_defrag)
1490                 max_to_defrag = last_index - i + 1;
1491
1492         /*
1493          * make writeback starts from i, so the defrag range can be
1494          * written sequentially.
1495          */
1496         if (i < inode->i_mapping->writeback_index)
1497                 inode->i_mapping->writeback_index = i;
1498
1499         while (i <= last_index && defrag_count < max_to_defrag &&
1500                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1501                 /*
1502                  * make sure we stop running if someone unmounts
1503                  * the FS
1504                  */
1505                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1506                         break;
1507
1508                 if (btrfs_defrag_cancelled(fs_info)) {
1509                         btrfs_debug(fs_info, "defrag_file cancelled");
1510                         ret = -EAGAIN;
1511                         break;
1512                 }
1513
1514                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1515                                          extent_thresh, &last_len, &skip,
1516                                          &defrag_end, do_compress)){
1517                         unsigned long next;
1518                         /*
1519                          * the should_defrag function tells us how much to skip
1520                          * bump our counter by the suggested amount
1521                          */
1522                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1523                         i = max(i + 1, next);
1524                         continue;
1525                 }
1526
1527                 if (!newer_than) {
1528                         cluster = (PAGE_ALIGN(defrag_end) >>
1529                                    PAGE_SHIFT) - i;
1530                         cluster = min(cluster, max_cluster);
1531                 } else {
1532                         cluster = max_cluster;
1533                 }
1534
1535                 if (i + cluster > ra_index) {
1536                         ra_index = max(i, ra_index);
1537                         if (ra)
1538                                 page_cache_sync_readahead(inode->i_mapping, ra,
1539                                                 file, ra_index, cluster);
1540                         ra_index += cluster;
1541                 }
1542
1543                 inode_lock(inode);
1544                 if (IS_SWAPFILE(inode)) {
1545                         ret = -ETXTBSY;
1546                 } else {
1547                         if (do_compress)
1548                                 BTRFS_I(inode)->defrag_compress = compress_type;
1549                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1550                 }
1551                 if (ret < 0) {
1552                         inode_unlock(inode);
1553                         goto out_ra;
1554                 }
1555
1556                 defrag_count += ret;
1557                 balance_dirty_pages_ratelimited(inode->i_mapping);
1558                 inode_unlock(inode);
1559
1560                 if (newer_than) {
1561                         if (newer_off == (u64)-1)
1562                                 break;
1563
1564                         if (ret > 0)
1565                                 i += ret;
1566
1567                         newer_off = max(newer_off + 1,
1568                                         (u64)i << PAGE_SHIFT);
1569
1570                         ret = find_new_extents(root, inode, newer_than,
1571                                                &newer_off, SZ_64K);
1572                         if (!ret) {
1573                                 range->start = newer_off;
1574                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1575                         } else {
1576                                 break;
1577                         }
1578                 } else {
1579                         if (ret > 0) {
1580                                 i += ret;
1581                                 last_len += ret << PAGE_SHIFT;
1582                         } else {
1583                                 i++;
1584                                 last_len = 0;
1585                         }
1586                 }
1587         }
1588
1589         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1590                 filemap_flush(inode->i_mapping);
1591                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1592                              &BTRFS_I(inode)->runtime_flags))
1593                         filemap_flush(inode->i_mapping);
1594         }
1595
1596         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1597                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1598         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1599                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1600         }
1601
1602         ret = defrag_count;
1603
1604 out_ra:
1605         if (do_compress) {
1606                 inode_lock(inode);
1607                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1608                 inode_unlock(inode);
1609         }
1610         if (!file)
1611                 kfree(ra);
1612         kfree(pages);
1613         return ret;
1614 }
1615
1616 static noinline int btrfs_ioctl_resize(struct file *file,
1617                                         void __user *arg)
1618 {
1619         struct inode *inode = file_inode(file);
1620         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1621         u64 new_size;
1622         u64 old_size;
1623         u64 devid = 1;
1624         struct btrfs_root *root = BTRFS_I(inode)->root;
1625         struct btrfs_ioctl_vol_args *vol_args;
1626         struct btrfs_trans_handle *trans;
1627         struct btrfs_device *device = NULL;
1628         char *sizestr;
1629         char *retptr;
1630         char *devstr = NULL;
1631         int ret = 0;
1632         int mod = 0;
1633
1634         if (!capable(CAP_SYS_ADMIN))
1635                 return -EPERM;
1636
1637         ret = mnt_want_write_file(file);
1638         if (ret)
1639                 return ret;
1640
1641         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1642                 mnt_drop_write_file(file);
1643                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1644         }
1645
1646         vol_args = memdup_user(arg, sizeof(*vol_args));
1647         if (IS_ERR(vol_args)) {
1648                 ret = PTR_ERR(vol_args);
1649                 goto out;
1650         }
1651
1652         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1653
1654         sizestr = vol_args->name;
1655         devstr = strchr(sizestr, ':');
1656         if (devstr) {
1657                 sizestr = devstr + 1;
1658                 *devstr = '\0';
1659                 devstr = vol_args->name;
1660                 ret = kstrtoull(devstr, 10, &devid);
1661                 if (ret)
1662                         goto out_free;
1663                 if (!devid) {
1664                         ret = -EINVAL;
1665                         goto out_free;
1666                 }
1667                 btrfs_info(fs_info, "resizing devid %llu", devid);
1668         }
1669
1670         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1671         if (!device) {
1672                 btrfs_info(fs_info, "resizer unable to find device %llu",
1673                            devid);
1674                 ret = -ENODEV;
1675                 goto out_free;
1676         }
1677
1678         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1679                 btrfs_info(fs_info,
1680                            "resizer unable to apply on readonly device %llu",
1681                        devid);
1682                 ret = -EPERM;
1683                 goto out_free;
1684         }
1685
1686         if (!strcmp(sizestr, "max"))
1687                 new_size = device->bdev->bd_inode->i_size;
1688         else {
1689                 if (sizestr[0] == '-') {
1690                         mod = -1;
1691                         sizestr++;
1692                 } else if (sizestr[0] == '+') {
1693                         mod = 1;
1694                         sizestr++;
1695                 }
1696                 new_size = memparse(sizestr, &retptr);
1697                 if (*retptr != '\0' || new_size == 0) {
1698                         ret = -EINVAL;
1699                         goto out_free;
1700                 }
1701         }
1702
1703         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1704                 ret = -EPERM;
1705                 goto out_free;
1706         }
1707
1708         old_size = btrfs_device_get_total_bytes(device);
1709
1710         if (mod < 0) {
1711                 if (new_size > old_size) {
1712                         ret = -EINVAL;
1713                         goto out_free;
1714                 }
1715                 new_size = old_size - new_size;
1716         } else if (mod > 0) {
1717                 if (new_size > ULLONG_MAX - old_size) {
1718                         ret = -ERANGE;
1719                         goto out_free;
1720                 }
1721                 new_size = old_size + new_size;
1722         }
1723
1724         if (new_size < SZ_256M) {
1725                 ret = -EINVAL;
1726                 goto out_free;
1727         }
1728         if (new_size > device->bdev->bd_inode->i_size) {
1729                 ret = -EFBIG;
1730                 goto out_free;
1731         }
1732
1733         new_size = round_down(new_size, fs_info->sectorsize);
1734
1735         if (new_size > old_size) {
1736                 trans = btrfs_start_transaction(root, 0);
1737                 if (IS_ERR(trans)) {
1738                         ret = PTR_ERR(trans);
1739                         goto out_free;
1740                 }
1741                 ret = btrfs_grow_device(trans, device, new_size);
1742                 btrfs_commit_transaction(trans);
1743         } else if (new_size < old_size) {
1744                 ret = btrfs_shrink_device(device, new_size);
1745         } /* equal, nothing need to do */
1746
1747         if (ret == 0 && new_size != old_size)
1748                 btrfs_info_in_rcu(fs_info,
1749                         "resize device %s (devid %llu) from %llu to %llu",
1750                         rcu_str_deref(device->name), device->devid,
1751                         old_size, new_size);
1752 out_free:
1753         kfree(vol_args);
1754 out:
1755         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1756         mnt_drop_write_file(file);
1757         return ret;
1758 }
1759
1760 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1761                                 const char *name, unsigned long fd, int subvol,
1762                                 bool readonly,
1763                                 struct btrfs_qgroup_inherit *inherit)
1764 {
1765         int namelen;
1766         int ret = 0;
1767
1768         if (!S_ISDIR(file_inode(file)->i_mode))
1769                 return -ENOTDIR;
1770
1771         ret = mnt_want_write_file(file);
1772         if (ret)
1773                 goto out;
1774
1775         namelen = strlen(name);
1776         if (strchr(name, '/')) {
1777                 ret = -EINVAL;
1778                 goto out_drop_write;
1779         }
1780
1781         if (name[0] == '.' &&
1782            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1783                 ret = -EEXIST;
1784                 goto out_drop_write;
1785         }
1786
1787         if (subvol) {
1788                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1789                                      NULL, readonly, inherit);
1790         } else {
1791                 struct fd src = fdget(fd);
1792                 struct inode *src_inode;
1793                 if (!src.file) {
1794                         ret = -EINVAL;
1795                         goto out_drop_write;
1796                 }
1797
1798                 src_inode = file_inode(src.file);
1799                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1800                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1801                                    "Snapshot src from another FS");
1802                         ret = -EXDEV;
1803                 } else if (!inode_owner_or_capable(src_inode)) {
1804                         /*
1805                          * Subvolume creation is not restricted, but snapshots
1806                          * are limited to own subvolumes only
1807                          */
1808                         ret = -EPERM;
1809                 } else {
1810                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1811                                              BTRFS_I(src_inode)->root,
1812                                              readonly, inherit);
1813                 }
1814                 fdput(src);
1815         }
1816 out_drop_write:
1817         mnt_drop_write_file(file);
1818 out:
1819         return ret;
1820 }
1821
1822 static noinline int btrfs_ioctl_snap_create(struct file *file,
1823                                             void __user *arg, int subvol)
1824 {
1825         struct btrfs_ioctl_vol_args *vol_args;
1826         int ret;
1827
1828         if (!S_ISDIR(file_inode(file)->i_mode))
1829                 return -ENOTDIR;
1830
1831         vol_args = memdup_user(arg, sizeof(*vol_args));
1832         if (IS_ERR(vol_args))
1833                 return PTR_ERR(vol_args);
1834         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1835
1836         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1837                                         subvol, false, NULL);
1838
1839         kfree(vol_args);
1840         return ret;
1841 }
1842
1843 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1844                                                void __user *arg, int subvol)
1845 {
1846         struct btrfs_ioctl_vol_args_v2 *vol_args;
1847         int ret;
1848         bool readonly = false;
1849         struct btrfs_qgroup_inherit *inherit = NULL;
1850
1851         if (!S_ISDIR(file_inode(file)->i_mode))
1852                 return -ENOTDIR;
1853
1854         vol_args = memdup_user(arg, sizeof(*vol_args));
1855         if (IS_ERR(vol_args))
1856                 return PTR_ERR(vol_args);
1857         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1858
1859         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1860                 ret = -EOPNOTSUPP;
1861                 goto free_args;
1862         }
1863
1864         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1865                 readonly = true;
1866         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1867                 if (vol_args->size > PAGE_SIZE) {
1868                         ret = -EINVAL;
1869                         goto free_args;
1870                 }
1871                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1872                 if (IS_ERR(inherit)) {
1873                         ret = PTR_ERR(inherit);
1874                         goto free_args;
1875                 }
1876         }
1877
1878         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1879                                         subvol, readonly, inherit);
1880         if (ret)
1881                 goto free_inherit;
1882 free_inherit:
1883         kfree(inherit);
1884 free_args:
1885         kfree(vol_args);
1886         return ret;
1887 }
1888
1889 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1890                                                 void __user *arg)
1891 {
1892         struct inode *inode = file_inode(file);
1893         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1894         struct btrfs_root *root = BTRFS_I(inode)->root;
1895         int ret = 0;
1896         u64 flags = 0;
1897
1898         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1899                 return -EINVAL;
1900
1901         down_read(&fs_info->subvol_sem);
1902         if (btrfs_root_readonly(root))
1903                 flags |= BTRFS_SUBVOL_RDONLY;
1904         up_read(&fs_info->subvol_sem);
1905
1906         if (copy_to_user(arg, &flags, sizeof(flags)))
1907                 ret = -EFAULT;
1908
1909         return ret;
1910 }
1911
1912 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1913                                               void __user *arg)
1914 {
1915         struct inode *inode = file_inode(file);
1916         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1917         struct btrfs_root *root = BTRFS_I(inode)->root;
1918         struct btrfs_trans_handle *trans;
1919         u64 root_flags;
1920         u64 flags;
1921         int ret = 0;
1922
1923         if (!inode_owner_or_capable(inode))
1924                 return -EPERM;
1925
1926         ret = mnt_want_write_file(file);
1927         if (ret)
1928                 goto out;
1929
1930         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1931                 ret = -EINVAL;
1932                 goto out_drop_write;
1933         }
1934
1935         if (copy_from_user(&flags, arg, sizeof(flags))) {
1936                 ret = -EFAULT;
1937                 goto out_drop_write;
1938         }
1939
1940         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1941                 ret = -EOPNOTSUPP;
1942                 goto out_drop_write;
1943         }
1944
1945         down_write(&fs_info->subvol_sem);
1946
1947         /* nothing to do */
1948         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1949                 goto out_drop_sem;
1950
1951         root_flags = btrfs_root_flags(&root->root_item);
1952         if (flags & BTRFS_SUBVOL_RDONLY) {
1953                 btrfs_set_root_flags(&root->root_item,
1954                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1955         } else {
1956                 /*
1957                  * Block RO -> RW transition if this subvolume is involved in
1958                  * send
1959                  */
1960                 spin_lock(&root->root_item_lock);
1961                 if (root->send_in_progress == 0) {
1962                         btrfs_set_root_flags(&root->root_item,
1963                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1964                         spin_unlock(&root->root_item_lock);
1965                 } else {
1966                         spin_unlock(&root->root_item_lock);
1967                         btrfs_warn(fs_info,
1968                                    "Attempt to set subvolume %llu read-write during send",
1969                                    root->root_key.objectid);
1970                         ret = -EPERM;
1971                         goto out_drop_sem;
1972                 }
1973         }
1974
1975         trans = btrfs_start_transaction(root, 1);
1976         if (IS_ERR(trans)) {
1977                 ret = PTR_ERR(trans);
1978                 goto out_reset;
1979         }
1980
1981         ret = btrfs_update_root(trans, fs_info->tree_root,
1982                                 &root->root_key, &root->root_item);
1983         if (ret < 0) {
1984                 btrfs_end_transaction(trans);
1985                 goto out_reset;
1986         }
1987
1988         ret = btrfs_commit_transaction(trans);
1989
1990 out_reset:
1991         if (ret)
1992                 btrfs_set_root_flags(&root->root_item, root_flags);
1993 out_drop_sem:
1994         up_write(&fs_info->subvol_sem);
1995 out_drop_write:
1996         mnt_drop_write_file(file);
1997 out:
1998         return ret;
1999 }
2000
2001 static noinline int key_in_sk(struct btrfs_key *key,
2002                               struct btrfs_ioctl_search_key *sk)
2003 {
2004         struct btrfs_key test;
2005         int ret;
2006
2007         test.objectid = sk->min_objectid;
2008         test.type = sk->min_type;
2009         test.offset = sk->min_offset;
2010
2011         ret = btrfs_comp_cpu_keys(key, &test);
2012         if (ret < 0)
2013                 return 0;
2014
2015         test.objectid = sk->max_objectid;
2016         test.type = sk->max_type;
2017         test.offset = sk->max_offset;
2018
2019         ret = btrfs_comp_cpu_keys(key, &test);
2020         if (ret > 0)
2021                 return 0;
2022         return 1;
2023 }
2024
2025 static noinline int copy_to_sk(struct btrfs_path *path,
2026                                struct btrfs_key *key,
2027                                struct btrfs_ioctl_search_key *sk,
2028                                size_t *buf_size,
2029                                char __user *ubuf,
2030                                unsigned long *sk_offset,
2031                                int *num_found)
2032 {
2033         u64 found_transid;
2034         struct extent_buffer *leaf;
2035         struct btrfs_ioctl_search_header sh;
2036         struct btrfs_key test;
2037         unsigned long item_off;
2038         unsigned long item_len;
2039         int nritems;
2040         int i;
2041         int slot;
2042         int ret = 0;
2043
2044         leaf = path->nodes[0];
2045         slot = path->slots[0];
2046         nritems = btrfs_header_nritems(leaf);
2047
2048         if (btrfs_header_generation(leaf) > sk->max_transid) {
2049                 i = nritems;
2050                 goto advance_key;
2051         }
2052         found_transid = btrfs_header_generation(leaf);
2053
2054         for (i = slot; i < nritems; i++) {
2055                 item_off = btrfs_item_ptr_offset(leaf, i);
2056                 item_len = btrfs_item_size_nr(leaf, i);
2057
2058                 btrfs_item_key_to_cpu(leaf, key, i);
2059                 if (!key_in_sk(key, sk))
2060                         continue;
2061
2062                 if (sizeof(sh) + item_len > *buf_size) {
2063                         if (*num_found) {
2064                                 ret = 1;
2065                                 goto out;
2066                         }
2067
2068                         /*
2069                          * return one empty item back for v1, which does not
2070                          * handle -EOVERFLOW
2071                          */
2072
2073                         *buf_size = sizeof(sh) + item_len;
2074                         item_len = 0;
2075                         ret = -EOVERFLOW;
2076                 }
2077
2078                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2079                         ret = 1;
2080                         goto out;
2081                 }
2082
2083                 sh.objectid = key->objectid;
2084                 sh.offset = key->offset;
2085                 sh.type = key->type;
2086                 sh.len = item_len;
2087                 sh.transid = found_transid;
2088
2089                 /*
2090                  * Copy search result header. If we fault then loop again so we
2091                  * can fault in the pages and -EFAULT there if there's a
2092                  * problem. Otherwise we'll fault and then copy the buffer in
2093                  * properly this next time through
2094                  */
2095                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2096                         ret = 0;
2097                         goto out;
2098                 }
2099
2100                 *sk_offset += sizeof(sh);
2101
2102                 if (item_len) {
2103                         char __user *up = ubuf + *sk_offset;
2104                         /*
2105                          * Copy the item, same behavior as above, but reset the
2106                          * * sk_offset so we copy the full thing again.
2107                          */
2108                         if (read_extent_buffer_to_user_nofault(leaf, up,
2109                                                 item_off, item_len)) {
2110                                 ret = 0;
2111                                 *sk_offset -= sizeof(sh);
2112                                 goto out;
2113                         }
2114
2115                         *sk_offset += item_len;
2116                 }
2117                 (*num_found)++;
2118
2119                 if (ret) /* -EOVERFLOW from above */
2120                         goto out;
2121
2122                 if (*num_found >= sk->nr_items) {
2123                         ret = 1;
2124                         goto out;
2125                 }
2126         }
2127 advance_key:
2128         ret = 0;
2129         test.objectid = sk->max_objectid;
2130         test.type = sk->max_type;
2131         test.offset = sk->max_offset;
2132         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2133                 ret = 1;
2134         else if (key->offset < (u64)-1)
2135                 key->offset++;
2136         else if (key->type < (u8)-1) {
2137                 key->offset = 0;
2138                 key->type++;
2139         } else if (key->objectid < (u64)-1) {
2140                 key->offset = 0;
2141                 key->type = 0;
2142                 key->objectid++;
2143         } else
2144                 ret = 1;
2145 out:
2146         /*
2147          *  0: all items from this leaf copied, continue with next
2148          *  1: * more items can be copied, but unused buffer is too small
2149          *     * all items were found
2150          *     Either way, it will stops the loop which iterates to the next
2151          *     leaf
2152          *  -EOVERFLOW: item was to large for buffer
2153          *  -EFAULT: could not copy extent buffer back to userspace
2154          */
2155         return ret;
2156 }
2157
2158 static noinline int search_ioctl(struct inode *inode,
2159                                  struct btrfs_ioctl_search_key *sk,
2160                                  size_t *buf_size,
2161                                  char __user *ubuf)
2162 {
2163         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2164         struct btrfs_root *root;
2165         struct btrfs_key key;
2166         struct btrfs_path *path;
2167         int ret;
2168         int num_found = 0;
2169         unsigned long sk_offset = 0;
2170
2171         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2172                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2173                 return -EOVERFLOW;
2174         }
2175
2176         path = btrfs_alloc_path();
2177         if (!path)
2178                 return -ENOMEM;
2179
2180         if (sk->tree_id == 0) {
2181                 /* search the root of the inode that was passed */
2182                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2183         } else {
2184                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2185                 if (IS_ERR(root)) {
2186                         btrfs_free_path(path);
2187                         return PTR_ERR(root);
2188                 }
2189         }
2190
2191         key.objectid = sk->min_objectid;
2192         key.type = sk->min_type;
2193         key.offset = sk->min_offset;
2194
2195         while (1) {
2196                 ret = fault_in_pages_writeable(ubuf, *buf_size - sk_offset);
2197                 if (ret)
2198                         break;
2199
2200                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2201                 if (ret != 0) {
2202                         if (ret > 0)
2203                                 ret = 0;
2204                         goto err;
2205                 }
2206                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2207                                  &sk_offset, &num_found);
2208                 btrfs_release_path(path);
2209                 if (ret)
2210                         break;
2211
2212         }
2213         if (ret > 0)
2214                 ret = 0;
2215 err:
2216         sk->nr_items = num_found;
2217         btrfs_put_root(root);
2218         btrfs_free_path(path);
2219         return ret;
2220 }
2221
2222 static noinline int btrfs_ioctl_tree_search(struct file *file,
2223                                            void __user *argp)
2224 {
2225         struct btrfs_ioctl_search_args __user *uargs;
2226         struct btrfs_ioctl_search_key sk;
2227         struct inode *inode;
2228         int ret;
2229         size_t buf_size;
2230
2231         if (!capable(CAP_SYS_ADMIN))
2232                 return -EPERM;
2233
2234         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2235
2236         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2237                 return -EFAULT;
2238
2239         buf_size = sizeof(uargs->buf);
2240
2241         inode = file_inode(file);
2242         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2243
2244         /*
2245          * In the origin implementation an overflow is handled by returning a
2246          * search header with a len of zero, so reset ret.
2247          */
2248         if (ret == -EOVERFLOW)
2249                 ret = 0;
2250
2251         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2252                 ret = -EFAULT;
2253         return ret;
2254 }
2255
2256 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2257                                                void __user *argp)
2258 {
2259         struct btrfs_ioctl_search_args_v2 __user *uarg;
2260         struct btrfs_ioctl_search_args_v2 args;
2261         struct inode *inode;
2262         int ret;
2263         size_t buf_size;
2264         const size_t buf_limit = SZ_16M;
2265
2266         if (!capable(CAP_SYS_ADMIN))
2267                 return -EPERM;
2268
2269         /* copy search header and buffer size */
2270         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2271         if (copy_from_user(&args, uarg, sizeof(args)))
2272                 return -EFAULT;
2273
2274         buf_size = args.buf_size;
2275
2276         /* limit result size to 16MB */
2277         if (buf_size > buf_limit)
2278                 buf_size = buf_limit;
2279
2280         inode = file_inode(file);
2281         ret = search_ioctl(inode, &args.key, &buf_size,
2282                            (char __user *)(&uarg->buf[0]));
2283         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2284                 ret = -EFAULT;
2285         else if (ret == -EOVERFLOW &&
2286                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2287                 ret = -EFAULT;
2288
2289         return ret;
2290 }
2291
2292 /*
2293  * Search INODE_REFs to identify path name of 'dirid' directory
2294  * in a 'tree_id' tree. and sets path name to 'name'.
2295  */
2296 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2297                                 u64 tree_id, u64 dirid, char *name)
2298 {
2299         struct btrfs_root *root;
2300         struct btrfs_key key;
2301         char *ptr;
2302         int ret = -1;
2303         int slot;
2304         int len;
2305         int total_len = 0;
2306         struct btrfs_inode_ref *iref;
2307         struct extent_buffer *l;
2308         struct btrfs_path *path;
2309
2310         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2311                 name[0]='\0';
2312                 return 0;
2313         }
2314
2315         path = btrfs_alloc_path();
2316         if (!path)
2317                 return -ENOMEM;
2318
2319         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2320
2321         root = btrfs_get_fs_root(info, tree_id, true);
2322         if (IS_ERR(root)) {
2323                 ret = PTR_ERR(root);
2324                 root = NULL;
2325                 goto out;
2326         }
2327
2328         key.objectid = dirid;
2329         key.type = BTRFS_INODE_REF_KEY;
2330         key.offset = (u64)-1;
2331
2332         while (1) {
2333                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2334                 if (ret < 0)
2335                         goto out;
2336                 else if (ret > 0) {
2337                         ret = btrfs_previous_item(root, path, dirid,
2338                                                   BTRFS_INODE_REF_KEY);
2339                         if (ret < 0)
2340                                 goto out;
2341                         else if (ret > 0) {
2342                                 ret = -ENOENT;
2343                                 goto out;
2344                         }
2345                 }
2346
2347                 l = path->nodes[0];
2348                 slot = path->slots[0];
2349                 btrfs_item_key_to_cpu(l, &key, slot);
2350
2351                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2352                 len = btrfs_inode_ref_name_len(l, iref);
2353                 ptr -= len + 1;
2354                 total_len += len + 1;
2355                 if (ptr < name) {
2356                         ret = -ENAMETOOLONG;
2357                         goto out;
2358                 }
2359
2360                 *(ptr + len) = '/';
2361                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2362
2363                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2364                         break;
2365
2366                 btrfs_release_path(path);
2367                 key.objectid = key.offset;
2368                 key.offset = (u64)-1;
2369                 dirid = key.objectid;
2370         }
2371         memmove(name, ptr, total_len);
2372         name[total_len] = '\0';
2373         ret = 0;
2374 out:
2375         btrfs_put_root(root);
2376         btrfs_free_path(path);
2377         return ret;
2378 }
2379
2380 static int btrfs_search_path_in_tree_user(struct inode *inode,
2381                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2382 {
2383         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2384         struct super_block *sb = inode->i_sb;
2385         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2386         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2387         u64 dirid = args->dirid;
2388         unsigned long item_off;
2389         unsigned long item_len;
2390         struct btrfs_inode_ref *iref;
2391         struct btrfs_root_ref *rref;
2392         struct btrfs_root *root = NULL;
2393         struct btrfs_path *path;
2394         struct btrfs_key key, key2;
2395         struct extent_buffer *leaf;
2396         struct inode *temp_inode;
2397         char *ptr;
2398         int slot;
2399         int len;
2400         int total_len = 0;
2401         int ret;
2402
2403         path = btrfs_alloc_path();
2404         if (!path)
2405                 return -ENOMEM;
2406
2407         /*
2408          * If the bottom subvolume does not exist directly under upper_limit,
2409          * construct the path in from the bottom up.
2410          */
2411         if (dirid != upper_limit.objectid) {
2412                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2413
2414                 root = btrfs_get_fs_root(fs_info, treeid, true);
2415                 if (IS_ERR(root)) {
2416                         ret = PTR_ERR(root);
2417                         goto out;
2418                 }
2419
2420                 key.objectid = dirid;
2421                 key.type = BTRFS_INODE_REF_KEY;
2422                 key.offset = (u64)-1;
2423                 while (1) {
2424                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2425                         if (ret < 0) {
2426                                 goto out_put;
2427                         } else if (ret > 0) {
2428                                 ret = btrfs_previous_item(root, path, dirid,
2429                                                           BTRFS_INODE_REF_KEY);
2430                                 if (ret < 0) {
2431                                         goto out_put;
2432                                 } else if (ret > 0) {
2433                                         ret = -ENOENT;
2434                                         goto out_put;
2435                                 }
2436                         }
2437
2438                         leaf = path->nodes[0];
2439                         slot = path->slots[0];
2440                         btrfs_item_key_to_cpu(leaf, &key, slot);
2441
2442                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2443                         len = btrfs_inode_ref_name_len(leaf, iref);
2444                         ptr -= len + 1;
2445                         total_len += len + 1;
2446                         if (ptr < args->path) {
2447                                 ret = -ENAMETOOLONG;
2448                                 goto out_put;
2449                         }
2450
2451                         *(ptr + len) = '/';
2452                         read_extent_buffer(leaf, ptr,
2453                                         (unsigned long)(iref + 1), len);
2454
2455                         /* Check the read+exec permission of this directory */
2456                         ret = btrfs_previous_item(root, path, dirid,
2457                                                   BTRFS_INODE_ITEM_KEY);
2458                         if (ret < 0) {
2459                                 goto out_put;
2460                         } else if (ret > 0) {
2461                                 ret = -ENOENT;
2462                                 goto out_put;
2463                         }
2464
2465                         leaf = path->nodes[0];
2466                         slot = path->slots[0];
2467                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2468                         if (key2.objectid != dirid) {
2469                                 ret = -ENOENT;
2470                                 goto out_put;
2471                         }
2472
2473                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2474                         if (IS_ERR(temp_inode)) {
2475                                 ret = PTR_ERR(temp_inode);
2476                                 goto out_put;
2477                         }
2478                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2479                         iput(temp_inode);
2480                         if (ret) {
2481                                 ret = -EACCES;
2482                                 goto out_put;
2483                         }
2484
2485                         if (key.offset == upper_limit.objectid)
2486                                 break;
2487                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2488                                 ret = -EACCES;
2489                                 goto out_put;
2490                         }
2491
2492                         btrfs_release_path(path);
2493                         key.objectid = key.offset;
2494                         key.offset = (u64)-1;
2495                         dirid = key.objectid;
2496                 }
2497
2498                 memmove(args->path, ptr, total_len);
2499                 args->path[total_len] = '\0';
2500                 btrfs_put_root(root);
2501                 root = NULL;
2502                 btrfs_release_path(path);
2503         }
2504
2505         /* Get the bottom subvolume's name from ROOT_REF */
2506         key.objectid = treeid;
2507         key.type = BTRFS_ROOT_REF_KEY;
2508         key.offset = args->treeid;
2509         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2510         if (ret < 0) {
2511                 goto out;
2512         } else if (ret > 0) {
2513                 ret = -ENOENT;
2514                 goto out;
2515         }
2516
2517         leaf = path->nodes[0];
2518         slot = path->slots[0];
2519         btrfs_item_key_to_cpu(leaf, &key, slot);
2520
2521         item_off = btrfs_item_ptr_offset(leaf, slot);
2522         item_len = btrfs_item_size_nr(leaf, slot);
2523         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2524         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2525         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2526                 ret = -EINVAL;
2527                 goto out;
2528         }
2529
2530         /* Copy subvolume's name */
2531         item_off += sizeof(struct btrfs_root_ref);
2532         item_len -= sizeof(struct btrfs_root_ref);
2533         read_extent_buffer(leaf, args->name, item_off, item_len);
2534         args->name[item_len] = 0;
2535
2536 out_put:
2537         btrfs_put_root(root);
2538 out:
2539         btrfs_free_path(path);
2540         return ret;
2541 }
2542
2543 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2544                                            void __user *argp)
2545 {
2546         struct btrfs_ioctl_ino_lookup_args *args;
2547         struct inode *inode;
2548         int ret = 0;
2549
2550         args = memdup_user(argp, sizeof(*args));
2551         if (IS_ERR(args))
2552                 return PTR_ERR(args);
2553
2554         inode = file_inode(file);
2555
2556         /*
2557          * Unprivileged query to obtain the containing subvolume root id. The
2558          * path is reset so it's consistent with btrfs_search_path_in_tree.
2559          */
2560         if (args->treeid == 0)
2561                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2562
2563         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2564                 args->name[0] = 0;
2565                 goto out;
2566         }
2567
2568         if (!capable(CAP_SYS_ADMIN)) {
2569                 ret = -EPERM;
2570                 goto out;
2571         }
2572
2573         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2574                                         args->treeid, args->objectid,
2575                                         args->name);
2576
2577 out:
2578         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2579                 ret = -EFAULT;
2580
2581         kfree(args);
2582         return ret;
2583 }
2584
2585 /*
2586  * Version of ino_lookup ioctl (unprivileged)
2587  *
2588  * The main differences from ino_lookup ioctl are:
2589  *
2590  *   1. Read + Exec permission will be checked using inode_permission() during
2591  *      path construction. -EACCES will be returned in case of failure.
2592  *   2. Path construction will be stopped at the inode number which corresponds
2593  *      to the fd with which this ioctl is called. If constructed path does not
2594  *      exist under fd's inode, -EACCES will be returned.
2595  *   3. The name of bottom subvolume is also searched and filled.
2596  */
2597 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2598 {
2599         struct btrfs_ioctl_ino_lookup_user_args *args;
2600         struct inode *inode;
2601         int ret;
2602
2603         args = memdup_user(argp, sizeof(*args));
2604         if (IS_ERR(args))
2605                 return PTR_ERR(args);
2606
2607         inode = file_inode(file);
2608
2609         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2610             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2611                 /*
2612                  * The subvolume does not exist under fd with which this is
2613                  * called
2614                  */
2615                 kfree(args);
2616                 return -EACCES;
2617         }
2618
2619         ret = btrfs_search_path_in_tree_user(inode, args);
2620
2621         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2622                 ret = -EFAULT;
2623
2624         kfree(args);
2625         return ret;
2626 }
2627
2628 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2629 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2630 {
2631         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2632         struct btrfs_fs_info *fs_info;
2633         struct btrfs_root *root;
2634         struct btrfs_path *path;
2635         struct btrfs_key key;
2636         struct btrfs_root_item *root_item;
2637         struct btrfs_root_ref *rref;
2638         struct extent_buffer *leaf;
2639         unsigned long item_off;
2640         unsigned long item_len;
2641         struct inode *inode;
2642         int slot;
2643         int ret = 0;
2644
2645         path = btrfs_alloc_path();
2646         if (!path)
2647                 return -ENOMEM;
2648
2649         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2650         if (!subvol_info) {
2651                 btrfs_free_path(path);
2652                 return -ENOMEM;
2653         }
2654
2655         inode = file_inode(file);
2656         fs_info = BTRFS_I(inode)->root->fs_info;
2657
2658         /* Get root_item of inode's subvolume */
2659         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2660         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2661         if (IS_ERR(root)) {
2662                 ret = PTR_ERR(root);
2663                 goto out_free;
2664         }
2665         root_item = &root->root_item;
2666
2667         subvol_info->treeid = key.objectid;
2668
2669         subvol_info->generation = btrfs_root_generation(root_item);
2670         subvol_info->flags = btrfs_root_flags(root_item);
2671
2672         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2673         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2674                                                     BTRFS_UUID_SIZE);
2675         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2676                                                     BTRFS_UUID_SIZE);
2677
2678         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2679         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2680         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2681
2682         subvol_info->otransid = btrfs_root_otransid(root_item);
2683         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2684         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2685
2686         subvol_info->stransid = btrfs_root_stransid(root_item);
2687         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2688         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2689
2690         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2691         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2692         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2693
2694         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2695                 /* Search root tree for ROOT_BACKREF of this subvolume */
2696                 key.type = BTRFS_ROOT_BACKREF_KEY;
2697                 key.offset = 0;
2698                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2699                 if (ret < 0) {
2700                         goto out;
2701                 } else if (path->slots[0] >=
2702                            btrfs_header_nritems(path->nodes[0])) {
2703                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2704                         if (ret < 0) {
2705                                 goto out;
2706                         } else if (ret > 0) {
2707                                 ret = -EUCLEAN;
2708                                 goto out;
2709                         }
2710                 }
2711
2712                 leaf = path->nodes[0];
2713                 slot = path->slots[0];
2714                 btrfs_item_key_to_cpu(leaf, &key, slot);
2715                 if (key.objectid == subvol_info->treeid &&
2716                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2717                         subvol_info->parent_id = key.offset;
2718
2719                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2720                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2721
2722                         item_off = btrfs_item_ptr_offset(leaf, slot)
2723                                         + sizeof(struct btrfs_root_ref);
2724                         item_len = btrfs_item_size_nr(leaf, slot)
2725                                         - sizeof(struct btrfs_root_ref);
2726                         read_extent_buffer(leaf, subvol_info->name,
2727                                            item_off, item_len);
2728                 } else {
2729                         ret = -ENOENT;
2730                         goto out;
2731                 }
2732         }
2733
2734         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2735                 ret = -EFAULT;
2736
2737 out:
2738         btrfs_put_root(root);
2739 out_free:
2740         btrfs_free_path(path);
2741         kfree(subvol_info);
2742         return ret;
2743 }
2744
2745 /*
2746  * Return ROOT_REF information of the subvolume containing this inode
2747  * except the subvolume name.
2748  */
2749 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2750 {
2751         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2752         struct btrfs_root_ref *rref;
2753         struct btrfs_root *root;
2754         struct btrfs_path *path;
2755         struct btrfs_key key;
2756         struct extent_buffer *leaf;
2757         struct inode *inode;
2758         u64 objectid;
2759         int slot;
2760         int ret;
2761         u8 found;
2762
2763         path = btrfs_alloc_path();
2764         if (!path)
2765                 return -ENOMEM;
2766
2767         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2768         if (IS_ERR(rootrefs)) {
2769                 btrfs_free_path(path);
2770                 return PTR_ERR(rootrefs);
2771         }
2772
2773         inode = file_inode(file);
2774         root = BTRFS_I(inode)->root->fs_info->tree_root;
2775         objectid = BTRFS_I(inode)->root->root_key.objectid;
2776
2777         key.objectid = objectid;
2778         key.type = BTRFS_ROOT_REF_KEY;
2779         key.offset = rootrefs->min_treeid;
2780         found = 0;
2781
2782         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2783         if (ret < 0) {
2784                 goto out;
2785         } else if (path->slots[0] >=
2786                    btrfs_header_nritems(path->nodes[0])) {
2787                 ret = btrfs_next_leaf(root, path);
2788                 if (ret < 0) {
2789                         goto out;
2790                 } else if (ret > 0) {
2791                         ret = -EUCLEAN;
2792                         goto out;
2793                 }
2794         }
2795         while (1) {
2796                 leaf = path->nodes[0];
2797                 slot = path->slots[0];
2798
2799                 btrfs_item_key_to_cpu(leaf, &key, slot);
2800                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2801                         ret = 0;
2802                         goto out;
2803                 }
2804
2805                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2806                         ret = -EOVERFLOW;
2807                         goto out;
2808                 }
2809
2810                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2811                 rootrefs->rootref[found].treeid = key.offset;
2812                 rootrefs->rootref[found].dirid =
2813                                   btrfs_root_ref_dirid(leaf, rref);
2814                 found++;
2815
2816                 ret = btrfs_next_item(root, path);
2817                 if (ret < 0) {
2818                         goto out;
2819                 } else if (ret > 0) {
2820                         ret = -EUCLEAN;
2821                         goto out;
2822                 }
2823         }
2824
2825 out:
2826         if (!ret || ret == -EOVERFLOW) {
2827                 rootrefs->num_items = found;
2828                 /* update min_treeid for next search */
2829                 if (found)
2830                         rootrefs->min_treeid =
2831                                 rootrefs->rootref[found - 1].treeid + 1;
2832                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2833                         ret = -EFAULT;
2834         }
2835
2836         kfree(rootrefs);
2837         btrfs_free_path(path);
2838
2839         return ret;
2840 }
2841
2842 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2843                                              void __user *arg,
2844                                              bool destroy_v2)
2845 {
2846         struct dentry *parent = file->f_path.dentry;
2847         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2848         struct dentry *dentry;
2849         struct inode *dir = d_inode(parent);
2850         struct inode *inode;
2851         struct btrfs_root *root = BTRFS_I(dir)->root;
2852         struct btrfs_root *dest = NULL;
2853         struct btrfs_ioctl_vol_args *vol_args = NULL;
2854         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2855         char *subvol_name, *subvol_name_ptr = NULL;
2856         int subvol_namelen;
2857         int err = 0;
2858         bool destroy_parent = false;
2859
2860         if (destroy_v2) {
2861                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2862                 if (IS_ERR(vol_args2))
2863                         return PTR_ERR(vol_args2);
2864
2865                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2866                         err = -EOPNOTSUPP;
2867                         goto out;
2868                 }
2869
2870                 /*
2871                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2872                  * name, same as v1 currently does.
2873                  */
2874                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2875                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2876                         subvol_name = vol_args2->name;
2877
2878                         err = mnt_want_write_file(file);
2879                         if (err)
2880                                 goto out;
2881                 } else {
2882                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2883                                 err = -EINVAL;
2884                                 goto out;
2885                         }
2886
2887                         err = mnt_want_write_file(file);
2888                         if (err)
2889                                 goto out;
2890
2891                         dentry = btrfs_get_dentry(fs_info->sb,
2892                                         BTRFS_FIRST_FREE_OBJECTID,
2893                                         vol_args2->subvolid, 0, 0);
2894                         if (IS_ERR(dentry)) {
2895                                 err = PTR_ERR(dentry);
2896                                 goto out_drop_write;
2897                         }
2898
2899                         /*
2900                          * Change the default parent since the subvolume being
2901                          * deleted can be outside of the current mount point.
2902                          */
2903                         parent = btrfs_get_parent(dentry);
2904
2905                         /*
2906                          * At this point dentry->d_name can point to '/' if the
2907                          * subvolume we want to destroy is outsite of the
2908                          * current mount point, so we need to release the
2909                          * current dentry and execute the lookup to return a new
2910                          * one with ->d_name pointing to the
2911                          * <mount point>/subvol_name.
2912                          */
2913                         dput(dentry);
2914                         if (IS_ERR(parent)) {
2915                                 err = PTR_ERR(parent);
2916                                 goto out_drop_write;
2917                         }
2918                         dir = d_inode(parent);
2919
2920                         /*
2921                          * If v2 was used with SPEC_BY_ID, a new parent was
2922                          * allocated since the subvolume can be outside of the
2923                          * current mount point. Later on we need to release this
2924                          * new parent dentry.
2925                          */
2926                         destroy_parent = true;
2927
2928                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2929                                                 fs_info, vol_args2->subvolid);
2930                         if (IS_ERR(subvol_name_ptr)) {
2931                                 err = PTR_ERR(subvol_name_ptr);
2932                                 goto free_parent;
2933                         }
2934                         /* subvol_name_ptr is already NULL termined */
2935                         subvol_name = (char *)kbasename(subvol_name_ptr);
2936                 }
2937         } else {
2938                 vol_args = memdup_user(arg, sizeof(*vol_args));
2939                 if (IS_ERR(vol_args))
2940                         return PTR_ERR(vol_args);
2941
2942                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2943                 subvol_name = vol_args->name;
2944
2945                 err = mnt_want_write_file(file);
2946                 if (err)
2947                         goto out;
2948         }
2949
2950         subvol_namelen = strlen(subvol_name);
2951
2952         if (strchr(subvol_name, '/') ||
2953             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2954                 err = -EINVAL;
2955                 goto free_subvol_name;
2956         }
2957
2958         if (!S_ISDIR(dir->i_mode)) {
2959                 err = -ENOTDIR;
2960                 goto free_subvol_name;
2961         }
2962
2963         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2964         if (err == -EINTR)
2965                 goto free_subvol_name;
2966         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
2967         if (IS_ERR(dentry)) {
2968                 err = PTR_ERR(dentry);
2969                 goto out_unlock_dir;
2970         }
2971
2972         if (d_really_is_negative(dentry)) {
2973                 err = -ENOENT;
2974                 goto out_dput;
2975         }
2976
2977         inode = d_inode(dentry);
2978         dest = BTRFS_I(inode)->root;
2979         if (!capable(CAP_SYS_ADMIN)) {
2980                 /*
2981                  * Regular user.  Only allow this with a special mount
2982                  * option, when the user has write+exec access to the
2983                  * subvol root, and when rmdir(2) would have been
2984                  * allowed.
2985                  *
2986                  * Note that this is _not_ check that the subvol is
2987                  * empty or doesn't contain data that we wouldn't
2988                  * otherwise be able to delete.
2989                  *
2990                  * Users who want to delete empty subvols should try
2991                  * rmdir(2).
2992                  */
2993                 err = -EPERM;
2994                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2995                         goto out_dput;
2996
2997                 /*
2998                  * Do not allow deletion if the parent dir is the same
2999                  * as the dir to be deleted.  That means the ioctl
3000                  * must be called on the dentry referencing the root
3001                  * of the subvol, not a random directory contained
3002                  * within it.
3003                  */
3004                 err = -EINVAL;
3005                 if (root == dest)
3006                         goto out_dput;
3007
3008                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
3009                 if (err)
3010                         goto out_dput;
3011         }
3012
3013         /* check if subvolume may be deleted by a user */
3014         err = btrfs_may_delete(dir, dentry, 1);
3015         if (err)
3016                 goto out_dput;
3017
3018         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3019                 err = -EINVAL;
3020                 goto out_dput;
3021         }
3022
3023         inode_lock(inode);
3024         err = btrfs_delete_subvolume(dir, dentry);
3025         inode_unlock(inode);
3026         if (!err) {
3027                 fsnotify_rmdir(dir, dentry);
3028                 d_delete(dentry);
3029         }
3030
3031 out_dput:
3032         dput(dentry);
3033 out_unlock_dir:
3034         inode_unlock(dir);
3035 free_subvol_name:
3036         kfree(subvol_name_ptr);
3037 free_parent:
3038         if (destroy_parent)
3039                 dput(parent);
3040 out_drop_write:
3041         mnt_drop_write_file(file);
3042 out:
3043         kfree(vol_args2);
3044         kfree(vol_args);
3045         return err;
3046 }
3047
3048 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3049 {
3050         struct inode *inode = file_inode(file);
3051         struct btrfs_root *root = BTRFS_I(inode)->root;
3052         struct btrfs_ioctl_defrag_range_args *range;
3053         int ret;
3054
3055         ret = mnt_want_write_file(file);
3056         if (ret)
3057                 return ret;
3058
3059         if (btrfs_root_readonly(root)) {
3060                 ret = -EROFS;
3061                 goto out;
3062         }
3063
3064         switch (inode->i_mode & S_IFMT) {
3065         case S_IFDIR:
3066                 if (!capable(CAP_SYS_ADMIN)) {
3067                         ret = -EPERM;
3068                         goto out;
3069                 }
3070                 ret = btrfs_defrag_root(root);
3071                 break;
3072         case S_IFREG:
3073                 /*
3074                  * Note that this does not check the file descriptor for write
3075                  * access. This prevents defragmenting executables that are
3076                  * running and allows defrag on files open in read-only mode.
3077                  */
3078                 if (!capable(CAP_SYS_ADMIN) &&
3079                     inode_permission(inode, MAY_WRITE)) {
3080                         ret = -EPERM;
3081                         goto out;
3082                 }
3083
3084                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3085                 if (!range) {
3086                         ret = -ENOMEM;
3087                         goto out;
3088                 }
3089
3090                 if (argp) {
3091                         if (copy_from_user(range, argp,
3092                                            sizeof(*range))) {
3093                                 ret = -EFAULT;
3094                                 kfree(range);
3095                                 goto out;
3096                         }
3097                         /* compression requires us to start the IO */
3098                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3099                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3100                                 range->extent_thresh = (u32)-1;
3101                         }
3102                 } else {
3103                         /* the rest are all set to zero by kzalloc */
3104                         range->len = (u64)-1;
3105                 }
3106                 ret = btrfs_defrag_file(file_inode(file), file,
3107                                         range, BTRFS_OLDEST_GENERATION, 0);
3108                 if (ret > 0)
3109                         ret = 0;
3110                 kfree(range);
3111                 break;
3112         default:
3113                 ret = -EINVAL;
3114         }
3115 out:
3116         mnt_drop_write_file(file);
3117         return ret;
3118 }
3119
3120 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3121 {
3122         struct btrfs_ioctl_vol_args *vol_args;
3123         int ret;
3124
3125         if (!capable(CAP_SYS_ADMIN))
3126                 return -EPERM;
3127
3128         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3129                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3130
3131         vol_args = memdup_user(arg, sizeof(*vol_args));
3132         if (IS_ERR(vol_args)) {
3133                 ret = PTR_ERR(vol_args);
3134                 goto out;
3135         }
3136
3137         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3138         ret = btrfs_init_new_device(fs_info, vol_args->name);
3139
3140         if (!ret)
3141                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3142
3143         kfree(vol_args);
3144 out:
3145         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3146         return ret;
3147 }
3148
3149 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3150 {
3151         struct inode *inode = file_inode(file);
3152         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3153         struct btrfs_ioctl_vol_args_v2 *vol_args;
3154         int ret;
3155
3156         if (!capable(CAP_SYS_ADMIN))
3157                 return -EPERM;
3158
3159         ret = mnt_want_write_file(file);
3160         if (ret)
3161                 return ret;
3162
3163         vol_args = memdup_user(arg, sizeof(*vol_args));
3164         if (IS_ERR(vol_args)) {
3165                 ret = PTR_ERR(vol_args);
3166                 goto err_drop;
3167         }
3168
3169         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3170                 ret = -EOPNOTSUPP;
3171                 goto out;
3172         }
3173
3174         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3175                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3176                 goto out;
3177         }
3178
3179         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3180                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3181         } else {
3182                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3183                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3184         }
3185         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3186
3187         if (!ret) {
3188                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3189                         btrfs_info(fs_info, "device deleted: id %llu",
3190                                         vol_args->devid);
3191                 else
3192                         btrfs_info(fs_info, "device deleted: %s",
3193                                         vol_args->name);
3194         }
3195 out:
3196         kfree(vol_args);
3197 err_drop:
3198         mnt_drop_write_file(file);
3199         return ret;
3200 }
3201
3202 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3203 {
3204         struct inode *inode = file_inode(file);
3205         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3206         struct btrfs_ioctl_vol_args *vol_args;
3207         int ret;
3208
3209         if (!capable(CAP_SYS_ADMIN))
3210                 return -EPERM;
3211
3212         ret = mnt_want_write_file(file);
3213         if (ret)
3214                 return ret;
3215
3216         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3217                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3218                 goto out_drop_write;
3219         }
3220
3221         vol_args = memdup_user(arg, sizeof(*vol_args));
3222         if (IS_ERR(vol_args)) {
3223                 ret = PTR_ERR(vol_args);
3224                 goto out;
3225         }
3226
3227         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3228         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3229
3230         if (!ret)
3231                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3232         kfree(vol_args);
3233 out:
3234         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3235 out_drop_write:
3236         mnt_drop_write_file(file);
3237
3238         return ret;
3239 }
3240
3241 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3242                                 void __user *arg)
3243 {
3244         struct btrfs_ioctl_fs_info_args *fi_args;
3245         struct btrfs_device *device;
3246         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3247         u64 flags_in;
3248         int ret = 0;
3249
3250         fi_args = memdup_user(arg, sizeof(*fi_args));
3251         if (IS_ERR(fi_args))
3252                 return PTR_ERR(fi_args);
3253
3254         flags_in = fi_args->flags;
3255         memset(fi_args, 0, sizeof(*fi_args));
3256
3257         rcu_read_lock();
3258         fi_args->num_devices = fs_devices->num_devices;
3259
3260         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3261                 if (device->devid > fi_args->max_id)
3262                         fi_args->max_id = device->devid;
3263         }
3264         rcu_read_unlock();
3265
3266         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3267         fi_args->nodesize = fs_info->nodesize;
3268         fi_args->sectorsize = fs_info->sectorsize;
3269         fi_args->clone_alignment = fs_info->sectorsize;
3270
3271         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3272                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3273                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3274                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3275         }
3276
3277         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3278                 fi_args->generation = fs_info->generation;
3279                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3280         }
3281
3282         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3283                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3284                        sizeof(fi_args->metadata_uuid));
3285                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3286         }
3287
3288         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3289                 ret = -EFAULT;
3290
3291         kfree(fi_args);
3292         return ret;
3293 }
3294
3295 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3296                                  void __user *arg)
3297 {
3298         struct btrfs_ioctl_dev_info_args *di_args;
3299         struct btrfs_device *dev;
3300         int ret = 0;
3301         char *s_uuid = NULL;
3302
3303         di_args = memdup_user(arg, sizeof(*di_args));
3304         if (IS_ERR(di_args))
3305                 return PTR_ERR(di_args);
3306
3307         if (!btrfs_is_empty_uuid(di_args->uuid))
3308                 s_uuid = di_args->uuid;
3309
3310         rcu_read_lock();
3311         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3312                                 NULL, true);
3313
3314         if (!dev) {
3315                 ret = -ENODEV;
3316                 goto out;
3317         }
3318
3319         di_args->devid = dev->devid;
3320         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3321         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3322         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3323         if (dev->name) {
3324                 strncpy(di_args->path, rcu_str_deref(dev->name),
3325                                 sizeof(di_args->path) - 1);
3326                 di_args->path[sizeof(di_args->path) - 1] = 0;
3327         } else {
3328                 di_args->path[0] = '\0';
3329         }
3330
3331 out:
3332         rcu_read_unlock();
3333         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3334                 ret = -EFAULT;
3335
3336         kfree(di_args);
3337         return ret;
3338 }
3339
3340 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3341 {
3342         struct inode *inode = file_inode(file);
3343         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3344         struct btrfs_root *root = BTRFS_I(inode)->root;
3345         struct btrfs_root *new_root;
3346         struct btrfs_dir_item *di;
3347         struct btrfs_trans_handle *trans;
3348         struct btrfs_path *path = NULL;
3349         struct btrfs_disk_key disk_key;
3350         u64 objectid = 0;
3351         u64 dir_id;
3352         int ret;
3353
3354         if (!capable(CAP_SYS_ADMIN))
3355                 return -EPERM;
3356
3357         ret = mnt_want_write_file(file);
3358         if (ret)
3359                 return ret;
3360
3361         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3362                 ret = -EFAULT;
3363                 goto out;
3364         }
3365
3366         if (!objectid)
3367                 objectid = BTRFS_FS_TREE_OBJECTID;
3368
3369         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3370         if (IS_ERR(new_root)) {
3371                 ret = PTR_ERR(new_root);
3372                 goto out;
3373         }
3374         if (!is_fstree(new_root->root_key.objectid)) {
3375                 ret = -ENOENT;
3376                 goto out_free;
3377         }
3378
3379         path = btrfs_alloc_path();
3380         if (!path) {
3381                 ret = -ENOMEM;
3382                 goto out_free;
3383         }
3384         path->leave_spinning = 1;
3385
3386         trans = btrfs_start_transaction(root, 1);
3387         if (IS_ERR(trans)) {
3388                 ret = PTR_ERR(trans);
3389                 goto out_free;
3390         }
3391
3392         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3393         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3394                                    dir_id, "default", 7, 1);
3395         if (IS_ERR_OR_NULL(di)) {
3396                 btrfs_release_path(path);
3397                 btrfs_end_transaction(trans);
3398                 btrfs_err(fs_info,
3399                           "Umm, you don't have the default diritem, this isn't going to work");
3400                 ret = -ENOENT;
3401                 goto out_free;
3402         }
3403
3404         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3405         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3406         btrfs_mark_buffer_dirty(path->nodes[0]);
3407         btrfs_release_path(path);
3408
3409         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3410         btrfs_end_transaction(trans);
3411 out_free:
3412         btrfs_put_root(new_root);
3413         btrfs_free_path(path);
3414 out:
3415         mnt_drop_write_file(file);
3416         return ret;
3417 }
3418
3419 static void get_block_group_info(struct list_head *groups_list,
3420                                  struct btrfs_ioctl_space_info *space)
3421 {
3422         struct btrfs_block_group *block_group;
3423
3424         space->total_bytes = 0;
3425         space->used_bytes = 0;
3426         space->flags = 0;
3427         list_for_each_entry(block_group, groups_list, list) {
3428                 space->flags = block_group->flags;
3429                 space->total_bytes += block_group->length;
3430                 space->used_bytes += block_group->used;
3431         }
3432 }
3433
3434 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3435                                    void __user *arg)
3436 {
3437         struct btrfs_ioctl_space_args space_args;
3438         struct btrfs_ioctl_space_info space;
3439         struct btrfs_ioctl_space_info *dest;
3440         struct btrfs_ioctl_space_info *dest_orig;
3441         struct btrfs_ioctl_space_info __user *user_dest;
3442         struct btrfs_space_info *info;
3443         static const u64 types[] = {
3444                 BTRFS_BLOCK_GROUP_DATA,
3445                 BTRFS_BLOCK_GROUP_SYSTEM,
3446                 BTRFS_BLOCK_GROUP_METADATA,
3447                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3448         };
3449         int num_types = 4;
3450         int alloc_size;
3451         int ret = 0;
3452         u64 slot_count = 0;
3453         int i, c;
3454
3455         if (copy_from_user(&space_args,
3456                            (struct btrfs_ioctl_space_args __user *)arg,
3457                            sizeof(space_args)))
3458                 return -EFAULT;
3459
3460         for (i = 0; i < num_types; i++) {
3461                 struct btrfs_space_info *tmp;
3462
3463                 info = NULL;
3464                 rcu_read_lock();
3465                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
3466                                         list) {
3467                         if (tmp->flags == types[i]) {
3468                                 info = tmp;
3469                                 break;
3470                         }
3471                 }
3472                 rcu_read_unlock();
3473
3474                 if (!info)
3475                         continue;
3476
3477                 down_read(&info->groups_sem);
3478                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3479                         if (!list_empty(&info->block_groups[c]))
3480                                 slot_count++;
3481                 }
3482                 up_read(&info->groups_sem);
3483         }
3484
3485         /*
3486          * Global block reserve, exported as a space_info
3487          */
3488         slot_count++;
3489
3490         /* space_slots == 0 means they are asking for a count */
3491         if (space_args.space_slots == 0) {
3492                 space_args.total_spaces = slot_count;
3493                 goto out;
3494         }
3495
3496         slot_count = min_t(u64, space_args.space_slots, slot_count);
3497
3498         alloc_size = sizeof(*dest) * slot_count;
3499
3500         /* we generally have at most 6 or so space infos, one for each raid
3501          * level.  So, a whole page should be more than enough for everyone
3502          */
3503         if (alloc_size > PAGE_SIZE)
3504                 return -ENOMEM;
3505
3506         space_args.total_spaces = 0;
3507         dest = kmalloc(alloc_size, GFP_KERNEL);
3508         if (!dest)
3509                 return -ENOMEM;
3510         dest_orig = dest;
3511
3512         /* now we have a buffer to copy into */
3513         for (i = 0; i < num_types; i++) {
3514                 struct btrfs_space_info *tmp;
3515
3516                 if (!slot_count)
3517                         break;
3518
3519                 info = NULL;
3520                 rcu_read_lock();
3521                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
3522                                         list) {
3523                         if (tmp->flags == types[i]) {
3524                                 info = tmp;
3525                                 break;
3526                         }
3527                 }
3528                 rcu_read_unlock();
3529
3530                 if (!info)
3531                         continue;
3532                 down_read(&info->groups_sem);
3533                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3534                         if (!list_empty(&info->block_groups[c])) {
3535                                 get_block_group_info(&info->block_groups[c],
3536                                                      &space);
3537                                 memcpy(dest, &space, sizeof(space));
3538                                 dest++;
3539                                 space_args.total_spaces++;
3540                                 slot_count--;
3541                         }
3542                         if (!slot_count)
3543                                 break;
3544                 }
3545                 up_read(&info->groups_sem);
3546         }
3547
3548         /*
3549          * Add global block reserve
3550          */
3551         if (slot_count) {
3552                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3553
3554                 spin_lock(&block_rsv->lock);
3555                 space.total_bytes = block_rsv->size;
3556                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3557                 spin_unlock(&block_rsv->lock);
3558                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3559                 memcpy(dest, &space, sizeof(space));
3560                 space_args.total_spaces++;
3561         }
3562
3563         user_dest = (struct btrfs_ioctl_space_info __user *)
3564                 (arg + sizeof(struct btrfs_ioctl_space_args));
3565
3566         if (copy_to_user(user_dest, dest_orig, alloc_size))
3567                 ret = -EFAULT;
3568
3569         kfree(dest_orig);
3570 out:
3571         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3572                 ret = -EFAULT;
3573
3574         return ret;
3575 }
3576
3577 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3578                                             void __user *argp)
3579 {
3580         struct btrfs_trans_handle *trans;
3581         u64 transid;
3582         int ret;
3583
3584         trans = btrfs_attach_transaction_barrier(root);
3585         if (IS_ERR(trans)) {
3586                 if (PTR_ERR(trans) != -ENOENT)
3587                         return PTR_ERR(trans);
3588
3589                 /* No running transaction, don't bother */
3590                 transid = root->fs_info->last_trans_committed;
3591                 goto out;
3592         }
3593         transid = trans->transid;
3594         ret = btrfs_commit_transaction_async(trans, 0);
3595         if (ret) {
3596                 btrfs_end_transaction(trans);
3597                 return ret;
3598         }
3599 out:
3600         if (argp)
3601                 if (copy_to_user(argp, &transid, sizeof(transid)))
3602                         return -EFAULT;
3603         return 0;
3604 }
3605
3606 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3607                                            void __user *argp)
3608 {
3609         u64 transid;
3610
3611         if (argp) {
3612                 if (copy_from_user(&transid, argp, sizeof(transid)))
3613                         return -EFAULT;
3614         } else {
3615                 transid = 0;  /* current trans */
3616         }
3617         return btrfs_wait_for_commit(fs_info, transid);
3618 }
3619
3620 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3621 {
3622         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3623         struct btrfs_ioctl_scrub_args *sa;
3624         int ret;
3625
3626         if (!capable(CAP_SYS_ADMIN))
3627                 return -EPERM;
3628
3629         sa = memdup_user(arg, sizeof(*sa));
3630         if (IS_ERR(sa))
3631                 return PTR_ERR(sa);
3632
3633         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3634                 ret = mnt_want_write_file(file);
3635                 if (ret)
3636                         goto out;
3637         }
3638
3639         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3640                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3641                               0);
3642
3643         /*
3644          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3645          * error. This is important as it allows user space to know how much
3646          * progress scrub has done. For example, if scrub is canceled we get
3647          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3648          * space. Later user space can inspect the progress from the structure
3649          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3650          * previously (btrfs-progs does this).
3651          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3652          * then return -EFAULT to signal the structure was not copied or it may
3653          * be corrupt and unreliable due to a partial copy.
3654          */
3655         if (copy_to_user(arg, sa, sizeof(*sa)))
3656                 ret = -EFAULT;
3657
3658         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3659                 mnt_drop_write_file(file);
3660 out:
3661         kfree(sa);
3662         return ret;
3663 }
3664
3665 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3666 {
3667         if (!capable(CAP_SYS_ADMIN))
3668                 return -EPERM;
3669
3670         return btrfs_scrub_cancel(fs_info);
3671 }
3672
3673 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3674                                        void __user *arg)
3675 {
3676         struct btrfs_ioctl_scrub_args *sa;
3677         int ret;
3678
3679         if (!capable(CAP_SYS_ADMIN))
3680                 return -EPERM;
3681
3682         sa = memdup_user(arg, sizeof(*sa));
3683         if (IS_ERR(sa))
3684                 return PTR_ERR(sa);
3685
3686         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3687
3688         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3689                 ret = -EFAULT;
3690
3691         kfree(sa);
3692         return ret;
3693 }
3694
3695 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3696                                       void __user *arg)
3697 {
3698         struct btrfs_ioctl_get_dev_stats *sa;
3699         int ret;
3700
3701         sa = memdup_user(arg, sizeof(*sa));
3702         if (IS_ERR(sa))
3703                 return PTR_ERR(sa);
3704
3705         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3706                 kfree(sa);
3707                 return -EPERM;
3708         }
3709
3710         ret = btrfs_get_dev_stats(fs_info, sa);
3711
3712         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3713                 ret = -EFAULT;
3714
3715         kfree(sa);
3716         return ret;
3717 }
3718
3719 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3720                                     void __user *arg)
3721 {
3722         struct btrfs_ioctl_dev_replace_args *p;
3723         int ret;
3724
3725         if (!capable(CAP_SYS_ADMIN))
3726                 return -EPERM;
3727
3728         p = memdup_user(arg, sizeof(*p));
3729         if (IS_ERR(p))
3730                 return PTR_ERR(p);
3731
3732         switch (p->cmd) {
3733         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3734                 if (sb_rdonly(fs_info->sb)) {
3735                         ret = -EROFS;
3736                         goto out;
3737                 }
3738                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3739                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3740                 } else {
3741                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3742                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3743                 }
3744                 break;
3745         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3746                 btrfs_dev_replace_status(fs_info, p);
3747                 ret = 0;
3748                 break;
3749         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3750                 p->result = btrfs_dev_replace_cancel(fs_info);
3751                 ret = 0;
3752                 break;
3753         default:
3754                 ret = -EINVAL;
3755                 break;
3756         }
3757
3758         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3759                 ret = -EFAULT;
3760 out:
3761         kfree(p);
3762         return ret;
3763 }
3764
3765 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3766 {
3767         int ret = 0;
3768         int i;
3769         u64 rel_ptr;
3770         int size;
3771         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3772         struct inode_fs_paths *ipath = NULL;
3773         struct btrfs_path *path;
3774
3775         if (!capable(CAP_DAC_READ_SEARCH))
3776                 return -EPERM;
3777
3778         path = btrfs_alloc_path();
3779         if (!path) {
3780                 ret = -ENOMEM;
3781                 goto out;
3782         }
3783
3784         ipa = memdup_user(arg, sizeof(*ipa));
3785         if (IS_ERR(ipa)) {
3786                 ret = PTR_ERR(ipa);
3787                 ipa = NULL;
3788                 goto out;
3789         }
3790
3791         size = min_t(u32, ipa->size, 4096);
3792         ipath = init_ipath(size, root, path);
3793         if (IS_ERR(ipath)) {
3794                 ret = PTR_ERR(ipath);
3795                 ipath = NULL;
3796                 goto out;
3797         }
3798
3799         ret = paths_from_inode(ipa->inum, ipath);
3800         if (ret < 0)
3801                 goto out;
3802
3803         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3804                 rel_ptr = ipath->fspath->val[i] -
3805                           (u64)(unsigned long)ipath->fspath->val;
3806                 ipath->fspath->val[i] = rel_ptr;
3807         }
3808
3809         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3810                            ipath->fspath, size);
3811         if (ret) {
3812                 ret = -EFAULT;
3813                 goto out;
3814         }
3815
3816 out:
3817         btrfs_free_path(path);
3818         free_ipath(ipath);
3819         kfree(ipa);
3820
3821         return ret;
3822 }
3823
3824 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3825 {
3826         struct btrfs_data_container *inodes = ctx;
3827         const size_t c = 3 * sizeof(u64);
3828
3829         if (inodes->bytes_left >= c) {
3830                 inodes->bytes_left -= c;
3831                 inodes->val[inodes->elem_cnt] = inum;
3832                 inodes->val[inodes->elem_cnt + 1] = offset;
3833                 inodes->val[inodes->elem_cnt + 2] = root;
3834                 inodes->elem_cnt += 3;
3835         } else {
3836                 inodes->bytes_missing += c - inodes->bytes_left;
3837                 inodes->bytes_left = 0;
3838                 inodes->elem_missed += 3;
3839         }
3840
3841         return 0;
3842 }
3843
3844 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3845                                         void __user *arg, int version)
3846 {
3847         int ret = 0;
3848         int size;
3849         struct btrfs_ioctl_logical_ino_args *loi;
3850         struct btrfs_data_container *inodes = NULL;
3851         struct btrfs_path *path = NULL;
3852         bool ignore_offset;
3853
3854         if (!capable(CAP_SYS_ADMIN))
3855                 return -EPERM;
3856
3857         loi = memdup_user(arg, sizeof(*loi));
3858         if (IS_ERR(loi))
3859                 return PTR_ERR(loi);
3860
3861         if (version == 1) {
3862                 ignore_offset = false;
3863                 size = min_t(u32, loi->size, SZ_64K);
3864         } else {
3865                 /* All reserved bits must be 0 for now */
3866                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3867                         ret = -EINVAL;
3868                         goto out_loi;
3869                 }
3870                 /* Only accept flags we have defined so far */
3871                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3872                         ret = -EINVAL;
3873                         goto out_loi;
3874                 }
3875                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3876                 size = min_t(u32, loi->size, SZ_16M);
3877         }
3878
3879         path = btrfs_alloc_path();
3880         if (!path) {
3881                 ret = -ENOMEM;
3882                 goto out;
3883         }
3884
3885         inodes = init_data_container(size);
3886         if (IS_ERR(inodes)) {
3887                 ret = PTR_ERR(inodes);
3888                 inodes = NULL;
3889                 goto out;
3890         }
3891
3892         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3893                                           build_ino_list, inodes, ignore_offset);
3894         if (ret == -EINVAL)
3895                 ret = -ENOENT;
3896         if (ret < 0)
3897                 goto out;
3898
3899         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3900                            size);
3901         if (ret)
3902                 ret = -EFAULT;
3903
3904 out:
3905         btrfs_free_path(path);
3906         kvfree(inodes);
3907 out_loi:
3908         kfree(loi);
3909
3910         return ret;
3911 }
3912
3913 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3914                                struct btrfs_ioctl_balance_args *bargs)
3915 {
3916         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3917
3918         bargs->flags = bctl->flags;
3919
3920         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3921                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3922         if (atomic_read(&fs_info->balance_pause_req))
3923                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3924         if (atomic_read(&fs_info->balance_cancel_req))
3925                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3926
3927         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3928         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3929         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3930
3931         spin_lock(&fs_info->balance_lock);
3932         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3933         spin_unlock(&fs_info->balance_lock);
3934 }
3935
3936 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3937 {
3938         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3939         struct btrfs_fs_info *fs_info = root->fs_info;
3940         struct btrfs_ioctl_balance_args *bargs;
3941         struct btrfs_balance_control *bctl;
3942         bool need_unlock; /* for mut. excl. ops lock */
3943         int ret;
3944
3945         if (!capable(CAP_SYS_ADMIN))
3946                 return -EPERM;
3947
3948         ret = mnt_want_write_file(file);
3949         if (ret)
3950                 return ret;
3951
3952 again:
3953         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3954                 mutex_lock(&fs_info->balance_mutex);
3955                 need_unlock = true;
3956                 goto locked;
3957         }
3958
3959         /*
3960          * mut. excl. ops lock is locked.  Three possibilities:
3961          *   (1) some other op is running
3962          *   (2) balance is running
3963          *   (3) balance is paused -- special case (think resume)
3964          */
3965         mutex_lock(&fs_info->balance_mutex);
3966         if (fs_info->balance_ctl) {
3967                 /* this is either (2) or (3) */
3968                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3969                         mutex_unlock(&fs_info->balance_mutex);
3970                         /*
3971                          * Lock released to allow other waiters to continue,
3972                          * we'll reexamine the status again.
3973                          */
3974                         mutex_lock(&fs_info->balance_mutex);
3975
3976                         if (fs_info->balance_ctl &&
3977                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3978                                 /* this is (3) */
3979                                 need_unlock = false;
3980                                 goto locked;
3981                         }
3982
3983                         mutex_unlock(&fs_info->balance_mutex);
3984                         goto again;
3985                 } else {
3986                         /* this is (2) */
3987                         mutex_unlock(&fs_info->balance_mutex);
3988                         ret = -EINPROGRESS;
3989                         goto out;
3990                 }
3991         } else {
3992                 /* this is (1) */
3993                 mutex_unlock(&fs_info->balance_mutex);
3994                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3995                 goto out;
3996         }
3997
3998 locked:
3999         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4000
4001         if (arg) {
4002                 bargs = memdup_user(arg, sizeof(*bargs));
4003                 if (IS_ERR(bargs)) {
4004                         ret = PTR_ERR(bargs);
4005                         goto out_unlock;
4006                 }
4007
4008                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4009                         if (!fs_info->balance_ctl) {
4010                                 ret = -ENOTCONN;
4011                                 goto out_bargs;
4012                         }
4013
4014                         bctl = fs_info->balance_ctl;
4015                         spin_lock(&fs_info->balance_lock);
4016                         bctl->flags |= BTRFS_BALANCE_RESUME;
4017                         spin_unlock(&fs_info->balance_lock);
4018
4019                         goto do_balance;
4020                 }
4021         } else {
4022                 bargs = NULL;
4023         }
4024
4025         if (fs_info->balance_ctl) {
4026                 ret = -EINPROGRESS;
4027                 goto out_bargs;
4028         }
4029
4030         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4031         if (!bctl) {
4032                 ret = -ENOMEM;
4033                 goto out_bargs;
4034         }
4035
4036         if (arg) {
4037                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4038                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4039                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4040
4041                 bctl->flags = bargs->flags;
4042         } else {
4043                 /* balance everything - no filters */
4044                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4045         }
4046
4047         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4048                 ret = -EINVAL;
4049                 goto out_bctl;
4050         }
4051
4052 do_balance:
4053         /*
4054          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4055          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4056          * restriper was paused all the way until unmount, in free_fs_info.
4057          * The flag should be cleared after reset_balance_state.
4058          */
4059         need_unlock = false;
4060
4061         ret = btrfs_balance(fs_info, bctl, bargs);
4062         bctl = NULL;
4063
4064         if ((ret == 0 || ret == -ECANCELED) && arg) {
4065                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4066                         ret = -EFAULT;
4067         }
4068
4069 out_bctl:
4070         kfree(bctl);
4071 out_bargs:
4072         kfree(bargs);
4073 out_unlock:
4074         mutex_unlock(&fs_info->balance_mutex);
4075         if (need_unlock)
4076                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4077 out:
4078         mnt_drop_write_file(file);
4079         return ret;
4080 }
4081
4082 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4083 {
4084         if (!capable(CAP_SYS_ADMIN))
4085                 return -EPERM;
4086
4087         switch (cmd) {
4088         case BTRFS_BALANCE_CTL_PAUSE:
4089                 return btrfs_pause_balance(fs_info);
4090         case BTRFS_BALANCE_CTL_CANCEL:
4091                 return btrfs_cancel_balance(fs_info);
4092         }
4093
4094         return -EINVAL;
4095 }
4096
4097 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4098                                          void __user *arg)
4099 {
4100         struct btrfs_ioctl_balance_args *bargs;
4101         int ret = 0;
4102
4103         if (!capable(CAP_SYS_ADMIN))
4104                 return -EPERM;
4105
4106         mutex_lock(&fs_info->balance_mutex);
4107         if (!fs_info->balance_ctl) {
4108                 ret = -ENOTCONN;
4109                 goto out;
4110         }
4111
4112         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4113         if (!bargs) {
4114                 ret = -ENOMEM;
4115                 goto out;
4116         }
4117
4118         btrfs_update_ioctl_balance_args(fs_info, bargs);
4119
4120         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4121                 ret = -EFAULT;
4122
4123         kfree(bargs);
4124 out:
4125         mutex_unlock(&fs_info->balance_mutex);
4126         return ret;
4127 }
4128
4129 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4130 {
4131         struct inode *inode = file_inode(file);
4132         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4133         struct btrfs_ioctl_quota_ctl_args *sa;
4134         int ret;
4135
4136         if (!capable(CAP_SYS_ADMIN))
4137                 return -EPERM;
4138
4139         ret = mnt_want_write_file(file);
4140         if (ret)
4141                 return ret;
4142
4143         sa = memdup_user(arg, sizeof(*sa));
4144         if (IS_ERR(sa)) {
4145                 ret = PTR_ERR(sa);
4146                 goto drop_write;
4147         }
4148
4149         down_write(&fs_info->subvol_sem);
4150
4151         switch (sa->cmd) {
4152         case BTRFS_QUOTA_CTL_ENABLE:
4153                 ret = btrfs_quota_enable(fs_info);
4154                 break;
4155         case BTRFS_QUOTA_CTL_DISABLE:
4156                 ret = btrfs_quota_disable(fs_info);
4157                 break;
4158         default:
4159                 ret = -EINVAL;
4160                 break;
4161         }
4162
4163         kfree(sa);
4164         up_write(&fs_info->subvol_sem);
4165 drop_write:
4166         mnt_drop_write_file(file);
4167         return ret;
4168 }
4169
4170 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4171 {
4172         struct inode *inode = file_inode(file);
4173         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4174         struct btrfs_root *root = BTRFS_I(inode)->root;
4175         struct btrfs_ioctl_qgroup_assign_args *sa;
4176         struct btrfs_trans_handle *trans;
4177         int ret;
4178         int err;
4179
4180         if (!capable(CAP_SYS_ADMIN))
4181                 return -EPERM;
4182
4183         ret = mnt_want_write_file(file);
4184         if (ret)
4185                 return ret;
4186
4187         sa = memdup_user(arg, sizeof(*sa));
4188         if (IS_ERR(sa)) {
4189                 ret = PTR_ERR(sa);
4190                 goto drop_write;
4191         }
4192
4193         trans = btrfs_join_transaction(root);
4194         if (IS_ERR(trans)) {
4195                 ret = PTR_ERR(trans);
4196                 goto out;
4197         }
4198
4199         if (sa->assign) {
4200                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4201         } else {
4202                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4203         }
4204
4205         /* update qgroup status and info */
4206         err = btrfs_run_qgroups(trans);
4207         if (err < 0)
4208                 btrfs_handle_fs_error(fs_info, err,
4209                                       "failed to update qgroup status and info");
4210         err = btrfs_end_transaction(trans);
4211         if (err && !ret)
4212                 ret = err;
4213
4214 out:
4215         kfree(sa);
4216 drop_write:
4217         mnt_drop_write_file(file);
4218         return ret;
4219 }
4220
4221 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4222 {
4223         struct inode *inode = file_inode(file);
4224         struct btrfs_root *root = BTRFS_I(inode)->root;
4225         struct btrfs_ioctl_qgroup_create_args *sa;
4226         struct btrfs_trans_handle *trans;
4227         int ret;
4228         int err;
4229
4230         if (!capable(CAP_SYS_ADMIN))
4231                 return -EPERM;
4232
4233         ret = mnt_want_write_file(file);
4234         if (ret)
4235                 return ret;
4236
4237         sa = memdup_user(arg, sizeof(*sa));
4238         if (IS_ERR(sa)) {
4239                 ret = PTR_ERR(sa);
4240                 goto drop_write;
4241         }
4242
4243         if (!sa->qgroupid) {
4244                 ret = -EINVAL;
4245                 goto out;
4246         }
4247
4248         trans = btrfs_join_transaction(root);
4249         if (IS_ERR(trans)) {
4250                 ret = PTR_ERR(trans);
4251                 goto out;
4252         }
4253
4254         if (sa->create) {
4255                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4256         } else {
4257                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4258         }
4259
4260         err = btrfs_end_transaction(trans);
4261         if (err && !ret)
4262                 ret = err;
4263
4264 out:
4265         kfree(sa);
4266 drop_write:
4267         mnt_drop_write_file(file);
4268         return ret;
4269 }
4270
4271 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4272 {
4273         struct inode *inode = file_inode(file);
4274         struct btrfs_root *root = BTRFS_I(inode)->root;
4275         struct btrfs_ioctl_qgroup_limit_args *sa;
4276         struct btrfs_trans_handle *trans;
4277         int ret;
4278         int err;
4279         u64 qgroupid;
4280
4281         if (!capable(CAP_SYS_ADMIN))
4282                 return -EPERM;
4283
4284         ret = mnt_want_write_file(file);
4285         if (ret)
4286                 return ret;
4287
4288         sa = memdup_user(arg, sizeof(*sa));
4289         if (IS_ERR(sa)) {
4290                 ret = PTR_ERR(sa);
4291                 goto drop_write;
4292         }
4293
4294         trans = btrfs_join_transaction(root);
4295         if (IS_ERR(trans)) {
4296                 ret = PTR_ERR(trans);
4297                 goto out;
4298         }
4299
4300         qgroupid = sa->qgroupid;
4301         if (!qgroupid) {
4302                 /* take the current subvol as qgroup */
4303                 qgroupid = root->root_key.objectid;
4304         }
4305
4306         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4307
4308         err = btrfs_end_transaction(trans);
4309         if (err && !ret)
4310                 ret = err;
4311
4312 out:
4313         kfree(sa);
4314 drop_write:
4315         mnt_drop_write_file(file);
4316         return ret;
4317 }
4318
4319 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4320 {
4321         struct inode *inode = file_inode(file);
4322         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4323         struct btrfs_ioctl_quota_rescan_args *qsa;
4324         int ret;
4325
4326         if (!capable(CAP_SYS_ADMIN))
4327                 return -EPERM;
4328
4329         ret = mnt_want_write_file(file);
4330         if (ret)
4331                 return ret;
4332
4333         qsa = memdup_user(arg, sizeof(*qsa));
4334         if (IS_ERR(qsa)) {
4335                 ret = PTR_ERR(qsa);
4336                 goto drop_write;
4337         }
4338
4339         if (qsa->flags) {
4340                 ret = -EINVAL;
4341                 goto out;
4342         }
4343
4344         ret = btrfs_qgroup_rescan(fs_info);
4345
4346 out:
4347         kfree(qsa);
4348 drop_write:
4349         mnt_drop_write_file(file);
4350         return ret;
4351 }
4352
4353 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4354                                                 void __user *arg)
4355 {
4356         struct btrfs_ioctl_quota_rescan_args *qsa;
4357         int ret = 0;
4358
4359         if (!capable(CAP_SYS_ADMIN))
4360                 return -EPERM;
4361
4362         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4363         if (!qsa)
4364                 return -ENOMEM;
4365
4366         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4367                 qsa->flags = 1;
4368                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4369         }
4370
4371         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4372                 ret = -EFAULT;
4373
4374         kfree(qsa);
4375         return ret;
4376 }
4377
4378 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4379                                                 void __user *arg)
4380 {
4381         if (!capable(CAP_SYS_ADMIN))
4382                 return -EPERM;
4383
4384         return btrfs_qgroup_wait_for_completion(fs_info, true);
4385 }
4386
4387 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4388                                             struct btrfs_ioctl_received_subvol_args *sa)
4389 {
4390         struct inode *inode = file_inode(file);
4391         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4392         struct btrfs_root *root = BTRFS_I(inode)->root;
4393         struct btrfs_root_item *root_item = &root->root_item;
4394         struct btrfs_trans_handle *trans;
4395         struct timespec64 ct = current_time(inode);
4396         int ret = 0;
4397         int received_uuid_changed;
4398
4399         if (!inode_owner_or_capable(inode))
4400                 return -EPERM;
4401
4402         ret = mnt_want_write_file(file);
4403         if (ret < 0)
4404                 return ret;
4405
4406         down_write(&fs_info->subvol_sem);
4407
4408         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4409                 ret = -EINVAL;
4410                 goto out;
4411         }
4412
4413         if (btrfs_root_readonly(root)) {
4414                 ret = -EROFS;
4415                 goto out;
4416         }
4417
4418         /*
4419          * 1 - root item
4420          * 2 - uuid items (received uuid + subvol uuid)
4421          */
4422         trans = btrfs_start_transaction(root, 3);
4423         if (IS_ERR(trans)) {
4424                 ret = PTR_ERR(trans);
4425                 trans = NULL;
4426                 goto out;
4427         }
4428
4429         sa->rtransid = trans->transid;
4430         sa->rtime.sec = ct.tv_sec;
4431         sa->rtime.nsec = ct.tv_nsec;
4432
4433         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4434                                        BTRFS_UUID_SIZE);
4435         if (received_uuid_changed &&
4436             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4437                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4438                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4439                                           root->root_key.objectid);
4440                 if (ret && ret != -ENOENT) {
4441                         btrfs_abort_transaction(trans, ret);
4442                         btrfs_end_transaction(trans);
4443                         goto out;
4444                 }
4445         }
4446         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4447         btrfs_set_root_stransid(root_item, sa->stransid);
4448         btrfs_set_root_rtransid(root_item, sa->rtransid);
4449         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4450         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4451         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4452         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4453
4454         ret = btrfs_update_root(trans, fs_info->tree_root,
4455                                 &root->root_key, &root->root_item);
4456         if (ret < 0) {
4457                 btrfs_end_transaction(trans);
4458                 goto out;
4459         }
4460         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4461                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4462                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4463                                           root->root_key.objectid);
4464                 if (ret < 0 && ret != -EEXIST) {
4465                         btrfs_abort_transaction(trans, ret);
4466                         btrfs_end_transaction(trans);
4467                         goto out;
4468                 }
4469         }
4470         ret = btrfs_commit_transaction(trans);
4471 out:
4472         up_write(&fs_info->subvol_sem);
4473         mnt_drop_write_file(file);
4474         return ret;
4475 }
4476
4477 #ifdef CONFIG_64BIT
4478 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4479                                                 void __user *arg)
4480 {
4481         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4482         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4483         int ret = 0;
4484
4485         args32 = memdup_user(arg, sizeof(*args32));
4486         if (IS_ERR(args32))
4487                 return PTR_ERR(args32);
4488
4489         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4490         if (!args64) {
4491                 ret = -ENOMEM;
4492                 goto out;
4493         }
4494
4495         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4496         args64->stransid = args32->stransid;
4497         args64->rtransid = args32->rtransid;
4498         args64->stime.sec = args32->stime.sec;
4499         args64->stime.nsec = args32->stime.nsec;
4500         args64->rtime.sec = args32->rtime.sec;
4501         args64->rtime.nsec = args32->rtime.nsec;
4502         args64->flags = args32->flags;
4503
4504         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4505         if (ret)
4506                 goto out;
4507
4508         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4509         args32->stransid = args64->stransid;
4510         args32->rtransid = args64->rtransid;
4511         args32->stime.sec = args64->stime.sec;
4512         args32->stime.nsec = args64->stime.nsec;
4513         args32->rtime.sec = args64->rtime.sec;
4514         args32->rtime.nsec = args64->rtime.nsec;
4515         args32->flags = args64->flags;
4516
4517         ret = copy_to_user(arg, args32, sizeof(*args32));
4518         if (ret)
4519                 ret = -EFAULT;
4520
4521 out:
4522         kfree(args32);
4523         kfree(args64);
4524         return ret;
4525 }
4526 #endif
4527
4528 static long btrfs_ioctl_set_received_subvol(struct file *file,
4529                                             void __user *arg)
4530 {
4531         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4532         int ret = 0;
4533
4534         sa = memdup_user(arg, sizeof(*sa));
4535         if (IS_ERR(sa))
4536                 return PTR_ERR(sa);
4537
4538         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4539
4540         if (ret)
4541                 goto out;
4542
4543         ret = copy_to_user(arg, sa, sizeof(*sa));
4544         if (ret)
4545                 ret = -EFAULT;
4546
4547 out:
4548         kfree(sa);
4549         return ret;
4550 }
4551
4552 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4553                                         void __user *arg)
4554 {
4555         size_t len;
4556         int ret;
4557         char label[BTRFS_LABEL_SIZE];
4558
4559         spin_lock(&fs_info->super_lock);
4560         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4561         spin_unlock(&fs_info->super_lock);
4562
4563         len = strnlen(label, BTRFS_LABEL_SIZE);
4564
4565         if (len == BTRFS_LABEL_SIZE) {
4566                 btrfs_warn(fs_info,
4567                            "label is too long, return the first %zu bytes",
4568                            --len);
4569         }
4570
4571         ret = copy_to_user(arg, label, len);
4572
4573         return ret ? -EFAULT : 0;
4574 }
4575
4576 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4577 {
4578         struct inode *inode = file_inode(file);
4579         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4580         struct btrfs_root *root = BTRFS_I(inode)->root;
4581         struct btrfs_super_block *super_block = fs_info->super_copy;
4582         struct btrfs_trans_handle *trans;
4583         char label[BTRFS_LABEL_SIZE];
4584         int ret;
4585
4586         if (!capable(CAP_SYS_ADMIN))
4587                 return -EPERM;
4588
4589         if (copy_from_user(label, arg, sizeof(label)))
4590                 return -EFAULT;
4591
4592         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4593                 btrfs_err(fs_info,
4594                           "unable to set label with more than %d bytes",
4595                           BTRFS_LABEL_SIZE - 1);
4596                 return -EINVAL;
4597         }
4598
4599         ret = mnt_want_write_file(file);
4600         if (ret)
4601                 return ret;
4602
4603         trans = btrfs_start_transaction(root, 0);
4604         if (IS_ERR(trans)) {
4605                 ret = PTR_ERR(trans);
4606                 goto out_unlock;
4607         }
4608
4609         spin_lock(&fs_info->super_lock);
4610         strcpy(super_block->label, label);
4611         spin_unlock(&fs_info->super_lock);
4612         ret = btrfs_commit_transaction(trans);
4613
4614 out_unlock:
4615         mnt_drop_write_file(file);
4616         return ret;
4617 }
4618
4619 #define INIT_FEATURE_FLAGS(suffix) \
4620         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4621           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4622           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4623
4624 int btrfs_ioctl_get_supported_features(void __user *arg)
4625 {
4626         static const struct btrfs_ioctl_feature_flags features[3] = {
4627                 INIT_FEATURE_FLAGS(SUPP),
4628                 INIT_FEATURE_FLAGS(SAFE_SET),
4629                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4630         };
4631
4632         if (copy_to_user(arg, &features, sizeof(features)))
4633                 return -EFAULT;
4634
4635         return 0;
4636 }
4637
4638 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4639                                         void __user *arg)
4640 {
4641         struct btrfs_super_block *super_block = fs_info->super_copy;
4642         struct btrfs_ioctl_feature_flags features;
4643
4644         features.compat_flags = btrfs_super_compat_flags(super_block);
4645         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4646         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4647
4648         if (copy_to_user(arg, &features, sizeof(features)))
4649                 return -EFAULT;
4650
4651         return 0;
4652 }
4653
4654 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4655                               enum btrfs_feature_set set,
4656                               u64 change_mask, u64 flags, u64 supported_flags,
4657                               u64 safe_set, u64 safe_clear)
4658 {
4659         const char *type = btrfs_feature_set_name(set);
4660         char *names;
4661         u64 disallowed, unsupported;
4662         u64 set_mask = flags & change_mask;
4663         u64 clear_mask = ~flags & change_mask;
4664
4665         unsupported = set_mask & ~supported_flags;
4666         if (unsupported) {
4667                 names = btrfs_printable_features(set, unsupported);
4668                 if (names) {
4669                         btrfs_warn(fs_info,
4670                                    "this kernel does not support the %s feature bit%s",
4671                                    names, strchr(names, ',') ? "s" : "");
4672                         kfree(names);
4673                 } else
4674                         btrfs_warn(fs_info,
4675                                    "this kernel does not support %s bits 0x%llx",
4676                                    type, unsupported);
4677                 return -EOPNOTSUPP;
4678         }
4679
4680         disallowed = set_mask & ~safe_set;
4681         if (disallowed) {
4682                 names = btrfs_printable_features(set, disallowed);
4683                 if (names) {
4684                         btrfs_warn(fs_info,
4685                                    "can't set the %s feature bit%s while mounted",
4686                                    names, strchr(names, ',') ? "s" : "");
4687                         kfree(names);
4688                 } else
4689                         btrfs_warn(fs_info,
4690                                    "can't set %s bits 0x%llx while mounted",
4691                                    type, disallowed);
4692                 return -EPERM;
4693         }
4694
4695         disallowed = clear_mask & ~safe_clear;
4696         if (disallowed) {
4697                 names = btrfs_printable_features(set, disallowed);
4698                 if (names) {
4699                         btrfs_warn(fs_info,
4700                                    "can't clear the %s feature bit%s while mounted",
4701                                    names, strchr(names, ',') ? "s" : "");
4702                         kfree(names);
4703                 } else
4704                         btrfs_warn(fs_info,
4705                                    "can't clear %s bits 0x%llx while mounted",
4706                                    type, disallowed);
4707                 return -EPERM;
4708         }
4709
4710         return 0;
4711 }
4712
4713 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4714 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4715                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4716                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4717                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4718
4719 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4720 {
4721         struct inode *inode = file_inode(file);
4722         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4723         struct btrfs_root *root = BTRFS_I(inode)->root;
4724         struct btrfs_super_block *super_block = fs_info->super_copy;
4725         struct btrfs_ioctl_feature_flags flags[2];
4726         struct btrfs_trans_handle *trans;
4727         u64 newflags;
4728         int ret;
4729
4730         if (!capable(CAP_SYS_ADMIN))
4731                 return -EPERM;
4732
4733         if (copy_from_user(flags, arg, sizeof(flags)))
4734                 return -EFAULT;
4735
4736         /* Nothing to do */
4737         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4738             !flags[0].incompat_flags)
4739                 return 0;
4740
4741         ret = check_feature(fs_info, flags[0].compat_flags,
4742                             flags[1].compat_flags, COMPAT);
4743         if (ret)
4744                 return ret;
4745
4746         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4747                             flags[1].compat_ro_flags, COMPAT_RO);
4748         if (ret)
4749                 return ret;
4750
4751         ret = check_feature(fs_info, flags[0].incompat_flags,
4752                             flags[1].incompat_flags, INCOMPAT);
4753         if (ret)
4754                 return ret;
4755
4756         ret = mnt_want_write_file(file);
4757         if (ret)
4758                 return ret;
4759
4760         trans = btrfs_start_transaction(root, 0);
4761         if (IS_ERR(trans)) {
4762                 ret = PTR_ERR(trans);
4763                 goto out_drop_write;
4764         }
4765
4766         spin_lock(&fs_info->super_lock);
4767         newflags = btrfs_super_compat_flags(super_block);
4768         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4769         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4770         btrfs_set_super_compat_flags(super_block, newflags);
4771
4772         newflags = btrfs_super_compat_ro_flags(super_block);
4773         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4774         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4775         btrfs_set_super_compat_ro_flags(super_block, newflags);
4776
4777         newflags = btrfs_super_incompat_flags(super_block);
4778         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4779         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4780         btrfs_set_super_incompat_flags(super_block, newflags);
4781         spin_unlock(&fs_info->super_lock);
4782
4783         ret = btrfs_commit_transaction(trans);
4784 out_drop_write:
4785         mnt_drop_write_file(file);
4786
4787         return ret;
4788 }
4789
4790 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4791 {
4792         struct btrfs_ioctl_send_args *arg;
4793         int ret;
4794
4795         if (compat) {
4796 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4797                 struct btrfs_ioctl_send_args_32 args32;
4798
4799                 ret = copy_from_user(&args32, argp, sizeof(args32));
4800                 if (ret)
4801                         return -EFAULT;
4802                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4803                 if (!arg)
4804                         return -ENOMEM;
4805                 arg->send_fd = args32.send_fd;
4806                 arg->clone_sources_count = args32.clone_sources_count;
4807                 arg->clone_sources = compat_ptr(args32.clone_sources);
4808                 arg->parent_root = args32.parent_root;
4809                 arg->flags = args32.flags;
4810                 memcpy(arg->reserved, args32.reserved,
4811                        sizeof(args32.reserved));
4812 #else
4813                 return -ENOTTY;
4814 #endif
4815         } else {
4816                 arg = memdup_user(argp, sizeof(*arg));
4817                 if (IS_ERR(arg))
4818                         return PTR_ERR(arg);
4819         }
4820         ret = btrfs_ioctl_send(file, arg);
4821         kfree(arg);
4822         return ret;
4823 }
4824
4825 long btrfs_ioctl(struct file *file, unsigned int
4826                 cmd, unsigned long arg)
4827 {
4828         struct inode *inode = file_inode(file);
4829         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4830         struct btrfs_root *root = BTRFS_I(inode)->root;
4831         void __user *argp = (void __user *)arg;
4832
4833         switch (cmd) {
4834         case FS_IOC_GETFLAGS:
4835                 return btrfs_ioctl_getflags(file, argp);
4836         case FS_IOC_SETFLAGS:
4837                 return btrfs_ioctl_setflags(file, argp);
4838         case FS_IOC_GETVERSION:
4839                 return btrfs_ioctl_getversion(file, argp);
4840         case FS_IOC_GETFSLABEL:
4841                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4842         case FS_IOC_SETFSLABEL:
4843                 return btrfs_ioctl_set_fslabel(file, argp);
4844         case FITRIM:
4845                 return btrfs_ioctl_fitrim(fs_info, argp);
4846         case BTRFS_IOC_SNAP_CREATE:
4847                 return btrfs_ioctl_snap_create(file, argp, 0);
4848         case BTRFS_IOC_SNAP_CREATE_V2:
4849                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4850         case BTRFS_IOC_SUBVOL_CREATE:
4851                 return btrfs_ioctl_snap_create(file, argp, 1);
4852         case BTRFS_IOC_SUBVOL_CREATE_V2:
4853                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4854         case BTRFS_IOC_SNAP_DESTROY:
4855                 return btrfs_ioctl_snap_destroy(file, argp, false);
4856         case BTRFS_IOC_SNAP_DESTROY_V2:
4857                 return btrfs_ioctl_snap_destroy(file, argp, true);
4858         case BTRFS_IOC_SUBVOL_GETFLAGS:
4859                 return btrfs_ioctl_subvol_getflags(file, argp);
4860         case BTRFS_IOC_SUBVOL_SETFLAGS:
4861                 return btrfs_ioctl_subvol_setflags(file, argp);
4862         case BTRFS_IOC_DEFAULT_SUBVOL:
4863                 return btrfs_ioctl_default_subvol(file, argp);
4864         case BTRFS_IOC_DEFRAG:
4865                 return btrfs_ioctl_defrag(file, NULL);
4866         case BTRFS_IOC_DEFRAG_RANGE:
4867                 return btrfs_ioctl_defrag(file, argp);
4868         case BTRFS_IOC_RESIZE:
4869                 return btrfs_ioctl_resize(file, argp);
4870         case BTRFS_IOC_ADD_DEV:
4871                 return btrfs_ioctl_add_dev(fs_info, argp);
4872         case BTRFS_IOC_RM_DEV:
4873                 return btrfs_ioctl_rm_dev(file, argp);
4874         case BTRFS_IOC_RM_DEV_V2:
4875                 return btrfs_ioctl_rm_dev_v2(file, argp);
4876         case BTRFS_IOC_FS_INFO:
4877                 return btrfs_ioctl_fs_info(fs_info, argp);
4878         case BTRFS_IOC_DEV_INFO:
4879                 return btrfs_ioctl_dev_info(fs_info, argp);
4880         case BTRFS_IOC_BALANCE:
4881                 return btrfs_ioctl_balance(file, NULL);
4882         case BTRFS_IOC_TREE_SEARCH:
4883                 return btrfs_ioctl_tree_search(file, argp);
4884         case BTRFS_IOC_TREE_SEARCH_V2:
4885                 return btrfs_ioctl_tree_search_v2(file, argp);
4886         case BTRFS_IOC_INO_LOOKUP:
4887                 return btrfs_ioctl_ino_lookup(file, argp);
4888         case BTRFS_IOC_INO_PATHS:
4889                 return btrfs_ioctl_ino_to_path(root, argp);
4890         case BTRFS_IOC_LOGICAL_INO:
4891                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4892         case BTRFS_IOC_LOGICAL_INO_V2:
4893                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4894         case BTRFS_IOC_SPACE_INFO:
4895                 return btrfs_ioctl_space_info(fs_info, argp);
4896         case BTRFS_IOC_SYNC: {
4897                 int ret;
4898
4899                 ret = btrfs_start_delalloc_roots(fs_info, -1);
4900                 if (ret)
4901                         return ret;
4902                 ret = btrfs_sync_fs(inode->i_sb, 1);
4903                 /*
4904                  * The transaction thread may want to do more work,
4905                  * namely it pokes the cleaner kthread that will start
4906                  * processing uncleaned subvols.
4907                  */
4908                 wake_up_process(fs_info->transaction_kthread);
4909                 return ret;
4910         }
4911         case BTRFS_IOC_START_SYNC:
4912                 return btrfs_ioctl_start_sync(root, argp);
4913         case BTRFS_IOC_WAIT_SYNC:
4914                 return btrfs_ioctl_wait_sync(fs_info, argp);
4915         case BTRFS_IOC_SCRUB:
4916                 return btrfs_ioctl_scrub(file, argp);
4917         case BTRFS_IOC_SCRUB_CANCEL:
4918                 return btrfs_ioctl_scrub_cancel(fs_info);
4919         case BTRFS_IOC_SCRUB_PROGRESS:
4920                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4921         case BTRFS_IOC_BALANCE_V2:
4922                 return btrfs_ioctl_balance(file, argp);
4923         case BTRFS_IOC_BALANCE_CTL:
4924                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4925         case BTRFS_IOC_BALANCE_PROGRESS:
4926                 return btrfs_ioctl_balance_progress(fs_info, argp);
4927         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4928                 return btrfs_ioctl_set_received_subvol(file, argp);
4929 #ifdef CONFIG_64BIT
4930         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4931                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4932 #endif
4933         case BTRFS_IOC_SEND:
4934                 return _btrfs_ioctl_send(file, argp, false);
4935 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4936         case BTRFS_IOC_SEND_32:
4937                 return _btrfs_ioctl_send(file, argp, true);
4938 #endif
4939         case BTRFS_IOC_GET_DEV_STATS:
4940                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4941         case BTRFS_IOC_QUOTA_CTL:
4942                 return btrfs_ioctl_quota_ctl(file, argp);
4943         case BTRFS_IOC_QGROUP_ASSIGN:
4944                 return btrfs_ioctl_qgroup_assign(file, argp);
4945         case BTRFS_IOC_QGROUP_CREATE:
4946                 return btrfs_ioctl_qgroup_create(file, argp);
4947         case BTRFS_IOC_QGROUP_LIMIT:
4948                 return btrfs_ioctl_qgroup_limit(file, argp);
4949         case BTRFS_IOC_QUOTA_RESCAN:
4950                 return btrfs_ioctl_quota_rescan(file, argp);
4951         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4952                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4953         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4954                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4955         case BTRFS_IOC_DEV_REPLACE:
4956                 return btrfs_ioctl_dev_replace(fs_info, argp);
4957         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4958                 return btrfs_ioctl_get_supported_features(argp);
4959         case BTRFS_IOC_GET_FEATURES:
4960                 return btrfs_ioctl_get_features(fs_info, argp);
4961         case BTRFS_IOC_SET_FEATURES:
4962                 return btrfs_ioctl_set_features(file, argp);
4963         case FS_IOC_FSGETXATTR:
4964                 return btrfs_ioctl_fsgetxattr(file, argp);
4965         case FS_IOC_FSSETXATTR:
4966                 return btrfs_ioctl_fssetxattr(file, argp);
4967         case BTRFS_IOC_GET_SUBVOL_INFO:
4968                 return btrfs_ioctl_get_subvol_info(file, argp);
4969         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4970                 return btrfs_ioctl_get_subvol_rootref(file, argp);
4971         case BTRFS_IOC_INO_LOOKUP_USER:
4972                 return btrfs_ioctl_ino_lookup_user(file, argp);
4973         }
4974
4975         return -ENOTTY;
4976 }
4977
4978 #ifdef CONFIG_COMPAT
4979 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4980 {
4981         /*
4982          * These all access 32-bit values anyway so no further
4983          * handling is necessary.
4984          */
4985         switch (cmd) {
4986         case FS_IOC32_GETFLAGS:
4987                 cmd = FS_IOC_GETFLAGS;
4988                 break;
4989         case FS_IOC32_SETFLAGS:
4990                 cmd = FS_IOC_SETFLAGS;
4991                 break;
4992         case FS_IOC32_GETVERSION:
4993                 cmd = FS_IOC_GETVERSION;
4994                 break;
4995         }
4996
4997         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
4998 }
4999 #endif