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 + sk_offset,
2197                                                *buf_size - sk_offset);
2198                 if (ret)
2199                         break;
2200
2201                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2202                 if (ret != 0) {
2203                         if (ret > 0)
2204                                 ret = 0;
2205                         goto err;
2206                 }
2207                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2208                                  &sk_offset, &num_found);
2209                 btrfs_release_path(path);
2210                 if (ret)
2211                         break;
2212
2213         }
2214         if (ret > 0)
2215                 ret = 0;
2216 err:
2217         sk->nr_items = num_found;
2218         btrfs_put_root(root);
2219         btrfs_free_path(path);
2220         return ret;
2221 }
2222
2223 static noinline int btrfs_ioctl_tree_search(struct file *file,
2224                                            void __user *argp)
2225 {
2226         struct btrfs_ioctl_search_args __user *uargs;
2227         struct btrfs_ioctl_search_key sk;
2228         struct inode *inode;
2229         int ret;
2230         size_t buf_size;
2231
2232         if (!capable(CAP_SYS_ADMIN))
2233                 return -EPERM;
2234
2235         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2236
2237         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2238                 return -EFAULT;
2239
2240         buf_size = sizeof(uargs->buf);
2241
2242         inode = file_inode(file);
2243         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2244
2245         /*
2246          * In the origin implementation an overflow is handled by returning a
2247          * search header with a len of zero, so reset ret.
2248          */
2249         if (ret == -EOVERFLOW)
2250                 ret = 0;
2251
2252         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2253                 ret = -EFAULT;
2254         return ret;
2255 }
2256
2257 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2258                                                void __user *argp)
2259 {
2260         struct btrfs_ioctl_search_args_v2 __user *uarg;
2261         struct btrfs_ioctl_search_args_v2 args;
2262         struct inode *inode;
2263         int ret;
2264         size_t buf_size;
2265         const size_t buf_limit = SZ_16M;
2266
2267         if (!capable(CAP_SYS_ADMIN))
2268                 return -EPERM;
2269
2270         /* copy search header and buffer size */
2271         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2272         if (copy_from_user(&args, uarg, sizeof(args)))
2273                 return -EFAULT;
2274
2275         buf_size = args.buf_size;
2276
2277         /* limit result size to 16MB */
2278         if (buf_size > buf_limit)
2279                 buf_size = buf_limit;
2280
2281         inode = file_inode(file);
2282         ret = search_ioctl(inode, &args.key, &buf_size,
2283                            (char __user *)(&uarg->buf[0]));
2284         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2285                 ret = -EFAULT;
2286         else if (ret == -EOVERFLOW &&
2287                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2288                 ret = -EFAULT;
2289
2290         return ret;
2291 }
2292
2293 /*
2294  * Search INODE_REFs to identify path name of 'dirid' directory
2295  * in a 'tree_id' tree. and sets path name to 'name'.
2296  */
2297 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2298                                 u64 tree_id, u64 dirid, char *name)
2299 {
2300         struct btrfs_root *root;
2301         struct btrfs_key key;
2302         char *ptr;
2303         int ret = -1;
2304         int slot;
2305         int len;
2306         int total_len = 0;
2307         struct btrfs_inode_ref *iref;
2308         struct extent_buffer *l;
2309         struct btrfs_path *path;
2310
2311         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2312                 name[0]='\0';
2313                 return 0;
2314         }
2315
2316         path = btrfs_alloc_path();
2317         if (!path)
2318                 return -ENOMEM;
2319
2320         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2321
2322         root = btrfs_get_fs_root(info, tree_id, true);
2323         if (IS_ERR(root)) {
2324                 ret = PTR_ERR(root);
2325                 root = NULL;
2326                 goto out;
2327         }
2328
2329         key.objectid = dirid;
2330         key.type = BTRFS_INODE_REF_KEY;
2331         key.offset = (u64)-1;
2332
2333         while (1) {
2334                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2335                 if (ret < 0)
2336                         goto out;
2337                 else if (ret > 0) {
2338                         ret = btrfs_previous_item(root, path, dirid,
2339                                                   BTRFS_INODE_REF_KEY);
2340                         if (ret < 0)
2341                                 goto out;
2342                         else if (ret > 0) {
2343                                 ret = -ENOENT;
2344                                 goto out;
2345                         }
2346                 }
2347
2348                 l = path->nodes[0];
2349                 slot = path->slots[0];
2350                 btrfs_item_key_to_cpu(l, &key, slot);
2351
2352                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2353                 len = btrfs_inode_ref_name_len(l, iref);
2354                 ptr -= len + 1;
2355                 total_len += len + 1;
2356                 if (ptr < name) {
2357                         ret = -ENAMETOOLONG;
2358                         goto out;
2359                 }
2360
2361                 *(ptr + len) = '/';
2362                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2363
2364                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2365                         break;
2366
2367                 btrfs_release_path(path);
2368                 key.objectid = key.offset;
2369                 key.offset = (u64)-1;
2370                 dirid = key.objectid;
2371         }
2372         memmove(name, ptr, total_len);
2373         name[total_len] = '\0';
2374         ret = 0;
2375 out:
2376         btrfs_put_root(root);
2377         btrfs_free_path(path);
2378         return ret;
2379 }
2380
2381 static int btrfs_search_path_in_tree_user(struct inode *inode,
2382                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2383 {
2384         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2385         struct super_block *sb = inode->i_sb;
2386         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2387         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2388         u64 dirid = args->dirid;
2389         unsigned long item_off;
2390         unsigned long item_len;
2391         struct btrfs_inode_ref *iref;
2392         struct btrfs_root_ref *rref;
2393         struct btrfs_root *root = NULL;
2394         struct btrfs_path *path;
2395         struct btrfs_key key, key2;
2396         struct extent_buffer *leaf;
2397         struct inode *temp_inode;
2398         char *ptr;
2399         int slot;
2400         int len;
2401         int total_len = 0;
2402         int ret;
2403
2404         path = btrfs_alloc_path();
2405         if (!path)
2406                 return -ENOMEM;
2407
2408         /*
2409          * If the bottom subvolume does not exist directly under upper_limit,
2410          * construct the path in from the bottom up.
2411          */
2412         if (dirid != upper_limit.objectid) {
2413                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2414
2415                 root = btrfs_get_fs_root(fs_info, treeid, true);
2416                 if (IS_ERR(root)) {
2417                         ret = PTR_ERR(root);
2418                         goto out;
2419                 }
2420
2421                 key.objectid = dirid;
2422                 key.type = BTRFS_INODE_REF_KEY;
2423                 key.offset = (u64)-1;
2424                 while (1) {
2425                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2426                         if (ret < 0) {
2427                                 goto out_put;
2428                         } else if (ret > 0) {
2429                                 ret = btrfs_previous_item(root, path, dirid,
2430                                                           BTRFS_INODE_REF_KEY);
2431                                 if (ret < 0) {
2432                                         goto out_put;
2433                                 } else if (ret > 0) {
2434                                         ret = -ENOENT;
2435                                         goto out_put;
2436                                 }
2437                         }
2438
2439                         leaf = path->nodes[0];
2440                         slot = path->slots[0];
2441                         btrfs_item_key_to_cpu(leaf, &key, slot);
2442
2443                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2444                         len = btrfs_inode_ref_name_len(leaf, iref);
2445                         ptr -= len + 1;
2446                         total_len += len + 1;
2447                         if (ptr < args->path) {
2448                                 ret = -ENAMETOOLONG;
2449                                 goto out_put;
2450                         }
2451
2452                         *(ptr + len) = '/';
2453                         read_extent_buffer(leaf, ptr,
2454                                         (unsigned long)(iref + 1), len);
2455
2456                         /* Check the read+exec permission of this directory */
2457                         ret = btrfs_previous_item(root, path, dirid,
2458                                                   BTRFS_INODE_ITEM_KEY);
2459                         if (ret < 0) {
2460                                 goto out_put;
2461                         } else if (ret > 0) {
2462                                 ret = -ENOENT;
2463                                 goto out_put;
2464                         }
2465
2466                         leaf = path->nodes[0];
2467                         slot = path->slots[0];
2468                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2469                         if (key2.objectid != dirid) {
2470                                 ret = -ENOENT;
2471                                 goto out_put;
2472                         }
2473
2474                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2475                         if (IS_ERR(temp_inode)) {
2476                                 ret = PTR_ERR(temp_inode);
2477                                 goto out_put;
2478                         }
2479                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2480                         iput(temp_inode);
2481                         if (ret) {
2482                                 ret = -EACCES;
2483                                 goto out_put;
2484                         }
2485
2486                         if (key.offset == upper_limit.objectid)
2487                                 break;
2488                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2489                                 ret = -EACCES;
2490                                 goto out_put;
2491                         }
2492
2493                         btrfs_release_path(path);
2494                         key.objectid = key.offset;
2495                         key.offset = (u64)-1;
2496                         dirid = key.objectid;
2497                 }
2498
2499                 memmove(args->path, ptr, total_len);
2500                 args->path[total_len] = '\0';
2501                 btrfs_put_root(root);
2502                 root = NULL;
2503                 btrfs_release_path(path);
2504         }
2505
2506         /* Get the bottom subvolume's name from ROOT_REF */
2507         key.objectid = treeid;
2508         key.type = BTRFS_ROOT_REF_KEY;
2509         key.offset = args->treeid;
2510         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2511         if (ret < 0) {
2512                 goto out;
2513         } else if (ret > 0) {
2514                 ret = -ENOENT;
2515                 goto out;
2516         }
2517
2518         leaf = path->nodes[0];
2519         slot = path->slots[0];
2520         btrfs_item_key_to_cpu(leaf, &key, slot);
2521
2522         item_off = btrfs_item_ptr_offset(leaf, slot);
2523         item_len = btrfs_item_size_nr(leaf, slot);
2524         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2525         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2526         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2527                 ret = -EINVAL;
2528                 goto out;
2529         }
2530
2531         /* Copy subvolume's name */
2532         item_off += sizeof(struct btrfs_root_ref);
2533         item_len -= sizeof(struct btrfs_root_ref);
2534         read_extent_buffer(leaf, args->name, item_off, item_len);
2535         args->name[item_len] = 0;
2536
2537 out_put:
2538         btrfs_put_root(root);
2539 out:
2540         btrfs_free_path(path);
2541         return ret;
2542 }
2543
2544 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2545                                            void __user *argp)
2546 {
2547         struct btrfs_ioctl_ino_lookup_args *args;
2548         struct inode *inode;
2549         int ret = 0;
2550
2551         args = memdup_user(argp, sizeof(*args));
2552         if (IS_ERR(args))
2553                 return PTR_ERR(args);
2554
2555         inode = file_inode(file);
2556
2557         /*
2558          * Unprivileged query to obtain the containing subvolume root id. The
2559          * path is reset so it's consistent with btrfs_search_path_in_tree.
2560          */
2561         if (args->treeid == 0)
2562                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2563
2564         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2565                 args->name[0] = 0;
2566                 goto out;
2567         }
2568
2569         if (!capable(CAP_SYS_ADMIN)) {
2570                 ret = -EPERM;
2571                 goto out;
2572         }
2573
2574         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2575                                         args->treeid, args->objectid,
2576                                         args->name);
2577
2578 out:
2579         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2580                 ret = -EFAULT;
2581
2582         kfree(args);
2583         return ret;
2584 }
2585
2586 /*
2587  * Version of ino_lookup ioctl (unprivileged)
2588  *
2589  * The main differences from ino_lookup ioctl are:
2590  *
2591  *   1. Read + Exec permission will be checked using inode_permission() during
2592  *      path construction. -EACCES will be returned in case of failure.
2593  *   2. Path construction will be stopped at the inode number which corresponds
2594  *      to the fd with which this ioctl is called. If constructed path does not
2595  *      exist under fd's inode, -EACCES will be returned.
2596  *   3. The name of bottom subvolume is also searched and filled.
2597  */
2598 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2599 {
2600         struct btrfs_ioctl_ino_lookup_user_args *args;
2601         struct inode *inode;
2602         int ret;
2603
2604         args = memdup_user(argp, sizeof(*args));
2605         if (IS_ERR(args))
2606                 return PTR_ERR(args);
2607
2608         inode = file_inode(file);
2609
2610         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2611             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2612                 /*
2613                  * The subvolume does not exist under fd with which this is
2614                  * called
2615                  */
2616                 kfree(args);
2617                 return -EACCES;
2618         }
2619
2620         ret = btrfs_search_path_in_tree_user(inode, args);
2621
2622         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2623                 ret = -EFAULT;
2624
2625         kfree(args);
2626         return ret;
2627 }
2628
2629 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2630 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2631 {
2632         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2633         struct btrfs_fs_info *fs_info;
2634         struct btrfs_root *root;
2635         struct btrfs_path *path;
2636         struct btrfs_key key;
2637         struct btrfs_root_item *root_item;
2638         struct btrfs_root_ref *rref;
2639         struct extent_buffer *leaf;
2640         unsigned long item_off;
2641         unsigned long item_len;
2642         struct inode *inode;
2643         int slot;
2644         int ret = 0;
2645
2646         path = btrfs_alloc_path();
2647         if (!path)
2648                 return -ENOMEM;
2649
2650         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2651         if (!subvol_info) {
2652                 btrfs_free_path(path);
2653                 return -ENOMEM;
2654         }
2655
2656         inode = file_inode(file);
2657         fs_info = BTRFS_I(inode)->root->fs_info;
2658
2659         /* Get root_item of inode's subvolume */
2660         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2661         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2662         if (IS_ERR(root)) {
2663                 ret = PTR_ERR(root);
2664                 goto out_free;
2665         }
2666         root_item = &root->root_item;
2667
2668         subvol_info->treeid = key.objectid;
2669
2670         subvol_info->generation = btrfs_root_generation(root_item);
2671         subvol_info->flags = btrfs_root_flags(root_item);
2672
2673         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2674         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2675                                                     BTRFS_UUID_SIZE);
2676         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2677                                                     BTRFS_UUID_SIZE);
2678
2679         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2680         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2681         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2682
2683         subvol_info->otransid = btrfs_root_otransid(root_item);
2684         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2685         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2686
2687         subvol_info->stransid = btrfs_root_stransid(root_item);
2688         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2689         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2690
2691         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2692         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2693         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2694
2695         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2696                 /* Search root tree for ROOT_BACKREF of this subvolume */
2697                 key.type = BTRFS_ROOT_BACKREF_KEY;
2698                 key.offset = 0;
2699                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2700                 if (ret < 0) {
2701                         goto out;
2702                 } else if (path->slots[0] >=
2703                            btrfs_header_nritems(path->nodes[0])) {
2704                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2705                         if (ret < 0) {
2706                                 goto out;
2707                         } else if (ret > 0) {
2708                                 ret = -EUCLEAN;
2709                                 goto out;
2710                         }
2711                 }
2712
2713                 leaf = path->nodes[0];
2714                 slot = path->slots[0];
2715                 btrfs_item_key_to_cpu(leaf, &key, slot);
2716                 if (key.objectid == subvol_info->treeid &&
2717                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2718                         subvol_info->parent_id = key.offset;
2719
2720                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2721                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2722
2723                         item_off = btrfs_item_ptr_offset(leaf, slot)
2724                                         + sizeof(struct btrfs_root_ref);
2725                         item_len = btrfs_item_size_nr(leaf, slot)
2726                                         - sizeof(struct btrfs_root_ref);
2727                         read_extent_buffer(leaf, subvol_info->name,
2728                                            item_off, item_len);
2729                 } else {
2730                         ret = -ENOENT;
2731                         goto out;
2732                 }
2733         }
2734
2735         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2736                 ret = -EFAULT;
2737
2738 out:
2739         btrfs_put_root(root);
2740 out_free:
2741         btrfs_free_path(path);
2742         kfree(subvol_info);
2743         return ret;
2744 }
2745
2746 /*
2747  * Return ROOT_REF information of the subvolume containing this inode
2748  * except the subvolume name.
2749  */
2750 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2751 {
2752         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2753         struct btrfs_root_ref *rref;
2754         struct btrfs_root *root;
2755         struct btrfs_path *path;
2756         struct btrfs_key key;
2757         struct extent_buffer *leaf;
2758         struct inode *inode;
2759         u64 objectid;
2760         int slot;
2761         int ret;
2762         u8 found;
2763
2764         path = btrfs_alloc_path();
2765         if (!path)
2766                 return -ENOMEM;
2767
2768         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2769         if (IS_ERR(rootrefs)) {
2770                 btrfs_free_path(path);
2771                 return PTR_ERR(rootrefs);
2772         }
2773
2774         inode = file_inode(file);
2775         root = BTRFS_I(inode)->root->fs_info->tree_root;
2776         objectid = BTRFS_I(inode)->root->root_key.objectid;
2777
2778         key.objectid = objectid;
2779         key.type = BTRFS_ROOT_REF_KEY;
2780         key.offset = rootrefs->min_treeid;
2781         found = 0;
2782
2783         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2784         if (ret < 0) {
2785                 goto out;
2786         } else if (path->slots[0] >=
2787                    btrfs_header_nritems(path->nodes[0])) {
2788                 ret = btrfs_next_leaf(root, path);
2789                 if (ret < 0) {
2790                         goto out;
2791                 } else if (ret > 0) {
2792                         ret = -EUCLEAN;
2793                         goto out;
2794                 }
2795         }
2796         while (1) {
2797                 leaf = path->nodes[0];
2798                 slot = path->slots[0];
2799
2800                 btrfs_item_key_to_cpu(leaf, &key, slot);
2801                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2802                         ret = 0;
2803                         goto out;
2804                 }
2805
2806                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2807                         ret = -EOVERFLOW;
2808                         goto out;
2809                 }
2810
2811                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2812                 rootrefs->rootref[found].treeid = key.offset;
2813                 rootrefs->rootref[found].dirid =
2814                                   btrfs_root_ref_dirid(leaf, rref);
2815                 found++;
2816
2817                 ret = btrfs_next_item(root, path);
2818                 if (ret < 0) {
2819                         goto out;
2820                 } else if (ret > 0) {
2821                         ret = -EUCLEAN;
2822                         goto out;
2823                 }
2824         }
2825
2826 out:
2827         if (!ret || ret == -EOVERFLOW) {
2828                 rootrefs->num_items = found;
2829                 /* update min_treeid for next search */
2830                 if (found)
2831                         rootrefs->min_treeid =
2832                                 rootrefs->rootref[found - 1].treeid + 1;
2833                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2834                         ret = -EFAULT;
2835         }
2836
2837         kfree(rootrefs);
2838         btrfs_free_path(path);
2839
2840         return ret;
2841 }
2842
2843 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2844                                              void __user *arg,
2845                                              bool destroy_v2)
2846 {
2847         struct dentry *parent = file->f_path.dentry;
2848         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2849         struct dentry *dentry;
2850         struct inode *dir = d_inode(parent);
2851         struct inode *inode;
2852         struct btrfs_root *root = BTRFS_I(dir)->root;
2853         struct btrfs_root *dest = NULL;
2854         struct btrfs_ioctl_vol_args *vol_args = NULL;
2855         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2856         char *subvol_name, *subvol_name_ptr = NULL;
2857         int subvol_namelen;
2858         int err = 0;
2859         bool destroy_parent = false;
2860
2861         if (destroy_v2) {
2862                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2863                 if (IS_ERR(vol_args2))
2864                         return PTR_ERR(vol_args2);
2865
2866                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2867                         err = -EOPNOTSUPP;
2868                         goto out;
2869                 }
2870
2871                 /*
2872                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2873                  * name, same as v1 currently does.
2874                  */
2875                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2876                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2877                         subvol_name = vol_args2->name;
2878
2879                         err = mnt_want_write_file(file);
2880                         if (err)
2881                                 goto out;
2882                 } else {
2883                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2884                                 err = -EINVAL;
2885                                 goto out;
2886                         }
2887
2888                         err = mnt_want_write_file(file);
2889                         if (err)
2890                                 goto out;
2891
2892                         dentry = btrfs_get_dentry(fs_info->sb,
2893                                         BTRFS_FIRST_FREE_OBJECTID,
2894                                         vol_args2->subvolid, 0, 0);
2895                         if (IS_ERR(dentry)) {
2896                                 err = PTR_ERR(dentry);
2897                                 goto out_drop_write;
2898                         }
2899
2900                         /*
2901                          * Change the default parent since the subvolume being
2902                          * deleted can be outside of the current mount point.
2903                          */
2904                         parent = btrfs_get_parent(dentry);
2905
2906                         /*
2907                          * At this point dentry->d_name can point to '/' if the
2908                          * subvolume we want to destroy is outsite of the
2909                          * current mount point, so we need to release the
2910                          * current dentry and execute the lookup to return a new
2911                          * one with ->d_name pointing to the
2912                          * <mount point>/subvol_name.
2913                          */
2914                         dput(dentry);
2915                         if (IS_ERR(parent)) {
2916                                 err = PTR_ERR(parent);
2917                                 goto out_drop_write;
2918                         }
2919                         dir = d_inode(parent);
2920
2921                         /*
2922                          * If v2 was used with SPEC_BY_ID, a new parent was
2923                          * allocated since the subvolume can be outside of the
2924                          * current mount point. Later on we need to release this
2925                          * new parent dentry.
2926                          */
2927                         destroy_parent = true;
2928
2929                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2930                                                 fs_info, vol_args2->subvolid);
2931                         if (IS_ERR(subvol_name_ptr)) {
2932                                 err = PTR_ERR(subvol_name_ptr);
2933                                 goto free_parent;
2934                         }
2935                         /* subvol_name_ptr is already NULL termined */
2936                         subvol_name = (char *)kbasename(subvol_name_ptr);
2937                 }
2938         } else {
2939                 vol_args = memdup_user(arg, sizeof(*vol_args));
2940                 if (IS_ERR(vol_args))
2941                         return PTR_ERR(vol_args);
2942
2943                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2944                 subvol_name = vol_args->name;
2945
2946                 err = mnt_want_write_file(file);
2947                 if (err)
2948                         goto out;
2949         }
2950
2951         subvol_namelen = strlen(subvol_name);
2952
2953         if (strchr(subvol_name, '/') ||
2954             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2955                 err = -EINVAL;
2956                 goto free_subvol_name;
2957         }
2958
2959         if (!S_ISDIR(dir->i_mode)) {
2960                 err = -ENOTDIR;
2961                 goto free_subvol_name;
2962         }
2963
2964         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2965         if (err == -EINTR)
2966                 goto free_subvol_name;
2967         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
2968         if (IS_ERR(dentry)) {
2969                 err = PTR_ERR(dentry);
2970                 goto out_unlock_dir;
2971         }
2972
2973         if (d_really_is_negative(dentry)) {
2974                 err = -ENOENT;
2975                 goto out_dput;
2976         }
2977
2978         inode = d_inode(dentry);
2979         dest = BTRFS_I(inode)->root;
2980         if (!capable(CAP_SYS_ADMIN)) {
2981                 /*
2982                  * Regular user.  Only allow this with a special mount
2983                  * option, when the user has write+exec access to the
2984                  * subvol root, and when rmdir(2) would have been
2985                  * allowed.
2986                  *
2987                  * Note that this is _not_ check that the subvol is
2988                  * empty or doesn't contain data that we wouldn't
2989                  * otherwise be able to delete.
2990                  *
2991                  * Users who want to delete empty subvols should try
2992                  * rmdir(2).
2993                  */
2994                 err = -EPERM;
2995                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2996                         goto out_dput;
2997
2998                 /*
2999                  * Do not allow deletion if the parent dir is the same
3000                  * as the dir to be deleted.  That means the ioctl
3001                  * must be called on the dentry referencing the root
3002                  * of the subvol, not a random directory contained
3003                  * within it.
3004                  */
3005                 err = -EINVAL;
3006                 if (root == dest)
3007                         goto out_dput;
3008
3009                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
3010                 if (err)
3011                         goto out_dput;
3012         }
3013
3014         /* check if subvolume may be deleted by a user */
3015         err = btrfs_may_delete(dir, dentry, 1);
3016         if (err)
3017                 goto out_dput;
3018
3019         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3020                 err = -EINVAL;
3021                 goto out_dput;
3022         }
3023
3024         inode_lock(inode);
3025         err = btrfs_delete_subvolume(dir, dentry);
3026         inode_unlock(inode);
3027         if (!err) {
3028                 fsnotify_rmdir(dir, dentry);
3029                 d_delete(dentry);
3030         }
3031
3032 out_dput:
3033         dput(dentry);
3034 out_unlock_dir:
3035         inode_unlock(dir);
3036 free_subvol_name:
3037         kfree(subvol_name_ptr);
3038 free_parent:
3039         if (destroy_parent)
3040                 dput(parent);
3041 out_drop_write:
3042         mnt_drop_write_file(file);
3043 out:
3044         kfree(vol_args2);
3045         kfree(vol_args);
3046         return err;
3047 }
3048
3049 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3050 {
3051         struct inode *inode = file_inode(file);
3052         struct btrfs_root *root = BTRFS_I(inode)->root;
3053         struct btrfs_ioctl_defrag_range_args *range;
3054         int ret;
3055
3056         ret = mnt_want_write_file(file);
3057         if (ret)
3058                 return ret;
3059
3060         if (btrfs_root_readonly(root)) {
3061                 ret = -EROFS;
3062                 goto out;
3063         }
3064
3065         switch (inode->i_mode & S_IFMT) {
3066         case S_IFDIR:
3067                 if (!capable(CAP_SYS_ADMIN)) {
3068                         ret = -EPERM;
3069                         goto out;
3070                 }
3071                 ret = btrfs_defrag_root(root);
3072                 break;
3073         case S_IFREG:
3074                 /*
3075                  * Note that this does not check the file descriptor for write
3076                  * access. This prevents defragmenting executables that are
3077                  * running and allows defrag on files open in read-only mode.
3078                  */
3079                 if (!capable(CAP_SYS_ADMIN) &&
3080                     inode_permission(inode, MAY_WRITE)) {
3081                         ret = -EPERM;
3082                         goto out;
3083                 }
3084
3085                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3086                 if (!range) {
3087                         ret = -ENOMEM;
3088                         goto out;
3089                 }
3090
3091                 if (argp) {
3092                         if (copy_from_user(range, argp,
3093                                            sizeof(*range))) {
3094                                 ret = -EFAULT;
3095                                 kfree(range);
3096                                 goto out;
3097                         }
3098                         /* compression requires us to start the IO */
3099                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3100                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3101                                 range->extent_thresh = (u32)-1;
3102                         }
3103                 } else {
3104                         /* the rest are all set to zero by kzalloc */
3105                         range->len = (u64)-1;
3106                 }
3107                 ret = btrfs_defrag_file(file_inode(file), file,
3108                                         range, BTRFS_OLDEST_GENERATION, 0);
3109                 if (ret > 0)
3110                         ret = 0;
3111                 kfree(range);
3112                 break;
3113         default:
3114                 ret = -EINVAL;
3115         }
3116 out:
3117         mnt_drop_write_file(file);
3118         return ret;
3119 }
3120
3121 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3122 {
3123         struct btrfs_ioctl_vol_args *vol_args;
3124         int ret;
3125
3126         if (!capable(CAP_SYS_ADMIN))
3127                 return -EPERM;
3128
3129         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3130                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3131
3132         vol_args = memdup_user(arg, sizeof(*vol_args));
3133         if (IS_ERR(vol_args)) {
3134                 ret = PTR_ERR(vol_args);
3135                 goto out;
3136         }
3137
3138         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3139         ret = btrfs_init_new_device(fs_info, vol_args->name);
3140
3141         if (!ret)
3142                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3143
3144         kfree(vol_args);
3145 out:
3146         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3147         return ret;
3148 }
3149
3150 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3151 {
3152         struct inode *inode = file_inode(file);
3153         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3154         struct btrfs_ioctl_vol_args_v2 *vol_args;
3155         int ret;
3156
3157         if (!capable(CAP_SYS_ADMIN))
3158                 return -EPERM;
3159
3160         ret = mnt_want_write_file(file);
3161         if (ret)
3162                 return ret;
3163
3164         vol_args = memdup_user(arg, sizeof(*vol_args));
3165         if (IS_ERR(vol_args)) {
3166                 ret = PTR_ERR(vol_args);
3167                 goto err_drop;
3168         }
3169
3170         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3171                 ret = -EOPNOTSUPP;
3172                 goto out;
3173         }
3174
3175         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3176                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3177                 goto out;
3178         }
3179
3180         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3181                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3182         } else {
3183                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3184                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3185         }
3186         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3187
3188         if (!ret) {
3189                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3190                         btrfs_info(fs_info, "device deleted: id %llu",
3191                                         vol_args->devid);
3192                 else
3193                         btrfs_info(fs_info, "device deleted: %s",
3194                                         vol_args->name);
3195         }
3196 out:
3197         kfree(vol_args);
3198 err_drop:
3199         mnt_drop_write_file(file);
3200         return ret;
3201 }
3202
3203 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3204 {
3205         struct inode *inode = file_inode(file);
3206         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3207         struct btrfs_ioctl_vol_args *vol_args;
3208         int ret;
3209
3210         if (!capable(CAP_SYS_ADMIN))
3211                 return -EPERM;
3212
3213         ret = mnt_want_write_file(file);
3214         if (ret)
3215                 return ret;
3216
3217         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3218                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3219                 goto out_drop_write;
3220         }
3221
3222         vol_args = memdup_user(arg, sizeof(*vol_args));
3223         if (IS_ERR(vol_args)) {
3224                 ret = PTR_ERR(vol_args);
3225                 goto out;
3226         }
3227
3228         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3229         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3230
3231         if (!ret)
3232                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3233         kfree(vol_args);
3234 out:
3235         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3236 out_drop_write:
3237         mnt_drop_write_file(file);
3238
3239         return ret;
3240 }
3241
3242 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3243                                 void __user *arg)
3244 {
3245         struct btrfs_ioctl_fs_info_args *fi_args;
3246         struct btrfs_device *device;
3247         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3248         u64 flags_in;
3249         int ret = 0;
3250
3251         fi_args = memdup_user(arg, sizeof(*fi_args));
3252         if (IS_ERR(fi_args))
3253                 return PTR_ERR(fi_args);
3254
3255         flags_in = fi_args->flags;
3256         memset(fi_args, 0, sizeof(*fi_args));
3257
3258         rcu_read_lock();
3259         fi_args->num_devices = fs_devices->num_devices;
3260
3261         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3262                 if (device->devid > fi_args->max_id)
3263                         fi_args->max_id = device->devid;
3264         }
3265         rcu_read_unlock();
3266
3267         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3268         fi_args->nodesize = fs_info->nodesize;
3269         fi_args->sectorsize = fs_info->sectorsize;
3270         fi_args->clone_alignment = fs_info->sectorsize;
3271
3272         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3273                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3274                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3275                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3276         }
3277
3278         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3279                 fi_args->generation = fs_info->generation;
3280                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3281         }
3282
3283         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3284                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3285                        sizeof(fi_args->metadata_uuid));
3286                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3287         }
3288
3289         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3290                 ret = -EFAULT;
3291
3292         kfree(fi_args);
3293         return ret;
3294 }
3295
3296 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3297                                  void __user *arg)
3298 {
3299         struct btrfs_ioctl_dev_info_args *di_args;
3300         struct btrfs_device *dev;
3301         int ret = 0;
3302         char *s_uuid = NULL;
3303
3304         di_args = memdup_user(arg, sizeof(*di_args));
3305         if (IS_ERR(di_args))
3306                 return PTR_ERR(di_args);
3307
3308         if (!btrfs_is_empty_uuid(di_args->uuid))
3309                 s_uuid = di_args->uuid;
3310
3311         rcu_read_lock();
3312         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3313                                 NULL, true);
3314
3315         if (!dev) {
3316                 ret = -ENODEV;
3317                 goto out;
3318         }
3319
3320         di_args->devid = dev->devid;
3321         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3322         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3323         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3324         if (dev->name) {
3325                 strncpy(di_args->path, rcu_str_deref(dev->name),
3326                                 sizeof(di_args->path) - 1);
3327                 di_args->path[sizeof(di_args->path) - 1] = 0;
3328         } else {
3329                 di_args->path[0] = '\0';
3330         }
3331
3332 out:
3333         rcu_read_unlock();
3334         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3335                 ret = -EFAULT;
3336
3337         kfree(di_args);
3338         return ret;
3339 }
3340
3341 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3342 {
3343         struct inode *inode = file_inode(file);
3344         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3345         struct btrfs_root *root = BTRFS_I(inode)->root;
3346         struct btrfs_root *new_root;
3347         struct btrfs_dir_item *di;
3348         struct btrfs_trans_handle *trans;
3349         struct btrfs_path *path = NULL;
3350         struct btrfs_disk_key disk_key;
3351         u64 objectid = 0;
3352         u64 dir_id;
3353         int ret;
3354
3355         if (!capable(CAP_SYS_ADMIN))
3356                 return -EPERM;
3357
3358         ret = mnt_want_write_file(file);
3359         if (ret)
3360                 return ret;
3361
3362         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3363                 ret = -EFAULT;
3364                 goto out;
3365         }
3366
3367         if (!objectid)
3368                 objectid = BTRFS_FS_TREE_OBJECTID;
3369
3370         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3371         if (IS_ERR(new_root)) {
3372                 ret = PTR_ERR(new_root);
3373                 goto out;
3374         }
3375         if (!is_fstree(new_root->root_key.objectid)) {
3376                 ret = -ENOENT;
3377                 goto out_free;
3378         }
3379
3380         path = btrfs_alloc_path();
3381         if (!path) {
3382                 ret = -ENOMEM;
3383                 goto out_free;
3384         }
3385         path->leave_spinning = 1;
3386
3387         trans = btrfs_start_transaction(root, 1);
3388         if (IS_ERR(trans)) {
3389                 ret = PTR_ERR(trans);
3390                 goto out_free;
3391         }
3392
3393         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3394         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3395                                    dir_id, "default", 7, 1);
3396         if (IS_ERR_OR_NULL(di)) {
3397                 btrfs_release_path(path);
3398                 btrfs_end_transaction(trans);
3399                 btrfs_err(fs_info,
3400                           "Umm, you don't have the default diritem, this isn't going to work");
3401                 ret = -ENOENT;
3402                 goto out_free;
3403         }
3404
3405         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3406         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3407         btrfs_mark_buffer_dirty(path->nodes[0]);
3408         btrfs_release_path(path);
3409
3410         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3411         btrfs_end_transaction(trans);
3412 out_free:
3413         btrfs_put_root(new_root);
3414         btrfs_free_path(path);
3415 out:
3416         mnt_drop_write_file(file);
3417         return ret;
3418 }
3419
3420 static void get_block_group_info(struct list_head *groups_list,
3421                                  struct btrfs_ioctl_space_info *space)
3422 {
3423         struct btrfs_block_group *block_group;
3424
3425         space->total_bytes = 0;
3426         space->used_bytes = 0;
3427         space->flags = 0;
3428         list_for_each_entry(block_group, groups_list, list) {
3429                 space->flags = block_group->flags;
3430                 space->total_bytes += block_group->length;
3431                 space->used_bytes += block_group->used;
3432         }
3433 }
3434
3435 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3436                                    void __user *arg)
3437 {
3438         struct btrfs_ioctl_space_args space_args;
3439         struct btrfs_ioctl_space_info space;
3440         struct btrfs_ioctl_space_info *dest;
3441         struct btrfs_ioctl_space_info *dest_orig;
3442         struct btrfs_ioctl_space_info __user *user_dest;
3443         struct btrfs_space_info *info;
3444         static const u64 types[] = {
3445                 BTRFS_BLOCK_GROUP_DATA,
3446                 BTRFS_BLOCK_GROUP_SYSTEM,
3447                 BTRFS_BLOCK_GROUP_METADATA,
3448                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3449         };
3450         int num_types = 4;
3451         int alloc_size;
3452         int ret = 0;
3453         u64 slot_count = 0;
3454         int i, c;
3455
3456         if (copy_from_user(&space_args,
3457                            (struct btrfs_ioctl_space_args __user *)arg,
3458                            sizeof(space_args)))
3459                 return -EFAULT;
3460
3461         for (i = 0; i < num_types; i++) {
3462                 struct btrfs_space_info *tmp;
3463
3464                 info = NULL;
3465                 rcu_read_lock();
3466                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
3467                                         list) {
3468                         if (tmp->flags == types[i]) {
3469                                 info = tmp;
3470                                 break;
3471                         }
3472                 }
3473                 rcu_read_unlock();
3474
3475                 if (!info)
3476                         continue;
3477
3478                 down_read(&info->groups_sem);
3479                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3480                         if (!list_empty(&info->block_groups[c]))
3481                                 slot_count++;
3482                 }
3483                 up_read(&info->groups_sem);
3484         }
3485
3486         /*
3487          * Global block reserve, exported as a space_info
3488          */
3489         slot_count++;
3490
3491         /* space_slots == 0 means they are asking for a count */
3492         if (space_args.space_slots == 0) {
3493                 space_args.total_spaces = slot_count;
3494                 goto out;
3495         }
3496
3497         slot_count = min_t(u64, space_args.space_slots, slot_count);
3498
3499         alloc_size = sizeof(*dest) * slot_count;
3500
3501         /* we generally have at most 6 or so space infos, one for each raid
3502          * level.  So, a whole page should be more than enough for everyone
3503          */
3504         if (alloc_size > PAGE_SIZE)
3505                 return -ENOMEM;
3506
3507         space_args.total_spaces = 0;
3508         dest = kmalloc(alloc_size, GFP_KERNEL);
3509         if (!dest)
3510                 return -ENOMEM;
3511         dest_orig = dest;
3512
3513         /* now we have a buffer to copy into */
3514         for (i = 0; i < num_types; i++) {
3515                 struct btrfs_space_info *tmp;
3516
3517                 if (!slot_count)
3518                         break;
3519
3520                 info = NULL;
3521                 rcu_read_lock();
3522                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
3523                                         list) {
3524                         if (tmp->flags == types[i]) {
3525                                 info = tmp;
3526                                 break;
3527                         }
3528                 }
3529                 rcu_read_unlock();
3530
3531                 if (!info)
3532                         continue;
3533                 down_read(&info->groups_sem);
3534                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3535                         if (!list_empty(&info->block_groups[c])) {
3536                                 get_block_group_info(&info->block_groups[c],
3537                                                      &space);
3538                                 memcpy(dest, &space, sizeof(space));
3539                                 dest++;
3540                                 space_args.total_spaces++;
3541                                 slot_count--;
3542                         }
3543                         if (!slot_count)
3544                                 break;
3545                 }
3546                 up_read(&info->groups_sem);
3547         }
3548
3549         /*
3550          * Add global block reserve
3551          */
3552         if (slot_count) {
3553                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3554
3555                 spin_lock(&block_rsv->lock);
3556                 space.total_bytes = block_rsv->size;
3557                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3558                 spin_unlock(&block_rsv->lock);
3559                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3560                 memcpy(dest, &space, sizeof(space));
3561                 space_args.total_spaces++;
3562         }
3563
3564         user_dest = (struct btrfs_ioctl_space_info __user *)
3565                 (arg + sizeof(struct btrfs_ioctl_space_args));
3566
3567         if (copy_to_user(user_dest, dest_orig, alloc_size))
3568                 ret = -EFAULT;
3569
3570         kfree(dest_orig);
3571 out:
3572         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3573                 ret = -EFAULT;
3574
3575         return ret;
3576 }
3577
3578 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3579                                             void __user *argp)
3580 {
3581         struct btrfs_trans_handle *trans;
3582         u64 transid;
3583         int ret;
3584
3585         trans = btrfs_attach_transaction_barrier(root);
3586         if (IS_ERR(trans)) {
3587                 if (PTR_ERR(trans) != -ENOENT)
3588                         return PTR_ERR(trans);
3589
3590                 /* No running transaction, don't bother */
3591                 transid = root->fs_info->last_trans_committed;
3592                 goto out;
3593         }
3594         transid = trans->transid;
3595         ret = btrfs_commit_transaction_async(trans, 0);
3596         if (ret) {
3597                 btrfs_end_transaction(trans);
3598                 return ret;
3599         }
3600 out:
3601         if (argp)
3602                 if (copy_to_user(argp, &transid, sizeof(transid)))
3603                         return -EFAULT;
3604         return 0;
3605 }
3606
3607 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3608                                            void __user *argp)
3609 {
3610         u64 transid;
3611
3612         if (argp) {
3613                 if (copy_from_user(&transid, argp, sizeof(transid)))
3614                         return -EFAULT;
3615         } else {
3616                 transid = 0;  /* current trans */
3617         }
3618         return btrfs_wait_for_commit(fs_info, transid);
3619 }
3620
3621 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3622 {
3623         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3624         struct btrfs_ioctl_scrub_args *sa;
3625         int ret;
3626
3627         if (!capable(CAP_SYS_ADMIN))
3628                 return -EPERM;
3629
3630         sa = memdup_user(arg, sizeof(*sa));
3631         if (IS_ERR(sa))
3632                 return PTR_ERR(sa);
3633
3634         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3635                 ret = mnt_want_write_file(file);
3636                 if (ret)
3637                         goto out;
3638         }
3639
3640         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3641                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3642                               0);
3643
3644         /*
3645          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3646          * error. This is important as it allows user space to know how much
3647          * progress scrub has done. For example, if scrub is canceled we get
3648          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3649          * space. Later user space can inspect the progress from the structure
3650          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3651          * previously (btrfs-progs does this).
3652          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3653          * then return -EFAULT to signal the structure was not copied or it may
3654          * be corrupt and unreliable due to a partial copy.
3655          */
3656         if (copy_to_user(arg, sa, sizeof(*sa)))
3657                 ret = -EFAULT;
3658
3659         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3660                 mnt_drop_write_file(file);
3661 out:
3662         kfree(sa);
3663         return ret;
3664 }
3665
3666 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3667 {
3668         if (!capable(CAP_SYS_ADMIN))
3669                 return -EPERM;
3670
3671         return btrfs_scrub_cancel(fs_info);
3672 }
3673
3674 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3675                                        void __user *arg)
3676 {
3677         struct btrfs_ioctl_scrub_args *sa;
3678         int ret;
3679
3680         if (!capable(CAP_SYS_ADMIN))
3681                 return -EPERM;
3682
3683         sa = memdup_user(arg, sizeof(*sa));
3684         if (IS_ERR(sa))
3685                 return PTR_ERR(sa);
3686
3687         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3688
3689         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3690                 ret = -EFAULT;
3691
3692         kfree(sa);
3693         return ret;
3694 }
3695
3696 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3697                                       void __user *arg)
3698 {
3699         struct btrfs_ioctl_get_dev_stats *sa;
3700         int ret;
3701
3702         sa = memdup_user(arg, sizeof(*sa));
3703         if (IS_ERR(sa))
3704                 return PTR_ERR(sa);
3705
3706         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3707                 kfree(sa);
3708                 return -EPERM;
3709         }
3710
3711         ret = btrfs_get_dev_stats(fs_info, sa);
3712
3713         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3714                 ret = -EFAULT;
3715
3716         kfree(sa);
3717         return ret;
3718 }
3719
3720 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3721                                     void __user *arg)
3722 {
3723         struct btrfs_ioctl_dev_replace_args *p;
3724         int ret;
3725
3726         if (!capable(CAP_SYS_ADMIN))
3727                 return -EPERM;
3728
3729         p = memdup_user(arg, sizeof(*p));
3730         if (IS_ERR(p))
3731                 return PTR_ERR(p);
3732
3733         switch (p->cmd) {
3734         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3735                 if (sb_rdonly(fs_info->sb)) {
3736                         ret = -EROFS;
3737                         goto out;
3738                 }
3739                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3740                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3741                 } else {
3742                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3743                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3744                 }
3745                 break;
3746         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3747                 btrfs_dev_replace_status(fs_info, p);
3748                 ret = 0;
3749                 break;
3750         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3751                 p->result = btrfs_dev_replace_cancel(fs_info);
3752                 ret = 0;
3753                 break;
3754         default:
3755                 ret = -EINVAL;
3756                 break;
3757         }
3758
3759         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3760                 ret = -EFAULT;
3761 out:
3762         kfree(p);
3763         return ret;
3764 }
3765
3766 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3767 {
3768         int ret = 0;
3769         int i;
3770         u64 rel_ptr;
3771         int size;
3772         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3773         struct inode_fs_paths *ipath = NULL;
3774         struct btrfs_path *path;
3775
3776         if (!capable(CAP_DAC_READ_SEARCH))
3777                 return -EPERM;
3778
3779         path = btrfs_alloc_path();
3780         if (!path) {
3781                 ret = -ENOMEM;
3782                 goto out;
3783         }
3784
3785         ipa = memdup_user(arg, sizeof(*ipa));
3786         if (IS_ERR(ipa)) {
3787                 ret = PTR_ERR(ipa);
3788                 ipa = NULL;
3789                 goto out;
3790         }
3791
3792         size = min_t(u32, ipa->size, 4096);
3793         ipath = init_ipath(size, root, path);
3794         if (IS_ERR(ipath)) {
3795                 ret = PTR_ERR(ipath);
3796                 ipath = NULL;
3797                 goto out;
3798         }
3799
3800         ret = paths_from_inode(ipa->inum, ipath);
3801         if (ret < 0)
3802                 goto out;
3803
3804         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3805                 rel_ptr = ipath->fspath->val[i] -
3806                           (u64)(unsigned long)ipath->fspath->val;
3807                 ipath->fspath->val[i] = rel_ptr;
3808         }
3809
3810         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3811                            ipath->fspath, size);
3812         if (ret) {
3813                 ret = -EFAULT;
3814                 goto out;
3815         }
3816
3817 out:
3818         btrfs_free_path(path);
3819         free_ipath(ipath);
3820         kfree(ipa);
3821
3822         return ret;
3823 }
3824
3825 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3826 {
3827         struct btrfs_data_container *inodes = ctx;
3828         const size_t c = 3 * sizeof(u64);
3829
3830         if (inodes->bytes_left >= c) {
3831                 inodes->bytes_left -= c;
3832                 inodes->val[inodes->elem_cnt] = inum;
3833                 inodes->val[inodes->elem_cnt + 1] = offset;
3834                 inodes->val[inodes->elem_cnt + 2] = root;
3835                 inodes->elem_cnt += 3;
3836         } else {
3837                 inodes->bytes_missing += c - inodes->bytes_left;
3838                 inodes->bytes_left = 0;
3839                 inodes->elem_missed += 3;
3840         }
3841
3842         return 0;
3843 }
3844
3845 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3846                                         void __user *arg, int version)
3847 {
3848         int ret = 0;
3849         int size;
3850         struct btrfs_ioctl_logical_ino_args *loi;
3851         struct btrfs_data_container *inodes = NULL;
3852         struct btrfs_path *path = NULL;
3853         bool ignore_offset;
3854
3855         if (!capable(CAP_SYS_ADMIN))
3856                 return -EPERM;
3857
3858         loi = memdup_user(arg, sizeof(*loi));
3859         if (IS_ERR(loi))
3860                 return PTR_ERR(loi);
3861
3862         if (version == 1) {
3863                 ignore_offset = false;
3864                 size = min_t(u32, loi->size, SZ_64K);
3865         } else {
3866                 /* All reserved bits must be 0 for now */
3867                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3868                         ret = -EINVAL;
3869                         goto out_loi;
3870                 }
3871                 /* Only accept flags we have defined so far */
3872                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3873                         ret = -EINVAL;
3874                         goto out_loi;
3875                 }
3876                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3877                 size = min_t(u32, loi->size, SZ_16M);
3878         }
3879
3880         path = btrfs_alloc_path();
3881         if (!path) {
3882                 ret = -ENOMEM;
3883                 goto out;
3884         }
3885
3886         inodes = init_data_container(size);
3887         if (IS_ERR(inodes)) {
3888                 ret = PTR_ERR(inodes);
3889                 inodes = NULL;
3890                 goto out;
3891         }
3892
3893         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3894                                           build_ino_list, inodes, ignore_offset);
3895         if (ret == -EINVAL)
3896                 ret = -ENOENT;
3897         if (ret < 0)
3898                 goto out;
3899
3900         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3901                            size);
3902         if (ret)
3903                 ret = -EFAULT;
3904
3905 out:
3906         btrfs_free_path(path);
3907         kvfree(inodes);
3908 out_loi:
3909         kfree(loi);
3910
3911         return ret;
3912 }
3913
3914 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3915                                struct btrfs_ioctl_balance_args *bargs)
3916 {
3917         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3918
3919         bargs->flags = bctl->flags;
3920
3921         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3922                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3923         if (atomic_read(&fs_info->balance_pause_req))
3924                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3925         if (atomic_read(&fs_info->balance_cancel_req))
3926                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3927
3928         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3929         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3930         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3931
3932         spin_lock(&fs_info->balance_lock);
3933         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3934         spin_unlock(&fs_info->balance_lock);
3935 }
3936
3937 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3938 {
3939         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3940         struct btrfs_fs_info *fs_info = root->fs_info;
3941         struct btrfs_ioctl_balance_args *bargs;
3942         struct btrfs_balance_control *bctl;
3943         bool need_unlock; /* for mut. excl. ops lock */
3944         int ret;
3945
3946         if (!capable(CAP_SYS_ADMIN))
3947                 return -EPERM;
3948
3949         ret = mnt_want_write_file(file);
3950         if (ret)
3951                 return ret;
3952
3953 again:
3954         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3955                 mutex_lock(&fs_info->balance_mutex);
3956                 need_unlock = true;
3957                 goto locked;
3958         }
3959
3960         /*
3961          * mut. excl. ops lock is locked.  Three possibilities:
3962          *   (1) some other op is running
3963          *   (2) balance is running
3964          *   (3) balance is paused -- special case (think resume)
3965          */
3966         mutex_lock(&fs_info->balance_mutex);
3967         if (fs_info->balance_ctl) {
3968                 /* this is either (2) or (3) */
3969                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3970                         mutex_unlock(&fs_info->balance_mutex);
3971                         /*
3972                          * Lock released to allow other waiters to continue,
3973                          * we'll reexamine the status again.
3974                          */
3975                         mutex_lock(&fs_info->balance_mutex);
3976
3977                         if (fs_info->balance_ctl &&
3978                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3979                                 /* this is (3) */
3980                                 need_unlock = false;
3981                                 goto locked;
3982                         }
3983
3984                         mutex_unlock(&fs_info->balance_mutex);
3985                         goto again;
3986                 } else {
3987                         /* this is (2) */
3988                         mutex_unlock(&fs_info->balance_mutex);
3989                         ret = -EINPROGRESS;
3990                         goto out;
3991                 }
3992         } else {
3993                 /* this is (1) */
3994                 mutex_unlock(&fs_info->balance_mutex);
3995                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3996                 goto out;
3997         }
3998
3999 locked:
4000         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4001
4002         if (arg) {
4003                 bargs = memdup_user(arg, sizeof(*bargs));
4004                 if (IS_ERR(bargs)) {
4005                         ret = PTR_ERR(bargs);
4006                         goto out_unlock;
4007                 }
4008
4009                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4010                         if (!fs_info->balance_ctl) {
4011                                 ret = -ENOTCONN;
4012                                 goto out_bargs;
4013                         }
4014
4015                         bctl = fs_info->balance_ctl;
4016                         spin_lock(&fs_info->balance_lock);
4017                         bctl->flags |= BTRFS_BALANCE_RESUME;
4018                         spin_unlock(&fs_info->balance_lock);
4019
4020                         goto do_balance;
4021                 }
4022         } else {
4023                 bargs = NULL;
4024         }
4025
4026         if (fs_info->balance_ctl) {
4027                 ret = -EINPROGRESS;
4028                 goto out_bargs;
4029         }
4030
4031         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4032         if (!bctl) {
4033                 ret = -ENOMEM;
4034                 goto out_bargs;
4035         }
4036
4037         if (arg) {
4038                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4039                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4040                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4041
4042                 bctl->flags = bargs->flags;
4043         } else {
4044                 /* balance everything - no filters */
4045                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4046         }
4047
4048         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4049                 ret = -EINVAL;
4050                 goto out_bctl;
4051         }
4052
4053 do_balance:
4054         /*
4055          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4056          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4057          * restriper was paused all the way until unmount, in free_fs_info.
4058          * The flag should be cleared after reset_balance_state.
4059          */
4060         need_unlock = false;
4061
4062         ret = btrfs_balance(fs_info, bctl, bargs);
4063         bctl = NULL;
4064
4065         if ((ret == 0 || ret == -ECANCELED) && arg) {
4066                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4067                         ret = -EFAULT;
4068         }
4069
4070 out_bctl:
4071         kfree(bctl);
4072 out_bargs:
4073         kfree(bargs);
4074 out_unlock:
4075         mutex_unlock(&fs_info->balance_mutex);
4076         if (need_unlock)
4077                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4078 out:
4079         mnt_drop_write_file(file);
4080         return ret;
4081 }
4082
4083 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4084 {
4085         if (!capable(CAP_SYS_ADMIN))
4086                 return -EPERM;
4087
4088         switch (cmd) {
4089         case BTRFS_BALANCE_CTL_PAUSE:
4090                 return btrfs_pause_balance(fs_info);
4091         case BTRFS_BALANCE_CTL_CANCEL:
4092                 return btrfs_cancel_balance(fs_info);
4093         }
4094
4095         return -EINVAL;
4096 }
4097
4098 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4099                                          void __user *arg)
4100 {
4101         struct btrfs_ioctl_balance_args *bargs;
4102         int ret = 0;
4103
4104         if (!capable(CAP_SYS_ADMIN))
4105                 return -EPERM;
4106
4107         mutex_lock(&fs_info->balance_mutex);
4108         if (!fs_info->balance_ctl) {
4109                 ret = -ENOTCONN;
4110                 goto out;
4111         }
4112
4113         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4114         if (!bargs) {
4115                 ret = -ENOMEM;
4116                 goto out;
4117         }
4118
4119         btrfs_update_ioctl_balance_args(fs_info, bargs);
4120
4121         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4122                 ret = -EFAULT;
4123
4124         kfree(bargs);
4125 out:
4126         mutex_unlock(&fs_info->balance_mutex);
4127         return ret;
4128 }
4129
4130 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4131 {
4132         struct inode *inode = file_inode(file);
4133         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4134         struct btrfs_ioctl_quota_ctl_args *sa;
4135         int ret;
4136
4137         if (!capable(CAP_SYS_ADMIN))
4138                 return -EPERM;
4139
4140         ret = mnt_want_write_file(file);
4141         if (ret)
4142                 return ret;
4143
4144         sa = memdup_user(arg, sizeof(*sa));
4145         if (IS_ERR(sa)) {
4146                 ret = PTR_ERR(sa);
4147                 goto drop_write;
4148         }
4149
4150         down_write(&fs_info->subvol_sem);
4151
4152         switch (sa->cmd) {
4153         case BTRFS_QUOTA_CTL_ENABLE:
4154                 ret = btrfs_quota_enable(fs_info);
4155                 break;
4156         case BTRFS_QUOTA_CTL_DISABLE:
4157                 ret = btrfs_quota_disable(fs_info);
4158                 break;
4159         default:
4160                 ret = -EINVAL;
4161                 break;
4162         }
4163
4164         kfree(sa);
4165         up_write(&fs_info->subvol_sem);
4166 drop_write:
4167         mnt_drop_write_file(file);
4168         return ret;
4169 }
4170
4171 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4172 {
4173         struct inode *inode = file_inode(file);
4174         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4175         struct btrfs_root *root = BTRFS_I(inode)->root;
4176         struct btrfs_ioctl_qgroup_assign_args *sa;
4177         struct btrfs_trans_handle *trans;
4178         int ret;
4179         int err;
4180
4181         if (!capable(CAP_SYS_ADMIN))
4182                 return -EPERM;
4183
4184         ret = mnt_want_write_file(file);
4185         if (ret)
4186                 return ret;
4187
4188         sa = memdup_user(arg, sizeof(*sa));
4189         if (IS_ERR(sa)) {
4190                 ret = PTR_ERR(sa);
4191                 goto drop_write;
4192         }
4193
4194         trans = btrfs_join_transaction(root);
4195         if (IS_ERR(trans)) {
4196                 ret = PTR_ERR(trans);
4197                 goto out;
4198         }
4199
4200         if (sa->assign) {
4201                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4202         } else {
4203                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4204         }
4205
4206         /* update qgroup status and info */
4207         err = btrfs_run_qgroups(trans);
4208         if (err < 0)
4209                 btrfs_handle_fs_error(fs_info, err,
4210                                       "failed to update qgroup status and info");
4211         err = btrfs_end_transaction(trans);
4212         if (err && !ret)
4213                 ret = err;
4214
4215 out:
4216         kfree(sa);
4217 drop_write:
4218         mnt_drop_write_file(file);
4219         return ret;
4220 }
4221
4222 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4223 {
4224         struct inode *inode = file_inode(file);
4225         struct btrfs_root *root = BTRFS_I(inode)->root;
4226         struct btrfs_ioctl_qgroup_create_args *sa;
4227         struct btrfs_trans_handle *trans;
4228         int ret;
4229         int err;
4230
4231         if (!capable(CAP_SYS_ADMIN))
4232                 return -EPERM;
4233
4234         ret = mnt_want_write_file(file);
4235         if (ret)
4236                 return ret;
4237
4238         sa = memdup_user(arg, sizeof(*sa));
4239         if (IS_ERR(sa)) {
4240                 ret = PTR_ERR(sa);
4241                 goto drop_write;
4242         }
4243
4244         if (!sa->qgroupid) {
4245                 ret = -EINVAL;
4246                 goto out;
4247         }
4248
4249         trans = btrfs_join_transaction(root);
4250         if (IS_ERR(trans)) {
4251                 ret = PTR_ERR(trans);
4252                 goto out;
4253         }
4254
4255         if (sa->create) {
4256                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4257         } else {
4258                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4259         }
4260
4261         err = btrfs_end_transaction(trans);
4262         if (err && !ret)
4263                 ret = err;
4264
4265 out:
4266         kfree(sa);
4267 drop_write:
4268         mnt_drop_write_file(file);
4269         return ret;
4270 }
4271
4272 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4273 {
4274         struct inode *inode = file_inode(file);
4275         struct btrfs_root *root = BTRFS_I(inode)->root;
4276         struct btrfs_ioctl_qgroup_limit_args *sa;
4277         struct btrfs_trans_handle *trans;
4278         int ret;
4279         int err;
4280         u64 qgroupid;
4281
4282         if (!capable(CAP_SYS_ADMIN))
4283                 return -EPERM;
4284
4285         ret = mnt_want_write_file(file);
4286         if (ret)
4287                 return ret;
4288
4289         sa = memdup_user(arg, sizeof(*sa));
4290         if (IS_ERR(sa)) {
4291                 ret = PTR_ERR(sa);
4292                 goto drop_write;
4293         }
4294
4295         trans = btrfs_join_transaction(root);
4296         if (IS_ERR(trans)) {
4297                 ret = PTR_ERR(trans);
4298                 goto out;
4299         }
4300
4301         qgroupid = sa->qgroupid;
4302         if (!qgroupid) {
4303                 /* take the current subvol as qgroup */
4304                 qgroupid = root->root_key.objectid;
4305         }
4306
4307         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4308
4309         err = btrfs_end_transaction(trans);
4310         if (err && !ret)
4311                 ret = err;
4312
4313 out:
4314         kfree(sa);
4315 drop_write:
4316         mnt_drop_write_file(file);
4317         return ret;
4318 }
4319
4320 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4321 {
4322         struct inode *inode = file_inode(file);
4323         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4324         struct btrfs_ioctl_quota_rescan_args *qsa;
4325         int ret;
4326
4327         if (!capable(CAP_SYS_ADMIN))
4328                 return -EPERM;
4329
4330         ret = mnt_want_write_file(file);
4331         if (ret)
4332                 return ret;
4333
4334         qsa = memdup_user(arg, sizeof(*qsa));
4335         if (IS_ERR(qsa)) {
4336                 ret = PTR_ERR(qsa);
4337                 goto drop_write;
4338         }
4339
4340         if (qsa->flags) {
4341                 ret = -EINVAL;
4342                 goto out;
4343         }
4344
4345         ret = btrfs_qgroup_rescan(fs_info);
4346
4347 out:
4348         kfree(qsa);
4349 drop_write:
4350         mnt_drop_write_file(file);
4351         return ret;
4352 }
4353
4354 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4355                                                 void __user *arg)
4356 {
4357         struct btrfs_ioctl_quota_rescan_args *qsa;
4358         int ret = 0;
4359
4360         if (!capable(CAP_SYS_ADMIN))
4361                 return -EPERM;
4362
4363         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4364         if (!qsa)
4365                 return -ENOMEM;
4366
4367         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4368                 qsa->flags = 1;
4369                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4370         }
4371
4372         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4373                 ret = -EFAULT;
4374
4375         kfree(qsa);
4376         return ret;
4377 }
4378
4379 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4380                                                 void __user *arg)
4381 {
4382         if (!capable(CAP_SYS_ADMIN))
4383                 return -EPERM;
4384
4385         return btrfs_qgroup_wait_for_completion(fs_info, true);
4386 }
4387
4388 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4389                                             struct btrfs_ioctl_received_subvol_args *sa)
4390 {
4391         struct inode *inode = file_inode(file);
4392         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4393         struct btrfs_root *root = BTRFS_I(inode)->root;
4394         struct btrfs_root_item *root_item = &root->root_item;
4395         struct btrfs_trans_handle *trans;
4396         struct timespec64 ct = current_time(inode);
4397         int ret = 0;
4398         int received_uuid_changed;
4399
4400         if (!inode_owner_or_capable(inode))
4401                 return -EPERM;
4402
4403         ret = mnt_want_write_file(file);
4404         if (ret < 0)
4405                 return ret;
4406
4407         down_write(&fs_info->subvol_sem);
4408
4409         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4410                 ret = -EINVAL;
4411                 goto out;
4412         }
4413
4414         if (btrfs_root_readonly(root)) {
4415                 ret = -EROFS;
4416                 goto out;
4417         }
4418
4419         /*
4420          * 1 - root item
4421          * 2 - uuid items (received uuid + subvol uuid)
4422          */
4423         trans = btrfs_start_transaction(root, 3);
4424         if (IS_ERR(trans)) {
4425                 ret = PTR_ERR(trans);
4426                 trans = NULL;
4427                 goto out;
4428         }
4429
4430         sa->rtransid = trans->transid;
4431         sa->rtime.sec = ct.tv_sec;
4432         sa->rtime.nsec = ct.tv_nsec;
4433
4434         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4435                                        BTRFS_UUID_SIZE);
4436         if (received_uuid_changed &&
4437             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4438                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4439                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4440                                           root->root_key.objectid);
4441                 if (ret && ret != -ENOENT) {
4442                         btrfs_abort_transaction(trans, ret);
4443                         btrfs_end_transaction(trans);
4444                         goto out;
4445                 }
4446         }
4447         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4448         btrfs_set_root_stransid(root_item, sa->stransid);
4449         btrfs_set_root_rtransid(root_item, sa->rtransid);
4450         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4451         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4452         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4453         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4454
4455         ret = btrfs_update_root(trans, fs_info->tree_root,
4456                                 &root->root_key, &root->root_item);
4457         if (ret < 0) {
4458                 btrfs_end_transaction(trans);
4459                 goto out;
4460         }
4461         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4462                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4463                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4464                                           root->root_key.objectid);
4465                 if (ret < 0 && ret != -EEXIST) {
4466                         btrfs_abort_transaction(trans, ret);
4467                         btrfs_end_transaction(trans);
4468                         goto out;
4469                 }
4470         }
4471         ret = btrfs_commit_transaction(trans);
4472 out:
4473         up_write(&fs_info->subvol_sem);
4474         mnt_drop_write_file(file);
4475         return ret;
4476 }
4477
4478 #ifdef CONFIG_64BIT
4479 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4480                                                 void __user *arg)
4481 {
4482         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4483         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4484         int ret = 0;
4485
4486         args32 = memdup_user(arg, sizeof(*args32));
4487         if (IS_ERR(args32))
4488                 return PTR_ERR(args32);
4489
4490         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4491         if (!args64) {
4492                 ret = -ENOMEM;
4493                 goto out;
4494         }
4495
4496         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4497         args64->stransid = args32->stransid;
4498         args64->rtransid = args32->rtransid;
4499         args64->stime.sec = args32->stime.sec;
4500         args64->stime.nsec = args32->stime.nsec;
4501         args64->rtime.sec = args32->rtime.sec;
4502         args64->rtime.nsec = args32->rtime.nsec;
4503         args64->flags = args32->flags;
4504
4505         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4506         if (ret)
4507                 goto out;
4508
4509         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4510         args32->stransid = args64->stransid;
4511         args32->rtransid = args64->rtransid;
4512         args32->stime.sec = args64->stime.sec;
4513         args32->stime.nsec = args64->stime.nsec;
4514         args32->rtime.sec = args64->rtime.sec;
4515         args32->rtime.nsec = args64->rtime.nsec;
4516         args32->flags = args64->flags;
4517
4518         ret = copy_to_user(arg, args32, sizeof(*args32));
4519         if (ret)
4520                 ret = -EFAULT;
4521
4522 out:
4523         kfree(args32);
4524         kfree(args64);
4525         return ret;
4526 }
4527 #endif
4528
4529 static long btrfs_ioctl_set_received_subvol(struct file *file,
4530                                             void __user *arg)
4531 {
4532         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4533         int ret = 0;
4534
4535         sa = memdup_user(arg, sizeof(*sa));
4536         if (IS_ERR(sa))
4537                 return PTR_ERR(sa);
4538
4539         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4540
4541         if (ret)
4542                 goto out;
4543
4544         ret = copy_to_user(arg, sa, sizeof(*sa));
4545         if (ret)
4546                 ret = -EFAULT;
4547
4548 out:
4549         kfree(sa);
4550         return ret;
4551 }
4552
4553 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4554                                         void __user *arg)
4555 {
4556         size_t len;
4557         int ret;
4558         char label[BTRFS_LABEL_SIZE];
4559
4560         spin_lock(&fs_info->super_lock);
4561         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4562         spin_unlock(&fs_info->super_lock);
4563
4564         len = strnlen(label, BTRFS_LABEL_SIZE);
4565
4566         if (len == BTRFS_LABEL_SIZE) {
4567                 btrfs_warn(fs_info,
4568                            "label is too long, return the first %zu bytes",
4569                            --len);
4570         }
4571
4572         ret = copy_to_user(arg, label, len);
4573
4574         return ret ? -EFAULT : 0;
4575 }
4576
4577 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4578 {
4579         struct inode *inode = file_inode(file);
4580         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4581         struct btrfs_root *root = BTRFS_I(inode)->root;
4582         struct btrfs_super_block *super_block = fs_info->super_copy;
4583         struct btrfs_trans_handle *trans;
4584         char label[BTRFS_LABEL_SIZE];
4585         int ret;
4586
4587         if (!capable(CAP_SYS_ADMIN))
4588                 return -EPERM;
4589
4590         if (copy_from_user(label, arg, sizeof(label)))
4591                 return -EFAULT;
4592
4593         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4594                 btrfs_err(fs_info,
4595                           "unable to set label with more than %d bytes",
4596                           BTRFS_LABEL_SIZE - 1);
4597                 return -EINVAL;
4598         }
4599
4600         ret = mnt_want_write_file(file);
4601         if (ret)
4602                 return ret;
4603
4604         trans = btrfs_start_transaction(root, 0);
4605         if (IS_ERR(trans)) {
4606                 ret = PTR_ERR(trans);
4607                 goto out_unlock;
4608         }
4609
4610         spin_lock(&fs_info->super_lock);
4611         strcpy(super_block->label, label);
4612         spin_unlock(&fs_info->super_lock);
4613         ret = btrfs_commit_transaction(trans);
4614
4615 out_unlock:
4616         mnt_drop_write_file(file);
4617         return ret;
4618 }
4619
4620 #define INIT_FEATURE_FLAGS(suffix) \
4621         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4622           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4623           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4624
4625 int btrfs_ioctl_get_supported_features(void __user *arg)
4626 {
4627         static const struct btrfs_ioctl_feature_flags features[3] = {
4628                 INIT_FEATURE_FLAGS(SUPP),
4629                 INIT_FEATURE_FLAGS(SAFE_SET),
4630                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4631         };
4632
4633         if (copy_to_user(arg, &features, sizeof(features)))
4634                 return -EFAULT;
4635
4636         return 0;
4637 }
4638
4639 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4640                                         void __user *arg)
4641 {
4642         struct btrfs_super_block *super_block = fs_info->super_copy;
4643         struct btrfs_ioctl_feature_flags features;
4644
4645         features.compat_flags = btrfs_super_compat_flags(super_block);
4646         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4647         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4648
4649         if (copy_to_user(arg, &features, sizeof(features)))
4650                 return -EFAULT;
4651
4652         return 0;
4653 }
4654
4655 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4656                               enum btrfs_feature_set set,
4657                               u64 change_mask, u64 flags, u64 supported_flags,
4658                               u64 safe_set, u64 safe_clear)
4659 {
4660         const char *type = btrfs_feature_set_name(set);
4661         char *names;
4662         u64 disallowed, unsupported;
4663         u64 set_mask = flags & change_mask;
4664         u64 clear_mask = ~flags & change_mask;
4665
4666         unsupported = set_mask & ~supported_flags;
4667         if (unsupported) {
4668                 names = btrfs_printable_features(set, unsupported);
4669                 if (names) {
4670                         btrfs_warn(fs_info,
4671                                    "this kernel does not support the %s feature bit%s",
4672                                    names, strchr(names, ',') ? "s" : "");
4673                         kfree(names);
4674                 } else
4675                         btrfs_warn(fs_info,
4676                                    "this kernel does not support %s bits 0x%llx",
4677                                    type, unsupported);
4678                 return -EOPNOTSUPP;
4679         }
4680
4681         disallowed = set_mask & ~safe_set;
4682         if (disallowed) {
4683                 names = btrfs_printable_features(set, disallowed);
4684                 if (names) {
4685                         btrfs_warn(fs_info,
4686                                    "can't set the %s feature bit%s while mounted",
4687                                    names, strchr(names, ',') ? "s" : "");
4688                         kfree(names);
4689                 } else
4690                         btrfs_warn(fs_info,
4691                                    "can't set %s bits 0x%llx while mounted",
4692                                    type, disallowed);
4693                 return -EPERM;
4694         }
4695
4696         disallowed = clear_mask & ~safe_clear;
4697         if (disallowed) {
4698                 names = btrfs_printable_features(set, disallowed);
4699                 if (names) {
4700                         btrfs_warn(fs_info,
4701                                    "can't clear the %s feature bit%s while mounted",
4702                                    names, strchr(names, ',') ? "s" : "");
4703                         kfree(names);
4704                 } else
4705                         btrfs_warn(fs_info,
4706                                    "can't clear %s bits 0x%llx while mounted",
4707                                    type, disallowed);
4708                 return -EPERM;
4709         }
4710
4711         return 0;
4712 }
4713
4714 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4715 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4716                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4717                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4718                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4719
4720 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4721 {
4722         struct inode *inode = file_inode(file);
4723         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4724         struct btrfs_root *root = BTRFS_I(inode)->root;
4725         struct btrfs_super_block *super_block = fs_info->super_copy;
4726         struct btrfs_ioctl_feature_flags flags[2];
4727         struct btrfs_trans_handle *trans;
4728         u64 newflags;
4729         int ret;
4730
4731         if (!capable(CAP_SYS_ADMIN))
4732                 return -EPERM;
4733
4734         if (copy_from_user(flags, arg, sizeof(flags)))
4735                 return -EFAULT;
4736
4737         /* Nothing to do */
4738         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4739             !flags[0].incompat_flags)
4740                 return 0;
4741
4742         ret = check_feature(fs_info, flags[0].compat_flags,
4743                             flags[1].compat_flags, COMPAT);
4744         if (ret)
4745                 return ret;
4746
4747         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4748                             flags[1].compat_ro_flags, COMPAT_RO);
4749         if (ret)
4750                 return ret;
4751
4752         ret = check_feature(fs_info, flags[0].incompat_flags,
4753                             flags[1].incompat_flags, INCOMPAT);
4754         if (ret)
4755                 return ret;
4756
4757         ret = mnt_want_write_file(file);
4758         if (ret)
4759                 return ret;
4760
4761         trans = btrfs_start_transaction(root, 0);
4762         if (IS_ERR(trans)) {
4763                 ret = PTR_ERR(trans);
4764                 goto out_drop_write;
4765         }
4766
4767         spin_lock(&fs_info->super_lock);
4768         newflags = btrfs_super_compat_flags(super_block);
4769         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4770         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4771         btrfs_set_super_compat_flags(super_block, newflags);
4772
4773         newflags = btrfs_super_compat_ro_flags(super_block);
4774         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4775         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4776         btrfs_set_super_compat_ro_flags(super_block, newflags);
4777
4778         newflags = btrfs_super_incompat_flags(super_block);
4779         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4780         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4781         btrfs_set_super_incompat_flags(super_block, newflags);
4782         spin_unlock(&fs_info->super_lock);
4783
4784         ret = btrfs_commit_transaction(trans);
4785 out_drop_write:
4786         mnt_drop_write_file(file);
4787
4788         return ret;
4789 }
4790
4791 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4792 {
4793         struct btrfs_ioctl_send_args *arg;
4794         int ret;
4795
4796         if (compat) {
4797 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4798                 struct btrfs_ioctl_send_args_32 args32;
4799
4800                 ret = copy_from_user(&args32, argp, sizeof(args32));
4801                 if (ret)
4802                         return -EFAULT;
4803                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4804                 if (!arg)
4805                         return -ENOMEM;
4806                 arg->send_fd = args32.send_fd;
4807                 arg->clone_sources_count = args32.clone_sources_count;
4808                 arg->clone_sources = compat_ptr(args32.clone_sources);
4809                 arg->parent_root = args32.parent_root;
4810                 arg->flags = args32.flags;
4811                 memcpy(arg->reserved, args32.reserved,
4812                        sizeof(args32.reserved));
4813 #else
4814                 return -ENOTTY;
4815 #endif
4816         } else {
4817                 arg = memdup_user(argp, sizeof(*arg));
4818                 if (IS_ERR(arg))
4819                         return PTR_ERR(arg);
4820         }
4821         ret = btrfs_ioctl_send(file, arg);
4822         kfree(arg);
4823         return ret;
4824 }
4825
4826 long btrfs_ioctl(struct file *file, unsigned int
4827                 cmd, unsigned long arg)
4828 {
4829         struct inode *inode = file_inode(file);
4830         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4831         struct btrfs_root *root = BTRFS_I(inode)->root;
4832         void __user *argp = (void __user *)arg;
4833
4834         switch (cmd) {
4835         case FS_IOC_GETFLAGS:
4836                 return btrfs_ioctl_getflags(file, argp);
4837         case FS_IOC_SETFLAGS:
4838                 return btrfs_ioctl_setflags(file, argp);
4839         case FS_IOC_GETVERSION:
4840                 return btrfs_ioctl_getversion(file, argp);
4841         case FS_IOC_GETFSLABEL:
4842                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4843         case FS_IOC_SETFSLABEL:
4844                 return btrfs_ioctl_set_fslabel(file, argp);
4845         case FITRIM:
4846                 return btrfs_ioctl_fitrim(fs_info, argp);
4847         case BTRFS_IOC_SNAP_CREATE:
4848                 return btrfs_ioctl_snap_create(file, argp, 0);
4849         case BTRFS_IOC_SNAP_CREATE_V2:
4850                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4851         case BTRFS_IOC_SUBVOL_CREATE:
4852                 return btrfs_ioctl_snap_create(file, argp, 1);
4853         case BTRFS_IOC_SUBVOL_CREATE_V2:
4854                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4855         case BTRFS_IOC_SNAP_DESTROY:
4856                 return btrfs_ioctl_snap_destroy(file, argp, false);
4857         case BTRFS_IOC_SNAP_DESTROY_V2:
4858                 return btrfs_ioctl_snap_destroy(file, argp, true);
4859         case BTRFS_IOC_SUBVOL_GETFLAGS:
4860                 return btrfs_ioctl_subvol_getflags(file, argp);
4861         case BTRFS_IOC_SUBVOL_SETFLAGS:
4862                 return btrfs_ioctl_subvol_setflags(file, argp);
4863         case BTRFS_IOC_DEFAULT_SUBVOL:
4864                 return btrfs_ioctl_default_subvol(file, argp);
4865         case BTRFS_IOC_DEFRAG:
4866                 return btrfs_ioctl_defrag(file, NULL);
4867         case BTRFS_IOC_DEFRAG_RANGE:
4868                 return btrfs_ioctl_defrag(file, argp);
4869         case BTRFS_IOC_RESIZE:
4870                 return btrfs_ioctl_resize(file, argp);
4871         case BTRFS_IOC_ADD_DEV:
4872                 return btrfs_ioctl_add_dev(fs_info, argp);
4873         case BTRFS_IOC_RM_DEV:
4874                 return btrfs_ioctl_rm_dev(file, argp);
4875         case BTRFS_IOC_RM_DEV_V2:
4876                 return btrfs_ioctl_rm_dev_v2(file, argp);
4877         case BTRFS_IOC_FS_INFO:
4878                 return btrfs_ioctl_fs_info(fs_info, argp);
4879         case BTRFS_IOC_DEV_INFO:
4880                 return btrfs_ioctl_dev_info(fs_info, argp);
4881         case BTRFS_IOC_BALANCE:
4882                 return btrfs_ioctl_balance(file, NULL);
4883         case BTRFS_IOC_TREE_SEARCH:
4884                 return btrfs_ioctl_tree_search(file, argp);
4885         case BTRFS_IOC_TREE_SEARCH_V2:
4886                 return btrfs_ioctl_tree_search_v2(file, argp);
4887         case BTRFS_IOC_INO_LOOKUP:
4888                 return btrfs_ioctl_ino_lookup(file, argp);
4889         case BTRFS_IOC_INO_PATHS:
4890                 return btrfs_ioctl_ino_to_path(root, argp);
4891         case BTRFS_IOC_LOGICAL_INO:
4892                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4893         case BTRFS_IOC_LOGICAL_INO_V2:
4894                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4895         case BTRFS_IOC_SPACE_INFO:
4896                 return btrfs_ioctl_space_info(fs_info, argp);
4897         case BTRFS_IOC_SYNC: {
4898                 int ret;
4899
4900                 ret = btrfs_start_delalloc_roots(fs_info, -1);
4901                 if (ret)
4902                         return ret;
4903                 ret = btrfs_sync_fs(inode->i_sb, 1);
4904                 /*
4905                  * The transaction thread may want to do more work,
4906                  * namely it pokes the cleaner kthread that will start
4907                  * processing uncleaned subvols.
4908                  */
4909                 wake_up_process(fs_info->transaction_kthread);
4910                 return ret;
4911         }
4912         case BTRFS_IOC_START_SYNC:
4913                 return btrfs_ioctl_start_sync(root, argp);
4914         case BTRFS_IOC_WAIT_SYNC:
4915                 return btrfs_ioctl_wait_sync(fs_info, argp);
4916         case BTRFS_IOC_SCRUB:
4917                 return btrfs_ioctl_scrub(file, argp);
4918         case BTRFS_IOC_SCRUB_CANCEL:
4919                 return btrfs_ioctl_scrub_cancel(fs_info);
4920         case BTRFS_IOC_SCRUB_PROGRESS:
4921                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4922         case BTRFS_IOC_BALANCE_V2:
4923                 return btrfs_ioctl_balance(file, argp);
4924         case BTRFS_IOC_BALANCE_CTL:
4925                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4926         case BTRFS_IOC_BALANCE_PROGRESS:
4927                 return btrfs_ioctl_balance_progress(fs_info, argp);
4928         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4929                 return btrfs_ioctl_set_received_subvol(file, argp);
4930 #ifdef CONFIG_64BIT
4931         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4932                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4933 #endif
4934         case BTRFS_IOC_SEND:
4935                 return _btrfs_ioctl_send(file, argp, false);
4936 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4937         case BTRFS_IOC_SEND_32:
4938                 return _btrfs_ioctl_send(file, argp, true);
4939 #endif
4940         case BTRFS_IOC_GET_DEV_STATS:
4941                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4942         case BTRFS_IOC_QUOTA_CTL:
4943                 return btrfs_ioctl_quota_ctl(file, argp);
4944         case BTRFS_IOC_QGROUP_ASSIGN:
4945                 return btrfs_ioctl_qgroup_assign(file, argp);
4946         case BTRFS_IOC_QGROUP_CREATE:
4947                 return btrfs_ioctl_qgroup_create(file, argp);
4948         case BTRFS_IOC_QGROUP_LIMIT:
4949                 return btrfs_ioctl_qgroup_limit(file, argp);
4950         case BTRFS_IOC_QUOTA_RESCAN:
4951                 return btrfs_ioctl_quota_rescan(file, argp);
4952         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4953                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4954         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4955                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4956         case BTRFS_IOC_DEV_REPLACE:
4957                 return btrfs_ioctl_dev_replace(fs_info, argp);
4958         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4959                 return btrfs_ioctl_get_supported_features(argp);
4960         case BTRFS_IOC_GET_FEATURES:
4961                 return btrfs_ioctl_get_features(fs_info, argp);
4962         case BTRFS_IOC_SET_FEATURES:
4963                 return btrfs_ioctl_set_features(file, argp);
4964         case FS_IOC_FSGETXATTR:
4965                 return btrfs_ioctl_fsgetxattr(file, argp);
4966         case FS_IOC_FSSETXATTR:
4967                 return btrfs_ioctl_fssetxattr(file, argp);
4968         case BTRFS_IOC_GET_SUBVOL_INFO:
4969                 return btrfs_ioctl_get_subvol_info(file, argp);
4970         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4971                 return btrfs_ioctl_get_subvol_rootref(file, argp);
4972         case BTRFS_IOC_INO_LOOKUP_USER:
4973                 return btrfs_ioctl_ino_lookup_user(file, argp);
4974         }
4975
4976         return -ENOTTY;
4977 }
4978
4979 #ifdef CONFIG_COMPAT
4980 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4981 {
4982         /*
4983          * These all access 32-bit values anyway so no further
4984          * handling is necessary.
4985          */
4986         switch (cmd) {
4987         case FS_IOC32_GETFLAGS:
4988                 cmd = FS_IOC_GETFLAGS;
4989                 break;
4990         case FS_IOC32_SETFLAGS:
4991                 cmd = FS_IOC_SETFLAGS;
4992                 break;
4993         case FS_IOC32_GETVERSION:
4994                 cmd = FS_IOC_GETVERSION;
4995                 break;
4996         }
4997
4998         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
4999 }
5000 #endif