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 "transaction.h"
  32 #include "btrfs_inode.h"
  33 #include "print-tree.h"
  34 #include "volumes.h"
  35 #include "locking.h"
  36 #include "inode-map.h"
  37 #include "backref.h"
  38 #include "rcu-string.h"
  39 #include "send.h"
  40 #include "dev-replace.h"
  41 #include "props.h"
  42 #include "sysfs.h"
  43 #include "qgroup.h"
  44 #include "tree-log.h"
  45 #include "compression.h"
  46
  47 #ifdef CONFIG_64BIT
  48 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  49  * structures are incorrect, as the timespec structure from userspace
  50  * is 4 bytes too small. We define these alternatives here to teach
  51  * the kernel about the 32-bit struct packing.
  52  */
  53 struct btrfs_ioctl_timespec_32 {
  54         __u64 sec;
  55         __u32 nsec;
  56 } __attribute__ ((__packed__));
  57
  58 struct btrfs_ioctl_received_subvol_args_32 {
  59         char    uuid[BTRFS_UUID_SIZE];  /* in */
  60         __u64   stransid;               /* in */
  61         __u64   rtransid;               /* out */
  62         struct btrfs_ioctl_timespec_32 stime; /* in */
  63         struct btrfs_ioctl_timespec_32 rtime; /* out */
  64         __u64   flags;                  /* in */
  65         __u64   reserved[16];           /* in */
  66 } __attribute__ ((__packed__));
  67
  68 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  69                                 struct btrfs_ioctl_received_subvol_args_32)
  70 #endif
  71
  72 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  73 struct btrfs_ioctl_send_args_32 {
  74         __s64 send_fd;                  /* in */
  75         __u64 clone_sources_count;      /* in */
  76         compat_uptr_t clone_sources;    /* in */
  77         __u64 parent_root;              /* in */
  78         __u64 flags;                    /* in */
  79         __u64 reserved[4];              /* in */
  80 } __attribute__ ((__packed__));
  81
  82 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  83                                struct btrfs_ioctl_send_args_32)
  84 #endif
  85
  86 static int btrfs_clone(struct inode *src, struct inode *inode,
  87                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  88                        int no_time_update);
  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 /* Check if @flags are a supported and valid set of FS_*_FL flags */
 168 static int check_fsflags(unsigned int flags)
 169 {
 170         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 171                       FS_NOATIME_FL | FS_NODUMP_FL | \
 172                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 173                       FS_NOCOMP_FL | FS_COMPR_FL |
 174                       FS_NOCOW_FL))
 175                 return -EOPNOTSUPP;
 176
 177         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 178                 return -EINVAL;
 179
 180         return 0;
 181 }
 182
 183 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 184 {
 185         struct inode *inode = file_inode(file);
 186         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 187         struct btrfs_inode *binode = BTRFS_I(inode);
 188         struct btrfs_root *root = binode->root;
 189         struct btrfs_trans_handle *trans;
 190         unsigned int fsflags, old_fsflags;
 191         int ret;
 192         u64 old_flags;
 193         unsigned int old_i_flags;
 194         umode_t mode;
 195
 196         if (!inode_owner_or_capable(inode))
 197                 return -EPERM;
 198
 199         if (btrfs_root_readonly(root))
 200                 return -EROFS;
 201
 202         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 203                 return -EFAULT;
 204
 205         ret = check_fsflags(fsflags);
 206         if (ret)
 207                 return ret;
 208
 209         ret = mnt_want_write_file(file);
 210         if (ret)
 211                 return ret;
 212
 213         inode_lock(inode);
 214
 215         old_flags = binode->flags;
 216         old_i_flags = inode->i_flags;
 217         mode = inode->i_mode;
 218
 219         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 220         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 221         if ((fsflags ^ old_fsflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 222                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 223                         ret = -EPERM;
 224                         goto out_unlock;
 225                 }
 226         }
 227
 228         if (fsflags & FS_SYNC_FL)
 229                 binode->flags |= BTRFS_INODE_SYNC;
 230         else
 231                 binode->flags &= ~BTRFS_INODE_SYNC;
 232         if (fsflags & FS_IMMUTABLE_FL)
 233                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 234         else
 235                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 236         if (fsflags & FS_APPEND_FL)
 237                 binode->flags |= BTRFS_INODE_APPEND;
 238         else
 239                 binode->flags &= ~BTRFS_INODE_APPEND;
 240         if (fsflags & FS_NODUMP_FL)
 241                 binode->flags |= BTRFS_INODE_NODUMP;
 242         else
 243                 binode->flags &= ~BTRFS_INODE_NODUMP;
 244         if (fsflags & FS_NOATIME_FL)
 245                 binode->flags |= BTRFS_INODE_NOATIME;
 246         else
 247                 binode->flags &= ~BTRFS_INODE_NOATIME;
 248         if (fsflags & FS_DIRSYNC_FL)
 249                 binode->flags |= BTRFS_INODE_DIRSYNC;
 250         else
 251                 binode->flags &= ~BTRFS_INODE_DIRSYNC;
 252         if (fsflags & FS_NOCOW_FL) {
 253                 if (S_ISREG(mode)) {
 254                         /*
 255                          * It's safe to turn csums off here, no extents exist.
 256                          * Otherwise we want the flag to reflect the real COW
 257                          * status of the file and will not set it.
 258                          */
 259                         if (inode->i_size == 0)
 260                                 binode->flags |= BTRFS_INODE_NODATACOW
 261                                               | BTRFS_INODE_NODATASUM;
 262                 } else {
 263                         binode->flags |= BTRFS_INODE_NODATACOW;
 264                 }
 265         } else {
 266                 /*
 267                  * Revert back under same assumptions as above
 268                  */
 269                 if (S_ISREG(mode)) {
 270                         if (inode->i_size == 0)
 271                                 binode->flags &= ~(BTRFS_INODE_NODATACOW
 272                                              | BTRFS_INODE_NODATASUM);
 273                 } else {
 274                         binode->flags &= ~BTRFS_INODE_NODATACOW;
 275                 }
 276         }
 277
 278         /*
 279          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 280          * flag may be changed automatically if compression code won't make
 281          * things smaller.
 282          */
 283         if (fsflags & FS_NOCOMP_FL) {
 284                 binode->flags &= ~BTRFS_INODE_COMPRESS;
 285                 binode->flags |= BTRFS_INODE_NOCOMPRESS;
 286
 287                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 288                 if (ret && ret != -ENODATA)
 289                         goto out_drop;
 290         } else if (fsflags & FS_COMPR_FL) {
 291                 const char *comp;
 292
 293                 binode->flags |= BTRFS_INODE_COMPRESS;
 294                 binode->flags &= ~BTRFS_INODE_NOCOMPRESS;
 295
 296                 comp = btrfs_compress_type2str(fs_info->compress_type);
 297                 if (!comp || comp[0] == 0)
 298                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 299
 300                 ret = btrfs_set_prop(inode, "btrfs.compression",
 301                                      comp, strlen(comp), 0);
 302                 if (ret)
 303                         goto out_drop;
 304
 305         } else {
 306                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 307                 if (ret && ret != -ENODATA)
 308                         goto out_drop;
 309                 binode->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 310         }
 311
 312         trans = btrfs_start_transaction(root, 1);
 313         if (IS_ERR(trans)) {
 314                 ret = PTR_ERR(trans);
 315                 goto out_drop;
 316         }
 317
 318         btrfs_sync_inode_flags_to_i_flags(inode);
 319         inode_inc_iversion(inode);
 320         inode->i_ctime = current_time(inode);
 321         ret = btrfs_update_inode(trans, root, inode);
 322
 323         btrfs_end_transaction(trans);
 324  out_drop:
 325         if (ret) {
 326                 binode->flags = old_flags;
 327                 inode->i_flags = old_i_flags;
 328         }
 329
 330  out_unlock:
 331         inode_unlock(inode);
 332         mnt_drop_write_file(file);
 333         return ret;
 334 }
 335
 336 /*
 337  * Translate btrfs internal inode flags to xflags as expected by the
 338  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 339  * silently dropped.
 340  */
 341 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 342 {
 343         unsigned int xflags = 0;
 344
 345         if (flags & BTRFS_INODE_APPEND)
 346                 xflags |= FS_XFLAG_APPEND;
 347         if (flags & BTRFS_INODE_IMMUTABLE)
 348                 xflags |= FS_XFLAG_IMMUTABLE;
 349         if (flags & BTRFS_INODE_NOATIME)
 350                 xflags |= FS_XFLAG_NOATIME;
 351         if (flags & BTRFS_INODE_NODUMP)
 352                 xflags |= FS_XFLAG_NODUMP;
 353         if (flags & BTRFS_INODE_SYNC)
 354                 xflags |= FS_XFLAG_SYNC;
 355
 356         return xflags;
 357 }
 358
 359 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 360 static int check_xflags(unsigned int flags)
 361 {
 362         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 363                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 364                 return -EOPNOTSUPP;
 365         return 0;
 366 }
 367
 368 /*
 369  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 370  * are zeroed.
 371  */
 372 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 373 {
 374         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 375         struct fsxattr fa;
 376
 377         memset(&fa, 0, sizeof(fa));
 378         fa.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags);
 379
 380         if (copy_to_user(arg, &fa, sizeof(fa)))
 381                 return -EFAULT;
 382
 383         return 0;
 384 }
 385
 386 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 387 {
 388         struct inode *inode = file_inode(file);
 389         struct btrfs_inode *binode = BTRFS_I(inode);
 390         struct btrfs_root *root = binode->root;
 391         struct btrfs_trans_handle *trans;
 392         struct fsxattr fa;
 393         unsigned old_flags;
 394         unsigned old_i_flags;
 395         int ret = 0;
 396
 397         if (!inode_owner_or_capable(inode))
 398                 return -EPERM;
 399
 400         if (btrfs_root_readonly(root))
 401                 return -EROFS;
 402
 403         memset(&fa, 0, sizeof(fa));
 404         if (copy_from_user(&fa, arg, sizeof(fa)))
 405                 return -EFAULT;
 406
 407         ret = check_xflags(fa.fsx_xflags);
 408         if (ret)
 409                 return ret;
 410
 411         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 412                 return -EOPNOTSUPP;
 413
 414         ret = mnt_want_write_file(file);
 415         if (ret)
 416                 return ret;
 417
 418         inode_lock(inode);
 419
 420         old_flags = binode->flags;
 421         old_i_flags = inode->i_flags;
 422
 423         /* We need the capabilities to change append-only or immutable inode */
 424         if (((old_flags & (BTRFS_INODE_APPEND | BTRFS_INODE_IMMUTABLE)) ||
 425              (fa.fsx_xflags & (FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE))) &&
 426             !capable(CAP_LINUX_IMMUTABLE)) {
 427                 ret = -EPERM;
 428                 goto out_unlock;
 429         }
 430
 431         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 432                 binode->flags |= BTRFS_INODE_SYNC;
 433         else
 434                 binode->flags &= ~BTRFS_INODE_SYNC;
 435         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 436                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 437         else
 438                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 439         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 440                 binode->flags |= BTRFS_INODE_APPEND;
 441         else
 442                 binode->flags &= ~BTRFS_INODE_APPEND;
 443         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 444                 binode->flags |= BTRFS_INODE_NODUMP;
 445         else
 446                 binode->flags &= ~BTRFS_INODE_NODUMP;
 447         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 448                 binode->flags |= BTRFS_INODE_NOATIME;
 449         else
 450                 binode->flags &= ~BTRFS_INODE_NOATIME;
 451
 452         /* 1 item for the inode */
 453         trans = btrfs_start_transaction(root, 1);
 454         if (IS_ERR(trans)) {
 455                 ret = PTR_ERR(trans);
 456                 goto out_unlock;
 457         }
 458
 459         btrfs_sync_inode_flags_to_i_flags(inode);
 460         inode_inc_iversion(inode);
 461         inode->i_ctime = current_time(inode);
 462         ret = btrfs_update_inode(trans, root, inode);
 463
 464         btrfs_end_transaction(trans);
 465
 466 out_unlock:
 467         if (ret) {
 468                 binode->flags = old_flags;
 469                 inode->i_flags = old_i_flags;
 470         }
 471
 472         inode_unlock(inode);
 473         mnt_drop_write_file(file);
 474
 475         return ret;
 476 }
 477
 478 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 479 {
 480         struct inode *inode = file_inode(file);
 481
 482         return put_user(inode->i_generation, arg);
 483 }
 484
 485 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 486 {
 487         struct inode *inode = file_inode(file);
 488         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 489         struct btrfs_device *device;
 490         struct request_queue *q;
 491         struct fstrim_range range;
 492         u64 minlen = ULLONG_MAX;
 493         u64 num_devices = 0;
 494         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 495         int ret;
 496
 497         if (!capable(CAP_SYS_ADMIN))
 498                 return -EPERM;
 499
 500         rcu_read_lock();
 501         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 502                                 dev_list) {
 503                 if (!device->bdev)
 504                         continue;
 505                 q = bdev_get_queue(device->bdev);
 506                 if (blk_queue_discard(q)) {
 507                         num_devices++;
 508                         minlen = min_t(u64, q->limits.discard_granularity,
 509                                      minlen);
 510                 }
 511         }
 512         rcu_read_unlock();
 513
 514         if (!num_devices)
 515                 return -EOPNOTSUPP;
 516         if (copy_from_user(&range, arg, sizeof(range)))
 517                 return -EFAULT;
 518         if (range.start > total_bytes ||
 519             range.len < fs_info->sb->s_blocksize)
 520                 return -EINVAL;
 521
 522         range.len = min(range.len, total_bytes - range.start);
 523         range.minlen = max(range.minlen, minlen);
 524         ret = btrfs_trim_fs(fs_info, &range);
 525         if (ret < 0)
 526                 return ret;
 527
 528         if (copy_to_user(arg, &range, sizeof(range)))
 529                 return -EFAULT;
 530
 531         return 0;
 532 }
 533
 534 int btrfs_is_empty_uuid(u8 *uuid)
 535 {
 536         int i;
 537
 538         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 539                 if (uuid[i])
 540                         return 0;
 541         }
 542         return 1;
 543 }
 544
 545 static noinline int create_subvol(struct inode *dir,
 546                                   struct dentry *dentry,
 547                                   const char *name, int namelen,
 548                                   u64 *async_transid,
 549                                   struct btrfs_qgroup_inherit *inherit)
 550 {
 551         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 552         struct btrfs_trans_handle *trans;
 553         struct btrfs_key key;
 554         struct btrfs_root_item *root_item;
 555         struct btrfs_inode_item *inode_item;
 556         struct extent_buffer *leaf;
 557         struct btrfs_root *root = BTRFS_I(dir)->root;
 558         struct btrfs_root *new_root;
 559         struct btrfs_block_rsv block_rsv;
 560         struct timespec64 cur_time = current_time(dir);
 561         struct inode *inode;
 562         int ret;
 563         int err;
 564         u64 objectid;
 565         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 566         u64 index = 0;
 567         uuid_le new_uuid;
 568
 569         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 570         if (!root_item)
 571                 return -ENOMEM;
 572
 573         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 574         if (ret)
 575                 goto fail_free;
 576
 577         /*
 578          * Don't create subvolume whose level is not zero. Or qgroup will be
 579          * screwed up since it assumes subvolume qgroup's level to be 0.
 580          */
 581         if (btrfs_qgroup_level(objectid)) {
 582                 ret = -ENOSPC;
 583                 goto fail_free;
 584         }
 585
 586         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 587         /*
 588          * The same as the snapshot creation, please see the comment
 589          * of create_snapshot().
 590          */
 591         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 592         if (ret)
 593                 goto fail_free;
 594
 595         trans = btrfs_start_transaction(root, 0);
 596         if (IS_ERR(trans)) {
 597                 ret = PTR_ERR(trans);
 598                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 599                 goto fail_free;
 600         }
 601         trans->block_rsv = &block_rsv;
 602         trans->bytes_reserved = block_rsv.size;
 603
 604         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 605         if (ret)
 606                 goto fail;
 607
 608         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 609         if (IS_ERR(leaf)) {
 610                 ret = PTR_ERR(leaf);
 611                 goto fail;
 612         }
 613
 614         btrfs_mark_buffer_dirty(leaf);
 615
 616         inode_item = &root_item->inode;
 617         btrfs_set_stack_inode_generation(inode_item, 1);
 618         btrfs_set_stack_inode_size(inode_item, 3);
 619         btrfs_set_stack_inode_nlink(inode_item, 1);
 620         btrfs_set_stack_inode_nbytes(inode_item,
 621                                      fs_info->nodesize);
 622         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 623
 624         btrfs_set_root_flags(root_item, 0);
 625         btrfs_set_root_limit(root_item, 0);
 626         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 627
 628         btrfs_set_root_bytenr(root_item, leaf->start);
 629         btrfs_set_root_generation(root_item, trans->transid);
 630         btrfs_set_root_level(root_item, 0);
 631         btrfs_set_root_refs(root_item, 1);
 632         btrfs_set_root_used(root_item, leaf->len);
 633         btrfs_set_root_last_snapshot(root_item, 0);
 634
 635         btrfs_set_root_generation_v2(root_item,
 636                         btrfs_root_generation(root_item));
 637         uuid_le_gen(&new_uuid);
 638         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 639         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 640         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 641         root_item->ctime = root_item->otime;
 642         btrfs_set_root_ctransid(root_item, trans->transid);
 643         btrfs_set_root_otransid(root_item, trans->transid);
 644
 645         btrfs_tree_unlock(leaf);
 646         free_extent_buffer(leaf);
 647         leaf = NULL;
 648
 649         btrfs_set_root_dirid(root_item, new_dirid);
 650
 651         key.objectid = objectid;
 652         key.offset = 0;
 653         key.type = BTRFS_ROOT_ITEM_KEY;
 654         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 655                                 root_item);
 656         if (ret)
 657                 goto fail;
 658
 659         key.offset = (u64)-1;
 660         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 661         if (IS_ERR(new_root)) {
 662                 ret = PTR_ERR(new_root);
 663                 btrfs_abort_transaction(trans, ret);
 664                 goto fail;
 665         }
 666
 667         btrfs_record_root_in_trans(trans, new_root);
 668
 669         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 670         if (ret) {
 671                 /* We potentially lose an unused inode item here */
 672                 btrfs_abort_transaction(trans, ret);
 673                 goto fail;
 674         }
 675
 676         mutex_lock(&new_root->objectid_mutex);
 677         new_root->highest_objectid = new_dirid;
 678         mutex_unlock(&new_root->objectid_mutex);
 679
 680         /*
 681          * insert the directory item
 682          */
 683         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 684         if (ret) {
 685                 btrfs_abort_transaction(trans, ret);
 686                 goto fail;
 687         }
 688
 689         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 690                                     BTRFS_FT_DIR, index);
 691         if (ret) {
 692                 btrfs_abort_transaction(trans, ret);
 693                 goto fail;
 694         }
 695
 696         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 697         ret = btrfs_update_inode(trans, root, dir);
 698         BUG_ON(ret);
 699
 700         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 701                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 702         BUG_ON(ret);
 703
 704         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 705                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 706         if (ret)
 707                 btrfs_abort_transaction(trans, ret);
 708
 709 fail:
 710         kfree(root_item);
 711         trans->block_rsv = NULL;
 712         trans->bytes_reserved = 0;
 713         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 714
 715         if (async_transid) {
 716                 *async_transid = trans->transid;
 717                 err = btrfs_commit_transaction_async(trans, 1);
 718                 if (err)
 719                         err = btrfs_commit_transaction(trans);
 720         } else {
 721                 err = btrfs_commit_transaction(trans);
 722         }
 723         if (err && !ret)
 724                 ret = err;
 725
 726         if (!ret) {
 727                 inode = btrfs_lookup_dentry(dir, dentry);
 728                 if (IS_ERR(inode))
 729                         return PTR_ERR(inode);
 730                 d_instantiate(dentry, inode);
 731         }
 732         return ret;
 733
 734 fail_free:
 735         kfree(root_item);
 736         return ret;
 737 }
 738
 739 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 740                            struct dentry *dentry,
 741                            u64 *async_transid, bool readonly,
 742                            struct btrfs_qgroup_inherit *inherit)
 743 {
 744         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 745         struct inode *inode;
 746         struct btrfs_pending_snapshot *pending_snapshot;
 747         struct btrfs_trans_handle *trans;
 748         int ret;
 749         bool snapshot_force_cow = false;
 750
 751         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 752                 return -EINVAL;
 753
 754         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 755         if (!pending_snapshot)
 756                 return -ENOMEM;
 757
 758         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 759                         GFP_KERNEL);
 760         pending_snapshot->path = btrfs_alloc_path();
 761         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 762                 ret = -ENOMEM;
 763                 goto free_pending;
 764         }
 765
 766         /*
 767          * Force new buffered writes to reserve space even when NOCOW is
 768          * possible. This is to avoid later writeback (running dealloc) to
 769          * fallback to COW mode and unexpectedly fail with ENOSPC.
 770          */
 771         atomic_inc(&root->will_be_snapshotted);
 772         smp_mb__after_atomic();
 773         /* wait for no snapshot writes */
 774         wait_event(root->subv_writers->wait,
 775                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 776
 777         ret = btrfs_start_delalloc_inodes(root);
 778         if (ret)
 779                 goto dec_and_free;
 780
 781         /*
 782          * All previous writes have started writeback in NOCOW mode, so now
 783          * we force future writes to fallback to COW mode during snapshot
 784          * creation.
 785          */
 786         atomic_inc(&root->snapshot_force_cow);
 787         snapshot_force_cow = true;
 788
 789         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 790
 791         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 792                              BTRFS_BLOCK_RSV_TEMP);
 793         /*
 794          * 1 - parent dir inode
 795          * 2 - dir entries
 796          * 1 - root item
 797          * 2 - root ref/backref
 798          * 1 - root of snapshot
 799          * 1 - UUID item
 800          */
 801         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 802                                         &pending_snapshot->block_rsv, 8,
 803                                         false);
 804         if (ret)
 805                 goto dec_and_free;
 806
 807         pending_snapshot->dentry = dentry;
 808         pending_snapshot->root = root;
 809         pending_snapshot->readonly = readonly;
 810         pending_snapshot->dir = dir;
 811         pending_snapshot->inherit = inherit;
 812
 813         trans = btrfs_start_transaction(root, 0);
 814         if (IS_ERR(trans)) {
 815                 ret = PTR_ERR(trans);
 816                 goto fail;
 817         }
 818
 819         spin_lock(&fs_info->trans_lock);
 820         list_add(&pending_snapshot->list,
 821                  &trans->transaction->pending_snapshots);
 822         spin_unlock(&fs_info->trans_lock);
 823         if (async_transid) {
 824                 *async_transid = trans->transid;
 825                 ret = btrfs_commit_transaction_async(trans, 1);
 826                 if (ret)
 827                         ret = btrfs_commit_transaction(trans);
 828         } else {
 829                 ret = btrfs_commit_transaction(trans);
 830         }
 831         if (ret)
 832                 goto fail;
 833
 834         ret = pending_snapshot->error;
 835         if (ret)
 836                 goto fail;
 837
 838         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 839         if (ret)
 840                 goto fail;
 841
 842         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 843         if (IS_ERR(inode)) {
 844                 ret = PTR_ERR(inode);
 845                 goto fail;
 846         }
 847
 848         d_instantiate(dentry, inode);
 849         ret = 0;
 850 fail:
 851         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 852 dec_and_free:
 853         if (snapshot_force_cow)
 854                 atomic_dec(&root->snapshot_force_cow);
 855         if (atomic_dec_and_test(&root->will_be_snapshotted))
 856                 wake_up_var(&root->will_be_snapshotted);
 857 free_pending:
 858         kfree(pending_snapshot->root_item);
 859         btrfs_free_path(pending_snapshot->path);
 860         kfree(pending_snapshot);
 861
 862         return ret;
 863 }
 864
 865 /*  copy of may_delete in fs/namei.c()
 866  *      Check whether we can remove a link victim from directory dir, check
 867  *  whether the type of victim is right.
 868  *  1. We can't do it if dir is read-only (done in permission())
 869  *  2. We should have write and exec permissions on dir
 870  *  3. We can't remove anything from append-only dir
 871  *  4. We can't do anything with immutable dir (done in permission())
 872  *  5. If the sticky bit on dir is set we should either
 873  *      a. be owner of dir, or
 874  *      b. be owner of victim, or
 875  *      c. have CAP_FOWNER capability
 876  *  6. If the victim is append-only or immutable we can't do anything with
 877  *     links pointing to it.
 878  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 879  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 880  *  9. We can't remove a root or mountpoint.
 881  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 882  *     nfs_async_unlink().
 883  */
 884
 885 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 886 {
 887         int error;
 888
 889         if (d_really_is_negative(victim))
 890                 return -ENOENT;
 891
 892         BUG_ON(d_inode(victim->d_parent) != dir);
 893         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 894
 895         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 896         if (error)
 897                 return error;
 898         if (IS_APPEND(dir))
 899                 return -EPERM;
 900         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 901             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 902                 return -EPERM;
 903         if (isdir) {
 904                 if (!d_is_dir(victim))
 905                         return -ENOTDIR;
 906                 if (IS_ROOT(victim))
 907                         return -EBUSY;
 908         } else if (d_is_dir(victim))
 909                 return -EISDIR;
 910         if (IS_DEADDIR(dir))
 911                 return -ENOENT;
 912         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 913                 return -EBUSY;
 914         return 0;
 915 }
 916
 917 /* copy of may_create in fs/namei.c() */
 918 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 919 {
 920         if (d_really_is_positive(child))
 921                 return -EEXIST;
 922         if (IS_DEADDIR(dir))
 923                 return -ENOENT;
 924         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 925 }
 926
 927 /*
 928  * Create a new subvolume below @parent.  This is largely modeled after
 929  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 930  * inside this filesystem so it's quite a bit simpler.
 931  */
 932 static noinline int btrfs_mksubvol(const struct path *parent,
 933                                    const char *name, int namelen,
 934                                    struct btrfs_root *snap_src,
 935                                    u64 *async_transid, bool readonly,
 936                                    struct btrfs_qgroup_inherit *inherit)
 937 {
 938         struct inode *dir = d_inode(parent->dentry);
 939         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 940         struct dentry *dentry;
 941         int error;
 942
 943         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 944         if (error == -EINTR)
 945                 return error;
 946
 947         dentry = lookup_one_len(name, parent->dentry, namelen);
 948         error = PTR_ERR(dentry);
 949         if (IS_ERR(dentry))
 950                 goto out_unlock;
 951
 952         error = btrfs_may_create(dir, dentry);
 953         if (error)
 954                 goto out_dput;
 955
 956         /*
 957          * even if this name doesn't exist, we may get hash collisions.
 958          * check for them now when we can safely fail
 959          */
 960         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 961                                                dir->i_ino, name,
 962                                                namelen);
 963         if (error)
 964                 goto out_dput;
 965
 966         down_read(&fs_info->subvol_sem);
 967
 968         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 969                 goto out_up_read;
 970
 971         if (snap_src) {
 972                 error = create_snapshot(snap_src, dir, dentry,
 973                                         async_transid, readonly, inherit);
 974         } else {
 975                 error = create_subvol(dir, dentry, name, namelen,
 976                                       async_transid, inherit);
 977         }
 978         if (!error)
 979                 fsnotify_mkdir(dir, dentry);
 980 out_up_read:
 981         up_read(&fs_info->subvol_sem);
 982 out_dput:
 983         dput(dentry);
 984 out_unlock:
 985         inode_unlock(dir);
 986         return error;
 987 }
 988
 989 /*
 990  * When we're defragging a range, we don't want to kick it off again
 991  * if it is really just waiting for delalloc to send it down.
 992  * If we find a nice big extent or delalloc range for the bytes in the
 993  * file you want to defrag, we return 0 to let you know to skip this
 994  * part of the file
 995  */
 996 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
 997 {
 998         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 999         struct extent_map *em = NULL;
1000         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1001         u64 end;
1002
1003         read_lock(&em_tree->lock);
1004         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1005         read_unlock(&em_tree->lock);
1006
1007         if (em) {
1008                 end = extent_map_end(em);
1009                 free_extent_map(em);
1010                 if (end - offset > thresh)
1011                         return 0;
1012         }
1013         /* if we already have a nice delalloc here, just stop */
1014         thresh /= 2;
1015         end = count_range_bits(io_tree, &offset, offset + thresh,
1016                                thresh, EXTENT_DELALLOC, 1);
1017         if (end >= thresh)
1018                 return 0;
1019         return 1;
1020 }
1021
1022 /*
1023  * helper function to walk through a file and find extents
1024  * newer than a specific transid, and smaller than thresh.
1025  *
1026  * This is used by the defragging code to find new and small
1027  * extents
1028  */
1029 static int find_new_extents(struct btrfs_root *root,
1030                             struct inode *inode, u64 newer_than,
1031                             u64 *off, u32 thresh)
1032 {
1033         struct btrfs_path *path;
1034         struct btrfs_key min_key;
1035         struct extent_buffer *leaf;
1036         struct btrfs_file_extent_item *extent;
1037         int type;
1038         int ret;
1039         u64 ino = btrfs_ino(BTRFS_I(inode));
1040
1041         path = btrfs_alloc_path();
1042         if (!path)
1043                 return -ENOMEM;
1044
1045         min_key.objectid = ino;
1046         min_key.type = BTRFS_EXTENT_DATA_KEY;
1047         min_key.offset = *off;
1048
1049         while (1) {
1050                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1051                 if (ret != 0)
1052                         goto none;
1053 process_slot:
1054                 if (min_key.objectid != ino)
1055                         goto none;
1056                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1057                         goto none;
1058
1059                 leaf = path->nodes[0];
1060                 extent = btrfs_item_ptr(leaf, path->slots[0],
1061                                         struct btrfs_file_extent_item);
1062
1063                 type = btrfs_file_extent_type(leaf, extent);
1064                 if (type == BTRFS_FILE_EXTENT_REG &&
1065                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1066                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1067                         *off = min_key.offset;
1068                         btrfs_free_path(path);
1069                         return 0;
1070                 }
1071
1072                 path->slots[0]++;
1073                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1074                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1075                         goto process_slot;
1076                 }
1077
1078                 if (min_key.offset == (u64)-1)
1079                         goto none;
1080
1081                 min_key.offset++;
1082                 btrfs_release_path(path);
1083         }
1084 none:
1085         btrfs_free_path(path);
1086         return -ENOENT;
1087 }
1088
1089 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1090 {
1091         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1092         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1093         struct extent_map *em;
1094         u64 len = PAGE_SIZE;
1095
1096         /*
1097          * hopefully we have this extent in the tree already, try without
1098          * the full extent lock
1099          */
1100         read_lock(&em_tree->lock);
1101         em = lookup_extent_mapping(em_tree, start, len);
1102         read_unlock(&em_tree->lock);
1103
1104         if (!em) {
1105                 struct extent_state *cached = NULL;
1106                 u64 end = start + len - 1;
1107
1108                 /* get the big lock and read metadata off disk */
1109                 lock_extent_bits(io_tree, start, end, &cached);
1110                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1111                 unlock_extent_cached(io_tree, start, end, &cached);
1112
1113                 if (IS_ERR(em))
1114                         return NULL;
1115         }
1116
1117         return em;
1118 }
1119
1120 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1121 {
1122         struct extent_map *next;
1123         bool ret = true;
1124
1125         /* this is the last extent */
1126         if (em->start + em->len >= i_size_read(inode))
1127                 return false;
1128
1129         next = defrag_lookup_extent(inode, em->start + em->len);
1130         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1131                 ret = false;
1132         else if ((em->block_start + em->block_len == next->block_start) &&
1133                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1134                 ret = false;
1135
1136         free_extent_map(next);
1137         return ret;
1138 }
1139
1140 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1141                                u64 *last_len, u64 *skip, u64 *defrag_end,
1142                                int compress)
1143 {
1144         struct extent_map *em;
1145         int ret = 1;
1146         bool next_mergeable = true;
1147         bool prev_mergeable = true;
1148
1149         /*
1150          * make sure that once we start defragging an extent, we keep on
1151          * defragging it
1152          */
1153         if (start < *defrag_end)
1154                 return 1;
1155
1156         *skip = 0;
1157
1158         em = defrag_lookup_extent(inode, start);
1159         if (!em)
1160                 return 0;
1161
1162         /* this will cover holes, and inline extents */
1163         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1164                 ret = 0;
1165                 goto out;
1166         }
1167
1168         if (!*defrag_end)
1169                 prev_mergeable = false;
1170
1171         next_mergeable = defrag_check_next_extent(inode, em);
1172         /*
1173          * we hit a real extent, if it is big or the next extent is not a
1174          * real extent, don't bother defragging it
1175          */
1176         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1177             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1178                 ret = 0;
1179 out:
1180         /*
1181          * last_len ends up being a counter of how many bytes we've defragged.
1182          * every time we choose not to defrag an extent, we reset *last_len
1183          * so that the next tiny extent will force a defrag.
1184          *
1185          * The end result of this is that tiny extents before a single big
1186          * extent will force at least part of that big extent to be defragged.
1187          */
1188         if (ret) {
1189                 *defrag_end = extent_map_end(em);
1190         } else {
1191                 *last_len = 0;
1192                 *skip = extent_map_end(em);
1193                 *defrag_end = 0;
1194         }
1195
1196         free_extent_map(em);
1197         return ret;
1198 }
1199
1200 /*
1201  * it doesn't do much good to defrag one or two pages
1202  * at a time.  This pulls in a nice chunk of pages
1203  * to COW and defrag.
1204  *
1205  * It also makes sure the delalloc code has enough
1206  * dirty data to avoid making new small extents as part
1207  * of the defrag
1208  *
1209  * It's a good idea to start RA on this range
1210  * before calling this.
1211  */
1212 static int cluster_pages_for_defrag(struct inode *inode,
1213                                     struct page **pages,
1214                                     unsigned long start_index,
1215                                     unsigned long num_pages)
1216 {
1217         unsigned long file_end;
1218         u64 isize = i_size_read(inode);
1219         u64 page_start;
1220         u64 page_end;
1221         u64 page_cnt;
1222         int ret;
1223         int i;
1224         int i_done;
1225         struct btrfs_ordered_extent *ordered;
1226         struct extent_state *cached_state = NULL;
1227         struct extent_io_tree *tree;
1228         struct extent_changeset *data_reserved = NULL;
1229         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1230
1231         file_end = (isize - 1) >> PAGE_SHIFT;
1232         if (!isize || start_index > file_end)
1233                 return 0;
1234
1235         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1236
1237         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1238                         start_index << PAGE_SHIFT,
1239                         page_cnt << PAGE_SHIFT);
1240         if (ret)
1241                 return ret;
1242         i_done = 0;
1243         tree = &BTRFS_I(inode)->io_tree;
1244
1245         /* step one, lock all the pages */
1246         for (i = 0; i < page_cnt; i++) {
1247                 struct page *page;
1248 again:
1249                 page = find_or_create_page(inode->i_mapping,
1250                                            start_index + i, mask);
1251                 if (!page)
1252                         break;
1253
1254                 page_start = page_offset(page);
1255                 page_end = page_start + PAGE_SIZE - 1;
1256                 while (1) {
1257                         lock_extent_bits(tree, page_start, page_end,
1258                                          &cached_state);
1259                         ordered = btrfs_lookup_ordered_extent(inode,
1260                                                               page_start);
1261                         unlock_extent_cached(tree, page_start, page_end,
1262                                              &cached_state);
1263                         if (!ordered)
1264                                 break;
1265
1266                         unlock_page(page);
1267                         btrfs_start_ordered_extent(inode, ordered, 1);
1268                         btrfs_put_ordered_extent(ordered);
1269                         lock_page(page);
1270                         /*
1271                          * we unlocked the page above, so we need check if
1272                          * it was released or not.
1273                          */
1274                         if (page->mapping != inode->i_mapping) {
1275                                 unlock_page(page);
1276                                 put_page(page);
1277                                 goto again;
1278                         }
1279                 }
1280
1281                 if (!PageUptodate(page)) {
1282                         btrfs_readpage(NULL, page);
1283                         lock_page(page);
1284                         if (!PageUptodate(page)) {
1285                                 unlock_page(page);
1286                                 put_page(page);
1287                                 ret = -EIO;
1288                                 break;
1289                         }
1290                 }
1291
1292                 if (page->mapping != inode->i_mapping) {
1293                         unlock_page(page);
1294                         put_page(page);
1295                         goto again;
1296                 }
1297
1298                 pages[i] = page;
1299                 i_done++;
1300         }
1301         if (!i_done || ret)
1302                 goto out;
1303
1304         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1305                 goto out;
1306
1307         /*
1308          * so now we have a nice long stream of locked
1309          * and up to date pages, lets wait on them
1310          */
1311         for (i = 0; i < i_done; i++)
1312                 wait_on_page_writeback(pages[i]);
1313
1314         page_start = page_offset(pages[0]);
1315         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1316
1317         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1318                          page_start, page_end - 1, &cached_state);
1319         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1320                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1321                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1322                           &cached_state);
1323
1324         if (i_done != page_cnt) {
1325                 spin_lock(&BTRFS_I(inode)->lock);
1326                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1327                 spin_unlock(&BTRFS_I(inode)->lock);
1328                 btrfs_delalloc_release_space(inode, data_reserved,
1329                                 start_index << PAGE_SHIFT,
1330                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1331         }
1332
1333
1334         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1335                           &cached_state);
1336
1337         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1338                              page_start, page_end - 1, &cached_state);
1339
1340         for (i = 0; i < i_done; i++) {
1341                 clear_page_dirty_for_io(pages[i]);
1342                 ClearPageChecked(pages[i]);
1343                 set_page_extent_mapped(pages[i]);
1344                 set_page_dirty(pages[i]);
1345                 unlock_page(pages[i]);
1346                 put_page(pages[i]);
1347         }
1348         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1349                                        false);
1350         extent_changeset_free(data_reserved);
1351         return i_done;
1352 out:
1353         for (i = 0; i < i_done; i++) {
1354                 unlock_page(pages[i]);
1355                 put_page(pages[i]);
1356         }
1357         btrfs_delalloc_release_space(inode, data_reserved,
1358                         start_index << PAGE_SHIFT,
1359                         page_cnt << PAGE_SHIFT, true);
1360         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1361                                        true);
1362         extent_changeset_free(data_reserved);
1363         return ret;
1364
1365 }
1366
1367 int btrfs_defrag_file(struct inode *inode, struct file *file,
1368                       struct btrfs_ioctl_defrag_range_args *range,
1369                       u64 newer_than, unsigned long max_to_defrag)
1370 {
1371         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1372         struct btrfs_root *root = BTRFS_I(inode)->root;
1373         struct file_ra_state *ra = NULL;
1374         unsigned long last_index;
1375         u64 isize = i_size_read(inode);
1376         u64 last_len = 0;
1377         u64 skip = 0;
1378         u64 defrag_end = 0;
1379         u64 newer_off = range->start;
1380         unsigned long i;
1381         unsigned long ra_index = 0;
1382         int ret;
1383         int defrag_count = 0;
1384         int compress_type = BTRFS_COMPRESS_ZLIB;
1385         u32 extent_thresh = range->extent_thresh;
1386         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1387         unsigned long cluster = max_cluster;
1388         u64 new_align = ~((u64)SZ_128K - 1);
1389         struct page **pages = NULL;
1390         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1391
1392         if (isize == 0)
1393                 return 0;
1394
1395         if (range->start >= isize)
1396                 return -EINVAL;
1397
1398         if (do_compress) {
1399                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1400                         return -EINVAL;
1401                 if (range->compress_type)
1402                         compress_type = range->compress_type;
1403         }
1404
1405         if (extent_thresh == 0)
1406                 extent_thresh = SZ_256K;
1407
1408         /*
1409          * If we were not given a file, allocate a readahead context. As
1410          * readahead is just an optimization, defrag will work without it so
1411          * we don't error out.
1412          */
1413         if (!file) {
1414                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1415                 if (ra)
1416                         file_ra_state_init(ra, inode->i_mapping);
1417         } else {
1418                 ra = &file->f_ra;
1419         }
1420
1421         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1422         if (!pages) {
1423                 ret = -ENOMEM;
1424                 goto out_ra;
1425         }
1426
1427         /* find the last page to defrag */
1428         if (range->start + range->len > range->start) {
1429                 last_index = min_t(u64, isize - 1,
1430                          range->start + range->len - 1) >> PAGE_SHIFT;
1431         } else {
1432                 last_index = (isize - 1) >> PAGE_SHIFT;
1433         }
1434
1435         if (newer_than) {
1436                 ret = find_new_extents(root, inode, newer_than,
1437                                        &newer_off, SZ_64K);
1438                 if (!ret) {
1439                         range->start = newer_off;
1440                         /*
1441                          * we always align our defrag to help keep
1442                          * the extents in the file evenly spaced
1443                          */
1444                         i = (newer_off & new_align) >> PAGE_SHIFT;
1445                 } else
1446                         goto out_ra;
1447         } else {
1448                 i = range->start >> PAGE_SHIFT;
1449         }
1450         if (!max_to_defrag)
1451                 max_to_defrag = last_index - i + 1;
1452
1453         /*
1454          * make writeback starts from i, so the defrag range can be
1455          * written sequentially.
1456          */
1457         if (i < inode->i_mapping->writeback_index)
1458                 inode->i_mapping->writeback_index = i;
1459
1460         while (i <= last_index && defrag_count < max_to_defrag &&
1461                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1462                 /*
1463                  * make sure we stop running if someone unmounts
1464                  * the FS
1465                  */
1466                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1467                         break;
1468
1469                 if (btrfs_defrag_cancelled(fs_info)) {
1470                         btrfs_debug(fs_info, "defrag_file cancelled");
1471                         ret = -EAGAIN;
1472                         break;
1473                 }
1474
1475                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1476                                          extent_thresh, &last_len, &skip,
1477                                          &defrag_end, do_compress)){
1478                         unsigned long next;
1479                         /*
1480                          * the should_defrag function tells us how much to skip
1481                          * bump our counter by the suggested amount
1482                          */
1483                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1484                         i = max(i + 1, next);
1485                         continue;
1486                 }
1487
1488                 if (!newer_than) {
1489                         cluster = (PAGE_ALIGN(defrag_end) >>
1490                                    PAGE_SHIFT) - i;
1491                         cluster = min(cluster, max_cluster);
1492                 } else {
1493                         cluster = max_cluster;
1494                 }
1495
1496                 if (i + cluster > ra_index) {
1497                         ra_index = max(i, ra_index);
1498                         if (ra)
1499                                 page_cache_sync_readahead(inode->i_mapping, ra,
1500                                                 file, ra_index, cluster);
1501                         ra_index += cluster;
1502                 }
1503
1504                 inode_lock(inode);
1505                 if (do_compress)
1506                         BTRFS_I(inode)->defrag_compress = compress_type;
1507                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1508                 if (ret < 0) {
1509                         inode_unlock(inode);
1510                         goto out_ra;
1511                 }
1512
1513                 defrag_count += ret;
1514                 balance_dirty_pages_ratelimited(inode->i_mapping);
1515                 inode_unlock(inode);
1516
1517                 if (newer_than) {
1518                         if (newer_off == (u64)-1)
1519                                 break;
1520
1521                         if (ret > 0)
1522                                 i += ret;
1523
1524                         newer_off = max(newer_off + 1,
1525                                         (u64)i << PAGE_SHIFT);
1526
1527                         ret = find_new_extents(root, inode, newer_than,
1528                                                &newer_off, SZ_64K);
1529                         if (!ret) {
1530                                 range->start = newer_off;
1531                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1532                         } else {
1533                                 break;
1534                         }
1535                 } else {
1536                         if (ret > 0) {
1537                                 i += ret;
1538                                 last_len += ret << PAGE_SHIFT;
1539                         } else {
1540                                 i++;
1541                                 last_len = 0;
1542                         }
1543                 }
1544         }
1545
1546         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1547                 filemap_flush(inode->i_mapping);
1548                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1549                              &BTRFS_I(inode)->runtime_flags))
1550                         filemap_flush(inode->i_mapping);
1551         }
1552
1553         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1554                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1555         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1556                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1557         }
1558
1559         ret = defrag_count;
1560
1561 out_ra:
1562         if (do_compress) {
1563                 inode_lock(inode);
1564                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1565                 inode_unlock(inode);
1566         }
1567         if (!file)
1568                 kfree(ra);
1569         kfree(pages);
1570         return ret;
1571 }
1572
1573 static noinline int btrfs_ioctl_resize(struct file *file,
1574                                         void __user *arg)
1575 {
1576         struct inode *inode = file_inode(file);
1577         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1578         u64 new_size;
1579         u64 old_size;
1580         u64 devid = 1;
1581         struct btrfs_root *root = BTRFS_I(inode)->root;
1582         struct btrfs_ioctl_vol_args *vol_args;
1583         struct btrfs_trans_handle *trans;
1584         struct btrfs_device *device = NULL;
1585         char *sizestr;
1586         char *retptr;
1587         char *devstr = NULL;
1588         int ret = 0;
1589         int mod = 0;
1590
1591         if (!capable(CAP_SYS_ADMIN))
1592                 return -EPERM;
1593
1594         ret = mnt_want_write_file(file);
1595         if (ret)
1596                 return ret;
1597
1598         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1599                 mnt_drop_write_file(file);
1600                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1601         }
1602
1603         vol_args = memdup_user(arg, sizeof(*vol_args));
1604         if (IS_ERR(vol_args)) {
1605                 ret = PTR_ERR(vol_args);
1606                 goto out;
1607         }
1608
1609         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1610
1611         sizestr = vol_args->name;
1612         devstr = strchr(sizestr, ':');
1613         if (devstr) {
1614                 sizestr = devstr + 1;
1615                 *devstr = '\0';
1616                 devstr = vol_args->name;
1617                 ret = kstrtoull(devstr, 10, &devid);
1618                 if (ret)
1619                         goto out_free;
1620                 if (!devid) {
1621                         ret = -EINVAL;
1622                         goto out_free;
1623                 }
1624                 btrfs_info(fs_info, "resizing devid %llu", devid);
1625         }
1626
1627         device = btrfs_find_device(fs_info, devid, NULL, NULL);
1628         if (!device) {
1629                 btrfs_info(fs_info, "resizer unable to find device %llu",
1630                            devid);
1631                 ret = -ENODEV;
1632                 goto out_free;
1633         }
1634
1635         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1636                 btrfs_info(fs_info,
1637                            "resizer unable to apply on readonly device %llu",
1638                        devid);
1639                 ret = -EPERM;
1640                 goto out_free;
1641         }
1642
1643         if (!strcmp(sizestr, "max"))
1644                 new_size = device->bdev->bd_inode->i_size;
1645         else {
1646                 if (sizestr[0] == '-') {
1647                         mod = -1;
1648                         sizestr++;
1649                 } else if (sizestr[0] == '+') {
1650                         mod = 1;
1651                         sizestr++;
1652                 }
1653                 new_size = memparse(sizestr, &retptr);
1654                 if (*retptr != '\0' || new_size == 0) {
1655                         ret = -EINVAL;
1656                         goto out_free;
1657                 }
1658         }
1659
1660         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1661                 ret = -EPERM;
1662                 goto out_free;
1663         }
1664
1665         old_size = btrfs_device_get_total_bytes(device);
1666
1667         if (mod < 0) {
1668                 if (new_size > old_size) {
1669                         ret = -EINVAL;
1670                         goto out_free;
1671                 }
1672                 new_size = old_size - new_size;
1673         } else if (mod > 0) {
1674                 if (new_size > ULLONG_MAX - old_size) {
1675                         ret = -ERANGE;
1676                         goto out_free;
1677                 }
1678                 new_size = old_size + new_size;
1679         }
1680
1681         if (new_size < SZ_256M) {
1682                 ret = -EINVAL;
1683                 goto out_free;
1684         }
1685         if (new_size > device->bdev->bd_inode->i_size) {
1686                 ret = -EFBIG;
1687                 goto out_free;
1688         }
1689
1690         new_size = round_down(new_size, fs_info->sectorsize);
1691
1692         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1693                           rcu_str_deref(device->name), new_size);
1694
1695         if (new_size > old_size) {
1696                 trans = btrfs_start_transaction(root, 0);
1697                 if (IS_ERR(trans)) {
1698                         ret = PTR_ERR(trans);
1699                         goto out_free;
1700                 }
1701                 ret = btrfs_grow_device(trans, device, new_size);
1702                 btrfs_commit_transaction(trans);
1703         } else if (new_size < old_size) {
1704                 ret = btrfs_shrink_device(device, new_size);
1705         } /* equal, nothing need to do */
1706
1707 out_free:
1708         kfree(vol_args);
1709 out:
1710         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1711         mnt_drop_write_file(file);
1712         return ret;
1713 }
1714
1715 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1716                                 const char *name, unsigned long fd, int subvol,
1717                                 u64 *transid, bool readonly,
1718                                 struct btrfs_qgroup_inherit *inherit)
1719 {
1720         int namelen;
1721         int ret = 0;
1722
1723         if (!S_ISDIR(file_inode(file)->i_mode))
1724                 return -ENOTDIR;
1725
1726         ret = mnt_want_write_file(file);
1727         if (ret)
1728                 goto out;
1729
1730         namelen = strlen(name);
1731         if (strchr(name, '/')) {
1732                 ret = -EINVAL;
1733                 goto out_drop_write;
1734         }
1735
1736         if (name[0] == '.' &&
1737            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1738                 ret = -EEXIST;
1739                 goto out_drop_write;
1740         }
1741
1742         if (subvol) {
1743                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1744                                      NULL, transid, readonly, inherit);
1745         } else {
1746                 struct fd src = fdget(fd);
1747                 struct inode *src_inode;
1748                 if (!src.file) {
1749                         ret = -EINVAL;
1750                         goto out_drop_write;
1751                 }
1752
1753                 src_inode = file_inode(src.file);
1754                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1755                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1756                                    "Snapshot src from another FS");
1757                         ret = -EXDEV;
1758                 } else if (!inode_owner_or_capable(src_inode)) {
1759                         /*
1760                          * Subvolume creation is not restricted, but snapshots
1761                          * are limited to own subvolumes only
1762                          */
1763                         ret = -EPERM;
1764                 } else {
1765                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1766                                              BTRFS_I(src_inode)->root,
1767                                              transid, readonly, inherit);
1768                 }
1769                 fdput(src);
1770         }
1771 out_drop_write:
1772         mnt_drop_write_file(file);
1773 out:
1774         return ret;
1775 }
1776
1777 static noinline int btrfs_ioctl_snap_create(struct file *file,
1778                                             void __user *arg, int subvol)
1779 {
1780         struct btrfs_ioctl_vol_args *vol_args;
1781         int ret;
1782
1783         if (!S_ISDIR(file_inode(file)->i_mode))
1784                 return -ENOTDIR;
1785
1786         vol_args = memdup_user(arg, sizeof(*vol_args));
1787         if (IS_ERR(vol_args))
1788                 return PTR_ERR(vol_args);
1789         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1790
1791         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1792                                               vol_args->fd, subvol,
1793                                               NULL, false, NULL);
1794
1795         kfree(vol_args);
1796         return ret;
1797 }
1798
1799 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1800                                                void __user *arg, int subvol)
1801 {
1802         struct btrfs_ioctl_vol_args_v2 *vol_args;
1803         int ret;
1804         u64 transid = 0;
1805         u64 *ptr = NULL;
1806         bool readonly = false;
1807         struct btrfs_qgroup_inherit *inherit = NULL;
1808
1809         if (!S_ISDIR(file_inode(file)->i_mode))
1810                 return -ENOTDIR;
1811
1812         vol_args = memdup_user(arg, sizeof(*vol_args));
1813         if (IS_ERR(vol_args))
1814                 return PTR_ERR(vol_args);
1815         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1816
1817         if (vol_args->flags &
1818             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1819               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1820                 ret = -EOPNOTSUPP;
1821                 goto free_args;
1822         }
1823
1824         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1825                 ptr = &transid;
1826         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1827                 readonly = true;
1828         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1829                 if (vol_args->size > PAGE_SIZE) {
1830                         ret = -EINVAL;
1831                         goto free_args;
1832                 }
1833                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1834                 if (IS_ERR(inherit)) {
1835                         ret = PTR_ERR(inherit);
1836                         goto free_args;
1837                 }
1838         }
1839
1840         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1841                                               vol_args->fd, subvol, ptr,
1842                                               readonly, inherit);
1843         if (ret)
1844                 goto free_inherit;
1845
1846         if (ptr && copy_to_user(arg +
1847                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1848                                         transid),
1849                                 ptr, sizeof(*ptr)))
1850                 ret = -EFAULT;
1851
1852 free_inherit:
1853         kfree(inherit);
1854 free_args:
1855         kfree(vol_args);
1856         return ret;
1857 }
1858
1859 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1860                                                 void __user *arg)
1861 {
1862         struct inode *inode = file_inode(file);
1863         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1864         struct btrfs_root *root = BTRFS_I(inode)->root;
1865         int ret = 0;
1866         u64 flags = 0;
1867
1868         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1869                 return -EINVAL;
1870
1871         down_read(&fs_info->subvol_sem);
1872         if (btrfs_root_readonly(root))
1873                 flags |= BTRFS_SUBVOL_RDONLY;
1874         up_read(&fs_info->subvol_sem);
1875
1876         if (copy_to_user(arg, &flags, sizeof(flags)))
1877                 ret = -EFAULT;
1878
1879         return ret;
1880 }
1881
1882 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1883                                               void __user *arg)
1884 {
1885         struct inode *inode = file_inode(file);
1886         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1887         struct btrfs_root *root = BTRFS_I(inode)->root;
1888         struct btrfs_trans_handle *trans;
1889         u64 root_flags;
1890         u64 flags;
1891         int ret = 0;
1892
1893         if (!inode_owner_or_capable(inode))
1894                 return -EPERM;
1895
1896         ret = mnt_want_write_file(file);
1897         if (ret)
1898                 goto out;
1899
1900         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1901                 ret = -EINVAL;
1902                 goto out_drop_write;
1903         }
1904
1905         if (copy_from_user(&flags, arg, sizeof(flags))) {
1906                 ret = -EFAULT;
1907                 goto out_drop_write;
1908         }
1909
1910         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1911                 ret = -EINVAL;
1912                 goto out_drop_write;
1913         }
1914
1915         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1916                 ret = -EOPNOTSUPP;
1917                 goto out_drop_write;
1918         }
1919
1920         down_write(&fs_info->subvol_sem);
1921
1922         /* nothing to do */
1923         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1924                 goto out_drop_sem;
1925
1926         root_flags = btrfs_root_flags(&root->root_item);
1927         if (flags & BTRFS_SUBVOL_RDONLY) {
1928                 btrfs_set_root_flags(&root->root_item,
1929                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1930         } else {
1931                 /*
1932                  * Block RO -> RW transition if this subvolume is involved in
1933                  * send
1934                  */
1935                 spin_lock(&root->root_item_lock);
1936                 if (root->send_in_progress == 0) {
1937                         btrfs_set_root_flags(&root->root_item,
1938                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1939                         spin_unlock(&root->root_item_lock);
1940                 } else {
1941                         spin_unlock(&root->root_item_lock);
1942                         btrfs_warn(fs_info,
1943                                    "Attempt to set subvolume %llu read-write during send",
1944                                    root->root_key.objectid);
1945                         ret = -EPERM;
1946                         goto out_drop_sem;
1947                 }
1948         }
1949
1950         trans = btrfs_start_transaction(root, 1);
1951         if (IS_ERR(trans)) {
1952                 ret = PTR_ERR(trans);
1953                 goto out_reset;
1954         }
1955
1956         ret = btrfs_update_root(trans, fs_info->tree_root,
1957                                 &root->root_key, &root->root_item);
1958         if (ret < 0) {
1959                 btrfs_end_transaction(trans);
1960                 goto out_reset;
1961         }
1962
1963         ret = btrfs_commit_transaction(trans);
1964
1965 out_reset:
1966         if (ret)
1967                 btrfs_set_root_flags(&root->root_item, root_flags);
1968 out_drop_sem:
1969         up_write(&fs_info->subvol_sem);
1970 out_drop_write:
1971         mnt_drop_write_file(file);
1972 out:
1973         return ret;
1974 }
1975
1976 static noinline int key_in_sk(struct btrfs_key *key,
1977                               struct btrfs_ioctl_search_key *sk)
1978 {
1979         struct btrfs_key test;
1980         int ret;
1981
1982         test.objectid = sk->min_objectid;
1983         test.type = sk->min_type;
1984         test.offset = sk->min_offset;
1985
1986         ret = btrfs_comp_cpu_keys(key, &test);
1987         if (ret < 0)
1988                 return 0;
1989
1990         test.objectid = sk->max_objectid;
1991         test.type = sk->max_type;
1992         test.offset = sk->max_offset;
1993
1994         ret = btrfs_comp_cpu_keys(key, &test);
1995         if (ret > 0)
1996                 return 0;
1997         return 1;
1998 }
1999
2000 static noinline int copy_to_sk(struct btrfs_path *path,
2001                                struct btrfs_key *key,
2002                                struct btrfs_ioctl_search_key *sk,
2003                                size_t *buf_size,
2004                                char __user *ubuf,
2005                                unsigned long *sk_offset,
2006                                int *num_found)
2007 {
2008         u64 found_transid;
2009         struct extent_buffer *leaf;
2010         struct btrfs_ioctl_search_header sh;
2011         struct btrfs_key test;
2012         unsigned long item_off;
2013         unsigned long item_len;
2014         int nritems;
2015         int i;
2016         int slot;
2017         int ret = 0;
2018
2019         leaf = path->nodes[0];
2020         slot = path->slots[0];
2021         nritems = btrfs_header_nritems(leaf);
2022
2023         if (btrfs_header_generation(leaf) > sk->max_transid) {
2024                 i = nritems;
2025                 goto advance_key;
2026         }
2027         found_transid = btrfs_header_generation(leaf);
2028
2029         for (i = slot; i < nritems; i++) {
2030                 item_off = btrfs_item_ptr_offset(leaf, i);
2031                 item_len = btrfs_item_size_nr(leaf, i);
2032
2033                 btrfs_item_key_to_cpu(leaf, key, i);
2034                 if (!key_in_sk(key, sk))
2035                         continue;
2036
2037                 if (sizeof(sh) + item_len > *buf_size) {
2038                         if (*num_found) {
2039                                 ret = 1;
2040                                 goto out;
2041                         }
2042
2043                         /*
2044                          * return one empty item back for v1, which does not
2045                          * handle -EOVERFLOW
2046                          */
2047
2048                         *buf_size = sizeof(sh) + item_len;
2049                         item_len = 0;
2050                         ret = -EOVERFLOW;
2051                 }
2052
2053                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2054                         ret = 1;
2055                         goto out;
2056                 }
2057
2058                 sh.objectid = key->objectid;
2059                 sh.offset = key->offset;
2060                 sh.type = key->type;
2061                 sh.len = item_len;
2062                 sh.transid = found_transid;
2063
2064                 /* copy search result header */
2065                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2066                         ret = -EFAULT;
2067                         goto out;
2068                 }
2069
2070                 *sk_offset += sizeof(sh);
2071
2072                 if (item_len) {
2073                         char __user *up = ubuf + *sk_offset;
2074                         /* copy the item */
2075                         if (read_extent_buffer_to_user(leaf, up,
2076                                                        item_off, item_len)) {
2077                                 ret = -EFAULT;
2078                                 goto out;
2079                         }
2080
2081                         *sk_offset += item_len;
2082                 }
2083                 (*num_found)++;
2084
2085                 if (ret) /* -EOVERFLOW from above */
2086                         goto out;
2087
2088                 if (*num_found >= sk->nr_items) {
2089                         ret = 1;
2090                         goto out;
2091                 }
2092         }
2093 advance_key:
2094         ret = 0;
2095         test.objectid = sk->max_objectid;
2096         test.type = sk->max_type;
2097         test.offset = sk->max_offset;
2098         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2099                 ret = 1;
2100         else if (key->offset < (u64)-1)
2101                 key->offset++;
2102         else if (key->type < (u8)-1) {
2103                 key->offset = 0;
2104                 key->type++;
2105         } else if (key->objectid < (u64)-1) {
2106                 key->offset = 0;
2107                 key->type = 0;
2108                 key->objectid++;
2109         } else
2110                 ret = 1;
2111 out:
2112         /*
2113          *  0: all items from this leaf copied, continue with next
2114          *  1: * more items can be copied, but unused buffer is too small
2115          *     * all items were found
2116          *     Either way, it will stops the loop which iterates to the next
2117          *     leaf
2118          *  -EOVERFLOW: item was to large for buffer
2119          *  -EFAULT: could not copy extent buffer back to userspace
2120          */
2121         return ret;
2122 }
2123
2124 static noinline int search_ioctl(struct inode *inode,
2125                                  struct btrfs_ioctl_search_key *sk,
2126                                  size_t *buf_size,
2127                                  char __user *ubuf)
2128 {
2129         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2130         struct btrfs_root *root;
2131         struct btrfs_key key;
2132         struct btrfs_path *path;
2133         int ret;
2134         int num_found = 0;
2135         unsigned long sk_offset = 0;
2136
2137         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2138                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2139                 return -EOVERFLOW;
2140         }
2141
2142         path = btrfs_alloc_path();
2143         if (!path)
2144                 return -ENOMEM;
2145
2146         if (sk->tree_id == 0) {
2147                 /* search the root of the inode that was passed */
2148                 root = BTRFS_I(inode)->root;
2149         } else {
2150                 key.objectid = sk->tree_id;
2151                 key.type = BTRFS_ROOT_ITEM_KEY;
2152                 key.offset = (u64)-1;
2153                 root = btrfs_read_fs_root_no_name(info, &key);
2154                 if (IS_ERR(root)) {
2155                         btrfs_free_path(path);
2156                         return PTR_ERR(root);
2157                 }
2158         }
2159
2160         key.objectid = sk->min_objectid;
2161         key.type = sk->min_type;
2162         key.offset = sk->min_offset;
2163
2164         while (1) {
2165                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2166                 if (ret != 0) {
2167                         if (ret > 0)
2168                                 ret = 0;
2169                         goto err;
2170                 }
2171                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2172                                  &sk_offset, &num_found);
2173                 btrfs_release_path(path);
2174                 if (ret)
2175                         break;
2176
2177         }
2178         if (ret > 0)
2179                 ret = 0;
2180 err:
2181         sk->nr_items = num_found;
2182         btrfs_free_path(path);
2183         return ret;
2184 }
2185
2186 static noinline int btrfs_ioctl_tree_search(struct file *file,
2187                                            void __user *argp)
2188 {
2189         struct btrfs_ioctl_search_args __user *uargs;
2190         struct btrfs_ioctl_search_key sk;
2191         struct inode *inode;
2192         int ret;
2193         size_t buf_size;
2194
2195         if (!capable(CAP_SYS_ADMIN))
2196                 return -EPERM;
2197
2198         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2199
2200         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2201                 return -EFAULT;
2202
2203         buf_size = sizeof(uargs->buf);
2204
2205         inode = file_inode(file);
2206         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2207
2208         /*
2209          * In the origin implementation an overflow is handled by returning a
2210          * search header with a len of zero, so reset ret.
2211          */
2212         if (ret == -EOVERFLOW)
2213                 ret = 0;
2214
2215         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2216                 ret = -EFAULT;
2217         return ret;
2218 }
2219
2220 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2221                                                void __user *argp)
2222 {
2223         struct btrfs_ioctl_search_args_v2 __user *uarg;
2224         struct btrfs_ioctl_search_args_v2 args;
2225         struct inode *inode;
2226         int ret;
2227         size_t buf_size;
2228         const size_t buf_limit = SZ_16M;
2229
2230         if (!capable(CAP_SYS_ADMIN))
2231                 return -EPERM;
2232
2233         /* copy search header and buffer size */
2234         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2235         if (copy_from_user(&args, uarg, sizeof(args)))
2236                 return -EFAULT;
2237
2238         buf_size = args.buf_size;
2239
2240         /* limit result size to 16MB */
2241         if (buf_size > buf_limit)
2242                 buf_size = buf_limit;
2243
2244         inode = file_inode(file);
2245         ret = search_ioctl(inode, &args.key, &buf_size,
2246                            (char __user *)(&uarg->buf[0]));
2247         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2248                 ret = -EFAULT;
2249         else if (ret == -EOVERFLOW &&
2250                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2251                 ret = -EFAULT;
2252
2253         return ret;
2254 }
2255
2256 /*
2257  * Search INODE_REFs to identify path name of 'dirid' directory
2258  * in a 'tree_id' tree. and sets path name to 'name'.
2259  */
2260 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2261                                 u64 tree_id, u64 dirid, char *name)
2262 {
2263         struct btrfs_root *root;
2264         struct btrfs_key key;
2265         char *ptr;
2266         int ret = -1;
2267         int slot;
2268         int len;
2269         int total_len = 0;
2270         struct btrfs_inode_ref *iref;
2271         struct extent_buffer *l;
2272         struct btrfs_path *path;
2273
2274         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2275                 name[0]='\0';
2276                 return 0;
2277         }
2278
2279         path = btrfs_alloc_path();
2280         if (!path)
2281                 return -ENOMEM;
2282
2283         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2284
2285         key.objectid = tree_id;
2286         key.type = BTRFS_ROOT_ITEM_KEY;
2287         key.offset = (u64)-1;
2288         root = btrfs_read_fs_root_no_name(info, &key);
2289         if (IS_ERR(root)) {
2290                 ret = PTR_ERR(root);
2291                 goto out;
2292         }
2293
2294         key.objectid = dirid;
2295         key.type = BTRFS_INODE_REF_KEY;
2296         key.offset = (u64)-1;
2297
2298         while (1) {
2299                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2300                 if (ret < 0)
2301                         goto out;
2302                 else if (ret > 0) {
2303                         ret = btrfs_previous_item(root, path, dirid,
2304                                                   BTRFS_INODE_REF_KEY);
2305                         if (ret < 0)
2306                                 goto out;
2307                         else if (ret > 0) {
2308                                 ret = -ENOENT;
2309                                 goto out;
2310                         }
2311                 }
2312
2313                 l = path->nodes[0];
2314                 slot = path->slots[0];
2315                 btrfs_item_key_to_cpu(l, &key, slot);
2316
2317                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2318                 len = btrfs_inode_ref_name_len(l, iref);
2319                 ptr -= len + 1;
2320                 total_len += len + 1;
2321                 if (ptr < name) {
2322                         ret = -ENAMETOOLONG;
2323                         goto out;
2324                 }
2325
2326                 *(ptr + len) = '/';
2327                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2328
2329                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2330                         break;
2331
2332                 btrfs_release_path(path);
2333                 key.objectid = key.offset;
2334                 key.offset = (u64)-1;
2335                 dirid = key.objectid;
2336         }
2337         memmove(name, ptr, total_len);
2338         name[total_len] = '\0';
2339         ret = 0;
2340 out:
2341         btrfs_free_path(path);
2342         return ret;
2343 }
2344
2345 static int btrfs_search_path_in_tree_user(struct inode *inode,
2346                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2347 {
2348         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2349         struct super_block *sb = inode->i_sb;
2350         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2351         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2352         u64 dirid = args->dirid;
2353         unsigned long item_off;
2354         unsigned long item_len;
2355         struct btrfs_inode_ref *iref;
2356         struct btrfs_root_ref *rref;
2357         struct btrfs_root *root;
2358         struct btrfs_path *path;
2359         struct btrfs_key key, key2;
2360         struct extent_buffer *leaf;
2361         struct inode *temp_inode;
2362         char *ptr;
2363         int slot;
2364         int len;
2365         int total_len = 0;
2366         int ret;
2367
2368         path = btrfs_alloc_path();
2369         if (!path)
2370                 return -ENOMEM;
2371
2372         /*
2373          * If the bottom subvolume does not exist directly under upper_limit,
2374          * construct the path in from the bottom up.
2375          */
2376         if (dirid != upper_limit.objectid) {
2377                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2378
2379                 key.objectid = treeid;
2380                 key.type = BTRFS_ROOT_ITEM_KEY;
2381                 key.offset = (u64)-1;
2382                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2383                 if (IS_ERR(root)) {
2384                         ret = PTR_ERR(root);
2385                         goto out;
2386                 }
2387
2388                 key.objectid = dirid;
2389                 key.type = BTRFS_INODE_REF_KEY;
2390                 key.offset = (u64)-1;
2391                 while (1) {
2392                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2393                         if (ret < 0) {
2394                                 goto out;
2395                         } else if (ret > 0) {
2396                                 ret = btrfs_previous_item(root, path, dirid,
2397                                                           BTRFS_INODE_REF_KEY);
2398                                 if (ret < 0) {
2399                                         goto out;
2400                                 } else if (ret > 0) {
2401                                         ret = -ENOENT;
2402                                         goto out;
2403                                 }
2404                         }
2405
2406                         leaf = path->nodes[0];
2407                         slot = path->slots[0];
2408                         btrfs_item_key_to_cpu(leaf, &key, slot);
2409
2410                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2411                         len = btrfs_inode_ref_name_len(leaf, iref);
2412                         ptr -= len + 1;
2413                         total_len += len + 1;
2414                         if (ptr < args->path) {
2415                                 ret = -ENAMETOOLONG;
2416                                 goto out;
2417                         }
2418
2419                         *(ptr + len) = '/';
2420                         read_extent_buffer(leaf, ptr,
2421                                         (unsigned long)(iref + 1), len);
2422
2423                         /* Check the read+exec permission of this directory */
2424                         ret = btrfs_previous_item(root, path, dirid,
2425                                                   BTRFS_INODE_ITEM_KEY);
2426                         if (ret < 0) {
2427                                 goto out;
2428                         } else if (ret > 0) {
2429                                 ret = -ENOENT;
2430                                 goto out;
2431                         }
2432
2433                         leaf = path->nodes[0];
2434                         slot = path->slots[0];
2435                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2436                         if (key2.objectid != dirid) {
2437                                 ret = -ENOENT;
2438                                 goto out;
2439                         }
2440
2441                         temp_inode = btrfs_iget(sb, &key2, root, NULL);
2442                         if (IS_ERR(temp_inode)) {
2443                                 ret = PTR_ERR(temp_inode);
2444                                 goto out;
2445                         }
2446                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2447                         iput(temp_inode);
2448                         if (ret) {
2449                                 ret = -EACCES;
2450                                 goto out;
2451                         }
2452
2453                         if (key.offset == upper_limit.objectid)
2454                                 break;
2455                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2456                                 ret = -EACCES;
2457                                 goto out;
2458                         }
2459
2460                         btrfs_release_path(path);
2461                         key.objectid = key.offset;
2462                         key.offset = (u64)-1;
2463                         dirid = key.objectid;
2464                 }
2465
2466                 memmove(args->path, ptr, total_len);
2467                 args->path[total_len] = '\0';
2468                 btrfs_release_path(path);
2469         }
2470
2471         /* Get the bottom subvolume's name from ROOT_REF */
2472         root = fs_info->tree_root;
2473         key.objectid = treeid;
2474         key.type = BTRFS_ROOT_REF_KEY;
2475         key.offset = args->treeid;
2476         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2477         if (ret < 0) {
2478                 goto out;
2479         } else if (ret > 0) {
2480                 ret = -ENOENT;
2481                 goto out;
2482         }
2483
2484         leaf = path->nodes[0];
2485         slot = path->slots[0];
2486         btrfs_item_key_to_cpu(leaf, &key, slot);
2487
2488         item_off = btrfs_item_ptr_offset(leaf, slot);
2489         item_len = btrfs_item_size_nr(leaf, slot);
2490         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2491         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2492         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2493                 ret = -EINVAL;
2494                 goto out;
2495         }
2496
2497         /* Copy subvolume's name */
2498         item_off += sizeof(struct btrfs_root_ref);
2499         item_len -= sizeof(struct btrfs_root_ref);
2500         read_extent_buffer(leaf, args->name, item_off, item_len);
2501         args->name[item_len] = 0;
2502
2503 out:
2504         btrfs_free_path(path);
2505         return ret;
2506 }
2507
2508 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2509                                            void __user *argp)
2510 {
2511         struct btrfs_ioctl_ino_lookup_args *args;
2512         struct inode *inode;
2513         int ret = 0;
2514
2515         args = memdup_user(argp, sizeof(*args));
2516         if (IS_ERR(args))
2517                 return PTR_ERR(args);
2518
2519         inode = file_inode(file);
2520
2521         /*
2522          * Unprivileged query to obtain the containing subvolume root id. The
2523          * path is reset so it's consistent with btrfs_search_path_in_tree.
2524          */
2525         if (args->treeid == 0)
2526                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2527
2528         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2529                 args->name[0] = 0;
2530                 goto out;
2531         }
2532
2533         if (!capable(CAP_SYS_ADMIN)) {
2534                 ret = -EPERM;
2535                 goto out;
2536         }
2537
2538         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2539                                         args->treeid, args->objectid,
2540                                         args->name);
2541
2542 out:
2543         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2544                 ret = -EFAULT;
2545
2546         kfree(args);
2547         return ret;
2548 }
2549
2550 /*
2551  * Version of ino_lookup ioctl (unprivileged)
2552  *
2553  * The main differences from ino_lookup ioctl are:
2554  *
2555  *   1. Read + Exec permission will be checked using inode_permission() during
2556  *      path construction. -EACCES will be returned in case of failure.
2557  *   2. Path construction will be stopped at the inode number which corresponds
2558  *      to the fd with which this ioctl is called. If constructed path does not
2559  *      exist under fd's inode, -EACCES will be returned.
2560  *   3. The name of bottom subvolume is also searched and filled.
2561  */
2562 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2563 {
2564         struct btrfs_ioctl_ino_lookup_user_args *args;
2565         struct inode *inode;
2566         int ret;
2567
2568         args = memdup_user(argp, sizeof(*args));
2569         if (IS_ERR(args))
2570                 return PTR_ERR(args);
2571
2572         inode = file_inode(file);
2573
2574         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2575             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2576                 /*
2577                  * The subvolume does not exist under fd with which this is
2578                  * called
2579                  */
2580                 kfree(args);
2581                 return -EACCES;
2582         }
2583
2584         ret = btrfs_search_path_in_tree_user(inode, args);
2585
2586         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2587                 ret = -EFAULT;
2588
2589         kfree(args);
2590         return ret;
2591 }
2592
2593 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2594 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2595 {
2596         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2597         struct btrfs_fs_info *fs_info;
2598         struct btrfs_root *root;
2599         struct btrfs_path *path;
2600         struct btrfs_key key;
2601         struct btrfs_root_item *root_item;
2602         struct btrfs_root_ref *rref;
2603         struct extent_buffer *leaf;
2604         unsigned long item_off;
2605         unsigned long item_len;
2606         struct inode *inode;
2607         int slot;
2608         int ret = 0;
2609
2610         path = btrfs_alloc_path();
2611         if (!path)
2612                 return -ENOMEM;
2613
2614         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2615         if (!subvol_info) {
2616                 btrfs_free_path(path);
2617                 return -ENOMEM;
2618         }
2619
2620         inode = file_inode(file);
2621         fs_info = BTRFS_I(inode)->root->fs_info;
2622
2623         /* Get root_item of inode's subvolume */
2624         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2625         key.type = BTRFS_ROOT_ITEM_KEY;
2626         key.offset = (u64)-1;
2627         root = btrfs_read_fs_root_no_name(fs_info, &key);
2628         if (IS_ERR(root)) {
2629                 ret = PTR_ERR(root);
2630                 goto out;
2631         }
2632         root_item = &root->root_item;
2633
2634         subvol_info->treeid = key.objectid;
2635
2636         subvol_info->generation = btrfs_root_generation(root_item);
2637         subvol_info->flags = btrfs_root_flags(root_item);
2638
2639         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2640         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2641                                                     BTRFS_UUID_SIZE);
2642         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2643                                                     BTRFS_UUID_SIZE);
2644
2645         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2646         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2647         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2648
2649         subvol_info->otransid = btrfs_root_otransid(root_item);
2650         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2651         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2652
2653         subvol_info->stransid = btrfs_root_stransid(root_item);
2654         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2655         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2656
2657         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2658         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2659         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2660
2661         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2662                 /* Search root tree for ROOT_BACKREF of this subvolume */
2663                 root = fs_info->tree_root;
2664
2665                 key.type = BTRFS_ROOT_BACKREF_KEY;
2666                 key.offset = 0;
2667                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2668                 if (ret < 0) {
2669                         goto out;
2670                 } else if (path->slots[0] >=
2671                            btrfs_header_nritems(path->nodes[0])) {
2672                         ret = btrfs_next_leaf(root, path);
2673                         if (ret < 0) {
2674                                 goto out;
2675                         } else if (ret > 0) {
2676                                 ret = -EUCLEAN;
2677                                 goto out;
2678                         }
2679                 }
2680
2681                 leaf = path->nodes[0];
2682                 slot = path->slots[0];
2683                 btrfs_item_key_to_cpu(leaf, &key, slot);
2684                 if (key.objectid == subvol_info->treeid &&
2685                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2686                         subvol_info->parent_id = key.offset;
2687
2688                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2689                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2690
2691                         item_off = btrfs_item_ptr_offset(leaf, slot)
2692                                         + sizeof(struct btrfs_root_ref);
2693                         item_len = btrfs_item_size_nr(leaf, slot)
2694                                         - sizeof(struct btrfs_root_ref);
2695                         read_extent_buffer(leaf, subvol_info->name,
2696                                            item_off, item_len);
2697                 } else {
2698                         ret = -ENOENT;
2699                         goto out;
2700                 }
2701         }
2702
2703         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2704                 ret = -EFAULT;
2705
2706 out:
2707         btrfs_free_path(path);
2708         kzfree(subvol_info);
2709         return ret;
2710 }
2711
2712 /*
2713  * Return ROOT_REF information of the subvolume containing this inode
2714  * except the subvolume name.
2715  */
2716 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2717 {
2718         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2719         struct btrfs_root_ref *rref;
2720         struct btrfs_root *root;
2721         struct btrfs_path *path;
2722         struct btrfs_key key;
2723         struct extent_buffer *leaf;
2724         struct inode *inode;
2725         u64 objectid;
2726         int slot;
2727         int ret;
2728         u8 found;
2729
2730         path = btrfs_alloc_path();
2731         if (!path)
2732                 return -ENOMEM;
2733
2734         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2735         if (IS_ERR(rootrefs)) {
2736                 btrfs_free_path(path);
2737                 return PTR_ERR(rootrefs);
2738         }
2739
2740         inode = file_inode(file);
2741         root = BTRFS_I(inode)->root->fs_info->tree_root;
2742         objectid = BTRFS_I(inode)->root->root_key.objectid;
2743
2744         key.objectid = objectid;
2745         key.type = BTRFS_ROOT_REF_KEY;
2746         key.offset = rootrefs->min_treeid;
2747         found = 0;
2748
2749         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2750         if (ret < 0) {
2751                 goto out;
2752         } else if (path->slots[0] >=
2753                    btrfs_header_nritems(path->nodes[0])) {
2754                 ret = btrfs_next_leaf(root, path);
2755                 if (ret < 0) {
2756                         goto out;
2757                 } else if (ret > 0) {
2758                         ret = -EUCLEAN;
2759                         goto out;
2760                 }
2761         }
2762         while (1) {
2763                 leaf = path->nodes[0];
2764                 slot = path->slots[0];
2765
2766                 btrfs_item_key_to_cpu(leaf, &key, slot);
2767                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2768                         ret = 0;
2769                         goto out;
2770                 }
2771
2772                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2773                         ret = -EOVERFLOW;
2774                         goto out;
2775                 }
2776
2777                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2778                 rootrefs->rootref[found].treeid = key.offset;
2779                 rootrefs->rootref[found].dirid =
2780                                   btrfs_root_ref_dirid(leaf, rref);
2781                 found++;
2782
2783                 ret = btrfs_next_item(root, path);
2784                 if (ret < 0) {
2785                         goto out;
2786                 } else if (ret > 0) {
2787                         ret = -EUCLEAN;
2788                         goto out;
2789                 }
2790         }
2791
2792 out:
2793         if (!ret || ret == -EOVERFLOW) {
2794                 rootrefs->num_items = found;
2795                 /* update min_treeid for next search */
2796                 if (found)
2797                         rootrefs->min_treeid =
2798                                 rootrefs->rootref[found - 1].treeid + 1;
2799                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2800                         ret = -EFAULT;
2801         }
2802
2803         kfree(rootrefs);
2804         btrfs_free_path(path);
2805
2806         return ret;
2807 }
2808
2809 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2810                                              void __user *arg)
2811 {
2812         struct dentry *parent = file->f_path.dentry;
2813         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2814         struct dentry *dentry;
2815         struct inode *dir = d_inode(parent);
2816         struct inode *inode;
2817         struct btrfs_root *root = BTRFS_I(dir)->root;
2818         struct btrfs_root *dest = NULL;
2819         struct btrfs_ioctl_vol_args *vol_args;
2820         int namelen;
2821         int err = 0;
2822
2823         if (!S_ISDIR(dir->i_mode))
2824                 return -ENOTDIR;
2825
2826         vol_args = memdup_user(arg, sizeof(*vol_args));
2827         if (IS_ERR(vol_args))
2828                 return PTR_ERR(vol_args);
2829
2830         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2831         namelen = strlen(vol_args->name);
2832         if (strchr(vol_args->name, '/') ||
2833             strncmp(vol_args->name, "..", namelen) == 0) {
2834                 err = -EINVAL;
2835                 goto out;
2836         }
2837
2838         err = mnt_want_write_file(file);
2839         if (err)
2840                 goto out;
2841
2842
2843         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2844         if (err == -EINTR)
2845                 goto out_drop_write;
2846         dentry = lookup_one_len(vol_args->name, parent, namelen);
2847         if (IS_ERR(dentry)) {
2848                 err = PTR_ERR(dentry);
2849                 goto out_unlock_dir;
2850         }
2851
2852         if (d_really_is_negative(dentry)) {
2853                 err = -ENOENT;
2854                 goto out_dput;
2855         }
2856
2857         inode = d_inode(dentry);
2858         dest = BTRFS_I(inode)->root;
2859         if (!capable(CAP_SYS_ADMIN)) {
2860                 /*
2861                  * Regular user.  Only allow this with a special mount
2862                  * option, when the user has write+exec access to the
2863                  * subvol root, and when rmdir(2) would have been
2864                  * allowed.
2865                  *
2866                  * Note that this is _not_ check that the subvol is
2867                  * empty or doesn't contain data that we wouldn't
2868                  * otherwise be able to delete.
2869                  *
2870                  * Users who want to delete empty subvols should try
2871                  * rmdir(2).
2872                  */
2873                 err = -EPERM;
2874                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2875                         goto out_dput;
2876
2877                 /*
2878                  * Do not allow deletion if the parent dir is the same
2879                  * as the dir to be deleted.  That means the ioctl
2880                  * must be called on the dentry referencing the root
2881                  * of the subvol, not a random directory contained
2882                  * within it.
2883                  */
2884                 err = -EINVAL;
2885                 if (root == dest)
2886                         goto out_dput;
2887
2888                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2889                 if (err)
2890                         goto out_dput;
2891         }
2892
2893         /* check if subvolume may be deleted by a user */
2894         err = btrfs_may_delete(dir, dentry, 1);
2895         if (err)
2896                 goto out_dput;
2897
2898         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2899                 err = -EINVAL;
2900                 goto out_dput;
2901         }
2902
2903         inode_lock(inode);
2904         err = btrfs_delete_subvolume(dir, dentry);
2905         inode_unlock(inode);
2906         if (!err)
2907                 d_delete(dentry);
2908
2909 out_dput:
2910         dput(dentry);
2911 out_unlock_dir:
2912         inode_unlock(dir);
2913 out_drop_write:
2914         mnt_drop_write_file(file);
2915 out:
2916         kfree(vol_args);
2917         return err;
2918 }
2919
2920 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2921 {
2922         struct inode *inode = file_inode(file);
2923         struct btrfs_root *root = BTRFS_I(inode)->root;
2924         struct btrfs_ioctl_defrag_range_args *range;
2925         int ret;
2926
2927         ret = mnt_want_write_file(file);
2928         if (ret)
2929                 return ret;
2930
2931         if (btrfs_root_readonly(root)) {
2932                 ret = -EROFS;
2933                 goto out;
2934         }
2935
2936         switch (inode->i_mode & S_IFMT) {
2937         case S_IFDIR:
2938                 if (!capable(CAP_SYS_ADMIN)) {
2939                         ret = -EPERM;
2940                         goto out;
2941                 }
2942                 ret = btrfs_defrag_root(root);
2943                 break;
2944         case S_IFREG:
2945                 /*
2946                  * Note that this does not check the file descriptor for write
2947                  * access. This prevents defragmenting executables that are
2948                  * running and allows defrag on files open in read-only mode.
2949                  */
2950                 if (!capable(CAP_SYS_ADMIN) &&
2951                     inode_permission(inode, MAY_WRITE)) {
2952                         ret = -EPERM;
2953                         goto out;
2954                 }
2955
2956                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2957                 if (!range) {
2958                         ret = -ENOMEM;
2959                         goto out;
2960                 }
2961
2962                 if (argp) {
2963                         if (copy_from_user(range, argp,
2964                                            sizeof(*range))) {
2965                                 ret = -EFAULT;
2966                                 kfree(range);
2967                                 goto out;
2968                         }
2969                         /* compression requires us to start the IO */
2970                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2971                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2972                                 range->extent_thresh = (u32)-1;
2973                         }
2974                 } else {
2975                         /* the rest are all set to zero by kzalloc */
2976                         range->len = (u64)-1;
2977                 }
2978                 ret = btrfs_defrag_file(file_inode(file), file,
2979                                         range, BTRFS_OLDEST_GENERATION, 0);
2980                 if (ret > 0)
2981                         ret = 0;
2982                 kfree(range);
2983                 break;
2984         default:
2985                 ret = -EINVAL;
2986         }
2987 out:
2988         mnt_drop_write_file(file);
2989         return ret;
2990 }
2991
2992 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
2993 {
2994         struct btrfs_ioctl_vol_args *vol_args;
2995         int ret;
2996
2997         if (!capable(CAP_SYS_ADMIN))
2998                 return -EPERM;
2999
3000         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3001                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3002
3003         vol_args = memdup_user(arg, sizeof(*vol_args));
3004         if (IS_ERR(vol_args)) {
3005                 ret = PTR_ERR(vol_args);
3006                 goto out;
3007         }
3008
3009         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3010         ret = btrfs_init_new_device(fs_info, vol_args->name);
3011
3012         if (!ret)
3013                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3014
3015         kfree(vol_args);
3016 out:
3017         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3018         return ret;
3019 }
3020
3021 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3022 {
3023         struct inode *inode = file_inode(file);
3024         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3025         struct btrfs_ioctl_vol_args_v2 *vol_args;
3026         int ret;
3027
3028         if (!capable(CAP_SYS_ADMIN))
3029                 return -EPERM;
3030
3031         ret = mnt_want_write_file(file);
3032         if (ret)
3033                 return ret;
3034
3035         vol_args = memdup_user(arg, sizeof(*vol_args));
3036         if (IS_ERR(vol_args)) {
3037                 ret = PTR_ERR(vol_args);
3038                 goto err_drop;
3039         }
3040
3041         /* Check for compatibility reject unknown flags */
3042         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3043                 ret = -EOPNOTSUPP;
3044                 goto out;
3045         }
3046
3047         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3048                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3049                 goto out;
3050         }
3051
3052         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3053                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3054         } else {
3055                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3056                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3057         }
3058         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3059
3060         if (!ret) {
3061                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3062                         btrfs_info(fs_info, "device deleted: id %llu",
3063                                         vol_args->devid);
3064                 else
3065                         btrfs_info(fs_info, "device deleted: %s",
3066                                         vol_args->name);
3067         }
3068 out:
3069         kfree(vol_args);
3070 err_drop:
3071         mnt_drop_write_file(file);
3072         return ret;
3073 }
3074
3075 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3076 {
3077         struct inode *inode = file_inode(file);
3078         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3079         struct btrfs_ioctl_vol_args *vol_args;
3080         int ret;
3081
3082         if (!capable(CAP_SYS_ADMIN))
3083                 return -EPERM;
3084
3085         ret = mnt_want_write_file(file);
3086         if (ret)
3087                 return ret;
3088
3089         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3090                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3091                 goto out_drop_write;
3092         }
3093
3094         vol_args = memdup_user(arg, sizeof(*vol_args));
3095         if (IS_ERR(vol_args)) {
3096                 ret = PTR_ERR(vol_args);
3097                 goto out;
3098         }
3099
3100         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3101         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3102
3103         if (!ret)
3104                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3105         kfree(vol_args);
3106 out:
3107         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3108 out_drop_write:
3109         mnt_drop_write_file(file);
3110
3111         return ret;
3112 }
3113
3114 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3115                                 void __user *arg)
3116 {
3117         struct btrfs_ioctl_fs_info_args *fi_args;
3118         struct btrfs_device *device;
3119         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3120         int ret = 0;
3121
3122         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3123         if (!fi_args)
3124                 return -ENOMEM;
3125
3126         rcu_read_lock();
3127         fi_args->num_devices = fs_devices->num_devices;
3128
3129         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3130                 if (device->devid > fi_args->max_id)
3131                         fi_args->max_id = device->devid;
3132         }
3133         rcu_read_unlock();
3134
3135         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
3136         fi_args->nodesize = fs_info->nodesize;
3137         fi_args->sectorsize = fs_info->sectorsize;
3138         fi_args->clone_alignment = fs_info->sectorsize;
3139
3140         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3141                 ret = -EFAULT;
3142
3143         kfree(fi_args);
3144         return ret;
3145 }
3146
3147 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3148                                  void __user *arg)
3149 {
3150         struct btrfs_ioctl_dev_info_args *di_args;
3151         struct btrfs_device *dev;
3152         int ret = 0;
3153         char *s_uuid = NULL;
3154
3155         di_args = memdup_user(arg, sizeof(*di_args));
3156         if (IS_ERR(di_args))
3157                 return PTR_ERR(di_args);
3158
3159         if (!btrfs_is_empty_uuid(di_args->uuid))
3160                 s_uuid = di_args->uuid;
3161
3162         rcu_read_lock();
3163         dev = btrfs_find_device(fs_info, di_args->devid, s_uuid, NULL);
3164
3165         if (!dev) {
3166                 ret = -ENODEV;
3167                 goto out;
3168         }
3169
3170         di_args->devid = dev->devid;
3171         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3172         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3173         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3174         if (dev->name) {
3175                 strncpy(di_args->path, rcu_str_deref(dev->name),
3176                                 sizeof(di_args->path) - 1);
3177                 di_args->path[sizeof(di_args->path) - 1] = 0;
3178         } else {
3179                 di_args->path[0] = '\0';
3180         }
3181
3182 out:
3183         rcu_read_unlock();
3184         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3185                 ret = -EFAULT;
3186
3187         kfree(di_args);
3188         return ret;
3189 }
3190
3191 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
3192 {
3193         struct page *page;
3194
3195         page = grab_cache_page(inode->i_mapping, index);
3196         if (!page)
3197                 return ERR_PTR(-ENOMEM);
3198
3199         if (!PageUptodate(page)) {
3200                 int ret;
3201
3202                 ret = btrfs_readpage(NULL, page);
3203                 if (ret)
3204                         return ERR_PTR(ret);
3205                 lock_page(page);
3206                 if (!PageUptodate(page)) {
3207                         unlock_page(page);
3208                         put_page(page);
3209                         return ERR_PTR(-EIO);
3210                 }
3211                 if (page->mapping != inode->i_mapping) {
3212                         unlock_page(page);
3213                         put_page(page);
3214                         return ERR_PTR(-EAGAIN);
3215                 }
3216         }
3217
3218         return page;
3219 }
3220
3221 static int gather_extent_pages(struct inode *inode, struct page **pages,
3222                                int num_pages, u64 off)
3223 {
3224         int i;
3225         pgoff_t index = off >> PAGE_SHIFT;
3226
3227         for (i = 0; i < num_pages; i++) {
3228 again:
3229                 pages[i] = extent_same_get_page(inode, index + i);
3230                 if (IS_ERR(pages[i])) {
3231                         int err = PTR_ERR(pages[i]);
3232
3233                         if (err == -EAGAIN)
3234                                 goto again;
3235                         pages[i] = NULL;
3236                         return err;
3237                 }
3238         }
3239         return 0;
3240 }
3241
3242 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
3243                              bool retry_range_locking)
3244 {
3245         /*
3246          * Do any pending delalloc/csum calculations on inode, one way or
3247          * another, and lock file content.
3248          * The locking order is:
3249          *
3250          *   1) pages
3251          *   2) range in the inode's io tree
3252          */
3253         while (1) {
3254                 struct btrfs_ordered_extent *ordered;
3255                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3256                 ordered = btrfs_lookup_first_ordered_extent(inode,
3257                                                             off + len - 1);
3258                 if ((!ordered ||
3259                      ordered->file_offset + ordered->len <= off ||
3260                      ordered->file_offset >= off + len) &&
3261                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
3262                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
3263                         if (ordered)
3264                                 btrfs_put_ordered_extent(ordered);
3265                         break;
3266                 }
3267                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3268                 if (ordered)
3269                         btrfs_put_ordered_extent(ordered);
3270                 if (!retry_range_locking)
3271                         return -EAGAIN;
3272                 btrfs_wait_ordered_range(inode, off, len);
3273         }
3274         return 0;
3275 }
3276
3277 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
3278 {
3279         inode_unlock(inode1);
3280         inode_unlock(inode2);
3281 }
3282
3283 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
3284 {
3285         if (inode1 < inode2)
3286                 swap(inode1, inode2);
3287
3288         inode_lock_nested(inode1, I_MUTEX_PARENT);
3289         inode_lock_nested(inode2, I_MUTEX_CHILD);
3290 }
3291
3292 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3293                                       struct inode *inode2, u64 loff2, u64 len)
3294 {
3295         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3296         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3297 }
3298
3299 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3300                                     struct inode *inode2, u64 loff2, u64 len,
3301                                     bool retry_range_locking)
3302 {
3303         int ret;
3304
3305         if (inode1 < inode2) {
3306                 swap(inode1, inode2);
3307                 swap(loff1, loff2);
3308         }
3309         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
3310         if (ret)
3311                 return ret;
3312         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
3313         if (ret)
3314                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
3315                               loff1 + len - 1);
3316         return ret;
3317 }
3318
3319 struct cmp_pages {
3320         int             num_pages;
3321         struct page     **src_pages;
3322         struct page     **dst_pages;
3323 };
3324
3325 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3326 {
3327         int i;
3328         struct page *pg;
3329
3330         for (i = 0; i < cmp->num_pages; i++) {
3331                 pg = cmp->src_pages[i];
3332                 if (pg) {
3333                         unlock_page(pg);
3334                         put_page(pg);
3335                         cmp->src_pages[i] = NULL;
3336                 }
3337                 pg = cmp->dst_pages[i];
3338                 if (pg) {
3339                         unlock_page(pg);
3340                         put_page(pg);
3341                         cmp->dst_pages[i] = NULL;
3342                 }
3343         }
3344 }
3345
3346 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3347                                   struct inode *dst, u64 dst_loff,
3348                                   u64 len, struct cmp_pages *cmp)
3349 {
3350         int ret;
3351         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3352
3353         cmp->num_pages = num_pages;
3354
3355         ret = gather_extent_pages(src, cmp->src_pages, num_pages, loff);
3356         if (ret)
3357                 goto out;
3358
3359         ret = gather_extent_pages(dst, cmp->dst_pages, num_pages, dst_loff);
3360
3361 out:
3362         if (ret)
3363                 btrfs_cmp_data_free(cmp);
3364         return ret;
3365 }
3366
3367 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3368 {
3369         int ret = 0;
3370         int i;
3371         struct page *src_page, *dst_page;
3372         unsigned int cmp_len = PAGE_SIZE;
3373         void *addr, *dst_addr;
3374
3375         i = 0;
3376         while (len) {
3377                 if (len < PAGE_SIZE)
3378                         cmp_len = len;
3379
3380                 BUG_ON(i >= cmp->num_pages);
3381
3382                 src_page = cmp->src_pages[i];
3383                 dst_page = cmp->dst_pages[i];
3384                 ASSERT(PageLocked(src_page));
3385                 ASSERT(PageLocked(dst_page));
3386
3387                 addr = kmap_atomic(src_page);
3388                 dst_addr = kmap_atomic(dst_page);
3389
3390                 flush_dcache_page(src_page);
3391                 flush_dcache_page(dst_page);
3392
3393                 if (memcmp(addr, dst_addr, cmp_len))
3394                         ret = -EBADE;
3395
3396                 kunmap_atomic(addr);
3397                 kunmap_atomic(dst_addr);
3398
3399                 if (ret)
3400                         break;
3401
3402                 len -= cmp_len;
3403                 i++;
3404         }
3405
3406         return ret;
3407 }
3408
3409 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3410                                      u64 olen)
3411 {
3412         u64 len = *plen;
3413         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3414
3415         if (off + olen > inode->i_size || off + olen < off)
3416                 return -EINVAL;
3417
3418         /* if we extend to eof, continue to block boundary */
3419         if (off + len == inode->i_size)
3420                 *plen = len = ALIGN(inode->i_size, bs) - off;
3421
3422         /* Check that we are block aligned - btrfs_clone() requires this */
3423         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3424                 return -EINVAL;
3425
3426         return 0;
3427 }
3428
3429 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 olen,
3430                                    struct inode *dst, u64 dst_loff,
3431                                    struct cmp_pages *cmp)
3432 {
3433         int ret;
3434         u64 len = olen;
3435         bool same_inode = (src == dst);
3436         u64 same_lock_start = 0;
3437         u64 same_lock_len = 0;
3438
3439         ret = extent_same_check_offsets(src, loff, &len, olen);
3440         if (ret)
3441                 return ret;
3442
3443         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3444         if (ret)
3445                 return ret;
3446
3447         if (same_inode) {
3448                 /*
3449                  * Single inode case wants the same checks, except we
3450                  * don't want our length pushed out past i_size as
3451                  * comparing that data range makes no sense.
3452                  *
3453                  * extent_same_check_offsets() will do this for an
3454                  * unaligned length at i_size, so catch it here and
3455                  * reject the request.
3456                  *
3457                  * This effectively means we require aligned extents
3458                  * for the single-inode case, whereas the other cases
3459                  * allow an unaligned length so long as it ends at
3460                  * i_size.
3461                  */
3462                 if (len != olen)
3463                         return -EINVAL;
3464
3465                 /* Check for overlapping ranges */
3466                 if (dst_loff + len > loff && dst_loff < loff + len)
3467                         return -EINVAL;
3468
3469                 same_lock_start = min_t(u64, loff, dst_loff);
3470                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3471         } else {
3472                 /*
3473                  * If the source and destination inodes are different, the
3474                  * source's range end offset matches the source's i_size, that
3475                  * i_size is not a multiple of the sector size, and the
3476                  * destination range does not go past the destination's i_size,
3477                  * we must round down the length to the nearest sector size
3478                  * multiple. If we don't do this adjustment we end replacing
3479                  * with zeroes the bytes in the range that starts at the
3480                  * deduplication range's end offset and ends at the next sector
3481                  * size multiple.
3482                  */
3483                 if (loff + olen == i_size_read(src) &&
3484                     dst_loff + len < i_size_read(dst)) {
3485                         const u64 sz = BTRFS_I(src)->root->fs_info->sectorsize;
3486
3487                         len = round_down(i_size_read(src), sz) - loff;
3488                         olen = len;
3489                 }
3490         }
3491
3492 again:
3493         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, cmp);
3494         if (ret)
3495                 return ret;
3496
3497         if (same_inode)
3498                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3499                                         false);
3500         else
3501                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3502                                                false);
3503         /*
3504          * If one of the inodes has dirty pages in the respective range or
3505          * ordered extents, we need to flush dellaloc and wait for all ordered
3506          * extents in the range. We must unlock the pages and the ranges in the
3507          * io trees to avoid deadlocks when flushing delalloc (requires locking
3508          * pages) and when waiting for ordered extents to complete (they require
3509          * range locking).
3510          */
3511         if (ret == -EAGAIN) {
3512                 /*
3513                  * Ranges in the io trees already unlocked. Now unlock all
3514                  * pages before waiting for all IO to complete.
3515                  */
3516                 btrfs_cmp_data_free(cmp);
3517                 if (same_inode) {
3518                         btrfs_wait_ordered_range(src, same_lock_start,
3519                                                  same_lock_len);
3520                 } else {
3521                         btrfs_wait_ordered_range(src, loff, len);
3522                         btrfs_wait_ordered_range(dst, dst_loff, len);
3523                 }
3524                 goto again;
3525         }
3526         ASSERT(ret == 0);
3527         if (WARN_ON(ret)) {
3528                 /* ranges in the io trees already unlocked */
3529                 btrfs_cmp_data_free(cmp);
3530                 return ret;
3531         }
3532
3533         /* pass original length for comparison so we stay within i_size */
3534         ret = btrfs_cmp_data(olen, cmp);
3535         if (ret == 0)
3536                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3537
3538         if (same_inode)
3539                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3540                               same_lock_start + same_lock_len - 1);
3541         else
3542                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3543
3544         btrfs_cmp_data_free(cmp);
3545
3546         return ret;
3547 }
3548
3549 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3550
3551 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3552                              struct inode *dst, u64 dst_loff)
3553 {
3554         int ret;
3555         struct cmp_pages cmp;
3556         int num_pages = PAGE_ALIGN(BTRFS_MAX_DEDUPE_LEN) >> PAGE_SHIFT;
3557         bool same_inode = (src == dst);
3558         u64 i, tail_len, chunk_count;
3559
3560         if (olen == 0)
3561                 return 0;
3562
3563         if (same_inode)
3564                 inode_lock(src);
3565         else
3566                 btrfs_double_inode_lock(src, dst);
3567
3568         /* don't make the dst file partly checksummed */
3569         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3570             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3571                 ret = -EINVAL;
3572                 goto out_unlock;
3573         }
3574
3575         tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3576         chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3577         if (chunk_count == 0)
3578                 num_pages = PAGE_ALIGN(tail_len) >> PAGE_SHIFT;
3579
3580         /*
3581          * If deduping ranges in the same inode, locking rules make it
3582          * mandatory to always lock pages in ascending order to avoid deadlocks
3583          * with concurrent tasks (such as starting writeback/delalloc).
3584          */
3585         if (same_inode && dst_loff < loff)
3586                 swap(loff, dst_loff);
3587
3588         /*
3589          * We must gather up all the pages before we initiate our extent
3590          * locking. We use an array for the page pointers. Size of the array is
3591          * bounded by len, which is in turn bounded by BTRFS_MAX_DEDUPE_LEN.
3592          */
3593         cmp.src_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3594                                        GFP_KERNEL | __GFP_ZERO);
3595         cmp.dst_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3596                                        GFP_KERNEL | __GFP_ZERO);
3597         if (!cmp.src_pages || !cmp.dst_pages) {
3598                 ret = -ENOMEM;
3599                 goto out_free;
3600         }
3601
3602         for (i = 0; i < chunk_count; i++) {
3603                 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3604                                               dst, dst_loff, &cmp);
3605                 if (ret)
3606                         goto out_free;
3607
3608                 loff += BTRFS_MAX_DEDUPE_LEN;
3609                 dst_loff += BTRFS_MAX_DEDUPE_LEN;
3610         }
3611
3612         if (tail_len > 0)
3613                 ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3614                                               dst_loff, &cmp);
3615
3616 out_free:
3617         kvfree(cmp.src_pages);
3618         kvfree(cmp.dst_pages);
3619
3620 out_unlock:
3621         if (same_inode)
3622                 inode_unlock(src);
3623         else
3624                 btrfs_double_inode_unlock(src, dst);
3625
3626         return ret;
3627 }
3628
3629 int btrfs_dedupe_file_range(struct file *src_file, loff_t src_loff,
3630                             struct file *dst_file, loff_t dst_loff,
3631                             u64 olen)
3632 {
3633         struct inode *src = file_inode(src_file);
3634         struct inode *dst = file_inode(dst_file);
3635         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3636
3637         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3638                 /*
3639                  * Btrfs does not support blocksize < page_size. As a
3640                  * result, btrfs_cmp_data() won't correctly handle
3641                  * this situation without an update.
3642                  */
3643                 return -EINVAL;
3644         }
3645
3646         return btrfs_extent_same(src, src_loff, olen, dst, dst_loff);
3647 }
3648
3649 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3650                                      struct inode *inode,
3651                                      u64 endoff,
3652                                      const u64 destoff,
3653                                      const u64 olen,
3654                                      int no_time_update)
3655 {
3656         struct btrfs_root *root = BTRFS_I(inode)->root;
3657         int ret;
3658
3659         inode_inc_iversion(inode);
3660         if (!no_time_update)
3661                 inode->i_mtime = inode->i_ctime = current_time(inode);
3662         /*
3663          * We round up to the block size at eof when determining which
3664          * extents to clone above, but shouldn't round up the file size.
3665          */
3666         if (endoff > destoff + olen)
3667                 endoff = destoff + olen;
3668         if (endoff > inode->i_size)
3669                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3670
3671         ret = btrfs_update_inode(trans, root, inode);
3672         if (ret) {
3673                 btrfs_abort_transaction(trans, ret);
3674                 btrfs_end_transaction(trans);
3675                 goto out;
3676         }
3677         ret = btrfs_end_transaction(trans);
3678 out:
3679         return ret;
3680 }
3681
3682 static void clone_update_extent_map(struct btrfs_inode *inode,
3683                                     const struct btrfs_trans_handle *trans,
3684                                     const struct btrfs_path *path,
3685                                     const u64 hole_offset,
3686                                     const u64 hole_len)
3687 {
3688         struct extent_map_tree *em_tree = &inode->extent_tree;
3689         struct extent_map *em;
3690         int ret;
3691
3692         em = alloc_extent_map();
3693         if (!em) {
3694                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3695                 return;
3696         }
3697
3698         if (path) {
3699                 struct btrfs_file_extent_item *fi;
3700
3701                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3702                                     struct btrfs_file_extent_item);
3703                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3704                 em->generation = -1;
3705                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3706                     BTRFS_FILE_EXTENT_INLINE)
3707                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3708                                         &inode->runtime_flags);
3709         } else {
3710                 em->start = hole_offset;
3711                 em->len = hole_len;
3712                 em->ram_bytes = em->len;
3713                 em->orig_start = hole_offset;
3714                 em->block_start = EXTENT_MAP_HOLE;
3715                 em->block_len = 0;
3716                 em->orig_block_len = 0;
3717                 em->compress_type = BTRFS_COMPRESS_NONE;
3718                 em->generation = trans->transid;
3719         }
3720
3721         while (1) {
3722                 write_lock(&em_tree->lock);
3723                 ret = add_extent_mapping(em_tree, em, 1);
3724                 write_unlock(&em_tree->lock);
3725                 if (ret != -EEXIST) {
3726                         free_extent_map(em);
3727                         break;
3728                 }
3729                 btrfs_drop_extent_cache(inode, em->start,
3730                                         em->start + em->len - 1, 0);
3731         }
3732
3733         if (ret)
3734                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3735 }
3736
3737 /*
3738  * Make sure we do not end up inserting an inline extent into a file that has
3739  * already other (non-inline) extents. If a file has an inline extent it can
3740  * not have any other extents and the (single) inline extent must start at the
3741  * file offset 0. Failing to respect these rules will lead to file corruption,
3742  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3743  *
3744  * We can have extents that have been already written to disk or we can have
3745  * dirty ranges still in delalloc, in which case the extent maps and items are
3746  * created only when we run delalloc, and the delalloc ranges might fall outside
3747  * the range we are currently locking in the inode's io tree. So we check the
3748  * inode's i_size because of that (i_size updates are done while holding the
3749  * i_mutex, which we are holding here).
3750  * We also check to see if the inode has a size not greater than "datal" but has
3751  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3752  * protected against such concurrent fallocate calls by the i_mutex).
3753  *
3754  * If the file has no extents but a size greater than datal, do not allow the
3755  * copy because we would need turn the inline extent into a non-inline one (even
3756  * with NO_HOLES enabled). If we find our destination inode only has one inline
3757  * extent, just overwrite it with the source inline extent if its size is less
3758  * than the source extent's size, or we could copy the source inline extent's
3759  * data into the destination inode's inline extent if the later is greater then
3760  * the former.
3761  */
3762 static int clone_copy_inline_extent(struct inode *dst,
3763                                     struct btrfs_trans_handle *trans,
3764                                     struct btrfs_path *path,
3765                                     struct btrfs_key *new_key,
3766                                     const u64 drop_start,
3767                                     const u64 datal,
3768                                     const u64 skip,
3769                                     const u64 size,
3770                                     char *inline_data)
3771 {
3772         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3773         struct btrfs_root *root = BTRFS_I(dst)->root;
3774         const u64 aligned_end = ALIGN(new_key->offset + datal,
3775                                       fs_info->sectorsize);
3776         int ret;
3777         struct btrfs_key key;
3778
3779         if (new_key->offset > 0)
3780                 return -EOPNOTSUPP;
3781
3782         key.objectid = btrfs_ino(BTRFS_I(dst));
3783         key.type = BTRFS_EXTENT_DATA_KEY;
3784         key.offset = 0;
3785         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3786         if (ret < 0) {
3787                 return ret;
3788         } else if (ret > 0) {
3789                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3790                         ret = btrfs_next_leaf(root, path);
3791                         if (ret < 0)
3792                                 return ret;
3793                         else if (ret > 0)
3794                                 goto copy_inline_extent;
3795                 }
3796                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3797                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3798                     key.type == BTRFS_EXTENT_DATA_KEY) {
3799                         ASSERT(key.offset > 0);
3800                         return -EOPNOTSUPP;
3801                 }
3802         } else if (i_size_read(dst) <= datal) {
3803                 struct btrfs_file_extent_item *ei;
3804                 u64 ext_len;
3805
3806                 /*
3807                  * If the file size is <= datal, make sure there are no other
3808                  * extents following (can happen do to an fallocate call with
3809                  * the flag FALLOC_FL_KEEP_SIZE).
3810                  */
3811                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3812                                     struct btrfs_file_extent_item);
3813                 /*
3814                  * If it's an inline extent, it can not have other extents
3815                  * following it.
3816                  */
3817                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3818                     BTRFS_FILE_EXTENT_INLINE)
3819                         goto copy_inline_extent;
3820
3821                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3822                 if (ext_len > aligned_end)
3823                         return -EOPNOTSUPP;
3824
3825                 ret = btrfs_next_item(root, path);
3826                 if (ret < 0) {
3827                         return ret;
3828                 } else if (ret == 0) {
3829                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3830                                               path->slots[0]);
3831                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3832                             key.type == BTRFS_EXTENT_DATA_KEY)
3833                                 return -EOPNOTSUPP;
3834                 }
3835         }
3836
3837 copy_inline_extent:
3838         /*
3839          * We have no extent items, or we have an extent at offset 0 which may
3840          * or may not be inlined. All these cases are dealt the same way.
3841          */
3842         if (i_size_read(dst) > datal) {
3843                 /*
3844                  * If the destination inode has an inline extent...
3845                  * This would require copying the data from the source inline
3846                  * extent into the beginning of the destination's inline extent.
3847                  * But this is really complex, both extents can be compressed
3848                  * or just one of them, which would require decompressing and
3849                  * re-compressing data (which could increase the new compressed
3850                  * size, not allowing the compressed data to fit anymore in an
3851                  * inline extent).
3852                  * So just don't support this case for now (it should be rare,
3853                  * we are not really saving space when cloning inline extents).
3854                  */
3855                 return -EOPNOTSUPP;
3856         }
3857
3858         btrfs_release_path(path);
3859         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3860         if (ret)
3861                 return ret;
3862         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3863         if (ret)
3864                 return ret;
3865
3866         if (skip) {
3867                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3868
3869                 memmove(inline_data + start, inline_data + start + skip, datal);
3870         }
3871
3872         write_extent_buffer(path->nodes[0], inline_data,
3873                             btrfs_item_ptr_offset(path->nodes[0],
3874                                                   path->slots[0]),
3875                             size);
3876         inode_add_bytes(dst, datal);
3877
3878         return 0;
3879 }
3880
3881 /**
3882  * btrfs_clone() - clone a range from inode file to another
3883  *
3884  * @src: Inode to clone from
3885  * @inode: Inode to clone to
3886  * @off: Offset within source to start clone from
3887  * @olen: Original length, passed by user, of range to clone
3888  * @olen_aligned: Block-aligned value of olen
3889  * @destoff: Offset within @inode to start clone
3890  * @no_time_update: Whether to update mtime/ctime on the target inode
3891  */
3892 static int btrfs_clone(struct inode *src, struct inode *inode,
3893                        const u64 off, const u64 olen, const u64 olen_aligned,
3894                        const u64 destoff, int no_time_update)
3895 {
3896         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3897         struct btrfs_root *root = BTRFS_I(inode)->root;
3898         struct btrfs_path *path = NULL;
3899         struct extent_buffer *leaf;
3900         struct btrfs_trans_handle *trans;
3901         char *buf = NULL;
3902         struct btrfs_key key;
3903         u32 nritems;
3904         int slot;
3905         int ret;
3906         const u64 len = olen_aligned;
3907         u64 last_dest_end = destoff;
3908
3909         ret = -ENOMEM;
3910         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3911         if (!buf)
3912                 return ret;
3913
3914         path = btrfs_alloc_path();
3915         if (!path) {
3916                 kvfree(buf);
3917                 return ret;
3918         }
3919
3920         path->reada = READA_FORWARD;
3921         /* clone data */
3922         key.objectid = btrfs_ino(BTRFS_I(src));
3923         key.type = BTRFS_EXTENT_DATA_KEY;
3924         key.offset = off;
3925
3926         while (1) {
3927                 u64 next_key_min_offset = key.offset + 1;
3928
3929                 /*
3930                  * note the key will change type as we walk through the
3931                  * tree.
3932                  */
3933                 path->leave_spinning = 1;
3934                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3935                                 0, 0);
3936                 if (ret < 0)
3937                         goto out;
3938                 /*
3939                  * First search, if no extent item that starts at offset off was
3940                  * found but the previous item is an extent item, it's possible
3941                  * it might overlap our target range, therefore process it.
3942                  */
3943                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3944                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3945                                               path->slots[0] - 1);
3946                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3947                                 path->slots[0]--;
3948                 }
3949
3950                 nritems = btrfs_header_nritems(path->nodes[0]);
3951 process_slot:
3952                 if (path->slots[0] >= nritems) {
3953                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3954                         if (ret < 0)
3955                                 goto out;
3956                         if (ret > 0)
3957                                 break;
3958                         nritems = btrfs_header_nritems(path->nodes[0]);
3959                 }
3960                 leaf = path->nodes[0];
3961                 slot = path->slots[0];
3962
3963                 btrfs_item_key_to_cpu(leaf, &key, slot);
3964                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3965                     key.objectid != btrfs_ino(BTRFS_I(src)))
3966                         break;
3967
3968                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3969                         struct btrfs_file_extent_item *extent;
3970                         int type;
3971                         u32 size;
3972                         struct btrfs_key new_key;
3973                         u64 disko = 0, diskl = 0;
3974                         u64 datao = 0, datal = 0;
3975                         u8 comp;
3976                         u64 drop_start;
3977
3978                         extent = btrfs_item_ptr(leaf, slot,
3979                                                 struct btrfs_file_extent_item);
3980                         comp = btrfs_file_extent_compression(leaf, extent);
3981                         type = btrfs_file_extent_type(leaf, extent);
3982                         if (type == BTRFS_FILE_EXTENT_REG ||
3983                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3984                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3985                                                                       extent);
3986                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3987                                                                  extent);
3988                                 datao = btrfs_file_extent_offset(leaf, extent);
3989                                 datal = btrfs_file_extent_num_bytes(leaf,
3990                                                                     extent);
3991                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3992                                 /* take upper bound, may be compressed */
3993                                 datal = btrfs_file_extent_ram_bytes(leaf,
3994                                                                     extent);
3995                         }
3996
3997                         /*
3998                          * The first search might have left us at an extent
3999                          * item that ends before our target range's start, can
4000                          * happen if we have holes and NO_HOLES feature enabled.
4001                          */
4002                         if (key.offset + datal <= off) {
4003                                 path->slots[0]++;
4004                                 goto process_slot;
4005                         } else if (key.offset >= off + len) {
4006                                 break;
4007                         }
4008                         next_key_min_offset = key.offset + datal;
4009                         size = btrfs_item_size_nr(leaf, slot);
4010                         read_extent_buffer(leaf, buf,
4011                                            btrfs_item_ptr_offset(leaf, slot),
4012                                            size);
4013
4014                         btrfs_release_path(path);
4015                         path->leave_spinning = 0;
4016
4017                         memcpy(&new_key, &key, sizeof(new_key));
4018                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
4019                         if (off <= key.offset)
4020                                 new_key.offset = key.offset + destoff - off;
4021                         else
4022                                 new_key.offset = destoff;
4023
4024                         /*
4025                          * Deal with a hole that doesn't have an extent item
4026                          * that represents it (NO_HOLES feature enabled).
4027                          * This hole is either in the middle of the cloning
4028                          * range or at the beginning (fully overlaps it or
4029                          * partially overlaps it).
4030                          */
4031                         if (new_key.offset != last_dest_end)
4032                                 drop_start = last_dest_end;
4033                         else
4034                                 drop_start = new_key.offset;
4035
4036                         /*
4037                          * 1 - adjusting old extent (we may have to split it)
4038                          * 1 - add new extent
4039                          * 1 - inode update
4040                          */
4041                         trans = btrfs_start_transaction(root, 3);
4042                         if (IS_ERR(trans)) {
4043                                 ret = PTR_ERR(trans);
4044                                 goto out;
4045                         }
4046
4047                         if (type == BTRFS_FILE_EXTENT_REG ||
4048                             type == BTRFS_FILE_EXTENT_PREALLOC) {
4049                                 /*
4050                                  *    a  | --- range to clone ---|  b
4051                                  * | ------------- extent ------------- |
4052                                  */
4053
4054                                 /* subtract range b */
4055                                 if (key.offset + datal > off + len)
4056                                         datal = off + len - key.offset;
4057
4058                                 /* subtract range a */
4059                                 if (off > key.offset) {
4060                                         datao += off - key.offset;
4061                                         datal -= off - key.offset;
4062                                 }
4063
4064                                 ret = btrfs_drop_extents(trans, root, inode,
4065                                                          drop_start,
4066                                                          new_key.offset + datal,
4067                                                          1);
4068                                 if (ret) {
4069                                         if (ret != -EOPNOTSUPP)
4070                                                 btrfs_abort_transaction(trans,
4071                                                                         ret);
4072                                         btrfs_end_transaction(trans);
4073                                         goto out;
4074                                 }
4075
4076                                 ret = btrfs_insert_empty_item(trans, root, path,
4077                                                               &new_key, size);
4078                                 if (ret) {
4079                                         btrfs_abort_transaction(trans, ret);
4080                                         btrfs_end_transaction(trans);
4081                                         goto out;
4082                                 }
4083
4084                                 leaf = path->nodes[0];
4085                                 slot = path->slots[0];
4086                                 write_extent_buffer(leaf, buf,
4087                                             btrfs_item_ptr_offset(leaf, slot),
4088                                             size);
4089
4090                                 extent = btrfs_item_ptr(leaf, slot,
4091                                                 struct btrfs_file_extent_item);
4092
4093                                 /* disko == 0 means it's a hole */
4094                                 if (!disko)
4095                                         datao = 0;
4096
4097                                 btrfs_set_file_extent_offset(leaf, extent,
4098                                                              datao);
4099                                 btrfs_set_file_extent_num_bytes(leaf, extent,
4100                                                                 datal);
4101
4102                                 if (disko) {
4103                                         inode_add_bytes(inode, datal);
4104                                         ret = btrfs_inc_extent_ref(trans,
4105                                                         root,
4106                                                         disko, diskl, 0,
4107                                                         root->root_key.objectid,
4108                                                         btrfs_ino(BTRFS_I(inode)),
4109                                                         new_key.offset - datao);
4110                                         if (ret) {
4111                                                 btrfs_abort_transaction(trans,
4112                                                                         ret);
4113                                                 btrfs_end_transaction(trans);
4114                                                 goto out;
4115
4116                                         }
4117                                 }
4118                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
4119                                 u64 skip = 0;
4120                                 u64 trim = 0;
4121
4122                                 if (off > key.offset) {
4123                                         skip = off - key.offset;
4124                                         new_key.offset += skip;
4125                                 }
4126
4127                                 if (key.offset + datal > off + len)
4128                                         trim = key.offset + datal - (off + len);
4129
4130                                 if (comp && (skip || trim)) {
4131                                         ret = -EINVAL;
4132                                         btrfs_end_transaction(trans);
4133                                         goto out;
4134                                 }
4135                                 size -= skip + trim;
4136                                 datal -= skip + trim;
4137
4138                                 ret = clone_copy_inline_extent(inode,
4139                                                                trans, path,
4140                                                                &new_key,
4141                                                                drop_start,
4142                                                                datal,
4143                                                                skip, size, buf);
4144                                 if (ret) {
4145                                         if (ret != -EOPNOTSUPP)
4146                                                 btrfs_abort_transaction(trans,
4147                                                                         ret);
4148                                         btrfs_end_transaction(trans);
4149                                         goto out;
4150                                 }
4151                                 leaf = path->nodes[0];
4152                                 slot = path->slots[0];
4153                         }
4154
4155                         /* If we have an implicit hole (NO_HOLES feature). */
4156                         if (drop_start < new_key.offset)
4157                                 clone_update_extent_map(BTRFS_I(inode), trans,
4158                                                 NULL, drop_start,
4159                                                 new_key.offset - drop_start);
4160
4161                         clone_update_extent_map(BTRFS_I(inode), trans,
4162                                         path, 0, 0);
4163
4164                         btrfs_mark_buffer_dirty(leaf);
4165                         btrfs_release_path(path);
4166
4167                         last_dest_end = ALIGN(new_key.offset + datal,
4168                                               fs_info->sectorsize);
4169                         ret = clone_finish_inode_update(trans, inode,
4170                                                         last_dest_end,
4171                                                         destoff, olen,
4172                                                         no_time_update);
4173                         if (ret)
4174                                 goto out;
4175                         if (new_key.offset + datal >= destoff + len)
4176                                 break;
4177                 }
4178                 btrfs_release_path(path);
4179                 key.offset = next_key_min_offset;
4180
4181                 if (fatal_signal_pending(current)) {
4182                         ret = -EINTR;
4183                         goto out;
4184                 }
4185         }
4186         ret = 0;
4187
4188         if (last_dest_end < destoff + len) {
4189                 /*
4190                  * We have an implicit hole (NO_HOLES feature is enabled) that
4191                  * fully or partially overlaps our cloning range at its end.
4192                  */
4193                 btrfs_release_path(path);
4194
4195                 /*
4196                  * 1 - remove extent(s)
4197                  * 1 - inode update
4198                  */
4199                 trans = btrfs_start_transaction(root, 2);
4200                 if (IS_ERR(trans)) {
4201                         ret = PTR_ERR(trans);
4202                         goto out;
4203                 }
4204                 ret = btrfs_drop_extents(trans, root, inode,
4205                                          last_dest_end, destoff + len, 1);
4206                 if (ret) {
4207                         if (ret != -EOPNOTSUPP)
4208                                 btrfs_abort_transaction(trans, ret);
4209                         btrfs_end_transaction(trans);
4210                         goto out;
4211                 }
4212                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
4213                                 last_dest_end,
4214                                 destoff + len - last_dest_end);
4215                 ret = clone_finish_inode_update(trans, inode, destoff + len,
4216                                                 destoff, olen, no_time_update);
4217         }
4218
4219 out:
4220         btrfs_free_path(path);
4221         kvfree(buf);
4222         return ret;
4223 }
4224
4225 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
4226                                         u64 off, u64 olen, u64 destoff)
4227 {
4228         struct inode *inode = file_inode(file);
4229         struct inode *src = file_inode(file_src);
4230         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4231         struct btrfs_root *root = BTRFS_I(inode)->root;
4232         int ret;
4233         u64 len = olen;
4234         u64 bs = fs_info->sb->s_blocksize;
4235         int same_inode = src == inode;
4236
4237         /*
4238          * TODO:
4239          * - split compressed inline extents.  annoying: we need to
4240          *   decompress into destination's address_space (the file offset
4241          *   may change, so source mapping won't do), then recompress (or
4242          *   otherwise reinsert) a subrange.
4243          *
4244          * - split destination inode's inline extents.  The inline extents can
4245          *   be either compressed or non-compressed.
4246          */
4247
4248         if (btrfs_root_readonly(root))
4249                 return -EROFS;
4250
4251         if (file_src->f_path.mnt != file->f_path.mnt ||
4252             src->i_sb != inode->i_sb)
4253                 return -EXDEV;
4254
4255         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
4256                 return -EISDIR;
4257
4258         if (!same_inode) {
4259                 btrfs_double_inode_lock(src, inode);
4260         } else {
4261                 inode_lock(src);
4262         }
4263
4264         /* don't make the dst file partly checksummed */
4265         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
4266             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
4267                 ret = -EINVAL;
4268                 goto out_unlock;
4269         }
4270
4271         /* determine range to clone */
4272         ret = -EINVAL;
4273         if (off + len > src->i_size || off + len < off)
4274                 goto out_unlock;
4275         if (len == 0)
4276                 olen = len = src->i_size - off;
4277         /* if we extend to eof, continue to block boundary */
4278         if (off + len == src->i_size)
4279                 len = ALIGN(src->i_size, bs) - off;
4280
4281         if (len == 0) {
4282                 ret = 0;
4283                 goto out_unlock;
4284         }
4285
4286         /* verify the end result is block aligned */
4287         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
4288             !IS_ALIGNED(destoff, bs))
4289                 goto out_unlock;
4290
4291         /* verify if ranges are overlapped within the same file */
4292         if (same_inode) {
4293                 if (destoff + len > off && destoff < off + len)
4294                         goto out_unlock;
4295         }
4296
4297         if (destoff > inode->i_size) {
4298                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
4299                 if (ret)
4300                         goto out_unlock;
4301         }
4302
4303         /*
4304          * Lock the target range too. Right after we replace the file extent
4305          * items in the fs tree (which now point to the cloned data), we might
4306          * have a worker replace them with extent items relative to a write
4307          * operation that was issued before this clone operation (i.e. confront
4308          * with inode.c:btrfs_finish_ordered_io).
4309          */
4310         if (same_inode) {
4311                 u64 lock_start = min_t(u64, off, destoff);
4312                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
4313
4314                 ret = lock_extent_range(src, lock_start, lock_len, true);
4315         } else {
4316                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
4317                                                true);
4318         }
4319         ASSERT(ret == 0);
4320         if (WARN_ON(ret)) {
4321                 /* ranges in the io trees already unlocked */
4322                 goto out_unlock;
4323         }
4324
4325         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
4326
4327         if (same_inode) {
4328                 u64 lock_start = min_t(u64, off, destoff);
4329                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
4330
4331                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
4332         } else {
4333                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
4334         }
4335         /*
4336          * Truncate page cache pages so that future reads will see the cloned
4337          * data immediately and not the previous data.
4338          */
4339         truncate_inode_pages_range(&inode->i_data,
4340                                 round_down(destoff, PAGE_SIZE),
4341                                 round_up(destoff + len, PAGE_SIZE) - 1);
4342 out_unlock:
4343         if (!same_inode)
4344                 btrfs_double_inode_unlock(src, inode);
4345         else
4346                 inode_unlock(src);
4347         return ret;
4348 }
4349
4350 int btrfs_clone_file_range(struct file *src_file, loff_t off,
4351                 struct file *dst_file, loff_t destoff, u64 len)
4352 {
4353         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
4354 }
4355
4356 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4357 {
4358         struct inode *inode = file_inode(file);
4359         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4360         struct btrfs_root *root = BTRFS_I(inode)->root;
4361         struct btrfs_root *new_root;
4362         struct btrfs_dir_item *di;
4363         struct btrfs_trans_handle *trans;
4364         struct btrfs_path *path;
4365         struct btrfs_key location;
4366         struct btrfs_disk_key disk_key;
4367         u64 objectid = 0;
4368         u64 dir_id;
4369         int ret;
4370
4371         if (!capable(CAP_SYS_ADMIN))
4372                 return -EPERM;
4373
4374         ret = mnt_want_write_file(file);
4375         if (ret)
4376                 return ret;
4377
4378         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4379                 ret = -EFAULT;
4380                 goto out;
4381         }
4382
4383         if (!objectid)
4384                 objectid = BTRFS_FS_TREE_OBJECTID;
4385
4386         location.objectid = objectid;
4387         location.type = BTRFS_ROOT_ITEM_KEY;
4388         location.offset = (u64)-1;
4389
4390         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4391         if (IS_ERR(new_root)) {
4392                 ret = PTR_ERR(new_root);
4393                 goto out;
4394         }
4395         if (!is_fstree(new_root->root_key.objectid)) {
4396                 ret = -ENOENT;
4397                 goto out;
4398         }
4399
4400         path = btrfs_alloc_path();
4401         if (!path) {
4402                 ret = -ENOMEM;
4403                 goto out;
4404         }
4405         path->leave_spinning = 1;
4406
4407         trans = btrfs_start_transaction(root, 1);
4408         if (IS_ERR(trans)) {
4409                 btrfs_free_path(path);
4410                 ret = PTR_ERR(trans);
4411                 goto out;
4412         }
4413
4414         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4415         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4416                                    dir_id, "default", 7, 1);
4417         if (IS_ERR_OR_NULL(di)) {
4418                 btrfs_free_path(path);
4419                 btrfs_end_transaction(trans);
4420                 btrfs_err(fs_info,
4421                           "Umm, you don't have the default diritem, this isn't going to work");
4422                 ret = -ENOENT;
4423                 goto out;
4424         }
4425
4426         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4427         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4428         btrfs_mark_buffer_dirty(path->nodes[0]);
4429         btrfs_free_path(path);
4430
4431         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4432         btrfs_end_transaction(trans);
4433 out:
4434         mnt_drop_write_file(file);
4435         return ret;
4436 }
4437
4438 static void get_block_group_info(struct list_head *groups_list,
4439                                  struct btrfs_ioctl_space_info *space)
4440 {
4441         struct btrfs_block_group_cache *block_group;
4442
4443         space->total_bytes = 0;
4444         space->used_bytes = 0;
4445         space->flags = 0;
4446         list_for_each_entry(block_group, groups_list, list) {
4447                 space->flags = block_group->flags;
4448                 space->total_bytes += block_group->key.offset;
4449                 space->used_bytes +=
4450                         btrfs_block_group_used(&block_group->item);
4451         }
4452 }
4453
4454 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4455                                    void __user *arg)
4456 {
4457         struct btrfs_ioctl_space_args space_args;
4458         struct btrfs_ioctl_space_info space;
4459         struct btrfs_ioctl_space_info *dest;
4460         struct btrfs_ioctl_space_info *dest_orig;
4461         struct btrfs_ioctl_space_info __user *user_dest;
4462         struct btrfs_space_info *info;
4463         static const u64 types[] = {
4464                 BTRFS_BLOCK_GROUP_DATA,
4465                 BTRFS_BLOCK_GROUP_SYSTEM,
4466                 BTRFS_BLOCK_GROUP_METADATA,
4467                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4468         };
4469         int num_types = 4;
4470         int alloc_size;
4471         int ret = 0;
4472         u64 slot_count = 0;
4473         int i, c;
4474
4475         if (copy_from_user(&space_args,
4476                            (struct btrfs_ioctl_space_args __user *)arg,
4477                            sizeof(space_args)))
4478                 return -EFAULT;
4479
4480         for (i = 0; i < num_types; i++) {
4481                 struct btrfs_space_info *tmp;
4482
4483                 info = NULL;
4484                 rcu_read_lock();
4485                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4486                                         list) {
4487                         if (tmp->flags == types[i]) {
4488                                 info = tmp;
4489                                 break;
4490                         }
4491                 }
4492                 rcu_read_unlock();
4493
4494                 if (!info)
4495                         continue;
4496
4497                 down_read(&info->groups_sem);
4498                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4499                         if (!list_empty(&info->block_groups[c]))
4500                                 slot_count++;
4501                 }
4502                 up_read(&info->groups_sem);
4503         }
4504
4505         /*
4506          * Global block reserve, exported as a space_info
4507          */
4508         slot_count++;
4509
4510         /* space_slots == 0 means they are asking for a count */
4511         if (space_args.space_slots == 0) {
4512                 space_args.total_spaces = slot_count;
4513                 goto out;
4514         }
4515
4516         slot_count = min_t(u64, space_args.space_slots, slot_count);
4517
4518         alloc_size = sizeof(*dest) * slot_count;
4519
4520         /* we generally have at most 6 or so space infos, one for each raid
4521          * level.  So, a whole page should be more than enough for everyone
4522          */
4523         if (alloc_size > PAGE_SIZE)
4524                 return -ENOMEM;
4525
4526         space_args.total_spaces = 0;
4527         dest = kmalloc(alloc_size, GFP_KERNEL);
4528         if (!dest)
4529                 return -ENOMEM;
4530         dest_orig = dest;
4531
4532         /* now we have a buffer to copy into */
4533         for (i = 0; i < num_types; i++) {
4534                 struct btrfs_space_info *tmp;
4535
4536                 if (!slot_count)
4537                         break;
4538
4539                 info = NULL;
4540                 rcu_read_lock();
4541                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4542                                         list) {
4543                         if (tmp->flags == types[i]) {
4544                                 info = tmp;
4545                                 break;
4546                         }
4547                 }
4548                 rcu_read_unlock();
4549
4550                 if (!info)
4551                         continue;
4552                 down_read(&info->groups_sem);
4553                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4554                         if (!list_empty(&info->block_groups[c])) {
4555                                 get_block_group_info(&info->block_groups[c],
4556                                                      &space);
4557                                 memcpy(dest, &space, sizeof(space));
4558                                 dest++;
4559                                 space_args.total_spaces++;
4560                                 slot_count--;
4561                         }
4562                         if (!slot_count)
4563                                 break;
4564                 }
4565                 up_read(&info->groups_sem);
4566         }
4567
4568         /*
4569          * Add global block reserve
4570          */
4571         if (slot_count) {
4572                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4573
4574                 spin_lock(&block_rsv->lock);
4575                 space.total_bytes = block_rsv->size;
4576                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4577                 spin_unlock(&block_rsv->lock);
4578                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4579                 memcpy(dest, &space, sizeof(space));
4580                 space_args.total_spaces++;
4581         }
4582
4583         user_dest = (struct btrfs_ioctl_space_info __user *)
4584                 (arg + sizeof(struct btrfs_ioctl_space_args));
4585
4586         if (copy_to_user(user_dest, dest_orig, alloc_size))
4587                 ret = -EFAULT;
4588
4589         kfree(dest_orig);
4590 out:
4591         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4592                 ret = -EFAULT;
4593
4594         return ret;
4595 }
4596
4597 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4598                                             void __user *argp)
4599 {
4600         struct btrfs_trans_handle *trans;
4601         u64 transid;
4602         int ret;
4603
4604         trans = btrfs_attach_transaction_barrier(root);
4605         if (IS_ERR(trans)) {
4606                 if (PTR_ERR(trans) != -ENOENT)
4607                         return PTR_ERR(trans);
4608
4609                 /* No running transaction, don't bother */
4610                 transid = root->fs_info->last_trans_committed;
4611                 goto out;
4612         }
4613         transid = trans->transid;
4614         ret = btrfs_commit_transaction_async(trans, 0);
4615         if (ret) {
4616                 btrfs_end_transaction(trans);
4617                 return ret;
4618         }
4619 out:
4620         if (argp)
4621                 if (copy_to_user(argp, &transid, sizeof(transid)))
4622                         return -EFAULT;
4623         return 0;
4624 }
4625
4626 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4627                                            void __user *argp)
4628 {
4629         u64 transid;
4630
4631         if (argp) {
4632                 if (copy_from_user(&transid, argp, sizeof(transid)))
4633                         return -EFAULT;
4634         } else {
4635                 transid = 0;  /* current trans */
4636         }
4637         return btrfs_wait_for_commit(fs_info, transid);
4638 }
4639
4640 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4641 {
4642         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4643         struct btrfs_ioctl_scrub_args *sa;
4644         int ret;
4645
4646         if (!capable(CAP_SYS_ADMIN))
4647                 return -EPERM;
4648
4649         sa = memdup_user(arg, sizeof(*sa));
4650         if (IS_ERR(sa))
4651                 return PTR_ERR(sa);
4652
4653         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4654                 ret = mnt_want_write_file(file);
4655                 if (ret)
4656                         goto out;
4657         }
4658
4659         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4660                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4661                               0);
4662
4663         if (copy_to_user(arg, sa, sizeof(*sa)))
4664                 ret = -EFAULT;
4665
4666         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4667                 mnt_drop_write_file(file);
4668 out:
4669         kfree(sa);
4670         return ret;
4671 }
4672
4673 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4674 {
4675         if (!capable(CAP_SYS_ADMIN))
4676                 return -EPERM;
4677
4678         return btrfs_scrub_cancel(fs_info);
4679 }
4680
4681 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4682                                        void __user *arg)
4683 {
4684         struct btrfs_ioctl_scrub_args *sa;
4685         int ret;
4686
4687         if (!capable(CAP_SYS_ADMIN))
4688                 return -EPERM;
4689
4690         sa = memdup_user(arg, sizeof(*sa));
4691         if (IS_ERR(sa))
4692                 return PTR_ERR(sa);
4693
4694         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4695
4696         if (copy_to_user(arg, sa, sizeof(*sa)))
4697                 ret = -EFAULT;
4698
4699         kfree(sa);
4700         return ret;
4701 }
4702
4703 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4704                                       void __user *arg)
4705 {
4706         struct btrfs_ioctl_get_dev_stats *sa;
4707         int ret;
4708
4709         sa = memdup_user(arg, sizeof(*sa));
4710         if (IS_ERR(sa))
4711                 return PTR_ERR(sa);
4712
4713         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4714                 kfree(sa);
4715                 return -EPERM;
4716         }
4717
4718         ret = btrfs_get_dev_stats(fs_info, sa);
4719
4720         if (copy_to_user(arg, sa, sizeof(*sa)))
4721                 ret = -EFAULT;
4722
4723         kfree(sa);
4724         return ret;
4725 }
4726
4727 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4728                                     void __user *arg)
4729 {
4730         struct btrfs_ioctl_dev_replace_args *p;
4731         int ret;
4732
4733         if (!capable(CAP_SYS_ADMIN))
4734                 return -EPERM;
4735
4736         p = memdup_user(arg, sizeof(*p));
4737         if (IS_ERR(p))
4738                 return PTR_ERR(p);
4739
4740         switch (p->cmd) {
4741         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4742                 if (sb_rdonly(fs_info->sb)) {
4743                         ret = -EROFS;
4744                         goto out;
4745                 }
4746                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4747                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4748                 } else {
4749                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4750                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4751                 }
4752                 break;
4753         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4754                 btrfs_dev_replace_status(fs_info, p);
4755                 ret = 0;
4756                 break;
4757         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4758                 p->result = btrfs_dev_replace_cancel(fs_info);
4759                 ret = 0;
4760                 break;
4761         default:
4762                 ret = -EINVAL;
4763                 break;
4764         }
4765
4766         if (copy_to_user(arg, p, sizeof(*p)))
4767                 ret = -EFAULT;
4768 out:
4769         kfree(p);
4770         return ret;
4771 }
4772
4773 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4774 {
4775         int ret = 0;
4776         int i;
4777         u64 rel_ptr;
4778         int size;
4779         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4780         struct inode_fs_paths *ipath = NULL;
4781         struct btrfs_path *path;
4782
4783         if (!capable(CAP_DAC_READ_SEARCH))
4784                 return -EPERM;
4785
4786         path = btrfs_alloc_path();
4787         if (!path) {
4788                 ret = -ENOMEM;
4789                 goto out;
4790         }
4791
4792         ipa = memdup_user(arg, sizeof(*ipa));
4793         if (IS_ERR(ipa)) {
4794                 ret = PTR_ERR(ipa);
4795                 ipa = NULL;
4796                 goto out;
4797         }
4798
4799         size = min_t(u32, ipa->size, 4096);
4800         ipath = init_ipath(size, root, path);
4801         if (IS_ERR(ipath)) {
4802                 ret = PTR_ERR(ipath);
4803                 ipath = NULL;
4804                 goto out;
4805         }
4806
4807         ret = paths_from_inode(ipa->inum, ipath);
4808         if (ret < 0)
4809                 goto out;
4810
4811         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4812                 rel_ptr = ipath->fspath->val[i] -
4813                           (u64)(unsigned long)ipath->fspath->val;
4814                 ipath->fspath->val[i] = rel_ptr;
4815         }
4816
4817         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4818                            ipath->fspath, size);
4819         if (ret) {
4820                 ret = -EFAULT;
4821                 goto out;
4822         }
4823
4824 out:
4825         btrfs_free_path(path);
4826         free_ipath(ipath);
4827         kfree(ipa);
4828
4829         return ret;
4830 }
4831
4832 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4833 {
4834         struct btrfs_data_container *inodes = ctx;
4835         const size_t c = 3 * sizeof(u64);
4836
4837         if (inodes->bytes_left >= c) {
4838                 inodes->bytes_left -= c;
4839                 inodes->val[inodes->elem_cnt] = inum;
4840                 inodes->val[inodes->elem_cnt + 1] = offset;
4841                 inodes->val[inodes->elem_cnt + 2] = root;
4842                 inodes->elem_cnt += 3;
4843         } else {
4844                 inodes->bytes_missing += c - inodes->bytes_left;
4845                 inodes->bytes_left = 0;
4846                 inodes->elem_missed += 3;
4847         }
4848
4849         return 0;
4850 }
4851
4852 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4853                                         void __user *arg, int version)
4854 {
4855         int ret = 0;
4856         int size;
4857         struct btrfs_ioctl_logical_ino_args *loi;
4858         struct btrfs_data_container *inodes = NULL;
4859         struct btrfs_path *path = NULL;
4860         bool ignore_offset;
4861
4862         if (!capable(CAP_SYS_ADMIN))
4863                 return -EPERM;
4864
4865         loi = memdup_user(arg, sizeof(*loi));
4866         if (IS_ERR(loi))
4867                 return PTR_ERR(loi);
4868
4869         if (version == 1) {
4870                 ignore_offset = false;
4871                 size = min_t(u32, loi->size, SZ_64K);
4872         } else {
4873                 /* All reserved bits must be 0 for now */
4874                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4875                         ret = -EINVAL;
4876                         goto out_loi;
4877                 }
4878                 /* Only accept flags we have defined so far */
4879                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4880                         ret = -EINVAL;
4881                         goto out_loi;
4882                 }
4883                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4884                 size = min_t(u32, loi->size, SZ_16M);
4885         }
4886
4887         path = btrfs_alloc_path();
4888         if (!path) {
4889                 ret = -ENOMEM;
4890                 goto out;
4891         }
4892
4893         inodes = init_data_container(size);
4894         if (IS_ERR(inodes)) {
4895                 ret = PTR_ERR(inodes);
4896                 inodes = NULL;
4897                 goto out;
4898         }
4899
4900         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4901                                           build_ino_list, inodes, ignore_offset);
4902         if (ret == -EINVAL)
4903                 ret = -ENOENT;
4904         if (ret < 0)
4905                 goto out;
4906
4907         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4908                            size);
4909         if (ret)
4910                 ret = -EFAULT;
4911
4912 out:
4913         btrfs_free_path(path);
4914         kvfree(inodes);
4915 out_loi:
4916         kfree(loi);
4917
4918         return ret;
4919 }
4920
4921 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4922                                struct btrfs_ioctl_balance_args *bargs)
4923 {
4924         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4925
4926         bargs->flags = bctl->flags;
4927
4928         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4929                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4930         if (atomic_read(&fs_info->balance_pause_req))
4931                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4932         if (atomic_read(&fs_info->balance_cancel_req))
4933                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4934
4935         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4936         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4937         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4938
4939         spin_lock(&fs_info->balance_lock);
4940         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4941         spin_unlock(&fs_info->balance_lock);
4942 }
4943
4944 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4945 {
4946         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4947         struct btrfs_fs_info *fs_info = root->fs_info;
4948         struct btrfs_ioctl_balance_args *bargs;
4949         struct btrfs_balance_control *bctl;
4950         bool need_unlock; /* for mut. excl. ops lock */
4951         int ret;
4952
4953         if (!capable(CAP_SYS_ADMIN))
4954                 return -EPERM;
4955
4956         ret = mnt_want_write_file(file);
4957         if (ret)
4958                 return ret;
4959
4960 again:
4961         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4962                 mutex_lock(&fs_info->balance_mutex);
4963                 need_unlock = true;
4964                 goto locked;
4965         }
4966
4967         /*
4968          * mut. excl. ops lock is locked.  Three possibilities:
4969          *   (1) some other op is running
4970          *   (2) balance is running
4971          *   (3) balance is paused -- special case (think resume)
4972          */
4973         mutex_lock(&fs_info->balance_mutex);
4974         if (fs_info->balance_ctl) {
4975                 /* this is either (2) or (3) */
4976                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4977                         mutex_unlock(&fs_info->balance_mutex);
4978                         /*
4979                          * Lock released to allow other waiters to continue,
4980                          * we'll reexamine the status again.
4981                          */
4982                         mutex_lock(&fs_info->balance_mutex);
4983
4984                         if (fs_info->balance_ctl &&
4985                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4986                                 /* this is (3) */
4987                                 need_unlock = false;
4988                                 goto locked;
4989                         }
4990
4991                         mutex_unlock(&fs_info->balance_mutex);
4992                         goto again;
4993                 } else {
4994                         /* this is (2) */
4995                         mutex_unlock(&fs_info->balance_mutex);
4996                         ret = -EINPROGRESS;
4997                         goto out;
4998                 }
4999         } else {
5000                 /* this is (1) */
5001                 mutex_unlock(&fs_info->balance_mutex);
5002                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
5003                 goto out;
5004         }
5005
5006 locked:
5007         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
5008
5009         if (arg) {
5010                 bargs = memdup_user(arg, sizeof(*bargs));
5011                 if (IS_ERR(bargs)) {
5012                         ret = PTR_ERR(bargs);
5013                         goto out_unlock;
5014                 }
5015
5016                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
5017                         if (!fs_info->balance_ctl) {
5018                                 ret = -ENOTCONN;
5019                                 goto out_bargs;
5020                         }
5021
5022                         bctl = fs_info->balance_ctl;
5023                         spin_lock(&fs_info->balance_lock);
5024                         bctl->flags |= BTRFS_BALANCE_RESUME;
5025                         spin_unlock(&fs_info->balance_lock);
5026
5027                         goto do_balance;
5028                 }
5029         } else {
5030                 bargs = NULL;
5031         }
5032
5033         if (fs_info->balance_ctl) {
5034                 ret = -EINPROGRESS;
5035                 goto out_bargs;
5036         }
5037
5038         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
5039         if (!bctl) {
5040                 ret = -ENOMEM;
5041                 goto out_bargs;
5042         }
5043
5044         if (arg) {
5045                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
5046                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
5047                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
5048
5049                 bctl->flags = bargs->flags;
5050         } else {
5051                 /* balance everything - no filters */
5052                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
5053         }
5054
5055         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
5056                 ret = -EINVAL;
5057                 goto out_bctl;
5058         }
5059
5060 do_balance:
5061         /*
5062          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
5063          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
5064          * restriper was paused all the way until unmount, in free_fs_info.
5065          * The flag should be cleared after reset_balance_state.
5066          */
5067         need_unlock = false;
5068
5069         ret = btrfs_balance(fs_info, bctl, bargs);
5070         bctl = NULL;
5071
5072         if (arg) {
5073                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
5074                         ret = -EFAULT;
5075         }
5076
5077 out_bctl:
5078         kfree(bctl);
5079 out_bargs:
5080         kfree(bargs);
5081 out_unlock:
5082         mutex_unlock(&fs_info->balance_mutex);
5083         if (need_unlock)
5084                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
5085 out:
5086         mnt_drop_write_file(file);
5087         return ret;
5088 }
5089
5090 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
5091 {
5092         if (!capable(CAP_SYS_ADMIN))
5093                 return -EPERM;
5094
5095         switch (cmd) {
5096         case BTRFS_BALANCE_CTL_PAUSE:
5097                 return btrfs_pause_balance(fs_info);
5098         case BTRFS_BALANCE_CTL_CANCEL:
5099                 return btrfs_cancel_balance(fs_info);
5100         }
5101
5102         return -EINVAL;
5103 }
5104
5105 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
5106                                          void __user *arg)
5107 {
5108         struct btrfs_ioctl_balance_args *bargs;
5109         int ret = 0;
5110
5111         if (!capable(CAP_SYS_ADMIN))
5112                 return -EPERM;
5113
5114         mutex_lock(&fs_info->balance_mutex);
5115         if (!fs_info->balance_ctl) {
5116                 ret = -ENOTCONN;
5117                 goto out;
5118         }
5119
5120         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
5121         if (!bargs) {
5122                 ret = -ENOMEM;
5123                 goto out;
5124         }
5125
5126         btrfs_update_ioctl_balance_args(fs_info, bargs);
5127
5128         if (copy_to_user(arg, bargs, sizeof(*bargs)))
5129                 ret = -EFAULT;
5130
5131         kfree(bargs);
5132 out:
5133         mutex_unlock(&fs_info->balance_mutex);
5134         return ret;
5135 }
5136
5137 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
5138 {
5139         struct inode *inode = file_inode(file);
5140         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5141         struct btrfs_ioctl_quota_ctl_args *sa;
5142         int ret;
5143
5144         if (!capable(CAP_SYS_ADMIN))
5145                 return -EPERM;
5146
5147         ret = mnt_want_write_file(file);
5148         if (ret)
5149                 return ret;
5150
5151         sa = memdup_user(arg, sizeof(*sa));
5152         if (IS_ERR(sa)) {
5153                 ret = PTR_ERR(sa);
5154                 goto drop_write;
5155         }
5156
5157         down_write(&fs_info->subvol_sem);
5158
5159         switch (sa->cmd) {
5160         case BTRFS_QUOTA_CTL_ENABLE:
5161                 ret = btrfs_quota_enable(fs_info);
5162                 break;
5163         case BTRFS_QUOTA_CTL_DISABLE:
5164                 ret = btrfs_quota_disable(fs_info);
5165                 break;
5166         default:
5167                 ret = -EINVAL;
5168                 break;
5169         }
5170
5171         kfree(sa);
5172         up_write(&fs_info->subvol_sem);
5173 drop_write:
5174         mnt_drop_write_file(file);
5175         return ret;
5176 }
5177
5178 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
5179 {
5180         struct inode *inode = file_inode(file);
5181         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5182         struct btrfs_root *root = BTRFS_I(inode)->root;
5183         struct btrfs_ioctl_qgroup_assign_args *sa;
5184         struct btrfs_trans_handle *trans;
5185         int ret;
5186         int err;
5187
5188         if (!capable(CAP_SYS_ADMIN))
5189                 return -EPERM;
5190
5191         ret = mnt_want_write_file(file);
5192         if (ret)
5193                 return ret;
5194
5195         sa = memdup_user(arg, sizeof(*sa));
5196         if (IS_ERR(sa)) {
5197                 ret = PTR_ERR(sa);
5198                 goto drop_write;
5199         }
5200
5201         trans = btrfs_join_transaction(root);
5202         if (IS_ERR(trans)) {
5203                 ret = PTR_ERR(trans);
5204                 goto out;
5205         }
5206
5207         if (sa->assign) {
5208                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
5209         } else {
5210                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
5211         }
5212
5213         /* update qgroup status and info */
5214         err = btrfs_run_qgroups(trans);
5215         if (err < 0)
5216                 btrfs_handle_fs_error(fs_info, err,
5217                                       "failed to update qgroup status and info");
5218         err = btrfs_end_transaction(trans);
5219         if (err && !ret)
5220                 ret = err;
5221
5222 out:
5223         kfree(sa);
5224 drop_write:
5225         mnt_drop_write_file(file);
5226         return ret;
5227 }
5228
5229 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
5230 {
5231         struct inode *inode = file_inode(file);
5232         struct btrfs_root *root = BTRFS_I(inode)->root;
5233         struct btrfs_ioctl_qgroup_create_args *sa;
5234         struct btrfs_trans_handle *trans;
5235         int ret;
5236         int err;
5237
5238         if (!capable(CAP_SYS_ADMIN))
5239                 return -EPERM;
5240
5241         ret = mnt_want_write_file(file);
5242         if (ret)
5243                 return ret;
5244
5245         sa = memdup_user(arg, sizeof(*sa));
5246         if (IS_ERR(sa)) {
5247                 ret = PTR_ERR(sa);
5248                 goto drop_write;
5249         }
5250
5251         if (!sa->qgroupid) {
5252                 ret = -EINVAL;
5253                 goto out;
5254         }
5255
5256         trans = btrfs_join_transaction(root);
5257         if (IS_ERR(trans)) {
5258                 ret = PTR_ERR(trans);
5259                 goto out;
5260         }
5261
5262         if (sa->create) {
5263                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
5264         } else {
5265                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
5266         }
5267
5268         err = btrfs_end_transaction(trans);
5269         if (err && !ret)
5270                 ret = err;
5271
5272 out:
5273         kfree(sa);
5274 drop_write:
5275         mnt_drop_write_file(file);
5276         return ret;
5277 }
5278
5279 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
5280 {
5281         struct inode *inode = file_inode(file);
5282         struct btrfs_root *root = BTRFS_I(inode)->root;
5283         struct btrfs_ioctl_qgroup_limit_args *sa;
5284         struct btrfs_trans_handle *trans;
5285         int ret;
5286         int err;
5287         u64 qgroupid;
5288
5289         if (!capable(CAP_SYS_ADMIN))
5290                 return -EPERM;
5291
5292         ret = mnt_want_write_file(file);
5293         if (ret)
5294                 return ret;
5295
5296         sa = memdup_user(arg, sizeof(*sa));
5297         if (IS_ERR(sa)) {
5298                 ret = PTR_ERR(sa);
5299                 goto drop_write;
5300         }
5301
5302         trans = btrfs_join_transaction(root);
5303         if (IS_ERR(trans)) {
5304                 ret = PTR_ERR(trans);
5305                 goto out;
5306         }
5307
5308         qgroupid = sa->qgroupid;
5309         if (!qgroupid) {
5310                 /* take the current subvol as qgroup */
5311                 qgroupid = root->root_key.objectid;
5312         }
5313
5314         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
5315
5316         err = btrfs_end_transaction(trans);
5317         if (err && !ret)
5318                 ret = err;
5319
5320 out:
5321         kfree(sa);
5322 drop_write:
5323         mnt_drop_write_file(file);
5324         return ret;
5325 }
5326
5327 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5328 {
5329         struct inode *inode = file_inode(file);
5330         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5331         struct btrfs_ioctl_quota_rescan_args *qsa;
5332         int ret;
5333
5334         if (!capable(CAP_SYS_ADMIN))
5335                 return -EPERM;
5336
5337         ret = mnt_want_write_file(file);
5338         if (ret)
5339                 return ret;
5340
5341         qsa = memdup_user(arg, sizeof(*qsa));
5342         if (IS_ERR(qsa)) {
5343                 ret = PTR_ERR(qsa);
5344                 goto drop_write;
5345         }
5346
5347         if (qsa->flags) {
5348                 ret = -EINVAL;
5349                 goto out;
5350         }
5351
5352         ret = btrfs_qgroup_rescan(fs_info);
5353
5354 out:
5355         kfree(qsa);
5356 drop_write:
5357         mnt_drop_write_file(file);
5358         return ret;
5359 }
5360
5361 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5362 {
5363         struct inode *inode = file_inode(file);
5364         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5365         struct btrfs_ioctl_quota_rescan_args *qsa;
5366         int ret = 0;
5367
5368         if (!capable(CAP_SYS_ADMIN))
5369                 return -EPERM;
5370
5371         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5372         if (!qsa)
5373                 return -ENOMEM;
5374
5375         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5376                 qsa->flags = 1;
5377                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5378         }
5379
5380         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5381                 ret = -EFAULT;
5382
5383         kfree(qsa);
5384         return ret;
5385 }
5386
5387 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5388 {
5389         struct inode *inode = file_inode(file);
5390         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5391
5392         if (!capable(CAP_SYS_ADMIN))
5393                 return -EPERM;
5394
5395         return btrfs_qgroup_wait_for_completion(fs_info, true);
5396 }
5397
5398 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5399                                             struct btrfs_ioctl_received_subvol_args *sa)
5400 {
5401         struct inode *inode = file_inode(file);
5402         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5403         struct btrfs_root *root = BTRFS_I(inode)->root;
5404         struct btrfs_root_item *root_item = &root->root_item;
5405         struct btrfs_trans_handle *trans;
5406         struct timespec64 ct = current_time(inode);
5407         int ret = 0;
5408         int received_uuid_changed;
5409
5410         if (!inode_owner_or_capable(inode))
5411                 return -EPERM;
5412
5413         ret = mnt_want_write_file(file);
5414         if (ret < 0)
5415                 return ret;
5416
5417         down_write(&fs_info->subvol_sem);
5418
5419         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5420                 ret = -EINVAL;
5421                 goto out;
5422         }
5423
5424         if (btrfs_root_readonly(root)) {
5425                 ret = -EROFS;
5426                 goto out;
5427         }
5428
5429         /*
5430          * 1 - root item
5431          * 2 - uuid items (received uuid + subvol uuid)
5432          */
5433         trans = btrfs_start_transaction(root, 3);
5434         if (IS_ERR(trans)) {
5435                 ret = PTR_ERR(trans);
5436                 trans = NULL;
5437                 goto out;
5438         }
5439
5440         sa->rtransid = trans->transid;
5441         sa->rtime.sec = ct.tv_sec;
5442         sa->rtime.nsec = ct.tv_nsec;
5443
5444         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5445                                        BTRFS_UUID_SIZE);
5446         if (received_uuid_changed &&
5447             !btrfs_is_empty_uuid(root_item->received_uuid)) {
5448                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5449                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5450                                           root->root_key.objectid);
5451                 if (ret && ret != -ENOENT) {
5452                         btrfs_abort_transaction(trans, ret);
5453                         btrfs_end_transaction(trans);
5454                         goto out;
5455                 }
5456         }
5457         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5458         btrfs_set_root_stransid(root_item, sa->stransid);
5459         btrfs_set_root_rtransid(root_item, sa->rtransid);
5460         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5461         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5462         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5463         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5464
5465         ret = btrfs_update_root(trans, fs_info->tree_root,
5466                                 &root->root_key, &root->root_item);
5467         if (ret < 0) {
5468                 btrfs_end_transaction(trans);
5469                 goto out;
5470         }
5471         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5472                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
5473                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5474                                           root->root_key.objectid);
5475                 if (ret < 0 && ret != -EEXIST) {
5476                         btrfs_abort_transaction(trans, ret);
5477                         btrfs_end_transaction(trans);
5478                         goto out;
5479                 }
5480         }
5481         ret = btrfs_commit_transaction(trans);
5482 out:
5483         up_write(&fs_info->subvol_sem);
5484         mnt_drop_write_file(file);
5485         return ret;
5486 }
5487
5488 #ifdef CONFIG_64BIT
5489 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5490                                                 void __user *arg)
5491 {
5492         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5493         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5494         int ret = 0;
5495
5496         args32 = memdup_user(arg, sizeof(*args32));
5497         if (IS_ERR(args32))
5498                 return PTR_ERR(args32);
5499
5500         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5501         if (!args64) {
5502                 ret = -ENOMEM;
5503                 goto out;
5504         }
5505
5506         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5507         args64->stransid = args32->stransid;
5508         args64->rtransid = args32->rtransid;
5509         args64->stime.sec = args32->stime.sec;
5510         args64->stime.nsec = args32->stime.nsec;
5511         args64->rtime.sec = args32->rtime.sec;
5512         args64->rtime.nsec = args32->rtime.nsec;
5513         args64->flags = args32->flags;
5514
5515         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5516         if (ret)
5517                 goto out;
5518
5519         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5520         args32->stransid = args64->stransid;
5521         args32->rtransid = args64->rtransid;
5522         args32->stime.sec = args64->stime.sec;
5523         args32->stime.nsec = args64->stime.nsec;
5524         args32->rtime.sec = args64->rtime.sec;
5525         args32->rtime.nsec = args64->rtime.nsec;
5526         args32->flags = args64->flags;
5527
5528         ret = copy_to_user(arg, args32, sizeof(*args32));
5529         if (ret)
5530                 ret = -EFAULT;
5531
5532 out:
5533         kfree(args32);
5534         kfree(args64);
5535         return ret;
5536 }
5537 #endif
5538
5539 static long btrfs_ioctl_set_received_subvol(struct file *file,
5540                                             void __user *arg)
5541 {
5542         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5543         int ret = 0;
5544
5545         sa = memdup_user(arg, sizeof(*sa));
5546         if (IS_ERR(sa))
5547                 return PTR_ERR(sa);
5548
5549         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5550
5551         if (ret)
5552                 goto out;
5553
5554         ret = copy_to_user(arg, sa, sizeof(*sa));
5555         if (ret)
5556                 ret = -EFAULT;
5557
5558 out:
5559         kfree(sa);
5560         return ret;
5561 }
5562
5563 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5564 {
5565         struct inode *inode = file_inode(file);
5566         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5567         size_t len;
5568         int ret;
5569         char label[BTRFS_LABEL_SIZE];
5570
5571         spin_lock(&fs_info->super_lock);
5572         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5573         spin_unlock(&fs_info->super_lock);
5574
5575         len = strnlen(label, BTRFS_LABEL_SIZE);
5576
5577         if (len == BTRFS_LABEL_SIZE) {
5578                 btrfs_warn(fs_info,
5579                            "label is too long, return the first %zu bytes",
5580                            --len);
5581         }
5582
5583         ret = copy_to_user(arg, label, len);
5584
5585         return ret ? -EFAULT : 0;
5586 }
5587
5588 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5589 {
5590         struct inode *inode = file_inode(file);
5591         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5592         struct btrfs_root *root = BTRFS_I(inode)->root;
5593         struct btrfs_super_block *super_block = fs_info->super_copy;
5594         struct btrfs_trans_handle *trans;
5595         char label[BTRFS_LABEL_SIZE];
5596         int ret;
5597
5598         if (!capable(CAP_SYS_ADMIN))
5599                 return -EPERM;
5600
5601         if (copy_from_user(label, arg, sizeof(label)))
5602                 return -EFAULT;
5603
5604         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5605                 btrfs_err(fs_info,
5606                           "unable to set label with more than %d bytes",
5607                           BTRFS_LABEL_SIZE - 1);
5608                 return -EINVAL;
5609         }
5610
5611         ret = mnt_want_write_file(file);
5612         if (ret)
5613                 return ret;
5614
5615         trans = btrfs_start_transaction(root, 0);
5616         if (IS_ERR(trans)) {
5617                 ret = PTR_ERR(trans);
5618                 goto out_unlock;
5619         }
5620
5621         spin_lock(&fs_info->super_lock);
5622         strcpy(super_block->label, label);
5623         spin_unlock(&fs_info->super_lock);
5624         ret = btrfs_commit_transaction(trans);
5625
5626 out_unlock:
5627         mnt_drop_write_file(file);
5628         return ret;
5629 }
5630
5631 #define INIT_FEATURE_FLAGS(suffix) \
5632         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5633           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5634           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5635
5636 int btrfs_ioctl_get_supported_features(void __user *arg)
5637 {
5638         static const struct btrfs_ioctl_feature_flags features[3] = {
5639                 INIT_FEATURE_FLAGS(SUPP),
5640                 INIT_FEATURE_FLAGS(SAFE_SET),
5641                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5642         };
5643
5644         if (copy_to_user(arg, &features, sizeof(features)))
5645                 return -EFAULT;
5646
5647         return 0;
5648 }
5649
5650 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5651 {
5652         struct inode *inode = file_inode(file);
5653         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5654         struct btrfs_super_block *super_block = fs_info->super_copy;
5655         struct btrfs_ioctl_feature_flags features;
5656
5657         features.compat_flags = btrfs_super_compat_flags(super_block);
5658         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5659         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5660
5661         if (copy_to_user(arg, &features, sizeof(features)))
5662                 return -EFAULT;
5663
5664         return 0;
5665 }
5666
5667 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5668                               enum btrfs_feature_set set,
5669                               u64 change_mask, u64 flags, u64 supported_flags,
5670                               u64 safe_set, u64 safe_clear)
5671 {
5672         const char *type = btrfs_feature_set_names[set];
5673         char *names;
5674         u64 disallowed, unsupported;
5675         u64 set_mask = flags & change_mask;
5676         u64 clear_mask = ~flags & change_mask;
5677
5678         unsupported = set_mask & ~supported_flags;
5679         if (unsupported) {
5680                 names = btrfs_printable_features(set, unsupported);
5681                 if (names) {
5682                         btrfs_warn(fs_info,
5683                                    "this kernel does not support the %s feature bit%s",
5684                                    names, strchr(names, ',') ? "s" : "");
5685                         kfree(names);
5686                 } else
5687                         btrfs_warn(fs_info,
5688                                    "this kernel does not support %s bits 0x%llx",
5689                                    type, unsupported);
5690                 return -EOPNOTSUPP;
5691         }
5692
5693         disallowed = set_mask & ~safe_set;
5694         if (disallowed) {
5695                 names = btrfs_printable_features(set, disallowed);
5696                 if (names) {
5697                         btrfs_warn(fs_info,
5698                                    "can't set the %s feature bit%s while mounted",
5699                                    names, strchr(names, ',') ? "s" : "");
5700                         kfree(names);
5701                 } else
5702                         btrfs_warn(fs_info,
5703                                    "can't set %s bits 0x%llx while mounted",
5704                                    type, disallowed);
5705                 return -EPERM;
5706         }
5707
5708         disallowed = clear_mask & ~safe_clear;
5709         if (disallowed) {
5710                 names = btrfs_printable_features(set, disallowed);
5711                 if (names) {
5712                         btrfs_warn(fs_info,
5713                                    "can't clear the %s feature bit%s while mounted",
5714                                    names, strchr(names, ',') ? "s" : "");
5715                         kfree(names);
5716                 } else
5717                         btrfs_warn(fs_info,
5718                                    "can't clear %s bits 0x%llx while mounted",
5719                                    type, disallowed);
5720                 return -EPERM;
5721         }
5722
5723         return 0;
5724 }
5725
5726 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5727 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5728                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5729                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5730                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5731
5732 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5733 {
5734         struct inode *inode = file_inode(file);
5735         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5736         struct btrfs_root *root = BTRFS_I(inode)->root;
5737         struct btrfs_super_block *super_block = fs_info->super_copy;
5738         struct btrfs_ioctl_feature_flags flags[2];
5739         struct btrfs_trans_handle *trans;
5740         u64 newflags;
5741         int ret;
5742
5743         if (!capable(CAP_SYS_ADMIN))
5744                 return -EPERM;
5745
5746         if (copy_from_user(flags, arg, sizeof(flags)))
5747                 return -EFAULT;
5748
5749         /* Nothing to do */
5750         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5751             !flags[0].incompat_flags)
5752                 return 0;
5753
5754         ret = check_feature(fs_info, flags[0].compat_flags,
5755                             flags[1].compat_flags, COMPAT);
5756         if (ret)
5757                 return ret;
5758
5759         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5760                             flags[1].compat_ro_flags, COMPAT_RO);
5761         if (ret)
5762                 return ret;
5763
5764         ret = check_feature(fs_info, flags[0].incompat_flags,
5765                             flags[1].incompat_flags, INCOMPAT);
5766         if (ret)
5767                 return ret;
5768
5769         ret = mnt_want_write_file(file);
5770         if (ret)
5771                 return ret;
5772
5773         trans = btrfs_start_transaction(root, 0);
5774         if (IS_ERR(trans)) {
5775                 ret = PTR_ERR(trans);
5776                 goto out_drop_write;
5777         }
5778
5779         spin_lock(&fs_info->super_lock);
5780         newflags = btrfs_super_compat_flags(super_block);
5781         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5782         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5783         btrfs_set_super_compat_flags(super_block, newflags);
5784
5785         newflags = btrfs_super_compat_ro_flags(super_block);
5786         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5787         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5788         btrfs_set_super_compat_ro_flags(super_block, newflags);
5789
5790         newflags = btrfs_super_incompat_flags(super_block);
5791         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5792         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5793         btrfs_set_super_incompat_flags(super_block, newflags);
5794         spin_unlock(&fs_info->super_lock);
5795
5796         ret = btrfs_commit_transaction(trans);
5797 out_drop_write:
5798         mnt_drop_write_file(file);
5799
5800         return ret;
5801 }
5802
5803 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5804 {
5805         struct btrfs_ioctl_send_args *arg;
5806         int ret;
5807
5808         if (compat) {
5809 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5810                 struct btrfs_ioctl_send_args_32 args32;
5811
5812                 ret = copy_from_user(&args32, argp, sizeof(args32));
5813                 if (ret)
5814                         return -EFAULT;
5815                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5816                 if (!arg)
5817                         return -ENOMEM;
5818                 arg->send_fd = args32.send_fd;
5819                 arg->clone_sources_count = args32.clone_sources_count;
5820                 arg->clone_sources = compat_ptr(args32.clone_sources);
5821                 arg->parent_root = args32.parent_root;
5822                 arg->flags = args32.flags;
5823                 memcpy(arg->reserved, args32.reserved,
5824                        sizeof(args32.reserved));
5825 #else
5826                 return -ENOTTY;
5827 #endif
5828         } else {
5829                 arg = memdup_user(argp, sizeof(*arg));
5830                 if (IS_ERR(arg))
5831                         return PTR_ERR(arg);
5832         }
5833         ret = btrfs_ioctl_send(file, arg);
5834         kfree(arg);
5835         return ret;
5836 }
5837
5838 long btrfs_ioctl(struct file *file, unsigned int
5839                 cmd, unsigned long arg)
5840 {
5841         struct inode *inode = file_inode(file);
5842         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5843         struct btrfs_root *root = BTRFS_I(inode)->root;
5844         void __user *argp = (void __user *)arg;
5845
5846         switch (cmd) {
5847         case FS_IOC_GETFLAGS:
5848                 return btrfs_ioctl_getflags(file, argp);
5849         case FS_IOC_SETFLAGS:
5850                 return btrfs_ioctl_setflags(file, argp);
5851         case FS_IOC_GETVERSION:
5852                 return btrfs_ioctl_getversion(file, argp);
5853         case FITRIM:
5854                 return btrfs_ioctl_fitrim(file, argp);
5855         case BTRFS_IOC_SNAP_CREATE:
5856                 return btrfs_ioctl_snap_create(file, argp, 0);
5857         case BTRFS_IOC_SNAP_CREATE_V2:
5858                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5859         case BTRFS_IOC_SUBVOL_CREATE:
5860                 return btrfs_ioctl_snap_create(file, argp, 1);
5861         case BTRFS_IOC_SUBVOL_CREATE_V2:
5862                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5863         case BTRFS_IOC_SNAP_DESTROY:
5864                 return btrfs_ioctl_snap_destroy(file, argp);
5865         case BTRFS_IOC_SUBVOL_GETFLAGS:
5866                 return btrfs_ioctl_subvol_getflags(file, argp);
5867         case BTRFS_IOC_SUBVOL_SETFLAGS:
5868                 return btrfs_ioctl_subvol_setflags(file, argp);
5869         case BTRFS_IOC_DEFAULT_SUBVOL:
5870                 return btrfs_ioctl_default_subvol(file, argp);
5871         case BTRFS_IOC_DEFRAG:
5872                 return btrfs_ioctl_defrag(file, NULL);
5873         case BTRFS_IOC_DEFRAG_RANGE:
5874                 return btrfs_ioctl_defrag(file, argp);
5875         case BTRFS_IOC_RESIZE:
5876                 return btrfs_ioctl_resize(file, argp);
5877         case BTRFS_IOC_ADD_DEV:
5878                 return btrfs_ioctl_add_dev(fs_info, argp);
5879         case BTRFS_IOC_RM_DEV:
5880                 return btrfs_ioctl_rm_dev(file, argp);
5881         case BTRFS_IOC_RM_DEV_V2:
5882                 return btrfs_ioctl_rm_dev_v2(file, argp);
5883         case BTRFS_IOC_FS_INFO:
5884                 return btrfs_ioctl_fs_info(fs_info, argp);
5885         case BTRFS_IOC_DEV_INFO:
5886                 return btrfs_ioctl_dev_info(fs_info, argp);
5887         case BTRFS_IOC_BALANCE:
5888                 return btrfs_ioctl_balance(file, NULL);
5889         case BTRFS_IOC_TREE_SEARCH:
5890                 return btrfs_ioctl_tree_search(file, argp);
5891         case BTRFS_IOC_TREE_SEARCH_V2:
5892                 return btrfs_ioctl_tree_search_v2(file, argp);
5893         case BTRFS_IOC_INO_LOOKUP:
5894                 return btrfs_ioctl_ino_lookup(file, argp);
5895         case BTRFS_IOC_INO_PATHS:
5896                 return btrfs_ioctl_ino_to_path(root, argp);
5897         case BTRFS_IOC_LOGICAL_INO:
5898                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5899         case BTRFS_IOC_LOGICAL_INO_V2:
5900                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5901         case BTRFS_IOC_SPACE_INFO:
5902                 return btrfs_ioctl_space_info(fs_info, argp);
5903         case BTRFS_IOC_SYNC: {
5904                 int ret;
5905
5906                 ret = btrfs_start_delalloc_roots(fs_info, -1);
5907                 if (ret)
5908                         return ret;
5909                 ret = btrfs_sync_fs(inode->i_sb, 1);
5910                 /*
5911                  * The transaction thread may want to do more work,
5912                  * namely it pokes the cleaner kthread that will start
5913                  * processing uncleaned subvols.
5914                  */
5915                 wake_up_process(fs_info->transaction_kthread);
5916                 return ret;
5917         }
5918         case BTRFS_IOC_START_SYNC:
5919                 return btrfs_ioctl_start_sync(root, argp);
5920         case BTRFS_IOC_WAIT_SYNC:
5921                 return btrfs_ioctl_wait_sync(fs_info, argp);
5922         case BTRFS_IOC_SCRUB:
5923                 return btrfs_ioctl_scrub(file, argp);
5924         case BTRFS_IOC_SCRUB_CANCEL:
5925                 return btrfs_ioctl_scrub_cancel(fs_info);
5926         case BTRFS_IOC_SCRUB_PROGRESS:
5927                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5928         case BTRFS_IOC_BALANCE_V2:
5929                 return btrfs_ioctl_balance(file, argp);
5930         case BTRFS_IOC_BALANCE_CTL:
5931                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5932         case BTRFS_IOC_BALANCE_PROGRESS:
5933                 return btrfs_ioctl_balance_progress(fs_info, argp);
5934         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5935                 return btrfs_ioctl_set_received_subvol(file, argp);
5936 #ifdef CONFIG_64BIT
5937         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5938                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5939 #endif
5940         case BTRFS_IOC_SEND:
5941                 return _btrfs_ioctl_send(file, argp, false);
5942 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5943         case BTRFS_IOC_SEND_32:
5944                 return _btrfs_ioctl_send(file, argp, true);
5945 #endif
5946         case BTRFS_IOC_GET_DEV_STATS:
5947                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5948         case BTRFS_IOC_QUOTA_CTL:
5949                 return btrfs_ioctl_quota_ctl(file, argp);
5950         case BTRFS_IOC_QGROUP_ASSIGN:
5951                 return btrfs_ioctl_qgroup_assign(file, argp);
5952         case BTRFS_IOC_QGROUP_CREATE:
5953                 return btrfs_ioctl_qgroup_create(file, argp);
5954         case BTRFS_IOC_QGROUP_LIMIT:
5955                 return btrfs_ioctl_qgroup_limit(file, argp);
5956         case BTRFS_IOC_QUOTA_RESCAN:
5957                 return btrfs_ioctl_quota_rescan(file, argp);
5958         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5959                 return btrfs_ioctl_quota_rescan_status(file, argp);
5960         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5961                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5962         case BTRFS_IOC_DEV_REPLACE:
5963                 return btrfs_ioctl_dev_replace(fs_info, argp);
5964         case BTRFS_IOC_GET_FSLABEL:
5965                 return btrfs_ioctl_get_fslabel(file, argp);
5966         case BTRFS_IOC_SET_FSLABEL:
5967                 return btrfs_ioctl_set_fslabel(file, argp);
5968         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5969                 return btrfs_ioctl_get_supported_features(argp);
5970         case BTRFS_IOC_GET_FEATURES:
5971                 return btrfs_ioctl_get_features(file, argp);
5972         case BTRFS_IOC_SET_FEATURES:
5973                 return btrfs_ioctl_set_features(file, argp);
5974         case FS_IOC_FSGETXATTR:
5975                 return btrfs_ioctl_fsgetxattr(file, argp);
5976         case FS_IOC_FSSETXATTR:
5977                 return btrfs_ioctl_fssetxattr(file, argp);
5978         case BTRFS_IOC_GET_SUBVOL_INFO:
5979                 return btrfs_ioctl_get_subvol_info(file, argp);
5980         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5981                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5982         case BTRFS_IOC_INO_LOOKUP_USER:
5983                 return btrfs_ioctl_ino_lookup_user(file, argp);
5984         }
5985
5986         return -ENOTTY;
5987 }
5988
5989 #ifdef CONFIG_COMPAT
5990 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5991 {
5992         /*
5993          * These all access 32-bit values anyway so no further
5994          * handling is necessary.
5995          */
5996         switch (cmd) {
5997         case FS_IOC32_GETFLAGS:
5998                 cmd = FS_IOC_GETFLAGS;
5999                 break;
6000         case FS_IOC32_SETFLAGS:
6001                 cmd = FS_IOC_SETFLAGS;
6002                 break;
6003         case FS_IOC32_GETVERSION:
6004                 cmd = FS_IOC_GETVERSION;
6005                 break;
6006         }
6007
6008         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
6009 }
6010 #endif