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         int ret;
 495
 496         if (!capable(CAP_SYS_ADMIN))
 497                 return -EPERM;
 498
 499         /*
 500          * If the fs is mounted with nologreplay, which requires it to be
 501          * mounted in RO mode as well, we can not allow discard on free space
 502          * inside block groups, because log trees refer to extents that are not
 503          * pinned in a block group's free space cache (pinning the extents is
 504          * precisely the first phase of replaying a log tree).
 505          */
 506         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 507                 return -EROFS;
 508
 509         rcu_read_lock();
 510         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 511                                 dev_list) {
 512                 if (!device->bdev)
 513                         continue;
 514                 q = bdev_get_queue(device->bdev);
 515                 if (blk_queue_discard(q)) {
 516                         num_devices++;
 517                         minlen = min_t(u64, q->limits.discard_granularity,
 518                                      minlen);
 519                 }
 520         }
 521         rcu_read_unlock();
 522
 523         if (!num_devices)
 524                 return -EOPNOTSUPP;
 525         if (copy_from_user(&range, arg, sizeof(range)))
 526                 return -EFAULT;
 527
 528         /*
 529          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 530          * block group is in the logical address space, which can be any
 531          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 532          */
 533         if (range.len < fs_info->sb->s_blocksize)
 534                 return -EINVAL;
 535
 536         range.minlen = max(range.minlen, minlen);
 537         ret = btrfs_trim_fs(fs_info, &range);
 538         if (ret < 0)
 539                 return ret;
 540
 541         if (copy_to_user(arg, &range, sizeof(range)))
 542                 return -EFAULT;
 543
 544         return 0;
 545 }
 546
 547 int btrfs_is_empty_uuid(u8 *uuid)
 548 {
 549         int i;
 550
 551         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 552                 if (uuid[i])
 553                         return 0;
 554         }
 555         return 1;
 556 }
 557
 558 static noinline int create_subvol(struct inode *dir,
 559                                   struct dentry *dentry,
 560                                   const char *name, int namelen,
 561                                   u64 *async_transid,
 562                                   struct btrfs_qgroup_inherit *inherit)
 563 {
 564         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 565         struct btrfs_trans_handle *trans;
 566         struct btrfs_key key;
 567         struct btrfs_root_item *root_item;
 568         struct btrfs_inode_item *inode_item;
 569         struct extent_buffer *leaf;
 570         struct btrfs_root *root = BTRFS_I(dir)->root;
 571         struct btrfs_root *new_root;
 572         struct btrfs_block_rsv block_rsv;
 573         struct timespec64 cur_time = current_time(dir);
 574         struct inode *inode;
 575         int ret;
 576         int err;
 577         u64 objectid;
 578         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 579         u64 index = 0;
 580         uuid_le new_uuid;
 581
 582         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 583         if (!root_item)
 584                 return -ENOMEM;
 585
 586         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 587         if (ret)
 588                 goto fail_free;
 589
 590         /*
 591          * Don't create subvolume whose level is not zero. Or qgroup will be
 592          * screwed up since it assumes subvolume qgroup's level to be 0.
 593          */
 594         if (btrfs_qgroup_level(objectid)) {
 595                 ret = -ENOSPC;
 596                 goto fail_free;
 597         }
 598
 599         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 600         /*
 601          * The same as the snapshot creation, please see the comment
 602          * of create_snapshot().
 603          */
 604         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 605         if (ret)
 606                 goto fail_free;
 607
 608         trans = btrfs_start_transaction(root, 0);
 609         if (IS_ERR(trans)) {
 610                 ret = PTR_ERR(trans);
 611                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 612                 goto fail_free;
 613         }
 614         trans->block_rsv = &block_rsv;
 615         trans->bytes_reserved = block_rsv.size;
 616
 617         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 618         if (ret)
 619                 goto fail;
 620
 621         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 622         if (IS_ERR(leaf)) {
 623                 ret = PTR_ERR(leaf);
 624                 goto fail;
 625         }
 626
 627         btrfs_mark_buffer_dirty(leaf);
 628
 629         inode_item = &root_item->inode;
 630         btrfs_set_stack_inode_generation(inode_item, 1);
 631         btrfs_set_stack_inode_size(inode_item, 3);
 632         btrfs_set_stack_inode_nlink(inode_item, 1);
 633         btrfs_set_stack_inode_nbytes(inode_item,
 634                                      fs_info->nodesize);
 635         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 636
 637         btrfs_set_root_flags(root_item, 0);
 638         btrfs_set_root_limit(root_item, 0);
 639         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 640
 641         btrfs_set_root_bytenr(root_item, leaf->start);
 642         btrfs_set_root_generation(root_item, trans->transid);
 643         btrfs_set_root_level(root_item, 0);
 644         btrfs_set_root_refs(root_item, 1);
 645         btrfs_set_root_used(root_item, leaf->len);
 646         btrfs_set_root_last_snapshot(root_item, 0);
 647
 648         btrfs_set_root_generation_v2(root_item,
 649                         btrfs_root_generation(root_item));
 650         uuid_le_gen(&new_uuid);
 651         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 652         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 653         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 654         root_item->ctime = root_item->otime;
 655         btrfs_set_root_ctransid(root_item, trans->transid);
 656         btrfs_set_root_otransid(root_item, trans->transid);
 657
 658         btrfs_tree_unlock(leaf);
 659
 660         btrfs_set_root_dirid(root_item, new_dirid);
 661
 662         key.objectid = objectid;
 663         key.offset = 0;
 664         key.type = BTRFS_ROOT_ITEM_KEY;
 665         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 666                                 root_item);
 667         if (ret) {
 668                 /*
 669                  * Since we don't abort the transaction in this case, free the
 670                  * tree block so that we don't leak space and leave the
 671                  * filesystem in an inconsistent state (an extent item in the
 672                  * extent tree without backreferences). Also no need to have
 673                  * the tree block locked since it is not in any tree at this
 674                  * point, so no other task can find it and use it.
 675                  */
 676                 btrfs_free_tree_block(trans, root, leaf, 0, 1);
 677                 free_extent_buffer(leaf);
 678                 goto fail;
 679         }
 680
 681         free_extent_buffer(leaf);
 682         leaf = NULL;
 683
 684         key.offset = (u64)-1;
 685         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 686         if (IS_ERR(new_root)) {
 687                 ret = PTR_ERR(new_root);
 688                 btrfs_abort_transaction(trans, ret);
 689                 goto fail;
 690         }
 691
 692         btrfs_record_root_in_trans(trans, new_root);
 693
 694         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 695         if (ret) {
 696                 /* We potentially lose an unused inode item here */
 697                 btrfs_abort_transaction(trans, ret);
 698                 goto fail;
 699         }
 700
 701         mutex_lock(&new_root->objectid_mutex);
 702         new_root->highest_objectid = new_dirid;
 703         mutex_unlock(&new_root->objectid_mutex);
 704
 705         /*
 706          * insert the directory item
 707          */
 708         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 709         if (ret) {
 710                 btrfs_abort_transaction(trans, ret);
 711                 goto fail;
 712         }
 713
 714         ret = btrfs_insert_dir_item(trans, root,
 715                                     name, namelen, BTRFS_I(dir), &key,
 716                                     BTRFS_FT_DIR, index);
 717         if (ret) {
 718                 btrfs_abort_transaction(trans, ret);
 719                 goto fail;
 720         }
 721
 722         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 723         ret = btrfs_update_inode(trans, root, dir);
 724         if (ret) {
 725                 btrfs_abort_transaction(trans, ret);
 726                 goto fail;
 727         }
 728
 729         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 730                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 731         if (ret) {
 732                 btrfs_abort_transaction(trans, ret);
 733                 goto fail;
 734         }
 735
 736         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 737                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 738         if (ret)
 739                 btrfs_abort_transaction(trans, ret);
 740
 741 fail:
 742         kfree(root_item);
 743         trans->block_rsv = NULL;
 744         trans->bytes_reserved = 0;
 745         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 746
 747         if (async_transid) {
 748                 *async_transid = trans->transid;
 749                 err = btrfs_commit_transaction_async(trans, 1);
 750                 if (err)
 751                         err = btrfs_commit_transaction(trans);
 752         } else {
 753                 err = btrfs_commit_transaction(trans);
 754         }
 755         if (err && !ret)
 756                 ret = err;
 757
 758         if (!ret) {
 759                 inode = btrfs_lookup_dentry(dir, dentry);
 760                 if (IS_ERR(inode))
 761                         return PTR_ERR(inode);
 762                 d_instantiate(dentry, inode);
 763         }
 764         return ret;
 765
 766 fail_free:
 767         kfree(root_item);
 768         return ret;
 769 }
 770
 771 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 772                            struct dentry *dentry,
 773                            u64 *async_transid, bool readonly,
 774                            struct btrfs_qgroup_inherit *inherit)
 775 {
 776         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 777         struct inode *inode;
 778         struct btrfs_pending_snapshot *pending_snapshot;
 779         struct btrfs_trans_handle *trans;
 780         int ret;
 781         bool snapshot_force_cow = false;
 782
 783         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 784                 return -EINVAL;
 785
 786         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 787         if (!pending_snapshot)
 788                 return -ENOMEM;
 789
 790         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 791                         GFP_KERNEL);
 792         pending_snapshot->path = btrfs_alloc_path();
 793         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 794                 ret = -ENOMEM;
 795                 goto free_pending;
 796         }
 797
 798         /*
 799          * Force new buffered writes to reserve space even when NOCOW is
 800          * possible. This is to avoid later writeback (running dealloc) to
 801          * fallback to COW mode and unexpectedly fail with ENOSPC.
 802          */
 803         atomic_inc(&root->will_be_snapshotted);
 804         smp_mb__after_atomic();
 805         /* wait for no snapshot writes */
 806         wait_event(root->subv_writers->wait,
 807                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 808
 809         ret = btrfs_start_delalloc_snapshot(root);
 810         if (ret)
 811                 goto dec_and_free;
 812
 813         /*
 814          * All previous writes have started writeback in NOCOW mode, so now
 815          * we force future writes to fallback to COW mode during snapshot
 816          * creation.
 817          */
 818         atomic_inc(&root->snapshot_force_cow);
 819         snapshot_force_cow = true;
 820
 821         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 822
 823         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 824                              BTRFS_BLOCK_RSV_TEMP);
 825         /*
 826          * 1 - parent dir inode
 827          * 2 - dir entries
 828          * 1 - root item
 829          * 2 - root ref/backref
 830          * 1 - root of snapshot
 831          * 1 - UUID item
 832          */
 833         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 834                                         &pending_snapshot->block_rsv, 8,
 835                                         false);
 836         if (ret)
 837                 goto dec_and_free;
 838
 839         pending_snapshot->dentry = dentry;
 840         pending_snapshot->root = root;
 841         pending_snapshot->readonly = readonly;
 842         pending_snapshot->dir = dir;
 843         pending_snapshot->inherit = inherit;
 844
 845         trans = btrfs_start_transaction(root, 0);
 846         if (IS_ERR(trans)) {
 847                 ret = PTR_ERR(trans);
 848                 goto fail;
 849         }
 850
 851         spin_lock(&fs_info->trans_lock);
 852         list_add(&pending_snapshot->list,
 853                  &trans->transaction->pending_snapshots);
 854         spin_unlock(&fs_info->trans_lock);
 855         if (async_transid) {
 856                 *async_transid = trans->transid;
 857                 ret = btrfs_commit_transaction_async(trans, 1);
 858                 if (ret)
 859                         ret = btrfs_commit_transaction(trans);
 860         } else {
 861                 ret = btrfs_commit_transaction(trans);
 862         }
 863         if (ret)
 864                 goto fail;
 865
 866         ret = pending_snapshot->error;
 867         if (ret)
 868                 goto fail;
 869
 870         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 871         if (ret)
 872                 goto fail;
 873
 874         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 875         if (IS_ERR(inode)) {
 876                 ret = PTR_ERR(inode);
 877                 goto fail;
 878         }
 879
 880         d_instantiate(dentry, inode);
 881         ret = 0;
 882 fail:
 883         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 884 dec_and_free:
 885         if (snapshot_force_cow)
 886                 atomic_dec(&root->snapshot_force_cow);
 887         if (atomic_dec_and_test(&root->will_be_snapshotted))
 888                 wake_up_var(&root->will_be_snapshotted);
 889 free_pending:
 890         kfree(pending_snapshot->root_item);
 891         btrfs_free_path(pending_snapshot->path);
 892         kfree(pending_snapshot);
 893
 894         return ret;
 895 }
 896
 897 /*  copy of may_delete in fs/namei.c()
 898  *      Check whether we can remove a link victim from directory dir, check
 899  *  whether the type of victim is right.
 900  *  1. We can't do it if dir is read-only (done in permission())
 901  *  2. We should have write and exec permissions on dir
 902  *  3. We can't remove anything from append-only dir
 903  *  4. We can't do anything with immutable dir (done in permission())
 904  *  5. If the sticky bit on dir is set we should either
 905  *      a. be owner of dir, or
 906  *      b. be owner of victim, or
 907  *      c. have CAP_FOWNER capability
 908  *  6. If the victim is append-only or immutable we can't do anything with
 909  *     links pointing to it.
 910  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 911  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 912  *  9. We can't remove a root or mountpoint.
 913  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 914  *     nfs_async_unlink().
 915  */
 916
 917 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 918 {
 919         int error;
 920
 921         if (d_really_is_negative(victim))
 922                 return -ENOENT;
 923
 924         BUG_ON(d_inode(victim->d_parent) != dir);
 925         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 926
 927         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 928         if (error)
 929                 return error;
 930         if (IS_APPEND(dir))
 931                 return -EPERM;
 932         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 933             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 934                 return -EPERM;
 935         if (isdir) {
 936                 if (!d_is_dir(victim))
 937                         return -ENOTDIR;
 938                 if (IS_ROOT(victim))
 939                         return -EBUSY;
 940         } else if (d_is_dir(victim))
 941                 return -EISDIR;
 942         if (IS_DEADDIR(dir))
 943                 return -ENOENT;
 944         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 945                 return -EBUSY;
 946         return 0;
 947 }
 948
 949 /* copy of may_create in fs/namei.c() */
 950 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 951 {
 952         if (d_really_is_positive(child))
 953                 return -EEXIST;
 954         if (IS_DEADDIR(dir))
 955                 return -ENOENT;
 956         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 957 }
 958
 959 /*
 960  * Create a new subvolume below @parent.  This is largely modeled after
 961  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 962  * inside this filesystem so it's quite a bit simpler.
 963  */
 964 static noinline int btrfs_mksubvol(const struct path *parent,
 965                                    const char *name, int namelen,
 966                                    struct btrfs_root *snap_src,
 967                                    u64 *async_transid, bool readonly,
 968                                    struct btrfs_qgroup_inherit *inherit)
 969 {
 970         struct inode *dir = d_inode(parent->dentry);
 971         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 972         struct dentry *dentry;
 973         int error;
 974
 975         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 976         if (error == -EINTR)
 977                 return error;
 978
 979         dentry = lookup_one_len(name, parent->dentry, namelen);
 980         error = PTR_ERR(dentry);
 981         if (IS_ERR(dentry))
 982                 goto out_unlock;
 983
 984         error = btrfs_may_create(dir, dentry);
 985         if (error)
 986                 goto out_dput;
 987
 988         /*
 989          * even if this name doesn't exist, we may get hash collisions.
 990          * check for them now when we can safely fail
 991          */
 992         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 993                                                dir->i_ino, name,
 994                                                namelen);
 995         if (error)
 996                 goto out_dput;
 997
 998         down_read(&fs_info->subvol_sem);
 999
1000         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
1001                 goto out_up_read;
1002
1003         if (snap_src) {
1004                 error = create_snapshot(snap_src, dir, dentry,
1005                                         async_transid, readonly, inherit);
1006         } else {
1007                 error = create_subvol(dir, dentry, name, namelen,
1008                                       async_transid, inherit);
1009         }
1010         if (!error)
1011                 fsnotify_mkdir(dir, dentry);
1012 out_up_read:
1013         up_read(&fs_info->subvol_sem);
1014 out_dput:
1015         dput(dentry);
1016 out_unlock:
1017         inode_unlock(dir);
1018         return error;
1019 }
1020
1021 /*
1022  * When we're defragging a range, we don't want to kick it off again
1023  * if it is really just waiting for delalloc to send it down.
1024  * If we find a nice big extent or delalloc range for the bytes in the
1025  * file you want to defrag, we return 0 to let you know to skip this
1026  * part of the file
1027  */
1028 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1029 {
1030         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1031         struct extent_map *em = NULL;
1032         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1033         u64 end;
1034
1035         read_lock(&em_tree->lock);
1036         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1037         read_unlock(&em_tree->lock);
1038
1039         if (em) {
1040                 end = extent_map_end(em);
1041                 free_extent_map(em);
1042                 if (end - offset > thresh)
1043                         return 0;
1044         }
1045         /* if we already have a nice delalloc here, just stop */
1046         thresh /= 2;
1047         end = count_range_bits(io_tree, &offset, offset + thresh,
1048                                thresh, EXTENT_DELALLOC, 1);
1049         if (end >= thresh)
1050                 return 0;
1051         return 1;
1052 }
1053
1054 /*
1055  * helper function to walk through a file and find extents
1056  * newer than a specific transid, and smaller than thresh.
1057  *
1058  * This is used by the defragging code to find new and small
1059  * extents
1060  */
1061 static int find_new_extents(struct btrfs_root *root,
1062                             struct inode *inode, u64 newer_than,
1063                             u64 *off, u32 thresh)
1064 {
1065         struct btrfs_path *path;
1066         struct btrfs_key min_key;
1067         struct extent_buffer *leaf;
1068         struct btrfs_file_extent_item *extent;
1069         int type;
1070         int ret;
1071         u64 ino = btrfs_ino(BTRFS_I(inode));
1072
1073         path = btrfs_alloc_path();
1074         if (!path)
1075                 return -ENOMEM;
1076
1077         min_key.objectid = ino;
1078         min_key.type = BTRFS_EXTENT_DATA_KEY;
1079         min_key.offset = *off;
1080
1081         while (1) {
1082                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1083                 if (ret != 0)
1084                         goto none;
1085 process_slot:
1086                 if (min_key.objectid != ino)
1087                         goto none;
1088                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1089                         goto none;
1090
1091                 leaf = path->nodes[0];
1092                 extent = btrfs_item_ptr(leaf, path->slots[0],
1093                                         struct btrfs_file_extent_item);
1094
1095                 type = btrfs_file_extent_type(leaf, extent);
1096                 if (type == BTRFS_FILE_EXTENT_REG &&
1097                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1098                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1099                         *off = min_key.offset;
1100                         btrfs_free_path(path);
1101                         return 0;
1102                 }
1103
1104                 path->slots[0]++;
1105                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1106                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1107                         goto process_slot;
1108                 }
1109
1110                 if (min_key.offset == (u64)-1)
1111                         goto none;
1112
1113                 min_key.offset++;
1114                 btrfs_release_path(path);
1115         }
1116 none:
1117         btrfs_free_path(path);
1118         return -ENOENT;
1119 }
1120
1121 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1122 {
1123         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1124         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1125         struct extent_map *em;
1126         u64 len = PAGE_SIZE;
1127
1128         /*
1129          * hopefully we have this extent in the tree already, try without
1130          * the full extent lock
1131          */
1132         read_lock(&em_tree->lock);
1133         em = lookup_extent_mapping(em_tree, start, len);
1134         read_unlock(&em_tree->lock);
1135
1136         if (!em) {
1137                 struct extent_state *cached = NULL;
1138                 u64 end = start + len - 1;
1139
1140                 /* get the big lock and read metadata off disk */
1141                 lock_extent_bits(io_tree, start, end, &cached);
1142                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1143                 unlock_extent_cached(io_tree, start, end, &cached);
1144
1145                 if (IS_ERR(em))
1146                         return NULL;
1147         }
1148
1149         return em;
1150 }
1151
1152 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1153 {
1154         struct extent_map *next;
1155         bool ret = true;
1156
1157         /* this is the last extent */
1158         if (em->start + em->len >= i_size_read(inode))
1159                 return false;
1160
1161         next = defrag_lookup_extent(inode, em->start + em->len);
1162         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1163                 ret = false;
1164         else if ((em->block_start + em->block_len == next->block_start) &&
1165                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1166                 ret = false;
1167
1168         free_extent_map(next);
1169         return ret;
1170 }
1171
1172 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1173                                u64 *last_len, u64 *skip, u64 *defrag_end,
1174                                int compress)
1175 {
1176         struct extent_map *em;
1177         int ret = 1;
1178         bool next_mergeable = true;
1179         bool prev_mergeable = true;
1180
1181         /*
1182          * make sure that once we start defragging an extent, we keep on
1183          * defragging it
1184          */
1185         if (start < *defrag_end)
1186                 return 1;
1187
1188         *skip = 0;
1189
1190         em = defrag_lookup_extent(inode, start);
1191         if (!em)
1192                 return 0;
1193
1194         /* this will cover holes, and inline extents */
1195         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1196                 ret = 0;
1197                 goto out;
1198         }
1199
1200         if (!*defrag_end)
1201                 prev_mergeable = false;
1202
1203         next_mergeable = defrag_check_next_extent(inode, em);
1204         /*
1205          * we hit a real extent, if it is big or the next extent is not a
1206          * real extent, don't bother defragging it
1207          */
1208         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1209             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1210                 ret = 0;
1211 out:
1212         /*
1213          * last_len ends up being a counter of how many bytes we've defragged.
1214          * every time we choose not to defrag an extent, we reset *last_len
1215          * so that the next tiny extent will force a defrag.
1216          *
1217          * The end result of this is that tiny extents before a single big
1218          * extent will force at least part of that big extent to be defragged.
1219          */
1220         if (ret) {
1221                 *defrag_end = extent_map_end(em);
1222         } else {
1223                 *last_len = 0;
1224                 *skip = extent_map_end(em);
1225                 *defrag_end = 0;
1226         }
1227
1228         free_extent_map(em);
1229         return ret;
1230 }
1231
1232 /*
1233  * it doesn't do much good to defrag one or two pages
1234  * at a time.  This pulls in a nice chunk of pages
1235  * to COW and defrag.
1236  *
1237  * It also makes sure the delalloc code has enough
1238  * dirty data to avoid making new small extents as part
1239  * of the defrag
1240  *
1241  * It's a good idea to start RA on this range
1242  * before calling this.
1243  */
1244 static int cluster_pages_for_defrag(struct inode *inode,
1245                                     struct page **pages,
1246                                     unsigned long start_index,
1247                                     unsigned long num_pages)
1248 {
1249         unsigned long file_end;
1250         u64 isize = i_size_read(inode);
1251         u64 page_start;
1252         u64 page_end;
1253         u64 page_cnt;
1254         u64 start = (u64)start_index << PAGE_SHIFT;
1255         int ret;
1256         int i;
1257         int i_done;
1258         struct btrfs_ordered_extent *ordered;
1259         struct extent_state *cached_state = NULL;
1260         struct extent_io_tree *tree;
1261         struct extent_changeset *data_reserved = NULL;
1262         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1263
1264         file_end = (isize - 1) >> PAGE_SHIFT;
1265         if (!isize || start_index > file_end)
1266                 return 0;
1267
1268         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1269
1270         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1271                         start, page_cnt << PAGE_SHIFT);
1272         if (ret)
1273                 return ret;
1274         i_done = 0;
1275         tree = &BTRFS_I(inode)->io_tree;
1276
1277         /* step one, lock all the pages */
1278         for (i = 0; i < page_cnt; i++) {
1279                 struct page *page;
1280 again:
1281                 page = find_or_create_page(inode->i_mapping,
1282                                            start_index + i, mask);
1283                 if (!page)
1284                         break;
1285
1286                 page_start = page_offset(page);
1287                 page_end = page_start + PAGE_SIZE - 1;
1288                 while (1) {
1289                         lock_extent_bits(tree, page_start, page_end,
1290                                          &cached_state);
1291                         ordered = btrfs_lookup_ordered_extent(inode,
1292                                                               page_start);
1293                         unlock_extent_cached(tree, page_start, page_end,
1294                                              &cached_state);
1295                         if (!ordered)
1296                                 break;
1297
1298                         unlock_page(page);
1299                         btrfs_start_ordered_extent(inode, ordered, 1);
1300                         btrfs_put_ordered_extent(ordered);
1301                         lock_page(page);
1302                         /*
1303                          * we unlocked the page above, so we need check if
1304                          * it was released or not.
1305                          */
1306                         if (page->mapping != inode->i_mapping) {
1307                                 unlock_page(page);
1308                                 put_page(page);
1309                                 goto again;
1310                         }
1311                 }
1312
1313                 if (!PageUptodate(page)) {
1314                         btrfs_readpage(NULL, page);
1315                         lock_page(page);
1316                         if (!PageUptodate(page)) {
1317                                 unlock_page(page);
1318                                 put_page(page);
1319                                 ret = -EIO;
1320                                 break;
1321                         }
1322                 }
1323
1324                 if (page->mapping != inode->i_mapping) {
1325                         unlock_page(page);
1326                         put_page(page);
1327                         goto again;
1328                 }
1329
1330                 pages[i] = page;
1331                 i_done++;
1332         }
1333         if (!i_done || ret)
1334                 goto out;
1335
1336         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1337                 goto out;
1338
1339         /*
1340          * so now we have a nice long stream of locked
1341          * and up to date pages, lets wait on them
1342          */
1343         for (i = 0; i < i_done; i++)
1344                 wait_on_page_writeback(pages[i]);
1345
1346         page_start = page_offset(pages[0]);
1347         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1348
1349         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1350                          page_start, page_end - 1, &cached_state);
1351         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1352                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1353                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1354                           &cached_state);
1355
1356         if (i_done != page_cnt) {
1357                 spin_lock(&BTRFS_I(inode)->lock);
1358                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1359                 spin_unlock(&BTRFS_I(inode)->lock);
1360                 btrfs_delalloc_release_space(inode, data_reserved,
1361                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1362         }
1363
1364
1365         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1366                           &cached_state);
1367
1368         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1369                              page_start, page_end - 1, &cached_state);
1370
1371         for (i = 0; i < i_done; i++) {
1372                 clear_page_dirty_for_io(pages[i]);
1373                 ClearPageChecked(pages[i]);
1374                 set_page_extent_mapped(pages[i]);
1375                 set_page_dirty(pages[i]);
1376                 unlock_page(pages[i]);
1377                 put_page(pages[i]);
1378         }
1379         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1380         extent_changeset_free(data_reserved);
1381         return i_done;
1382 out:
1383         for (i = 0; i < i_done; i++) {
1384                 unlock_page(pages[i]);
1385                 put_page(pages[i]);
1386         }
1387         btrfs_delalloc_release_space(inode, data_reserved,
1388                         start, page_cnt << PAGE_SHIFT, true);
1389         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1390         extent_changeset_free(data_reserved);
1391         return ret;
1392
1393 }
1394
1395 int btrfs_defrag_file(struct inode *inode, struct file *file,
1396                       struct btrfs_ioctl_defrag_range_args *range,
1397                       u64 newer_than, unsigned long max_to_defrag)
1398 {
1399         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1400         struct btrfs_root *root = BTRFS_I(inode)->root;
1401         struct file_ra_state *ra = NULL;
1402         unsigned long last_index;
1403         u64 isize = i_size_read(inode);
1404         u64 last_len = 0;
1405         u64 skip = 0;
1406         u64 defrag_end = 0;
1407         u64 newer_off = range->start;
1408         unsigned long i;
1409         unsigned long ra_index = 0;
1410         int ret;
1411         int defrag_count = 0;
1412         int compress_type = BTRFS_COMPRESS_ZLIB;
1413         u32 extent_thresh = range->extent_thresh;
1414         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1415         unsigned long cluster = max_cluster;
1416         u64 new_align = ~((u64)SZ_128K - 1);
1417         struct page **pages = NULL;
1418         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1419
1420         if (isize == 0)
1421                 return 0;
1422
1423         if (range->start >= isize)
1424                 return -EINVAL;
1425
1426         if (do_compress) {
1427                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1428                         return -EINVAL;
1429                 if (range->compress_type)
1430                         compress_type = range->compress_type;
1431         }
1432
1433         if (extent_thresh == 0)
1434                 extent_thresh = SZ_256K;
1435
1436         /*
1437          * If we were not given a file, allocate a readahead context. As
1438          * readahead is just an optimization, defrag will work without it so
1439          * we don't error out.
1440          */
1441         if (!file) {
1442                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1443                 if (ra)
1444                         file_ra_state_init(ra, inode->i_mapping);
1445         } else {
1446                 ra = &file->f_ra;
1447         }
1448
1449         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1450         if (!pages) {
1451                 ret = -ENOMEM;
1452                 goto out_ra;
1453         }
1454
1455         /* find the last page to defrag */
1456         if (range->start + range->len > range->start) {
1457                 last_index = min_t(u64, isize - 1,
1458                          range->start + range->len - 1) >> PAGE_SHIFT;
1459         } else {
1460                 last_index = (isize - 1) >> PAGE_SHIFT;
1461         }
1462
1463         if (newer_than) {
1464                 ret = find_new_extents(root, inode, newer_than,
1465                                        &newer_off, SZ_64K);
1466                 if (!ret) {
1467                         range->start = newer_off;
1468                         /*
1469                          * we always align our defrag to help keep
1470                          * the extents in the file evenly spaced
1471                          */
1472                         i = (newer_off & new_align) >> PAGE_SHIFT;
1473                 } else
1474                         goto out_ra;
1475         } else {
1476                 i = range->start >> PAGE_SHIFT;
1477         }
1478         if (!max_to_defrag)
1479                 max_to_defrag = last_index - i + 1;
1480
1481         /*
1482          * make writeback starts from i, so the defrag range can be
1483          * written sequentially.
1484          */
1485         if (i < inode->i_mapping->writeback_index)
1486                 inode->i_mapping->writeback_index = i;
1487
1488         while (i <= last_index && defrag_count < max_to_defrag &&
1489                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1490                 /*
1491                  * make sure we stop running if someone unmounts
1492                  * the FS
1493                  */
1494                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1495                         break;
1496
1497                 if (btrfs_defrag_cancelled(fs_info)) {
1498                         btrfs_debug(fs_info, "defrag_file cancelled");
1499                         ret = -EAGAIN;
1500                         break;
1501                 }
1502
1503                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1504                                          extent_thresh, &last_len, &skip,
1505                                          &defrag_end, do_compress)){
1506                         unsigned long next;
1507                         /*
1508                          * the should_defrag function tells us how much to skip
1509                          * bump our counter by the suggested amount
1510                          */
1511                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1512                         i = max(i + 1, next);
1513                         continue;
1514                 }
1515
1516                 if (!newer_than) {
1517                         cluster = (PAGE_ALIGN(defrag_end) >>
1518                                    PAGE_SHIFT) - i;
1519                         cluster = min(cluster, max_cluster);
1520                 } else {
1521                         cluster = max_cluster;
1522                 }
1523
1524                 if (i + cluster > ra_index) {
1525                         ra_index = max(i, ra_index);
1526                         if (ra)
1527                                 page_cache_sync_readahead(inode->i_mapping, ra,
1528                                                 file, ra_index, cluster);
1529                         ra_index += cluster;
1530                 }
1531
1532                 inode_lock(inode);
1533                 if (do_compress)
1534                         BTRFS_I(inode)->defrag_compress = compress_type;
1535                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1536                 if (ret < 0) {
1537                         inode_unlock(inode);
1538                         goto out_ra;
1539                 }
1540
1541                 defrag_count += ret;
1542                 balance_dirty_pages_ratelimited(inode->i_mapping);
1543                 inode_unlock(inode);
1544
1545                 if (newer_than) {
1546                         if (newer_off == (u64)-1)
1547                                 break;
1548
1549                         if (ret > 0)
1550                                 i += ret;
1551
1552                         newer_off = max(newer_off + 1,
1553                                         (u64)i << PAGE_SHIFT);
1554
1555                         ret = find_new_extents(root, inode, newer_than,
1556                                                &newer_off, SZ_64K);
1557                         if (!ret) {
1558                                 range->start = newer_off;
1559                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1560                         } else {
1561                                 break;
1562                         }
1563                 } else {
1564                         if (ret > 0) {
1565                                 i += ret;
1566                                 last_len += ret << PAGE_SHIFT;
1567                         } else {
1568                                 i++;
1569                                 last_len = 0;
1570                         }
1571                 }
1572         }
1573
1574         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1575                 filemap_flush(inode->i_mapping);
1576                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1577                              &BTRFS_I(inode)->runtime_flags))
1578                         filemap_flush(inode->i_mapping);
1579         }
1580
1581         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1582                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1583         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1584                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1585         }
1586
1587         ret = defrag_count;
1588
1589 out_ra:
1590         if (do_compress) {
1591                 inode_lock(inode);
1592                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1593                 inode_unlock(inode);
1594         }
1595         if (!file)
1596                 kfree(ra);
1597         kfree(pages);
1598         return ret;
1599 }
1600
1601 static noinline int btrfs_ioctl_resize(struct file *file,
1602                                         void __user *arg)
1603 {
1604         struct inode *inode = file_inode(file);
1605         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1606         u64 new_size;
1607         u64 old_size;
1608         u64 devid = 1;
1609         struct btrfs_root *root = BTRFS_I(inode)->root;
1610         struct btrfs_ioctl_vol_args *vol_args;
1611         struct btrfs_trans_handle *trans;
1612         struct btrfs_device *device = NULL;
1613         char *sizestr;
1614         char *retptr;
1615         char *devstr = NULL;
1616         int ret = 0;
1617         int mod = 0;
1618
1619         if (!capable(CAP_SYS_ADMIN))
1620                 return -EPERM;
1621
1622         ret = mnt_want_write_file(file);
1623         if (ret)
1624                 return ret;
1625
1626         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1627                 mnt_drop_write_file(file);
1628                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1629         }
1630
1631         vol_args = memdup_user(arg, sizeof(*vol_args));
1632         if (IS_ERR(vol_args)) {
1633                 ret = PTR_ERR(vol_args);
1634                 goto out;
1635         }
1636
1637         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1638
1639         sizestr = vol_args->name;
1640         devstr = strchr(sizestr, ':');
1641         if (devstr) {
1642                 sizestr = devstr + 1;
1643                 *devstr = '\0';
1644                 devstr = vol_args->name;
1645                 ret = kstrtoull(devstr, 10, &devid);
1646                 if (ret)
1647                         goto out_free;
1648                 if (!devid) {
1649                         ret = -EINVAL;
1650                         goto out_free;
1651                 }
1652                 btrfs_info(fs_info, "resizing devid %llu", devid);
1653         }
1654
1655         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1656         if (!device) {
1657                 btrfs_info(fs_info, "resizer unable to find device %llu",
1658                            devid);
1659                 ret = -ENODEV;
1660                 goto out_free;
1661         }
1662
1663         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1664                 btrfs_info(fs_info,
1665                            "resizer unable to apply on readonly device %llu",
1666                        devid);
1667                 ret = -EPERM;
1668                 goto out_free;
1669         }
1670
1671         if (!strcmp(sizestr, "max"))
1672                 new_size = device->bdev->bd_inode->i_size;
1673         else {
1674                 if (sizestr[0] == '-') {
1675                         mod = -1;
1676                         sizestr++;
1677                 } else if (sizestr[0] == '+') {
1678                         mod = 1;
1679                         sizestr++;
1680                 }
1681                 new_size = memparse(sizestr, &retptr);
1682                 if (*retptr != '\0' || new_size == 0) {
1683                         ret = -EINVAL;
1684                         goto out_free;
1685                 }
1686         }
1687
1688         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1689                 ret = -EPERM;
1690                 goto out_free;
1691         }
1692
1693         old_size = btrfs_device_get_total_bytes(device);
1694
1695         if (mod < 0) {
1696                 if (new_size > old_size) {
1697                         ret = -EINVAL;
1698                         goto out_free;
1699                 }
1700                 new_size = old_size - new_size;
1701         } else if (mod > 0) {
1702                 if (new_size > ULLONG_MAX - old_size) {
1703                         ret = -ERANGE;
1704                         goto out_free;
1705                 }
1706                 new_size = old_size + new_size;
1707         }
1708
1709         if (new_size < SZ_256M) {
1710                 ret = -EINVAL;
1711                 goto out_free;
1712         }
1713         if (new_size > device->bdev->bd_inode->i_size) {
1714                 ret = -EFBIG;
1715                 goto out_free;
1716         }
1717
1718         new_size = round_down(new_size, fs_info->sectorsize);
1719
1720         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1721                           rcu_str_deref(device->name), new_size);
1722
1723         if (new_size > old_size) {
1724                 trans = btrfs_start_transaction(root, 0);
1725                 if (IS_ERR(trans)) {
1726                         ret = PTR_ERR(trans);
1727                         goto out_free;
1728                 }
1729                 ret = btrfs_grow_device(trans, device, new_size);
1730                 btrfs_commit_transaction(trans);
1731         } else if (new_size < old_size) {
1732                 ret = btrfs_shrink_device(device, new_size);
1733         } /* equal, nothing need to do */
1734
1735 out_free:
1736         kfree(vol_args);
1737 out:
1738         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1739         mnt_drop_write_file(file);
1740         return ret;
1741 }
1742
1743 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1744                                 const char *name, unsigned long fd, int subvol,
1745                                 u64 *transid, bool readonly,
1746                                 struct btrfs_qgroup_inherit *inherit)
1747 {
1748         int namelen;
1749         int ret = 0;
1750
1751         if (!S_ISDIR(file_inode(file)->i_mode))
1752                 return -ENOTDIR;
1753
1754         ret = mnt_want_write_file(file);
1755         if (ret)
1756                 goto out;
1757
1758         namelen = strlen(name);
1759         if (strchr(name, '/')) {
1760                 ret = -EINVAL;
1761                 goto out_drop_write;
1762         }
1763
1764         if (name[0] == '.' &&
1765            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1766                 ret = -EEXIST;
1767                 goto out_drop_write;
1768         }
1769
1770         if (subvol) {
1771                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1772                                      NULL, transid, readonly, inherit);
1773         } else {
1774                 struct fd src = fdget(fd);
1775                 struct inode *src_inode;
1776                 if (!src.file) {
1777                         ret = -EINVAL;
1778                         goto out_drop_write;
1779                 }
1780
1781                 src_inode = file_inode(src.file);
1782                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1783                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1784                                    "Snapshot src from another FS");
1785                         ret = -EXDEV;
1786                 } else if (!inode_owner_or_capable(src_inode)) {
1787                         /*
1788                          * Subvolume creation is not restricted, but snapshots
1789                          * are limited to own subvolumes only
1790                          */
1791                         ret = -EPERM;
1792                 } else {
1793                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1794                                              BTRFS_I(src_inode)->root,
1795                                              transid, readonly, inherit);
1796                 }
1797                 fdput(src);
1798         }
1799 out_drop_write:
1800         mnt_drop_write_file(file);
1801 out:
1802         return ret;
1803 }
1804
1805 static noinline int btrfs_ioctl_snap_create(struct file *file,
1806                                             void __user *arg, int subvol)
1807 {
1808         struct btrfs_ioctl_vol_args *vol_args;
1809         int ret;
1810
1811         if (!S_ISDIR(file_inode(file)->i_mode))
1812                 return -ENOTDIR;
1813
1814         vol_args = memdup_user(arg, sizeof(*vol_args));
1815         if (IS_ERR(vol_args))
1816                 return PTR_ERR(vol_args);
1817         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1818
1819         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1820                                               vol_args->fd, subvol,
1821                                               NULL, false, NULL);
1822
1823         kfree(vol_args);
1824         return ret;
1825 }
1826
1827 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1828                                                void __user *arg, int subvol)
1829 {
1830         struct btrfs_ioctl_vol_args_v2 *vol_args;
1831         int ret;
1832         u64 transid = 0;
1833         u64 *ptr = NULL;
1834         bool readonly = false;
1835         struct btrfs_qgroup_inherit *inherit = NULL;
1836
1837         if (!S_ISDIR(file_inode(file)->i_mode))
1838                 return -ENOTDIR;
1839
1840         vol_args = memdup_user(arg, sizeof(*vol_args));
1841         if (IS_ERR(vol_args))
1842                 return PTR_ERR(vol_args);
1843         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1844
1845         if (vol_args->flags &
1846             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1847               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1848                 ret = -EOPNOTSUPP;
1849                 goto free_args;
1850         }
1851
1852         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1853                 ptr = &transid;
1854         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1855                 readonly = true;
1856         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1857                 u64 nums;
1858
1859                 if (vol_args->size < sizeof(*inherit) ||
1860                     vol_args->size > PAGE_SIZE) {
1861                         ret = -EINVAL;
1862                         goto free_args;
1863                 }
1864                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1865                 if (IS_ERR(inherit)) {
1866                         ret = PTR_ERR(inherit);
1867                         goto free_args;
1868                 }
1869
1870                 if (inherit->num_qgroups > PAGE_SIZE ||
1871                     inherit->num_ref_copies > PAGE_SIZE ||
1872                     inherit->num_excl_copies > PAGE_SIZE) {
1873                         ret = -EINVAL;
1874                         goto free_inherit;
1875                 }
1876
1877                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1878                        2 * inherit->num_excl_copies;
1879                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1880                         ret = -EINVAL;
1881                         goto free_inherit;
1882                 }
1883         }
1884
1885         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1886                                               vol_args->fd, subvol, ptr,
1887                                               readonly, inherit);
1888         if (ret)
1889                 goto free_inherit;
1890
1891         if (ptr && copy_to_user(arg +
1892                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1893                                         transid),
1894                                 ptr, sizeof(*ptr)))
1895                 ret = -EFAULT;
1896
1897 free_inherit:
1898         kfree(inherit);
1899 free_args:
1900         kfree(vol_args);
1901         return ret;
1902 }
1903
1904 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1905                                                 void __user *arg)
1906 {
1907         struct inode *inode = file_inode(file);
1908         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1909         struct btrfs_root *root = BTRFS_I(inode)->root;
1910         int ret = 0;
1911         u64 flags = 0;
1912
1913         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1914                 return -EINVAL;
1915
1916         down_read(&fs_info->subvol_sem);
1917         if (btrfs_root_readonly(root))
1918                 flags |= BTRFS_SUBVOL_RDONLY;
1919         up_read(&fs_info->subvol_sem);
1920
1921         if (copy_to_user(arg, &flags, sizeof(flags)))
1922                 ret = -EFAULT;
1923
1924         return ret;
1925 }
1926
1927 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1928                                               void __user *arg)
1929 {
1930         struct inode *inode = file_inode(file);
1931         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1932         struct btrfs_root *root = BTRFS_I(inode)->root;
1933         struct btrfs_trans_handle *trans;
1934         u64 root_flags;
1935         u64 flags;
1936         int ret = 0;
1937
1938         if (!inode_owner_or_capable(inode))
1939                 return -EPERM;
1940
1941         ret = mnt_want_write_file(file);
1942         if (ret)
1943                 goto out;
1944
1945         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1946                 ret = -EINVAL;
1947                 goto out_drop_write;
1948         }
1949
1950         if (copy_from_user(&flags, arg, sizeof(flags))) {
1951                 ret = -EFAULT;
1952                 goto out_drop_write;
1953         }
1954
1955         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1956                 ret = -EINVAL;
1957                 goto out_drop_write;
1958         }
1959
1960         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1961                 ret = -EOPNOTSUPP;
1962                 goto out_drop_write;
1963         }
1964
1965         down_write(&fs_info->subvol_sem);
1966
1967         /* nothing to do */
1968         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1969                 goto out_drop_sem;
1970
1971         root_flags = btrfs_root_flags(&root->root_item);
1972         if (flags & BTRFS_SUBVOL_RDONLY) {
1973                 btrfs_set_root_flags(&root->root_item,
1974                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1975         } else {
1976                 /*
1977                  * Block RO -> RW transition if this subvolume is involved in
1978                  * send
1979                  */
1980                 spin_lock(&root->root_item_lock);
1981                 if (root->send_in_progress == 0) {
1982                         btrfs_set_root_flags(&root->root_item,
1983                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1984                         spin_unlock(&root->root_item_lock);
1985                 } else {
1986                         spin_unlock(&root->root_item_lock);
1987                         btrfs_warn(fs_info,
1988                                    "Attempt to set subvolume %llu read-write during send",
1989                                    root->root_key.objectid);
1990                         ret = -EPERM;
1991                         goto out_drop_sem;
1992                 }
1993         }
1994
1995         trans = btrfs_start_transaction(root, 1);
1996         if (IS_ERR(trans)) {
1997                 ret = PTR_ERR(trans);
1998                 goto out_reset;
1999         }
2000
2001         ret = btrfs_update_root(trans, fs_info->tree_root,
2002                                 &root->root_key, &root->root_item);
2003         if (ret < 0) {
2004                 btrfs_end_transaction(trans);
2005                 goto out_reset;
2006         }
2007
2008         ret = btrfs_commit_transaction(trans);
2009
2010 out_reset:
2011         if (ret)
2012                 btrfs_set_root_flags(&root->root_item, root_flags);
2013 out_drop_sem:
2014         up_write(&fs_info->subvol_sem);
2015 out_drop_write:
2016         mnt_drop_write_file(file);
2017 out:
2018         return ret;
2019 }
2020
2021 static noinline int key_in_sk(struct btrfs_key *key,
2022                               struct btrfs_ioctl_search_key *sk)
2023 {
2024         struct btrfs_key test;
2025         int ret;
2026
2027         test.objectid = sk->min_objectid;
2028         test.type = sk->min_type;
2029         test.offset = sk->min_offset;
2030
2031         ret = btrfs_comp_cpu_keys(key, &test);
2032         if (ret < 0)
2033                 return 0;
2034
2035         test.objectid = sk->max_objectid;
2036         test.type = sk->max_type;
2037         test.offset = sk->max_offset;
2038
2039         ret = btrfs_comp_cpu_keys(key, &test);
2040         if (ret > 0)
2041                 return 0;
2042         return 1;
2043 }
2044
2045 static noinline int copy_to_sk(struct btrfs_path *path,
2046                                struct btrfs_key *key,
2047                                struct btrfs_ioctl_search_key *sk,
2048                                size_t *buf_size,
2049                                char __user *ubuf,
2050                                unsigned long *sk_offset,
2051                                int *num_found)
2052 {
2053         u64 found_transid;
2054         struct extent_buffer *leaf;
2055         struct btrfs_ioctl_search_header sh;
2056         struct btrfs_key test;
2057         unsigned long item_off;
2058         unsigned long item_len;
2059         int nritems;
2060         int i;
2061         int slot;
2062         int ret = 0;
2063
2064         leaf = path->nodes[0];
2065         slot = path->slots[0];
2066         nritems = btrfs_header_nritems(leaf);
2067
2068         if (btrfs_header_generation(leaf) > sk->max_transid) {
2069                 i = nritems;
2070                 goto advance_key;
2071         }
2072         found_transid = btrfs_header_generation(leaf);
2073
2074         for (i = slot; i < nritems; i++) {
2075                 item_off = btrfs_item_ptr_offset(leaf, i);
2076                 item_len = btrfs_item_size_nr(leaf, i);
2077
2078                 btrfs_item_key_to_cpu(leaf, key, i);
2079                 if (!key_in_sk(key, sk))
2080                         continue;
2081
2082                 if (sizeof(sh) + item_len > *buf_size) {
2083                         if (*num_found) {
2084                                 ret = 1;
2085                                 goto out;
2086                         }
2087
2088                         /*
2089                          * return one empty item back for v1, which does not
2090                          * handle -EOVERFLOW
2091                          */
2092
2093                         *buf_size = sizeof(sh) + item_len;
2094                         item_len = 0;
2095                         ret = -EOVERFLOW;
2096                 }
2097
2098                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2099                         ret = 1;
2100                         goto out;
2101                 }
2102
2103                 sh.objectid = key->objectid;
2104                 sh.offset = key->offset;
2105                 sh.type = key->type;
2106                 sh.len = item_len;
2107                 sh.transid = found_transid;
2108
2109                 /*
2110                  * Copy search result header. If we fault then loop again so we
2111                  * can fault in the pages and -EFAULT there if there's a
2112                  * problem. Otherwise we'll fault and then copy the buffer in
2113                  * properly this next time through
2114                  */
2115                 if (probe_user_write(ubuf + *sk_offset, &sh, sizeof(sh))) {
2116                         ret = 0;
2117                         goto out;
2118                 }
2119
2120                 *sk_offset += sizeof(sh);
2121
2122                 if (item_len) {
2123                         char __user *up = ubuf + *sk_offset;
2124                         /*
2125                          * Copy the item, same behavior as above, but reset the
2126                          * * sk_offset so we copy the full thing again.
2127                          */
2128                         if (read_extent_buffer_to_user_nofault(leaf, up,
2129                                                 item_off, item_len)) {
2130                                 ret = 0;
2131                                 *sk_offset -= sizeof(sh);
2132                                 goto out;
2133                         }
2134
2135                         *sk_offset += item_len;
2136                 }
2137                 (*num_found)++;
2138
2139                 if (ret) /* -EOVERFLOW from above */
2140                         goto out;
2141
2142                 if (*num_found >= sk->nr_items) {
2143                         ret = 1;
2144                         goto out;
2145                 }
2146         }
2147 advance_key:
2148         ret = 0;
2149         test.objectid = sk->max_objectid;
2150         test.type = sk->max_type;
2151         test.offset = sk->max_offset;
2152         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2153                 ret = 1;
2154         else if (key->offset < (u64)-1)
2155                 key->offset++;
2156         else if (key->type < (u8)-1) {
2157                 key->offset = 0;
2158                 key->type++;
2159         } else if (key->objectid < (u64)-1) {
2160                 key->offset = 0;
2161                 key->type = 0;
2162                 key->objectid++;
2163         } else
2164                 ret = 1;
2165 out:
2166         /*
2167          *  0: all items from this leaf copied, continue with next
2168          *  1: * more items can be copied, but unused buffer is too small
2169          *     * all items were found
2170          *     Either way, it will stops the loop which iterates to the next
2171          *     leaf
2172          *  -EOVERFLOW: item was to large for buffer
2173          *  -EFAULT: could not copy extent buffer back to userspace
2174          */
2175         return ret;
2176 }
2177
2178 static noinline int search_ioctl(struct inode *inode,
2179                                  struct btrfs_ioctl_search_key *sk,
2180                                  size_t *buf_size,
2181                                  char __user *ubuf)
2182 {
2183         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2184         struct btrfs_root *root;
2185         struct btrfs_key key;
2186         struct btrfs_path *path;
2187         int ret;
2188         int num_found = 0;
2189         unsigned long sk_offset = 0;
2190
2191         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2192                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2193                 return -EOVERFLOW;
2194         }
2195
2196         path = btrfs_alloc_path();
2197         if (!path)
2198                 return -ENOMEM;
2199
2200         if (sk->tree_id == 0) {
2201                 /* search the root of the inode that was passed */
2202                 root = BTRFS_I(inode)->root;
2203         } else {
2204                 key.objectid = sk->tree_id;
2205                 key.type = BTRFS_ROOT_ITEM_KEY;
2206                 key.offset = (u64)-1;
2207                 root = btrfs_read_fs_root_no_name(info, &key);
2208                 if (IS_ERR(root)) {
2209                         btrfs_free_path(path);
2210                         return PTR_ERR(root);
2211                 }
2212         }
2213
2214         key.objectid = sk->min_objectid;
2215         key.type = sk->min_type;
2216         key.offset = sk->min_offset;
2217
2218         while (1) {
2219                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2220                                                *buf_size - sk_offset);
2221                 if (ret)
2222                         break;
2223
2224                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2225                 if (ret != 0) {
2226                         if (ret > 0)
2227                                 ret = 0;
2228                         goto err;
2229                 }
2230                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2231                                  &sk_offset, &num_found);
2232                 btrfs_release_path(path);
2233                 if (ret)
2234                         break;
2235
2236         }
2237         if (ret > 0)
2238                 ret = 0;
2239 err:
2240         sk->nr_items = num_found;
2241         btrfs_free_path(path);
2242         return ret;
2243 }
2244
2245 static noinline int btrfs_ioctl_tree_search(struct file *file,
2246                                            void __user *argp)
2247 {
2248         struct btrfs_ioctl_search_args __user *uargs;
2249         struct btrfs_ioctl_search_key sk;
2250         struct inode *inode;
2251         int ret;
2252         size_t buf_size;
2253
2254         if (!capable(CAP_SYS_ADMIN))
2255                 return -EPERM;
2256
2257         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2258
2259         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2260                 return -EFAULT;
2261
2262         buf_size = sizeof(uargs->buf);
2263
2264         inode = file_inode(file);
2265         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2266
2267         /*
2268          * In the origin implementation an overflow is handled by returning a
2269          * search header with a len of zero, so reset ret.
2270          */
2271         if (ret == -EOVERFLOW)
2272                 ret = 0;
2273
2274         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2275                 ret = -EFAULT;
2276         return ret;
2277 }
2278
2279 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2280                                                void __user *argp)
2281 {
2282         struct btrfs_ioctl_search_args_v2 __user *uarg;
2283         struct btrfs_ioctl_search_args_v2 args;
2284         struct inode *inode;
2285         int ret;
2286         size_t buf_size;
2287         const size_t buf_limit = SZ_16M;
2288
2289         if (!capable(CAP_SYS_ADMIN))
2290                 return -EPERM;
2291
2292         /* copy search header and buffer size */
2293         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2294         if (copy_from_user(&args, uarg, sizeof(args)))
2295                 return -EFAULT;
2296
2297         buf_size = args.buf_size;
2298
2299         /* limit result size to 16MB */
2300         if (buf_size > buf_limit)
2301                 buf_size = buf_limit;
2302
2303         inode = file_inode(file);
2304         ret = search_ioctl(inode, &args.key, &buf_size,
2305                            (char __user *)(&uarg->buf[0]));
2306         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2307                 ret = -EFAULT;
2308         else if (ret == -EOVERFLOW &&
2309                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2310                 ret = -EFAULT;
2311
2312         return ret;
2313 }
2314
2315 /*
2316  * Search INODE_REFs to identify path name of 'dirid' directory
2317  * in a 'tree_id' tree. and sets path name to 'name'.
2318  */
2319 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2320                                 u64 tree_id, u64 dirid, char *name)
2321 {
2322         struct btrfs_root *root;
2323         struct btrfs_key key;
2324         char *ptr;
2325         int ret = -1;
2326         int slot;
2327         int len;
2328         int total_len = 0;
2329         struct btrfs_inode_ref *iref;
2330         struct extent_buffer *l;
2331         struct btrfs_path *path;
2332
2333         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2334                 name[0]='\0';
2335                 return 0;
2336         }
2337
2338         path = btrfs_alloc_path();
2339         if (!path)
2340                 return -ENOMEM;
2341
2342         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2343
2344         key.objectid = tree_id;
2345         key.type = BTRFS_ROOT_ITEM_KEY;
2346         key.offset = (u64)-1;
2347         root = btrfs_read_fs_root_no_name(info, &key);
2348         if (IS_ERR(root)) {
2349                 ret = PTR_ERR(root);
2350                 goto out;
2351         }
2352
2353         key.objectid = dirid;
2354         key.type = BTRFS_INODE_REF_KEY;
2355         key.offset = (u64)-1;
2356
2357         while (1) {
2358                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2359                 if (ret < 0)
2360                         goto out;
2361                 else if (ret > 0) {
2362                         ret = btrfs_previous_item(root, path, dirid,
2363                                                   BTRFS_INODE_REF_KEY);
2364                         if (ret < 0)
2365                                 goto out;
2366                         else if (ret > 0) {
2367                                 ret = -ENOENT;
2368                                 goto out;
2369                         }
2370                 }
2371
2372                 l = path->nodes[0];
2373                 slot = path->slots[0];
2374                 btrfs_item_key_to_cpu(l, &key, slot);
2375
2376                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2377                 len = btrfs_inode_ref_name_len(l, iref);
2378                 ptr -= len + 1;
2379                 total_len += len + 1;
2380                 if (ptr < name) {
2381                         ret = -ENAMETOOLONG;
2382                         goto out;
2383                 }
2384
2385                 *(ptr + len) = '/';
2386                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2387
2388                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2389                         break;
2390
2391                 btrfs_release_path(path);
2392                 key.objectid = key.offset;
2393                 key.offset = (u64)-1;
2394                 dirid = key.objectid;
2395         }
2396         memmove(name, ptr, total_len);
2397         name[total_len] = '\0';
2398         ret = 0;
2399 out:
2400         btrfs_free_path(path);
2401         return ret;
2402 }
2403
2404 static int btrfs_search_path_in_tree_user(struct inode *inode,
2405                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2406 {
2407         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2408         struct super_block *sb = inode->i_sb;
2409         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2410         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2411         u64 dirid = args->dirid;
2412         unsigned long item_off;
2413         unsigned long item_len;
2414         struct btrfs_inode_ref *iref;
2415         struct btrfs_root_ref *rref;
2416         struct btrfs_root *root;
2417         struct btrfs_path *path;
2418         struct btrfs_key key, key2;
2419         struct extent_buffer *leaf;
2420         struct inode *temp_inode;
2421         char *ptr;
2422         int slot;
2423         int len;
2424         int total_len = 0;
2425         int ret;
2426
2427         path = btrfs_alloc_path();
2428         if (!path)
2429                 return -ENOMEM;
2430
2431         /*
2432          * If the bottom subvolume does not exist directly under upper_limit,
2433          * construct the path in from the bottom up.
2434          */
2435         if (dirid != upper_limit.objectid) {
2436                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2437
2438                 key.objectid = treeid;
2439                 key.type = BTRFS_ROOT_ITEM_KEY;
2440                 key.offset = (u64)-1;
2441                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2442                 if (IS_ERR(root)) {
2443                         ret = PTR_ERR(root);
2444                         goto out;
2445                 }
2446
2447                 key.objectid = dirid;
2448                 key.type = BTRFS_INODE_REF_KEY;
2449                 key.offset = (u64)-1;
2450                 while (1) {
2451                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2452                         if (ret < 0) {
2453                                 goto out;
2454                         } else if (ret > 0) {
2455                                 ret = btrfs_previous_item(root, path, dirid,
2456                                                           BTRFS_INODE_REF_KEY);
2457                                 if (ret < 0) {
2458                                         goto out;
2459                                 } else if (ret > 0) {
2460                                         ret = -ENOENT;
2461                                         goto out;
2462                                 }
2463                         }
2464
2465                         leaf = path->nodes[0];
2466                         slot = path->slots[0];
2467                         btrfs_item_key_to_cpu(leaf, &key, slot);
2468
2469                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2470                         len = btrfs_inode_ref_name_len(leaf, iref);
2471                         ptr -= len + 1;
2472                         total_len += len + 1;
2473                         if (ptr < args->path) {
2474                                 ret = -ENAMETOOLONG;
2475                                 goto out;
2476                         }
2477
2478                         *(ptr + len) = '/';
2479                         read_extent_buffer(leaf, ptr,
2480                                         (unsigned long)(iref + 1), len);
2481
2482                         /* Check the read+exec permission of this directory */
2483                         ret = btrfs_previous_item(root, path, dirid,
2484                                                   BTRFS_INODE_ITEM_KEY);
2485                         if (ret < 0) {
2486                                 goto out;
2487                         } else if (ret > 0) {
2488                                 ret = -ENOENT;
2489                                 goto out;
2490                         }
2491
2492                         leaf = path->nodes[0];
2493                         slot = path->slots[0];
2494                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2495                         if (key2.objectid != dirid) {
2496                                 ret = -ENOENT;
2497                                 goto out;
2498                         }
2499
2500                         temp_inode = btrfs_iget(sb, &key2, root, NULL);
2501                         if (IS_ERR(temp_inode)) {
2502                                 ret = PTR_ERR(temp_inode);
2503                                 goto out;
2504                         }
2505                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2506                         iput(temp_inode);
2507                         if (ret) {
2508                                 ret = -EACCES;
2509                                 goto out;
2510                         }
2511
2512                         if (key.offset == upper_limit.objectid)
2513                                 break;
2514                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2515                                 ret = -EACCES;
2516                                 goto out;
2517                         }
2518
2519                         btrfs_release_path(path);
2520                         key.objectid = key.offset;
2521                         key.offset = (u64)-1;
2522                         dirid = key.objectid;
2523                 }
2524
2525                 memmove(args->path, ptr, total_len);
2526                 args->path[total_len] = '\0';
2527                 btrfs_release_path(path);
2528         }
2529
2530         /* Get the bottom subvolume's name from ROOT_REF */
2531         root = fs_info->tree_root;
2532         key.objectid = treeid;
2533         key.type = BTRFS_ROOT_REF_KEY;
2534         key.offset = args->treeid;
2535         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2536         if (ret < 0) {
2537                 goto out;
2538         } else if (ret > 0) {
2539                 ret = -ENOENT;
2540                 goto out;
2541         }
2542
2543         leaf = path->nodes[0];
2544         slot = path->slots[0];
2545         btrfs_item_key_to_cpu(leaf, &key, slot);
2546
2547         item_off = btrfs_item_ptr_offset(leaf, slot);
2548         item_len = btrfs_item_size_nr(leaf, slot);
2549         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2550         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2551         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2552                 ret = -EINVAL;
2553                 goto out;
2554         }
2555
2556         /* Copy subvolume's name */
2557         item_off += sizeof(struct btrfs_root_ref);
2558         item_len -= sizeof(struct btrfs_root_ref);
2559         read_extent_buffer(leaf, args->name, item_off, item_len);
2560         args->name[item_len] = 0;
2561
2562 out:
2563         btrfs_free_path(path);
2564         return ret;
2565 }
2566
2567 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2568                                            void __user *argp)
2569 {
2570         struct btrfs_ioctl_ino_lookup_args *args;
2571         struct inode *inode;
2572         int ret = 0;
2573
2574         args = memdup_user(argp, sizeof(*args));
2575         if (IS_ERR(args))
2576                 return PTR_ERR(args);
2577
2578         inode = file_inode(file);
2579
2580         /*
2581          * Unprivileged query to obtain the containing subvolume root id. The
2582          * path is reset so it's consistent with btrfs_search_path_in_tree.
2583          */
2584         if (args->treeid == 0)
2585                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2586
2587         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2588                 args->name[0] = 0;
2589                 goto out;
2590         }
2591
2592         if (!capable(CAP_SYS_ADMIN)) {
2593                 ret = -EPERM;
2594                 goto out;
2595         }
2596
2597         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2598                                         args->treeid, args->objectid,
2599                                         args->name);
2600
2601 out:
2602         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2603                 ret = -EFAULT;
2604
2605         kfree(args);
2606         return ret;
2607 }
2608
2609 /*
2610  * Version of ino_lookup ioctl (unprivileged)
2611  *
2612  * The main differences from ino_lookup ioctl are:
2613  *
2614  *   1. Read + Exec permission will be checked using inode_permission() during
2615  *      path construction. -EACCES will be returned in case of failure.
2616  *   2. Path construction will be stopped at the inode number which corresponds
2617  *      to the fd with which this ioctl is called. If constructed path does not
2618  *      exist under fd's inode, -EACCES will be returned.
2619  *   3. The name of bottom subvolume is also searched and filled.
2620  */
2621 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2622 {
2623         struct btrfs_ioctl_ino_lookup_user_args *args;
2624         struct inode *inode;
2625         int ret;
2626
2627         args = memdup_user(argp, sizeof(*args));
2628         if (IS_ERR(args))
2629                 return PTR_ERR(args);
2630
2631         inode = file_inode(file);
2632
2633         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2634             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2635                 /*
2636                  * The subvolume does not exist under fd with which this is
2637                  * called
2638                  */
2639                 kfree(args);
2640                 return -EACCES;
2641         }
2642
2643         ret = btrfs_search_path_in_tree_user(inode, args);
2644
2645         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2646                 ret = -EFAULT;
2647
2648         kfree(args);
2649         return ret;
2650 }
2651
2652 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2653 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2654 {
2655         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2656         struct btrfs_fs_info *fs_info;
2657         struct btrfs_root *root;
2658         struct btrfs_path *path;
2659         struct btrfs_key key;
2660         struct btrfs_root_item *root_item;
2661         struct btrfs_root_ref *rref;
2662         struct extent_buffer *leaf;
2663         unsigned long item_off;
2664         unsigned long item_len;
2665         struct inode *inode;
2666         int slot;
2667         int ret = 0;
2668
2669         path = btrfs_alloc_path();
2670         if (!path)
2671                 return -ENOMEM;
2672
2673         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2674         if (!subvol_info) {
2675                 btrfs_free_path(path);
2676                 return -ENOMEM;
2677         }
2678
2679         inode = file_inode(file);
2680         fs_info = BTRFS_I(inode)->root->fs_info;
2681
2682         /* Get root_item of inode's subvolume */
2683         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2684         key.type = BTRFS_ROOT_ITEM_KEY;
2685         key.offset = (u64)-1;
2686         root = btrfs_read_fs_root_no_name(fs_info, &key);
2687         if (IS_ERR(root)) {
2688                 ret = PTR_ERR(root);
2689                 goto out;
2690         }
2691         root_item = &root->root_item;
2692
2693         subvol_info->treeid = key.objectid;
2694
2695         subvol_info->generation = btrfs_root_generation(root_item);
2696         subvol_info->flags = btrfs_root_flags(root_item);
2697
2698         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2699         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2700                                                     BTRFS_UUID_SIZE);
2701         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2702                                                     BTRFS_UUID_SIZE);
2703
2704         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2705         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2706         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2707
2708         subvol_info->otransid = btrfs_root_otransid(root_item);
2709         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2710         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2711
2712         subvol_info->stransid = btrfs_root_stransid(root_item);
2713         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2714         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2715
2716         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2717         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2718         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2719
2720         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2721                 /* Search root tree for ROOT_BACKREF of this subvolume */
2722                 root = fs_info->tree_root;
2723
2724                 key.type = BTRFS_ROOT_BACKREF_KEY;
2725                 key.offset = 0;
2726                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2727                 if (ret < 0) {
2728                         goto out;
2729                 } else if (path->slots[0] >=
2730                            btrfs_header_nritems(path->nodes[0])) {
2731                         ret = btrfs_next_leaf(root, path);
2732                         if (ret < 0) {
2733                                 goto out;
2734                         } else if (ret > 0) {
2735                                 ret = -EUCLEAN;
2736                                 goto out;
2737                         }
2738                 }
2739
2740                 leaf = path->nodes[0];
2741                 slot = path->slots[0];
2742                 btrfs_item_key_to_cpu(leaf, &key, slot);
2743                 if (key.objectid == subvol_info->treeid &&
2744                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2745                         subvol_info->parent_id = key.offset;
2746
2747                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2748                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2749
2750                         item_off = btrfs_item_ptr_offset(leaf, slot)
2751                                         + sizeof(struct btrfs_root_ref);
2752                         item_len = btrfs_item_size_nr(leaf, slot)
2753                                         - sizeof(struct btrfs_root_ref);
2754                         read_extent_buffer(leaf, subvol_info->name,
2755                                            item_off, item_len);
2756                 } else {
2757                         ret = -ENOENT;
2758                         goto out;
2759                 }
2760         }
2761
2762         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2763                 ret = -EFAULT;
2764
2765 out:
2766         btrfs_free_path(path);
2767         kzfree(subvol_info);
2768         return ret;
2769 }
2770
2771 /*
2772  * Return ROOT_REF information of the subvolume containing this inode
2773  * except the subvolume name.
2774  */
2775 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2776 {
2777         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2778         struct btrfs_root_ref *rref;
2779         struct btrfs_root *root;
2780         struct btrfs_path *path;
2781         struct btrfs_key key;
2782         struct extent_buffer *leaf;
2783         struct inode *inode;
2784         u64 objectid;
2785         int slot;
2786         int ret;
2787         u8 found;
2788
2789         path = btrfs_alloc_path();
2790         if (!path)
2791                 return -ENOMEM;
2792
2793         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2794         if (IS_ERR(rootrefs)) {
2795                 btrfs_free_path(path);
2796                 return PTR_ERR(rootrefs);
2797         }
2798
2799         inode = file_inode(file);
2800         root = BTRFS_I(inode)->root->fs_info->tree_root;
2801         objectid = BTRFS_I(inode)->root->root_key.objectid;
2802
2803         key.objectid = objectid;
2804         key.type = BTRFS_ROOT_REF_KEY;
2805         key.offset = rootrefs->min_treeid;
2806         found = 0;
2807
2808         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2809         if (ret < 0) {
2810                 goto out;
2811         } else if (path->slots[0] >=
2812                    btrfs_header_nritems(path->nodes[0])) {
2813                 ret = btrfs_next_leaf(root, path);
2814                 if (ret < 0) {
2815                         goto out;
2816                 } else if (ret > 0) {
2817                         ret = -EUCLEAN;
2818                         goto out;
2819                 }
2820         }
2821         while (1) {
2822                 leaf = path->nodes[0];
2823                 slot = path->slots[0];
2824
2825                 btrfs_item_key_to_cpu(leaf, &key, slot);
2826                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2827                         ret = 0;
2828                         goto out;
2829                 }
2830
2831                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2832                         ret = -EOVERFLOW;
2833                         goto out;
2834                 }
2835
2836                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2837                 rootrefs->rootref[found].treeid = key.offset;
2838                 rootrefs->rootref[found].dirid =
2839                                   btrfs_root_ref_dirid(leaf, rref);
2840                 found++;
2841
2842                 ret = btrfs_next_item(root, path);
2843                 if (ret < 0) {
2844                         goto out;
2845                 } else if (ret > 0) {
2846                         ret = -EUCLEAN;
2847                         goto out;
2848                 }
2849         }
2850
2851 out:
2852         if (!ret || ret == -EOVERFLOW) {
2853                 rootrefs->num_items = found;
2854                 /* update min_treeid for next search */
2855                 if (found)
2856                         rootrefs->min_treeid =
2857                                 rootrefs->rootref[found - 1].treeid + 1;
2858                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2859                         ret = -EFAULT;
2860         }
2861
2862         kfree(rootrefs);
2863         btrfs_free_path(path);
2864
2865         return ret;
2866 }
2867
2868 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2869                                              void __user *arg)
2870 {
2871         struct dentry *parent = file->f_path.dentry;
2872         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2873         struct dentry *dentry;
2874         struct inode *dir = d_inode(parent);
2875         struct inode *inode;
2876         struct btrfs_root *root = BTRFS_I(dir)->root;
2877         struct btrfs_root *dest = NULL;
2878         struct btrfs_ioctl_vol_args *vol_args;
2879         int namelen;
2880         int err = 0;
2881
2882         if (!S_ISDIR(dir->i_mode))
2883                 return -ENOTDIR;
2884
2885         vol_args = memdup_user(arg, sizeof(*vol_args));
2886         if (IS_ERR(vol_args))
2887                 return PTR_ERR(vol_args);
2888
2889         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2890         namelen = strlen(vol_args->name);
2891         if (strchr(vol_args->name, '/') ||
2892             strncmp(vol_args->name, "..", namelen) == 0) {
2893                 err = -EINVAL;
2894                 goto out;
2895         }
2896
2897         err = mnt_want_write_file(file);
2898         if (err)
2899                 goto out;
2900
2901
2902         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2903         if (err == -EINTR)
2904                 goto out_drop_write;
2905         dentry = lookup_one_len(vol_args->name, parent, namelen);
2906         if (IS_ERR(dentry)) {
2907                 err = PTR_ERR(dentry);
2908                 goto out_unlock_dir;
2909         }
2910
2911         if (d_really_is_negative(dentry)) {
2912                 err = -ENOENT;
2913                 goto out_dput;
2914         }
2915
2916         inode = d_inode(dentry);
2917         dest = BTRFS_I(inode)->root;
2918         if (!capable(CAP_SYS_ADMIN)) {
2919                 /*
2920                  * Regular user.  Only allow this with a special mount
2921                  * option, when the user has write+exec access to the
2922                  * subvol root, and when rmdir(2) would have been
2923                  * allowed.
2924                  *
2925                  * Note that this is _not_ check that the subvol is
2926                  * empty or doesn't contain data that we wouldn't
2927                  * otherwise be able to delete.
2928                  *
2929                  * Users who want to delete empty subvols should try
2930                  * rmdir(2).
2931                  */
2932                 err = -EPERM;
2933                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2934                         goto out_dput;
2935
2936                 /*
2937                  * Do not allow deletion if the parent dir is the same
2938                  * as the dir to be deleted.  That means the ioctl
2939                  * must be called on the dentry referencing the root
2940                  * of the subvol, not a random directory contained
2941                  * within it.
2942                  */
2943                 err = -EINVAL;
2944                 if (root == dest)
2945                         goto out_dput;
2946
2947                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2948                 if (err)
2949                         goto out_dput;
2950         }
2951
2952         /* check if subvolume may be deleted by a user */
2953         err = btrfs_may_delete(dir, dentry, 1);
2954         if (err)
2955                 goto out_dput;
2956
2957         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2958                 err = -EINVAL;
2959                 goto out_dput;
2960         }
2961
2962         inode_lock(inode);
2963         err = btrfs_delete_subvolume(dir, dentry);
2964         inode_unlock(inode);
2965         if (!err)
2966                 d_delete(dentry);
2967
2968 out_dput:
2969         dput(dentry);
2970 out_unlock_dir:
2971         inode_unlock(dir);
2972 out_drop_write:
2973         mnt_drop_write_file(file);
2974 out:
2975         kfree(vol_args);
2976         return err;
2977 }
2978
2979 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2980 {
2981         struct inode *inode = file_inode(file);
2982         struct btrfs_root *root = BTRFS_I(inode)->root;
2983         struct btrfs_ioctl_defrag_range_args *range;
2984         int ret;
2985
2986         ret = mnt_want_write_file(file);
2987         if (ret)
2988                 return ret;
2989
2990         if (btrfs_root_readonly(root)) {
2991                 ret = -EROFS;
2992                 goto out;
2993         }
2994
2995         switch (inode->i_mode & S_IFMT) {
2996         case S_IFDIR:
2997                 if (!capable(CAP_SYS_ADMIN)) {
2998                         ret = -EPERM;
2999                         goto out;
3000                 }
3001                 ret = btrfs_defrag_root(root);
3002                 break;
3003         case S_IFREG:
3004                 /*
3005                  * Note that this does not check the file descriptor for write
3006                  * access. This prevents defragmenting executables that are
3007                  * running and allows defrag on files open in read-only mode.
3008                  */
3009                 if (!capable(CAP_SYS_ADMIN) &&
3010                     inode_permission(inode, MAY_WRITE)) {
3011                         ret = -EPERM;
3012                         goto out;
3013                 }
3014
3015                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3016                 if (!range) {
3017                         ret = -ENOMEM;
3018                         goto out;
3019                 }
3020
3021                 if (argp) {
3022                         if (copy_from_user(range, argp,
3023                                            sizeof(*range))) {
3024                                 ret = -EFAULT;
3025                                 kfree(range);
3026                                 goto out;
3027                         }
3028                         /* compression requires us to start the IO */
3029                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3030                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3031                                 range->extent_thresh = (u32)-1;
3032                         }
3033                 } else {
3034                         /* the rest are all set to zero by kzalloc */
3035                         range->len = (u64)-1;
3036                 }
3037                 ret = btrfs_defrag_file(file_inode(file), file,
3038                                         range, BTRFS_OLDEST_GENERATION, 0);
3039                 if (ret > 0)
3040                         ret = 0;
3041                 kfree(range);
3042                 break;
3043         default:
3044                 ret = -EINVAL;
3045         }
3046 out:
3047         mnt_drop_write_file(file);
3048         return ret;
3049 }
3050
3051 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3052 {
3053         struct btrfs_ioctl_vol_args *vol_args;
3054         int ret;
3055
3056         if (!capable(CAP_SYS_ADMIN))
3057                 return -EPERM;
3058
3059         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3060                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3061
3062         vol_args = memdup_user(arg, sizeof(*vol_args));
3063         if (IS_ERR(vol_args)) {
3064                 ret = PTR_ERR(vol_args);
3065                 goto out;
3066         }
3067
3068         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3069         ret = btrfs_init_new_device(fs_info, vol_args->name);
3070
3071         if (!ret)
3072                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3073
3074         kfree(vol_args);
3075 out:
3076         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3077         return ret;
3078 }
3079
3080 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3081 {
3082         struct inode *inode = file_inode(file);
3083         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3084         struct btrfs_ioctl_vol_args_v2 *vol_args;
3085         int ret;
3086
3087         if (!capable(CAP_SYS_ADMIN))
3088                 return -EPERM;
3089
3090         ret = mnt_want_write_file(file);
3091         if (ret)
3092                 return ret;
3093
3094         vol_args = memdup_user(arg, sizeof(*vol_args));
3095         if (IS_ERR(vol_args)) {
3096                 ret = PTR_ERR(vol_args);
3097                 goto err_drop;
3098         }
3099
3100         /* Check for compatibility reject unknown flags */
3101         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3102                 ret = -EOPNOTSUPP;
3103                 goto out;
3104         }
3105
3106         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3107                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3108                 goto out;
3109         }
3110
3111         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3112                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3113         } else {
3114                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3115                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3116         }
3117         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3118
3119         if (!ret) {
3120                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3121                         btrfs_info(fs_info, "device deleted: id %llu",
3122                                         vol_args->devid);
3123                 else
3124                         btrfs_info(fs_info, "device deleted: %s",
3125                                         vol_args->name);
3126         }
3127 out:
3128         kfree(vol_args);
3129 err_drop:
3130         mnt_drop_write_file(file);
3131         return ret;
3132 }
3133
3134 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3135 {
3136         struct inode *inode = file_inode(file);
3137         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3138         struct btrfs_ioctl_vol_args *vol_args;
3139         int ret;
3140
3141         if (!capable(CAP_SYS_ADMIN))
3142                 return -EPERM;
3143
3144         ret = mnt_want_write_file(file);
3145         if (ret)
3146                 return ret;
3147
3148         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3149                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3150                 goto out_drop_write;
3151         }
3152
3153         vol_args = memdup_user(arg, sizeof(*vol_args));
3154         if (IS_ERR(vol_args)) {
3155                 ret = PTR_ERR(vol_args);
3156                 goto out;
3157         }
3158
3159         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3160         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3161
3162         if (!ret)
3163                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3164         kfree(vol_args);
3165 out:
3166         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3167 out_drop_write:
3168         mnt_drop_write_file(file);
3169
3170         return ret;
3171 }
3172
3173 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3174                                 void __user *arg)
3175 {
3176         struct btrfs_ioctl_fs_info_args *fi_args;
3177         struct btrfs_device *device;
3178         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3179         int ret = 0;
3180
3181         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3182         if (!fi_args)
3183                 return -ENOMEM;
3184
3185         rcu_read_lock();
3186         fi_args->num_devices = fs_devices->num_devices;
3187
3188         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3189                 if (device->devid > fi_args->max_id)
3190                         fi_args->max_id = device->devid;
3191         }
3192         rcu_read_unlock();
3193
3194         memcpy(&fi_args->fsid, fs_info->fsid, sizeof(fi_args->fsid));
3195         fi_args->nodesize = fs_info->nodesize;
3196         fi_args->sectorsize = fs_info->sectorsize;
3197         fi_args->clone_alignment = fs_info->sectorsize;
3198
3199         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3200                 ret = -EFAULT;
3201
3202         kfree(fi_args);
3203         return ret;
3204 }
3205
3206 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3207                                  void __user *arg)
3208 {
3209         struct btrfs_ioctl_dev_info_args *di_args;
3210         struct btrfs_device *dev;
3211         int ret = 0;
3212         char *s_uuid = NULL;
3213
3214         di_args = memdup_user(arg, sizeof(*di_args));
3215         if (IS_ERR(di_args))
3216                 return PTR_ERR(di_args);
3217
3218         if (!btrfs_is_empty_uuid(di_args->uuid))
3219                 s_uuid = di_args->uuid;
3220
3221         rcu_read_lock();
3222         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3223                                 NULL, true);
3224
3225         if (!dev) {
3226                 ret = -ENODEV;
3227                 goto out;
3228         }
3229
3230         di_args->devid = dev->devid;
3231         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3232         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3233         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3234         if (dev->name) {
3235                 strncpy(di_args->path, rcu_str_deref(dev->name),
3236                                 sizeof(di_args->path) - 1);
3237                 di_args->path[sizeof(di_args->path) - 1] = 0;
3238         } else {
3239                 di_args->path[0] = '\0';
3240         }
3241
3242 out:
3243         rcu_read_unlock();
3244         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3245                 ret = -EFAULT;
3246
3247         kfree(di_args);
3248         return ret;
3249 }
3250
3251 static struct page *extent_same_get_page(struct inode *inode, pgoff_t index)
3252 {
3253         struct page *page;
3254
3255         page = grab_cache_page(inode->i_mapping, index);
3256         if (!page)
3257                 return ERR_PTR(-ENOMEM);
3258
3259         if (!PageUptodate(page)) {
3260                 int ret;
3261
3262                 ret = btrfs_readpage(NULL, page);
3263                 if (ret)
3264                         return ERR_PTR(ret);
3265                 lock_page(page);
3266                 if (!PageUptodate(page)) {
3267                         unlock_page(page);
3268                         put_page(page);
3269                         return ERR_PTR(-EIO);
3270                 }
3271                 if (page->mapping != inode->i_mapping) {
3272                         unlock_page(page);
3273                         put_page(page);
3274                         return ERR_PTR(-EAGAIN);
3275                 }
3276         }
3277
3278         return page;
3279 }
3280
3281 static int gather_extent_pages(struct inode *inode, struct page **pages,
3282                                int num_pages, u64 off)
3283 {
3284         int i;
3285         pgoff_t index = off >> PAGE_SHIFT;
3286
3287         for (i = 0; i < num_pages; i++) {
3288 again:
3289                 pages[i] = extent_same_get_page(inode, index + i);
3290                 if (IS_ERR(pages[i])) {
3291                         int err = PTR_ERR(pages[i]);
3292
3293                         if (err == -EAGAIN)
3294                                 goto again;
3295                         pages[i] = NULL;
3296                         return err;
3297                 }
3298         }
3299         return 0;
3300 }
3301
3302 static int lock_extent_range(struct inode *inode, u64 off, u64 len,
3303                              bool retry_range_locking)
3304 {
3305         /*
3306          * Do any pending delalloc/csum calculations on inode, one way or
3307          * another, and lock file content.
3308          * The locking order is:
3309          *
3310          *   1) pages
3311          *   2) range in the inode's io tree
3312          */
3313         while (1) {
3314                 struct btrfs_ordered_extent *ordered;
3315                 lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3316                 ordered = btrfs_lookup_first_ordered_extent(inode,
3317                                                             off + len - 1);
3318                 if ((!ordered ||
3319                      ordered->file_offset + ordered->len <= off ||
3320                      ordered->file_offset >= off + len) &&
3321                     !test_range_bit(&BTRFS_I(inode)->io_tree, off,
3322                                     off + len - 1, EXTENT_DELALLOC, 0, NULL)) {
3323                         if (ordered)
3324                                 btrfs_put_ordered_extent(ordered);
3325                         break;
3326                 }
3327                 unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1);
3328                 if (ordered)
3329                         btrfs_put_ordered_extent(ordered);
3330                 if (!retry_range_locking)
3331                         return -EAGAIN;
3332                 btrfs_wait_ordered_range(inode, off, len);
3333         }
3334         return 0;
3335 }
3336
3337 static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2)
3338 {
3339         inode_unlock(inode1);
3340         inode_unlock(inode2);
3341 }
3342
3343 static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2)
3344 {
3345         if (inode1 < inode2)
3346                 swap(inode1, inode2);
3347
3348         inode_lock_nested(inode1, I_MUTEX_PARENT);
3349         inode_lock_nested(inode2, I_MUTEX_CHILD);
3350 }
3351
3352 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3353                                       struct inode *inode2, u64 loff2, u64 len)
3354 {
3355         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3356         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3357 }
3358
3359 static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3360                                     struct inode *inode2, u64 loff2, u64 len,
3361                                     bool retry_range_locking)
3362 {
3363         int ret;
3364
3365         if (inode1 < inode2) {
3366                 swap(inode1, inode2);
3367                 swap(loff1, loff2);
3368         }
3369         ret = lock_extent_range(inode1, loff1, len, retry_range_locking);
3370         if (ret)
3371                 return ret;
3372         ret = lock_extent_range(inode2, loff2, len, retry_range_locking);
3373         if (ret)
3374                 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1,
3375                               loff1 + len - 1);
3376         return ret;
3377 }
3378
3379 struct cmp_pages {
3380         int             num_pages;
3381         struct page     **src_pages;
3382         struct page     **dst_pages;
3383 };
3384
3385 static void btrfs_cmp_data_free(struct cmp_pages *cmp)
3386 {
3387         int i;
3388         struct page *pg;
3389
3390         for (i = 0; i < cmp->num_pages; i++) {
3391                 pg = cmp->src_pages[i];
3392                 if (pg) {
3393                         unlock_page(pg);
3394                         put_page(pg);
3395                         cmp->src_pages[i] = NULL;
3396                 }
3397                 pg = cmp->dst_pages[i];
3398                 if (pg) {
3399                         unlock_page(pg);
3400                         put_page(pg);
3401                         cmp->dst_pages[i] = NULL;
3402                 }
3403         }
3404 }
3405
3406 static int btrfs_cmp_data_prepare(struct inode *src, u64 loff,
3407                                   struct inode *dst, u64 dst_loff,
3408                                   u64 len, struct cmp_pages *cmp)
3409 {
3410         int ret;
3411         int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT;
3412
3413         cmp->num_pages = num_pages;
3414
3415         ret = gather_extent_pages(src, cmp->src_pages, num_pages, loff);
3416         if (ret)
3417                 goto out;
3418
3419         ret = gather_extent_pages(dst, cmp->dst_pages, num_pages, dst_loff);
3420
3421 out:
3422         if (ret)
3423                 btrfs_cmp_data_free(cmp);
3424         return ret;
3425 }
3426
3427 static int btrfs_cmp_data(u64 len, struct cmp_pages *cmp)
3428 {
3429         int ret = 0;
3430         int i;
3431         struct page *src_page, *dst_page;
3432         unsigned int cmp_len = PAGE_SIZE;
3433         void *addr, *dst_addr;
3434
3435         i = 0;
3436         while (len) {
3437                 if (len < PAGE_SIZE)
3438                         cmp_len = len;
3439
3440                 BUG_ON(i >= cmp->num_pages);
3441
3442                 src_page = cmp->src_pages[i];
3443                 dst_page = cmp->dst_pages[i];
3444                 ASSERT(PageLocked(src_page));
3445                 ASSERT(PageLocked(dst_page));
3446
3447                 addr = kmap_atomic(src_page);
3448                 dst_addr = kmap_atomic(dst_page);
3449
3450                 flush_dcache_page(src_page);
3451                 flush_dcache_page(dst_page);
3452
3453                 if (memcmp(addr, dst_addr, cmp_len))
3454                         ret = -EBADE;
3455
3456                 kunmap_atomic(addr);
3457                 kunmap_atomic(dst_addr);
3458
3459                 if (ret)
3460                         break;
3461
3462                 len -= cmp_len;
3463                 i++;
3464         }
3465
3466         return ret;
3467 }
3468
3469 static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen,
3470                                      u64 olen)
3471 {
3472         u64 len = *plen;
3473         u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize;
3474
3475         if (off + olen > inode->i_size || off + olen < off)
3476                 return -EINVAL;
3477
3478         /* if we extend to eof, continue to block boundary */
3479         if (off + len == inode->i_size)
3480                 *plen = len = ALIGN(inode->i_size, bs) - off;
3481
3482         /* Check that we are block aligned - btrfs_clone() requires this */
3483         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs))
3484                 return -EINVAL;
3485
3486         return 0;
3487 }
3488
3489 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 olen,
3490                                    struct inode *dst, u64 dst_loff,
3491                                    struct cmp_pages *cmp)
3492 {
3493         int ret;
3494         u64 len = olen;
3495         bool same_inode = (src == dst);
3496         u64 same_lock_start = 0;
3497         u64 same_lock_len = 0;
3498
3499         ret = extent_same_check_offsets(src, loff, &len, olen);
3500         if (ret)
3501                 return ret;
3502
3503         ret = extent_same_check_offsets(dst, dst_loff, &len, olen);
3504         if (ret)
3505                 return ret;
3506
3507         if (same_inode) {
3508                 /*
3509                  * Single inode case wants the same checks, except we
3510                  * don't want our length pushed out past i_size as
3511                  * comparing that data range makes no sense.
3512                  *
3513                  * extent_same_check_offsets() will do this for an
3514                  * unaligned length at i_size, so catch it here and
3515                  * reject the request.
3516                  *
3517                  * This effectively means we require aligned extents
3518                  * for the single-inode case, whereas the other cases
3519                  * allow an unaligned length so long as it ends at
3520                  * i_size.
3521                  */
3522                 if (len != olen)
3523                         return -EINVAL;
3524
3525                 /* Check for overlapping ranges */
3526                 if (dst_loff + len > loff && dst_loff < loff + len)
3527                         return -EINVAL;
3528
3529                 same_lock_start = min_t(u64, loff, dst_loff);
3530                 same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start;
3531         } else {
3532                 /*
3533                  * If the source and destination inodes are different, the
3534                  * source's range end offset matches the source's i_size, that
3535                  * i_size is not a multiple of the sector size, and the
3536                  * destination range does not go past the destination's i_size,
3537                  * we must round down the length to the nearest sector size
3538                  * multiple. If we don't do this adjustment we end replacing
3539                  * with zeroes the bytes in the range that starts at the
3540                  * deduplication range's end offset and ends at the next sector
3541                  * size multiple.
3542                  */
3543                 if (loff + olen == i_size_read(src) &&
3544                     dst_loff + len < i_size_read(dst)) {
3545                         const u64 sz = BTRFS_I(src)->root->fs_info->sectorsize;
3546
3547                         len = round_down(i_size_read(src), sz) - loff;
3548                         if (len == 0)
3549                                 return 0;
3550                         olen = len;
3551                 }
3552         }
3553
3554 again:
3555         ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, cmp);
3556         if (ret)
3557                 return ret;
3558
3559         if (same_inode)
3560                 ret = lock_extent_range(src, same_lock_start, same_lock_len,
3561                                         false);
3562         else
3563                 ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len,
3564                                                false);
3565         /*
3566          * If one of the inodes has dirty pages in the respective range or
3567          * ordered extents, we need to flush dellaloc and wait for all ordered
3568          * extents in the range. We must unlock the pages and the ranges in the
3569          * io trees to avoid deadlocks when flushing delalloc (requires locking
3570          * pages) and when waiting for ordered extents to complete (they require
3571          * range locking).
3572          */
3573         if (ret == -EAGAIN) {
3574                 /*
3575                  * Ranges in the io trees already unlocked. Now unlock all
3576                  * pages before waiting for all IO to complete.
3577                  */
3578                 btrfs_cmp_data_free(cmp);
3579                 if (same_inode) {
3580                         btrfs_wait_ordered_range(src, same_lock_start,
3581                                                  same_lock_len);
3582                 } else {
3583                         btrfs_wait_ordered_range(src, loff, len);
3584                         btrfs_wait_ordered_range(dst, dst_loff, len);
3585                 }
3586                 goto again;
3587         }
3588         ASSERT(ret == 0);
3589         if (WARN_ON(ret)) {
3590                 /* ranges in the io trees already unlocked */
3591                 btrfs_cmp_data_free(cmp);
3592                 return ret;
3593         }
3594
3595         /* pass original length for comparison so we stay within i_size */
3596         ret = btrfs_cmp_data(olen, cmp);
3597         if (ret == 0)
3598                 ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1);
3599
3600         if (same_inode)
3601                 unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start,
3602                               same_lock_start + same_lock_len - 1);
3603         else
3604                 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3605
3606         btrfs_cmp_data_free(cmp);
3607
3608         return ret;
3609 }
3610
3611 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3612
3613 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3614                              struct inode *dst, u64 dst_loff)
3615 {
3616         int ret;
3617         struct cmp_pages cmp;
3618         int num_pages = PAGE_ALIGN(BTRFS_MAX_DEDUPE_LEN) >> PAGE_SHIFT;
3619         bool same_inode = (src == dst);
3620         u64 i, tail_len, chunk_count;
3621
3622         if (olen == 0)
3623                 return 0;
3624
3625         if (same_inode)
3626                 inode_lock(src);
3627         else
3628                 btrfs_double_inode_lock(src, dst);
3629
3630         /* don't make the dst file partly checksummed */
3631         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
3632             (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) {
3633                 ret = -EINVAL;
3634                 goto out_unlock;
3635         }
3636
3637         tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3638         chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3639         if (chunk_count == 0)
3640                 num_pages = PAGE_ALIGN(tail_len) >> PAGE_SHIFT;
3641
3642         /*
3643          * If deduping ranges in the same inode, locking rules make it
3644          * mandatory to always lock pages in ascending order to avoid deadlocks
3645          * with concurrent tasks (such as starting writeback/delalloc).
3646          */
3647         if (same_inode && dst_loff < loff)
3648                 swap(loff, dst_loff);
3649
3650         /*
3651          * We must gather up all the pages before we initiate our extent
3652          * locking. We use an array for the page pointers. Size of the array is
3653          * bounded by len, which is in turn bounded by BTRFS_MAX_DEDUPE_LEN.
3654          */
3655         cmp.src_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3656                                        GFP_KERNEL | __GFP_ZERO);
3657         cmp.dst_pages = kvmalloc_array(num_pages, sizeof(struct page *),
3658                                        GFP_KERNEL | __GFP_ZERO);
3659         if (!cmp.src_pages || !cmp.dst_pages) {
3660                 ret = -ENOMEM;
3661                 goto out_free;
3662         }
3663
3664         for (i = 0; i < chunk_count; i++) {
3665                 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3666                                               dst, dst_loff, &cmp);
3667                 if (ret)
3668                         goto out_free;
3669
3670                 loff += BTRFS_MAX_DEDUPE_LEN;
3671                 dst_loff += BTRFS_MAX_DEDUPE_LEN;
3672         }
3673
3674         if (tail_len > 0)
3675                 ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3676                                               dst_loff, &cmp);
3677
3678 out_free:
3679         kvfree(cmp.src_pages);
3680         kvfree(cmp.dst_pages);
3681
3682 out_unlock:
3683         if (same_inode)
3684                 inode_unlock(src);
3685         else
3686                 btrfs_double_inode_unlock(src, dst);
3687
3688         return ret;
3689 }
3690
3691 int btrfs_dedupe_file_range(struct file *src_file, loff_t src_loff,
3692                             struct file *dst_file, loff_t dst_loff,
3693                             u64 olen)
3694 {
3695         struct inode *src = file_inode(src_file);
3696         struct inode *dst = file_inode(dst_file);
3697         u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3698
3699         if (WARN_ON_ONCE(bs < PAGE_SIZE)) {
3700                 /*
3701                  * Btrfs does not support blocksize < page_size. As a
3702                  * result, btrfs_cmp_data() won't correctly handle
3703                  * this situation without an update.
3704                  */
3705                 return -EINVAL;
3706         }
3707
3708         return btrfs_extent_same(src, src_loff, olen, dst, dst_loff);
3709 }
3710
3711 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3712                                      struct inode *inode,
3713                                      u64 endoff,
3714                                      const u64 destoff,
3715                                      const u64 olen,
3716                                      int no_time_update)
3717 {
3718         struct btrfs_root *root = BTRFS_I(inode)->root;
3719         int ret;
3720
3721         inode_inc_iversion(inode);
3722         if (!no_time_update)
3723                 inode->i_mtime = inode->i_ctime = current_time(inode);
3724         /*
3725          * We round up to the block size at eof when determining which
3726          * extents to clone above, but shouldn't round up the file size.
3727          */
3728         if (endoff > destoff + olen)
3729                 endoff = destoff + olen;
3730         if (endoff > inode->i_size)
3731                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3732
3733         ret = btrfs_update_inode(trans, root, inode);
3734         if (ret) {
3735                 btrfs_abort_transaction(trans, ret);
3736                 btrfs_end_transaction(trans);
3737                 goto out;
3738         }
3739         ret = btrfs_end_transaction(trans);
3740 out:
3741         return ret;
3742 }
3743
3744 static void clone_update_extent_map(struct btrfs_inode *inode,
3745                                     const struct btrfs_trans_handle *trans,
3746                                     const struct btrfs_path *path,
3747                                     const u64 hole_offset,
3748                                     const u64 hole_len)
3749 {
3750         struct extent_map_tree *em_tree = &inode->extent_tree;
3751         struct extent_map *em;
3752         int ret;
3753
3754         em = alloc_extent_map();
3755         if (!em) {
3756                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3757                 return;
3758         }
3759
3760         if (path) {
3761                 struct btrfs_file_extent_item *fi;
3762
3763                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3764                                     struct btrfs_file_extent_item);
3765                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3766                 em->generation = -1;
3767                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3768                     BTRFS_FILE_EXTENT_INLINE)
3769                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3770                                         &inode->runtime_flags);
3771         } else {
3772                 em->start = hole_offset;
3773                 em->len = hole_len;
3774                 em->ram_bytes = em->len;
3775                 em->orig_start = hole_offset;
3776                 em->block_start = EXTENT_MAP_HOLE;
3777                 em->block_len = 0;
3778                 em->orig_block_len = 0;
3779                 em->compress_type = BTRFS_COMPRESS_NONE;
3780                 em->generation = trans->transid;
3781         }
3782
3783         while (1) {
3784                 write_lock(&em_tree->lock);
3785                 ret = add_extent_mapping(em_tree, em, 1);
3786                 write_unlock(&em_tree->lock);
3787                 if (ret != -EEXIST) {
3788                         free_extent_map(em);
3789                         break;
3790                 }
3791                 btrfs_drop_extent_cache(inode, em->start,
3792                                         em->start + em->len - 1, 0);
3793         }
3794
3795         if (ret)
3796                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3797 }
3798
3799 /*
3800  * Make sure we do not end up inserting an inline extent into a file that has
3801  * already other (non-inline) extents. If a file has an inline extent it can
3802  * not have any other extents and the (single) inline extent must start at the
3803  * file offset 0. Failing to respect these rules will lead to file corruption,
3804  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3805  *
3806  * We can have extents that have been already written to disk or we can have
3807  * dirty ranges still in delalloc, in which case the extent maps and items are
3808  * created only when we run delalloc, and the delalloc ranges might fall outside
3809  * the range we are currently locking in the inode's io tree. So we check the
3810  * inode's i_size because of that (i_size updates are done while holding the
3811  * i_mutex, which we are holding here).
3812  * We also check to see if the inode has a size not greater than "datal" but has
3813  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3814  * protected against such concurrent fallocate calls by the i_mutex).
3815  *
3816  * If the file has no extents but a size greater than datal, do not allow the
3817  * copy because we would need turn the inline extent into a non-inline one (even
3818  * with NO_HOLES enabled). If we find our destination inode only has one inline
3819  * extent, just overwrite it with the source inline extent if its size is less
3820  * than the source extent's size, or we could copy the source inline extent's
3821  * data into the destination inode's inline extent if the later is greater then
3822  * the former.
3823  */
3824 static int clone_copy_inline_extent(struct inode *dst,
3825                                     struct btrfs_trans_handle *trans,
3826                                     struct btrfs_path *path,
3827                                     struct btrfs_key *new_key,
3828                                     const u64 drop_start,
3829                                     const u64 datal,
3830                                     const u64 skip,
3831                                     const u64 size,
3832                                     char *inline_data)
3833 {
3834         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3835         struct btrfs_root *root = BTRFS_I(dst)->root;
3836         const u64 aligned_end = ALIGN(new_key->offset + datal,
3837                                       fs_info->sectorsize);
3838         int ret;
3839         struct btrfs_key key;
3840
3841         if (new_key->offset > 0)
3842                 return -EOPNOTSUPP;
3843
3844         key.objectid = btrfs_ino(BTRFS_I(dst));
3845         key.type = BTRFS_EXTENT_DATA_KEY;
3846         key.offset = 0;
3847         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3848         if (ret < 0) {
3849                 return ret;
3850         } else if (ret > 0) {
3851                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3852                         ret = btrfs_next_leaf(root, path);
3853                         if (ret < 0)
3854                                 return ret;
3855                         else if (ret > 0)
3856                                 goto copy_inline_extent;
3857                 }
3858                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3859                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3860                     key.type == BTRFS_EXTENT_DATA_KEY) {
3861                         ASSERT(key.offset > 0);
3862                         return -EOPNOTSUPP;
3863                 }
3864         } else if (i_size_read(dst) <= datal) {
3865                 struct btrfs_file_extent_item *ei;
3866                 u64 ext_len;
3867
3868                 /*
3869                  * If the file size is <= datal, make sure there are no other
3870                  * extents following (can happen do to an fallocate call with
3871                  * the flag FALLOC_FL_KEEP_SIZE).
3872                  */
3873                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3874                                     struct btrfs_file_extent_item);
3875                 /*
3876                  * If it's an inline extent, it can not have other extents
3877                  * following it.
3878                  */
3879                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3880                     BTRFS_FILE_EXTENT_INLINE)
3881                         goto copy_inline_extent;
3882
3883                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3884                 if (ext_len > aligned_end)
3885                         return -EOPNOTSUPP;
3886
3887                 ret = btrfs_next_item(root, path);
3888                 if (ret < 0) {
3889                         return ret;
3890                 } else if (ret == 0) {
3891                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3892                                               path->slots[0]);
3893                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3894                             key.type == BTRFS_EXTENT_DATA_KEY)
3895                                 return -EOPNOTSUPP;
3896                 }
3897         }
3898
3899 copy_inline_extent:
3900         /*
3901          * We have no extent items, or we have an extent at offset 0 which may
3902          * or may not be inlined. All these cases are dealt the same way.
3903          */
3904         if (i_size_read(dst) > datal) {
3905                 /*
3906                  * If the destination inode has an inline extent...
3907                  * This would require copying the data from the source inline
3908                  * extent into the beginning of the destination's inline extent.
3909                  * But this is really complex, both extents can be compressed
3910                  * or just one of them, which would require decompressing and
3911                  * re-compressing data (which could increase the new compressed
3912                  * size, not allowing the compressed data to fit anymore in an
3913                  * inline extent).
3914                  * So just don't support this case for now (it should be rare,
3915                  * we are not really saving space when cloning inline extents).
3916                  */
3917                 return -EOPNOTSUPP;
3918         }
3919
3920         btrfs_release_path(path);
3921         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3922         if (ret)
3923                 return ret;
3924         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3925         if (ret)
3926                 return ret;
3927
3928         if (skip) {
3929                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3930
3931                 memmove(inline_data + start, inline_data + start + skip, datal);
3932         }
3933
3934         write_extent_buffer(path->nodes[0], inline_data,
3935                             btrfs_item_ptr_offset(path->nodes[0],
3936                                                   path->slots[0]),
3937                             size);
3938         inode_add_bytes(dst, datal);
3939
3940         return 0;
3941 }
3942
3943 /**
3944  * btrfs_clone() - clone a range from inode file to another
3945  *
3946  * @src: Inode to clone from
3947  * @inode: Inode to clone to
3948  * @off: Offset within source to start clone from
3949  * @olen: Original length, passed by user, of range to clone
3950  * @olen_aligned: Block-aligned value of olen
3951  * @destoff: Offset within @inode to start clone
3952  * @no_time_update: Whether to update mtime/ctime on the target inode
3953  */
3954 static int btrfs_clone(struct inode *src, struct inode *inode,
3955                        const u64 off, const u64 olen, const u64 olen_aligned,
3956                        const u64 destoff, int no_time_update)
3957 {
3958         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3959         struct btrfs_root *root = BTRFS_I(inode)->root;
3960         struct btrfs_path *path = NULL;
3961         struct extent_buffer *leaf;
3962         struct btrfs_trans_handle *trans;
3963         char *buf = NULL;
3964         struct btrfs_key key;
3965         u32 nritems;
3966         int slot;
3967         int ret;
3968         const u64 len = olen_aligned;
3969         u64 last_dest_end = destoff;
3970
3971         ret = -ENOMEM;
3972         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3973         if (!buf)
3974                 return ret;
3975
3976         path = btrfs_alloc_path();
3977         if (!path) {
3978                 kvfree(buf);
3979                 return ret;
3980         }
3981
3982         path->reada = READA_FORWARD;
3983         /* clone data */
3984         key.objectid = btrfs_ino(BTRFS_I(src));
3985         key.type = BTRFS_EXTENT_DATA_KEY;
3986         key.offset = off;
3987
3988         while (1) {
3989                 u64 next_key_min_offset = key.offset + 1;
3990
3991                 /*
3992                  * note the key will change type as we walk through the
3993                  * tree.
3994                  */
3995                 path->leave_spinning = 1;
3996                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3997                                 0, 0);
3998                 if (ret < 0)
3999                         goto out;
4000                 /*
4001                  * First search, if no extent item that starts at offset off was
4002                  * found but the previous item is an extent item, it's possible
4003                  * it might overlap our target range, therefore process it.
4004                  */
4005                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
4006                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4007                                               path->slots[0] - 1);
4008                         if (key.type == BTRFS_EXTENT_DATA_KEY)
4009                                 path->slots[0]--;
4010                 }
4011
4012                 nritems = btrfs_header_nritems(path->nodes[0]);
4013 process_slot:
4014                 if (path->slots[0] >= nritems) {
4015                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
4016                         if (ret < 0)
4017                                 goto out;
4018                         if (ret > 0)
4019                                 break;
4020                         nritems = btrfs_header_nritems(path->nodes[0]);
4021                 }
4022                 leaf = path->nodes[0];
4023                 slot = path->slots[0];
4024
4025                 btrfs_item_key_to_cpu(leaf, &key, slot);
4026                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
4027                     key.objectid != btrfs_ino(BTRFS_I(src)))
4028                         break;
4029
4030                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4031                         struct btrfs_file_extent_item *extent;
4032                         int type;
4033                         u32 size;
4034                         struct btrfs_key new_key;
4035                         u64 disko = 0, diskl = 0;
4036                         u64 datao = 0, datal = 0;
4037                         u8 comp;
4038                         u64 drop_start;
4039
4040                         extent = btrfs_item_ptr(leaf, slot,
4041                                                 struct btrfs_file_extent_item);
4042                         comp = btrfs_file_extent_compression(leaf, extent);
4043                         type = btrfs_file_extent_type(leaf, extent);
4044                         if (type == BTRFS_FILE_EXTENT_REG ||
4045                             type == BTRFS_FILE_EXTENT_PREALLOC) {
4046                                 disko = btrfs_file_extent_disk_bytenr(leaf,
4047                                                                       extent);
4048                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
4049                                                                  extent);
4050                                 datao = btrfs_file_extent_offset(leaf, extent);
4051                                 datal = btrfs_file_extent_num_bytes(leaf,
4052                                                                     extent);
4053                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
4054                                 /* take upper bound, may be compressed */
4055                                 datal = btrfs_file_extent_ram_bytes(leaf,
4056                                                                     extent);
4057                         }
4058
4059                         /*
4060                          * The first search might have left us at an extent
4061                          * item that ends before our target range's start, can
4062                          * happen if we have holes and NO_HOLES feature enabled.
4063                          */
4064                         if (key.offset + datal <= off) {
4065                                 path->slots[0]++;
4066                                 goto process_slot;
4067                         } else if (key.offset >= off + len) {
4068                                 break;
4069                         }
4070                         next_key_min_offset = key.offset + datal;
4071                         size = btrfs_item_size_nr(leaf, slot);
4072                         read_extent_buffer(leaf, buf,
4073                                            btrfs_item_ptr_offset(leaf, slot),
4074                                            size);
4075
4076                         btrfs_release_path(path);
4077                         path->leave_spinning = 0;
4078
4079                         memcpy(&new_key, &key, sizeof(new_key));
4080                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
4081                         if (off <= key.offset)
4082                                 new_key.offset = key.offset + destoff - off;
4083                         else
4084                                 new_key.offset = destoff;
4085
4086                         /*
4087                          * Deal with a hole that doesn't have an extent item
4088                          * that represents it (NO_HOLES feature enabled).
4089                          * This hole is either in the middle of the cloning
4090                          * range or at the beginning (fully overlaps it or
4091                          * partially overlaps it).
4092                          */
4093                         if (new_key.offset != last_dest_end)
4094                                 drop_start = last_dest_end;
4095                         else
4096                                 drop_start = new_key.offset;
4097
4098                         /*
4099                          * 1 - adjusting old extent (we may have to split it)
4100                          * 1 - add new extent
4101                          * 1 - inode update
4102                          */
4103                         trans = btrfs_start_transaction(root, 3);
4104                         if (IS_ERR(trans)) {
4105                                 ret = PTR_ERR(trans);
4106                                 goto out;
4107                         }
4108
4109                         if (type == BTRFS_FILE_EXTENT_REG ||
4110                             type == BTRFS_FILE_EXTENT_PREALLOC) {
4111                                 /*
4112                                  *    a  | --- range to clone ---|  b
4113                                  * | ------------- extent ------------- |
4114                                  */
4115
4116                                 /* subtract range b */
4117                                 if (key.offset + datal > off + len)
4118                                         datal = off + len - key.offset;
4119
4120                                 /* subtract range a */
4121                                 if (off > key.offset) {
4122                                         datao += off - key.offset;
4123                                         datal -= off - key.offset;
4124                                 }
4125
4126                                 ret = btrfs_drop_extents(trans, root, inode,
4127                                                          drop_start,
4128                                                          new_key.offset + datal,
4129                                                          1);
4130                                 if (ret) {
4131                                         if (ret != -EOPNOTSUPP)
4132                                                 btrfs_abort_transaction(trans,
4133                                                                         ret);
4134                                         btrfs_end_transaction(trans);
4135                                         goto out;
4136                                 }
4137
4138                                 ret = btrfs_insert_empty_item(trans, root, path,
4139                                                               &new_key, size);
4140                                 if (ret) {
4141                                         btrfs_abort_transaction(trans, ret);
4142                                         btrfs_end_transaction(trans);
4143                                         goto out;
4144                                 }
4145
4146                                 leaf = path->nodes[0];
4147                                 slot = path->slots[0];
4148                                 write_extent_buffer(leaf, buf,
4149                                             btrfs_item_ptr_offset(leaf, slot),
4150                                             size);
4151
4152                                 extent = btrfs_item_ptr(leaf, slot,
4153                                                 struct btrfs_file_extent_item);
4154
4155                                 /* disko == 0 means it's a hole */
4156                                 if (!disko)
4157                                         datao = 0;
4158
4159                                 btrfs_set_file_extent_offset(leaf, extent,
4160                                                              datao);
4161                                 btrfs_set_file_extent_num_bytes(leaf, extent,
4162                                                                 datal);
4163
4164                                 if (disko) {
4165                                         inode_add_bytes(inode, datal);
4166                                         ret = btrfs_inc_extent_ref(trans,
4167                                                         root,
4168                                                         disko, diskl, 0,
4169                                                         root->root_key.objectid,
4170                                                         btrfs_ino(BTRFS_I(inode)),
4171                                                         new_key.offset - datao);
4172                                         if (ret) {
4173                                                 btrfs_abort_transaction(trans,
4174                                                                         ret);
4175                                                 btrfs_end_transaction(trans);
4176                                                 goto out;
4177
4178                                         }
4179                                 }
4180                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
4181                                 u64 skip = 0;
4182                                 u64 trim = 0;
4183
4184                                 if (off > key.offset) {
4185                                         skip = off - key.offset;
4186                                         new_key.offset += skip;
4187                                 }
4188
4189                                 if (key.offset + datal > off + len)
4190                                         trim = key.offset + datal - (off + len);
4191
4192                                 if (comp && (skip || trim)) {
4193                                         ret = -EINVAL;
4194                                         btrfs_end_transaction(trans);
4195                                         goto out;
4196                                 }
4197                                 size -= skip + trim;
4198                                 datal -= skip + trim;
4199
4200                                 ret = clone_copy_inline_extent(inode,
4201                                                                trans, path,
4202                                                                &new_key,
4203                                                                drop_start,
4204                                                                datal,
4205                                                                skip, size, buf);
4206                                 if (ret) {
4207                                         if (ret != -EOPNOTSUPP)
4208                                                 btrfs_abort_transaction(trans,
4209                                                                         ret);
4210                                         btrfs_end_transaction(trans);
4211                                         goto out;
4212                                 }
4213                                 leaf = path->nodes[0];
4214                                 slot = path->slots[0];
4215                         }
4216
4217                         /* If we have an implicit hole (NO_HOLES feature). */
4218                         if (drop_start < new_key.offset)
4219                                 clone_update_extent_map(BTRFS_I(inode), trans,
4220                                                 NULL, drop_start,
4221                                                 new_key.offset - drop_start);
4222
4223                         clone_update_extent_map(BTRFS_I(inode), trans,
4224                                         path, 0, 0);
4225
4226                         btrfs_mark_buffer_dirty(leaf);
4227                         btrfs_release_path(path);
4228
4229                         last_dest_end = ALIGN(new_key.offset + datal,
4230                                               fs_info->sectorsize);
4231                         ret = clone_finish_inode_update(trans, inode,
4232                                                         last_dest_end,
4233                                                         destoff, olen,
4234                                                         no_time_update);
4235                         if (ret)
4236                                 goto out;
4237                         if (new_key.offset + datal >= destoff + len)
4238                                 break;
4239                 }
4240                 btrfs_release_path(path);
4241                 key.offset = next_key_min_offset;
4242
4243                 if (fatal_signal_pending(current)) {
4244                         ret = -EINTR;
4245                         goto out;
4246                 }
4247
4248                 cond_resched();
4249         }
4250         ret = 0;
4251
4252         if (last_dest_end < destoff + len) {
4253                 /*
4254                  * We have an implicit hole (NO_HOLES feature is enabled) that
4255                  * fully or partially overlaps our cloning range at its end.
4256                  */
4257                 btrfs_release_path(path);
4258
4259                 /*
4260                  * 1 - remove extent(s)
4261                  * 1 - inode update
4262                  */
4263                 trans = btrfs_start_transaction(root, 2);
4264                 if (IS_ERR(trans)) {
4265                         ret = PTR_ERR(trans);
4266                         goto out;
4267                 }
4268                 ret = btrfs_drop_extents(trans, root, inode,
4269                                          last_dest_end, destoff + len, 1);
4270                 if (ret) {
4271                         if (ret != -EOPNOTSUPP)
4272                                 btrfs_abort_transaction(trans, ret);
4273                         btrfs_end_transaction(trans);
4274                         goto out;
4275                 }
4276                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
4277                                 last_dest_end,
4278                                 destoff + len - last_dest_end);
4279                 ret = clone_finish_inode_update(trans, inode, destoff + len,
4280                                                 destoff, olen, no_time_update);
4281         }
4282
4283 out:
4284         btrfs_free_path(path);
4285         kvfree(buf);
4286         return ret;
4287 }
4288
4289 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
4290                                         u64 off, u64 olen, u64 destoff)
4291 {
4292         struct inode *inode = file_inode(file);
4293         struct inode *src = file_inode(file_src);
4294         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4295         struct btrfs_root *root = BTRFS_I(inode)->root;
4296         int ret;
4297         u64 len = olen;
4298         u64 bs = fs_info->sb->s_blocksize;
4299         int same_inode = src == inode;
4300
4301         /*
4302          * TODO:
4303          * - split compressed inline extents.  annoying: we need to
4304          *   decompress into destination's address_space (the file offset
4305          *   may change, so source mapping won't do), then recompress (or
4306          *   otherwise reinsert) a subrange.
4307          *
4308          * - split destination inode's inline extents.  The inline extents can
4309          *   be either compressed or non-compressed.
4310          */
4311
4312         if (btrfs_root_readonly(root))
4313                 return -EROFS;
4314
4315         if (file_src->f_path.mnt != file->f_path.mnt ||
4316             src->i_sb != inode->i_sb)
4317                 return -EXDEV;
4318
4319         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
4320                 return -EISDIR;
4321
4322         if (!same_inode) {
4323                 btrfs_double_inode_lock(src, inode);
4324         } else {
4325                 inode_lock(src);
4326         }
4327
4328         /* don't make the dst file partly checksummed */
4329         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
4330             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
4331                 ret = -EINVAL;
4332                 goto out_unlock;
4333         }
4334
4335         /* determine range to clone */
4336         ret = -EINVAL;
4337         if (off + len > src->i_size || off + len < off)
4338                 goto out_unlock;
4339         if (len == 0)
4340                 olen = len = src->i_size - off;
4341         /*
4342          * If we extend to eof, continue to block boundary if and only if the
4343          * destination end offset matches the destination file's size, otherwise
4344          * we would be corrupting data by placing the eof block into the middle
4345          * of a file.
4346          */
4347         if (off + len == src->i_size) {
4348                 if (!IS_ALIGNED(len, bs) && destoff + len < inode->i_size)
4349                         goto out_unlock;
4350                 len = ALIGN(src->i_size, bs) - off;
4351         }
4352
4353         if (len == 0) {
4354                 ret = 0;
4355                 goto out_unlock;
4356         }
4357
4358         /* verify the end result is block aligned */
4359         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
4360             !IS_ALIGNED(destoff, bs))
4361                 goto out_unlock;
4362
4363         /* verify if ranges are overlapped within the same file */
4364         if (same_inode) {
4365                 if (destoff + len > off && destoff < off + len)
4366                         goto out_unlock;
4367         }
4368
4369         if (destoff > inode->i_size) {
4370                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
4371                 if (ret)
4372                         goto out_unlock;
4373         }
4374
4375         /*
4376          * Lock the target range too. Right after we replace the file extent
4377          * items in the fs tree (which now point to the cloned data), we might
4378          * have a worker replace them with extent items relative to a write
4379          * operation that was issued before this clone operation (i.e. confront
4380          * with inode.c:btrfs_finish_ordered_io).
4381          */
4382         if (same_inode) {
4383                 u64 lock_start = min_t(u64, off, destoff);
4384                 u64 lock_len = max_t(u64, off, destoff) + len - lock_start;
4385
4386                 ret = lock_extent_range(src, lock_start, lock_len, true);
4387         } else {
4388                 ret = btrfs_double_extent_lock(src, off, inode, destoff, len,
4389                                                true);
4390         }
4391         ASSERT(ret == 0);
4392         if (WARN_ON(ret)) {
4393                 /* ranges in the io trees already unlocked */
4394                 goto out_unlock;
4395         }
4396
4397         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
4398
4399         if (same_inode) {
4400                 u64 lock_start = min_t(u64, off, destoff);
4401                 u64 lock_end = max_t(u64, off, destoff) + len - 1;
4402
4403                 unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end);
4404         } else {
4405                 btrfs_double_extent_unlock(src, off, inode, destoff, len);
4406         }
4407         /*
4408          * Truncate page cache pages so that future reads will see the cloned
4409          * data immediately and not the previous data.
4410          */
4411         truncate_inode_pages_range(&inode->i_data,
4412                                 round_down(destoff, PAGE_SIZE),
4413                                 round_up(destoff + len, PAGE_SIZE) - 1);
4414 out_unlock:
4415         if (!same_inode)
4416                 btrfs_double_inode_unlock(src, inode);
4417         else
4418                 inode_unlock(src);
4419         return ret;
4420 }
4421
4422 int btrfs_clone_file_range(struct file *src_file, loff_t off,
4423                 struct file *dst_file, loff_t destoff, u64 len)
4424 {
4425         return btrfs_clone_files(dst_file, src_file, off, len, destoff);
4426 }
4427
4428 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4429 {
4430         struct inode *inode = file_inode(file);
4431         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4432         struct btrfs_root *root = BTRFS_I(inode)->root;
4433         struct btrfs_root *new_root;
4434         struct btrfs_dir_item *di;
4435         struct btrfs_trans_handle *trans;
4436         struct btrfs_path *path;
4437         struct btrfs_key location;
4438         struct btrfs_disk_key disk_key;
4439         u64 objectid = 0;
4440         u64 dir_id;
4441         int ret;
4442
4443         if (!capable(CAP_SYS_ADMIN))
4444                 return -EPERM;
4445
4446         ret = mnt_want_write_file(file);
4447         if (ret)
4448                 return ret;
4449
4450         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4451                 ret = -EFAULT;
4452                 goto out;
4453         }
4454
4455         if (!objectid)
4456                 objectid = BTRFS_FS_TREE_OBJECTID;
4457
4458         location.objectid = objectid;
4459         location.type = BTRFS_ROOT_ITEM_KEY;
4460         location.offset = (u64)-1;
4461
4462         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4463         if (IS_ERR(new_root)) {
4464                 ret = PTR_ERR(new_root);
4465                 goto out;
4466         }
4467         if (!is_fstree(new_root->objectid)) {
4468                 ret = -ENOENT;
4469                 goto out;
4470         }
4471
4472         path = btrfs_alloc_path();
4473         if (!path) {
4474                 ret = -ENOMEM;
4475                 goto out;
4476         }
4477         path->leave_spinning = 1;
4478
4479         trans = btrfs_start_transaction(root, 1);
4480         if (IS_ERR(trans)) {
4481                 btrfs_free_path(path);
4482                 ret = PTR_ERR(trans);
4483                 goto out;
4484         }
4485
4486         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4487         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4488                                    dir_id, "default", 7, 1);
4489         if (IS_ERR_OR_NULL(di)) {
4490                 btrfs_free_path(path);
4491                 btrfs_end_transaction(trans);
4492                 btrfs_err(fs_info,
4493                           "Umm, you don't have the default diritem, this isn't going to work");
4494                 ret = -ENOENT;
4495                 goto out;
4496         }
4497
4498         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4499         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4500         btrfs_mark_buffer_dirty(path->nodes[0]);
4501         btrfs_free_path(path);
4502
4503         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4504         btrfs_end_transaction(trans);
4505 out:
4506         mnt_drop_write_file(file);
4507         return ret;
4508 }
4509
4510 static void get_block_group_info(struct list_head *groups_list,
4511                                  struct btrfs_ioctl_space_info *space)
4512 {
4513         struct btrfs_block_group_cache *block_group;
4514
4515         space->total_bytes = 0;
4516         space->used_bytes = 0;
4517         space->flags = 0;
4518         list_for_each_entry(block_group, groups_list, list) {
4519                 space->flags = block_group->flags;
4520                 space->total_bytes += block_group->key.offset;
4521                 space->used_bytes +=
4522                         btrfs_block_group_used(&block_group->item);
4523         }
4524 }
4525
4526 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4527                                    void __user *arg)
4528 {
4529         struct btrfs_ioctl_space_args space_args;
4530         struct btrfs_ioctl_space_info space;
4531         struct btrfs_ioctl_space_info *dest;
4532         struct btrfs_ioctl_space_info *dest_orig;
4533         struct btrfs_ioctl_space_info __user *user_dest;
4534         struct btrfs_space_info *info;
4535         static const u64 types[] = {
4536                 BTRFS_BLOCK_GROUP_DATA,
4537                 BTRFS_BLOCK_GROUP_SYSTEM,
4538                 BTRFS_BLOCK_GROUP_METADATA,
4539                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4540         };
4541         int num_types = 4;
4542         int alloc_size;
4543         int ret = 0;
4544         u64 slot_count = 0;
4545         int i, c;
4546
4547         if (copy_from_user(&space_args,
4548                            (struct btrfs_ioctl_space_args __user *)arg,
4549                            sizeof(space_args)))
4550                 return -EFAULT;
4551
4552         for (i = 0; i < num_types; i++) {
4553                 struct btrfs_space_info *tmp;
4554
4555                 info = NULL;
4556                 rcu_read_lock();
4557                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4558                                         list) {
4559                         if (tmp->flags == types[i]) {
4560                                 info = tmp;
4561                                 break;
4562                         }
4563                 }
4564                 rcu_read_unlock();
4565
4566                 if (!info)
4567                         continue;
4568
4569                 down_read(&info->groups_sem);
4570                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4571                         if (!list_empty(&info->block_groups[c]))
4572                                 slot_count++;
4573                 }
4574                 up_read(&info->groups_sem);
4575         }
4576
4577         /*
4578          * Global block reserve, exported as a space_info
4579          */
4580         slot_count++;
4581
4582         /* space_slots == 0 means they are asking for a count */
4583         if (space_args.space_slots == 0) {
4584                 space_args.total_spaces = slot_count;
4585                 goto out;
4586         }
4587
4588         slot_count = min_t(u64, space_args.space_slots, slot_count);
4589
4590         alloc_size = sizeof(*dest) * slot_count;
4591
4592         /* we generally have at most 6 or so space infos, one for each raid
4593          * level.  So, a whole page should be more than enough for everyone
4594          */
4595         if (alloc_size > PAGE_SIZE)
4596                 return -ENOMEM;
4597
4598         space_args.total_spaces = 0;
4599         dest = kmalloc(alloc_size, GFP_KERNEL);
4600         if (!dest)
4601                 return -ENOMEM;
4602         dest_orig = dest;
4603
4604         /* now we have a buffer to copy into */
4605         for (i = 0; i < num_types; i++) {
4606                 struct btrfs_space_info *tmp;
4607
4608                 if (!slot_count)
4609                         break;
4610
4611                 info = NULL;
4612                 rcu_read_lock();
4613                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4614                                         list) {
4615                         if (tmp->flags == types[i]) {
4616                                 info = tmp;
4617                                 break;
4618                         }
4619                 }
4620                 rcu_read_unlock();
4621
4622                 if (!info)
4623                         continue;
4624                 down_read(&info->groups_sem);
4625                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4626                         if (!list_empty(&info->block_groups[c])) {
4627                                 get_block_group_info(&info->block_groups[c],
4628                                                      &space);
4629                                 memcpy(dest, &space, sizeof(space));
4630                                 dest++;
4631                                 space_args.total_spaces++;
4632                                 slot_count--;
4633                         }
4634                         if (!slot_count)
4635                                 break;
4636                 }
4637                 up_read(&info->groups_sem);
4638         }
4639
4640         /*
4641          * Add global block reserve
4642          */
4643         if (slot_count) {
4644                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4645
4646                 spin_lock(&block_rsv->lock);
4647                 space.total_bytes = block_rsv->size;
4648                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4649                 spin_unlock(&block_rsv->lock);
4650                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4651                 memcpy(dest, &space, sizeof(space));
4652                 space_args.total_spaces++;
4653         }
4654
4655         user_dest = (struct btrfs_ioctl_space_info __user *)
4656                 (arg + sizeof(struct btrfs_ioctl_space_args));
4657
4658         if (copy_to_user(user_dest, dest_orig, alloc_size))
4659                 ret = -EFAULT;
4660
4661         kfree(dest_orig);
4662 out:
4663         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4664                 ret = -EFAULT;
4665
4666         return ret;
4667 }
4668
4669 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4670                                             void __user *argp)
4671 {
4672         struct btrfs_trans_handle *trans;
4673         u64 transid;
4674         int ret;
4675
4676         trans = btrfs_attach_transaction_barrier(root);
4677         if (IS_ERR(trans)) {
4678                 if (PTR_ERR(trans) != -ENOENT)
4679                         return PTR_ERR(trans);
4680
4681                 /* No running transaction, don't bother */
4682                 transid = root->fs_info->last_trans_committed;
4683                 goto out;
4684         }
4685         transid = trans->transid;
4686         ret = btrfs_commit_transaction_async(trans, 0);
4687         if (ret) {
4688                 btrfs_end_transaction(trans);
4689                 return ret;
4690         }
4691 out:
4692         if (argp)
4693                 if (copy_to_user(argp, &transid, sizeof(transid)))
4694                         return -EFAULT;
4695         return 0;
4696 }
4697
4698 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4699                                            void __user *argp)
4700 {
4701         u64 transid;
4702
4703         if (argp) {
4704                 if (copy_from_user(&transid, argp, sizeof(transid)))
4705                         return -EFAULT;
4706         } else {
4707                 transid = 0;  /* current trans */
4708         }
4709         return btrfs_wait_for_commit(fs_info, transid);
4710 }
4711
4712 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4713 {
4714         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4715         struct btrfs_ioctl_scrub_args *sa;
4716         int ret;
4717
4718         if (!capable(CAP_SYS_ADMIN))
4719                 return -EPERM;
4720
4721         sa = memdup_user(arg, sizeof(*sa));
4722         if (IS_ERR(sa))
4723                 return PTR_ERR(sa);
4724
4725         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4726                 ret = mnt_want_write_file(file);
4727                 if (ret)
4728                         goto out;
4729         }
4730
4731         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4732                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4733                               0);
4734
4735         if (copy_to_user(arg, sa, sizeof(*sa)))
4736                 ret = -EFAULT;
4737
4738         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4739                 mnt_drop_write_file(file);
4740 out:
4741         kfree(sa);
4742         return ret;
4743 }
4744
4745 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4746 {
4747         if (!capable(CAP_SYS_ADMIN))
4748                 return -EPERM;
4749
4750         return btrfs_scrub_cancel(fs_info);
4751 }
4752
4753 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4754                                        void __user *arg)
4755 {
4756         struct btrfs_ioctl_scrub_args *sa;
4757         int ret;
4758
4759         if (!capable(CAP_SYS_ADMIN))
4760                 return -EPERM;
4761
4762         sa = memdup_user(arg, sizeof(*sa));
4763         if (IS_ERR(sa))
4764                 return PTR_ERR(sa);
4765
4766         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4767
4768         if (copy_to_user(arg, sa, sizeof(*sa)))
4769                 ret = -EFAULT;
4770
4771         kfree(sa);
4772         return ret;
4773 }
4774
4775 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4776                                       void __user *arg)
4777 {
4778         struct btrfs_ioctl_get_dev_stats *sa;
4779         int ret;
4780
4781         sa = memdup_user(arg, sizeof(*sa));
4782         if (IS_ERR(sa))
4783                 return PTR_ERR(sa);
4784
4785         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4786                 kfree(sa);
4787                 return -EPERM;
4788         }
4789
4790         ret = btrfs_get_dev_stats(fs_info, sa);
4791
4792         if (copy_to_user(arg, sa, sizeof(*sa)))
4793                 ret = -EFAULT;
4794
4795         kfree(sa);
4796         return ret;
4797 }
4798
4799 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4800                                     void __user *arg)
4801 {
4802         struct btrfs_ioctl_dev_replace_args *p;
4803         int ret;
4804
4805         if (!capable(CAP_SYS_ADMIN))
4806                 return -EPERM;
4807
4808         p = memdup_user(arg, sizeof(*p));
4809         if (IS_ERR(p))
4810                 return PTR_ERR(p);
4811
4812         switch (p->cmd) {
4813         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4814                 if (sb_rdonly(fs_info->sb)) {
4815                         ret = -EROFS;
4816                         goto out;
4817                 }
4818                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4819                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4820                 } else {
4821                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4822                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4823                 }
4824                 break;
4825         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4826                 btrfs_dev_replace_status(fs_info, p);
4827                 ret = 0;
4828                 break;
4829         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4830                 p->result = btrfs_dev_replace_cancel(fs_info);
4831                 ret = 0;
4832                 break;
4833         default:
4834                 ret = -EINVAL;
4835                 break;
4836         }
4837
4838         if (copy_to_user(arg, p, sizeof(*p)))
4839                 ret = -EFAULT;
4840 out:
4841         kfree(p);
4842         return ret;
4843 }
4844
4845 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4846 {
4847         int ret = 0;
4848         int i;
4849         u64 rel_ptr;
4850         int size;
4851         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4852         struct inode_fs_paths *ipath = NULL;
4853         struct btrfs_path *path;
4854
4855         if (!capable(CAP_DAC_READ_SEARCH))
4856                 return -EPERM;
4857
4858         path = btrfs_alloc_path();
4859         if (!path) {
4860                 ret = -ENOMEM;
4861                 goto out;
4862         }
4863
4864         ipa = memdup_user(arg, sizeof(*ipa));
4865         if (IS_ERR(ipa)) {
4866                 ret = PTR_ERR(ipa);
4867                 ipa = NULL;
4868                 goto out;
4869         }
4870
4871         size = min_t(u32, ipa->size, 4096);
4872         ipath = init_ipath(size, root, path);
4873         if (IS_ERR(ipath)) {
4874                 ret = PTR_ERR(ipath);
4875                 ipath = NULL;
4876                 goto out;
4877         }
4878
4879         ret = paths_from_inode(ipa->inum, ipath);
4880         if (ret < 0)
4881                 goto out;
4882
4883         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4884                 rel_ptr = ipath->fspath->val[i] -
4885                           (u64)(unsigned long)ipath->fspath->val;
4886                 ipath->fspath->val[i] = rel_ptr;
4887         }
4888
4889         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4890                            ipath->fspath, size);
4891         if (ret) {
4892                 ret = -EFAULT;
4893                 goto out;
4894         }
4895
4896 out:
4897         btrfs_free_path(path);
4898         free_ipath(ipath);
4899         kfree(ipa);
4900
4901         return ret;
4902 }
4903
4904 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4905 {
4906         struct btrfs_data_container *inodes = ctx;
4907         const size_t c = 3 * sizeof(u64);
4908
4909         if (inodes->bytes_left >= c) {
4910                 inodes->bytes_left -= c;
4911                 inodes->val[inodes->elem_cnt] = inum;
4912                 inodes->val[inodes->elem_cnt + 1] = offset;
4913                 inodes->val[inodes->elem_cnt + 2] = root;
4914                 inodes->elem_cnt += 3;
4915         } else {
4916                 inodes->bytes_missing += c - inodes->bytes_left;
4917                 inodes->bytes_left = 0;
4918                 inodes->elem_missed += 3;
4919         }
4920
4921         return 0;
4922 }
4923
4924 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4925                                         void __user *arg, int version)
4926 {
4927         int ret = 0;
4928         int size;
4929         struct btrfs_ioctl_logical_ino_args *loi;
4930         struct btrfs_data_container *inodes = NULL;
4931         struct btrfs_path *path = NULL;
4932         bool ignore_offset;
4933
4934         if (!capable(CAP_SYS_ADMIN))
4935                 return -EPERM;
4936
4937         loi = memdup_user(arg, sizeof(*loi));
4938         if (IS_ERR(loi))
4939                 return PTR_ERR(loi);
4940
4941         if (version == 1) {
4942                 ignore_offset = false;
4943                 size = min_t(u32, loi->size, SZ_64K);
4944         } else {
4945                 /* All reserved bits must be 0 for now */
4946                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4947                         ret = -EINVAL;
4948                         goto out_loi;
4949                 }
4950                 /* Only accept flags we have defined so far */
4951                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4952                         ret = -EINVAL;
4953                         goto out_loi;
4954                 }
4955                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4956                 size = min_t(u32, loi->size, SZ_16M);
4957         }
4958
4959         path = btrfs_alloc_path();
4960         if (!path) {
4961                 ret = -ENOMEM;
4962                 goto out;
4963         }
4964
4965         inodes = init_data_container(size);
4966         if (IS_ERR(inodes)) {
4967                 ret = PTR_ERR(inodes);
4968                 inodes = NULL;
4969                 goto out;
4970         }
4971
4972         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4973                                           build_ino_list, inodes, ignore_offset);
4974         if (ret == -EINVAL)
4975                 ret = -ENOENT;
4976         if (ret < 0)
4977                 goto out;
4978
4979         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4980                            size);
4981         if (ret)
4982                 ret = -EFAULT;
4983
4984 out:
4985         btrfs_free_path(path);
4986         kvfree(inodes);
4987 out_loi:
4988         kfree(loi);
4989
4990         return ret;
4991 }
4992
4993 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4994                                struct btrfs_ioctl_balance_args *bargs)
4995 {
4996         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4997
4998         bargs->flags = bctl->flags;
4999
5000         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
5001                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
5002         if (atomic_read(&fs_info->balance_pause_req))
5003                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
5004         if (atomic_read(&fs_info->balance_cancel_req))
5005                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
5006
5007         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
5008         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
5009         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
5010
5011         spin_lock(&fs_info->balance_lock);
5012         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
5013         spin_unlock(&fs_info->balance_lock);
5014 }
5015
5016 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
5017 {
5018         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
5019         struct btrfs_fs_info *fs_info = root->fs_info;
5020         struct btrfs_ioctl_balance_args *bargs;
5021         struct btrfs_balance_control *bctl;
5022         bool need_unlock; /* for mut. excl. ops lock */
5023         int ret;
5024
5025         if (!capable(CAP_SYS_ADMIN))
5026                 return -EPERM;
5027
5028         ret = mnt_want_write_file(file);
5029         if (ret)
5030                 return ret;
5031
5032 again:
5033         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
5034                 mutex_lock(&fs_info->balance_mutex);
5035                 need_unlock = true;
5036                 goto locked;
5037         }
5038
5039         /*
5040          * mut. excl. ops lock is locked.  Three possibilities:
5041          *   (1) some other op is running
5042          *   (2) balance is running
5043          *   (3) balance is paused -- special case (think resume)
5044          */
5045         mutex_lock(&fs_info->balance_mutex);
5046         if (fs_info->balance_ctl) {
5047                 /* this is either (2) or (3) */
5048                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
5049                         mutex_unlock(&fs_info->balance_mutex);
5050                         /*
5051                          * Lock released to allow other waiters to continue,
5052                          * we'll reexamine the status again.
5053                          */
5054                         mutex_lock(&fs_info->balance_mutex);
5055
5056                         if (fs_info->balance_ctl &&
5057                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
5058                                 /* this is (3) */
5059                                 need_unlock = false;
5060                                 goto locked;
5061                         }
5062
5063                         mutex_unlock(&fs_info->balance_mutex);
5064                         goto again;
5065                 } else {
5066                         /* this is (2) */
5067                         mutex_unlock(&fs_info->balance_mutex);
5068                         ret = -EINPROGRESS;
5069                         goto out;
5070                 }
5071         } else {
5072                 /* this is (1) */
5073                 mutex_unlock(&fs_info->balance_mutex);
5074                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
5075                 goto out;
5076         }
5077
5078 locked:
5079         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
5080
5081         if (arg) {
5082                 bargs = memdup_user(arg, sizeof(*bargs));
5083                 if (IS_ERR(bargs)) {
5084                         ret = PTR_ERR(bargs);
5085                         goto out_unlock;
5086                 }
5087
5088                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
5089                         if (!fs_info->balance_ctl) {
5090                                 ret = -ENOTCONN;
5091                                 goto out_bargs;
5092                         }
5093
5094                         bctl = fs_info->balance_ctl;
5095                         spin_lock(&fs_info->balance_lock);
5096                         bctl->flags |= BTRFS_BALANCE_RESUME;
5097                         spin_unlock(&fs_info->balance_lock);
5098
5099                         goto do_balance;
5100                 }
5101         } else {
5102                 bargs = NULL;
5103         }
5104
5105         if (fs_info->balance_ctl) {
5106                 ret = -EINPROGRESS;
5107                 goto out_bargs;
5108         }
5109
5110         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
5111         if (!bctl) {
5112                 ret = -ENOMEM;
5113                 goto out_bargs;
5114         }
5115
5116         if (arg) {
5117                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
5118                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
5119                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
5120
5121                 bctl->flags = bargs->flags;
5122         } else {
5123                 /* balance everything - no filters */
5124                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
5125         }
5126
5127         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
5128                 ret = -EINVAL;
5129                 goto out_bctl;
5130         }
5131
5132 do_balance:
5133         /*
5134          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
5135          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
5136          * restriper was paused all the way until unmount, in free_fs_info.
5137          * The flag should be cleared after reset_balance_state.
5138          */
5139         need_unlock = false;
5140
5141         ret = btrfs_balance(fs_info, bctl, bargs);
5142         bctl = NULL;
5143
5144         if (arg) {
5145                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
5146                         ret = -EFAULT;
5147         }
5148
5149 out_bctl:
5150         kfree(bctl);
5151 out_bargs:
5152         kfree(bargs);
5153 out_unlock:
5154         mutex_unlock(&fs_info->balance_mutex);
5155         if (need_unlock)
5156                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
5157 out:
5158         mnt_drop_write_file(file);
5159         return ret;
5160 }
5161
5162 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
5163 {
5164         if (!capable(CAP_SYS_ADMIN))
5165                 return -EPERM;
5166
5167         switch (cmd) {
5168         case BTRFS_BALANCE_CTL_PAUSE:
5169                 return btrfs_pause_balance(fs_info);
5170         case BTRFS_BALANCE_CTL_CANCEL:
5171                 return btrfs_cancel_balance(fs_info);
5172         }
5173
5174         return -EINVAL;
5175 }
5176
5177 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
5178                                          void __user *arg)
5179 {
5180         struct btrfs_ioctl_balance_args *bargs;
5181         int ret = 0;
5182
5183         if (!capable(CAP_SYS_ADMIN))
5184                 return -EPERM;
5185
5186         mutex_lock(&fs_info->balance_mutex);
5187         if (!fs_info->balance_ctl) {
5188                 ret = -ENOTCONN;
5189                 goto out;
5190         }
5191
5192         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
5193         if (!bargs) {
5194                 ret = -ENOMEM;
5195                 goto out;
5196         }
5197
5198         btrfs_update_ioctl_balance_args(fs_info, bargs);
5199
5200         if (copy_to_user(arg, bargs, sizeof(*bargs)))
5201                 ret = -EFAULT;
5202
5203         kfree(bargs);
5204 out:
5205         mutex_unlock(&fs_info->balance_mutex);
5206         return ret;
5207 }
5208
5209 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
5210 {
5211         struct inode *inode = file_inode(file);
5212         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5213         struct btrfs_ioctl_quota_ctl_args *sa;
5214         int ret;
5215
5216         if (!capable(CAP_SYS_ADMIN))
5217                 return -EPERM;
5218
5219         ret = mnt_want_write_file(file);
5220         if (ret)
5221                 return ret;
5222
5223         sa = memdup_user(arg, sizeof(*sa));
5224         if (IS_ERR(sa)) {
5225                 ret = PTR_ERR(sa);
5226                 goto drop_write;
5227         }
5228
5229         down_write(&fs_info->subvol_sem);
5230
5231         switch (sa->cmd) {
5232         case BTRFS_QUOTA_CTL_ENABLE:
5233                 ret = btrfs_quota_enable(fs_info);
5234                 break;
5235         case BTRFS_QUOTA_CTL_DISABLE:
5236                 ret = btrfs_quota_disable(fs_info);
5237                 break;
5238         default:
5239                 ret = -EINVAL;
5240                 break;
5241         }
5242
5243         kfree(sa);
5244         up_write(&fs_info->subvol_sem);
5245 drop_write:
5246         mnt_drop_write_file(file);
5247         return ret;
5248 }
5249
5250 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
5251 {
5252         struct inode *inode = file_inode(file);
5253         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5254         struct btrfs_root *root = BTRFS_I(inode)->root;
5255         struct btrfs_ioctl_qgroup_assign_args *sa;
5256         struct btrfs_trans_handle *trans;
5257         int ret;
5258         int err;
5259
5260         if (!capable(CAP_SYS_ADMIN))
5261                 return -EPERM;
5262
5263         ret = mnt_want_write_file(file);
5264         if (ret)
5265                 return ret;
5266
5267         sa = memdup_user(arg, sizeof(*sa));
5268         if (IS_ERR(sa)) {
5269                 ret = PTR_ERR(sa);
5270                 goto drop_write;
5271         }
5272
5273         trans = btrfs_join_transaction(root);
5274         if (IS_ERR(trans)) {
5275                 ret = PTR_ERR(trans);
5276                 goto out;
5277         }
5278
5279         if (sa->assign) {
5280                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
5281         } else {
5282                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
5283         }
5284
5285         /* update qgroup status and info */
5286         err = btrfs_run_qgroups(trans);
5287         if (err < 0)
5288                 btrfs_handle_fs_error(fs_info, err,
5289                                       "failed to update qgroup status and info");
5290         err = btrfs_end_transaction(trans);
5291         if (err && !ret)
5292                 ret = err;
5293
5294 out:
5295         kfree(sa);
5296 drop_write:
5297         mnt_drop_write_file(file);
5298         return ret;
5299 }
5300
5301 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
5302 {
5303         struct inode *inode = file_inode(file);
5304         struct btrfs_root *root = BTRFS_I(inode)->root;
5305         struct btrfs_ioctl_qgroup_create_args *sa;
5306         struct btrfs_trans_handle *trans;
5307         int ret;
5308         int err;
5309
5310         if (!capable(CAP_SYS_ADMIN))
5311                 return -EPERM;
5312
5313         ret = mnt_want_write_file(file);
5314         if (ret)
5315                 return ret;
5316
5317         sa = memdup_user(arg, sizeof(*sa));
5318         if (IS_ERR(sa)) {
5319                 ret = PTR_ERR(sa);
5320                 goto drop_write;
5321         }
5322
5323         if (!sa->qgroupid) {
5324                 ret = -EINVAL;
5325                 goto out;
5326         }
5327
5328         trans = btrfs_join_transaction(root);
5329         if (IS_ERR(trans)) {
5330                 ret = PTR_ERR(trans);
5331                 goto out;
5332         }
5333
5334         if (sa->create) {
5335                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
5336         } else {
5337                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
5338         }
5339
5340         err = btrfs_end_transaction(trans);
5341         if (err && !ret)
5342                 ret = err;
5343
5344 out:
5345         kfree(sa);
5346 drop_write:
5347         mnt_drop_write_file(file);
5348         return ret;
5349 }
5350
5351 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
5352 {
5353         struct inode *inode = file_inode(file);
5354         struct btrfs_root *root = BTRFS_I(inode)->root;
5355         struct btrfs_ioctl_qgroup_limit_args *sa;
5356         struct btrfs_trans_handle *trans;
5357         int ret;
5358         int err;
5359         u64 qgroupid;
5360
5361         if (!capable(CAP_SYS_ADMIN))
5362                 return -EPERM;
5363
5364         ret = mnt_want_write_file(file);
5365         if (ret)
5366                 return ret;
5367
5368         sa = memdup_user(arg, sizeof(*sa));
5369         if (IS_ERR(sa)) {
5370                 ret = PTR_ERR(sa);
5371                 goto drop_write;
5372         }
5373
5374         trans = btrfs_join_transaction(root);
5375         if (IS_ERR(trans)) {
5376                 ret = PTR_ERR(trans);
5377                 goto out;
5378         }
5379
5380         qgroupid = sa->qgroupid;
5381         if (!qgroupid) {
5382                 /* take the current subvol as qgroup */
5383                 qgroupid = root->root_key.objectid;
5384         }
5385
5386         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
5387
5388         err = btrfs_end_transaction(trans);
5389         if (err && !ret)
5390                 ret = err;
5391
5392 out:
5393         kfree(sa);
5394 drop_write:
5395         mnt_drop_write_file(file);
5396         return ret;
5397 }
5398
5399 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5400 {
5401         struct inode *inode = file_inode(file);
5402         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5403         struct btrfs_ioctl_quota_rescan_args *qsa;
5404         int ret;
5405
5406         if (!capable(CAP_SYS_ADMIN))
5407                 return -EPERM;
5408
5409         ret = mnt_want_write_file(file);
5410         if (ret)
5411                 return ret;
5412
5413         qsa = memdup_user(arg, sizeof(*qsa));
5414         if (IS_ERR(qsa)) {
5415                 ret = PTR_ERR(qsa);
5416                 goto drop_write;
5417         }
5418
5419         if (qsa->flags) {
5420                 ret = -EINVAL;
5421                 goto out;
5422         }
5423
5424         ret = btrfs_qgroup_rescan(fs_info);
5425
5426 out:
5427         kfree(qsa);
5428 drop_write:
5429         mnt_drop_write_file(file);
5430         return ret;
5431 }
5432
5433 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5434 {
5435         struct inode *inode = file_inode(file);
5436         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5437         struct btrfs_ioctl_quota_rescan_args *qsa;
5438         int ret = 0;
5439
5440         if (!capable(CAP_SYS_ADMIN))
5441                 return -EPERM;
5442
5443         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5444         if (!qsa)
5445                 return -ENOMEM;
5446
5447         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5448                 qsa->flags = 1;
5449                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5450         }
5451
5452         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5453                 ret = -EFAULT;
5454
5455         kfree(qsa);
5456         return ret;
5457 }
5458
5459 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5460 {
5461         struct inode *inode = file_inode(file);
5462         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5463
5464         if (!capable(CAP_SYS_ADMIN))
5465                 return -EPERM;
5466
5467         return btrfs_qgroup_wait_for_completion(fs_info, true);
5468 }
5469
5470 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5471                                             struct btrfs_ioctl_received_subvol_args *sa)
5472 {
5473         struct inode *inode = file_inode(file);
5474         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5475         struct btrfs_root *root = BTRFS_I(inode)->root;
5476         struct btrfs_root_item *root_item = &root->root_item;
5477         struct btrfs_trans_handle *trans;
5478         struct timespec64 ct = current_time(inode);
5479         int ret = 0;
5480         int received_uuid_changed;
5481
5482         if (!inode_owner_or_capable(inode))
5483                 return -EPERM;
5484
5485         ret = mnt_want_write_file(file);
5486         if (ret < 0)
5487                 return ret;
5488
5489         down_write(&fs_info->subvol_sem);
5490
5491         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5492                 ret = -EINVAL;
5493                 goto out;
5494         }
5495
5496         if (btrfs_root_readonly(root)) {
5497                 ret = -EROFS;
5498                 goto out;
5499         }
5500
5501         /*
5502          * 1 - root item
5503          * 2 - uuid items (received uuid + subvol uuid)
5504          */
5505         trans = btrfs_start_transaction(root, 3);
5506         if (IS_ERR(trans)) {
5507                 ret = PTR_ERR(trans);
5508                 trans = NULL;
5509                 goto out;
5510         }
5511
5512         sa->rtransid = trans->transid;
5513         sa->rtime.sec = ct.tv_sec;
5514         sa->rtime.nsec = ct.tv_nsec;
5515
5516         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5517                                        BTRFS_UUID_SIZE);
5518         if (received_uuid_changed &&
5519             !btrfs_is_empty_uuid(root_item->received_uuid)) {
5520                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5521                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5522                                           root->root_key.objectid);
5523                 if (ret && ret != -ENOENT) {
5524                         btrfs_abort_transaction(trans, ret);
5525                         btrfs_end_transaction(trans);
5526                         goto out;
5527                 }
5528         }
5529         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5530         btrfs_set_root_stransid(root_item, sa->stransid);
5531         btrfs_set_root_rtransid(root_item, sa->rtransid);
5532         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5533         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5534         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5535         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5536
5537         ret = btrfs_update_root(trans, fs_info->tree_root,
5538                                 &root->root_key, &root->root_item);
5539         if (ret < 0) {
5540                 btrfs_end_transaction(trans);
5541                 goto out;
5542         }
5543         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5544                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
5545                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5546                                           root->root_key.objectid);
5547                 if (ret < 0 && ret != -EEXIST) {
5548                         btrfs_abort_transaction(trans, ret);
5549                         btrfs_end_transaction(trans);
5550                         goto out;
5551                 }
5552         }
5553         ret = btrfs_commit_transaction(trans);
5554 out:
5555         up_write(&fs_info->subvol_sem);
5556         mnt_drop_write_file(file);
5557         return ret;
5558 }
5559
5560 #ifdef CONFIG_64BIT
5561 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5562                                                 void __user *arg)
5563 {
5564         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5565         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5566         int ret = 0;
5567
5568         args32 = memdup_user(arg, sizeof(*args32));
5569         if (IS_ERR(args32))
5570                 return PTR_ERR(args32);
5571
5572         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5573         if (!args64) {
5574                 ret = -ENOMEM;
5575                 goto out;
5576         }
5577
5578         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5579         args64->stransid = args32->stransid;
5580         args64->rtransid = args32->rtransid;
5581         args64->stime.sec = args32->stime.sec;
5582         args64->stime.nsec = args32->stime.nsec;
5583         args64->rtime.sec = args32->rtime.sec;
5584         args64->rtime.nsec = args32->rtime.nsec;
5585         args64->flags = args32->flags;
5586
5587         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5588         if (ret)
5589                 goto out;
5590
5591         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5592         args32->stransid = args64->stransid;
5593         args32->rtransid = args64->rtransid;
5594         args32->stime.sec = args64->stime.sec;
5595         args32->stime.nsec = args64->stime.nsec;
5596         args32->rtime.sec = args64->rtime.sec;
5597         args32->rtime.nsec = args64->rtime.nsec;
5598         args32->flags = args64->flags;
5599
5600         ret = copy_to_user(arg, args32, sizeof(*args32));
5601         if (ret)
5602                 ret = -EFAULT;
5603
5604 out:
5605         kfree(args32);
5606         kfree(args64);
5607         return ret;
5608 }
5609 #endif
5610
5611 static long btrfs_ioctl_set_received_subvol(struct file *file,
5612                                             void __user *arg)
5613 {
5614         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5615         int ret = 0;
5616
5617         sa = memdup_user(arg, sizeof(*sa));
5618         if (IS_ERR(sa))
5619                 return PTR_ERR(sa);
5620
5621         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5622
5623         if (ret)
5624                 goto out;
5625
5626         ret = copy_to_user(arg, sa, sizeof(*sa));
5627         if (ret)
5628                 ret = -EFAULT;
5629
5630 out:
5631         kfree(sa);
5632         return ret;
5633 }
5634
5635 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5636 {
5637         struct inode *inode = file_inode(file);
5638         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5639         size_t len;
5640         int ret;
5641         char label[BTRFS_LABEL_SIZE];
5642
5643         spin_lock(&fs_info->super_lock);
5644         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5645         spin_unlock(&fs_info->super_lock);
5646
5647         len = strnlen(label, BTRFS_LABEL_SIZE);
5648
5649         if (len == BTRFS_LABEL_SIZE) {
5650                 btrfs_warn(fs_info,
5651                            "label is too long, return the first %zu bytes",
5652                            --len);
5653         }
5654
5655         ret = copy_to_user(arg, label, len);
5656
5657         return ret ? -EFAULT : 0;
5658 }
5659
5660 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5661 {
5662         struct inode *inode = file_inode(file);
5663         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5664         struct btrfs_root *root = BTRFS_I(inode)->root;
5665         struct btrfs_super_block *super_block = fs_info->super_copy;
5666         struct btrfs_trans_handle *trans;
5667         char label[BTRFS_LABEL_SIZE];
5668         int ret;
5669
5670         if (!capable(CAP_SYS_ADMIN))
5671                 return -EPERM;
5672
5673         if (copy_from_user(label, arg, sizeof(label)))
5674                 return -EFAULT;
5675
5676         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5677                 btrfs_err(fs_info,
5678                           "unable to set label with more than %d bytes",
5679                           BTRFS_LABEL_SIZE - 1);
5680                 return -EINVAL;
5681         }
5682
5683         ret = mnt_want_write_file(file);
5684         if (ret)
5685                 return ret;
5686
5687         trans = btrfs_start_transaction(root, 0);
5688         if (IS_ERR(trans)) {
5689                 ret = PTR_ERR(trans);
5690                 goto out_unlock;
5691         }
5692
5693         spin_lock(&fs_info->super_lock);
5694         strcpy(super_block->label, label);
5695         spin_unlock(&fs_info->super_lock);
5696         ret = btrfs_commit_transaction(trans);
5697
5698 out_unlock:
5699         mnt_drop_write_file(file);
5700         return ret;
5701 }
5702
5703 #define INIT_FEATURE_FLAGS(suffix) \
5704         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5705           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5706           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5707
5708 int btrfs_ioctl_get_supported_features(void __user *arg)
5709 {
5710         static const struct btrfs_ioctl_feature_flags features[3] = {
5711                 INIT_FEATURE_FLAGS(SUPP),
5712                 INIT_FEATURE_FLAGS(SAFE_SET),
5713                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5714         };
5715
5716         if (copy_to_user(arg, &features, sizeof(features)))
5717                 return -EFAULT;
5718
5719         return 0;
5720 }
5721
5722 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5723 {
5724         struct inode *inode = file_inode(file);
5725         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5726         struct btrfs_super_block *super_block = fs_info->super_copy;
5727         struct btrfs_ioctl_feature_flags features;
5728
5729         features.compat_flags = btrfs_super_compat_flags(super_block);
5730         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5731         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5732
5733         if (copy_to_user(arg, &features, sizeof(features)))
5734                 return -EFAULT;
5735
5736         return 0;
5737 }
5738
5739 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5740                               enum btrfs_feature_set set,
5741                               u64 change_mask, u64 flags, u64 supported_flags,
5742                               u64 safe_set, u64 safe_clear)
5743 {
5744         const char *type = btrfs_feature_set_names[set];
5745         char *names;
5746         u64 disallowed, unsupported;
5747         u64 set_mask = flags & change_mask;
5748         u64 clear_mask = ~flags & change_mask;
5749
5750         unsupported = set_mask & ~supported_flags;
5751         if (unsupported) {
5752                 names = btrfs_printable_features(set, unsupported);
5753                 if (names) {
5754                         btrfs_warn(fs_info,
5755                                    "this kernel does not support the %s feature bit%s",
5756                                    names, strchr(names, ',') ? "s" : "");
5757                         kfree(names);
5758                 } else
5759                         btrfs_warn(fs_info,
5760                                    "this kernel does not support %s bits 0x%llx",
5761                                    type, unsupported);
5762                 return -EOPNOTSUPP;
5763         }
5764
5765         disallowed = set_mask & ~safe_set;
5766         if (disallowed) {
5767                 names = btrfs_printable_features(set, disallowed);
5768                 if (names) {
5769                         btrfs_warn(fs_info,
5770                                    "can't set the %s feature bit%s while mounted",
5771                                    names, strchr(names, ',') ? "s" : "");
5772                         kfree(names);
5773                 } else
5774                         btrfs_warn(fs_info,
5775                                    "can't set %s bits 0x%llx while mounted",
5776                                    type, disallowed);
5777                 return -EPERM;
5778         }
5779
5780         disallowed = clear_mask & ~safe_clear;
5781         if (disallowed) {
5782                 names = btrfs_printable_features(set, disallowed);
5783                 if (names) {
5784                         btrfs_warn(fs_info,
5785                                    "can't clear the %s feature bit%s while mounted",
5786                                    names, strchr(names, ',') ? "s" : "");
5787                         kfree(names);
5788                 } else
5789                         btrfs_warn(fs_info,
5790                                    "can't clear %s bits 0x%llx while mounted",
5791                                    type, disallowed);
5792                 return -EPERM;
5793         }
5794
5795         return 0;
5796 }
5797
5798 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5799 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5800                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5801                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5802                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5803
5804 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5805 {
5806         struct inode *inode = file_inode(file);
5807         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5808         struct btrfs_root *root = BTRFS_I(inode)->root;
5809         struct btrfs_super_block *super_block = fs_info->super_copy;
5810         struct btrfs_ioctl_feature_flags flags[2];
5811         struct btrfs_trans_handle *trans;
5812         u64 newflags;
5813         int ret;
5814
5815         if (!capable(CAP_SYS_ADMIN))
5816                 return -EPERM;
5817
5818         if (copy_from_user(flags, arg, sizeof(flags)))
5819                 return -EFAULT;
5820
5821         /* Nothing to do */
5822         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5823             !flags[0].incompat_flags)
5824                 return 0;
5825
5826         ret = check_feature(fs_info, flags[0].compat_flags,
5827                             flags[1].compat_flags, COMPAT);
5828         if (ret)
5829                 return ret;
5830
5831         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5832                             flags[1].compat_ro_flags, COMPAT_RO);
5833         if (ret)
5834                 return ret;
5835
5836         ret = check_feature(fs_info, flags[0].incompat_flags,
5837                             flags[1].incompat_flags, INCOMPAT);
5838         if (ret)
5839                 return ret;
5840
5841         ret = mnt_want_write_file(file);
5842         if (ret)
5843                 return ret;
5844
5845         trans = btrfs_start_transaction(root, 0);
5846         if (IS_ERR(trans)) {
5847                 ret = PTR_ERR(trans);
5848                 goto out_drop_write;
5849         }
5850
5851         spin_lock(&fs_info->super_lock);
5852         newflags = btrfs_super_compat_flags(super_block);
5853         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5854         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5855         btrfs_set_super_compat_flags(super_block, newflags);
5856
5857         newflags = btrfs_super_compat_ro_flags(super_block);
5858         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5859         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5860         btrfs_set_super_compat_ro_flags(super_block, newflags);
5861
5862         newflags = btrfs_super_incompat_flags(super_block);
5863         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5864         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5865         btrfs_set_super_incompat_flags(super_block, newflags);
5866         spin_unlock(&fs_info->super_lock);
5867
5868         ret = btrfs_commit_transaction(trans);
5869 out_drop_write:
5870         mnt_drop_write_file(file);
5871
5872         return ret;
5873 }
5874
5875 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5876 {
5877         struct btrfs_ioctl_send_args *arg;
5878         int ret;
5879
5880         if (compat) {
5881 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5882                 struct btrfs_ioctl_send_args_32 args32;
5883
5884                 ret = copy_from_user(&args32, argp, sizeof(args32));
5885                 if (ret)
5886                         return -EFAULT;
5887                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5888                 if (!arg)
5889                         return -ENOMEM;
5890                 arg->send_fd = args32.send_fd;
5891                 arg->clone_sources_count = args32.clone_sources_count;
5892                 arg->clone_sources = compat_ptr(args32.clone_sources);
5893                 arg->parent_root = args32.parent_root;
5894                 arg->flags = args32.flags;
5895                 memcpy(arg->reserved, args32.reserved,
5896                        sizeof(args32.reserved));
5897 #else
5898                 return -ENOTTY;
5899 #endif
5900         } else {
5901                 arg = memdup_user(argp, sizeof(*arg));
5902                 if (IS_ERR(arg))
5903                         return PTR_ERR(arg);
5904         }
5905         ret = btrfs_ioctl_send(file, arg);
5906         kfree(arg);
5907         return ret;
5908 }
5909
5910 long btrfs_ioctl(struct file *file, unsigned int
5911                 cmd, unsigned long arg)
5912 {
5913         struct inode *inode = file_inode(file);
5914         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5915         struct btrfs_root *root = BTRFS_I(inode)->root;
5916         void __user *argp = (void __user *)arg;
5917
5918         switch (cmd) {
5919         case FS_IOC_GETFLAGS:
5920                 return btrfs_ioctl_getflags(file, argp);
5921         case FS_IOC_SETFLAGS:
5922                 return btrfs_ioctl_setflags(file, argp);
5923         case FS_IOC_GETVERSION:
5924                 return btrfs_ioctl_getversion(file, argp);
5925         case FITRIM:
5926                 return btrfs_ioctl_fitrim(file, argp);
5927         case BTRFS_IOC_SNAP_CREATE:
5928                 return btrfs_ioctl_snap_create(file, argp, 0);
5929         case BTRFS_IOC_SNAP_CREATE_V2:
5930                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5931         case BTRFS_IOC_SUBVOL_CREATE:
5932                 return btrfs_ioctl_snap_create(file, argp, 1);
5933         case BTRFS_IOC_SUBVOL_CREATE_V2:
5934                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5935         case BTRFS_IOC_SNAP_DESTROY:
5936                 return btrfs_ioctl_snap_destroy(file, argp);
5937         case BTRFS_IOC_SUBVOL_GETFLAGS:
5938                 return btrfs_ioctl_subvol_getflags(file, argp);
5939         case BTRFS_IOC_SUBVOL_SETFLAGS:
5940                 return btrfs_ioctl_subvol_setflags(file, argp);
5941         case BTRFS_IOC_DEFAULT_SUBVOL:
5942                 return btrfs_ioctl_default_subvol(file, argp);
5943         case BTRFS_IOC_DEFRAG:
5944                 return btrfs_ioctl_defrag(file, NULL);
5945         case BTRFS_IOC_DEFRAG_RANGE:
5946                 return btrfs_ioctl_defrag(file, argp);
5947         case BTRFS_IOC_RESIZE:
5948                 return btrfs_ioctl_resize(file, argp);
5949         case BTRFS_IOC_ADD_DEV:
5950                 return btrfs_ioctl_add_dev(fs_info, argp);
5951         case BTRFS_IOC_RM_DEV:
5952                 return btrfs_ioctl_rm_dev(file, argp);
5953         case BTRFS_IOC_RM_DEV_V2:
5954                 return btrfs_ioctl_rm_dev_v2(file, argp);
5955         case BTRFS_IOC_FS_INFO:
5956                 return btrfs_ioctl_fs_info(fs_info, argp);
5957         case BTRFS_IOC_DEV_INFO:
5958                 return btrfs_ioctl_dev_info(fs_info, argp);
5959         case BTRFS_IOC_BALANCE:
5960                 return btrfs_ioctl_balance(file, NULL);
5961         case BTRFS_IOC_TREE_SEARCH:
5962                 return btrfs_ioctl_tree_search(file, argp);
5963         case BTRFS_IOC_TREE_SEARCH_V2:
5964                 return btrfs_ioctl_tree_search_v2(file, argp);
5965         case BTRFS_IOC_INO_LOOKUP:
5966                 return btrfs_ioctl_ino_lookup(file, argp);
5967         case BTRFS_IOC_INO_PATHS:
5968                 return btrfs_ioctl_ino_to_path(root, argp);
5969         case BTRFS_IOC_LOGICAL_INO:
5970                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5971         case BTRFS_IOC_LOGICAL_INO_V2:
5972                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5973         case BTRFS_IOC_SPACE_INFO:
5974                 return btrfs_ioctl_space_info(fs_info, argp);
5975         case BTRFS_IOC_SYNC: {
5976                 int ret;
5977
5978                 ret = btrfs_start_delalloc_roots(fs_info, -1);
5979                 if (ret)
5980                         return ret;
5981                 ret = btrfs_sync_fs(inode->i_sb, 1);
5982                 /*
5983                  * The transaction thread may want to do more work,
5984                  * namely it pokes the cleaner kthread that will start
5985                  * processing uncleaned subvols.
5986                  */
5987                 wake_up_process(fs_info->transaction_kthread);
5988                 return ret;
5989         }
5990         case BTRFS_IOC_START_SYNC:
5991                 return btrfs_ioctl_start_sync(root, argp);
5992         case BTRFS_IOC_WAIT_SYNC:
5993                 return btrfs_ioctl_wait_sync(fs_info, argp);
5994         case BTRFS_IOC_SCRUB:
5995                 return btrfs_ioctl_scrub(file, argp);
5996         case BTRFS_IOC_SCRUB_CANCEL:
5997                 return btrfs_ioctl_scrub_cancel(fs_info);
5998         case BTRFS_IOC_SCRUB_PROGRESS:
5999                 return btrfs_ioctl_scrub_progress(fs_info, argp);
6000         case BTRFS_IOC_BALANCE_V2:
6001                 return btrfs_ioctl_balance(file, argp);
6002         case BTRFS_IOC_BALANCE_CTL:
6003                 return btrfs_ioctl_balance_ctl(fs_info, arg);
6004         case BTRFS_IOC_BALANCE_PROGRESS:
6005                 return btrfs_ioctl_balance_progress(fs_info, argp);
6006         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
6007                 return btrfs_ioctl_set_received_subvol(file, argp);
6008 #ifdef CONFIG_64BIT
6009         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
6010                 return btrfs_ioctl_set_received_subvol_32(file, argp);
6011 #endif
6012         case BTRFS_IOC_SEND:
6013                 return _btrfs_ioctl_send(file, argp, false);
6014 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
6015         case BTRFS_IOC_SEND_32:
6016                 return _btrfs_ioctl_send(file, argp, true);
6017 #endif
6018         case BTRFS_IOC_GET_DEV_STATS:
6019                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
6020         case BTRFS_IOC_QUOTA_CTL:
6021                 return btrfs_ioctl_quota_ctl(file, argp);
6022         case BTRFS_IOC_QGROUP_ASSIGN:
6023                 return btrfs_ioctl_qgroup_assign(file, argp);
6024         case BTRFS_IOC_QGROUP_CREATE:
6025                 return btrfs_ioctl_qgroup_create(file, argp);
6026         case BTRFS_IOC_QGROUP_LIMIT:
6027                 return btrfs_ioctl_qgroup_limit(file, argp);
6028         case BTRFS_IOC_QUOTA_RESCAN:
6029                 return btrfs_ioctl_quota_rescan(file, argp);
6030         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
6031                 return btrfs_ioctl_quota_rescan_status(file, argp);
6032         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
6033                 return btrfs_ioctl_quota_rescan_wait(file, argp);
6034         case BTRFS_IOC_DEV_REPLACE:
6035                 return btrfs_ioctl_dev_replace(fs_info, argp);
6036         case BTRFS_IOC_GET_FSLABEL:
6037                 return btrfs_ioctl_get_fslabel(file, argp);
6038         case BTRFS_IOC_SET_FSLABEL:
6039                 return btrfs_ioctl_set_fslabel(file, argp);
6040         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
6041                 return btrfs_ioctl_get_supported_features(argp);
6042         case BTRFS_IOC_GET_FEATURES:
6043                 return btrfs_ioctl_get_features(file, argp);
6044         case BTRFS_IOC_SET_FEATURES:
6045                 return btrfs_ioctl_set_features(file, argp);
6046         case FS_IOC_FSGETXATTR:
6047                 return btrfs_ioctl_fsgetxattr(file, argp);
6048         case FS_IOC_FSSETXATTR:
6049                 return btrfs_ioctl_fssetxattr(file, argp);
6050         case BTRFS_IOC_GET_SUBVOL_INFO:
6051                 return btrfs_ioctl_get_subvol_info(file, argp);
6052         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
6053                 return btrfs_ioctl_get_subvol_rootref(file, argp);
6054         case BTRFS_IOC_INO_LOOKUP_USER:
6055                 return btrfs_ioctl_ino_lookup_user(file, argp);
6056         }
6057
6058         return -ENOTTY;
6059 }
6060
6061 #ifdef CONFIG_COMPAT
6062 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
6063 {
6064         /*
6065          * These all access 32-bit values anyway so no further
6066          * handling is necessary.
6067          */
6068         switch (cmd) {
6069         case FS_IOC32_GETFLAGS:
6070                 cmd = FS_IOC_GETFLAGS;
6071                 break;
6072         case FS_IOC32_SETFLAGS:
6073                 cmd = FS_IOC_SETFLAGS;
6074                 break;
6075         case FS_IOC32_GETVERSION:
6076                 cmd = FS_IOC_GETVERSION;
6077                 break;
6078         }
6079
6080         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
6081 }
6082 #endif