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 #include "space-info.h"
  47 #include "delalloc-space.h"
  48 #include "block-group.h"
  49
  50 #ifdef CONFIG_64BIT
  51 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  52  * structures are incorrect, as the timespec structure from userspace
  53  * is 4 bytes too small. We define these alternatives here to teach
  54  * the kernel about the 32-bit struct packing.
  55  */
  56 struct btrfs_ioctl_timespec_32 {
  57         __u64 sec;
  58         __u32 nsec;
  59 } __attribute__ ((__packed__));
  60
  61 struct btrfs_ioctl_received_subvol_args_32 {
  62         char    uuid[BTRFS_UUID_SIZE];  /* in */
  63         __u64   stransid;               /* in */
  64         __u64   rtransid;               /* out */
  65         struct btrfs_ioctl_timespec_32 stime; /* in */
  66         struct btrfs_ioctl_timespec_32 rtime; /* out */
  67         __u64   flags;                  /* in */
  68         __u64   reserved[16];           /* in */
  69 } __attribute__ ((__packed__));
  70
  71 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  72                                 struct btrfs_ioctl_received_subvol_args_32)
  73 #endif
  74
  75 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  76 struct btrfs_ioctl_send_args_32 {
  77         __s64 send_fd;                  /* in */
  78         __u64 clone_sources_count;      /* in */
  79         compat_uptr_t clone_sources;    /* in */
  80         __u64 parent_root;              /* in */
  81         __u64 flags;                    /* in */
  82         __u64 reserved[4];              /* in */
  83 } __attribute__ ((__packed__));
  84
  85 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  86                                struct btrfs_ioctl_send_args_32)
  87 #endif
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  91                        int no_time_update);
  92
  93 /* Mask out flags that are inappropriate for the given type of inode. */
  94 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  95                 unsigned int flags)
  96 {
  97         if (S_ISDIR(inode->i_mode))
  98                 return flags;
  99         else if (S_ISREG(inode->i_mode))
 100                 return flags & ~FS_DIRSYNC_FL;
 101         else
 102                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 103 }
 104
 105 /*
 106  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 107  * ioctl.
 108  */
 109 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 110 {
 111         unsigned int iflags = 0;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127
 128         if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130         else if (flags & BTRFS_INODE_COMPRESS)
 131                 iflags |= FS_COMPR_FL;
 132
 133         return iflags;
 134 }
 135
 136 /*
 137  * Update inode->i_flags based on the btrfs internal flags.
 138  */
 139 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 140 {
 141         struct btrfs_inode *binode = BTRFS_I(inode);
 142         unsigned int new_fl = 0;
 143
 144         if (binode->flags & BTRFS_INODE_SYNC)
 145                 new_fl |= S_SYNC;
 146         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 147                 new_fl |= S_IMMUTABLE;
 148         if (binode->flags & BTRFS_INODE_APPEND)
 149                 new_fl |= S_APPEND;
 150         if (binode->flags & BTRFS_INODE_NOATIME)
 151                 new_fl |= S_NOATIME;
 152         if (binode->flags & BTRFS_INODE_DIRSYNC)
 153                 new_fl |= S_DIRSYNC;
 154
 155         set_mask_bits(&inode->i_flags,
 156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157                       new_fl);
 158 }
 159
 160 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 161 {
 162         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 163         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 164
 165         if (copy_to_user(arg, &flags, sizeof(flags)))
 166                 return -EFAULT;
 167         return 0;
 168 }
 169
 170 /* Check if @flags are a supported and valid set of FS_*_FL flags */
 171 static int check_fsflags(unsigned int flags)
 172 {
 173         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 174                       FS_NOATIME_FL | FS_NODUMP_FL | \
 175                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 176                       FS_NOCOMP_FL | FS_COMPR_FL |
 177                       FS_NOCOW_FL))
 178                 return -EOPNOTSUPP;
 179
 180         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 181                 return -EINVAL;
 182
 183         return 0;
 184 }
 185
 186 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 187 {
 188         struct inode *inode = file_inode(file);
 189         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 190         struct btrfs_inode *binode = BTRFS_I(inode);
 191         struct btrfs_root *root = binode->root;
 192         struct btrfs_trans_handle *trans;
 193         unsigned int fsflags, old_fsflags;
 194         int ret;
 195         const char *comp = NULL;
 196         u32 binode_flags = binode->flags;
 197
 198         if (!inode_owner_or_capable(inode))
 199                 return -EPERM;
 200
 201         if (btrfs_root_readonly(root))
 202                 return -EROFS;
 203
 204         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 205                 return -EFAULT;
 206
 207         ret = check_fsflags(fsflags);
 208         if (ret)
 209                 return ret;
 210
 211         ret = mnt_want_write_file(file);
 212         if (ret)
 213                 return ret;
 214
 215         inode_lock(inode);
 216
 217         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 218         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 219         ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 220         if (ret)
 221                 goto out_unlock;
 222
 223         if (fsflags & FS_SYNC_FL)
 224                 binode_flags |= BTRFS_INODE_SYNC;
 225         else
 226                 binode_flags &= ~BTRFS_INODE_SYNC;
 227         if (fsflags & FS_IMMUTABLE_FL)
 228                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 229         else
 230                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 231         if (fsflags & FS_APPEND_FL)
 232                 binode_flags |= BTRFS_INODE_APPEND;
 233         else
 234                 binode_flags &= ~BTRFS_INODE_APPEND;
 235         if (fsflags & FS_NODUMP_FL)
 236                 binode_flags |= BTRFS_INODE_NODUMP;
 237         else
 238                 binode_flags &= ~BTRFS_INODE_NODUMP;
 239         if (fsflags & FS_NOATIME_FL)
 240                 binode_flags |= BTRFS_INODE_NOATIME;
 241         else
 242                 binode_flags &= ~BTRFS_INODE_NOATIME;
 243         if (fsflags & FS_DIRSYNC_FL)
 244                 binode_flags |= BTRFS_INODE_DIRSYNC;
 245         else
 246                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 247         if (fsflags & FS_NOCOW_FL) {
 248                 if (S_ISREG(inode->i_mode)) {
 249                         /*
 250                          * It's safe to turn csums off here, no extents exist.
 251                          * Otherwise we want the flag to reflect the real COW
 252                          * status of the file and will not set it.
 253                          */
 254                         if (inode->i_size == 0)
 255                                 binode_flags |= BTRFS_INODE_NODATACOW |
 256                                                 BTRFS_INODE_NODATASUM;
 257                 } else {
 258                         binode_flags |= BTRFS_INODE_NODATACOW;
 259                 }
 260         } else {
 261                 /*
 262                  * Revert back under same assumptions as above
 263                  */
 264                 if (S_ISREG(inode->i_mode)) {
 265                         if (inode->i_size == 0)
 266                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 267                                                   BTRFS_INODE_NODATASUM);
 268                 } else {
 269                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 270                 }
 271         }
 272
 273         /*
 274          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 275          * flag may be changed automatically if compression code won't make
 276          * things smaller.
 277          */
 278         if (fsflags & FS_NOCOMP_FL) {
 279                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 280                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 281         } else if (fsflags & FS_COMPR_FL) {
 282
 283                 if (IS_SWAPFILE(inode)) {
 284                         ret = -ETXTBSY;
 285                         goto out_unlock;
 286                 }
 287
 288                 binode_flags |= BTRFS_INODE_COMPRESS;
 289                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 290
 291                 comp = btrfs_compress_type2str(fs_info->compress_type);
 292                 if (!comp || comp[0] == 0)
 293                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 294         } else {
 295                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 296         }
 297
 298         /*
 299          * 1 for inode item
 300          * 2 for properties
 301          */
 302         trans = btrfs_start_transaction(root, 3);
 303         if (IS_ERR(trans)) {
 304                 ret = PTR_ERR(trans);
 305                 goto out_unlock;
 306         }
 307
 308         if (comp) {
 309                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 310                                      strlen(comp), 0);
 311                 if (ret) {
 312                         btrfs_abort_transaction(trans, ret);
 313                         goto out_end_trans;
 314                 }
 315         } else {
 316                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 317                                      0, 0);
 318                 if (ret && ret != -ENODATA) {
 319                         btrfs_abort_transaction(trans, ret);
 320                         goto out_end_trans;
 321                 }
 322         }
 323
 324         binode->flags = binode_flags;
 325         btrfs_sync_inode_flags_to_i_flags(inode);
 326         inode_inc_iversion(inode);
 327         inode->i_ctime = current_time(inode);
 328         ret = btrfs_update_inode(trans, root, inode);
 329
 330  out_end_trans:
 331         btrfs_end_transaction(trans);
 332  out_unlock:
 333         inode_unlock(inode);
 334         mnt_drop_write_file(file);
 335         return ret;
 336 }
 337
 338 /*
 339  * Translate btrfs internal inode flags to xflags as expected by the
 340  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 341  * silently dropped.
 342  */
 343 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 344 {
 345         unsigned int xflags = 0;
 346
 347         if (flags & BTRFS_INODE_APPEND)
 348                 xflags |= FS_XFLAG_APPEND;
 349         if (flags & BTRFS_INODE_IMMUTABLE)
 350                 xflags |= FS_XFLAG_IMMUTABLE;
 351         if (flags & BTRFS_INODE_NOATIME)
 352                 xflags |= FS_XFLAG_NOATIME;
 353         if (flags & BTRFS_INODE_NODUMP)
 354                 xflags |= FS_XFLAG_NODUMP;
 355         if (flags & BTRFS_INODE_SYNC)
 356                 xflags |= FS_XFLAG_SYNC;
 357
 358         return xflags;
 359 }
 360
 361 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 362 static int check_xflags(unsigned int flags)
 363 {
 364         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 365                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 366                 return -EOPNOTSUPP;
 367         return 0;
 368 }
 369
 370 /*
 371  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 372  * are zeroed.
 373  */
 374 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 375 {
 376         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 377         struct fsxattr fa;
 378
 379         simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 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, old_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         if (copy_from_user(&fa, arg, sizeof(fa)))
 404                 return -EFAULT;
 405
 406         ret = check_xflags(fa.fsx_xflags);
 407         if (ret)
 408                 return ret;
 409
 410         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 411                 return -EOPNOTSUPP;
 412
 413         ret = mnt_want_write_file(file);
 414         if (ret)
 415                 return ret;
 416
 417         inode_lock(inode);
 418
 419         old_flags = binode->flags;
 420         old_i_flags = inode->i_flags;
 421
 422         simple_fill_fsxattr(&old_fa,
 423                             btrfs_inode_flags_to_xflags(binode->flags));
 424         ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 425         if (ret)
 426                 goto out_unlock;
 427
 428         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 429                 binode->flags |= BTRFS_INODE_SYNC;
 430         else
 431                 binode->flags &= ~BTRFS_INODE_SYNC;
 432         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 433                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 434         else
 435                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 436         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 437                 binode->flags |= BTRFS_INODE_APPEND;
 438         else
 439                 binode->flags &= ~BTRFS_INODE_APPEND;
 440         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 441                 binode->flags |= BTRFS_INODE_NODUMP;
 442         else
 443                 binode->flags &= ~BTRFS_INODE_NODUMP;
 444         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 445                 binode->flags |= BTRFS_INODE_NOATIME;
 446         else
 447                 binode->flags &= ~BTRFS_INODE_NOATIME;
 448
 449         /* 1 item for the inode */
 450         trans = btrfs_start_transaction(root, 1);
 451         if (IS_ERR(trans)) {
 452                 ret = PTR_ERR(trans);
 453                 goto out_unlock;
 454         }
 455
 456         btrfs_sync_inode_flags_to_i_flags(inode);
 457         inode_inc_iversion(inode);
 458         inode->i_ctime = current_time(inode);
 459         ret = btrfs_update_inode(trans, root, inode);
 460
 461         btrfs_end_transaction(trans);
 462
 463 out_unlock:
 464         if (ret) {
 465                 binode->flags = old_flags;
 466                 inode->i_flags = old_i_flags;
 467         }
 468
 469         inode_unlock(inode);
 470         mnt_drop_write_file(file);
 471
 472         return ret;
 473 }
 474
 475 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 476 {
 477         struct inode *inode = file_inode(file);
 478
 479         return put_user(inode->i_generation, arg);
 480 }
 481
 482 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 483                                         void __user *arg)
 484 {
 485         struct btrfs_device *device;
 486         struct request_queue *q;
 487         struct fstrim_range range;
 488         u64 minlen = ULLONG_MAX;
 489         u64 num_devices = 0;
 490         int ret;
 491
 492         if (!capable(CAP_SYS_ADMIN))
 493                 return -EPERM;
 494
 495         /*
 496          * If the fs is mounted with nologreplay, which requires it to be
 497          * mounted in RO mode as well, we can not allow discard on free space
 498          * inside block groups, because log trees refer to extents that are not
 499          * pinned in a block group's free space cache (pinning the extents is
 500          * precisely the first phase of replaying a log tree).
 501          */
 502         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 503                 return -EROFS;
 504
 505         rcu_read_lock();
 506         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 507                                 dev_list) {
 508                 if (!device->bdev)
 509                         continue;
 510                 q = bdev_get_queue(device->bdev);
 511                 if (blk_queue_discard(q)) {
 512                         num_devices++;
 513                         minlen = min_t(u64, q->limits.discard_granularity,
 514                                      minlen);
 515                 }
 516         }
 517         rcu_read_unlock();
 518
 519         if (!num_devices)
 520                 return -EOPNOTSUPP;
 521         if (copy_from_user(&range, arg, sizeof(range)))
 522                 return -EFAULT;
 523
 524         /*
 525          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 526          * block group is in the logical address space, which can be any
 527          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 528          */
 529         if (range.len < fs_info->sb->s_blocksize)
 530                 return -EINVAL;
 531
 532         range.minlen = max(range.minlen, minlen);
 533         ret = btrfs_trim_fs(fs_info, &range);
 534         if (ret < 0)
 535                 return ret;
 536
 537         if (copy_to_user(arg, &range, sizeof(range)))
 538                 return -EFAULT;
 539
 540         return 0;
 541 }
 542
 543 int __pure btrfs_is_empty_uuid(u8 *uuid)
 544 {
 545         int i;
 546
 547         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 548                 if (uuid[i])
 549                         return 0;
 550         }
 551         return 1;
 552 }
 553
 554 static noinline int create_subvol(struct inode *dir,
 555                                   struct dentry *dentry,
 556                                   const char *name, int namelen,
 557                                   u64 *async_transid,
 558                                   struct btrfs_qgroup_inherit *inherit)
 559 {
 560         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 561         struct btrfs_trans_handle *trans;
 562         struct btrfs_key key;
 563         struct btrfs_root_item *root_item;
 564         struct btrfs_inode_item *inode_item;
 565         struct extent_buffer *leaf;
 566         struct btrfs_root *root = BTRFS_I(dir)->root;
 567         struct btrfs_root *new_root;
 568         struct btrfs_block_rsv block_rsv;
 569         struct timespec64 cur_time = current_time(dir);
 570         struct inode *inode;
 571         int ret;
 572         int err;
 573         u64 objectid;
 574         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 575         u64 index = 0;
 576         uuid_le new_uuid;
 577
 578         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 579         if (!root_item)
 580                 return -ENOMEM;
 581
 582         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 583         if (ret)
 584                 goto fail_free;
 585
 586         /*
 587          * Don't create subvolume whose level is not zero. Or qgroup will be
 588          * screwed up since it assumes subvolume qgroup's level to be 0.
 589          */
 590         if (btrfs_qgroup_level(objectid)) {
 591                 ret = -ENOSPC;
 592                 goto fail_free;
 593         }
 594
 595         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 596         /*
 597          * The same as the snapshot creation, please see the comment
 598          * of create_snapshot().
 599          */
 600         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 601         if (ret)
 602                 goto fail_free;
 603
 604         trans = btrfs_start_transaction(root, 0);
 605         if (IS_ERR(trans)) {
 606                 ret = PTR_ERR(trans);
 607                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 608                 goto fail_free;
 609         }
 610         trans->block_rsv = &block_rsv;
 611         trans->bytes_reserved = block_rsv.size;
 612
 613         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 614         if (ret)
 615                 goto fail;
 616
 617         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 618         if (IS_ERR(leaf)) {
 619                 ret = PTR_ERR(leaf);
 620                 goto fail;
 621         }
 622
 623         btrfs_mark_buffer_dirty(leaf);
 624
 625         inode_item = &root_item->inode;
 626         btrfs_set_stack_inode_generation(inode_item, 1);
 627         btrfs_set_stack_inode_size(inode_item, 3);
 628         btrfs_set_stack_inode_nlink(inode_item, 1);
 629         btrfs_set_stack_inode_nbytes(inode_item,
 630                                      fs_info->nodesize);
 631         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 632
 633         btrfs_set_root_flags(root_item, 0);
 634         btrfs_set_root_limit(root_item, 0);
 635         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 636
 637         btrfs_set_root_bytenr(root_item, leaf->start);
 638         btrfs_set_root_generation(root_item, trans->transid);
 639         btrfs_set_root_level(root_item, 0);
 640         btrfs_set_root_refs(root_item, 1);
 641         btrfs_set_root_used(root_item, leaf->len);
 642         btrfs_set_root_last_snapshot(root_item, 0);
 643
 644         btrfs_set_root_generation_v2(root_item,
 645                         btrfs_root_generation(root_item));
 646         uuid_le_gen(&new_uuid);
 647         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 648         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 649         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 650         root_item->ctime = root_item->otime;
 651         btrfs_set_root_ctransid(root_item, trans->transid);
 652         btrfs_set_root_otransid(root_item, trans->transid);
 653
 654         btrfs_tree_unlock(leaf);
 655         free_extent_buffer(leaf);
 656         leaf = NULL;
 657
 658         btrfs_set_root_dirid(root_item, new_dirid);
 659
 660         key.objectid = objectid;
 661         key.offset = 0;
 662         key.type = BTRFS_ROOT_ITEM_KEY;
 663         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 664                                 root_item);
 665         if (ret)
 666                 goto fail;
 667
 668         key.offset = (u64)-1;
 669         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 670         if (IS_ERR(new_root)) {
 671                 ret = PTR_ERR(new_root);
 672                 btrfs_abort_transaction(trans, ret);
 673                 goto fail;
 674         }
 675
 676         btrfs_record_root_in_trans(trans, new_root);
 677
 678         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 679         if (ret) {
 680                 /* We potentially lose an unused inode item here */
 681                 btrfs_abort_transaction(trans, ret);
 682                 goto fail;
 683         }
 684
 685         mutex_lock(&new_root->objectid_mutex);
 686         new_root->highest_objectid = new_dirid;
 687         mutex_unlock(&new_root->objectid_mutex);
 688
 689         /*
 690          * insert the directory item
 691          */
 692         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 693         if (ret) {
 694                 btrfs_abort_transaction(trans, ret);
 695                 goto fail;
 696         }
 697
 698         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 699                                     BTRFS_FT_DIR, index);
 700         if (ret) {
 701                 btrfs_abort_transaction(trans, ret);
 702                 goto fail;
 703         }
 704
 705         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 706         ret = btrfs_update_inode(trans, root, dir);
 707         if (ret) {
 708                 btrfs_abort_transaction(trans, ret);
 709                 goto fail;
 710         }
 711
 712         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 713                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 714         if (ret) {
 715                 btrfs_abort_transaction(trans, ret);
 716                 goto fail;
 717         }
 718
 719         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 720                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 721         if (ret)
 722                 btrfs_abort_transaction(trans, ret);
 723
 724 fail:
 725         kfree(root_item);
 726         trans->block_rsv = NULL;
 727         trans->bytes_reserved = 0;
 728         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 729
 730         if (async_transid) {
 731                 *async_transid = trans->transid;
 732                 err = btrfs_commit_transaction_async(trans, 1);
 733                 if (err)
 734                         err = btrfs_commit_transaction(trans);
 735         } else {
 736                 err = btrfs_commit_transaction(trans);
 737         }
 738         if (err && !ret)
 739                 ret = err;
 740
 741         if (!ret) {
 742                 inode = btrfs_lookup_dentry(dir, dentry);
 743                 if (IS_ERR(inode))
 744                         return PTR_ERR(inode);
 745                 d_instantiate(dentry, inode);
 746         }
 747         return ret;
 748
 749 fail_free:
 750         kfree(root_item);
 751         return ret;
 752 }
 753
 754 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 755                            struct dentry *dentry,
 756                            u64 *async_transid, bool readonly,
 757                            struct btrfs_qgroup_inherit *inherit)
 758 {
 759         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 760         struct inode *inode;
 761         struct btrfs_pending_snapshot *pending_snapshot;
 762         struct btrfs_trans_handle *trans;
 763         int ret;
 764         bool snapshot_force_cow = false;
 765
 766         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 767                 return -EINVAL;
 768
 769         if (atomic_read(&root->nr_swapfiles)) {
 770                 btrfs_warn(fs_info,
 771                            "cannot snapshot subvolume with active swapfile");
 772                 return -ETXTBSY;
 773         }
 774
 775         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 776         if (!pending_snapshot)
 777                 return -ENOMEM;
 778
 779         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 780                         GFP_KERNEL);
 781         pending_snapshot->path = btrfs_alloc_path();
 782         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 783                 ret = -ENOMEM;
 784                 goto free_pending;
 785         }
 786
 787         /*
 788          * Force new buffered writes to reserve space even when NOCOW is
 789          * possible. This is to avoid later writeback (running dealloc) to
 790          * fallback to COW mode and unexpectedly fail with ENOSPC.
 791          */
 792         atomic_inc(&root->will_be_snapshotted);
 793         smp_mb__after_atomic();
 794         /* wait for no snapshot writes */
 795         wait_event(root->subv_writers->wait,
 796                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 797
 798         ret = btrfs_start_delalloc_snapshot(root);
 799         if (ret)
 800                 goto dec_and_free;
 801
 802         /*
 803          * All previous writes have started writeback in NOCOW mode, so now
 804          * we force future writes to fallback to COW mode during snapshot
 805          * creation.
 806          */
 807         atomic_inc(&root->snapshot_force_cow);
 808         snapshot_force_cow = true;
 809
 810         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 811
 812         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 813                              BTRFS_BLOCK_RSV_TEMP);
 814         /*
 815          * 1 - parent dir inode
 816          * 2 - dir entries
 817          * 1 - root item
 818          * 2 - root ref/backref
 819          * 1 - root of snapshot
 820          * 1 - UUID item
 821          */
 822         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 823                                         &pending_snapshot->block_rsv, 8,
 824                                         false);
 825         if (ret)
 826                 goto dec_and_free;
 827
 828         pending_snapshot->dentry = dentry;
 829         pending_snapshot->root = root;
 830         pending_snapshot->readonly = readonly;
 831         pending_snapshot->dir = dir;
 832         pending_snapshot->inherit = inherit;
 833
 834         trans = btrfs_start_transaction(root, 0);
 835         if (IS_ERR(trans)) {
 836                 ret = PTR_ERR(trans);
 837                 goto fail;
 838         }
 839
 840         spin_lock(&fs_info->trans_lock);
 841         list_add(&pending_snapshot->list,
 842                  &trans->transaction->pending_snapshots);
 843         spin_unlock(&fs_info->trans_lock);
 844         if (async_transid) {
 845                 *async_transid = trans->transid;
 846                 ret = btrfs_commit_transaction_async(trans, 1);
 847                 if (ret)
 848                         ret = btrfs_commit_transaction(trans);
 849         } else {
 850                 ret = btrfs_commit_transaction(trans);
 851         }
 852         if (ret)
 853                 goto fail;
 854
 855         ret = pending_snapshot->error;
 856         if (ret)
 857                 goto fail;
 858
 859         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 860         if (ret)
 861                 goto fail;
 862
 863         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 864         if (IS_ERR(inode)) {
 865                 ret = PTR_ERR(inode);
 866                 goto fail;
 867         }
 868
 869         d_instantiate(dentry, inode);
 870         ret = 0;
 871 fail:
 872         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 873 dec_and_free:
 874         if (snapshot_force_cow)
 875                 atomic_dec(&root->snapshot_force_cow);
 876         if (atomic_dec_and_test(&root->will_be_snapshotted))
 877                 wake_up_var(&root->will_be_snapshotted);
 878 free_pending:
 879         kfree(pending_snapshot->root_item);
 880         btrfs_free_path(pending_snapshot->path);
 881         kfree(pending_snapshot);
 882
 883         return ret;
 884 }
 885
 886 /*  copy of may_delete in fs/namei.c()
 887  *      Check whether we can remove a link victim from directory dir, check
 888  *  whether the type of victim is right.
 889  *  1. We can't do it if dir is read-only (done in permission())
 890  *  2. We should have write and exec permissions on dir
 891  *  3. We can't remove anything from append-only dir
 892  *  4. We can't do anything with immutable dir (done in permission())
 893  *  5. If the sticky bit on dir is set we should either
 894  *      a. be owner of dir, or
 895  *      b. be owner of victim, or
 896  *      c. have CAP_FOWNER capability
 897  *  6. If the victim is append-only or immutable we can't do anything with
 898  *     links pointing to it.
 899  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 900  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 901  *  9. We can't remove a root or mountpoint.
 902  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 903  *     nfs_async_unlink().
 904  */
 905
 906 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 907 {
 908         int error;
 909
 910         if (d_really_is_negative(victim))
 911                 return -ENOENT;
 912
 913         BUG_ON(d_inode(victim->d_parent) != dir);
 914         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 915
 916         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 917         if (error)
 918                 return error;
 919         if (IS_APPEND(dir))
 920                 return -EPERM;
 921         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 922             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 923                 return -EPERM;
 924         if (isdir) {
 925                 if (!d_is_dir(victim))
 926                         return -ENOTDIR;
 927                 if (IS_ROOT(victim))
 928                         return -EBUSY;
 929         } else if (d_is_dir(victim))
 930                 return -EISDIR;
 931         if (IS_DEADDIR(dir))
 932                 return -ENOENT;
 933         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 934                 return -EBUSY;
 935         return 0;
 936 }
 937
 938 /* copy of may_create in fs/namei.c() */
 939 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 940 {
 941         if (d_really_is_positive(child))
 942                 return -EEXIST;
 943         if (IS_DEADDIR(dir))
 944                 return -ENOENT;
 945         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 946 }
 947
 948 /*
 949  * Create a new subvolume below @parent.  This is largely modeled after
 950  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 951  * inside this filesystem so it's quite a bit simpler.
 952  */
 953 static noinline int btrfs_mksubvol(const struct path *parent,
 954                                    const char *name, int namelen,
 955                                    struct btrfs_root *snap_src,
 956                                    u64 *async_transid, bool readonly,
 957                                    struct btrfs_qgroup_inherit *inherit)
 958 {
 959         struct inode *dir = d_inode(parent->dentry);
 960         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 961         struct dentry *dentry;
 962         int error;
 963
 964         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 965         if (error == -EINTR)
 966                 return error;
 967
 968         dentry = lookup_one_len(name, parent->dentry, namelen);
 969         error = PTR_ERR(dentry);
 970         if (IS_ERR(dentry))
 971                 goto out_unlock;
 972
 973         error = btrfs_may_create(dir, dentry);
 974         if (error)
 975                 goto out_dput;
 976
 977         /*
 978          * even if this name doesn't exist, we may get hash collisions.
 979          * check for them now when we can safely fail
 980          */
 981         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 982                                                dir->i_ino, name,
 983                                                namelen);
 984         if (error)
 985                 goto out_dput;
 986
 987         down_read(&fs_info->subvol_sem);
 988
 989         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 990                 goto out_up_read;
 991
 992         if (snap_src) {
 993                 error = create_snapshot(snap_src, dir, dentry,
 994                                         async_transid, readonly, inherit);
 995         } else {
 996                 error = create_subvol(dir, dentry, name, namelen,
 997                                       async_transid, inherit);
 998         }
 999         if (!error)
1000                 fsnotify_mkdir(dir, dentry);
1001 out_up_read:
1002         up_read(&fs_info->subvol_sem);
1003 out_dput:
1004         dput(dentry);
1005 out_unlock:
1006         inode_unlock(dir);
1007         return error;
1008 }
1009
1010 /*
1011  * When we're defragging a range, we don't want to kick it off again
1012  * if it is really just waiting for delalloc to send it down.
1013  * If we find a nice big extent or delalloc range for the bytes in the
1014  * file you want to defrag, we return 0 to let you know to skip this
1015  * part of the file
1016  */
1017 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1018 {
1019         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1020         struct extent_map *em = NULL;
1021         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1022         u64 end;
1023
1024         read_lock(&em_tree->lock);
1025         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1026         read_unlock(&em_tree->lock);
1027
1028         if (em) {
1029                 end = extent_map_end(em);
1030                 free_extent_map(em);
1031                 if (end - offset > thresh)
1032                         return 0;
1033         }
1034         /* if we already have a nice delalloc here, just stop */
1035         thresh /= 2;
1036         end = count_range_bits(io_tree, &offset, offset + thresh,
1037                                thresh, EXTENT_DELALLOC, 1);
1038         if (end >= thresh)
1039                 return 0;
1040         return 1;
1041 }
1042
1043 /*
1044  * helper function to walk through a file and find extents
1045  * newer than a specific transid, and smaller than thresh.
1046  *
1047  * This is used by the defragging code to find new and small
1048  * extents
1049  */
1050 static int find_new_extents(struct btrfs_root *root,
1051                             struct inode *inode, u64 newer_than,
1052                             u64 *off, u32 thresh)
1053 {
1054         struct btrfs_path *path;
1055         struct btrfs_key min_key;
1056         struct extent_buffer *leaf;
1057         struct btrfs_file_extent_item *extent;
1058         int type;
1059         int ret;
1060         u64 ino = btrfs_ino(BTRFS_I(inode));
1061
1062         path = btrfs_alloc_path();
1063         if (!path)
1064                 return -ENOMEM;
1065
1066         min_key.objectid = ino;
1067         min_key.type = BTRFS_EXTENT_DATA_KEY;
1068         min_key.offset = *off;
1069
1070         while (1) {
1071                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1072                 if (ret != 0)
1073                         goto none;
1074 process_slot:
1075                 if (min_key.objectid != ino)
1076                         goto none;
1077                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1078                         goto none;
1079
1080                 leaf = path->nodes[0];
1081                 extent = btrfs_item_ptr(leaf, path->slots[0],
1082                                         struct btrfs_file_extent_item);
1083
1084                 type = btrfs_file_extent_type(leaf, extent);
1085                 if (type == BTRFS_FILE_EXTENT_REG &&
1086                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1087                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1088                         *off = min_key.offset;
1089                         btrfs_free_path(path);
1090                         return 0;
1091                 }
1092
1093                 path->slots[0]++;
1094                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1095                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1096                         goto process_slot;
1097                 }
1098
1099                 if (min_key.offset == (u64)-1)
1100                         goto none;
1101
1102                 min_key.offset++;
1103                 btrfs_release_path(path);
1104         }
1105 none:
1106         btrfs_free_path(path);
1107         return -ENOENT;
1108 }
1109
1110 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1111 {
1112         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1113         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1114         struct extent_map *em;
1115         u64 len = PAGE_SIZE;
1116
1117         /*
1118          * hopefully we have this extent in the tree already, try without
1119          * the full extent lock
1120          */
1121         read_lock(&em_tree->lock);
1122         em = lookup_extent_mapping(em_tree, start, len);
1123         read_unlock(&em_tree->lock);
1124
1125         if (!em) {
1126                 struct extent_state *cached = NULL;
1127                 u64 end = start + len - 1;
1128
1129                 /* get the big lock and read metadata off disk */
1130                 lock_extent_bits(io_tree, start, end, &cached);
1131                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1132                 unlock_extent_cached(io_tree, start, end, &cached);
1133
1134                 if (IS_ERR(em))
1135                         return NULL;
1136         }
1137
1138         return em;
1139 }
1140
1141 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1142 {
1143         struct extent_map *next;
1144         bool ret = true;
1145
1146         /* this is the last extent */
1147         if (em->start + em->len >= i_size_read(inode))
1148                 return false;
1149
1150         next = defrag_lookup_extent(inode, em->start + em->len);
1151         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1152                 ret = false;
1153         else if ((em->block_start + em->block_len == next->block_start) &&
1154                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1155                 ret = false;
1156
1157         free_extent_map(next);
1158         return ret;
1159 }
1160
1161 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1162                                u64 *last_len, u64 *skip, u64 *defrag_end,
1163                                int compress)
1164 {
1165         struct extent_map *em;
1166         int ret = 1;
1167         bool next_mergeable = true;
1168         bool prev_mergeable = true;
1169
1170         /*
1171          * make sure that once we start defragging an extent, we keep on
1172          * defragging it
1173          */
1174         if (start < *defrag_end)
1175                 return 1;
1176
1177         *skip = 0;
1178
1179         em = defrag_lookup_extent(inode, start);
1180         if (!em)
1181                 return 0;
1182
1183         /* this will cover holes, and inline extents */
1184         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1185                 ret = 0;
1186                 goto out;
1187         }
1188
1189         if (!*defrag_end)
1190                 prev_mergeable = false;
1191
1192         next_mergeable = defrag_check_next_extent(inode, em);
1193         /*
1194          * we hit a real extent, if it is big or the next extent is not a
1195          * real extent, don't bother defragging it
1196          */
1197         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1198             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1199                 ret = 0;
1200 out:
1201         /*
1202          * last_len ends up being a counter of how many bytes we've defragged.
1203          * every time we choose not to defrag an extent, we reset *last_len
1204          * so that the next tiny extent will force a defrag.
1205          *
1206          * The end result of this is that tiny extents before a single big
1207          * extent will force at least part of that big extent to be defragged.
1208          */
1209         if (ret) {
1210                 *defrag_end = extent_map_end(em);
1211         } else {
1212                 *last_len = 0;
1213                 *skip = extent_map_end(em);
1214                 *defrag_end = 0;
1215         }
1216
1217         free_extent_map(em);
1218         return ret;
1219 }
1220
1221 /*
1222  * it doesn't do much good to defrag one or two pages
1223  * at a time.  This pulls in a nice chunk of pages
1224  * to COW and defrag.
1225  *
1226  * It also makes sure the delalloc code has enough
1227  * dirty data to avoid making new small extents as part
1228  * of the defrag
1229  *
1230  * It's a good idea to start RA on this range
1231  * before calling this.
1232  */
1233 static int cluster_pages_for_defrag(struct inode *inode,
1234                                     struct page **pages,
1235                                     unsigned long start_index,
1236                                     unsigned long num_pages)
1237 {
1238         unsigned long file_end;
1239         u64 isize = i_size_read(inode);
1240         u64 page_start;
1241         u64 page_end;
1242         u64 page_cnt;
1243         int ret;
1244         int i;
1245         int i_done;
1246         struct btrfs_ordered_extent *ordered;
1247         struct extent_state *cached_state = NULL;
1248         struct extent_io_tree *tree;
1249         struct extent_changeset *data_reserved = NULL;
1250         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1251
1252         file_end = (isize - 1) >> PAGE_SHIFT;
1253         if (!isize || start_index > file_end)
1254                 return 0;
1255
1256         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1257
1258         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1259                         start_index << PAGE_SHIFT,
1260                         page_cnt << PAGE_SHIFT);
1261         if (ret)
1262                 return ret;
1263         i_done = 0;
1264         tree = &BTRFS_I(inode)->io_tree;
1265
1266         /* step one, lock all the pages */
1267         for (i = 0; i < page_cnt; i++) {
1268                 struct page *page;
1269 again:
1270                 page = find_or_create_page(inode->i_mapping,
1271                                            start_index + i, mask);
1272                 if (!page)
1273                         break;
1274
1275                 page_start = page_offset(page);
1276                 page_end = page_start + PAGE_SIZE - 1;
1277                 while (1) {
1278                         lock_extent_bits(tree, page_start, page_end,
1279                                          &cached_state);
1280                         ordered = btrfs_lookup_ordered_extent(inode,
1281                                                               page_start);
1282                         unlock_extent_cached(tree, page_start, page_end,
1283                                              &cached_state);
1284                         if (!ordered)
1285                                 break;
1286
1287                         unlock_page(page);
1288                         btrfs_start_ordered_extent(inode, ordered, 1);
1289                         btrfs_put_ordered_extent(ordered);
1290                         lock_page(page);
1291                         /*
1292                          * we unlocked the page above, so we need check if
1293                          * it was released or not.
1294                          */
1295                         if (page->mapping != inode->i_mapping) {
1296                                 unlock_page(page);
1297                                 put_page(page);
1298                                 goto again;
1299                         }
1300                 }
1301
1302                 if (!PageUptodate(page)) {
1303                         btrfs_readpage(NULL, page);
1304                         lock_page(page);
1305                         if (!PageUptodate(page)) {
1306                                 unlock_page(page);
1307                                 put_page(page);
1308                                 ret = -EIO;
1309                                 break;
1310                         }
1311                 }
1312
1313                 if (page->mapping != inode->i_mapping) {
1314                         unlock_page(page);
1315                         put_page(page);
1316                         goto again;
1317                 }
1318
1319                 pages[i] = page;
1320                 i_done++;
1321         }
1322         if (!i_done || ret)
1323                 goto out;
1324
1325         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1326                 goto out;
1327
1328         /*
1329          * so now we have a nice long stream of locked
1330          * and up to date pages, lets wait on them
1331          */
1332         for (i = 0; i < i_done; i++)
1333                 wait_on_page_writeback(pages[i]);
1334
1335         page_start = page_offset(pages[0]);
1336         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1337
1338         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1339                          page_start, page_end - 1, &cached_state);
1340         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1341                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1342                           EXTENT_DEFRAG, 0, 0, &cached_state);
1343
1344         if (i_done != page_cnt) {
1345                 spin_lock(&BTRFS_I(inode)->lock);
1346                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1347                 spin_unlock(&BTRFS_I(inode)->lock);
1348                 btrfs_delalloc_release_space(inode, data_reserved,
1349                                 start_index << PAGE_SHIFT,
1350                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1351         }
1352
1353
1354         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1355                           &cached_state);
1356
1357         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1358                              page_start, page_end - 1, &cached_state);
1359
1360         for (i = 0; i < i_done; i++) {
1361                 clear_page_dirty_for_io(pages[i]);
1362                 ClearPageChecked(pages[i]);
1363                 set_page_extent_mapped(pages[i]);
1364                 set_page_dirty(pages[i]);
1365                 unlock_page(pages[i]);
1366                 put_page(pages[i]);
1367         }
1368         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1369         extent_changeset_free(data_reserved);
1370         return i_done;
1371 out:
1372         for (i = 0; i < i_done; i++) {
1373                 unlock_page(pages[i]);
1374                 put_page(pages[i]);
1375         }
1376         btrfs_delalloc_release_space(inode, data_reserved,
1377                         start_index << PAGE_SHIFT,
1378                         page_cnt << PAGE_SHIFT, true);
1379         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1380         extent_changeset_free(data_reserved);
1381         return ret;
1382
1383 }
1384
1385 int btrfs_defrag_file(struct inode *inode, struct file *file,
1386                       struct btrfs_ioctl_defrag_range_args *range,
1387                       u64 newer_than, unsigned long max_to_defrag)
1388 {
1389         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1390         struct btrfs_root *root = BTRFS_I(inode)->root;
1391         struct file_ra_state *ra = NULL;
1392         unsigned long last_index;
1393         u64 isize = i_size_read(inode);
1394         u64 last_len = 0;
1395         u64 skip = 0;
1396         u64 defrag_end = 0;
1397         u64 newer_off = range->start;
1398         unsigned long i;
1399         unsigned long ra_index = 0;
1400         int ret;
1401         int defrag_count = 0;
1402         int compress_type = BTRFS_COMPRESS_ZLIB;
1403         u32 extent_thresh = range->extent_thresh;
1404         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1405         unsigned long cluster = max_cluster;
1406         u64 new_align = ~((u64)SZ_128K - 1);
1407         struct page **pages = NULL;
1408         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1409
1410         if (isize == 0)
1411                 return 0;
1412
1413         if (range->start >= isize)
1414                 return -EINVAL;
1415
1416         if (do_compress) {
1417                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1418                         return -EINVAL;
1419                 if (range->compress_type)
1420                         compress_type = range->compress_type;
1421         }
1422
1423         if (extent_thresh == 0)
1424                 extent_thresh = SZ_256K;
1425
1426         /*
1427          * If we were not given a file, allocate a readahead context. As
1428          * readahead is just an optimization, defrag will work without it so
1429          * we don't error out.
1430          */
1431         if (!file) {
1432                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1433                 if (ra)
1434                         file_ra_state_init(ra, inode->i_mapping);
1435         } else {
1436                 ra = &file->f_ra;
1437         }
1438
1439         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1440         if (!pages) {
1441                 ret = -ENOMEM;
1442                 goto out_ra;
1443         }
1444
1445         /* find the last page to defrag */
1446         if (range->start + range->len > range->start) {
1447                 last_index = min_t(u64, isize - 1,
1448                          range->start + range->len - 1) >> PAGE_SHIFT;
1449         } else {
1450                 last_index = (isize - 1) >> PAGE_SHIFT;
1451         }
1452
1453         if (newer_than) {
1454                 ret = find_new_extents(root, inode, newer_than,
1455                                        &newer_off, SZ_64K);
1456                 if (!ret) {
1457                         range->start = newer_off;
1458                         /*
1459                          * we always align our defrag to help keep
1460                          * the extents in the file evenly spaced
1461                          */
1462                         i = (newer_off & new_align) >> PAGE_SHIFT;
1463                 } else
1464                         goto out_ra;
1465         } else {
1466                 i = range->start >> PAGE_SHIFT;
1467         }
1468         if (!max_to_defrag)
1469                 max_to_defrag = last_index - i + 1;
1470
1471         /*
1472          * make writeback starts from i, so the defrag range can be
1473          * written sequentially.
1474          */
1475         if (i < inode->i_mapping->writeback_index)
1476                 inode->i_mapping->writeback_index = i;
1477
1478         while (i <= last_index && defrag_count < max_to_defrag &&
1479                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1480                 /*
1481                  * make sure we stop running if someone unmounts
1482                  * the FS
1483                  */
1484                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1485                         break;
1486
1487                 if (btrfs_defrag_cancelled(fs_info)) {
1488                         btrfs_debug(fs_info, "defrag_file cancelled");
1489                         ret = -EAGAIN;
1490                         break;
1491                 }
1492
1493                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1494                                          extent_thresh, &last_len, &skip,
1495                                          &defrag_end, do_compress)){
1496                         unsigned long next;
1497                         /*
1498                          * the should_defrag function tells us how much to skip
1499                          * bump our counter by the suggested amount
1500                          */
1501                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1502                         i = max(i + 1, next);
1503                         continue;
1504                 }
1505
1506                 if (!newer_than) {
1507                         cluster = (PAGE_ALIGN(defrag_end) >>
1508                                    PAGE_SHIFT) - i;
1509                         cluster = min(cluster, max_cluster);
1510                 } else {
1511                         cluster = max_cluster;
1512                 }
1513
1514                 if (i + cluster > ra_index) {
1515                         ra_index = max(i, ra_index);
1516                         if (ra)
1517                                 page_cache_sync_readahead(inode->i_mapping, ra,
1518                                                 file, ra_index, cluster);
1519                         ra_index += cluster;
1520                 }
1521
1522                 inode_lock(inode);
1523                 if (IS_SWAPFILE(inode)) {
1524                         ret = -ETXTBSY;
1525                 } else {
1526                         if (do_compress)
1527                                 BTRFS_I(inode)->defrag_compress = compress_type;
1528                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1529                 }
1530                 if (ret < 0) {
1531                         inode_unlock(inode);
1532                         goto out_ra;
1533                 }
1534
1535                 defrag_count += ret;
1536                 balance_dirty_pages_ratelimited(inode->i_mapping);
1537                 inode_unlock(inode);
1538
1539                 if (newer_than) {
1540                         if (newer_off == (u64)-1)
1541                                 break;
1542
1543                         if (ret > 0)
1544                                 i += ret;
1545
1546                         newer_off = max(newer_off + 1,
1547                                         (u64)i << PAGE_SHIFT);
1548
1549                         ret = find_new_extents(root, inode, newer_than,
1550                                                &newer_off, SZ_64K);
1551                         if (!ret) {
1552                                 range->start = newer_off;
1553                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1554                         } else {
1555                                 break;
1556                         }
1557                 } else {
1558                         if (ret > 0) {
1559                                 i += ret;
1560                                 last_len += ret << PAGE_SHIFT;
1561                         } else {
1562                                 i++;
1563                                 last_len = 0;
1564                         }
1565                 }
1566         }
1567
1568         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1569                 filemap_flush(inode->i_mapping);
1570                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1571                              &BTRFS_I(inode)->runtime_flags))
1572                         filemap_flush(inode->i_mapping);
1573         }
1574
1575         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1576                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1577         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1578                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1579         }
1580
1581         ret = defrag_count;
1582
1583 out_ra:
1584         if (do_compress) {
1585                 inode_lock(inode);
1586                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1587                 inode_unlock(inode);
1588         }
1589         if (!file)
1590                 kfree(ra);
1591         kfree(pages);
1592         return ret;
1593 }
1594
1595 static noinline int btrfs_ioctl_resize(struct file *file,
1596                                         void __user *arg)
1597 {
1598         struct inode *inode = file_inode(file);
1599         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1600         u64 new_size;
1601         u64 old_size;
1602         u64 devid = 1;
1603         struct btrfs_root *root = BTRFS_I(inode)->root;
1604         struct btrfs_ioctl_vol_args *vol_args;
1605         struct btrfs_trans_handle *trans;
1606         struct btrfs_device *device = NULL;
1607         char *sizestr;
1608         char *retptr;
1609         char *devstr = NULL;
1610         int ret = 0;
1611         int mod = 0;
1612
1613         if (!capable(CAP_SYS_ADMIN))
1614                 return -EPERM;
1615
1616         ret = mnt_want_write_file(file);
1617         if (ret)
1618                 return ret;
1619
1620         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1621                 mnt_drop_write_file(file);
1622                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1623         }
1624
1625         vol_args = memdup_user(arg, sizeof(*vol_args));
1626         if (IS_ERR(vol_args)) {
1627                 ret = PTR_ERR(vol_args);
1628                 goto out;
1629         }
1630
1631         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1632
1633         sizestr = vol_args->name;
1634         devstr = strchr(sizestr, ':');
1635         if (devstr) {
1636                 sizestr = devstr + 1;
1637                 *devstr = '\0';
1638                 devstr = vol_args->name;
1639                 ret = kstrtoull(devstr, 10, &devid);
1640                 if (ret)
1641                         goto out_free;
1642                 if (!devid) {
1643                         ret = -EINVAL;
1644                         goto out_free;
1645                 }
1646                 btrfs_info(fs_info, "resizing devid %llu", devid);
1647         }
1648
1649         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1650         if (!device) {
1651                 btrfs_info(fs_info, "resizer unable to find device %llu",
1652                            devid);
1653                 ret = -ENODEV;
1654                 goto out_free;
1655         }
1656
1657         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1658                 btrfs_info(fs_info,
1659                            "resizer unable to apply on readonly device %llu",
1660                        devid);
1661                 ret = -EPERM;
1662                 goto out_free;
1663         }
1664
1665         if (!strcmp(sizestr, "max"))
1666                 new_size = device->bdev->bd_inode->i_size;
1667         else {
1668                 if (sizestr[0] == '-') {
1669                         mod = -1;
1670                         sizestr++;
1671                 } else if (sizestr[0] == '+') {
1672                         mod = 1;
1673                         sizestr++;
1674                 }
1675                 new_size = memparse(sizestr, &retptr);
1676                 if (*retptr != '\0' || new_size == 0) {
1677                         ret = -EINVAL;
1678                         goto out_free;
1679                 }
1680         }
1681
1682         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1683                 ret = -EPERM;
1684                 goto out_free;
1685         }
1686
1687         old_size = btrfs_device_get_total_bytes(device);
1688
1689         if (mod < 0) {
1690                 if (new_size > old_size) {
1691                         ret = -EINVAL;
1692                         goto out_free;
1693                 }
1694                 new_size = old_size - new_size;
1695         } else if (mod > 0) {
1696                 if (new_size > ULLONG_MAX - old_size) {
1697                         ret = -ERANGE;
1698                         goto out_free;
1699                 }
1700                 new_size = old_size + new_size;
1701         }
1702
1703         if (new_size < SZ_256M) {
1704                 ret = -EINVAL;
1705                 goto out_free;
1706         }
1707         if (new_size > device->bdev->bd_inode->i_size) {
1708                 ret = -EFBIG;
1709                 goto out_free;
1710         }
1711
1712         new_size = round_down(new_size, fs_info->sectorsize);
1713
1714         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1715                           rcu_str_deref(device->name), new_size);
1716
1717         if (new_size > old_size) {
1718                 trans = btrfs_start_transaction(root, 0);
1719                 if (IS_ERR(trans)) {
1720                         ret = PTR_ERR(trans);
1721                         goto out_free;
1722                 }
1723                 ret = btrfs_grow_device(trans, device, new_size);
1724                 btrfs_commit_transaction(trans);
1725         } else if (new_size < old_size) {
1726                 ret = btrfs_shrink_device(device, new_size);
1727         } /* equal, nothing need to do */
1728
1729 out_free:
1730         kfree(vol_args);
1731 out:
1732         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1733         mnt_drop_write_file(file);
1734         return ret;
1735 }
1736
1737 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1738                                 const char *name, unsigned long fd, int subvol,
1739                                 u64 *transid, bool readonly,
1740                                 struct btrfs_qgroup_inherit *inherit)
1741 {
1742         int namelen;
1743         int ret = 0;
1744
1745         if (!S_ISDIR(file_inode(file)->i_mode))
1746                 return -ENOTDIR;
1747
1748         ret = mnt_want_write_file(file);
1749         if (ret)
1750                 goto out;
1751
1752         namelen = strlen(name);
1753         if (strchr(name, '/')) {
1754                 ret = -EINVAL;
1755                 goto out_drop_write;
1756         }
1757
1758         if (name[0] == '.' &&
1759            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1760                 ret = -EEXIST;
1761                 goto out_drop_write;
1762         }
1763
1764         if (subvol) {
1765                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1766                                      NULL, transid, readonly, inherit);
1767         } else {
1768                 struct fd src = fdget(fd);
1769                 struct inode *src_inode;
1770                 if (!src.file) {
1771                         ret = -EINVAL;
1772                         goto out_drop_write;
1773                 }
1774
1775                 src_inode = file_inode(src.file);
1776                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1777                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1778                                    "Snapshot src from another FS");
1779                         ret = -EXDEV;
1780                 } else if (!inode_owner_or_capable(src_inode)) {
1781                         /*
1782                          * Subvolume creation is not restricted, but snapshots
1783                          * are limited to own subvolumes only
1784                          */
1785                         ret = -EPERM;
1786                 } else {
1787                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1788                                              BTRFS_I(src_inode)->root,
1789                                              transid, readonly, inherit);
1790                 }
1791                 fdput(src);
1792         }
1793 out_drop_write:
1794         mnt_drop_write_file(file);
1795 out:
1796         return ret;
1797 }
1798
1799 static noinline int btrfs_ioctl_snap_create(struct file *file,
1800                                             void __user *arg, int subvol)
1801 {
1802         struct btrfs_ioctl_vol_args *vol_args;
1803         int ret;
1804
1805         if (!S_ISDIR(file_inode(file)->i_mode))
1806                 return -ENOTDIR;
1807
1808         vol_args = memdup_user(arg, sizeof(*vol_args));
1809         if (IS_ERR(vol_args))
1810                 return PTR_ERR(vol_args);
1811         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1812
1813         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1814                                               vol_args->fd, subvol,
1815                                               NULL, false, NULL);
1816
1817         kfree(vol_args);
1818         return ret;
1819 }
1820
1821 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1822                                                void __user *arg, int subvol)
1823 {
1824         struct btrfs_ioctl_vol_args_v2 *vol_args;
1825         int ret;
1826         u64 transid = 0;
1827         u64 *ptr = NULL;
1828         bool readonly = false;
1829         struct btrfs_qgroup_inherit *inherit = NULL;
1830
1831         if (!S_ISDIR(file_inode(file)->i_mode))
1832                 return -ENOTDIR;
1833
1834         vol_args = memdup_user(arg, sizeof(*vol_args));
1835         if (IS_ERR(vol_args))
1836                 return PTR_ERR(vol_args);
1837         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1838
1839         if (vol_args->flags &
1840             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1841               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1842                 ret = -EOPNOTSUPP;
1843                 goto free_args;
1844         }
1845
1846         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1847                 struct inode *inode = file_inode(file);
1848                 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1849
1850                 btrfs_warn(fs_info,
1851 "SNAP_CREATE_V2 ioctl with CREATE_ASYNC is deprecated and will be removed in kernel 5.7");
1852
1853                 ptr = &transid;
1854         }
1855         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1856                 readonly = true;
1857         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1858                 if (vol_args->size > PAGE_SIZE) {
1859                         ret = -EINVAL;
1860                         goto free_args;
1861                 }
1862                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1863                 if (IS_ERR(inherit)) {
1864                         ret = PTR_ERR(inherit);
1865                         goto free_args;
1866                 }
1867         }
1868
1869         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1870                                               vol_args->fd, subvol, ptr,
1871                                               readonly, inherit);
1872         if (ret)
1873                 goto free_inherit;
1874
1875         if (ptr && copy_to_user(arg +
1876                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1877                                         transid),
1878                                 ptr, sizeof(*ptr)))
1879                 ret = -EFAULT;
1880
1881 free_inherit:
1882         kfree(inherit);
1883 free_args:
1884         kfree(vol_args);
1885         return ret;
1886 }
1887
1888 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1889                                                 void __user *arg)
1890 {
1891         struct inode *inode = file_inode(file);
1892         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1893         struct btrfs_root *root = BTRFS_I(inode)->root;
1894         int ret = 0;
1895         u64 flags = 0;
1896
1897         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1898                 return -EINVAL;
1899
1900         down_read(&fs_info->subvol_sem);
1901         if (btrfs_root_readonly(root))
1902                 flags |= BTRFS_SUBVOL_RDONLY;
1903         up_read(&fs_info->subvol_sem);
1904
1905         if (copy_to_user(arg, &flags, sizeof(flags)))
1906                 ret = -EFAULT;
1907
1908         return ret;
1909 }
1910
1911 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1912                                               void __user *arg)
1913 {
1914         struct inode *inode = file_inode(file);
1915         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1916         struct btrfs_root *root = BTRFS_I(inode)->root;
1917         struct btrfs_trans_handle *trans;
1918         u64 root_flags;
1919         u64 flags;
1920         int ret = 0;
1921
1922         if (!inode_owner_or_capable(inode))
1923                 return -EPERM;
1924
1925         ret = mnt_want_write_file(file);
1926         if (ret)
1927                 goto out;
1928
1929         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1930                 ret = -EINVAL;
1931                 goto out_drop_write;
1932         }
1933
1934         if (copy_from_user(&flags, arg, sizeof(flags))) {
1935                 ret = -EFAULT;
1936                 goto out_drop_write;
1937         }
1938
1939         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1940                 ret = -EINVAL;
1941                 goto out_drop_write;
1942         }
1943
1944         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1945                 ret = -EOPNOTSUPP;
1946                 goto out_drop_write;
1947         }
1948
1949         down_write(&fs_info->subvol_sem);
1950
1951         /* nothing to do */
1952         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1953                 goto out_drop_sem;
1954
1955         root_flags = btrfs_root_flags(&root->root_item);
1956         if (flags & BTRFS_SUBVOL_RDONLY) {
1957                 btrfs_set_root_flags(&root->root_item,
1958                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1959         } else {
1960                 /*
1961                  * Block RO -> RW transition if this subvolume is involved in
1962                  * send
1963                  */
1964                 spin_lock(&root->root_item_lock);
1965                 if (root->send_in_progress == 0) {
1966                         btrfs_set_root_flags(&root->root_item,
1967                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1968                         spin_unlock(&root->root_item_lock);
1969                 } else {
1970                         spin_unlock(&root->root_item_lock);
1971                         btrfs_warn(fs_info,
1972                                    "Attempt to set subvolume %llu read-write during send",
1973                                    root->root_key.objectid);
1974                         ret = -EPERM;
1975                         goto out_drop_sem;
1976                 }
1977         }
1978
1979         trans = btrfs_start_transaction(root, 1);
1980         if (IS_ERR(trans)) {
1981                 ret = PTR_ERR(trans);
1982                 goto out_reset;
1983         }
1984
1985         ret = btrfs_update_root(trans, fs_info->tree_root,
1986                                 &root->root_key, &root->root_item);
1987         if (ret < 0) {
1988                 btrfs_end_transaction(trans);
1989                 goto out_reset;
1990         }
1991
1992         ret = btrfs_commit_transaction(trans);
1993
1994 out_reset:
1995         if (ret)
1996                 btrfs_set_root_flags(&root->root_item, root_flags);
1997 out_drop_sem:
1998         up_write(&fs_info->subvol_sem);
1999 out_drop_write:
2000         mnt_drop_write_file(file);
2001 out:
2002         return ret;
2003 }
2004
2005 static noinline int key_in_sk(struct btrfs_key *key,
2006                               struct btrfs_ioctl_search_key *sk)
2007 {
2008         struct btrfs_key test;
2009         int ret;
2010
2011         test.objectid = sk->min_objectid;
2012         test.type = sk->min_type;
2013         test.offset = sk->min_offset;
2014
2015         ret = btrfs_comp_cpu_keys(key, &test);
2016         if (ret < 0)
2017                 return 0;
2018
2019         test.objectid = sk->max_objectid;
2020         test.type = sk->max_type;
2021         test.offset = sk->max_offset;
2022
2023         ret = btrfs_comp_cpu_keys(key, &test);
2024         if (ret > 0)
2025                 return 0;
2026         return 1;
2027 }
2028
2029 static noinline int copy_to_sk(struct btrfs_path *path,
2030                                struct btrfs_key *key,
2031                                struct btrfs_ioctl_search_key *sk,
2032                                size_t *buf_size,
2033                                char __user *ubuf,
2034                                unsigned long *sk_offset,
2035                                int *num_found)
2036 {
2037         u64 found_transid;
2038         struct extent_buffer *leaf;
2039         struct btrfs_ioctl_search_header sh;
2040         struct btrfs_key test;
2041         unsigned long item_off;
2042         unsigned long item_len;
2043         int nritems;
2044         int i;
2045         int slot;
2046         int ret = 0;
2047
2048         leaf = path->nodes[0];
2049         slot = path->slots[0];
2050         nritems = btrfs_header_nritems(leaf);
2051
2052         if (btrfs_header_generation(leaf) > sk->max_transid) {
2053                 i = nritems;
2054                 goto advance_key;
2055         }
2056         found_transid = btrfs_header_generation(leaf);
2057
2058         for (i = slot; i < nritems; i++) {
2059                 item_off = btrfs_item_ptr_offset(leaf, i);
2060                 item_len = btrfs_item_size_nr(leaf, i);
2061
2062                 btrfs_item_key_to_cpu(leaf, key, i);
2063                 if (!key_in_sk(key, sk))
2064                         continue;
2065
2066                 if (sizeof(sh) + item_len > *buf_size) {
2067                         if (*num_found) {
2068                                 ret = 1;
2069                                 goto out;
2070                         }
2071
2072                         /*
2073                          * return one empty item back for v1, which does not
2074                          * handle -EOVERFLOW
2075                          */
2076
2077                         *buf_size = sizeof(sh) + item_len;
2078                         item_len = 0;
2079                         ret = -EOVERFLOW;
2080                 }
2081
2082                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2083                         ret = 1;
2084                         goto out;
2085                 }
2086
2087                 sh.objectid = key->objectid;
2088                 sh.offset = key->offset;
2089                 sh.type = key->type;
2090                 sh.len = item_len;
2091                 sh.transid = found_transid;
2092
2093                 /* copy search result header */
2094                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2095                         ret = -EFAULT;
2096                         goto out;
2097                 }
2098
2099                 *sk_offset += sizeof(sh);
2100
2101                 if (item_len) {
2102                         char __user *up = ubuf + *sk_offset;
2103                         /* copy the item */
2104                         if (read_extent_buffer_to_user(leaf, up,
2105                                                        item_off, item_len)) {
2106                                 ret = -EFAULT;
2107                                 goto out;
2108                         }
2109
2110                         *sk_offset += item_len;
2111                 }
2112                 (*num_found)++;
2113
2114                 if (ret) /* -EOVERFLOW from above */
2115                         goto out;
2116
2117                 if (*num_found >= sk->nr_items) {
2118                         ret = 1;
2119                         goto out;
2120                 }
2121         }
2122 advance_key:
2123         ret = 0;
2124         test.objectid = sk->max_objectid;
2125         test.type = sk->max_type;
2126         test.offset = sk->max_offset;
2127         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2128                 ret = 1;
2129         else if (key->offset < (u64)-1)
2130                 key->offset++;
2131         else if (key->type < (u8)-1) {
2132                 key->offset = 0;
2133                 key->type++;
2134         } else if (key->objectid < (u64)-1) {
2135                 key->offset = 0;
2136                 key->type = 0;
2137                 key->objectid++;
2138         } else
2139                 ret = 1;
2140 out:
2141         /*
2142          *  0: all items from this leaf copied, continue with next
2143          *  1: * more items can be copied, but unused buffer is too small
2144          *     * all items were found
2145          *     Either way, it will stops the loop which iterates to the next
2146          *     leaf
2147          *  -EOVERFLOW: item was to large for buffer
2148          *  -EFAULT: could not copy extent buffer back to userspace
2149          */
2150         return ret;
2151 }
2152
2153 static noinline int search_ioctl(struct inode *inode,
2154                                  struct btrfs_ioctl_search_key *sk,
2155                                  size_t *buf_size,
2156                                  char __user *ubuf)
2157 {
2158         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2159         struct btrfs_root *root;
2160         struct btrfs_key key;
2161         struct btrfs_path *path;
2162         int ret;
2163         int num_found = 0;
2164         unsigned long sk_offset = 0;
2165
2166         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2167                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2168                 return -EOVERFLOW;
2169         }
2170
2171         path = btrfs_alloc_path();
2172         if (!path)
2173                 return -ENOMEM;
2174
2175         if (sk->tree_id == 0) {
2176                 /* search the root of the inode that was passed */
2177                 root = BTRFS_I(inode)->root;
2178         } else {
2179                 key.objectid = sk->tree_id;
2180                 key.type = BTRFS_ROOT_ITEM_KEY;
2181                 key.offset = (u64)-1;
2182                 root = btrfs_read_fs_root_no_name(info, &key);
2183                 if (IS_ERR(root)) {
2184                         btrfs_free_path(path);
2185                         return PTR_ERR(root);
2186                 }
2187         }
2188
2189         key.objectid = sk->min_objectid;
2190         key.type = sk->min_type;
2191         key.offset = sk->min_offset;
2192
2193         while (1) {
2194                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2195                 if (ret != 0) {
2196                         if (ret > 0)
2197                                 ret = 0;
2198                         goto err;
2199                 }
2200                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2201                                  &sk_offset, &num_found);
2202                 btrfs_release_path(path);
2203                 if (ret)
2204                         break;
2205
2206         }
2207         if (ret > 0)
2208                 ret = 0;
2209 err:
2210         sk->nr_items = num_found;
2211         btrfs_free_path(path);
2212         return ret;
2213 }
2214
2215 static noinline int btrfs_ioctl_tree_search(struct file *file,
2216                                            void __user *argp)
2217 {
2218         struct btrfs_ioctl_search_args __user *uargs;
2219         struct btrfs_ioctl_search_key sk;
2220         struct inode *inode;
2221         int ret;
2222         size_t buf_size;
2223
2224         if (!capable(CAP_SYS_ADMIN))
2225                 return -EPERM;
2226
2227         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2228
2229         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2230                 return -EFAULT;
2231
2232         buf_size = sizeof(uargs->buf);
2233
2234         inode = file_inode(file);
2235         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2236
2237         /*
2238          * In the origin implementation an overflow is handled by returning a
2239          * search header with a len of zero, so reset ret.
2240          */
2241         if (ret == -EOVERFLOW)
2242                 ret = 0;
2243
2244         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2245                 ret = -EFAULT;
2246         return ret;
2247 }
2248
2249 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2250                                                void __user *argp)
2251 {
2252         struct btrfs_ioctl_search_args_v2 __user *uarg;
2253         struct btrfs_ioctl_search_args_v2 args;
2254         struct inode *inode;
2255         int ret;
2256         size_t buf_size;
2257         const size_t buf_limit = SZ_16M;
2258
2259         if (!capable(CAP_SYS_ADMIN))
2260                 return -EPERM;
2261
2262         /* copy search header and buffer size */
2263         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2264         if (copy_from_user(&args, uarg, sizeof(args)))
2265                 return -EFAULT;
2266
2267         buf_size = args.buf_size;
2268
2269         /* limit result size to 16MB */
2270         if (buf_size > buf_limit)
2271                 buf_size = buf_limit;
2272
2273         inode = file_inode(file);
2274         ret = search_ioctl(inode, &args.key, &buf_size,
2275                            (char __user *)(&uarg->buf[0]));
2276         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2277                 ret = -EFAULT;
2278         else if (ret == -EOVERFLOW &&
2279                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2280                 ret = -EFAULT;
2281
2282         return ret;
2283 }
2284
2285 /*
2286  * Search INODE_REFs to identify path name of 'dirid' directory
2287  * in a 'tree_id' tree. and sets path name to 'name'.
2288  */
2289 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2290                                 u64 tree_id, u64 dirid, char *name)
2291 {
2292         struct btrfs_root *root;
2293         struct btrfs_key key;
2294         char *ptr;
2295         int ret = -1;
2296         int slot;
2297         int len;
2298         int total_len = 0;
2299         struct btrfs_inode_ref *iref;
2300         struct extent_buffer *l;
2301         struct btrfs_path *path;
2302
2303         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2304                 name[0]='\0';
2305                 return 0;
2306         }
2307
2308         path = btrfs_alloc_path();
2309         if (!path)
2310                 return -ENOMEM;
2311
2312         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2313
2314         key.objectid = tree_id;
2315         key.type = BTRFS_ROOT_ITEM_KEY;
2316         key.offset = (u64)-1;
2317         root = btrfs_read_fs_root_no_name(info, &key);
2318         if (IS_ERR(root)) {
2319                 ret = PTR_ERR(root);
2320                 goto out;
2321         }
2322
2323         key.objectid = dirid;
2324         key.type = BTRFS_INODE_REF_KEY;
2325         key.offset = (u64)-1;
2326
2327         while (1) {
2328                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2329                 if (ret < 0)
2330                         goto out;
2331                 else if (ret > 0) {
2332                         ret = btrfs_previous_item(root, path, dirid,
2333                                                   BTRFS_INODE_REF_KEY);
2334                         if (ret < 0)
2335                                 goto out;
2336                         else if (ret > 0) {
2337                                 ret = -ENOENT;
2338                                 goto out;
2339                         }
2340                 }
2341
2342                 l = path->nodes[0];
2343                 slot = path->slots[0];
2344                 btrfs_item_key_to_cpu(l, &key, slot);
2345
2346                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2347                 len = btrfs_inode_ref_name_len(l, iref);
2348                 ptr -= len + 1;
2349                 total_len += len + 1;
2350                 if (ptr < name) {
2351                         ret = -ENAMETOOLONG;
2352                         goto out;
2353                 }
2354
2355                 *(ptr + len) = '/';
2356                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2357
2358                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2359                         break;
2360
2361                 btrfs_release_path(path);
2362                 key.objectid = key.offset;
2363                 key.offset = (u64)-1;
2364                 dirid = key.objectid;
2365         }
2366         memmove(name, ptr, total_len);
2367         name[total_len] = '\0';
2368         ret = 0;
2369 out:
2370         btrfs_free_path(path);
2371         return ret;
2372 }
2373
2374 static int btrfs_search_path_in_tree_user(struct inode *inode,
2375                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2376 {
2377         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2378         struct super_block *sb = inode->i_sb;
2379         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2380         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2381         u64 dirid = args->dirid;
2382         unsigned long item_off;
2383         unsigned long item_len;
2384         struct btrfs_inode_ref *iref;
2385         struct btrfs_root_ref *rref;
2386         struct btrfs_root *root;
2387         struct btrfs_path *path;
2388         struct btrfs_key key, key2;
2389         struct extent_buffer *leaf;
2390         struct inode *temp_inode;
2391         char *ptr;
2392         int slot;
2393         int len;
2394         int total_len = 0;
2395         int ret;
2396
2397         path = btrfs_alloc_path();
2398         if (!path)
2399                 return -ENOMEM;
2400
2401         /*
2402          * If the bottom subvolume does not exist directly under upper_limit,
2403          * construct the path in from the bottom up.
2404          */
2405         if (dirid != upper_limit.objectid) {
2406                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2407
2408                 key.objectid = treeid;
2409                 key.type = BTRFS_ROOT_ITEM_KEY;
2410                 key.offset = (u64)-1;
2411                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2412                 if (IS_ERR(root)) {
2413                         ret = PTR_ERR(root);
2414                         goto out;
2415                 }
2416
2417                 key.objectid = dirid;
2418                 key.type = BTRFS_INODE_REF_KEY;
2419                 key.offset = (u64)-1;
2420                 while (1) {
2421                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2422                         if (ret < 0) {
2423                                 goto out;
2424                         } else if (ret > 0) {
2425                                 ret = btrfs_previous_item(root, path, dirid,
2426                                                           BTRFS_INODE_REF_KEY);
2427                                 if (ret < 0) {
2428                                         goto out;
2429                                 } else if (ret > 0) {
2430                                         ret = -ENOENT;
2431                                         goto out;
2432                                 }
2433                         }
2434
2435                         leaf = path->nodes[0];
2436                         slot = path->slots[0];
2437                         btrfs_item_key_to_cpu(leaf, &key, slot);
2438
2439                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2440                         len = btrfs_inode_ref_name_len(leaf, iref);
2441                         ptr -= len + 1;
2442                         total_len += len + 1;
2443                         if (ptr < args->path) {
2444                                 ret = -ENAMETOOLONG;
2445                                 goto out;
2446                         }
2447
2448                         *(ptr + len) = '/';
2449                         read_extent_buffer(leaf, ptr,
2450                                         (unsigned long)(iref + 1), len);
2451
2452                         /* Check the read+exec permission of this directory */
2453                         ret = btrfs_previous_item(root, path, dirid,
2454                                                   BTRFS_INODE_ITEM_KEY);
2455                         if (ret < 0) {
2456                                 goto out;
2457                         } else if (ret > 0) {
2458                                 ret = -ENOENT;
2459                                 goto out;
2460                         }
2461
2462                         leaf = path->nodes[0];
2463                         slot = path->slots[0];
2464                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2465                         if (key2.objectid != dirid) {
2466                                 ret = -ENOENT;
2467                                 goto out;
2468                         }
2469
2470                         temp_inode = btrfs_iget(sb, &key2, root);
2471                         if (IS_ERR(temp_inode)) {
2472                                 ret = PTR_ERR(temp_inode);
2473                                 goto out;
2474                         }
2475                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2476                         iput(temp_inode);
2477                         if (ret) {
2478                                 ret = -EACCES;
2479                                 goto out;
2480                         }
2481
2482                         if (key.offset == upper_limit.objectid)
2483                                 break;
2484                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2485                                 ret = -EACCES;
2486                                 goto out;
2487                         }
2488
2489                         btrfs_release_path(path);
2490                         key.objectid = key.offset;
2491                         key.offset = (u64)-1;
2492                         dirid = key.objectid;
2493                 }
2494
2495                 memmove(args->path, ptr, total_len);
2496                 args->path[total_len] = '\0';
2497                 btrfs_release_path(path);
2498         }
2499
2500         /* Get the bottom subvolume's name from ROOT_REF */
2501         root = fs_info->tree_root;
2502         key.objectid = treeid;
2503         key.type = BTRFS_ROOT_REF_KEY;
2504         key.offset = args->treeid;
2505         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2506         if (ret < 0) {
2507                 goto out;
2508         } else if (ret > 0) {
2509                 ret = -ENOENT;
2510                 goto out;
2511         }
2512
2513         leaf = path->nodes[0];
2514         slot = path->slots[0];
2515         btrfs_item_key_to_cpu(leaf, &key, slot);
2516
2517         item_off = btrfs_item_ptr_offset(leaf, slot);
2518         item_len = btrfs_item_size_nr(leaf, slot);
2519         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2520         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2521         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2522                 ret = -EINVAL;
2523                 goto out;
2524         }
2525
2526         /* Copy subvolume's name */
2527         item_off += sizeof(struct btrfs_root_ref);
2528         item_len -= sizeof(struct btrfs_root_ref);
2529         read_extent_buffer(leaf, args->name, item_off, item_len);
2530         args->name[item_len] = 0;
2531
2532 out:
2533         btrfs_free_path(path);
2534         return ret;
2535 }
2536
2537 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2538                                            void __user *argp)
2539 {
2540         struct btrfs_ioctl_ino_lookup_args *args;
2541         struct inode *inode;
2542         int ret = 0;
2543
2544         args = memdup_user(argp, sizeof(*args));
2545         if (IS_ERR(args))
2546                 return PTR_ERR(args);
2547
2548         inode = file_inode(file);
2549
2550         /*
2551          * Unprivileged query to obtain the containing subvolume root id. The
2552          * path is reset so it's consistent with btrfs_search_path_in_tree.
2553          */
2554         if (args->treeid == 0)
2555                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2556
2557         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2558                 args->name[0] = 0;
2559                 goto out;
2560         }
2561
2562         if (!capable(CAP_SYS_ADMIN)) {
2563                 ret = -EPERM;
2564                 goto out;
2565         }
2566
2567         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2568                                         args->treeid, args->objectid,
2569                                         args->name);
2570
2571 out:
2572         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2573                 ret = -EFAULT;
2574
2575         kfree(args);
2576         return ret;
2577 }
2578
2579 /*
2580  * Version of ino_lookup ioctl (unprivileged)
2581  *
2582  * The main differences from ino_lookup ioctl are:
2583  *
2584  *   1. Read + Exec permission will be checked using inode_permission() during
2585  *      path construction. -EACCES will be returned in case of failure.
2586  *   2. Path construction will be stopped at the inode number which corresponds
2587  *      to the fd with which this ioctl is called. If constructed path does not
2588  *      exist under fd's inode, -EACCES will be returned.
2589  *   3. The name of bottom subvolume is also searched and filled.
2590  */
2591 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2592 {
2593         struct btrfs_ioctl_ino_lookup_user_args *args;
2594         struct inode *inode;
2595         int ret;
2596
2597         args = memdup_user(argp, sizeof(*args));
2598         if (IS_ERR(args))
2599                 return PTR_ERR(args);
2600
2601         inode = file_inode(file);
2602
2603         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2604             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2605                 /*
2606                  * The subvolume does not exist under fd with which this is
2607                  * called
2608                  */
2609                 kfree(args);
2610                 return -EACCES;
2611         }
2612
2613         ret = btrfs_search_path_in_tree_user(inode, args);
2614
2615         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2616                 ret = -EFAULT;
2617
2618         kfree(args);
2619         return ret;
2620 }
2621
2622 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2623 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2624 {
2625         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2626         struct btrfs_fs_info *fs_info;
2627         struct btrfs_root *root;
2628         struct btrfs_path *path;
2629         struct btrfs_key key;
2630         struct btrfs_root_item *root_item;
2631         struct btrfs_root_ref *rref;
2632         struct extent_buffer *leaf;
2633         unsigned long item_off;
2634         unsigned long item_len;
2635         struct inode *inode;
2636         int slot;
2637         int ret = 0;
2638
2639         path = btrfs_alloc_path();
2640         if (!path)
2641                 return -ENOMEM;
2642
2643         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2644         if (!subvol_info) {
2645                 btrfs_free_path(path);
2646                 return -ENOMEM;
2647         }
2648
2649         inode = file_inode(file);
2650         fs_info = BTRFS_I(inode)->root->fs_info;
2651
2652         /* Get root_item of inode's subvolume */
2653         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2654         key.type = BTRFS_ROOT_ITEM_KEY;
2655         key.offset = (u64)-1;
2656         root = btrfs_read_fs_root_no_name(fs_info, &key);
2657         if (IS_ERR(root)) {
2658                 ret = PTR_ERR(root);
2659                 goto out;
2660         }
2661         root_item = &root->root_item;
2662
2663         subvol_info->treeid = key.objectid;
2664
2665         subvol_info->generation = btrfs_root_generation(root_item);
2666         subvol_info->flags = btrfs_root_flags(root_item);
2667
2668         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2669         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2670                                                     BTRFS_UUID_SIZE);
2671         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2672                                                     BTRFS_UUID_SIZE);
2673
2674         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2675         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2676         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2677
2678         subvol_info->otransid = btrfs_root_otransid(root_item);
2679         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2680         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2681
2682         subvol_info->stransid = btrfs_root_stransid(root_item);
2683         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2684         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2685
2686         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2687         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2688         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2689
2690         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2691                 /* Search root tree for ROOT_BACKREF of this subvolume */
2692                 root = fs_info->tree_root;
2693
2694                 key.type = BTRFS_ROOT_BACKREF_KEY;
2695                 key.offset = 0;
2696                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2697                 if (ret < 0) {
2698                         goto out;
2699                 } else if (path->slots[0] >=
2700                            btrfs_header_nritems(path->nodes[0])) {
2701                         ret = btrfs_next_leaf(root, path);
2702                         if (ret < 0) {
2703                                 goto out;
2704                         } else if (ret > 0) {
2705                                 ret = -EUCLEAN;
2706                                 goto out;
2707                         }
2708                 }
2709
2710                 leaf = path->nodes[0];
2711                 slot = path->slots[0];
2712                 btrfs_item_key_to_cpu(leaf, &key, slot);
2713                 if (key.objectid == subvol_info->treeid &&
2714                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2715                         subvol_info->parent_id = key.offset;
2716
2717                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2718                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2719
2720                         item_off = btrfs_item_ptr_offset(leaf, slot)
2721                                         + sizeof(struct btrfs_root_ref);
2722                         item_len = btrfs_item_size_nr(leaf, slot)
2723                                         - sizeof(struct btrfs_root_ref);
2724                         read_extent_buffer(leaf, subvol_info->name,
2725                                            item_off, item_len);
2726                 } else {
2727                         ret = -ENOENT;
2728                         goto out;
2729                 }
2730         }
2731
2732         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2733                 ret = -EFAULT;
2734
2735 out:
2736         btrfs_free_path(path);
2737         kzfree(subvol_info);
2738         return ret;
2739 }
2740
2741 /*
2742  * Return ROOT_REF information of the subvolume containing this inode
2743  * except the subvolume name.
2744  */
2745 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2746 {
2747         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2748         struct btrfs_root_ref *rref;
2749         struct btrfs_root *root;
2750         struct btrfs_path *path;
2751         struct btrfs_key key;
2752         struct extent_buffer *leaf;
2753         struct inode *inode;
2754         u64 objectid;
2755         int slot;
2756         int ret;
2757         u8 found;
2758
2759         path = btrfs_alloc_path();
2760         if (!path)
2761                 return -ENOMEM;
2762
2763         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2764         if (IS_ERR(rootrefs)) {
2765                 btrfs_free_path(path);
2766                 return PTR_ERR(rootrefs);
2767         }
2768
2769         inode = file_inode(file);
2770         root = BTRFS_I(inode)->root->fs_info->tree_root;
2771         objectid = BTRFS_I(inode)->root->root_key.objectid;
2772
2773         key.objectid = objectid;
2774         key.type = BTRFS_ROOT_REF_KEY;
2775         key.offset = rootrefs->min_treeid;
2776         found = 0;
2777
2778         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2779         if (ret < 0) {
2780                 goto out;
2781         } else if (path->slots[0] >=
2782                    btrfs_header_nritems(path->nodes[0])) {
2783                 ret = btrfs_next_leaf(root, path);
2784                 if (ret < 0) {
2785                         goto out;
2786                 } else if (ret > 0) {
2787                         ret = -EUCLEAN;
2788                         goto out;
2789                 }
2790         }
2791         while (1) {
2792                 leaf = path->nodes[0];
2793                 slot = path->slots[0];
2794
2795                 btrfs_item_key_to_cpu(leaf, &key, slot);
2796                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2797                         ret = 0;
2798                         goto out;
2799                 }
2800
2801                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2802                         ret = -EOVERFLOW;
2803                         goto out;
2804                 }
2805
2806                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2807                 rootrefs->rootref[found].treeid = key.offset;
2808                 rootrefs->rootref[found].dirid =
2809                                   btrfs_root_ref_dirid(leaf, rref);
2810                 found++;
2811
2812                 ret = btrfs_next_item(root, path);
2813                 if (ret < 0) {
2814                         goto out;
2815                 } else if (ret > 0) {
2816                         ret = -EUCLEAN;
2817                         goto out;
2818                 }
2819         }
2820
2821 out:
2822         if (!ret || ret == -EOVERFLOW) {
2823                 rootrefs->num_items = found;
2824                 /* update min_treeid for next search */
2825                 if (found)
2826                         rootrefs->min_treeid =
2827                                 rootrefs->rootref[found - 1].treeid + 1;
2828                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2829                         ret = -EFAULT;
2830         }
2831
2832         kfree(rootrefs);
2833         btrfs_free_path(path);
2834
2835         return ret;
2836 }
2837
2838 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2839                                              void __user *arg)
2840 {
2841         struct dentry *parent = file->f_path.dentry;
2842         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2843         struct dentry *dentry;
2844         struct inode *dir = d_inode(parent);
2845         struct inode *inode;
2846         struct btrfs_root *root = BTRFS_I(dir)->root;
2847         struct btrfs_root *dest = NULL;
2848         struct btrfs_ioctl_vol_args *vol_args;
2849         int namelen;
2850         int err = 0;
2851
2852         if (!S_ISDIR(dir->i_mode))
2853                 return -ENOTDIR;
2854
2855         vol_args = memdup_user(arg, sizeof(*vol_args));
2856         if (IS_ERR(vol_args))
2857                 return PTR_ERR(vol_args);
2858
2859         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2860         namelen = strlen(vol_args->name);
2861         if (strchr(vol_args->name, '/') ||
2862             strncmp(vol_args->name, "..", namelen) == 0) {
2863                 err = -EINVAL;
2864                 goto out;
2865         }
2866
2867         err = mnt_want_write_file(file);
2868         if (err)
2869                 goto out;
2870
2871
2872         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2873         if (err == -EINTR)
2874                 goto out_drop_write;
2875         dentry = lookup_one_len(vol_args->name, parent, namelen);
2876         if (IS_ERR(dentry)) {
2877                 err = PTR_ERR(dentry);
2878                 goto out_unlock_dir;
2879         }
2880
2881         if (d_really_is_negative(dentry)) {
2882                 err = -ENOENT;
2883                 goto out_dput;
2884         }
2885
2886         inode = d_inode(dentry);
2887         dest = BTRFS_I(inode)->root;
2888         if (!capable(CAP_SYS_ADMIN)) {
2889                 /*
2890                  * Regular user.  Only allow this with a special mount
2891                  * option, when the user has write+exec access to the
2892                  * subvol root, and when rmdir(2) would have been
2893                  * allowed.
2894                  *
2895                  * Note that this is _not_ check that the subvol is
2896                  * empty or doesn't contain data that we wouldn't
2897                  * otherwise be able to delete.
2898                  *
2899                  * Users who want to delete empty subvols should try
2900                  * rmdir(2).
2901                  */
2902                 err = -EPERM;
2903                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2904                         goto out_dput;
2905
2906                 /*
2907                  * Do not allow deletion if the parent dir is the same
2908                  * as the dir to be deleted.  That means the ioctl
2909                  * must be called on the dentry referencing the root
2910                  * of the subvol, not a random directory contained
2911                  * within it.
2912                  */
2913                 err = -EINVAL;
2914                 if (root == dest)
2915                         goto out_dput;
2916
2917                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2918                 if (err)
2919                         goto out_dput;
2920         }
2921
2922         /* check if subvolume may be deleted by a user */
2923         err = btrfs_may_delete(dir, dentry, 1);
2924         if (err)
2925                 goto out_dput;
2926
2927         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2928                 err = -EINVAL;
2929                 goto out_dput;
2930         }
2931
2932         inode_lock(inode);
2933         err = btrfs_delete_subvolume(dir, dentry);
2934         inode_unlock(inode);
2935         if (!err) {
2936                 fsnotify_rmdir(dir, dentry);
2937                 d_delete(dentry);
2938         }
2939
2940 out_dput:
2941         dput(dentry);
2942 out_unlock_dir:
2943         inode_unlock(dir);
2944 out_drop_write:
2945         mnt_drop_write_file(file);
2946 out:
2947         kfree(vol_args);
2948         return err;
2949 }
2950
2951 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2952 {
2953         struct inode *inode = file_inode(file);
2954         struct btrfs_root *root = BTRFS_I(inode)->root;
2955         struct btrfs_ioctl_defrag_range_args *range;
2956         int ret;
2957
2958         ret = mnt_want_write_file(file);
2959         if (ret)
2960                 return ret;
2961
2962         if (btrfs_root_readonly(root)) {
2963                 ret = -EROFS;
2964                 goto out;
2965         }
2966
2967         switch (inode->i_mode & S_IFMT) {
2968         case S_IFDIR:
2969                 if (!capable(CAP_SYS_ADMIN)) {
2970                         ret = -EPERM;
2971                         goto out;
2972                 }
2973                 ret = btrfs_defrag_root(root);
2974                 break;
2975         case S_IFREG:
2976                 /*
2977                  * Note that this does not check the file descriptor for write
2978                  * access. This prevents defragmenting executables that are
2979                  * running and allows defrag on files open in read-only mode.
2980                  */
2981                 if (!capable(CAP_SYS_ADMIN) &&
2982                     inode_permission(inode, MAY_WRITE)) {
2983                         ret = -EPERM;
2984                         goto out;
2985                 }
2986
2987                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2988                 if (!range) {
2989                         ret = -ENOMEM;
2990                         goto out;
2991                 }
2992
2993                 if (argp) {
2994                         if (copy_from_user(range, argp,
2995                                            sizeof(*range))) {
2996                                 ret = -EFAULT;
2997                                 kfree(range);
2998                                 goto out;
2999                         }
3000                         /* compression requires us to start the IO */
3001                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3002                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3003                                 range->extent_thresh = (u32)-1;
3004                         }
3005                 } else {
3006                         /* the rest are all set to zero by kzalloc */
3007                         range->len = (u64)-1;
3008                 }
3009                 ret = btrfs_defrag_file(file_inode(file), file,
3010                                         range, BTRFS_OLDEST_GENERATION, 0);
3011                 if (ret > 0)
3012                         ret = 0;
3013                 kfree(range);
3014                 break;
3015         default:
3016                 ret = -EINVAL;
3017         }
3018 out:
3019         mnt_drop_write_file(file);
3020         return ret;
3021 }
3022
3023 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3024 {
3025         struct btrfs_ioctl_vol_args *vol_args;
3026         int ret;
3027
3028         if (!capable(CAP_SYS_ADMIN))
3029                 return -EPERM;
3030
3031         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3032                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3033
3034         vol_args = memdup_user(arg, sizeof(*vol_args));
3035         if (IS_ERR(vol_args)) {
3036                 ret = PTR_ERR(vol_args);
3037                 goto out;
3038         }
3039
3040         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3041         ret = btrfs_init_new_device(fs_info, vol_args->name);
3042
3043         if (!ret)
3044                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3045
3046         kfree(vol_args);
3047 out:
3048         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3049         return ret;
3050 }
3051
3052 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3053 {
3054         struct inode *inode = file_inode(file);
3055         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3056         struct btrfs_ioctl_vol_args_v2 *vol_args;
3057         int ret;
3058
3059         if (!capable(CAP_SYS_ADMIN))
3060                 return -EPERM;
3061
3062         ret = mnt_want_write_file(file);
3063         if (ret)
3064                 return ret;
3065
3066         vol_args = memdup_user(arg, sizeof(*vol_args));
3067         if (IS_ERR(vol_args)) {
3068                 ret = PTR_ERR(vol_args);
3069                 goto err_drop;
3070         }
3071
3072         /* Check for compatibility reject unknown flags */
3073         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3074                 ret = -EOPNOTSUPP;
3075                 goto out;
3076         }
3077
3078         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3079                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3080                 goto out;
3081         }
3082
3083         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3084                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3085         } else {
3086                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3087                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3088         }
3089         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3090
3091         if (!ret) {
3092                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3093                         btrfs_info(fs_info, "device deleted: id %llu",
3094                                         vol_args->devid);
3095                 else
3096                         btrfs_info(fs_info, "device deleted: %s",
3097                                         vol_args->name);
3098         }
3099 out:
3100         kfree(vol_args);
3101 err_drop:
3102         mnt_drop_write_file(file);
3103         return ret;
3104 }
3105
3106 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3107 {
3108         struct inode *inode = file_inode(file);
3109         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3110         struct btrfs_ioctl_vol_args *vol_args;
3111         int ret;
3112
3113         if (!capable(CAP_SYS_ADMIN))
3114                 return -EPERM;
3115
3116         ret = mnt_want_write_file(file);
3117         if (ret)
3118                 return ret;
3119
3120         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3121                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3122                 goto out_drop_write;
3123         }
3124
3125         vol_args = memdup_user(arg, sizeof(*vol_args));
3126         if (IS_ERR(vol_args)) {
3127                 ret = PTR_ERR(vol_args);
3128                 goto out;
3129         }
3130
3131         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3132         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3133
3134         if (!ret)
3135                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3136         kfree(vol_args);
3137 out:
3138         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3139 out_drop_write:
3140         mnt_drop_write_file(file);
3141
3142         return ret;
3143 }
3144
3145 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3146                                 void __user *arg)
3147 {
3148         struct btrfs_ioctl_fs_info_args *fi_args;
3149         struct btrfs_device *device;
3150         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3151         int ret = 0;
3152
3153         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3154         if (!fi_args)
3155                 return -ENOMEM;
3156
3157         rcu_read_lock();
3158         fi_args->num_devices = fs_devices->num_devices;
3159
3160         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3161                 if (device->devid > fi_args->max_id)
3162                         fi_args->max_id = device->devid;
3163         }
3164         rcu_read_unlock();
3165
3166         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3167         fi_args->nodesize = fs_info->nodesize;
3168         fi_args->sectorsize = fs_info->sectorsize;
3169         fi_args->clone_alignment = fs_info->sectorsize;
3170
3171         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3172                 ret = -EFAULT;
3173
3174         kfree(fi_args);
3175         return ret;
3176 }
3177
3178 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3179                                  void __user *arg)
3180 {
3181         struct btrfs_ioctl_dev_info_args *di_args;
3182         struct btrfs_device *dev;
3183         int ret = 0;
3184         char *s_uuid = NULL;
3185
3186         di_args = memdup_user(arg, sizeof(*di_args));
3187         if (IS_ERR(di_args))
3188                 return PTR_ERR(di_args);
3189
3190         if (!btrfs_is_empty_uuid(di_args->uuid))
3191                 s_uuid = di_args->uuid;
3192
3193         rcu_read_lock();
3194         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3195                                 NULL, true);
3196
3197         if (!dev) {
3198                 ret = -ENODEV;
3199                 goto out;
3200         }
3201
3202         di_args->devid = dev->devid;
3203         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3204         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3205         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3206         if (dev->name) {
3207                 strncpy(di_args->path, rcu_str_deref(dev->name),
3208                                 sizeof(di_args->path) - 1);
3209                 di_args->path[sizeof(di_args->path) - 1] = 0;
3210         } else {
3211                 di_args->path[0] = '\0';
3212         }
3213
3214 out:
3215         rcu_read_unlock();
3216         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3217                 ret = -EFAULT;
3218
3219         kfree(di_args);
3220         return ret;
3221 }
3222
3223 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3224                                        struct inode *inode2, u64 loff2, u64 len)
3225 {
3226         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3227         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3228 }
3229
3230 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3231                                      struct inode *inode2, u64 loff2, u64 len)
3232 {
3233         if (inode1 < inode2) {
3234                 swap(inode1, inode2);
3235                 swap(loff1, loff2);
3236         } else if (inode1 == inode2 && loff2 < loff1) {
3237                 swap(loff1, loff2);
3238         }
3239         lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3240         lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3241 }
3242
3243 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
3244                                    struct inode *dst, u64 dst_loff)
3245 {
3246         int ret;
3247
3248         /*
3249          * Lock destination range to serialize with concurrent readpages() and
3250          * source range to serialize with relocation.
3251          */
3252         btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3253         ret = btrfs_clone(src, dst, loff, len, len, dst_loff, 1);
3254         btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3255
3256         return ret;
3257 }
3258
3259 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3260
3261 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3262                              struct inode *dst, u64 dst_loff)
3263 {
3264         int ret;
3265         u64 i, tail_len, chunk_count;
3266         struct btrfs_root *root_dst = BTRFS_I(dst)->root;
3267
3268         spin_lock(&root_dst->root_item_lock);
3269         if (root_dst->send_in_progress) {
3270                 btrfs_warn_rl(root_dst->fs_info,
3271 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
3272                               root_dst->root_key.objectid,
3273                               root_dst->send_in_progress);
3274                 spin_unlock(&root_dst->root_item_lock);
3275                 return -EAGAIN;
3276         }
3277         root_dst->dedupe_in_progress++;
3278         spin_unlock(&root_dst->root_item_lock);
3279
3280         tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3281         chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3282
3283         for (i = 0; i < chunk_count; i++) {
3284                 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3285                                               dst, dst_loff);
3286                 if (ret)
3287                         goto out;
3288
3289                 loff += BTRFS_MAX_DEDUPE_LEN;
3290                 dst_loff += BTRFS_MAX_DEDUPE_LEN;
3291         }
3292
3293         if (tail_len > 0)
3294                 ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3295                                               dst_loff);
3296 out:
3297         spin_lock(&root_dst->root_item_lock);
3298         root_dst->dedupe_in_progress--;
3299         spin_unlock(&root_dst->root_item_lock);
3300
3301         return ret;
3302 }
3303
3304 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3305                                      struct inode *inode,
3306                                      u64 endoff,
3307                                      const u64 destoff,
3308                                      const u64 olen,
3309                                      int no_time_update)
3310 {
3311         struct btrfs_root *root = BTRFS_I(inode)->root;
3312         int ret;
3313
3314         inode_inc_iversion(inode);
3315         if (!no_time_update)
3316                 inode->i_mtime = inode->i_ctime = current_time(inode);
3317         /*
3318          * We round up to the block size at eof when determining which
3319          * extents to clone above, but shouldn't round up the file size.
3320          */
3321         if (endoff > destoff + olen)
3322                 endoff = destoff + olen;
3323         if (endoff > inode->i_size)
3324                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3325
3326         ret = btrfs_update_inode(trans, root, inode);
3327         if (ret) {
3328                 btrfs_abort_transaction(trans, ret);
3329                 btrfs_end_transaction(trans);
3330                 goto out;
3331         }
3332         ret = btrfs_end_transaction(trans);
3333 out:
3334         return ret;
3335 }
3336
3337 /*
3338  * Make sure we do not end up inserting an inline extent into a file that has
3339  * already other (non-inline) extents. If a file has an inline extent it can
3340  * not have any other extents and the (single) inline extent must start at the
3341  * file offset 0. Failing to respect these rules will lead to file corruption,
3342  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3343  *
3344  * We can have extents that have been already written to disk or we can have
3345  * dirty ranges still in delalloc, in which case the extent maps and items are
3346  * created only when we run delalloc, and the delalloc ranges might fall outside
3347  * the range we are currently locking in the inode's io tree. So we check the
3348  * inode's i_size because of that (i_size updates are done while holding the
3349  * i_mutex, which we are holding here).
3350  * We also check to see if the inode has a size not greater than "datal" but has
3351  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3352  * protected against such concurrent fallocate calls by the i_mutex).
3353  *
3354  * If the file has no extents but a size greater than datal, do not allow the
3355  * copy because we would need turn the inline extent into a non-inline one (even
3356  * with NO_HOLES enabled). If we find our destination inode only has one inline
3357  * extent, just overwrite it with the source inline extent if its size is less
3358  * than the source extent's size, or we could copy the source inline extent's
3359  * data into the destination inode's inline extent if the later is greater then
3360  * the former.
3361  */
3362 static int clone_copy_inline_extent(struct inode *dst,
3363                                     struct btrfs_trans_handle *trans,
3364                                     struct btrfs_path *path,
3365                                     struct btrfs_key *new_key,
3366                                     const u64 drop_start,
3367                                     const u64 datal,
3368                                     const u64 skip,
3369                                     const u64 size,
3370                                     char *inline_data)
3371 {
3372         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3373         struct btrfs_root *root = BTRFS_I(dst)->root;
3374         const u64 aligned_end = ALIGN(new_key->offset + datal,
3375                                       fs_info->sectorsize);
3376         int ret;
3377         struct btrfs_key key;
3378
3379         if (new_key->offset > 0)
3380                 return -EOPNOTSUPP;
3381
3382         key.objectid = btrfs_ino(BTRFS_I(dst));
3383         key.type = BTRFS_EXTENT_DATA_KEY;
3384         key.offset = 0;
3385         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3386         if (ret < 0) {
3387                 return ret;
3388         } else if (ret > 0) {
3389                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3390                         ret = btrfs_next_leaf(root, path);
3391                         if (ret < 0)
3392                                 return ret;
3393                         else if (ret > 0)
3394                                 goto copy_inline_extent;
3395                 }
3396                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3397                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3398                     key.type == BTRFS_EXTENT_DATA_KEY) {
3399                         ASSERT(key.offset > 0);
3400                         return -EOPNOTSUPP;
3401                 }
3402         } else if (i_size_read(dst) <= datal) {
3403                 struct btrfs_file_extent_item *ei;
3404                 u64 ext_len;
3405
3406                 /*
3407                  * If the file size is <= datal, make sure there are no other
3408                  * extents following (can happen do to an fallocate call with
3409                  * the flag FALLOC_FL_KEEP_SIZE).
3410                  */
3411                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3412                                     struct btrfs_file_extent_item);
3413                 /*
3414                  * If it's an inline extent, it can not have other extents
3415                  * following it.
3416                  */
3417                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3418                     BTRFS_FILE_EXTENT_INLINE)
3419                         goto copy_inline_extent;
3420
3421                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3422                 if (ext_len > aligned_end)
3423                         return -EOPNOTSUPP;
3424
3425                 ret = btrfs_next_item(root, path);
3426                 if (ret < 0) {
3427                         return ret;
3428                 } else if (ret == 0) {
3429                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3430                                               path->slots[0]);
3431                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3432                             key.type == BTRFS_EXTENT_DATA_KEY)
3433                                 return -EOPNOTSUPP;
3434                 }
3435         }
3436
3437 copy_inline_extent:
3438         /*
3439          * We have no extent items, or we have an extent at offset 0 which may
3440          * or may not be inlined. All these cases are dealt the same way.
3441          */
3442         if (i_size_read(dst) > datal) {
3443                 /*
3444                  * If the destination inode has an inline extent...
3445                  * This would require copying the data from the source inline
3446                  * extent into the beginning of the destination's inline extent.
3447                  * But this is really complex, both extents can be compressed
3448                  * or just one of them, which would require decompressing and
3449                  * re-compressing data (which could increase the new compressed
3450                  * size, not allowing the compressed data to fit anymore in an
3451                  * inline extent).
3452                  * So just don't support this case for now (it should be rare,
3453                  * we are not really saving space when cloning inline extents).
3454                  */
3455                 return -EOPNOTSUPP;
3456         }
3457
3458         btrfs_release_path(path);
3459         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3460         if (ret)
3461                 return ret;
3462         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3463         if (ret)
3464                 return ret;
3465
3466         if (skip) {
3467                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3468
3469                 memmove(inline_data + start, inline_data + start + skip, datal);
3470         }
3471
3472         write_extent_buffer(path->nodes[0], inline_data,
3473                             btrfs_item_ptr_offset(path->nodes[0],
3474                                                   path->slots[0]),
3475                             size);
3476         inode_add_bytes(dst, datal);
3477         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags);
3478
3479         return 0;
3480 }
3481
3482 /**
3483  * btrfs_clone() - clone a range from inode file to another
3484  *
3485  * @src: Inode to clone from
3486  * @inode: Inode to clone to
3487  * @off: Offset within source to start clone from
3488  * @olen: Original length, passed by user, of range to clone
3489  * @olen_aligned: Block-aligned value of olen
3490  * @destoff: Offset within @inode to start clone
3491  * @no_time_update: Whether to update mtime/ctime on the target inode
3492  */
3493 static int btrfs_clone(struct inode *src, struct inode *inode,
3494                        const u64 off, const u64 olen, const u64 olen_aligned,
3495                        const u64 destoff, int no_time_update)
3496 {
3497         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3498         struct btrfs_root *root = BTRFS_I(inode)->root;
3499         struct btrfs_path *path = NULL;
3500         struct extent_buffer *leaf;
3501         struct btrfs_trans_handle *trans;
3502         char *buf = NULL;
3503         struct btrfs_key key;
3504         u32 nritems;
3505         int slot;
3506         int ret;
3507         const u64 len = olen_aligned;
3508         u64 last_dest_end = destoff;
3509
3510         ret = -ENOMEM;
3511         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3512         if (!buf)
3513                 return ret;
3514
3515         path = btrfs_alloc_path();
3516         if (!path) {
3517                 kvfree(buf);
3518                 return ret;
3519         }
3520
3521         path->reada = READA_FORWARD;
3522         /* clone data */
3523         key.objectid = btrfs_ino(BTRFS_I(src));
3524         key.type = BTRFS_EXTENT_DATA_KEY;
3525         key.offset = off;
3526
3527         while (1) {
3528                 u64 next_key_min_offset = key.offset + 1;
3529                 struct btrfs_file_extent_item *extent;
3530                 int type;
3531                 u32 size;
3532                 struct btrfs_key new_key;
3533                 u64 disko = 0, diskl = 0;
3534                 u64 datao = 0, datal = 0;
3535                 u8 comp;
3536                 u64 drop_start;
3537
3538                 /*
3539                  * note the key will change type as we walk through the
3540                  * tree.
3541                  */
3542                 path->leave_spinning = 1;
3543                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3544                                 0, 0);
3545                 if (ret < 0)
3546                         goto out;
3547                 /*
3548                  * First search, if no extent item that starts at offset off was
3549                  * found but the previous item is an extent item, it's possible
3550                  * it might overlap our target range, therefore process it.
3551                  */
3552                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3553                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3554                                               path->slots[0] - 1);
3555                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3556                                 path->slots[0]--;
3557                 }
3558
3559                 nritems = btrfs_header_nritems(path->nodes[0]);
3560 process_slot:
3561                 if (path->slots[0] >= nritems) {
3562                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3563                         if (ret < 0)
3564                                 goto out;
3565                         if (ret > 0)
3566                                 break;
3567                         nritems = btrfs_header_nritems(path->nodes[0]);
3568                 }
3569                 leaf = path->nodes[0];
3570                 slot = path->slots[0];
3571
3572                 btrfs_item_key_to_cpu(leaf, &key, slot);
3573                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3574                     key.objectid != btrfs_ino(BTRFS_I(src)))
3575                         break;
3576
3577                 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
3578
3579                 extent = btrfs_item_ptr(leaf, slot,
3580                                         struct btrfs_file_extent_item);
3581                 comp = btrfs_file_extent_compression(leaf, extent);
3582                 type = btrfs_file_extent_type(leaf, extent);
3583                 if (type == BTRFS_FILE_EXTENT_REG ||
3584                     type == BTRFS_FILE_EXTENT_PREALLOC) {
3585                         disko = btrfs_file_extent_disk_bytenr(leaf, extent);
3586                         diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
3587                         datao = btrfs_file_extent_offset(leaf, extent);
3588                         datal = btrfs_file_extent_num_bytes(leaf, extent);
3589                 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3590                         /* Take upper bound, may be compressed */
3591                         datal = btrfs_file_extent_ram_bytes(leaf, extent);
3592                 }
3593
3594                 /*
3595                  * The first search might have left us at an extent item that
3596                  * ends before our target range's start, can happen if we have
3597                  * holes and NO_HOLES feature enabled.
3598                  */
3599                 if (key.offset + datal <= off) {
3600                         path->slots[0]++;
3601                         goto process_slot;
3602                 } else if (key.offset >= off + len) {
3603                         break;
3604                 }
3605                 next_key_min_offset = key.offset + datal;
3606                 size = btrfs_item_size_nr(leaf, slot);
3607                 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
3608                                    size);
3609
3610                 btrfs_release_path(path);
3611                 path->leave_spinning = 0;
3612
3613                 memcpy(&new_key, &key, sizeof(new_key));
3614                 new_key.objectid = btrfs_ino(BTRFS_I(inode));
3615                 if (off <= key.offset)
3616                         new_key.offset = key.offset + destoff - off;
3617                 else
3618                         new_key.offset = destoff;
3619
3620                 /*
3621                  * Deal with a hole that doesn't have an extent item that
3622                  * represents it (NO_HOLES feature enabled).
3623                  * This hole is either in the middle of the cloning range or at
3624                  * the beginning (fully overlaps it or partially overlaps it).
3625                  */
3626                 if (new_key.offset != last_dest_end)
3627                         drop_start = last_dest_end;
3628                 else
3629                         drop_start = new_key.offset;
3630
3631                 if (type == BTRFS_FILE_EXTENT_REG ||
3632                     type == BTRFS_FILE_EXTENT_PREALLOC) {
3633                         struct btrfs_clone_extent_info clone_info;
3634
3635                         /*
3636                          *    a  | --- range to clone ---|  b
3637                          * | ------------- extent ------------- |
3638                          */
3639
3640                         /* Subtract range b */
3641                         if (key.offset + datal > off + len)
3642                                 datal = off + len - key.offset;
3643
3644                         /* Subtract range a */
3645                         if (off > key.offset) {
3646                                 datao += off - key.offset;
3647                                 datal -= off - key.offset;
3648                         }
3649
3650                         clone_info.disk_offset = disko;
3651                         clone_info.disk_len = diskl;
3652                         clone_info.data_offset = datao;
3653                         clone_info.data_len = datal;
3654                         clone_info.file_offset = new_key.offset;
3655                         clone_info.extent_buf = buf;
3656                         clone_info.item_size = size;
3657                         ret = btrfs_punch_hole_range(inode, path,
3658                                                      drop_start,
3659                                                      new_key.offset + datal - 1,
3660                                                      &clone_info, &trans);
3661                         if (ret)
3662                                 goto out;
3663                 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3664                         u64 skip = 0;
3665                         u64 trim = 0;
3666
3667                         if (off > key.offset) {
3668                                 skip = off - key.offset;
3669                                 new_key.offset += skip;
3670                         }
3671
3672                         if (key.offset + datal > off + len)
3673                                 trim = key.offset + datal - (off + len);
3674
3675                         if (comp && (skip || trim)) {
3676                                 ret = -EINVAL;
3677                                 goto out;
3678                         }
3679                         size -= skip + trim;
3680                         datal -= skip + trim;
3681
3682                         /*
3683                          * If our extent is inline, we know we will drop or
3684                          * adjust at most 1 extent item in the destination root.
3685                          *
3686                          * 1 - adjusting old extent (we may have to split it)
3687                          * 1 - add new extent
3688                          * 1 - inode update
3689                          */
3690                         trans = btrfs_start_transaction(root, 3);
3691                         if (IS_ERR(trans)) {
3692                                 ret = PTR_ERR(trans);
3693                                 goto out;
3694                         }
3695
3696                         ret = clone_copy_inline_extent(inode, trans, path,
3697                                                        &new_key, drop_start,
3698                                                        datal, skip, size, buf);
3699                         if (ret) {
3700                                 if (ret != -EOPNOTSUPP)
3701                                         btrfs_abort_transaction(trans, ret);
3702                                 btrfs_end_transaction(trans);
3703                                 goto out;
3704                         }
3705                 }
3706
3707                 btrfs_release_path(path);
3708
3709                 last_dest_end = ALIGN(new_key.offset + datal,
3710                                       fs_info->sectorsize);
3711                 ret = clone_finish_inode_update(trans, inode, last_dest_end,
3712                                                 destoff, olen, no_time_update);
3713                 if (ret)
3714                         goto out;
3715                 if (new_key.offset + datal >= destoff + len)
3716                         break;
3717
3718                 btrfs_release_path(path);
3719                 key.offset = next_key_min_offset;
3720
3721                 if (fatal_signal_pending(current)) {
3722                         ret = -EINTR;
3723                         goto out;
3724                 }
3725         }
3726         ret = 0;
3727
3728         if (last_dest_end < destoff + len) {
3729                 /*
3730                  * We have an implicit hole that fully or partially overlaps our
3731                  * cloning range at its end. This means that we either have the
3732                  * NO_HOLES feature enabled or the implicit hole happened due to
3733                  * mixing buffered and direct IO writes against this file.
3734                  */
3735                 btrfs_release_path(path);
3736                 path->leave_spinning = 0;
3737
3738                 ret = btrfs_punch_hole_range(inode, path,
3739                                              last_dest_end, destoff + len - 1,
3740                                              NULL, &trans);
3741                 if (ret)
3742                         goto out;
3743
3744                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3745                                                 destoff, olen, no_time_update);
3746         }
3747
3748 out:
3749         btrfs_free_path(path);
3750         kvfree(buf);
3751         return ret;
3752 }
3753
3754 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3755                                         u64 off, u64 olen, u64 destoff)
3756 {
3757         struct inode *inode = file_inode(file);
3758         struct inode *src = file_inode(file_src);
3759         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3760         int ret;
3761         u64 len = olen;
3762         u64 bs = fs_info->sb->s_blocksize;
3763
3764         /*
3765          * TODO:
3766          * - split compressed inline extents.  annoying: we need to
3767          *   decompress into destination's address_space (the file offset
3768          *   may change, so source mapping won't do), then recompress (or
3769          *   otherwise reinsert) a subrange.
3770          *
3771          * - split destination inode's inline extents.  The inline extents can
3772          *   be either compressed or non-compressed.
3773          */
3774
3775         /*
3776          * VFS's generic_remap_file_range_prep() protects us from cloning the
3777          * eof block into the middle of a file, which would result in corruption
3778          * if the file size is not blocksize aligned. So we don't need to check
3779          * for that case here.
3780          */
3781         if (off + len == src->i_size)
3782                 len = ALIGN(src->i_size, bs) - off;
3783
3784         if (destoff > inode->i_size) {
3785                 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
3786
3787                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3788                 if (ret)
3789                         return ret;
3790                 /*
3791                  * We may have truncated the last block if the inode's size is
3792                  * not sector size aligned, so we need to wait for writeback to
3793                  * complete before proceeding further, otherwise we can race
3794                  * with cloning and attempt to increment a reference to an
3795                  * extent that no longer exists (writeback completed right after
3796                  * we found the previous extent covering eof and before we
3797                  * attempted to increment its reference count).
3798                  */
3799                 ret = btrfs_wait_ordered_range(inode, wb_start,
3800                                                destoff - wb_start);
3801                 if (ret)
3802                         return ret;
3803         }
3804
3805         /*
3806          * Lock destination range to serialize with concurrent readpages() and
3807          * source range to serialize with relocation.
3808          */
3809         btrfs_double_extent_lock(src, off, inode, destoff, len);
3810         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3811         btrfs_double_extent_unlock(src, off, inode, destoff, len);
3812         /*
3813          * Truncate page cache pages so that future reads will see the cloned
3814          * data immediately and not the previous data.
3815          */
3816         truncate_inode_pages_range(&inode->i_data,
3817                                 round_down(destoff, PAGE_SIZE),
3818                                 round_up(destoff + len, PAGE_SIZE) - 1);
3819
3820         return ret;
3821 }
3822
3823 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
3824                                        struct file *file_out, loff_t pos_out,
3825                                        loff_t *len, unsigned int remap_flags)
3826 {
3827         struct inode *inode_in = file_inode(file_in);
3828         struct inode *inode_out = file_inode(file_out);
3829         u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
3830         bool same_inode = inode_out == inode_in;
3831         u64 wb_len;
3832         int ret;
3833
3834         if (!(remap_flags & REMAP_FILE_DEDUP)) {
3835                 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
3836
3837                 if (btrfs_root_readonly(root_out))
3838                         return -EROFS;
3839
3840                 if (file_in->f_path.mnt != file_out->f_path.mnt ||
3841                     inode_in->i_sb != inode_out->i_sb)
3842                         return -EXDEV;
3843         }
3844
3845         /* don't make the dst file partly checksummed */
3846         if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
3847             (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
3848                 return -EINVAL;
3849         }
3850
3851         /*
3852          * Now that the inodes are locked, we need to start writeback ourselves
3853          * and can not rely on the writeback from the VFS's generic helper
3854          * generic_remap_file_range_prep() because:
3855          *
3856          * 1) For compression we must call filemap_fdatawrite_range() range
3857          *    twice (btrfs_fdatawrite_range() does it for us), and the generic
3858          *    helper only calls it once;
3859          *
3860          * 2) filemap_fdatawrite_range(), called by the generic helper only
3861          *    waits for the writeback to complete, i.e. for IO to be done, and
3862          *    not for the ordered extents to complete. We need to wait for them
3863          *    to complete so that new file extent items are in the fs tree.
3864          */
3865         if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
3866                 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
3867         else
3868                 wb_len = ALIGN(*len, bs);
3869
3870         /*
3871          * Since we don't lock ranges, wait for ongoing lockless dio writes (as
3872          * any in progress could create its ordered extents after we wait for
3873          * existing ordered extents below).
3874          */
3875         inode_dio_wait(inode_in);
3876         if (!same_inode)
3877                 inode_dio_wait(inode_out);
3878
3879         /*
3880          * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
3881          *
3882          * Btrfs' back references do not have a block level granularity, they
3883          * work at the whole extent level.
3884          * NOCOW buffered write without data space reserved may not be able
3885          * to fall back to CoW due to lack of data space, thus could cause
3886          * data loss.
3887          *
3888          * Here we take a shortcut by flushing the whole inode, so that all
3889          * nocow write should reach disk as nocow before we increase the
3890          * reference of the extent. We could do better by only flushing NOCOW
3891          * data, but that needs extra accounting.
3892          *
3893          * Also we don't need to check ASYNC_EXTENT, as async extent will be
3894          * CoWed anyway, not affecting nocow part.
3895          */
3896         ret = filemap_flush(inode_in->i_mapping);
3897         if (ret < 0)
3898                 return ret;
3899
3900         ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
3901                                        wb_len);
3902         if (ret < 0)
3903                 return ret;
3904         ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
3905                                        wb_len);
3906         if (ret < 0)
3907                 return ret;
3908
3909         return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
3910                                             len, remap_flags);
3911 }
3912
3913 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
3914                 struct file *dst_file, loff_t destoff, loff_t len,
3915                 unsigned int remap_flags)
3916 {
3917         struct inode *src_inode = file_inode(src_file);
3918         struct inode *dst_inode = file_inode(dst_file);
3919         bool same_inode = dst_inode == src_inode;
3920         int ret;
3921
3922         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
3923                 return -EINVAL;
3924
3925         if (same_inode)
3926                 inode_lock(src_inode);
3927         else
3928                 lock_two_nondirectories(src_inode, dst_inode);
3929
3930         ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
3931                                           &len, remap_flags);
3932         if (ret < 0 || len == 0)
3933                 goto out_unlock;
3934
3935         if (remap_flags & REMAP_FILE_DEDUP)
3936                 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
3937         else
3938                 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
3939
3940 out_unlock:
3941         if (same_inode)
3942                 inode_unlock(src_inode);
3943         else
3944                 unlock_two_nondirectories(src_inode, dst_inode);
3945
3946         return ret < 0 ? ret : len;
3947 }
3948
3949 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3950 {
3951         struct inode *inode = file_inode(file);
3952         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3953         struct btrfs_root *root = BTRFS_I(inode)->root;
3954         struct btrfs_root *new_root;
3955         struct btrfs_dir_item *di;
3956         struct btrfs_trans_handle *trans;
3957         struct btrfs_path *path;
3958         struct btrfs_key location;
3959         struct btrfs_disk_key disk_key;
3960         u64 objectid = 0;
3961         u64 dir_id;
3962         int ret;
3963
3964         if (!capable(CAP_SYS_ADMIN))
3965                 return -EPERM;
3966
3967         ret = mnt_want_write_file(file);
3968         if (ret)
3969                 return ret;
3970
3971         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3972                 ret = -EFAULT;
3973                 goto out;
3974         }
3975
3976         if (!objectid)
3977                 objectid = BTRFS_FS_TREE_OBJECTID;
3978
3979         location.objectid = objectid;
3980         location.type = BTRFS_ROOT_ITEM_KEY;
3981         location.offset = (u64)-1;
3982
3983         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
3984         if (IS_ERR(new_root)) {
3985                 ret = PTR_ERR(new_root);
3986                 goto out;
3987         }
3988         if (!is_fstree(new_root->root_key.objectid)) {
3989                 ret = -ENOENT;
3990                 goto out;
3991         }
3992
3993         path = btrfs_alloc_path();
3994         if (!path) {
3995                 ret = -ENOMEM;
3996                 goto out;
3997         }
3998         path->leave_spinning = 1;
3999
4000         trans = btrfs_start_transaction(root, 1);
4001         if (IS_ERR(trans)) {
4002                 btrfs_free_path(path);
4003                 ret = PTR_ERR(trans);
4004                 goto out;
4005         }
4006
4007         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4008         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4009                                    dir_id, "default", 7, 1);
4010         if (IS_ERR_OR_NULL(di)) {
4011                 btrfs_free_path(path);
4012                 btrfs_end_transaction(trans);
4013                 btrfs_err(fs_info,
4014                           "Umm, you don't have the default diritem, this isn't going to work");
4015                 ret = -ENOENT;
4016                 goto out;
4017         }
4018
4019         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4020         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4021         btrfs_mark_buffer_dirty(path->nodes[0]);
4022         btrfs_free_path(path);
4023
4024         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4025         btrfs_end_transaction(trans);
4026 out:
4027         mnt_drop_write_file(file);
4028         return ret;
4029 }
4030
4031 static void get_block_group_info(struct list_head *groups_list,
4032                                  struct btrfs_ioctl_space_info *space)
4033 {
4034         struct btrfs_block_group *block_group;
4035
4036         space->total_bytes = 0;
4037         space->used_bytes = 0;
4038         space->flags = 0;
4039         list_for_each_entry(block_group, groups_list, list) {
4040                 space->flags = block_group->flags;
4041                 space->total_bytes += block_group->length;
4042                 space->used_bytes += block_group->used;
4043         }
4044 }
4045
4046 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4047                                    void __user *arg)
4048 {
4049         struct btrfs_ioctl_space_args space_args;
4050         struct btrfs_ioctl_space_info space;
4051         struct btrfs_ioctl_space_info *dest;
4052         struct btrfs_ioctl_space_info *dest_orig;
4053         struct btrfs_ioctl_space_info __user *user_dest;
4054         struct btrfs_space_info *info;
4055         static const u64 types[] = {
4056                 BTRFS_BLOCK_GROUP_DATA,
4057                 BTRFS_BLOCK_GROUP_SYSTEM,
4058                 BTRFS_BLOCK_GROUP_METADATA,
4059                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4060         };
4061         int num_types = 4;
4062         int alloc_size;
4063         int ret = 0;
4064         u64 slot_count = 0;
4065         int i, c;
4066
4067         if (copy_from_user(&space_args,
4068                            (struct btrfs_ioctl_space_args __user *)arg,
4069                            sizeof(space_args)))
4070                 return -EFAULT;
4071
4072         for (i = 0; i < num_types; i++) {
4073                 struct btrfs_space_info *tmp;
4074
4075                 info = NULL;
4076                 rcu_read_lock();
4077                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4078                                         list) {
4079                         if (tmp->flags == types[i]) {
4080                                 info = tmp;
4081                                 break;
4082                         }
4083                 }
4084                 rcu_read_unlock();
4085
4086                 if (!info)
4087                         continue;
4088
4089                 down_read(&info->groups_sem);
4090                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4091                         if (!list_empty(&info->block_groups[c]))
4092                                 slot_count++;
4093                 }
4094                 up_read(&info->groups_sem);
4095         }
4096
4097         /*
4098          * Global block reserve, exported as a space_info
4099          */
4100         slot_count++;
4101
4102         /* space_slots == 0 means they are asking for a count */
4103         if (space_args.space_slots == 0) {
4104                 space_args.total_spaces = slot_count;
4105                 goto out;
4106         }
4107
4108         slot_count = min_t(u64, space_args.space_slots, slot_count);
4109
4110         alloc_size = sizeof(*dest) * slot_count;
4111
4112         /* we generally have at most 6 or so space infos, one for each raid
4113          * level.  So, a whole page should be more than enough for everyone
4114          */
4115         if (alloc_size > PAGE_SIZE)
4116                 return -ENOMEM;
4117
4118         space_args.total_spaces = 0;
4119         dest = kmalloc(alloc_size, GFP_KERNEL);
4120         if (!dest)
4121                 return -ENOMEM;
4122         dest_orig = dest;
4123
4124         /* now we have a buffer to copy into */
4125         for (i = 0; i < num_types; i++) {
4126                 struct btrfs_space_info *tmp;
4127
4128                 if (!slot_count)
4129                         break;
4130
4131                 info = NULL;
4132                 rcu_read_lock();
4133                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4134                                         list) {
4135                         if (tmp->flags == types[i]) {
4136                                 info = tmp;
4137                                 break;
4138                         }
4139                 }
4140                 rcu_read_unlock();
4141
4142                 if (!info)
4143                         continue;
4144                 down_read(&info->groups_sem);
4145                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4146                         if (!list_empty(&info->block_groups[c])) {
4147                                 get_block_group_info(&info->block_groups[c],
4148                                                      &space);
4149                                 memcpy(dest, &space, sizeof(space));
4150                                 dest++;
4151                                 space_args.total_spaces++;
4152                                 slot_count--;
4153                         }
4154                         if (!slot_count)
4155                                 break;
4156                 }
4157                 up_read(&info->groups_sem);
4158         }
4159
4160         /*
4161          * Add global block reserve
4162          */
4163         if (slot_count) {
4164                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4165
4166                 spin_lock(&block_rsv->lock);
4167                 space.total_bytes = block_rsv->size;
4168                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4169                 spin_unlock(&block_rsv->lock);
4170                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4171                 memcpy(dest, &space, sizeof(space));
4172                 space_args.total_spaces++;
4173         }
4174
4175         user_dest = (struct btrfs_ioctl_space_info __user *)
4176                 (arg + sizeof(struct btrfs_ioctl_space_args));
4177
4178         if (copy_to_user(user_dest, dest_orig, alloc_size))
4179                 ret = -EFAULT;
4180
4181         kfree(dest_orig);
4182 out:
4183         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4184                 ret = -EFAULT;
4185
4186         return ret;
4187 }
4188
4189 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4190                                             void __user *argp)
4191 {
4192         struct btrfs_trans_handle *trans;
4193         u64 transid;
4194         int ret;
4195
4196         trans = btrfs_attach_transaction_barrier(root);
4197         if (IS_ERR(trans)) {
4198                 if (PTR_ERR(trans) != -ENOENT)
4199                         return PTR_ERR(trans);
4200
4201                 /* No running transaction, don't bother */
4202                 transid = root->fs_info->last_trans_committed;
4203                 goto out;
4204         }
4205         transid = trans->transid;
4206         ret = btrfs_commit_transaction_async(trans, 0);
4207         if (ret) {
4208                 btrfs_end_transaction(trans);
4209                 return ret;
4210         }
4211 out:
4212         if (argp)
4213                 if (copy_to_user(argp, &transid, sizeof(transid)))
4214                         return -EFAULT;
4215         return 0;
4216 }
4217
4218 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4219                                            void __user *argp)
4220 {
4221         u64 transid;
4222
4223         if (argp) {
4224                 if (copy_from_user(&transid, argp, sizeof(transid)))
4225                         return -EFAULT;
4226         } else {
4227                 transid = 0;  /* current trans */
4228         }
4229         return btrfs_wait_for_commit(fs_info, transid);
4230 }
4231
4232 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4233 {
4234         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4235         struct btrfs_ioctl_scrub_args *sa;
4236         int ret;
4237
4238         if (!capable(CAP_SYS_ADMIN))
4239                 return -EPERM;
4240
4241         sa = memdup_user(arg, sizeof(*sa));
4242         if (IS_ERR(sa))
4243                 return PTR_ERR(sa);
4244
4245         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4246                 ret = mnt_want_write_file(file);
4247                 if (ret)
4248                         goto out;
4249         }
4250
4251         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4252                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4253                               0);
4254
4255         /*
4256          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4257          * error. This is important as it allows user space to know how much
4258          * progress scrub has done. For example, if scrub is canceled we get
4259          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4260          * space. Later user space can inspect the progress from the structure
4261          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4262          * previously (btrfs-progs does this).
4263          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4264          * then return -EFAULT to signal the structure was not copied or it may
4265          * be corrupt and unreliable due to a partial copy.
4266          */
4267         if (copy_to_user(arg, sa, sizeof(*sa)))
4268                 ret = -EFAULT;
4269
4270         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4271                 mnt_drop_write_file(file);
4272 out:
4273         kfree(sa);
4274         return ret;
4275 }
4276
4277 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4278 {
4279         if (!capable(CAP_SYS_ADMIN))
4280                 return -EPERM;
4281
4282         return btrfs_scrub_cancel(fs_info);
4283 }
4284
4285 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4286                                        void __user *arg)
4287 {
4288         struct btrfs_ioctl_scrub_args *sa;
4289         int ret;
4290
4291         if (!capable(CAP_SYS_ADMIN))
4292                 return -EPERM;
4293
4294         sa = memdup_user(arg, sizeof(*sa));
4295         if (IS_ERR(sa))
4296                 return PTR_ERR(sa);
4297
4298         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4299
4300         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4301                 ret = -EFAULT;
4302
4303         kfree(sa);
4304         return ret;
4305 }
4306
4307 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4308                                       void __user *arg)
4309 {
4310         struct btrfs_ioctl_get_dev_stats *sa;
4311         int ret;
4312
4313         sa = memdup_user(arg, sizeof(*sa));
4314         if (IS_ERR(sa))
4315                 return PTR_ERR(sa);
4316
4317         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4318                 kfree(sa);
4319                 return -EPERM;
4320         }
4321
4322         ret = btrfs_get_dev_stats(fs_info, sa);
4323
4324         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4325                 ret = -EFAULT;
4326
4327         kfree(sa);
4328         return ret;
4329 }
4330
4331 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4332                                     void __user *arg)
4333 {
4334         struct btrfs_ioctl_dev_replace_args *p;
4335         int ret;
4336
4337         if (!capable(CAP_SYS_ADMIN))
4338                 return -EPERM;
4339
4340         p = memdup_user(arg, sizeof(*p));
4341         if (IS_ERR(p))
4342                 return PTR_ERR(p);
4343
4344         switch (p->cmd) {
4345         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4346                 if (sb_rdonly(fs_info->sb)) {
4347                         ret = -EROFS;
4348                         goto out;
4349                 }
4350                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4351                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4352                 } else {
4353                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4354                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4355                 }
4356                 break;
4357         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4358                 btrfs_dev_replace_status(fs_info, p);
4359                 ret = 0;
4360                 break;
4361         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4362                 p->result = btrfs_dev_replace_cancel(fs_info);
4363                 ret = 0;
4364                 break;
4365         default:
4366                 ret = -EINVAL;
4367                 break;
4368         }
4369
4370         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4371                 ret = -EFAULT;
4372 out:
4373         kfree(p);
4374         return ret;
4375 }
4376
4377 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4378 {
4379         int ret = 0;
4380         int i;
4381         u64 rel_ptr;
4382         int size;
4383         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4384         struct inode_fs_paths *ipath = NULL;
4385         struct btrfs_path *path;
4386
4387         if (!capable(CAP_DAC_READ_SEARCH))
4388                 return -EPERM;
4389
4390         path = btrfs_alloc_path();
4391         if (!path) {
4392                 ret = -ENOMEM;
4393                 goto out;
4394         }
4395
4396         ipa = memdup_user(arg, sizeof(*ipa));
4397         if (IS_ERR(ipa)) {
4398                 ret = PTR_ERR(ipa);
4399                 ipa = NULL;
4400                 goto out;
4401         }
4402
4403         size = min_t(u32, ipa->size, 4096);
4404         ipath = init_ipath(size, root, path);
4405         if (IS_ERR(ipath)) {
4406                 ret = PTR_ERR(ipath);
4407                 ipath = NULL;
4408                 goto out;
4409         }
4410
4411         ret = paths_from_inode(ipa->inum, ipath);
4412         if (ret < 0)
4413                 goto out;
4414
4415         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4416                 rel_ptr = ipath->fspath->val[i] -
4417                           (u64)(unsigned long)ipath->fspath->val;
4418                 ipath->fspath->val[i] = rel_ptr;
4419         }
4420
4421         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4422                            ipath->fspath, size);
4423         if (ret) {
4424                 ret = -EFAULT;
4425                 goto out;
4426         }
4427
4428 out:
4429         btrfs_free_path(path);
4430         free_ipath(ipath);
4431         kfree(ipa);
4432
4433         return ret;
4434 }
4435
4436 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4437 {
4438         struct btrfs_data_container *inodes = ctx;
4439         const size_t c = 3 * sizeof(u64);
4440
4441         if (inodes->bytes_left >= c) {
4442                 inodes->bytes_left -= c;
4443                 inodes->val[inodes->elem_cnt] = inum;
4444                 inodes->val[inodes->elem_cnt + 1] = offset;
4445                 inodes->val[inodes->elem_cnt + 2] = root;
4446                 inodes->elem_cnt += 3;
4447         } else {
4448                 inodes->bytes_missing += c - inodes->bytes_left;
4449                 inodes->bytes_left = 0;
4450                 inodes->elem_missed += 3;
4451         }
4452
4453         return 0;
4454 }
4455
4456 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4457                                         void __user *arg, int version)
4458 {
4459         int ret = 0;
4460         int size;
4461         struct btrfs_ioctl_logical_ino_args *loi;
4462         struct btrfs_data_container *inodes = NULL;
4463         struct btrfs_path *path = NULL;
4464         bool ignore_offset;
4465
4466         if (!capable(CAP_SYS_ADMIN))
4467                 return -EPERM;
4468
4469         loi = memdup_user(arg, sizeof(*loi));
4470         if (IS_ERR(loi))
4471                 return PTR_ERR(loi);
4472
4473         if (version == 1) {
4474                 ignore_offset = false;
4475                 size = min_t(u32, loi->size, SZ_64K);
4476         } else {
4477                 /* All reserved bits must be 0 for now */
4478                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4479                         ret = -EINVAL;
4480                         goto out_loi;
4481                 }
4482                 /* Only accept flags we have defined so far */
4483                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4484                         ret = -EINVAL;
4485                         goto out_loi;
4486                 }
4487                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4488                 size = min_t(u32, loi->size, SZ_16M);
4489         }
4490
4491         path = btrfs_alloc_path();
4492         if (!path) {
4493                 ret = -ENOMEM;
4494                 goto out;
4495         }
4496
4497         inodes = init_data_container(size);
4498         if (IS_ERR(inodes)) {
4499                 ret = PTR_ERR(inodes);
4500                 inodes = NULL;
4501                 goto out;
4502         }
4503
4504         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4505                                           build_ino_list, inodes, ignore_offset);
4506         if (ret == -EINVAL)
4507                 ret = -ENOENT;
4508         if (ret < 0)
4509                 goto out;
4510
4511         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4512                            size);
4513         if (ret)
4514                 ret = -EFAULT;
4515
4516 out:
4517         btrfs_free_path(path);
4518         kvfree(inodes);
4519 out_loi:
4520         kfree(loi);
4521
4522         return ret;
4523 }
4524
4525 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4526                                struct btrfs_ioctl_balance_args *bargs)
4527 {
4528         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4529
4530         bargs->flags = bctl->flags;
4531
4532         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4533                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4534         if (atomic_read(&fs_info->balance_pause_req))
4535                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4536         if (atomic_read(&fs_info->balance_cancel_req))
4537                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4538
4539         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4540         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4541         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4542
4543         spin_lock(&fs_info->balance_lock);
4544         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4545         spin_unlock(&fs_info->balance_lock);
4546 }
4547
4548 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4549 {
4550         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4551         struct btrfs_fs_info *fs_info = root->fs_info;
4552         struct btrfs_ioctl_balance_args *bargs;
4553         struct btrfs_balance_control *bctl;
4554         bool need_unlock; /* for mut. excl. ops lock */
4555         int ret;
4556
4557         if (!capable(CAP_SYS_ADMIN))
4558                 return -EPERM;
4559
4560         ret = mnt_want_write_file(file);
4561         if (ret)
4562                 return ret;
4563
4564 again:
4565         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4566                 mutex_lock(&fs_info->balance_mutex);
4567                 need_unlock = true;
4568                 goto locked;
4569         }
4570
4571         /*
4572          * mut. excl. ops lock is locked.  Three possibilities:
4573          *   (1) some other op is running
4574          *   (2) balance is running
4575          *   (3) balance is paused -- special case (think resume)
4576          */
4577         mutex_lock(&fs_info->balance_mutex);
4578         if (fs_info->balance_ctl) {
4579                 /* this is either (2) or (3) */
4580                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4581                         mutex_unlock(&fs_info->balance_mutex);
4582                         /*
4583                          * Lock released to allow other waiters to continue,
4584                          * we'll reexamine the status again.
4585                          */
4586                         mutex_lock(&fs_info->balance_mutex);
4587
4588                         if (fs_info->balance_ctl &&
4589                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4590                                 /* this is (3) */
4591                                 need_unlock = false;
4592                                 goto locked;
4593                         }
4594
4595                         mutex_unlock(&fs_info->balance_mutex);
4596                         goto again;
4597                 } else {
4598                         /* this is (2) */
4599                         mutex_unlock(&fs_info->balance_mutex);
4600                         ret = -EINPROGRESS;
4601                         goto out;
4602                 }
4603         } else {
4604                 /* this is (1) */
4605                 mutex_unlock(&fs_info->balance_mutex);
4606                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4607                 goto out;
4608         }
4609
4610 locked:
4611         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4612
4613         if (arg) {
4614                 bargs = memdup_user(arg, sizeof(*bargs));
4615                 if (IS_ERR(bargs)) {
4616                         ret = PTR_ERR(bargs);
4617                         goto out_unlock;
4618                 }
4619
4620                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4621                         if (!fs_info->balance_ctl) {
4622                                 ret = -ENOTCONN;
4623                                 goto out_bargs;
4624                         }
4625
4626                         bctl = fs_info->balance_ctl;
4627                         spin_lock(&fs_info->balance_lock);
4628                         bctl->flags |= BTRFS_BALANCE_RESUME;
4629                         spin_unlock(&fs_info->balance_lock);
4630
4631                         goto do_balance;
4632                 }
4633         } else {
4634                 bargs = NULL;
4635         }
4636
4637         if (fs_info->balance_ctl) {
4638                 ret = -EINPROGRESS;
4639                 goto out_bargs;
4640         }
4641
4642         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4643         if (!bctl) {
4644                 ret = -ENOMEM;
4645                 goto out_bargs;
4646         }
4647
4648         if (arg) {
4649                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4650                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4651                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4652
4653                 bctl->flags = bargs->flags;
4654         } else {
4655                 /* balance everything - no filters */
4656                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4657         }
4658
4659         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4660                 ret = -EINVAL;
4661                 goto out_bctl;
4662         }
4663
4664 do_balance:
4665         /*
4666          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4667          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4668          * restriper was paused all the way until unmount, in free_fs_info.
4669          * The flag should be cleared after reset_balance_state.
4670          */
4671         need_unlock = false;
4672
4673         ret = btrfs_balance(fs_info, bctl, bargs);
4674         bctl = NULL;
4675
4676         if ((ret == 0 || ret == -ECANCELED) && arg) {
4677                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4678                         ret = -EFAULT;
4679         }
4680
4681 out_bctl:
4682         kfree(bctl);
4683 out_bargs:
4684         kfree(bargs);
4685 out_unlock:
4686         mutex_unlock(&fs_info->balance_mutex);
4687         if (need_unlock)
4688                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4689 out:
4690         mnt_drop_write_file(file);
4691         return ret;
4692 }
4693
4694 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4695 {
4696         if (!capable(CAP_SYS_ADMIN))
4697                 return -EPERM;
4698
4699         switch (cmd) {
4700         case BTRFS_BALANCE_CTL_PAUSE:
4701                 return btrfs_pause_balance(fs_info);
4702         case BTRFS_BALANCE_CTL_CANCEL:
4703                 return btrfs_cancel_balance(fs_info);
4704         }
4705
4706         return -EINVAL;
4707 }
4708
4709 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4710                                          void __user *arg)
4711 {
4712         struct btrfs_ioctl_balance_args *bargs;
4713         int ret = 0;
4714
4715         if (!capable(CAP_SYS_ADMIN))
4716                 return -EPERM;
4717
4718         mutex_lock(&fs_info->balance_mutex);
4719         if (!fs_info->balance_ctl) {
4720                 ret = -ENOTCONN;
4721                 goto out;
4722         }
4723
4724         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4725         if (!bargs) {
4726                 ret = -ENOMEM;
4727                 goto out;
4728         }
4729
4730         btrfs_update_ioctl_balance_args(fs_info, bargs);
4731
4732         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4733                 ret = -EFAULT;
4734
4735         kfree(bargs);
4736 out:
4737         mutex_unlock(&fs_info->balance_mutex);
4738         return ret;
4739 }
4740
4741 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4742 {
4743         struct inode *inode = file_inode(file);
4744         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4745         struct btrfs_ioctl_quota_ctl_args *sa;
4746         int ret;
4747
4748         if (!capable(CAP_SYS_ADMIN))
4749                 return -EPERM;
4750
4751         ret = mnt_want_write_file(file);
4752         if (ret)
4753                 return ret;
4754
4755         sa = memdup_user(arg, sizeof(*sa));
4756         if (IS_ERR(sa)) {
4757                 ret = PTR_ERR(sa);
4758                 goto drop_write;
4759         }
4760
4761         down_write(&fs_info->subvol_sem);
4762
4763         switch (sa->cmd) {
4764         case BTRFS_QUOTA_CTL_ENABLE:
4765                 ret = btrfs_quota_enable(fs_info);
4766                 break;
4767         case BTRFS_QUOTA_CTL_DISABLE:
4768                 ret = btrfs_quota_disable(fs_info);
4769                 break;
4770         default:
4771                 ret = -EINVAL;
4772                 break;
4773         }
4774
4775         kfree(sa);
4776         up_write(&fs_info->subvol_sem);
4777 drop_write:
4778         mnt_drop_write_file(file);
4779         return ret;
4780 }
4781
4782 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4783 {
4784         struct inode *inode = file_inode(file);
4785         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4786         struct btrfs_root *root = BTRFS_I(inode)->root;
4787         struct btrfs_ioctl_qgroup_assign_args *sa;
4788         struct btrfs_trans_handle *trans;
4789         int ret;
4790         int err;
4791
4792         if (!capable(CAP_SYS_ADMIN))
4793                 return -EPERM;
4794
4795         ret = mnt_want_write_file(file);
4796         if (ret)
4797                 return ret;
4798
4799         sa = memdup_user(arg, sizeof(*sa));
4800         if (IS_ERR(sa)) {
4801                 ret = PTR_ERR(sa);
4802                 goto drop_write;
4803         }
4804
4805         trans = btrfs_join_transaction(root);
4806         if (IS_ERR(trans)) {
4807                 ret = PTR_ERR(trans);
4808                 goto out;
4809         }
4810
4811         if (sa->assign) {
4812                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4813         } else {
4814                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4815         }
4816
4817         /* update qgroup status and info */
4818         err = btrfs_run_qgroups(trans);
4819         if (err < 0)
4820                 btrfs_handle_fs_error(fs_info, err,
4821                                       "failed to update qgroup status and info");
4822         err = btrfs_end_transaction(trans);
4823         if (err && !ret)
4824                 ret = err;
4825
4826 out:
4827         kfree(sa);
4828 drop_write:
4829         mnt_drop_write_file(file);
4830         return ret;
4831 }
4832
4833 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4834 {
4835         struct inode *inode = file_inode(file);
4836         struct btrfs_root *root = BTRFS_I(inode)->root;
4837         struct btrfs_ioctl_qgroup_create_args *sa;
4838         struct btrfs_trans_handle *trans;
4839         int ret;
4840         int err;
4841
4842         if (!capable(CAP_SYS_ADMIN))
4843                 return -EPERM;
4844
4845         ret = mnt_want_write_file(file);
4846         if (ret)
4847                 return ret;
4848
4849         sa = memdup_user(arg, sizeof(*sa));
4850         if (IS_ERR(sa)) {
4851                 ret = PTR_ERR(sa);
4852                 goto drop_write;
4853         }
4854
4855         if (!sa->qgroupid) {
4856                 ret = -EINVAL;
4857                 goto out;
4858         }
4859
4860         trans = btrfs_join_transaction(root);
4861         if (IS_ERR(trans)) {
4862                 ret = PTR_ERR(trans);
4863                 goto out;
4864         }
4865
4866         if (sa->create) {
4867                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4868         } else {
4869                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4870         }
4871
4872         err = btrfs_end_transaction(trans);
4873         if (err && !ret)
4874                 ret = err;
4875
4876 out:
4877         kfree(sa);
4878 drop_write:
4879         mnt_drop_write_file(file);
4880         return ret;
4881 }
4882
4883 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4884 {
4885         struct inode *inode = file_inode(file);
4886         struct btrfs_root *root = BTRFS_I(inode)->root;
4887         struct btrfs_ioctl_qgroup_limit_args *sa;
4888         struct btrfs_trans_handle *trans;
4889         int ret;
4890         int err;
4891         u64 qgroupid;
4892
4893         if (!capable(CAP_SYS_ADMIN))
4894                 return -EPERM;
4895
4896         ret = mnt_want_write_file(file);
4897         if (ret)
4898                 return ret;
4899
4900         sa = memdup_user(arg, sizeof(*sa));
4901         if (IS_ERR(sa)) {
4902                 ret = PTR_ERR(sa);
4903                 goto drop_write;
4904         }
4905
4906         trans = btrfs_join_transaction(root);
4907         if (IS_ERR(trans)) {
4908                 ret = PTR_ERR(trans);
4909                 goto out;
4910         }
4911
4912         qgroupid = sa->qgroupid;
4913         if (!qgroupid) {
4914                 /* take the current subvol as qgroup */
4915                 qgroupid = root->root_key.objectid;
4916         }
4917
4918         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4919
4920         err = btrfs_end_transaction(trans);
4921         if (err && !ret)
4922                 ret = err;
4923
4924 out:
4925         kfree(sa);
4926 drop_write:
4927         mnt_drop_write_file(file);
4928         return ret;
4929 }
4930
4931 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4932 {
4933         struct inode *inode = file_inode(file);
4934         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4935         struct btrfs_ioctl_quota_rescan_args *qsa;
4936         int ret;
4937
4938         if (!capable(CAP_SYS_ADMIN))
4939                 return -EPERM;
4940
4941         ret = mnt_want_write_file(file);
4942         if (ret)
4943                 return ret;
4944
4945         qsa = memdup_user(arg, sizeof(*qsa));
4946         if (IS_ERR(qsa)) {
4947                 ret = PTR_ERR(qsa);
4948                 goto drop_write;
4949         }
4950
4951         if (qsa->flags) {
4952                 ret = -EINVAL;
4953                 goto out;
4954         }
4955
4956         ret = btrfs_qgroup_rescan(fs_info);
4957
4958 out:
4959         kfree(qsa);
4960 drop_write:
4961         mnt_drop_write_file(file);
4962         return ret;
4963 }
4964
4965 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4966                                                 void __user *arg)
4967 {
4968         struct btrfs_ioctl_quota_rescan_args *qsa;
4969         int ret = 0;
4970
4971         if (!capable(CAP_SYS_ADMIN))
4972                 return -EPERM;
4973
4974         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4975         if (!qsa)
4976                 return -ENOMEM;
4977
4978         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4979                 qsa->flags = 1;
4980                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4981         }
4982
4983         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4984                 ret = -EFAULT;
4985
4986         kfree(qsa);
4987         return ret;
4988 }
4989
4990 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4991                                                 void __user *arg)
4992 {
4993         if (!capable(CAP_SYS_ADMIN))
4994                 return -EPERM;
4995
4996         return btrfs_qgroup_wait_for_completion(fs_info, true);
4997 }
4998
4999 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5000                                             struct btrfs_ioctl_received_subvol_args *sa)
5001 {
5002         struct inode *inode = file_inode(file);
5003         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5004         struct btrfs_root *root = BTRFS_I(inode)->root;
5005         struct btrfs_root_item *root_item = &root->root_item;
5006         struct btrfs_trans_handle *trans;
5007         struct timespec64 ct = current_time(inode);
5008         int ret = 0;
5009         int received_uuid_changed;
5010
5011         if (!inode_owner_or_capable(inode))
5012                 return -EPERM;
5013
5014         ret = mnt_want_write_file(file);
5015         if (ret < 0)
5016                 return ret;
5017
5018         down_write(&fs_info->subvol_sem);
5019
5020         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5021                 ret = -EINVAL;
5022                 goto out;
5023         }
5024
5025         if (btrfs_root_readonly(root)) {
5026                 ret = -EROFS;
5027                 goto out;
5028         }
5029
5030         /*
5031          * 1 - root item
5032          * 2 - uuid items (received uuid + subvol uuid)
5033          */
5034         trans = btrfs_start_transaction(root, 3);
5035         if (IS_ERR(trans)) {
5036                 ret = PTR_ERR(trans);
5037                 trans = NULL;
5038                 goto out;
5039         }
5040
5041         sa->rtransid = trans->transid;
5042         sa->rtime.sec = ct.tv_sec;
5043         sa->rtime.nsec = ct.tv_nsec;
5044
5045         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5046                                        BTRFS_UUID_SIZE);
5047         if (received_uuid_changed &&
5048             !btrfs_is_empty_uuid(root_item->received_uuid)) {
5049                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5050                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5051                                           root->root_key.objectid);
5052                 if (ret && ret != -ENOENT) {
5053                         btrfs_abort_transaction(trans, ret);
5054                         btrfs_end_transaction(trans);
5055                         goto out;
5056                 }
5057         }
5058         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5059         btrfs_set_root_stransid(root_item, sa->stransid);
5060         btrfs_set_root_rtransid(root_item, sa->rtransid);
5061         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5062         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5063         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5064         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5065
5066         ret = btrfs_update_root(trans, fs_info->tree_root,
5067                                 &root->root_key, &root->root_item);
5068         if (ret < 0) {
5069                 btrfs_end_transaction(trans);
5070                 goto out;
5071         }
5072         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5073                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
5074                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5075                                           root->root_key.objectid);
5076                 if (ret < 0 && ret != -EEXIST) {
5077                         btrfs_abort_transaction(trans, ret);
5078                         btrfs_end_transaction(trans);
5079                         goto out;
5080                 }
5081         }
5082         ret = btrfs_commit_transaction(trans);
5083 out:
5084         up_write(&fs_info->subvol_sem);
5085         mnt_drop_write_file(file);
5086         return ret;
5087 }
5088
5089 #ifdef CONFIG_64BIT
5090 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5091                                                 void __user *arg)
5092 {
5093         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5094         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5095         int ret = 0;
5096
5097         args32 = memdup_user(arg, sizeof(*args32));
5098         if (IS_ERR(args32))
5099                 return PTR_ERR(args32);
5100
5101         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5102         if (!args64) {
5103                 ret = -ENOMEM;
5104                 goto out;
5105         }
5106
5107         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5108         args64->stransid = args32->stransid;
5109         args64->rtransid = args32->rtransid;
5110         args64->stime.sec = args32->stime.sec;
5111         args64->stime.nsec = args32->stime.nsec;
5112         args64->rtime.sec = args32->rtime.sec;
5113         args64->rtime.nsec = args32->rtime.nsec;
5114         args64->flags = args32->flags;
5115
5116         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5117         if (ret)
5118                 goto out;
5119
5120         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5121         args32->stransid = args64->stransid;
5122         args32->rtransid = args64->rtransid;
5123         args32->stime.sec = args64->stime.sec;
5124         args32->stime.nsec = args64->stime.nsec;
5125         args32->rtime.sec = args64->rtime.sec;
5126         args32->rtime.nsec = args64->rtime.nsec;
5127         args32->flags = args64->flags;
5128
5129         ret = copy_to_user(arg, args32, sizeof(*args32));
5130         if (ret)
5131                 ret = -EFAULT;
5132
5133 out:
5134         kfree(args32);
5135         kfree(args64);
5136         return ret;
5137 }
5138 #endif
5139
5140 static long btrfs_ioctl_set_received_subvol(struct file *file,
5141                                             void __user *arg)
5142 {
5143         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5144         int ret = 0;
5145
5146         sa = memdup_user(arg, sizeof(*sa));
5147         if (IS_ERR(sa))
5148                 return PTR_ERR(sa);
5149
5150         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5151
5152         if (ret)
5153                 goto out;
5154
5155         ret = copy_to_user(arg, sa, sizeof(*sa));
5156         if (ret)
5157                 ret = -EFAULT;
5158
5159 out:
5160         kfree(sa);
5161         return ret;
5162 }
5163
5164 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
5165                                         void __user *arg)
5166 {
5167         size_t len;
5168         int ret;
5169         char label[BTRFS_LABEL_SIZE];
5170
5171         spin_lock(&fs_info->super_lock);
5172         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5173         spin_unlock(&fs_info->super_lock);
5174
5175         len = strnlen(label, BTRFS_LABEL_SIZE);
5176
5177         if (len == BTRFS_LABEL_SIZE) {
5178                 btrfs_warn(fs_info,
5179                            "label is too long, return the first %zu bytes",
5180                            --len);
5181         }
5182
5183         ret = copy_to_user(arg, label, len);
5184
5185         return ret ? -EFAULT : 0;
5186 }
5187
5188 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5189 {
5190         struct inode *inode = file_inode(file);
5191         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5192         struct btrfs_root *root = BTRFS_I(inode)->root;
5193         struct btrfs_super_block *super_block = fs_info->super_copy;
5194         struct btrfs_trans_handle *trans;
5195         char label[BTRFS_LABEL_SIZE];
5196         int ret;
5197
5198         if (!capable(CAP_SYS_ADMIN))
5199                 return -EPERM;
5200
5201         if (copy_from_user(label, arg, sizeof(label)))
5202                 return -EFAULT;
5203
5204         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5205                 btrfs_err(fs_info,
5206                           "unable to set label with more than %d bytes",
5207                           BTRFS_LABEL_SIZE - 1);
5208                 return -EINVAL;
5209         }
5210
5211         ret = mnt_want_write_file(file);
5212         if (ret)
5213                 return ret;
5214
5215         trans = btrfs_start_transaction(root, 0);
5216         if (IS_ERR(trans)) {
5217                 ret = PTR_ERR(trans);
5218                 goto out_unlock;
5219         }
5220
5221         spin_lock(&fs_info->super_lock);
5222         strcpy(super_block->label, label);
5223         spin_unlock(&fs_info->super_lock);
5224         ret = btrfs_commit_transaction(trans);
5225
5226 out_unlock:
5227         mnt_drop_write_file(file);
5228         return ret;
5229 }
5230
5231 #define INIT_FEATURE_FLAGS(suffix) \
5232         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5233           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5234           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5235
5236 int btrfs_ioctl_get_supported_features(void __user *arg)
5237 {
5238         static const struct btrfs_ioctl_feature_flags features[3] = {
5239                 INIT_FEATURE_FLAGS(SUPP),
5240                 INIT_FEATURE_FLAGS(SAFE_SET),
5241                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5242         };
5243
5244         if (copy_to_user(arg, &features, sizeof(features)))
5245                 return -EFAULT;
5246
5247         return 0;
5248 }
5249
5250 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5251                                         void __user *arg)
5252 {
5253         struct btrfs_super_block *super_block = fs_info->super_copy;
5254         struct btrfs_ioctl_feature_flags features;
5255
5256         features.compat_flags = btrfs_super_compat_flags(super_block);
5257         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5258         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5259
5260         if (copy_to_user(arg, &features, sizeof(features)))
5261                 return -EFAULT;
5262
5263         return 0;
5264 }
5265
5266 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5267                               enum btrfs_feature_set set,
5268                               u64 change_mask, u64 flags, u64 supported_flags,
5269                               u64 safe_set, u64 safe_clear)
5270 {
5271         const char *type = btrfs_feature_set_name(set);
5272         char *names;
5273         u64 disallowed, unsupported;
5274         u64 set_mask = flags & change_mask;
5275         u64 clear_mask = ~flags & change_mask;
5276
5277         unsupported = set_mask & ~supported_flags;
5278         if (unsupported) {
5279                 names = btrfs_printable_features(set, unsupported);
5280                 if (names) {
5281                         btrfs_warn(fs_info,
5282                                    "this kernel does not support the %s feature bit%s",
5283                                    names, strchr(names, ',') ? "s" : "");
5284                         kfree(names);
5285                 } else
5286                         btrfs_warn(fs_info,
5287                                    "this kernel does not support %s bits 0x%llx",
5288                                    type, unsupported);
5289                 return -EOPNOTSUPP;
5290         }
5291
5292         disallowed = set_mask & ~safe_set;
5293         if (disallowed) {
5294                 names = btrfs_printable_features(set, disallowed);
5295                 if (names) {
5296                         btrfs_warn(fs_info,
5297                                    "can't set the %s feature bit%s while mounted",
5298                                    names, strchr(names, ',') ? "s" : "");
5299                         kfree(names);
5300                 } else
5301                         btrfs_warn(fs_info,
5302                                    "can't set %s bits 0x%llx while mounted",
5303                                    type, disallowed);
5304                 return -EPERM;
5305         }
5306
5307         disallowed = clear_mask & ~safe_clear;
5308         if (disallowed) {
5309                 names = btrfs_printable_features(set, disallowed);
5310                 if (names) {
5311                         btrfs_warn(fs_info,
5312                                    "can't clear the %s feature bit%s while mounted",
5313                                    names, strchr(names, ',') ? "s" : "");
5314                         kfree(names);
5315                 } else
5316                         btrfs_warn(fs_info,
5317                                    "can't clear %s bits 0x%llx while mounted",
5318                                    type, disallowed);
5319                 return -EPERM;
5320         }
5321
5322         return 0;
5323 }
5324
5325 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5326 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5327                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5328                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5329                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5330
5331 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5332 {
5333         struct inode *inode = file_inode(file);
5334         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5335         struct btrfs_root *root = BTRFS_I(inode)->root;
5336         struct btrfs_super_block *super_block = fs_info->super_copy;
5337         struct btrfs_ioctl_feature_flags flags[2];
5338         struct btrfs_trans_handle *trans;
5339         u64 newflags;
5340         int ret;
5341
5342         if (!capable(CAP_SYS_ADMIN))
5343                 return -EPERM;
5344
5345         if (copy_from_user(flags, arg, sizeof(flags)))
5346                 return -EFAULT;
5347
5348         /* Nothing to do */
5349         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5350             !flags[0].incompat_flags)
5351                 return 0;
5352
5353         ret = check_feature(fs_info, flags[0].compat_flags,
5354                             flags[1].compat_flags, COMPAT);
5355         if (ret)
5356                 return ret;
5357
5358         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5359                             flags[1].compat_ro_flags, COMPAT_RO);
5360         if (ret)
5361                 return ret;
5362
5363         ret = check_feature(fs_info, flags[0].incompat_flags,
5364                             flags[1].incompat_flags, INCOMPAT);
5365         if (ret)
5366                 return ret;
5367
5368         ret = mnt_want_write_file(file);
5369         if (ret)
5370                 return ret;
5371
5372         trans = btrfs_start_transaction(root, 0);
5373         if (IS_ERR(trans)) {
5374                 ret = PTR_ERR(trans);
5375                 goto out_drop_write;
5376         }
5377
5378         spin_lock(&fs_info->super_lock);
5379         newflags = btrfs_super_compat_flags(super_block);
5380         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5381         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5382         btrfs_set_super_compat_flags(super_block, newflags);
5383
5384         newflags = btrfs_super_compat_ro_flags(super_block);
5385         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5386         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5387         btrfs_set_super_compat_ro_flags(super_block, newflags);
5388
5389         newflags = btrfs_super_incompat_flags(super_block);
5390         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5391         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5392         btrfs_set_super_incompat_flags(super_block, newflags);
5393         spin_unlock(&fs_info->super_lock);
5394
5395         ret = btrfs_commit_transaction(trans);
5396 out_drop_write:
5397         mnt_drop_write_file(file);
5398
5399         return ret;
5400 }
5401
5402 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5403 {
5404         struct btrfs_ioctl_send_args *arg;
5405         int ret;
5406
5407         if (compat) {
5408 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5409                 struct btrfs_ioctl_send_args_32 args32;
5410
5411                 ret = copy_from_user(&args32, argp, sizeof(args32));
5412                 if (ret)
5413                         return -EFAULT;
5414                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5415                 if (!arg)
5416                         return -ENOMEM;
5417                 arg->send_fd = args32.send_fd;
5418                 arg->clone_sources_count = args32.clone_sources_count;
5419                 arg->clone_sources = compat_ptr(args32.clone_sources);
5420                 arg->parent_root = args32.parent_root;
5421                 arg->flags = args32.flags;
5422                 memcpy(arg->reserved, args32.reserved,
5423                        sizeof(args32.reserved));
5424 #else
5425                 return -ENOTTY;
5426 #endif
5427         } else {
5428                 arg = memdup_user(argp, sizeof(*arg));
5429                 if (IS_ERR(arg))
5430                         return PTR_ERR(arg);
5431         }
5432         ret = btrfs_ioctl_send(file, arg);
5433         kfree(arg);
5434         return ret;
5435 }
5436
5437 long btrfs_ioctl(struct file *file, unsigned int
5438                 cmd, unsigned long arg)
5439 {
5440         struct inode *inode = file_inode(file);
5441         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5442         struct btrfs_root *root = BTRFS_I(inode)->root;
5443         void __user *argp = (void __user *)arg;
5444
5445         switch (cmd) {
5446         case FS_IOC_GETFLAGS:
5447                 return btrfs_ioctl_getflags(file, argp);
5448         case FS_IOC_SETFLAGS:
5449                 return btrfs_ioctl_setflags(file, argp);
5450         case FS_IOC_GETVERSION:
5451                 return btrfs_ioctl_getversion(file, argp);
5452         case FS_IOC_GETFSLABEL:
5453                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5454         case FS_IOC_SETFSLABEL:
5455                 return btrfs_ioctl_set_fslabel(file, argp);
5456         case FITRIM:
5457                 return btrfs_ioctl_fitrim(fs_info, argp);
5458         case BTRFS_IOC_SNAP_CREATE:
5459                 return btrfs_ioctl_snap_create(file, argp, 0);
5460         case BTRFS_IOC_SNAP_CREATE_V2:
5461                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5462         case BTRFS_IOC_SUBVOL_CREATE:
5463                 return btrfs_ioctl_snap_create(file, argp, 1);
5464         case BTRFS_IOC_SUBVOL_CREATE_V2:
5465                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5466         case BTRFS_IOC_SNAP_DESTROY:
5467                 return btrfs_ioctl_snap_destroy(file, argp);
5468         case BTRFS_IOC_SUBVOL_GETFLAGS:
5469                 return btrfs_ioctl_subvol_getflags(file, argp);
5470         case BTRFS_IOC_SUBVOL_SETFLAGS:
5471                 return btrfs_ioctl_subvol_setflags(file, argp);
5472         case BTRFS_IOC_DEFAULT_SUBVOL:
5473                 return btrfs_ioctl_default_subvol(file, argp);
5474         case BTRFS_IOC_DEFRAG:
5475                 return btrfs_ioctl_defrag(file, NULL);
5476         case BTRFS_IOC_DEFRAG_RANGE:
5477                 return btrfs_ioctl_defrag(file, argp);
5478         case BTRFS_IOC_RESIZE:
5479                 return btrfs_ioctl_resize(file, argp);
5480         case BTRFS_IOC_ADD_DEV:
5481                 return btrfs_ioctl_add_dev(fs_info, argp);
5482         case BTRFS_IOC_RM_DEV:
5483                 return btrfs_ioctl_rm_dev(file, argp);
5484         case BTRFS_IOC_RM_DEV_V2:
5485                 return btrfs_ioctl_rm_dev_v2(file, argp);
5486         case BTRFS_IOC_FS_INFO:
5487                 return btrfs_ioctl_fs_info(fs_info, argp);
5488         case BTRFS_IOC_DEV_INFO:
5489                 return btrfs_ioctl_dev_info(fs_info, argp);
5490         case BTRFS_IOC_BALANCE:
5491                 return btrfs_ioctl_balance(file, NULL);
5492         case BTRFS_IOC_TREE_SEARCH:
5493                 return btrfs_ioctl_tree_search(file, argp);
5494         case BTRFS_IOC_TREE_SEARCH_V2:
5495                 return btrfs_ioctl_tree_search_v2(file, argp);
5496         case BTRFS_IOC_INO_LOOKUP:
5497                 return btrfs_ioctl_ino_lookup(file, argp);
5498         case BTRFS_IOC_INO_PATHS:
5499                 return btrfs_ioctl_ino_to_path(root, argp);
5500         case BTRFS_IOC_LOGICAL_INO:
5501                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5502         case BTRFS_IOC_LOGICAL_INO_V2:
5503                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5504         case BTRFS_IOC_SPACE_INFO:
5505                 return btrfs_ioctl_space_info(fs_info, argp);
5506         case BTRFS_IOC_SYNC: {
5507                 int ret;
5508
5509                 ret = btrfs_start_delalloc_roots(fs_info, -1);
5510                 if (ret)
5511                         return ret;
5512                 ret = btrfs_sync_fs(inode->i_sb, 1);
5513                 /*
5514                  * The transaction thread may want to do more work,
5515                  * namely it pokes the cleaner kthread that will start
5516                  * processing uncleaned subvols.
5517                  */
5518                 wake_up_process(fs_info->transaction_kthread);
5519                 return ret;
5520         }
5521         case BTRFS_IOC_START_SYNC:
5522                 return btrfs_ioctl_start_sync(root, argp);
5523         case BTRFS_IOC_WAIT_SYNC:
5524                 return btrfs_ioctl_wait_sync(fs_info, argp);
5525         case BTRFS_IOC_SCRUB:
5526                 return btrfs_ioctl_scrub(file, argp);
5527         case BTRFS_IOC_SCRUB_CANCEL:
5528                 return btrfs_ioctl_scrub_cancel(fs_info);
5529         case BTRFS_IOC_SCRUB_PROGRESS:
5530                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5531         case BTRFS_IOC_BALANCE_V2:
5532                 return btrfs_ioctl_balance(file, argp);
5533         case BTRFS_IOC_BALANCE_CTL:
5534                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5535         case BTRFS_IOC_BALANCE_PROGRESS:
5536                 return btrfs_ioctl_balance_progress(fs_info, argp);
5537         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5538                 return btrfs_ioctl_set_received_subvol(file, argp);
5539 #ifdef CONFIG_64BIT
5540         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5541                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5542 #endif
5543         case BTRFS_IOC_SEND:
5544                 return _btrfs_ioctl_send(file, argp, false);
5545 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5546         case BTRFS_IOC_SEND_32:
5547                 return _btrfs_ioctl_send(file, argp, true);
5548 #endif
5549         case BTRFS_IOC_GET_DEV_STATS:
5550                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5551         case BTRFS_IOC_QUOTA_CTL:
5552                 return btrfs_ioctl_quota_ctl(file, argp);
5553         case BTRFS_IOC_QGROUP_ASSIGN:
5554                 return btrfs_ioctl_qgroup_assign(file, argp);
5555         case BTRFS_IOC_QGROUP_CREATE:
5556                 return btrfs_ioctl_qgroup_create(file, argp);
5557         case BTRFS_IOC_QGROUP_LIMIT:
5558                 return btrfs_ioctl_qgroup_limit(file, argp);
5559         case BTRFS_IOC_QUOTA_RESCAN:
5560                 return btrfs_ioctl_quota_rescan(file, argp);
5561         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5562                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5563         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5564                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5565         case BTRFS_IOC_DEV_REPLACE:
5566                 return btrfs_ioctl_dev_replace(fs_info, argp);
5567         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5568                 return btrfs_ioctl_get_supported_features(argp);
5569         case BTRFS_IOC_GET_FEATURES:
5570                 return btrfs_ioctl_get_features(fs_info, argp);
5571         case BTRFS_IOC_SET_FEATURES:
5572                 return btrfs_ioctl_set_features(file, argp);
5573         case FS_IOC_FSGETXATTR:
5574                 return btrfs_ioctl_fsgetxattr(file, argp);
5575         case FS_IOC_FSSETXATTR:
5576                 return btrfs_ioctl_fssetxattr(file, argp);
5577         case BTRFS_IOC_GET_SUBVOL_INFO:
5578                 return btrfs_ioctl_get_subvol_info(file, argp);
5579         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5580                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5581         case BTRFS_IOC_INO_LOOKUP_USER:
5582                 return btrfs_ioctl_ino_lookup_user(file, argp);
5583         }
5584
5585         return -ENOTTY;
5586 }
5587
5588 #ifdef CONFIG_COMPAT
5589 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5590 {
5591         /*
5592          * These all access 32-bit values anyway so no further
5593          * handling is necessary.
5594          */
5595         switch (cmd) {
5596         case FS_IOC32_GETFLAGS:
5597                 cmd = FS_IOC_GETFLAGS;
5598                 break;
5599         case FS_IOC32_SETFLAGS:
5600                 cmd = FS_IOC_SETFLAGS;
5601                 break;
5602         case FS_IOC32_GETVERSION:
5603                 cmd = FS_IOC_GETVERSION;
5604                 break;
5605         }
5606
5607         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5608 }
5609 #endif