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         BUG_ON(ret);
 708
 709         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 710                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 711         BUG_ON(ret);
 712
 713         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 714                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 715         if (ret)
 716                 btrfs_abort_transaction(trans, ret);
 717
 718 fail:
 719         kfree(root_item);
 720         trans->block_rsv = NULL;
 721         trans->bytes_reserved = 0;
 722         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 723
 724         if (async_transid) {
 725                 *async_transid = trans->transid;
 726                 err = btrfs_commit_transaction_async(trans, 1);
 727                 if (err)
 728                         err = btrfs_commit_transaction(trans);
 729         } else {
 730                 err = btrfs_commit_transaction(trans);
 731         }
 732         if (err && !ret)
 733                 ret = err;
 734
 735         if (!ret) {
 736                 inode = btrfs_lookup_dentry(dir, dentry);
 737                 if (IS_ERR(inode))
 738                         return PTR_ERR(inode);
 739                 d_instantiate(dentry, inode);
 740         }
 741         return ret;
 742
 743 fail_free:
 744         kfree(root_item);
 745         return ret;
 746 }
 747
 748 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 749                            struct dentry *dentry,
 750                            u64 *async_transid, bool readonly,
 751                            struct btrfs_qgroup_inherit *inherit)
 752 {
 753         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 754         struct inode *inode;
 755         struct btrfs_pending_snapshot *pending_snapshot;
 756         struct btrfs_trans_handle *trans;
 757         int ret;
 758         bool snapshot_force_cow = false;
 759
 760         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 761                 return -EINVAL;
 762
 763         if (atomic_read(&root->nr_swapfiles)) {
 764                 btrfs_warn(fs_info,
 765                            "cannot snapshot subvolume with active swapfile");
 766                 return -ETXTBSY;
 767         }
 768
 769         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 770         if (!pending_snapshot)
 771                 return -ENOMEM;
 772
 773         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 774                         GFP_KERNEL);
 775         pending_snapshot->path = btrfs_alloc_path();
 776         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 777                 ret = -ENOMEM;
 778                 goto free_pending;
 779         }
 780
 781         /*
 782          * Force new buffered writes to reserve space even when NOCOW is
 783          * possible. This is to avoid later writeback (running dealloc) to
 784          * fallback to COW mode and unexpectedly fail with ENOSPC.
 785          */
 786         atomic_inc(&root->will_be_snapshotted);
 787         smp_mb__after_atomic();
 788         /* wait for no snapshot writes */
 789         wait_event(root->subv_writers->wait,
 790                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 791
 792         ret = btrfs_start_delalloc_snapshot(root);
 793         if (ret)
 794                 goto dec_and_free;
 795
 796         /*
 797          * All previous writes have started writeback in NOCOW mode, so now
 798          * we force future writes to fallback to COW mode during snapshot
 799          * creation.
 800          */
 801         atomic_inc(&root->snapshot_force_cow);
 802         snapshot_force_cow = true;
 803
 804         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 805
 806         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 807                              BTRFS_BLOCK_RSV_TEMP);
 808         /*
 809          * 1 - parent dir inode
 810          * 2 - dir entries
 811          * 1 - root item
 812          * 2 - root ref/backref
 813          * 1 - root of snapshot
 814          * 1 - UUID item
 815          */
 816         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 817                                         &pending_snapshot->block_rsv, 8,
 818                                         false);
 819         if (ret)
 820                 goto dec_and_free;
 821
 822         pending_snapshot->dentry = dentry;
 823         pending_snapshot->root = root;
 824         pending_snapshot->readonly = readonly;
 825         pending_snapshot->dir = dir;
 826         pending_snapshot->inherit = inherit;
 827
 828         trans = btrfs_start_transaction(root, 0);
 829         if (IS_ERR(trans)) {
 830                 ret = PTR_ERR(trans);
 831                 goto fail;
 832         }
 833
 834         spin_lock(&fs_info->trans_lock);
 835         list_add(&pending_snapshot->list,
 836                  &trans->transaction->pending_snapshots);
 837         spin_unlock(&fs_info->trans_lock);
 838         if (async_transid) {
 839                 *async_transid = trans->transid;
 840                 ret = btrfs_commit_transaction_async(trans, 1);
 841                 if (ret)
 842                         ret = btrfs_commit_transaction(trans);
 843         } else {
 844                 ret = btrfs_commit_transaction(trans);
 845         }
 846         if (ret)
 847                 goto fail;
 848
 849         ret = pending_snapshot->error;
 850         if (ret)
 851                 goto fail;
 852
 853         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 854         if (ret)
 855                 goto fail;
 856
 857         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 858         if (IS_ERR(inode)) {
 859                 ret = PTR_ERR(inode);
 860                 goto fail;
 861         }
 862
 863         d_instantiate(dentry, inode);
 864         ret = 0;
 865 fail:
 866         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 867 dec_and_free:
 868         if (snapshot_force_cow)
 869                 atomic_dec(&root->snapshot_force_cow);
 870         if (atomic_dec_and_test(&root->will_be_snapshotted))
 871                 wake_up_var(&root->will_be_snapshotted);
 872 free_pending:
 873         kfree(pending_snapshot->root_item);
 874         btrfs_free_path(pending_snapshot->path);
 875         kfree(pending_snapshot);
 876
 877         return ret;
 878 }
 879
 880 /*  copy of may_delete in fs/namei.c()
 881  *      Check whether we can remove a link victim from directory dir, check
 882  *  whether the type of victim is right.
 883  *  1. We can't do it if dir is read-only (done in permission())
 884  *  2. We should have write and exec permissions on dir
 885  *  3. We can't remove anything from append-only dir
 886  *  4. We can't do anything with immutable dir (done in permission())
 887  *  5. If the sticky bit on dir is set we should either
 888  *      a. be owner of dir, or
 889  *      b. be owner of victim, or
 890  *      c. have CAP_FOWNER capability
 891  *  6. If the victim is append-only or immutable we can't do anything with
 892  *     links pointing to it.
 893  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 894  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 895  *  9. We can't remove a root or mountpoint.
 896  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 897  *     nfs_async_unlink().
 898  */
 899
 900 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 901 {
 902         int error;
 903
 904         if (d_really_is_negative(victim))
 905                 return -ENOENT;
 906
 907         BUG_ON(d_inode(victim->d_parent) != dir);
 908         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 909
 910         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 911         if (error)
 912                 return error;
 913         if (IS_APPEND(dir))
 914                 return -EPERM;
 915         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 916             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 917                 return -EPERM;
 918         if (isdir) {
 919                 if (!d_is_dir(victim))
 920                         return -ENOTDIR;
 921                 if (IS_ROOT(victim))
 922                         return -EBUSY;
 923         } else if (d_is_dir(victim))
 924                 return -EISDIR;
 925         if (IS_DEADDIR(dir))
 926                 return -ENOENT;
 927         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 928                 return -EBUSY;
 929         return 0;
 930 }
 931
 932 /* copy of may_create in fs/namei.c() */
 933 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 934 {
 935         if (d_really_is_positive(child))
 936                 return -EEXIST;
 937         if (IS_DEADDIR(dir))
 938                 return -ENOENT;
 939         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 940 }
 941
 942 /*
 943  * Create a new subvolume below @parent.  This is largely modeled after
 944  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 945  * inside this filesystem so it's quite a bit simpler.
 946  */
 947 static noinline int btrfs_mksubvol(const struct path *parent,
 948                                    const char *name, int namelen,
 949                                    struct btrfs_root *snap_src,
 950                                    u64 *async_transid, bool readonly,
 951                                    struct btrfs_qgroup_inherit *inherit)
 952 {
 953         struct inode *dir = d_inode(parent->dentry);
 954         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 955         struct dentry *dentry;
 956         int error;
 957
 958         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 959         if (error == -EINTR)
 960                 return error;
 961
 962         dentry = lookup_one_len(name, parent->dentry, namelen);
 963         error = PTR_ERR(dentry);
 964         if (IS_ERR(dentry))
 965                 goto out_unlock;
 966
 967         error = btrfs_may_create(dir, dentry);
 968         if (error)
 969                 goto out_dput;
 970
 971         /*
 972          * even if this name doesn't exist, we may get hash collisions.
 973          * check for them now when we can safely fail
 974          */
 975         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 976                                                dir->i_ino, name,
 977                                                namelen);
 978         if (error)
 979                 goto out_dput;
 980
 981         down_read(&fs_info->subvol_sem);
 982
 983         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 984                 goto out_up_read;
 985
 986         if (snap_src) {
 987                 error = create_snapshot(snap_src, dir, dentry,
 988                                         async_transid, readonly, inherit);
 989         } else {
 990                 error = create_subvol(dir, dentry, name, namelen,
 991                                       async_transid, inherit);
 992         }
 993         if (!error)
 994                 fsnotify_mkdir(dir, dentry);
 995 out_up_read:
 996         up_read(&fs_info->subvol_sem);
 997 out_dput:
 998         dput(dentry);
 999 out_unlock:
1000         inode_unlock(dir);
1001         return error;
1002 }
1003
1004 /*
1005  * When we're defragging a range, we don't want to kick it off again
1006  * if it is really just waiting for delalloc to send it down.
1007  * If we find a nice big extent or delalloc range for the bytes in the
1008  * file you want to defrag, we return 0 to let you know to skip this
1009  * part of the file
1010  */
1011 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1012 {
1013         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1014         struct extent_map *em = NULL;
1015         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1016         u64 end;
1017
1018         read_lock(&em_tree->lock);
1019         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1020         read_unlock(&em_tree->lock);
1021
1022         if (em) {
1023                 end = extent_map_end(em);
1024                 free_extent_map(em);
1025                 if (end - offset > thresh)
1026                         return 0;
1027         }
1028         /* if we already have a nice delalloc here, just stop */
1029         thresh /= 2;
1030         end = count_range_bits(io_tree, &offset, offset + thresh,
1031                                thresh, EXTENT_DELALLOC, 1);
1032         if (end >= thresh)
1033                 return 0;
1034         return 1;
1035 }
1036
1037 /*
1038  * helper function to walk through a file and find extents
1039  * newer than a specific transid, and smaller than thresh.
1040  *
1041  * This is used by the defragging code to find new and small
1042  * extents
1043  */
1044 static int find_new_extents(struct btrfs_root *root,
1045                             struct inode *inode, u64 newer_than,
1046                             u64 *off, u32 thresh)
1047 {
1048         struct btrfs_path *path;
1049         struct btrfs_key min_key;
1050         struct extent_buffer *leaf;
1051         struct btrfs_file_extent_item *extent;
1052         int type;
1053         int ret;
1054         u64 ino = btrfs_ino(BTRFS_I(inode));
1055
1056         path = btrfs_alloc_path();
1057         if (!path)
1058                 return -ENOMEM;
1059
1060         min_key.objectid = ino;
1061         min_key.type = BTRFS_EXTENT_DATA_KEY;
1062         min_key.offset = *off;
1063
1064         while (1) {
1065                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1066                 if (ret != 0)
1067                         goto none;
1068 process_slot:
1069                 if (min_key.objectid != ino)
1070                         goto none;
1071                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1072                         goto none;
1073
1074                 leaf = path->nodes[0];
1075                 extent = btrfs_item_ptr(leaf, path->slots[0],
1076                                         struct btrfs_file_extent_item);
1077
1078                 type = btrfs_file_extent_type(leaf, extent);
1079                 if (type == BTRFS_FILE_EXTENT_REG &&
1080                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1081                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1082                         *off = min_key.offset;
1083                         btrfs_free_path(path);
1084                         return 0;
1085                 }
1086
1087                 path->slots[0]++;
1088                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1089                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1090                         goto process_slot;
1091                 }
1092
1093                 if (min_key.offset == (u64)-1)
1094                         goto none;
1095
1096                 min_key.offset++;
1097                 btrfs_release_path(path);
1098         }
1099 none:
1100         btrfs_free_path(path);
1101         return -ENOENT;
1102 }
1103
1104 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1105 {
1106         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1107         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1108         struct extent_map *em;
1109         u64 len = PAGE_SIZE;
1110
1111         /*
1112          * hopefully we have this extent in the tree already, try without
1113          * the full extent lock
1114          */
1115         read_lock(&em_tree->lock);
1116         em = lookup_extent_mapping(em_tree, start, len);
1117         read_unlock(&em_tree->lock);
1118
1119         if (!em) {
1120                 struct extent_state *cached = NULL;
1121                 u64 end = start + len - 1;
1122
1123                 /* get the big lock and read metadata off disk */
1124                 lock_extent_bits(io_tree, start, end, &cached);
1125                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1126                 unlock_extent_cached(io_tree, start, end, &cached);
1127
1128                 if (IS_ERR(em))
1129                         return NULL;
1130         }
1131
1132         return em;
1133 }
1134
1135 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1136 {
1137         struct extent_map *next;
1138         bool ret = true;
1139
1140         /* this is the last extent */
1141         if (em->start + em->len >= i_size_read(inode))
1142                 return false;
1143
1144         next = defrag_lookup_extent(inode, em->start + em->len);
1145         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1146                 ret = false;
1147         else if ((em->block_start + em->block_len == next->block_start) &&
1148                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1149                 ret = false;
1150
1151         free_extent_map(next);
1152         return ret;
1153 }
1154
1155 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1156                                u64 *last_len, u64 *skip, u64 *defrag_end,
1157                                int compress)
1158 {
1159         struct extent_map *em;
1160         int ret = 1;
1161         bool next_mergeable = true;
1162         bool prev_mergeable = true;
1163
1164         /*
1165          * make sure that once we start defragging an extent, we keep on
1166          * defragging it
1167          */
1168         if (start < *defrag_end)
1169                 return 1;
1170
1171         *skip = 0;
1172
1173         em = defrag_lookup_extent(inode, start);
1174         if (!em)
1175                 return 0;
1176
1177         /* this will cover holes, and inline extents */
1178         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1179                 ret = 0;
1180                 goto out;
1181         }
1182
1183         if (!*defrag_end)
1184                 prev_mergeable = false;
1185
1186         next_mergeable = defrag_check_next_extent(inode, em);
1187         /*
1188          * we hit a real extent, if it is big or the next extent is not a
1189          * real extent, don't bother defragging it
1190          */
1191         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1192             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1193                 ret = 0;
1194 out:
1195         /*
1196          * last_len ends up being a counter of how many bytes we've defragged.
1197          * every time we choose not to defrag an extent, we reset *last_len
1198          * so that the next tiny extent will force a defrag.
1199          *
1200          * The end result of this is that tiny extents before a single big
1201          * extent will force at least part of that big extent to be defragged.
1202          */
1203         if (ret) {
1204                 *defrag_end = extent_map_end(em);
1205         } else {
1206                 *last_len = 0;
1207                 *skip = extent_map_end(em);
1208                 *defrag_end = 0;
1209         }
1210
1211         free_extent_map(em);
1212         return ret;
1213 }
1214
1215 /*
1216  * it doesn't do much good to defrag one or two pages
1217  * at a time.  This pulls in a nice chunk of pages
1218  * to COW and defrag.
1219  *
1220  * It also makes sure the delalloc code has enough
1221  * dirty data to avoid making new small extents as part
1222  * of the defrag
1223  *
1224  * It's a good idea to start RA on this range
1225  * before calling this.
1226  */
1227 static int cluster_pages_for_defrag(struct inode *inode,
1228                                     struct page **pages,
1229                                     unsigned long start_index,
1230                                     unsigned long num_pages)
1231 {
1232         unsigned long file_end;
1233         u64 isize = i_size_read(inode);
1234         u64 page_start;
1235         u64 page_end;
1236         u64 page_cnt;
1237         int ret;
1238         int i;
1239         int i_done;
1240         struct btrfs_ordered_extent *ordered;
1241         struct extent_state *cached_state = NULL;
1242         struct extent_io_tree *tree;
1243         struct extent_changeset *data_reserved = NULL;
1244         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1245
1246         file_end = (isize - 1) >> PAGE_SHIFT;
1247         if (!isize || start_index > file_end)
1248                 return 0;
1249
1250         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1251
1252         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1253                         start_index << PAGE_SHIFT,
1254                         page_cnt << PAGE_SHIFT);
1255         if (ret)
1256                 return ret;
1257         i_done = 0;
1258         tree = &BTRFS_I(inode)->io_tree;
1259
1260         /* step one, lock all the pages */
1261         for (i = 0; i < page_cnt; i++) {
1262                 struct page *page;
1263 again:
1264                 page = find_or_create_page(inode->i_mapping,
1265                                            start_index + i, mask);
1266                 if (!page)
1267                         break;
1268
1269                 page_start = page_offset(page);
1270                 page_end = page_start + PAGE_SIZE - 1;
1271                 while (1) {
1272                         lock_extent_bits(tree, page_start, page_end,
1273                                          &cached_state);
1274                         ordered = btrfs_lookup_ordered_extent(inode,
1275                                                               page_start);
1276                         unlock_extent_cached(tree, page_start, page_end,
1277                                              &cached_state);
1278                         if (!ordered)
1279                                 break;
1280
1281                         unlock_page(page);
1282                         btrfs_start_ordered_extent(inode, ordered, 1);
1283                         btrfs_put_ordered_extent(ordered);
1284                         lock_page(page);
1285                         /*
1286                          * we unlocked the page above, so we need check if
1287                          * it was released or not.
1288                          */
1289                         if (page->mapping != inode->i_mapping) {
1290                                 unlock_page(page);
1291                                 put_page(page);
1292                                 goto again;
1293                         }
1294                 }
1295
1296                 if (!PageUptodate(page)) {
1297                         btrfs_readpage(NULL, page);
1298                         lock_page(page);
1299                         if (!PageUptodate(page)) {
1300                                 unlock_page(page);
1301                                 put_page(page);
1302                                 ret = -EIO;
1303                                 break;
1304                         }
1305                 }
1306
1307                 if (page->mapping != inode->i_mapping) {
1308                         unlock_page(page);
1309                         put_page(page);
1310                         goto again;
1311                 }
1312
1313                 pages[i] = page;
1314                 i_done++;
1315         }
1316         if (!i_done || ret)
1317                 goto out;
1318
1319         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1320                 goto out;
1321
1322         /*
1323          * so now we have a nice long stream of locked
1324          * and up to date pages, lets wait on them
1325          */
1326         for (i = 0; i < i_done; i++)
1327                 wait_on_page_writeback(pages[i]);
1328
1329         page_start = page_offset(pages[0]);
1330         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1331
1332         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1333                          page_start, page_end - 1, &cached_state);
1334         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1335                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1336                           EXTENT_DEFRAG, 0, 0, &cached_state);
1337
1338         if (i_done != page_cnt) {
1339                 spin_lock(&BTRFS_I(inode)->lock);
1340                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1341                 spin_unlock(&BTRFS_I(inode)->lock);
1342                 btrfs_delalloc_release_space(inode, data_reserved,
1343                                 start_index << PAGE_SHIFT,
1344                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1345         }
1346
1347
1348         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1349                           &cached_state);
1350
1351         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1352                              page_start, page_end - 1, &cached_state);
1353
1354         for (i = 0; i < i_done; i++) {
1355                 clear_page_dirty_for_io(pages[i]);
1356                 ClearPageChecked(pages[i]);
1357                 set_page_extent_mapped(pages[i]);
1358                 set_page_dirty(pages[i]);
1359                 unlock_page(pages[i]);
1360                 put_page(pages[i]);
1361         }
1362         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1363         extent_changeset_free(data_reserved);
1364         return i_done;
1365 out:
1366         for (i = 0; i < i_done; i++) {
1367                 unlock_page(pages[i]);
1368                 put_page(pages[i]);
1369         }
1370         btrfs_delalloc_release_space(inode, data_reserved,
1371                         start_index << PAGE_SHIFT,
1372                         page_cnt << PAGE_SHIFT, true);
1373         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1374         extent_changeset_free(data_reserved);
1375         return ret;
1376
1377 }
1378
1379 int btrfs_defrag_file(struct inode *inode, struct file *file,
1380                       struct btrfs_ioctl_defrag_range_args *range,
1381                       u64 newer_than, unsigned long max_to_defrag)
1382 {
1383         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1384         struct btrfs_root *root = BTRFS_I(inode)->root;
1385         struct file_ra_state *ra = NULL;
1386         unsigned long last_index;
1387         u64 isize = i_size_read(inode);
1388         u64 last_len = 0;
1389         u64 skip = 0;
1390         u64 defrag_end = 0;
1391         u64 newer_off = range->start;
1392         unsigned long i;
1393         unsigned long ra_index = 0;
1394         int ret;
1395         int defrag_count = 0;
1396         int compress_type = BTRFS_COMPRESS_ZLIB;
1397         u32 extent_thresh = range->extent_thresh;
1398         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1399         unsigned long cluster = max_cluster;
1400         u64 new_align = ~((u64)SZ_128K - 1);
1401         struct page **pages = NULL;
1402         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1403
1404         if (isize == 0)
1405                 return 0;
1406
1407         if (range->start >= isize)
1408                 return -EINVAL;
1409
1410         if (do_compress) {
1411                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1412                         return -EINVAL;
1413                 if (range->compress_type)
1414                         compress_type = range->compress_type;
1415         }
1416
1417         if (extent_thresh == 0)
1418                 extent_thresh = SZ_256K;
1419
1420         /*
1421          * If we were not given a file, allocate a readahead context. As
1422          * readahead is just an optimization, defrag will work without it so
1423          * we don't error out.
1424          */
1425         if (!file) {
1426                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1427                 if (ra)
1428                         file_ra_state_init(ra, inode->i_mapping);
1429         } else {
1430                 ra = &file->f_ra;
1431         }
1432
1433         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1434         if (!pages) {
1435                 ret = -ENOMEM;
1436                 goto out_ra;
1437         }
1438
1439         /* find the last page to defrag */
1440         if (range->start + range->len > range->start) {
1441                 last_index = min_t(u64, isize - 1,
1442                          range->start + range->len - 1) >> PAGE_SHIFT;
1443         } else {
1444                 last_index = (isize - 1) >> PAGE_SHIFT;
1445         }
1446
1447         if (newer_than) {
1448                 ret = find_new_extents(root, inode, newer_than,
1449                                        &newer_off, SZ_64K);
1450                 if (!ret) {
1451                         range->start = newer_off;
1452                         /*
1453                          * we always align our defrag to help keep
1454                          * the extents in the file evenly spaced
1455                          */
1456                         i = (newer_off & new_align) >> PAGE_SHIFT;
1457                 } else
1458                         goto out_ra;
1459         } else {
1460                 i = range->start >> PAGE_SHIFT;
1461         }
1462         if (!max_to_defrag)
1463                 max_to_defrag = last_index - i + 1;
1464
1465         /*
1466          * make writeback starts from i, so the defrag range can be
1467          * written sequentially.
1468          */
1469         if (i < inode->i_mapping->writeback_index)
1470                 inode->i_mapping->writeback_index = i;
1471
1472         while (i <= last_index && defrag_count < max_to_defrag &&
1473                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1474                 /*
1475                  * make sure we stop running if someone unmounts
1476                  * the FS
1477                  */
1478                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1479                         break;
1480
1481                 if (btrfs_defrag_cancelled(fs_info)) {
1482                         btrfs_debug(fs_info, "defrag_file cancelled");
1483                         ret = -EAGAIN;
1484                         break;
1485                 }
1486
1487                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1488                                          extent_thresh, &last_len, &skip,
1489                                          &defrag_end, do_compress)){
1490                         unsigned long next;
1491                         /*
1492                          * the should_defrag function tells us how much to skip
1493                          * bump our counter by the suggested amount
1494                          */
1495                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1496                         i = max(i + 1, next);
1497                         continue;
1498                 }
1499
1500                 if (!newer_than) {
1501                         cluster = (PAGE_ALIGN(defrag_end) >>
1502                                    PAGE_SHIFT) - i;
1503                         cluster = min(cluster, max_cluster);
1504                 } else {
1505                         cluster = max_cluster;
1506                 }
1507
1508                 if (i + cluster > ra_index) {
1509                         ra_index = max(i, ra_index);
1510                         if (ra)
1511                                 page_cache_sync_readahead(inode->i_mapping, ra,
1512                                                 file, ra_index, cluster);
1513                         ra_index += cluster;
1514                 }
1515
1516                 inode_lock(inode);
1517                 if (IS_SWAPFILE(inode)) {
1518                         ret = -ETXTBSY;
1519                 } else {
1520                         if (do_compress)
1521                                 BTRFS_I(inode)->defrag_compress = compress_type;
1522                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1523                 }
1524                 if (ret < 0) {
1525                         inode_unlock(inode);
1526                         goto out_ra;
1527                 }
1528
1529                 defrag_count += ret;
1530                 balance_dirty_pages_ratelimited(inode->i_mapping);
1531                 inode_unlock(inode);
1532
1533                 if (newer_than) {
1534                         if (newer_off == (u64)-1)
1535                                 break;
1536
1537                         if (ret > 0)
1538                                 i += ret;
1539
1540                         newer_off = max(newer_off + 1,
1541                                         (u64)i << PAGE_SHIFT);
1542
1543                         ret = find_new_extents(root, inode, newer_than,
1544                                                &newer_off, SZ_64K);
1545                         if (!ret) {
1546                                 range->start = newer_off;
1547                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1548                         } else {
1549                                 break;
1550                         }
1551                 } else {
1552                         if (ret > 0) {
1553                                 i += ret;
1554                                 last_len += ret << PAGE_SHIFT;
1555                         } else {
1556                                 i++;
1557                                 last_len = 0;
1558                         }
1559                 }
1560         }
1561
1562         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1563                 filemap_flush(inode->i_mapping);
1564                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1565                              &BTRFS_I(inode)->runtime_flags))
1566                         filemap_flush(inode->i_mapping);
1567         }
1568
1569         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1570                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1571         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1572                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1573         }
1574
1575         ret = defrag_count;
1576
1577 out_ra:
1578         if (do_compress) {
1579                 inode_lock(inode);
1580                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1581                 inode_unlock(inode);
1582         }
1583         if (!file)
1584                 kfree(ra);
1585         kfree(pages);
1586         return ret;
1587 }
1588
1589 static noinline int btrfs_ioctl_resize(struct file *file,
1590                                         void __user *arg)
1591 {
1592         struct inode *inode = file_inode(file);
1593         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1594         u64 new_size;
1595         u64 old_size;
1596         u64 devid = 1;
1597         struct btrfs_root *root = BTRFS_I(inode)->root;
1598         struct btrfs_ioctl_vol_args *vol_args;
1599         struct btrfs_trans_handle *trans;
1600         struct btrfs_device *device = NULL;
1601         char *sizestr;
1602         char *retptr;
1603         char *devstr = NULL;
1604         int ret = 0;
1605         int mod = 0;
1606
1607         if (!capable(CAP_SYS_ADMIN))
1608                 return -EPERM;
1609
1610         ret = mnt_want_write_file(file);
1611         if (ret)
1612                 return ret;
1613
1614         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1615                 mnt_drop_write_file(file);
1616                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1617         }
1618
1619         vol_args = memdup_user(arg, sizeof(*vol_args));
1620         if (IS_ERR(vol_args)) {
1621                 ret = PTR_ERR(vol_args);
1622                 goto out;
1623         }
1624
1625         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1626
1627         sizestr = vol_args->name;
1628         devstr = strchr(sizestr, ':');
1629         if (devstr) {
1630                 sizestr = devstr + 1;
1631                 *devstr = '\0';
1632                 devstr = vol_args->name;
1633                 ret = kstrtoull(devstr, 10, &devid);
1634                 if (ret)
1635                         goto out_free;
1636                 if (!devid) {
1637                         ret = -EINVAL;
1638                         goto out_free;
1639                 }
1640                 btrfs_info(fs_info, "resizing devid %llu", devid);
1641         }
1642
1643         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1644         if (!device) {
1645                 btrfs_info(fs_info, "resizer unable to find device %llu",
1646                            devid);
1647                 ret = -ENODEV;
1648                 goto out_free;
1649         }
1650
1651         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1652                 btrfs_info(fs_info,
1653                            "resizer unable to apply on readonly device %llu",
1654                        devid);
1655                 ret = -EPERM;
1656                 goto out_free;
1657         }
1658
1659         if (!strcmp(sizestr, "max"))
1660                 new_size = device->bdev->bd_inode->i_size;
1661         else {
1662                 if (sizestr[0] == '-') {
1663                         mod = -1;
1664                         sizestr++;
1665                 } else if (sizestr[0] == '+') {
1666                         mod = 1;
1667                         sizestr++;
1668                 }
1669                 new_size = memparse(sizestr, &retptr);
1670                 if (*retptr != '\0' || new_size == 0) {
1671                         ret = -EINVAL;
1672                         goto out_free;
1673                 }
1674         }
1675
1676         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1677                 ret = -EPERM;
1678                 goto out_free;
1679         }
1680
1681         old_size = btrfs_device_get_total_bytes(device);
1682
1683         if (mod < 0) {
1684                 if (new_size > old_size) {
1685                         ret = -EINVAL;
1686                         goto out_free;
1687                 }
1688                 new_size = old_size - new_size;
1689         } else if (mod > 0) {
1690                 if (new_size > ULLONG_MAX - old_size) {
1691                         ret = -ERANGE;
1692                         goto out_free;
1693                 }
1694                 new_size = old_size + new_size;
1695         }
1696
1697         if (new_size < SZ_256M) {
1698                 ret = -EINVAL;
1699                 goto out_free;
1700         }
1701         if (new_size > device->bdev->bd_inode->i_size) {
1702                 ret = -EFBIG;
1703                 goto out_free;
1704         }
1705
1706         new_size = round_down(new_size, fs_info->sectorsize);
1707
1708         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1709                           rcu_str_deref(device->name), new_size);
1710
1711         if (new_size > old_size) {
1712                 trans = btrfs_start_transaction(root, 0);
1713                 if (IS_ERR(trans)) {
1714                         ret = PTR_ERR(trans);
1715                         goto out_free;
1716                 }
1717                 ret = btrfs_grow_device(trans, device, new_size);
1718                 btrfs_commit_transaction(trans);
1719         } else if (new_size < old_size) {
1720                 ret = btrfs_shrink_device(device, new_size);
1721         } /* equal, nothing need to do */
1722
1723 out_free:
1724         kfree(vol_args);
1725 out:
1726         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1727         mnt_drop_write_file(file);
1728         return ret;
1729 }
1730
1731 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1732                                 const char *name, unsigned long fd, int subvol,
1733                                 u64 *transid, bool readonly,
1734                                 struct btrfs_qgroup_inherit *inherit)
1735 {
1736         int namelen;
1737         int ret = 0;
1738
1739         if (!S_ISDIR(file_inode(file)->i_mode))
1740                 return -ENOTDIR;
1741
1742         ret = mnt_want_write_file(file);
1743         if (ret)
1744                 goto out;
1745
1746         namelen = strlen(name);
1747         if (strchr(name, '/')) {
1748                 ret = -EINVAL;
1749                 goto out_drop_write;
1750         }
1751
1752         if (name[0] == '.' &&
1753            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1754                 ret = -EEXIST;
1755                 goto out_drop_write;
1756         }
1757
1758         if (subvol) {
1759                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1760                                      NULL, transid, readonly, inherit);
1761         } else {
1762                 struct fd src = fdget(fd);
1763                 struct inode *src_inode;
1764                 if (!src.file) {
1765                         ret = -EINVAL;
1766                         goto out_drop_write;
1767                 }
1768
1769                 src_inode = file_inode(src.file);
1770                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1771                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1772                                    "Snapshot src from another FS");
1773                         ret = -EXDEV;
1774                 } else if (!inode_owner_or_capable(src_inode)) {
1775                         /*
1776                          * Subvolume creation is not restricted, but snapshots
1777                          * are limited to own subvolumes only
1778                          */
1779                         ret = -EPERM;
1780                 } else {
1781                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1782                                              BTRFS_I(src_inode)->root,
1783                                              transid, readonly, inherit);
1784                 }
1785                 fdput(src);
1786         }
1787 out_drop_write:
1788         mnt_drop_write_file(file);
1789 out:
1790         return ret;
1791 }
1792
1793 static noinline int btrfs_ioctl_snap_create(struct file *file,
1794                                             void __user *arg, int subvol)
1795 {
1796         struct btrfs_ioctl_vol_args *vol_args;
1797         int ret;
1798
1799         if (!S_ISDIR(file_inode(file)->i_mode))
1800                 return -ENOTDIR;
1801
1802         vol_args = memdup_user(arg, sizeof(*vol_args));
1803         if (IS_ERR(vol_args))
1804                 return PTR_ERR(vol_args);
1805         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1806
1807         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1808                                               vol_args->fd, subvol,
1809                                               NULL, false, NULL);
1810
1811         kfree(vol_args);
1812         return ret;
1813 }
1814
1815 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1816                                                void __user *arg, int subvol)
1817 {
1818         struct btrfs_ioctl_vol_args_v2 *vol_args;
1819         int ret;
1820         u64 transid = 0;
1821         u64 *ptr = NULL;
1822         bool readonly = false;
1823         struct btrfs_qgroup_inherit *inherit = NULL;
1824
1825         if (!S_ISDIR(file_inode(file)->i_mode))
1826                 return -ENOTDIR;
1827
1828         vol_args = memdup_user(arg, sizeof(*vol_args));
1829         if (IS_ERR(vol_args))
1830                 return PTR_ERR(vol_args);
1831         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1832
1833         if (vol_args->flags &
1834             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1835               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1836                 ret = -EOPNOTSUPP;
1837                 goto free_args;
1838         }
1839
1840         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1841                 struct inode *inode = file_inode(file);
1842                 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1843
1844                 btrfs_warn(fs_info,
1845 "SNAP_CREATE_V2 ioctl with CREATE_ASYNC is deprecated and will be removed in kernel 5.7");
1846
1847                 ptr = &transid;
1848         }
1849         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1850                 readonly = true;
1851         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1852                 if (vol_args->size > PAGE_SIZE) {
1853                         ret = -EINVAL;
1854                         goto free_args;
1855                 }
1856                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1857                 if (IS_ERR(inherit)) {
1858                         ret = PTR_ERR(inherit);
1859                         goto free_args;
1860                 }
1861         }
1862
1863         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1864                                               vol_args->fd, subvol, ptr,
1865                                               readonly, inherit);
1866         if (ret)
1867                 goto free_inherit;
1868
1869         if (ptr && copy_to_user(arg +
1870                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1871                                         transid),
1872                                 ptr, sizeof(*ptr)))
1873                 ret = -EFAULT;
1874
1875 free_inherit:
1876         kfree(inherit);
1877 free_args:
1878         kfree(vol_args);
1879         return ret;
1880 }
1881
1882 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1883                                                 void __user *arg)
1884 {
1885         struct inode *inode = file_inode(file);
1886         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1887         struct btrfs_root *root = BTRFS_I(inode)->root;
1888         int ret = 0;
1889         u64 flags = 0;
1890
1891         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1892                 return -EINVAL;
1893
1894         down_read(&fs_info->subvol_sem);
1895         if (btrfs_root_readonly(root))
1896                 flags |= BTRFS_SUBVOL_RDONLY;
1897         up_read(&fs_info->subvol_sem);
1898
1899         if (copy_to_user(arg, &flags, sizeof(flags)))
1900                 ret = -EFAULT;
1901
1902         return ret;
1903 }
1904
1905 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1906                                               void __user *arg)
1907 {
1908         struct inode *inode = file_inode(file);
1909         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1910         struct btrfs_root *root = BTRFS_I(inode)->root;
1911         struct btrfs_trans_handle *trans;
1912         u64 root_flags;
1913         u64 flags;
1914         int ret = 0;
1915
1916         if (!inode_owner_or_capable(inode))
1917                 return -EPERM;
1918
1919         ret = mnt_want_write_file(file);
1920         if (ret)
1921                 goto out;
1922
1923         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1924                 ret = -EINVAL;
1925                 goto out_drop_write;
1926         }
1927
1928         if (copy_from_user(&flags, arg, sizeof(flags))) {
1929                 ret = -EFAULT;
1930                 goto out_drop_write;
1931         }
1932
1933         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1934                 ret = -EINVAL;
1935                 goto out_drop_write;
1936         }
1937
1938         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1939                 ret = -EOPNOTSUPP;
1940                 goto out_drop_write;
1941         }
1942
1943         down_write(&fs_info->subvol_sem);
1944
1945         /* nothing to do */
1946         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1947                 goto out_drop_sem;
1948
1949         root_flags = btrfs_root_flags(&root->root_item);
1950         if (flags & BTRFS_SUBVOL_RDONLY) {
1951                 btrfs_set_root_flags(&root->root_item,
1952                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1953         } else {
1954                 /*
1955                  * Block RO -> RW transition if this subvolume is involved in
1956                  * send
1957                  */
1958                 spin_lock(&root->root_item_lock);
1959                 if (root->send_in_progress == 0) {
1960                         btrfs_set_root_flags(&root->root_item,
1961                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1962                         spin_unlock(&root->root_item_lock);
1963                 } else {
1964                         spin_unlock(&root->root_item_lock);
1965                         btrfs_warn(fs_info,
1966                                    "Attempt to set subvolume %llu read-write during send",
1967                                    root->root_key.objectid);
1968                         ret = -EPERM;
1969                         goto out_drop_sem;
1970                 }
1971         }
1972
1973         trans = btrfs_start_transaction(root, 1);
1974         if (IS_ERR(trans)) {
1975                 ret = PTR_ERR(trans);
1976                 goto out_reset;
1977         }
1978
1979         ret = btrfs_update_root(trans, fs_info->tree_root,
1980                                 &root->root_key, &root->root_item);
1981         if (ret < 0) {
1982                 btrfs_end_transaction(trans);
1983                 goto out_reset;
1984         }
1985
1986         ret = btrfs_commit_transaction(trans);
1987
1988 out_reset:
1989         if (ret)
1990                 btrfs_set_root_flags(&root->root_item, root_flags);
1991 out_drop_sem:
1992         up_write(&fs_info->subvol_sem);
1993 out_drop_write:
1994         mnt_drop_write_file(file);
1995 out:
1996         return ret;
1997 }
1998
1999 static noinline int key_in_sk(struct btrfs_key *key,
2000                               struct btrfs_ioctl_search_key *sk)
2001 {
2002         struct btrfs_key test;
2003         int ret;
2004
2005         test.objectid = sk->min_objectid;
2006         test.type = sk->min_type;
2007         test.offset = sk->min_offset;
2008
2009         ret = btrfs_comp_cpu_keys(key, &test);
2010         if (ret < 0)
2011                 return 0;
2012
2013         test.objectid = sk->max_objectid;
2014         test.type = sk->max_type;
2015         test.offset = sk->max_offset;
2016
2017         ret = btrfs_comp_cpu_keys(key, &test);
2018         if (ret > 0)
2019                 return 0;
2020         return 1;
2021 }
2022
2023 static noinline int copy_to_sk(struct btrfs_path *path,
2024                                struct btrfs_key *key,
2025                                struct btrfs_ioctl_search_key *sk,
2026                                size_t *buf_size,
2027                                char __user *ubuf,
2028                                unsigned long *sk_offset,
2029                                int *num_found)
2030 {
2031         u64 found_transid;
2032         struct extent_buffer *leaf;
2033         struct btrfs_ioctl_search_header sh;
2034         struct btrfs_key test;
2035         unsigned long item_off;
2036         unsigned long item_len;
2037         int nritems;
2038         int i;
2039         int slot;
2040         int ret = 0;
2041
2042         leaf = path->nodes[0];
2043         slot = path->slots[0];
2044         nritems = btrfs_header_nritems(leaf);
2045
2046         if (btrfs_header_generation(leaf) > sk->max_transid) {
2047                 i = nritems;
2048                 goto advance_key;
2049         }
2050         found_transid = btrfs_header_generation(leaf);
2051
2052         for (i = slot; i < nritems; i++) {
2053                 item_off = btrfs_item_ptr_offset(leaf, i);
2054                 item_len = btrfs_item_size_nr(leaf, i);
2055
2056                 btrfs_item_key_to_cpu(leaf, key, i);
2057                 if (!key_in_sk(key, sk))
2058                         continue;
2059
2060                 if (sizeof(sh) + item_len > *buf_size) {
2061                         if (*num_found) {
2062                                 ret = 1;
2063                                 goto out;
2064                         }
2065
2066                         /*
2067                          * return one empty item back for v1, which does not
2068                          * handle -EOVERFLOW
2069                          */
2070
2071                         *buf_size = sizeof(sh) + item_len;
2072                         item_len = 0;
2073                         ret = -EOVERFLOW;
2074                 }
2075
2076                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2077                         ret = 1;
2078                         goto out;
2079                 }
2080
2081                 sh.objectid = key->objectid;
2082                 sh.offset = key->offset;
2083                 sh.type = key->type;
2084                 sh.len = item_len;
2085                 sh.transid = found_transid;
2086
2087                 /* copy search result header */
2088                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2089                         ret = -EFAULT;
2090                         goto out;
2091                 }
2092
2093                 *sk_offset += sizeof(sh);
2094
2095                 if (item_len) {
2096                         char __user *up = ubuf + *sk_offset;
2097                         /* copy the item */
2098                         if (read_extent_buffer_to_user(leaf, up,
2099                                                        item_off, item_len)) {
2100                                 ret = -EFAULT;
2101                                 goto out;
2102                         }
2103
2104                         *sk_offset += item_len;
2105                 }
2106                 (*num_found)++;
2107
2108                 if (ret) /* -EOVERFLOW from above */
2109                         goto out;
2110
2111                 if (*num_found >= sk->nr_items) {
2112                         ret = 1;
2113                         goto out;
2114                 }
2115         }
2116 advance_key:
2117         ret = 0;
2118         test.objectid = sk->max_objectid;
2119         test.type = sk->max_type;
2120         test.offset = sk->max_offset;
2121         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2122                 ret = 1;
2123         else if (key->offset < (u64)-1)
2124                 key->offset++;
2125         else if (key->type < (u8)-1) {
2126                 key->offset = 0;
2127                 key->type++;
2128         } else if (key->objectid < (u64)-1) {
2129                 key->offset = 0;
2130                 key->type = 0;
2131                 key->objectid++;
2132         } else
2133                 ret = 1;
2134 out:
2135         /*
2136          *  0: all items from this leaf copied, continue with next
2137          *  1: * more items can be copied, but unused buffer is too small
2138          *     * all items were found
2139          *     Either way, it will stops the loop which iterates to the next
2140          *     leaf
2141          *  -EOVERFLOW: item was to large for buffer
2142          *  -EFAULT: could not copy extent buffer back to userspace
2143          */
2144         return ret;
2145 }
2146
2147 static noinline int search_ioctl(struct inode *inode,
2148                                  struct btrfs_ioctl_search_key *sk,
2149                                  size_t *buf_size,
2150                                  char __user *ubuf)
2151 {
2152         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2153         struct btrfs_root *root;
2154         struct btrfs_key key;
2155         struct btrfs_path *path;
2156         int ret;
2157         int num_found = 0;
2158         unsigned long sk_offset = 0;
2159
2160         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2161                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2162                 return -EOVERFLOW;
2163         }
2164
2165         path = btrfs_alloc_path();
2166         if (!path)
2167                 return -ENOMEM;
2168
2169         if (sk->tree_id == 0) {
2170                 /* search the root of the inode that was passed */
2171                 root = BTRFS_I(inode)->root;
2172         } else {
2173                 key.objectid = sk->tree_id;
2174                 key.type = BTRFS_ROOT_ITEM_KEY;
2175                 key.offset = (u64)-1;
2176                 root = btrfs_read_fs_root_no_name(info, &key);
2177                 if (IS_ERR(root)) {
2178                         btrfs_free_path(path);
2179                         return PTR_ERR(root);
2180                 }
2181         }
2182
2183         key.objectid = sk->min_objectid;
2184         key.type = sk->min_type;
2185         key.offset = sk->min_offset;
2186
2187         while (1) {
2188                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2189                 if (ret != 0) {
2190                         if (ret > 0)
2191                                 ret = 0;
2192                         goto err;
2193                 }
2194                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2195                                  &sk_offset, &num_found);
2196                 btrfs_release_path(path);
2197                 if (ret)
2198                         break;
2199
2200         }
2201         if (ret > 0)
2202                 ret = 0;
2203 err:
2204         sk->nr_items = num_found;
2205         btrfs_free_path(path);
2206         return ret;
2207 }
2208
2209 static noinline int btrfs_ioctl_tree_search(struct file *file,
2210                                            void __user *argp)
2211 {
2212         struct btrfs_ioctl_search_args __user *uargs;
2213         struct btrfs_ioctl_search_key sk;
2214         struct inode *inode;
2215         int ret;
2216         size_t buf_size;
2217
2218         if (!capable(CAP_SYS_ADMIN))
2219                 return -EPERM;
2220
2221         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2222
2223         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2224                 return -EFAULT;
2225
2226         buf_size = sizeof(uargs->buf);
2227
2228         inode = file_inode(file);
2229         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2230
2231         /*
2232          * In the origin implementation an overflow is handled by returning a
2233          * search header with a len of zero, so reset ret.
2234          */
2235         if (ret == -EOVERFLOW)
2236                 ret = 0;
2237
2238         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2239                 ret = -EFAULT;
2240         return ret;
2241 }
2242
2243 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2244                                                void __user *argp)
2245 {
2246         struct btrfs_ioctl_search_args_v2 __user *uarg;
2247         struct btrfs_ioctl_search_args_v2 args;
2248         struct inode *inode;
2249         int ret;
2250         size_t buf_size;
2251         const size_t buf_limit = SZ_16M;
2252
2253         if (!capable(CAP_SYS_ADMIN))
2254                 return -EPERM;
2255
2256         /* copy search header and buffer size */
2257         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2258         if (copy_from_user(&args, uarg, sizeof(args)))
2259                 return -EFAULT;
2260
2261         buf_size = args.buf_size;
2262
2263         /* limit result size to 16MB */
2264         if (buf_size > buf_limit)
2265                 buf_size = buf_limit;
2266
2267         inode = file_inode(file);
2268         ret = search_ioctl(inode, &args.key, &buf_size,
2269                            (char __user *)(&uarg->buf[0]));
2270         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2271                 ret = -EFAULT;
2272         else if (ret == -EOVERFLOW &&
2273                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2274                 ret = -EFAULT;
2275
2276         return ret;
2277 }
2278
2279 /*
2280  * Search INODE_REFs to identify path name of 'dirid' directory
2281  * in a 'tree_id' tree. and sets path name to 'name'.
2282  */
2283 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2284                                 u64 tree_id, u64 dirid, char *name)
2285 {
2286         struct btrfs_root *root;
2287         struct btrfs_key key;
2288         char *ptr;
2289         int ret = -1;
2290         int slot;
2291         int len;
2292         int total_len = 0;
2293         struct btrfs_inode_ref *iref;
2294         struct extent_buffer *l;
2295         struct btrfs_path *path;
2296
2297         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2298                 name[0]='\0';
2299                 return 0;
2300         }
2301
2302         path = btrfs_alloc_path();
2303         if (!path)
2304                 return -ENOMEM;
2305
2306         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2307
2308         key.objectid = tree_id;
2309         key.type = BTRFS_ROOT_ITEM_KEY;
2310         key.offset = (u64)-1;
2311         root = btrfs_read_fs_root_no_name(info, &key);
2312         if (IS_ERR(root)) {
2313                 ret = PTR_ERR(root);
2314                 goto out;
2315         }
2316
2317         key.objectid = dirid;
2318         key.type = BTRFS_INODE_REF_KEY;
2319         key.offset = (u64)-1;
2320
2321         while (1) {
2322                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2323                 if (ret < 0)
2324                         goto out;
2325                 else if (ret > 0) {
2326                         ret = btrfs_previous_item(root, path, dirid,
2327                                                   BTRFS_INODE_REF_KEY);
2328                         if (ret < 0)
2329                                 goto out;
2330                         else if (ret > 0) {
2331                                 ret = -ENOENT;
2332                                 goto out;
2333                         }
2334                 }
2335
2336                 l = path->nodes[0];
2337                 slot = path->slots[0];
2338                 btrfs_item_key_to_cpu(l, &key, slot);
2339
2340                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2341                 len = btrfs_inode_ref_name_len(l, iref);
2342                 ptr -= len + 1;
2343                 total_len += len + 1;
2344                 if (ptr < name) {
2345                         ret = -ENAMETOOLONG;
2346                         goto out;
2347                 }
2348
2349                 *(ptr + len) = '/';
2350                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2351
2352                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2353                         break;
2354
2355                 btrfs_release_path(path);
2356                 key.objectid = key.offset;
2357                 key.offset = (u64)-1;
2358                 dirid = key.objectid;
2359         }
2360         memmove(name, ptr, total_len);
2361         name[total_len] = '\0';
2362         ret = 0;
2363 out:
2364         btrfs_free_path(path);
2365         return ret;
2366 }
2367
2368 static int btrfs_search_path_in_tree_user(struct inode *inode,
2369                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2370 {
2371         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2372         struct super_block *sb = inode->i_sb;
2373         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2374         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2375         u64 dirid = args->dirid;
2376         unsigned long item_off;
2377         unsigned long item_len;
2378         struct btrfs_inode_ref *iref;
2379         struct btrfs_root_ref *rref;
2380         struct btrfs_root *root;
2381         struct btrfs_path *path;
2382         struct btrfs_key key, key2;
2383         struct extent_buffer *leaf;
2384         struct inode *temp_inode;
2385         char *ptr;
2386         int slot;
2387         int len;
2388         int total_len = 0;
2389         int ret;
2390
2391         path = btrfs_alloc_path();
2392         if (!path)
2393                 return -ENOMEM;
2394
2395         /*
2396          * If the bottom subvolume does not exist directly under upper_limit,
2397          * construct the path in from the bottom up.
2398          */
2399         if (dirid != upper_limit.objectid) {
2400                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2401
2402                 key.objectid = treeid;
2403                 key.type = BTRFS_ROOT_ITEM_KEY;
2404                 key.offset = (u64)-1;
2405                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2406                 if (IS_ERR(root)) {
2407                         ret = PTR_ERR(root);
2408                         goto out;
2409                 }
2410
2411                 key.objectid = dirid;
2412                 key.type = BTRFS_INODE_REF_KEY;
2413                 key.offset = (u64)-1;
2414                 while (1) {
2415                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2416                         if (ret < 0) {
2417                                 goto out;
2418                         } else if (ret > 0) {
2419                                 ret = btrfs_previous_item(root, path, dirid,
2420                                                           BTRFS_INODE_REF_KEY);
2421                                 if (ret < 0) {
2422                                         goto out;
2423                                 } else if (ret > 0) {
2424                                         ret = -ENOENT;
2425                                         goto out;
2426                                 }
2427                         }
2428
2429                         leaf = path->nodes[0];
2430                         slot = path->slots[0];
2431                         btrfs_item_key_to_cpu(leaf, &key, slot);
2432
2433                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2434                         len = btrfs_inode_ref_name_len(leaf, iref);
2435                         ptr -= len + 1;
2436                         total_len += len + 1;
2437                         if (ptr < args->path) {
2438                                 ret = -ENAMETOOLONG;
2439                                 goto out;
2440                         }
2441
2442                         *(ptr + len) = '/';
2443                         read_extent_buffer(leaf, ptr,
2444                                         (unsigned long)(iref + 1), len);
2445
2446                         /* Check the read+exec permission of this directory */
2447                         ret = btrfs_previous_item(root, path, dirid,
2448                                                   BTRFS_INODE_ITEM_KEY);
2449                         if (ret < 0) {
2450                                 goto out;
2451                         } else if (ret > 0) {
2452                                 ret = -ENOENT;
2453                                 goto out;
2454                         }
2455
2456                         leaf = path->nodes[0];
2457                         slot = path->slots[0];
2458                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2459                         if (key2.objectid != dirid) {
2460                                 ret = -ENOENT;
2461                                 goto out;
2462                         }
2463
2464                         temp_inode = btrfs_iget(sb, &key2, root);
2465                         if (IS_ERR(temp_inode)) {
2466                                 ret = PTR_ERR(temp_inode);
2467                                 goto out;
2468                         }
2469                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2470                         iput(temp_inode);
2471                         if (ret) {
2472                                 ret = -EACCES;
2473                                 goto out;
2474                         }
2475
2476                         if (key.offset == upper_limit.objectid)
2477                                 break;
2478                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2479                                 ret = -EACCES;
2480                                 goto out;
2481                         }
2482
2483                         btrfs_release_path(path);
2484                         key.objectid = key.offset;
2485                         key.offset = (u64)-1;
2486                         dirid = key.objectid;
2487                 }
2488
2489                 memmove(args->path, ptr, total_len);
2490                 args->path[total_len] = '\0';
2491                 btrfs_release_path(path);
2492         }
2493
2494         /* Get the bottom subvolume's name from ROOT_REF */
2495         root = fs_info->tree_root;
2496         key.objectid = treeid;
2497         key.type = BTRFS_ROOT_REF_KEY;
2498         key.offset = args->treeid;
2499         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2500         if (ret < 0) {
2501                 goto out;
2502         } else if (ret > 0) {
2503                 ret = -ENOENT;
2504                 goto out;
2505         }
2506
2507         leaf = path->nodes[0];
2508         slot = path->slots[0];
2509         btrfs_item_key_to_cpu(leaf, &key, slot);
2510
2511         item_off = btrfs_item_ptr_offset(leaf, slot);
2512         item_len = btrfs_item_size_nr(leaf, slot);
2513         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2514         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2515         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2516                 ret = -EINVAL;
2517                 goto out;
2518         }
2519
2520         /* Copy subvolume's name */
2521         item_off += sizeof(struct btrfs_root_ref);
2522         item_len -= sizeof(struct btrfs_root_ref);
2523         read_extent_buffer(leaf, args->name, item_off, item_len);
2524         args->name[item_len] = 0;
2525
2526 out:
2527         btrfs_free_path(path);
2528         return ret;
2529 }
2530
2531 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2532                                            void __user *argp)
2533 {
2534         struct btrfs_ioctl_ino_lookup_args *args;
2535         struct inode *inode;
2536         int ret = 0;
2537
2538         args = memdup_user(argp, sizeof(*args));
2539         if (IS_ERR(args))
2540                 return PTR_ERR(args);
2541
2542         inode = file_inode(file);
2543
2544         /*
2545          * Unprivileged query to obtain the containing subvolume root id. The
2546          * path is reset so it's consistent with btrfs_search_path_in_tree.
2547          */
2548         if (args->treeid == 0)
2549                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2550
2551         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2552                 args->name[0] = 0;
2553                 goto out;
2554         }
2555
2556         if (!capable(CAP_SYS_ADMIN)) {
2557                 ret = -EPERM;
2558                 goto out;
2559         }
2560
2561         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2562                                         args->treeid, args->objectid,
2563                                         args->name);
2564
2565 out:
2566         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2567                 ret = -EFAULT;
2568
2569         kfree(args);
2570         return ret;
2571 }
2572
2573 /*
2574  * Version of ino_lookup ioctl (unprivileged)
2575  *
2576  * The main differences from ino_lookup ioctl are:
2577  *
2578  *   1. Read + Exec permission will be checked using inode_permission() during
2579  *      path construction. -EACCES will be returned in case of failure.
2580  *   2. Path construction will be stopped at the inode number which corresponds
2581  *      to the fd with which this ioctl is called. If constructed path does not
2582  *      exist under fd's inode, -EACCES will be returned.
2583  *   3. The name of bottom subvolume is also searched and filled.
2584  */
2585 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2586 {
2587         struct btrfs_ioctl_ino_lookup_user_args *args;
2588         struct inode *inode;
2589         int ret;
2590
2591         args = memdup_user(argp, sizeof(*args));
2592         if (IS_ERR(args))
2593                 return PTR_ERR(args);
2594
2595         inode = file_inode(file);
2596
2597         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2598             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2599                 /*
2600                  * The subvolume does not exist under fd with which this is
2601                  * called
2602                  */
2603                 kfree(args);
2604                 return -EACCES;
2605         }
2606
2607         ret = btrfs_search_path_in_tree_user(inode, args);
2608
2609         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2610                 ret = -EFAULT;
2611
2612         kfree(args);
2613         return ret;
2614 }
2615
2616 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2617 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2618 {
2619         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2620         struct btrfs_fs_info *fs_info;
2621         struct btrfs_root *root;
2622         struct btrfs_path *path;
2623         struct btrfs_key key;
2624         struct btrfs_root_item *root_item;
2625         struct btrfs_root_ref *rref;
2626         struct extent_buffer *leaf;
2627         unsigned long item_off;
2628         unsigned long item_len;
2629         struct inode *inode;
2630         int slot;
2631         int ret = 0;
2632
2633         path = btrfs_alloc_path();
2634         if (!path)
2635                 return -ENOMEM;
2636
2637         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2638         if (!subvol_info) {
2639                 btrfs_free_path(path);
2640                 return -ENOMEM;
2641         }
2642
2643         inode = file_inode(file);
2644         fs_info = BTRFS_I(inode)->root->fs_info;
2645
2646         /* Get root_item of inode's subvolume */
2647         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2648         key.type = BTRFS_ROOT_ITEM_KEY;
2649         key.offset = (u64)-1;
2650         root = btrfs_read_fs_root_no_name(fs_info, &key);
2651         if (IS_ERR(root)) {
2652                 ret = PTR_ERR(root);
2653                 goto out;
2654         }
2655         root_item = &root->root_item;
2656
2657         subvol_info->treeid = key.objectid;
2658
2659         subvol_info->generation = btrfs_root_generation(root_item);
2660         subvol_info->flags = btrfs_root_flags(root_item);
2661
2662         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2663         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2664                                                     BTRFS_UUID_SIZE);
2665         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2666                                                     BTRFS_UUID_SIZE);
2667
2668         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2669         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2670         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2671
2672         subvol_info->otransid = btrfs_root_otransid(root_item);
2673         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2674         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2675
2676         subvol_info->stransid = btrfs_root_stransid(root_item);
2677         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2678         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2679
2680         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2681         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2682         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2683
2684         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2685                 /* Search root tree for ROOT_BACKREF of this subvolume */
2686                 root = fs_info->tree_root;
2687
2688                 key.type = BTRFS_ROOT_BACKREF_KEY;
2689                 key.offset = 0;
2690                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2691                 if (ret < 0) {
2692                         goto out;
2693                 } else if (path->slots[0] >=
2694                            btrfs_header_nritems(path->nodes[0])) {
2695                         ret = btrfs_next_leaf(root, path);
2696                         if (ret < 0) {
2697                                 goto out;
2698                         } else if (ret > 0) {
2699                                 ret = -EUCLEAN;
2700                                 goto out;
2701                         }
2702                 }
2703
2704                 leaf = path->nodes[0];
2705                 slot = path->slots[0];
2706                 btrfs_item_key_to_cpu(leaf, &key, slot);
2707                 if (key.objectid == subvol_info->treeid &&
2708                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2709                         subvol_info->parent_id = key.offset;
2710
2711                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2712                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2713
2714                         item_off = btrfs_item_ptr_offset(leaf, slot)
2715                                         + sizeof(struct btrfs_root_ref);
2716                         item_len = btrfs_item_size_nr(leaf, slot)
2717                                         - sizeof(struct btrfs_root_ref);
2718                         read_extent_buffer(leaf, subvol_info->name,
2719                                            item_off, item_len);
2720                 } else {
2721                         ret = -ENOENT;
2722                         goto out;
2723                 }
2724         }
2725
2726         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2727                 ret = -EFAULT;
2728
2729 out:
2730         btrfs_free_path(path);
2731         kzfree(subvol_info);
2732         return ret;
2733 }
2734
2735 /*
2736  * Return ROOT_REF information of the subvolume containing this inode
2737  * except the subvolume name.
2738  */
2739 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2740 {
2741         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2742         struct btrfs_root_ref *rref;
2743         struct btrfs_root *root;
2744         struct btrfs_path *path;
2745         struct btrfs_key key;
2746         struct extent_buffer *leaf;
2747         struct inode *inode;
2748         u64 objectid;
2749         int slot;
2750         int ret;
2751         u8 found;
2752
2753         path = btrfs_alloc_path();
2754         if (!path)
2755                 return -ENOMEM;
2756
2757         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2758         if (IS_ERR(rootrefs)) {
2759                 btrfs_free_path(path);
2760                 return PTR_ERR(rootrefs);
2761         }
2762
2763         inode = file_inode(file);
2764         root = BTRFS_I(inode)->root->fs_info->tree_root;
2765         objectid = BTRFS_I(inode)->root->root_key.objectid;
2766
2767         key.objectid = objectid;
2768         key.type = BTRFS_ROOT_REF_KEY;
2769         key.offset = rootrefs->min_treeid;
2770         found = 0;
2771
2772         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2773         if (ret < 0) {
2774                 goto out;
2775         } else if (path->slots[0] >=
2776                    btrfs_header_nritems(path->nodes[0])) {
2777                 ret = btrfs_next_leaf(root, path);
2778                 if (ret < 0) {
2779                         goto out;
2780                 } else if (ret > 0) {
2781                         ret = -EUCLEAN;
2782                         goto out;
2783                 }
2784         }
2785         while (1) {
2786                 leaf = path->nodes[0];
2787                 slot = path->slots[0];
2788
2789                 btrfs_item_key_to_cpu(leaf, &key, slot);
2790                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2791                         ret = 0;
2792                         goto out;
2793                 }
2794
2795                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2796                         ret = -EOVERFLOW;
2797                         goto out;
2798                 }
2799
2800                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2801                 rootrefs->rootref[found].treeid = key.offset;
2802                 rootrefs->rootref[found].dirid =
2803                                   btrfs_root_ref_dirid(leaf, rref);
2804                 found++;
2805
2806                 ret = btrfs_next_item(root, path);
2807                 if (ret < 0) {
2808                         goto out;
2809                 } else if (ret > 0) {
2810                         ret = -EUCLEAN;
2811                         goto out;
2812                 }
2813         }
2814
2815 out:
2816         if (!ret || ret == -EOVERFLOW) {
2817                 rootrefs->num_items = found;
2818                 /* update min_treeid for next search */
2819                 if (found)
2820                         rootrefs->min_treeid =
2821                                 rootrefs->rootref[found - 1].treeid + 1;
2822                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2823                         ret = -EFAULT;
2824         }
2825
2826         kfree(rootrefs);
2827         btrfs_free_path(path);
2828
2829         return ret;
2830 }
2831
2832 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2833                                              void __user *arg)
2834 {
2835         struct dentry *parent = file->f_path.dentry;
2836         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2837         struct dentry *dentry;
2838         struct inode *dir = d_inode(parent);
2839         struct inode *inode;
2840         struct btrfs_root *root = BTRFS_I(dir)->root;
2841         struct btrfs_root *dest = NULL;
2842         struct btrfs_ioctl_vol_args *vol_args;
2843         int namelen;
2844         int err = 0;
2845
2846         if (!S_ISDIR(dir->i_mode))
2847                 return -ENOTDIR;
2848
2849         vol_args = memdup_user(arg, sizeof(*vol_args));
2850         if (IS_ERR(vol_args))
2851                 return PTR_ERR(vol_args);
2852
2853         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2854         namelen = strlen(vol_args->name);
2855         if (strchr(vol_args->name, '/') ||
2856             strncmp(vol_args->name, "..", namelen) == 0) {
2857                 err = -EINVAL;
2858                 goto out;
2859         }
2860
2861         err = mnt_want_write_file(file);
2862         if (err)
2863                 goto out;
2864
2865
2866         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2867         if (err == -EINTR)
2868                 goto out_drop_write;
2869         dentry = lookup_one_len(vol_args->name, parent, namelen);
2870         if (IS_ERR(dentry)) {
2871                 err = PTR_ERR(dentry);
2872                 goto out_unlock_dir;
2873         }
2874
2875         if (d_really_is_negative(dentry)) {
2876                 err = -ENOENT;
2877                 goto out_dput;
2878         }
2879
2880         inode = d_inode(dentry);
2881         dest = BTRFS_I(inode)->root;
2882         if (!capable(CAP_SYS_ADMIN)) {
2883                 /*
2884                  * Regular user.  Only allow this with a special mount
2885                  * option, when the user has write+exec access to the
2886                  * subvol root, and when rmdir(2) would have been
2887                  * allowed.
2888                  *
2889                  * Note that this is _not_ check that the subvol is
2890                  * empty or doesn't contain data that we wouldn't
2891                  * otherwise be able to delete.
2892                  *
2893                  * Users who want to delete empty subvols should try
2894                  * rmdir(2).
2895                  */
2896                 err = -EPERM;
2897                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2898                         goto out_dput;
2899
2900                 /*
2901                  * Do not allow deletion if the parent dir is the same
2902                  * as the dir to be deleted.  That means the ioctl
2903                  * must be called on the dentry referencing the root
2904                  * of the subvol, not a random directory contained
2905                  * within it.
2906                  */
2907                 err = -EINVAL;
2908                 if (root == dest)
2909                         goto out_dput;
2910
2911                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2912                 if (err)
2913                         goto out_dput;
2914         }
2915
2916         /* check if subvolume may be deleted by a user */
2917         err = btrfs_may_delete(dir, dentry, 1);
2918         if (err)
2919                 goto out_dput;
2920
2921         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2922                 err = -EINVAL;
2923                 goto out_dput;
2924         }
2925
2926         inode_lock(inode);
2927         err = btrfs_delete_subvolume(dir, dentry);
2928         inode_unlock(inode);
2929         if (!err) {
2930                 fsnotify_rmdir(dir, dentry);
2931                 d_delete(dentry);
2932         }
2933
2934 out_dput:
2935         dput(dentry);
2936 out_unlock_dir:
2937         inode_unlock(dir);
2938 out_drop_write:
2939         mnt_drop_write_file(file);
2940 out:
2941         kfree(vol_args);
2942         return err;
2943 }
2944
2945 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2946 {
2947         struct inode *inode = file_inode(file);
2948         struct btrfs_root *root = BTRFS_I(inode)->root;
2949         struct btrfs_ioctl_defrag_range_args *range;
2950         int ret;
2951
2952         ret = mnt_want_write_file(file);
2953         if (ret)
2954                 return ret;
2955
2956         if (btrfs_root_readonly(root)) {
2957                 ret = -EROFS;
2958                 goto out;
2959         }
2960
2961         switch (inode->i_mode & S_IFMT) {
2962         case S_IFDIR:
2963                 if (!capable(CAP_SYS_ADMIN)) {
2964                         ret = -EPERM;
2965                         goto out;
2966                 }
2967                 ret = btrfs_defrag_root(root);
2968                 break;
2969         case S_IFREG:
2970                 /*
2971                  * Note that this does not check the file descriptor for write
2972                  * access. This prevents defragmenting executables that are
2973                  * running and allows defrag on files open in read-only mode.
2974                  */
2975                 if (!capable(CAP_SYS_ADMIN) &&
2976                     inode_permission(inode, MAY_WRITE)) {
2977                         ret = -EPERM;
2978                         goto out;
2979                 }
2980
2981                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2982                 if (!range) {
2983                         ret = -ENOMEM;
2984                         goto out;
2985                 }
2986
2987                 if (argp) {
2988                         if (copy_from_user(range, argp,
2989                                            sizeof(*range))) {
2990                                 ret = -EFAULT;
2991                                 kfree(range);
2992                                 goto out;
2993                         }
2994                         /* compression requires us to start the IO */
2995                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2996                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2997                                 range->extent_thresh = (u32)-1;
2998                         }
2999                 } else {
3000                         /* the rest are all set to zero by kzalloc */
3001                         range->len = (u64)-1;
3002                 }
3003                 ret = btrfs_defrag_file(file_inode(file), file,
3004                                         range, BTRFS_OLDEST_GENERATION, 0);
3005                 if (ret > 0)
3006                         ret = 0;
3007                 kfree(range);
3008                 break;
3009         default:
3010                 ret = -EINVAL;
3011         }
3012 out:
3013         mnt_drop_write_file(file);
3014         return ret;
3015 }
3016
3017 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3018 {
3019         struct btrfs_ioctl_vol_args *vol_args;
3020         int ret;
3021
3022         if (!capable(CAP_SYS_ADMIN))
3023                 return -EPERM;
3024
3025         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3026                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3027
3028         vol_args = memdup_user(arg, sizeof(*vol_args));
3029         if (IS_ERR(vol_args)) {
3030                 ret = PTR_ERR(vol_args);
3031                 goto out;
3032         }
3033
3034         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3035         ret = btrfs_init_new_device(fs_info, vol_args->name);
3036
3037         if (!ret)
3038                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3039
3040         kfree(vol_args);
3041 out:
3042         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3043         return ret;
3044 }
3045
3046 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3047 {
3048         struct inode *inode = file_inode(file);
3049         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3050         struct btrfs_ioctl_vol_args_v2 *vol_args;
3051         int ret;
3052
3053         if (!capable(CAP_SYS_ADMIN))
3054                 return -EPERM;
3055
3056         ret = mnt_want_write_file(file);
3057         if (ret)
3058                 return ret;
3059
3060         vol_args = memdup_user(arg, sizeof(*vol_args));
3061         if (IS_ERR(vol_args)) {
3062                 ret = PTR_ERR(vol_args);
3063                 goto err_drop;
3064         }
3065
3066         /* Check for compatibility reject unknown flags */
3067         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3068                 ret = -EOPNOTSUPP;
3069                 goto out;
3070         }
3071
3072         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3073                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3074                 goto out;
3075         }
3076
3077         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3078                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3079         } else {
3080                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3081                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3082         }
3083         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3084
3085         if (!ret) {
3086                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3087                         btrfs_info(fs_info, "device deleted: id %llu",
3088                                         vol_args->devid);
3089                 else
3090                         btrfs_info(fs_info, "device deleted: %s",
3091                                         vol_args->name);
3092         }
3093 out:
3094         kfree(vol_args);
3095 err_drop:
3096         mnt_drop_write_file(file);
3097         return ret;
3098 }
3099
3100 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3101 {
3102         struct inode *inode = file_inode(file);
3103         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3104         struct btrfs_ioctl_vol_args *vol_args;
3105         int ret;
3106
3107         if (!capable(CAP_SYS_ADMIN))
3108                 return -EPERM;
3109
3110         ret = mnt_want_write_file(file);
3111         if (ret)
3112                 return ret;
3113
3114         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3115                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3116                 goto out_drop_write;
3117         }
3118
3119         vol_args = memdup_user(arg, sizeof(*vol_args));
3120         if (IS_ERR(vol_args)) {
3121                 ret = PTR_ERR(vol_args);
3122                 goto out;
3123         }
3124
3125         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3126         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3127
3128         if (!ret)
3129                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3130         kfree(vol_args);
3131 out:
3132         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3133 out_drop_write:
3134         mnt_drop_write_file(file);
3135
3136         return ret;
3137 }
3138
3139 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3140                                 void __user *arg)
3141 {
3142         struct btrfs_ioctl_fs_info_args *fi_args;
3143         struct btrfs_device *device;
3144         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3145         int ret = 0;
3146
3147         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3148         if (!fi_args)
3149                 return -ENOMEM;
3150
3151         rcu_read_lock();
3152         fi_args->num_devices = fs_devices->num_devices;
3153
3154         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3155                 if (device->devid > fi_args->max_id)
3156                         fi_args->max_id = device->devid;
3157         }
3158         rcu_read_unlock();
3159
3160         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3161         fi_args->nodesize = fs_info->nodesize;
3162         fi_args->sectorsize = fs_info->sectorsize;
3163         fi_args->clone_alignment = fs_info->sectorsize;
3164
3165         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3166                 ret = -EFAULT;
3167
3168         kfree(fi_args);
3169         return ret;
3170 }
3171
3172 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3173                                  void __user *arg)
3174 {
3175         struct btrfs_ioctl_dev_info_args *di_args;
3176         struct btrfs_device *dev;
3177         int ret = 0;
3178         char *s_uuid = NULL;
3179
3180         di_args = memdup_user(arg, sizeof(*di_args));
3181         if (IS_ERR(di_args))
3182                 return PTR_ERR(di_args);
3183
3184         if (!btrfs_is_empty_uuid(di_args->uuid))
3185                 s_uuid = di_args->uuid;
3186
3187         rcu_read_lock();
3188         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3189                                 NULL, true);
3190
3191         if (!dev) {
3192                 ret = -ENODEV;
3193                 goto out;
3194         }
3195
3196         di_args->devid = dev->devid;
3197         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3198         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3199         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3200         if (dev->name) {
3201                 strncpy(di_args->path, rcu_str_deref(dev->name),
3202                                 sizeof(di_args->path) - 1);
3203                 di_args->path[sizeof(di_args->path) - 1] = 0;
3204         } else {
3205                 di_args->path[0] = '\0';
3206         }
3207
3208 out:
3209         rcu_read_unlock();
3210         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3211                 ret = -EFAULT;
3212
3213         kfree(di_args);
3214         return ret;
3215 }
3216
3217 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3218                                        struct inode *inode2, u64 loff2, u64 len)
3219 {
3220         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3221         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3222 }
3223
3224 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3225                                      struct inode *inode2, u64 loff2, u64 len)
3226 {
3227         if (inode1 < inode2) {
3228                 swap(inode1, inode2);
3229                 swap(loff1, loff2);
3230         } else if (inode1 == inode2 && loff2 < loff1) {
3231                 swap(loff1, loff2);
3232         }
3233         lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3234         lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3235 }
3236
3237 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
3238                                    struct inode *dst, u64 dst_loff)
3239 {
3240         int ret;
3241
3242         /*
3243          * Lock destination range to serialize with concurrent readpages() and
3244          * source range to serialize with relocation.
3245          */
3246         btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3247         ret = btrfs_clone(src, dst, loff, len, len, dst_loff, 1);
3248         btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3249
3250         return ret;
3251 }
3252
3253 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3254
3255 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3256                              struct inode *dst, u64 dst_loff)
3257 {
3258         int ret;
3259         u64 i, tail_len, chunk_count;
3260         struct btrfs_root *root_dst = BTRFS_I(dst)->root;
3261
3262         spin_lock(&root_dst->root_item_lock);
3263         if (root_dst->send_in_progress) {
3264                 btrfs_warn_rl(root_dst->fs_info,
3265 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
3266                               root_dst->root_key.objectid,
3267                               root_dst->send_in_progress);
3268                 spin_unlock(&root_dst->root_item_lock);
3269                 return -EAGAIN;
3270         }
3271         root_dst->dedupe_in_progress++;
3272         spin_unlock(&root_dst->root_item_lock);
3273
3274         tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3275         chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3276
3277         for (i = 0; i < chunk_count; i++) {
3278                 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3279                                               dst, dst_loff);
3280                 if (ret)
3281                         goto out;
3282
3283                 loff += BTRFS_MAX_DEDUPE_LEN;
3284                 dst_loff += BTRFS_MAX_DEDUPE_LEN;
3285         }
3286
3287         if (tail_len > 0)
3288                 ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3289                                               dst_loff);
3290 out:
3291         spin_lock(&root_dst->root_item_lock);
3292         root_dst->dedupe_in_progress--;
3293         spin_unlock(&root_dst->root_item_lock);
3294
3295         return ret;
3296 }
3297
3298 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3299                                      struct inode *inode,
3300                                      u64 endoff,
3301                                      const u64 destoff,
3302                                      const u64 olen,
3303                                      int no_time_update)
3304 {
3305         struct btrfs_root *root = BTRFS_I(inode)->root;
3306         int ret;
3307
3308         inode_inc_iversion(inode);
3309         if (!no_time_update)
3310                 inode->i_mtime = inode->i_ctime = current_time(inode);
3311         /*
3312          * We round up to the block size at eof when determining which
3313          * extents to clone above, but shouldn't round up the file size.
3314          */
3315         if (endoff > destoff + olen)
3316                 endoff = destoff + olen;
3317         if (endoff > inode->i_size)
3318                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3319
3320         ret = btrfs_update_inode(trans, root, inode);
3321         if (ret) {
3322                 btrfs_abort_transaction(trans, ret);
3323                 btrfs_end_transaction(trans);
3324                 goto out;
3325         }
3326         ret = btrfs_end_transaction(trans);
3327 out:
3328         return ret;
3329 }
3330
3331 /*
3332  * Make sure we do not end up inserting an inline extent into a file that has
3333  * already other (non-inline) extents. If a file has an inline extent it can
3334  * not have any other extents and the (single) inline extent must start at the
3335  * file offset 0. Failing to respect these rules will lead to file corruption,
3336  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3337  *
3338  * We can have extents that have been already written to disk or we can have
3339  * dirty ranges still in delalloc, in which case the extent maps and items are
3340  * created only when we run delalloc, and the delalloc ranges might fall outside
3341  * the range we are currently locking in the inode's io tree. So we check the
3342  * inode's i_size because of that (i_size updates are done while holding the
3343  * i_mutex, which we are holding here).
3344  * We also check to see if the inode has a size not greater than "datal" but has
3345  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3346  * protected against such concurrent fallocate calls by the i_mutex).
3347  *
3348  * If the file has no extents but a size greater than datal, do not allow the
3349  * copy because we would need turn the inline extent into a non-inline one (even
3350  * with NO_HOLES enabled). If we find our destination inode only has one inline
3351  * extent, just overwrite it with the source inline extent if its size is less
3352  * than the source extent's size, or we could copy the source inline extent's
3353  * data into the destination inode's inline extent if the later is greater then
3354  * the former.
3355  */
3356 static int clone_copy_inline_extent(struct inode *dst,
3357                                     struct btrfs_trans_handle *trans,
3358                                     struct btrfs_path *path,
3359                                     struct btrfs_key *new_key,
3360                                     const u64 drop_start,
3361                                     const u64 datal,
3362                                     const u64 skip,
3363                                     const u64 size,
3364                                     char *inline_data)
3365 {
3366         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3367         struct btrfs_root *root = BTRFS_I(dst)->root;
3368         const u64 aligned_end = ALIGN(new_key->offset + datal,
3369                                       fs_info->sectorsize);
3370         int ret;
3371         struct btrfs_key key;
3372
3373         if (new_key->offset > 0)
3374                 return -EOPNOTSUPP;
3375
3376         key.objectid = btrfs_ino(BTRFS_I(dst));
3377         key.type = BTRFS_EXTENT_DATA_KEY;
3378         key.offset = 0;
3379         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3380         if (ret < 0) {
3381                 return ret;
3382         } else if (ret > 0) {
3383                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3384                         ret = btrfs_next_leaf(root, path);
3385                         if (ret < 0)
3386                                 return ret;
3387                         else if (ret > 0)
3388                                 goto copy_inline_extent;
3389                 }
3390                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3391                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3392                     key.type == BTRFS_EXTENT_DATA_KEY) {
3393                         ASSERT(key.offset > 0);
3394                         return -EOPNOTSUPP;
3395                 }
3396         } else if (i_size_read(dst) <= datal) {
3397                 struct btrfs_file_extent_item *ei;
3398                 u64 ext_len;
3399
3400                 /*
3401                  * If the file size is <= datal, make sure there are no other
3402                  * extents following (can happen do to an fallocate call with
3403                  * the flag FALLOC_FL_KEEP_SIZE).
3404                  */
3405                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3406                                     struct btrfs_file_extent_item);
3407                 /*
3408                  * If it's an inline extent, it can not have other extents
3409                  * following it.
3410                  */
3411                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3412                     BTRFS_FILE_EXTENT_INLINE)
3413                         goto copy_inline_extent;
3414
3415                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3416                 if (ext_len > aligned_end)
3417                         return -EOPNOTSUPP;
3418
3419                 ret = btrfs_next_item(root, path);
3420                 if (ret < 0) {
3421                         return ret;
3422                 } else if (ret == 0) {
3423                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3424                                               path->slots[0]);
3425                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3426                             key.type == BTRFS_EXTENT_DATA_KEY)
3427                                 return -EOPNOTSUPP;
3428                 }
3429         }
3430
3431 copy_inline_extent:
3432         /*
3433          * We have no extent items, or we have an extent at offset 0 which may
3434          * or may not be inlined. All these cases are dealt the same way.
3435          */
3436         if (i_size_read(dst) > datal) {
3437                 /*
3438                  * If the destination inode has an inline extent...
3439                  * This would require copying the data from the source inline
3440                  * extent into the beginning of the destination's inline extent.
3441                  * But this is really complex, both extents can be compressed
3442                  * or just one of them, which would require decompressing and
3443                  * re-compressing data (which could increase the new compressed
3444                  * size, not allowing the compressed data to fit anymore in an
3445                  * inline extent).
3446                  * So just don't support this case for now (it should be rare,
3447                  * we are not really saving space when cloning inline extents).
3448                  */
3449                 return -EOPNOTSUPP;
3450         }
3451
3452         btrfs_release_path(path);
3453         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3454         if (ret)
3455                 return ret;
3456         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3457         if (ret)
3458                 return ret;
3459
3460         if (skip) {
3461                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3462
3463                 memmove(inline_data + start, inline_data + start + skip, datal);
3464         }
3465
3466         write_extent_buffer(path->nodes[0], inline_data,
3467                             btrfs_item_ptr_offset(path->nodes[0],
3468                                                   path->slots[0]),
3469                             size);
3470         inode_add_bytes(dst, datal);
3471         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags);
3472
3473         return 0;
3474 }
3475
3476 /**
3477  * btrfs_clone() - clone a range from inode file to another
3478  *
3479  * @src: Inode to clone from
3480  * @inode: Inode to clone to
3481  * @off: Offset within source to start clone from
3482  * @olen: Original length, passed by user, of range to clone
3483  * @olen_aligned: Block-aligned value of olen
3484  * @destoff: Offset within @inode to start clone
3485  * @no_time_update: Whether to update mtime/ctime on the target inode
3486  */
3487 static int btrfs_clone(struct inode *src, struct inode *inode,
3488                        const u64 off, const u64 olen, const u64 olen_aligned,
3489                        const u64 destoff, int no_time_update)
3490 {
3491         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3492         struct btrfs_root *root = BTRFS_I(inode)->root;
3493         struct btrfs_path *path = NULL;
3494         struct extent_buffer *leaf;
3495         struct btrfs_trans_handle *trans;
3496         char *buf = NULL;
3497         struct btrfs_key key;
3498         u32 nritems;
3499         int slot;
3500         int ret;
3501         const u64 len = olen_aligned;
3502         u64 last_dest_end = destoff;
3503
3504         ret = -ENOMEM;
3505         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3506         if (!buf)
3507                 return ret;
3508
3509         path = btrfs_alloc_path();
3510         if (!path) {
3511                 kvfree(buf);
3512                 return ret;
3513         }
3514
3515         path->reada = READA_FORWARD;
3516         /* clone data */
3517         key.objectid = btrfs_ino(BTRFS_I(src));
3518         key.type = BTRFS_EXTENT_DATA_KEY;
3519         key.offset = off;
3520
3521         while (1) {
3522                 u64 next_key_min_offset = key.offset + 1;
3523                 struct btrfs_file_extent_item *extent;
3524                 int type;
3525                 u32 size;
3526                 struct btrfs_key new_key;
3527                 u64 disko = 0, diskl = 0;
3528                 u64 datao = 0, datal = 0;
3529                 u8 comp;
3530                 u64 drop_start;
3531
3532                 /*
3533                  * note the key will change type as we walk through the
3534                  * tree.
3535                  */
3536                 path->leave_spinning = 1;
3537                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3538                                 0, 0);
3539                 if (ret < 0)
3540                         goto out;
3541                 /*
3542                  * First search, if no extent item that starts at offset off was
3543                  * found but the previous item is an extent item, it's possible
3544                  * it might overlap our target range, therefore process it.
3545                  */
3546                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3547                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3548                                               path->slots[0] - 1);
3549                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3550                                 path->slots[0]--;
3551                 }
3552
3553                 nritems = btrfs_header_nritems(path->nodes[0]);
3554 process_slot:
3555                 if (path->slots[0] >= nritems) {
3556                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3557                         if (ret < 0)
3558                                 goto out;
3559                         if (ret > 0)
3560                                 break;
3561                         nritems = btrfs_header_nritems(path->nodes[0]);
3562                 }
3563                 leaf = path->nodes[0];
3564                 slot = path->slots[0];
3565
3566                 btrfs_item_key_to_cpu(leaf, &key, slot);
3567                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3568                     key.objectid != btrfs_ino(BTRFS_I(src)))
3569                         break;
3570
3571                 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
3572
3573                 extent = btrfs_item_ptr(leaf, slot,
3574                                         struct btrfs_file_extent_item);
3575                 comp = btrfs_file_extent_compression(leaf, extent);
3576                 type = btrfs_file_extent_type(leaf, extent);
3577                 if (type == BTRFS_FILE_EXTENT_REG ||
3578                     type == BTRFS_FILE_EXTENT_PREALLOC) {
3579                         disko = btrfs_file_extent_disk_bytenr(leaf, extent);
3580                         diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
3581                         datao = btrfs_file_extent_offset(leaf, extent);
3582                         datal = btrfs_file_extent_num_bytes(leaf, extent);
3583                 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3584                         /* Take upper bound, may be compressed */
3585                         datal = btrfs_file_extent_ram_bytes(leaf, extent);
3586                 }
3587
3588                 /*
3589                  * The first search might have left us at an extent item that
3590                  * ends before our target range's start, can happen if we have
3591                  * holes and NO_HOLES feature enabled.
3592                  */
3593                 if (key.offset + datal <= off) {
3594                         path->slots[0]++;
3595                         goto process_slot;
3596                 } else if (key.offset >= off + len) {
3597                         break;
3598                 }
3599                 next_key_min_offset = key.offset + datal;
3600                 size = btrfs_item_size_nr(leaf, slot);
3601                 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
3602                                    size);
3603
3604                 btrfs_release_path(path);
3605                 path->leave_spinning = 0;
3606
3607                 memcpy(&new_key, &key, sizeof(new_key));
3608                 new_key.objectid = btrfs_ino(BTRFS_I(inode));
3609                 if (off <= key.offset)
3610                         new_key.offset = key.offset + destoff - off;
3611                 else
3612                         new_key.offset = destoff;
3613
3614                 /*
3615                  * Deal with a hole that doesn't have an extent item that
3616                  * represents it (NO_HOLES feature enabled).
3617                  * This hole is either in the middle of the cloning range or at
3618                  * the beginning (fully overlaps it or partially overlaps it).
3619                  */
3620                 if (new_key.offset != last_dest_end)
3621                         drop_start = last_dest_end;
3622                 else
3623                         drop_start = new_key.offset;
3624
3625                 if (type == BTRFS_FILE_EXTENT_REG ||
3626                     type == BTRFS_FILE_EXTENT_PREALLOC) {
3627                         struct btrfs_clone_extent_info clone_info;
3628
3629                         /*
3630                          *    a  | --- range to clone ---|  b
3631                          * | ------------- extent ------------- |
3632                          */
3633
3634                         /* Subtract range b */
3635                         if (key.offset + datal > off + len)
3636                                 datal = off + len - key.offset;
3637
3638                         /* Subtract range a */
3639                         if (off > key.offset) {
3640                                 datao += off - key.offset;
3641                                 datal -= off - key.offset;
3642                         }
3643
3644                         clone_info.disk_offset = disko;
3645                         clone_info.disk_len = diskl;
3646                         clone_info.data_offset = datao;
3647                         clone_info.data_len = datal;
3648                         clone_info.file_offset = new_key.offset;
3649                         clone_info.extent_buf = buf;
3650                         clone_info.item_size = size;
3651                         ret = btrfs_punch_hole_range(inode, path,
3652                                                      drop_start,
3653                                                      new_key.offset + datal - 1,
3654                                                      &clone_info, &trans);
3655                         if (ret)
3656                                 goto out;
3657                 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3658                         u64 skip = 0;
3659                         u64 trim = 0;
3660
3661                         if (off > key.offset) {
3662                                 skip = off - key.offset;
3663                                 new_key.offset += skip;
3664                         }
3665
3666                         if (key.offset + datal > off + len)
3667                                 trim = key.offset + datal - (off + len);
3668
3669                         if (comp && (skip || trim)) {
3670                                 ret = -EINVAL;
3671                                 goto out;
3672                         }
3673                         size -= skip + trim;
3674                         datal -= skip + trim;
3675
3676                         /*
3677                          * If our extent is inline, we know we will drop or
3678                          * adjust at most 1 extent item in the destination root.
3679                          *
3680                          * 1 - adjusting old extent (we may have to split it)
3681                          * 1 - add new extent
3682                          * 1 - inode update
3683                          */
3684                         trans = btrfs_start_transaction(root, 3);
3685                         if (IS_ERR(trans)) {
3686                                 ret = PTR_ERR(trans);
3687                                 goto out;
3688                         }
3689
3690                         ret = clone_copy_inline_extent(inode, trans, path,
3691                                                        &new_key, drop_start,
3692                                                        datal, skip, size, buf);
3693                         if (ret) {
3694                                 if (ret != -EOPNOTSUPP)
3695                                         btrfs_abort_transaction(trans, ret);
3696                                 btrfs_end_transaction(trans);
3697                                 goto out;
3698                         }
3699                 }
3700
3701                 btrfs_release_path(path);
3702
3703                 last_dest_end = ALIGN(new_key.offset + datal,
3704                                       fs_info->sectorsize);
3705                 ret = clone_finish_inode_update(trans, inode, last_dest_end,
3706                                                 destoff, olen, no_time_update);
3707                 if (ret)
3708                         goto out;
3709                 if (new_key.offset + datal >= destoff + len)
3710                         break;
3711
3712                 btrfs_release_path(path);
3713                 key.offset = next_key_min_offset;
3714
3715                 if (fatal_signal_pending(current)) {
3716                         ret = -EINTR;
3717                         goto out;
3718                 }
3719         }
3720         ret = 0;
3721
3722         if (last_dest_end < destoff + len) {
3723                 struct btrfs_clone_extent_info clone_info = { 0 };
3724                 /*
3725                  * We have an implicit hole (NO_HOLES feature is enabled) that
3726                  * fully or partially overlaps our cloning range at its end.
3727                  */
3728                 btrfs_release_path(path);
3729                 path->leave_spinning = 0;
3730
3731                 /*
3732                  * We are dealing with a hole and our clone_info already has a
3733                  * disk_offset of 0, we only need to fill the data length and
3734                  * file offset.
3735                  */
3736                 clone_info.data_len = destoff + len - last_dest_end;
3737                 clone_info.file_offset = last_dest_end;
3738                 ret = btrfs_punch_hole_range(inode, path,
3739                                              last_dest_end, destoff + len - 1,
3740                                              &clone_info, &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_cache *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         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4256                 ret = -EFAULT;
4257
4258         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4259                 mnt_drop_write_file(file);
4260 out:
4261         kfree(sa);
4262         return ret;
4263 }
4264
4265 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4266 {
4267         if (!capable(CAP_SYS_ADMIN))
4268                 return -EPERM;
4269
4270         return btrfs_scrub_cancel(fs_info);
4271 }
4272
4273 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4274                                        void __user *arg)
4275 {
4276         struct btrfs_ioctl_scrub_args *sa;
4277         int ret;
4278
4279         if (!capable(CAP_SYS_ADMIN))
4280                 return -EPERM;
4281
4282         sa = memdup_user(arg, sizeof(*sa));
4283         if (IS_ERR(sa))
4284                 return PTR_ERR(sa);
4285
4286         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4287
4288         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4289                 ret = -EFAULT;
4290
4291         kfree(sa);
4292         return ret;
4293 }
4294
4295 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4296                                       void __user *arg)
4297 {
4298         struct btrfs_ioctl_get_dev_stats *sa;
4299         int ret;
4300
4301         sa = memdup_user(arg, sizeof(*sa));
4302         if (IS_ERR(sa))
4303                 return PTR_ERR(sa);
4304
4305         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4306                 kfree(sa);
4307                 return -EPERM;
4308         }
4309
4310         ret = btrfs_get_dev_stats(fs_info, sa);
4311
4312         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4313                 ret = -EFAULT;
4314
4315         kfree(sa);
4316         return ret;
4317 }
4318
4319 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4320                                     void __user *arg)
4321 {
4322         struct btrfs_ioctl_dev_replace_args *p;
4323         int ret;
4324
4325         if (!capable(CAP_SYS_ADMIN))
4326                 return -EPERM;
4327
4328         p = memdup_user(arg, sizeof(*p));
4329         if (IS_ERR(p))
4330                 return PTR_ERR(p);
4331
4332         switch (p->cmd) {
4333         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4334                 if (sb_rdonly(fs_info->sb)) {
4335                         ret = -EROFS;
4336                         goto out;
4337                 }
4338                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4339                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4340                 } else {
4341                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4342                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4343                 }
4344                 break;
4345         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4346                 btrfs_dev_replace_status(fs_info, p);
4347                 ret = 0;
4348                 break;
4349         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4350                 p->result = btrfs_dev_replace_cancel(fs_info);
4351                 ret = 0;
4352                 break;
4353         default:
4354                 ret = -EINVAL;
4355                 break;
4356         }
4357
4358         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4359                 ret = -EFAULT;
4360 out:
4361         kfree(p);
4362         return ret;
4363 }
4364
4365 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4366 {
4367         int ret = 0;
4368         int i;
4369         u64 rel_ptr;
4370         int size;
4371         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4372         struct inode_fs_paths *ipath = NULL;
4373         struct btrfs_path *path;
4374
4375         if (!capable(CAP_DAC_READ_SEARCH))
4376                 return -EPERM;
4377
4378         path = btrfs_alloc_path();
4379         if (!path) {
4380                 ret = -ENOMEM;
4381                 goto out;
4382         }
4383
4384         ipa = memdup_user(arg, sizeof(*ipa));
4385         if (IS_ERR(ipa)) {
4386                 ret = PTR_ERR(ipa);
4387                 ipa = NULL;
4388                 goto out;
4389         }
4390
4391         size = min_t(u32, ipa->size, 4096);
4392         ipath = init_ipath(size, root, path);
4393         if (IS_ERR(ipath)) {
4394                 ret = PTR_ERR(ipath);
4395                 ipath = NULL;
4396                 goto out;
4397         }
4398
4399         ret = paths_from_inode(ipa->inum, ipath);
4400         if (ret < 0)
4401                 goto out;
4402
4403         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4404                 rel_ptr = ipath->fspath->val[i] -
4405                           (u64)(unsigned long)ipath->fspath->val;
4406                 ipath->fspath->val[i] = rel_ptr;
4407         }
4408
4409         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4410                            ipath->fspath, size);
4411         if (ret) {
4412                 ret = -EFAULT;
4413                 goto out;
4414         }
4415
4416 out:
4417         btrfs_free_path(path);
4418         free_ipath(ipath);
4419         kfree(ipa);
4420
4421         return ret;
4422 }
4423
4424 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4425 {
4426         struct btrfs_data_container *inodes = ctx;
4427         const size_t c = 3 * sizeof(u64);
4428
4429         if (inodes->bytes_left >= c) {
4430                 inodes->bytes_left -= c;
4431                 inodes->val[inodes->elem_cnt] = inum;
4432                 inodes->val[inodes->elem_cnt + 1] = offset;
4433                 inodes->val[inodes->elem_cnt + 2] = root;
4434                 inodes->elem_cnt += 3;
4435         } else {
4436                 inodes->bytes_missing += c - inodes->bytes_left;
4437                 inodes->bytes_left = 0;
4438                 inodes->elem_missed += 3;
4439         }
4440
4441         return 0;
4442 }
4443
4444 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4445                                         void __user *arg, int version)
4446 {
4447         int ret = 0;
4448         int size;
4449         struct btrfs_ioctl_logical_ino_args *loi;
4450         struct btrfs_data_container *inodes = NULL;
4451         struct btrfs_path *path = NULL;
4452         bool ignore_offset;
4453
4454         if (!capable(CAP_SYS_ADMIN))
4455                 return -EPERM;
4456
4457         loi = memdup_user(arg, sizeof(*loi));
4458         if (IS_ERR(loi))
4459                 return PTR_ERR(loi);
4460
4461         if (version == 1) {
4462                 ignore_offset = false;
4463                 size = min_t(u32, loi->size, SZ_64K);
4464         } else {
4465                 /* All reserved bits must be 0 for now */
4466                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4467                         ret = -EINVAL;
4468                         goto out_loi;
4469                 }
4470                 /* Only accept flags we have defined so far */
4471                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4472                         ret = -EINVAL;
4473                         goto out_loi;
4474                 }
4475                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4476                 size = min_t(u32, loi->size, SZ_16M);
4477         }
4478
4479         path = btrfs_alloc_path();
4480         if (!path) {
4481                 ret = -ENOMEM;
4482                 goto out;
4483         }
4484
4485         inodes = init_data_container(size);
4486         if (IS_ERR(inodes)) {
4487                 ret = PTR_ERR(inodes);
4488                 inodes = NULL;
4489                 goto out;
4490         }
4491
4492         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4493                                           build_ino_list, inodes, ignore_offset);
4494         if (ret == -EINVAL)
4495                 ret = -ENOENT;
4496         if (ret < 0)
4497                 goto out;
4498
4499         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4500                            size);
4501         if (ret)
4502                 ret = -EFAULT;
4503
4504 out:
4505         btrfs_free_path(path);
4506         kvfree(inodes);
4507 out_loi:
4508         kfree(loi);
4509
4510         return ret;
4511 }
4512
4513 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4514                                struct btrfs_ioctl_balance_args *bargs)
4515 {
4516         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4517
4518         bargs->flags = bctl->flags;
4519
4520         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4521                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4522         if (atomic_read(&fs_info->balance_pause_req))
4523                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4524         if (atomic_read(&fs_info->balance_cancel_req))
4525                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4526
4527         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4528         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4529         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4530
4531         spin_lock(&fs_info->balance_lock);
4532         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4533         spin_unlock(&fs_info->balance_lock);
4534 }
4535
4536 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4537 {
4538         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4539         struct btrfs_fs_info *fs_info = root->fs_info;
4540         struct btrfs_ioctl_balance_args *bargs;
4541         struct btrfs_balance_control *bctl;
4542         bool need_unlock; /* for mut. excl. ops lock */
4543         int ret;
4544
4545         if (!capable(CAP_SYS_ADMIN))
4546                 return -EPERM;
4547
4548         ret = mnt_want_write_file(file);
4549         if (ret)
4550                 return ret;
4551
4552 again:
4553         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4554                 mutex_lock(&fs_info->balance_mutex);
4555                 need_unlock = true;
4556                 goto locked;
4557         }
4558
4559         /*
4560          * mut. excl. ops lock is locked.  Three possibilities:
4561          *   (1) some other op is running
4562          *   (2) balance is running
4563          *   (3) balance is paused -- special case (think resume)
4564          */
4565         mutex_lock(&fs_info->balance_mutex);
4566         if (fs_info->balance_ctl) {
4567                 /* this is either (2) or (3) */
4568                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4569                         mutex_unlock(&fs_info->balance_mutex);
4570                         /*
4571                          * Lock released to allow other waiters to continue,
4572                          * we'll reexamine the status again.
4573                          */
4574                         mutex_lock(&fs_info->balance_mutex);
4575
4576                         if (fs_info->balance_ctl &&
4577                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4578                                 /* this is (3) */
4579                                 need_unlock = false;
4580                                 goto locked;
4581                         }
4582
4583                         mutex_unlock(&fs_info->balance_mutex);
4584                         goto again;
4585                 } else {
4586                         /* this is (2) */
4587                         mutex_unlock(&fs_info->balance_mutex);
4588                         ret = -EINPROGRESS;
4589                         goto out;
4590                 }
4591         } else {
4592                 /* this is (1) */
4593                 mutex_unlock(&fs_info->balance_mutex);
4594                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4595                 goto out;
4596         }
4597
4598 locked:
4599         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4600
4601         if (arg) {
4602                 bargs = memdup_user(arg, sizeof(*bargs));
4603                 if (IS_ERR(bargs)) {
4604                         ret = PTR_ERR(bargs);
4605                         goto out_unlock;
4606                 }
4607
4608                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4609                         if (!fs_info->balance_ctl) {
4610                                 ret = -ENOTCONN;
4611                                 goto out_bargs;
4612                         }
4613
4614                         bctl = fs_info->balance_ctl;
4615                         spin_lock(&fs_info->balance_lock);
4616                         bctl->flags |= BTRFS_BALANCE_RESUME;
4617                         spin_unlock(&fs_info->balance_lock);
4618
4619                         goto do_balance;
4620                 }
4621         } else {
4622                 bargs = NULL;
4623         }
4624
4625         if (fs_info->balance_ctl) {
4626                 ret = -EINPROGRESS;
4627                 goto out_bargs;
4628         }
4629
4630         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4631         if (!bctl) {
4632                 ret = -ENOMEM;
4633                 goto out_bargs;
4634         }
4635
4636         if (arg) {
4637                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4638                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4639                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4640
4641                 bctl->flags = bargs->flags;
4642         } else {
4643                 /* balance everything - no filters */
4644                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4645         }
4646
4647         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4648                 ret = -EINVAL;
4649                 goto out_bctl;
4650         }
4651
4652 do_balance:
4653         /*
4654          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4655          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4656          * restriper was paused all the way until unmount, in free_fs_info.
4657          * The flag should be cleared after reset_balance_state.
4658          */
4659         need_unlock = false;
4660
4661         ret = btrfs_balance(fs_info, bctl, bargs);
4662         bctl = NULL;
4663
4664         if ((ret == 0 || ret == -ECANCELED) && arg) {
4665                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4666                         ret = -EFAULT;
4667         }
4668
4669 out_bctl:
4670         kfree(bctl);
4671 out_bargs:
4672         kfree(bargs);
4673 out_unlock:
4674         mutex_unlock(&fs_info->balance_mutex);
4675         if (need_unlock)
4676                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4677 out:
4678         mnt_drop_write_file(file);
4679         return ret;
4680 }
4681
4682 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4683 {
4684         if (!capable(CAP_SYS_ADMIN))
4685                 return -EPERM;
4686
4687         switch (cmd) {
4688         case BTRFS_BALANCE_CTL_PAUSE:
4689                 return btrfs_pause_balance(fs_info);
4690         case BTRFS_BALANCE_CTL_CANCEL:
4691                 return btrfs_cancel_balance(fs_info);
4692         }
4693
4694         return -EINVAL;
4695 }
4696
4697 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4698                                          void __user *arg)
4699 {
4700         struct btrfs_ioctl_balance_args *bargs;
4701         int ret = 0;
4702
4703         if (!capable(CAP_SYS_ADMIN))
4704                 return -EPERM;
4705
4706         mutex_lock(&fs_info->balance_mutex);
4707         if (!fs_info->balance_ctl) {
4708                 ret = -ENOTCONN;
4709                 goto out;
4710         }
4711
4712         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4713         if (!bargs) {
4714                 ret = -ENOMEM;
4715                 goto out;
4716         }
4717
4718         btrfs_update_ioctl_balance_args(fs_info, bargs);
4719
4720         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4721                 ret = -EFAULT;
4722
4723         kfree(bargs);
4724 out:
4725         mutex_unlock(&fs_info->balance_mutex);
4726         return ret;
4727 }
4728
4729 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4730 {
4731         struct inode *inode = file_inode(file);
4732         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4733         struct btrfs_ioctl_quota_ctl_args *sa;
4734         int ret;
4735
4736         if (!capable(CAP_SYS_ADMIN))
4737                 return -EPERM;
4738
4739         ret = mnt_want_write_file(file);
4740         if (ret)
4741                 return ret;
4742
4743         sa = memdup_user(arg, sizeof(*sa));
4744         if (IS_ERR(sa)) {
4745                 ret = PTR_ERR(sa);
4746                 goto drop_write;
4747         }
4748
4749         down_write(&fs_info->subvol_sem);
4750
4751         switch (sa->cmd) {
4752         case BTRFS_QUOTA_CTL_ENABLE:
4753                 ret = btrfs_quota_enable(fs_info);
4754                 break;
4755         case BTRFS_QUOTA_CTL_DISABLE:
4756                 ret = btrfs_quota_disable(fs_info);
4757                 break;
4758         default:
4759                 ret = -EINVAL;
4760                 break;
4761         }
4762
4763         kfree(sa);
4764         up_write(&fs_info->subvol_sem);
4765 drop_write:
4766         mnt_drop_write_file(file);
4767         return ret;
4768 }
4769
4770 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4771 {
4772         struct inode *inode = file_inode(file);
4773         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4774         struct btrfs_root *root = BTRFS_I(inode)->root;
4775         struct btrfs_ioctl_qgroup_assign_args *sa;
4776         struct btrfs_trans_handle *trans;
4777         int ret;
4778         int err;
4779
4780         if (!capable(CAP_SYS_ADMIN))
4781                 return -EPERM;
4782
4783         ret = mnt_want_write_file(file);
4784         if (ret)
4785                 return ret;
4786
4787         sa = memdup_user(arg, sizeof(*sa));
4788         if (IS_ERR(sa)) {
4789                 ret = PTR_ERR(sa);
4790                 goto drop_write;
4791         }
4792
4793         trans = btrfs_join_transaction(root);
4794         if (IS_ERR(trans)) {
4795                 ret = PTR_ERR(trans);
4796                 goto out;
4797         }
4798
4799         if (sa->assign) {
4800                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4801         } else {
4802                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4803         }
4804
4805         /* update qgroup status and info */
4806         err = btrfs_run_qgroups(trans);
4807         if (err < 0)
4808                 btrfs_handle_fs_error(fs_info, err,
4809                                       "failed to update qgroup status and info");
4810         err = btrfs_end_transaction(trans);
4811         if (err && !ret)
4812                 ret = err;
4813
4814 out:
4815         kfree(sa);
4816 drop_write:
4817         mnt_drop_write_file(file);
4818         return ret;
4819 }
4820
4821 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4822 {
4823         struct inode *inode = file_inode(file);
4824         struct btrfs_root *root = BTRFS_I(inode)->root;
4825         struct btrfs_ioctl_qgroup_create_args *sa;
4826         struct btrfs_trans_handle *trans;
4827         int ret;
4828         int err;
4829
4830         if (!capable(CAP_SYS_ADMIN))
4831                 return -EPERM;
4832
4833         ret = mnt_want_write_file(file);
4834         if (ret)
4835                 return ret;
4836
4837         sa = memdup_user(arg, sizeof(*sa));
4838         if (IS_ERR(sa)) {
4839                 ret = PTR_ERR(sa);
4840                 goto drop_write;
4841         }
4842
4843         if (!sa->qgroupid) {
4844                 ret = -EINVAL;
4845                 goto out;
4846         }
4847
4848         trans = btrfs_join_transaction(root);
4849         if (IS_ERR(trans)) {
4850                 ret = PTR_ERR(trans);
4851                 goto out;
4852         }
4853
4854         if (sa->create) {
4855                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4856         } else {
4857                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4858         }
4859
4860         err = btrfs_end_transaction(trans);
4861         if (err && !ret)
4862                 ret = err;
4863
4864 out:
4865         kfree(sa);
4866 drop_write:
4867         mnt_drop_write_file(file);
4868         return ret;
4869 }
4870
4871 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4872 {
4873         struct inode *inode = file_inode(file);
4874         struct btrfs_root *root = BTRFS_I(inode)->root;
4875         struct btrfs_ioctl_qgroup_limit_args *sa;
4876         struct btrfs_trans_handle *trans;
4877         int ret;
4878         int err;
4879         u64 qgroupid;
4880
4881         if (!capable(CAP_SYS_ADMIN))
4882                 return -EPERM;
4883
4884         ret = mnt_want_write_file(file);
4885         if (ret)
4886                 return ret;
4887
4888         sa = memdup_user(arg, sizeof(*sa));
4889         if (IS_ERR(sa)) {
4890                 ret = PTR_ERR(sa);
4891                 goto drop_write;
4892         }
4893
4894         trans = btrfs_join_transaction(root);
4895         if (IS_ERR(trans)) {
4896                 ret = PTR_ERR(trans);
4897                 goto out;
4898         }
4899
4900         qgroupid = sa->qgroupid;
4901         if (!qgroupid) {
4902                 /* take the current subvol as qgroup */
4903                 qgroupid = root->root_key.objectid;
4904         }
4905
4906         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4907
4908         err = btrfs_end_transaction(trans);
4909         if (err && !ret)
4910                 ret = err;
4911
4912 out:
4913         kfree(sa);
4914 drop_write:
4915         mnt_drop_write_file(file);
4916         return ret;
4917 }
4918
4919 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4920 {
4921         struct inode *inode = file_inode(file);
4922         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4923         struct btrfs_ioctl_quota_rescan_args *qsa;
4924         int ret;
4925
4926         if (!capable(CAP_SYS_ADMIN))
4927                 return -EPERM;
4928
4929         ret = mnt_want_write_file(file);
4930         if (ret)
4931                 return ret;
4932
4933         qsa = memdup_user(arg, sizeof(*qsa));
4934         if (IS_ERR(qsa)) {
4935                 ret = PTR_ERR(qsa);
4936                 goto drop_write;
4937         }
4938
4939         if (qsa->flags) {
4940                 ret = -EINVAL;
4941                 goto out;
4942         }
4943
4944         ret = btrfs_qgroup_rescan(fs_info);
4945
4946 out:
4947         kfree(qsa);
4948 drop_write:
4949         mnt_drop_write_file(file);
4950         return ret;
4951 }
4952
4953 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4954                                                 void __user *arg)
4955 {
4956         struct btrfs_ioctl_quota_rescan_args *qsa;
4957         int ret = 0;
4958
4959         if (!capable(CAP_SYS_ADMIN))
4960                 return -EPERM;
4961
4962         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4963         if (!qsa)
4964                 return -ENOMEM;
4965
4966         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4967                 qsa->flags = 1;
4968                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4969         }
4970
4971         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4972                 ret = -EFAULT;
4973
4974         kfree(qsa);
4975         return ret;
4976 }
4977
4978 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4979                                                 void __user *arg)
4980 {
4981         if (!capable(CAP_SYS_ADMIN))
4982                 return -EPERM;
4983
4984         return btrfs_qgroup_wait_for_completion(fs_info, true);
4985 }
4986
4987 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4988                                             struct btrfs_ioctl_received_subvol_args *sa)
4989 {
4990         struct inode *inode = file_inode(file);
4991         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4992         struct btrfs_root *root = BTRFS_I(inode)->root;
4993         struct btrfs_root_item *root_item = &root->root_item;
4994         struct btrfs_trans_handle *trans;
4995         struct timespec64 ct = current_time(inode);
4996         int ret = 0;
4997         int received_uuid_changed;
4998
4999         if (!inode_owner_or_capable(inode))
5000                 return -EPERM;
5001
5002         ret = mnt_want_write_file(file);
5003         if (ret < 0)
5004                 return ret;
5005
5006         down_write(&fs_info->subvol_sem);
5007
5008         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5009                 ret = -EINVAL;
5010                 goto out;
5011         }
5012
5013         if (btrfs_root_readonly(root)) {
5014                 ret = -EROFS;
5015                 goto out;
5016         }
5017
5018         /*
5019          * 1 - root item
5020          * 2 - uuid items (received uuid + subvol uuid)
5021          */
5022         trans = btrfs_start_transaction(root, 3);
5023         if (IS_ERR(trans)) {
5024                 ret = PTR_ERR(trans);
5025                 trans = NULL;
5026                 goto out;
5027         }
5028
5029         sa->rtransid = trans->transid;
5030         sa->rtime.sec = ct.tv_sec;
5031         sa->rtime.nsec = ct.tv_nsec;
5032
5033         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5034                                        BTRFS_UUID_SIZE);
5035         if (received_uuid_changed &&
5036             !btrfs_is_empty_uuid(root_item->received_uuid)) {
5037                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5038                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5039                                           root->root_key.objectid);
5040                 if (ret && ret != -ENOENT) {
5041                         btrfs_abort_transaction(trans, ret);
5042                         btrfs_end_transaction(trans);
5043                         goto out;
5044                 }
5045         }
5046         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5047         btrfs_set_root_stransid(root_item, sa->stransid);
5048         btrfs_set_root_rtransid(root_item, sa->rtransid);
5049         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5050         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5051         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5052         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5053
5054         ret = btrfs_update_root(trans, fs_info->tree_root,
5055                                 &root->root_key, &root->root_item);
5056         if (ret < 0) {
5057                 btrfs_end_transaction(trans);
5058                 goto out;
5059         }
5060         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5061                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
5062                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5063                                           root->root_key.objectid);
5064                 if (ret < 0 && ret != -EEXIST) {
5065                         btrfs_abort_transaction(trans, ret);
5066                         btrfs_end_transaction(trans);
5067                         goto out;
5068                 }
5069         }
5070         ret = btrfs_commit_transaction(trans);
5071 out:
5072         up_write(&fs_info->subvol_sem);
5073         mnt_drop_write_file(file);
5074         return ret;
5075 }
5076
5077 #ifdef CONFIG_64BIT
5078 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5079                                                 void __user *arg)
5080 {
5081         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5082         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5083         int ret = 0;
5084
5085         args32 = memdup_user(arg, sizeof(*args32));
5086         if (IS_ERR(args32))
5087                 return PTR_ERR(args32);
5088
5089         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5090         if (!args64) {
5091                 ret = -ENOMEM;
5092                 goto out;
5093         }
5094
5095         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5096         args64->stransid = args32->stransid;
5097         args64->rtransid = args32->rtransid;
5098         args64->stime.sec = args32->stime.sec;
5099         args64->stime.nsec = args32->stime.nsec;
5100         args64->rtime.sec = args32->rtime.sec;
5101         args64->rtime.nsec = args32->rtime.nsec;
5102         args64->flags = args32->flags;
5103
5104         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5105         if (ret)
5106                 goto out;
5107
5108         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5109         args32->stransid = args64->stransid;
5110         args32->rtransid = args64->rtransid;
5111         args32->stime.sec = args64->stime.sec;
5112         args32->stime.nsec = args64->stime.nsec;
5113         args32->rtime.sec = args64->rtime.sec;
5114         args32->rtime.nsec = args64->rtime.nsec;
5115         args32->flags = args64->flags;
5116
5117         ret = copy_to_user(arg, args32, sizeof(*args32));
5118         if (ret)
5119                 ret = -EFAULT;
5120
5121 out:
5122         kfree(args32);
5123         kfree(args64);
5124         return ret;
5125 }
5126 #endif
5127
5128 static long btrfs_ioctl_set_received_subvol(struct file *file,
5129                                             void __user *arg)
5130 {
5131         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5132         int ret = 0;
5133
5134         sa = memdup_user(arg, sizeof(*sa));
5135         if (IS_ERR(sa))
5136                 return PTR_ERR(sa);
5137
5138         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5139
5140         if (ret)
5141                 goto out;
5142
5143         ret = copy_to_user(arg, sa, sizeof(*sa));
5144         if (ret)
5145                 ret = -EFAULT;
5146
5147 out:
5148         kfree(sa);
5149         return ret;
5150 }
5151
5152 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
5153                                         void __user *arg)
5154 {
5155         size_t len;
5156         int ret;
5157         char label[BTRFS_LABEL_SIZE];
5158
5159         spin_lock(&fs_info->super_lock);
5160         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5161         spin_unlock(&fs_info->super_lock);
5162
5163         len = strnlen(label, BTRFS_LABEL_SIZE);
5164
5165         if (len == BTRFS_LABEL_SIZE) {
5166                 btrfs_warn(fs_info,
5167                            "label is too long, return the first %zu bytes",
5168                            --len);
5169         }
5170
5171         ret = copy_to_user(arg, label, len);
5172
5173         return ret ? -EFAULT : 0;
5174 }
5175
5176 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5177 {
5178         struct inode *inode = file_inode(file);
5179         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5180         struct btrfs_root *root = BTRFS_I(inode)->root;
5181         struct btrfs_super_block *super_block = fs_info->super_copy;
5182         struct btrfs_trans_handle *trans;
5183         char label[BTRFS_LABEL_SIZE];
5184         int ret;
5185
5186         if (!capable(CAP_SYS_ADMIN))
5187                 return -EPERM;
5188
5189         if (copy_from_user(label, arg, sizeof(label)))
5190                 return -EFAULT;
5191
5192         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5193                 btrfs_err(fs_info,
5194                           "unable to set label with more than %d bytes",
5195                           BTRFS_LABEL_SIZE - 1);
5196                 return -EINVAL;
5197         }
5198
5199         ret = mnt_want_write_file(file);
5200         if (ret)
5201                 return ret;
5202
5203         trans = btrfs_start_transaction(root, 0);
5204         if (IS_ERR(trans)) {
5205                 ret = PTR_ERR(trans);
5206                 goto out_unlock;
5207         }
5208
5209         spin_lock(&fs_info->super_lock);
5210         strcpy(super_block->label, label);
5211         spin_unlock(&fs_info->super_lock);
5212         ret = btrfs_commit_transaction(trans);
5213
5214 out_unlock:
5215         mnt_drop_write_file(file);
5216         return ret;
5217 }
5218
5219 #define INIT_FEATURE_FLAGS(suffix) \
5220         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5221           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5222           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5223
5224 int btrfs_ioctl_get_supported_features(void __user *arg)
5225 {
5226         static const struct btrfs_ioctl_feature_flags features[3] = {
5227                 INIT_FEATURE_FLAGS(SUPP),
5228                 INIT_FEATURE_FLAGS(SAFE_SET),
5229                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5230         };
5231
5232         if (copy_to_user(arg, &features, sizeof(features)))
5233                 return -EFAULT;
5234
5235         return 0;
5236 }
5237
5238 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
5239                                         void __user *arg)
5240 {
5241         struct btrfs_super_block *super_block = fs_info->super_copy;
5242         struct btrfs_ioctl_feature_flags features;
5243
5244         features.compat_flags = btrfs_super_compat_flags(super_block);
5245         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5246         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5247
5248         if (copy_to_user(arg, &features, sizeof(features)))
5249                 return -EFAULT;
5250
5251         return 0;
5252 }
5253
5254 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5255                               enum btrfs_feature_set set,
5256                               u64 change_mask, u64 flags, u64 supported_flags,
5257                               u64 safe_set, u64 safe_clear)
5258 {
5259         const char *type = btrfs_feature_set_name(set);
5260         char *names;
5261         u64 disallowed, unsupported;
5262         u64 set_mask = flags & change_mask;
5263         u64 clear_mask = ~flags & change_mask;
5264
5265         unsupported = set_mask & ~supported_flags;
5266         if (unsupported) {
5267                 names = btrfs_printable_features(set, unsupported);
5268                 if (names) {
5269                         btrfs_warn(fs_info,
5270                                    "this kernel does not support the %s feature bit%s",
5271                                    names, strchr(names, ',') ? "s" : "");
5272                         kfree(names);
5273                 } else
5274                         btrfs_warn(fs_info,
5275                                    "this kernel does not support %s bits 0x%llx",
5276                                    type, unsupported);
5277                 return -EOPNOTSUPP;
5278         }
5279
5280         disallowed = set_mask & ~safe_set;
5281         if (disallowed) {
5282                 names = btrfs_printable_features(set, disallowed);
5283                 if (names) {
5284                         btrfs_warn(fs_info,
5285                                    "can't set the %s feature bit%s while mounted",
5286                                    names, strchr(names, ',') ? "s" : "");
5287                         kfree(names);
5288                 } else
5289                         btrfs_warn(fs_info,
5290                                    "can't set %s bits 0x%llx while mounted",
5291                                    type, disallowed);
5292                 return -EPERM;
5293         }
5294
5295         disallowed = clear_mask & ~safe_clear;
5296         if (disallowed) {
5297                 names = btrfs_printable_features(set, disallowed);
5298                 if (names) {
5299                         btrfs_warn(fs_info,
5300                                    "can't clear the %s feature bit%s while mounted",
5301                                    names, strchr(names, ',') ? "s" : "");
5302                         kfree(names);
5303                 } else
5304                         btrfs_warn(fs_info,
5305                                    "can't clear %s bits 0x%llx while mounted",
5306                                    type, disallowed);
5307                 return -EPERM;
5308         }
5309
5310         return 0;
5311 }
5312
5313 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5314 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5315                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5316                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5317                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5318
5319 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5320 {
5321         struct inode *inode = file_inode(file);
5322         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5323         struct btrfs_root *root = BTRFS_I(inode)->root;
5324         struct btrfs_super_block *super_block = fs_info->super_copy;
5325         struct btrfs_ioctl_feature_flags flags[2];
5326         struct btrfs_trans_handle *trans;
5327         u64 newflags;
5328         int ret;
5329
5330         if (!capable(CAP_SYS_ADMIN))
5331                 return -EPERM;
5332
5333         if (copy_from_user(flags, arg, sizeof(flags)))
5334                 return -EFAULT;
5335
5336         /* Nothing to do */
5337         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5338             !flags[0].incompat_flags)
5339                 return 0;
5340
5341         ret = check_feature(fs_info, flags[0].compat_flags,
5342                             flags[1].compat_flags, COMPAT);
5343         if (ret)
5344                 return ret;
5345
5346         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5347                             flags[1].compat_ro_flags, COMPAT_RO);
5348         if (ret)
5349                 return ret;
5350
5351         ret = check_feature(fs_info, flags[0].incompat_flags,
5352                             flags[1].incompat_flags, INCOMPAT);
5353         if (ret)
5354                 return ret;
5355
5356         ret = mnt_want_write_file(file);
5357         if (ret)
5358                 return ret;
5359
5360         trans = btrfs_start_transaction(root, 0);
5361         if (IS_ERR(trans)) {
5362                 ret = PTR_ERR(trans);
5363                 goto out_drop_write;
5364         }
5365
5366         spin_lock(&fs_info->super_lock);
5367         newflags = btrfs_super_compat_flags(super_block);
5368         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5369         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5370         btrfs_set_super_compat_flags(super_block, newflags);
5371
5372         newflags = btrfs_super_compat_ro_flags(super_block);
5373         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5374         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5375         btrfs_set_super_compat_ro_flags(super_block, newflags);
5376
5377         newflags = btrfs_super_incompat_flags(super_block);
5378         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5379         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5380         btrfs_set_super_incompat_flags(super_block, newflags);
5381         spin_unlock(&fs_info->super_lock);
5382
5383         ret = btrfs_commit_transaction(trans);
5384 out_drop_write:
5385         mnt_drop_write_file(file);
5386
5387         return ret;
5388 }
5389
5390 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5391 {
5392         struct btrfs_ioctl_send_args *arg;
5393         int ret;
5394
5395         if (compat) {
5396 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5397                 struct btrfs_ioctl_send_args_32 args32;
5398
5399                 ret = copy_from_user(&args32, argp, sizeof(args32));
5400                 if (ret)
5401                         return -EFAULT;
5402                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5403                 if (!arg)
5404                         return -ENOMEM;
5405                 arg->send_fd = args32.send_fd;
5406                 arg->clone_sources_count = args32.clone_sources_count;
5407                 arg->clone_sources = compat_ptr(args32.clone_sources);
5408                 arg->parent_root = args32.parent_root;
5409                 arg->flags = args32.flags;
5410                 memcpy(arg->reserved, args32.reserved,
5411                        sizeof(args32.reserved));
5412 #else
5413                 return -ENOTTY;
5414 #endif
5415         } else {
5416                 arg = memdup_user(argp, sizeof(*arg));
5417                 if (IS_ERR(arg))
5418                         return PTR_ERR(arg);
5419         }
5420         ret = btrfs_ioctl_send(file, arg);
5421         kfree(arg);
5422         return ret;
5423 }
5424
5425 long btrfs_ioctl(struct file *file, unsigned int
5426                 cmd, unsigned long arg)
5427 {
5428         struct inode *inode = file_inode(file);
5429         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5430         struct btrfs_root *root = BTRFS_I(inode)->root;
5431         void __user *argp = (void __user *)arg;
5432
5433         switch (cmd) {
5434         case FS_IOC_GETFLAGS:
5435                 return btrfs_ioctl_getflags(file, argp);
5436         case FS_IOC_SETFLAGS:
5437                 return btrfs_ioctl_setflags(file, argp);
5438         case FS_IOC_GETVERSION:
5439                 return btrfs_ioctl_getversion(file, argp);
5440         case FS_IOC_GETFSLABEL:
5441                 return btrfs_ioctl_get_fslabel(fs_info, argp);
5442         case FS_IOC_SETFSLABEL:
5443                 return btrfs_ioctl_set_fslabel(file, argp);
5444         case FITRIM:
5445                 return btrfs_ioctl_fitrim(fs_info, argp);
5446         case BTRFS_IOC_SNAP_CREATE:
5447                 return btrfs_ioctl_snap_create(file, argp, 0);
5448         case BTRFS_IOC_SNAP_CREATE_V2:
5449                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5450         case BTRFS_IOC_SUBVOL_CREATE:
5451                 return btrfs_ioctl_snap_create(file, argp, 1);
5452         case BTRFS_IOC_SUBVOL_CREATE_V2:
5453                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5454         case BTRFS_IOC_SNAP_DESTROY:
5455                 return btrfs_ioctl_snap_destroy(file, argp);
5456         case BTRFS_IOC_SUBVOL_GETFLAGS:
5457                 return btrfs_ioctl_subvol_getflags(file, argp);
5458         case BTRFS_IOC_SUBVOL_SETFLAGS:
5459                 return btrfs_ioctl_subvol_setflags(file, argp);
5460         case BTRFS_IOC_DEFAULT_SUBVOL:
5461                 return btrfs_ioctl_default_subvol(file, argp);
5462         case BTRFS_IOC_DEFRAG:
5463                 return btrfs_ioctl_defrag(file, NULL);
5464         case BTRFS_IOC_DEFRAG_RANGE:
5465                 return btrfs_ioctl_defrag(file, argp);
5466         case BTRFS_IOC_RESIZE:
5467                 return btrfs_ioctl_resize(file, argp);
5468         case BTRFS_IOC_ADD_DEV:
5469                 return btrfs_ioctl_add_dev(fs_info, argp);
5470         case BTRFS_IOC_RM_DEV:
5471                 return btrfs_ioctl_rm_dev(file, argp);
5472         case BTRFS_IOC_RM_DEV_V2:
5473                 return btrfs_ioctl_rm_dev_v2(file, argp);
5474         case BTRFS_IOC_FS_INFO:
5475                 return btrfs_ioctl_fs_info(fs_info, argp);
5476         case BTRFS_IOC_DEV_INFO:
5477                 return btrfs_ioctl_dev_info(fs_info, argp);
5478         case BTRFS_IOC_BALANCE:
5479                 return btrfs_ioctl_balance(file, NULL);
5480         case BTRFS_IOC_TREE_SEARCH:
5481                 return btrfs_ioctl_tree_search(file, argp);
5482         case BTRFS_IOC_TREE_SEARCH_V2:
5483                 return btrfs_ioctl_tree_search_v2(file, argp);
5484         case BTRFS_IOC_INO_LOOKUP:
5485                 return btrfs_ioctl_ino_lookup(file, argp);
5486         case BTRFS_IOC_INO_PATHS:
5487                 return btrfs_ioctl_ino_to_path(root, argp);
5488         case BTRFS_IOC_LOGICAL_INO:
5489                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5490         case BTRFS_IOC_LOGICAL_INO_V2:
5491                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5492         case BTRFS_IOC_SPACE_INFO:
5493                 return btrfs_ioctl_space_info(fs_info, argp);
5494         case BTRFS_IOC_SYNC: {
5495                 int ret;
5496
5497                 ret = btrfs_start_delalloc_roots(fs_info, -1);
5498                 if (ret)
5499                         return ret;
5500                 ret = btrfs_sync_fs(inode->i_sb, 1);
5501                 /*
5502                  * The transaction thread may want to do more work,
5503                  * namely it pokes the cleaner kthread that will start
5504                  * processing uncleaned subvols.
5505                  */
5506                 wake_up_process(fs_info->transaction_kthread);
5507                 return ret;
5508         }
5509         case BTRFS_IOC_START_SYNC:
5510                 return btrfs_ioctl_start_sync(root, argp);
5511         case BTRFS_IOC_WAIT_SYNC:
5512                 return btrfs_ioctl_wait_sync(fs_info, argp);
5513         case BTRFS_IOC_SCRUB:
5514                 return btrfs_ioctl_scrub(file, argp);
5515         case BTRFS_IOC_SCRUB_CANCEL:
5516                 return btrfs_ioctl_scrub_cancel(fs_info);
5517         case BTRFS_IOC_SCRUB_PROGRESS:
5518                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5519         case BTRFS_IOC_BALANCE_V2:
5520                 return btrfs_ioctl_balance(file, argp);
5521         case BTRFS_IOC_BALANCE_CTL:
5522                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5523         case BTRFS_IOC_BALANCE_PROGRESS:
5524                 return btrfs_ioctl_balance_progress(fs_info, argp);
5525         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5526                 return btrfs_ioctl_set_received_subvol(file, argp);
5527 #ifdef CONFIG_64BIT
5528         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5529                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5530 #endif
5531         case BTRFS_IOC_SEND:
5532                 return _btrfs_ioctl_send(file, argp, false);
5533 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5534         case BTRFS_IOC_SEND_32:
5535                 return _btrfs_ioctl_send(file, argp, true);
5536 #endif
5537         case BTRFS_IOC_GET_DEV_STATS:
5538                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5539         case BTRFS_IOC_QUOTA_CTL:
5540                 return btrfs_ioctl_quota_ctl(file, argp);
5541         case BTRFS_IOC_QGROUP_ASSIGN:
5542                 return btrfs_ioctl_qgroup_assign(file, argp);
5543         case BTRFS_IOC_QGROUP_CREATE:
5544                 return btrfs_ioctl_qgroup_create(file, argp);
5545         case BTRFS_IOC_QGROUP_LIMIT:
5546                 return btrfs_ioctl_qgroup_limit(file, argp);
5547         case BTRFS_IOC_QUOTA_RESCAN:
5548                 return btrfs_ioctl_quota_rescan(file, argp);
5549         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5550                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5551         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5552                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5553         case BTRFS_IOC_DEV_REPLACE:
5554                 return btrfs_ioctl_dev_replace(fs_info, argp);
5555         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5556                 return btrfs_ioctl_get_supported_features(argp);
5557         case BTRFS_IOC_GET_FEATURES:
5558                 return btrfs_ioctl_get_features(fs_info, argp);
5559         case BTRFS_IOC_SET_FEATURES:
5560                 return btrfs_ioctl_set_features(file, argp);
5561         case FS_IOC_FSGETXATTR:
5562                 return btrfs_ioctl_fsgetxattr(file, argp);
5563         case FS_IOC_FSSETXATTR:
5564                 return btrfs_ioctl_fssetxattr(file, argp);
5565         case BTRFS_IOC_GET_SUBVOL_INFO:
5566                 return btrfs_ioctl_get_subvol_info(file, argp);
5567         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5568                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5569         case BTRFS_IOC_INO_LOOKUP_USER:
5570                 return btrfs_ioctl_ino_lookup_user(file, argp);
5571         }
5572
5573         return -ENOTTY;
5574 }
5575
5576 #ifdef CONFIG_COMPAT
5577 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5578 {
5579         /*
5580          * These all access 32-bit values anyway so no further
5581          * handling is necessary.
5582          */
5583         switch (cmd) {
5584         case FS_IOC32_GETFLAGS:
5585                 cmd = FS_IOC_GETFLAGS;
5586                 break;
5587         case FS_IOC32_SETFLAGS:
5588                 cmd = FS_IOC_SETFLAGS;
5589                 break;
5590         case FS_IOC32_GETVERSION:
5591                 cmd = FS_IOC_GETVERSION;
5592                 break;
5593         }
5594
5595         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5596 }
5597 #endif