fs/btrfs/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/file.h>
   9 #include <linux/fs.h>
  10 #include <linux/fsnotify.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/writeback.h>
  19 #include <linux/compat.h>
  20 #include <linux/security.h>
  21 #include <linux/xattr.h>
  22 #include <linux/mm.h>
  23 #include <linux/slab.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/uuid.h>
  26 #include <linux/btrfs.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/iversion.h>
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32 #include "btrfs_inode.h"
  33 #include "print-tree.h"
  34 #include "volumes.h"
  35 #include "locking.h"
  36 #include "inode-map.h"
  37 #include "backref.h"
  38 #include "rcu-string.h"
  39 #include "send.h"
  40 #include "dev-replace.h"
  41 #include "props.h"
  42 #include "sysfs.h"
  43 #include "qgroup.h"
  44 #include "tree-log.h"
  45 #include "compression.h"
  46
  47 #ifdef CONFIG_64BIT
  48 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  49  * structures are incorrect, as the timespec structure from userspace
  50  * is 4 bytes too small. We define these alternatives here to teach
  51  * the kernel about the 32-bit struct packing.
  52  */
  53 struct btrfs_ioctl_timespec_32 {
  54         __u64 sec;
  55         __u32 nsec;
  56 } __attribute__ ((__packed__));
  57
  58 struct btrfs_ioctl_received_subvol_args_32 {
  59         char    uuid[BTRFS_UUID_SIZE];  /* in */
  60         __u64   stransid;               /* in */
  61         __u64   rtransid;               /* out */
  62         struct btrfs_ioctl_timespec_32 stime; /* in */
  63         struct btrfs_ioctl_timespec_32 rtime; /* out */
  64         __u64   flags;                  /* in */
  65         __u64   reserved[16];           /* in */
  66 } __attribute__ ((__packed__));
  67
  68 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  69                                 struct btrfs_ioctl_received_subvol_args_32)
  70 #endif
  71
  72 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  73 struct btrfs_ioctl_send_args_32 {
  74         __s64 send_fd;                  /* in */
  75         __u64 clone_sources_count;      /* in */
  76         compat_uptr_t clone_sources;    /* in */
  77         __u64 parent_root;              /* in */
  78         __u64 flags;                    /* in */
  79         __u64 reserved[4];              /* in */
  80 } __attribute__ ((__packed__));
  81
  82 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  83                                struct btrfs_ioctl_send_args_32)
  84 #endif
  85
  86 static int btrfs_clone(struct inode *src, struct inode *inode,
  87                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  88                        int no_time_update);
  89
  90 /* Mask out flags that are inappropriate for the given type of inode. */
  91 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  92                 unsigned int flags)
  93 {
  94         if (S_ISDIR(inode->i_mode))
  95                 return flags;
  96         else if (S_ISREG(inode->i_mode))
  97                 return flags & ~FS_DIRSYNC_FL;
  98         else
  99                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 100 }
 101
 102 /*
 103  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 104  * ioctl.
 105  */
 106 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if (flags & BTRFS_INODE_NOCOMPRESS)
 126                 iflags |= FS_NOCOMP_FL;
 127         else if (flags & BTRFS_INODE_COMPRESS)
 128                 iflags |= FS_COMPR_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 137 {
 138         struct btrfs_inode *binode = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (binode->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (binode->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (binode->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (binode->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 158 {
 159         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 160         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 161
 162         if (copy_to_user(arg, &flags, sizeof(flags)))
 163                 return -EFAULT;
 164         return 0;
 165 }
 166
 167 /* Check if @flags are a supported and valid set of FS_*_FL flags */
 168 static int check_fsflags(unsigned int flags)
 169 {
 170         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 171                       FS_NOATIME_FL | FS_NODUMP_FL | \
 172                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 173                       FS_NOCOMP_FL | FS_COMPR_FL |
 174                       FS_NOCOW_FL))
 175                 return -EOPNOTSUPP;
 176
 177         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 178                 return -EINVAL;
 179
 180         return 0;
 181 }
 182
 183 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 184 {
 185         struct inode *inode = file_inode(file);
 186         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 187         struct btrfs_inode *binode = BTRFS_I(inode);
 188         struct btrfs_root *root = binode->root;
 189         struct btrfs_trans_handle *trans;
 190         unsigned int fsflags, old_fsflags;
 191         int ret;
 192         u64 old_flags;
 193         unsigned int old_i_flags;
 194         umode_t mode;
 195
 196         if (!inode_owner_or_capable(inode))
 197                 return -EPERM;
 198
 199         if (btrfs_root_readonly(root))
 200                 return -EROFS;
 201
 202         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 203                 return -EFAULT;
 204
 205         ret = check_fsflags(fsflags);
 206         if (ret)
 207                 return ret;
 208
 209         ret = mnt_want_write_file(file);
 210         if (ret)
 211                 return ret;
 212
 213         inode_lock(inode);
 214
 215         old_flags = binode->flags;
 216         old_i_flags = inode->i_flags;
 217         mode = inode->i_mode;
 218
 219         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 220         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 221         if ((fsflags ^ old_fsflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 222                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 223                         ret = -EPERM;
 224                         goto out_unlock;
 225                 }
 226         }
 227
 228         if (fsflags & FS_SYNC_FL)
 229                 binode->flags |= BTRFS_INODE_SYNC;
 230         else
 231                 binode->flags &= ~BTRFS_INODE_SYNC;
 232         if (fsflags & FS_IMMUTABLE_FL)
 233                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 234         else
 235                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 236         if (fsflags & FS_APPEND_FL)
 237                 binode->flags |= BTRFS_INODE_APPEND;
 238         else
 239                 binode->flags &= ~BTRFS_INODE_APPEND;
 240         if (fsflags & FS_NODUMP_FL)
 241                 binode->flags |= BTRFS_INODE_NODUMP;
 242         else
 243                 binode->flags &= ~BTRFS_INODE_NODUMP;
 244         if (fsflags & FS_NOATIME_FL)
 245                 binode->flags |= BTRFS_INODE_NOATIME;
 246         else
 247                 binode->flags &= ~BTRFS_INODE_NOATIME;
 248         if (fsflags & FS_DIRSYNC_FL)
 249                 binode->flags |= BTRFS_INODE_DIRSYNC;
 250         else
 251                 binode->flags &= ~BTRFS_INODE_DIRSYNC;
 252         if (fsflags & FS_NOCOW_FL) {
 253                 if (S_ISREG(mode)) {
 254                         /*
 255                          * It's safe to turn csums off here, no extents exist.
 256                          * Otherwise we want the flag to reflect the real COW
 257                          * status of the file and will not set it.
 258                          */
 259                         if (inode->i_size == 0)
 260                                 binode->flags |= BTRFS_INODE_NODATACOW
 261                                               | BTRFS_INODE_NODATASUM;
 262                 } else {
 263                         binode->flags |= BTRFS_INODE_NODATACOW;
 264                 }
 265         } else {
 266                 /*
 267                  * Revert back under same assumptions as above
 268                  */
 269                 if (S_ISREG(mode)) {
 270                         if (inode->i_size == 0)
 271                                 binode->flags &= ~(BTRFS_INODE_NODATACOW
 272                                              | BTRFS_INODE_NODATASUM);
 273                 } else {
 274                         binode->flags &= ~BTRFS_INODE_NODATACOW;
 275                 }
 276         }
 277
 278         /*
 279          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 280          * flag may be changed automatically if compression code won't make
 281          * things smaller.
 282          */
 283         if (fsflags & FS_NOCOMP_FL) {
 284                 binode->flags &= ~BTRFS_INODE_COMPRESS;
 285                 binode->flags |= BTRFS_INODE_NOCOMPRESS;
 286
 287                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 288                 if (ret && ret != -ENODATA)
 289                         goto out_drop;
 290         } else if (fsflags & FS_COMPR_FL) {
 291                 const char *comp;
 292
 293                 if (IS_SWAPFILE(inode)) {
 294                         ret = -ETXTBSY;
 295                         goto out_unlock;
 296                 }
 297
 298                 binode->flags |= BTRFS_INODE_COMPRESS;
 299                 binode->flags &= ~BTRFS_INODE_NOCOMPRESS;
 300
 301                 comp = btrfs_compress_type2str(fs_info->compress_type);
 302                 if (!comp || comp[0] == 0)
 303                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 304
 305                 ret = btrfs_set_prop(inode, "btrfs.compression",
 306                                      comp, strlen(comp), 0);
 307                 if (ret)
 308                         goto out_drop;
 309
 310         } else {
 311                 ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0);
 312                 if (ret && ret != -ENODATA)
 313                         goto out_drop;
 314                 binode->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 315         }
 316
 317         trans = btrfs_start_transaction(root, 1);
 318         if (IS_ERR(trans)) {
 319                 ret = PTR_ERR(trans);
 320                 goto out_drop;
 321         }
 322
 323         btrfs_sync_inode_flags_to_i_flags(inode);
 324         inode_inc_iversion(inode);
 325         inode->i_ctime = current_time(inode);
 326         ret = btrfs_update_inode(trans, root, inode);
 327
 328         btrfs_end_transaction(trans);
 329  out_drop:
 330         if (ret) {
 331                 binode->flags = old_flags;
 332                 inode->i_flags = old_i_flags;
 333         }
 334
 335  out_unlock:
 336         inode_unlock(inode);
 337         mnt_drop_write_file(file);
 338         return ret;
 339 }
 340
 341 /*
 342  * Translate btrfs internal inode flags to xflags as expected by the
 343  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 344  * silently dropped.
 345  */
 346 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 347 {
 348         unsigned int xflags = 0;
 349
 350         if (flags & BTRFS_INODE_APPEND)
 351                 xflags |= FS_XFLAG_APPEND;
 352         if (flags & BTRFS_INODE_IMMUTABLE)
 353                 xflags |= FS_XFLAG_IMMUTABLE;
 354         if (flags & BTRFS_INODE_NOATIME)
 355                 xflags |= FS_XFLAG_NOATIME;
 356         if (flags & BTRFS_INODE_NODUMP)
 357                 xflags |= FS_XFLAG_NODUMP;
 358         if (flags & BTRFS_INODE_SYNC)
 359                 xflags |= FS_XFLAG_SYNC;
 360
 361         return xflags;
 362 }
 363
 364 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 365 static int check_xflags(unsigned int flags)
 366 {
 367         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 368                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 369                 return -EOPNOTSUPP;
 370         return 0;
 371 }
 372
 373 /*
 374  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 375  * are zeroed.
 376  */
 377 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 378 {
 379         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 380         struct fsxattr fa;
 381
 382         memset(&fa, 0, sizeof(fa));
 383         fa.fsx_xflags = btrfs_inode_flags_to_xflags(binode->flags);
 384
 385         if (copy_to_user(arg, &fa, sizeof(fa)))
 386                 return -EFAULT;
 387
 388         return 0;
 389 }
 390
 391 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 392 {
 393         struct inode *inode = file_inode(file);
 394         struct btrfs_inode *binode = BTRFS_I(inode);
 395         struct btrfs_root *root = binode->root;
 396         struct btrfs_trans_handle *trans;
 397         struct fsxattr fa;
 398         unsigned old_flags;
 399         unsigned old_i_flags;
 400         int ret = 0;
 401
 402         if (!inode_owner_or_capable(inode))
 403                 return -EPERM;
 404
 405         if (btrfs_root_readonly(root))
 406                 return -EROFS;
 407
 408         memset(&fa, 0, sizeof(fa));
 409         if (copy_from_user(&fa, arg, sizeof(fa)))
 410                 return -EFAULT;
 411
 412         ret = check_xflags(fa.fsx_xflags);
 413         if (ret)
 414                 return ret;
 415
 416         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 417                 return -EOPNOTSUPP;
 418
 419         ret = mnt_want_write_file(file);
 420         if (ret)
 421                 return ret;
 422
 423         inode_lock(inode);
 424
 425         old_flags = binode->flags;
 426         old_i_flags = inode->i_flags;
 427
 428         /* We need the capabilities to change append-only or immutable inode */
 429         if (((old_flags & (BTRFS_INODE_APPEND | BTRFS_INODE_IMMUTABLE)) ||
 430              (fa.fsx_xflags & (FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE))) &&
 431             !capable(CAP_LINUX_IMMUTABLE)) {
 432                 ret = -EPERM;
 433                 goto out_unlock;
 434         }
 435
 436         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 437                 binode->flags |= BTRFS_INODE_SYNC;
 438         else
 439                 binode->flags &= ~BTRFS_INODE_SYNC;
 440         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 441                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 442         else
 443                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 444         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 445                 binode->flags |= BTRFS_INODE_APPEND;
 446         else
 447                 binode->flags &= ~BTRFS_INODE_APPEND;
 448         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 449                 binode->flags |= BTRFS_INODE_NODUMP;
 450         else
 451                 binode->flags &= ~BTRFS_INODE_NODUMP;
 452         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 453                 binode->flags |= BTRFS_INODE_NOATIME;
 454         else
 455                 binode->flags &= ~BTRFS_INODE_NOATIME;
 456
 457         /* 1 item for the inode */
 458         trans = btrfs_start_transaction(root, 1);
 459         if (IS_ERR(trans)) {
 460                 ret = PTR_ERR(trans);
 461                 goto out_unlock;
 462         }
 463
 464         btrfs_sync_inode_flags_to_i_flags(inode);
 465         inode_inc_iversion(inode);
 466         inode->i_ctime = current_time(inode);
 467         ret = btrfs_update_inode(trans, root, inode);
 468
 469         btrfs_end_transaction(trans);
 470
 471 out_unlock:
 472         if (ret) {
 473                 binode->flags = old_flags;
 474                 inode->i_flags = old_i_flags;
 475         }
 476
 477         inode_unlock(inode);
 478         mnt_drop_write_file(file);
 479
 480         return ret;
 481 }
 482
 483 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 484 {
 485         struct inode *inode = file_inode(file);
 486
 487         return put_user(inode->i_generation, arg);
 488 }
 489
 490 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 491 {
 492         struct inode *inode = file_inode(file);
 493         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 494         struct btrfs_device *device;
 495         struct request_queue *q;
 496         struct fstrim_range range;
 497         u64 minlen = ULLONG_MAX;
 498         u64 num_devices = 0;
 499         int ret;
 500
 501         if (!capable(CAP_SYS_ADMIN))
 502                 return -EPERM;
 503
 504         /*
 505          * If the fs is mounted with nologreplay, which requires it to be
 506          * mounted in RO mode as well, we can not allow discard on free space
 507          * inside block groups, because log trees refer to extents that are not
 508          * pinned in a block group's free space cache (pinning the extents is
 509          * precisely the first phase of replaying a log tree).
 510          */
 511         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 512                 return -EROFS;
 513
 514         rcu_read_lock();
 515         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 516                                 dev_list) {
 517                 if (!device->bdev)
 518                         continue;
 519                 q = bdev_get_queue(device->bdev);
 520                 if (blk_queue_discard(q)) {
 521                         num_devices++;
 522                         minlen = min_t(u64, q->limits.discard_granularity,
 523                                      minlen);
 524                 }
 525         }
 526         rcu_read_unlock();
 527
 528         if (!num_devices)
 529                 return -EOPNOTSUPP;
 530         if (copy_from_user(&range, arg, sizeof(range)))
 531                 return -EFAULT;
 532
 533         /*
 534          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 535          * block group is in the logical address space, which can be any
 536          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 537          */
 538         if (range.len < fs_info->sb->s_blocksize)
 539                 return -EINVAL;
 540
 541         range.minlen = max(range.minlen, minlen);
 542         ret = btrfs_trim_fs(fs_info, &range);
 543         if (ret < 0)
 544                 return ret;
 545
 546         if (copy_to_user(arg, &range, sizeof(range)))
 547                 return -EFAULT;
 548
 549         return 0;
 550 }
 551
 552 int btrfs_is_empty_uuid(u8 *uuid)
 553 {
 554         int i;
 555
 556         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 557                 if (uuid[i])
 558                         return 0;
 559         }
 560         return 1;
 561 }
 562
 563 static noinline int create_subvol(struct inode *dir,
 564                                   struct dentry *dentry,
 565                                   const char *name, int namelen,
 566                                   u64 *async_transid,
 567                                   struct btrfs_qgroup_inherit *inherit)
 568 {
 569         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 570         struct btrfs_trans_handle *trans;
 571         struct btrfs_key key;
 572         struct btrfs_root_item *root_item;
 573         struct btrfs_inode_item *inode_item;
 574         struct extent_buffer *leaf;
 575         struct btrfs_root *root = BTRFS_I(dir)->root;
 576         struct btrfs_root *new_root;
 577         struct btrfs_block_rsv block_rsv;
 578         struct timespec64 cur_time = current_time(dir);
 579         struct inode *inode;
 580         int ret;
 581         int err;
 582         u64 objectid;
 583         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 584         u64 index = 0;
 585         uuid_le new_uuid;
 586
 587         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 588         if (!root_item)
 589                 return -ENOMEM;
 590
 591         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 592         if (ret)
 593                 goto fail_free;
 594
 595         /*
 596          * Don't create subvolume whose level is not zero. Or qgroup will be
 597          * screwed up since it assumes subvolume qgroup's level to be 0.
 598          */
 599         if (btrfs_qgroup_level(objectid)) {
 600                 ret = -ENOSPC;
 601                 goto fail_free;
 602         }
 603
 604         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 605         /*
 606          * The same as the snapshot creation, please see the comment
 607          * of create_snapshot().
 608          */
 609         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 610         if (ret)
 611                 goto fail_free;
 612
 613         trans = btrfs_start_transaction(root, 0);
 614         if (IS_ERR(trans)) {
 615                 ret = PTR_ERR(trans);
 616                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 617                 goto fail_free;
 618         }
 619         trans->block_rsv = &block_rsv;
 620         trans->bytes_reserved = block_rsv.size;
 621
 622         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 623         if (ret)
 624                 goto fail;
 625
 626         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 627         if (IS_ERR(leaf)) {
 628                 ret = PTR_ERR(leaf);
 629                 goto fail;
 630         }
 631
 632         btrfs_mark_buffer_dirty(leaf);
 633
 634         inode_item = &root_item->inode;
 635         btrfs_set_stack_inode_generation(inode_item, 1);
 636         btrfs_set_stack_inode_size(inode_item, 3);
 637         btrfs_set_stack_inode_nlink(inode_item, 1);
 638         btrfs_set_stack_inode_nbytes(inode_item,
 639                                      fs_info->nodesize);
 640         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 641
 642         btrfs_set_root_flags(root_item, 0);
 643         btrfs_set_root_limit(root_item, 0);
 644         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 645
 646         btrfs_set_root_bytenr(root_item, leaf->start);
 647         btrfs_set_root_generation(root_item, trans->transid);
 648         btrfs_set_root_level(root_item, 0);
 649         btrfs_set_root_refs(root_item, 1);
 650         btrfs_set_root_used(root_item, leaf->len);
 651         btrfs_set_root_last_snapshot(root_item, 0);
 652
 653         btrfs_set_root_generation_v2(root_item,
 654                         btrfs_root_generation(root_item));
 655         uuid_le_gen(&new_uuid);
 656         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 657         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 658         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 659         root_item->ctime = root_item->otime;
 660         btrfs_set_root_ctransid(root_item, trans->transid);
 661         btrfs_set_root_otransid(root_item, trans->transid);
 662
 663         btrfs_tree_unlock(leaf);
 664         free_extent_buffer(leaf);
 665         leaf = NULL;
 666
 667         btrfs_set_root_dirid(root_item, new_dirid);
 668
 669         key.objectid = objectid;
 670         key.offset = 0;
 671         key.type = BTRFS_ROOT_ITEM_KEY;
 672         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 673                                 root_item);
 674         if (ret)
 675                 goto fail;
 676
 677         key.offset = (u64)-1;
 678         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 679         if (IS_ERR(new_root)) {
 680                 ret = PTR_ERR(new_root);
 681                 btrfs_abort_transaction(trans, ret);
 682                 goto fail;
 683         }
 684
 685         btrfs_record_root_in_trans(trans, new_root);
 686
 687         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 688         if (ret) {
 689                 /* We potentially lose an unused inode item here */
 690                 btrfs_abort_transaction(trans, ret);
 691                 goto fail;
 692         }
 693
 694         mutex_lock(&new_root->objectid_mutex);
 695         new_root->highest_objectid = new_dirid;
 696         mutex_unlock(&new_root->objectid_mutex);
 697
 698         /*
 699          * insert the directory item
 700          */
 701         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 702         if (ret) {
 703                 btrfs_abort_transaction(trans, ret);
 704                 goto fail;
 705         }
 706
 707         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 708                                     BTRFS_FT_DIR, index);
 709         if (ret) {
 710                 btrfs_abort_transaction(trans, ret);
 711                 goto fail;
 712         }
 713
 714         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 715         ret = btrfs_update_inode(trans, root, dir);
 716         BUG_ON(ret);
 717
 718         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 719                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 720         BUG_ON(ret);
 721
 722         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 723                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 724         if (ret)
 725                 btrfs_abort_transaction(trans, ret);
 726
 727 fail:
 728         kfree(root_item);
 729         trans->block_rsv = NULL;
 730         trans->bytes_reserved = 0;
 731         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 732
 733         if (async_transid) {
 734                 *async_transid = trans->transid;
 735                 err = btrfs_commit_transaction_async(trans, 1);
 736                 if (err)
 737                         err = btrfs_commit_transaction(trans);
 738         } else {
 739                 err = btrfs_commit_transaction(trans);
 740         }
 741         if (err && !ret)
 742                 ret = err;
 743
 744         if (!ret) {
 745                 inode = btrfs_lookup_dentry(dir, dentry);
 746                 if (IS_ERR(inode))
 747                         return PTR_ERR(inode);
 748                 d_instantiate(dentry, inode);
 749         }
 750         return ret;
 751
 752 fail_free:
 753         kfree(root_item);
 754         return ret;
 755 }
 756
 757 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 758                            struct dentry *dentry,
 759                            u64 *async_transid, bool readonly,
 760                            struct btrfs_qgroup_inherit *inherit)
 761 {
 762         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 763         struct inode *inode;
 764         struct btrfs_pending_snapshot *pending_snapshot;
 765         struct btrfs_trans_handle *trans;
 766         int ret;
 767         bool snapshot_force_cow = false;
 768
 769         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 770                 return -EINVAL;
 771
 772         if (atomic_read(&root->nr_swapfiles)) {
 773                 btrfs_warn(fs_info,
 774                            "cannot snapshot subvolume with active swapfile");
 775                 return -ETXTBSY;
 776         }
 777
 778         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 779         if (!pending_snapshot)
 780                 return -ENOMEM;
 781
 782         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 783                         GFP_KERNEL);
 784         pending_snapshot->path = btrfs_alloc_path();
 785         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 786                 ret = -ENOMEM;
 787                 goto free_pending;
 788         }
 789
 790         /*
 791          * Force new buffered writes to reserve space even when NOCOW is
 792          * possible. This is to avoid later writeback (running dealloc) to
 793          * fallback to COW mode and unexpectedly fail with ENOSPC.
 794          */
 795         atomic_inc(&root->will_be_snapshotted);
 796         smp_mb__after_atomic();
 797         /* wait for no snapshot writes */
 798         wait_event(root->subv_writers->wait,
 799                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 800
 801         ret = btrfs_start_delalloc_snapshot(root);
 802         if (ret)
 803                 goto dec_and_free;
 804
 805         /*
 806          * All previous writes have started writeback in NOCOW mode, so now
 807          * we force future writes to fallback to COW mode during snapshot
 808          * creation.
 809          */
 810         atomic_inc(&root->snapshot_force_cow);
 811         snapshot_force_cow = true;
 812
 813         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 814
 815         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 816                              BTRFS_BLOCK_RSV_TEMP);
 817         /*
 818          * 1 - parent dir inode
 819          * 2 - dir entries
 820          * 1 - root item
 821          * 2 - root ref/backref
 822          * 1 - root of snapshot
 823          * 1 - UUID item
 824          */
 825         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 826                                         &pending_snapshot->block_rsv, 8,
 827                                         false);
 828         if (ret)
 829                 goto dec_and_free;
 830
 831         pending_snapshot->dentry = dentry;
 832         pending_snapshot->root = root;
 833         pending_snapshot->readonly = readonly;
 834         pending_snapshot->dir = dir;
 835         pending_snapshot->inherit = inherit;
 836
 837         trans = btrfs_start_transaction(root, 0);
 838         if (IS_ERR(trans)) {
 839                 ret = PTR_ERR(trans);
 840                 goto fail;
 841         }
 842
 843         spin_lock(&fs_info->trans_lock);
 844         list_add(&pending_snapshot->list,
 845                  &trans->transaction->pending_snapshots);
 846         spin_unlock(&fs_info->trans_lock);
 847         if (async_transid) {
 848                 *async_transid = trans->transid;
 849                 ret = btrfs_commit_transaction_async(trans, 1);
 850                 if (ret)
 851                         ret = btrfs_commit_transaction(trans);
 852         } else {
 853                 ret = btrfs_commit_transaction(trans);
 854         }
 855         if (ret)
 856                 goto fail;
 857
 858         ret = pending_snapshot->error;
 859         if (ret)
 860                 goto fail;
 861
 862         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 863         if (ret)
 864                 goto fail;
 865
 866         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 867         if (IS_ERR(inode)) {
 868                 ret = PTR_ERR(inode);
 869                 goto fail;
 870         }
 871
 872         d_instantiate(dentry, inode);
 873         ret = 0;
 874 fail:
 875         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 876 dec_and_free:
 877         if (snapshot_force_cow)
 878                 atomic_dec(&root->snapshot_force_cow);
 879         if (atomic_dec_and_test(&root->will_be_snapshotted))
 880                 wake_up_var(&root->will_be_snapshotted);
 881 free_pending:
 882         kfree(pending_snapshot->root_item);
 883         btrfs_free_path(pending_snapshot->path);
 884         kfree(pending_snapshot);
 885
 886         return ret;
 887 }
 888
 889 /*  copy of may_delete in fs/namei.c()
 890  *      Check whether we can remove a link victim from directory dir, check
 891  *  whether the type of victim is right.
 892  *  1. We can't do it if dir is read-only (done in permission())
 893  *  2. We should have write and exec permissions on dir
 894  *  3. We can't remove anything from append-only dir
 895  *  4. We can't do anything with immutable dir (done in permission())
 896  *  5. If the sticky bit on dir is set we should either
 897  *      a. be owner of dir, or
 898  *      b. be owner of victim, or
 899  *      c. have CAP_FOWNER capability
 900  *  6. If the victim is append-only or immutable we can't do anything with
 901  *     links pointing to it.
 902  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 903  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 904  *  9. We can't remove a root or mountpoint.
 905  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 906  *     nfs_async_unlink().
 907  */
 908
 909 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 910 {
 911         int error;
 912
 913         if (d_really_is_negative(victim))
 914                 return -ENOENT;
 915
 916         BUG_ON(d_inode(victim->d_parent) != dir);
 917         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 918
 919         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 920         if (error)
 921                 return error;
 922         if (IS_APPEND(dir))
 923                 return -EPERM;
 924         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 925             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 926                 return -EPERM;
 927         if (isdir) {
 928                 if (!d_is_dir(victim))
 929                         return -ENOTDIR;
 930                 if (IS_ROOT(victim))
 931                         return -EBUSY;
 932         } else if (d_is_dir(victim))
 933                 return -EISDIR;
 934         if (IS_DEADDIR(dir))
 935                 return -ENOENT;
 936         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 937                 return -EBUSY;
 938         return 0;
 939 }
 940
 941 /* copy of may_create in fs/namei.c() */
 942 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 943 {
 944         if (d_really_is_positive(child))
 945                 return -EEXIST;
 946         if (IS_DEADDIR(dir))
 947                 return -ENOENT;
 948         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 949 }
 950
 951 /*
 952  * Create a new subvolume below @parent.  This is largely modeled after
 953  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 954  * inside this filesystem so it's quite a bit simpler.
 955  */
 956 static noinline int btrfs_mksubvol(const struct path *parent,
 957                                    const char *name, int namelen,
 958                                    struct btrfs_root *snap_src,
 959                                    u64 *async_transid, bool readonly,
 960                                    struct btrfs_qgroup_inherit *inherit)
 961 {
 962         struct inode *dir = d_inode(parent->dentry);
 963         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 964         struct dentry *dentry;
 965         int error;
 966
 967         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 968         if (error == -EINTR)
 969                 return error;
 970
 971         dentry = lookup_one_len(name, parent->dentry, namelen);
 972         error = PTR_ERR(dentry);
 973         if (IS_ERR(dentry))
 974                 goto out_unlock;
 975
 976         error = btrfs_may_create(dir, dentry);
 977         if (error)
 978                 goto out_dput;
 979
 980         /*
 981          * even if this name doesn't exist, we may get hash collisions.
 982          * check for them now when we can safely fail
 983          */
 984         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 985                                                dir->i_ino, name,
 986                                                namelen);
 987         if (error)
 988                 goto out_dput;
 989
 990         down_read(&fs_info->subvol_sem);
 991
 992         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 993                 goto out_up_read;
 994
 995         if (snap_src) {
 996                 error = create_snapshot(snap_src, dir, dentry,
 997                                         async_transid, readonly, inherit);
 998         } else {
 999                 error = create_subvol(dir, dentry, name, namelen,
1000                                       async_transid, inherit);
1001         }
1002         if (!error)
1003                 fsnotify_mkdir(dir, dentry);
1004 out_up_read:
1005         up_read(&fs_info->subvol_sem);
1006 out_dput:
1007         dput(dentry);
1008 out_unlock:
1009         inode_unlock(dir);
1010         return error;
1011 }
1012
1013 /*
1014  * When we're defragging a range, we don't want to kick it off again
1015  * if it is really just waiting for delalloc to send it down.
1016  * If we find a nice big extent or delalloc range for the bytes in the
1017  * file you want to defrag, we return 0 to let you know to skip this
1018  * part of the file
1019  */
1020 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1021 {
1022         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1023         struct extent_map *em = NULL;
1024         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1025         u64 end;
1026
1027         read_lock(&em_tree->lock);
1028         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1029         read_unlock(&em_tree->lock);
1030
1031         if (em) {
1032                 end = extent_map_end(em);
1033                 free_extent_map(em);
1034                 if (end - offset > thresh)
1035                         return 0;
1036         }
1037         /* if we already have a nice delalloc here, just stop */
1038         thresh /= 2;
1039         end = count_range_bits(io_tree, &offset, offset + thresh,
1040                                thresh, EXTENT_DELALLOC, 1);
1041         if (end >= thresh)
1042                 return 0;
1043         return 1;
1044 }
1045
1046 /*
1047  * helper function to walk through a file and find extents
1048  * newer than a specific transid, and smaller than thresh.
1049  *
1050  * This is used by the defragging code to find new and small
1051  * extents
1052  */
1053 static int find_new_extents(struct btrfs_root *root,
1054                             struct inode *inode, u64 newer_than,
1055                             u64 *off, u32 thresh)
1056 {
1057         struct btrfs_path *path;
1058         struct btrfs_key min_key;
1059         struct extent_buffer *leaf;
1060         struct btrfs_file_extent_item *extent;
1061         int type;
1062         int ret;
1063         u64 ino = btrfs_ino(BTRFS_I(inode));
1064
1065         path = btrfs_alloc_path();
1066         if (!path)
1067                 return -ENOMEM;
1068
1069         min_key.objectid = ino;
1070         min_key.type = BTRFS_EXTENT_DATA_KEY;
1071         min_key.offset = *off;
1072
1073         while (1) {
1074                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1075                 if (ret != 0)
1076                         goto none;
1077 process_slot:
1078                 if (min_key.objectid != ino)
1079                         goto none;
1080                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1081                         goto none;
1082
1083                 leaf = path->nodes[0];
1084                 extent = btrfs_item_ptr(leaf, path->slots[0],
1085                                         struct btrfs_file_extent_item);
1086
1087                 type = btrfs_file_extent_type(leaf, extent);
1088                 if (type == BTRFS_FILE_EXTENT_REG &&
1089                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1090                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1091                         *off = min_key.offset;
1092                         btrfs_free_path(path);
1093                         return 0;
1094                 }
1095
1096                 path->slots[0]++;
1097                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1098                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1099                         goto process_slot;
1100                 }
1101
1102                 if (min_key.offset == (u64)-1)
1103                         goto none;
1104
1105                 min_key.offset++;
1106                 btrfs_release_path(path);
1107         }
1108 none:
1109         btrfs_free_path(path);
1110         return -ENOENT;
1111 }
1112
1113 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1114 {
1115         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1116         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1117         struct extent_map *em;
1118         u64 len = PAGE_SIZE;
1119
1120         /*
1121          * hopefully we have this extent in the tree already, try without
1122          * the full extent lock
1123          */
1124         read_lock(&em_tree->lock);
1125         em = lookup_extent_mapping(em_tree, start, len);
1126         read_unlock(&em_tree->lock);
1127
1128         if (!em) {
1129                 struct extent_state *cached = NULL;
1130                 u64 end = start + len - 1;
1131
1132                 /* get the big lock and read metadata off disk */
1133                 lock_extent_bits(io_tree, start, end, &cached);
1134                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1135                 unlock_extent_cached(io_tree, start, end, &cached);
1136
1137                 if (IS_ERR(em))
1138                         return NULL;
1139         }
1140
1141         return em;
1142 }
1143
1144 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1145 {
1146         struct extent_map *next;
1147         bool ret = true;
1148
1149         /* this is the last extent */
1150         if (em->start + em->len >= i_size_read(inode))
1151                 return false;
1152
1153         next = defrag_lookup_extent(inode, em->start + em->len);
1154         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1155                 ret = false;
1156         else if ((em->block_start + em->block_len == next->block_start) &&
1157                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1158                 ret = false;
1159
1160         free_extent_map(next);
1161         return ret;
1162 }
1163
1164 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1165                                u64 *last_len, u64 *skip, u64 *defrag_end,
1166                                int compress)
1167 {
1168         struct extent_map *em;
1169         int ret = 1;
1170         bool next_mergeable = true;
1171         bool prev_mergeable = true;
1172
1173         /*
1174          * make sure that once we start defragging an extent, we keep on
1175          * defragging it
1176          */
1177         if (start < *defrag_end)
1178                 return 1;
1179
1180         *skip = 0;
1181
1182         em = defrag_lookup_extent(inode, start);
1183         if (!em)
1184                 return 0;
1185
1186         /* this will cover holes, and inline extents */
1187         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1188                 ret = 0;
1189                 goto out;
1190         }
1191
1192         if (!*defrag_end)
1193                 prev_mergeable = false;
1194
1195         next_mergeable = defrag_check_next_extent(inode, em);
1196         /*
1197          * we hit a real extent, if it is big or the next extent is not a
1198          * real extent, don't bother defragging it
1199          */
1200         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1201             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1202                 ret = 0;
1203 out:
1204         /*
1205          * last_len ends up being a counter of how many bytes we've defragged.
1206          * every time we choose not to defrag an extent, we reset *last_len
1207          * so that the next tiny extent will force a defrag.
1208          *
1209          * The end result of this is that tiny extents before a single big
1210          * extent will force at least part of that big extent to be defragged.
1211          */
1212         if (ret) {
1213                 *defrag_end = extent_map_end(em);
1214         } else {
1215                 *last_len = 0;
1216                 *skip = extent_map_end(em);
1217                 *defrag_end = 0;
1218         }
1219
1220         free_extent_map(em);
1221         return ret;
1222 }
1223
1224 /*
1225  * it doesn't do much good to defrag one or two pages
1226  * at a time.  This pulls in a nice chunk of pages
1227  * to COW and defrag.
1228  *
1229  * It also makes sure the delalloc code has enough
1230  * dirty data to avoid making new small extents as part
1231  * of the defrag
1232  *
1233  * It's a good idea to start RA on this range
1234  * before calling this.
1235  */
1236 static int cluster_pages_for_defrag(struct inode *inode,
1237                                     struct page **pages,
1238                                     unsigned long start_index,
1239                                     unsigned long num_pages)
1240 {
1241         unsigned long file_end;
1242         u64 isize = i_size_read(inode);
1243         u64 page_start;
1244         u64 page_end;
1245         u64 page_cnt;
1246         int ret;
1247         int i;
1248         int i_done;
1249         struct btrfs_ordered_extent *ordered;
1250         struct extent_state *cached_state = NULL;
1251         struct extent_io_tree *tree;
1252         struct extent_changeset *data_reserved = NULL;
1253         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1254
1255         file_end = (isize - 1) >> PAGE_SHIFT;
1256         if (!isize || start_index > file_end)
1257                 return 0;
1258
1259         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1260
1261         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1262                         start_index << PAGE_SHIFT,
1263                         page_cnt << PAGE_SHIFT);
1264         if (ret)
1265                 return ret;
1266         i_done = 0;
1267         tree = &BTRFS_I(inode)->io_tree;
1268
1269         /* step one, lock all the pages */
1270         for (i = 0; i < page_cnt; i++) {
1271                 struct page *page;
1272 again:
1273                 page = find_or_create_page(inode->i_mapping,
1274                                            start_index + i, mask);
1275                 if (!page)
1276                         break;
1277
1278                 page_start = page_offset(page);
1279                 page_end = page_start + PAGE_SIZE - 1;
1280                 while (1) {
1281                         lock_extent_bits(tree, page_start, page_end,
1282                                          &cached_state);
1283                         ordered = btrfs_lookup_ordered_extent(inode,
1284                                                               page_start);
1285                         unlock_extent_cached(tree, page_start, page_end,
1286                                              &cached_state);
1287                         if (!ordered)
1288                                 break;
1289
1290                         unlock_page(page);
1291                         btrfs_start_ordered_extent(inode, ordered, 1);
1292                         btrfs_put_ordered_extent(ordered);
1293                         lock_page(page);
1294                         /*
1295                          * we unlocked the page above, so we need check if
1296                          * it was released or not.
1297                          */
1298                         if (page->mapping != inode->i_mapping) {
1299                                 unlock_page(page);
1300                                 put_page(page);
1301                                 goto again;
1302                         }
1303                 }
1304
1305                 if (!PageUptodate(page)) {
1306                         btrfs_readpage(NULL, page);
1307                         lock_page(page);
1308                         if (!PageUptodate(page)) {
1309                                 unlock_page(page);
1310                                 put_page(page);
1311                                 ret = -EIO;
1312                                 break;
1313                         }
1314                 }
1315
1316                 if (page->mapping != inode->i_mapping) {
1317                         unlock_page(page);
1318                         put_page(page);
1319                         goto again;
1320                 }
1321
1322                 pages[i] = page;
1323                 i_done++;
1324         }
1325         if (!i_done || ret)
1326                 goto out;
1327
1328         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1329                 goto out;
1330
1331         /*
1332          * so now we have a nice long stream of locked
1333          * and up to date pages, lets wait on them
1334          */
1335         for (i = 0; i < i_done; i++)
1336                 wait_on_page_writeback(pages[i]);
1337
1338         page_start = page_offset(pages[0]);
1339         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1340
1341         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1342                          page_start, page_end - 1, &cached_state);
1343         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1344                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1345                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1346                           &cached_state);
1347
1348         if (i_done != page_cnt) {
1349                 spin_lock(&BTRFS_I(inode)->lock);
1350                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1351                 spin_unlock(&BTRFS_I(inode)->lock);
1352                 btrfs_delalloc_release_space(inode, data_reserved,
1353                                 start_index << PAGE_SHIFT,
1354                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1355         }
1356
1357
1358         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1359                           &cached_state);
1360
1361         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1362                              page_start, page_end - 1, &cached_state);
1363
1364         for (i = 0; i < i_done; i++) {
1365                 clear_page_dirty_for_io(pages[i]);
1366                 ClearPageChecked(pages[i]);
1367                 set_page_extent_mapped(pages[i]);
1368                 set_page_dirty(pages[i]);
1369                 unlock_page(pages[i]);
1370                 put_page(pages[i]);
1371         }
1372         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1373                                        false);
1374         extent_changeset_free(data_reserved);
1375         return i_done;
1376 out:
1377         for (i = 0; i < i_done; i++) {
1378                 unlock_page(pages[i]);
1379                 put_page(pages[i]);
1380         }
1381         btrfs_delalloc_release_space(inode, data_reserved,
1382                         start_index << PAGE_SHIFT,
1383                         page_cnt << PAGE_SHIFT, true);
1384         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT,
1385                                        true);
1386         extent_changeset_free(data_reserved);
1387         return ret;
1388
1389 }
1390
1391 int btrfs_defrag_file(struct inode *inode, struct file *file,
1392                       struct btrfs_ioctl_defrag_range_args *range,
1393                       u64 newer_than, unsigned long max_to_defrag)
1394 {
1395         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1396         struct btrfs_root *root = BTRFS_I(inode)->root;
1397         struct file_ra_state *ra = NULL;
1398         unsigned long last_index;
1399         u64 isize = i_size_read(inode);
1400         u64 last_len = 0;
1401         u64 skip = 0;
1402         u64 defrag_end = 0;
1403         u64 newer_off = range->start;
1404         unsigned long i;
1405         unsigned long ra_index = 0;
1406         int ret;
1407         int defrag_count = 0;
1408         int compress_type = BTRFS_COMPRESS_ZLIB;
1409         u32 extent_thresh = range->extent_thresh;
1410         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1411         unsigned long cluster = max_cluster;
1412         u64 new_align = ~((u64)SZ_128K - 1);
1413         struct page **pages = NULL;
1414         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1415
1416         if (isize == 0)
1417                 return 0;
1418
1419         if (range->start >= isize)
1420                 return -EINVAL;
1421
1422         if (do_compress) {
1423                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1424                         return -EINVAL;
1425                 if (range->compress_type)
1426                         compress_type = range->compress_type;
1427         }
1428
1429         if (extent_thresh == 0)
1430                 extent_thresh = SZ_256K;
1431
1432         /*
1433          * If we were not given a file, allocate a readahead context. As
1434          * readahead is just an optimization, defrag will work without it so
1435          * we don't error out.
1436          */
1437         if (!file) {
1438                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1439                 if (ra)
1440                         file_ra_state_init(ra, inode->i_mapping);
1441         } else {
1442                 ra = &file->f_ra;
1443         }
1444
1445         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1446         if (!pages) {
1447                 ret = -ENOMEM;
1448                 goto out_ra;
1449         }
1450
1451         /* find the last page to defrag */
1452         if (range->start + range->len > range->start) {
1453                 last_index = min_t(u64, isize - 1,
1454                          range->start + range->len - 1) >> PAGE_SHIFT;
1455         } else {
1456                 last_index = (isize - 1) >> PAGE_SHIFT;
1457         }
1458
1459         if (newer_than) {
1460                 ret = find_new_extents(root, inode, newer_than,
1461                                        &newer_off, SZ_64K);
1462                 if (!ret) {
1463                         range->start = newer_off;
1464                         /*
1465                          * we always align our defrag to help keep
1466                          * the extents in the file evenly spaced
1467                          */
1468                         i = (newer_off & new_align) >> PAGE_SHIFT;
1469                 } else
1470                         goto out_ra;
1471         } else {
1472                 i = range->start >> PAGE_SHIFT;
1473         }
1474         if (!max_to_defrag)
1475                 max_to_defrag = last_index - i + 1;
1476
1477         /*
1478          * make writeback starts from i, so the defrag range can be
1479          * written sequentially.
1480          */
1481         if (i < inode->i_mapping->writeback_index)
1482                 inode->i_mapping->writeback_index = i;
1483
1484         while (i <= last_index && defrag_count < max_to_defrag &&
1485                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1486                 /*
1487                  * make sure we stop running if someone unmounts
1488                  * the FS
1489                  */
1490                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1491                         break;
1492
1493                 if (btrfs_defrag_cancelled(fs_info)) {
1494                         btrfs_debug(fs_info, "defrag_file cancelled");
1495                         ret = -EAGAIN;
1496                         break;
1497                 }
1498
1499                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1500                                          extent_thresh, &last_len, &skip,
1501                                          &defrag_end, do_compress)){
1502                         unsigned long next;
1503                         /*
1504                          * the should_defrag function tells us how much to skip
1505                          * bump our counter by the suggested amount
1506                          */
1507                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1508                         i = max(i + 1, next);
1509                         continue;
1510                 }
1511
1512                 if (!newer_than) {
1513                         cluster = (PAGE_ALIGN(defrag_end) >>
1514                                    PAGE_SHIFT) - i;
1515                         cluster = min(cluster, max_cluster);
1516                 } else {
1517                         cluster = max_cluster;
1518                 }
1519
1520                 if (i + cluster > ra_index) {
1521                         ra_index = max(i, ra_index);
1522                         if (ra)
1523                                 page_cache_sync_readahead(inode->i_mapping, ra,
1524                                                 file, ra_index, cluster);
1525                         ra_index += cluster;
1526                 }
1527
1528                 inode_lock(inode);
1529                 if (IS_SWAPFILE(inode)) {
1530                         ret = -ETXTBSY;
1531                 } else {
1532                         if (do_compress)
1533                                 BTRFS_I(inode)->defrag_compress = compress_type;
1534                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1535                 }
1536                 if (ret < 0) {
1537                         inode_unlock(inode);
1538                         goto out_ra;
1539                 }
1540
1541                 defrag_count += ret;
1542                 balance_dirty_pages_ratelimited(inode->i_mapping);
1543                 inode_unlock(inode);
1544
1545                 if (newer_than) {
1546                         if (newer_off == (u64)-1)
1547                                 break;
1548
1549                         if (ret > 0)
1550                                 i += ret;
1551
1552                         newer_off = max(newer_off + 1,
1553                                         (u64)i << PAGE_SHIFT);
1554
1555                         ret = find_new_extents(root, inode, newer_than,
1556                                                &newer_off, SZ_64K);
1557                         if (!ret) {
1558                                 range->start = newer_off;
1559                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1560                         } else {
1561                                 break;
1562                         }
1563                 } else {
1564                         if (ret > 0) {
1565                                 i += ret;
1566                                 last_len += ret << PAGE_SHIFT;
1567                         } else {
1568                                 i++;
1569                                 last_len = 0;
1570                         }
1571                 }
1572         }
1573
1574         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1575                 filemap_flush(inode->i_mapping);
1576                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1577                              &BTRFS_I(inode)->runtime_flags))
1578                         filemap_flush(inode->i_mapping);
1579         }
1580
1581         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1582                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1583         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1584                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1585         }
1586
1587         ret = defrag_count;
1588
1589 out_ra:
1590         if (do_compress) {
1591                 inode_lock(inode);
1592                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1593                 inode_unlock(inode);
1594         }
1595         if (!file)
1596                 kfree(ra);
1597         kfree(pages);
1598         return ret;
1599 }
1600
1601 static noinline int btrfs_ioctl_resize(struct file *file,
1602                                         void __user *arg)
1603 {
1604         struct inode *inode = file_inode(file);
1605         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1606         u64 new_size;
1607         u64 old_size;
1608         u64 devid = 1;
1609         struct btrfs_root *root = BTRFS_I(inode)->root;
1610         struct btrfs_ioctl_vol_args *vol_args;
1611         struct btrfs_trans_handle *trans;
1612         struct btrfs_device *device = NULL;
1613         char *sizestr;
1614         char *retptr;
1615         char *devstr = NULL;
1616         int ret = 0;
1617         int mod = 0;
1618
1619         if (!capable(CAP_SYS_ADMIN))
1620                 return -EPERM;
1621
1622         ret = mnt_want_write_file(file);
1623         if (ret)
1624                 return ret;
1625
1626         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1627                 mnt_drop_write_file(file);
1628                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1629         }
1630
1631         vol_args = memdup_user(arg, sizeof(*vol_args));
1632         if (IS_ERR(vol_args)) {
1633                 ret = PTR_ERR(vol_args);
1634                 goto out;
1635         }
1636
1637         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1638
1639         sizestr = vol_args->name;
1640         devstr = strchr(sizestr, ':');
1641         if (devstr) {
1642                 sizestr = devstr + 1;
1643                 *devstr = '\0';
1644                 devstr = vol_args->name;
1645                 ret = kstrtoull(devstr, 10, &devid);
1646                 if (ret)
1647                         goto out_free;
1648                 if (!devid) {
1649                         ret = -EINVAL;
1650                         goto out_free;
1651                 }
1652                 btrfs_info(fs_info, "resizing devid %llu", devid);
1653         }
1654
1655         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1656         if (!device) {
1657                 btrfs_info(fs_info, "resizer unable to find device %llu",
1658                            devid);
1659                 ret = -ENODEV;
1660                 goto out_free;
1661         }
1662
1663         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1664                 btrfs_info(fs_info,
1665                            "resizer unable to apply on readonly device %llu",
1666                        devid);
1667                 ret = -EPERM;
1668                 goto out_free;
1669         }
1670
1671         if (!strcmp(sizestr, "max"))
1672                 new_size = device->bdev->bd_inode->i_size;
1673         else {
1674                 if (sizestr[0] == '-') {
1675                         mod = -1;
1676                         sizestr++;
1677                 } else if (sizestr[0] == '+') {
1678                         mod = 1;
1679                         sizestr++;
1680                 }
1681                 new_size = memparse(sizestr, &retptr);
1682                 if (*retptr != '\0' || new_size == 0) {
1683                         ret = -EINVAL;
1684                         goto out_free;
1685                 }
1686         }
1687
1688         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1689                 ret = -EPERM;
1690                 goto out_free;
1691         }
1692
1693         old_size = btrfs_device_get_total_bytes(device);
1694
1695         if (mod < 0) {
1696                 if (new_size > old_size) {
1697                         ret = -EINVAL;
1698                         goto out_free;
1699                 }
1700                 new_size = old_size - new_size;
1701         } else if (mod > 0) {
1702                 if (new_size > ULLONG_MAX - old_size) {
1703                         ret = -ERANGE;
1704                         goto out_free;
1705                 }
1706                 new_size = old_size + new_size;
1707         }
1708
1709         if (new_size < SZ_256M) {
1710                 ret = -EINVAL;
1711                 goto out_free;
1712         }
1713         if (new_size > device->bdev->bd_inode->i_size) {
1714                 ret = -EFBIG;
1715                 goto out_free;
1716         }
1717
1718         new_size = round_down(new_size, fs_info->sectorsize);
1719
1720         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1721                           rcu_str_deref(device->name), new_size);
1722
1723         if (new_size > old_size) {
1724                 trans = btrfs_start_transaction(root, 0);
1725                 if (IS_ERR(trans)) {
1726                         ret = PTR_ERR(trans);
1727                         goto out_free;
1728                 }
1729                 ret = btrfs_grow_device(trans, device, new_size);
1730                 btrfs_commit_transaction(trans);
1731         } else if (new_size < old_size) {
1732                 ret = btrfs_shrink_device(device, new_size);
1733         } /* equal, nothing need to do */
1734
1735 out_free:
1736         kfree(vol_args);
1737 out:
1738         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1739         mnt_drop_write_file(file);
1740         return ret;
1741 }
1742
1743 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1744                                 const char *name, unsigned long fd, int subvol,
1745                                 u64 *transid, bool readonly,
1746                                 struct btrfs_qgroup_inherit *inherit)
1747 {
1748         int namelen;
1749         int ret = 0;
1750
1751         if (!S_ISDIR(file_inode(file)->i_mode))
1752                 return -ENOTDIR;
1753
1754         ret = mnt_want_write_file(file);
1755         if (ret)
1756                 goto out;
1757
1758         namelen = strlen(name);
1759         if (strchr(name, '/')) {
1760                 ret = -EINVAL;
1761                 goto out_drop_write;
1762         }
1763
1764         if (name[0] == '.' &&
1765            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1766                 ret = -EEXIST;
1767                 goto out_drop_write;
1768         }
1769
1770         if (subvol) {
1771                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1772                                      NULL, transid, readonly, inherit);
1773         } else {
1774                 struct fd src = fdget(fd);
1775                 struct inode *src_inode;
1776                 if (!src.file) {
1777                         ret = -EINVAL;
1778                         goto out_drop_write;
1779                 }
1780
1781                 src_inode = file_inode(src.file);
1782                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1783                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1784                                    "Snapshot src from another FS");
1785                         ret = -EXDEV;
1786                 } else if (!inode_owner_or_capable(src_inode)) {
1787                         /*
1788                          * Subvolume creation is not restricted, but snapshots
1789                          * are limited to own subvolumes only
1790                          */
1791                         ret = -EPERM;
1792                 } else {
1793                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1794                                              BTRFS_I(src_inode)->root,
1795                                              transid, readonly, inherit);
1796                 }
1797                 fdput(src);
1798         }
1799 out_drop_write:
1800         mnt_drop_write_file(file);
1801 out:
1802         return ret;
1803 }
1804
1805 static noinline int btrfs_ioctl_snap_create(struct file *file,
1806                                             void __user *arg, int subvol)
1807 {
1808         struct btrfs_ioctl_vol_args *vol_args;
1809         int ret;
1810
1811         if (!S_ISDIR(file_inode(file)->i_mode))
1812                 return -ENOTDIR;
1813
1814         vol_args = memdup_user(arg, sizeof(*vol_args));
1815         if (IS_ERR(vol_args))
1816                 return PTR_ERR(vol_args);
1817         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1818
1819         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1820                                               vol_args->fd, subvol,
1821                                               NULL, false, NULL);
1822
1823         kfree(vol_args);
1824         return ret;
1825 }
1826
1827 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1828                                                void __user *arg, int subvol)
1829 {
1830         struct btrfs_ioctl_vol_args_v2 *vol_args;
1831         int ret;
1832         u64 transid = 0;
1833         u64 *ptr = NULL;
1834         bool readonly = false;
1835         struct btrfs_qgroup_inherit *inherit = NULL;
1836
1837         if (!S_ISDIR(file_inode(file)->i_mode))
1838                 return -ENOTDIR;
1839
1840         vol_args = memdup_user(arg, sizeof(*vol_args));
1841         if (IS_ERR(vol_args))
1842                 return PTR_ERR(vol_args);
1843         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1844
1845         if (vol_args->flags &
1846             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1847               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1848                 ret = -EOPNOTSUPP;
1849                 goto free_args;
1850         }
1851
1852         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1853                 ptr = &transid;
1854         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1855                 readonly = true;
1856         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1857                 if (vol_args->size > PAGE_SIZE) {
1858                         ret = -EINVAL;
1859                         goto free_args;
1860                 }
1861                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1862                 if (IS_ERR(inherit)) {
1863                         ret = PTR_ERR(inherit);
1864                         goto free_args;
1865                 }
1866         }
1867
1868         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1869                                               vol_args->fd, subvol, ptr,
1870                                               readonly, inherit);
1871         if (ret)
1872                 goto free_inherit;
1873
1874         if (ptr && copy_to_user(arg +
1875                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1876                                         transid),
1877                                 ptr, sizeof(*ptr)))
1878                 ret = -EFAULT;
1879
1880 free_inherit:
1881         kfree(inherit);
1882 free_args:
1883         kfree(vol_args);
1884         return ret;
1885 }
1886
1887 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1888                                                 void __user *arg)
1889 {
1890         struct inode *inode = file_inode(file);
1891         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1892         struct btrfs_root *root = BTRFS_I(inode)->root;
1893         int ret = 0;
1894         u64 flags = 0;
1895
1896         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1897                 return -EINVAL;
1898
1899         down_read(&fs_info->subvol_sem);
1900         if (btrfs_root_readonly(root))
1901                 flags |= BTRFS_SUBVOL_RDONLY;
1902         up_read(&fs_info->subvol_sem);
1903
1904         if (copy_to_user(arg, &flags, sizeof(flags)))
1905                 ret = -EFAULT;
1906
1907         return ret;
1908 }
1909
1910 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1911                                               void __user *arg)
1912 {
1913         struct inode *inode = file_inode(file);
1914         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1915         struct btrfs_root *root = BTRFS_I(inode)->root;
1916         struct btrfs_trans_handle *trans;
1917         u64 root_flags;
1918         u64 flags;
1919         int ret = 0;
1920
1921         if (!inode_owner_or_capable(inode))
1922                 return -EPERM;
1923
1924         ret = mnt_want_write_file(file);
1925         if (ret)
1926                 goto out;
1927
1928         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1929                 ret = -EINVAL;
1930                 goto out_drop_write;
1931         }
1932
1933         if (copy_from_user(&flags, arg, sizeof(flags))) {
1934                 ret = -EFAULT;
1935                 goto out_drop_write;
1936         }
1937
1938         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1939                 ret = -EINVAL;
1940                 goto out_drop_write;
1941         }
1942
1943         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1944                 ret = -EOPNOTSUPP;
1945                 goto out_drop_write;
1946         }
1947
1948         down_write(&fs_info->subvol_sem);
1949
1950         /* nothing to do */
1951         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1952                 goto out_drop_sem;
1953
1954         root_flags = btrfs_root_flags(&root->root_item);
1955         if (flags & BTRFS_SUBVOL_RDONLY) {
1956                 btrfs_set_root_flags(&root->root_item,
1957                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1958         } else {
1959                 /*
1960                  * Block RO -> RW transition if this subvolume is involved in
1961                  * send
1962                  */
1963                 spin_lock(&root->root_item_lock);
1964                 if (root->send_in_progress == 0) {
1965                         btrfs_set_root_flags(&root->root_item,
1966                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1967                         spin_unlock(&root->root_item_lock);
1968                 } else {
1969                         spin_unlock(&root->root_item_lock);
1970                         btrfs_warn(fs_info,
1971                                    "Attempt to set subvolume %llu read-write during send",
1972                                    root->root_key.objectid);
1973                         ret = -EPERM;
1974                         goto out_drop_sem;
1975                 }
1976         }
1977
1978         trans = btrfs_start_transaction(root, 1);
1979         if (IS_ERR(trans)) {
1980                 ret = PTR_ERR(trans);
1981                 goto out_reset;
1982         }
1983
1984         ret = btrfs_update_root(trans, fs_info->tree_root,
1985                                 &root->root_key, &root->root_item);
1986         if (ret < 0) {
1987                 btrfs_end_transaction(trans);
1988                 goto out_reset;
1989         }
1990
1991         ret = btrfs_commit_transaction(trans);
1992
1993 out_reset:
1994         if (ret)
1995                 btrfs_set_root_flags(&root->root_item, root_flags);
1996 out_drop_sem:
1997         up_write(&fs_info->subvol_sem);
1998 out_drop_write:
1999         mnt_drop_write_file(file);
2000 out:
2001         return ret;
2002 }
2003
2004 static noinline int key_in_sk(struct btrfs_key *key,
2005                               struct btrfs_ioctl_search_key *sk)
2006 {
2007         struct btrfs_key test;
2008         int ret;
2009
2010         test.objectid = sk->min_objectid;
2011         test.type = sk->min_type;
2012         test.offset = sk->min_offset;
2013
2014         ret = btrfs_comp_cpu_keys(key, &test);
2015         if (ret < 0)
2016                 return 0;
2017
2018         test.objectid = sk->max_objectid;
2019         test.type = sk->max_type;
2020         test.offset = sk->max_offset;
2021
2022         ret = btrfs_comp_cpu_keys(key, &test);
2023         if (ret > 0)
2024                 return 0;
2025         return 1;
2026 }
2027
2028 static noinline int copy_to_sk(struct btrfs_path *path,
2029                                struct btrfs_key *key,
2030                                struct btrfs_ioctl_search_key *sk,
2031                                size_t *buf_size,
2032                                char __user *ubuf,
2033                                unsigned long *sk_offset,
2034                                int *num_found)
2035 {
2036         u64 found_transid;
2037         struct extent_buffer *leaf;
2038         struct btrfs_ioctl_search_header sh;
2039         struct btrfs_key test;
2040         unsigned long item_off;
2041         unsigned long item_len;
2042         int nritems;
2043         int i;
2044         int slot;
2045         int ret = 0;
2046
2047         leaf = path->nodes[0];
2048         slot = path->slots[0];
2049         nritems = btrfs_header_nritems(leaf);
2050
2051         if (btrfs_header_generation(leaf) > sk->max_transid) {
2052                 i = nritems;
2053                 goto advance_key;
2054         }
2055         found_transid = btrfs_header_generation(leaf);
2056
2057         for (i = slot; i < nritems; i++) {
2058                 item_off = btrfs_item_ptr_offset(leaf, i);
2059                 item_len = btrfs_item_size_nr(leaf, i);
2060
2061                 btrfs_item_key_to_cpu(leaf, key, i);
2062                 if (!key_in_sk(key, sk))
2063                         continue;
2064
2065                 if (sizeof(sh) + item_len > *buf_size) {
2066                         if (*num_found) {
2067                                 ret = 1;
2068                                 goto out;
2069                         }
2070
2071                         /*
2072                          * return one empty item back for v1, which does not
2073                          * handle -EOVERFLOW
2074                          */
2075
2076                         *buf_size = sizeof(sh) + item_len;
2077                         item_len = 0;
2078                         ret = -EOVERFLOW;
2079                 }
2080
2081                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2082                         ret = 1;
2083                         goto out;
2084                 }
2085
2086                 sh.objectid = key->objectid;
2087                 sh.offset = key->offset;
2088                 sh.type = key->type;
2089                 sh.len = item_len;
2090                 sh.transid = found_transid;
2091
2092                 /* copy search result header */
2093                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2094                         ret = -EFAULT;
2095                         goto out;
2096                 }
2097
2098                 *sk_offset += sizeof(sh);
2099
2100                 if (item_len) {
2101                         char __user *up = ubuf + *sk_offset;
2102                         /* copy the item */
2103                         if (read_extent_buffer_to_user(leaf, up,
2104                                                        item_off, item_len)) {
2105                                 ret = -EFAULT;
2106                                 goto out;
2107                         }
2108
2109                         *sk_offset += item_len;
2110                 }
2111                 (*num_found)++;
2112
2113                 if (ret) /* -EOVERFLOW from above */
2114                         goto out;
2115
2116                 if (*num_found >= sk->nr_items) {
2117                         ret = 1;
2118                         goto out;
2119                 }
2120         }
2121 advance_key:
2122         ret = 0;
2123         test.objectid = sk->max_objectid;
2124         test.type = sk->max_type;
2125         test.offset = sk->max_offset;
2126         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2127                 ret = 1;
2128         else if (key->offset < (u64)-1)
2129                 key->offset++;
2130         else if (key->type < (u8)-1) {
2131                 key->offset = 0;
2132                 key->type++;
2133         } else if (key->objectid < (u64)-1) {
2134                 key->offset = 0;
2135                 key->type = 0;
2136                 key->objectid++;
2137         } else
2138                 ret = 1;
2139 out:
2140         /*
2141          *  0: all items from this leaf copied, continue with next
2142          *  1: * more items can be copied, but unused buffer is too small
2143          *     * all items were found
2144          *     Either way, it will stops the loop which iterates to the next
2145          *     leaf
2146          *  -EOVERFLOW: item was to large for buffer
2147          *  -EFAULT: could not copy extent buffer back to userspace
2148          */
2149         return ret;
2150 }
2151
2152 static noinline int search_ioctl(struct inode *inode,
2153                                  struct btrfs_ioctl_search_key *sk,
2154                                  size_t *buf_size,
2155                                  char __user *ubuf)
2156 {
2157         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2158         struct btrfs_root *root;
2159         struct btrfs_key key;
2160         struct btrfs_path *path;
2161         int ret;
2162         int num_found = 0;
2163         unsigned long sk_offset = 0;
2164
2165         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2166                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2167                 return -EOVERFLOW;
2168         }
2169
2170         path = btrfs_alloc_path();
2171         if (!path)
2172                 return -ENOMEM;
2173
2174         if (sk->tree_id == 0) {
2175                 /* search the root of the inode that was passed */
2176                 root = BTRFS_I(inode)->root;
2177         } else {
2178                 key.objectid = sk->tree_id;
2179                 key.type = BTRFS_ROOT_ITEM_KEY;
2180                 key.offset = (u64)-1;
2181                 root = btrfs_read_fs_root_no_name(info, &key);
2182                 if (IS_ERR(root)) {
2183                         btrfs_free_path(path);
2184                         return PTR_ERR(root);
2185                 }
2186         }
2187
2188         key.objectid = sk->min_objectid;
2189         key.type = sk->min_type;
2190         key.offset = sk->min_offset;
2191
2192         while (1) {
2193                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2194                 if (ret != 0) {
2195                         if (ret > 0)
2196                                 ret = 0;
2197                         goto err;
2198                 }
2199                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2200                                  &sk_offset, &num_found);
2201                 btrfs_release_path(path);
2202                 if (ret)
2203                         break;
2204
2205         }
2206         if (ret > 0)
2207                 ret = 0;
2208 err:
2209         sk->nr_items = num_found;
2210         btrfs_free_path(path);
2211         return ret;
2212 }
2213
2214 static noinline int btrfs_ioctl_tree_search(struct file *file,
2215                                            void __user *argp)
2216 {
2217         struct btrfs_ioctl_search_args __user *uargs;
2218         struct btrfs_ioctl_search_key sk;
2219         struct inode *inode;
2220         int ret;
2221         size_t buf_size;
2222
2223         if (!capable(CAP_SYS_ADMIN))
2224                 return -EPERM;
2225
2226         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2227
2228         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2229                 return -EFAULT;
2230
2231         buf_size = sizeof(uargs->buf);
2232
2233         inode = file_inode(file);
2234         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2235
2236         /*
2237          * In the origin implementation an overflow is handled by returning a
2238          * search header with a len of zero, so reset ret.
2239          */
2240         if (ret == -EOVERFLOW)
2241                 ret = 0;
2242
2243         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2244                 ret = -EFAULT;
2245         return ret;
2246 }
2247
2248 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2249                                                void __user *argp)
2250 {
2251         struct btrfs_ioctl_search_args_v2 __user *uarg;
2252         struct btrfs_ioctl_search_args_v2 args;
2253         struct inode *inode;
2254         int ret;
2255         size_t buf_size;
2256         const size_t buf_limit = SZ_16M;
2257
2258         if (!capable(CAP_SYS_ADMIN))
2259                 return -EPERM;
2260
2261         /* copy search header and buffer size */
2262         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2263         if (copy_from_user(&args, uarg, sizeof(args)))
2264                 return -EFAULT;
2265
2266         buf_size = args.buf_size;
2267
2268         /* limit result size to 16MB */
2269         if (buf_size > buf_limit)
2270                 buf_size = buf_limit;
2271
2272         inode = file_inode(file);
2273         ret = search_ioctl(inode, &args.key, &buf_size,
2274                            (char __user *)(&uarg->buf[0]));
2275         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2276                 ret = -EFAULT;
2277         else if (ret == -EOVERFLOW &&
2278                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2279                 ret = -EFAULT;
2280
2281         return ret;
2282 }
2283
2284 /*
2285  * Search INODE_REFs to identify path name of 'dirid' directory
2286  * in a 'tree_id' tree. and sets path name to 'name'.
2287  */
2288 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2289                                 u64 tree_id, u64 dirid, char *name)
2290 {
2291         struct btrfs_root *root;
2292         struct btrfs_key key;
2293         char *ptr;
2294         int ret = -1;
2295         int slot;
2296         int len;
2297         int total_len = 0;
2298         struct btrfs_inode_ref *iref;
2299         struct extent_buffer *l;
2300         struct btrfs_path *path;
2301
2302         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2303                 name[0]='\0';
2304                 return 0;
2305         }
2306
2307         path = btrfs_alloc_path();
2308         if (!path)
2309                 return -ENOMEM;
2310
2311         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2312
2313         key.objectid = tree_id;
2314         key.type = BTRFS_ROOT_ITEM_KEY;
2315         key.offset = (u64)-1;
2316         root = btrfs_read_fs_root_no_name(info, &key);
2317         if (IS_ERR(root)) {
2318                 ret = PTR_ERR(root);
2319                 goto out;
2320         }
2321
2322         key.objectid = dirid;
2323         key.type = BTRFS_INODE_REF_KEY;
2324         key.offset = (u64)-1;
2325
2326         while (1) {
2327                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2328                 if (ret < 0)
2329                         goto out;
2330                 else if (ret > 0) {
2331                         ret = btrfs_previous_item(root, path, dirid,
2332                                                   BTRFS_INODE_REF_KEY);
2333                         if (ret < 0)
2334                                 goto out;
2335                         else if (ret > 0) {
2336                                 ret = -ENOENT;
2337                                 goto out;
2338                         }
2339                 }
2340
2341                 l = path->nodes[0];
2342                 slot = path->slots[0];
2343                 btrfs_item_key_to_cpu(l, &key, slot);
2344
2345                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2346                 len = btrfs_inode_ref_name_len(l, iref);
2347                 ptr -= len + 1;
2348                 total_len += len + 1;
2349                 if (ptr < name) {
2350                         ret = -ENAMETOOLONG;
2351                         goto out;
2352                 }
2353
2354                 *(ptr + len) = '/';
2355                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2356
2357                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2358                         break;
2359
2360                 btrfs_release_path(path);
2361                 key.objectid = key.offset;
2362                 key.offset = (u64)-1;
2363                 dirid = key.objectid;
2364         }
2365         memmove(name, ptr, total_len);
2366         name[total_len] = '\0';
2367         ret = 0;
2368 out:
2369         btrfs_free_path(path);
2370         return ret;
2371 }
2372
2373 static int btrfs_search_path_in_tree_user(struct inode *inode,
2374                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2375 {
2376         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2377         struct super_block *sb = inode->i_sb;
2378         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2379         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2380         u64 dirid = args->dirid;
2381         unsigned long item_off;
2382         unsigned long item_len;
2383         struct btrfs_inode_ref *iref;
2384         struct btrfs_root_ref *rref;
2385         struct btrfs_root *root;
2386         struct btrfs_path *path;
2387         struct btrfs_key key, key2;
2388         struct extent_buffer *leaf;
2389         struct inode *temp_inode;
2390         char *ptr;
2391         int slot;
2392         int len;
2393         int total_len = 0;
2394         int ret;
2395
2396         path = btrfs_alloc_path();
2397         if (!path)
2398                 return -ENOMEM;
2399
2400         /*
2401          * If the bottom subvolume does not exist directly under upper_limit,
2402          * construct the path in from the bottom up.
2403          */
2404         if (dirid != upper_limit.objectid) {
2405                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2406
2407                 key.objectid = treeid;
2408                 key.type = BTRFS_ROOT_ITEM_KEY;
2409                 key.offset = (u64)-1;
2410                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2411                 if (IS_ERR(root)) {
2412                         ret = PTR_ERR(root);
2413                         goto out;
2414                 }
2415
2416                 key.objectid = dirid;
2417                 key.type = BTRFS_INODE_REF_KEY;
2418                 key.offset = (u64)-1;
2419                 while (1) {
2420                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2421                         if (ret < 0) {
2422                                 goto out;
2423                         } else if (ret > 0) {
2424                                 ret = btrfs_previous_item(root, path, dirid,
2425                                                           BTRFS_INODE_REF_KEY);
2426                                 if (ret < 0) {
2427                                         goto out;
2428                                 } else if (ret > 0) {
2429                                         ret = -ENOENT;
2430                                         goto out;
2431                                 }
2432                         }
2433
2434                         leaf = path->nodes[0];
2435                         slot = path->slots[0];
2436                         btrfs_item_key_to_cpu(leaf, &key, slot);
2437
2438                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2439                         len = btrfs_inode_ref_name_len(leaf, iref);
2440                         ptr -= len + 1;
2441                         total_len += len + 1;
2442                         if (ptr < args->path) {
2443                                 ret = -ENAMETOOLONG;
2444                                 goto out;
2445                         }
2446
2447                         *(ptr + len) = '/';
2448                         read_extent_buffer(leaf, ptr,
2449                                         (unsigned long)(iref + 1), len);
2450
2451                         /* Check the read+exec permission of this directory */
2452                         ret = btrfs_previous_item(root, path, dirid,
2453                                                   BTRFS_INODE_ITEM_KEY);
2454                         if (ret < 0) {
2455                                 goto out;
2456                         } else if (ret > 0) {
2457                                 ret = -ENOENT;
2458                                 goto out;
2459                         }
2460
2461                         leaf = path->nodes[0];
2462                         slot = path->slots[0];
2463                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2464                         if (key2.objectid != dirid) {
2465                                 ret = -ENOENT;
2466                                 goto out;
2467                         }
2468
2469                         temp_inode = btrfs_iget(sb, &key2, root, NULL);
2470                         if (IS_ERR(temp_inode)) {
2471                                 ret = PTR_ERR(temp_inode);
2472                                 goto out;
2473                         }
2474                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2475                         iput(temp_inode);
2476                         if (ret) {
2477                                 ret = -EACCES;
2478                                 goto out;
2479                         }
2480
2481                         if (key.offset == upper_limit.objectid)
2482                                 break;
2483                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2484                                 ret = -EACCES;
2485                                 goto out;
2486                         }
2487
2488                         btrfs_release_path(path);
2489                         key.objectid = key.offset;
2490                         key.offset = (u64)-1;
2491                         dirid = key.objectid;
2492                 }
2493
2494                 memmove(args->path, ptr, total_len);
2495                 args->path[total_len] = '\0';
2496                 btrfs_release_path(path);
2497         }
2498
2499         /* Get the bottom subvolume's name from ROOT_REF */
2500         root = fs_info->tree_root;
2501         key.objectid = treeid;
2502         key.type = BTRFS_ROOT_REF_KEY;
2503         key.offset = args->treeid;
2504         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2505         if (ret < 0) {
2506                 goto out;
2507         } else if (ret > 0) {
2508                 ret = -ENOENT;
2509                 goto out;
2510         }
2511
2512         leaf = path->nodes[0];
2513         slot = path->slots[0];
2514         btrfs_item_key_to_cpu(leaf, &key, slot);
2515
2516         item_off = btrfs_item_ptr_offset(leaf, slot);
2517         item_len = btrfs_item_size_nr(leaf, slot);
2518         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2519         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2520         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2521                 ret = -EINVAL;
2522                 goto out;
2523         }
2524
2525         /* Copy subvolume's name */
2526         item_off += sizeof(struct btrfs_root_ref);
2527         item_len -= sizeof(struct btrfs_root_ref);
2528         read_extent_buffer(leaf, args->name, item_off, item_len);
2529         args->name[item_len] = 0;
2530
2531 out:
2532         btrfs_free_path(path);
2533         return ret;
2534 }
2535
2536 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2537                                            void __user *argp)
2538 {
2539         struct btrfs_ioctl_ino_lookup_args *args;
2540         struct inode *inode;
2541         int ret = 0;
2542
2543         args = memdup_user(argp, sizeof(*args));
2544         if (IS_ERR(args))
2545                 return PTR_ERR(args);
2546
2547         inode = file_inode(file);
2548
2549         /*
2550          * Unprivileged query to obtain the containing subvolume root id. The
2551          * path is reset so it's consistent with btrfs_search_path_in_tree.
2552          */
2553         if (args->treeid == 0)
2554                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2555
2556         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2557                 args->name[0] = 0;
2558                 goto out;
2559         }
2560
2561         if (!capable(CAP_SYS_ADMIN)) {
2562                 ret = -EPERM;
2563                 goto out;
2564         }
2565
2566         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2567                                         args->treeid, args->objectid,
2568                                         args->name);
2569
2570 out:
2571         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2572                 ret = -EFAULT;
2573
2574         kfree(args);
2575         return ret;
2576 }
2577
2578 /*
2579  * Version of ino_lookup ioctl (unprivileged)
2580  *
2581  * The main differences from ino_lookup ioctl are:
2582  *
2583  *   1. Read + Exec permission will be checked using inode_permission() during
2584  *      path construction. -EACCES will be returned in case of failure.
2585  *   2. Path construction will be stopped at the inode number which corresponds
2586  *      to the fd with which this ioctl is called. If constructed path does not
2587  *      exist under fd's inode, -EACCES will be returned.
2588  *   3. The name of bottom subvolume is also searched and filled.
2589  */
2590 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2591 {
2592         struct btrfs_ioctl_ino_lookup_user_args *args;
2593         struct inode *inode;
2594         int ret;
2595
2596         args = memdup_user(argp, sizeof(*args));
2597         if (IS_ERR(args))
2598                 return PTR_ERR(args);
2599
2600         inode = file_inode(file);
2601
2602         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2603             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2604                 /*
2605                  * The subvolume does not exist under fd with which this is
2606                  * called
2607                  */
2608                 kfree(args);
2609                 return -EACCES;
2610         }
2611
2612         ret = btrfs_search_path_in_tree_user(inode, args);
2613
2614         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2615                 ret = -EFAULT;
2616
2617         kfree(args);
2618         return ret;
2619 }
2620
2621 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2622 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2623 {
2624         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2625         struct btrfs_fs_info *fs_info;
2626         struct btrfs_root *root;
2627         struct btrfs_path *path;
2628         struct btrfs_key key;
2629         struct btrfs_root_item *root_item;
2630         struct btrfs_root_ref *rref;
2631         struct extent_buffer *leaf;
2632         unsigned long item_off;
2633         unsigned long item_len;
2634         struct inode *inode;
2635         int slot;
2636         int ret = 0;
2637
2638         path = btrfs_alloc_path();
2639         if (!path)
2640                 return -ENOMEM;
2641
2642         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2643         if (!subvol_info) {
2644                 btrfs_free_path(path);
2645                 return -ENOMEM;
2646         }
2647
2648         inode = file_inode(file);
2649         fs_info = BTRFS_I(inode)->root->fs_info;
2650
2651         /* Get root_item of inode's subvolume */
2652         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2653         key.type = BTRFS_ROOT_ITEM_KEY;
2654         key.offset = (u64)-1;
2655         root = btrfs_read_fs_root_no_name(fs_info, &key);
2656         if (IS_ERR(root)) {
2657                 ret = PTR_ERR(root);
2658                 goto out;
2659         }
2660         root_item = &root->root_item;
2661
2662         subvol_info->treeid = key.objectid;
2663
2664         subvol_info->generation = btrfs_root_generation(root_item);
2665         subvol_info->flags = btrfs_root_flags(root_item);
2666
2667         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2668         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2669                                                     BTRFS_UUID_SIZE);
2670         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2671                                                     BTRFS_UUID_SIZE);
2672
2673         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2674         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2675         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2676
2677         subvol_info->otransid = btrfs_root_otransid(root_item);
2678         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2679         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2680
2681         subvol_info->stransid = btrfs_root_stransid(root_item);
2682         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2683         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2684
2685         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2686         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2687         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2688
2689         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2690                 /* Search root tree for ROOT_BACKREF of this subvolume */
2691                 root = fs_info->tree_root;
2692
2693                 key.type = BTRFS_ROOT_BACKREF_KEY;
2694                 key.offset = 0;
2695                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2696                 if (ret < 0) {
2697                         goto out;
2698                 } else if (path->slots[0] >=
2699                            btrfs_header_nritems(path->nodes[0])) {
2700                         ret = btrfs_next_leaf(root, path);
2701                         if (ret < 0) {
2702                                 goto out;
2703                         } else if (ret > 0) {
2704                                 ret = -EUCLEAN;
2705                                 goto out;
2706                         }
2707                 }
2708
2709                 leaf = path->nodes[0];
2710                 slot = path->slots[0];
2711                 btrfs_item_key_to_cpu(leaf, &key, slot);
2712                 if (key.objectid == subvol_info->treeid &&
2713                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2714                         subvol_info->parent_id = key.offset;
2715
2716                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2717                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2718
2719                         item_off = btrfs_item_ptr_offset(leaf, slot)
2720                                         + sizeof(struct btrfs_root_ref);
2721                         item_len = btrfs_item_size_nr(leaf, slot)
2722                                         - sizeof(struct btrfs_root_ref);
2723                         read_extent_buffer(leaf, subvol_info->name,
2724                                            item_off, item_len);
2725                 } else {
2726                         ret = -ENOENT;
2727                         goto out;
2728                 }
2729         }
2730
2731         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2732                 ret = -EFAULT;
2733
2734 out:
2735         btrfs_free_path(path);
2736         kzfree(subvol_info);
2737         return ret;
2738 }
2739
2740 /*
2741  * Return ROOT_REF information of the subvolume containing this inode
2742  * except the subvolume name.
2743  */
2744 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2745 {
2746         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2747         struct btrfs_root_ref *rref;
2748         struct btrfs_root *root;
2749         struct btrfs_path *path;
2750         struct btrfs_key key;
2751         struct extent_buffer *leaf;
2752         struct inode *inode;
2753         u64 objectid;
2754         int slot;
2755         int ret;
2756         u8 found;
2757
2758         path = btrfs_alloc_path();
2759         if (!path)
2760                 return -ENOMEM;
2761
2762         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2763         if (IS_ERR(rootrefs)) {
2764                 btrfs_free_path(path);
2765                 return PTR_ERR(rootrefs);
2766         }
2767
2768         inode = file_inode(file);
2769         root = BTRFS_I(inode)->root->fs_info->tree_root;
2770         objectid = BTRFS_I(inode)->root->root_key.objectid;
2771
2772         key.objectid = objectid;
2773         key.type = BTRFS_ROOT_REF_KEY;
2774         key.offset = rootrefs->min_treeid;
2775         found = 0;
2776
2777         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2778         if (ret < 0) {
2779                 goto out;
2780         } else if (path->slots[0] >=
2781                    btrfs_header_nritems(path->nodes[0])) {
2782                 ret = btrfs_next_leaf(root, path);
2783                 if (ret < 0) {
2784                         goto out;
2785                 } else if (ret > 0) {
2786                         ret = -EUCLEAN;
2787                         goto out;
2788                 }
2789         }
2790         while (1) {
2791                 leaf = path->nodes[0];
2792                 slot = path->slots[0];
2793
2794                 btrfs_item_key_to_cpu(leaf, &key, slot);
2795                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2796                         ret = 0;
2797                         goto out;
2798                 }
2799
2800                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2801                         ret = -EOVERFLOW;
2802                         goto out;
2803                 }
2804
2805                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2806                 rootrefs->rootref[found].treeid = key.offset;
2807                 rootrefs->rootref[found].dirid =
2808                                   btrfs_root_ref_dirid(leaf, rref);
2809                 found++;
2810
2811                 ret = btrfs_next_item(root, path);
2812                 if (ret < 0) {
2813                         goto out;
2814                 } else if (ret > 0) {
2815                         ret = -EUCLEAN;
2816                         goto out;
2817                 }
2818         }
2819
2820 out:
2821         if (!ret || ret == -EOVERFLOW) {
2822                 rootrefs->num_items = found;
2823                 /* update min_treeid for next search */
2824                 if (found)
2825                         rootrefs->min_treeid =
2826                                 rootrefs->rootref[found - 1].treeid + 1;
2827                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2828                         ret = -EFAULT;
2829         }
2830
2831         kfree(rootrefs);
2832         btrfs_free_path(path);
2833
2834         return ret;
2835 }
2836
2837 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2838                                              void __user *arg)
2839 {
2840         struct dentry *parent = file->f_path.dentry;
2841         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2842         struct dentry *dentry;
2843         struct inode *dir = d_inode(parent);
2844         struct inode *inode;
2845         struct btrfs_root *root = BTRFS_I(dir)->root;
2846         struct btrfs_root *dest = NULL;
2847         struct btrfs_ioctl_vol_args *vol_args;
2848         int namelen;
2849         int err = 0;
2850
2851         if (!S_ISDIR(dir->i_mode))
2852                 return -ENOTDIR;
2853
2854         vol_args = memdup_user(arg, sizeof(*vol_args));
2855         if (IS_ERR(vol_args))
2856                 return PTR_ERR(vol_args);
2857
2858         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2859         namelen = strlen(vol_args->name);
2860         if (strchr(vol_args->name, '/') ||
2861             strncmp(vol_args->name, "..", namelen) == 0) {
2862                 err = -EINVAL;
2863                 goto out;
2864         }
2865
2866         err = mnt_want_write_file(file);
2867         if (err)
2868                 goto out;
2869
2870
2871         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2872         if (err == -EINTR)
2873                 goto out_drop_write;
2874         dentry = lookup_one_len(vol_args->name, parent, namelen);
2875         if (IS_ERR(dentry)) {
2876                 err = PTR_ERR(dentry);
2877                 goto out_unlock_dir;
2878         }
2879
2880         if (d_really_is_negative(dentry)) {
2881                 err = -ENOENT;
2882                 goto out_dput;
2883         }
2884
2885         inode = d_inode(dentry);
2886         dest = BTRFS_I(inode)->root;
2887         if (!capable(CAP_SYS_ADMIN)) {
2888                 /*
2889                  * Regular user.  Only allow this with a special mount
2890                  * option, when the user has write+exec access to the
2891                  * subvol root, and when rmdir(2) would have been
2892                  * allowed.
2893                  *
2894                  * Note that this is _not_ check that the subvol is
2895                  * empty or doesn't contain data that we wouldn't
2896                  * otherwise be able to delete.
2897                  *
2898                  * Users who want to delete empty subvols should try
2899                  * rmdir(2).
2900                  */
2901                 err = -EPERM;
2902                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2903                         goto out_dput;
2904
2905                 /*
2906                  * Do not allow deletion if the parent dir is the same
2907                  * as the dir to be deleted.  That means the ioctl
2908                  * must be called on the dentry referencing the root
2909                  * of the subvol, not a random directory contained
2910                  * within it.
2911                  */
2912                 err = -EINVAL;
2913                 if (root == dest)
2914                         goto out_dput;
2915
2916                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2917                 if (err)
2918                         goto out_dput;
2919         }
2920
2921         /* check if subvolume may be deleted by a user */
2922         err = btrfs_may_delete(dir, dentry, 1);
2923         if (err)
2924                 goto out_dput;
2925
2926         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2927                 err = -EINVAL;
2928                 goto out_dput;
2929         }
2930
2931         inode_lock(inode);
2932         err = btrfs_delete_subvolume(dir, dentry);
2933         inode_unlock(inode);
2934         if (!err)
2935                 d_delete(dentry);
2936
2937 out_dput:
2938         dput(dentry);
2939 out_unlock_dir:
2940         inode_unlock(dir);
2941 out_drop_write:
2942         mnt_drop_write_file(file);
2943 out:
2944         kfree(vol_args);
2945         return err;
2946 }
2947
2948 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2949 {
2950         struct inode *inode = file_inode(file);
2951         struct btrfs_root *root = BTRFS_I(inode)->root;
2952         struct btrfs_ioctl_defrag_range_args *range;
2953         int ret;
2954
2955         ret = mnt_want_write_file(file);
2956         if (ret)
2957                 return ret;
2958
2959         if (btrfs_root_readonly(root)) {
2960                 ret = -EROFS;
2961                 goto out;
2962         }
2963
2964         switch (inode->i_mode & S_IFMT) {
2965         case S_IFDIR:
2966                 if (!capable(CAP_SYS_ADMIN)) {
2967                         ret = -EPERM;
2968                         goto out;
2969                 }
2970                 ret = btrfs_defrag_root(root);
2971                 break;
2972         case S_IFREG:
2973                 /*
2974                  * Note that this does not check the file descriptor for write
2975                  * access. This prevents defragmenting executables that are
2976                  * running and allows defrag on files open in read-only mode.
2977                  */
2978                 if (!capable(CAP_SYS_ADMIN) &&
2979                     inode_permission(inode, MAY_WRITE)) {
2980                         ret = -EPERM;
2981                         goto out;
2982                 }
2983
2984                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2985                 if (!range) {
2986                         ret = -ENOMEM;
2987                         goto out;
2988                 }
2989
2990                 if (argp) {
2991                         if (copy_from_user(range, argp,
2992                                            sizeof(*range))) {
2993                                 ret = -EFAULT;
2994                                 kfree(range);
2995                                 goto out;
2996                         }
2997                         /* compression requires us to start the IO */
2998                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2999                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3000                                 range->extent_thresh = (u32)-1;
3001                         }
3002                 } else {
3003                         /* the rest are all set to zero by kzalloc */
3004                         range->len = (u64)-1;
3005                 }
3006                 ret = btrfs_defrag_file(file_inode(file), file,
3007                                         range, BTRFS_OLDEST_GENERATION, 0);
3008                 if (ret > 0)
3009                         ret = 0;
3010                 kfree(range);
3011                 break;
3012         default:
3013                 ret = -EINVAL;
3014         }
3015 out:
3016         mnt_drop_write_file(file);
3017         return ret;
3018 }
3019
3020 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3021 {
3022         struct btrfs_ioctl_vol_args *vol_args;
3023         int ret;
3024
3025         if (!capable(CAP_SYS_ADMIN))
3026                 return -EPERM;
3027
3028         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3029                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3030
3031         vol_args = memdup_user(arg, sizeof(*vol_args));
3032         if (IS_ERR(vol_args)) {
3033                 ret = PTR_ERR(vol_args);
3034                 goto out;
3035         }
3036
3037         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3038         ret = btrfs_init_new_device(fs_info, vol_args->name);
3039
3040         if (!ret)
3041                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3042
3043         kfree(vol_args);
3044 out:
3045         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3046         return ret;
3047 }
3048
3049 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3050 {
3051         struct inode *inode = file_inode(file);
3052         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3053         struct btrfs_ioctl_vol_args_v2 *vol_args;
3054         int ret;
3055
3056         if (!capable(CAP_SYS_ADMIN))
3057                 return -EPERM;
3058
3059         ret = mnt_want_write_file(file);
3060         if (ret)
3061                 return ret;
3062
3063         vol_args = memdup_user(arg, sizeof(*vol_args));
3064         if (IS_ERR(vol_args)) {
3065                 ret = PTR_ERR(vol_args);
3066                 goto err_drop;
3067         }
3068
3069         /* Check for compatibility reject unknown flags */
3070         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3071                 ret = -EOPNOTSUPP;
3072                 goto out;
3073         }
3074
3075         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3076                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3077                 goto out;
3078         }
3079
3080         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3081                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3082         } else {
3083                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3084                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3085         }
3086         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3087
3088         if (!ret) {
3089                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3090                         btrfs_info(fs_info, "device deleted: id %llu",
3091                                         vol_args->devid);
3092                 else
3093                         btrfs_info(fs_info, "device deleted: %s",
3094                                         vol_args->name);
3095         }
3096 out:
3097         kfree(vol_args);
3098 err_drop:
3099         mnt_drop_write_file(file);
3100         return ret;
3101 }
3102
3103 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3104 {
3105         struct inode *inode = file_inode(file);
3106         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3107         struct btrfs_ioctl_vol_args *vol_args;
3108         int ret;
3109
3110         if (!capable(CAP_SYS_ADMIN))
3111                 return -EPERM;
3112
3113         ret = mnt_want_write_file(file);
3114         if (ret)
3115                 return ret;
3116
3117         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3118                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3119                 goto out_drop_write;
3120         }
3121
3122         vol_args = memdup_user(arg, sizeof(*vol_args));
3123         if (IS_ERR(vol_args)) {
3124                 ret = PTR_ERR(vol_args);
3125                 goto out;
3126         }
3127
3128         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3129         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3130
3131         if (!ret)
3132                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3133         kfree(vol_args);
3134 out:
3135         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3136 out_drop_write:
3137         mnt_drop_write_file(file);
3138
3139         return ret;
3140 }
3141
3142 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3143                                 void __user *arg)
3144 {
3145         struct btrfs_ioctl_fs_info_args *fi_args;
3146         struct btrfs_device *device;
3147         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3148         int ret = 0;
3149
3150         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3151         if (!fi_args)
3152                 return -ENOMEM;
3153
3154         rcu_read_lock();
3155         fi_args->num_devices = fs_devices->num_devices;
3156
3157         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3158                 if (device->devid > fi_args->max_id)
3159                         fi_args->max_id = device->devid;
3160         }
3161         rcu_read_unlock();
3162
3163         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3164         fi_args->nodesize = fs_info->nodesize;
3165         fi_args->sectorsize = fs_info->sectorsize;
3166         fi_args->clone_alignment = fs_info->sectorsize;
3167
3168         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3169                 ret = -EFAULT;
3170
3171         kfree(fi_args);
3172         return ret;
3173 }
3174
3175 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3176                                  void __user *arg)
3177 {
3178         struct btrfs_ioctl_dev_info_args *di_args;
3179         struct btrfs_device *dev;
3180         int ret = 0;
3181         char *s_uuid = NULL;
3182
3183         di_args = memdup_user(arg, sizeof(*di_args));
3184         if (IS_ERR(di_args))
3185                 return PTR_ERR(di_args);
3186
3187         if (!btrfs_is_empty_uuid(di_args->uuid))
3188                 s_uuid = di_args->uuid;
3189
3190         rcu_read_lock();
3191         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3192                                 NULL, true);
3193
3194         if (!dev) {
3195                 ret = -ENODEV;
3196                 goto out;
3197         }
3198
3199         di_args->devid = dev->devid;
3200         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3201         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3202         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3203         if (dev->name) {
3204                 strncpy(di_args->path, rcu_str_deref(dev->name),
3205                                 sizeof(di_args->path) - 1);
3206                 di_args->path[sizeof(di_args->path) - 1] = 0;
3207         } else {
3208                 di_args->path[0] = '\0';
3209         }
3210
3211 out:
3212         rcu_read_unlock();
3213         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3214                 ret = -EFAULT;
3215
3216         kfree(di_args);
3217         return ret;
3218 }
3219
3220 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3221                                        struct inode *inode2, u64 loff2, u64 len)
3222 {
3223         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3224         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3225 }
3226
3227 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3228                                      struct inode *inode2, u64 loff2, u64 len)
3229 {
3230         if (inode1 < inode2) {
3231                 swap(inode1, inode2);
3232                 swap(loff1, loff2);
3233         } else if (inode1 == inode2 && loff2 < loff1) {
3234                 swap(loff1, loff2);
3235         }
3236         lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3237         lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3238 }
3239
3240 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
3241                                    struct inode *dst, u64 dst_loff)
3242 {
3243         int ret;
3244
3245         /*
3246          * Lock destination range to serialize with concurrent readpages() and
3247          * source range to serialize with relocation.
3248          */
3249         btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3250         ret = btrfs_clone(src, dst, loff, len, len, dst_loff, 1);
3251         btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3252
3253         return ret;
3254 }
3255
3256 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3257
3258 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3259                              struct inode *dst, u64 dst_loff)
3260 {
3261         int ret;
3262         u64 i, tail_len, chunk_count;
3263
3264         tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3265         chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3266
3267         for (i = 0; i < chunk_count; i++) {
3268                 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3269                                               dst, dst_loff);
3270                 if (ret)
3271                         return ret;
3272
3273                 loff += BTRFS_MAX_DEDUPE_LEN;
3274                 dst_loff += BTRFS_MAX_DEDUPE_LEN;
3275         }
3276
3277         if (tail_len > 0)
3278                 ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3279                                               dst_loff);
3280
3281         return ret;
3282 }
3283
3284 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3285                                      struct inode *inode,
3286                                      u64 endoff,
3287                                      const u64 destoff,
3288                                      const u64 olen,
3289                                      int no_time_update)
3290 {
3291         struct btrfs_root *root = BTRFS_I(inode)->root;
3292         int ret;
3293
3294         inode_inc_iversion(inode);
3295         if (!no_time_update)
3296                 inode->i_mtime = inode->i_ctime = current_time(inode);
3297         /*
3298          * We round up to the block size at eof when determining which
3299          * extents to clone above, but shouldn't round up the file size.
3300          */
3301         if (endoff > destoff + olen)
3302                 endoff = destoff + olen;
3303         if (endoff > inode->i_size)
3304                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3305
3306         ret = btrfs_update_inode(trans, root, inode);
3307         if (ret) {
3308                 btrfs_abort_transaction(trans, ret);
3309                 btrfs_end_transaction(trans);
3310                 goto out;
3311         }
3312         ret = btrfs_end_transaction(trans);
3313 out:
3314         return ret;
3315 }
3316
3317 static void clone_update_extent_map(struct btrfs_inode *inode,
3318                                     const struct btrfs_trans_handle *trans,
3319                                     const struct btrfs_path *path,
3320                                     const u64 hole_offset,
3321                                     const u64 hole_len)
3322 {
3323         struct extent_map_tree *em_tree = &inode->extent_tree;
3324         struct extent_map *em;
3325         int ret;
3326
3327         em = alloc_extent_map();
3328         if (!em) {
3329                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3330                 return;
3331         }
3332
3333         if (path) {
3334                 struct btrfs_file_extent_item *fi;
3335
3336                 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3337                                     struct btrfs_file_extent_item);
3338                 btrfs_extent_item_to_extent_map(inode, path, fi, false, em);
3339                 em->generation = -1;
3340                 if (btrfs_file_extent_type(path->nodes[0], fi) ==
3341                     BTRFS_FILE_EXTENT_INLINE)
3342                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3343                                         &inode->runtime_flags);
3344         } else {
3345                 em->start = hole_offset;
3346                 em->len = hole_len;
3347                 em->ram_bytes = em->len;
3348                 em->orig_start = hole_offset;
3349                 em->block_start = EXTENT_MAP_HOLE;
3350                 em->block_len = 0;
3351                 em->orig_block_len = 0;
3352                 em->compress_type = BTRFS_COMPRESS_NONE;
3353                 em->generation = trans->transid;
3354         }
3355
3356         while (1) {
3357                 write_lock(&em_tree->lock);
3358                 ret = add_extent_mapping(em_tree, em, 1);
3359                 write_unlock(&em_tree->lock);
3360                 if (ret != -EEXIST) {
3361                         free_extent_map(em);
3362                         break;
3363                 }
3364                 btrfs_drop_extent_cache(inode, em->start,
3365                                         em->start + em->len - 1, 0);
3366         }
3367
3368         if (ret)
3369                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
3370 }
3371
3372 /*
3373  * Make sure we do not end up inserting an inline extent into a file that has
3374  * already other (non-inline) extents. If a file has an inline extent it can
3375  * not have any other extents and the (single) inline extent must start at the
3376  * file offset 0. Failing to respect these rules will lead to file corruption,
3377  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3378  *
3379  * We can have extents that have been already written to disk or we can have
3380  * dirty ranges still in delalloc, in which case the extent maps and items are
3381  * created only when we run delalloc, and the delalloc ranges might fall outside
3382  * the range we are currently locking in the inode's io tree. So we check the
3383  * inode's i_size because of that (i_size updates are done while holding the
3384  * i_mutex, which we are holding here).
3385  * We also check to see if the inode has a size not greater than "datal" but has
3386  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3387  * protected against such concurrent fallocate calls by the i_mutex).
3388  *
3389  * If the file has no extents but a size greater than datal, do not allow the
3390  * copy because we would need turn the inline extent into a non-inline one (even
3391  * with NO_HOLES enabled). If we find our destination inode only has one inline
3392  * extent, just overwrite it with the source inline extent if its size is less
3393  * than the source extent's size, or we could copy the source inline extent's
3394  * data into the destination inode's inline extent if the later is greater then
3395  * the former.
3396  */
3397 static int clone_copy_inline_extent(struct inode *dst,
3398                                     struct btrfs_trans_handle *trans,
3399                                     struct btrfs_path *path,
3400                                     struct btrfs_key *new_key,
3401                                     const u64 drop_start,
3402                                     const u64 datal,
3403                                     const u64 skip,
3404                                     const u64 size,
3405                                     char *inline_data)
3406 {
3407         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3408         struct btrfs_root *root = BTRFS_I(dst)->root;
3409         const u64 aligned_end = ALIGN(new_key->offset + datal,
3410                                       fs_info->sectorsize);
3411         int ret;
3412         struct btrfs_key key;
3413
3414         if (new_key->offset > 0)
3415                 return -EOPNOTSUPP;
3416
3417         key.objectid = btrfs_ino(BTRFS_I(dst));
3418         key.type = BTRFS_EXTENT_DATA_KEY;
3419         key.offset = 0;
3420         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3421         if (ret < 0) {
3422                 return ret;
3423         } else if (ret > 0) {
3424                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3425                         ret = btrfs_next_leaf(root, path);
3426                         if (ret < 0)
3427                                 return ret;
3428                         else if (ret > 0)
3429                                 goto copy_inline_extent;
3430                 }
3431                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3432                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3433                     key.type == BTRFS_EXTENT_DATA_KEY) {
3434                         ASSERT(key.offset > 0);
3435                         return -EOPNOTSUPP;
3436                 }
3437         } else if (i_size_read(dst) <= datal) {
3438                 struct btrfs_file_extent_item *ei;
3439                 u64 ext_len;
3440
3441                 /*
3442                  * If the file size is <= datal, make sure there are no other
3443                  * extents following (can happen do to an fallocate call with
3444                  * the flag FALLOC_FL_KEEP_SIZE).
3445                  */
3446                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3447                                     struct btrfs_file_extent_item);
3448                 /*
3449                  * If it's an inline extent, it can not have other extents
3450                  * following it.
3451                  */
3452                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3453                     BTRFS_FILE_EXTENT_INLINE)
3454                         goto copy_inline_extent;
3455
3456                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3457                 if (ext_len > aligned_end)
3458                         return -EOPNOTSUPP;
3459
3460                 ret = btrfs_next_item(root, path);
3461                 if (ret < 0) {
3462                         return ret;
3463                 } else if (ret == 0) {
3464                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3465                                               path->slots[0]);
3466                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3467                             key.type == BTRFS_EXTENT_DATA_KEY)
3468                                 return -EOPNOTSUPP;
3469                 }
3470         }
3471
3472 copy_inline_extent:
3473         /*
3474          * We have no extent items, or we have an extent at offset 0 which may
3475          * or may not be inlined. All these cases are dealt the same way.
3476          */
3477         if (i_size_read(dst) > datal) {
3478                 /*
3479                  * If the destination inode has an inline extent...
3480                  * This would require copying the data from the source inline
3481                  * extent into the beginning of the destination's inline extent.
3482                  * But this is really complex, both extents can be compressed
3483                  * or just one of them, which would require decompressing and
3484                  * re-compressing data (which could increase the new compressed
3485                  * size, not allowing the compressed data to fit anymore in an
3486                  * inline extent).
3487                  * So just don't support this case for now (it should be rare,
3488                  * we are not really saving space when cloning inline extents).
3489                  */
3490                 return -EOPNOTSUPP;
3491         }
3492
3493         btrfs_release_path(path);
3494         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3495         if (ret)
3496                 return ret;
3497         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3498         if (ret)
3499                 return ret;
3500
3501         if (skip) {
3502                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3503
3504                 memmove(inline_data + start, inline_data + start + skip, datal);
3505         }
3506
3507         write_extent_buffer(path->nodes[0], inline_data,
3508                             btrfs_item_ptr_offset(path->nodes[0],
3509                                                   path->slots[0]),
3510                             size);
3511         inode_add_bytes(dst, datal);
3512
3513         return 0;
3514 }
3515
3516 /**
3517  * btrfs_clone() - clone a range from inode file to another
3518  *
3519  * @src: Inode to clone from
3520  * @inode: Inode to clone to
3521  * @off: Offset within source to start clone from
3522  * @olen: Original length, passed by user, of range to clone
3523  * @olen_aligned: Block-aligned value of olen
3524  * @destoff: Offset within @inode to start clone
3525  * @no_time_update: Whether to update mtime/ctime on the target inode
3526  */
3527 static int btrfs_clone(struct inode *src, struct inode *inode,
3528                        const u64 off, const u64 olen, const u64 olen_aligned,
3529                        const u64 destoff, int no_time_update)
3530 {
3531         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3532         struct btrfs_root *root = BTRFS_I(inode)->root;
3533         struct btrfs_path *path = NULL;
3534         struct extent_buffer *leaf;
3535         struct btrfs_trans_handle *trans;
3536         char *buf = NULL;
3537         struct btrfs_key key;
3538         u32 nritems;
3539         int slot;
3540         int ret;
3541         const u64 len = olen_aligned;
3542         u64 last_dest_end = destoff;
3543
3544         ret = -ENOMEM;
3545         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3546         if (!buf)
3547                 return ret;
3548
3549         path = btrfs_alloc_path();
3550         if (!path) {
3551                 kvfree(buf);
3552                 return ret;
3553         }
3554
3555         path->reada = READA_FORWARD;
3556         /* clone data */
3557         key.objectid = btrfs_ino(BTRFS_I(src));
3558         key.type = BTRFS_EXTENT_DATA_KEY;
3559         key.offset = off;
3560
3561         while (1) {
3562                 u64 next_key_min_offset = key.offset + 1;
3563
3564                 /*
3565                  * note the key will change type as we walk through the
3566                  * tree.
3567                  */
3568                 path->leave_spinning = 1;
3569                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3570                                 0, 0);
3571                 if (ret < 0)
3572                         goto out;
3573                 /*
3574                  * First search, if no extent item that starts at offset off was
3575                  * found but the previous item is an extent item, it's possible
3576                  * it might overlap our target range, therefore process it.
3577                  */
3578                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3579                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3580                                               path->slots[0] - 1);
3581                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3582                                 path->slots[0]--;
3583                 }
3584
3585                 nritems = btrfs_header_nritems(path->nodes[0]);
3586 process_slot:
3587                 if (path->slots[0] >= nritems) {
3588                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3589                         if (ret < 0)
3590                                 goto out;
3591                         if (ret > 0)
3592                                 break;
3593                         nritems = btrfs_header_nritems(path->nodes[0]);
3594                 }
3595                 leaf = path->nodes[0];
3596                 slot = path->slots[0];
3597
3598                 btrfs_item_key_to_cpu(leaf, &key, slot);
3599                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3600                     key.objectid != btrfs_ino(BTRFS_I(src)))
3601                         break;
3602
3603                 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3604                         struct btrfs_file_extent_item *extent;
3605                         int type;
3606                         u32 size;
3607                         struct btrfs_key new_key;
3608                         u64 disko = 0, diskl = 0;
3609                         u64 datao = 0, datal = 0;
3610                         u8 comp;
3611                         u64 drop_start;
3612
3613                         extent = btrfs_item_ptr(leaf, slot,
3614                                                 struct btrfs_file_extent_item);
3615                         comp = btrfs_file_extent_compression(leaf, extent);
3616                         type = btrfs_file_extent_type(leaf, extent);
3617                         if (type == BTRFS_FILE_EXTENT_REG ||
3618                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3619                                 disko = btrfs_file_extent_disk_bytenr(leaf,
3620                                                                       extent);
3621                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
3622                                                                  extent);
3623                                 datao = btrfs_file_extent_offset(leaf, extent);
3624                                 datal = btrfs_file_extent_num_bytes(leaf,
3625                                                                     extent);
3626                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3627                                 /* take upper bound, may be compressed */
3628                                 datal = btrfs_file_extent_ram_bytes(leaf,
3629                                                                     extent);
3630                         }
3631
3632                         /*
3633                          * The first search might have left us at an extent
3634                          * item that ends before our target range's start, can
3635                          * happen if we have holes and NO_HOLES feature enabled.
3636                          */
3637                         if (key.offset + datal <= off) {
3638                                 path->slots[0]++;
3639                                 goto process_slot;
3640                         } else if (key.offset >= off + len) {
3641                                 break;
3642                         }
3643                         next_key_min_offset = key.offset + datal;
3644                         size = btrfs_item_size_nr(leaf, slot);
3645                         read_extent_buffer(leaf, buf,
3646                                            btrfs_item_ptr_offset(leaf, slot),
3647                                            size);
3648
3649                         btrfs_release_path(path);
3650                         path->leave_spinning = 0;
3651
3652                         memcpy(&new_key, &key, sizeof(new_key));
3653                         new_key.objectid = btrfs_ino(BTRFS_I(inode));
3654                         if (off <= key.offset)
3655                                 new_key.offset = key.offset + destoff - off;
3656                         else
3657                                 new_key.offset = destoff;
3658
3659                         /*
3660                          * Deal with a hole that doesn't have an extent item
3661                          * that represents it (NO_HOLES feature enabled).
3662                          * This hole is either in the middle of the cloning
3663                          * range or at the beginning (fully overlaps it or
3664                          * partially overlaps it).
3665                          */
3666                         if (new_key.offset != last_dest_end)
3667                                 drop_start = last_dest_end;
3668                         else
3669                                 drop_start = new_key.offset;
3670
3671                         /*
3672                          * 1 - adjusting old extent (we may have to split it)
3673                          * 1 - add new extent
3674                          * 1 - inode update
3675                          */
3676                         trans = btrfs_start_transaction(root, 3);
3677                         if (IS_ERR(trans)) {
3678                                 ret = PTR_ERR(trans);
3679                                 goto out;
3680                         }
3681
3682                         if (type == BTRFS_FILE_EXTENT_REG ||
3683                             type == BTRFS_FILE_EXTENT_PREALLOC) {
3684                                 /*
3685                                  *    a  | --- range to clone ---|  b
3686                                  * | ------------- extent ------------- |
3687                                  */
3688
3689                                 /* subtract range b */
3690                                 if (key.offset + datal > off + len)
3691                                         datal = off + len - key.offset;
3692
3693                                 /* subtract range a */
3694                                 if (off > key.offset) {
3695                                         datao += off - key.offset;
3696                                         datal -= off - key.offset;
3697                                 }
3698
3699                                 ret = btrfs_drop_extents(trans, root, inode,
3700                                                          drop_start,
3701                                                          new_key.offset + datal,
3702                                                          1);
3703                                 if (ret) {
3704                                         if (ret != -EOPNOTSUPP)
3705                                                 btrfs_abort_transaction(trans,
3706                                                                         ret);
3707                                         btrfs_end_transaction(trans);
3708                                         goto out;
3709                                 }
3710
3711                                 ret = btrfs_insert_empty_item(trans, root, path,
3712                                                               &new_key, size);
3713                                 if (ret) {
3714                                         btrfs_abort_transaction(trans, ret);
3715                                         btrfs_end_transaction(trans);
3716                                         goto out;
3717                                 }
3718
3719                                 leaf = path->nodes[0];
3720                                 slot = path->slots[0];
3721                                 write_extent_buffer(leaf, buf,
3722                                             btrfs_item_ptr_offset(leaf, slot),
3723                                             size);
3724
3725                                 extent = btrfs_item_ptr(leaf, slot,
3726                                                 struct btrfs_file_extent_item);
3727
3728                                 /* disko == 0 means it's a hole */
3729                                 if (!disko)
3730                                         datao = 0;
3731
3732                                 btrfs_set_file_extent_offset(leaf, extent,
3733                                                              datao);
3734                                 btrfs_set_file_extent_num_bytes(leaf, extent,
3735                                                                 datal);
3736
3737                                 if (disko) {
3738                                         inode_add_bytes(inode, datal);
3739                                         ret = btrfs_inc_extent_ref(trans,
3740                                                         root,
3741                                                         disko, diskl, 0,
3742                                                         root->root_key.objectid,
3743                                                         btrfs_ino(BTRFS_I(inode)),
3744                                                         new_key.offset - datao);
3745                                         if (ret) {
3746                                                 btrfs_abort_transaction(trans,
3747                                                                         ret);
3748                                                 btrfs_end_transaction(trans);
3749                                                 goto out;
3750
3751                                         }
3752                                 }
3753                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3754                                 u64 skip = 0;
3755                                 u64 trim = 0;
3756
3757                                 if (off > key.offset) {
3758                                         skip = off - key.offset;
3759                                         new_key.offset += skip;
3760                                 }
3761
3762                                 if (key.offset + datal > off + len)
3763                                         trim = key.offset + datal - (off + len);
3764
3765                                 if (comp && (skip || trim)) {
3766                                         ret = -EINVAL;
3767                                         btrfs_end_transaction(trans);
3768                                         goto out;
3769                                 }
3770                                 size -= skip + trim;
3771                                 datal -= skip + trim;
3772
3773                                 ret = clone_copy_inline_extent(inode,
3774                                                                trans, path,
3775                                                                &new_key,
3776                                                                drop_start,
3777                                                                datal,
3778                                                                skip, size, buf);
3779                                 if (ret) {
3780                                         if (ret != -EOPNOTSUPP)
3781                                                 btrfs_abort_transaction(trans,
3782                                                                         ret);
3783                                         btrfs_end_transaction(trans);
3784                                         goto out;
3785                                 }
3786                                 leaf = path->nodes[0];
3787                                 slot = path->slots[0];
3788                         }
3789
3790                         /* If we have an implicit hole (NO_HOLES feature). */
3791                         if (drop_start < new_key.offset)
3792                                 clone_update_extent_map(BTRFS_I(inode), trans,
3793                                                 NULL, drop_start,
3794                                                 new_key.offset - drop_start);
3795
3796                         clone_update_extent_map(BTRFS_I(inode), trans,
3797                                         path, 0, 0);
3798
3799                         btrfs_mark_buffer_dirty(leaf);
3800                         btrfs_release_path(path);
3801
3802                         last_dest_end = ALIGN(new_key.offset + datal,
3803                                               fs_info->sectorsize);
3804                         ret = clone_finish_inode_update(trans, inode,
3805                                                         last_dest_end,
3806                                                         destoff, olen,
3807                                                         no_time_update);
3808                         if (ret)
3809                                 goto out;
3810                         if (new_key.offset + datal >= destoff + len)
3811                                 break;
3812                 }
3813                 btrfs_release_path(path);
3814                 key.offset = next_key_min_offset;
3815
3816                 if (fatal_signal_pending(current)) {
3817                         ret = -EINTR;
3818                         goto out;
3819                 }
3820         }
3821         ret = 0;
3822
3823         if (last_dest_end < destoff + len) {
3824                 /*
3825                  * We have an implicit hole (NO_HOLES feature is enabled) that
3826                  * fully or partially overlaps our cloning range at its end.
3827                  */
3828                 btrfs_release_path(path);
3829
3830                 /*
3831                  * 1 - remove extent(s)
3832                  * 1 - inode update
3833                  */
3834                 trans = btrfs_start_transaction(root, 2);
3835                 if (IS_ERR(trans)) {
3836                         ret = PTR_ERR(trans);
3837                         goto out;
3838                 }
3839                 ret = btrfs_drop_extents(trans, root, inode,
3840                                          last_dest_end, destoff + len, 1);
3841                 if (ret) {
3842                         if (ret != -EOPNOTSUPP)
3843                                 btrfs_abort_transaction(trans, ret);
3844                         btrfs_end_transaction(trans);
3845                         goto out;
3846                 }
3847                 clone_update_extent_map(BTRFS_I(inode), trans, NULL,
3848                                 last_dest_end,
3849                                 destoff + len - last_dest_end);
3850                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3851                                                 destoff, olen, no_time_update);
3852         }
3853
3854 out:
3855         btrfs_free_path(path);
3856         kvfree(buf);
3857         return ret;
3858 }
3859
3860 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3861                                         u64 off, u64 olen, u64 destoff)
3862 {
3863         struct inode *inode = file_inode(file);
3864         struct inode *src = file_inode(file_src);
3865         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3866         int ret;
3867         u64 len = olen;
3868         u64 bs = fs_info->sb->s_blocksize;
3869
3870         /*
3871          * TODO:
3872          * - split compressed inline extents.  annoying: we need to
3873          *   decompress into destination's address_space (the file offset
3874          *   may change, so source mapping won't do), then recompress (or
3875          *   otherwise reinsert) a subrange.
3876          *
3877          * - split destination inode's inline extents.  The inline extents can
3878          *   be either compressed or non-compressed.
3879          */
3880
3881         /*
3882          * VFS's generic_remap_file_range_prep() protects us from cloning the
3883          * eof block into the middle of a file, which would result in corruption
3884          * if the file size is not blocksize aligned. So we don't need to check
3885          * for that case here.
3886          */
3887         if (off + len == src->i_size)
3888                 len = ALIGN(src->i_size, bs) - off;
3889
3890         if (destoff > inode->i_size) {
3891                 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
3892
3893                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3894                 if (ret)
3895                         return ret;
3896                 /*
3897                  * We may have truncated the last block if the inode's size is
3898                  * not sector size aligned, so we need to wait for writeback to
3899                  * complete before proceeding further, otherwise we can race
3900                  * with cloning and attempt to increment a reference to an
3901                  * extent that no longer exists (writeback completed right after
3902                  * we found the previous extent covering eof and before we
3903                  * attempted to increment its reference count).
3904                  */
3905                 ret = btrfs_wait_ordered_range(inode, wb_start,
3906                                                destoff - wb_start);
3907                 if (ret)
3908                         return ret;
3909         }
3910
3911         /*
3912          * Lock destination range to serialize with concurrent readpages() and
3913          * source range to serialize with relocation.
3914          */
3915         btrfs_double_extent_lock(src, off, inode, destoff, len);
3916         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3917         btrfs_double_extent_unlock(src, off, inode, destoff, len);
3918         /*
3919          * Truncate page cache pages so that future reads will see the cloned
3920          * data immediately and not the previous data.
3921          */
3922         truncate_inode_pages_range(&inode->i_data,
3923                                 round_down(destoff, PAGE_SIZE),
3924                                 round_up(destoff + len, PAGE_SIZE) - 1);
3925
3926         return ret;
3927 }
3928
3929 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
3930                                        struct file *file_out, loff_t pos_out,
3931                                        loff_t *len, unsigned int remap_flags)
3932 {
3933         struct inode *inode_in = file_inode(file_in);
3934         struct inode *inode_out = file_inode(file_out);
3935         u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
3936         bool same_inode = inode_out == inode_in;
3937         u64 wb_len;
3938         int ret;
3939
3940         if (!(remap_flags & REMAP_FILE_DEDUP)) {
3941                 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
3942
3943                 if (btrfs_root_readonly(root_out))
3944                         return -EROFS;
3945
3946                 if (file_in->f_path.mnt != file_out->f_path.mnt ||
3947                     inode_in->i_sb != inode_out->i_sb)
3948                         return -EXDEV;
3949         }
3950
3951         if (same_inode)
3952                 inode_lock(inode_in);
3953         else
3954                 lock_two_nondirectories(inode_in, inode_out);
3955
3956         /* don't make the dst file partly checksummed */
3957         if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
3958             (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
3959                 ret = -EINVAL;
3960                 goto out_unlock;
3961         }
3962
3963         /*
3964          * Now that the inodes are locked, we need to start writeback ourselves
3965          * and can not rely on the writeback from the VFS's generic helper
3966          * generic_remap_file_range_prep() because:
3967          *
3968          * 1) For compression we must call filemap_fdatawrite_range() range
3969          *    twice (btrfs_fdatawrite_range() does it for us), and the generic
3970          *    helper only calls it once;
3971          *
3972          * 2) filemap_fdatawrite_range(), called by the generic helper only
3973          *    waits for the writeback to complete, i.e. for IO to be done, and
3974          *    not for the ordered extents to complete. We need to wait for them
3975          *    to complete so that new file extent items are in the fs tree.
3976          */
3977         if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
3978                 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
3979         else
3980                 wb_len = ALIGN(*len, bs);
3981
3982         /*
3983          * Since we don't lock ranges, wait for ongoing lockless dio writes (as
3984          * any in progress could create its ordered extents after we wait for
3985          * existing ordered extents below).
3986          */
3987         inode_dio_wait(inode_in);
3988         if (!same_inode)
3989                 inode_dio_wait(inode_out);
3990
3991         ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
3992                                        wb_len);
3993         if (ret < 0)
3994                 goto out_unlock;
3995         ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
3996                                        wb_len);
3997         if (ret < 0)
3998                 goto out_unlock;
3999
4000         ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
4001                                             len, remap_flags);
4002         if (ret < 0 || *len == 0)
4003                 goto out_unlock;
4004
4005         return 0;
4006
4007  out_unlock:
4008         if (same_inode)
4009                 inode_unlock(inode_in);
4010         else
4011                 unlock_two_nondirectories(inode_in, inode_out);
4012
4013         return ret;
4014 }
4015
4016 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
4017                 struct file *dst_file, loff_t destoff, loff_t len,
4018                 unsigned int remap_flags)
4019 {
4020         struct inode *src_inode = file_inode(src_file);
4021         struct inode *dst_inode = file_inode(dst_file);
4022         bool same_inode = dst_inode == src_inode;
4023         int ret;
4024
4025         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
4026                 return -EINVAL;
4027
4028         ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
4029                                           &len, remap_flags);
4030         if (ret < 0 || len == 0)
4031                 return ret;
4032
4033         if (remap_flags & REMAP_FILE_DEDUP)
4034                 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
4035         else
4036                 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
4037
4038         if (same_inode)
4039                 inode_unlock(src_inode);
4040         else
4041                 unlock_two_nondirectories(src_inode, dst_inode);
4042
4043         return ret < 0 ? ret : len;
4044 }
4045
4046 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4047 {
4048         struct inode *inode = file_inode(file);
4049         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4050         struct btrfs_root *root = BTRFS_I(inode)->root;
4051         struct btrfs_root *new_root;
4052         struct btrfs_dir_item *di;
4053         struct btrfs_trans_handle *trans;
4054         struct btrfs_path *path;
4055         struct btrfs_key location;
4056         struct btrfs_disk_key disk_key;
4057         u64 objectid = 0;
4058         u64 dir_id;
4059         int ret;
4060
4061         if (!capable(CAP_SYS_ADMIN))
4062                 return -EPERM;
4063
4064         ret = mnt_want_write_file(file);
4065         if (ret)
4066                 return ret;
4067
4068         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4069                 ret = -EFAULT;
4070                 goto out;
4071         }
4072
4073         if (!objectid)
4074                 objectid = BTRFS_FS_TREE_OBJECTID;
4075
4076         location.objectid = objectid;
4077         location.type = BTRFS_ROOT_ITEM_KEY;
4078         location.offset = (u64)-1;
4079
4080         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4081         if (IS_ERR(new_root)) {
4082                 ret = PTR_ERR(new_root);
4083                 goto out;
4084         }
4085         if (!is_fstree(new_root->root_key.objectid)) {
4086                 ret = -ENOENT;
4087                 goto out;
4088         }
4089
4090         path = btrfs_alloc_path();
4091         if (!path) {
4092                 ret = -ENOMEM;
4093                 goto out;
4094         }
4095         path->leave_spinning = 1;
4096
4097         trans = btrfs_start_transaction(root, 1);
4098         if (IS_ERR(trans)) {
4099                 btrfs_free_path(path);
4100                 ret = PTR_ERR(trans);
4101                 goto out;
4102         }
4103
4104         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4105         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4106                                    dir_id, "default", 7, 1);
4107         if (IS_ERR_OR_NULL(di)) {
4108                 btrfs_free_path(path);
4109                 btrfs_end_transaction(trans);
4110                 btrfs_err(fs_info,
4111                           "Umm, you don't have the default diritem, this isn't going to work");
4112                 ret = -ENOENT;
4113                 goto out;
4114         }
4115
4116         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4117         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4118         btrfs_mark_buffer_dirty(path->nodes[0]);
4119         btrfs_free_path(path);
4120
4121         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4122         btrfs_end_transaction(trans);
4123 out:
4124         mnt_drop_write_file(file);
4125         return ret;
4126 }
4127
4128 static void get_block_group_info(struct list_head *groups_list,
4129                                  struct btrfs_ioctl_space_info *space)
4130 {
4131         struct btrfs_block_group_cache *block_group;
4132
4133         space->total_bytes = 0;
4134         space->used_bytes = 0;
4135         space->flags = 0;
4136         list_for_each_entry(block_group, groups_list, list) {
4137                 space->flags = block_group->flags;
4138                 space->total_bytes += block_group->key.offset;
4139                 space->used_bytes +=
4140                         btrfs_block_group_used(&block_group->item);
4141         }
4142 }
4143
4144 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4145                                    void __user *arg)
4146 {
4147         struct btrfs_ioctl_space_args space_args;
4148         struct btrfs_ioctl_space_info space;
4149         struct btrfs_ioctl_space_info *dest;
4150         struct btrfs_ioctl_space_info *dest_orig;
4151         struct btrfs_ioctl_space_info __user *user_dest;
4152         struct btrfs_space_info *info;
4153         static const u64 types[] = {
4154                 BTRFS_BLOCK_GROUP_DATA,
4155                 BTRFS_BLOCK_GROUP_SYSTEM,
4156                 BTRFS_BLOCK_GROUP_METADATA,
4157                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4158         };
4159         int num_types = 4;
4160         int alloc_size;
4161         int ret = 0;
4162         u64 slot_count = 0;
4163         int i, c;
4164
4165         if (copy_from_user(&space_args,
4166                            (struct btrfs_ioctl_space_args __user *)arg,
4167                            sizeof(space_args)))
4168                 return -EFAULT;
4169
4170         for (i = 0; i < num_types; i++) {
4171                 struct btrfs_space_info *tmp;
4172
4173                 info = NULL;
4174                 rcu_read_lock();
4175                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4176                                         list) {
4177                         if (tmp->flags == types[i]) {
4178                                 info = tmp;
4179                                 break;
4180                         }
4181                 }
4182                 rcu_read_unlock();
4183
4184                 if (!info)
4185                         continue;
4186
4187                 down_read(&info->groups_sem);
4188                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4189                         if (!list_empty(&info->block_groups[c]))
4190                                 slot_count++;
4191                 }
4192                 up_read(&info->groups_sem);
4193         }
4194
4195         /*
4196          * Global block reserve, exported as a space_info
4197          */
4198         slot_count++;
4199
4200         /* space_slots == 0 means they are asking for a count */
4201         if (space_args.space_slots == 0) {
4202                 space_args.total_spaces = slot_count;
4203                 goto out;
4204         }
4205
4206         slot_count = min_t(u64, space_args.space_slots, slot_count);
4207
4208         alloc_size = sizeof(*dest) * slot_count;
4209
4210         /* we generally have at most 6 or so space infos, one for each raid
4211          * level.  So, a whole page should be more than enough for everyone
4212          */
4213         if (alloc_size > PAGE_SIZE)
4214                 return -ENOMEM;
4215
4216         space_args.total_spaces = 0;
4217         dest = kmalloc(alloc_size, GFP_KERNEL);
4218         if (!dest)
4219                 return -ENOMEM;
4220         dest_orig = dest;
4221
4222         /* now we have a buffer to copy into */
4223         for (i = 0; i < num_types; i++) {
4224                 struct btrfs_space_info *tmp;
4225
4226                 if (!slot_count)
4227                         break;
4228
4229                 info = NULL;
4230                 rcu_read_lock();
4231                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4232                                         list) {
4233                         if (tmp->flags == types[i]) {
4234                                 info = tmp;
4235                                 break;
4236                         }
4237                 }
4238                 rcu_read_unlock();
4239
4240                 if (!info)
4241                         continue;
4242                 down_read(&info->groups_sem);
4243                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4244                         if (!list_empty(&info->block_groups[c])) {
4245                                 get_block_group_info(&info->block_groups[c],
4246                                                      &space);
4247                                 memcpy(dest, &space, sizeof(space));
4248                                 dest++;
4249                                 space_args.total_spaces++;
4250                                 slot_count--;
4251                         }
4252                         if (!slot_count)
4253                                 break;
4254                 }
4255                 up_read(&info->groups_sem);
4256         }
4257
4258         /*
4259          * Add global block reserve
4260          */
4261         if (slot_count) {
4262                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4263
4264                 spin_lock(&block_rsv->lock);
4265                 space.total_bytes = block_rsv->size;
4266                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4267                 spin_unlock(&block_rsv->lock);
4268                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4269                 memcpy(dest, &space, sizeof(space));
4270                 space_args.total_spaces++;
4271         }
4272
4273         user_dest = (struct btrfs_ioctl_space_info __user *)
4274                 (arg + sizeof(struct btrfs_ioctl_space_args));
4275
4276         if (copy_to_user(user_dest, dest_orig, alloc_size))
4277                 ret = -EFAULT;
4278
4279         kfree(dest_orig);
4280 out:
4281         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4282                 ret = -EFAULT;
4283
4284         return ret;
4285 }
4286
4287 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4288                                             void __user *argp)
4289 {
4290         struct btrfs_trans_handle *trans;
4291         u64 transid;
4292         int ret;
4293
4294         trans = btrfs_attach_transaction_barrier(root);
4295         if (IS_ERR(trans)) {
4296                 if (PTR_ERR(trans) != -ENOENT)
4297                         return PTR_ERR(trans);
4298
4299                 /* No running transaction, don't bother */
4300                 transid = root->fs_info->last_trans_committed;
4301                 goto out;
4302         }
4303         transid = trans->transid;
4304         ret = btrfs_commit_transaction_async(trans, 0);
4305         if (ret) {
4306                 btrfs_end_transaction(trans);
4307                 return ret;
4308         }
4309 out:
4310         if (argp)
4311                 if (copy_to_user(argp, &transid, sizeof(transid)))
4312                         return -EFAULT;
4313         return 0;
4314 }
4315
4316 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4317                                            void __user *argp)
4318 {
4319         u64 transid;
4320
4321         if (argp) {
4322                 if (copy_from_user(&transid, argp, sizeof(transid)))
4323                         return -EFAULT;
4324         } else {
4325                 transid = 0;  /* current trans */
4326         }
4327         return btrfs_wait_for_commit(fs_info, transid);
4328 }
4329
4330 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4331 {
4332         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4333         struct btrfs_ioctl_scrub_args *sa;
4334         int ret;
4335
4336         if (!capable(CAP_SYS_ADMIN))
4337                 return -EPERM;
4338
4339         sa = memdup_user(arg, sizeof(*sa));
4340         if (IS_ERR(sa))
4341                 return PTR_ERR(sa);
4342
4343         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4344                 ret = mnt_want_write_file(file);
4345                 if (ret)
4346                         goto out;
4347         }
4348
4349         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4350                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4351                               0);
4352
4353         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4354                 ret = -EFAULT;
4355
4356         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4357                 mnt_drop_write_file(file);
4358 out:
4359         kfree(sa);
4360         return ret;
4361 }
4362
4363 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4364 {
4365         if (!capable(CAP_SYS_ADMIN))
4366                 return -EPERM;
4367
4368         return btrfs_scrub_cancel(fs_info);
4369 }
4370
4371 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4372                                        void __user *arg)
4373 {
4374         struct btrfs_ioctl_scrub_args *sa;
4375         int ret;
4376
4377         if (!capable(CAP_SYS_ADMIN))
4378                 return -EPERM;
4379
4380         sa = memdup_user(arg, sizeof(*sa));
4381         if (IS_ERR(sa))
4382                 return PTR_ERR(sa);
4383
4384         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4385
4386         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4387                 ret = -EFAULT;
4388
4389         kfree(sa);
4390         return ret;
4391 }
4392
4393 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4394                                       void __user *arg)
4395 {
4396         struct btrfs_ioctl_get_dev_stats *sa;
4397         int ret;
4398
4399         sa = memdup_user(arg, sizeof(*sa));
4400         if (IS_ERR(sa))
4401                 return PTR_ERR(sa);
4402
4403         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4404                 kfree(sa);
4405                 return -EPERM;
4406         }
4407
4408         ret = btrfs_get_dev_stats(fs_info, sa);
4409
4410         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4411                 ret = -EFAULT;
4412
4413         kfree(sa);
4414         return ret;
4415 }
4416
4417 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4418                                     void __user *arg)
4419 {
4420         struct btrfs_ioctl_dev_replace_args *p;
4421         int ret;
4422
4423         if (!capable(CAP_SYS_ADMIN))
4424                 return -EPERM;
4425
4426         p = memdup_user(arg, sizeof(*p));
4427         if (IS_ERR(p))
4428                 return PTR_ERR(p);
4429
4430         switch (p->cmd) {
4431         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4432                 if (sb_rdonly(fs_info->sb)) {
4433                         ret = -EROFS;
4434                         goto out;
4435                 }
4436                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4437                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4438                 } else {
4439                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4440                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4441                 }
4442                 break;
4443         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4444                 btrfs_dev_replace_status(fs_info, p);
4445                 ret = 0;
4446                 break;
4447         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4448                 p->result = btrfs_dev_replace_cancel(fs_info);
4449                 ret = 0;
4450                 break;
4451         default:
4452                 ret = -EINVAL;
4453                 break;
4454         }
4455
4456         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4457                 ret = -EFAULT;
4458 out:
4459         kfree(p);
4460         return ret;
4461 }
4462
4463 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4464 {
4465         int ret = 0;
4466         int i;
4467         u64 rel_ptr;
4468         int size;
4469         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4470         struct inode_fs_paths *ipath = NULL;
4471         struct btrfs_path *path;
4472
4473         if (!capable(CAP_DAC_READ_SEARCH))
4474                 return -EPERM;
4475
4476         path = btrfs_alloc_path();
4477         if (!path) {
4478                 ret = -ENOMEM;
4479                 goto out;
4480         }
4481
4482         ipa = memdup_user(arg, sizeof(*ipa));
4483         if (IS_ERR(ipa)) {
4484                 ret = PTR_ERR(ipa);
4485                 ipa = NULL;
4486                 goto out;
4487         }
4488
4489         size = min_t(u32, ipa->size, 4096);
4490         ipath = init_ipath(size, root, path);
4491         if (IS_ERR(ipath)) {
4492                 ret = PTR_ERR(ipath);
4493                 ipath = NULL;
4494                 goto out;
4495         }
4496
4497         ret = paths_from_inode(ipa->inum, ipath);
4498         if (ret < 0)
4499                 goto out;
4500
4501         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4502                 rel_ptr = ipath->fspath->val[i] -
4503                           (u64)(unsigned long)ipath->fspath->val;
4504                 ipath->fspath->val[i] = rel_ptr;
4505         }
4506
4507         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4508                            ipath->fspath, size);
4509         if (ret) {
4510                 ret = -EFAULT;
4511                 goto out;
4512         }
4513
4514 out:
4515         btrfs_free_path(path);
4516         free_ipath(ipath);
4517         kfree(ipa);
4518
4519         return ret;
4520 }
4521
4522 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4523 {
4524         struct btrfs_data_container *inodes = ctx;
4525         const size_t c = 3 * sizeof(u64);
4526
4527         if (inodes->bytes_left >= c) {
4528                 inodes->bytes_left -= c;
4529                 inodes->val[inodes->elem_cnt] = inum;
4530                 inodes->val[inodes->elem_cnt + 1] = offset;
4531                 inodes->val[inodes->elem_cnt + 2] = root;
4532                 inodes->elem_cnt += 3;
4533         } else {
4534                 inodes->bytes_missing += c - inodes->bytes_left;
4535                 inodes->bytes_left = 0;
4536                 inodes->elem_missed += 3;
4537         }
4538
4539         return 0;
4540 }
4541
4542 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4543                                         void __user *arg, int version)
4544 {
4545         int ret = 0;
4546         int size;
4547         struct btrfs_ioctl_logical_ino_args *loi;
4548         struct btrfs_data_container *inodes = NULL;
4549         struct btrfs_path *path = NULL;
4550         bool ignore_offset;
4551
4552         if (!capable(CAP_SYS_ADMIN))
4553                 return -EPERM;
4554
4555         loi = memdup_user(arg, sizeof(*loi));
4556         if (IS_ERR(loi))
4557                 return PTR_ERR(loi);
4558
4559         if (version == 1) {
4560                 ignore_offset = false;
4561                 size = min_t(u32, loi->size, SZ_64K);
4562         } else {
4563                 /* All reserved bits must be 0 for now */
4564                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4565                         ret = -EINVAL;
4566                         goto out_loi;
4567                 }
4568                 /* Only accept flags we have defined so far */
4569                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4570                         ret = -EINVAL;
4571                         goto out_loi;
4572                 }
4573                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4574                 size = min_t(u32, loi->size, SZ_16M);
4575         }
4576
4577         path = btrfs_alloc_path();
4578         if (!path) {
4579                 ret = -ENOMEM;
4580                 goto out;
4581         }
4582
4583         inodes = init_data_container(size);
4584         if (IS_ERR(inodes)) {
4585                 ret = PTR_ERR(inodes);
4586                 inodes = NULL;
4587                 goto out;
4588         }
4589
4590         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4591                                           build_ino_list, inodes, ignore_offset);
4592         if (ret == -EINVAL)
4593                 ret = -ENOENT;
4594         if (ret < 0)
4595                 goto out;
4596
4597         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4598                            size);
4599         if (ret)
4600                 ret = -EFAULT;
4601
4602 out:
4603         btrfs_free_path(path);
4604         kvfree(inodes);
4605 out_loi:
4606         kfree(loi);
4607
4608         return ret;
4609 }
4610
4611 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4612                                struct btrfs_ioctl_balance_args *bargs)
4613 {
4614         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4615
4616         bargs->flags = bctl->flags;
4617
4618         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4619                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4620         if (atomic_read(&fs_info->balance_pause_req))
4621                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4622         if (atomic_read(&fs_info->balance_cancel_req))
4623                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4624
4625         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4626         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4627         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4628
4629         spin_lock(&fs_info->balance_lock);
4630         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4631         spin_unlock(&fs_info->balance_lock);
4632 }
4633
4634 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4635 {
4636         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4637         struct btrfs_fs_info *fs_info = root->fs_info;
4638         struct btrfs_ioctl_balance_args *bargs;
4639         struct btrfs_balance_control *bctl;
4640         bool need_unlock; /* for mut. excl. ops lock */
4641         int ret;
4642
4643         if (!capable(CAP_SYS_ADMIN))
4644                 return -EPERM;
4645
4646         ret = mnt_want_write_file(file);
4647         if (ret)
4648                 return ret;
4649
4650 again:
4651         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4652                 mutex_lock(&fs_info->balance_mutex);
4653                 need_unlock = true;
4654                 goto locked;
4655         }
4656
4657         /*
4658          * mut. excl. ops lock is locked.  Three possibilities:
4659          *   (1) some other op is running
4660          *   (2) balance is running
4661          *   (3) balance is paused -- special case (think resume)
4662          */
4663         mutex_lock(&fs_info->balance_mutex);
4664         if (fs_info->balance_ctl) {
4665                 /* this is either (2) or (3) */
4666                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4667                         mutex_unlock(&fs_info->balance_mutex);
4668                         /*
4669                          * Lock released to allow other waiters to continue,
4670                          * we'll reexamine the status again.
4671                          */
4672                         mutex_lock(&fs_info->balance_mutex);
4673
4674                         if (fs_info->balance_ctl &&
4675                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4676                                 /* this is (3) */
4677                                 need_unlock = false;
4678                                 goto locked;
4679                         }
4680
4681                         mutex_unlock(&fs_info->balance_mutex);
4682                         goto again;
4683                 } else {
4684                         /* this is (2) */
4685                         mutex_unlock(&fs_info->balance_mutex);
4686                         ret = -EINPROGRESS;
4687                         goto out;
4688                 }
4689         } else {
4690                 /* this is (1) */
4691                 mutex_unlock(&fs_info->balance_mutex);
4692                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4693                 goto out;
4694         }
4695
4696 locked:
4697         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4698
4699         if (arg) {
4700                 bargs = memdup_user(arg, sizeof(*bargs));
4701                 if (IS_ERR(bargs)) {
4702                         ret = PTR_ERR(bargs);
4703                         goto out_unlock;
4704                 }
4705
4706                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4707                         if (!fs_info->balance_ctl) {
4708                                 ret = -ENOTCONN;
4709                                 goto out_bargs;
4710                         }
4711
4712                         bctl = fs_info->balance_ctl;
4713                         spin_lock(&fs_info->balance_lock);
4714                         bctl->flags |= BTRFS_BALANCE_RESUME;
4715                         spin_unlock(&fs_info->balance_lock);
4716
4717                         goto do_balance;
4718                 }
4719         } else {
4720                 bargs = NULL;
4721         }
4722
4723         if (fs_info->balance_ctl) {
4724                 ret = -EINPROGRESS;
4725                 goto out_bargs;
4726         }
4727
4728         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4729         if (!bctl) {
4730                 ret = -ENOMEM;
4731                 goto out_bargs;
4732         }
4733
4734         if (arg) {
4735                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4736                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4737                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4738
4739                 bctl->flags = bargs->flags;
4740         } else {
4741                 /* balance everything - no filters */
4742                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4743         }
4744
4745         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4746                 ret = -EINVAL;
4747                 goto out_bctl;
4748         }
4749
4750 do_balance:
4751         /*
4752          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4753          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4754          * restriper was paused all the way until unmount, in free_fs_info.
4755          * The flag should be cleared after reset_balance_state.
4756          */
4757         need_unlock = false;
4758
4759         ret = btrfs_balance(fs_info, bctl, bargs);
4760         bctl = NULL;
4761
4762         if ((ret == 0 || ret == -ECANCELED) && arg) {
4763                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4764                         ret = -EFAULT;
4765         }
4766
4767 out_bctl:
4768         kfree(bctl);
4769 out_bargs:
4770         kfree(bargs);
4771 out_unlock:
4772         mutex_unlock(&fs_info->balance_mutex);
4773         if (need_unlock)
4774                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4775 out:
4776         mnt_drop_write_file(file);
4777         return ret;
4778 }
4779
4780 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4781 {
4782         if (!capable(CAP_SYS_ADMIN))
4783                 return -EPERM;
4784
4785         switch (cmd) {
4786         case BTRFS_BALANCE_CTL_PAUSE:
4787                 return btrfs_pause_balance(fs_info);
4788         case BTRFS_BALANCE_CTL_CANCEL:
4789                 return btrfs_cancel_balance(fs_info);
4790         }
4791
4792         return -EINVAL;
4793 }
4794
4795 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4796                                          void __user *arg)
4797 {
4798         struct btrfs_ioctl_balance_args *bargs;
4799         int ret = 0;
4800
4801         if (!capable(CAP_SYS_ADMIN))
4802                 return -EPERM;
4803
4804         mutex_lock(&fs_info->balance_mutex);
4805         if (!fs_info->balance_ctl) {
4806                 ret = -ENOTCONN;
4807                 goto out;
4808         }
4809
4810         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4811         if (!bargs) {
4812                 ret = -ENOMEM;
4813                 goto out;
4814         }
4815
4816         btrfs_update_ioctl_balance_args(fs_info, bargs);
4817
4818         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4819                 ret = -EFAULT;
4820
4821         kfree(bargs);
4822 out:
4823         mutex_unlock(&fs_info->balance_mutex);
4824         return ret;
4825 }
4826
4827 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4828 {
4829         struct inode *inode = file_inode(file);
4830         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4831         struct btrfs_ioctl_quota_ctl_args *sa;
4832         int ret;
4833
4834         if (!capable(CAP_SYS_ADMIN))
4835                 return -EPERM;
4836
4837         ret = mnt_want_write_file(file);
4838         if (ret)
4839                 return ret;
4840
4841         sa = memdup_user(arg, sizeof(*sa));
4842         if (IS_ERR(sa)) {
4843                 ret = PTR_ERR(sa);
4844                 goto drop_write;
4845         }
4846
4847         down_write(&fs_info->subvol_sem);
4848
4849         switch (sa->cmd) {
4850         case BTRFS_QUOTA_CTL_ENABLE:
4851                 ret = btrfs_quota_enable(fs_info);
4852                 break;
4853         case BTRFS_QUOTA_CTL_DISABLE:
4854                 ret = btrfs_quota_disable(fs_info);
4855                 break;
4856         default:
4857                 ret = -EINVAL;
4858                 break;
4859         }
4860
4861         kfree(sa);
4862         up_write(&fs_info->subvol_sem);
4863 drop_write:
4864         mnt_drop_write_file(file);
4865         return ret;
4866 }
4867
4868 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4869 {
4870         struct inode *inode = file_inode(file);
4871         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4872         struct btrfs_root *root = BTRFS_I(inode)->root;
4873         struct btrfs_ioctl_qgroup_assign_args *sa;
4874         struct btrfs_trans_handle *trans;
4875         int ret;
4876         int err;
4877
4878         if (!capable(CAP_SYS_ADMIN))
4879                 return -EPERM;
4880
4881         ret = mnt_want_write_file(file);
4882         if (ret)
4883                 return ret;
4884
4885         sa = memdup_user(arg, sizeof(*sa));
4886         if (IS_ERR(sa)) {
4887                 ret = PTR_ERR(sa);
4888                 goto drop_write;
4889         }
4890
4891         trans = btrfs_join_transaction(root);
4892         if (IS_ERR(trans)) {
4893                 ret = PTR_ERR(trans);
4894                 goto out;
4895         }
4896
4897         if (sa->assign) {
4898                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4899         } else {
4900                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4901         }
4902
4903         /* update qgroup status and info */
4904         err = btrfs_run_qgroups(trans);
4905         if (err < 0)
4906                 btrfs_handle_fs_error(fs_info, err,
4907                                       "failed to update qgroup status and info");
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_qgroup_create(struct file *file, void __user *arg)
4920 {
4921         struct inode *inode = file_inode(file);
4922         struct btrfs_root *root = BTRFS_I(inode)->root;
4923         struct btrfs_ioctl_qgroup_create_args *sa;
4924         struct btrfs_trans_handle *trans;
4925         int ret;
4926         int err;
4927
4928         if (!capable(CAP_SYS_ADMIN))
4929                 return -EPERM;
4930
4931         ret = mnt_want_write_file(file);
4932         if (ret)
4933                 return ret;
4934
4935         sa = memdup_user(arg, sizeof(*sa));
4936         if (IS_ERR(sa)) {
4937                 ret = PTR_ERR(sa);
4938                 goto drop_write;
4939         }
4940
4941         if (!sa->qgroupid) {
4942                 ret = -EINVAL;
4943                 goto out;
4944         }
4945
4946         trans = btrfs_join_transaction(root);
4947         if (IS_ERR(trans)) {
4948                 ret = PTR_ERR(trans);
4949                 goto out;
4950         }
4951
4952         if (sa->create) {
4953                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4954         } else {
4955                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4956         }
4957
4958         err = btrfs_end_transaction(trans);
4959         if (err && !ret)
4960                 ret = err;
4961
4962 out:
4963         kfree(sa);
4964 drop_write:
4965         mnt_drop_write_file(file);
4966         return ret;
4967 }
4968
4969 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4970 {
4971         struct inode *inode = file_inode(file);
4972         struct btrfs_root *root = BTRFS_I(inode)->root;
4973         struct btrfs_ioctl_qgroup_limit_args *sa;
4974         struct btrfs_trans_handle *trans;
4975         int ret;
4976         int err;
4977         u64 qgroupid;
4978
4979         if (!capable(CAP_SYS_ADMIN))
4980                 return -EPERM;
4981
4982         ret = mnt_want_write_file(file);
4983         if (ret)
4984                 return ret;
4985
4986         sa = memdup_user(arg, sizeof(*sa));
4987         if (IS_ERR(sa)) {
4988                 ret = PTR_ERR(sa);
4989                 goto drop_write;
4990         }
4991
4992         trans = btrfs_join_transaction(root);
4993         if (IS_ERR(trans)) {
4994                 ret = PTR_ERR(trans);
4995                 goto out;
4996         }
4997
4998         qgroupid = sa->qgroupid;
4999         if (!qgroupid) {
5000                 /* take the current subvol as qgroup */
5001                 qgroupid = root->root_key.objectid;
5002         }
5003
5004         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
5005
5006         err = btrfs_end_transaction(trans);
5007         if (err && !ret)
5008                 ret = err;
5009
5010 out:
5011         kfree(sa);
5012 drop_write:
5013         mnt_drop_write_file(file);
5014         return ret;
5015 }
5016
5017 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5018 {
5019         struct inode *inode = file_inode(file);
5020         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5021         struct btrfs_ioctl_quota_rescan_args *qsa;
5022         int ret;
5023
5024         if (!capable(CAP_SYS_ADMIN))
5025                 return -EPERM;
5026
5027         ret = mnt_want_write_file(file);
5028         if (ret)
5029                 return ret;
5030
5031         qsa = memdup_user(arg, sizeof(*qsa));
5032         if (IS_ERR(qsa)) {
5033                 ret = PTR_ERR(qsa);
5034                 goto drop_write;
5035         }
5036
5037         if (qsa->flags) {
5038                 ret = -EINVAL;
5039                 goto out;
5040         }
5041
5042         ret = btrfs_qgroup_rescan(fs_info);
5043
5044 out:
5045         kfree(qsa);
5046 drop_write:
5047         mnt_drop_write_file(file);
5048         return ret;
5049 }
5050
5051 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5052 {
5053         struct inode *inode = file_inode(file);
5054         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5055         struct btrfs_ioctl_quota_rescan_args *qsa;
5056         int ret = 0;
5057
5058         if (!capable(CAP_SYS_ADMIN))
5059                 return -EPERM;
5060
5061         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5062         if (!qsa)
5063                 return -ENOMEM;
5064
5065         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5066                 qsa->flags = 1;
5067                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5068         }
5069
5070         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5071                 ret = -EFAULT;
5072
5073         kfree(qsa);
5074         return ret;
5075 }
5076
5077 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5078 {
5079         struct inode *inode = file_inode(file);
5080         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5081
5082         if (!capable(CAP_SYS_ADMIN))
5083                 return -EPERM;
5084
5085         return btrfs_qgroup_wait_for_completion(fs_info, true);
5086 }
5087
5088 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5089                                             struct btrfs_ioctl_received_subvol_args *sa)
5090 {
5091         struct inode *inode = file_inode(file);
5092         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5093         struct btrfs_root *root = BTRFS_I(inode)->root;
5094         struct btrfs_root_item *root_item = &root->root_item;
5095         struct btrfs_trans_handle *trans;
5096         struct timespec64 ct = current_time(inode);
5097         int ret = 0;
5098         int received_uuid_changed;
5099
5100         if (!inode_owner_or_capable(inode))
5101                 return -EPERM;
5102
5103         ret = mnt_want_write_file(file);
5104         if (ret < 0)
5105                 return ret;
5106
5107         down_write(&fs_info->subvol_sem);
5108
5109         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5110                 ret = -EINVAL;
5111                 goto out;
5112         }
5113
5114         if (btrfs_root_readonly(root)) {
5115                 ret = -EROFS;
5116                 goto out;
5117         }
5118
5119         /*
5120          * 1 - root item
5121          * 2 - uuid items (received uuid + subvol uuid)
5122          */
5123         trans = btrfs_start_transaction(root, 3);
5124         if (IS_ERR(trans)) {
5125                 ret = PTR_ERR(trans);
5126                 trans = NULL;
5127                 goto out;
5128         }
5129
5130         sa->rtransid = trans->transid;
5131         sa->rtime.sec = ct.tv_sec;
5132         sa->rtime.nsec = ct.tv_nsec;
5133
5134         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5135                                        BTRFS_UUID_SIZE);
5136         if (received_uuid_changed &&
5137             !btrfs_is_empty_uuid(root_item->received_uuid)) {
5138                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5139                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5140                                           root->root_key.objectid);
5141                 if (ret && ret != -ENOENT) {
5142                         btrfs_abort_transaction(trans, ret);
5143                         btrfs_end_transaction(trans);
5144                         goto out;
5145                 }
5146         }
5147         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5148         btrfs_set_root_stransid(root_item, sa->stransid);
5149         btrfs_set_root_rtransid(root_item, sa->rtransid);
5150         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5151         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5152         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5153         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5154
5155         ret = btrfs_update_root(trans, fs_info->tree_root,
5156                                 &root->root_key, &root->root_item);
5157         if (ret < 0) {
5158                 btrfs_end_transaction(trans);
5159                 goto out;
5160         }
5161         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5162                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
5163                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5164                                           root->root_key.objectid);
5165                 if (ret < 0 && ret != -EEXIST) {
5166                         btrfs_abort_transaction(trans, ret);
5167                         btrfs_end_transaction(trans);
5168                         goto out;
5169                 }
5170         }
5171         ret = btrfs_commit_transaction(trans);
5172 out:
5173         up_write(&fs_info->subvol_sem);
5174         mnt_drop_write_file(file);
5175         return ret;
5176 }
5177
5178 #ifdef CONFIG_64BIT
5179 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5180                                                 void __user *arg)
5181 {
5182         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5183         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5184         int ret = 0;
5185
5186         args32 = memdup_user(arg, sizeof(*args32));
5187         if (IS_ERR(args32))
5188                 return PTR_ERR(args32);
5189
5190         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5191         if (!args64) {
5192                 ret = -ENOMEM;
5193                 goto out;
5194         }
5195
5196         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5197         args64->stransid = args32->stransid;
5198         args64->rtransid = args32->rtransid;
5199         args64->stime.sec = args32->stime.sec;
5200         args64->stime.nsec = args32->stime.nsec;
5201         args64->rtime.sec = args32->rtime.sec;
5202         args64->rtime.nsec = args32->rtime.nsec;
5203         args64->flags = args32->flags;
5204
5205         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5206         if (ret)
5207                 goto out;
5208
5209         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5210         args32->stransid = args64->stransid;
5211         args32->rtransid = args64->rtransid;
5212         args32->stime.sec = args64->stime.sec;
5213         args32->stime.nsec = args64->stime.nsec;
5214         args32->rtime.sec = args64->rtime.sec;
5215         args32->rtime.nsec = args64->rtime.nsec;
5216         args32->flags = args64->flags;
5217
5218         ret = copy_to_user(arg, args32, sizeof(*args32));
5219         if (ret)
5220                 ret = -EFAULT;
5221
5222 out:
5223         kfree(args32);
5224         kfree(args64);
5225         return ret;
5226 }
5227 #endif
5228
5229 static long btrfs_ioctl_set_received_subvol(struct file *file,
5230                                             void __user *arg)
5231 {
5232         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5233         int ret = 0;
5234
5235         sa = memdup_user(arg, sizeof(*sa));
5236         if (IS_ERR(sa))
5237                 return PTR_ERR(sa);
5238
5239         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5240
5241         if (ret)
5242                 goto out;
5243
5244         ret = copy_to_user(arg, sa, sizeof(*sa));
5245         if (ret)
5246                 ret = -EFAULT;
5247
5248 out:
5249         kfree(sa);
5250         return ret;
5251 }
5252
5253 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5254 {
5255         struct inode *inode = file_inode(file);
5256         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5257         size_t len;
5258         int ret;
5259         char label[BTRFS_LABEL_SIZE];
5260
5261         spin_lock(&fs_info->super_lock);
5262         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5263         spin_unlock(&fs_info->super_lock);
5264
5265         len = strnlen(label, BTRFS_LABEL_SIZE);
5266
5267         if (len == BTRFS_LABEL_SIZE) {
5268                 btrfs_warn(fs_info,
5269                            "label is too long, return the first %zu bytes",
5270                            --len);
5271         }
5272
5273         ret = copy_to_user(arg, label, len);
5274
5275         return ret ? -EFAULT : 0;
5276 }
5277
5278 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5279 {
5280         struct inode *inode = file_inode(file);
5281         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5282         struct btrfs_root *root = BTRFS_I(inode)->root;
5283         struct btrfs_super_block *super_block = fs_info->super_copy;
5284         struct btrfs_trans_handle *trans;
5285         char label[BTRFS_LABEL_SIZE];
5286         int ret;
5287
5288         if (!capable(CAP_SYS_ADMIN))
5289                 return -EPERM;
5290
5291         if (copy_from_user(label, arg, sizeof(label)))
5292                 return -EFAULT;
5293
5294         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5295                 btrfs_err(fs_info,
5296                           "unable to set label with more than %d bytes",
5297                           BTRFS_LABEL_SIZE - 1);
5298                 return -EINVAL;
5299         }
5300
5301         ret = mnt_want_write_file(file);
5302         if (ret)
5303                 return ret;
5304
5305         trans = btrfs_start_transaction(root, 0);
5306         if (IS_ERR(trans)) {
5307                 ret = PTR_ERR(trans);
5308                 goto out_unlock;
5309         }
5310
5311         spin_lock(&fs_info->super_lock);
5312         strcpy(super_block->label, label);
5313         spin_unlock(&fs_info->super_lock);
5314         ret = btrfs_commit_transaction(trans);
5315
5316 out_unlock:
5317         mnt_drop_write_file(file);
5318         return ret;
5319 }
5320
5321 #define INIT_FEATURE_FLAGS(suffix) \
5322         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5323           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5324           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5325
5326 int btrfs_ioctl_get_supported_features(void __user *arg)
5327 {
5328         static const struct btrfs_ioctl_feature_flags features[3] = {
5329                 INIT_FEATURE_FLAGS(SUPP),
5330                 INIT_FEATURE_FLAGS(SAFE_SET),
5331                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5332         };
5333
5334         if (copy_to_user(arg, &features, sizeof(features)))
5335                 return -EFAULT;
5336
5337         return 0;
5338 }
5339
5340 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5341 {
5342         struct inode *inode = file_inode(file);
5343         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5344         struct btrfs_super_block *super_block = fs_info->super_copy;
5345         struct btrfs_ioctl_feature_flags features;
5346
5347         features.compat_flags = btrfs_super_compat_flags(super_block);
5348         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5349         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5350
5351         if (copy_to_user(arg, &features, sizeof(features)))
5352                 return -EFAULT;
5353
5354         return 0;
5355 }
5356
5357 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5358                               enum btrfs_feature_set set,
5359                               u64 change_mask, u64 flags, u64 supported_flags,
5360                               u64 safe_set, u64 safe_clear)
5361 {
5362         const char *type = btrfs_feature_set_names[set];
5363         char *names;
5364         u64 disallowed, unsupported;
5365         u64 set_mask = flags & change_mask;
5366         u64 clear_mask = ~flags & change_mask;
5367
5368         unsupported = set_mask & ~supported_flags;
5369         if (unsupported) {
5370                 names = btrfs_printable_features(set, unsupported);
5371                 if (names) {
5372                         btrfs_warn(fs_info,
5373                                    "this kernel does not support the %s feature bit%s",
5374                                    names, strchr(names, ',') ? "s" : "");
5375                         kfree(names);
5376                 } else
5377                         btrfs_warn(fs_info,
5378                                    "this kernel does not support %s bits 0x%llx",
5379                                    type, unsupported);
5380                 return -EOPNOTSUPP;
5381         }
5382
5383         disallowed = set_mask & ~safe_set;
5384         if (disallowed) {
5385                 names = btrfs_printable_features(set, disallowed);
5386                 if (names) {
5387                         btrfs_warn(fs_info,
5388                                    "can't set the %s feature bit%s while mounted",
5389                                    names, strchr(names, ',') ? "s" : "");
5390                         kfree(names);
5391                 } else
5392                         btrfs_warn(fs_info,
5393                                    "can't set %s bits 0x%llx while mounted",
5394                                    type, disallowed);
5395                 return -EPERM;
5396         }
5397
5398         disallowed = clear_mask & ~safe_clear;
5399         if (disallowed) {
5400                 names = btrfs_printable_features(set, disallowed);
5401                 if (names) {
5402                         btrfs_warn(fs_info,
5403                                    "can't clear the %s feature bit%s while mounted",
5404                                    names, strchr(names, ',') ? "s" : "");
5405                         kfree(names);
5406                 } else
5407                         btrfs_warn(fs_info,
5408                                    "can't clear %s bits 0x%llx while mounted",
5409                                    type, disallowed);
5410                 return -EPERM;
5411         }
5412
5413         return 0;
5414 }
5415
5416 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5417 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5418                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5419                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5420                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5421
5422 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5423 {
5424         struct inode *inode = file_inode(file);
5425         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5426         struct btrfs_root *root = BTRFS_I(inode)->root;
5427         struct btrfs_super_block *super_block = fs_info->super_copy;
5428         struct btrfs_ioctl_feature_flags flags[2];
5429         struct btrfs_trans_handle *trans;
5430         u64 newflags;
5431         int ret;
5432
5433         if (!capable(CAP_SYS_ADMIN))
5434                 return -EPERM;
5435
5436         if (copy_from_user(flags, arg, sizeof(flags)))
5437                 return -EFAULT;
5438
5439         /* Nothing to do */
5440         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5441             !flags[0].incompat_flags)
5442                 return 0;
5443
5444         ret = check_feature(fs_info, flags[0].compat_flags,
5445                             flags[1].compat_flags, COMPAT);
5446         if (ret)
5447                 return ret;
5448
5449         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5450                             flags[1].compat_ro_flags, COMPAT_RO);
5451         if (ret)
5452                 return ret;
5453
5454         ret = check_feature(fs_info, flags[0].incompat_flags,
5455                             flags[1].incompat_flags, INCOMPAT);
5456         if (ret)
5457                 return ret;
5458
5459         ret = mnt_want_write_file(file);
5460         if (ret)
5461                 return ret;
5462
5463         trans = btrfs_start_transaction(root, 0);
5464         if (IS_ERR(trans)) {
5465                 ret = PTR_ERR(trans);
5466                 goto out_drop_write;
5467         }
5468
5469         spin_lock(&fs_info->super_lock);
5470         newflags = btrfs_super_compat_flags(super_block);
5471         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5472         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5473         btrfs_set_super_compat_flags(super_block, newflags);
5474
5475         newflags = btrfs_super_compat_ro_flags(super_block);
5476         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5477         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5478         btrfs_set_super_compat_ro_flags(super_block, newflags);
5479
5480         newflags = btrfs_super_incompat_flags(super_block);
5481         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5482         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5483         btrfs_set_super_incompat_flags(super_block, newflags);
5484         spin_unlock(&fs_info->super_lock);
5485
5486         ret = btrfs_commit_transaction(trans);
5487 out_drop_write:
5488         mnt_drop_write_file(file);
5489
5490         return ret;
5491 }
5492
5493 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5494 {
5495         struct btrfs_ioctl_send_args *arg;
5496         int ret;
5497
5498         if (compat) {
5499 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5500                 struct btrfs_ioctl_send_args_32 args32;
5501
5502                 ret = copy_from_user(&args32, argp, sizeof(args32));
5503                 if (ret)
5504                         return -EFAULT;
5505                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5506                 if (!arg)
5507                         return -ENOMEM;
5508                 arg->send_fd = args32.send_fd;
5509                 arg->clone_sources_count = args32.clone_sources_count;
5510                 arg->clone_sources = compat_ptr(args32.clone_sources);
5511                 arg->parent_root = args32.parent_root;
5512                 arg->flags = args32.flags;
5513                 memcpy(arg->reserved, args32.reserved,
5514                        sizeof(args32.reserved));
5515 #else
5516                 return -ENOTTY;
5517 #endif
5518         } else {
5519                 arg = memdup_user(argp, sizeof(*arg));
5520                 if (IS_ERR(arg))
5521                         return PTR_ERR(arg);
5522         }
5523         ret = btrfs_ioctl_send(file, arg);
5524         kfree(arg);
5525         return ret;
5526 }
5527
5528 long btrfs_ioctl(struct file *file, unsigned int
5529                 cmd, unsigned long arg)
5530 {
5531         struct inode *inode = file_inode(file);
5532         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5533         struct btrfs_root *root = BTRFS_I(inode)->root;
5534         void __user *argp = (void __user *)arg;
5535
5536         switch (cmd) {
5537         case FS_IOC_GETFLAGS:
5538                 return btrfs_ioctl_getflags(file, argp);
5539         case FS_IOC_SETFLAGS:
5540                 return btrfs_ioctl_setflags(file, argp);
5541         case FS_IOC_GETVERSION:
5542                 return btrfs_ioctl_getversion(file, argp);
5543         case FITRIM:
5544                 return btrfs_ioctl_fitrim(file, argp);
5545         case BTRFS_IOC_SNAP_CREATE:
5546                 return btrfs_ioctl_snap_create(file, argp, 0);
5547         case BTRFS_IOC_SNAP_CREATE_V2:
5548                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5549         case BTRFS_IOC_SUBVOL_CREATE:
5550                 return btrfs_ioctl_snap_create(file, argp, 1);
5551         case BTRFS_IOC_SUBVOL_CREATE_V2:
5552                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5553         case BTRFS_IOC_SNAP_DESTROY:
5554                 return btrfs_ioctl_snap_destroy(file, argp);
5555         case BTRFS_IOC_SUBVOL_GETFLAGS:
5556                 return btrfs_ioctl_subvol_getflags(file, argp);
5557         case BTRFS_IOC_SUBVOL_SETFLAGS:
5558                 return btrfs_ioctl_subvol_setflags(file, argp);
5559         case BTRFS_IOC_DEFAULT_SUBVOL:
5560                 return btrfs_ioctl_default_subvol(file, argp);
5561         case BTRFS_IOC_DEFRAG:
5562                 return btrfs_ioctl_defrag(file, NULL);
5563         case BTRFS_IOC_DEFRAG_RANGE:
5564                 return btrfs_ioctl_defrag(file, argp);
5565         case BTRFS_IOC_RESIZE:
5566                 return btrfs_ioctl_resize(file, argp);
5567         case BTRFS_IOC_ADD_DEV:
5568                 return btrfs_ioctl_add_dev(fs_info, argp);
5569         case BTRFS_IOC_RM_DEV:
5570                 return btrfs_ioctl_rm_dev(file, argp);
5571         case BTRFS_IOC_RM_DEV_V2:
5572                 return btrfs_ioctl_rm_dev_v2(file, argp);
5573         case BTRFS_IOC_FS_INFO:
5574                 return btrfs_ioctl_fs_info(fs_info, argp);
5575         case BTRFS_IOC_DEV_INFO:
5576                 return btrfs_ioctl_dev_info(fs_info, argp);
5577         case BTRFS_IOC_BALANCE:
5578                 return btrfs_ioctl_balance(file, NULL);
5579         case BTRFS_IOC_TREE_SEARCH:
5580                 return btrfs_ioctl_tree_search(file, argp);
5581         case BTRFS_IOC_TREE_SEARCH_V2:
5582                 return btrfs_ioctl_tree_search_v2(file, argp);
5583         case BTRFS_IOC_INO_LOOKUP:
5584                 return btrfs_ioctl_ino_lookup(file, argp);
5585         case BTRFS_IOC_INO_PATHS:
5586                 return btrfs_ioctl_ino_to_path(root, argp);
5587         case BTRFS_IOC_LOGICAL_INO:
5588                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5589         case BTRFS_IOC_LOGICAL_INO_V2:
5590                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5591         case BTRFS_IOC_SPACE_INFO:
5592                 return btrfs_ioctl_space_info(fs_info, argp);
5593         case BTRFS_IOC_SYNC: {
5594                 int ret;
5595
5596                 ret = btrfs_start_delalloc_roots(fs_info, -1);
5597                 if (ret)
5598                         return ret;
5599                 ret = btrfs_sync_fs(inode->i_sb, 1);
5600                 /*
5601                  * The transaction thread may want to do more work,
5602                  * namely it pokes the cleaner kthread that will start
5603                  * processing uncleaned subvols.
5604                  */
5605                 wake_up_process(fs_info->transaction_kthread);
5606                 return ret;
5607         }
5608         case BTRFS_IOC_START_SYNC:
5609                 return btrfs_ioctl_start_sync(root, argp);
5610         case BTRFS_IOC_WAIT_SYNC:
5611                 return btrfs_ioctl_wait_sync(fs_info, argp);
5612         case BTRFS_IOC_SCRUB:
5613                 return btrfs_ioctl_scrub(file, argp);
5614         case BTRFS_IOC_SCRUB_CANCEL:
5615                 return btrfs_ioctl_scrub_cancel(fs_info);
5616         case BTRFS_IOC_SCRUB_PROGRESS:
5617                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5618         case BTRFS_IOC_BALANCE_V2:
5619                 return btrfs_ioctl_balance(file, argp);
5620         case BTRFS_IOC_BALANCE_CTL:
5621                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5622         case BTRFS_IOC_BALANCE_PROGRESS:
5623                 return btrfs_ioctl_balance_progress(fs_info, argp);
5624         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5625                 return btrfs_ioctl_set_received_subvol(file, argp);
5626 #ifdef CONFIG_64BIT
5627         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5628                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5629 #endif
5630         case BTRFS_IOC_SEND:
5631                 return _btrfs_ioctl_send(file, argp, false);
5632 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5633         case BTRFS_IOC_SEND_32:
5634                 return _btrfs_ioctl_send(file, argp, true);
5635 #endif
5636         case BTRFS_IOC_GET_DEV_STATS:
5637                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5638         case BTRFS_IOC_QUOTA_CTL:
5639                 return btrfs_ioctl_quota_ctl(file, argp);
5640         case BTRFS_IOC_QGROUP_ASSIGN:
5641                 return btrfs_ioctl_qgroup_assign(file, argp);
5642         case BTRFS_IOC_QGROUP_CREATE:
5643                 return btrfs_ioctl_qgroup_create(file, argp);
5644         case BTRFS_IOC_QGROUP_LIMIT:
5645                 return btrfs_ioctl_qgroup_limit(file, argp);
5646         case BTRFS_IOC_QUOTA_RESCAN:
5647                 return btrfs_ioctl_quota_rescan(file, argp);
5648         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5649                 return btrfs_ioctl_quota_rescan_status(file, argp);
5650         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5651                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5652         case BTRFS_IOC_DEV_REPLACE:
5653                 return btrfs_ioctl_dev_replace(fs_info, argp);
5654         case BTRFS_IOC_GET_FSLABEL:
5655                 return btrfs_ioctl_get_fslabel(file, argp);
5656         case BTRFS_IOC_SET_FSLABEL:
5657                 return btrfs_ioctl_set_fslabel(file, argp);
5658         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5659                 return btrfs_ioctl_get_supported_features(argp);
5660         case BTRFS_IOC_GET_FEATURES:
5661                 return btrfs_ioctl_get_features(file, argp);
5662         case BTRFS_IOC_SET_FEATURES:
5663                 return btrfs_ioctl_set_features(file, argp);
5664         case FS_IOC_FSGETXATTR:
5665                 return btrfs_ioctl_fsgetxattr(file, argp);
5666         case FS_IOC_FSSETXATTR:
5667                 return btrfs_ioctl_fssetxattr(file, argp);
5668         case BTRFS_IOC_GET_SUBVOL_INFO:
5669                 return btrfs_ioctl_get_subvol_info(file, argp);
5670         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5671                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5672         case BTRFS_IOC_INO_LOOKUP_USER:
5673                 return btrfs_ioctl_ino_lookup_user(file, argp);
5674         }
5675
5676         return -ENOTTY;
5677 }
5678
5679 #ifdef CONFIG_COMPAT
5680 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5681 {
5682         /*
5683          * These all access 32-bit values anyway so no further
5684          * handling is necessary.
5685          */
5686         switch (cmd) {
5687         case FS_IOC32_GETFLAGS:
5688                 cmd = FS_IOC_GETFLAGS;
5689                 break;
5690         case FS_IOC32_SETFLAGS:
5691                 cmd = FS_IOC_SETFLAGS;
5692                 break;
5693         case FS_IOC32_GETVERSION:
5694                 cmd = FS_IOC_GETVERSION;
5695                 break;
5696         }
5697
5698         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5699 }
5700 #endif