fs/btrfs/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/file.h>
   9 #include <linux/fs.h>
  10 #include <linux/fsnotify.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/writeback.h>
  19 #include <linux/compat.h>
  20 #include <linux/security.h>
  21 #include <linux/xattr.h>
  22 #include <linux/mm.h>
  23 #include <linux/slab.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/uuid.h>
  26 #include <linux/btrfs.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/iversion.h>
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32 #include "btrfs_inode.h"
  33 #include "print-tree.h"
  34 #include "volumes.h"
  35 #include "locking.h"
  36 #include "inode-map.h"
  37 #include "backref.h"
  38 #include "rcu-string.h"
  39 #include "send.h"
  40 #include "dev-replace.h"
  41 #include "props.h"
  42 #include "sysfs.h"
  43 #include "qgroup.h"
  44 #include "tree-log.h"
  45 #include "compression.h"
  46 #include "space-info.h"
  47 #include "delalloc-space.h"
  48 #include "block-group.h"
  49
  50 #ifdef CONFIG_64BIT
  51 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  52  * structures are incorrect, as the timespec structure from userspace
  53  * is 4 bytes too small. We define these alternatives here to teach
  54  * the kernel about the 32-bit struct packing.
  55  */
  56 struct btrfs_ioctl_timespec_32 {
  57         __u64 sec;
  58         __u32 nsec;
  59 } __attribute__ ((__packed__));
  60
  61 struct btrfs_ioctl_received_subvol_args_32 {
  62         char    uuid[BTRFS_UUID_SIZE];  /* in */
  63         __u64   stransid;               /* in */
  64         __u64   rtransid;               /* out */
  65         struct btrfs_ioctl_timespec_32 stime; /* in */
  66         struct btrfs_ioctl_timespec_32 rtime; /* out */
  67         __u64   flags;                  /* in */
  68         __u64   reserved[16];           /* in */
  69 } __attribute__ ((__packed__));
  70
  71 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  72                                 struct btrfs_ioctl_received_subvol_args_32)
  73 #endif
  74
  75 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  76 struct btrfs_ioctl_send_args_32 {
  77         __s64 send_fd;                  /* in */
  78         __u64 clone_sources_count;      /* in */
  79         compat_uptr_t clone_sources;    /* in */
  80         __u64 parent_root;              /* in */
  81         __u64 flags;                    /* in */
  82         __u64 reserved[4];              /* in */
  83 } __attribute__ ((__packed__));
  84
  85 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  86                                struct btrfs_ioctl_send_args_32)
  87 #endif
  88
  89 static int btrfs_clone(struct inode *src, struct inode *inode,
  90                        u64 off, u64 olen, u64 olen_aligned, u64 destoff,
  91                        int no_time_update);
  92
  93 /* Mask out flags that are inappropriate for the given type of inode. */
  94 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  95                 unsigned int flags)
  96 {
  97         if (S_ISDIR(inode->i_mode))
  98                 return flags;
  99         else if (S_ISREG(inode->i_mode))
 100                 return flags & ~FS_DIRSYNC_FL;
 101         else
 102                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 103 }
 104
 105 /*
 106  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 107  * ioctl.
 108  */
 109 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 110 {
 111         unsigned int iflags = 0;
 112
 113         if (flags & BTRFS_INODE_SYNC)
 114                 iflags |= FS_SYNC_FL;
 115         if (flags & BTRFS_INODE_IMMUTABLE)
 116                 iflags |= FS_IMMUTABLE_FL;
 117         if (flags & BTRFS_INODE_APPEND)
 118                 iflags |= FS_APPEND_FL;
 119         if (flags & BTRFS_INODE_NODUMP)
 120                 iflags |= FS_NODUMP_FL;
 121         if (flags & BTRFS_INODE_NOATIME)
 122                 iflags |= FS_NOATIME_FL;
 123         if (flags & BTRFS_INODE_DIRSYNC)
 124                 iflags |= FS_DIRSYNC_FL;
 125         if (flags & BTRFS_INODE_NODATACOW)
 126                 iflags |= FS_NOCOW_FL;
 127
 128         if (flags & BTRFS_INODE_NOCOMPRESS)
 129                 iflags |= FS_NOCOMP_FL;
 130         else if (flags & BTRFS_INODE_COMPRESS)
 131                 iflags |= FS_COMPR_FL;
 132
 133         return iflags;
 134 }
 135
 136 /*
 137  * Update inode->i_flags based on the btrfs internal flags.
 138  */
 139 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 140 {
 141         struct btrfs_inode *binode = BTRFS_I(inode);
 142         unsigned int new_fl = 0;
 143
 144         if (binode->flags & BTRFS_INODE_SYNC)
 145                 new_fl |= S_SYNC;
 146         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 147                 new_fl |= S_IMMUTABLE;
 148         if (binode->flags & BTRFS_INODE_APPEND)
 149                 new_fl |= S_APPEND;
 150         if (binode->flags & BTRFS_INODE_NOATIME)
 151                 new_fl |= S_NOATIME;
 152         if (binode->flags & BTRFS_INODE_DIRSYNC)
 153                 new_fl |= S_DIRSYNC;
 154
 155         set_mask_bits(&inode->i_flags,
 156                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 157                       new_fl);
 158 }
 159
 160 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 161 {
 162         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 163         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 164
 165         if (copy_to_user(arg, &flags, sizeof(flags)))
 166                 return -EFAULT;
 167         return 0;
 168 }
 169
 170 /*
 171  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 172  * the old and new flags are not conflicting
 173  */
 174 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 175 {
 176         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 177                       FS_NOATIME_FL | FS_NODUMP_FL | \
 178                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 179                       FS_NOCOMP_FL | FS_COMPR_FL |
 180                       FS_NOCOW_FL))
 181                 return -EOPNOTSUPP;
 182
 183         /* COMPR and NOCOMP on new/old are valid */
 184         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 185                 return -EINVAL;
 186
 187         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 188                 return -EINVAL;
 189
 190         /* NOCOW and compression options are mutually exclusive */
 191         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 192                 return -EINVAL;
 193         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 194                 return -EINVAL;
 195
 196         return 0;
 197 }
 198
 199 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 200 {
 201         struct inode *inode = file_inode(file);
 202         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 203         struct btrfs_inode *binode = BTRFS_I(inode);
 204         struct btrfs_root *root = binode->root;
 205         struct btrfs_trans_handle *trans;
 206         unsigned int fsflags, old_fsflags;
 207         int ret;
 208         const char *comp = NULL;
 209         u32 binode_flags;
 210
 211         if (!inode_owner_or_capable(inode))
 212                 return -EPERM;
 213
 214         if (btrfs_root_readonly(root))
 215                 return -EROFS;
 216
 217         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 218                 return -EFAULT;
 219
 220         ret = mnt_want_write_file(file);
 221         if (ret)
 222                 return ret;
 223
 224         inode_lock(inode);
 225         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 226         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 227
 228         ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 229         if (ret)
 230                 goto out_unlock;
 231
 232         ret = check_fsflags(old_fsflags, fsflags);
 233         if (ret)
 234                 goto out_unlock;
 235
 236         binode_flags = binode->flags;
 237         if (fsflags & FS_SYNC_FL)
 238                 binode_flags |= BTRFS_INODE_SYNC;
 239         else
 240                 binode_flags &= ~BTRFS_INODE_SYNC;
 241         if (fsflags & FS_IMMUTABLE_FL)
 242                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 243         else
 244                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 245         if (fsflags & FS_APPEND_FL)
 246                 binode_flags |= BTRFS_INODE_APPEND;
 247         else
 248                 binode_flags &= ~BTRFS_INODE_APPEND;
 249         if (fsflags & FS_NODUMP_FL)
 250                 binode_flags |= BTRFS_INODE_NODUMP;
 251         else
 252                 binode_flags &= ~BTRFS_INODE_NODUMP;
 253         if (fsflags & FS_NOATIME_FL)
 254                 binode_flags |= BTRFS_INODE_NOATIME;
 255         else
 256                 binode_flags &= ~BTRFS_INODE_NOATIME;
 257         if (fsflags & FS_DIRSYNC_FL)
 258                 binode_flags |= BTRFS_INODE_DIRSYNC;
 259         else
 260                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 261         if (fsflags & FS_NOCOW_FL) {
 262                 if (S_ISREG(inode->i_mode)) {
 263                         /*
 264                          * It's safe to turn csums off here, no extents exist.
 265                          * Otherwise we want the flag to reflect the real COW
 266                          * status of the file and will not set it.
 267                          */
 268                         if (inode->i_size == 0)
 269                                 binode_flags |= BTRFS_INODE_NODATACOW |
 270                                                 BTRFS_INODE_NODATASUM;
 271                 } else {
 272                         binode_flags |= BTRFS_INODE_NODATACOW;
 273                 }
 274         } else {
 275                 /*
 276                  * Revert back under same assumptions as above
 277                  */
 278                 if (S_ISREG(inode->i_mode)) {
 279                         if (inode->i_size == 0)
 280                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 281                                                   BTRFS_INODE_NODATASUM);
 282                 } else {
 283                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 284                 }
 285         }
 286
 287         /*
 288          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 289          * flag may be changed automatically if compression code won't make
 290          * things smaller.
 291          */
 292         if (fsflags & FS_NOCOMP_FL) {
 293                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 294                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 295         } else if (fsflags & FS_COMPR_FL) {
 296
 297                 if (IS_SWAPFILE(inode)) {
 298                         ret = -ETXTBSY;
 299                         goto out_unlock;
 300                 }
 301
 302                 binode_flags |= BTRFS_INODE_COMPRESS;
 303                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 304
 305                 comp = btrfs_compress_type2str(fs_info->compress_type);
 306                 if (!comp || comp[0] == 0)
 307                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 308         } else {
 309                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 310         }
 311
 312         /*
 313          * 1 for inode item
 314          * 2 for properties
 315          */
 316         trans = btrfs_start_transaction(root, 3);
 317         if (IS_ERR(trans)) {
 318                 ret = PTR_ERR(trans);
 319                 goto out_unlock;
 320         }
 321
 322         if (comp) {
 323                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 324                                      strlen(comp), 0);
 325                 if (ret) {
 326                         btrfs_abort_transaction(trans, ret);
 327                         goto out_end_trans;
 328                 }
 329         } else {
 330                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 331                                      0, 0);
 332                 if (ret && ret != -ENODATA) {
 333                         btrfs_abort_transaction(trans, ret);
 334                         goto out_end_trans;
 335                 }
 336         }
 337
 338         binode->flags = binode_flags;
 339         btrfs_sync_inode_flags_to_i_flags(inode);
 340         inode_inc_iversion(inode);
 341         inode->i_ctime = current_time(inode);
 342         ret = btrfs_update_inode(trans, root, inode);
 343
 344  out_end_trans:
 345         btrfs_end_transaction(trans);
 346  out_unlock:
 347         inode_unlock(inode);
 348         mnt_drop_write_file(file);
 349         return ret;
 350 }
 351
 352 /*
 353  * Translate btrfs internal inode flags to xflags as expected by the
 354  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 355  * silently dropped.
 356  */
 357 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 358 {
 359         unsigned int xflags = 0;
 360
 361         if (flags & BTRFS_INODE_APPEND)
 362                 xflags |= FS_XFLAG_APPEND;
 363         if (flags & BTRFS_INODE_IMMUTABLE)
 364                 xflags |= FS_XFLAG_IMMUTABLE;
 365         if (flags & BTRFS_INODE_NOATIME)
 366                 xflags |= FS_XFLAG_NOATIME;
 367         if (flags & BTRFS_INODE_NODUMP)
 368                 xflags |= FS_XFLAG_NODUMP;
 369         if (flags & BTRFS_INODE_SYNC)
 370                 xflags |= FS_XFLAG_SYNC;
 371
 372         return xflags;
 373 }
 374
 375 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 376 static int check_xflags(unsigned int flags)
 377 {
 378         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 379                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 380                 return -EOPNOTSUPP;
 381         return 0;
 382 }
 383
 384 /*
 385  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 386  * are zeroed.
 387  */
 388 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 389 {
 390         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 391         struct fsxattr fa;
 392
 393         simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 394         if (copy_to_user(arg, &fa, sizeof(fa)))
 395                 return -EFAULT;
 396
 397         return 0;
 398 }
 399
 400 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 401 {
 402         struct inode *inode = file_inode(file);
 403         struct btrfs_inode *binode = BTRFS_I(inode);
 404         struct btrfs_root *root = binode->root;
 405         struct btrfs_trans_handle *trans;
 406         struct fsxattr fa, old_fa;
 407         unsigned old_flags;
 408         unsigned old_i_flags;
 409         int ret = 0;
 410
 411         if (!inode_owner_or_capable(inode))
 412                 return -EPERM;
 413
 414         if (btrfs_root_readonly(root))
 415                 return -EROFS;
 416
 417         if (copy_from_user(&fa, arg, sizeof(fa)))
 418                 return -EFAULT;
 419
 420         ret = check_xflags(fa.fsx_xflags);
 421         if (ret)
 422                 return ret;
 423
 424         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 425                 return -EOPNOTSUPP;
 426
 427         ret = mnt_want_write_file(file);
 428         if (ret)
 429                 return ret;
 430
 431         inode_lock(inode);
 432
 433         old_flags = binode->flags;
 434         old_i_flags = inode->i_flags;
 435
 436         simple_fill_fsxattr(&old_fa,
 437                             btrfs_inode_flags_to_xflags(binode->flags));
 438         ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 439         if (ret)
 440                 goto out_unlock;
 441
 442         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 443                 binode->flags |= BTRFS_INODE_SYNC;
 444         else
 445                 binode->flags &= ~BTRFS_INODE_SYNC;
 446         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 447                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 448         else
 449                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 450         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 451                 binode->flags |= BTRFS_INODE_APPEND;
 452         else
 453                 binode->flags &= ~BTRFS_INODE_APPEND;
 454         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 455                 binode->flags |= BTRFS_INODE_NODUMP;
 456         else
 457                 binode->flags &= ~BTRFS_INODE_NODUMP;
 458         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 459                 binode->flags |= BTRFS_INODE_NOATIME;
 460         else
 461                 binode->flags &= ~BTRFS_INODE_NOATIME;
 462
 463         /* 1 item for the inode */
 464         trans = btrfs_start_transaction(root, 1);
 465         if (IS_ERR(trans)) {
 466                 ret = PTR_ERR(trans);
 467                 goto out_unlock;
 468         }
 469
 470         btrfs_sync_inode_flags_to_i_flags(inode);
 471         inode_inc_iversion(inode);
 472         inode->i_ctime = current_time(inode);
 473         ret = btrfs_update_inode(trans, root, inode);
 474
 475         btrfs_end_transaction(trans);
 476
 477 out_unlock:
 478         if (ret) {
 479                 binode->flags = old_flags;
 480                 inode->i_flags = old_i_flags;
 481         }
 482
 483         inode_unlock(inode);
 484         mnt_drop_write_file(file);
 485
 486         return ret;
 487 }
 488
 489 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 490 {
 491         struct inode *inode = file_inode(file);
 492
 493         return put_user(inode->i_generation, arg);
 494 }
 495
 496 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 497 {
 498         struct inode *inode = file_inode(file);
 499         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 500         struct btrfs_device *device;
 501         struct request_queue *q;
 502         struct fstrim_range range;
 503         u64 minlen = ULLONG_MAX;
 504         u64 num_devices = 0;
 505         int ret;
 506
 507         if (!capable(CAP_SYS_ADMIN))
 508                 return -EPERM;
 509
 510         /*
 511          * If the fs is mounted with nologreplay, which requires it to be
 512          * mounted in RO mode as well, we can not allow discard on free space
 513          * inside block groups, because log trees refer to extents that are not
 514          * pinned in a block group's free space cache (pinning the extents is
 515          * precisely the first phase of replaying a log tree).
 516          */
 517         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 518                 return -EROFS;
 519
 520         rcu_read_lock();
 521         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 522                                 dev_list) {
 523                 if (!device->bdev)
 524                         continue;
 525                 q = bdev_get_queue(device->bdev);
 526                 if (blk_queue_discard(q)) {
 527                         num_devices++;
 528                         minlen = min_t(u64, q->limits.discard_granularity,
 529                                      minlen);
 530                 }
 531         }
 532         rcu_read_unlock();
 533
 534         if (!num_devices)
 535                 return -EOPNOTSUPP;
 536         if (copy_from_user(&range, arg, sizeof(range)))
 537                 return -EFAULT;
 538
 539         /*
 540          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 541          * block group is in the logical address space, which can be any
 542          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 543          */
 544         if (range.len < fs_info->sb->s_blocksize)
 545                 return -EINVAL;
 546
 547         range.minlen = max(range.minlen, minlen);
 548         ret = btrfs_trim_fs(fs_info, &range);
 549         if (ret < 0)
 550                 return ret;
 551
 552         if (copy_to_user(arg, &range, sizeof(range)))
 553                 return -EFAULT;
 554
 555         return 0;
 556 }
 557
 558 int btrfs_is_empty_uuid(u8 *uuid)
 559 {
 560         int i;
 561
 562         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 563                 if (uuid[i])
 564                         return 0;
 565         }
 566         return 1;
 567 }
 568
 569 static noinline int create_subvol(struct inode *dir,
 570                                   struct dentry *dentry,
 571                                   const char *name, int namelen,
 572                                   u64 *async_transid,
 573                                   struct btrfs_qgroup_inherit *inherit)
 574 {
 575         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 576         struct btrfs_trans_handle *trans;
 577         struct btrfs_key key;
 578         struct btrfs_root_item *root_item;
 579         struct btrfs_inode_item *inode_item;
 580         struct extent_buffer *leaf;
 581         struct btrfs_root *root = BTRFS_I(dir)->root;
 582         struct btrfs_root *new_root;
 583         struct btrfs_block_rsv block_rsv;
 584         struct timespec64 cur_time = current_time(dir);
 585         struct inode *inode;
 586         int ret;
 587         int err;
 588         u64 objectid;
 589         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 590         u64 index = 0;
 591         uuid_le new_uuid;
 592
 593         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 594         if (!root_item)
 595                 return -ENOMEM;
 596
 597         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 598         if (ret)
 599                 goto fail_free;
 600
 601         /*
 602          * Don't create subvolume whose level is not zero. Or qgroup will be
 603          * screwed up since it assumes subvolume qgroup's level to be 0.
 604          */
 605         if (btrfs_qgroup_level(objectid)) {
 606                 ret = -ENOSPC;
 607                 goto fail_free;
 608         }
 609
 610         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 611         /*
 612          * The same as the snapshot creation, please see the comment
 613          * of create_snapshot().
 614          */
 615         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 616         if (ret)
 617                 goto fail_free;
 618
 619         trans = btrfs_start_transaction(root, 0);
 620         if (IS_ERR(trans)) {
 621                 ret = PTR_ERR(trans);
 622                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 623                 goto fail_free;
 624         }
 625         trans->block_rsv = &block_rsv;
 626         trans->bytes_reserved = block_rsv.size;
 627
 628         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 629         if (ret)
 630                 goto fail;
 631
 632         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 633         if (IS_ERR(leaf)) {
 634                 ret = PTR_ERR(leaf);
 635                 goto fail;
 636         }
 637
 638         btrfs_mark_buffer_dirty(leaf);
 639
 640         inode_item = &root_item->inode;
 641         btrfs_set_stack_inode_generation(inode_item, 1);
 642         btrfs_set_stack_inode_size(inode_item, 3);
 643         btrfs_set_stack_inode_nlink(inode_item, 1);
 644         btrfs_set_stack_inode_nbytes(inode_item,
 645                                      fs_info->nodesize);
 646         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 647
 648         btrfs_set_root_flags(root_item, 0);
 649         btrfs_set_root_limit(root_item, 0);
 650         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 651
 652         btrfs_set_root_bytenr(root_item, leaf->start);
 653         btrfs_set_root_generation(root_item, trans->transid);
 654         btrfs_set_root_level(root_item, 0);
 655         btrfs_set_root_refs(root_item, 1);
 656         btrfs_set_root_used(root_item, leaf->len);
 657         btrfs_set_root_last_snapshot(root_item, 0);
 658
 659         btrfs_set_root_generation_v2(root_item,
 660                         btrfs_root_generation(root_item));
 661         uuid_le_gen(&new_uuid);
 662         memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
 663         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 664         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 665         root_item->ctime = root_item->otime;
 666         btrfs_set_root_ctransid(root_item, trans->transid);
 667         btrfs_set_root_otransid(root_item, trans->transid);
 668
 669         btrfs_tree_unlock(leaf);
 670
 671         btrfs_set_root_dirid(root_item, new_dirid);
 672
 673         key.objectid = objectid;
 674         key.offset = 0;
 675         key.type = BTRFS_ROOT_ITEM_KEY;
 676         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 677                                 root_item);
 678         if (ret) {
 679                 /*
 680                  * Since we don't abort the transaction in this case, free the
 681                  * tree block so that we don't leak space and leave the
 682                  * filesystem in an inconsistent state (an extent item in the
 683                  * extent tree without backreferences). Also no need to have
 684                  * the tree block locked since it is not in any tree at this
 685                  * point, so no other task can find it and use it.
 686                  */
 687                 btrfs_free_tree_block(trans, root, leaf, 0, 1);
 688                 free_extent_buffer(leaf);
 689                 goto fail;
 690         }
 691
 692         free_extent_buffer(leaf);
 693         leaf = NULL;
 694
 695         key.offset = (u64)-1;
 696         new_root = btrfs_read_fs_root_no_name(fs_info, &key);
 697         if (IS_ERR(new_root)) {
 698                 ret = PTR_ERR(new_root);
 699                 btrfs_abort_transaction(trans, ret);
 700                 goto fail;
 701         }
 702
 703         btrfs_record_root_in_trans(trans, new_root);
 704
 705         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 706         if (ret) {
 707                 /* We potentially lose an unused inode item here */
 708                 btrfs_abort_transaction(trans, ret);
 709                 goto fail;
 710         }
 711
 712         mutex_lock(&new_root->objectid_mutex);
 713         new_root->highest_objectid = new_dirid;
 714         mutex_unlock(&new_root->objectid_mutex);
 715
 716         /*
 717          * insert the directory item
 718          */
 719         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 720         if (ret) {
 721                 btrfs_abort_transaction(trans, ret);
 722                 goto fail;
 723         }
 724
 725         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 726                                     BTRFS_FT_DIR, index);
 727         if (ret) {
 728                 btrfs_abort_transaction(trans, ret);
 729                 goto fail;
 730         }
 731
 732         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 733         ret = btrfs_update_inode(trans, root, dir);
 734         if (ret) {
 735                 btrfs_abort_transaction(trans, ret);
 736                 goto fail;
 737         }
 738
 739         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 740                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 741         if (ret) {
 742                 btrfs_abort_transaction(trans, ret);
 743                 goto fail;
 744         }
 745
 746         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 747                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 748         if (ret)
 749                 btrfs_abort_transaction(trans, ret);
 750
 751 fail:
 752         kfree(root_item);
 753         trans->block_rsv = NULL;
 754         trans->bytes_reserved = 0;
 755         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 756
 757         if (async_transid) {
 758                 *async_transid = trans->transid;
 759                 err = btrfs_commit_transaction_async(trans, 1);
 760                 if (err)
 761                         err = btrfs_commit_transaction(trans);
 762         } else {
 763                 err = btrfs_commit_transaction(trans);
 764         }
 765         if (err && !ret)
 766                 ret = err;
 767
 768         if (!ret) {
 769                 inode = btrfs_lookup_dentry(dir, dentry);
 770                 if (IS_ERR(inode))
 771                         return PTR_ERR(inode);
 772                 d_instantiate(dentry, inode);
 773         }
 774         return ret;
 775
 776 fail_free:
 777         kfree(root_item);
 778         return ret;
 779 }
 780
 781 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 782                            struct dentry *dentry,
 783                            u64 *async_transid, bool readonly,
 784                            struct btrfs_qgroup_inherit *inherit)
 785 {
 786         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 787         struct inode *inode;
 788         struct btrfs_pending_snapshot *pending_snapshot;
 789         struct btrfs_trans_handle *trans;
 790         int ret;
 791         bool snapshot_force_cow = false;
 792
 793         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 794                 return -EINVAL;
 795
 796         if (atomic_read(&root->nr_swapfiles)) {
 797                 btrfs_warn(fs_info,
 798                            "cannot snapshot subvolume with active swapfile");
 799                 return -ETXTBSY;
 800         }
 801
 802         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 803         if (!pending_snapshot)
 804                 return -ENOMEM;
 805
 806         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 807                         GFP_KERNEL);
 808         pending_snapshot->path = btrfs_alloc_path();
 809         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 810                 ret = -ENOMEM;
 811                 goto free_pending;
 812         }
 813
 814         /*
 815          * Force new buffered writes to reserve space even when NOCOW is
 816          * possible. This is to avoid later writeback (running dealloc) to
 817          * fallback to COW mode and unexpectedly fail with ENOSPC.
 818          */
 819         atomic_inc(&root->will_be_snapshotted);
 820         smp_mb__after_atomic();
 821         /* wait for no snapshot writes */
 822         wait_event(root->subv_writers->wait,
 823                    percpu_counter_sum(&root->subv_writers->counter) == 0);
 824
 825         ret = btrfs_start_delalloc_snapshot(root);
 826         if (ret)
 827                 goto dec_and_free;
 828
 829         /*
 830          * All previous writes have started writeback in NOCOW mode, so now
 831          * we force future writes to fallback to COW mode during snapshot
 832          * creation.
 833          */
 834         atomic_inc(&root->snapshot_force_cow);
 835         snapshot_force_cow = true;
 836
 837         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
 838
 839         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 840                              BTRFS_BLOCK_RSV_TEMP);
 841         /*
 842          * 1 - parent dir inode
 843          * 2 - dir entries
 844          * 1 - root item
 845          * 2 - root ref/backref
 846          * 1 - root of snapshot
 847          * 1 - UUID item
 848          */
 849         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 850                                         &pending_snapshot->block_rsv, 8,
 851                                         false);
 852         if (ret)
 853                 goto dec_and_free;
 854
 855         pending_snapshot->dentry = dentry;
 856         pending_snapshot->root = root;
 857         pending_snapshot->readonly = readonly;
 858         pending_snapshot->dir = dir;
 859         pending_snapshot->inherit = inherit;
 860
 861         trans = btrfs_start_transaction(root, 0);
 862         if (IS_ERR(trans)) {
 863                 ret = PTR_ERR(trans);
 864                 goto fail;
 865         }
 866
 867         spin_lock(&fs_info->trans_lock);
 868         list_add(&pending_snapshot->list,
 869                  &trans->transaction->pending_snapshots);
 870         spin_unlock(&fs_info->trans_lock);
 871         if (async_transid) {
 872                 *async_transid = trans->transid;
 873                 ret = btrfs_commit_transaction_async(trans, 1);
 874                 if (ret)
 875                         ret = btrfs_commit_transaction(trans);
 876         } else {
 877                 ret = btrfs_commit_transaction(trans);
 878         }
 879         if (ret)
 880                 goto fail;
 881
 882         ret = pending_snapshot->error;
 883         if (ret)
 884                 goto fail;
 885
 886         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 887         if (ret)
 888                 goto fail;
 889
 890         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 891         if (IS_ERR(inode)) {
 892                 ret = PTR_ERR(inode);
 893                 goto fail;
 894         }
 895
 896         d_instantiate(dentry, inode);
 897         ret = 0;
 898 fail:
 899         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 900 dec_and_free:
 901         if (snapshot_force_cow)
 902                 atomic_dec(&root->snapshot_force_cow);
 903         if (atomic_dec_and_test(&root->will_be_snapshotted))
 904                 wake_up_var(&root->will_be_snapshotted);
 905 free_pending:
 906         kfree(pending_snapshot->root_item);
 907         btrfs_free_path(pending_snapshot->path);
 908         kfree(pending_snapshot);
 909
 910         return ret;
 911 }
 912
 913 /*  copy of may_delete in fs/namei.c()
 914  *      Check whether we can remove a link victim from directory dir, check
 915  *  whether the type of victim is right.
 916  *  1. We can't do it if dir is read-only (done in permission())
 917  *  2. We should have write and exec permissions on dir
 918  *  3. We can't remove anything from append-only dir
 919  *  4. We can't do anything with immutable dir (done in permission())
 920  *  5. If the sticky bit on dir is set we should either
 921  *      a. be owner of dir, or
 922  *      b. be owner of victim, or
 923  *      c. have CAP_FOWNER capability
 924  *  6. If the victim is append-only or immutable we can't do anything with
 925  *     links pointing to it.
 926  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 927  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 928  *  9. We can't remove a root or mountpoint.
 929  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 930  *     nfs_async_unlink().
 931  */
 932
 933 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 934 {
 935         int error;
 936
 937         if (d_really_is_negative(victim))
 938                 return -ENOENT;
 939
 940         BUG_ON(d_inode(victim->d_parent) != dir);
 941         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 942
 943         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 944         if (error)
 945                 return error;
 946         if (IS_APPEND(dir))
 947                 return -EPERM;
 948         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 949             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 950                 return -EPERM;
 951         if (isdir) {
 952                 if (!d_is_dir(victim))
 953                         return -ENOTDIR;
 954                 if (IS_ROOT(victim))
 955                         return -EBUSY;
 956         } else if (d_is_dir(victim))
 957                 return -EISDIR;
 958         if (IS_DEADDIR(dir))
 959                 return -ENOENT;
 960         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 961                 return -EBUSY;
 962         return 0;
 963 }
 964
 965 /* copy of may_create in fs/namei.c() */
 966 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 967 {
 968         if (d_really_is_positive(child))
 969                 return -EEXIST;
 970         if (IS_DEADDIR(dir))
 971                 return -ENOENT;
 972         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 973 }
 974
 975 /*
 976  * Create a new subvolume below @parent.  This is largely modeled after
 977  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 978  * inside this filesystem so it's quite a bit simpler.
 979  */
 980 static noinline int btrfs_mksubvol(const struct path *parent,
 981                                    const char *name, int namelen,
 982                                    struct btrfs_root *snap_src,
 983                                    u64 *async_transid, bool readonly,
 984                                    struct btrfs_qgroup_inherit *inherit)
 985 {
 986         struct inode *dir = d_inode(parent->dentry);
 987         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 988         struct dentry *dentry;
 989         int error;
 990
 991         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 992         if (error == -EINTR)
 993                 return error;
 994
 995         dentry = lookup_one_len(name, parent->dentry, namelen);
 996         error = PTR_ERR(dentry);
 997         if (IS_ERR(dentry))
 998                 goto out_unlock;
 999
1000         error = btrfs_may_create(dir, dentry);
1001         if (error)
1002                 goto out_dput;
1003
1004         /*
1005          * even if this name doesn't exist, we may get hash collisions.
1006          * check for them now when we can safely fail
1007          */
1008         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
1009                                                dir->i_ino, name,
1010                                                namelen);
1011         if (error)
1012                 goto out_dput;
1013
1014         down_read(&fs_info->subvol_sem);
1015
1016         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
1017                 goto out_up_read;
1018
1019         if (snap_src) {
1020                 error = create_snapshot(snap_src, dir, dentry,
1021                                         async_transid, readonly, inherit);
1022         } else {
1023                 error = create_subvol(dir, dentry, name, namelen,
1024                                       async_transid, inherit);
1025         }
1026         if (!error)
1027                 fsnotify_mkdir(dir, dentry);
1028 out_up_read:
1029         up_read(&fs_info->subvol_sem);
1030 out_dput:
1031         dput(dentry);
1032 out_unlock:
1033         inode_unlock(dir);
1034         return error;
1035 }
1036
1037 /*
1038  * When we're defragging a range, we don't want to kick it off again
1039  * if it is really just waiting for delalloc to send it down.
1040  * If we find a nice big extent or delalloc range for the bytes in the
1041  * file you want to defrag, we return 0 to let you know to skip this
1042  * part of the file
1043  */
1044 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1045 {
1046         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1047         struct extent_map *em = NULL;
1048         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1049         u64 end;
1050
1051         read_lock(&em_tree->lock);
1052         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1053         read_unlock(&em_tree->lock);
1054
1055         if (em) {
1056                 end = extent_map_end(em);
1057                 free_extent_map(em);
1058                 if (end - offset > thresh)
1059                         return 0;
1060         }
1061         /* if we already have a nice delalloc here, just stop */
1062         thresh /= 2;
1063         end = count_range_bits(io_tree, &offset, offset + thresh,
1064                                thresh, EXTENT_DELALLOC, 1);
1065         if (end >= thresh)
1066                 return 0;
1067         return 1;
1068 }
1069
1070 /*
1071  * helper function to walk through a file and find extents
1072  * newer than a specific transid, and smaller than thresh.
1073  *
1074  * This is used by the defragging code to find new and small
1075  * extents
1076  */
1077 static int find_new_extents(struct btrfs_root *root,
1078                             struct inode *inode, u64 newer_than,
1079                             u64 *off, u32 thresh)
1080 {
1081         struct btrfs_path *path;
1082         struct btrfs_key min_key;
1083         struct extent_buffer *leaf;
1084         struct btrfs_file_extent_item *extent;
1085         int type;
1086         int ret;
1087         u64 ino = btrfs_ino(BTRFS_I(inode));
1088
1089         path = btrfs_alloc_path();
1090         if (!path)
1091                 return -ENOMEM;
1092
1093         min_key.objectid = ino;
1094         min_key.type = BTRFS_EXTENT_DATA_KEY;
1095         min_key.offset = *off;
1096
1097         while (1) {
1098                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1099                 if (ret != 0)
1100                         goto none;
1101 process_slot:
1102                 if (min_key.objectid != ino)
1103                         goto none;
1104                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1105                         goto none;
1106
1107                 leaf = path->nodes[0];
1108                 extent = btrfs_item_ptr(leaf, path->slots[0],
1109                                         struct btrfs_file_extent_item);
1110
1111                 type = btrfs_file_extent_type(leaf, extent);
1112                 if (type == BTRFS_FILE_EXTENT_REG &&
1113                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1114                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1115                         *off = min_key.offset;
1116                         btrfs_free_path(path);
1117                         return 0;
1118                 }
1119
1120                 path->slots[0]++;
1121                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1122                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1123                         goto process_slot;
1124                 }
1125
1126                 if (min_key.offset == (u64)-1)
1127                         goto none;
1128
1129                 min_key.offset++;
1130                 btrfs_release_path(path);
1131         }
1132 none:
1133         btrfs_free_path(path);
1134         return -ENOENT;
1135 }
1136
1137 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1138 {
1139         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1140         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1141         struct extent_map *em;
1142         u64 len = PAGE_SIZE;
1143
1144         /*
1145          * hopefully we have this extent in the tree already, try without
1146          * the full extent lock
1147          */
1148         read_lock(&em_tree->lock);
1149         em = lookup_extent_mapping(em_tree, start, len);
1150         read_unlock(&em_tree->lock);
1151
1152         if (!em) {
1153                 struct extent_state *cached = NULL;
1154                 u64 end = start + len - 1;
1155
1156                 /* get the big lock and read metadata off disk */
1157                 lock_extent_bits(io_tree, start, end, &cached);
1158                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
1159                 unlock_extent_cached(io_tree, start, end, &cached);
1160
1161                 if (IS_ERR(em))
1162                         return NULL;
1163         }
1164
1165         return em;
1166 }
1167
1168 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1169 {
1170         struct extent_map *next;
1171         bool ret = true;
1172
1173         /* this is the last extent */
1174         if (em->start + em->len >= i_size_read(inode))
1175                 return false;
1176
1177         next = defrag_lookup_extent(inode, em->start + em->len);
1178         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1179                 ret = false;
1180         else if ((em->block_start + em->block_len == next->block_start) &&
1181                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1182                 ret = false;
1183
1184         free_extent_map(next);
1185         return ret;
1186 }
1187
1188 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1189                                u64 *last_len, u64 *skip, u64 *defrag_end,
1190                                int compress)
1191 {
1192         struct extent_map *em;
1193         int ret = 1;
1194         bool next_mergeable = true;
1195         bool prev_mergeable = true;
1196
1197         /*
1198          * make sure that once we start defragging an extent, we keep on
1199          * defragging it
1200          */
1201         if (start < *defrag_end)
1202                 return 1;
1203
1204         *skip = 0;
1205
1206         em = defrag_lookup_extent(inode, start);
1207         if (!em)
1208                 return 0;
1209
1210         /* this will cover holes, and inline extents */
1211         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1212                 ret = 0;
1213                 goto out;
1214         }
1215
1216         if (!*defrag_end)
1217                 prev_mergeable = false;
1218
1219         next_mergeable = defrag_check_next_extent(inode, em);
1220         /*
1221          * we hit a real extent, if it is big or the next extent is not a
1222          * real extent, don't bother defragging it
1223          */
1224         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1225             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1226                 ret = 0;
1227 out:
1228         /*
1229          * last_len ends up being a counter of how many bytes we've defragged.
1230          * every time we choose not to defrag an extent, we reset *last_len
1231          * so that the next tiny extent will force a defrag.
1232          *
1233          * The end result of this is that tiny extents before a single big
1234          * extent will force at least part of that big extent to be defragged.
1235          */
1236         if (ret) {
1237                 *defrag_end = extent_map_end(em);
1238         } else {
1239                 *last_len = 0;
1240                 *skip = extent_map_end(em);
1241                 *defrag_end = 0;
1242         }
1243
1244         free_extent_map(em);
1245         return ret;
1246 }
1247
1248 /*
1249  * it doesn't do much good to defrag one or two pages
1250  * at a time.  This pulls in a nice chunk of pages
1251  * to COW and defrag.
1252  *
1253  * It also makes sure the delalloc code has enough
1254  * dirty data to avoid making new small extents as part
1255  * of the defrag
1256  *
1257  * It's a good idea to start RA on this range
1258  * before calling this.
1259  */
1260 static int cluster_pages_for_defrag(struct inode *inode,
1261                                     struct page **pages,
1262                                     unsigned long start_index,
1263                                     unsigned long num_pages)
1264 {
1265         unsigned long file_end;
1266         u64 isize = i_size_read(inode);
1267         u64 page_start;
1268         u64 page_end;
1269         u64 page_cnt;
1270         u64 start = (u64)start_index << PAGE_SHIFT;
1271         u64 search_start;
1272         int ret;
1273         int i;
1274         int i_done;
1275         struct btrfs_ordered_extent *ordered;
1276         struct extent_state *cached_state = NULL;
1277         struct extent_io_tree *tree;
1278         struct extent_changeset *data_reserved = NULL;
1279         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1280
1281         file_end = (isize - 1) >> PAGE_SHIFT;
1282         if (!isize || start_index > file_end)
1283                 return 0;
1284
1285         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1286
1287         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
1288                         start, page_cnt << PAGE_SHIFT);
1289         if (ret)
1290                 return ret;
1291         i_done = 0;
1292         tree = &BTRFS_I(inode)->io_tree;
1293
1294         /* step one, lock all the pages */
1295         for (i = 0; i < page_cnt; i++) {
1296                 struct page *page;
1297 again:
1298                 page = find_or_create_page(inode->i_mapping,
1299                                            start_index + i, mask);
1300                 if (!page)
1301                         break;
1302
1303                 page_start = page_offset(page);
1304                 page_end = page_start + PAGE_SIZE - 1;
1305                 while (1) {
1306                         lock_extent_bits(tree, page_start, page_end,
1307                                          &cached_state);
1308                         ordered = btrfs_lookup_ordered_extent(inode,
1309                                                               page_start);
1310                         unlock_extent_cached(tree, page_start, page_end,
1311                                              &cached_state);
1312                         if (!ordered)
1313                                 break;
1314
1315                         unlock_page(page);
1316                         btrfs_start_ordered_extent(inode, ordered, 1);
1317                         btrfs_put_ordered_extent(ordered);
1318                         lock_page(page);
1319                         /*
1320                          * we unlocked the page above, so we need check if
1321                          * it was released or not.
1322                          */
1323                         if (page->mapping != inode->i_mapping) {
1324                                 unlock_page(page);
1325                                 put_page(page);
1326                                 goto again;
1327                         }
1328                 }
1329
1330                 if (!PageUptodate(page)) {
1331                         btrfs_readpage(NULL, page);
1332                         lock_page(page);
1333                         if (!PageUptodate(page)) {
1334                                 unlock_page(page);
1335                                 put_page(page);
1336                                 ret = -EIO;
1337                                 break;
1338                         }
1339                 }
1340
1341                 if (page->mapping != inode->i_mapping) {
1342                         unlock_page(page);
1343                         put_page(page);
1344                         goto again;
1345                 }
1346
1347                 pages[i] = page;
1348                 i_done++;
1349         }
1350         if (!i_done || ret)
1351                 goto out;
1352
1353         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1354                 goto out;
1355
1356         /*
1357          * so now we have a nice long stream of locked
1358          * and up to date pages, lets wait on them
1359          */
1360         for (i = 0; i < i_done; i++)
1361                 wait_on_page_writeback(pages[i]);
1362
1363         page_start = page_offset(pages[0]);
1364         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1365
1366         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1367                          page_start, page_end - 1, &cached_state);
1368
1369         /*
1370          * When defragmenting we skip ranges that have holes or inline extents,
1371          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1372          * space. At btrfs_defrag_file(), we check if a range should be defragged
1373          * before locking the inode and then, if it should, we trigger a sync
1374          * page cache readahead - we lock the inode only after that to avoid
1375          * blocking for too long other tasks that possibly want to operate on
1376          * other file ranges. But before we were able to get the inode lock,
1377          * some other task may have punched a hole in the range, or we may have
1378          * now an inline extent, in which case we should not defrag. So check
1379          * for that here, where we have the inode and the range locked, and bail
1380          * out if that happened.
1381          */
1382         search_start = page_start;
1383         while (search_start < page_end) {
1384                 struct extent_map *em;
1385
1386                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1387                                       page_end - search_start, 0);
1388                 if (IS_ERR(em)) {
1389                         ret = PTR_ERR(em);
1390                         goto out_unlock_range;
1391                 }
1392                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1393                         free_extent_map(em);
1394                         /* Ok, 0 means we did not defrag anything */
1395                         ret = 0;
1396                         goto out_unlock_range;
1397                 }
1398                 search_start = extent_map_end(em);
1399                 free_extent_map(em);
1400         }
1401
1402         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1403                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1404                           EXTENT_DEFRAG, 0, 0, &cached_state);
1405
1406         if (i_done != page_cnt) {
1407                 spin_lock(&BTRFS_I(inode)->lock);
1408                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1409                 spin_unlock(&BTRFS_I(inode)->lock);
1410                 btrfs_delalloc_release_space(inode, data_reserved,
1411                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1412         }
1413
1414
1415         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1416                           &cached_state);
1417
1418         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1419                              page_start, page_end - 1, &cached_state);
1420
1421         for (i = 0; i < i_done; i++) {
1422                 clear_page_dirty_for_io(pages[i]);
1423                 ClearPageChecked(pages[i]);
1424                 set_page_extent_mapped(pages[i]);
1425                 set_page_dirty(pages[i]);
1426                 unlock_page(pages[i]);
1427                 put_page(pages[i]);
1428         }
1429         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1430         extent_changeset_free(data_reserved);
1431         return i_done;
1432
1433 out_unlock_range:
1434         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1435                              page_start, page_end - 1, &cached_state);
1436 out:
1437         for (i = 0; i < i_done; i++) {
1438                 unlock_page(pages[i]);
1439                 put_page(pages[i]);
1440         }
1441         btrfs_delalloc_release_space(inode, data_reserved,
1442                         start, page_cnt << PAGE_SHIFT, true);
1443         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1444         extent_changeset_free(data_reserved);
1445         return ret;
1446
1447 }
1448
1449 int btrfs_defrag_file(struct inode *inode, struct file *file,
1450                       struct btrfs_ioctl_defrag_range_args *range,
1451                       u64 newer_than, unsigned long max_to_defrag)
1452 {
1453         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         struct file_ra_state *ra = NULL;
1456         unsigned long last_index;
1457         u64 isize = i_size_read(inode);
1458         u64 last_len = 0;
1459         u64 skip = 0;
1460         u64 defrag_end = 0;
1461         u64 newer_off = range->start;
1462         unsigned long i;
1463         unsigned long ra_index = 0;
1464         int ret;
1465         int defrag_count = 0;
1466         int compress_type = BTRFS_COMPRESS_ZLIB;
1467         u32 extent_thresh = range->extent_thresh;
1468         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1469         unsigned long cluster = max_cluster;
1470         u64 new_align = ~((u64)SZ_128K - 1);
1471         struct page **pages = NULL;
1472         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1473
1474         if (isize == 0)
1475                 return 0;
1476
1477         if (range->start >= isize)
1478                 return -EINVAL;
1479
1480         if (do_compress) {
1481                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1482                         return -EINVAL;
1483                 if (range->compress_type)
1484                         compress_type = range->compress_type;
1485         }
1486
1487         if (extent_thresh == 0)
1488                 extent_thresh = SZ_256K;
1489
1490         /*
1491          * If we were not given a file, allocate a readahead context. As
1492          * readahead is just an optimization, defrag will work without it so
1493          * we don't error out.
1494          */
1495         if (!file) {
1496                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1497                 if (ra)
1498                         file_ra_state_init(ra, inode->i_mapping);
1499         } else {
1500                 ra = &file->f_ra;
1501         }
1502
1503         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1504         if (!pages) {
1505                 ret = -ENOMEM;
1506                 goto out_ra;
1507         }
1508
1509         /* find the last page to defrag */
1510         if (range->start + range->len > range->start) {
1511                 last_index = min_t(u64, isize - 1,
1512                          range->start + range->len - 1) >> PAGE_SHIFT;
1513         } else {
1514                 last_index = (isize - 1) >> PAGE_SHIFT;
1515         }
1516
1517         if (newer_than) {
1518                 ret = find_new_extents(root, inode, newer_than,
1519                                        &newer_off, SZ_64K);
1520                 if (!ret) {
1521                         range->start = newer_off;
1522                         /*
1523                          * we always align our defrag to help keep
1524                          * the extents in the file evenly spaced
1525                          */
1526                         i = (newer_off & new_align) >> PAGE_SHIFT;
1527                 } else
1528                         goto out_ra;
1529         } else {
1530                 i = range->start >> PAGE_SHIFT;
1531         }
1532         if (!max_to_defrag)
1533                 max_to_defrag = last_index - i + 1;
1534
1535         /*
1536          * make writeback starts from i, so the defrag range can be
1537          * written sequentially.
1538          */
1539         if (i < inode->i_mapping->writeback_index)
1540                 inode->i_mapping->writeback_index = i;
1541
1542         while (i <= last_index && defrag_count < max_to_defrag &&
1543                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1544                 /*
1545                  * make sure we stop running if someone unmounts
1546                  * the FS
1547                  */
1548                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1549                         break;
1550
1551                 if (btrfs_defrag_cancelled(fs_info)) {
1552                         btrfs_debug(fs_info, "defrag_file cancelled");
1553                         ret = -EAGAIN;
1554                         break;
1555                 }
1556
1557                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1558                                          extent_thresh, &last_len, &skip,
1559                                          &defrag_end, do_compress)){
1560                         unsigned long next;
1561                         /*
1562                          * the should_defrag function tells us how much to skip
1563                          * bump our counter by the suggested amount
1564                          */
1565                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1566                         i = max(i + 1, next);
1567                         continue;
1568                 }
1569
1570                 if (!newer_than) {
1571                         cluster = (PAGE_ALIGN(defrag_end) >>
1572                                    PAGE_SHIFT) - i;
1573                         cluster = min(cluster, max_cluster);
1574                 } else {
1575                         cluster = max_cluster;
1576                 }
1577
1578                 if (i + cluster > ra_index) {
1579                         ra_index = max(i, ra_index);
1580                         if (ra)
1581                                 page_cache_sync_readahead(inode->i_mapping, ra,
1582                                                 file, ra_index, cluster);
1583                         ra_index += cluster;
1584                 }
1585
1586                 inode_lock(inode);
1587                 if (IS_SWAPFILE(inode)) {
1588                         ret = -ETXTBSY;
1589                 } else {
1590                         if (do_compress)
1591                                 BTRFS_I(inode)->defrag_compress = compress_type;
1592                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1593                 }
1594                 if (ret < 0) {
1595                         inode_unlock(inode);
1596                         goto out_ra;
1597                 }
1598
1599                 defrag_count += ret;
1600                 balance_dirty_pages_ratelimited(inode->i_mapping);
1601                 inode_unlock(inode);
1602
1603                 if (newer_than) {
1604                         if (newer_off == (u64)-1)
1605                                 break;
1606
1607                         if (ret > 0)
1608                                 i += ret;
1609
1610                         newer_off = max(newer_off + 1,
1611                                         (u64)i << PAGE_SHIFT);
1612
1613                         ret = find_new_extents(root, inode, newer_than,
1614                                                &newer_off, SZ_64K);
1615                         if (!ret) {
1616                                 range->start = newer_off;
1617                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1618                         } else {
1619                                 break;
1620                         }
1621                 } else {
1622                         if (ret > 0) {
1623                                 i += ret;
1624                                 last_len += ret << PAGE_SHIFT;
1625                         } else {
1626                                 i++;
1627                                 last_len = 0;
1628                         }
1629                 }
1630         }
1631
1632         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1633                 filemap_flush(inode->i_mapping);
1634                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1635                              &BTRFS_I(inode)->runtime_flags))
1636                         filemap_flush(inode->i_mapping);
1637         }
1638
1639         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1640                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1641         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1642                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1643         }
1644
1645         ret = defrag_count;
1646
1647 out_ra:
1648         if (do_compress) {
1649                 inode_lock(inode);
1650                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1651                 inode_unlock(inode);
1652         }
1653         if (!file)
1654                 kfree(ra);
1655         kfree(pages);
1656         return ret;
1657 }
1658
1659 static noinline int btrfs_ioctl_resize(struct file *file,
1660                                         void __user *arg)
1661 {
1662         struct inode *inode = file_inode(file);
1663         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1664         u64 new_size;
1665         u64 old_size;
1666         u64 devid = 1;
1667         struct btrfs_root *root = BTRFS_I(inode)->root;
1668         struct btrfs_ioctl_vol_args *vol_args;
1669         struct btrfs_trans_handle *trans;
1670         struct btrfs_device *device = NULL;
1671         char *sizestr;
1672         char *retptr;
1673         char *devstr = NULL;
1674         int ret = 0;
1675         int mod = 0;
1676
1677         if (!capable(CAP_SYS_ADMIN))
1678                 return -EPERM;
1679
1680         ret = mnt_want_write_file(file);
1681         if (ret)
1682                 return ret;
1683
1684         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1685                 mnt_drop_write_file(file);
1686                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1687         }
1688
1689         vol_args = memdup_user(arg, sizeof(*vol_args));
1690         if (IS_ERR(vol_args)) {
1691                 ret = PTR_ERR(vol_args);
1692                 goto out;
1693         }
1694
1695         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1696
1697         sizestr = vol_args->name;
1698         devstr = strchr(sizestr, ':');
1699         if (devstr) {
1700                 sizestr = devstr + 1;
1701                 *devstr = '\0';
1702                 devstr = vol_args->name;
1703                 ret = kstrtoull(devstr, 10, &devid);
1704                 if (ret)
1705                         goto out_free;
1706                 if (!devid) {
1707                         ret = -EINVAL;
1708                         goto out_free;
1709                 }
1710                 btrfs_info(fs_info, "resizing devid %llu", devid);
1711         }
1712
1713         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1714         if (!device) {
1715                 btrfs_info(fs_info, "resizer unable to find device %llu",
1716                            devid);
1717                 ret = -ENODEV;
1718                 goto out_free;
1719         }
1720
1721         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1722                 btrfs_info(fs_info,
1723                            "resizer unable to apply on readonly device %llu",
1724                        devid);
1725                 ret = -EPERM;
1726                 goto out_free;
1727         }
1728
1729         if (!strcmp(sizestr, "max"))
1730                 new_size = device->bdev->bd_inode->i_size;
1731         else {
1732                 if (sizestr[0] == '-') {
1733                         mod = -1;
1734                         sizestr++;
1735                 } else if (sizestr[0] == '+') {
1736                         mod = 1;
1737                         sizestr++;
1738                 }
1739                 new_size = memparse(sizestr, &retptr);
1740                 if (*retptr != '\0' || new_size == 0) {
1741                         ret = -EINVAL;
1742                         goto out_free;
1743                 }
1744         }
1745
1746         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1747                 ret = -EPERM;
1748                 goto out_free;
1749         }
1750
1751         old_size = btrfs_device_get_total_bytes(device);
1752
1753         if (mod < 0) {
1754                 if (new_size > old_size) {
1755                         ret = -EINVAL;
1756                         goto out_free;
1757                 }
1758                 new_size = old_size - new_size;
1759         } else if (mod > 0) {
1760                 if (new_size > ULLONG_MAX - old_size) {
1761                         ret = -ERANGE;
1762                         goto out_free;
1763                 }
1764                 new_size = old_size + new_size;
1765         }
1766
1767         if (new_size < SZ_256M) {
1768                 ret = -EINVAL;
1769                 goto out_free;
1770         }
1771         if (new_size > device->bdev->bd_inode->i_size) {
1772                 ret = -EFBIG;
1773                 goto out_free;
1774         }
1775
1776         new_size = round_down(new_size, fs_info->sectorsize);
1777
1778         btrfs_info_in_rcu(fs_info, "new size for %s is %llu",
1779                           rcu_str_deref(device->name), new_size);
1780
1781         if (new_size > old_size) {
1782                 trans = btrfs_start_transaction(root, 0);
1783                 if (IS_ERR(trans)) {
1784                         ret = PTR_ERR(trans);
1785                         goto out_free;
1786                 }
1787                 ret = btrfs_grow_device(trans, device, new_size);
1788                 btrfs_commit_transaction(trans);
1789         } else if (new_size < old_size) {
1790                 ret = btrfs_shrink_device(device, new_size);
1791         } /* equal, nothing need to do */
1792
1793 out_free:
1794         kfree(vol_args);
1795 out:
1796         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1797         mnt_drop_write_file(file);
1798         return ret;
1799 }
1800
1801 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1802                                 const char *name, unsigned long fd, int subvol,
1803                                 u64 *transid, bool readonly,
1804                                 struct btrfs_qgroup_inherit *inherit)
1805 {
1806         int namelen;
1807         int ret = 0;
1808
1809         if (!S_ISDIR(file_inode(file)->i_mode))
1810                 return -ENOTDIR;
1811
1812         ret = mnt_want_write_file(file);
1813         if (ret)
1814                 goto out;
1815
1816         namelen = strlen(name);
1817         if (strchr(name, '/')) {
1818                 ret = -EINVAL;
1819                 goto out_drop_write;
1820         }
1821
1822         if (name[0] == '.' &&
1823            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1824                 ret = -EEXIST;
1825                 goto out_drop_write;
1826         }
1827
1828         if (subvol) {
1829                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1830                                      NULL, transid, readonly, inherit);
1831         } else {
1832                 struct fd src = fdget(fd);
1833                 struct inode *src_inode;
1834                 if (!src.file) {
1835                         ret = -EINVAL;
1836                         goto out_drop_write;
1837                 }
1838
1839                 src_inode = file_inode(src.file);
1840                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1841                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1842                                    "Snapshot src from another FS");
1843                         ret = -EXDEV;
1844                 } else if (!inode_owner_or_capable(src_inode)) {
1845                         /*
1846                          * Subvolume creation is not restricted, but snapshots
1847                          * are limited to own subvolumes only
1848                          */
1849                         ret = -EPERM;
1850                 } else {
1851                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1852                                              BTRFS_I(src_inode)->root,
1853                                              transid, readonly, inherit);
1854                 }
1855                 fdput(src);
1856         }
1857 out_drop_write:
1858         mnt_drop_write_file(file);
1859 out:
1860         return ret;
1861 }
1862
1863 static noinline int btrfs_ioctl_snap_create(struct file *file,
1864                                             void __user *arg, int subvol)
1865 {
1866         struct btrfs_ioctl_vol_args *vol_args;
1867         int ret;
1868
1869         if (!S_ISDIR(file_inode(file)->i_mode))
1870                 return -ENOTDIR;
1871
1872         vol_args = memdup_user(arg, sizeof(*vol_args));
1873         if (IS_ERR(vol_args))
1874                 return PTR_ERR(vol_args);
1875         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1876
1877         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1878                                               vol_args->fd, subvol,
1879                                               NULL, false, NULL);
1880
1881         kfree(vol_args);
1882         return ret;
1883 }
1884
1885 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1886                                                void __user *arg, int subvol)
1887 {
1888         struct btrfs_ioctl_vol_args_v2 *vol_args;
1889         int ret;
1890         u64 transid = 0;
1891         u64 *ptr = NULL;
1892         bool readonly = false;
1893         struct btrfs_qgroup_inherit *inherit = NULL;
1894
1895         if (!S_ISDIR(file_inode(file)->i_mode))
1896                 return -ENOTDIR;
1897
1898         vol_args = memdup_user(arg, sizeof(*vol_args));
1899         if (IS_ERR(vol_args))
1900                 return PTR_ERR(vol_args);
1901         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1902
1903         if (vol_args->flags &
1904             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1905               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1906                 ret = -EOPNOTSUPP;
1907                 goto free_args;
1908         }
1909
1910         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1911                 struct inode *inode = file_inode(file);
1912                 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1913
1914                 btrfs_warn(fs_info,
1915 "SNAP_CREATE_V2 ioctl with CREATE_ASYNC is deprecated and will be removed in kernel 5.7");
1916
1917                 ptr = &transid;
1918         }
1919         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1920                 readonly = true;
1921         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1922                 u64 nums;
1923
1924                 if (vol_args->size < sizeof(*inherit) ||
1925                     vol_args->size > PAGE_SIZE) {
1926                         ret = -EINVAL;
1927                         goto free_args;
1928                 }
1929                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1930                 if (IS_ERR(inherit)) {
1931                         ret = PTR_ERR(inherit);
1932                         goto free_args;
1933                 }
1934
1935                 if (inherit->num_qgroups > PAGE_SIZE ||
1936                     inherit->num_ref_copies > PAGE_SIZE ||
1937                     inherit->num_excl_copies > PAGE_SIZE) {
1938                         ret = -EINVAL;
1939                         goto free_inherit;
1940                 }
1941
1942                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1943                        2 * inherit->num_excl_copies;
1944                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1945                         ret = -EINVAL;
1946                         goto free_inherit;
1947                 }
1948         }
1949
1950         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1951                                               vol_args->fd, subvol, ptr,
1952                                               readonly, inherit);
1953         if (ret)
1954                 goto free_inherit;
1955
1956         if (ptr && copy_to_user(arg +
1957                                 offsetof(struct btrfs_ioctl_vol_args_v2,
1958                                         transid),
1959                                 ptr, sizeof(*ptr)))
1960                 ret = -EFAULT;
1961
1962 free_inherit:
1963         kfree(inherit);
1964 free_args:
1965         kfree(vol_args);
1966         return ret;
1967 }
1968
1969 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1970                                                 void __user *arg)
1971 {
1972         struct inode *inode = file_inode(file);
1973         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1974         struct btrfs_root *root = BTRFS_I(inode)->root;
1975         int ret = 0;
1976         u64 flags = 0;
1977
1978         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1979                 return -EINVAL;
1980
1981         down_read(&fs_info->subvol_sem);
1982         if (btrfs_root_readonly(root))
1983                 flags |= BTRFS_SUBVOL_RDONLY;
1984         up_read(&fs_info->subvol_sem);
1985
1986         if (copy_to_user(arg, &flags, sizeof(flags)))
1987                 ret = -EFAULT;
1988
1989         return ret;
1990 }
1991
1992 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1993                                               void __user *arg)
1994 {
1995         struct inode *inode = file_inode(file);
1996         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1997         struct btrfs_root *root = BTRFS_I(inode)->root;
1998         struct btrfs_trans_handle *trans;
1999         u64 root_flags;
2000         u64 flags;
2001         int ret = 0;
2002
2003         if (!inode_owner_or_capable(inode))
2004                 return -EPERM;
2005
2006         ret = mnt_want_write_file(file);
2007         if (ret)
2008                 goto out;
2009
2010         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2011                 ret = -EINVAL;
2012                 goto out_drop_write;
2013         }
2014
2015         if (copy_from_user(&flags, arg, sizeof(flags))) {
2016                 ret = -EFAULT;
2017                 goto out_drop_write;
2018         }
2019
2020         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
2021                 ret = -EINVAL;
2022                 goto out_drop_write;
2023         }
2024
2025         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2026                 ret = -EOPNOTSUPP;
2027                 goto out_drop_write;
2028         }
2029
2030         down_write(&fs_info->subvol_sem);
2031
2032         /* nothing to do */
2033         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2034                 goto out_drop_sem;
2035
2036         root_flags = btrfs_root_flags(&root->root_item);
2037         if (flags & BTRFS_SUBVOL_RDONLY) {
2038                 btrfs_set_root_flags(&root->root_item,
2039                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2040         } else {
2041                 /*
2042                  * Block RO -> RW transition if this subvolume is involved in
2043                  * send
2044                  */
2045                 spin_lock(&root->root_item_lock);
2046                 if (root->send_in_progress == 0) {
2047                         btrfs_set_root_flags(&root->root_item,
2048                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2049                         spin_unlock(&root->root_item_lock);
2050                 } else {
2051                         spin_unlock(&root->root_item_lock);
2052                         btrfs_warn(fs_info,
2053                                    "Attempt to set subvolume %llu read-write during send",
2054                                    root->root_key.objectid);
2055                         ret = -EPERM;
2056                         goto out_drop_sem;
2057                 }
2058         }
2059
2060         trans = btrfs_start_transaction(root, 1);
2061         if (IS_ERR(trans)) {
2062                 ret = PTR_ERR(trans);
2063                 goto out_reset;
2064         }
2065
2066         ret = btrfs_update_root(trans, fs_info->tree_root,
2067                                 &root->root_key, &root->root_item);
2068         if (ret < 0) {
2069                 btrfs_end_transaction(trans);
2070                 goto out_reset;
2071         }
2072
2073         ret = btrfs_commit_transaction(trans);
2074
2075 out_reset:
2076         if (ret)
2077                 btrfs_set_root_flags(&root->root_item, root_flags);
2078 out_drop_sem:
2079         up_write(&fs_info->subvol_sem);
2080 out_drop_write:
2081         mnt_drop_write_file(file);
2082 out:
2083         return ret;
2084 }
2085
2086 static noinline int key_in_sk(struct btrfs_key *key,
2087                               struct btrfs_ioctl_search_key *sk)
2088 {
2089         struct btrfs_key test;
2090         int ret;
2091
2092         test.objectid = sk->min_objectid;
2093         test.type = sk->min_type;
2094         test.offset = sk->min_offset;
2095
2096         ret = btrfs_comp_cpu_keys(key, &test);
2097         if (ret < 0)
2098                 return 0;
2099
2100         test.objectid = sk->max_objectid;
2101         test.type = sk->max_type;
2102         test.offset = sk->max_offset;
2103
2104         ret = btrfs_comp_cpu_keys(key, &test);
2105         if (ret > 0)
2106                 return 0;
2107         return 1;
2108 }
2109
2110 static noinline int copy_to_sk(struct btrfs_path *path,
2111                                struct btrfs_key *key,
2112                                struct btrfs_ioctl_search_key *sk,
2113                                size_t *buf_size,
2114                                char __user *ubuf,
2115                                unsigned long *sk_offset,
2116                                int *num_found)
2117 {
2118         u64 found_transid;
2119         struct extent_buffer *leaf;
2120         struct btrfs_ioctl_search_header sh;
2121         struct btrfs_key test;
2122         unsigned long item_off;
2123         unsigned long item_len;
2124         int nritems;
2125         int i;
2126         int slot;
2127         int ret = 0;
2128
2129         leaf = path->nodes[0];
2130         slot = path->slots[0];
2131         nritems = btrfs_header_nritems(leaf);
2132
2133         if (btrfs_header_generation(leaf) > sk->max_transid) {
2134                 i = nritems;
2135                 goto advance_key;
2136         }
2137         found_transid = btrfs_header_generation(leaf);
2138
2139         for (i = slot; i < nritems; i++) {
2140                 item_off = btrfs_item_ptr_offset(leaf, i);
2141                 item_len = btrfs_item_size_nr(leaf, i);
2142
2143                 btrfs_item_key_to_cpu(leaf, key, i);
2144                 if (!key_in_sk(key, sk))
2145                         continue;
2146
2147                 if (sizeof(sh) + item_len > *buf_size) {
2148                         if (*num_found) {
2149                                 ret = 1;
2150                                 goto out;
2151                         }
2152
2153                         /*
2154                          * return one empty item back for v1, which does not
2155                          * handle -EOVERFLOW
2156                          */
2157
2158                         *buf_size = sizeof(sh) + item_len;
2159                         item_len = 0;
2160                         ret = -EOVERFLOW;
2161                 }
2162
2163                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2164                         ret = 1;
2165                         goto out;
2166                 }
2167
2168                 sh.objectid = key->objectid;
2169                 sh.offset = key->offset;
2170                 sh.type = key->type;
2171                 sh.len = item_len;
2172                 sh.transid = found_transid;
2173
2174                 /*
2175                  * Copy search result header. If we fault then loop again so we
2176                  * can fault in the pages and -EFAULT there if there's a
2177                  * problem. Otherwise we'll fault and then copy the buffer in
2178                  * properly this next time through
2179                  */
2180                 if (probe_user_write(ubuf + *sk_offset, &sh, sizeof(sh))) {
2181                         ret = 0;
2182                         goto out;
2183                 }
2184
2185                 *sk_offset += sizeof(sh);
2186
2187                 if (item_len) {
2188                         char __user *up = ubuf + *sk_offset;
2189                         /*
2190                          * Copy the item, same behavior as above, but reset the
2191                          * * sk_offset so we copy the full thing again.
2192                          */
2193                         if (read_extent_buffer_to_user_nofault(leaf, up,
2194                                                 item_off, item_len)) {
2195                                 ret = 0;
2196                                 *sk_offset -= sizeof(sh);
2197                                 goto out;
2198                         }
2199
2200                         *sk_offset += item_len;
2201                 }
2202                 (*num_found)++;
2203
2204                 if (ret) /* -EOVERFLOW from above */
2205                         goto out;
2206
2207                 if (*num_found >= sk->nr_items) {
2208                         ret = 1;
2209                         goto out;
2210                 }
2211         }
2212 advance_key:
2213         ret = 0;
2214         test.objectid = sk->max_objectid;
2215         test.type = sk->max_type;
2216         test.offset = sk->max_offset;
2217         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2218                 ret = 1;
2219         else if (key->offset < (u64)-1)
2220                 key->offset++;
2221         else if (key->type < (u8)-1) {
2222                 key->offset = 0;
2223                 key->type++;
2224         } else if (key->objectid < (u64)-1) {
2225                 key->offset = 0;
2226                 key->type = 0;
2227                 key->objectid++;
2228         } else
2229                 ret = 1;
2230 out:
2231         /*
2232          *  0: all items from this leaf copied, continue with next
2233          *  1: * more items can be copied, but unused buffer is too small
2234          *     * all items were found
2235          *     Either way, it will stops the loop which iterates to the next
2236          *     leaf
2237          *  -EOVERFLOW: item was to large for buffer
2238          *  -EFAULT: could not copy extent buffer back to userspace
2239          */
2240         return ret;
2241 }
2242
2243 static noinline int search_ioctl(struct inode *inode,
2244                                  struct btrfs_ioctl_search_key *sk,
2245                                  size_t *buf_size,
2246                                  char __user *ubuf)
2247 {
2248         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2249         struct btrfs_root *root;
2250         struct btrfs_key key;
2251         struct btrfs_path *path;
2252         int ret;
2253         int num_found = 0;
2254         unsigned long sk_offset = 0;
2255
2256         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2257                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2258                 return -EOVERFLOW;
2259         }
2260
2261         path = btrfs_alloc_path();
2262         if (!path)
2263                 return -ENOMEM;
2264
2265         if (sk->tree_id == 0) {
2266                 /* search the root of the inode that was passed */
2267                 root = BTRFS_I(inode)->root;
2268         } else {
2269                 key.objectid = sk->tree_id;
2270                 key.type = BTRFS_ROOT_ITEM_KEY;
2271                 key.offset = (u64)-1;
2272                 root = btrfs_read_fs_root_no_name(info, &key);
2273                 if (IS_ERR(root)) {
2274                         btrfs_free_path(path);
2275                         return PTR_ERR(root);
2276                 }
2277         }
2278
2279         key.objectid = sk->min_objectid;
2280         key.type = sk->min_type;
2281         key.offset = sk->min_offset;
2282
2283         while (1) {
2284                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2285                                                *buf_size - sk_offset);
2286                 if (ret)
2287                         break;
2288
2289                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2290                 if (ret != 0) {
2291                         if (ret > 0)
2292                                 ret = 0;
2293                         goto err;
2294                 }
2295                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2296                                  &sk_offset, &num_found);
2297                 btrfs_release_path(path);
2298                 if (ret)
2299                         break;
2300
2301         }
2302         if (ret > 0)
2303                 ret = 0;
2304 err:
2305         sk->nr_items = num_found;
2306         btrfs_free_path(path);
2307         return ret;
2308 }
2309
2310 static noinline int btrfs_ioctl_tree_search(struct file *file,
2311                                            void __user *argp)
2312 {
2313         struct btrfs_ioctl_search_args __user *uargs;
2314         struct btrfs_ioctl_search_key sk;
2315         struct inode *inode;
2316         int ret;
2317         size_t buf_size;
2318
2319         if (!capable(CAP_SYS_ADMIN))
2320                 return -EPERM;
2321
2322         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2323
2324         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2325                 return -EFAULT;
2326
2327         buf_size = sizeof(uargs->buf);
2328
2329         inode = file_inode(file);
2330         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2331
2332         /*
2333          * In the origin implementation an overflow is handled by returning a
2334          * search header with a len of zero, so reset ret.
2335          */
2336         if (ret == -EOVERFLOW)
2337                 ret = 0;
2338
2339         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2340                 ret = -EFAULT;
2341         return ret;
2342 }
2343
2344 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2345                                                void __user *argp)
2346 {
2347         struct btrfs_ioctl_search_args_v2 __user *uarg;
2348         struct btrfs_ioctl_search_args_v2 args;
2349         struct inode *inode;
2350         int ret;
2351         size_t buf_size;
2352         const size_t buf_limit = SZ_16M;
2353
2354         if (!capable(CAP_SYS_ADMIN))
2355                 return -EPERM;
2356
2357         /* copy search header and buffer size */
2358         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2359         if (copy_from_user(&args, uarg, sizeof(args)))
2360                 return -EFAULT;
2361
2362         buf_size = args.buf_size;
2363
2364         /* limit result size to 16MB */
2365         if (buf_size > buf_limit)
2366                 buf_size = buf_limit;
2367
2368         inode = file_inode(file);
2369         ret = search_ioctl(inode, &args.key, &buf_size,
2370                            (char __user *)(&uarg->buf[0]));
2371         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2372                 ret = -EFAULT;
2373         else if (ret == -EOVERFLOW &&
2374                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2375                 ret = -EFAULT;
2376
2377         return ret;
2378 }
2379
2380 /*
2381  * Search INODE_REFs to identify path name of 'dirid' directory
2382  * in a 'tree_id' tree. and sets path name to 'name'.
2383  */
2384 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2385                                 u64 tree_id, u64 dirid, char *name)
2386 {
2387         struct btrfs_root *root;
2388         struct btrfs_key key;
2389         char *ptr;
2390         int ret = -1;
2391         int slot;
2392         int len;
2393         int total_len = 0;
2394         struct btrfs_inode_ref *iref;
2395         struct extent_buffer *l;
2396         struct btrfs_path *path;
2397
2398         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2399                 name[0]='\0';
2400                 return 0;
2401         }
2402
2403         path = btrfs_alloc_path();
2404         if (!path)
2405                 return -ENOMEM;
2406
2407         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2408
2409         key.objectid = tree_id;
2410         key.type = BTRFS_ROOT_ITEM_KEY;
2411         key.offset = (u64)-1;
2412         root = btrfs_read_fs_root_no_name(info, &key);
2413         if (IS_ERR(root)) {
2414                 ret = PTR_ERR(root);
2415                 goto out;
2416         }
2417
2418         key.objectid = dirid;
2419         key.type = BTRFS_INODE_REF_KEY;
2420         key.offset = (u64)-1;
2421
2422         while (1) {
2423                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2424                 if (ret < 0)
2425                         goto out;
2426                 else if (ret > 0) {
2427                         ret = btrfs_previous_item(root, path, dirid,
2428                                                   BTRFS_INODE_REF_KEY);
2429                         if (ret < 0)
2430                                 goto out;
2431                         else if (ret > 0) {
2432                                 ret = -ENOENT;
2433                                 goto out;
2434                         }
2435                 }
2436
2437                 l = path->nodes[0];
2438                 slot = path->slots[0];
2439                 btrfs_item_key_to_cpu(l, &key, slot);
2440
2441                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2442                 len = btrfs_inode_ref_name_len(l, iref);
2443                 ptr -= len + 1;
2444                 total_len += len + 1;
2445                 if (ptr < name) {
2446                         ret = -ENAMETOOLONG;
2447                         goto out;
2448                 }
2449
2450                 *(ptr + len) = '/';
2451                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2452
2453                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2454                         break;
2455
2456                 btrfs_release_path(path);
2457                 key.objectid = key.offset;
2458                 key.offset = (u64)-1;
2459                 dirid = key.objectid;
2460         }
2461         memmove(name, ptr, total_len);
2462         name[total_len] = '\0';
2463         ret = 0;
2464 out:
2465         btrfs_free_path(path);
2466         return ret;
2467 }
2468
2469 static int btrfs_search_path_in_tree_user(struct inode *inode,
2470                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2471 {
2472         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2473         struct super_block *sb = inode->i_sb;
2474         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2475         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2476         u64 dirid = args->dirid;
2477         unsigned long item_off;
2478         unsigned long item_len;
2479         struct btrfs_inode_ref *iref;
2480         struct btrfs_root_ref *rref;
2481         struct btrfs_root *root;
2482         struct btrfs_path *path;
2483         struct btrfs_key key, key2;
2484         struct extent_buffer *leaf;
2485         struct inode *temp_inode;
2486         char *ptr;
2487         int slot;
2488         int len;
2489         int total_len = 0;
2490         int ret;
2491
2492         path = btrfs_alloc_path();
2493         if (!path)
2494                 return -ENOMEM;
2495
2496         /*
2497          * If the bottom subvolume does not exist directly under upper_limit,
2498          * construct the path in from the bottom up.
2499          */
2500         if (dirid != upper_limit.objectid) {
2501                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2502
2503                 key.objectid = treeid;
2504                 key.type = BTRFS_ROOT_ITEM_KEY;
2505                 key.offset = (u64)-1;
2506                 root = btrfs_read_fs_root_no_name(fs_info, &key);
2507                 if (IS_ERR(root)) {
2508                         ret = PTR_ERR(root);
2509                         goto out;
2510                 }
2511
2512                 key.objectid = dirid;
2513                 key.type = BTRFS_INODE_REF_KEY;
2514                 key.offset = (u64)-1;
2515                 while (1) {
2516                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2517                         if (ret < 0) {
2518                                 goto out;
2519                         } else if (ret > 0) {
2520                                 ret = btrfs_previous_item(root, path, dirid,
2521                                                           BTRFS_INODE_REF_KEY);
2522                                 if (ret < 0) {
2523                                         goto out;
2524                                 } else if (ret > 0) {
2525                                         ret = -ENOENT;
2526                                         goto out;
2527                                 }
2528                         }
2529
2530                         leaf = path->nodes[0];
2531                         slot = path->slots[0];
2532                         btrfs_item_key_to_cpu(leaf, &key, slot);
2533
2534                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2535                         len = btrfs_inode_ref_name_len(leaf, iref);
2536                         ptr -= len + 1;
2537                         total_len += len + 1;
2538                         if (ptr < args->path) {
2539                                 ret = -ENAMETOOLONG;
2540                                 goto out;
2541                         }
2542
2543                         *(ptr + len) = '/';
2544                         read_extent_buffer(leaf, ptr,
2545                                         (unsigned long)(iref + 1), len);
2546
2547                         /* Check the read+exec permission of this directory */
2548                         ret = btrfs_previous_item(root, path, dirid,
2549                                                   BTRFS_INODE_ITEM_KEY);
2550                         if (ret < 0) {
2551                                 goto out;
2552                         } else if (ret > 0) {
2553                                 ret = -ENOENT;
2554                                 goto out;
2555                         }
2556
2557                         leaf = path->nodes[0];
2558                         slot = path->slots[0];
2559                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2560                         if (key2.objectid != dirid) {
2561                                 ret = -ENOENT;
2562                                 goto out;
2563                         }
2564
2565                         temp_inode = btrfs_iget(sb, &key2, root, NULL);
2566                         if (IS_ERR(temp_inode)) {
2567                                 ret = PTR_ERR(temp_inode);
2568                                 goto out;
2569                         }
2570                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2571                         iput(temp_inode);
2572                         if (ret) {
2573                                 ret = -EACCES;
2574                                 goto out;
2575                         }
2576
2577                         if (key.offset == upper_limit.objectid)
2578                                 break;
2579                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2580                                 ret = -EACCES;
2581                                 goto out;
2582                         }
2583
2584                         btrfs_release_path(path);
2585                         key.objectid = key.offset;
2586                         key.offset = (u64)-1;
2587                         dirid = key.objectid;
2588                 }
2589
2590                 memmove(args->path, ptr, total_len);
2591                 args->path[total_len] = '\0';
2592                 btrfs_release_path(path);
2593         }
2594
2595         /* Get the bottom subvolume's name from ROOT_REF */
2596         root = fs_info->tree_root;
2597         key.objectid = treeid;
2598         key.type = BTRFS_ROOT_REF_KEY;
2599         key.offset = args->treeid;
2600         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2601         if (ret < 0) {
2602                 goto out;
2603         } else if (ret > 0) {
2604                 ret = -ENOENT;
2605                 goto out;
2606         }
2607
2608         leaf = path->nodes[0];
2609         slot = path->slots[0];
2610         btrfs_item_key_to_cpu(leaf, &key, slot);
2611
2612         item_off = btrfs_item_ptr_offset(leaf, slot);
2613         item_len = btrfs_item_size_nr(leaf, slot);
2614         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2615         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2616         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2617                 ret = -EINVAL;
2618                 goto out;
2619         }
2620
2621         /* Copy subvolume's name */
2622         item_off += sizeof(struct btrfs_root_ref);
2623         item_len -= sizeof(struct btrfs_root_ref);
2624         read_extent_buffer(leaf, args->name, item_off, item_len);
2625         args->name[item_len] = 0;
2626
2627 out:
2628         btrfs_free_path(path);
2629         return ret;
2630 }
2631
2632 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2633                                            void __user *argp)
2634 {
2635         struct btrfs_ioctl_ino_lookup_args *args;
2636         struct inode *inode;
2637         int ret = 0;
2638
2639         args = memdup_user(argp, sizeof(*args));
2640         if (IS_ERR(args))
2641                 return PTR_ERR(args);
2642
2643         inode = file_inode(file);
2644
2645         /*
2646          * Unprivileged query to obtain the containing subvolume root id. The
2647          * path is reset so it's consistent with btrfs_search_path_in_tree.
2648          */
2649         if (args->treeid == 0)
2650                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2651
2652         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2653                 args->name[0] = 0;
2654                 goto out;
2655         }
2656
2657         if (!capable(CAP_SYS_ADMIN)) {
2658                 ret = -EPERM;
2659                 goto out;
2660         }
2661
2662         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2663                                         args->treeid, args->objectid,
2664                                         args->name);
2665
2666 out:
2667         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2668                 ret = -EFAULT;
2669
2670         kfree(args);
2671         return ret;
2672 }
2673
2674 /*
2675  * Version of ino_lookup ioctl (unprivileged)
2676  *
2677  * The main differences from ino_lookup ioctl are:
2678  *
2679  *   1. Read + Exec permission will be checked using inode_permission() during
2680  *      path construction. -EACCES will be returned in case of failure.
2681  *   2. Path construction will be stopped at the inode number which corresponds
2682  *      to the fd with which this ioctl is called. If constructed path does not
2683  *      exist under fd's inode, -EACCES will be returned.
2684  *   3. The name of bottom subvolume is also searched and filled.
2685  */
2686 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2687 {
2688         struct btrfs_ioctl_ino_lookup_user_args *args;
2689         struct inode *inode;
2690         int ret;
2691
2692         args = memdup_user(argp, sizeof(*args));
2693         if (IS_ERR(args))
2694                 return PTR_ERR(args);
2695
2696         inode = file_inode(file);
2697
2698         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2699             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2700                 /*
2701                  * The subvolume does not exist under fd with which this is
2702                  * called
2703                  */
2704                 kfree(args);
2705                 return -EACCES;
2706         }
2707
2708         ret = btrfs_search_path_in_tree_user(inode, args);
2709
2710         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2711                 ret = -EFAULT;
2712
2713         kfree(args);
2714         return ret;
2715 }
2716
2717 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2718 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2719 {
2720         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2721         struct btrfs_fs_info *fs_info;
2722         struct btrfs_root *root;
2723         struct btrfs_path *path;
2724         struct btrfs_key key;
2725         struct btrfs_root_item *root_item;
2726         struct btrfs_root_ref *rref;
2727         struct extent_buffer *leaf;
2728         unsigned long item_off;
2729         unsigned long item_len;
2730         struct inode *inode;
2731         int slot;
2732         int ret = 0;
2733
2734         path = btrfs_alloc_path();
2735         if (!path)
2736                 return -ENOMEM;
2737
2738         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2739         if (!subvol_info) {
2740                 btrfs_free_path(path);
2741                 return -ENOMEM;
2742         }
2743
2744         inode = file_inode(file);
2745         fs_info = BTRFS_I(inode)->root->fs_info;
2746
2747         /* Get root_item of inode's subvolume */
2748         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2749         key.type = BTRFS_ROOT_ITEM_KEY;
2750         key.offset = (u64)-1;
2751         root = btrfs_read_fs_root_no_name(fs_info, &key);
2752         if (IS_ERR(root)) {
2753                 ret = PTR_ERR(root);
2754                 goto out;
2755         }
2756         root_item = &root->root_item;
2757
2758         subvol_info->treeid = key.objectid;
2759
2760         subvol_info->generation = btrfs_root_generation(root_item);
2761         subvol_info->flags = btrfs_root_flags(root_item);
2762
2763         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2764         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2765                                                     BTRFS_UUID_SIZE);
2766         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2767                                                     BTRFS_UUID_SIZE);
2768
2769         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2770         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2771         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2772
2773         subvol_info->otransid = btrfs_root_otransid(root_item);
2774         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2775         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2776
2777         subvol_info->stransid = btrfs_root_stransid(root_item);
2778         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2779         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2780
2781         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2782         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2783         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2784
2785         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2786                 /* Search root tree for ROOT_BACKREF of this subvolume */
2787                 root = fs_info->tree_root;
2788
2789                 key.type = BTRFS_ROOT_BACKREF_KEY;
2790                 key.offset = 0;
2791                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2792                 if (ret < 0) {
2793                         goto out;
2794                 } else if (path->slots[0] >=
2795                            btrfs_header_nritems(path->nodes[0])) {
2796                         ret = btrfs_next_leaf(root, path);
2797                         if (ret < 0) {
2798                                 goto out;
2799                         } else if (ret > 0) {
2800                                 ret = -EUCLEAN;
2801                                 goto out;
2802                         }
2803                 }
2804
2805                 leaf = path->nodes[0];
2806                 slot = path->slots[0];
2807                 btrfs_item_key_to_cpu(leaf, &key, slot);
2808                 if (key.objectid == subvol_info->treeid &&
2809                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2810                         subvol_info->parent_id = key.offset;
2811
2812                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2813                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2814
2815                         item_off = btrfs_item_ptr_offset(leaf, slot)
2816                                         + sizeof(struct btrfs_root_ref);
2817                         item_len = btrfs_item_size_nr(leaf, slot)
2818                                         - sizeof(struct btrfs_root_ref);
2819                         read_extent_buffer(leaf, subvol_info->name,
2820                                            item_off, item_len);
2821                 } else {
2822                         ret = -ENOENT;
2823                         goto out;
2824                 }
2825         }
2826
2827         btrfs_free_path(path);
2828         path = NULL;
2829         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2830                 ret = -EFAULT;
2831
2832 out:
2833         btrfs_free_path(path);
2834         kzfree(subvol_info);
2835         return ret;
2836 }
2837
2838 /*
2839  * Return ROOT_REF information of the subvolume containing this inode
2840  * except the subvolume name.
2841  */
2842 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2843 {
2844         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2845         struct btrfs_root_ref *rref;
2846         struct btrfs_root *root;
2847         struct btrfs_path *path;
2848         struct btrfs_key key;
2849         struct extent_buffer *leaf;
2850         struct inode *inode;
2851         u64 objectid;
2852         int slot;
2853         int ret;
2854         u8 found;
2855
2856         path = btrfs_alloc_path();
2857         if (!path)
2858                 return -ENOMEM;
2859
2860         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2861         if (IS_ERR(rootrefs)) {
2862                 btrfs_free_path(path);
2863                 return PTR_ERR(rootrefs);
2864         }
2865
2866         inode = file_inode(file);
2867         root = BTRFS_I(inode)->root->fs_info->tree_root;
2868         objectid = BTRFS_I(inode)->root->root_key.objectid;
2869
2870         key.objectid = objectid;
2871         key.type = BTRFS_ROOT_REF_KEY;
2872         key.offset = rootrefs->min_treeid;
2873         found = 0;
2874
2875         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2876         if (ret < 0) {
2877                 goto out;
2878         } else if (path->slots[0] >=
2879                    btrfs_header_nritems(path->nodes[0])) {
2880                 ret = btrfs_next_leaf(root, path);
2881                 if (ret < 0) {
2882                         goto out;
2883                 } else if (ret > 0) {
2884                         ret = -EUCLEAN;
2885                         goto out;
2886                 }
2887         }
2888         while (1) {
2889                 leaf = path->nodes[0];
2890                 slot = path->slots[0];
2891
2892                 btrfs_item_key_to_cpu(leaf, &key, slot);
2893                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2894                         ret = 0;
2895                         goto out;
2896                 }
2897
2898                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2899                         ret = -EOVERFLOW;
2900                         goto out;
2901                 }
2902
2903                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2904                 rootrefs->rootref[found].treeid = key.offset;
2905                 rootrefs->rootref[found].dirid =
2906                                   btrfs_root_ref_dirid(leaf, rref);
2907                 found++;
2908
2909                 ret = btrfs_next_item(root, path);
2910                 if (ret < 0) {
2911                         goto out;
2912                 } else if (ret > 0) {
2913                         ret = -EUCLEAN;
2914                         goto out;
2915                 }
2916         }
2917
2918 out:
2919         btrfs_free_path(path);
2920
2921         if (!ret || ret == -EOVERFLOW) {
2922                 rootrefs->num_items = found;
2923                 /* update min_treeid for next search */
2924                 if (found)
2925                         rootrefs->min_treeid =
2926                                 rootrefs->rootref[found - 1].treeid + 1;
2927                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2928                         ret = -EFAULT;
2929         }
2930
2931         kfree(rootrefs);
2932
2933         return ret;
2934 }
2935
2936 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2937                                              void __user *arg)
2938 {
2939         struct dentry *parent = file->f_path.dentry;
2940         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2941         struct dentry *dentry;
2942         struct inode *dir = d_inode(parent);
2943         struct inode *inode;
2944         struct btrfs_root *root = BTRFS_I(dir)->root;
2945         struct btrfs_root *dest = NULL;
2946         struct btrfs_ioctl_vol_args *vol_args;
2947         int namelen;
2948         int err = 0;
2949
2950         if (!S_ISDIR(dir->i_mode))
2951                 return -ENOTDIR;
2952
2953         vol_args = memdup_user(arg, sizeof(*vol_args));
2954         if (IS_ERR(vol_args))
2955                 return PTR_ERR(vol_args);
2956
2957         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2958         namelen = strlen(vol_args->name);
2959         if (strchr(vol_args->name, '/') ||
2960             strncmp(vol_args->name, "..", namelen) == 0) {
2961                 err = -EINVAL;
2962                 goto out;
2963         }
2964
2965         err = mnt_want_write_file(file);
2966         if (err)
2967                 goto out;
2968
2969
2970         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2971         if (err == -EINTR)
2972                 goto out_drop_write;
2973         dentry = lookup_one_len(vol_args->name, parent, namelen);
2974         if (IS_ERR(dentry)) {
2975                 err = PTR_ERR(dentry);
2976                 goto out_unlock_dir;
2977         }
2978
2979         if (d_really_is_negative(dentry)) {
2980                 err = -ENOENT;
2981                 goto out_dput;
2982         }
2983
2984         inode = d_inode(dentry);
2985         dest = BTRFS_I(inode)->root;
2986         if (!capable(CAP_SYS_ADMIN)) {
2987                 /*
2988                  * Regular user.  Only allow this with a special mount
2989                  * option, when the user has write+exec access to the
2990                  * subvol root, and when rmdir(2) would have been
2991                  * allowed.
2992                  *
2993                  * Note that this is _not_ check that the subvol is
2994                  * empty or doesn't contain data that we wouldn't
2995                  * otherwise be able to delete.
2996                  *
2997                  * Users who want to delete empty subvols should try
2998                  * rmdir(2).
2999                  */
3000                 err = -EPERM;
3001                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3002                         goto out_dput;
3003
3004                 /*
3005                  * Do not allow deletion if the parent dir is the same
3006                  * as the dir to be deleted.  That means the ioctl
3007                  * must be called on the dentry referencing the root
3008                  * of the subvol, not a random directory contained
3009                  * within it.
3010                  */
3011                 err = -EINVAL;
3012                 if (root == dest)
3013                         goto out_dput;
3014
3015                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
3016                 if (err)
3017                         goto out_dput;
3018         }
3019
3020         /* check if subvolume may be deleted by a user */
3021         err = btrfs_may_delete(dir, dentry, 1);
3022         if (err)
3023                 goto out_dput;
3024
3025         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3026                 err = -EINVAL;
3027                 goto out_dput;
3028         }
3029
3030         inode_lock(inode);
3031         err = btrfs_delete_subvolume(dir, dentry);
3032         inode_unlock(inode);
3033         if (!err)
3034                 d_delete_notify(dir, dentry);
3035
3036 out_dput:
3037         dput(dentry);
3038 out_unlock_dir:
3039         inode_unlock(dir);
3040 out_drop_write:
3041         mnt_drop_write_file(file);
3042 out:
3043         kfree(vol_args);
3044         return err;
3045 }
3046
3047 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3048 {
3049         struct inode *inode = file_inode(file);
3050         struct btrfs_root *root = BTRFS_I(inode)->root;
3051         struct btrfs_ioctl_defrag_range_args *range;
3052         int ret;
3053
3054         ret = mnt_want_write_file(file);
3055         if (ret)
3056                 return ret;
3057
3058         if (btrfs_root_readonly(root)) {
3059                 ret = -EROFS;
3060                 goto out;
3061         }
3062
3063         switch (inode->i_mode & S_IFMT) {
3064         case S_IFDIR:
3065                 if (!capable(CAP_SYS_ADMIN)) {
3066                         ret = -EPERM;
3067                         goto out;
3068                 }
3069                 ret = btrfs_defrag_root(root);
3070                 break;
3071         case S_IFREG:
3072                 /*
3073                  * Note that this does not check the file descriptor for write
3074                  * access. This prevents defragmenting executables that are
3075                  * running and allows defrag on files open in read-only mode.
3076                  */
3077                 if (!capable(CAP_SYS_ADMIN) &&
3078                     inode_permission(inode, MAY_WRITE)) {
3079                         ret = -EPERM;
3080                         goto out;
3081                 }
3082
3083                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3084                 if (!range) {
3085                         ret = -ENOMEM;
3086                         goto out;
3087                 }
3088
3089                 if (argp) {
3090                         if (copy_from_user(range, argp,
3091                                            sizeof(*range))) {
3092                                 ret = -EFAULT;
3093                                 kfree(range);
3094                                 goto out;
3095                         }
3096                         /* compression requires us to start the IO */
3097                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3098                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3099                                 range->extent_thresh = (u32)-1;
3100                         }
3101                 } else {
3102                         /* the rest are all set to zero by kzalloc */
3103                         range->len = (u64)-1;
3104                 }
3105                 ret = btrfs_defrag_file(file_inode(file), file,
3106                                         range, BTRFS_OLDEST_GENERATION, 0);
3107                 if (ret > 0)
3108                         ret = 0;
3109                 kfree(range);
3110                 break;
3111         default:
3112                 ret = -EINVAL;
3113         }
3114 out:
3115         mnt_drop_write_file(file);
3116         return ret;
3117 }
3118
3119 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3120 {
3121         struct btrfs_ioctl_vol_args *vol_args;
3122         int ret;
3123
3124         if (!capable(CAP_SYS_ADMIN))
3125                 return -EPERM;
3126
3127         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3128                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3129
3130         vol_args = memdup_user(arg, sizeof(*vol_args));
3131         if (IS_ERR(vol_args)) {
3132                 ret = PTR_ERR(vol_args);
3133                 goto out;
3134         }
3135
3136         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3137         ret = btrfs_init_new_device(fs_info, vol_args->name);
3138
3139         if (!ret)
3140                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3141
3142         kfree(vol_args);
3143 out:
3144         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3145         return ret;
3146 }
3147
3148 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3149 {
3150         struct inode *inode = file_inode(file);
3151         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3152         struct btrfs_ioctl_vol_args_v2 *vol_args;
3153         int ret;
3154
3155         if (!capable(CAP_SYS_ADMIN))
3156                 return -EPERM;
3157
3158         ret = mnt_want_write_file(file);
3159         if (ret)
3160                 return ret;
3161
3162         vol_args = memdup_user(arg, sizeof(*vol_args));
3163         if (IS_ERR(vol_args)) {
3164                 ret = PTR_ERR(vol_args);
3165                 goto err_drop;
3166         }
3167
3168         /* Check for compatibility reject unknown flags */
3169         if (vol_args->flags & ~BTRFS_VOL_ARG_V2_FLAGS_SUPPORTED) {
3170                 ret = -EOPNOTSUPP;
3171                 goto out;
3172         }
3173
3174         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3175                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3176                 goto out;
3177         }
3178
3179         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3180                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3181         } else {
3182                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3183                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3184         }
3185         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3186
3187         if (!ret) {
3188                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3189                         btrfs_info(fs_info, "device deleted: id %llu",
3190                                         vol_args->devid);
3191                 else
3192                         btrfs_info(fs_info, "device deleted: %s",
3193                                         vol_args->name);
3194         }
3195 out:
3196         kfree(vol_args);
3197 err_drop:
3198         mnt_drop_write_file(file);
3199         return ret;
3200 }
3201
3202 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3203 {
3204         struct inode *inode = file_inode(file);
3205         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3206         struct btrfs_ioctl_vol_args *vol_args;
3207         int ret;
3208
3209         if (!capable(CAP_SYS_ADMIN))
3210                 return -EPERM;
3211
3212         ret = mnt_want_write_file(file);
3213         if (ret)
3214                 return ret;
3215
3216         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3217                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3218                 goto out_drop_write;
3219         }
3220
3221         vol_args = memdup_user(arg, sizeof(*vol_args));
3222         if (IS_ERR(vol_args)) {
3223                 ret = PTR_ERR(vol_args);
3224                 goto out;
3225         }
3226
3227         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3228         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3229
3230         if (!ret)
3231                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3232         kfree(vol_args);
3233 out:
3234         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3235 out_drop_write:
3236         mnt_drop_write_file(file);
3237
3238         return ret;
3239 }
3240
3241 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3242                                 void __user *arg)
3243 {
3244         struct btrfs_ioctl_fs_info_args *fi_args;
3245         struct btrfs_device *device;
3246         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3247         int ret = 0;
3248
3249         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
3250         if (!fi_args)
3251                 return -ENOMEM;
3252
3253         rcu_read_lock();
3254         fi_args->num_devices = fs_devices->num_devices;
3255
3256         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3257                 if (device->devid > fi_args->max_id)
3258                         fi_args->max_id = device->devid;
3259         }
3260         rcu_read_unlock();
3261
3262         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3263         fi_args->nodesize = fs_info->nodesize;
3264         fi_args->sectorsize = fs_info->sectorsize;
3265         fi_args->clone_alignment = fs_info->sectorsize;
3266
3267         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3268                 ret = -EFAULT;
3269
3270         kfree(fi_args);
3271         return ret;
3272 }
3273
3274 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3275                                  void __user *arg)
3276 {
3277         struct btrfs_ioctl_dev_info_args *di_args;
3278         struct btrfs_device *dev;
3279         int ret = 0;
3280         char *s_uuid = NULL;
3281
3282         di_args = memdup_user(arg, sizeof(*di_args));
3283         if (IS_ERR(di_args))
3284                 return PTR_ERR(di_args);
3285
3286         if (!btrfs_is_empty_uuid(di_args->uuid))
3287                 s_uuid = di_args->uuid;
3288
3289         rcu_read_lock();
3290         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3291                                 NULL, true);
3292
3293         if (!dev) {
3294                 ret = -ENODEV;
3295                 goto out;
3296         }
3297
3298         di_args->devid = dev->devid;
3299         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3300         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3301         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3302         if (dev->name)
3303                 strscpy(di_args->path, rcu_str_deref(dev->name), sizeof(di_args->path));
3304         else
3305                 di_args->path[0] = '\0';
3306
3307 out:
3308         rcu_read_unlock();
3309         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3310                 ret = -EFAULT;
3311
3312         kfree(di_args);
3313         return ret;
3314 }
3315
3316 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
3317                                        struct inode *inode2, u64 loff2, u64 len)
3318 {
3319         unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3320         unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3321 }
3322
3323 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
3324                                      struct inode *inode2, u64 loff2, u64 len)
3325 {
3326         if (inode1 < inode2) {
3327                 swap(inode1, inode2);
3328                 swap(loff1, loff2);
3329         } else if (inode1 == inode2 && loff2 < loff1) {
3330                 swap(loff1, loff2);
3331         }
3332         lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
3333         lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
3334 }
3335
3336 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
3337                                    struct inode *dst, u64 dst_loff)
3338 {
3339         const u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
3340         int ret;
3341
3342         /*
3343          * Lock destination range to serialize with concurrent readpages() and
3344          * source range to serialize with relocation.
3345          */
3346         btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
3347         ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1);
3348         btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
3349
3350         return ret;
3351 }
3352
3353 #define BTRFS_MAX_DEDUPE_LEN    SZ_16M
3354
3355 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
3356                              struct inode *dst, u64 dst_loff)
3357 {
3358         int ret;
3359         u64 i, tail_len, chunk_count;
3360         struct btrfs_root *root_dst = BTRFS_I(dst)->root;
3361
3362         spin_lock(&root_dst->root_item_lock);
3363         if (root_dst->send_in_progress) {
3364                 btrfs_warn_rl(root_dst->fs_info,
3365 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
3366                               root_dst->root_key.objectid,
3367                               root_dst->send_in_progress);
3368                 spin_unlock(&root_dst->root_item_lock);
3369                 return -EAGAIN;
3370         }
3371         root_dst->dedupe_in_progress++;
3372         spin_unlock(&root_dst->root_item_lock);
3373
3374         tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
3375         chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
3376
3377         for (i = 0; i < chunk_count; i++) {
3378                 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
3379                                               dst, dst_loff);
3380                 if (ret)
3381                         goto out;
3382
3383                 loff += BTRFS_MAX_DEDUPE_LEN;
3384                 dst_loff += BTRFS_MAX_DEDUPE_LEN;
3385         }
3386
3387         if (tail_len > 0)
3388                 ret = btrfs_extent_same_range(src, loff, tail_len, dst,
3389                                               dst_loff);
3390 out:
3391         spin_lock(&root_dst->root_item_lock);
3392         root_dst->dedupe_in_progress--;
3393         spin_unlock(&root_dst->root_item_lock);
3394
3395         return ret;
3396 }
3397
3398 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
3399                                      struct inode *inode,
3400                                      u64 endoff,
3401                                      const u64 destoff,
3402                                      const u64 olen,
3403                                      int no_time_update)
3404 {
3405         struct btrfs_root *root = BTRFS_I(inode)->root;
3406         int ret;
3407
3408         inode_inc_iversion(inode);
3409         if (!no_time_update)
3410                 inode->i_mtime = inode->i_ctime = current_time(inode);
3411         /*
3412          * We round up to the block size at eof when determining which
3413          * extents to clone above, but shouldn't round up the file size.
3414          */
3415         if (endoff > destoff + olen)
3416                 endoff = destoff + olen;
3417         if (endoff > inode->i_size)
3418                 btrfs_i_size_write(BTRFS_I(inode), endoff);
3419
3420         ret = btrfs_update_inode(trans, root, inode);
3421         if (ret) {
3422                 btrfs_abort_transaction(trans, ret);
3423                 btrfs_end_transaction(trans);
3424                 goto out;
3425         }
3426         ret = btrfs_end_transaction(trans);
3427 out:
3428         return ret;
3429 }
3430
3431 /*
3432  * Make sure we do not end up inserting an inline extent into a file that has
3433  * already other (non-inline) extents. If a file has an inline extent it can
3434  * not have any other extents and the (single) inline extent must start at the
3435  * file offset 0. Failing to respect these rules will lead to file corruption,
3436  * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc
3437  *
3438  * We can have extents that have been already written to disk or we can have
3439  * dirty ranges still in delalloc, in which case the extent maps and items are
3440  * created only when we run delalloc, and the delalloc ranges might fall outside
3441  * the range we are currently locking in the inode's io tree. So we check the
3442  * inode's i_size because of that (i_size updates are done while holding the
3443  * i_mutex, which we are holding here).
3444  * We also check to see if the inode has a size not greater than "datal" but has
3445  * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are
3446  * protected against such concurrent fallocate calls by the i_mutex).
3447  *
3448  * If the file has no extents but a size greater than datal, do not allow the
3449  * copy because we would need turn the inline extent into a non-inline one (even
3450  * with NO_HOLES enabled). If we find our destination inode only has one inline
3451  * extent, just overwrite it with the source inline extent if its size is less
3452  * than the source extent's size, or we could copy the source inline extent's
3453  * data into the destination inode's inline extent if the later is greater then
3454  * the former.
3455  */
3456 static int clone_copy_inline_extent(struct inode *dst,
3457                                     struct btrfs_trans_handle *trans,
3458                                     struct btrfs_path *path,
3459                                     struct btrfs_key *new_key,
3460                                     const u64 drop_start,
3461                                     const u64 datal,
3462                                     const u64 skip,
3463                                     const u64 size,
3464                                     char *inline_data)
3465 {
3466         struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
3467         struct btrfs_root *root = BTRFS_I(dst)->root;
3468         const u64 aligned_end = ALIGN(new_key->offset + datal,
3469                                       fs_info->sectorsize);
3470         int ret;
3471         struct btrfs_key key;
3472
3473         if (new_key->offset > 0)
3474                 return -EOPNOTSUPP;
3475
3476         key.objectid = btrfs_ino(BTRFS_I(dst));
3477         key.type = BTRFS_EXTENT_DATA_KEY;
3478         key.offset = 0;
3479         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3480         if (ret < 0) {
3481                 return ret;
3482         } else if (ret > 0) {
3483                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3484                         ret = btrfs_next_leaf(root, path);
3485                         if (ret < 0)
3486                                 return ret;
3487                         else if (ret > 0)
3488                                 goto copy_inline_extent;
3489                 }
3490                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3491                 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3492                     key.type == BTRFS_EXTENT_DATA_KEY) {
3493                         ASSERT(key.offset > 0);
3494                         return -EOPNOTSUPP;
3495                 }
3496         } else if (i_size_read(dst) <= datal) {
3497                 struct btrfs_file_extent_item *ei;
3498                 u64 ext_len;
3499
3500                 /*
3501                  * If the file size is <= datal, make sure there are no other
3502                  * extents following (can happen do to an fallocate call with
3503                  * the flag FALLOC_FL_KEEP_SIZE).
3504                  */
3505                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3506                                     struct btrfs_file_extent_item);
3507                 /*
3508                  * If it's an inline extent, it can not have other extents
3509                  * following it.
3510                  */
3511                 if (btrfs_file_extent_type(path->nodes[0], ei) ==
3512                     BTRFS_FILE_EXTENT_INLINE)
3513                         goto copy_inline_extent;
3514
3515                 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3516                 if (ext_len > aligned_end)
3517                         return -EOPNOTSUPP;
3518
3519                 ret = btrfs_next_item(root, path);
3520                 if (ret < 0) {
3521                         return ret;
3522                 } else if (ret == 0) {
3523                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3524                                               path->slots[0]);
3525                         if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
3526                             key.type == BTRFS_EXTENT_DATA_KEY)
3527                                 return -EOPNOTSUPP;
3528                 }
3529         }
3530
3531 copy_inline_extent:
3532         /*
3533          * We have no extent items, or we have an extent at offset 0 which may
3534          * or may not be inlined. All these cases are dealt the same way.
3535          */
3536         if (i_size_read(dst) > datal) {
3537                 /*
3538                  * If the destination inode has an inline extent...
3539                  * This would require copying the data from the source inline
3540                  * extent into the beginning of the destination's inline extent.
3541                  * But this is really complex, both extents can be compressed
3542                  * or just one of them, which would require decompressing and
3543                  * re-compressing data (which could increase the new compressed
3544                  * size, not allowing the compressed data to fit anymore in an
3545                  * inline extent).
3546                  * So just don't support this case for now (it should be rare,
3547                  * we are not really saving space when cloning inline extents).
3548                  */
3549                 return -EOPNOTSUPP;
3550         }
3551
3552         btrfs_release_path(path);
3553         ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
3554         if (ret)
3555                 return ret;
3556         ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
3557         if (ret)
3558                 return ret;
3559
3560         if (skip) {
3561                 const u32 start = btrfs_file_extent_calc_inline_size(0);
3562
3563                 memmove(inline_data + start, inline_data + start + skip, datal);
3564         }
3565
3566         write_extent_buffer(path->nodes[0], inline_data,
3567                             btrfs_item_ptr_offset(path->nodes[0],
3568                                                   path->slots[0]),
3569                             size);
3570         inode_add_bytes(dst, datal);
3571         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags);
3572
3573         return 0;
3574 }
3575
3576 /**
3577  * btrfs_clone() - clone a range from inode file to another
3578  *
3579  * @src: Inode to clone from
3580  * @inode: Inode to clone to
3581  * @off: Offset within source to start clone from
3582  * @olen: Original length, passed by user, of range to clone
3583  * @olen_aligned: Block-aligned value of olen
3584  * @destoff: Offset within @inode to start clone
3585  * @no_time_update: Whether to update mtime/ctime on the target inode
3586  */
3587 static int btrfs_clone(struct inode *src, struct inode *inode,
3588                        const u64 off, const u64 olen, const u64 olen_aligned,
3589                        const u64 destoff, int no_time_update)
3590 {
3591         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3592         struct btrfs_root *root = BTRFS_I(inode)->root;
3593         struct btrfs_path *path = NULL;
3594         struct extent_buffer *leaf;
3595         struct btrfs_trans_handle *trans;
3596         char *buf = NULL;
3597         struct btrfs_key key;
3598         u32 nritems;
3599         int slot;
3600         int ret;
3601         const u64 len = olen_aligned;
3602         u64 last_dest_end = destoff;
3603
3604         ret = -ENOMEM;
3605         buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
3606         if (!buf)
3607                 return ret;
3608
3609         path = btrfs_alloc_path();
3610         if (!path) {
3611                 kvfree(buf);
3612                 return ret;
3613         }
3614
3615         path->reada = READA_FORWARD;
3616         /* clone data */
3617         key.objectid = btrfs_ino(BTRFS_I(src));
3618         key.type = BTRFS_EXTENT_DATA_KEY;
3619         key.offset = off;
3620
3621         while (1) {
3622                 u64 next_key_min_offset = key.offset + 1;
3623                 struct btrfs_file_extent_item *extent;
3624                 int type;
3625                 u32 size;
3626                 struct btrfs_key new_key;
3627                 u64 disko = 0, diskl = 0;
3628                 u64 datao = 0, datal = 0;
3629                 u8 comp;
3630                 u64 drop_start;
3631
3632                 /*
3633                  * note the key will change type as we walk through the
3634                  * tree.
3635                  */
3636                 path->leave_spinning = 1;
3637                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
3638                                 0, 0);
3639                 if (ret < 0)
3640                         goto out;
3641                 /*
3642                  * First search, if no extent item that starts at offset off was
3643                  * found but the previous item is an extent item, it's possible
3644                  * it might overlap our target range, therefore process it.
3645                  */
3646                 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
3647                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3648                                               path->slots[0] - 1);
3649                         if (key.type == BTRFS_EXTENT_DATA_KEY)
3650                                 path->slots[0]--;
3651                 }
3652
3653                 nritems = btrfs_header_nritems(path->nodes[0]);
3654 process_slot:
3655                 if (path->slots[0] >= nritems) {
3656                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
3657                         if (ret < 0)
3658                                 goto out;
3659                         if (ret > 0)
3660                                 break;
3661                         nritems = btrfs_header_nritems(path->nodes[0]);
3662                 }
3663                 leaf = path->nodes[0];
3664                 slot = path->slots[0];
3665
3666                 btrfs_item_key_to_cpu(leaf, &key, slot);
3667                 if (key.type > BTRFS_EXTENT_DATA_KEY ||
3668                     key.objectid != btrfs_ino(BTRFS_I(src)))
3669                         break;
3670
3671                 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
3672
3673                 extent = btrfs_item_ptr(leaf, slot,
3674                                         struct btrfs_file_extent_item);
3675                 comp = btrfs_file_extent_compression(leaf, extent);
3676                 type = btrfs_file_extent_type(leaf, extent);
3677                 if (type == BTRFS_FILE_EXTENT_REG ||
3678                     type == BTRFS_FILE_EXTENT_PREALLOC) {
3679                         disko = btrfs_file_extent_disk_bytenr(leaf, extent);
3680                         diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
3681                         datao = btrfs_file_extent_offset(leaf, extent);
3682                         datal = btrfs_file_extent_num_bytes(leaf, extent);
3683                 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3684                         /* Take upper bound, may be compressed */
3685                         datal = btrfs_file_extent_ram_bytes(leaf, extent);
3686                 }
3687
3688                 /*
3689                  * The first search might have left us at an extent item that
3690                  * ends before our target range's start, can happen if we have
3691                  * holes and NO_HOLES feature enabled.
3692                  */
3693                 if (key.offset + datal <= off) {
3694                         path->slots[0]++;
3695                         goto process_slot;
3696                 } else if (key.offset >= off + len) {
3697                         break;
3698                 }
3699                 next_key_min_offset = key.offset + datal;
3700                 size = btrfs_item_size_nr(leaf, slot);
3701                 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
3702                                    size);
3703
3704                 btrfs_release_path(path);
3705                 path->leave_spinning = 0;
3706
3707                 memcpy(&new_key, &key, sizeof(new_key));
3708                 new_key.objectid = btrfs_ino(BTRFS_I(inode));
3709                 if (off <= key.offset)
3710                         new_key.offset = key.offset + destoff - off;
3711                 else
3712                         new_key.offset = destoff;
3713
3714                 /*
3715                  * Deal with a hole that doesn't have an extent item that
3716                  * represents it (NO_HOLES feature enabled).
3717                  * This hole is either in the middle of the cloning range or at
3718                  * the beginning (fully overlaps it or partially overlaps it).
3719                  */
3720                 if (new_key.offset != last_dest_end)
3721                         drop_start = last_dest_end;
3722                 else
3723                         drop_start = new_key.offset;
3724
3725                 if (type == BTRFS_FILE_EXTENT_REG ||
3726                     type == BTRFS_FILE_EXTENT_PREALLOC) {
3727                         struct btrfs_clone_extent_info clone_info;
3728
3729                         /*
3730                          *    a  | --- range to clone ---|  b
3731                          * | ------------- extent ------------- |
3732                          */
3733
3734                         /* Subtract range b */
3735                         if (key.offset + datal > off + len)
3736                                 datal = off + len - key.offset;
3737
3738                         /* Subtract range a */
3739                         if (off > key.offset) {
3740                                 datao += off - key.offset;
3741                                 datal -= off - key.offset;
3742                         }
3743
3744                         clone_info.disk_offset = disko;
3745                         clone_info.disk_len = diskl;
3746                         clone_info.data_offset = datao;
3747                         clone_info.data_len = datal;
3748                         clone_info.file_offset = new_key.offset;
3749                         clone_info.extent_buf = buf;
3750                         clone_info.item_size = size;
3751                         ret = btrfs_punch_hole_range(inode, path,
3752                                                      drop_start,
3753                                                      new_key.offset + datal - 1,
3754                                                      &clone_info, &trans);
3755                         if (ret)
3756                                 goto out;
3757                 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
3758                         u64 skip = 0;
3759                         u64 trim = 0;
3760
3761                         if (off > key.offset) {
3762                                 skip = off - key.offset;
3763                                 new_key.offset += skip;
3764                         }
3765
3766                         if (key.offset + datal > off + len)
3767                                 trim = key.offset + datal - (off + len);
3768
3769                         if (comp && (skip || trim)) {
3770                                 ret = -EINVAL;
3771                                 goto out;
3772                         }
3773                         size -= skip + trim;
3774                         datal -= skip + trim;
3775
3776                         /*
3777                          * If our extent is inline, we know we will drop or
3778                          * adjust at most 1 extent item in the destination root.
3779                          *
3780                          * 1 - adjusting old extent (we may have to split it)
3781                          * 1 - add new extent
3782                          * 1 - inode update
3783                          */
3784                         trans = btrfs_start_transaction(root, 3);
3785                         if (IS_ERR(trans)) {
3786                                 ret = PTR_ERR(trans);
3787                                 goto out;
3788                         }
3789
3790                         ret = clone_copy_inline_extent(inode, trans, path,
3791                                                        &new_key, drop_start,
3792                                                        datal, skip, size, buf);
3793                         if (ret) {
3794                                 if (ret != -EOPNOTSUPP)
3795                                         btrfs_abort_transaction(trans, ret);
3796                                 btrfs_end_transaction(trans);
3797                                 goto out;
3798                         }
3799                 }
3800
3801                 btrfs_release_path(path);
3802
3803                 last_dest_end = ALIGN(new_key.offset + datal,
3804                                       fs_info->sectorsize);
3805                 ret = clone_finish_inode_update(trans, inode, last_dest_end,
3806                                                 destoff, olen, no_time_update);
3807                 if (ret)
3808                         goto out;
3809                 if (new_key.offset + datal >= destoff + len)
3810                         break;
3811
3812                 btrfs_release_path(path);
3813                 key.offset = next_key_min_offset;
3814
3815                 if (fatal_signal_pending(current)) {
3816                         ret = -EINTR;
3817                         goto out;
3818                 }
3819
3820                 cond_resched();
3821         }
3822         ret = 0;
3823
3824         if (last_dest_end < destoff + len) {
3825                 /*
3826                  * We have an implicit hole that fully or partially overlaps our
3827                  * cloning range at its end. This means that we either have the
3828                  * NO_HOLES feature enabled or the implicit hole happened due to
3829                  * mixing buffered and direct IO writes against this file.
3830                  */
3831                 btrfs_release_path(path);
3832                 path->leave_spinning = 0;
3833
3834                 ret = btrfs_punch_hole_range(inode, path,
3835                                              last_dest_end, destoff + len - 1,
3836                                              NULL, &trans);
3837                 if (ret)
3838                         goto out;
3839
3840                 ret = clone_finish_inode_update(trans, inode, destoff + len,
3841                                                 destoff, olen, no_time_update);
3842         }
3843
3844 out:
3845         btrfs_free_path(path);
3846         kvfree(buf);
3847         return ret;
3848 }
3849
3850 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
3851                                         u64 off, u64 olen, u64 destoff)
3852 {
3853         struct inode *inode = file_inode(file);
3854         struct inode *src = file_inode(file_src);
3855         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3856         int ret;
3857         u64 len = olen;
3858         u64 bs = fs_info->sb->s_blocksize;
3859
3860         /*
3861          * TODO:
3862          * - split compressed inline extents.  annoying: we need to
3863          *   decompress into destination's address_space (the file offset
3864          *   may change, so source mapping won't do), then recompress (or
3865          *   otherwise reinsert) a subrange.
3866          *
3867          * - split destination inode's inline extents.  The inline extents can
3868          *   be either compressed or non-compressed.
3869          */
3870
3871         /*
3872          * VFS's generic_remap_file_range_prep() protects us from cloning the
3873          * eof block into the middle of a file, which would result in corruption
3874          * if the file size is not blocksize aligned. So we don't need to check
3875          * for that case here.
3876          */
3877         if (off + len == src->i_size)
3878                 len = ALIGN(src->i_size, bs) - off;
3879
3880         if (destoff > inode->i_size) {
3881                 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
3882
3883                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
3884                 if (ret)
3885                         return ret;
3886                 /*
3887                  * We may have truncated the last block if the inode's size is
3888                  * not sector size aligned, so we need to wait for writeback to
3889                  * complete before proceeding further, otherwise we can race
3890                  * with cloning and attempt to increment a reference to an
3891                  * extent that no longer exists (writeback completed right after
3892                  * we found the previous extent covering eof and before we
3893                  * attempted to increment its reference count).
3894                  */
3895                 ret = btrfs_wait_ordered_range(inode, wb_start,
3896                                                destoff - wb_start);
3897                 if (ret)
3898                         return ret;
3899         }
3900
3901         /*
3902          * Lock destination range to serialize with concurrent readpages() and
3903          * source range to serialize with relocation.
3904          */
3905         btrfs_double_extent_lock(src, off, inode, destoff, len);
3906         ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
3907         btrfs_double_extent_unlock(src, off, inode, destoff, len);
3908         /*
3909          * Truncate page cache pages so that future reads will see the cloned
3910          * data immediately and not the previous data.
3911          */
3912         truncate_inode_pages_range(&inode->i_data,
3913                                 round_down(destoff, PAGE_SIZE),
3914                                 round_up(destoff + len, PAGE_SIZE) - 1);
3915
3916         return ret;
3917 }
3918
3919 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
3920                                        struct file *file_out, loff_t pos_out,
3921                                        loff_t *len, unsigned int remap_flags)
3922 {
3923         struct inode *inode_in = file_inode(file_in);
3924         struct inode *inode_out = file_inode(file_out);
3925         u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
3926         bool same_inode = inode_out == inode_in;
3927         u64 wb_len;
3928         int ret;
3929
3930         if (!(remap_flags & REMAP_FILE_DEDUP)) {
3931                 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
3932
3933                 if (btrfs_root_readonly(root_out))
3934                         return -EROFS;
3935
3936                 if (file_in->f_path.mnt != file_out->f_path.mnt ||
3937                     inode_in->i_sb != inode_out->i_sb)
3938                         return -EXDEV;
3939         }
3940
3941         /* don't make the dst file partly checksummed */
3942         if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
3943             (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
3944                 return -EINVAL;
3945         }
3946
3947         /*
3948          * Now that the inodes are locked, we need to start writeback ourselves
3949          * and can not rely on the writeback from the VFS's generic helper
3950          * generic_remap_file_range_prep() because:
3951          *
3952          * 1) For compression we must call filemap_fdatawrite_range() range
3953          *    twice (btrfs_fdatawrite_range() does it for us), and the generic
3954          *    helper only calls it once;
3955          *
3956          * 2) filemap_fdatawrite_range(), called by the generic helper only
3957          *    waits for the writeback to complete, i.e. for IO to be done, and
3958          *    not for the ordered extents to complete. We need to wait for them
3959          *    to complete so that new file extent items are in the fs tree.
3960          */
3961         if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
3962                 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
3963         else
3964                 wb_len = ALIGN(*len, bs);
3965
3966         /*
3967          * Since we don't lock ranges, wait for ongoing lockless dio writes (as
3968          * any in progress could create its ordered extents after we wait for
3969          * existing ordered extents below).
3970          */
3971         inode_dio_wait(inode_in);
3972         if (!same_inode)
3973                 inode_dio_wait(inode_out);
3974
3975         /*
3976          * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
3977          *
3978          * Btrfs' back references do not have a block level granularity, they
3979          * work at the whole extent level.
3980          * NOCOW buffered write without data space reserved may not be able
3981          * to fall back to CoW due to lack of data space, thus could cause
3982          * data loss.
3983          *
3984          * Here we take a shortcut by flushing the whole inode, so that all
3985          * nocow write should reach disk as nocow before we increase the
3986          * reference of the extent. We could do better by only flushing NOCOW
3987          * data, but that needs extra accounting.
3988          *
3989          * Also we don't need to check ASYNC_EXTENT, as async extent will be
3990          * CoWed anyway, not affecting nocow part.
3991          */
3992         ret = filemap_flush(inode_in->i_mapping);
3993         if (ret < 0)
3994                 return ret;
3995
3996         ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
3997                                        wb_len);
3998         if (ret < 0)
3999                 return ret;
4000         ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
4001                                        wb_len);
4002         if (ret < 0)
4003                 return ret;
4004
4005         return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
4006                                             len, remap_flags);
4007 }
4008
4009 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
4010                 struct file *dst_file, loff_t destoff, loff_t len,
4011                 unsigned int remap_flags)
4012 {
4013         struct inode *src_inode = file_inode(src_file);
4014         struct inode *dst_inode = file_inode(dst_file);
4015         bool same_inode = dst_inode == src_inode;
4016         int ret;
4017
4018         if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
4019                 return -EINVAL;
4020
4021         if (same_inode)
4022                 inode_lock(src_inode);
4023         else
4024                 lock_two_nondirectories(src_inode, dst_inode);
4025
4026         ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
4027                                           &len, remap_flags);
4028         if (ret < 0 || len == 0)
4029                 goto out_unlock;
4030
4031         if (remap_flags & REMAP_FILE_DEDUP)
4032                 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
4033         else
4034                 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
4035
4036 out_unlock:
4037         if (same_inode)
4038                 inode_unlock(src_inode);
4039         else
4040                 unlock_two_nondirectories(src_inode, dst_inode);
4041
4042         return ret < 0 ? ret : len;
4043 }
4044
4045 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
4046 {
4047         struct inode *inode = file_inode(file);
4048         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4049         struct btrfs_root *root = BTRFS_I(inode)->root;
4050         struct btrfs_root *new_root;
4051         struct btrfs_dir_item *di;
4052         struct btrfs_trans_handle *trans;
4053         struct btrfs_path *path;
4054         struct btrfs_key location;
4055         struct btrfs_disk_key disk_key;
4056         u64 objectid = 0;
4057         u64 dir_id;
4058         int ret;
4059
4060         if (!capable(CAP_SYS_ADMIN))
4061                 return -EPERM;
4062
4063         ret = mnt_want_write_file(file);
4064         if (ret)
4065                 return ret;
4066
4067         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
4068                 ret = -EFAULT;
4069                 goto out;
4070         }
4071
4072         if (!objectid)
4073                 objectid = BTRFS_FS_TREE_OBJECTID;
4074
4075         location.objectid = objectid;
4076         location.type = BTRFS_ROOT_ITEM_KEY;
4077         location.offset = (u64)-1;
4078
4079         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
4080         if (IS_ERR(new_root)) {
4081                 ret = PTR_ERR(new_root);
4082                 goto out;
4083         }
4084         if (!is_fstree(new_root->root_key.objectid)) {
4085                 ret = -ENOENT;
4086                 goto out;
4087         }
4088
4089         path = btrfs_alloc_path();
4090         if (!path) {
4091                 ret = -ENOMEM;
4092                 goto out;
4093         }
4094         path->leave_spinning = 1;
4095
4096         trans = btrfs_start_transaction(root, 1);
4097         if (IS_ERR(trans)) {
4098                 btrfs_free_path(path);
4099                 ret = PTR_ERR(trans);
4100                 goto out;
4101         }
4102
4103         dir_id = btrfs_super_root_dir(fs_info->super_copy);
4104         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
4105                                    dir_id, "default", 7, 1);
4106         if (IS_ERR_OR_NULL(di)) {
4107                 btrfs_free_path(path);
4108                 btrfs_end_transaction(trans);
4109                 btrfs_err(fs_info,
4110                           "Umm, you don't have the default diritem, this isn't going to work");
4111                 ret = -ENOENT;
4112                 goto out;
4113         }
4114
4115         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
4116         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
4117         btrfs_mark_buffer_dirty(path->nodes[0]);
4118         btrfs_free_path(path);
4119
4120         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
4121         btrfs_end_transaction(trans);
4122 out:
4123         mnt_drop_write_file(file);
4124         return ret;
4125 }
4126
4127 static void get_block_group_info(struct list_head *groups_list,
4128                                  struct btrfs_ioctl_space_info *space)
4129 {
4130         struct btrfs_block_group_cache *block_group;
4131
4132         space->total_bytes = 0;
4133         space->used_bytes = 0;
4134         space->flags = 0;
4135         list_for_each_entry(block_group, groups_list, list) {
4136                 space->flags = block_group->flags;
4137                 space->total_bytes += block_group->key.offset;
4138                 space->used_bytes +=
4139                         btrfs_block_group_used(&block_group->item);
4140         }
4141 }
4142
4143 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
4144                                    void __user *arg)
4145 {
4146         struct btrfs_ioctl_space_args space_args;
4147         struct btrfs_ioctl_space_info space;
4148         struct btrfs_ioctl_space_info *dest;
4149         struct btrfs_ioctl_space_info *dest_orig;
4150         struct btrfs_ioctl_space_info __user *user_dest;
4151         struct btrfs_space_info *info;
4152         static const u64 types[] = {
4153                 BTRFS_BLOCK_GROUP_DATA,
4154                 BTRFS_BLOCK_GROUP_SYSTEM,
4155                 BTRFS_BLOCK_GROUP_METADATA,
4156                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
4157         };
4158         int num_types = 4;
4159         int alloc_size;
4160         int ret = 0;
4161         u64 slot_count = 0;
4162         int i, c;
4163
4164         if (copy_from_user(&space_args,
4165                            (struct btrfs_ioctl_space_args __user *)arg,
4166                            sizeof(space_args)))
4167                 return -EFAULT;
4168
4169         for (i = 0; i < num_types; i++) {
4170                 struct btrfs_space_info *tmp;
4171
4172                 info = NULL;
4173                 rcu_read_lock();
4174                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4175                                         list) {
4176                         if (tmp->flags == types[i]) {
4177                                 info = tmp;
4178                                 break;
4179                         }
4180                 }
4181                 rcu_read_unlock();
4182
4183                 if (!info)
4184                         continue;
4185
4186                 down_read(&info->groups_sem);
4187                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4188                         if (!list_empty(&info->block_groups[c]))
4189                                 slot_count++;
4190                 }
4191                 up_read(&info->groups_sem);
4192         }
4193
4194         /*
4195          * Global block reserve, exported as a space_info
4196          */
4197         slot_count++;
4198
4199         /* space_slots == 0 means they are asking for a count */
4200         if (space_args.space_slots == 0) {
4201                 space_args.total_spaces = slot_count;
4202                 goto out;
4203         }
4204
4205         slot_count = min_t(u64, space_args.space_slots, slot_count);
4206
4207         alloc_size = sizeof(*dest) * slot_count;
4208
4209         /* we generally have at most 6 or so space infos, one for each raid
4210          * level.  So, a whole page should be more than enough for everyone
4211          */
4212         if (alloc_size > PAGE_SIZE)
4213                 return -ENOMEM;
4214
4215         space_args.total_spaces = 0;
4216         dest = kmalloc(alloc_size, GFP_KERNEL);
4217         if (!dest)
4218                 return -ENOMEM;
4219         dest_orig = dest;
4220
4221         /* now we have a buffer to copy into */
4222         for (i = 0; i < num_types; i++) {
4223                 struct btrfs_space_info *tmp;
4224
4225                 if (!slot_count)
4226                         break;
4227
4228                 info = NULL;
4229                 rcu_read_lock();
4230                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
4231                                         list) {
4232                         if (tmp->flags == types[i]) {
4233                                 info = tmp;
4234                                 break;
4235                         }
4236                 }
4237                 rcu_read_unlock();
4238
4239                 if (!info)
4240                         continue;
4241                 down_read(&info->groups_sem);
4242                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
4243                         if (!list_empty(&info->block_groups[c])) {
4244                                 get_block_group_info(&info->block_groups[c],
4245                                                      &space);
4246                                 memcpy(dest, &space, sizeof(space));
4247                                 dest++;
4248                                 space_args.total_spaces++;
4249                                 slot_count--;
4250                         }
4251                         if (!slot_count)
4252                                 break;
4253                 }
4254                 up_read(&info->groups_sem);
4255         }
4256
4257         /*
4258          * Add global block reserve
4259          */
4260         if (slot_count) {
4261                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4262
4263                 spin_lock(&block_rsv->lock);
4264                 space.total_bytes = block_rsv->size;
4265                 space.used_bytes = block_rsv->size - block_rsv->reserved;
4266                 spin_unlock(&block_rsv->lock);
4267                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
4268                 memcpy(dest, &space, sizeof(space));
4269                 space_args.total_spaces++;
4270         }
4271
4272         user_dest = (struct btrfs_ioctl_space_info __user *)
4273                 (arg + sizeof(struct btrfs_ioctl_space_args));
4274
4275         if (copy_to_user(user_dest, dest_orig, alloc_size))
4276                 ret = -EFAULT;
4277
4278         kfree(dest_orig);
4279 out:
4280         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
4281                 ret = -EFAULT;
4282
4283         return ret;
4284 }
4285
4286 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
4287                                             void __user *argp)
4288 {
4289         struct btrfs_trans_handle *trans;
4290         u64 transid;
4291         int ret;
4292
4293         trans = btrfs_attach_transaction_barrier(root);
4294         if (IS_ERR(trans)) {
4295                 if (PTR_ERR(trans) != -ENOENT)
4296                         return PTR_ERR(trans);
4297
4298                 /* No running transaction, don't bother */
4299                 transid = root->fs_info->last_trans_committed;
4300                 goto out;
4301         }
4302         transid = trans->transid;
4303         ret = btrfs_commit_transaction_async(trans, 0);
4304         if (ret) {
4305                 btrfs_end_transaction(trans);
4306                 return ret;
4307         }
4308 out:
4309         if (argp)
4310                 if (copy_to_user(argp, &transid, sizeof(transid)))
4311                         return -EFAULT;
4312         return 0;
4313 }
4314
4315 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
4316                                            void __user *argp)
4317 {
4318         u64 transid;
4319
4320         if (argp) {
4321                 if (copy_from_user(&transid, argp, sizeof(transid)))
4322                         return -EFAULT;
4323         } else {
4324                 transid = 0;  /* current trans */
4325         }
4326         return btrfs_wait_for_commit(fs_info, transid);
4327 }
4328
4329 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
4330 {
4331         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
4332         struct btrfs_ioctl_scrub_args *sa;
4333         int ret;
4334
4335         if (!capable(CAP_SYS_ADMIN))
4336                 return -EPERM;
4337
4338         sa = memdup_user(arg, sizeof(*sa));
4339         if (IS_ERR(sa))
4340                 return PTR_ERR(sa);
4341
4342         if (sa->flags & ~BTRFS_SCRUB_SUPPORTED_FLAGS) {
4343                 ret = -EOPNOTSUPP;
4344                 goto out;
4345         }
4346
4347         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
4348                 ret = mnt_want_write_file(file);
4349                 if (ret)
4350                         goto out;
4351         }
4352
4353         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
4354                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
4355                               0);
4356
4357         /*
4358          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
4359          * error. This is important as it allows user space to know how much
4360          * progress scrub has done. For example, if scrub is canceled we get
4361          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
4362          * space. Later user space can inspect the progress from the structure
4363          * btrfs_ioctl_scrub_args and resume scrub from where it left off
4364          * previously (btrfs-progs does this).
4365          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
4366          * then return -EFAULT to signal the structure was not copied or it may
4367          * be corrupt and unreliable due to a partial copy.
4368          */
4369         if (copy_to_user(arg, sa, sizeof(*sa)))
4370                 ret = -EFAULT;
4371
4372         if (!(sa->flags & BTRFS_SCRUB_READONLY))
4373                 mnt_drop_write_file(file);
4374 out:
4375         kfree(sa);
4376         return ret;
4377 }
4378
4379 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
4380 {
4381         if (!capable(CAP_SYS_ADMIN))
4382                 return -EPERM;
4383
4384         return btrfs_scrub_cancel(fs_info);
4385 }
4386
4387 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
4388                                        void __user *arg)
4389 {
4390         struct btrfs_ioctl_scrub_args *sa;
4391         int ret;
4392
4393         if (!capable(CAP_SYS_ADMIN))
4394                 return -EPERM;
4395
4396         sa = memdup_user(arg, sizeof(*sa));
4397         if (IS_ERR(sa))
4398                 return PTR_ERR(sa);
4399
4400         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
4401
4402         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4403                 ret = -EFAULT;
4404
4405         kfree(sa);
4406         return ret;
4407 }
4408
4409 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
4410                                       void __user *arg)
4411 {
4412         struct btrfs_ioctl_get_dev_stats *sa;
4413         int ret;
4414
4415         sa = memdup_user(arg, sizeof(*sa));
4416         if (IS_ERR(sa))
4417                 return PTR_ERR(sa);
4418
4419         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
4420                 kfree(sa);
4421                 return -EPERM;
4422         }
4423
4424         ret = btrfs_get_dev_stats(fs_info, sa);
4425
4426         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
4427                 ret = -EFAULT;
4428
4429         kfree(sa);
4430         return ret;
4431 }
4432
4433 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
4434                                     void __user *arg)
4435 {
4436         struct btrfs_ioctl_dev_replace_args *p;
4437         int ret;
4438
4439         if (!capable(CAP_SYS_ADMIN))
4440                 return -EPERM;
4441
4442         p = memdup_user(arg, sizeof(*p));
4443         if (IS_ERR(p))
4444                 return PTR_ERR(p);
4445
4446         switch (p->cmd) {
4447         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
4448                 if (sb_rdonly(fs_info->sb)) {
4449                         ret = -EROFS;
4450                         goto out;
4451                 }
4452                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4453                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4454                 } else {
4455                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
4456                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4457                 }
4458                 break;
4459         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
4460                 btrfs_dev_replace_status(fs_info, p);
4461                 ret = 0;
4462                 break;
4463         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
4464                 p->result = btrfs_dev_replace_cancel(fs_info);
4465                 ret = 0;
4466                 break;
4467         default:
4468                 ret = -EINVAL;
4469                 break;
4470         }
4471
4472         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
4473                 ret = -EFAULT;
4474 out:
4475         kfree(p);
4476         return ret;
4477 }
4478
4479 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
4480 {
4481         int ret = 0;
4482         int i;
4483         u64 rel_ptr;
4484         int size;
4485         struct btrfs_ioctl_ino_path_args *ipa = NULL;
4486         struct inode_fs_paths *ipath = NULL;
4487         struct btrfs_path *path;
4488
4489         if (!capable(CAP_DAC_READ_SEARCH))
4490                 return -EPERM;
4491
4492         path = btrfs_alloc_path();
4493         if (!path) {
4494                 ret = -ENOMEM;
4495                 goto out;
4496         }
4497
4498         ipa = memdup_user(arg, sizeof(*ipa));
4499         if (IS_ERR(ipa)) {
4500                 ret = PTR_ERR(ipa);
4501                 ipa = NULL;
4502                 goto out;
4503         }
4504
4505         size = min_t(u32, ipa->size, 4096);
4506         ipath = init_ipath(size, root, path);
4507         if (IS_ERR(ipath)) {
4508                 ret = PTR_ERR(ipath);
4509                 ipath = NULL;
4510                 goto out;
4511         }
4512
4513         ret = paths_from_inode(ipa->inum, ipath);
4514         if (ret < 0)
4515                 goto out;
4516
4517         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
4518                 rel_ptr = ipath->fspath->val[i] -
4519                           (u64)(unsigned long)ipath->fspath->val;
4520                 ipath->fspath->val[i] = rel_ptr;
4521         }
4522
4523         btrfs_free_path(path);
4524         path = NULL;
4525         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
4526                            ipath->fspath, size);
4527         if (ret) {
4528                 ret = -EFAULT;
4529                 goto out;
4530         }
4531
4532 out:
4533         btrfs_free_path(path);
4534         free_ipath(ipath);
4535         kfree(ipa);
4536
4537         return ret;
4538 }
4539
4540 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
4541 {
4542         struct btrfs_data_container *inodes = ctx;
4543         const size_t c = 3 * sizeof(u64);
4544
4545         if (inodes->bytes_left >= c) {
4546                 inodes->bytes_left -= c;
4547                 inodes->val[inodes->elem_cnt] = inum;
4548                 inodes->val[inodes->elem_cnt + 1] = offset;
4549                 inodes->val[inodes->elem_cnt + 2] = root;
4550                 inodes->elem_cnt += 3;
4551         } else {
4552                 inodes->bytes_missing += c - inodes->bytes_left;
4553                 inodes->bytes_left = 0;
4554                 inodes->elem_missed += 3;
4555         }
4556
4557         return 0;
4558 }
4559
4560 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
4561                                         void __user *arg, int version)
4562 {
4563         int ret = 0;
4564         int size;
4565         struct btrfs_ioctl_logical_ino_args *loi;
4566         struct btrfs_data_container *inodes = NULL;
4567         struct btrfs_path *path = NULL;
4568         bool ignore_offset;
4569
4570         if (!capable(CAP_SYS_ADMIN))
4571                 return -EPERM;
4572
4573         loi = memdup_user(arg, sizeof(*loi));
4574         if (IS_ERR(loi))
4575                 return PTR_ERR(loi);
4576
4577         if (version == 1) {
4578                 ignore_offset = false;
4579                 size = min_t(u32, loi->size, SZ_64K);
4580         } else {
4581                 /* All reserved bits must be 0 for now */
4582                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
4583                         ret = -EINVAL;
4584                         goto out_loi;
4585                 }
4586                 /* Only accept flags we have defined so far */
4587                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
4588                         ret = -EINVAL;
4589                         goto out_loi;
4590                 }
4591                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
4592                 size = min_t(u32, loi->size, SZ_16M);
4593         }
4594
4595         inodes = init_data_container(size);
4596         if (IS_ERR(inodes)) {
4597                 ret = PTR_ERR(inodes);
4598                 goto out_loi;
4599         }
4600
4601         path = btrfs_alloc_path();
4602         if (!path) {
4603                 ret = -ENOMEM;
4604                 goto out;
4605         }
4606         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
4607                                           build_ino_list, inodes, ignore_offset);
4608         btrfs_free_path(path);
4609         if (ret == -EINVAL)
4610                 ret = -ENOENT;
4611         if (ret < 0)
4612                 goto out;
4613
4614         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
4615                            size);
4616         if (ret)
4617                 ret = -EFAULT;
4618
4619 out:
4620         kvfree(inodes);
4621 out_loi:
4622         kfree(loi);
4623
4624         return ret;
4625 }
4626
4627 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
4628                                struct btrfs_ioctl_balance_args *bargs)
4629 {
4630         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4631
4632         bargs->flags = bctl->flags;
4633
4634         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4635                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4636         if (atomic_read(&fs_info->balance_pause_req))
4637                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4638         if (atomic_read(&fs_info->balance_cancel_req))
4639                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4640
4641         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4642         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4643         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4644
4645         spin_lock(&fs_info->balance_lock);
4646         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4647         spin_unlock(&fs_info->balance_lock);
4648 }
4649
4650 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4651 {
4652         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4653         struct btrfs_fs_info *fs_info = root->fs_info;
4654         struct btrfs_ioctl_balance_args *bargs;
4655         struct btrfs_balance_control *bctl;
4656         bool need_unlock; /* for mut. excl. ops lock */
4657         int ret;
4658
4659         if (!capable(CAP_SYS_ADMIN))
4660                 return -EPERM;
4661
4662         ret = mnt_want_write_file(file);
4663         if (ret)
4664                 return ret;
4665
4666 again:
4667         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
4668                 mutex_lock(&fs_info->balance_mutex);
4669                 need_unlock = true;
4670                 goto locked;
4671         }
4672
4673         /*
4674          * mut. excl. ops lock is locked.  Three possibilities:
4675          *   (1) some other op is running
4676          *   (2) balance is running
4677          *   (3) balance is paused -- special case (think resume)
4678          */
4679         mutex_lock(&fs_info->balance_mutex);
4680         if (fs_info->balance_ctl) {
4681                 /* this is either (2) or (3) */
4682                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4683                         mutex_unlock(&fs_info->balance_mutex);
4684                         /*
4685                          * Lock released to allow other waiters to continue,
4686                          * we'll reexamine the status again.
4687                          */
4688                         mutex_lock(&fs_info->balance_mutex);
4689
4690                         if (fs_info->balance_ctl &&
4691                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4692                                 /* this is (3) */
4693                                 need_unlock = false;
4694                                 goto locked;
4695                         }
4696
4697                         mutex_unlock(&fs_info->balance_mutex);
4698                         goto again;
4699                 } else {
4700                         /* this is (2) */
4701                         mutex_unlock(&fs_info->balance_mutex);
4702                         ret = -EINPROGRESS;
4703                         goto out;
4704                 }
4705         } else {
4706                 /* this is (1) */
4707                 mutex_unlock(&fs_info->balance_mutex);
4708                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4709                 goto out;
4710         }
4711
4712 locked:
4713         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4714
4715         if (arg) {
4716                 bargs = memdup_user(arg, sizeof(*bargs));
4717                 if (IS_ERR(bargs)) {
4718                         ret = PTR_ERR(bargs);
4719                         goto out_unlock;
4720                 }
4721
4722                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4723                         if (!fs_info->balance_ctl) {
4724                                 ret = -ENOTCONN;
4725                                 goto out_bargs;
4726                         }
4727
4728                         bctl = fs_info->balance_ctl;
4729                         spin_lock(&fs_info->balance_lock);
4730                         bctl->flags |= BTRFS_BALANCE_RESUME;
4731                         spin_unlock(&fs_info->balance_lock);
4732
4733                         goto do_balance;
4734                 }
4735         } else {
4736                 bargs = NULL;
4737         }
4738
4739         if (fs_info->balance_ctl) {
4740                 ret = -EINPROGRESS;
4741                 goto out_bargs;
4742         }
4743
4744         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4745         if (!bctl) {
4746                 ret = -ENOMEM;
4747                 goto out_bargs;
4748         }
4749
4750         if (arg) {
4751                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4752                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4753                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4754
4755                 bctl->flags = bargs->flags;
4756         } else {
4757                 /* balance everything - no filters */
4758                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4759         }
4760
4761         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4762                 ret = -EINVAL;
4763                 goto out_bctl;
4764         }
4765
4766 do_balance:
4767         /*
4768          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4769          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4770          * restriper was paused all the way until unmount, in free_fs_info.
4771          * The flag should be cleared after reset_balance_state.
4772          */
4773         need_unlock = false;
4774
4775         ret = btrfs_balance(fs_info, bctl, bargs);
4776         bctl = NULL;
4777
4778         if ((ret == 0 || ret == -ECANCELED) && arg) {
4779                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4780                         ret = -EFAULT;
4781         }
4782
4783 out_bctl:
4784         kfree(bctl);
4785 out_bargs:
4786         kfree(bargs);
4787 out_unlock:
4788         mutex_unlock(&fs_info->balance_mutex);
4789         if (need_unlock)
4790                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4791 out:
4792         mnt_drop_write_file(file);
4793         return ret;
4794 }
4795
4796 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4797 {
4798         if (!capable(CAP_SYS_ADMIN))
4799                 return -EPERM;
4800
4801         switch (cmd) {
4802         case BTRFS_BALANCE_CTL_PAUSE:
4803                 return btrfs_pause_balance(fs_info);
4804         case BTRFS_BALANCE_CTL_CANCEL:
4805                 return btrfs_cancel_balance(fs_info);
4806         }
4807
4808         return -EINVAL;
4809 }
4810
4811 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4812                                          void __user *arg)
4813 {
4814         struct btrfs_ioctl_balance_args *bargs;
4815         int ret = 0;
4816
4817         if (!capable(CAP_SYS_ADMIN))
4818                 return -EPERM;
4819
4820         mutex_lock(&fs_info->balance_mutex);
4821         if (!fs_info->balance_ctl) {
4822                 ret = -ENOTCONN;
4823                 goto out;
4824         }
4825
4826         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4827         if (!bargs) {
4828                 ret = -ENOMEM;
4829                 goto out;
4830         }
4831
4832         btrfs_update_ioctl_balance_args(fs_info, bargs);
4833
4834         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4835                 ret = -EFAULT;
4836
4837         kfree(bargs);
4838 out:
4839         mutex_unlock(&fs_info->balance_mutex);
4840         return ret;
4841 }
4842
4843 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4844 {
4845         struct inode *inode = file_inode(file);
4846         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4847         struct btrfs_ioctl_quota_ctl_args *sa;
4848         int ret;
4849
4850         if (!capable(CAP_SYS_ADMIN))
4851                 return -EPERM;
4852
4853         ret = mnt_want_write_file(file);
4854         if (ret)
4855                 return ret;
4856
4857         sa = memdup_user(arg, sizeof(*sa));
4858         if (IS_ERR(sa)) {
4859                 ret = PTR_ERR(sa);
4860                 goto drop_write;
4861         }
4862
4863         down_write(&fs_info->subvol_sem);
4864
4865         switch (sa->cmd) {
4866         case BTRFS_QUOTA_CTL_ENABLE:
4867                 ret = btrfs_quota_enable(fs_info);
4868                 break;
4869         case BTRFS_QUOTA_CTL_DISABLE:
4870                 ret = btrfs_quota_disable(fs_info);
4871                 break;
4872         default:
4873                 ret = -EINVAL;
4874                 break;
4875         }
4876
4877         kfree(sa);
4878         up_write(&fs_info->subvol_sem);
4879 drop_write:
4880         mnt_drop_write_file(file);
4881         return ret;
4882 }
4883
4884 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4885 {
4886         struct inode *inode = file_inode(file);
4887         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4888         struct btrfs_root *root = BTRFS_I(inode)->root;
4889         struct btrfs_ioctl_qgroup_assign_args *sa;
4890         struct btrfs_trans_handle *trans;
4891         int ret;
4892         int err;
4893
4894         if (!capable(CAP_SYS_ADMIN))
4895                 return -EPERM;
4896
4897         ret = mnt_want_write_file(file);
4898         if (ret)
4899                 return ret;
4900
4901         sa = memdup_user(arg, sizeof(*sa));
4902         if (IS_ERR(sa)) {
4903                 ret = PTR_ERR(sa);
4904                 goto drop_write;
4905         }
4906
4907         trans = btrfs_join_transaction(root);
4908         if (IS_ERR(trans)) {
4909                 ret = PTR_ERR(trans);
4910                 goto out;
4911         }
4912
4913         if (sa->assign) {
4914                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4915         } else {
4916                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4917         }
4918
4919         /* update qgroup status and info */
4920         mutex_lock(&fs_info->qgroup_ioctl_lock);
4921         err = btrfs_run_qgroups(trans);
4922         mutex_unlock(&fs_info->qgroup_ioctl_lock);
4923         if (err < 0)
4924                 btrfs_handle_fs_error(fs_info, err,
4925                                       "failed to update qgroup status and info");
4926         err = btrfs_end_transaction(trans);
4927         if (err && !ret)
4928                 ret = err;
4929
4930 out:
4931         kfree(sa);
4932 drop_write:
4933         mnt_drop_write_file(file);
4934         return ret;
4935 }
4936
4937 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4938 {
4939         struct inode *inode = file_inode(file);
4940         struct btrfs_root *root = BTRFS_I(inode)->root;
4941         struct btrfs_ioctl_qgroup_create_args *sa;
4942         struct btrfs_trans_handle *trans;
4943         int ret;
4944         int err;
4945
4946         if (!capable(CAP_SYS_ADMIN))
4947                 return -EPERM;
4948
4949         ret = mnt_want_write_file(file);
4950         if (ret)
4951                 return ret;
4952
4953         sa = memdup_user(arg, sizeof(*sa));
4954         if (IS_ERR(sa)) {
4955                 ret = PTR_ERR(sa);
4956                 goto drop_write;
4957         }
4958
4959         if (!sa->qgroupid) {
4960                 ret = -EINVAL;
4961                 goto out;
4962         }
4963
4964         trans = btrfs_join_transaction(root);
4965         if (IS_ERR(trans)) {
4966                 ret = PTR_ERR(trans);
4967                 goto out;
4968         }
4969
4970         if (sa->create) {
4971                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4972         } else {
4973                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4974         }
4975
4976         err = btrfs_end_transaction(trans);
4977         if (err && !ret)
4978                 ret = err;
4979
4980 out:
4981         kfree(sa);
4982 drop_write:
4983         mnt_drop_write_file(file);
4984         return ret;
4985 }
4986
4987 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4988 {
4989         struct inode *inode = file_inode(file);
4990         struct btrfs_root *root = BTRFS_I(inode)->root;
4991         struct btrfs_ioctl_qgroup_limit_args *sa;
4992         struct btrfs_trans_handle *trans;
4993         int ret;
4994         int err;
4995         u64 qgroupid;
4996
4997         if (!capable(CAP_SYS_ADMIN))
4998                 return -EPERM;
4999
5000         ret = mnt_want_write_file(file);
5001         if (ret)
5002                 return ret;
5003
5004         sa = memdup_user(arg, sizeof(*sa));
5005         if (IS_ERR(sa)) {
5006                 ret = PTR_ERR(sa);
5007                 goto drop_write;
5008         }
5009
5010         trans = btrfs_join_transaction(root);
5011         if (IS_ERR(trans)) {
5012                 ret = PTR_ERR(trans);
5013                 goto out;
5014         }
5015
5016         qgroupid = sa->qgroupid;
5017         if (!qgroupid) {
5018                 /* take the current subvol as qgroup */
5019                 qgroupid = root->root_key.objectid;
5020         }
5021
5022         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
5023
5024         err = btrfs_end_transaction(trans);
5025         if (err && !ret)
5026                 ret = err;
5027
5028 out:
5029         kfree(sa);
5030 drop_write:
5031         mnt_drop_write_file(file);
5032         return ret;
5033 }
5034
5035 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
5036 {
5037         struct inode *inode = file_inode(file);
5038         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5039         struct btrfs_ioctl_quota_rescan_args *qsa;
5040         int ret;
5041
5042         if (!capable(CAP_SYS_ADMIN))
5043                 return -EPERM;
5044
5045         ret = mnt_want_write_file(file);
5046         if (ret)
5047                 return ret;
5048
5049         qsa = memdup_user(arg, sizeof(*qsa));
5050         if (IS_ERR(qsa)) {
5051                 ret = PTR_ERR(qsa);
5052                 goto drop_write;
5053         }
5054
5055         if (qsa->flags) {
5056                 ret = -EINVAL;
5057                 goto out;
5058         }
5059
5060         ret = btrfs_qgroup_rescan(fs_info);
5061
5062 out:
5063         kfree(qsa);
5064 drop_write:
5065         mnt_drop_write_file(file);
5066         return ret;
5067 }
5068
5069 static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg)
5070 {
5071         struct inode *inode = file_inode(file);
5072         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5073         struct btrfs_ioctl_quota_rescan_args *qsa;
5074         int ret = 0;
5075
5076         if (!capable(CAP_SYS_ADMIN))
5077                 return -EPERM;
5078
5079         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
5080         if (!qsa)
5081                 return -ENOMEM;
5082
5083         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
5084                 qsa->flags = 1;
5085                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
5086         }
5087
5088         if (copy_to_user(arg, qsa, sizeof(*qsa)))
5089                 ret = -EFAULT;
5090
5091         kfree(qsa);
5092         return ret;
5093 }
5094
5095 static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg)
5096 {
5097         struct inode *inode = file_inode(file);
5098         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5099
5100         if (!capable(CAP_SYS_ADMIN))
5101                 return -EPERM;
5102
5103         return btrfs_qgroup_wait_for_completion(fs_info, true);
5104 }
5105
5106 static long _btrfs_ioctl_set_received_subvol(struct file *file,
5107                                             struct btrfs_ioctl_received_subvol_args *sa)
5108 {
5109         struct inode *inode = file_inode(file);
5110         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5111         struct btrfs_root *root = BTRFS_I(inode)->root;
5112         struct btrfs_root_item *root_item = &root->root_item;
5113         struct btrfs_trans_handle *trans;
5114         struct timespec64 ct = current_time(inode);
5115         int ret = 0;
5116         int received_uuid_changed;
5117
5118         if (!inode_owner_or_capable(inode))
5119                 return -EPERM;
5120
5121         ret = mnt_want_write_file(file);
5122         if (ret < 0)
5123                 return ret;
5124
5125         down_write(&fs_info->subvol_sem);
5126
5127         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
5128                 ret = -EINVAL;
5129                 goto out;
5130         }
5131
5132         if (btrfs_root_readonly(root)) {
5133                 ret = -EROFS;
5134                 goto out;
5135         }
5136
5137         /*
5138          * 1 - root item
5139          * 2 - uuid items (received uuid + subvol uuid)
5140          */
5141         trans = btrfs_start_transaction(root, 3);
5142         if (IS_ERR(trans)) {
5143                 ret = PTR_ERR(trans);
5144                 trans = NULL;
5145                 goto out;
5146         }
5147
5148         sa->rtransid = trans->transid;
5149         sa->rtime.sec = ct.tv_sec;
5150         sa->rtime.nsec = ct.tv_nsec;
5151
5152         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
5153                                        BTRFS_UUID_SIZE);
5154         if (received_uuid_changed &&
5155             !btrfs_is_empty_uuid(root_item->received_uuid)) {
5156                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
5157                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5158                                           root->root_key.objectid);
5159                 if (ret && ret != -ENOENT) {
5160                         btrfs_abort_transaction(trans, ret);
5161                         btrfs_end_transaction(trans);
5162                         goto out;
5163                 }
5164         }
5165         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
5166         btrfs_set_root_stransid(root_item, sa->stransid);
5167         btrfs_set_root_rtransid(root_item, sa->rtransid);
5168         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
5169         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
5170         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
5171         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
5172
5173         ret = btrfs_update_root(trans, fs_info->tree_root,
5174                                 &root->root_key, &root->root_item);
5175         if (ret < 0) {
5176                 btrfs_end_transaction(trans);
5177                 goto out;
5178         }
5179         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
5180                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
5181                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
5182                                           root->root_key.objectid);
5183                 if (ret < 0 && ret != -EEXIST) {
5184                         btrfs_abort_transaction(trans, ret);
5185                         btrfs_end_transaction(trans);
5186                         goto out;
5187                 }
5188         }
5189         ret = btrfs_commit_transaction(trans);
5190 out:
5191         up_write(&fs_info->subvol_sem);
5192         mnt_drop_write_file(file);
5193         return ret;
5194 }
5195
5196 #ifdef CONFIG_64BIT
5197 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
5198                                                 void __user *arg)
5199 {
5200         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
5201         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
5202         int ret = 0;
5203
5204         args32 = memdup_user(arg, sizeof(*args32));
5205         if (IS_ERR(args32))
5206                 return PTR_ERR(args32);
5207
5208         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
5209         if (!args64) {
5210                 ret = -ENOMEM;
5211                 goto out;
5212         }
5213
5214         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
5215         args64->stransid = args32->stransid;
5216         args64->rtransid = args32->rtransid;
5217         args64->stime.sec = args32->stime.sec;
5218         args64->stime.nsec = args32->stime.nsec;
5219         args64->rtime.sec = args32->rtime.sec;
5220         args64->rtime.nsec = args32->rtime.nsec;
5221         args64->flags = args32->flags;
5222
5223         ret = _btrfs_ioctl_set_received_subvol(file, args64);
5224         if (ret)
5225                 goto out;
5226
5227         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
5228         args32->stransid = args64->stransid;
5229         args32->rtransid = args64->rtransid;
5230         args32->stime.sec = args64->stime.sec;
5231         args32->stime.nsec = args64->stime.nsec;
5232         args32->rtime.sec = args64->rtime.sec;
5233         args32->rtime.nsec = args64->rtime.nsec;
5234         args32->flags = args64->flags;
5235
5236         ret = copy_to_user(arg, args32, sizeof(*args32));
5237         if (ret)
5238                 ret = -EFAULT;
5239
5240 out:
5241         kfree(args32);
5242         kfree(args64);
5243         return ret;
5244 }
5245 #endif
5246
5247 static long btrfs_ioctl_set_received_subvol(struct file *file,
5248                                             void __user *arg)
5249 {
5250         struct btrfs_ioctl_received_subvol_args *sa = NULL;
5251         int ret = 0;
5252
5253         sa = memdup_user(arg, sizeof(*sa));
5254         if (IS_ERR(sa))
5255                 return PTR_ERR(sa);
5256
5257         ret = _btrfs_ioctl_set_received_subvol(file, sa);
5258
5259         if (ret)
5260                 goto out;
5261
5262         ret = copy_to_user(arg, sa, sizeof(*sa));
5263         if (ret)
5264                 ret = -EFAULT;
5265
5266 out:
5267         kfree(sa);
5268         return ret;
5269 }
5270
5271 static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg)
5272 {
5273         struct inode *inode = file_inode(file);
5274         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5275         size_t len;
5276         int ret;
5277         char label[BTRFS_LABEL_SIZE];
5278
5279         spin_lock(&fs_info->super_lock);
5280         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
5281         spin_unlock(&fs_info->super_lock);
5282
5283         len = strnlen(label, BTRFS_LABEL_SIZE);
5284
5285         if (len == BTRFS_LABEL_SIZE) {
5286                 btrfs_warn(fs_info,
5287                            "label is too long, return the first %zu bytes",
5288                            --len);
5289         }
5290
5291         ret = copy_to_user(arg, label, len);
5292
5293         return ret ? -EFAULT : 0;
5294 }
5295
5296 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
5297 {
5298         struct inode *inode = file_inode(file);
5299         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5300         struct btrfs_root *root = BTRFS_I(inode)->root;
5301         struct btrfs_super_block *super_block = fs_info->super_copy;
5302         struct btrfs_trans_handle *trans;
5303         char label[BTRFS_LABEL_SIZE];
5304         int ret;
5305
5306         if (!capable(CAP_SYS_ADMIN))
5307                 return -EPERM;
5308
5309         if (copy_from_user(label, arg, sizeof(label)))
5310                 return -EFAULT;
5311
5312         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
5313                 btrfs_err(fs_info,
5314                           "unable to set label with more than %d bytes",
5315                           BTRFS_LABEL_SIZE - 1);
5316                 return -EINVAL;
5317         }
5318
5319         ret = mnt_want_write_file(file);
5320         if (ret)
5321                 return ret;
5322
5323         trans = btrfs_start_transaction(root, 0);
5324         if (IS_ERR(trans)) {
5325                 ret = PTR_ERR(trans);
5326                 goto out_unlock;
5327         }
5328
5329         spin_lock(&fs_info->super_lock);
5330         strcpy(super_block->label, label);
5331         spin_unlock(&fs_info->super_lock);
5332         ret = btrfs_commit_transaction(trans);
5333
5334 out_unlock:
5335         mnt_drop_write_file(file);
5336         return ret;
5337 }
5338
5339 #define INIT_FEATURE_FLAGS(suffix) \
5340         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
5341           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
5342           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
5343
5344 int btrfs_ioctl_get_supported_features(void __user *arg)
5345 {
5346         static const struct btrfs_ioctl_feature_flags features[3] = {
5347                 INIT_FEATURE_FLAGS(SUPP),
5348                 INIT_FEATURE_FLAGS(SAFE_SET),
5349                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
5350         };
5351
5352         if (copy_to_user(arg, &features, sizeof(features)))
5353                 return -EFAULT;
5354
5355         return 0;
5356 }
5357
5358 static int btrfs_ioctl_get_features(struct file *file, void __user *arg)
5359 {
5360         struct inode *inode = file_inode(file);
5361         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5362         struct btrfs_super_block *super_block = fs_info->super_copy;
5363         struct btrfs_ioctl_feature_flags features;
5364
5365         features.compat_flags = btrfs_super_compat_flags(super_block);
5366         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
5367         features.incompat_flags = btrfs_super_incompat_flags(super_block);
5368
5369         if (copy_to_user(arg, &features, sizeof(features)))
5370                 return -EFAULT;
5371
5372         return 0;
5373 }
5374
5375 static int check_feature_bits(struct btrfs_fs_info *fs_info,
5376                               enum btrfs_feature_set set,
5377                               u64 change_mask, u64 flags, u64 supported_flags,
5378                               u64 safe_set, u64 safe_clear)
5379 {
5380         const char *type = btrfs_feature_set_name(set);
5381         char *names;
5382         u64 disallowed, unsupported;
5383         u64 set_mask = flags & change_mask;
5384         u64 clear_mask = ~flags & change_mask;
5385
5386         unsupported = set_mask & ~supported_flags;
5387         if (unsupported) {
5388                 names = btrfs_printable_features(set, unsupported);
5389                 if (names) {
5390                         btrfs_warn(fs_info,
5391                                    "this kernel does not support the %s feature bit%s",
5392                                    names, strchr(names, ',') ? "s" : "");
5393                         kfree(names);
5394                 } else
5395                         btrfs_warn(fs_info,
5396                                    "this kernel does not support %s bits 0x%llx",
5397                                    type, unsupported);
5398                 return -EOPNOTSUPP;
5399         }
5400
5401         disallowed = set_mask & ~safe_set;
5402         if (disallowed) {
5403                 names = btrfs_printable_features(set, disallowed);
5404                 if (names) {
5405                         btrfs_warn(fs_info,
5406                                    "can't set the %s feature bit%s while mounted",
5407                                    names, strchr(names, ',') ? "s" : "");
5408                         kfree(names);
5409                 } else
5410                         btrfs_warn(fs_info,
5411                                    "can't set %s bits 0x%llx while mounted",
5412                                    type, disallowed);
5413                 return -EPERM;
5414         }
5415
5416         disallowed = clear_mask & ~safe_clear;
5417         if (disallowed) {
5418                 names = btrfs_printable_features(set, disallowed);
5419                 if (names) {
5420                         btrfs_warn(fs_info,
5421                                    "can't clear the %s feature bit%s while mounted",
5422                                    names, strchr(names, ',') ? "s" : "");
5423                         kfree(names);
5424                 } else
5425                         btrfs_warn(fs_info,
5426                                    "can't clear %s bits 0x%llx while mounted",
5427                                    type, disallowed);
5428                 return -EPERM;
5429         }
5430
5431         return 0;
5432 }
5433
5434 #define check_feature(fs_info, change_mask, flags, mask_base)   \
5435 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
5436                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
5437                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
5438                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
5439
5440 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
5441 {
5442         struct inode *inode = file_inode(file);
5443         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5444         struct btrfs_root *root = BTRFS_I(inode)->root;
5445         struct btrfs_super_block *super_block = fs_info->super_copy;
5446         struct btrfs_ioctl_feature_flags flags[2];
5447         struct btrfs_trans_handle *trans;
5448         u64 newflags;
5449         int ret;
5450
5451         if (!capable(CAP_SYS_ADMIN))
5452                 return -EPERM;
5453
5454         if (copy_from_user(flags, arg, sizeof(flags)))
5455                 return -EFAULT;
5456
5457         /* Nothing to do */
5458         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
5459             !flags[0].incompat_flags)
5460                 return 0;
5461
5462         ret = check_feature(fs_info, flags[0].compat_flags,
5463                             flags[1].compat_flags, COMPAT);
5464         if (ret)
5465                 return ret;
5466
5467         ret = check_feature(fs_info, flags[0].compat_ro_flags,
5468                             flags[1].compat_ro_flags, COMPAT_RO);
5469         if (ret)
5470                 return ret;
5471
5472         ret = check_feature(fs_info, flags[0].incompat_flags,
5473                             flags[1].incompat_flags, INCOMPAT);
5474         if (ret)
5475                 return ret;
5476
5477         ret = mnt_want_write_file(file);
5478         if (ret)
5479                 return ret;
5480
5481         trans = btrfs_start_transaction(root, 0);
5482         if (IS_ERR(trans)) {
5483                 ret = PTR_ERR(trans);
5484                 goto out_drop_write;
5485         }
5486
5487         spin_lock(&fs_info->super_lock);
5488         newflags = btrfs_super_compat_flags(super_block);
5489         newflags |= flags[0].compat_flags & flags[1].compat_flags;
5490         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
5491         btrfs_set_super_compat_flags(super_block, newflags);
5492
5493         newflags = btrfs_super_compat_ro_flags(super_block);
5494         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
5495         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
5496         btrfs_set_super_compat_ro_flags(super_block, newflags);
5497
5498         newflags = btrfs_super_incompat_flags(super_block);
5499         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
5500         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
5501         btrfs_set_super_incompat_flags(super_block, newflags);
5502         spin_unlock(&fs_info->super_lock);
5503
5504         ret = btrfs_commit_transaction(trans);
5505 out_drop_write:
5506         mnt_drop_write_file(file);
5507
5508         return ret;
5509 }
5510
5511 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
5512 {
5513         struct btrfs_ioctl_send_args *arg;
5514         int ret;
5515
5516         if (compat) {
5517 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5518                 struct btrfs_ioctl_send_args_32 args32;
5519
5520                 ret = copy_from_user(&args32, argp, sizeof(args32));
5521                 if (ret)
5522                         return -EFAULT;
5523                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
5524                 if (!arg)
5525                         return -ENOMEM;
5526                 arg->send_fd = args32.send_fd;
5527                 arg->clone_sources_count = args32.clone_sources_count;
5528                 arg->clone_sources = compat_ptr(args32.clone_sources);
5529                 arg->parent_root = args32.parent_root;
5530                 arg->flags = args32.flags;
5531                 memcpy(arg->reserved, args32.reserved,
5532                        sizeof(args32.reserved));
5533 #else
5534                 return -ENOTTY;
5535 #endif
5536         } else {
5537                 arg = memdup_user(argp, sizeof(*arg));
5538                 if (IS_ERR(arg))
5539                         return PTR_ERR(arg);
5540         }
5541         ret = btrfs_ioctl_send(file, arg);
5542         kfree(arg);
5543         return ret;
5544 }
5545
5546 long btrfs_ioctl(struct file *file, unsigned int
5547                 cmd, unsigned long arg)
5548 {
5549         struct inode *inode = file_inode(file);
5550         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5551         struct btrfs_root *root = BTRFS_I(inode)->root;
5552         void __user *argp = (void __user *)arg;
5553
5554         switch (cmd) {
5555         case FS_IOC_GETFLAGS:
5556                 return btrfs_ioctl_getflags(file, argp);
5557         case FS_IOC_SETFLAGS:
5558                 return btrfs_ioctl_setflags(file, argp);
5559         case FS_IOC_GETVERSION:
5560                 return btrfs_ioctl_getversion(file, argp);
5561         case FS_IOC_GETFSLABEL:
5562                 return btrfs_ioctl_get_fslabel(file, argp);
5563         case FS_IOC_SETFSLABEL:
5564                 return btrfs_ioctl_set_fslabel(file, argp);
5565         case FITRIM:
5566                 return btrfs_ioctl_fitrim(file, argp);
5567         case BTRFS_IOC_SNAP_CREATE:
5568                 return btrfs_ioctl_snap_create(file, argp, 0);
5569         case BTRFS_IOC_SNAP_CREATE_V2:
5570                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
5571         case BTRFS_IOC_SUBVOL_CREATE:
5572                 return btrfs_ioctl_snap_create(file, argp, 1);
5573         case BTRFS_IOC_SUBVOL_CREATE_V2:
5574                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
5575         case BTRFS_IOC_SNAP_DESTROY:
5576                 return btrfs_ioctl_snap_destroy(file, argp);
5577         case BTRFS_IOC_SUBVOL_GETFLAGS:
5578                 return btrfs_ioctl_subvol_getflags(file, argp);
5579         case BTRFS_IOC_SUBVOL_SETFLAGS:
5580                 return btrfs_ioctl_subvol_setflags(file, argp);
5581         case BTRFS_IOC_DEFAULT_SUBVOL:
5582                 return btrfs_ioctl_default_subvol(file, argp);
5583         case BTRFS_IOC_DEFRAG:
5584                 return btrfs_ioctl_defrag(file, NULL);
5585         case BTRFS_IOC_DEFRAG_RANGE:
5586                 return btrfs_ioctl_defrag(file, argp);
5587         case BTRFS_IOC_RESIZE:
5588                 return btrfs_ioctl_resize(file, argp);
5589         case BTRFS_IOC_ADD_DEV:
5590                 return btrfs_ioctl_add_dev(fs_info, argp);
5591         case BTRFS_IOC_RM_DEV:
5592                 return btrfs_ioctl_rm_dev(file, argp);
5593         case BTRFS_IOC_RM_DEV_V2:
5594                 return btrfs_ioctl_rm_dev_v2(file, argp);
5595         case BTRFS_IOC_FS_INFO:
5596                 return btrfs_ioctl_fs_info(fs_info, argp);
5597         case BTRFS_IOC_DEV_INFO:
5598                 return btrfs_ioctl_dev_info(fs_info, argp);
5599         case BTRFS_IOC_BALANCE:
5600                 return btrfs_ioctl_balance(file, NULL);
5601         case BTRFS_IOC_TREE_SEARCH:
5602                 return btrfs_ioctl_tree_search(file, argp);
5603         case BTRFS_IOC_TREE_SEARCH_V2:
5604                 return btrfs_ioctl_tree_search_v2(file, argp);
5605         case BTRFS_IOC_INO_LOOKUP:
5606                 return btrfs_ioctl_ino_lookup(file, argp);
5607         case BTRFS_IOC_INO_PATHS:
5608                 return btrfs_ioctl_ino_to_path(root, argp);
5609         case BTRFS_IOC_LOGICAL_INO:
5610                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
5611         case BTRFS_IOC_LOGICAL_INO_V2:
5612                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
5613         case BTRFS_IOC_SPACE_INFO:
5614                 return btrfs_ioctl_space_info(fs_info, argp);
5615         case BTRFS_IOC_SYNC: {
5616                 int ret;
5617
5618                 ret = btrfs_start_delalloc_roots(fs_info, -1);
5619                 if (ret)
5620                         return ret;
5621                 ret = btrfs_sync_fs(inode->i_sb, 1);
5622                 /*
5623                  * The transaction thread may want to do more work,
5624                  * namely it pokes the cleaner kthread that will start
5625                  * processing uncleaned subvols.
5626                  */
5627                 wake_up_process(fs_info->transaction_kthread);
5628                 return ret;
5629         }
5630         case BTRFS_IOC_START_SYNC:
5631                 return btrfs_ioctl_start_sync(root, argp);
5632         case BTRFS_IOC_WAIT_SYNC:
5633                 return btrfs_ioctl_wait_sync(fs_info, argp);
5634         case BTRFS_IOC_SCRUB:
5635                 return btrfs_ioctl_scrub(file, argp);
5636         case BTRFS_IOC_SCRUB_CANCEL:
5637                 return btrfs_ioctl_scrub_cancel(fs_info);
5638         case BTRFS_IOC_SCRUB_PROGRESS:
5639                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5640         case BTRFS_IOC_BALANCE_V2:
5641                 return btrfs_ioctl_balance(file, argp);
5642         case BTRFS_IOC_BALANCE_CTL:
5643                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5644         case BTRFS_IOC_BALANCE_PROGRESS:
5645                 return btrfs_ioctl_balance_progress(fs_info, argp);
5646         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5647                 return btrfs_ioctl_set_received_subvol(file, argp);
5648 #ifdef CONFIG_64BIT
5649         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5650                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5651 #endif
5652         case BTRFS_IOC_SEND:
5653                 return _btrfs_ioctl_send(file, argp, false);
5654 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5655         case BTRFS_IOC_SEND_32:
5656                 return _btrfs_ioctl_send(file, argp, true);
5657 #endif
5658         case BTRFS_IOC_GET_DEV_STATS:
5659                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5660         case BTRFS_IOC_QUOTA_CTL:
5661                 return btrfs_ioctl_quota_ctl(file, argp);
5662         case BTRFS_IOC_QGROUP_ASSIGN:
5663                 return btrfs_ioctl_qgroup_assign(file, argp);
5664         case BTRFS_IOC_QGROUP_CREATE:
5665                 return btrfs_ioctl_qgroup_create(file, argp);
5666         case BTRFS_IOC_QGROUP_LIMIT:
5667                 return btrfs_ioctl_qgroup_limit(file, argp);
5668         case BTRFS_IOC_QUOTA_RESCAN:
5669                 return btrfs_ioctl_quota_rescan(file, argp);
5670         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5671                 return btrfs_ioctl_quota_rescan_status(file, argp);
5672         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5673                 return btrfs_ioctl_quota_rescan_wait(file, argp);
5674         case BTRFS_IOC_DEV_REPLACE:
5675                 return btrfs_ioctl_dev_replace(fs_info, argp);
5676         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5677                 return btrfs_ioctl_get_supported_features(argp);
5678         case BTRFS_IOC_GET_FEATURES:
5679                 return btrfs_ioctl_get_features(file, argp);
5680         case BTRFS_IOC_SET_FEATURES:
5681                 return btrfs_ioctl_set_features(file, argp);
5682         case FS_IOC_FSGETXATTR:
5683                 return btrfs_ioctl_fsgetxattr(file, argp);
5684         case FS_IOC_FSSETXATTR:
5685                 return btrfs_ioctl_fssetxattr(file, argp);
5686         case BTRFS_IOC_GET_SUBVOL_INFO:
5687                 return btrfs_ioctl_get_subvol_info(file, argp);
5688         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5689                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5690         case BTRFS_IOC_INO_LOOKUP_USER:
5691                 return btrfs_ioctl_ino_lookup_user(file, argp);
5692         }
5693
5694         return -ENOTTY;
5695 }
5696
5697 #ifdef CONFIG_COMPAT
5698 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5699 {
5700         /*
5701          * These all access 32-bit values anyway so no further
5702          * handling is necessary.
5703          */
5704         switch (cmd) {
5705         case FS_IOC32_GETFLAGS:
5706                 cmd = FS_IOC_GETFLAGS;
5707                 break;
5708         case FS_IOC32_SETFLAGS:
5709                 cmd = FS_IOC_SETFLAGS;
5710                 break;
5711         case FS_IOC32_GETVERSION:
5712                 cmd = FS_IOC_GETVERSION;
5713                 break;
5714         }
5715
5716         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5717 }
5718 #endif