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 "export.h"
  32 #include "transaction.h"
  33 #include "btrfs_inode.h"
  34 #include "print-tree.h"
  35 #include "volumes.h"
  36 #include "locking.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 /* Mask out flags that are inappropriate for the given type of inode. */
  90 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  91                 unsigned int flags)
  92 {
  93         if (S_ISDIR(inode->i_mode))
  94                 return flags;
  95         else if (S_ISREG(inode->i_mode))
  96                 return flags & ~FS_DIRSYNC_FL;
  97         else
  98                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  99 }
 100
 101 /*
 102  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 103  * ioctl.
 104  */
 105 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 106 {
 107         unsigned int iflags = 0;
 108
 109         if (flags & BTRFS_INODE_SYNC)
 110                 iflags |= FS_SYNC_FL;
 111         if (flags & BTRFS_INODE_IMMUTABLE)
 112                 iflags |= FS_IMMUTABLE_FL;
 113         if (flags & BTRFS_INODE_APPEND)
 114                 iflags |= FS_APPEND_FL;
 115         if (flags & BTRFS_INODE_NODUMP)
 116                 iflags |= FS_NODUMP_FL;
 117         if (flags & BTRFS_INODE_NOATIME)
 118                 iflags |= FS_NOATIME_FL;
 119         if (flags & BTRFS_INODE_DIRSYNC)
 120                 iflags |= FS_DIRSYNC_FL;
 121         if (flags & BTRFS_INODE_NODATACOW)
 122                 iflags |= FS_NOCOW_FL;
 123
 124         if (flags & BTRFS_INODE_NOCOMPRESS)
 125                 iflags |= FS_NOCOMP_FL;
 126         else if (flags & BTRFS_INODE_COMPRESS)
 127                 iflags |= FS_COMPR_FL;
 128
 129         return iflags;
 130 }
 131
 132 /*
 133  * Update inode->i_flags based on the btrfs internal flags.
 134  */
 135 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 136 {
 137         struct btrfs_inode *binode = BTRFS_I(inode);
 138         unsigned int new_fl = 0;
 139
 140         if (binode->flags & BTRFS_INODE_SYNC)
 141                 new_fl |= S_SYNC;
 142         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 143                 new_fl |= S_IMMUTABLE;
 144         if (binode->flags & BTRFS_INODE_APPEND)
 145                 new_fl |= S_APPEND;
 146         if (binode->flags & BTRFS_INODE_NOATIME)
 147                 new_fl |= S_NOATIME;
 148         if (binode->flags & BTRFS_INODE_DIRSYNC)
 149                 new_fl |= S_DIRSYNC;
 150
 151         set_mask_bits(&inode->i_flags,
 152                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 153                       new_fl);
 154 }
 155
 156 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 157 {
 158         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 159         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 160
 161         if (copy_to_user(arg, &flags, sizeof(flags)))
 162                 return -EFAULT;
 163         return 0;
 164 }
 165
 166 /*
 167  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 168  * the old and new flags are not conflicting
 169  */
 170 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 171 {
 172         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 173                       FS_NOATIME_FL | FS_NODUMP_FL | \
 174                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 175                       FS_NOCOMP_FL | FS_COMPR_FL |
 176                       FS_NOCOW_FL))
 177                 return -EOPNOTSUPP;
 178
 179         /* COMPR and NOCOMP on new/old are valid */
 180         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 181                 return -EINVAL;
 182
 183         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 184                 return -EINVAL;
 185
 186         /* NOCOW and compression options are mutually exclusive */
 187         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 188                 return -EINVAL;
 189         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 190                 return -EINVAL;
 191
 192         return 0;
 193 }
 194
 195 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 196                                     unsigned int flags)
 197 {
 198         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 199                 return -EPERM;
 200
 201         return 0;
 202 }
 203
 204 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 205 {
 206         struct inode *inode = file_inode(file);
 207         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 208         struct btrfs_inode *binode = BTRFS_I(inode);
 209         struct btrfs_root *root = binode->root;
 210         struct btrfs_trans_handle *trans;
 211         unsigned int fsflags, old_fsflags;
 212         int ret;
 213         const char *comp = NULL;
 214         u32 binode_flags;
 215
 216         if (!inode_owner_or_capable(inode))
 217                 return -EPERM;
 218
 219         if (btrfs_root_readonly(root))
 220                 return -EROFS;
 221
 222         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 223                 return -EFAULT;
 224
 225         ret = mnt_want_write_file(file);
 226         if (ret)
 227                 return ret;
 228
 229         inode_lock(inode);
 230         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 231         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 232
 233         ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 234         if (ret)
 235                 goto out_unlock;
 236
 237         ret = check_fsflags(old_fsflags, fsflags);
 238         if (ret)
 239                 goto out_unlock;
 240
 241         ret = check_fsflags_compatible(fs_info, fsflags);
 242         if (ret)
 243                 goto out_unlock;
 244
 245         binode_flags = binode->flags;
 246         if (fsflags & FS_SYNC_FL)
 247                 binode_flags |= BTRFS_INODE_SYNC;
 248         else
 249                 binode_flags &= ~BTRFS_INODE_SYNC;
 250         if (fsflags & FS_IMMUTABLE_FL)
 251                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 252         else
 253                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 254         if (fsflags & FS_APPEND_FL)
 255                 binode_flags |= BTRFS_INODE_APPEND;
 256         else
 257                 binode_flags &= ~BTRFS_INODE_APPEND;
 258         if (fsflags & FS_NODUMP_FL)
 259                 binode_flags |= BTRFS_INODE_NODUMP;
 260         else
 261                 binode_flags &= ~BTRFS_INODE_NODUMP;
 262         if (fsflags & FS_NOATIME_FL)
 263                 binode_flags |= BTRFS_INODE_NOATIME;
 264         else
 265                 binode_flags &= ~BTRFS_INODE_NOATIME;
 266         if (fsflags & FS_DIRSYNC_FL)
 267                 binode_flags |= BTRFS_INODE_DIRSYNC;
 268         else
 269                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 270         if (fsflags & FS_NOCOW_FL) {
 271                 if (S_ISREG(inode->i_mode)) {
 272                         /*
 273                          * It's safe to turn csums off here, no extents exist.
 274                          * Otherwise we want the flag to reflect the real COW
 275                          * status of the file and will not set it.
 276                          */
 277                         if (inode->i_size == 0)
 278                                 binode_flags |= BTRFS_INODE_NODATACOW |
 279                                                 BTRFS_INODE_NODATASUM;
 280                 } else {
 281                         binode_flags |= BTRFS_INODE_NODATACOW;
 282                 }
 283         } else {
 284                 /*
 285                  * Revert back under same assumptions as above
 286                  */
 287                 if (S_ISREG(inode->i_mode)) {
 288                         if (inode->i_size == 0)
 289                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 290                                                   BTRFS_INODE_NODATASUM);
 291                 } else {
 292                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 293                 }
 294         }
 295
 296         /*
 297          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 298          * flag may be changed automatically if compression code won't make
 299          * things smaller.
 300          */
 301         if (fsflags & FS_NOCOMP_FL) {
 302                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 303                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 304         } else if (fsflags & FS_COMPR_FL) {
 305
 306                 if (IS_SWAPFILE(inode)) {
 307                         ret = -ETXTBSY;
 308                         goto out_unlock;
 309                 }
 310
 311                 binode_flags |= BTRFS_INODE_COMPRESS;
 312                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 313
 314                 comp = btrfs_compress_type2str(fs_info->compress_type);
 315                 if (!comp || comp[0] == 0)
 316                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 317         } else {
 318                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 319         }
 320
 321         /*
 322          * 1 for inode item
 323          * 2 for properties
 324          */
 325         trans = btrfs_start_transaction(root, 3);
 326         if (IS_ERR(trans)) {
 327                 ret = PTR_ERR(trans);
 328                 goto out_unlock;
 329         }
 330
 331         if (comp) {
 332                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 333                                      strlen(comp), 0);
 334                 if (ret) {
 335                         btrfs_abort_transaction(trans, ret);
 336                         goto out_end_trans;
 337                 }
 338         } else {
 339                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 340                                      0, 0);
 341                 if (ret && ret != -ENODATA) {
 342                         btrfs_abort_transaction(trans, ret);
 343                         goto out_end_trans;
 344                 }
 345         }
 346
 347         binode->flags = binode_flags;
 348         btrfs_sync_inode_flags_to_i_flags(inode);
 349         inode_inc_iversion(inode);
 350         inode->i_ctime = current_time(inode);
 351         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 352
 353  out_end_trans:
 354         btrfs_end_transaction(trans);
 355  out_unlock:
 356         inode_unlock(inode);
 357         mnt_drop_write_file(file);
 358         return ret;
 359 }
 360
 361 /*
 362  * Translate btrfs internal inode flags to xflags as expected by the
 363  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 364  * silently dropped.
 365  */
 366 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 367 {
 368         unsigned int xflags = 0;
 369
 370         if (flags & BTRFS_INODE_APPEND)
 371                 xflags |= FS_XFLAG_APPEND;
 372         if (flags & BTRFS_INODE_IMMUTABLE)
 373                 xflags |= FS_XFLAG_IMMUTABLE;
 374         if (flags & BTRFS_INODE_NOATIME)
 375                 xflags |= FS_XFLAG_NOATIME;
 376         if (flags & BTRFS_INODE_NODUMP)
 377                 xflags |= FS_XFLAG_NODUMP;
 378         if (flags & BTRFS_INODE_SYNC)
 379                 xflags |= FS_XFLAG_SYNC;
 380
 381         return xflags;
 382 }
 383
 384 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 385 static int check_xflags(unsigned int flags)
 386 {
 387         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 388                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 389                 return -EOPNOTSUPP;
 390         return 0;
 391 }
 392
 393 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 394                         enum btrfs_exclusive_operation type)
 395 {
 396         return !cmpxchg(&fs_info->exclusive_operation, BTRFS_EXCLOP_NONE, type);
 397 }
 398
 399 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 400 {
 401         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 402         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 403 }
 404
 405 /*
 406  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 407  * are zeroed.
 408  */
 409 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 410 {
 411         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 412         struct fsxattr fa;
 413
 414         simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 415         if (copy_to_user(arg, &fa, sizeof(fa)))
 416                 return -EFAULT;
 417
 418         return 0;
 419 }
 420
 421 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 422 {
 423         struct inode *inode = file_inode(file);
 424         struct btrfs_inode *binode = BTRFS_I(inode);
 425         struct btrfs_root *root = binode->root;
 426         struct btrfs_trans_handle *trans;
 427         struct fsxattr fa, old_fa;
 428         unsigned old_flags;
 429         unsigned old_i_flags;
 430         int ret = 0;
 431
 432         if (!inode_owner_or_capable(inode))
 433                 return -EPERM;
 434
 435         if (btrfs_root_readonly(root))
 436                 return -EROFS;
 437
 438         if (copy_from_user(&fa, arg, sizeof(fa)))
 439                 return -EFAULT;
 440
 441         ret = check_xflags(fa.fsx_xflags);
 442         if (ret)
 443                 return ret;
 444
 445         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 446                 return -EOPNOTSUPP;
 447
 448         ret = mnt_want_write_file(file);
 449         if (ret)
 450                 return ret;
 451
 452         inode_lock(inode);
 453
 454         old_flags = binode->flags;
 455         old_i_flags = inode->i_flags;
 456
 457         simple_fill_fsxattr(&old_fa,
 458                             btrfs_inode_flags_to_xflags(binode->flags));
 459         ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 460         if (ret)
 461                 goto out_unlock;
 462
 463         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 464                 binode->flags |= BTRFS_INODE_SYNC;
 465         else
 466                 binode->flags &= ~BTRFS_INODE_SYNC;
 467         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 468                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 469         else
 470                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 471         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 472                 binode->flags |= BTRFS_INODE_APPEND;
 473         else
 474                 binode->flags &= ~BTRFS_INODE_APPEND;
 475         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 476                 binode->flags |= BTRFS_INODE_NODUMP;
 477         else
 478                 binode->flags &= ~BTRFS_INODE_NODUMP;
 479         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 480                 binode->flags |= BTRFS_INODE_NOATIME;
 481         else
 482                 binode->flags &= ~BTRFS_INODE_NOATIME;
 483
 484         /* 1 item for the inode */
 485         trans = btrfs_start_transaction(root, 1);
 486         if (IS_ERR(trans)) {
 487                 ret = PTR_ERR(trans);
 488                 goto out_unlock;
 489         }
 490
 491         btrfs_sync_inode_flags_to_i_flags(inode);
 492         inode_inc_iversion(inode);
 493         inode->i_ctime = current_time(inode);
 494         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 495
 496         btrfs_end_transaction(trans);
 497
 498 out_unlock:
 499         if (ret) {
 500                 binode->flags = old_flags;
 501                 inode->i_flags = old_i_flags;
 502         }
 503
 504         inode_unlock(inode);
 505         mnt_drop_write_file(file);
 506
 507         return ret;
 508 }
 509
 510 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 511 {
 512         struct inode *inode = file_inode(file);
 513
 514         return put_user(inode->i_generation, arg);
 515 }
 516
 517 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 518                                         void __user *arg)
 519 {
 520         struct btrfs_device *device;
 521         struct request_queue *q;
 522         struct fstrim_range range;
 523         u64 minlen = ULLONG_MAX;
 524         u64 num_devices = 0;
 525         int ret;
 526
 527         if (!capable(CAP_SYS_ADMIN))
 528                 return -EPERM;
 529
 530         /*
 531          * btrfs_trim_block_group() depends on space cache, which is not
 532          * available in zoned filesystem. So, disallow fitrim on a zoned
 533          * filesystem for now.
 534          */
 535         if (btrfs_is_zoned(fs_info))
 536                 return -EOPNOTSUPP;
 537
 538         /*
 539          * If the fs is mounted with nologreplay, which requires it to be
 540          * mounted in RO mode as well, we can not allow discard on free space
 541          * inside block groups, because log trees refer to extents that are not
 542          * pinned in a block group's free space cache (pinning the extents is
 543          * precisely the first phase of replaying a log tree).
 544          */
 545         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 546                 return -EROFS;
 547
 548         rcu_read_lock();
 549         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 550                                 dev_list) {
 551                 if (!device->bdev)
 552                         continue;
 553                 q = bdev_get_queue(device->bdev);
 554                 if (blk_queue_discard(q)) {
 555                         num_devices++;
 556                         minlen = min_t(u64, q->limits.discard_granularity,
 557                                      minlen);
 558                 }
 559         }
 560         rcu_read_unlock();
 561
 562         if (!num_devices)
 563                 return -EOPNOTSUPP;
 564         if (copy_from_user(&range, arg, sizeof(range)))
 565                 return -EFAULT;
 566
 567         /*
 568          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 569          * block group is in the logical address space, which can be any
 570          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 571          */
 572         if (range.len < fs_info->sb->s_blocksize)
 573                 return -EINVAL;
 574
 575         range.minlen = max(range.minlen, minlen);
 576         ret = btrfs_trim_fs(fs_info, &range);
 577         if (ret < 0)
 578                 return ret;
 579
 580         if (copy_to_user(arg, &range, sizeof(range)))
 581                 return -EFAULT;
 582
 583         return 0;
 584 }
 585
 586 int __pure btrfs_is_empty_uuid(u8 *uuid)
 587 {
 588         int i;
 589
 590         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 591                 if (uuid[i])
 592                         return 0;
 593         }
 594         return 1;
 595 }
 596
 597 static noinline int create_subvol(struct inode *dir,
 598                                   struct dentry *dentry,
 599                                   const char *name, int namelen,
 600                                   struct btrfs_qgroup_inherit *inherit)
 601 {
 602         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 603         struct btrfs_trans_handle *trans;
 604         struct btrfs_key key;
 605         struct btrfs_root_item *root_item;
 606         struct btrfs_inode_item *inode_item;
 607         struct extent_buffer *leaf;
 608         struct btrfs_root *root = BTRFS_I(dir)->root;
 609         struct btrfs_root *new_root;
 610         struct btrfs_block_rsv block_rsv;
 611         struct timespec64 cur_time = current_time(dir);
 612         struct inode *inode;
 613         int ret;
 614         int err;
 615         dev_t anon_dev = 0;
 616         u64 objectid;
 617         u64 index = 0;
 618
 619         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 620         if (!root_item)
 621                 return -ENOMEM;
 622
 623         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 624         if (ret)
 625                 goto fail_free;
 626
 627         ret = get_anon_bdev(&anon_dev);
 628         if (ret < 0)
 629                 goto fail_free;
 630
 631         /*
 632          * Don't create subvolume whose level is not zero. Or qgroup will be
 633          * screwed up since it assumes subvolume qgroup's level to be 0.
 634          */
 635         if (btrfs_qgroup_level(objectid)) {
 636                 ret = -ENOSPC;
 637                 goto fail_free;
 638         }
 639
 640         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 641         /*
 642          * The same as the snapshot creation, please see the comment
 643          * of create_snapshot().
 644          */
 645         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 646         if (ret)
 647                 goto fail_free;
 648
 649         trans = btrfs_start_transaction(root, 0);
 650         if (IS_ERR(trans)) {
 651                 ret = PTR_ERR(trans);
 652                 btrfs_subvolume_release_metadata(root, &block_rsv);
 653                 goto fail_free;
 654         }
 655         trans->block_rsv = &block_rsv;
 656         trans->bytes_reserved = block_rsv.size;
 657
 658         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 659         if (ret)
 660                 goto fail;
 661
 662         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 663                                       BTRFS_NESTING_NORMAL);
 664         if (IS_ERR(leaf)) {
 665                 ret = PTR_ERR(leaf);
 666                 goto fail;
 667         }
 668
 669         btrfs_mark_buffer_dirty(leaf);
 670
 671         inode_item = &root_item->inode;
 672         btrfs_set_stack_inode_generation(inode_item, 1);
 673         btrfs_set_stack_inode_size(inode_item, 3);
 674         btrfs_set_stack_inode_nlink(inode_item, 1);
 675         btrfs_set_stack_inode_nbytes(inode_item,
 676                                      fs_info->nodesize);
 677         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 678
 679         btrfs_set_root_flags(root_item, 0);
 680         btrfs_set_root_limit(root_item, 0);
 681         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 682
 683         btrfs_set_root_bytenr(root_item, leaf->start);
 684         btrfs_set_root_generation(root_item, trans->transid);
 685         btrfs_set_root_level(root_item, 0);
 686         btrfs_set_root_refs(root_item, 1);
 687         btrfs_set_root_used(root_item, leaf->len);
 688         btrfs_set_root_last_snapshot(root_item, 0);
 689
 690         btrfs_set_root_generation_v2(root_item,
 691                         btrfs_root_generation(root_item));
 692         generate_random_guid(root_item->uuid);
 693         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 694         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 695         root_item->ctime = root_item->otime;
 696         btrfs_set_root_ctransid(root_item, trans->transid);
 697         btrfs_set_root_otransid(root_item, trans->transid);
 698
 699         btrfs_tree_unlock(leaf);
 700         free_extent_buffer(leaf);
 701         leaf = NULL;
 702
 703         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 704
 705         key.objectid = objectid;
 706         key.offset = 0;
 707         key.type = BTRFS_ROOT_ITEM_KEY;
 708         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 709                                 root_item);
 710         if (ret)
 711                 goto fail;
 712
 713         key.offset = (u64)-1;
 714         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 715         if (IS_ERR(new_root)) {
 716                 free_anon_bdev(anon_dev);
 717                 ret = PTR_ERR(new_root);
 718                 btrfs_abort_transaction(trans, ret);
 719                 goto fail;
 720         }
 721         /* Freeing will be done in btrfs_put_root() of new_root */
 722         anon_dev = 0;
 723
 724         btrfs_record_root_in_trans(trans, new_root);
 725
 726         ret = btrfs_create_subvol_root(trans, new_root, root);
 727         btrfs_put_root(new_root);
 728         if (ret) {
 729                 /* We potentially lose an unused inode item here */
 730                 btrfs_abort_transaction(trans, ret);
 731                 goto fail;
 732         }
 733
 734         /*
 735          * insert the directory item
 736          */
 737         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 738         if (ret) {
 739                 btrfs_abort_transaction(trans, ret);
 740                 goto fail;
 741         }
 742
 743         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 744                                     BTRFS_FT_DIR, index);
 745         if (ret) {
 746                 btrfs_abort_transaction(trans, ret);
 747                 goto fail;
 748         }
 749
 750         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 751         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 752         if (ret) {
 753                 btrfs_abort_transaction(trans, ret);
 754                 goto fail;
 755         }
 756
 757         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 758                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 759         if (ret) {
 760                 btrfs_abort_transaction(trans, ret);
 761                 goto fail;
 762         }
 763
 764         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 765                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 766         if (ret)
 767                 btrfs_abort_transaction(trans, ret);
 768
 769 fail:
 770         kfree(root_item);
 771         trans->block_rsv = NULL;
 772         trans->bytes_reserved = 0;
 773         btrfs_subvolume_release_metadata(root, &block_rsv);
 774
 775         err = btrfs_commit_transaction(trans);
 776         if (err && !ret)
 777                 ret = err;
 778
 779         if (!ret) {
 780                 inode = btrfs_lookup_dentry(dir, dentry);
 781                 if (IS_ERR(inode))
 782                         return PTR_ERR(inode);
 783                 d_instantiate(dentry, inode);
 784         }
 785         return ret;
 786
 787 fail_free:
 788         if (anon_dev)
 789                 free_anon_bdev(anon_dev);
 790         kfree(root_item);
 791         return ret;
 792 }
 793
 794 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 795                            struct dentry *dentry, bool readonly,
 796                            struct btrfs_qgroup_inherit *inherit)
 797 {
 798         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 799         struct inode *inode;
 800         struct btrfs_pending_snapshot *pending_snapshot;
 801         struct btrfs_trans_handle *trans;
 802         int ret;
 803
 804         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 805                 return -EINVAL;
 806
 807         if (atomic_read(&root->nr_swapfiles)) {
 808                 btrfs_warn(fs_info,
 809                            "cannot snapshot subvolume with active swapfile");
 810                 return -ETXTBSY;
 811         }
 812
 813         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 814         if (!pending_snapshot)
 815                 return -ENOMEM;
 816
 817         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 818         if (ret < 0)
 819                 goto free_pending;
 820         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 821                         GFP_KERNEL);
 822         pending_snapshot->path = btrfs_alloc_path();
 823         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 824                 ret = -ENOMEM;
 825                 goto free_pending;
 826         }
 827
 828         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 829                              BTRFS_BLOCK_RSV_TEMP);
 830         /*
 831          * 1 - parent dir inode
 832          * 2 - dir entries
 833          * 1 - root item
 834          * 2 - root ref/backref
 835          * 1 - root of snapshot
 836          * 1 - UUID item
 837          */
 838         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 839                                         &pending_snapshot->block_rsv, 8,
 840                                         false);
 841         if (ret)
 842                 goto free_pending;
 843
 844         pending_snapshot->dentry = dentry;
 845         pending_snapshot->root = root;
 846         pending_snapshot->readonly = readonly;
 847         pending_snapshot->dir = dir;
 848         pending_snapshot->inherit = inherit;
 849
 850         trans = btrfs_start_transaction(root, 0);
 851         if (IS_ERR(trans)) {
 852                 ret = PTR_ERR(trans);
 853                 goto fail;
 854         }
 855
 856         spin_lock(&fs_info->trans_lock);
 857         list_add(&pending_snapshot->list,
 858                  &trans->transaction->pending_snapshots);
 859         spin_unlock(&fs_info->trans_lock);
 860
 861         ret = btrfs_commit_transaction(trans);
 862         if (ret)
 863                 goto fail;
 864
 865         ret = pending_snapshot->error;
 866         if (ret)
 867                 goto fail;
 868
 869         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 870         if (ret)
 871                 goto fail;
 872
 873         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 874         if (IS_ERR(inode)) {
 875                 ret = PTR_ERR(inode);
 876                 goto fail;
 877         }
 878
 879         d_instantiate(dentry, inode);
 880         ret = 0;
 881         pending_snapshot->anon_dev = 0;
 882 fail:
 883         /* Prevent double freeing of anon_dev */
 884         if (ret && pending_snapshot->snap)
 885                 pending_snapshot->snap->anon_dev = 0;
 886         btrfs_put_root(pending_snapshot->snap);
 887         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 888 free_pending:
 889         if (pending_snapshot->anon_dev)
 890                 free_anon_bdev(pending_snapshot->anon_dev);
 891         kfree(pending_snapshot->root_item);
 892         btrfs_free_path(pending_snapshot->path);
 893         kfree(pending_snapshot);
 894
 895         return ret;
 896 }
 897
 898 /*  copy of may_delete in fs/namei.c()
 899  *      Check whether we can remove a link victim from directory dir, check
 900  *  whether the type of victim is right.
 901  *  1. We can't do it if dir is read-only (done in permission())
 902  *  2. We should have write and exec permissions on dir
 903  *  3. We can't remove anything from append-only dir
 904  *  4. We can't do anything with immutable dir (done in permission())
 905  *  5. If the sticky bit on dir is set we should either
 906  *      a. be owner of dir, or
 907  *      b. be owner of victim, or
 908  *      c. have CAP_FOWNER capability
 909  *  6. If the victim is append-only or immutable we can't do anything with
 910  *     links pointing to it.
 911  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 912  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 913  *  9. We can't remove a root or mountpoint.
 914  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 915  *     nfs_async_unlink().
 916  */
 917
 918 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 919 {
 920         int error;
 921
 922         if (d_really_is_negative(victim))
 923                 return -ENOENT;
 924
 925         BUG_ON(d_inode(victim->d_parent) != dir);
 926         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 927
 928         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 929         if (error)
 930                 return error;
 931         if (IS_APPEND(dir))
 932                 return -EPERM;
 933         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 934             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 935                 return -EPERM;
 936         if (isdir) {
 937                 if (!d_is_dir(victim))
 938                         return -ENOTDIR;
 939                 if (IS_ROOT(victim))
 940                         return -EBUSY;
 941         } else if (d_is_dir(victim))
 942                 return -EISDIR;
 943         if (IS_DEADDIR(dir))
 944                 return -ENOENT;
 945         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 946                 return -EBUSY;
 947         return 0;
 948 }
 949
 950 /* copy of may_create in fs/namei.c() */
 951 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 952 {
 953         if (d_really_is_positive(child))
 954                 return -EEXIST;
 955         if (IS_DEADDIR(dir))
 956                 return -ENOENT;
 957         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 958 }
 959
 960 /*
 961  * Create a new subvolume below @parent.  This is largely modeled after
 962  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 963  * inside this filesystem so it's quite a bit simpler.
 964  */
 965 static noinline int btrfs_mksubvol(const struct path *parent,
 966                                    const char *name, int namelen,
 967                                    struct btrfs_root *snap_src,
 968                                    bool readonly,
 969                                    struct btrfs_qgroup_inherit *inherit)
 970 {
 971         struct inode *dir = d_inode(parent->dentry);
 972         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 973         struct dentry *dentry;
 974         int error;
 975
 976         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 977         if (error == -EINTR)
 978                 return error;
 979
 980         dentry = lookup_one_len(name, parent->dentry, namelen);
 981         error = PTR_ERR(dentry);
 982         if (IS_ERR(dentry))
 983                 goto out_unlock;
 984
 985         error = btrfs_may_create(dir, dentry);
 986         if (error)
 987                 goto out_dput;
 988
 989         /*
 990          * even if this name doesn't exist, we may get hash collisions.
 991          * check for them now when we can safely fail
 992          */
 993         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 994                                                dir->i_ino, name,
 995                                                namelen);
 996         if (error)
 997                 goto out_dput;
 998
 999         down_read(&fs_info->subvol_sem);
1000
1001         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
1002                 goto out_up_read;
1003
1004         if (snap_src)
1005                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
1006         else
1007                 error = create_subvol(dir, dentry, name, namelen, inherit);
1008
1009         if (!error)
1010                 fsnotify_mkdir(dir, dentry);
1011 out_up_read:
1012         up_read(&fs_info->subvol_sem);
1013 out_dput:
1014         dput(dentry);
1015 out_unlock:
1016         inode_unlock(dir);
1017         return error;
1018 }
1019
1020 static noinline int btrfs_mksnapshot(const struct path *parent,
1021                                    const char *name, int namelen,
1022                                    struct btrfs_root *root,
1023                                    bool readonly,
1024                                    struct btrfs_qgroup_inherit *inherit)
1025 {
1026         int ret;
1027         bool snapshot_force_cow = false;
1028
1029         /*
1030          * Force new buffered writes to reserve space even when NOCOW is
1031          * possible. This is to avoid later writeback (running dealloc) to
1032          * fallback to COW mode and unexpectedly fail with ENOSPC.
1033          */
1034         btrfs_drew_read_lock(&root->snapshot_lock);
1035
1036         ret = btrfs_start_delalloc_snapshot(root);
1037         if (ret)
1038                 goto out;
1039
1040         /*
1041          * All previous writes have started writeback in NOCOW mode, so now
1042          * we force future writes to fallback to COW mode during snapshot
1043          * creation.
1044          */
1045         atomic_inc(&root->snapshot_force_cow);
1046         snapshot_force_cow = true;
1047
1048         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1049
1050         ret = btrfs_mksubvol(parent, name, namelen,
1051                              root, readonly, inherit);
1052 out:
1053         if (snapshot_force_cow)
1054                 atomic_dec(&root->snapshot_force_cow);
1055         btrfs_drew_read_unlock(&root->snapshot_lock);
1056         return ret;
1057 }
1058
1059 /*
1060  * When we're defragging a range, we don't want to kick it off again
1061  * if it is really just waiting for delalloc to send it down.
1062  * If we find a nice big extent or delalloc range for the bytes in the
1063  * file you want to defrag, we return 0 to let you know to skip this
1064  * part of the file
1065  */
1066 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1067 {
1068         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1069         struct extent_map *em = NULL;
1070         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1071         u64 end;
1072
1073         read_lock(&em_tree->lock);
1074         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1075         read_unlock(&em_tree->lock);
1076
1077         if (em) {
1078                 end = extent_map_end(em);
1079                 free_extent_map(em);
1080                 if (end - offset > thresh)
1081                         return 0;
1082         }
1083         /* if we already have a nice delalloc here, just stop */
1084         thresh /= 2;
1085         end = count_range_bits(io_tree, &offset, offset + thresh,
1086                                thresh, EXTENT_DELALLOC, 1);
1087         if (end >= thresh)
1088                 return 0;
1089         return 1;
1090 }
1091
1092 /*
1093  * helper function to walk through a file and find extents
1094  * newer than a specific transid, and smaller than thresh.
1095  *
1096  * This is used by the defragging code to find new and small
1097  * extents
1098  */
1099 static int find_new_extents(struct btrfs_root *root,
1100                             struct inode *inode, u64 newer_than,
1101                             u64 *off, u32 thresh)
1102 {
1103         struct btrfs_path *path;
1104         struct btrfs_key min_key;
1105         struct extent_buffer *leaf;
1106         struct btrfs_file_extent_item *extent;
1107         int type;
1108         int ret;
1109         u64 ino = btrfs_ino(BTRFS_I(inode));
1110
1111         path = btrfs_alloc_path();
1112         if (!path)
1113                 return -ENOMEM;
1114
1115         min_key.objectid = ino;
1116         min_key.type = BTRFS_EXTENT_DATA_KEY;
1117         min_key.offset = *off;
1118
1119         while (1) {
1120                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1121                 if (ret != 0)
1122                         goto none;
1123 process_slot:
1124                 if (min_key.objectid != ino)
1125                         goto none;
1126                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1127                         goto none;
1128
1129                 leaf = path->nodes[0];
1130                 extent = btrfs_item_ptr(leaf, path->slots[0],
1131                                         struct btrfs_file_extent_item);
1132
1133                 type = btrfs_file_extent_type(leaf, extent);
1134                 if (type == BTRFS_FILE_EXTENT_REG &&
1135                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1136                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1137                         *off = min_key.offset;
1138                         btrfs_free_path(path);
1139                         return 0;
1140                 }
1141
1142                 path->slots[0]++;
1143                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1144                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1145                         goto process_slot;
1146                 }
1147
1148                 if (min_key.offset == (u64)-1)
1149                         goto none;
1150
1151                 min_key.offset++;
1152                 btrfs_release_path(path);
1153         }
1154 none:
1155         btrfs_free_path(path);
1156         return -ENOENT;
1157 }
1158
1159 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1160 {
1161         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1162         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1163         struct extent_map *em;
1164         u64 len = PAGE_SIZE;
1165
1166         /*
1167          * hopefully we have this extent in the tree already, try without
1168          * the full extent lock
1169          */
1170         read_lock(&em_tree->lock);
1171         em = lookup_extent_mapping(em_tree, start, len);
1172         read_unlock(&em_tree->lock);
1173
1174         if (!em) {
1175                 struct extent_state *cached = NULL;
1176                 u64 end = start + len - 1;
1177
1178                 /* get the big lock and read metadata off disk */
1179                 lock_extent_bits(io_tree, start, end, &cached);
1180                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1181                 unlock_extent_cached(io_tree, start, end, &cached);
1182
1183                 if (IS_ERR(em))
1184                         return NULL;
1185         }
1186
1187         return em;
1188 }
1189
1190 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1191 {
1192         struct extent_map *next;
1193         bool ret = true;
1194
1195         /* this is the last extent */
1196         if (em->start + em->len >= i_size_read(inode))
1197                 return false;
1198
1199         next = defrag_lookup_extent(inode, em->start + em->len);
1200         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1201                 ret = false;
1202         else if ((em->block_start + em->block_len == next->block_start) &&
1203                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1204                 ret = false;
1205
1206         free_extent_map(next);
1207         return ret;
1208 }
1209
1210 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1211                                u64 *last_len, u64 *skip, u64 *defrag_end,
1212                                int compress)
1213 {
1214         struct extent_map *em;
1215         int ret = 1;
1216         bool next_mergeable = true;
1217         bool prev_mergeable = true;
1218
1219         /*
1220          * make sure that once we start defragging an extent, we keep on
1221          * defragging it
1222          */
1223         if (start < *defrag_end)
1224                 return 1;
1225
1226         *skip = 0;
1227
1228         em = defrag_lookup_extent(inode, start);
1229         if (!em)
1230                 return 0;
1231
1232         /* this will cover holes, and inline extents */
1233         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1234                 ret = 0;
1235                 goto out;
1236         }
1237
1238         if (!*defrag_end)
1239                 prev_mergeable = false;
1240
1241         next_mergeable = defrag_check_next_extent(inode, em);
1242         /*
1243          * we hit a real extent, if it is big or the next extent is not a
1244          * real extent, don't bother defragging it
1245          */
1246         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1247             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1248                 ret = 0;
1249 out:
1250         /*
1251          * last_len ends up being a counter of how many bytes we've defragged.
1252          * every time we choose not to defrag an extent, we reset *last_len
1253          * so that the next tiny extent will force a defrag.
1254          *
1255          * The end result of this is that tiny extents before a single big
1256          * extent will force at least part of that big extent to be defragged.
1257          */
1258         if (ret) {
1259                 *defrag_end = extent_map_end(em);
1260         } else {
1261                 *last_len = 0;
1262                 *skip = extent_map_end(em);
1263                 *defrag_end = 0;
1264         }
1265
1266         free_extent_map(em);
1267         return ret;
1268 }
1269
1270 /*
1271  * it doesn't do much good to defrag one or two pages
1272  * at a time.  This pulls in a nice chunk of pages
1273  * to COW and defrag.
1274  *
1275  * It also makes sure the delalloc code has enough
1276  * dirty data to avoid making new small extents as part
1277  * of the defrag
1278  *
1279  * It's a good idea to start RA on this range
1280  * before calling this.
1281  */
1282 static int cluster_pages_for_defrag(struct inode *inode,
1283                                     struct page **pages,
1284                                     unsigned long start_index,
1285                                     unsigned long num_pages)
1286 {
1287         unsigned long file_end;
1288         u64 isize = i_size_read(inode);
1289         u64 page_start;
1290         u64 page_end;
1291         u64 page_cnt;
1292         u64 start = (u64)start_index << PAGE_SHIFT;
1293         u64 search_start;
1294         int ret;
1295         int i;
1296         int i_done;
1297         struct btrfs_ordered_extent *ordered;
1298         struct extent_state *cached_state = NULL;
1299         struct extent_io_tree *tree;
1300         struct extent_changeset *data_reserved = NULL;
1301         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1302
1303         file_end = (isize - 1) >> PAGE_SHIFT;
1304         if (!isize || start_index > file_end)
1305                 return 0;
1306
1307         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1308
1309         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1310                         start, page_cnt << PAGE_SHIFT);
1311         if (ret)
1312                 return ret;
1313         i_done = 0;
1314         tree = &BTRFS_I(inode)->io_tree;
1315
1316         /* step one, lock all the pages */
1317         for (i = 0; i < page_cnt; i++) {
1318                 struct page *page;
1319 again:
1320                 page = find_or_create_page(inode->i_mapping,
1321                                            start_index + i, mask);
1322                 if (!page)
1323                         break;
1324
1325                 ret = set_page_extent_mapped(page);
1326                 if (ret < 0) {
1327                         unlock_page(page);
1328                         put_page(page);
1329                         break;
1330                 }
1331
1332                 page_start = page_offset(page);
1333                 page_end = page_start + PAGE_SIZE - 1;
1334                 while (1) {
1335                         lock_extent_bits(tree, page_start, page_end,
1336                                          &cached_state);
1337                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1338                                                               page_start);
1339                         unlock_extent_cached(tree, page_start, page_end,
1340                                              &cached_state);
1341                         if (!ordered)
1342                                 break;
1343
1344                         unlock_page(page);
1345                         btrfs_start_ordered_extent(ordered, 1);
1346                         btrfs_put_ordered_extent(ordered);
1347                         lock_page(page);
1348                         /*
1349                          * we unlocked the page above, so we need check if
1350                          * it was released or not.
1351                          */
1352                         if (page->mapping != inode->i_mapping) {
1353                                 unlock_page(page);
1354                                 put_page(page);
1355                                 goto again;
1356                         }
1357                 }
1358
1359                 if (!PageUptodate(page)) {
1360                         btrfs_readpage(NULL, page);
1361                         lock_page(page);
1362                         if (!PageUptodate(page)) {
1363                                 unlock_page(page);
1364                                 put_page(page);
1365                                 ret = -EIO;
1366                                 break;
1367                         }
1368                 }
1369
1370                 if (page->mapping != inode->i_mapping) {
1371                         unlock_page(page);
1372                         put_page(page);
1373                         goto again;
1374                 }
1375
1376                 pages[i] = page;
1377                 i_done++;
1378         }
1379         if (!i_done || ret)
1380                 goto out;
1381
1382         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1383                 goto out;
1384
1385         /*
1386          * so now we have a nice long stream of locked
1387          * and up to date pages, lets wait on them
1388          */
1389         for (i = 0; i < i_done; i++)
1390                 wait_on_page_writeback(pages[i]);
1391
1392         page_start = page_offset(pages[0]);
1393         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1394
1395         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1396                          page_start, page_end - 1, &cached_state);
1397
1398         /*
1399          * When defragmenting we skip ranges that have holes or inline extents,
1400          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1401          * space. At btrfs_defrag_file(), we check if a range should be defragged
1402          * before locking the inode and then, if it should, we trigger a sync
1403          * page cache readahead - we lock the inode only after that to avoid
1404          * blocking for too long other tasks that possibly want to operate on
1405          * other file ranges. But before we were able to get the inode lock,
1406          * some other task may have punched a hole in the range, or we may have
1407          * now an inline extent, in which case we should not defrag. So check
1408          * for that here, where we have the inode and the range locked, and bail
1409          * out if that happened.
1410          */
1411         search_start = page_start;
1412         while (search_start < page_end) {
1413                 struct extent_map *em;
1414
1415                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1416                                       page_end - search_start);
1417                 if (IS_ERR(em)) {
1418                         ret = PTR_ERR(em);
1419                         goto out_unlock_range;
1420                 }
1421                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1422                         free_extent_map(em);
1423                         /* Ok, 0 means we did not defrag anything */
1424                         ret = 0;
1425                         goto out_unlock_range;
1426                 }
1427                 search_start = extent_map_end(em);
1428                 free_extent_map(em);
1429         }
1430
1431         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1432                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1433                           EXTENT_DEFRAG, 0, 0, &cached_state);
1434
1435         if (i_done != page_cnt) {
1436                 spin_lock(&BTRFS_I(inode)->lock);
1437                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1438                 spin_unlock(&BTRFS_I(inode)->lock);
1439                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1440                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1441         }
1442
1443
1444         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1445                           &cached_state);
1446
1447         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1448                              page_start, page_end - 1, &cached_state);
1449
1450         for (i = 0; i < i_done; i++) {
1451                 clear_page_dirty_for_io(pages[i]);
1452                 ClearPageChecked(pages[i]);
1453                 set_page_dirty(pages[i]);
1454                 unlock_page(pages[i]);
1455                 put_page(pages[i]);
1456         }
1457         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1458         extent_changeset_free(data_reserved);
1459         return i_done;
1460
1461 out_unlock_range:
1462         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1463                              page_start, page_end - 1, &cached_state);
1464 out:
1465         for (i = 0; i < i_done; i++) {
1466                 unlock_page(pages[i]);
1467                 put_page(pages[i]);
1468         }
1469         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1470                         start, page_cnt << PAGE_SHIFT, true);
1471         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1472         extent_changeset_free(data_reserved);
1473         return ret;
1474
1475 }
1476
1477 int btrfs_defrag_file(struct inode *inode, struct file *file,
1478                       struct btrfs_ioctl_defrag_range_args *range,
1479                       u64 newer_than, unsigned long max_to_defrag)
1480 {
1481         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1482         struct btrfs_root *root = BTRFS_I(inode)->root;
1483         struct file_ra_state *ra = NULL;
1484         unsigned long last_index;
1485         u64 isize = i_size_read(inode);
1486         u64 last_len = 0;
1487         u64 skip = 0;
1488         u64 defrag_end = 0;
1489         u64 newer_off = range->start;
1490         unsigned long i;
1491         unsigned long ra_index = 0;
1492         int ret;
1493         int defrag_count = 0;
1494         int compress_type = BTRFS_COMPRESS_ZLIB;
1495         u32 extent_thresh = range->extent_thresh;
1496         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1497         unsigned long cluster = max_cluster;
1498         u64 new_align = ~((u64)SZ_128K - 1);
1499         struct page **pages = NULL;
1500         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1501
1502         if (isize == 0)
1503                 return 0;
1504
1505         if (range->start >= isize)
1506                 return -EINVAL;
1507
1508         if (do_compress) {
1509                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1510                         return -EINVAL;
1511                 if (range->compress_type)
1512                         compress_type = range->compress_type;
1513         }
1514
1515         if (extent_thresh == 0)
1516                 extent_thresh = SZ_256K;
1517
1518         /*
1519          * If we were not given a file, allocate a readahead context. As
1520          * readahead is just an optimization, defrag will work without it so
1521          * we don't error out.
1522          */
1523         if (!file) {
1524                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1525                 if (ra)
1526                         file_ra_state_init(ra, inode->i_mapping);
1527         } else {
1528                 ra = &file->f_ra;
1529         }
1530
1531         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1532         if (!pages) {
1533                 ret = -ENOMEM;
1534                 goto out_ra;
1535         }
1536
1537         /* find the last page to defrag */
1538         if (range->start + range->len > range->start) {
1539                 last_index = min_t(u64, isize - 1,
1540                          range->start + range->len - 1) >> PAGE_SHIFT;
1541         } else {
1542                 last_index = (isize - 1) >> PAGE_SHIFT;
1543         }
1544
1545         if (newer_than) {
1546                 ret = find_new_extents(root, inode, newer_than,
1547                                        &newer_off, SZ_64K);
1548                 if (!ret) {
1549                         range->start = newer_off;
1550                         /*
1551                          * we always align our defrag to help keep
1552                          * the extents in the file evenly spaced
1553                          */
1554                         i = (newer_off & new_align) >> PAGE_SHIFT;
1555                 } else
1556                         goto out_ra;
1557         } else {
1558                 i = range->start >> PAGE_SHIFT;
1559         }
1560         if (!max_to_defrag)
1561                 max_to_defrag = last_index - i + 1;
1562
1563         /*
1564          * make writeback starts from i, so the defrag range can be
1565          * written sequentially.
1566          */
1567         if (i < inode->i_mapping->writeback_index)
1568                 inode->i_mapping->writeback_index = i;
1569
1570         while (i <= last_index && defrag_count < max_to_defrag &&
1571                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1572                 /*
1573                  * make sure we stop running if someone unmounts
1574                  * the FS
1575                  */
1576                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1577                         break;
1578
1579                 if (btrfs_defrag_cancelled(fs_info)) {
1580                         btrfs_debug(fs_info, "defrag_file cancelled");
1581                         ret = -EAGAIN;
1582                         break;
1583                 }
1584
1585                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1586                                          extent_thresh, &last_len, &skip,
1587                                          &defrag_end, do_compress)){
1588                         unsigned long next;
1589                         /*
1590                          * the should_defrag function tells us how much to skip
1591                          * bump our counter by the suggested amount
1592                          */
1593                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1594                         i = max(i + 1, next);
1595                         continue;
1596                 }
1597
1598                 if (!newer_than) {
1599                         cluster = (PAGE_ALIGN(defrag_end) >>
1600                                    PAGE_SHIFT) - i;
1601                         cluster = min(cluster, max_cluster);
1602                 } else {
1603                         cluster = max_cluster;
1604                 }
1605
1606                 if (i + cluster > ra_index) {
1607                         ra_index = max(i, ra_index);
1608                         if (ra)
1609                                 page_cache_sync_readahead(inode->i_mapping, ra,
1610                                                 file, ra_index, cluster);
1611                         ra_index += cluster;
1612                 }
1613
1614                 inode_lock(inode);
1615                 if (IS_SWAPFILE(inode)) {
1616                         ret = -ETXTBSY;
1617                 } else {
1618                         if (do_compress)
1619                                 BTRFS_I(inode)->defrag_compress = compress_type;
1620                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1621                 }
1622                 if (ret < 0) {
1623                         inode_unlock(inode);
1624                         goto out_ra;
1625                 }
1626
1627                 defrag_count += ret;
1628                 balance_dirty_pages_ratelimited(inode->i_mapping);
1629                 inode_unlock(inode);
1630
1631                 if (newer_than) {
1632                         if (newer_off == (u64)-1)
1633                                 break;
1634
1635                         if (ret > 0)
1636                                 i += ret;
1637
1638                         newer_off = max(newer_off + 1,
1639                                         (u64)i << PAGE_SHIFT);
1640
1641                         ret = find_new_extents(root, inode, newer_than,
1642                                                &newer_off, SZ_64K);
1643                         if (!ret) {
1644                                 range->start = newer_off;
1645                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1646                         } else {
1647                                 break;
1648                         }
1649                 } else {
1650                         if (ret > 0) {
1651                                 i += ret;
1652                                 last_len += ret << PAGE_SHIFT;
1653                         } else {
1654                                 i++;
1655                                 last_len = 0;
1656                         }
1657                 }
1658         }
1659
1660         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1661                 filemap_flush(inode->i_mapping);
1662                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1663                              &BTRFS_I(inode)->runtime_flags))
1664                         filemap_flush(inode->i_mapping);
1665         }
1666
1667         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1668                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1669         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1670                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1671         }
1672
1673         ret = defrag_count;
1674
1675 out_ra:
1676         if (do_compress) {
1677                 inode_lock(inode);
1678                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1679                 inode_unlock(inode);
1680         }
1681         if (!file)
1682                 kfree(ra);
1683         kfree(pages);
1684         return ret;
1685 }
1686
1687 static noinline int btrfs_ioctl_resize(struct file *file,
1688                                         void __user *arg)
1689 {
1690         struct inode *inode = file_inode(file);
1691         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1692         u64 new_size;
1693         u64 old_size;
1694         u64 devid = 1;
1695         struct btrfs_root *root = BTRFS_I(inode)->root;
1696         struct btrfs_ioctl_vol_args *vol_args;
1697         struct btrfs_trans_handle *trans;
1698         struct btrfs_device *device = NULL;
1699         char *sizestr;
1700         char *retptr;
1701         char *devstr = NULL;
1702         int ret = 0;
1703         int mod = 0;
1704
1705         if (!capable(CAP_SYS_ADMIN))
1706                 return -EPERM;
1707
1708         ret = mnt_want_write_file(file);
1709         if (ret)
1710                 return ret;
1711
1712         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_RESIZE)) {
1713                 mnt_drop_write_file(file);
1714                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1715         }
1716
1717         vol_args = memdup_user(arg, sizeof(*vol_args));
1718         if (IS_ERR(vol_args)) {
1719                 ret = PTR_ERR(vol_args);
1720                 goto out;
1721         }
1722
1723         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1724
1725         sizestr = vol_args->name;
1726         devstr = strchr(sizestr, ':');
1727         if (devstr) {
1728                 sizestr = devstr + 1;
1729                 *devstr = '\0';
1730                 devstr = vol_args->name;
1731                 ret = kstrtoull(devstr, 10, &devid);
1732                 if (ret)
1733                         goto out_free;
1734                 if (!devid) {
1735                         ret = -EINVAL;
1736                         goto out_free;
1737                 }
1738                 btrfs_info(fs_info, "resizing devid %llu", devid);
1739         }
1740
1741         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
1742         if (!device) {
1743                 btrfs_info(fs_info, "resizer unable to find device %llu",
1744                            devid);
1745                 ret = -ENODEV;
1746                 goto out_free;
1747         }
1748
1749         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1750                 btrfs_info(fs_info,
1751                            "resizer unable to apply on readonly device %llu",
1752                        devid);
1753                 ret = -EPERM;
1754                 goto out_free;
1755         }
1756
1757         if (!strcmp(sizestr, "max"))
1758                 new_size = device->bdev->bd_inode->i_size;
1759         else {
1760                 if (sizestr[0] == '-') {
1761                         mod = -1;
1762                         sizestr++;
1763                 } else if (sizestr[0] == '+') {
1764                         mod = 1;
1765                         sizestr++;
1766                 }
1767                 new_size = memparse(sizestr, &retptr);
1768                 if (*retptr != '\0' || new_size == 0) {
1769                         ret = -EINVAL;
1770                         goto out_free;
1771                 }
1772         }
1773
1774         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1775                 ret = -EPERM;
1776                 goto out_free;
1777         }
1778
1779         old_size = btrfs_device_get_total_bytes(device);
1780
1781         if (mod < 0) {
1782                 if (new_size > old_size) {
1783                         ret = -EINVAL;
1784                         goto out_free;
1785                 }
1786                 new_size = old_size - new_size;
1787         } else if (mod > 0) {
1788                 if (new_size > ULLONG_MAX - old_size) {
1789                         ret = -ERANGE;
1790                         goto out_free;
1791                 }
1792                 new_size = old_size + new_size;
1793         }
1794
1795         if (new_size < SZ_256M) {
1796                 ret = -EINVAL;
1797                 goto out_free;
1798         }
1799         if (new_size > device->bdev->bd_inode->i_size) {
1800                 ret = -EFBIG;
1801                 goto out_free;
1802         }
1803
1804         new_size = round_down(new_size, fs_info->sectorsize);
1805
1806         if (new_size > old_size) {
1807                 trans = btrfs_start_transaction(root, 0);
1808                 if (IS_ERR(trans)) {
1809                         ret = PTR_ERR(trans);
1810                         goto out_free;
1811                 }
1812                 ret = btrfs_grow_device(trans, device, new_size);
1813                 btrfs_commit_transaction(trans);
1814         } else if (new_size < old_size) {
1815                 ret = btrfs_shrink_device(device, new_size);
1816         } /* equal, nothing need to do */
1817
1818         if (ret == 0 && new_size != old_size)
1819                 btrfs_info_in_rcu(fs_info,
1820                         "resize device %s (devid %llu) from %llu to %llu",
1821                         rcu_str_deref(device->name), device->devid,
1822                         old_size, new_size);
1823 out_free:
1824         kfree(vol_args);
1825 out:
1826         btrfs_exclop_finish(fs_info);
1827         mnt_drop_write_file(file);
1828         return ret;
1829 }
1830
1831 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1832                                 const char *name, unsigned long fd, int subvol,
1833                                 bool readonly,
1834                                 struct btrfs_qgroup_inherit *inherit)
1835 {
1836         int namelen;
1837         int ret = 0;
1838
1839         if (!S_ISDIR(file_inode(file)->i_mode))
1840                 return -ENOTDIR;
1841
1842         ret = mnt_want_write_file(file);
1843         if (ret)
1844                 goto out;
1845
1846         namelen = strlen(name);
1847         if (strchr(name, '/')) {
1848                 ret = -EINVAL;
1849                 goto out_drop_write;
1850         }
1851
1852         if (name[0] == '.' &&
1853            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1854                 ret = -EEXIST;
1855                 goto out_drop_write;
1856         }
1857
1858         if (subvol) {
1859                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1860                                      NULL, readonly, inherit);
1861         } else {
1862                 struct fd src = fdget(fd);
1863                 struct inode *src_inode;
1864                 if (!src.file) {
1865                         ret = -EINVAL;
1866                         goto out_drop_write;
1867                 }
1868
1869                 src_inode = file_inode(src.file);
1870                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1871                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1872                                    "Snapshot src from another FS");
1873                         ret = -EXDEV;
1874                 } else if (!inode_owner_or_capable(src_inode)) {
1875                         /*
1876                          * Subvolume creation is not restricted, but snapshots
1877                          * are limited to own subvolumes only
1878                          */
1879                         ret = -EPERM;
1880                 } else {
1881                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1882                                              BTRFS_I(src_inode)->root,
1883                                              readonly, inherit);
1884                 }
1885                 fdput(src);
1886         }
1887 out_drop_write:
1888         mnt_drop_write_file(file);
1889 out:
1890         return ret;
1891 }
1892
1893 static noinline int btrfs_ioctl_snap_create(struct file *file,
1894                                             void __user *arg, int subvol)
1895 {
1896         struct btrfs_ioctl_vol_args *vol_args;
1897         int ret;
1898
1899         if (!S_ISDIR(file_inode(file)->i_mode))
1900                 return -ENOTDIR;
1901
1902         vol_args = memdup_user(arg, sizeof(*vol_args));
1903         if (IS_ERR(vol_args))
1904                 return PTR_ERR(vol_args);
1905         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1906
1907         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1908                                         subvol, false, NULL);
1909
1910         kfree(vol_args);
1911         return ret;
1912 }
1913
1914 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1915                                                void __user *arg, int subvol)
1916 {
1917         struct btrfs_ioctl_vol_args_v2 *vol_args;
1918         int ret;
1919         bool readonly = false;
1920         struct btrfs_qgroup_inherit *inherit = NULL;
1921
1922         if (!S_ISDIR(file_inode(file)->i_mode))
1923                 return -ENOTDIR;
1924
1925         vol_args = memdup_user(arg, sizeof(*vol_args));
1926         if (IS_ERR(vol_args))
1927                 return PTR_ERR(vol_args);
1928         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1929
1930         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1931                 ret = -EOPNOTSUPP;
1932                 goto free_args;
1933         }
1934
1935         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1936                 readonly = true;
1937         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1938                 u64 nums;
1939
1940                 if (vol_args->size < sizeof(*inherit) ||
1941                     vol_args->size > PAGE_SIZE) {
1942                         ret = -EINVAL;
1943                         goto free_args;
1944                 }
1945                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1946                 if (IS_ERR(inherit)) {
1947                         ret = PTR_ERR(inherit);
1948                         goto free_args;
1949                 }
1950
1951                 if (inherit->num_qgroups > PAGE_SIZE ||
1952                     inherit->num_ref_copies > PAGE_SIZE ||
1953                     inherit->num_excl_copies > PAGE_SIZE) {
1954                         ret = -EINVAL;
1955                         goto free_inherit;
1956                 }
1957
1958                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1959                        2 * inherit->num_excl_copies;
1960                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1961                         ret = -EINVAL;
1962                         goto free_inherit;
1963                 }
1964         }
1965
1966         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1967                                         subvol, readonly, inherit);
1968         if (ret)
1969                 goto free_inherit;
1970 free_inherit:
1971         kfree(inherit);
1972 free_args:
1973         kfree(vol_args);
1974         return ret;
1975 }
1976
1977 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1978                                                 void __user *arg)
1979 {
1980         struct inode *inode = file_inode(file);
1981         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1982         struct btrfs_root *root = BTRFS_I(inode)->root;
1983         int ret = 0;
1984         u64 flags = 0;
1985
1986         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1987                 return -EINVAL;
1988
1989         down_read(&fs_info->subvol_sem);
1990         if (btrfs_root_readonly(root))
1991                 flags |= BTRFS_SUBVOL_RDONLY;
1992         up_read(&fs_info->subvol_sem);
1993
1994         if (copy_to_user(arg, &flags, sizeof(flags)))
1995                 ret = -EFAULT;
1996
1997         return ret;
1998 }
1999
2000 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2001                                               void __user *arg)
2002 {
2003         struct inode *inode = file_inode(file);
2004         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2005         struct btrfs_root *root = BTRFS_I(inode)->root;
2006         struct btrfs_trans_handle *trans;
2007         u64 root_flags;
2008         u64 flags;
2009         int ret = 0;
2010
2011         if (!inode_owner_or_capable(inode))
2012                 return -EPERM;
2013
2014         ret = mnt_want_write_file(file);
2015         if (ret)
2016                 goto out;
2017
2018         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2019                 ret = -EINVAL;
2020                 goto out_drop_write;
2021         }
2022
2023         if (copy_from_user(&flags, arg, sizeof(flags))) {
2024                 ret = -EFAULT;
2025                 goto out_drop_write;
2026         }
2027
2028         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2029                 ret = -EOPNOTSUPP;
2030                 goto out_drop_write;
2031         }
2032
2033         down_write(&fs_info->subvol_sem);
2034
2035         /* nothing to do */
2036         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2037                 goto out_drop_sem;
2038
2039         root_flags = btrfs_root_flags(&root->root_item);
2040         if (flags & BTRFS_SUBVOL_RDONLY) {
2041                 btrfs_set_root_flags(&root->root_item,
2042                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2043         } else {
2044                 /*
2045                  * Block RO -> RW transition if this subvolume is involved in
2046                  * send
2047                  */
2048                 spin_lock(&root->root_item_lock);
2049                 if (root->send_in_progress == 0) {
2050                         btrfs_set_root_flags(&root->root_item,
2051                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2052                         spin_unlock(&root->root_item_lock);
2053                 } else {
2054                         spin_unlock(&root->root_item_lock);
2055                         btrfs_warn(fs_info,
2056                                    "Attempt to set subvolume %llu read-write during send",
2057                                    root->root_key.objectid);
2058                         ret = -EPERM;
2059                         goto out_drop_sem;
2060                 }
2061         }
2062
2063         trans = btrfs_start_transaction(root, 1);
2064         if (IS_ERR(trans)) {
2065                 ret = PTR_ERR(trans);
2066                 goto out_reset;
2067         }
2068
2069         ret = btrfs_update_root(trans, fs_info->tree_root,
2070                                 &root->root_key, &root->root_item);
2071         if (ret < 0) {
2072                 btrfs_end_transaction(trans);
2073                 goto out_reset;
2074         }
2075
2076         ret = btrfs_commit_transaction(trans);
2077
2078 out_reset:
2079         if (ret)
2080                 btrfs_set_root_flags(&root->root_item, root_flags);
2081 out_drop_sem:
2082         up_write(&fs_info->subvol_sem);
2083 out_drop_write:
2084         mnt_drop_write_file(file);
2085 out:
2086         return ret;
2087 }
2088
2089 static noinline int key_in_sk(struct btrfs_key *key,
2090                               struct btrfs_ioctl_search_key *sk)
2091 {
2092         struct btrfs_key test;
2093         int ret;
2094
2095         test.objectid = sk->min_objectid;
2096         test.type = sk->min_type;
2097         test.offset = sk->min_offset;
2098
2099         ret = btrfs_comp_cpu_keys(key, &test);
2100         if (ret < 0)
2101                 return 0;
2102
2103         test.objectid = sk->max_objectid;
2104         test.type = sk->max_type;
2105         test.offset = sk->max_offset;
2106
2107         ret = btrfs_comp_cpu_keys(key, &test);
2108         if (ret > 0)
2109                 return 0;
2110         return 1;
2111 }
2112
2113 static noinline int copy_to_sk(struct btrfs_path *path,
2114                                struct btrfs_key *key,
2115                                struct btrfs_ioctl_search_key *sk,
2116                                size_t *buf_size,
2117                                char __user *ubuf,
2118                                unsigned long *sk_offset,
2119                                int *num_found)
2120 {
2121         u64 found_transid;
2122         struct extent_buffer *leaf;
2123         struct btrfs_ioctl_search_header sh;
2124         struct btrfs_key test;
2125         unsigned long item_off;
2126         unsigned long item_len;
2127         int nritems;
2128         int i;
2129         int slot;
2130         int ret = 0;
2131
2132         leaf = path->nodes[0];
2133         slot = path->slots[0];
2134         nritems = btrfs_header_nritems(leaf);
2135
2136         if (btrfs_header_generation(leaf) > sk->max_transid) {
2137                 i = nritems;
2138                 goto advance_key;
2139         }
2140         found_transid = btrfs_header_generation(leaf);
2141
2142         for (i = slot; i < nritems; i++) {
2143                 item_off = btrfs_item_ptr_offset(leaf, i);
2144                 item_len = btrfs_item_size_nr(leaf, i);
2145
2146                 btrfs_item_key_to_cpu(leaf, key, i);
2147                 if (!key_in_sk(key, sk))
2148                         continue;
2149
2150                 if (sizeof(sh) + item_len > *buf_size) {
2151                         if (*num_found) {
2152                                 ret = 1;
2153                                 goto out;
2154                         }
2155
2156                         /*
2157                          * return one empty item back for v1, which does not
2158                          * handle -EOVERFLOW
2159                          */
2160
2161                         *buf_size = sizeof(sh) + item_len;
2162                         item_len = 0;
2163                         ret = -EOVERFLOW;
2164                 }
2165
2166                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2167                         ret = 1;
2168                         goto out;
2169                 }
2170
2171                 sh.objectid = key->objectid;
2172                 sh.offset = key->offset;
2173                 sh.type = key->type;
2174                 sh.len = item_len;
2175                 sh.transid = found_transid;
2176
2177                 /*
2178                  * Copy search result header. If we fault then loop again so we
2179                  * can fault in the pages and -EFAULT there if there's a
2180                  * problem. Otherwise we'll fault and then copy the buffer in
2181                  * properly this next time through
2182                  */
2183                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2184                         ret = 0;
2185                         goto out;
2186                 }
2187
2188                 *sk_offset += sizeof(sh);
2189
2190                 if (item_len) {
2191                         char __user *up = ubuf + *sk_offset;
2192                         /*
2193                          * Copy the item, same behavior as above, but reset the
2194                          * * sk_offset so we copy the full thing again.
2195                          */
2196                         if (read_extent_buffer_to_user_nofault(leaf, up,
2197                                                 item_off, item_len)) {
2198                                 ret = 0;
2199                                 *sk_offset -= sizeof(sh);
2200                                 goto out;
2201                         }
2202
2203                         *sk_offset += item_len;
2204                 }
2205                 (*num_found)++;
2206
2207                 if (ret) /* -EOVERFLOW from above */
2208                         goto out;
2209
2210                 if (*num_found >= sk->nr_items) {
2211                         ret = 1;
2212                         goto out;
2213                 }
2214         }
2215 advance_key:
2216         ret = 0;
2217         test.objectid = sk->max_objectid;
2218         test.type = sk->max_type;
2219         test.offset = sk->max_offset;
2220         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2221                 ret = 1;
2222         else if (key->offset < (u64)-1)
2223                 key->offset++;
2224         else if (key->type < (u8)-1) {
2225                 key->offset = 0;
2226                 key->type++;
2227         } else if (key->objectid < (u64)-1) {
2228                 key->offset = 0;
2229                 key->type = 0;
2230                 key->objectid++;
2231         } else
2232                 ret = 1;
2233 out:
2234         /*
2235          *  0: all items from this leaf copied, continue with next
2236          *  1: * more items can be copied, but unused buffer is too small
2237          *     * all items were found
2238          *     Either way, it will stops the loop which iterates to the next
2239          *     leaf
2240          *  -EOVERFLOW: item was to large for buffer
2241          *  -EFAULT: could not copy extent buffer back to userspace
2242          */
2243         return ret;
2244 }
2245
2246 static noinline int search_ioctl(struct inode *inode,
2247                                  struct btrfs_ioctl_search_key *sk,
2248                                  size_t *buf_size,
2249                                  char __user *ubuf)
2250 {
2251         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2252         struct btrfs_root *root;
2253         struct btrfs_key key;
2254         struct btrfs_path *path;
2255         int ret;
2256         int num_found = 0;
2257         unsigned long sk_offset = 0;
2258
2259         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2260                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2261                 return -EOVERFLOW;
2262         }
2263
2264         path = btrfs_alloc_path();
2265         if (!path)
2266                 return -ENOMEM;
2267
2268         if (sk->tree_id == 0) {
2269                 /* search the root of the inode that was passed */
2270                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2271         } else {
2272                 root = btrfs_get_fs_root(info, sk->tree_id, true);
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_put_root(root);
2307         btrfs_free_path(path);
2308         return ret;
2309 }
2310
2311 static noinline int btrfs_ioctl_tree_search(struct file *file,
2312                                            void __user *argp)
2313 {
2314         struct btrfs_ioctl_search_args __user *uargs;
2315         struct btrfs_ioctl_search_key sk;
2316         struct inode *inode;
2317         int ret;
2318         size_t buf_size;
2319
2320         if (!capable(CAP_SYS_ADMIN))
2321                 return -EPERM;
2322
2323         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2324
2325         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2326                 return -EFAULT;
2327
2328         buf_size = sizeof(uargs->buf);
2329
2330         inode = file_inode(file);
2331         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2332
2333         /*
2334          * In the origin implementation an overflow is handled by returning a
2335          * search header with a len of zero, so reset ret.
2336          */
2337         if (ret == -EOVERFLOW)
2338                 ret = 0;
2339
2340         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2341                 ret = -EFAULT;
2342         return ret;
2343 }
2344
2345 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2346                                                void __user *argp)
2347 {
2348         struct btrfs_ioctl_search_args_v2 __user *uarg;
2349         struct btrfs_ioctl_search_args_v2 args;
2350         struct inode *inode;
2351         int ret;
2352         size_t buf_size;
2353         const size_t buf_limit = SZ_16M;
2354
2355         if (!capable(CAP_SYS_ADMIN))
2356                 return -EPERM;
2357
2358         /* copy search header and buffer size */
2359         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2360         if (copy_from_user(&args, uarg, sizeof(args)))
2361                 return -EFAULT;
2362
2363         buf_size = args.buf_size;
2364
2365         /* limit result size to 16MB */
2366         if (buf_size > buf_limit)
2367                 buf_size = buf_limit;
2368
2369         inode = file_inode(file);
2370         ret = search_ioctl(inode, &args.key, &buf_size,
2371                            (char __user *)(&uarg->buf[0]));
2372         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2373                 ret = -EFAULT;
2374         else if (ret == -EOVERFLOW &&
2375                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2376                 ret = -EFAULT;
2377
2378         return ret;
2379 }
2380
2381 /*
2382  * Search INODE_REFs to identify path name of 'dirid' directory
2383  * in a 'tree_id' tree. and sets path name to 'name'.
2384  */
2385 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2386                                 u64 tree_id, u64 dirid, char *name)
2387 {
2388         struct btrfs_root *root;
2389         struct btrfs_key key;
2390         char *ptr;
2391         int ret = -1;
2392         int slot;
2393         int len;
2394         int total_len = 0;
2395         struct btrfs_inode_ref *iref;
2396         struct extent_buffer *l;
2397         struct btrfs_path *path;
2398
2399         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2400                 name[0]='\0';
2401                 return 0;
2402         }
2403
2404         path = btrfs_alloc_path();
2405         if (!path)
2406                 return -ENOMEM;
2407
2408         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2409
2410         root = btrfs_get_fs_root(info, tree_id, true);
2411         if (IS_ERR(root)) {
2412                 ret = PTR_ERR(root);
2413                 root = NULL;
2414                 goto out;
2415         }
2416
2417         key.objectid = dirid;
2418         key.type = BTRFS_INODE_REF_KEY;
2419         key.offset = (u64)-1;
2420
2421         while (1) {
2422                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2423                 if (ret < 0)
2424                         goto out;
2425                 else if (ret > 0) {
2426                         ret = btrfs_previous_item(root, path, dirid,
2427                                                   BTRFS_INODE_REF_KEY);
2428                         if (ret < 0)
2429                                 goto out;
2430                         else if (ret > 0) {
2431                                 ret = -ENOENT;
2432                                 goto out;
2433                         }
2434                 }
2435
2436                 l = path->nodes[0];
2437                 slot = path->slots[0];
2438                 btrfs_item_key_to_cpu(l, &key, slot);
2439
2440                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2441                 len = btrfs_inode_ref_name_len(l, iref);
2442                 ptr -= len + 1;
2443                 total_len += len + 1;
2444                 if (ptr < name) {
2445                         ret = -ENAMETOOLONG;
2446                         goto out;
2447                 }
2448
2449                 *(ptr + len) = '/';
2450                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2451
2452                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2453                         break;
2454
2455                 btrfs_release_path(path);
2456                 key.objectid = key.offset;
2457                 key.offset = (u64)-1;
2458                 dirid = key.objectid;
2459         }
2460         memmove(name, ptr, total_len);
2461         name[total_len] = '\0';
2462         ret = 0;
2463 out:
2464         btrfs_put_root(root);
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 = NULL;
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                 root = btrfs_get_fs_root(fs_info, treeid, true);
2504                 if (IS_ERR(root)) {
2505                         ret = PTR_ERR(root);
2506                         goto out;
2507                 }
2508
2509                 key.objectid = dirid;
2510                 key.type = BTRFS_INODE_REF_KEY;
2511                 key.offset = (u64)-1;
2512                 while (1) {
2513                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2514                         if (ret < 0) {
2515                                 goto out_put;
2516                         } else if (ret > 0) {
2517                                 ret = btrfs_previous_item(root, path, dirid,
2518                                                           BTRFS_INODE_REF_KEY);
2519                                 if (ret < 0) {
2520                                         goto out_put;
2521                                 } else if (ret > 0) {
2522                                         ret = -ENOENT;
2523                                         goto out_put;
2524                                 }
2525                         }
2526
2527                         leaf = path->nodes[0];
2528                         slot = path->slots[0];
2529                         btrfs_item_key_to_cpu(leaf, &key, slot);
2530
2531                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2532                         len = btrfs_inode_ref_name_len(leaf, iref);
2533                         ptr -= len + 1;
2534                         total_len += len + 1;
2535                         if (ptr < args->path) {
2536                                 ret = -ENAMETOOLONG;
2537                                 goto out_put;
2538                         }
2539
2540                         *(ptr + len) = '/';
2541                         read_extent_buffer(leaf, ptr,
2542                                         (unsigned long)(iref + 1), len);
2543
2544                         /* Check the read+exec permission of this directory */
2545                         ret = btrfs_previous_item(root, path, dirid,
2546                                                   BTRFS_INODE_ITEM_KEY);
2547                         if (ret < 0) {
2548                                 goto out_put;
2549                         } else if (ret > 0) {
2550                                 ret = -ENOENT;
2551                                 goto out_put;
2552                         }
2553
2554                         leaf = path->nodes[0];
2555                         slot = path->slots[0];
2556                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2557                         if (key2.objectid != dirid) {
2558                                 ret = -ENOENT;
2559                                 goto out_put;
2560                         }
2561
2562                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2563                         if (IS_ERR(temp_inode)) {
2564                                 ret = PTR_ERR(temp_inode);
2565                                 goto out_put;
2566                         }
2567                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2568                         iput(temp_inode);
2569                         if (ret) {
2570                                 ret = -EACCES;
2571                                 goto out_put;
2572                         }
2573
2574                         if (key.offset == upper_limit.objectid)
2575                                 break;
2576                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2577                                 ret = -EACCES;
2578                                 goto out_put;
2579                         }
2580
2581                         btrfs_release_path(path);
2582                         key.objectid = key.offset;
2583                         key.offset = (u64)-1;
2584                         dirid = key.objectid;
2585                 }
2586
2587                 memmove(args->path, ptr, total_len);
2588                 args->path[total_len] = '\0';
2589                 btrfs_put_root(root);
2590                 root = NULL;
2591                 btrfs_release_path(path);
2592         }
2593
2594         /* Get the bottom subvolume's name from ROOT_REF */
2595         key.objectid = treeid;
2596         key.type = BTRFS_ROOT_REF_KEY;
2597         key.offset = args->treeid;
2598         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2599         if (ret < 0) {
2600                 goto out;
2601         } else if (ret > 0) {
2602                 ret = -ENOENT;
2603                 goto out;
2604         }
2605
2606         leaf = path->nodes[0];
2607         slot = path->slots[0];
2608         btrfs_item_key_to_cpu(leaf, &key, slot);
2609
2610         item_off = btrfs_item_ptr_offset(leaf, slot);
2611         item_len = btrfs_item_size_nr(leaf, slot);
2612         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2613         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2614         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2615                 ret = -EINVAL;
2616                 goto out;
2617         }
2618
2619         /* Copy subvolume's name */
2620         item_off += sizeof(struct btrfs_root_ref);
2621         item_len -= sizeof(struct btrfs_root_ref);
2622         read_extent_buffer(leaf, args->name, item_off, item_len);
2623         args->name[item_len] = 0;
2624
2625 out_put:
2626         btrfs_put_root(root);
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         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2750         if (IS_ERR(root)) {
2751                 ret = PTR_ERR(root);
2752                 goto out_free;
2753         }
2754         root_item = &root->root_item;
2755
2756         subvol_info->treeid = key.objectid;
2757
2758         subvol_info->generation = btrfs_root_generation(root_item);
2759         subvol_info->flags = btrfs_root_flags(root_item);
2760
2761         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2762         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2763                                                     BTRFS_UUID_SIZE);
2764         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2765                                                     BTRFS_UUID_SIZE);
2766
2767         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2768         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2769         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2770
2771         subvol_info->otransid = btrfs_root_otransid(root_item);
2772         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2773         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2774
2775         subvol_info->stransid = btrfs_root_stransid(root_item);
2776         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2777         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2778
2779         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2780         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2781         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2782
2783         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2784                 /* Search root tree for ROOT_BACKREF of this subvolume */
2785                 key.type = BTRFS_ROOT_BACKREF_KEY;
2786                 key.offset = 0;
2787                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2788                 if (ret < 0) {
2789                         goto out;
2790                 } else if (path->slots[0] >=
2791                            btrfs_header_nritems(path->nodes[0])) {
2792                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2793                         if (ret < 0) {
2794                                 goto out;
2795                         } else if (ret > 0) {
2796                                 ret = -EUCLEAN;
2797                                 goto out;
2798                         }
2799                 }
2800
2801                 leaf = path->nodes[0];
2802                 slot = path->slots[0];
2803                 btrfs_item_key_to_cpu(leaf, &key, slot);
2804                 if (key.objectid == subvol_info->treeid &&
2805                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2806                         subvol_info->parent_id = key.offset;
2807
2808                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2809                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2810
2811                         item_off = btrfs_item_ptr_offset(leaf, slot)
2812                                         + sizeof(struct btrfs_root_ref);
2813                         item_len = btrfs_item_size_nr(leaf, slot)
2814                                         - sizeof(struct btrfs_root_ref);
2815                         read_extent_buffer(leaf, subvol_info->name,
2816                                            item_off, item_len);
2817                 } else {
2818                         ret = -ENOENT;
2819                         goto out;
2820                 }
2821         }
2822
2823         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2824                 ret = -EFAULT;
2825
2826 out:
2827         btrfs_put_root(root);
2828 out_free:
2829         btrfs_free_path(path);
2830         kfree(subvol_info);
2831         return ret;
2832 }
2833
2834 /*
2835  * Return ROOT_REF information of the subvolume containing this inode
2836  * except the subvolume name.
2837  */
2838 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2839 {
2840         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2841         struct btrfs_root_ref *rref;
2842         struct btrfs_root *root;
2843         struct btrfs_path *path;
2844         struct btrfs_key key;
2845         struct extent_buffer *leaf;
2846         struct inode *inode;
2847         u64 objectid;
2848         int slot;
2849         int ret;
2850         u8 found;
2851
2852         path = btrfs_alloc_path();
2853         if (!path)
2854                 return -ENOMEM;
2855
2856         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2857         if (IS_ERR(rootrefs)) {
2858                 btrfs_free_path(path);
2859                 return PTR_ERR(rootrefs);
2860         }
2861
2862         inode = file_inode(file);
2863         root = BTRFS_I(inode)->root->fs_info->tree_root;
2864         objectid = BTRFS_I(inode)->root->root_key.objectid;
2865
2866         key.objectid = objectid;
2867         key.type = BTRFS_ROOT_REF_KEY;
2868         key.offset = rootrefs->min_treeid;
2869         found = 0;
2870
2871         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2872         if (ret < 0) {
2873                 goto out;
2874         } else if (path->slots[0] >=
2875                    btrfs_header_nritems(path->nodes[0])) {
2876                 ret = btrfs_next_leaf(root, path);
2877                 if (ret < 0) {
2878                         goto out;
2879                 } else if (ret > 0) {
2880                         ret = -EUCLEAN;
2881                         goto out;
2882                 }
2883         }
2884         while (1) {
2885                 leaf = path->nodes[0];
2886                 slot = path->slots[0];
2887
2888                 btrfs_item_key_to_cpu(leaf, &key, slot);
2889                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2890                         ret = 0;
2891                         goto out;
2892                 }
2893
2894                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2895                         ret = -EOVERFLOW;
2896                         goto out;
2897                 }
2898
2899                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2900                 rootrefs->rootref[found].treeid = key.offset;
2901                 rootrefs->rootref[found].dirid =
2902                                   btrfs_root_ref_dirid(leaf, rref);
2903                 found++;
2904
2905                 ret = btrfs_next_item(root, path);
2906                 if (ret < 0) {
2907                         goto out;
2908                 } else if (ret > 0) {
2909                         ret = -EUCLEAN;
2910                         goto out;
2911                 }
2912         }
2913
2914 out:
2915         if (!ret || ret == -EOVERFLOW) {
2916                 rootrefs->num_items = found;
2917                 /* update min_treeid for next search */
2918                 if (found)
2919                         rootrefs->min_treeid =
2920                                 rootrefs->rootref[found - 1].treeid + 1;
2921                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2922                         ret = -EFAULT;
2923         }
2924
2925         kfree(rootrefs);
2926         btrfs_free_path(path);
2927
2928         return ret;
2929 }
2930
2931 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2932                                              void __user *arg,
2933                                              bool destroy_v2)
2934 {
2935         struct dentry *parent = file->f_path.dentry;
2936         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2937         struct dentry *dentry;
2938         struct inode *dir = d_inode(parent);
2939         struct inode *inode;
2940         struct btrfs_root *root = BTRFS_I(dir)->root;
2941         struct btrfs_root *dest = NULL;
2942         struct btrfs_ioctl_vol_args *vol_args = NULL;
2943         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2944         char *subvol_name, *subvol_name_ptr = NULL;
2945         int subvol_namelen;
2946         int err = 0;
2947         bool destroy_parent = false;
2948
2949         if (destroy_v2) {
2950                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2951                 if (IS_ERR(vol_args2))
2952                         return PTR_ERR(vol_args2);
2953
2954                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2955                         err = -EOPNOTSUPP;
2956                         goto out;
2957                 }
2958
2959                 /*
2960                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2961                  * name, same as v1 currently does.
2962                  */
2963                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2964                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2965                         subvol_name = vol_args2->name;
2966
2967                         err = mnt_want_write_file(file);
2968                         if (err)
2969                                 goto out;
2970                 } else {
2971                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2972                                 err = -EINVAL;
2973                                 goto out;
2974                         }
2975
2976                         err = mnt_want_write_file(file);
2977                         if (err)
2978                                 goto out;
2979
2980                         dentry = btrfs_get_dentry(fs_info->sb,
2981                                         BTRFS_FIRST_FREE_OBJECTID,
2982                                         vol_args2->subvolid, 0, 0);
2983                         if (IS_ERR(dentry)) {
2984                                 err = PTR_ERR(dentry);
2985                                 goto out_drop_write;
2986                         }
2987
2988                         /*
2989                          * Change the default parent since the subvolume being
2990                          * deleted can be outside of the current mount point.
2991                          */
2992                         parent = btrfs_get_parent(dentry);
2993
2994                         /*
2995                          * At this point dentry->d_name can point to '/' if the
2996                          * subvolume we want to destroy is outsite of the
2997                          * current mount point, so we need to release the
2998                          * current dentry and execute the lookup to return a new
2999                          * one with ->d_name pointing to the
3000                          * <mount point>/subvol_name.
3001                          */
3002                         dput(dentry);
3003                         if (IS_ERR(parent)) {
3004                                 err = PTR_ERR(parent);
3005                                 goto out_drop_write;
3006                         }
3007                         dir = d_inode(parent);
3008
3009                         /*
3010                          * If v2 was used with SPEC_BY_ID, a new parent was
3011                          * allocated since the subvolume can be outside of the
3012                          * current mount point. Later on we need to release this
3013                          * new parent dentry.
3014                          */
3015                         destroy_parent = true;
3016
3017                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3018                                                 fs_info, vol_args2->subvolid);
3019                         if (IS_ERR(subvol_name_ptr)) {
3020                                 err = PTR_ERR(subvol_name_ptr);
3021                                 goto free_parent;
3022                         }
3023                         /* subvol_name_ptr is already NULL termined */
3024                         subvol_name = (char *)kbasename(subvol_name_ptr);
3025                 }
3026         } else {
3027                 vol_args = memdup_user(arg, sizeof(*vol_args));
3028                 if (IS_ERR(vol_args))
3029                         return PTR_ERR(vol_args);
3030
3031                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3032                 subvol_name = vol_args->name;
3033
3034                 err = mnt_want_write_file(file);
3035                 if (err)
3036                         goto out;
3037         }
3038
3039         subvol_namelen = strlen(subvol_name);
3040
3041         if (strchr(subvol_name, '/') ||
3042             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3043                 err = -EINVAL;
3044                 goto free_subvol_name;
3045         }
3046
3047         if (!S_ISDIR(dir->i_mode)) {
3048                 err = -ENOTDIR;
3049                 goto free_subvol_name;
3050         }
3051
3052         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3053         if (err == -EINTR)
3054                 goto free_subvol_name;
3055         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
3056         if (IS_ERR(dentry)) {
3057                 err = PTR_ERR(dentry);
3058                 goto out_unlock_dir;
3059         }
3060
3061         if (d_really_is_negative(dentry)) {
3062                 err = -ENOENT;
3063                 goto out_dput;
3064         }
3065
3066         inode = d_inode(dentry);
3067         dest = BTRFS_I(inode)->root;
3068         if (!capable(CAP_SYS_ADMIN)) {
3069                 /*
3070                  * Regular user.  Only allow this with a special mount
3071                  * option, when the user has write+exec access to the
3072                  * subvol root, and when rmdir(2) would have been
3073                  * allowed.
3074                  *
3075                  * Note that this is _not_ check that the subvol is
3076                  * empty or doesn't contain data that we wouldn't
3077                  * otherwise be able to delete.
3078                  *
3079                  * Users who want to delete empty subvols should try
3080                  * rmdir(2).
3081                  */
3082                 err = -EPERM;
3083                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3084                         goto out_dput;
3085
3086                 /*
3087                  * Do not allow deletion if the parent dir is the same
3088                  * as the dir to be deleted.  That means the ioctl
3089                  * must be called on the dentry referencing the root
3090                  * of the subvol, not a random directory contained
3091                  * within it.
3092                  */
3093                 err = -EINVAL;
3094                 if (root == dest)
3095                         goto out_dput;
3096
3097                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
3098                 if (err)
3099                         goto out_dput;
3100         }
3101
3102         /* check if subvolume may be deleted by a user */
3103         err = btrfs_may_delete(dir, dentry, 1);
3104         if (err)
3105                 goto out_dput;
3106
3107         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3108                 err = -EINVAL;
3109                 goto out_dput;
3110         }
3111
3112         inode_lock(inode);
3113         err = btrfs_delete_subvolume(dir, dentry);
3114         inode_unlock(inode);
3115         if (!err) {
3116                 fsnotify_rmdir(dir, dentry);
3117                 d_delete(dentry);
3118         }
3119
3120 out_dput:
3121         dput(dentry);
3122 out_unlock_dir:
3123         inode_unlock(dir);
3124 free_subvol_name:
3125         kfree(subvol_name_ptr);
3126 free_parent:
3127         if (destroy_parent)
3128                 dput(parent);
3129 out_drop_write:
3130         mnt_drop_write_file(file);
3131 out:
3132         kfree(vol_args2);
3133         kfree(vol_args);
3134         return err;
3135 }
3136
3137 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3138 {
3139         struct inode *inode = file_inode(file);
3140         struct btrfs_root *root = BTRFS_I(inode)->root;
3141         struct btrfs_ioctl_defrag_range_args *range;
3142         int ret;
3143
3144         ret = mnt_want_write_file(file);
3145         if (ret)
3146                 return ret;
3147
3148         if (btrfs_root_readonly(root)) {
3149                 ret = -EROFS;
3150                 goto out;
3151         }
3152
3153         switch (inode->i_mode & S_IFMT) {
3154         case S_IFDIR:
3155                 if (!capable(CAP_SYS_ADMIN)) {
3156                         ret = -EPERM;
3157                         goto out;
3158                 }
3159                 ret = btrfs_defrag_root(root);
3160                 break;
3161         case S_IFREG:
3162                 /*
3163                  * Note that this does not check the file descriptor for write
3164                  * access. This prevents defragmenting executables that are
3165                  * running and allows defrag on files open in read-only mode.
3166                  */
3167                 if (!capable(CAP_SYS_ADMIN) &&
3168                     inode_permission(inode, MAY_WRITE)) {
3169                         ret = -EPERM;
3170                         goto out;
3171                 }
3172
3173                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3174                 if (!range) {
3175                         ret = -ENOMEM;
3176                         goto out;
3177                 }
3178
3179                 if (argp) {
3180                         if (copy_from_user(range, argp,
3181                                            sizeof(*range))) {
3182                                 ret = -EFAULT;
3183                                 kfree(range);
3184                                 goto out;
3185                         }
3186                         /* compression requires us to start the IO */
3187                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3188                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3189                                 range->extent_thresh = (u32)-1;
3190                         }
3191                 } else {
3192                         /* the rest are all set to zero by kzalloc */
3193                         range->len = (u64)-1;
3194                 }
3195                 ret = btrfs_defrag_file(file_inode(file), file,
3196                                         range, BTRFS_OLDEST_GENERATION, 0);
3197                 if (ret > 0)
3198                         ret = 0;
3199                 kfree(range);
3200                 break;
3201         default:
3202                 ret = -EINVAL;
3203         }
3204 out:
3205         mnt_drop_write_file(file);
3206         return ret;
3207 }
3208
3209 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3210 {
3211         struct btrfs_ioctl_vol_args *vol_args;
3212         int ret;
3213
3214         if (!capable(CAP_SYS_ADMIN))
3215                 return -EPERM;
3216
3217         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3218                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3219
3220         vol_args = memdup_user(arg, sizeof(*vol_args));
3221         if (IS_ERR(vol_args)) {
3222                 ret = PTR_ERR(vol_args);
3223                 goto out;
3224         }
3225
3226         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3227         ret = btrfs_init_new_device(fs_info, vol_args->name);
3228
3229         if (!ret)
3230                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3231
3232         kfree(vol_args);
3233 out:
3234         btrfs_exclop_finish(fs_info);
3235         return ret;
3236 }
3237
3238 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3239 {
3240         struct inode *inode = file_inode(file);
3241         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3242         struct btrfs_ioctl_vol_args_v2 *vol_args;
3243         int ret;
3244
3245         if (!capable(CAP_SYS_ADMIN))
3246                 return -EPERM;
3247
3248         ret = mnt_want_write_file(file);
3249         if (ret)
3250                 return ret;
3251
3252         vol_args = memdup_user(arg, sizeof(*vol_args));
3253         if (IS_ERR(vol_args)) {
3254                 ret = PTR_ERR(vol_args);
3255                 goto err_drop;
3256         }
3257
3258         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3259                 ret = -EOPNOTSUPP;
3260                 goto out;
3261         }
3262
3263         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3264                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3265                 goto out;
3266         }
3267
3268         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3269                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3270         } else {
3271                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3272                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3273         }
3274         btrfs_exclop_finish(fs_info);
3275
3276         if (!ret) {
3277                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3278                         btrfs_info(fs_info, "device deleted: id %llu",
3279                                         vol_args->devid);
3280                 else
3281                         btrfs_info(fs_info, "device deleted: %s",
3282                                         vol_args->name);
3283         }
3284 out:
3285         kfree(vol_args);
3286 err_drop:
3287         mnt_drop_write_file(file);
3288         return ret;
3289 }
3290
3291 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3292 {
3293         struct inode *inode = file_inode(file);
3294         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3295         struct btrfs_ioctl_vol_args *vol_args;
3296         int ret;
3297
3298         if (!capable(CAP_SYS_ADMIN))
3299                 return -EPERM;
3300
3301         ret = mnt_want_write_file(file);
3302         if (ret)
3303                 return ret;
3304
3305         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3306                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3307                 goto out_drop_write;
3308         }
3309
3310         vol_args = memdup_user(arg, sizeof(*vol_args));
3311         if (IS_ERR(vol_args)) {
3312                 ret = PTR_ERR(vol_args);
3313                 goto out;
3314         }
3315
3316         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3317         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3318
3319         if (!ret)
3320                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3321         kfree(vol_args);
3322 out:
3323         btrfs_exclop_finish(fs_info);
3324 out_drop_write:
3325         mnt_drop_write_file(file);
3326
3327         return ret;
3328 }
3329
3330 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3331                                 void __user *arg)
3332 {
3333         struct btrfs_ioctl_fs_info_args *fi_args;
3334         struct btrfs_device *device;
3335         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3336         u64 flags_in;
3337         int ret = 0;
3338
3339         fi_args = memdup_user(arg, sizeof(*fi_args));
3340         if (IS_ERR(fi_args))
3341                 return PTR_ERR(fi_args);
3342
3343         flags_in = fi_args->flags;
3344         memset(fi_args, 0, sizeof(*fi_args));
3345
3346         rcu_read_lock();
3347         fi_args->num_devices = fs_devices->num_devices;
3348
3349         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3350                 if (device->devid > fi_args->max_id)
3351                         fi_args->max_id = device->devid;
3352         }
3353         rcu_read_unlock();
3354
3355         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3356         fi_args->nodesize = fs_info->nodesize;
3357         fi_args->sectorsize = fs_info->sectorsize;
3358         fi_args->clone_alignment = fs_info->sectorsize;
3359
3360         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3361                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3362                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3363                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3364         }
3365
3366         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3367                 fi_args->generation = fs_info->generation;
3368                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3369         }
3370
3371         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3372                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3373                        sizeof(fi_args->metadata_uuid));
3374                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3375         }
3376
3377         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3378                 ret = -EFAULT;
3379
3380         kfree(fi_args);
3381         return ret;
3382 }
3383
3384 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3385                                  void __user *arg)
3386 {
3387         struct btrfs_ioctl_dev_info_args *di_args;
3388         struct btrfs_device *dev;
3389         int ret = 0;
3390         char *s_uuid = NULL;
3391
3392         di_args = memdup_user(arg, sizeof(*di_args));
3393         if (IS_ERR(di_args))
3394                 return PTR_ERR(di_args);
3395
3396         if (!btrfs_is_empty_uuid(di_args->uuid))
3397                 s_uuid = di_args->uuid;
3398
3399         rcu_read_lock();
3400         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3401                                 NULL);
3402
3403         if (!dev) {
3404                 ret = -ENODEV;
3405                 goto out;
3406         }
3407
3408         di_args->devid = dev->devid;
3409         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3410         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3411         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3412         if (dev->name) {
3413                 strncpy(di_args->path, rcu_str_deref(dev->name),
3414                                 sizeof(di_args->path) - 1);
3415                 di_args->path[sizeof(di_args->path) - 1] = 0;
3416         } else {
3417                 di_args->path[0] = '\0';
3418         }
3419
3420 out:
3421         rcu_read_unlock();
3422         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3423                 ret = -EFAULT;
3424
3425         kfree(di_args);
3426         return ret;
3427 }
3428
3429 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3430 {
3431         struct inode *inode = file_inode(file);
3432         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3433         struct btrfs_root *root = BTRFS_I(inode)->root;
3434         struct btrfs_root *new_root;
3435         struct btrfs_dir_item *di;
3436         struct btrfs_trans_handle *trans;
3437         struct btrfs_path *path = NULL;
3438         struct btrfs_disk_key disk_key;
3439         u64 objectid = 0;
3440         u64 dir_id;
3441         int ret;
3442
3443         if (!capable(CAP_SYS_ADMIN))
3444                 return -EPERM;
3445
3446         ret = mnt_want_write_file(file);
3447         if (ret)
3448                 return ret;
3449
3450         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3451                 ret = -EFAULT;
3452                 goto out;
3453         }
3454
3455         if (!objectid)
3456                 objectid = BTRFS_FS_TREE_OBJECTID;
3457
3458         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3459         if (IS_ERR(new_root)) {
3460                 ret = PTR_ERR(new_root);
3461                 goto out;
3462         }
3463         if (!is_fstree(new_root->root_key.objectid)) {
3464                 ret = -ENOENT;
3465                 goto out_free;
3466         }
3467
3468         path = btrfs_alloc_path();
3469         if (!path) {
3470                 ret = -ENOMEM;
3471                 goto out_free;
3472         }
3473
3474         trans = btrfs_start_transaction(root, 1);
3475         if (IS_ERR(trans)) {
3476                 ret = PTR_ERR(trans);
3477                 goto out_free;
3478         }
3479
3480         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3481         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3482                                    dir_id, "default", 7, 1);
3483         if (IS_ERR_OR_NULL(di)) {
3484                 btrfs_release_path(path);
3485                 btrfs_end_transaction(trans);
3486                 btrfs_err(fs_info,
3487                           "Umm, you don't have the default diritem, this isn't going to work");
3488                 ret = -ENOENT;
3489                 goto out_free;
3490         }
3491
3492         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3493         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3494         btrfs_mark_buffer_dirty(path->nodes[0]);
3495         btrfs_release_path(path);
3496
3497         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3498         btrfs_end_transaction(trans);
3499 out_free:
3500         btrfs_put_root(new_root);
3501         btrfs_free_path(path);
3502 out:
3503         mnt_drop_write_file(file);
3504         return ret;
3505 }
3506
3507 static void get_block_group_info(struct list_head *groups_list,
3508                                  struct btrfs_ioctl_space_info *space)
3509 {
3510         struct btrfs_block_group *block_group;
3511
3512         space->total_bytes = 0;
3513         space->used_bytes = 0;
3514         space->flags = 0;
3515         list_for_each_entry(block_group, groups_list, list) {
3516                 space->flags = block_group->flags;
3517                 space->total_bytes += block_group->length;
3518                 space->used_bytes += block_group->used;
3519         }
3520 }
3521
3522 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3523                                    void __user *arg)
3524 {
3525         struct btrfs_ioctl_space_args space_args;
3526         struct btrfs_ioctl_space_info space;
3527         struct btrfs_ioctl_space_info *dest;
3528         struct btrfs_ioctl_space_info *dest_orig;
3529         struct btrfs_ioctl_space_info __user *user_dest;
3530         struct btrfs_space_info *info;
3531         static const u64 types[] = {
3532                 BTRFS_BLOCK_GROUP_DATA,
3533                 BTRFS_BLOCK_GROUP_SYSTEM,
3534                 BTRFS_BLOCK_GROUP_METADATA,
3535                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3536         };
3537         int num_types = 4;
3538         int alloc_size;
3539         int ret = 0;
3540         u64 slot_count = 0;
3541         int i, c;
3542
3543         if (copy_from_user(&space_args,
3544                            (struct btrfs_ioctl_space_args __user *)arg,
3545                            sizeof(space_args)))
3546                 return -EFAULT;
3547
3548         for (i = 0; i < num_types; i++) {
3549                 struct btrfs_space_info *tmp;
3550
3551                 info = NULL;
3552                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3553                         if (tmp->flags == types[i]) {
3554                                 info = tmp;
3555                                 break;
3556                         }
3557                 }
3558
3559                 if (!info)
3560                         continue;
3561
3562                 down_read(&info->groups_sem);
3563                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3564                         if (!list_empty(&info->block_groups[c]))
3565                                 slot_count++;
3566                 }
3567                 up_read(&info->groups_sem);
3568         }
3569
3570         /*
3571          * Global block reserve, exported as a space_info
3572          */
3573         slot_count++;
3574
3575         /* space_slots == 0 means they are asking for a count */
3576         if (space_args.space_slots == 0) {
3577                 space_args.total_spaces = slot_count;
3578                 goto out;
3579         }
3580
3581         slot_count = min_t(u64, space_args.space_slots, slot_count);
3582
3583         alloc_size = sizeof(*dest) * slot_count;
3584
3585         /* we generally have at most 6 or so space infos, one for each raid
3586          * level.  So, a whole page should be more than enough for everyone
3587          */
3588         if (alloc_size > PAGE_SIZE)
3589                 return -ENOMEM;
3590
3591         space_args.total_spaces = 0;
3592         dest = kmalloc(alloc_size, GFP_KERNEL);
3593         if (!dest)
3594                 return -ENOMEM;
3595         dest_orig = dest;
3596
3597         /* now we have a buffer to copy into */
3598         for (i = 0; i < num_types; i++) {
3599                 struct btrfs_space_info *tmp;
3600
3601                 if (!slot_count)
3602                         break;
3603
3604                 info = NULL;
3605                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3606                         if (tmp->flags == types[i]) {
3607                                 info = tmp;
3608                                 break;
3609                         }
3610                 }
3611
3612                 if (!info)
3613                         continue;
3614                 down_read(&info->groups_sem);
3615                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3616                         if (!list_empty(&info->block_groups[c])) {
3617                                 get_block_group_info(&info->block_groups[c],
3618                                                      &space);
3619                                 memcpy(dest, &space, sizeof(space));
3620                                 dest++;
3621                                 space_args.total_spaces++;
3622                                 slot_count--;
3623                         }
3624                         if (!slot_count)
3625                                 break;
3626                 }
3627                 up_read(&info->groups_sem);
3628         }
3629
3630         /*
3631          * Add global block reserve
3632          */
3633         if (slot_count) {
3634                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3635
3636                 spin_lock(&block_rsv->lock);
3637                 space.total_bytes = block_rsv->size;
3638                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3639                 spin_unlock(&block_rsv->lock);
3640                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3641                 memcpy(dest, &space, sizeof(space));
3642                 space_args.total_spaces++;
3643         }
3644
3645         user_dest = (struct btrfs_ioctl_space_info __user *)
3646                 (arg + sizeof(struct btrfs_ioctl_space_args));
3647
3648         if (copy_to_user(user_dest, dest_orig, alloc_size))
3649                 ret = -EFAULT;
3650
3651         kfree(dest_orig);
3652 out:
3653         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3654                 ret = -EFAULT;
3655
3656         return ret;
3657 }
3658
3659 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3660                                             void __user *argp)
3661 {
3662         struct btrfs_trans_handle *trans;
3663         u64 transid;
3664         int ret;
3665
3666         trans = btrfs_attach_transaction_barrier(root);
3667         if (IS_ERR(trans)) {
3668                 if (PTR_ERR(trans) != -ENOENT)
3669                         return PTR_ERR(trans);
3670
3671                 /* No running transaction, don't bother */
3672                 transid = root->fs_info->last_trans_committed;
3673                 goto out;
3674         }
3675         transid = trans->transid;
3676         ret = btrfs_commit_transaction_async(trans, 0);
3677         if (ret) {
3678                 btrfs_end_transaction(trans);
3679                 return ret;
3680         }
3681 out:
3682         if (argp)
3683                 if (copy_to_user(argp, &transid, sizeof(transid)))
3684                         return -EFAULT;
3685         return 0;
3686 }
3687
3688 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3689                                            void __user *argp)
3690 {
3691         u64 transid;
3692
3693         if (argp) {
3694                 if (copy_from_user(&transid, argp, sizeof(transid)))
3695                         return -EFAULT;
3696         } else {
3697                 transid = 0;  /* current trans */
3698         }
3699         return btrfs_wait_for_commit(fs_info, transid);
3700 }
3701
3702 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3703 {
3704         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3705         struct btrfs_ioctl_scrub_args *sa;
3706         int ret;
3707
3708         if (!capable(CAP_SYS_ADMIN))
3709                 return -EPERM;
3710
3711         sa = memdup_user(arg, sizeof(*sa));
3712         if (IS_ERR(sa))
3713                 return PTR_ERR(sa);
3714
3715         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3716                 ret = mnt_want_write_file(file);
3717                 if (ret)
3718                         goto out;
3719         }
3720
3721         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3722                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3723                               0);
3724
3725         /*
3726          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3727          * error. This is important as it allows user space to know how much
3728          * progress scrub has done. For example, if scrub is canceled we get
3729          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3730          * space. Later user space can inspect the progress from the structure
3731          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3732          * previously (btrfs-progs does this).
3733          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3734          * then return -EFAULT to signal the structure was not copied or it may
3735          * be corrupt and unreliable due to a partial copy.
3736          */
3737         if (copy_to_user(arg, sa, sizeof(*sa)))
3738                 ret = -EFAULT;
3739
3740         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3741                 mnt_drop_write_file(file);
3742 out:
3743         kfree(sa);
3744         return ret;
3745 }
3746
3747 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3748 {
3749         if (!capable(CAP_SYS_ADMIN))
3750                 return -EPERM;
3751
3752         return btrfs_scrub_cancel(fs_info);
3753 }
3754
3755 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3756                                        void __user *arg)
3757 {
3758         struct btrfs_ioctl_scrub_args *sa;
3759         int ret;
3760
3761         if (!capable(CAP_SYS_ADMIN))
3762                 return -EPERM;
3763
3764         sa = memdup_user(arg, sizeof(*sa));
3765         if (IS_ERR(sa))
3766                 return PTR_ERR(sa);
3767
3768         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3769
3770         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3771                 ret = -EFAULT;
3772
3773         kfree(sa);
3774         return ret;
3775 }
3776
3777 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3778                                       void __user *arg)
3779 {
3780         struct btrfs_ioctl_get_dev_stats *sa;
3781         int ret;
3782
3783         sa = memdup_user(arg, sizeof(*sa));
3784         if (IS_ERR(sa))
3785                 return PTR_ERR(sa);
3786
3787         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3788                 kfree(sa);
3789                 return -EPERM;
3790         }
3791
3792         ret = btrfs_get_dev_stats(fs_info, sa);
3793
3794         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3795                 ret = -EFAULT;
3796
3797         kfree(sa);
3798         return ret;
3799 }
3800
3801 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3802                                     void __user *arg)
3803 {
3804         struct btrfs_ioctl_dev_replace_args *p;
3805         int ret;
3806
3807         if (!capable(CAP_SYS_ADMIN))
3808                 return -EPERM;
3809
3810         p = memdup_user(arg, sizeof(*p));
3811         if (IS_ERR(p))
3812                 return PTR_ERR(p);
3813
3814         switch (p->cmd) {
3815         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3816                 if (sb_rdonly(fs_info->sb)) {
3817                         ret = -EROFS;
3818                         goto out;
3819                 }
3820                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3821                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3822                 } else {
3823                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3824                         btrfs_exclop_finish(fs_info);
3825                 }
3826                 break;
3827         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3828                 btrfs_dev_replace_status(fs_info, p);
3829                 ret = 0;
3830                 break;
3831         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3832                 p->result = btrfs_dev_replace_cancel(fs_info);
3833                 ret = 0;
3834                 break;
3835         default:
3836                 ret = -EINVAL;
3837                 break;
3838         }
3839
3840         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3841                 ret = -EFAULT;
3842 out:
3843         kfree(p);
3844         return ret;
3845 }
3846
3847 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3848 {
3849         int ret = 0;
3850         int i;
3851         u64 rel_ptr;
3852         int size;
3853         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3854         struct inode_fs_paths *ipath = NULL;
3855         struct btrfs_path *path;
3856
3857         if (!capable(CAP_DAC_READ_SEARCH))
3858                 return -EPERM;
3859
3860         path = btrfs_alloc_path();
3861         if (!path) {
3862                 ret = -ENOMEM;
3863                 goto out;
3864         }
3865
3866         ipa = memdup_user(arg, sizeof(*ipa));
3867         if (IS_ERR(ipa)) {
3868                 ret = PTR_ERR(ipa);
3869                 ipa = NULL;
3870                 goto out;
3871         }
3872
3873         size = min_t(u32, ipa->size, 4096);
3874         ipath = init_ipath(size, root, path);
3875         if (IS_ERR(ipath)) {
3876                 ret = PTR_ERR(ipath);
3877                 ipath = NULL;
3878                 goto out;
3879         }
3880
3881         ret = paths_from_inode(ipa->inum, ipath);
3882         if (ret < 0)
3883                 goto out;
3884
3885         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3886                 rel_ptr = ipath->fspath->val[i] -
3887                           (u64)(unsigned long)ipath->fspath->val;
3888                 ipath->fspath->val[i] = rel_ptr;
3889         }
3890
3891         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3892                            ipath->fspath, size);
3893         if (ret) {
3894                 ret = -EFAULT;
3895                 goto out;
3896         }
3897
3898 out:
3899         btrfs_free_path(path);
3900         free_ipath(ipath);
3901         kfree(ipa);
3902
3903         return ret;
3904 }
3905
3906 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3907 {
3908         struct btrfs_data_container *inodes = ctx;
3909         const size_t c = 3 * sizeof(u64);
3910
3911         if (inodes->bytes_left >= c) {
3912                 inodes->bytes_left -= c;
3913                 inodes->val[inodes->elem_cnt] = inum;
3914                 inodes->val[inodes->elem_cnt + 1] = offset;
3915                 inodes->val[inodes->elem_cnt + 2] = root;
3916                 inodes->elem_cnt += 3;
3917         } else {
3918                 inodes->bytes_missing += c - inodes->bytes_left;
3919                 inodes->bytes_left = 0;
3920                 inodes->elem_missed += 3;
3921         }
3922
3923         return 0;
3924 }
3925
3926 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3927                                         void __user *arg, int version)
3928 {
3929         int ret = 0;
3930         int size;
3931         struct btrfs_ioctl_logical_ino_args *loi;
3932         struct btrfs_data_container *inodes = NULL;
3933         struct btrfs_path *path = NULL;
3934         bool ignore_offset;
3935
3936         if (!capable(CAP_SYS_ADMIN))
3937                 return -EPERM;
3938
3939         loi = memdup_user(arg, sizeof(*loi));
3940         if (IS_ERR(loi))
3941                 return PTR_ERR(loi);
3942
3943         if (version == 1) {
3944                 ignore_offset = false;
3945                 size = min_t(u32, loi->size, SZ_64K);
3946         } else {
3947                 /* All reserved bits must be 0 for now */
3948                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3949                         ret = -EINVAL;
3950                         goto out_loi;
3951                 }
3952                 /* Only accept flags we have defined so far */
3953                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3954                         ret = -EINVAL;
3955                         goto out_loi;
3956                 }
3957                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3958                 size = min_t(u32, loi->size, SZ_16M);
3959         }
3960
3961         path = btrfs_alloc_path();
3962         if (!path) {
3963                 ret = -ENOMEM;
3964                 goto out;
3965         }
3966
3967         inodes = init_data_container(size);
3968         if (IS_ERR(inodes)) {
3969                 ret = PTR_ERR(inodes);
3970                 inodes = NULL;
3971                 goto out;
3972         }
3973
3974         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3975                                           build_ino_list, inodes, ignore_offset);
3976         if (ret == -EINVAL)
3977                 ret = -ENOENT;
3978         if (ret < 0)
3979                 goto out;
3980
3981         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3982                            size);
3983         if (ret)
3984                 ret = -EFAULT;
3985
3986 out:
3987         btrfs_free_path(path);
3988         kvfree(inodes);
3989 out_loi:
3990         kfree(loi);
3991
3992         return ret;
3993 }
3994
3995 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3996                                struct btrfs_ioctl_balance_args *bargs)
3997 {
3998         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3999
4000         bargs->flags = bctl->flags;
4001
4002         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4003                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4004         if (atomic_read(&fs_info->balance_pause_req))
4005                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4006         if (atomic_read(&fs_info->balance_cancel_req))
4007                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4008
4009         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4010         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4011         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4012
4013         spin_lock(&fs_info->balance_lock);
4014         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4015         spin_unlock(&fs_info->balance_lock);
4016 }
4017
4018 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4019 {
4020         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4021         struct btrfs_fs_info *fs_info = root->fs_info;
4022         struct btrfs_ioctl_balance_args *bargs;
4023         struct btrfs_balance_control *bctl;
4024         bool need_unlock; /* for mut. excl. ops lock */
4025         int ret;
4026
4027         if (!capable(CAP_SYS_ADMIN))
4028                 return -EPERM;
4029
4030         ret = mnt_want_write_file(file);
4031         if (ret)
4032                 return ret;
4033
4034 again:
4035         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4036                 mutex_lock(&fs_info->balance_mutex);
4037                 need_unlock = true;
4038                 goto locked;
4039         }
4040
4041         /*
4042          * mut. excl. ops lock is locked.  Three possibilities:
4043          *   (1) some other op is running
4044          *   (2) balance is running
4045          *   (3) balance is paused -- special case (think resume)
4046          */
4047         mutex_lock(&fs_info->balance_mutex);
4048         if (fs_info->balance_ctl) {
4049                 /* this is either (2) or (3) */
4050                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4051                         mutex_unlock(&fs_info->balance_mutex);
4052                         /*
4053                          * Lock released to allow other waiters to continue,
4054                          * we'll reexamine the status again.
4055                          */
4056                         mutex_lock(&fs_info->balance_mutex);
4057
4058                         if (fs_info->balance_ctl &&
4059                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4060                                 /* this is (3) */
4061                                 need_unlock = false;
4062                                 goto locked;
4063                         }
4064
4065                         mutex_unlock(&fs_info->balance_mutex);
4066                         goto again;
4067                 } else {
4068                         /* this is (2) */
4069                         mutex_unlock(&fs_info->balance_mutex);
4070                         ret = -EINPROGRESS;
4071                         goto out;
4072                 }
4073         } else {
4074                 /* this is (1) */
4075                 mutex_unlock(&fs_info->balance_mutex);
4076                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4077                 goto out;
4078         }
4079
4080 locked:
4081
4082         if (arg) {
4083                 bargs = memdup_user(arg, sizeof(*bargs));
4084                 if (IS_ERR(bargs)) {
4085                         ret = PTR_ERR(bargs);
4086                         goto out_unlock;
4087                 }
4088
4089                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4090                         if (!fs_info->balance_ctl) {
4091                                 ret = -ENOTCONN;
4092                                 goto out_bargs;
4093                         }
4094
4095                         bctl = fs_info->balance_ctl;
4096                         spin_lock(&fs_info->balance_lock);
4097                         bctl->flags |= BTRFS_BALANCE_RESUME;
4098                         spin_unlock(&fs_info->balance_lock);
4099
4100                         goto do_balance;
4101                 }
4102         } else {
4103                 bargs = NULL;
4104         }
4105
4106         if (fs_info->balance_ctl) {
4107                 ret = -EINPROGRESS;
4108                 goto out_bargs;
4109         }
4110
4111         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4112         if (!bctl) {
4113                 ret = -ENOMEM;
4114                 goto out_bargs;
4115         }
4116
4117         if (arg) {
4118                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4119                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4120                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4121
4122                 bctl->flags = bargs->flags;
4123         } else {
4124                 /* balance everything - no filters */
4125                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4126         }
4127
4128         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4129                 ret = -EINVAL;
4130                 goto out_bctl;
4131         }
4132
4133 do_balance:
4134         /*
4135          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4136          * bctl is freed in reset_balance_state, or, if restriper was paused
4137          * all the way until unmount, in free_fs_info.  The flag should be
4138          * cleared after reset_balance_state.
4139          */
4140         need_unlock = false;
4141
4142         ret = btrfs_balance(fs_info, bctl, bargs);
4143         bctl = NULL;
4144
4145         if ((ret == 0 || ret == -ECANCELED) && arg) {
4146                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4147                         ret = -EFAULT;
4148         }
4149
4150 out_bctl:
4151         kfree(bctl);
4152 out_bargs:
4153         kfree(bargs);
4154 out_unlock:
4155         mutex_unlock(&fs_info->balance_mutex);
4156         if (need_unlock)
4157                 btrfs_exclop_finish(fs_info);
4158 out:
4159         mnt_drop_write_file(file);
4160         return ret;
4161 }
4162
4163 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4164 {
4165         if (!capable(CAP_SYS_ADMIN))
4166                 return -EPERM;
4167
4168         switch (cmd) {
4169         case BTRFS_BALANCE_CTL_PAUSE:
4170                 return btrfs_pause_balance(fs_info);
4171         case BTRFS_BALANCE_CTL_CANCEL:
4172                 return btrfs_cancel_balance(fs_info);
4173         }
4174
4175         return -EINVAL;
4176 }
4177
4178 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4179                                          void __user *arg)
4180 {
4181         struct btrfs_ioctl_balance_args *bargs;
4182         int ret = 0;
4183
4184         if (!capable(CAP_SYS_ADMIN))
4185                 return -EPERM;
4186
4187         mutex_lock(&fs_info->balance_mutex);
4188         if (!fs_info->balance_ctl) {
4189                 ret = -ENOTCONN;
4190                 goto out;
4191         }
4192
4193         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4194         if (!bargs) {
4195                 ret = -ENOMEM;
4196                 goto out;
4197         }
4198
4199         btrfs_update_ioctl_balance_args(fs_info, bargs);
4200
4201         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4202                 ret = -EFAULT;
4203
4204         kfree(bargs);
4205 out:
4206         mutex_unlock(&fs_info->balance_mutex);
4207         return ret;
4208 }
4209
4210 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4211 {
4212         struct inode *inode = file_inode(file);
4213         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4214         struct btrfs_ioctl_quota_ctl_args *sa;
4215         int ret;
4216
4217         if (!capable(CAP_SYS_ADMIN))
4218                 return -EPERM;
4219
4220         ret = mnt_want_write_file(file);
4221         if (ret)
4222                 return ret;
4223
4224         sa = memdup_user(arg, sizeof(*sa));
4225         if (IS_ERR(sa)) {
4226                 ret = PTR_ERR(sa);
4227                 goto drop_write;
4228         }
4229
4230         down_write(&fs_info->subvol_sem);
4231
4232         switch (sa->cmd) {
4233         case BTRFS_QUOTA_CTL_ENABLE:
4234                 ret = btrfs_quota_enable(fs_info);
4235                 break;
4236         case BTRFS_QUOTA_CTL_DISABLE:
4237                 ret = btrfs_quota_disable(fs_info);
4238                 break;
4239         default:
4240                 ret = -EINVAL;
4241                 break;
4242         }
4243
4244         kfree(sa);
4245         up_write(&fs_info->subvol_sem);
4246 drop_write:
4247         mnt_drop_write_file(file);
4248         return ret;
4249 }
4250
4251 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4252 {
4253         struct inode *inode = file_inode(file);
4254         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4255         struct btrfs_root *root = BTRFS_I(inode)->root;
4256         struct btrfs_ioctl_qgroup_assign_args *sa;
4257         struct btrfs_trans_handle *trans;
4258         int ret;
4259         int err;
4260
4261         if (!capable(CAP_SYS_ADMIN))
4262                 return -EPERM;
4263
4264         ret = mnt_want_write_file(file);
4265         if (ret)
4266                 return ret;
4267
4268         sa = memdup_user(arg, sizeof(*sa));
4269         if (IS_ERR(sa)) {
4270                 ret = PTR_ERR(sa);
4271                 goto drop_write;
4272         }
4273
4274         trans = btrfs_join_transaction(root);
4275         if (IS_ERR(trans)) {
4276                 ret = PTR_ERR(trans);
4277                 goto out;
4278         }
4279
4280         if (sa->assign) {
4281                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4282         } else {
4283                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4284         }
4285
4286         /* update qgroup status and info */
4287         err = btrfs_run_qgroups(trans);
4288         if (err < 0)
4289                 btrfs_handle_fs_error(fs_info, err,
4290                                       "failed to update qgroup status and info");
4291         err = btrfs_end_transaction(trans);
4292         if (err && !ret)
4293                 ret = err;
4294
4295 out:
4296         kfree(sa);
4297 drop_write:
4298         mnt_drop_write_file(file);
4299         return ret;
4300 }
4301
4302 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4303 {
4304         struct inode *inode = file_inode(file);
4305         struct btrfs_root *root = BTRFS_I(inode)->root;
4306         struct btrfs_ioctl_qgroup_create_args *sa;
4307         struct btrfs_trans_handle *trans;
4308         int ret;
4309         int err;
4310
4311         if (!capable(CAP_SYS_ADMIN))
4312                 return -EPERM;
4313
4314         ret = mnt_want_write_file(file);
4315         if (ret)
4316                 return ret;
4317
4318         sa = memdup_user(arg, sizeof(*sa));
4319         if (IS_ERR(sa)) {
4320                 ret = PTR_ERR(sa);
4321                 goto drop_write;
4322         }
4323
4324         if (!sa->qgroupid) {
4325                 ret = -EINVAL;
4326                 goto out;
4327         }
4328
4329         trans = btrfs_join_transaction(root);
4330         if (IS_ERR(trans)) {
4331                 ret = PTR_ERR(trans);
4332                 goto out;
4333         }
4334
4335         if (sa->create) {
4336                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4337         } else {
4338                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4339         }
4340
4341         err = btrfs_end_transaction(trans);
4342         if (err && !ret)
4343                 ret = err;
4344
4345 out:
4346         kfree(sa);
4347 drop_write:
4348         mnt_drop_write_file(file);
4349         return ret;
4350 }
4351
4352 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4353 {
4354         struct inode *inode = file_inode(file);
4355         struct btrfs_root *root = BTRFS_I(inode)->root;
4356         struct btrfs_ioctl_qgroup_limit_args *sa;
4357         struct btrfs_trans_handle *trans;
4358         int ret;
4359         int err;
4360         u64 qgroupid;
4361
4362         if (!capable(CAP_SYS_ADMIN))
4363                 return -EPERM;
4364
4365         ret = mnt_want_write_file(file);
4366         if (ret)
4367                 return ret;
4368
4369         sa = memdup_user(arg, sizeof(*sa));
4370         if (IS_ERR(sa)) {
4371                 ret = PTR_ERR(sa);
4372                 goto drop_write;
4373         }
4374
4375         trans = btrfs_join_transaction(root);
4376         if (IS_ERR(trans)) {
4377                 ret = PTR_ERR(trans);
4378                 goto out;
4379         }
4380
4381         qgroupid = sa->qgroupid;
4382         if (!qgroupid) {
4383                 /* take the current subvol as qgroup */
4384                 qgroupid = root->root_key.objectid;
4385         }
4386
4387         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4388
4389         err = btrfs_end_transaction(trans);
4390         if (err && !ret)
4391                 ret = err;
4392
4393 out:
4394         kfree(sa);
4395 drop_write:
4396         mnt_drop_write_file(file);
4397         return ret;
4398 }
4399
4400 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4401 {
4402         struct inode *inode = file_inode(file);
4403         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4404         struct btrfs_ioctl_quota_rescan_args *qsa;
4405         int ret;
4406
4407         if (!capable(CAP_SYS_ADMIN))
4408                 return -EPERM;
4409
4410         ret = mnt_want_write_file(file);
4411         if (ret)
4412                 return ret;
4413
4414         qsa = memdup_user(arg, sizeof(*qsa));
4415         if (IS_ERR(qsa)) {
4416                 ret = PTR_ERR(qsa);
4417                 goto drop_write;
4418         }
4419
4420         if (qsa->flags) {
4421                 ret = -EINVAL;
4422                 goto out;
4423         }
4424
4425         ret = btrfs_qgroup_rescan(fs_info);
4426
4427 out:
4428         kfree(qsa);
4429 drop_write:
4430         mnt_drop_write_file(file);
4431         return ret;
4432 }
4433
4434 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4435                                                 void __user *arg)
4436 {
4437         struct btrfs_ioctl_quota_rescan_args *qsa;
4438         int ret = 0;
4439
4440         if (!capable(CAP_SYS_ADMIN))
4441                 return -EPERM;
4442
4443         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4444         if (!qsa)
4445                 return -ENOMEM;
4446
4447         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4448                 qsa->flags = 1;
4449                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4450         }
4451
4452         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4453                 ret = -EFAULT;
4454
4455         kfree(qsa);
4456         return ret;
4457 }
4458
4459 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4460                                                 void __user *arg)
4461 {
4462         if (!capable(CAP_SYS_ADMIN))
4463                 return -EPERM;
4464
4465         return btrfs_qgroup_wait_for_completion(fs_info, true);
4466 }
4467
4468 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4469                                             struct btrfs_ioctl_received_subvol_args *sa)
4470 {
4471         struct inode *inode = file_inode(file);
4472         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4473         struct btrfs_root *root = BTRFS_I(inode)->root;
4474         struct btrfs_root_item *root_item = &root->root_item;
4475         struct btrfs_trans_handle *trans;
4476         struct timespec64 ct = current_time(inode);
4477         int ret = 0;
4478         int received_uuid_changed;
4479
4480         if (!inode_owner_or_capable(inode))
4481                 return -EPERM;
4482
4483         ret = mnt_want_write_file(file);
4484         if (ret < 0)
4485                 return ret;
4486
4487         down_write(&fs_info->subvol_sem);
4488
4489         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4490                 ret = -EINVAL;
4491                 goto out;
4492         }
4493
4494         if (btrfs_root_readonly(root)) {
4495                 ret = -EROFS;
4496                 goto out;
4497         }
4498
4499         /*
4500          * 1 - root item
4501          * 2 - uuid items (received uuid + subvol uuid)
4502          */
4503         trans = btrfs_start_transaction(root, 3);
4504         if (IS_ERR(trans)) {
4505                 ret = PTR_ERR(trans);
4506                 trans = NULL;
4507                 goto out;
4508         }
4509
4510         sa->rtransid = trans->transid;
4511         sa->rtime.sec = ct.tv_sec;
4512         sa->rtime.nsec = ct.tv_nsec;
4513
4514         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4515                                        BTRFS_UUID_SIZE);
4516         if (received_uuid_changed &&
4517             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4518                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4519                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4520                                           root->root_key.objectid);
4521                 if (ret && ret != -ENOENT) {
4522                         btrfs_abort_transaction(trans, ret);
4523                         btrfs_end_transaction(trans);
4524                         goto out;
4525                 }
4526         }
4527         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4528         btrfs_set_root_stransid(root_item, sa->stransid);
4529         btrfs_set_root_rtransid(root_item, sa->rtransid);
4530         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4531         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4532         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4533         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4534
4535         ret = btrfs_update_root(trans, fs_info->tree_root,
4536                                 &root->root_key, &root->root_item);
4537         if (ret < 0) {
4538                 btrfs_end_transaction(trans);
4539                 goto out;
4540         }
4541         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4542                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4543                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4544                                           root->root_key.objectid);
4545                 if (ret < 0 && ret != -EEXIST) {
4546                         btrfs_abort_transaction(trans, ret);
4547                         btrfs_end_transaction(trans);
4548                         goto out;
4549                 }
4550         }
4551         ret = btrfs_commit_transaction(trans);
4552 out:
4553         up_write(&fs_info->subvol_sem);
4554         mnt_drop_write_file(file);
4555         return ret;
4556 }
4557
4558 #ifdef CONFIG_64BIT
4559 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4560                                                 void __user *arg)
4561 {
4562         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4563         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4564         int ret = 0;
4565
4566         args32 = memdup_user(arg, sizeof(*args32));
4567         if (IS_ERR(args32))
4568                 return PTR_ERR(args32);
4569
4570         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4571         if (!args64) {
4572                 ret = -ENOMEM;
4573                 goto out;
4574         }
4575
4576         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4577         args64->stransid = args32->stransid;
4578         args64->rtransid = args32->rtransid;
4579         args64->stime.sec = args32->stime.sec;
4580         args64->stime.nsec = args32->stime.nsec;
4581         args64->rtime.sec = args32->rtime.sec;
4582         args64->rtime.nsec = args32->rtime.nsec;
4583         args64->flags = args32->flags;
4584
4585         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4586         if (ret)
4587                 goto out;
4588
4589         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4590         args32->stransid = args64->stransid;
4591         args32->rtransid = args64->rtransid;
4592         args32->stime.sec = args64->stime.sec;
4593         args32->stime.nsec = args64->stime.nsec;
4594         args32->rtime.sec = args64->rtime.sec;
4595         args32->rtime.nsec = args64->rtime.nsec;
4596         args32->flags = args64->flags;
4597
4598         ret = copy_to_user(arg, args32, sizeof(*args32));
4599         if (ret)
4600                 ret = -EFAULT;
4601
4602 out:
4603         kfree(args32);
4604         kfree(args64);
4605         return ret;
4606 }
4607 #endif
4608
4609 static long btrfs_ioctl_set_received_subvol(struct file *file,
4610                                             void __user *arg)
4611 {
4612         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4613         int ret = 0;
4614
4615         sa = memdup_user(arg, sizeof(*sa));
4616         if (IS_ERR(sa))
4617                 return PTR_ERR(sa);
4618
4619         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4620
4621         if (ret)
4622                 goto out;
4623
4624         ret = copy_to_user(arg, sa, sizeof(*sa));
4625         if (ret)
4626                 ret = -EFAULT;
4627
4628 out:
4629         kfree(sa);
4630         return ret;
4631 }
4632
4633 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4634                                         void __user *arg)
4635 {
4636         size_t len;
4637         int ret;
4638         char label[BTRFS_LABEL_SIZE];
4639
4640         spin_lock(&fs_info->super_lock);
4641         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4642         spin_unlock(&fs_info->super_lock);
4643
4644         len = strnlen(label, BTRFS_LABEL_SIZE);
4645
4646         if (len == BTRFS_LABEL_SIZE) {
4647                 btrfs_warn(fs_info,
4648                            "label is too long, return the first %zu bytes",
4649                            --len);
4650         }
4651
4652         ret = copy_to_user(arg, label, len);
4653
4654         return ret ? -EFAULT : 0;
4655 }
4656
4657 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4658 {
4659         struct inode *inode = file_inode(file);
4660         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4661         struct btrfs_root *root = BTRFS_I(inode)->root;
4662         struct btrfs_super_block *super_block = fs_info->super_copy;
4663         struct btrfs_trans_handle *trans;
4664         char label[BTRFS_LABEL_SIZE];
4665         int ret;
4666
4667         if (!capable(CAP_SYS_ADMIN))
4668                 return -EPERM;
4669
4670         if (copy_from_user(label, arg, sizeof(label)))
4671                 return -EFAULT;
4672
4673         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4674                 btrfs_err(fs_info,
4675                           "unable to set label with more than %d bytes",
4676                           BTRFS_LABEL_SIZE - 1);
4677                 return -EINVAL;
4678         }
4679
4680         ret = mnt_want_write_file(file);
4681         if (ret)
4682                 return ret;
4683
4684         trans = btrfs_start_transaction(root, 0);
4685         if (IS_ERR(trans)) {
4686                 ret = PTR_ERR(trans);
4687                 goto out_unlock;
4688         }
4689
4690         spin_lock(&fs_info->super_lock);
4691         strcpy(super_block->label, label);
4692         spin_unlock(&fs_info->super_lock);
4693         ret = btrfs_commit_transaction(trans);
4694
4695 out_unlock:
4696         mnt_drop_write_file(file);
4697         return ret;
4698 }
4699
4700 #define INIT_FEATURE_FLAGS(suffix) \
4701         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4702           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4703           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4704
4705 int btrfs_ioctl_get_supported_features(void __user *arg)
4706 {
4707         static const struct btrfs_ioctl_feature_flags features[3] = {
4708                 INIT_FEATURE_FLAGS(SUPP),
4709                 INIT_FEATURE_FLAGS(SAFE_SET),
4710                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4711         };
4712
4713         if (copy_to_user(arg, &features, sizeof(features)))
4714                 return -EFAULT;
4715
4716         return 0;
4717 }
4718
4719 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4720                                         void __user *arg)
4721 {
4722         struct btrfs_super_block *super_block = fs_info->super_copy;
4723         struct btrfs_ioctl_feature_flags features;
4724
4725         features.compat_flags = btrfs_super_compat_flags(super_block);
4726         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4727         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4728
4729         if (copy_to_user(arg, &features, sizeof(features)))
4730                 return -EFAULT;
4731
4732         return 0;
4733 }
4734
4735 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4736                               enum btrfs_feature_set set,
4737                               u64 change_mask, u64 flags, u64 supported_flags,
4738                               u64 safe_set, u64 safe_clear)
4739 {
4740         const char *type = btrfs_feature_set_name(set);
4741         char *names;
4742         u64 disallowed, unsupported;
4743         u64 set_mask = flags & change_mask;
4744         u64 clear_mask = ~flags & change_mask;
4745
4746         unsupported = set_mask & ~supported_flags;
4747         if (unsupported) {
4748                 names = btrfs_printable_features(set, unsupported);
4749                 if (names) {
4750                         btrfs_warn(fs_info,
4751                                    "this kernel does not support the %s feature bit%s",
4752                                    names, strchr(names, ',') ? "s" : "");
4753                         kfree(names);
4754                 } else
4755                         btrfs_warn(fs_info,
4756                                    "this kernel does not support %s bits 0x%llx",
4757                                    type, unsupported);
4758                 return -EOPNOTSUPP;
4759         }
4760
4761         disallowed = set_mask & ~safe_set;
4762         if (disallowed) {
4763                 names = btrfs_printable_features(set, disallowed);
4764                 if (names) {
4765                         btrfs_warn(fs_info,
4766                                    "can't set the %s feature bit%s while mounted",
4767                                    names, strchr(names, ',') ? "s" : "");
4768                         kfree(names);
4769                 } else
4770                         btrfs_warn(fs_info,
4771                                    "can't set %s bits 0x%llx while mounted",
4772                                    type, disallowed);
4773                 return -EPERM;
4774         }
4775
4776         disallowed = clear_mask & ~safe_clear;
4777         if (disallowed) {
4778                 names = btrfs_printable_features(set, disallowed);
4779                 if (names) {
4780                         btrfs_warn(fs_info,
4781                                    "can't clear the %s feature bit%s while mounted",
4782                                    names, strchr(names, ',') ? "s" : "");
4783                         kfree(names);
4784                 } else
4785                         btrfs_warn(fs_info,
4786                                    "can't clear %s bits 0x%llx while mounted",
4787                                    type, disallowed);
4788                 return -EPERM;
4789         }
4790
4791         return 0;
4792 }
4793
4794 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4795 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4796                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4797                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4798                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4799
4800 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4801 {
4802         struct inode *inode = file_inode(file);
4803         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4804         struct btrfs_root *root = BTRFS_I(inode)->root;
4805         struct btrfs_super_block *super_block = fs_info->super_copy;
4806         struct btrfs_ioctl_feature_flags flags[2];
4807         struct btrfs_trans_handle *trans;
4808         u64 newflags;
4809         int ret;
4810
4811         if (!capable(CAP_SYS_ADMIN))
4812                 return -EPERM;
4813
4814         if (copy_from_user(flags, arg, sizeof(flags)))
4815                 return -EFAULT;
4816
4817         /* Nothing to do */
4818         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4819             !flags[0].incompat_flags)
4820                 return 0;
4821
4822         ret = check_feature(fs_info, flags[0].compat_flags,
4823                             flags[1].compat_flags, COMPAT);
4824         if (ret)
4825                 return ret;
4826
4827         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4828                             flags[1].compat_ro_flags, COMPAT_RO);
4829         if (ret)
4830                 return ret;
4831
4832         ret = check_feature(fs_info, flags[0].incompat_flags,
4833                             flags[1].incompat_flags, INCOMPAT);
4834         if (ret)
4835                 return ret;
4836
4837         ret = mnt_want_write_file(file);
4838         if (ret)
4839                 return ret;
4840
4841         trans = btrfs_start_transaction(root, 0);
4842         if (IS_ERR(trans)) {
4843                 ret = PTR_ERR(trans);
4844                 goto out_drop_write;
4845         }
4846
4847         spin_lock(&fs_info->super_lock);
4848         newflags = btrfs_super_compat_flags(super_block);
4849         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4850         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4851         btrfs_set_super_compat_flags(super_block, newflags);
4852
4853         newflags = btrfs_super_compat_ro_flags(super_block);
4854         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4855         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4856         btrfs_set_super_compat_ro_flags(super_block, newflags);
4857
4858         newflags = btrfs_super_incompat_flags(super_block);
4859         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4860         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4861         btrfs_set_super_incompat_flags(super_block, newflags);
4862         spin_unlock(&fs_info->super_lock);
4863
4864         ret = btrfs_commit_transaction(trans);
4865 out_drop_write:
4866         mnt_drop_write_file(file);
4867
4868         return ret;
4869 }
4870
4871 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4872 {
4873         struct btrfs_ioctl_send_args *arg;
4874         int ret;
4875
4876         if (compat) {
4877 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4878                 struct btrfs_ioctl_send_args_32 args32;
4879
4880                 ret = copy_from_user(&args32, argp, sizeof(args32));
4881                 if (ret)
4882                         return -EFAULT;
4883                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4884                 if (!arg)
4885                         return -ENOMEM;
4886                 arg->send_fd = args32.send_fd;
4887                 arg->clone_sources_count = args32.clone_sources_count;
4888                 arg->clone_sources = compat_ptr(args32.clone_sources);
4889                 arg->parent_root = args32.parent_root;
4890                 arg->flags = args32.flags;
4891                 memcpy(arg->reserved, args32.reserved,
4892                        sizeof(args32.reserved));
4893 #else
4894                 return -ENOTTY;
4895 #endif
4896         } else {
4897                 arg = memdup_user(argp, sizeof(*arg));
4898                 if (IS_ERR(arg))
4899                         return PTR_ERR(arg);
4900         }
4901         ret = btrfs_ioctl_send(file, arg);
4902         kfree(arg);
4903         return ret;
4904 }
4905
4906 long btrfs_ioctl(struct file *file, unsigned int
4907                 cmd, unsigned long arg)
4908 {
4909         struct inode *inode = file_inode(file);
4910         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4911         struct btrfs_root *root = BTRFS_I(inode)->root;
4912         void __user *argp = (void __user *)arg;
4913
4914         switch (cmd) {
4915         case FS_IOC_GETFLAGS:
4916                 return btrfs_ioctl_getflags(file, argp);
4917         case FS_IOC_SETFLAGS:
4918                 return btrfs_ioctl_setflags(file, argp);
4919         case FS_IOC_GETVERSION:
4920                 return btrfs_ioctl_getversion(file, argp);
4921         case FS_IOC_GETFSLABEL:
4922                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4923         case FS_IOC_SETFSLABEL:
4924                 return btrfs_ioctl_set_fslabel(file, argp);
4925         case FITRIM:
4926                 return btrfs_ioctl_fitrim(fs_info, argp);
4927         case BTRFS_IOC_SNAP_CREATE:
4928                 return btrfs_ioctl_snap_create(file, argp, 0);
4929         case BTRFS_IOC_SNAP_CREATE_V2:
4930                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4931         case BTRFS_IOC_SUBVOL_CREATE:
4932                 return btrfs_ioctl_snap_create(file, argp, 1);
4933         case BTRFS_IOC_SUBVOL_CREATE_V2:
4934                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4935         case BTRFS_IOC_SNAP_DESTROY:
4936                 return btrfs_ioctl_snap_destroy(file, argp, false);
4937         case BTRFS_IOC_SNAP_DESTROY_V2:
4938                 return btrfs_ioctl_snap_destroy(file, argp, true);
4939         case BTRFS_IOC_SUBVOL_GETFLAGS:
4940                 return btrfs_ioctl_subvol_getflags(file, argp);
4941         case BTRFS_IOC_SUBVOL_SETFLAGS:
4942                 return btrfs_ioctl_subvol_setflags(file, argp);
4943         case BTRFS_IOC_DEFAULT_SUBVOL:
4944                 return btrfs_ioctl_default_subvol(file, argp);
4945         case BTRFS_IOC_DEFRAG:
4946                 return btrfs_ioctl_defrag(file, NULL);
4947         case BTRFS_IOC_DEFRAG_RANGE:
4948                 return btrfs_ioctl_defrag(file, argp);
4949         case BTRFS_IOC_RESIZE:
4950                 return btrfs_ioctl_resize(file, argp);
4951         case BTRFS_IOC_ADD_DEV:
4952                 return btrfs_ioctl_add_dev(fs_info, argp);
4953         case BTRFS_IOC_RM_DEV:
4954                 return btrfs_ioctl_rm_dev(file, argp);
4955         case BTRFS_IOC_RM_DEV_V2:
4956                 return btrfs_ioctl_rm_dev_v2(file, argp);
4957         case BTRFS_IOC_FS_INFO:
4958                 return btrfs_ioctl_fs_info(fs_info, argp);
4959         case BTRFS_IOC_DEV_INFO:
4960                 return btrfs_ioctl_dev_info(fs_info, argp);
4961         case BTRFS_IOC_BALANCE:
4962                 return btrfs_ioctl_balance(file, NULL);
4963         case BTRFS_IOC_TREE_SEARCH:
4964                 return btrfs_ioctl_tree_search(file, argp);
4965         case BTRFS_IOC_TREE_SEARCH_V2:
4966                 return btrfs_ioctl_tree_search_v2(file, argp);
4967         case BTRFS_IOC_INO_LOOKUP:
4968                 return btrfs_ioctl_ino_lookup(file, argp);
4969         case BTRFS_IOC_INO_PATHS:
4970                 return btrfs_ioctl_ino_to_path(root, argp);
4971         case BTRFS_IOC_LOGICAL_INO:
4972                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4973         case BTRFS_IOC_LOGICAL_INO_V2:
4974                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4975         case BTRFS_IOC_SPACE_INFO:
4976                 return btrfs_ioctl_space_info(fs_info, argp);
4977         case BTRFS_IOC_SYNC: {
4978                 int ret;
4979
4980                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4981                 if (ret)
4982                         return ret;
4983                 ret = btrfs_sync_fs(inode->i_sb, 1);
4984                 /*
4985                  * The transaction thread may want to do more work,
4986                  * namely it pokes the cleaner kthread that will start
4987                  * processing uncleaned subvols.
4988                  */
4989                 wake_up_process(fs_info->transaction_kthread);
4990                 return ret;
4991         }
4992         case BTRFS_IOC_START_SYNC:
4993                 return btrfs_ioctl_start_sync(root, argp);
4994         case BTRFS_IOC_WAIT_SYNC:
4995                 return btrfs_ioctl_wait_sync(fs_info, argp);
4996         case BTRFS_IOC_SCRUB:
4997                 return btrfs_ioctl_scrub(file, argp);
4998         case BTRFS_IOC_SCRUB_CANCEL:
4999                 return btrfs_ioctl_scrub_cancel(fs_info);
5000         case BTRFS_IOC_SCRUB_PROGRESS:
5001                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5002         case BTRFS_IOC_BALANCE_V2:
5003                 return btrfs_ioctl_balance(file, argp);
5004         case BTRFS_IOC_BALANCE_CTL:
5005                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5006         case BTRFS_IOC_BALANCE_PROGRESS:
5007                 return btrfs_ioctl_balance_progress(fs_info, argp);
5008         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5009                 return btrfs_ioctl_set_received_subvol(file, argp);
5010 #ifdef CONFIG_64BIT
5011         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5012                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5013 #endif
5014         case BTRFS_IOC_SEND:
5015                 return _btrfs_ioctl_send(file, argp, false);
5016 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5017         case BTRFS_IOC_SEND_32:
5018                 return _btrfs_ioctl_send(file, argp, true);
5019 #endif
5020         case BTRFS_IOC_GET_DEV_STATS:
5021                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5022         case BTRFS_IOC_QUOTA_CTL:
5023                 return btrfs_ioctl_quota_ctl(file, argp);
5024         case BTRFS_IOC_QGROUP_ASSIGN:
5025                 return btrfs_ioctl_qgroup_assign(file, argp);
5026         case BTRFS_IOC_QGROUP_CREATE:
5027                 return btrfs_ioctl_qgroup_create(file, argp);
5028         case BTRFS_IOC_QGROUP_LIMIT:
5029                 return btrfs_ioctl_qgroup_limit(file, argp);
5030         case BTRFS_IOC_QUOTA_RESCAN:
5031                 return btrfs_ioctl_quota_rescan(file, argp);
5032         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5033                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5034         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5035                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5036         case BTRFS_IOC_DEV_REPLACE:
5037                 return btrfs_ioctl_dev_replace(fs_info, argp);
5038         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5039                 return btrfs_ioctl_get_supported_features(argp);
5040         case BTRFS_IOC_GET_FEATURES:
5041                 return btrfs_ioctl_get_features(fs_info, argp);
5042         case BTRFS_IOC_SET_FEATURES:
5043                 return btrfs_ioctl_set_features(file, argp);
5044         case FS_IOC_FSGETXATTR:
5045                 return btrfs_ioctl_fsgetxattr(file, argp);
5046         case FS_IOC_FSSETXATTR:
5047                 return btrfs_ioctl_fssetxattr(file, argp);
5048         case BTRFS_IOC_GET_SUBVOL_INFO:
5049                 return btrfs_ioctl_get_subvol_info(file, argp);
5050         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5051                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5052         case BTRFS_IOC_INO_LOOKUP_USER:
5053                 return btrfs_ioctl_ino_lookup_user(file, argp);
5054         }
5055
5056         return -ENOTTY;
5057 }
5058
5059 #ifdef CONFIG_COMPAT
5060 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5061 {
5062         /*
5063          * These all access 32-bit values anyway so no further
5064          * handling is necessary.
5065          */
5066         switch (cmd) {
5067         case FS_IOC32_GETFLAGS:
5068                 cmd = FS_IOC_GETFLAGS;
5069                 break;
5070         case FS_IOC32_SETFLAGS:
5071                 cmd = FS_IOC_SETFLAGS;
5072                 break;
5073         case FS_IOC32_GETVERSION:
5074                 cmd = FS_IOC_GETVERSION;
5075                 break;
5076         }
5077
5078         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5079 }
5080 #endif