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          * If the fs is mounted with nologreplay, which requires it to be
 532          * mounted in RO mode as well, we can not allow discard on free space
 533          * inside block groups, because log trees refer to extents that are not
 534          * pinned in a block group's free space cache (pinning the extents is
 535          * precisely the first phase of replaying a log tree).
 536          */
 537         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 538                 return -EROFS;
 539
 540         rcu_read_lock();
 541         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 542                                 dev_list) {
 543                 if (!device->bdev)
 544                         continue;
 545                 q = bdev_get_queue(device->bdev);
 546                 if (blk_queue_discard(q)) {
 547                         num_devices++;
 548                         minlen = min_t(u64, q->limits.discard_granularity,
 549                                      minlen);
 550                 }
 551         }
 552         rcu_read_unlock();
 553
 554         if (!num_devices)
 555                 return -EOPNOTSUPP;
 556         if (copy_from_user(&range, arg, sizeof(range)))
 557                 return -EFAULT;
 558
 559         /*
 560          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 561          * block group is in the logical address space, which can be any
 562          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 563          */
 564         if (range.len < fs_info->sb->s_blocksize)
 565                 return -EINVAL;
 566
 567         range.minlen = max(range.minlen, minlen);
 568         ret = btrfs_trim_fs(fs_info, &range);
 569         if (ret < 0)
 570                 return ret;
 571
 572         if (copy_to_user(arg, &range, sizeof(range)))
 573                 return -EFAULT;
 574
 575         return 0;
 576 }
 577
 578 int __pure btrfs_is_empty_uuid(u8 *uuid)
 579 {
 580         int i;
 581
 582         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 583                 if (uuid[i])
 584                         return 0;
 585         }
 586         return 1;
 587 }
 588
 589 static noinline int create_subvol(struct inode *dir,
 590                                   struct dentry *dentry,
 591                                   const char *name, int namelen,
 592                                   struct btrfs_qgroup_inherit *inherit)
 593 {
 594         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 595         struct btrfs_trans_handle *trans;
 596         struct btrfs_key key;
 597         struct btrfs_root_item *root_item;
 598         struct btrfs_inode_item *inode_item;
 599         struct extent_buffer *leaf;
 600         struct btrfs_root *root = BTRFS_I(dir)->root;
 601         struct btrfs_root *new_root;
 602         struct btrfs_block_rsv block_rsv;
 603         struct timespec64 cur_time = current_time(dir);
 604         struct inode *inode;
 605         int ret;
 606         int err;
 607         dev_t anon_dev = 0;
 608         u64 objectid;
 609         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 610         u64 index = 0;
 611
 612         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 613         if (!root_item)
 614                 return -ENOMEM;
 615
 616         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 617         if (ret)
 618                 goto fail_free;
 619
 620         ret = get_anon_bdev(&anon_dev);
 621         if (ret < 0)
 622                 goto fail_free;
 623
 624         /*
 625          * Don't create subvolume whose level is not zero. Or qgroup will be
 626          * screwed up since it assumes subvolume qgroup's level to be 0.
 627          */
 628         if (btrfs_qgroup_level(objectid)) {
 629                 ret = -ENOSPC;
 630                 goto fail_free;
 631         }
 632
 633         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 634         /*
 635          * The same as the snapshot creation, please see the comment
 636          * of create_snapshot().
 637          */
 638         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 639         if (ret)
 640                 goto fail_free;
 641
 642         trans = btrfs_start_transaction(root, 0);
 643         if (IS_ERR(trans)) {
 644                 ret = PTR_ERR(trans);
 645                 btrfs_subvolume_release_metadata(root, &block_rsv);
 646                 goto fail_free;
 647         }
 648         trans->block_rsv = &block_rsv;
 649         trans->bytes_reserved = block_rsv.size;
 650
 651         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 652         if (ret)
 653                 goto fail;
 654
 655         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 656                                       BTRFS_NESTING_NORMAL);
 657         if (IS_ERR(leaf)) {
 658                 ret = PTR_ERR(leaf);
 659                 goto fail;
 660         }
 661
 662         btrfs_mark_buffer_dirty(leaf);
 663
 664         inode_item = &root_item->inode;
 665         btrfs_set_stack_inode_generation(inode_item, 1);
 666         btrfs_set_stack_inode_size(inode_item, 3);
 667         btrfs_set_stack_inode_nlink(inode_item, 1);
 668         btrfs_set_stack_inode_nbytes(inode_item,
 669                                      fs_info->nodesize);
 670         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 671
 672         btrfs_set_root_flags(root_item, 0);
 673         btrfs_set_root_limit(root_item, 0);
 674         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 675
 676         btrfs_set_root_bytenr(root_item, leaf->start);
 677         btrfs_set_root_generation(root_item, trans->transid);
 678         btrfs_set_root_level(root_item, 0);
 679         btrfs_set_root_refs(root_item, 1);
 680         btrfs_set_root_used(root_item, leaf->len);
 681         btrfs_set_root_last_snapshot(root_item, 0);
 682
 683         btrfs_set_root_generation_v2(root_item,
 684                         btrfs_root_generation(root_item));
 685         generate_random_guid(root_item->uuid);
 686         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 687         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 688         root_item->ctime = root_item->otime;
 689         btrfs_set_root_ctransid(root_item, trans->transid);
 690         btrfs_set_root_otransid(root_item, trans->transid);
 691
 692         btrfs_tree_unlock(leaf);
 693         free_extent_buffer(leaf);
 694         leaf = NULL;
 695
 696         btrfs_set_root_dirid(root_item, new_dirid);
 697
 698         key.objectid = objectid;
 699         key.offset = 0;
 700         key.type = BTRFS_ROOT_ITEM_KEY;
 701         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 702                                 root_item);
 703         if (ret)
 704                 goto fail;
 705
 706         key.offset = (u64)-1;
 707         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 708         if (IS_ERR(new_root)) {
 709                 free_anon_bdev(anon_dev);
 710                 ret = PTR_ERR(new_root);
 711                 btrfs_abort_transaction(trans, ret);
 712                 goto fail;
 713         }
 714         /* Freeing will be done in btrfs_put_root() of new_root */
 715         anon_dev = 0;
 716
 717         btrfs_record_root_in_trans(trans, new_root);
 718
 719         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 720         btrfs_put_root(new_root);
 721         if (ret) {
 722                 /* We potentially lose an unused inode item here */
 723                 btrfs_abort_transaction(trans, ret);
 724                 goto fail;
 725         }
 726
 727         /*
 728          * insert the directory item
 729          */
 730         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 731         if (ret) {
 732                 btrfs_abort_transaction(trans, ret);
 733                 goto fail;
 734         }
 735
 736         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 737                                     BTRFS_FT_DIR, index);
 738         if (ret) {
 739                 btrfs_abort_transaction(trans, ret);
 740                 goto fail;
 741         }
 742
 743         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 744         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 745         if (ret) {
 746                 btrfs_abort_transaction(trans, ret);
 747                 goto fail;
 748         }
 749
 750         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 751                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 752         if (ret) {
 753                 btrfs_abort_transaction(trans, ret);
 754                 goto fail;
 755         }
 756
 757         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 758                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 759         if (ret)
 760                 btrfs_abort_transaction(trans, ret);
 761
 762 fail:
 763         kfree(root_item);
 764         trans->block_rsv = NULL;
 765         trans->bytes_reserved = 0;
 766         btrfs_subvolume_release_metadata(root, &block_rsv);
 767
 768         err = btrfs_commit_transaction(trans);
 769         if (err && !ret)
 770                 ret = err;
 771
 772         if (!ret) {
 773                 inode = btrfs_lookup_dentry(dir, dentry);
 774                 if (IS_ERR(inode))
 775                         return PTR_ERR(inode);
 776                 d_instantiate(dentry, inode);
 777         }
 778         return ret;
 779
 780 fail_free:
 781         if (anon_dev)
 782                 free_anon_bdev(anon_dev);
 783         kfree(root_item);
 784         return ret;
 785 }
 786
 787 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 788                            struct dentry *dentry, bool readonly,
 789                            struct btrfs_qgroup_inherit *inherit)
 790 {
 791         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 792         struct inode *inode;
 793         struct btrfs_pending_snapshot *pending_snapshot;
 794         struct btrfs_trans_handle *trans;
 795         int ret;
 796
 797         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 798                 return -EINVAL;
 799
 800         if (atomic_read(&root->nr_swapfiles)) {
 801                 btrfs_warn(fs_info,
 802                            "cannot snapshot subvolume with active swapfile");
 803                 return -ETXTBSY;
 804         }
 805
 806         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 807         if (!pending_snapshot)
 808                 return -ENOMEM;
 809
 810         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 811         if (ret < 0)
 812                 goto free_pending;
 813         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 814                         GFP_KERNEL);
 815         pending_snapshot->path = btrfs_alloc_path();
 816         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 817                 ret = -ENOMEM;
 818                 goto free_pending;
 819         }
 820
 821         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 822                              BTRFS_BLOCK_RSV_TEMP);
 823         /*
 824          * 1 - parent dir inode
 825          * 2 - dir entries
 826          * 1 - root item
 827          * 2 - root ref/backref
 828          * 1 - root of snapshot
 829          * 1 - UUID item
 830          */
 831         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 832                                         &pending_snapshot->block_rsv, 8,
 833                                         false);
 834         if (ret)
 835                 goto free_pending;
 836
 837         pending_snapshot->dentry = dentry;
 838         pending_snapshot->root = root;
 839         pending_snapshot->readonly = readonly;
 840         pending_snapshot->dir = dir;
 841         pending_snapshot->inherit = inherit;
 842
 843         trans = btrfs_start_transaction(root, 0);
 844         if (IS_ERR(trans)) {
 845                 ret = PTR_ERR(trans);
 846                 goto fail;
 847         }
 848
 849         spin_lock(&fs_info->trans_lock);
 850         list_add(&pending_snapshot->list,
 851                  &trans->transaction->pending_snapshots);
 852         spin_unlock(&fs_info->trans_lock);
 853
 854         ret = btrfs_commit_transaction(trans);
 855         if (ret)
 856                 goto fail;
 857
 858         ret = pending_snapshot->error;
 859         if (ret)
 860                 goto fail;
 861
 862         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 863         if (ret)
 864                 goto fail;
 865
 866         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 867         if (IS_ERR(inode)) {
 868                 ret = PTR_ERR(inode);
 869                 goto fail;
 870         }
 871
 872         d_instantiate(dentry, inode);
 873         ret = 0;
 874         pending_snapshot->anon_dev = 0;
 875 fail:
 876         /* Prevent double freeing of anon_dev */
 877         if (ret && pending_snapshot->snap)
 878                 pending_snapshot->snap->anon_dev = 0;
 879         btrfs_put_root(pending_snapshot->snap);
 880         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 881 free_pending:
 882         if (pending_snapshot->anon_dev)
 883                 free_anon_bdev(pending_snapshot->anon_dev);
 884         kfree(pending_snapshot->root_item);
 885         btrfs_free_path(pending_snapshot->path);
 886         kfree(pending_snapshot);
 887
 888         return ret;
 889 }
 890
 891 /*  copy of may_delete in fs/namei.c()
 892  *      Check whether we can remove a link victim from directory dir, check
 893  *  whether the type of victim is right.
 894  *  1. We can't do it if dir is read-only (done in permission())
 895  *  2. We should have write and exec permissions on dir
 896  *  3. We can't remove anything from append-only dir
 897  *  4. We can't do anything with immutable dir (done in permission())
 898  *  5. If the sticky bit on dir is set we should either
 899  *      a. be owner of dir, or
 900  *      b. be owner of victim, or
 901  *      c. have CAP_FOWNER capability
 902  *  6. If the victim is append-only or immutable we can't do anything with
 903  *     links pointing to it.
 904  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 905  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 906  *  9. We can't remove a root or mountpoint.
 907  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 908  *     nfs_async_unlink().
 909  */
 910
 911 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 912 {
 913         int error;
 914
 915         if (d_really_is_negative(victim))
 916                 return -ENOENT;
 917
 918         BUG_ON(d_inode(victim->d_parent) != dir);
 919         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 920
 921         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 922         if (error)
 923                 return error;
 924         if (IS_APPEND(dir))
 925                 return -EPERM;
 926         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 927             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 928                 return -EPERM;
 929         if (isdir) {
 930                 if (!d_is_dir(victim))
 931                         return -ENOTDIR;
 932                 if (IS_ROOT(victim))
 933                         return -EBUSY;
 934         } else if (d_is_dir(victim))
 935                 return -EISDIR;
 936         if (IS_DEADDIR(dir))
 937                 return -ENOENT;
 938         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 939                 return -EBUSY;
 940         return 0;
 941 }
 942
 943 /* copy of may_create in fs/namei.c() */
 944 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 945 {
 946         if (d_really_is_positive(child))
 947                 return -EEXIST;
 948         if (IS_DEADDIR(dir))
 949                 return -ENOENT;
 950         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 951 }
 952
 953 /*
 954  * Create a new subvolume below @parent.  This is largely modeled after
 955  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 956  * inside this filesystem so it's quite a bit simpler.
 957  */
 958 static noinline int btrfs_mksubvol(const struct path *parent,
 959                                    const char *name, int namelen,
 960                                    struct btrfs_root *snap_src,
 961                                    bool readonly,
 962                                    struct btrfs_qgroup_inherit *inherit)
 963 {
 964         struct inode *dir = d_inode(parent->dentry);
 965         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 966         struct dentry *dentry;
 967         int error;
 968
 969         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 970         if (error == -EINTR)
 971                 return error;
 972
 973         dentry = lookup_one_len(name, parent->dentry, namelen);
 974         error = PTR_ERR(dentry);
 975         if (IS_ERR(dentry))
 976                 goto out_unlock;
 977
 978         error = btrfs_may_create(dir, dentry);
 979         if (error)
 980                 goto out_dput;
 981
 982         /*
 983          * even if this name doesn't exist, we may get hash collisions.
 984          * check for them now when we can safely fail
 985          */
 986         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 987                                                dir->i_ino, name,
 988                                                namelen);
 989         if (error)
 990                 goto out_dput;
 991
 992         down_read(&fs_info->subvol_sem);
 993
 994         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 995                 goto out_up_read;
 996
 997         if (snap_src)
 998                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 999         else
1000                 error = create_subvol(dir, dentry, name, namelen, inherit);
1001
1002         if (!error)
1003                 fsnotify_mkdir(dir, dentry);
1004 out_up_read:
1005         up_read(&fs_info->subvol_sem);
1006 out_dput:
1007         dput(dentry);
1008 out_unlock:
1009         inode_unlock(dir);
1010         return error;
1011 }
1012
1013 static noinline int btrfs_mksnapshot(const struct path *parent,
1014                                    const char *name, int namelen,
1015                                    struct btrfs_root *root,
1016                                    bool readonly,
1017                                    struct btrfs_qgroup_inherit *inherit)
1018 {
1019         int ret;
1020         bool snapshot_force_cow = false;
1021
1022         /*
1023          * Force new buffered writes to reserve space even when NOCOW is
1024          * possible. This is to avoid later writeback (running dealloc) to
1025          * fallback to COW mode and unexpectedly fail with ENOSPC.
1026          */
1027         btrfs_drew_read_lock(&root->snapshot_lock);
1028
1029         ret = btrfs_start_delalloc_snapshot(root);
1030         if (ret)
1031                 goto out;
1032
1033         /*
1034          * All previous writes have started writeback in NOCOW mode, so now
1035          * we force future writes to fallback to COW mode during snapshot
1036          * creation.
1037          */
1038         atomic_inc(&root->snapshot_force_cow);
1039         snapshot_force_cow = true;
1040
1041         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1042
1043         ret = btrfs_mksubvol(parent, name, namelen,
1044                              root, readonly, inherit);
1045 out:
1046         if (snapshot_force_cow)
1047                 atomic_dec(&root->snapshot_force_cow);
1048         btrfs_drew_read_unlock(&root->snapshot_lock);
1049         return ret;
1050 }
1051
1052 /*
1053  * When we're defragging a range, we don't want to kick it off again
1054  * if it is really just waiting for delalloc to send it down.
1055  * If we find a nice big extent or delalloc range for the bytes in the
1056  * file you want to defrag, we return 0 to let you know to skip this
1057  * part of the file
1058  */
1059 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1060 {
1061         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1062         struct extent_map *em = NULL;
1063         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1064         u64 end;
1065
1066         read_lock(&em_tree->lock);
1067         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1068         read_unlock(&em_tree->lock);
1069
1070         if (em) {
1071                 end = extent_map_end(em);
1072                 free_extent_map(em);
1073                 if (end - offset > thresh)
1074                         return 0;
1075         }
1076         /* if we already have a nice delalloc here, just stop */
1077         thresh /= 2;
1078         end = count_range_bits(io_tree, &offset, offset + thresh,
1079                                thresh, EXTENT_DELALLOC, 1);
1080         if (end >= thresh)
1081                 return 0;
1082         return 1;
1083 }
1084
1085 /*
1086  * helper function to walk through a file and find extents
1087  * newer than a specific transid, and smaller than thresh.
1088  *
1089  * This is used by the defragging code to find new and small
1090  * extents
1091  */
1092 static int find_new_extents(struct btrfs_root *root,
1093                             struct inode *inode, u64 newer_than,
1094                             u64 *off, u32 thresh)
1095 {
1096         struct btrfs_path *path;
1097         struct btrfs_key min_key;
1098         struct extent_buffer *leaf;
1099         struct btrfs_file_extent_item *extent;
1100         int type;
1101         int ret;
1102         u64 ino = btrfs_ino(BTRFS_I(inode));
1103
1104         path = btrfs_alloc_path();
1105         if (!path)
1106                 return -ENOMEM;
1107
1108         min_key.objectid = ino;
1109         min_key.type = BTRFS_EXTENT_DATA_KEY;
1110         min_key.offset = *off;
1111
1112         while (1) {
1113                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1114                 if (ret != 0)
1115                         goto none;
1116 process_slot:
1117                 if (min_key.objectid != ino)
1118                         goto none;
1119                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1120                         goto none;
1121
1122                 leaf = path->nodes[0];
1123                 extent = btrfs_item_ptr(leaf, path->slots[0],
1124                                         struct btrfs_file_extent_item);
1125
1126                 type = btrfs_file_extent_type(leaf, extent);
1127                 if (type == BTRFS_FILE_EXTENT_REG &&
1128                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1129                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1130                         *off = min_key.offset;
1131                         btrfs_free_path(path);
1132                         return 0;
1133                 }
1134
1135                 path->slots[0]++;
1136                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1137                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1138                         goto process_slot;
1139                 }
1140
1141                 if (min_key.offset == (u64)-1)
1142                         goto none;
1143
1144                 min_key.offset++;
1145                 btrfs_release_path(path);
1146         }
1147 none:
1148         btrfs_free_path(path);
1149         return -ENOENT;
1150 }
1151
1152 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1153 {
1154         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1155         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1156         struct extent_map *em;
1157         u64 len = PAGE_SIZE;
1158
1159         /*
1160          * hopefully we have this extent in the tree already, try without
1161          * the full extent lock
1162          */
1163         read_lock(&em_tree->lock);
1164         em = lookup_extent_mapping(em_tree, start, len);
1165         read_unlock(&em_tree->lock);
1166
1167         if (!em) {
1168                 struct extent_state *cached = NULL;
1169                 u64 end = start + len - 1;
1170
1171                 /* get the big lock and read metadata off disk */
1172                 lock_extent_bits(io_tree, start, end, &cached);
1173                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1174                 unlock_extent_cached(io_tree, start, end, &cached);
1175
1176                 if (IS_ERR(em))
1177                         return NULL;
1178         }
1179
1180         return em;
1181 }
1182
1183 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1184 {
1185         struct extent_map *next;
1186         bool ret = true;
1187
1188         /* this is the last extent */
1189         if (em->start + em->len >= i_size_read(inode))
1190                 return false;
1191
1192         next = defrag_lookup_extent(inode, em->start + em->len);
1193         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1194                 ret = false;
1195         else if ((em->block_start + em->block_len == next->block_start) &&
1196                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1197                 ret = false;
1198
1199         free_extent_map(next);
1200         return ret;
1201 }
1202
1203 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1204                                u64 *last_len, u64 *skip, u64 *defrag_end,
1205                                int compress)
1206 {
1207         struct extent_map *em;
1208         int ret = 1;
1209         bool next_mergeable = true;
1210         bool prev_mergeable = true;
1211
1212         /*
1213          * make sure that once we start defragging an extent, we keep on
1214          * defragging it
1215          */
1216         if (start < *defrag_end)
1217                 return 1;
1218
1219         *skip = 0;
1220
1221         em = defrag_lookup_extent(inode, start);
1222         if (!em)
1223                 return 0;
1224
1225         /* this will cover holes, and inline extents */
1226         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1227                 ret = 0;
1228                 goto out;
1229         }
1230
1231         if (!*defrag_end)
1232                 prev_mergeable = false;
1233
1234         next_mergeable = defrag_check_next_extent(inode, em);
1235         /*
1236          * we hit a real extent, if it is big or the next extent is not a
1237          * real extent, don't bother defragging it
1238          */
1239         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1240             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1241                 ret = 0;
1242 out:
1243         /*
1244          * last_len ends up being a counter of how many bytes we've defragged.
1245          * every time we choose not to defrag an extent, we reset *last_len
1246          * so that the next tiny extent will force a defrag.
1247          *
1248          * The end result of this is that tiny extents before a single big
1249          * extent will force at least part of that big extent to be defragged.
1250          */
1251         if (ret) {
1252                 *defrag_end = extent_map_end(em);
1253         } else {
1254                 *last_len = 0;
1255                 *skip = extent_map_end(em);
1256                 *defrag_end = 0;
1257         }
1258
1259         free_extent_map(em);
1260         return ret;
1261 }
1262
1263 /*
1264  * it doesn't do much good to defrag one or two pages
1265  * at a time.  This pulls in a nice chunk of pages
1266  * to COW and defrag.
1267  *
1268  * It also makes sure the delalloc code has enough
1269  * dirty data to avoid making new small extents as part
1270  * of the defrag
1271  *
1272  * It's a good idea to start RA on this range
1273  * before calling this.
1274  */
1275 static int cluster_pages_for_defrag(struct inode *inode,
1276                                     struct page **pages,
1277                                     unsigned long start_index,
1278                                     unsigned long num_pages)
1279 {
1280         unsigned long file_end;
1281         u64 isize = i_size_read(inode);
1282         u64 page_start;
1283         u64 page_end;
1284         u64 page_cnt;
1285         u64 start = (u64)start_index << PAGE_SHIFT;
1286         u64 search_start;
1287         int ret;
1288         int i;
1289         int i_done;
1290         struct btrfs_ordered_extent *ordered;
1291         struct extent_state *cached_state = NULL;
1292         struct extent_io_tree *tree;
1293         struct extent_changeset *data_reserved = NULL;
1294         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1295
1296         file_end = (isize - 1) >> PAGE_SHIFT;
1297         if (!isize || start_index > file_end)
1298                 return 0;
1299
1300         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1301
1302         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1303                         start, page_cnt << PAGE_SHIFT);
1304         if (ret)
1305                 return ret;
1306         i_done = 0;
1307         tree = &BTRFS_I(inode)->io_tree;
1308
1309         /* step one, lock all the pages */
1310         for (i = 0; i < page_cnt; i++) {
1311                 struct page *page;
1312 again:
1313                 page = find_or_create_page(inode->i_mapping,
1314                                            start_index + i, mask);
1315                 if (!page)
1316                         break;
1317
1318                 page_start = page_offset(page);
1319                 page_end = page_start + PAGE_SIZE - 1;
1320                 while (1) {
1321                         lock_extent_bits(tree, page_start, page_end,
1322                                          &cached_state);
1323                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1324                                                               page_start);
1325                         unlock_extent_cached(tree, page_start, page_end,
1326                                              &cached_state);
1327                         if (!ordered)
1328                                 break;
1329
1330                         unlock_page(page);
1331                         btrfs_start_ordered_extent(ordered, 1);
1332                         btrfs_put_ordered_extent(ordered);
1333                         lock_page(page);
1334                         /*
1335                          * we unlocked the page above, so we need check if
1336                          * it was released or not.
1337                          */
1338                         if (page->mapping != inode->i_mapping) {
1339                                 unlock_page(page);
1340                                 put_page(page);
1341                                 goto again;
1342                         }
1343                 }
1344
1345                 if (!PageUptodate(page)) {
1346                         btrfs_readpage(NULL, page);
1347                         lock_page(page);
1348                         if (!PageUptodate(page)) {
1349                                 unlock_page(page);
1350                                 put_page(page);
1351                                 ret = -EIO;
1352                                 break;
1353                         }
1354                 }
1355
1356                 if (page->mapping != inode->i_mapping) {
1357                         unlock_page(page);
1358                         put_page(page);
1359                         goto again;
1360                 }
1361
1362                 pages[i] = page;
1363                 i_done++;
1364         }
1365         if (!i_done || ret)
1366                 goto out;
1367
1368         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1369                 goto out;
1370
1371         /*
1372          * so now we have a nice long stream of locked
1373          * and up to date pages, lets wait on them
1374          */
1375         for (i = 0; i < i_done; i++)
1376                 wait_on_page_writeback(pages[i]);
1377
1378         page_start = page_offset(pages[0]);
1379         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1380
1381         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1382                          page_start, page_end - 1, &cached_state);
1383
1384         /*
1385          * When defragmenting we skip ranges that have holes or inline extents,
1386          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1387          * space. At btrfs_defrag_file(), we check if a range should be defragged
1388          * before locking the inode and then, if it should, we trigger a sync
1389          * page cache readahead - we lock the inode only after that to avoid
1390          * blocking for too long other tasks that possibly want to operate on
1391          * other file ranges. But before we were able to get the inode lock,
1392          * some other task may have punched a hole in the range, or we may have
1393          * now an inline extent, in which case we should not defrag. So check
1394          * for that here, where we have the inode and the range locked, and bail
1395          * out if that happened.
1396          */
1397         search_start = page_start;
1398         while (search_start < page_end) {
1399                 struct extent_map *em;
1400
1401                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1402                                       page_end - search_start);
1403                 if (IS_ERR(em)) {
1404                         ret = PTR_ERR(em);
1405                         goto out_unlock_range;
1406                 }
1407                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1408                         free_extent_map(em);
1409                         /* Ok, 0 means we did not defrag anything */
1410                         ret = 0;
1411                         goto out_unlock_range;
1412                 }
1413                 search_start = extent_map_end(em);
1414                 free_extent_map(em);
1415         }
1416
1417         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1418                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1419                           EXTENT_DEFRAG, 0, 0, &cached_state);
1420
1421         if (i_done != page_cnt) {
1422                 spin_lock(&BTRFS_I(inode)->lock);
1423                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1424                 spin_unlock(&BTRFS_I(inode)->lock);
1425                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1426                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1427         }
1428
1429
1430         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1431                           &cached_state);
1432
1433         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1434                              page_start, page_end - 1, &cached_state);
1435
1436         for (i = 0; i < i_done; i++) {
1437                 clear_page_dirty_for_io(pages[i]);
1438                 ClearPageChecked(pages[i]);
1439                 set_page_extent_mapped(pages[i]);
1440                 set_page_dirty(pages[i]);
1441                 unlock_page(pages[i]);
1442                 put_page(pages[i]);
1443         }
1444         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1445         extent_changeset_free(data_reserved);
1446         return i_done;
1447
1448 out_unlock_range:
1449         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1450                              page_start, page_end - 1, &cached_state);
1451 out:
1452         for (i = 0; i < i_done; i++) {
1453                 unlock_page(pages[i]);
1454                 put_page(pages[i]);
1455         }
1456         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1457                         start, page_cnt << PAGE_SHIFT, true);
1458         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1459         extent_changeset_free(data_reserved);
1460         return ret;
1461
1462 }
1463
1464 int btrfs_defrag_file(struct inode *inode, struct file *file,
1465                       struct btrfs_ioctl_defrag_range_args *range,
1466                       u64 newer_than, unsigned long max_to_defrag)
1467 {
1468         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1469         struct btrfs_root *root = BTRFS_I(inode)->root;
1470         struct file_ra_state *ra = NULL;
1471         unsigned long last_index;
1472         u64 isize = i_size_read(inode);
1473         u64 last_len = 0;
1474         u64 skip = 0;
1475         u64 defrag_end = 0;
1476         u64 newer_off = range->start;
1477         unsigned long i;
1478         unsigned long ra_index = 0;
1479         int ret;
1480         int defrag_count = 0;
1481         int compress_type = BTRFS_COMPRESS_ZLIB;
1482         u32 extent_thresh = range->extent_thresh;
1483         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1484         unsigned long cluster = max_cluster;
1485         u64 new_align = ~((u64)SZ_128K - 1);
1486         struct page **pages = NULL;
1487         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1488
1489         if (isize == 0)
1490                 return 0;
1491
1492         if (range->start >= isize)
1493                 return -EINVAL;
1494
1495         if (do_compress) {
1496                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1497                         return -EINVAL;
1498                 if (range->compress_type)
1499                         compress_type = range->compress_type;
1500         }
1501
1502         if (extent_thresh == 0)
1503                 extent_thresh = SZ_256K;
1504
1505         /*
1506          * If we were not given a file, allocate a readahead context. As
1507          * readahead is just an optimization, defrag will work without it so
1508          * we don't error out.
1509          */
1510         if (!file) {
1511                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1512                 if (ra)
1513                         file_ra_state_init(ra, inode->i_mapping);
1514         } else {
1515                 ra = &file->f_ra;
1516         }
1517
1518         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1519         if (!pages) {
1520                 ret = -ENOMEM;
1521                 goto out_ra;
1522         }
1523
1524         /* find the last page to defrag */
1525         if (range->start + range->len > range->start) {
1526                 last_index = min_t(u64, isize - 1,
1527                          range->start + range->len - 1) >> PAGE_SHIFT;
1528         } else {
1529                 last_index = (isize - 1) >> PAGE_SHIFT;
1530         }
1531
1532         if (newer_than) {
1533                 ret = find_new_extents(root, inode, newer_than,
1534                                        &newer_off, SZ_64K);
1535                 if (!ret) {
1536                         range->start = newer_off;
1537                         /*
1538                          * we always align our defrag to help keep
1539                          * the extents in the file evenly spaced
1540                          */
1541                         i = (newer_off & new_align) >> PAGE_SHIFT;
1542                 } else
1543                         goto out_ra;
1544         } else {
1545                 i = range->start >> PAGE_SHIFT;
1546         }
1547         if (!max_to_defrag)
1548                 max_to_defrag = last_index - i + 1;
1549
1550         /*
1551          * make writeback starts from i, so the defrag range can be
1552          * written sequentially.
1553          */
1554         if (i < inode->i_mapping->writeback_index)
1555                 inode->i_mapping->writeback_index = i;
1556
1557         while (i <= last_index && defrag_count < max_to_defrag &&
1558                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1559                 /*
1560                  * make sure we stop running if someone unmounts
1561                  * the FS
1562                  */
1563                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1564                         break;
1565
1566                 if (btrfs_defrag_cancelled(fs_info)) {
1567                         btrfs_debug(fs_info, "defrag_file cancelled");
1568                         ret = -EAGAIN;
1569                         break;
1570                 }
1571
1572                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1573                                          extent_thresh, &last_len, &skip,
1574                                          &defrag_end, do_compress)){
1575                         unsigned long next;
1576                         /*
1577                          * the should_defrag function tells us how much to skip
1578                          * bump our counter by the suggested amount
1579                          */
1580                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1581                         i = max(i + 1, next);
1582                         continue;
1583                 }
1584
1585                 if (!newer_than) {
1586                         cluster = (PAGE_ALIGN(defrag_end) >>
1587                                    PAGE_SHIFT) - i;
1588                         cluster = min(cluster, max_cluster);
1589                 } else {
1590                         cluster = max_cluster;
1591                 }
1592
1593                 if (i + cluster > ra_index) {
1594                         ra_index = max(i, ra_index);
1595                         if (ra)
1596                                 page_cache_sync_readahead(inode->i_mapping, ra,
1597                                                 file, ra_index, cluster);
1598                         ra_index += cluster;
1599                 }
1600
1601                 inode_lock(inode);
1602                 if (IS_SWAPFILE(inode)) {
1603                         ret = -ETXTBSY;
1604                 } else {
1605                         if (do_compress)
1606                                 BTRFS_I(inode)->defrag_compress = compress_type;
1607                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1608                 }
1609                 if (ret < 0) {
1610                         inode_unlock(inode);
1611                         goto out_ra;
1612                 }
1613
1614                 defrag_count += ret;
1615                 balance_dirty_pages_ratelimited(inode->i_mapping);
1616                 inode_unlock(inode);
1617
1618                 if (newer_than) {
1619                         if (newer_off == (u64)-1)
1620                                 break;
1621
1622                         if (ret > 0)
1623                                 i += ret;
1624
1625                         newer_off = max(newer_off + 1,
1626                                         (u64)i << PAGE_SHIFT);
1627
1628                         ret = find_new_extents(root, inode, newer_than,
1629                                                &newer_off, SZ_64K);
1630                         if (!ret) {
1631                                 range->start = newer_off;
1632                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1633                         } else {
1634                                 break;
1635                         }
1636                 } else {
1637                         if (ret > 0) {
1638                                 i += ret;
1639                                 last_len += ret << PAGE_SHIFT;
1640                         } else {
1641                                 i++;
1642                                 last_len = 0;
1643                         }
1644                 }
1645         }
1646
1647         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1648                 filemap_flush(inode->i_mapping);
1649                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1650                              &BTRFS_I(inode)->runtime_flags))
1651                         filemap_flush(inode->i_mapping);
1652         }
1653
1654         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1655                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1656         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1657                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1658         }
1659
1660         ret = defrag_count;
1661
1662 out_ra:
1663         if (do_compress) {
1664                 inode_lock(inode);
1665                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1666                 inode_unlock(inode);
1667         }
1668         if (!file)
1669                 kfree(ra);
1670         kfree(pages);
1671         return ret;
1672 }
1673
1674 static noinline int btrfs_ioctl_resize(struct file *file,
1675                                         void __user *arg)
1676 {
1677         struct inode *inode = file_inode(file);
1678         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1679         u64 new_size;
1680         u64 old_size;
1681         u64 devid = 1;
1682         struct btrfs_root *root = BTRFS_I(inode)->root;
1683         struct btrfs_ioctl_vol_args *vol_args;
1684         struct btrfs_trans_handle *trans;
1685         struct btrfs_device *device = NULL;
1686         char *sizestr;
1687         char *retptr;
1688         char *devstr = NULL;
1689         int ret = 0;
1690         int mod = 0;
1691
1692         if (!capable(CAP_SYS_ADMIN))
1693                 return -EPERM;
1694
1695         ret = mnt_want_write_file(file);
1696         if (ret)
1697                 return ret;
1698
1699         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_RESIZE)) {
1700                 mnt_drop_write_file(file);
1701                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1702         }
1703
1704         vol_args = memdup_user(arg, sizeof(*vol_args));
1705         if (IS_ERR(vol_args)) {
1706                 ret = PTR_ERR(vol_args);
1707                 goto out;
1708         }
1709
1710         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1711
1712         sizestr = vol_args->name;
1713         devstr = strchr(sizestr, ':');
1714         if (devstr) {
1715                 sizestr = devstr + 1;
1716                 *devstr = '\0';
1717                 devstr = vol_args->name;
1718                 ret = kstrtoull(devstr, 10, &devid);
1719                 if (ret)
1720                         goto out_free;
1721                 if (!devid) {
1722                         ret = -EINVAL;
1723                         goto out_free;
1724                 }
1725                 btrfs_info(fs_info, "resizing devid %llu", devid);
1726         }
1727
1728         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
1729         if (!device) {
1730                 btrfs_info(fs_info, "resizer unable to find device %llu",
1731                            devid);
1732                 ret = -ENODEV;
1733                 goto out_free;
1734         }
1735
1736         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1737                 btrfs_info(fs_info,
1738                            "resizer unable to apply on readonly device %llu",
1739                        devid);
1740                 ret = -EPERM;
1741                 goto out_free;
1742         }
1743
1744         if (!strcmp(sizestr, "max"))
1745                 new_size = device->bdev->bd_inode->i_size;
1746         else {
1747                 if (sizestr[0] == '-') {
1748                         mod = -1;
1749                         sizestr++;
1750                 } else if (sizestr[0] == '+') {
1751                         mod = 1;
1752                         sizestr++;
1753                 }
1754                 new_size = memparse(sizestr, &retptr);
1755                 if (*retptr != '\0' || new_size == 0) {
1756                         ret = -EINVAL;
1757                         goto out_free;
1758                 }
1759         }
1760
1761         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1762                 ret = -EPERM;
1763                 goto out_free;
1764         }
1765
1766         old_size = btrfs_device_get_total_bytes(device);
1767
1768         if (mod < 0) {
1769                 if (new_size > old_size) {
1770                         ret = -EINVAL;
1771                         goto out_free;
1772                 }
1773                 new_size = old_size - new_size;
1774         } else if (mod > 0) {
1775                 if (new_size > ULLONG_MAX - old_size) {
1776                         ret = -ERANGE;
1777                         goto out_free;
1778                 }
1779                 new_size = old_size + new_size;
1780         }
1781
1782         if (new_size < SZ_256M) {
1783                 ret = -EINVAL;
1784                 goto out_free;
1785         }
1786         if (new_size > device->bdev->bd_inode->i_size) {
1787                 ret = -EFBIG;
1788                 goto out_free;
1789         }
1790
1791         new_size = round_down(new_size, fs_info->sectorsize);
1792
1793         if (new_size > old_size) {
1794                 trans = btrfs_start_transaction(root, 0);
1795                 if (IS_ERR(trans)) {
1796                         ret = PTR_ERR(trans);
1797                         goto out_free;
1798                 }
1799                 ret = btrfs_grow_device(trans, device, new_size);
1800                 btrfs_commit_transaction(trans);
1801         } else if (new_size < old_size) {
1802                 ret = btrfs_shrink_device(device, new_size);
1803         } /* equal, nothing need to do */
1804
1805         if (ret == 0 && new_size != old_size)
1806                 btrfs_info_in_rcu(fs_info,
1807                         "resize device %s (devid %llu) from %llu to %llu",
1808                         rcu_str_deref(device->name), device->devid,
1809                         old_size, new_size);
1810 out_free:
1811         kfree(vol_args);
1812 out:
1813         btrfs_exclop_finish(fs_info);
1814         mnt_drop_write_file(file);
1815         return ret;
1816 }
1817
1818 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1819                                 const char *name, unsigned long fd, int subvol,
1820                                 bool readonly,
1821                                 struct btrfs_qgroup_inherit *inherit)
1822 {
1823         int namelen;
1824         int ret = 0;
1825
1826         if (!S_ISDIR(file_inode(file)->i_mode))
1827                 return -ENOTDIR;
1828
1829         ret = mnt_want_write_file(file);
1830         if (ret)
1831                 goto out;
1832
1833         namelen = strlen(name);
1834         if (strchr(name, '/')) {
1835                 ret = -EINVAL;
1836                 goto out_drop_write;
1837         }
1838
1839         if (name[0] == '.' &&
1840            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1841                 ret = -EEXIST;
1842                 goto out_drop_write;
1843         }
1844
1845         if (subvol) {
1846                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1847                                      NULL, readonly, inherit);
1848         } else {
1849                 struct fd src = fdget(fd);
1850                 struct inode *src_inode;
1851                 if (!src.file) {
1852                         ret = -EINVAL;
1853                         goto out_drop_write;
1854                 }
1855
1856                 src_inode = file_inode(src.file);
1857                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1858                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1859                                    "Snapshot src from another FS");
1860                         ret = -EXDEV;
1861                 } else if (!inode_owner_or_capable(src_inode)) {
1862                         /*
1863                          * Subvolume creation is not restricted, but snapshots
1864                          * are limited to own subvolumes only
1865                          */
1866                         ret = -EPERM;
1867                 } else {
1868                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1869                                              BTRFS_I(src_inode)->root,
1870                                              readonly, inherit);
1871                 }
1872                 fdput(src);
1873         }
1874 out_drop_write:
1875         mnt_drop_write_file(file);
1876 out:
1877         return ret;
1878 }
1879
1880 static noinline int btrfs_ioctl_snap_create(struct file *file,
1881                                             void __user *arg, int subvol)
1882 {
1883         struct btrfs_ioctl_vol_args *vol_args;
1884         int ret;
1885
1886         if (!S_ISDIR(file_inode(file)->i_mode))
1887                 return -ENOTDIR;
1888
1889         vol_args = memdup_user(arg, sizeof(*vol_args));
1890         if (IS_ERR(vol_args))
1891                 return PTR_ERR(vol_args);
1892         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1893
1894         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1895                                         subvol, false, NULL);
1896
1897         kfree(vol_args);
1898         return ret;
1899 }
1900
1901 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1902                                                void __user *arg, int subvol)
1903 {
1904         struct btrfs_ioctl_vol_args_v2 *vol_args;
1905         int ret;
1906         bool readonly = false;
1907         struct btrfs_qgroup_inherit *inherit = NULL;
1908
1909         if (!S_ISDIR(file_inode(file)->i_mode))
1910                 return -ENOTDIR;
1911
1912         vol_args = memdup_user(arg, sizeof(*vol_args));
1913         if (IS_ERR(vol_args))
1914                 return PTR_ERR(vol_args);
1915         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1916
1917         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1918                 ret = -EOPNOTSUPP;
1919                 goto free_args;
1920         }
1921
1922         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1923                 readonly = true;
1924         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1925                 if (vol_args->size > PAGE_SIZE) {
1926                         ret = -EINVAL;
1927                         goto free_args;
1928                 }
1929                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1930                 if (IS_ERR(inherit)) {
1931                         ret = PTR_ERR(inherit);
1932                         goto free_args;
1933                 }
1934         }
1935
1936         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1937                                         subvol, readonly, inherit);
1938         if (ret)
1939                 goto free_inherit;
1940 free_inherit:
1941         kfree(inherit);
1942 free_args:
1943         kfree(vol_args);
1944         return ret;
1945 }
1946
1947 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1948                                                 void __user *arg)
1949 {
1950         struct inode *inode = file_inode(file);
1951         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1952         struct btrfs_root *root = BTRFS_I(inode)->root;
1953         int ret = 0;
1954         u64 flags = 0;
1955
1956         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1957                 return -EINVAL;
1958
1959         down_read(&fs_info->subvol_sem);
1960         if (btrfs_root_readonly(root))
1961                 flags |= BTRFS_SUBVOL_RDONLY;
1962         up_read(&fs_info->subvol_sem);
1963
1964         if (copy_to_user(arg, &flags, sizeof(flags)))
1965                 ret = -EFAULT;
1966
1967         return ret;
1968 }
1969
1970 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1971                                               void __user *arg)
1972 {
1973         struct inode *inode = file_inode(file);
1974         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1975         struct btrfs_root *root = BTRFS_I(inode)->root;
1976         struct btrfs_trans_handle *trans;
1977         u64 root_flags;
1978         u64 flags;
1979         int ret = 0;
1980
1981         if (!inode_owner_or_capable(inode))
1982                 return -EPERM;
1983
1984         ret = mnt_want_write_file(file);
1985         if (ret)
1986                 goto out;
1987
1988         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1989                 ret = -EINVAL;
1990                 goto out_drop_write;
1991         }
1992
1993         if (copy_from_user(&flags, arg, sizeof(flags))) {
1994                 ret = -EFAULT;
1995                 goto out_drop_write;
1996         }
1997
1998         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1999                 ret = -EOPNOTSUPP;
2000                 goto out_drop_write;
2001         }
2002
2003         down_write(&fs_info->subvol_sem);
2004
2005         /* nothing to do */
2006         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2007                 goto out_drop_sem;
2008
2009         root_flags = btrfs_root_flags(&root->root_item);
2010         if (flags & BTRFS_SUBVOL_RDONLY) {
2011                 btrfs_set_root_flags(&root->root_item,
2012                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2013         } else {
2014                 /*
2015                  * Block RO -> RW transition if this subvolume is involved in
2016                  * send
2017                  */
2018                 spin_lock(&root->root_item_lock);
2019                 if (root->send_in_progress == 0) {
2020                         btrfs_set_root_flags(&root->root_item,
2021                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2022                         spin_unlock(&root->root_item_lock);
2023                 } else {
2024                         spin_unlock(&root->root_item_lock);
2025                         btrfs_warn(fs_info,
2026                                    "Attempt to set subvolume %llu read-write during send",
2027                                    root->root_key.objectid);
2028                         ret = -EPERM;
2029                         goto out_drop_sem;
2030                 }
2031         }
2032
2033         trans = btrfs_start_transaction(root, 1);
2034         if (IS_ERR(trans)) {
2035                 ret = PTR_ERR(trans);
2036                 goto out_reset;
2037         }
2038
2039         ret = btrfs_update_root(trans, fs_info->tree_root,
2040                                 &root->root_key, &root->root_item);
2041         if (ret < 0) {
2042                 btrfs_end_transaction(trans);
2043                 goto out_reset;
2044         }
2045
2046         ret = btrfs_commit_transaction(trans);
2047
2048 out_reset:
2049         if (ret)
2050                 btrfs_set_root_flags(&root->root_item, root_flags);
2051 out_drop_sem:
2052         up_write(&fs_info->subvol_sem);
2053 out_drop_write:
2054         mnt_drop_write_file(file);
2055 out:
2056         return ret;
2057 }
2058
2059 static noinline int key_in_sk(struct btrfs_key *key,
2060                               struct btrfs_ioctl_search_key *sk)
2061 {
2062         struct btrfs_key test;
2063         int ret;
2064
2065         test.objectid = sk->min_objectid;
2066         test.type = sk->min_type;
2067         test.offset = sk->min_offset;
2068
2069         ret = btrfs_comp_cpu_keys(key, &test);
2070         if (ret < 0)
2071                 return 0;
2072
2073         test.objectid = sk->max_objectid;
2074         test.type = sk->max_type;
2075         test.offset = sk->max_offset;
2076
2077         ret = btrfs_comp_cpu_keys(key, &test);
2078         if (ret > 0)
2079                 return 0;
2080         return 1;
2081 }
2082
2083 static noinline int copy_to_sk(struct btrfs_path *path,
2084                                struct btrfs_key *key,
2085                                struct btrfs_ioctl_search_key *sk,
2086                                size_t *buf_size,
2087                                char __user *ubuf,
2088                                unsigned long *sk_offset,
2089                                int *num_found)
2090 {
2091         u64 found_transid;
2092         struct extent_buffer *leaf;
2093         struct btrfs_ioctl_search_header sh;
2094         struct btrfs_key test;
2095         unsigned long item_off;
2096         unsigned long item_len;
2097         int nritems;
2098         int i;
2099         int slot;
2100         int ret = 0;
2101
2102         leaf = path->nodes[0];
2103         slot = path->slots[0];
2104         nritems = btrfs_header_nritems(leaf);
2105
2106         if (btrfs_header_generation(leaf) > sk->max_transid) {
2107                 i = nritems;
2108                 goto advance_key;
2109         }
2110         found_transid = btrfs_header_generation(leaf);
2111
2112         for (i = slot; i < nritems; i++) {
2113                 item_off = btrfs_item_ptr_offset(leaf, i);
2114                 item_len = btrfs_item_size_nr(leaf, i);
2115
2116                 btrfs_item_key_to_cpu(leaf, key, i);
2117                 if (!key_in_sk(key, sk))
2118                         continue;
2119
2120                 if (sizeof(sh) + item_len > *buf_size) {
2121                         if (*num_found) {
2122                                 ret = 1;
2123                                 goto out;
2124                         }
2125
2126                         /*
2127                          * return one empty item back for v1, which does not
2128                          * handle -EOVERFLOW
2129                          */
2130
2131                         *buf_size = sizeof(sh) + item_len;
2132                         item_len = 0;
2133                         ret = -EOVERFLOW;
2134                 }
2135
2136                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2137                         ret = 1;
2138                         goto out;
2139                 }
2140
2141                 sh.objectid = key->objectid;
2142                 sh.offset = key->offset;
2143                 sh.type = key->type;
2144                 sh.len = item_len;
2145                 sh.transid = found_transid;
2146
2147                 /*
2148                  * Copy search result header. If we fault then loop again so we
2149                  * can fault in the pages and -EFAULT there if there's a
2150                  * problem. Otherwise we'll fault and then copy the buffer in
2151                  * properly this next time through
2152                  */
2153                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2154                         ret = 0;
2155                         goto out;
2156                 }
2157
2158                 *sk_offset += sizeof(sh);
2159
2160                 if (item_len) {
2161                         char __user *up = ubuf + *sk_offset;
2162                         /*
2163                          * Copy the item, same behavior as above, but reset the
2164                          * * sk_offset so we copy the full thing again.
2165                          */
2166                         if (read_extent_buffer_to_user_nofault(leaf, up,
2167                                                 item_off, item_len)) {
2168                                 ret = 0;
2169                                 *sk_offset -= sizeof(sh);
2170                                 goto out;
2171                         }
2172
2173                         *sk_offset += item_len;
2174                 }
2175                 (*num_found)++;
2176
2177                 if (ret) /* -EOVERFLOW from above */
2178                         goto out;
2179
2180                 if (*num_found >= sk->nr_items) {
2181                         ret = 1;
2182                         goto out;
2183                 }
2184         }
2185 advance_key:
2186         ret = 0;
2187         test.objectid = sk->max_objectid;
2188         test.type = sk->max_type;
2189         test.offset = sk->max_offset;
2190         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2191                 ret = 1;
2192         else if (key->offset < (u64)-1)
2193                 key->offset++;
2194         else if (key->type < (u8)-1) {
2195                 key->offset = 0;
2196                 key->type++;
2197         } else if (key->objectid < (u64)-1) {
2198                 key->offset = 0;
2199                 key->type = 0;
2200                 key->objectid++;
2201         } else
2202                 ret = 1;
2203 out:
2204         /*
2205          *  0: all items from this leaf copied, continue with next
2206          *  1: * more items can be copied, but unused buffer is too small
2207          *     * all items were found
2208          *     Either way, it will stops the loop which iterates to the next
2209          *     leaf
2210          *  -EOVERFLOW: item was to large for buffer
2211          *  -EFAULT: could not copy extent buffer back to userspace
2212          */
2213         return ret;
2214 }
2215
2216 static noinline int search_ioctl(struct inode *inode,
2217                                  struct btrfs_ioctl_search_key *sk,
2218                                  size_t *buf_size,
2219                                  char __user *ubuf)
2220 {
2221         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2222         struct btrfs_root *root;
2223         struct btrfs_key key;
2224         struct btrfs_path *path;
2225         int ret;
2226         int num_found = 0;
2227         unsigned long sk_offset = 0;
2228
2229         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2230                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2231                 return -EOVERFLOW;
2232         }
2233
2234         path = btrfs_alloc_path();
2235         if (!path)
2236                 return -ENOMEM;
2237
2238         if (sk->tree_id == 0) {
2239                 /* search the root of the inode that was passed */
2240                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2241         } else {
2242                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2243                 if (IS_ERR(root)) {
2244                         btrfs_free_path(path);
2245                         return PTR_ERR(root);
2246                 }
2247         }
2248
2249         key.objectid = sk->min_objectid;
2250         key.type = sk->min_type;
2251         key.offset = sk->min_offset;
2252
2253         while (1) {
2254                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2255                                                *buf_size - sk_offset);
2256                 if (ret)
2257                         break;
2258
2259                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2260                 if (ret != 0) {
2261                         if (ret > 0)
2262                                 ret = 0;
2263                         goto err;
2264                 }
2265                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2266                                  &sk_offset, &num_found);
2267                 btrfs_release_path(path);
2268                 if (ret)
2269                         break;
2270
2271         }
2272         if (ret > 0)
2273                 ret = 0;
2274 err:
2275         sk->nr_items = num_found;
2276         btrfs_put_root(root);
2277         btrfs_free_path(path);
2278         return ret;
2279 }
2280
2281 static noinline int btrfs_ioctl_tree_search(struct file *file,
2282                                            void __user *argp)
2283 {
2284         struct btrfs_ioctl_search_args __user *uargs;
2285         struct btrfs_ioctl_search_key sk;
2286         struct inode *inode;
2287         int ret;
2288         size_t buf_size;
2289
2290         if (!capable(CAP_SYS_ADMIN))
2291                 return -EPERM;
2292
2293         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2294
2295         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2296                 return -EFAULT;
2297
2298         buf_size = sizeof(uargs->buf);
2299
2300         inode = file_inode(file);
2301         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2302
2303         /*
2304          * In the origin implementation an overflow is handled by returning a
2305          * search header with a len of zero, so reset ret.
2306          */
2307         if (ret == -EOVERFLOW)
2308                 ret = 0;
2309
2310         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2311                 ret = -EFAULT;
2312         return ret;
2313 }
2314
2315 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2316                                                void __user *argp)
2317 {
2318         struct btrfs_ioctl_search_args_v2 __user *uarg;
2319         struct btrfs_ioctl_search_args_v2 args;
2320         struct inode *inode;
2321         int ret;
2322         size_t buf_size;
2323         const size_t buf_limit = SZ_16M;
2324
2325         if (!capable(CAP_SYS_ADMIN))
2326                 return -EPERM;
2327
2328         /* copy search header and buffer size */
2329         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2330         if (copy_from_user(&args, uarg, sizeof(args)))
2331                 return -EFAULT;
2332
2333         buf_size = args.buf_size;
2334
2335         /* limit result size to 16MB */
2336         if (buf_size > buf_limit)
2337                 buf_size = buf_limit;
2338
2339         inode = file_inode(file);
2340         ret = search_ioctl(inode, &args.key, &buf_size,
2341                            (char __user *)(&uarg->buf[0]));
2342         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2343                 ret = -EFAULT;
2344         else if (ret == -EOVERFLOW &&
2345                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2346                 ret = -EFAULT;
2347
2348         return ret;
2349 }
2350
2351 /*
2352  * Search INODE_REFs to identify path name of 'dirid' directory
2353  * in a 'tree_id' tree. and sets path name to 'name'.
2354  */
2355 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2356                                 u64 tree_id, u64 dirid, char *name)
2357 {
2358         struct btrfs_root *root;
2359         struct btrfs_key key;
2360         char *ptr;
2361         int ret = -1;
2362         int slot;
2363         int len;
2364         int total_len = 0;
2365         struct btrfs_inode_ref *iref;
2366         struct extent_buffer *l;
2367         struct btrfs_path *path;
2368
2369         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2370                 name[0]='\0';
2371                 return 0;
2372         }
2373
2374         path = btrfs_alloc_path();
2375         if (!path)
2376                 return -ENOMEM;
2377
2378         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2379
2380         root = btrfs_get_fs_root(info, tree_id, true);
2381         if (IS_ERR(root)) {
2382                 ret = PTR_ERR(root);
2383                 root = NULL;
2384                 goto out;
2385         }
2386
2387         key.objectid = dirid;
2388         key.type = BTRFS_INODE_REF_KEY;
2389         key.offset = (u64)-1;
2390
2391         while (1) {
2392                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2393                 if (ret < 0)
2394                         goto out;
2395                 else if (ret > 0) {
2396                         ret = btrfs_previous_item(root, path, dirid,
2397                                                   BTRFS_INODE_REF_KEY);
2398                         if (ret < 0)
2399                                 goto out;
2400                         else if (ret > 0) {
2401                                 ret = -ENOENT;
2402                                 goto out;
2403                         }
2404                 }
2405
2406                 l = path->nodes[0];
2407                 slot = path->slots[0];
2408                 btrfs_item_key_to_cpu(l, &key, slot);
2409
2410                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2411                 len = btrfs_inode_ref_name_len(l, iref);
2412                 ptr -= len + 1;
2413                 total_len += len + 1;
2414                 if (ptr < name) {
2415                         ret = -ENAMETOOLONG;
2416                         goto out;
2417                 }
2418
2419                 *(ptr + len) = '/';
2420                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2421
2422                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2423                         break;
2424
2425                 btrfs_release_path(path);
2426                 key.objectid = key.offset;
2427                 key.offset = (u64)-1;
2428                 dirid = key.objectid;
2429         }
2430         memmove(name, ptr, total_len);
2431         name[total_len] = '\0';
2432         ret = 0;
2433 out:
2434         btrfs_put_root(root);
2435         btrfs_free_path(path);
2436         return ret;
2437 }
2438
2439 static int btrfs_search_path_in_tree_user(struct inode *inode,
2440                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2441 {
2442         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2443         struct super_block *sb = inode->i_sb;
2444         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2445         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2446         u64 dirid = args->dirid;
2447         unsigned long item_off;
2448         unsigned long item_len;
2449         struct btrfs_inode_ref *iref;
2450         struct btrfs_root_ref *rref;
2451         struct btrfs_root *root = NULL;
2452         struct btrfs_path *path;
2453         struct btrfs_key key, key2;
2454         struct extent_buffer *leaf;
2455         struct inode *temp_inode;
2456         char *ptr;
2457         int slot;
2458         int len;
2459         int total_len = 0;
2460         int ret;
2461
2462         path = btrfs_alloc_path();
2463         if (!path)
2464                 return -ENOMEM;
2465
2466         /*
2467          * If the bottom subvolume does not exist directly under upper_limit,
2468          * construct the path in from the bottom up.
2469          */
2470         if (dirid != upper_limit.objectid) {
2471                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2472
2473                 root = btrfs_get_fs_root(fs_info, treeid, true);
2474                 if (IS_ERR(root)) {
2475                         ret = PTR_ERR(root);
2476                         goto out;
2477                 }
2478
2479                 key.objectid = dirid;
2480                 key.type = BTRFS_INODE_REF_KEY;
2481                 key.offset = (u64)-1;
2482                 while (1) {
2483                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2484                         if (ret < 0) {
2485                                 goto out_put;
2486                         } else if (ret > 0) {
2487                                 ret = btrfs_previous_item(root, path, dirid,
2488                                                           BTRFS_INODE_REF_KEY);
2489                                 if (ret < 0) {
2490                                         goto out_put;
2491                                 } else if (ret > 0) {
2492                                         ret = -ENOENT;
2493                                         goto out_put;
2494                                 }
2495                         }
2496
2497                         leaf = path->nodes[0];
2498                         slot = path->slots[0];
2499                         btrfs_item_key_to_cpu(leaf, &key, slot);
2500
2501                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2502                         len = btrfs_inode_ref_name_len(leaf, iref);
2503                         ptr -= len + 1;
2504                         total_len += len + 1;
2505                         if (ptr < args->path) {
2506                                 ret = -ENAMETOOLONG;
2507                                 goto out_put;
2508                         }
2509
2510                         *(ptr + len) = '/';
2511                         read_extent_buffer(leaf, ptr,
2512                                         (unsigned long)(iref + 1), len);
2513
2514                         /* Check the read+exec permission of this directory */
2515                         ret = btrfs_previous_item(root, path, dirid,
2516                                                   BTRFS_INODE_ITEM_KEY);
2517                         if (ret < 0) {
2518                                 goto out_put;
2519                         } else if (ret > 0) {
2520                                 ret = -ENOENT;
2521                                 goto out_put;
2522                         }
2523
2524                         leaf = path->nodes[0];
2525                         slot = path->slots[0];
2526                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2527                         if (key2.objectid != dirid) {
2528                                 ret = -ENOENT;
2529                                 goto out_put;
2530                         }
2531
2532                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2533                         if (IS_ERR(temp_inode)) {
2534                                 ret = PTR_ERR(temp_inode);
2535                                 goto out_put;
2536                         }
2537                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2538                         iput(temp_inode);
2539                         if (ret) {
2540                                 ret = -EACCES;
2541                                 goto out_put;
2542                         }
2543
2544                         if (key.offset == upper_limit.objectid)
2545                                 break;
2546                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2547                                 ret = -EACCES;
2548                                 goto out_put;
2549                         }
2550
2551                         btrfs_release_path(path);
2552                         key.objectid = key.offset;
2553                         key.offset = (u64)-1;
2554                         dirid = key.objectid;
2555                 }
2556
2557                 memmove(args->path, ptr, total_len);
2558                 args->path[total_len] = '\0';
2559                 btrfs_put_root(root);
2560                 root = NULL;
2561                 btrfs_release_path(path);
2562         }
2563
2564         /* Get the bottom subvolume's name from ROOT_REF */
2565         key.objectid = treeid;
2566         key.type = BTRFS_ROOT_REF_KEY;
2567         key.offset = args->treeid;
2568         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2569         if (ret < 0) {
2570                 goto out;
2571         } else if (ret > 0) {
2572                 ret = -ENOENT;
2573                 goto out;
2574         }
2575
2576         leaf = path->nodes[0];
2577         slot = path->slots[0];
2578         btrfs_item_key_to_cpu(leaf, &key, slot);
2579
2580         item_off = btrfs_item_ptr_offset(leaf, slot);
2581         item_len = btrfs_item_size_nr(leaf, slot);
2582         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2583         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2584         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2585                 ret = -EINVAL;
2586                 goto out;
2587         }
2588
2589         /* Copy subvolume's name */
2590         item_off += sizeof(struct btrfs_root_ref);
2591         item_len -= sizeof(struct btrfs_root_ref);
2592         read_extent_buffer(leaf, args->name, item_off, item_len);
2593         args->name[item_len] = 0;
2594
2595 out_put:
2596         btrfs_put_root(root);
2597 out:
2598         btrfs_free_path(path);
2599         return ret;
2600 }
2601
2602 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2603                                            void __user *argp)
2604 {
2605         struct btrfs_ioctl_ino_lookup_args *args;
2606         struct inode *inode;
2607         int ret = 0;
2608
2609         args = memdup_user(argp, sizeof(*args));
2610         if (IS_ERR(args))
2611                 return PTR_ERR(args);
2612
2613         inode = file_inode(file);
2614
2615         /*
2616          * Unprivileged query to obtain the containing subvolume root id. The
2617          * path is reset so it's consistent with btrfs_search_path_in_tree.
2618          */
2619         if (args->treeid == 0)
2620                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2621
2622         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2623                 args->name[0] = 0;
2624                 goto out;
2625         }
2626
2627         if (!capable(CAP_SYS_ADMIN)) {
2628                 ret = -EPERM;
2629                 goto out;
2630         }
2631
2632         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2633                                         args->treeid, args->objectid,
2634                                         args->name);
2635
2636 out:
2637         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2638                 ret = -EFAULT;
2639
2640         kfree(args);
2641         return ret;
2642 }
2643
2644 /*
2645  * Version of ino_lookup ioctl (unprivileged)
2646  *
2647  * The main differences from ino_lookup ioctl are:
2648  *
2649  *   1. Read + Exec permission will be checked using inode_permission() during
2650  *      path construction. -EACCES will be returned in case of failure.
2651  *   2. Path construction will be stopped at the inode number which corresponds
2652  *      to the fd with which this ioctl is called. If constructed path does not
2653  *      exist under fd's inode, -EACCES will be returned.
2654  *   3. The name of bottom subvolume is also searched and filled.
2655  */
2656 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2657 {
2658         struct btrfs_ioctl_ino_lookup_user_args *args;
2659         struct inode *inode;
2660         int ret;
2661
2662         args = memdup_user(argp, sizeof(*args));
2663         if (IS_ERR(args))
2664                 return PTR_ERR(args);
2665
2666         inode = file_inode(file);
2667
2668         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2669             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2670                 /*
2671                  * The subvolume does not exist under fd with which this is
2672                  * called
2673                  */
2674                 kfree(args);
2675                 return -EACCES;
2676         }
2677
2678         ret = btrfs_search_path_in_tree_user(inode, args);
2679
2680         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2681                 ret = -EFAULT;
2682
2683         kfree(args);
2684         return ret;
2685 }
2686
2687 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2688 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2689 {
2690         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2691         struct btrfs_fs_info *fs_info;
2692         struct btrfs_root *root;
2693         struct btrfs_path *path;
2694         struct btrfs_key key;
2695         struct btrfs_root_item *root_item;
2696         struct btrfs_root_ref *rref;
2697         struct extent_buffer *leaf;
2698         unsigned long item_off;
2699         unsigned long item_len;
2700         struct inode *inode;
2701         int slot;
2702         int ret = 0;
2703
2704         path = btrfs_alloc_path();
2705         if (!path)
2706                 return -ENOMEM;
2707
2708         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2709         if (!subvol_info) {
2710                 btrfs_free_path(path);
2711                 return -ENOMEM;
2712         }
2713
2714         inode = file_inode(file);
2715         fs_info = BTRFS_I(inode)->root->fs_info;
2716
2717         /* Get root_item of inode's subvolume */
2718         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2719         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2720         if (IS_ERR(root)) {
2721                 ret = PTR_ERR(root);
2722                 goto out_free;
2723         }
2724         root_item = &root->root_item;
2725
2726         subvol_info->treeid = key.objectid;
2727
2728         subvol_info->generation = btrfs_root_generation(root_item);
2729         subvol_info->flags = btrfs_root_flags(root_item);
2730
2731         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2732         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2733                                                     BTRFS_UUID_SIZE);
2734         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2735                                                     BTRFS_UUID_SIZE);
2736
2737         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2738         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2739         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2740
2741         subvol_info->otransid = btrfs_root_otransid(root_item);
2742         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2743         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2744
2745         subvol_info->stransid = btrfs_root_stransid(root_item);
2746         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2747         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2748
2749         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2750         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2751         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2752
2753         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2754                 /* Search root tree for ROOT_BACKREF of this subvolume */
2755                 key.type = BTRFS_ROOT_BACKREF_KEY;
2756                 key.offset = 0;
2757                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2758                 if (ret < 0) {
2759                         goto out;
2760                 } else if (path->slots[0] >=
2761                            btrfs_header_nritems(path->nodes[0])) {
2762                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2763                         if (ret < 0) {
2764                                 goto out;
2765                         } else if (ret > 0) {
2766                                 ret = -EUCLEAN;
2767                                 goto out;
2768                         }
2769                 }
2770
2771                 leaf = path->nodes[0];
2772                 slot = path->slots[0];
2773                 btrfs_item_key_to_cpu(leaf, &key, slot);
2774                 if (key.objectid == subvol_info->treeid &&
2775                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2776                         subvol_info->parent_id = key.offset;
2777
2778                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2779                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2780
2781                         item_off = btrfs_item_ptr_offset(leaf, slot)
2782                                         + sizeof(struct btrfs_root_ref);
2783                         item_len = btrfs_item_size_nr(leaf, slot)
2784                                         - sizeof(struct btrfs_root_ref);
2785                         read_extent_buffer(leaf, subvol_info->name,
2786                                            item_off, item_len);
2787                 } else {
2788                         ret = -ENOENT;
2789                         goto out;
2790                 }
2791         }
2792
2793         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2794                 ret = -EFAULT;
2795
2796 out:
2797         btrfs_put_root(root);
2798 out_free:
2799         btrfs_free_path(path);
2800         kfree(subvol_info);
2801         return ret;
2802 }
2803
2804 /*
2805  * Return ROOT_REF information of the subvolume containing this inode
2806  * except the subvolume name.
2807  */
2808 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2809 {
2810         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2811         struct btrfs_root_ref *rref;
2812         struct btrfs_root *root;
2813         struct btrfs_path *path;
2814         struct btrfs_key key;
2815         struct extent_buffer *leaf;
2816         struct inode *inode;
2817         u64 objectid;
2818         int slot;
2819         int ret;
2820         u8 found;
2821
2822         path = btrfs_alloc_path();
2823         if (!path)
2824                 return -ENOMEM;
2825
2826         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2827         if (IS_ERR(rootrefs)) {
2828                 btrfs_free_path(path);
2829                 return PTR_ERR(rootrefs);
2830         }
2831
2832         inode = file_inode(file);
2833         root = BTRFS_I(inode)->root->fs_info->tree_root;
2834         objectid = BTRFS_I(inode)->root->root_key.objectid;
2835
2836         key.objectid = objectid;
2837         key.type = BTRFS_ROOT_REF_KEY;
2838         key.offset = rootrefs->min_treeid;
2839         found = 0;
2840
2841         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2842         if (ret < 0) {
2843                 goto out;
2844         } else if (path->slots[0] >=
2845                    btrfs_header_nritems(path->nodes[0])) {
2846                 ret = btrfs_next_leaf(root, path);
2847                 if (ret < 0) {
2848                         goto out;
2849                 } else if (ret > 0) {
2850                         ret = -EUCLEAN;
2851                         goto out;
2852                 }
2853         }
2854         while (1) {
2855                 leaf = path->nodes[0];
2856                 slot = path->slots[0];
2857
2858                 btrfs_item_key_to_cpu(leaf, &key, slot);
2859                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2860                         ret = 0;
2861                         goto out;
2862                 }
2863
2864                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2865                         ret = -EOVERFLOW;
2866                         goto out;
2867                 }
2868
2869                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2870                 rootrefs->rootref[found].treeid = key.offset;
2871                 rootrefs->rootref[found].dirid =
2872                                   btrfs_root_ref_dirid(leaf, rref);
2873                 found++;
2874
2875                 ret = btrfs_next_item(root, path);
2876                 if (ret < 0) {
2877                         goto out;
2878                 } else if (ret > 0) {
2879                         ret = -EUCLEAN;
2880                         goto out;
2881                 }
2882         }
2883
2884 out:
2885         if (!ret || ret == -EOVERFLOW) {
2886                 rootrefs->num_items = found;
2887                 /* update min_treeid for next search */
2888                 if (found)
2889                         rootrefs->min_treeid =
2890                                 rootrefs->rootref[found - 1].treeid + 1;
2891                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2892                         ret = -EFAULT;
2893         }
2894
2895         kfree(rootrefs);
2896         btrfs_free_path(path);
2897
2898         return ret;
2899 }
2900
2901 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2902                                              void __user *arg,
2903                                              bool destroy_v2)
2904 {
2905         struct dentry *parent = file->f_path.dentry;
2906         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2907         struct dentry *dentry;
2908         struct inode *dir = d_inode(parent);
2909         struct inode *inode;
2910         struct btrfs_root *root = BTRFS_I(dir)->root;
2911         struct btrfs_root *dest = NULL;
2912         struct btrfs_ioctl_vol_args *vol_args = NULL;
2913         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2914         char *subvol_name, *subvol_name_ptr = NULL;
2915         int subvol_namelen;
2916         int err = 0;
2917         bool destroy_parent = false;
2918
2919         if (destroy_v2) {
2920                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2921                 if (IS_ERR(vol_args2))
2922                         return PTR_ERR(vol_args2);
2923
2924                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2925                         err = -EOPNOTSUPP;
2926                         goto out;
2927                 }
2928
2929                 /*
2930                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2931                  * name, same as v1 currently does.
2932                  */
2933                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2934                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2935                         subvol_name = vol_args2->name;
2936
2937                         err = mnt_want_write_file(file);
2938                         if (err)
2939                                 goto out;
2940                 } else {
2941                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2942                                 err = -EINVAL;
2943                                 goto out;
2944                         }
2945
2946                         err = mnt_want_write_file(file);
2947                         if (err)
2948                                 goto out;
2949
2950                         dentry = btrfs_get_dentry(fs_info->sb,
2951                                         BTRFS_FIRST_FREE_OBJECTID,
2952                                         vol_args2->subvolid, 0, 0);
2953                         if (IS_ERR(dentry)) {
2954                                 err = PTR_ERR(dentry);
2955                                 goto out_drop_write;
2956                         }
2957
2958                         /*
2959                          * Change the default parent since the subvolume being
2960                          * deleted can be outside of the current mount point.
2961                          */
2962                         parent = btrfs_get_parent(dentry);
2963
2964                         /*
2965                          * At this point dentry->d_name can point to '/' if the
2966                          * subvolume we want to destroy is outsite of the
2967                          * current mount point, so we need to release the
2968                          * current dentry and execute the lookup to return a new
2969                          * one with ->d_name pointing to the
2970                          * <mount point>/subvol_name.
2971                          */
2972                         dput(dentry);
2973                         if (IS_ERR(parent)) {
2974                                 err = PTR_ERR(parent);
2975                                 goto out_drop_write;
2976                         }
2977                         dir = d_inode(parent);
2978
2979                         /*
2980                          * If v2 was used with SPEC_BY_ID, a new parent was
2981                          * allocated since the subvolume can be outside of the
2982                          * current mount point. Later on we need to release this
2983                          * new parent dentry.
2984                          */
2985                         destroy_parent = true;
2986
2987                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2988                                                 fs_info, vol_args2->subvolid);
2989                         if (IS_ERR(subvol_name_ptr)) {
2990                                 err = PTR_ERR(subvol_name_ptr);
2991                                 goto free_parent;
2992                         }
2993                         /* subvol_name_ptr is already NULL termined */
2994                         subvol_name = (char *)kbasename(subvol_name_ptr);
2995                 }
2996         } else {
2997                 vol_args = memdup_user(arg, sizeof(*vol_args));
2998                 if (IS_ERR(vol_args))
2999                         return PTR_ERR(vol_args);
3000
3001                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3002                 subvol_name = vol_args->name;
3003
3004                 err = mnt_want_write_file(file);
3005                 if (err)
3006                         goto out;
3007         }
3008
3009         subvol_namelen = strlen(subvol_name);
3010
3011         if (strchr(subvol_name, '/') ||
3012             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3013                 err = -EINVAL;
3014                 goto free_subvol_name;
3015         }
3016
3017         if (!S_ISDIR(dir->i_mode)) {
3018                 err = -ENOTDIR;
3019                 goto free_subvol_name;
3020         }
3021
3022         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3023         if (err == -EINTR)
3024                 goto free_subvol_name;
3025         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
3026         if (IS_ERR(dentry)) {
3027                 err = PTR_ERR(dentry);
3028                 goto out_unlock_dir;
3029         }
3030
3031         if (d_really_is_negative(dentry)) {
3032                 err = -ENOENT;
3033                 goto out_dput;
3034         }
3035
3036         inode = d_inode(dentry);
3037         dest = BTRFS_I(inode)->root;
3038         if (!capable(CAP_SYS_ADMIN)) {
3039                 /*
3040                  * Regular user.  Only allow this with a special mount
3041                  * option, when the user has write+exec access to the
3042                  * subvol root, and when rmdir(2) would have been
3043                  * allowed.
3044                  *
3045                  * Note that this is _not_ check that the subvol is
3046                  * empty or doesn't contain data that we wouldn't
3047                  * otherwise be able to delete.
3048                  *
3049                  * Users who want to delete empty subvols should try
3050                  * rmdir(2).
3051                  */
3052                 err = -EPERM;
3053                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3054                         goto out_dput;
3055
3056                 /*
3057                  * Do not allow deletion if the parent dir is the same
3058                  * as the dir to be deleted.  That means the ioctl
3059                  * must be called on the dentry referencing the root
3060                  * of the subvol, not a random directory contained
3061                  * within it.
3062                  */
3063                 err = -EINVAL;
3064                 if (root == dest)
3065                         goto out_dput;
3066
3067                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
3068                 if (err)
3069                         goto out_dput;
3070         }
3071
3072         /* check if subvolume may be deleted by a user */
3073         err = btrfs_may_delete(dir, dentry, 1);
3074         if (err)
3075                 goto out_dput;
3076
3077         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3078                 err = -EINVAL;
3079                 goto out_dput;
3080         }
3081
3082         inode_lock(inode);
3083         err = btrfs_delete_subvolume(dir, dentry);
3084         inode_unlock(inode);
3085         if (!err) {
3086                 fsnotify_rmdir(dir, dentry);
3087                 d_delete(dentry);
3088         }
3089
3090 out_dput:
3091         dput(dentry);
3092 out_unlock_dir:
3093         inode_unlock(dir);
3094 free_subvol_name:
3095         kfree(subvol_name_ptr);
3096 free_parent:
3097         if (destroy_parent)
3098                 dput(parent);
3099 out_drop_write:
3100         mnt_drop_write_file(file);
3101 out:
3102         kfree(vol_args2);
3103         kfree(vol_args);
3104         return err;
3105 }
3106
3107 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3108 {
3109         struct inode *inode = file_inode(file);
3110         struct btrfs_root *root = BTRFS_I(inode)->root;
3111         struct btrfs_ioctl_defrag_range_args *range;
3112         int ret;
3113
3114         ret = mnt_want_write_file(file);
3115         if (ret)
3116                 return ret;
3117
3118         if (btrfs_root_readonly(root)) {
3119                 ret = -EROFS;
3120                 goto out;
3121         }
3122
3123         switch (inode->i_mode & S_IFMT) {
3124         case S_IFDIR:
3125                 if (!capable(CAP_SYS_ADMIN)) {
3126                         ret = -EPERM;
3127                         goto out;
3128                 }
3129                 ret = btrfs_defrag_root(root);
3130                 break;
3131         case S_IFREG:
3132                 /*
3133                  * Note that this does not check the file descriptor for write
3134                  * access. This prevents defragmenting executables that are
3135                  * running and allows defrag on files open in read-only mode.
3136                  */
3137                 if (!capable(CAP_SYS_ADMIN) &&
3138                     inode_permission(inode, MAY_WRITE)) {
3139                         ret = -EPERM;
3140                         goto out;
3141                 }
3142
3143                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3144                 if (!range) {
3145                         ret = -ENOMEM;
3146                         goto out;
3147                 }
3148
3149                 if (argp) {
3150                         if (copy_from_user(range, argp,
3151                                            sizeof(*range))) {
3152                                 ret = -EFAULT;
3153                                 kfree(range);
3154                                 goto out;
3155                         }
3156                         /* compression requires us to start the IO */
3157                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3158                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3159                                 range->extent_thresh = (u32)-1;
3160                         }
3161                 } else {
3162                         /* the rest are all set to zero by kzalloc */
3163                         range->len = (u64)-1;
3164                 }
3165                 ret = btrfs_defrag_file(file_inode(file), file,
3166                                         range, BTRFS_OLDEST_GENERATION, 0);
3167                 if (ret > 0)
3168                         ret = 0;
3169                 kfree(range);
3170                 break;
3171         default:
3172                 ret = -EINVAL;
3173         }
3174 out:
3175         mnt_drop_write_file(file);
3176         return ret;
3177 }
3178
3179 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3180 {
3181         struct btrfs_ioctl_vol_args *vol_args;
3182         int ret;
3183
3184         if (!capable(CAP_SYS_ADMIN))
3185                 return -EPERM;
3186
3187         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3188                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3189
3190         vol_args = memdup_user(arg, sizeof(*vol_args));
3191         if (IS_ERR(vol_args)) {
3192                 ret = PTR_ERR(vol_args);
3193                 goto out;
3194         }
3195
3196         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3197         ret = btrfs_init_new_device(fs_info, vol_args->name);
3198
3199         if (!ret)
3200                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3201
3202         kfree(vol_args);
3203 out:
3204         btrfs_exclop_finish(fs_info);
3205         return ret;
3206 }
3207
3208 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3209 {
3210         struct inode *inode = file_inode(file);
3211         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3212         struct btrfs_ioctl_vol_args_v2 *vol_args;
3213         int ret;
3214
3215         if (!capable(CAP_SYS_ADMIN))
3216                 return -EPERM;
3217
3218         ret = mnt_want_write_file(file);
3219         if (ret)
3220                 return ret;
3221
3222         vol_args = memdup_user(arg, sizeof(*vol_args));
3223         if (IS_ERR(vol_args)) {
3224                 ret = PTR_ERR(vol_args);
3225                 goto err_drop;
3226         }
3227
3228         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3229                 ret = -EOPNOTSUPP;
3230                 goto out;
3231         }
3232
3233         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3234                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3235                 goto out;
3236         }
3237
3238         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3239                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3240         } else {
3241                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3242                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3243         }
3244         btrfs_exclop_finish(fs_info);
3245
3246         if (!ret) {
3247                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3248                         btrfs_info(fs_info, "device deleted: id %llu",
3249                                         vol_args->devid);
3250                 else
3251                         btrfs_info(fs_info, "device deleted: %s",
3252                                         vol_args->name);
3253         }
3254 out:
3255         kfree(vol_args);
3256 err_drop:
3257         mnt_drop_write_file(file);
3258         return ret;
3259 }
3260
3261 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3262 {
3263         struct inode *inode = file_inode(file);
3264         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3265         struct btrfs_ioctl_vol_args *vol_args;
3266         int ret;
3267
3268         if (!capable(CAP_SYS_ADMIN))
3269                 return -EPERM;
3270
3271         ret = mnt_want_write_file(file);
3272         if (ret)
3273                 return ret;
3274
3275         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3276                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3277                 goto out_drop_write;
3278         }
3279
3280         vol_args = memdup_user(arg, sizeof(*vol_args));
3281         if (IS_ERR(vol_args)) {
3282                 ret = PTR_ERR(vol_args);
3283                 goto out;
3284         }
3285
3286         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3287         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3288
3289         if (!ret)
3290                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3291         kfree(vol_args);
3292 out:
3293         btrfs_exclop_finish(fs_info);
3294 out_drop_write:
3295         mnt_drop_write_file(file);
3296
3297         return ret;
3298 }
3299
3300 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3301                                 void __user *arg)
3302 {
3303         struct btrfs_ioctl_fs_info_args *fi_args;
3304         struct btrfs_device *device;
3305         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3306         u64 flags_in;
3307         int ret = 0;
3308
3309         fi_args = memdup_user(arg, sizeof(*fi_args));
3310         if (IS_ERR(fi_args))
3311                 return PTR_ERR(fi_args);
3312
3313         flags_in = fi_args->flags;
3314         memset(fi_args, 0, sizeof(*fi_args));
3315
3316         rcu_read_lock();
3317         fi_args->num_devices = fs_devices->num_devices;
3318
3319         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3320                 if (device->devid > fi_args->max_id)
3321                         fi_args->max_id = device->devid;
3322         }
3323         rcu_read_unlock();
3324
3325         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3326         fi_args->nodesize = fs_info->nodesize;
3327         fi_args->sectorsize = fs_info->sectorsize;
3328         fi_args->clone_alignment = fs_info->sectorsize;
3329
3330         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3331                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3332                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3333                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3334         }
3335
3336         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3337                 fi_args->generation = fs_info->generation;
3338                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3339         }
3340
3341         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3342                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3343                        sizeof(fi_args->metadata_uuid));
3344                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3345         }
3346
3347         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3348                 ret = -EFAULT;
3349
3350         kfree(fi_args);
3351         return ret;
3352 }
3353
3354 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3355                                  void __user *arg)
3356 {
3357         struct btrfs_ioctl_dev_info_args *di_args;
3358         struct btrfs_device *dev;
3359         int ret = 0;
3360         char *s_uuid = NULL;
3361
3362         di_args = memdup_user(arg, sizeof(*di_args));
3363         if (IS_ERR(di_args))
3364                 return PTR_ERR(di_args);
3365
3366         if (!btrfs_is_empty_uuid(di_args->uuid))
3367                 s_uuid = di_args->uuid;
3368
3369         rcu_read_lock();
3370         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3371                                 NULL);
3372
3373         if (!dev) {
3374                 ret = -ENODEV;
3375                 goto out;
3376         }
3377
3378         di_args->devid = dev->devid;
3379         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3380         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3381         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3382         if (dev->name) {
3383                 strncpy(di_args->path, rcu_str_deref(dev->name),
3384                                 sizeof(di_args->path) - 1);
3385                 di_args->path[sizeof(di_args->path) - 1] = 0;
3386         } else {
3387                 di_args->path[0] = '\0';
3388         }
3389
3390 out:
3391         rcu_read_unlock();
3392         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3393                 ret = -EFAULT;
3394
3395         kfree(di_args);
3396         return ret;
3397 }
3398
3399 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3400 {
3401         struct inode *inode = file_inode(file);
3402         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3403         struct btrfs_root *root = BTRFS_I(inode)->root;
3404         struct btrfs_root *new_root;
3405         struct btrfs_dir_item *di;
3406         struct btrfs_trans_handle *trans;
3407         struct btrfs_path *path = NULL;
3408         struct btrfs_disk_key disk_key;
3409         u64 objectid = 0;
3410         u64 dir_id;
3411         int ret;
3412
3413         if (!capable(CAP_SYS_ADMIN))
3414                 return -EPERM;
3415
3416         ret = mnt_want_write_file(file);
3417         if (ret)
3418                 return ret;
3419
3420         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3421                 ret = -EFAULT;
3422                 goto out;
3423         }
3424
3425         if (!objectid)
3426                 objectid = BTRFS_FS_TREE_OBJECTID;
3427
3428         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3429         if (IS_ERR(new_root)) {
3430                 ret = PTR_ERR(new_root);
3431                 goto out;
3432         }
3433         if (!is_fstree(new_root->root_key.objectid)) {
3434                 ret = -ENOENT;
3435                 goto out_free;
3436         }
3437
3438         path = btrfs_alloc_path();
3439         if (!path) {
3440                 ret = -ENOMEM;
3441                 goto out_free;
3442         }
3443
3444         trans = btrfs_start_transaction(root, 1);
3445         if (IS_ERR(trans)) {
3446                 ret = PTR_ERR(trans);
3447                 goto out_free;
3448         }
3449
3450         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3451         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3452                                    dir_id, "default", 7, 1);
3453         if (IS_ERR_OR_NULL(di)) {
3454                 btrfs_release_path(path);
3455                 btrfs_end_transaction(trans);
3456                 btrfs_err(fs_info,
3457                           "Umm, you don't have the default diritem, this isn't going to work");
3458                 ret = -ENOENT;
3459                 goto out_free;
3460         }
3461
3462         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3463         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3464         btrfs_mark_buffer_dirty(path->nodes[0]);
3465         btrfs_release_path(path);
3466
3467         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3468         btrfs_end_transaction(trans);
3469 out_free:
3470         btrfs_put_root(new_root);
3471         btrfs_free_path(path);
3472 out:
3473         mnt_drop_write_file(file);
3474         return ret;
3475 }
3476
3477 static void get_block_group_info(struct list_head *groups_list,
3478                                  struct btrfs_ioctl_space_info *space)
3479 {
3480         struct btrfs_block_group *block_group;
3481
3482         space->total_bytes = 0;
3483         space->used_bytes = 0;
3484         space->flags = 0;
3485         list_for_each_entry(block_group, groups_list, list) {
3486                 space->flags = block_group->flags;
3487                 space->total_bytes += block_group->length;
3488                 space->used_bytes += block_group->used;
3489         }
3490 }
3491
3492 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3493                                    void __user *arg)
3494 {
3495         struct btrfs_ioctl_space_args space_args;
3496         struct btrfs_ioctl_space_info space;
3497         struct btrfs_ioctl_space_info *dest;
3498         struct btrfs_ioctl_space_info *dest_orig;
3499         struct btrfs_ioctl_space_info __user *user_dest;
3500         struct btrfs_space_info *info;
3501         static const u64 types[] = {
3502                 BTRFS_BLOCK_GROUP_DATA,
3503                 BTRFS_BLOCK_GROUP_SYSTEM,
3504                 BTRFS_BLOCK_GROUP_METADATA,
3505                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3506         };
3507         int num_types = 4;
3508         int alloc_size;
3509         int ret = 0;
3510         u64 slot_count = 0;
3511         int i, c;
3512
3513         if (copy_from_user(&space_args,
3514                            (struct btrfs_ioctl_space_args __user *)arg,
3515                            sizeof(space_args)))
3516                 return -EFAULT;
3517
3518         for (i = 0; i < num_types; i++) {
3519                 struct btrfs_space_info *tmp;
3520
3521                 info = NULL;
3522                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3523                         if (tmp->flags == types[i]) {
3524                                 info = tmp;
3525                                 break;
3526                         }
3527                 }
3528
3529                 if (!info)
3530                         continue;
3531
3532                 down_read(&info->groups_sem);
3533                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3534                         if (!list_empty(&info->block_groups[c]))
3535                                 slot_count++;
3536                 }
3537                 up_read(&info->groups_sem);
3538         }
3539
3540         /*
3541          * Global block reserve, exported as a space_info
3542          */
3543         slot_count++;
3544
3545         /* space_slots == 0 means they are asking for a count */
3546         if (space_args.space_slots == 0) {
3547                 space_args.total_spaces = slot_count;
3548                 goto out;
3549         }
3550
3551         slot_count = min_t(u64, space_args.space_slots, slot_count);
3552
3553         alloc_size = sizeof(*dest) * slot_count;
3554
3555         /* we generally have at most 6 or so space infos, one for each raid
3556          * level.  So, a whole page should be more than enough for everyone
3557          */
3558         if (alloc_size > PAGE_SIZE)
3559                 return -ENOMEM;
3560
3561         space_args.total_spaces = 0;
3562         dest = kmalloc(alloc_size, GFP_KERNEL);
3563         if (!dest)
3564                 return -ENOMEM;
3565         dest_orig = dest;
3566
3567         /* now we have a buffer to copy into */
3568         for (i = 0; i < num_types; i++) {
3569                 struct btrfs_space_info *tmp;
3570
3571                 if (!slot_count)
3572                         break;
3573
3574                 info = NULL;
3575                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3576                         if (tmp->flags == types[i]) {
3577                                 info = tmp;
3578                                 break;
3579                         }
3580                 }
3581
3582                 if (!info)
3583                         continue;
3584                 down_read(&info->groups_sem);
3585                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3586                         if (!list_empty(&info->block_groups[c])) {
3587                                 get_block_group_info(&info->block_groups[c],
3588                                                      &space);
3589                                 memcpy(dest, &space, sizeof(space));
3590                                 dest++;
3591                                 space_args.total_spaces++;
3592                                 slot_count--;
3593                         }
3594                         if (!slot_count)
3595                                 break;
3596                 }
3597                 up_read(&info->groups_sem);
3598         }
3599
3600         /*
3601          * Add global block reserve
3602          */
3603         if (slot_count) {
3604                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3605
3606                 spin_lock(&block_rsv->lock);
3607                 space.total_bytes = block_rsv->size;
3608                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3609                 spin_unlock(&block_rsv->lock);
3610                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3611                 memcpy(dest, &space, sizeof(space));
3612                 space_args.total_spaces++;
3613         }
3614
3615         user_dest = (struct btrfs_ioctl_space_info __user *)
3616                 (arg + sizeof(struct btrfs_ioctl_space_args));
3617
3618         if (copy_to_user(user_dest, dest_orig, alloc_size))
3619                 ret = -EFAULT;
3620
3621         kfree(dest_orig);
3622 out:
3623         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3624                 ret = -EFAULT;
3625
3626         return ret;
3627 }
3628
3629 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3630                                             void __user *argp)
3631 {
3632         struct btrfs_trans_handle *trans;
3633         u64 transid;
3634         int ret;
3635
3636         trans = btrfs_attach_transaction_barrier(root);
3637         if (IS_ERR(trans)) {
3638                 if (PTR_ERR(trans) != -ENOENT)
3639                         return PTR_ERR(trans);
3640
3641                 /* No running transaction, don't bother */
3642                 transid = root->fs_info->last_trans_committed;
3643                 goto out;
3644         }
3645         transid = trans->transid;
3646         ret = btrfs_commit_transaction_async(trans, 0);
3647         if (ret) {
3648                 btrfs_end_transaction(trans);
3649                 return ret;
3650         }
3651 out:
3652         if (argp)
3653                 if (copy_to_user(argp, &transid, sizeof(transid)))
3654                         return -EFAULT;
3655         return 0;
3656 }
3657
3658 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3659                                            void __user *argp)
3660 {
3661         u64 transid;
3662
3663         if (argp) {
3664                 if (copy_from_user(&transid, argp, sizeof(transid)))
3665                         return -EFAULT;
3666         } else {
3667                 transid = 0;  /* current trans */
3668         }
3669         return btrfs_wait_for_commit(fs_info, transid);
3670 }
3671
3672 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3673 {
3674         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3675         struct btrfs_ioctl_scrub_args *sa;
3676         int ret;
3677
3678         if (!capable(CAP_SYS_ADMIN))
3679                 return -EPERM;
3680
3681         sa = memdup_user(arg, sizeof(*sa));
3682         if (IS_ERR(sa))
3683                 return PTR_ERR(sa);
3684
3685         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3686                 ret = mnt_want_write_file(file);
3687                 if (ret)
3688                         goto out;
3689         }
3690
3691         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3692                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3693                               0);
3694
3695         /*
3696          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3697          * error. This is important as it allows user space to know how much
3698          * progress scrub has done. For example, if scrub is canceled we get
3699          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3700          * space. Later user space can inspect the progress from the structure
3701          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3702          * previously (btrfs-progs does this).
3703          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3704          * then return -EFAULT to signal the structure was not copied or it may
3705          * be corrupt and unreliable due to a partial copy.
3706          */
3707         if (copy_to_user(arg, sa, sizeof(*sa)))
3708                 ret = -EFAULT;
3709
3710         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3711                 mnt_drop_write_file(file);
3712 out:
3713         kfree(sa);
3714         return ret;
3715 }
3716
3717 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3718 {
3719         if (!capable(CAP_SYS_ADMIN))
3720                 return -EPERM;
3721
3722         return btrfs_scrub_cancel(fs_info);
3723 }
3724
3725 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3726                                        void __user *arg)
3727 {
3728         struct btrfs_ioctl_scrub_args *sa;
3729         int ret;
3730
3731         if (!capable(CAP_SYS_ADMIN))
3732                 return -EPERM;
3733
3734         sa = memdup_user(arg, sizeof(*sa));
3735         if (IS_ERR(sa))
3736                 return PTR_ERR(sa);
3737
3738         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3739
3740         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3741                 ret = -EFAULT;
3742
3743         kfree(sa);
3744         return ret;
3745 }
3746
3747 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3748                                       void __user *arg)
3749 {
3750         struct btrfs_ioctl_get_dev_stats *sa;
3751         int ret;
3752
3753         sa = memdup_user(arg, sizeof(*sa));
3754         if (IS_ERR(sa))
3755                 return PTR_ERR(sa);
3756
3757         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3758                 kfree(sa);
3759                 return -EPERM;
3760         }
3761
3762         ret = btrfs_get_dev_stats(fs_info, sa);
3763
3764         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3765                 ret = -EFAULT;
3766
3767         kfree(sa);
3768         return ret;
3769 }
3770
3771 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3772                                     void __user *arg)
3773 {
3774         struct btrfs_ioctl_dev_replace_args *p;
3775         int ret;
3776
3777         if (!capable(CAP_SYS_ADMIN))
3778                 return -EPERM;
3779
3780         p = memdup_user(arg, sizeof(*p));
3781         if (IS_ERR(p))
3782                 return PTR_ERR(p);
3783
3784         switch (p->cmd) {
3785         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3786                 if (sb_rdonly(fs_info->sb)) {
3787                         ret = -EROFS;
3788                         goto out;
3789                 }
3790                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3791                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3792                 } else {
3793                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3794                         btrfs_exclop_finish(fs_info);
3795                 }
3796                 break;
3797         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3798                 btrfs_dev_replace_status(fs_info, p);
3799                 ret = 0;
3800                 break;
3801         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3802                 p->result = btrfs_dev_replace_cancel(fs_info);
3803                 ret = 0;
3804                 break;
3805         default:
3806                 ret = -EINVAL;
3807                 break;
3808         }
3809
3810         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3811                 ret = -EFAULT;
3812 out:
3813         kfree(p);
3814         return ret;
3815 }
3816
3817 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3818 {
3819         int ret = 0;
3820         int i;
3821         u64 rel_ptr;
3822         int size;
3823         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3824         struct inode_fs_paths *ipath = NULL;
3825         struct btrfs_path *path;
3826
3827         if (!capable(CAP_DAC_READ_SEARCH))
3828                 return -EPERM;
3829
3830         path = btrfs_alloc_path();
3831         if (!path) {
3832                 ret = -ENOMEM;
3833                 goto out;
3834         }
3835
3836         ipa = memdup_user(arg, sizeof(*ipa));
3837         if (IS_ERR(ipa)) {
3838                 ret = PTR_ERR(ipa);
3839                 ipa = NULL;
3840                 goto out;
3841         }
3842
3843         size = min_t(u32, ipa->size, 4096);
3844         ipath = init_ipath(size, root, path);
3845         if (IS_ERR(ipath)) {
3846                 ret = PTR_ERR(ipath);
3847                 ipath = NULL;
3848                 goto out;
3849         }
3850
3851         ret = paths_from_inode(ipa->inum, ipath);
3852         if (ret < 0)
3853                 goto out;
3854
3855         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3856                 rel_ptr = ipath->fspath->val[i] -
3857                           (u64)(unsigned long)ipath->fspath->val;
3858                 ipath->fspath->val[i] = rel_ptr;
3859         }
3860
3861         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3862                            ipath->fspath, size);
3863         if (ret) {
3864                 ret = -EFAULT;
3865                 goto out;
3866         }
3867
3868 out:
3869         btrfs_free_path(path);
3870         free_ipath(ipath);
3871         kfree(ipa);
3872
3873         return ret;
3874 }
3875
3876 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3877 {
3878         struct btrfs_data_container *inodes = ctx;
3879         const size_t c = 3 * sizeof(u64);
3880
3881         if (inodes->bytes_left >= c) {
3882                 inodes->bytes_left -= c;
3883                 inodes->val[inodes->elem_cnt] = inum;
3884                 inodes->val[inodes->elem_cnt + 1] = offset;
3885                 inodes->val[inodes->elem_cnt + 2] = root;
3886                 inodes->elem_cnt += 3;
3887         } else {
3888                 inodes->bytes_missing += c - inodes->bytes_left;
3889                 inodes->bytes_left = 0;
3890                 inodes->elem_missed += 3;
3891         }
3892
3893         return 0;
3894 }
3895
3896 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3897                                         void __user *arg, int version)
3898 {
3899         int ret = 0;
3900         int size;
3901         struct btrfs_ioctl_logical_ino_args *loi;
3902         struct btrfs_data_container *inodes = NULL;
3903         struct btrfs_path *path = NULL;
3904         bool ignore_offset;
3905
3906         if (!capable(CAP_SYS_ADMIN))
3907                 return -EPERM;
3908
3909         loi = memdup_user(arg, sizeof(*loi));
3910         if (IS_ERR(loi))
3911                 return PTR_ERR(loi);
3912
3913         if (version == 1) {
3914                 ignore_offset = false;
3915                 size = min_t(u32, loi->size, SZ_64K);
3916         } else {
3917                 /* All reserved bits must be 0 for now */
3918                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3919                         ret = -EINVAL;
3920                         goto out_loi;
3921                 }
3922                 /* Only accept flags we have defined so far */
3923                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3924                         ret = -EINVAL;
3925                         goto out_loi;
3926                 }
3927                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3928                 size = min_t(u32, loi->size, SZ_16M);
3929         }
3930
3931         path = btrfs_alloc_path();
3932         if (!path) {
3933                 ret = -ENOMEM;
3934                 goto out;
3935         }
3936
3937         inodes = init_data_container(size);
3938         if (IS_ERR(inodes)) {
3939                 ret = PTR_ERR(inodes);
3940                 inodes = NULL;
3941                 goto out;
3942         }
3943
3944         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3945                                           build_ino_list, inodes, ignore_offset);
3946         if (ret == -EINVAL)
3947                 ret = -ENOENT;
3948         if (ret < 0)
3949                 goto out;
3950
3951         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3952                            size);
3953         if (ret)
3954                 ret = -EFAULT;
3955
3956 out:
3957         btrfs_free_path(path);
3958         kvfree(inodes);
3959 out_loi:
3960         kfree(loi);
3961
3962         return ret;
3963 }
3964
3965 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3966                                struct btrfs_ioctl_balance_args *bargs)
3967 {
3968         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3969
3970         bargs->flags = bctl->flags;
3971
3972         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3973                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3974         if (atomic_read(&fs_info->balance_pause_req))
3975                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3976         if (atomic_read(&fs_info->balance_cancel_req))
3977                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3978
3979         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3980         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3981         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3982
3983         spin_lock(&fs_info->balance_lock);
3984         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3985         spin_unlock(&fs_info->balance_lock);
3986 }
3987
3988 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3989 {
3990         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3991         struct btrfs_fs_info *fs_info = root->fs_info;
3992         struct btrfs_ioctl_balance_args *bargs;
3993         struct btrfs_balance_control *bctl;
3994         bool need_unlock; /* for mut. excl. ops lock */
3995         int ret;
3996
3997         if (!capable(CAP_SYS_ADMIN))
3998                 return -EPERM;
3999
4000         ret = mnt_want_write_file(file);
4001         if (ret)
4002                 return ret;
4003
4004 again:
4005         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4006                 mutex_lock(&fs_info->balance_mutex);
4007                 need_unlock = true;
4008                 goto locked;
4009         }
4010
4011         /*
4012          * mut. excl. ops lock is locked.  Three possibilities:
4013          *   (1) some other op is running
4014          *   (2) balance is running
4015          *   (3) balance is paused -- special case (think resume)
4016          */
4017         mutex_lock(&fs_info->balance_mutex);
4018         if (fs_info->balance_ctl) {
4019                 /* this is either (2) or (3) */
4020                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4021                         mutex_unlock(&fs_info->balance_mutex);
4022                         /*
4023                          * Lock released to allow other waiters to continue,
4024                          * we'll reexamine the status again.
4025                          */
4026                         mutex_lock(&fs_info->balance_mutex);
4027
4028                         if (fs_info->balance_ctl &&
4029                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4030                                 /* this is (3) */
4031                                 need_unlock = false;
4032                                 goto locked;
4033                         }
4034
4035                         mutex_unlock(&fs_info->balance_mutex);
4036                         goto again;
4037                 } else {
4038                         /* this is (2) */
4039                         mutex_unlock(&fs_info->balance_mutex);
4040                         ret = -EINPROGRESS;
4041                         goto out;
4042                 }
4043         } else {
4044                 /* this is (1) */
4045                 mutex_unlock(&fs_info->balance_mutex);
4046                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4047                 goto out;
4048         }
4049
4050 locked:
4051
4052         if (arg) {
4053                 bargs = memdup_user(arg, sizeof(*bargs));
4054                 if (IS_ERR(bargs)) {
4055                         ret = PTR_ERR(bargs);
4056                         goto out_unlock;
4057                 }
4058
4059                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4060                         if (!fs_info->balance_ctl) {
4061                                 ret = -ENOTCONN;
4062                                 goto out_bargs;
4063                         }
4064
4065                         bctl = fs_info->balance_ctl;
4066                         spin_lock(&fs_info->balance_lock);
4067                         bctl->flags |= BTRFS_BALANCE_RESUME;
4068                         spin_unlock(&fs_info->balance_lock);
4069
4070                         goto do_balance;
4071                 }
4072         } else {
4073                 bargs = NULL;
4074         }
4075
4076         if (fs_info->balance_ctl) {
4077                 ret = -EINPROGRESS;
4078                 goto out_bargs;
4079         }
4080
4081         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4082         if (!bctl) {
4083                 ret = -ENOMEM;
4084                 goto out_bargs;
4085         }
4086
4087         if (arg) {
4088                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4089                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4090                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4091
4092                 bctl->flags = bargs->flags;
4093         } else {
4094                 /* balance everything - no filters */
4095                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4096         }
4097
4098         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4099                 ret = -EINVAL;
4100                 goto out_bctl;
4101         }
4102
4103 do_balance:
4104         /*
4105          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4106          * bctl is freed in reset_balance_state, or, if restriper was paused
4107          * all the way until unmount, in free_fs_info.  The flag should be
4108          * cleared after reset_balance_state.
4109          */
4110         need_unlock = false;
4111
4112         ret = btrfs_balance(fs_info, bctl, bargs);
4113         bctl = NULL;
4114
4115         if ((ret == 0 || ret == -ECANCELED) && arg) {
4116                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4117                         ret = -EFAULT;
4118         }
4119
4120 out_bctl:
4121         kfree(bctl);
4122 out_bargs:
4123         kfree(bargs);
4124 out_unlock:
4125         mutex_unlock(&fs_info->balance_mutex);
4126         if (need_unlock)
4127                 btrfs_exclop_finish(fs_info);
4128 out:
4129         mnt_drop_write_file(file);
4130         return ret;
4131 }
4132
4133 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4134 {
4135         if (!capable(CAP_SYS_ADMIN))
4136                 return -EPERM;
4137
4138         switch (cmd) {
4139         case BTRFS_BALANCE_CTL_PAUSE:
4140                 return btrfs_pause_balance(fs_info);
4141         case BTRFS_BALANCE_CTL_CANCEL:
4142                 return btrfs_cancel_balance(fs_info);
4143         }
4144
4145         return -EINVAL;
4146 }
4147
4148 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4149                                          void __user *arg)
4150 {
4151         struct btrfs_ioctl_balance_args *bargs;
4152         int ret = 0;
4153
4154         if (!capable(CAP_SYS_ADMIN))
4155                 return -EPERM;
4156
4157         mutex_lock(&fs_info->balance_mutex);
4158         if (!fs_info->balance_ctl) {
4159                 ret = -ENOTCONN;
4160                 goto out;
4161         }
4162
4163         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4164         if (!bargs) {
4165                 ret = -ENOMEM;
4166                 goto out;
4167         }
4168
4169         btrfs_update_ioctl_balance_args(fs_info, bargs);
4170
4171         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4172                 ret = -EFAULT;
4173
4174         kfree(bargs);
4175 out:
4176         mutex_unlock(&fs_info->balance_mutex);
4177         return ret;
4178 }
4179
4180 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4181 {
4182         struct inode *inode = file_inode(file);
4183         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4184         struct btrfs_ioctl_quota_ctl_args *sa;
4185         int ret;
4186
4187         if (!capable(CAP_SYS_ADMIN))
4188                 return -EPERM;
4189
4190         ret = mnt_want_write_file(file);
4191         if (ret)
4192                 return ret;
4193
4194         sa = memdup_user(arg, sizeof(*sa));
4195         if (IS_ERR(sa)) {
4196                 ret = PTR_ERR(sa);
4197                 goto drop_write;
4198         }
4199
4200         down_write(&fs_info->subvol_sem);
4201
4202         switch (sa->cmd) {
4203         case BTRFS_QUOTA_CTL_ENABLE:
4204                 ret = btrfs_quota_enable(fs_info);
4205                 break;
4206         case BTRFS_QUOTA_CTL_DISABLE:
4207                 ret = btrfs_quota_disable(fs_info);
4208                 break;
4209         default:
4210                 ret = -EINVAL;
4211                 break;
4212         }
4213
4214         kfree(sa);
4215         up_write(&fs_info->subvol_sem);
4216 drop_write:
4217         mnt_drop_write_file(file);
4218         return ret;
4219 }
4220
4221 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4222 {
4223         struct inode *inode = file_inode(file);
4224         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4225         struct btrfs_root *root = BTRFS_I(inode)->root;
4226         struct btrfs_ioctl_qgroup_assign_args *sa;
4227         struct btrfs_trans_handle *trans;
4228         int ret;
4229         int err;
4230
4231         if (!capable(CAP_SYS_ADMIN))
4232                 return -EPERM;
4233
4234         ret = mnt_want_write_file(file);
4235         if (ret)
4236                 return ret;
4237
4238         sa = memdup_user(arg, sizeof(*sa));
4239         if (IS_ERR(sa)) {
4240                 ret = PTR_ERR(sa);
4241                 goto drop_write;
4242         }
4243
4244         trans = btrfs_join_transaction(root);
4245         if (IS_ERR(trans)) {
4246                 ret = PTR_ERR(trans);
4247                 goto out;
4248         }
4249
4250         if (sa->assign) {
4251                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4252         } else {
4253                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4254         }
4255
4256         /* update qgroup status and info */
4257         err = btrfs_run_qgroups(trans);
4258         if (err < 0)
4259                 btrfs_handle_fs_error(fs_info, err,
4260                                       "failed to update qgroup status and info");
4261         err = btrfs_end_transaction(trans);
4262         if (err && !ret)
4263                 ret = err;
4264
4265 out:
4266         kfree(sa);
4267 drop_write:
4268         mnt_drop_write_file(file);
4269         return ret;
4270 }
4271
4272 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4273 {
4274         struct inode *inode = file_inode(file);
4275         struct btrfs_root *root = BTRFS_I(inode)->root;
4276         struct btrfs_ioctl_qgroup_create_args *sa;
4277         struct btrfs_trans_handle *trans;
4278         int ret;
4279         int err;
4280
4281         if (!capable(CAP_SYS_ADMIN))
4282                 return -EPERM;
4283
4284         ret = mnt_want_write_file(file);
4285         if (ret)
4286                 return ret;
4287
4288         sa = memdup_user(arg, sizeof(*sa));
4289         if (IS_ERR(sa)) {
4290                 ret = PTR_ERR(sa);
4291                 goto drop_write;
4292         }
4293
4294         if (!sa->qgroupid) {
4295                 ret = -EINVAL;
4296                 goto out;
4297         }
4298
4299         trans = btrfs_join_transaction(root);
4300         if (IS_ERR(trans)) {
4301                 ret = PTR_ERR(trans);
4302                 goto out;
4303         }
4304
4305         if (sa->create) {
4306                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4307         } else {
4308                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4309         }
4310
4311         err = btrfs_end_transaction(trans);
4312         if (err && !ret)
4313                 ret = err;
4314
4315 out:
4316         kfree(sa);
4317 drop_write:
4318         mnt_drop_write_file(file);
4319         return ret;
4320 }
4321
4322 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4323 {
4324         struct inode *inode = file_inode(file);
4325         struct btrfs_root *root = BTRFS_I(inode)->root;
4326         struct btrfs_ioctl_qgroup_limit_args *sa;
4327         struct btrfs_trans_handle *trans;
4328         int ret;
4329         int err;
4330         u64 qgroupid;
4331
4332         if (!capable(CAP_SYS_ADMIN))
4333                 return -EPERM;
4334
4335         ret = mnt_want_write_file(file);
4336         if (ret)
4337                 return ret;
4338
4339         sa = memdup_user(arg, sizeof(*sa));
4340         if (IS_ERR(sa)) {
4341                 ret = PTR_ERR(sa);
4342                 goto drop_write;
4343         }
4344
4345         trans = btrfs_join_transaction(root);
4346         if (IS_ERR(trans)) {
4347                 ret = PTR_ERR(trans);
4348                 goto out;
4349         }
4350
4351         qgroupid = sa->qgroupid;
4352         if (!qgroupid) {
4353                 /* take the current subvol as qgroup */
4354                 qgroupid = root->root_key.objectid;
4355         }
4356
4357         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4358
4359         err = btrfs_end_transaction(trans);
4360         if (err && !ret)
4361                 ret = err;
4362
4363 out:
4364         kfree(sa);
4365 drop_write:
4366         mnt_drop_write_file(file);
4367         return ret;
4368 }
4369
4370 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4371 {
4372         struct inode *inode = file_inode(file);
4373         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4374         struct btrfs_ioctl_quota_rescan_args *qsa;
4375         int ret;
4376
4377         if (!capable(CAP_SYS_ADMIN))
4378                 return -EPERM;
4379
4380         ret = mnt_want_write_file(file);
4381         if (ret)
4382                 return ret;
4383
4384         qsa = memdup_user(arg, sizeof(*qsa));
4385         if (IS_ERR(qsa)) {
4386                 ret = PTR_ERR(qsa);
4387                 goto drop_write;
4388         }
4389
4390         if (qsa->flags) {
4391                 ret = -EINVAL;
4392                 goto out;
4393         }
4394
4395         ret = btrfs_qgroup_rescan(fs_info);
4396
4397 out:
4398         kfree(qsa);
4399 drop_write:
4400         mnt_drop_write_file(file);
4401         return ret;
4402 }
4403
4404 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4405                                                 void __user *arg)
4406 {
4407         struct btrfs_ioctl_quota_rescan_args *qsa;
4408         int ret = 0;
4409
4410         if (!capable(CAP_SYS_ADMIN))
4411                 return -EPERM;
4412
4413         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4414         if (!qsa)
4415                 return -ENOMEM;
4416
4417         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4418                 qsa->flags = 1;
4419                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4420         }
4421
4422         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4423                 ret = -EFAULT;
4424
4425         kfree(qsa);
4426         return ret;
4427 }
4428
4429 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4430                                                 void __user *arg)
4431 {
4432         if (!capable(CAP_SYS_ADMIN))
4433                 return -EPERM;
4434
4435         return btrfs_qgroup_wait_for_completion(fs_info, true);
4436 }
4437
4438 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4439                                             struct btrfs_ioctl_received_subvol_args *sa)
4440 {
4441         struct inode *inode = file_inode(file);
4442         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4443         struct btrfs_root *root = BTRFS_I(inode)->root;
4444         struct btrfs_root_item *root_item = &root->root_item;
4445         struct btrfs_trans_handle *trans;
4446         struct timespec64 ct = current_time(inode);
4447         int ret = 0;
4448         int received_uuid_changed;
4449
4450         if (!inode_owner_or_capable(inode))
4451                 return -EPERM;
4452
4453         ret = mnt_want_write_file(file);
4454         if (ret < 0)
4455                 return ret;
4456
4457         down_write(&fs_info->subvol_sem);
4458
4459         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4460                 ret = -EINVAL;
4461                 goto out;
4462         }
4463
4464         if (btrfs_root_readonly(root)) {
4465                 ret = -EROFS;
4466                 goto out;
4467         }
4468
4469         /*
4470          * 1 - root item
4471          * 2 - uuid items (received uuid + subvol uuid)
4472          */
4473         trans = btrfs_start_transaction(root, 3);
4474         if (IS_ERR(trans)) {
4475                 ret = PTR_ERR(trans);
4476                 trans = NULL;
4477                 goto out;
4478         }
4479
4480         sa->rtransid = trans->transid;
4481         sa->rtime.sec = ct.tv_sec;
4482         sa->rtime.nsec = ct.tv_nsec;
4483
4484         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4485                                        BTRFS_UUID_SIZE);
4486         if (received_uuid_changed &&
4487             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4488                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4489                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4490                                           root->root_key.objectid);
4491                 if (ret && ret != -ENOENT) {
4492                         btrfs_abort_transaction(trans, ret);
4493                         btrfs_end_transaction(trans);
4494                         goto out;
4495                 }
4496         }
4497         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4498         btrfs_set_root_stransid(root_item, sa->stransid);
4499         btrfs_set_root_rtransid(root_item, sa->rtransid);
4500         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4501         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4502         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4503         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4504
4505         ret = btrfs_update_root(trans, fs_info->tree_root,
4506                                 &root->root_key, &root->root_item);
4507         if (ret < 0) {
4508                 btrfs_end_transaction(trans);
4509                 goto out;
4510         }
4511         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4512                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4513                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4514                                           root->root_key.objectid);
4515                 if (ret < 0 && ret != -EEXIST) {
4516                         btrfs_abort_transaction(trans, ret);
4517                         btrfs_end_transaction(trans);
4518                         goto out;
4519                 }
4520         }
4521         ret = btrfs_commit_transaction(trans);
4522 out:
4523         up_write(&fs_info->subvol_sem);
4524         mnt_drop_write_file(file);
4525         return ret;
4526 }
4527
4528 #ifdef CONFIG_64BIT
4529 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4530                                                 void __user *arg)
4531 {
4532         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4533         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4534         int ret = 0;
4535
4536         args32 = memdup_user(arg, sizeof(*args32));
4537         if (IS_ERR(args32))
4538                 return PTR_ERR(args32);
4539
4540         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4541         if (!args64) {
4542                 ret = -ENOMEM;
4543                 goto out;
4544         }
4545
4546         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4547         args64->stransid = args32->stransid;
4548         args64->rtransid = args32->rtransid;
4549         args64->stime.sec = args32->stime.sec;
4550         args64->stime.nsec = args32->stime.nsec;
4551         args64->rtime.sec = args32->rtime.sec;
4552         args64->rtime.nsec = args32->rtime.nsec;
4553         args64->flags = args32->flags;
4554
4555         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4556         if (ret)
4557                 goto out;
4558
4559         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4560         args32->stransid = args64->stransid;
4561         args32->rtransid = args64->rtransid;
4562         args32->stime.sec = args64->stime.sec;
4563         args32->stime.nsec = args64->stime.nsec;
4564         args32->rtime.sec = args64->rtime.sec;
4565         args32->rtime.nsec = args64->rtime.nsec;
4566         args32->flags = args64->flags;
4567
4568         ret = copy_to_user(arg, args32, sizeof(*args32));
4569         if (ret)
4570                 ret = -EFAULT;
4571
4572 out:
4573         kfree(args32);
4574         kfree(args64);
4575         return ret;
4576 }
4577 #endif
4578
4579 static long btrfs_ioctl_set_received_subvol(struct file *file,
4580                                             void __user *arg)
4581 {
4582         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4583         int ret = 0;
4584
4585         sa = memdup_user(arg, sizeof(*sa));
4586         if (IS_ERR(sa))
4587                 return PTR_ERR(sa);
4588
4589         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4590
4591         if (ret)
4592                 goto out;
4593
4594         ret = copy_to_user(arg, sa, sizeof(*sa));
4595         if (ret)
4596                 ret = -EFAULT;
4597
4598 out:
4599         kfree(sa);
4600         return ret;
4601 }
4602
4603 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4604                                         void __user *arg)
4605 {
4606         size_t len;
4607         int ret;
4608         char label[BTRFS_LABEL_SIZE];
4609
4610         spin_lock(&fs_info->super_lock);
4611         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4612         spin_unlock(&fs_info->super_lock);
4613
4614         len = strnlen(label, BTRFS_LABEL_SIZE);
4615
4616         if (len == BTRFS_LABEL_SIZE) {
4617                 btrfs_warn(fs_info,
4618                            "label is too long, return the first %zu bytes",
4619                            --len);
4620         }
4621
4622         ret = copy_to_user(arg, label, len);
4623
4624         return ret ? -EFAULT : 0;
4625 }
4626
4627 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4628 {
4629         struct inode *inode = file_inode(file);
4630         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4631         struct btrfs_root *root = BTRFS_I(inode)->root;
4632         struct btrfs_super_block *super_block = fs_info->super_copy;
4633         struct btrfs_trans_handle *trans;
4634         char label[BTRFS_LABEL_SIZE];
4635         int ret;
4636
4637         if (!capable(CAP_SYS_ADMIN))
4638                 return -EPERM;
4639
4640         if (copy_from_user(label, arg, sizeof(label)))
4641                 return -EFAULT;
4642
4643         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4644                 btrfs_err(fs_info,
4645                           "unable to set label with more than %d bytes",
4646                           BTRFS_LABEL_SIZE - 1);
4647                 return -EINVAL;
4648         }
4649
4650         ret = mnt_want_write_file(file);
4651         if (ret)
4652                 return ret;
4653
4654         trans = btrfs_start_transaction(root, 0);
4655         if (IS_ERR(trans)) {
4656                 ret = PTR_ERR(trans);
4657                 goto out_unlock;
4658         }
4659
4660         spin_lock(&fs_info->super_lock);
4661         strcpy(super_block->label, label);
4662         spin_unlock(&fs_info->super_lock);
4663         ret = btrfs_commit_transaction(trans);
4664
4665 out_unlock:
4666         mnt_drop_write_file(file);
4667         return ret;
4668 }
4669
4670 #define INIT_FEATURE_FLAGS(suffix) \
4671         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4672           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4673           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4674
4675 int btrfs_ioctl_get_supported_features(void __user *arg)
4676 {
4677         static const struct btrfs_ioctl_feature_flags features[3] = {
4678                 INIT_FEATURE_FLAGS(SUPP),
4679                 INIT_FEATURE_FLAGS(SAFE_SET),
4680                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4681         };
4682
4683         if (copy_to_user(arg, &features, sizeof(features)))
4684                 return -EFAULT;
4685
4686         return 0;
4687 }
4688
4689 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4690                                         void __user *arg)
4691 {
4692         struct btrfs_super_block *super_block = fs_info->super_copy;
4693         struct btrfs_ioctl_feature_flags features;
4694
4695         features.compat_flags = btrfs_super_compat_flags(super_block);
4696         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4697         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4698
4699         if (copy_to_user(arg, &features, sizeof(features)))
4700                 return -EFAULT;
4701
4702         return 0;
4703 }
4704
4705 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4706                               enum btrfs_feature_set set,
4707                               u64 change_mask, u64 flags, u64 supported_flags,
4708                               u64 safe_set, u64 safe_clear)
4709 {
4710         const char *type = btrfs_feature_set_name(set);
4711         char *names;
4712         u64 disallowed, unsupported;
4713         u64 set_mask = flags & change_mask;
4714         u64 clear_mask = ~flags & change_mask;
4715
4716         unsupported = set_mask & ~supported_flags;
4717         if (unsupported) {
4718                 names = btrfs_printable_features(set, unsupported);
4719                 if (names) {
4720                         btrfs_warn(fs_info,
4721                                    "this kernel does not support the %s feature bit%s",
4722                                    names, strchr(names, ',') ? "s" : "");
4723                         kfree(names);
4724                 } else
4725                         btrfs_warn(fs_info,
4726                                    "this kernel does not support %s bits 0x%llx",
4727                                    type, unsupported);
4728                 return -EOPNOTSUPP;
4729         }
4730
4731         disallowed = set_mask & ~safe_set;
4732         if (disallowed) {
4733                 names = btrfs_printable_features(set, disallowed);
4734                 if (names) {
4735                         btrfs_warn(fs_info,
4736                                    "can't set the %s feature bit%s while mounted",
4737                                    names, strchr(names, ',') ? "s" : "");
4738                         kfree(names);
4739                 } else
4740                         btrfs_warn(fs_info,
4741                                    "can't set %s bits 0x%llx while mounted",
4742                                    type, disallowed);
4743                 return -EPERM;
4744         }
4745
4746         disallowed = clear_mask & ~safe_clear;
4747         if (disallowed) {
4748                 names = btrfs_printable_features(set, disallowed);
4749                 if (names) {
4750                         btrfs_warn(fs_info,
4751                                    "can't clear the %s feature bit%s while mounted",
4752                                    names, strchr(names, ',') ? "s" : "");
4753                         kfree(names);
4754                 } else
4755                         btrfs_warn(fs_info,
4756                                    "can't clear %s bits 0x%llx while mounted",
4757                                    type, disallowed);
4758                 return -EPERM;
4759         }
4760
4761         return 0;
4762 }
4763
4764 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4765 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4766                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4767                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4768                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4769
4770 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4771 {
4772         struct inode *inode = file_inode(file);
4773         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4774         struct btrfs_root *root = BTRFS_I(inode)->root;
4775         struct btrfs_super_block *super_block = fs_info->super_copy;
4776         struct btrfs_ioctl_feature_flags flags[2];
4777         struct btrfs_trans_handle *trans;
4778         u64 newflags;
4779         int ret;
4780
4781         if (!capable(CAP_SYS_ADMIN))
4782                 return -EPERM;
4783
4784         if (copy_from_user(flags, arg, sizeof(flags)))
4785                 return -EFAULT;
4786
4787         /* Nothing to do */
4788         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4789             !flags[0].incompat_flags)
4790                 return 0;
4791
4792         ret = check_feature(fs_info, flags[0].compat_flags,
4793                             flags[1].compat_flags, COMPAT);
4794         if (ret)
4795                 return ret;
4796
4797         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4798                             flags[1].compat_ro_flags, COMPAT_RO);
4799         if (ret)
4800                 return ret;
4801
4802         ret = check_feature(fs_info, flags[0].incompat_flags,
4803                             flags[1].incompat_flags, INCOMPAT);
4804         if (ret)
4805                 return ret;
4806
4807         ret = mnt_want_write_file(file);
4808         if (ret)
4809                 return ret;
4810
4811         trans = btrfs_start_transaction(root, 0);
4812         if (IS_ERR(trans)) {
4813                 ret = PTR_ERR(trans);
4814                 goto out_drop_write;
4815         }
4816
4817         spin_lock(&fs_info->super_lock);
4818         newflags = btrfs_super_compat_flags(super_block);
4819         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4820         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4821         btrfs_set_super_compat_flags(super_block, newflags);
4822
4823         newflags = btrfs_super_compat_ro_flags(super_block);
4824         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4825         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4826         btrfs_set_super_compat_ro_flags(super_block, newflags);
4827
4828         newflags = btrfs_super_incompat_flags(super_block);
4829         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4830         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4831         btrfs_set_super_incompat_flags(super_block, newflags);
4832         spin_unlock(&fs_info->super_lock);
4833
4834         ret = btrfs_commit_transaction(trans);
4835 out_drop_write:
4836         mnt_drop_write_file(file);
4837
4838         return ret;
4839 }
4840
4841 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4842 {
4843         struct btrfs_ioctl_send_args *arg;
4844         int ret;
4845
4846         if (compat) {
4847 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4848                 struct btrfs_ioctl_send_args_32 args32;
4849
4850                 ret = copy_from_user(&args32, argp, sizeof(args32));
4851                 if (ret)
4852                         return -EFAULT;
4853                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4854                 if (!arg)
4855                         return -ENOMEM;
4856                 arg->send_fd = args32.send_fd;
4857                 arg->clone_sources_count = args32.clone_sources_count;
4858                 arg->clone_sources = compat_ptr(args32.clone_sources);
4859                 arg->parent_root = args32.parent_root;
4860                 arg->flags = args32.flags;
4861                 memcpy(arg->reserved, args32.reserved,
4862                        sizeof(args32.reserved));
4863 #else
4864                 return -ENOTTY;
4865 #endif
4866         } else {
4867                 arg = memdup_user(argp, sizeof(*arg));
4868                 if (IS_ERR(arg))
4869                         return PTR_ERR(arg);
4870         }
4871         ret = btrfs_ioctl_send(file, arg);
4872         kfree(arg);
4873         return ret;
4874 }
4875
4876 long btrfs_ioctl(struct file *file, unsigned int
4877                 cmd, unsigned long arg)
4878 {
4879         struct inode *inode = file_inode(file);
4880         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4881         struct btrfs_root *root = BTRFS_I(inode)->root;
4882         void __user *argp = (void __user *)arg;
4883
4884         switch (cmd) {
4885         case FS_IOC_GETFLAGS:
4886                 return btrfs_ioctl_getflags(file, argp);
4887         case FS_IOC_SETFLAGS:
4888                 return btrfs_ioctl_setflags(file, argp);
4889         case FS_IOC_GETVERSION:
4890                 return btrfs_ioctl_getversion(file, argp);
4891         case FS_IOC_GETFSLABEL:
4892                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4893         case FS_IOC_SETFSLABEL:
4894                 return btrfs_ioctl_set_fslabel(file, argp);
4895         case FITRIM:
4896                 return btrfs_ioctl_fitrim(fs_info, argp);
4897         case BTRFS_IOC_SNAP_CREATE:
4898                 return btrfs_ioctl_snap_create(file, argp, 0);
4899         case BTRFS_IOC_SNAP_CREATE_V2:
4900                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4901         case BTRFS_IOC_SUBVOL_CREATE:
4902                 return btrfs_ioctl_snap_create(file, argp, 1);
4903         case BTRFS_IOC_SUBVOL_CREATE_V2:
4904                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4905         case BTRFS_IOC_SNAP_DESTROY:
4906                 return btrfs_ioctl_snap_destroy(file, argp, false);
4907         case BTRFS_IOC_SNAP_DESTROY_V2:
4908                 return btrfs_ioctl_snap_destroy(file, argp, true);
4909         case BTRFS_IOC_SUBVOL_GETFLAGS:
4910                 return btrfs_ioctl_subvol_getflags(file, argp);
4911         case BTRFS_IOC_SUBVOL_SETFLAGS:
4912                 return btrfs_ioctl_subvol_setflags(file, argp);
4913         case BTRFS_IOC_DEFAULT_SUBVOL:
4914                 return btrfs_ioctl_default_subvol(file, argp);
4915         case BTRFS_IOC_DEFRAG:
4916                 return btrfs_ioctl_defrag(file, NULL);
4917         case BTRFS_IOC_DEFRAG_RANGE:
4918                 return btrfs_ioctl_defrag(file, argp);
4919         case BTRFS_IOC_RESIZE:
4920                 return btrfs_ioctl_resize(file, argp);
4921         case BTRFS_IOC_ADD_DEV:
4922                 return btrfs_ioctl_add_dev(fs_info, argp);
4923         case BTRFS_IOC_RM_DEV:
4924                 return btrfs_ioctl_rm_dev(file, argp);
4925         case BTRFS_IOC_RM_DEV_V2:
4926                 return btrfs_ioctl_rm_dev_v2(file, argp);
4927         case BTRFS_IOC_FS_INFO:
4928                 return btrfs_ioctl_fs_info(fs_info, argp);
4929         case BTRFS_IOC_DEV_INFO:
4930                 return btrfs_ioctl_dev_info(fs_info, argp);
4931         case BTRFS_IOC_BALANCE:
4932                 return btrfs_ioctl_balance(file, NULL);
4933         case BTRFS_IOC_TREE_SEARCH:
4934                 return btrfs_ioctl_tree_search(file, argp);
4935         case BTRFS_IOC_TREE_SEARCH_V2:
4936                 return btrfs_ioctl_tree_search_v2(file, argp);
4937         case BTRFS_IOC_INO_LOOKUP:
4938                 return btrfs_ioctl_ino_lookup(file, argp);
4939         case BTRFS_IOC_INO_PATHS:
4940                 return btrfs_ioctl_ino_to_path(root, argp);
4941         case BTRFS_IOC_LOGICAL_INO:
4942                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4943         case BTRFS_IOC_LOGICAL_INO_V2:
4944                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4945         case BTRFS_IOC_SPACE_INFO:
4946                 return btrfs_ioctl_space_info(fs_info, argp);
4947         case BTRFS_IOC_SYNC: {
4948                 int ret;
4949
4950                 ret = btrfs_start_delalloc_roots(fs_info, U64_MAX, false);
4951                 if (ret)
4952                         return ret;
4953                 ret = btrfs_sync_fs(inode->i_sb, 1);
4954                 /*
4955                  * The transaction thread may want to do more work,
4956                  * namely it pokes the cleaner kthread that will start
4957                  * processing uncleaned subvols.
4958                  */
4959                 wake_up_process(fs_info->transaction_kthread);
4960                 return ret;
4961         }
4962         case BTRFS_IOC_START_SYNC:
4963                 return btrfs_ioctl_start_sync(root, argp);
4964         case BTRFS_IOC_WAIT_SYNC:
4965                 return btrfs_ioctl_wait_sync(fs_info, argp);
4966         case BTRFS_IOC_SCRUB:
4967                 return btrfs_ioctl_scrub(file, argp);
4968         case BTRFS_IOC_SCRUB_CANCEL:
4969                 return btrfs_ioctl_scrub_cancel(fs_info);
4970         case BTRFS_IOC_SCRUB_PROGRESS:
4971                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4972         case BTRFS_IOC_BALANCE_V2:
4973                 return btrfs_ioctl_balance(file, argp);
4974         case BTRFS_IOC_BALANCE_CTL:
4975                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4976         case BTRFS_IOC_BALANCE_PROGRESS:
4977                 return btrfs_ioctl_balance_progress(fs_info, argp);
4978         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4979                 return btrfs_ioctl_set_received_subvol(file, argp);
4980 #ifdef CONFIG_64BIT
4981         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4982                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4983 #endif
4984         case BTRFS_IOC_SEND:
4985                 return _btrfs_ioctl_send(file, argp, false);
4986 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4987         case BTRFS_IOC_SEND_32:
4988                 return _btrfs_ioctl_send(file, argp, true);
4989 #endif
4990         case BTRFS_IOC_GET_DEV_STATS:
4991                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4992         case BTRFS_IOC_QUOTA_CTL:
4993                 return btrfs_ioctl_quota_ctl(file, argp);
4994         case BTRFS_IOC_QGROUP_ASSIGN:
4995                 return btrfs_ioctl_qgroup_assign(file, argp);
4996         case BTRFS_IOC_QGROUP_CREATE:
4997                 return btrfs_ioctl_qgroup_create(file, argp);
4998         case BTRFS_IOC_QGROUP_LIMIT:
4999                 return btrfs_ioctl_qgroup_limit(file, argp);
5000         case BTRFS_IOC_QUOTA_RESCAN:
5001                 return btrfs_ioctl_quota_rescan(file, argp);
5002         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5003                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5004         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5005                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5006         case BTRFS_IOC_DEV_REPLACE:
5007                 return btrfs_ioctl_dev_replace(fs_info, argp);
5008         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5009                 return btrfs_ioctl_get_supported_features(argp);
5010         case BTRFS_IOC_GET_FEATURES:
5011                 return btrfs_ioctl_get_features(fs_info, argp);
5012         case BTRFS_IOC_SET_FEATURES:
5013                 return btrfs_ioctl_set_features(file, argp);
5014         case FS_IOC_FSGETXATTR:
5015                 return btrfs_ioctl_fsgetxattr(file, argp);
5016         case FS_IOC_FSSETXATTR:
5017                 return btrfs_ioctl_fssetxattr(file, argp);
5018         case BTRFS_IOC_GET_SUBVOL_INFO:
5019                 return btrfs_ioctl_get_subvol_info(file, argp);
5020         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5021                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5022         case BTRFS_IOC_INO_LOOKUP_USER:
5023                 return btrfs_ioctl_ino_lookup_user(file, argp);
5024         }
5025
5026         return -ENOTTY;
5027 }
5028
5029 #ifdef CONFIG_COMPAT
5030 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5031 {
5032         /*
5033          * These all access 32-bit values anyway so no further
5034          * handling is necessary.
5035          */
5036         switch (cmd) {
5037         case FS_IOC32_GETFLAGS:
5038                 cmd = FS_IOC_GETFLAGS;
5039                 break;
5040         case FS_IOC32_SETFLAGS:
5041                 cmd = FS_IOC_SETFLAGS;
5042                 break;
5043         case FS_IOC32_GETVERSION:
5044                 cmd = FS_IOC_GETVERSION;
5045                 break;
5046         }
5047
5048         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5049 }
5050 #endif