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_find_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         mutex_lock(&new_root->objectid_mutex);
 728         new_root->highest_objectid = new_dirid;
 729         mutex_unlock(&new_root->objectid_mutex);
 730
 731         /*
 732          * insert the directory item
 733          */
 734         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 735         if (ret) {
 736                 btrfs_abort_transaction(trans, ret);
 737                 goto fail;
 738         }
 739
 740         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 741                                     BTRFS_FT_DIR, index);
 742         if (ret) {
 743                 btrfs_abort_transaction(trans, ret);
 744                 goto fail;
 745         }
 746
 747         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 748         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 749         if (ret) {
 750                 btrfs_abort_transaction(trans, ret);
 751                 goto fail;
 752         }
 753
 754         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 755                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 756         if (ret) {
 757                 btrfs_abort_transaction(trans, ret);
 758                 goto fail;
 759         }
 760
 761         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 762                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 763         if (ret)
 764                 btrfs_abort_transaction(trans, ret);
 765
 766 fail:
 767         kfree(root_item);
 768         trans->block_rsv = NULL;
 769         trans->bytes_reserved = 0;
 770         btrfs_subvolume_release_metadata(root, &block_rsv);
 771
 772         err = btrfs_commit_transaction(trans);
 773         if (err && !ret)
 774                 ret = err;
 775
 776         if (!ret) {
 777                 inode = btrfs_lookup_dentry(dir, dentry);
 778                 if (IS_ERR(inode))
 779                         return PTR_ERR(inode);
 780                 d_instantiate(dentry, inode);
 781         }
 782         return ret;
 783
 784 fail_free:
 785         if (anon_dev)
 786                 free_anon_bdev(anon_dev);
 787         kfree(root_item);
 788         return ret;
 789 }
 790
 791 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 792                            struct dentry *dentry, bool readonly,
 793                            struct btrfs_qgroup_inherit *inherit)
 794 {
 795         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 796         struct inode *inode;
 797         struct btrfs_pending_snapshot *pending_snapshot;
 798         struct btrfs_trans_handle *trans;
 799         int ret;
 800
 801         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 802                 return -EINVAL;
 803
 804         if (atomic_read(&root->nr_swapfiles)) {
 805                 btrfs_warn(fs_info,
 806                            "cannot snapshot subvolume with active swapfile");
 807                 return -ETXTBSY;
 808         }
 809
 810         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 811         if (!pending_snapshot)
 812                 return -ENOMEM;
 813
 814         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 815         if (ret < 0)
 816                 goto free_pending;
 817         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 818                         GFP_KERNEL);
 819         pending_snapshot->path = btrfs_alloc_path();
 820         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 821                 ret = -ENOMEM;
 822                 goto free_pending;
 823         }
 824
 825         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 826                              BTRFS_BLOCK_RSV_TEMP);
 827         /*
 828          * 1 - parent dir inode
 829          * 2 - dir entries
 830          * 1 - root item
 831          * 2 - root ref/backref
 832          * 1 - root of snapshot
 833          * 1 - UUID item
 834          */
 835         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 836                                         &pending_snapshot->block_rsv, 8,
 837                                         false);
 838         if (ret)
 839                 goto free_pending;
 840
 841         pending_snapshot->dentry = dentry;
 842         pending_snapshot->root = root;
 843         pending_snapshot->readonly = readonly;
 844         pending_snapshot->dir = dir;
 845         pending_snapshot->inherit = inherit;
 846
 847         trans = btrfs_start_transaction(root, 0);
 848         if (IS_ERR(trans)) {
 849                 ret = PTR_ERR(trans);
 850                 goto fail;
 851         }
 852
 853         spin_lock(&fs_info->trans_lock);
 854         list_add(&pending_snapshot->list,
 855                  &trans->transaction->pending_snapshots);
 856         spin_unlock(&fs_info->trans_lock);
 857
 858         ret = btrfs_commit_transaction(trans);
 859         if (ret)
 860                 goto fail;
 861
 862         ret = pending_snapshot->error;
 863         if (ret)
 864                 goto fail;
 865
 866         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 867         if (ret)
 868                 goto fail;
 869
 870         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 871         if (IS_ERR(inode)) {
 872                 ret = PTR_ERR(inode);
 873                 goto fail;
 874         }
 875
 876         d_instantiate(dentry, inode);
 877         ret = 0;
 878         pending_snapshot->anon_dev = 0;
 879 fail:
 880         /* Prevent double freeing of anon_dev */
 881         if (ret && pending_snapshot->snap)
 882                 pending_snapshot->snap->anon_dev = 0;
 883         btrfs_put_root(pending_snapshot->snap);
 884         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 885 free_pending:
 886         if (pending_snapshot->anon_dev)
 887                 free_anon_bdev(pending_snapshot->anon_dev);
 888         kfree(pending_snapshot->root_item);
 889         btrfs_free_path(pending_snapshot->path);
 890         kfree(pending_snapshot);
 891
 892         return ret;
 893 }
 894
 895 /*  copy of may_delete in fs/namei.c()
 896  *      Check whether we can remove a link victim from directory dir, check
 897  *  whether the type of victim is right.
 898  *  1. We can't do it if dir is read-only (done in permission())
 899  *  2. We should have write and exec permissions on dir
 900  *  3. We can't remove anything from append-only dir
 901  *  4. We can't do anything with immutable dir (done in permission())
 902  *  5. If the sticky bit on dir is set we should either
 903  *      a. be owner of dir, or
 904  *      b. be owner of victim, or
 905  *      c. have CAP_FOWNER capability
 906  *  6. If the victim is append-only or immutable we can't do anything with
 907  *     links pointing to it.
 908  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 909  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 910  *  9. We can't remove a root or mountpoint.
 911  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 912  *     nfs_async_unlink().
 913  */
 914
 915 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 916 {
 917         int error;
 918
 919         if (d_really_is_negative(victim))
 920                 return -ENOENT;
 921
 922         BUG_ON(d_inode(victim->d_parent) != dir);
 923         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 924
 925         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 926         if (error)
 927                 return error;
 928         if (IS_APPEND(dir))
 929                 return -EPERM;
 930         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 931             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 932                 return -EPERM;
 933         if (isdir) {
 934                 if (!d_is_dir(victim))
 935                         return -ENOTDIR;
 936                 if (IS_ROOT(victim))
 937                         return -EBUSY;
 938         } else if (d_is_dir(victim))
 939                 return -EISDIR;
 940         if (IS_DEADDIR(dir))
 941                 return -ENOENT;
 942         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 943                 return -EBUSY;
 944         return 0;
 945 }
 946
 947 /* copy of may_create in fs/namei.c() */
 948 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 949 {
 950         if (d_really_is_positive(child))
 951                 return -EEXIST;
 952         if (IS_DEADDIR(dir))
 953                 return -ENOENT;
 954         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 955 }
 956
 957 /*
 958  * Create a new subvolume below @parent.  This is largely modeled after
 959  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 960  * inside this filesystem so it's quite a bit simpler.
 961  */
 962 static noinline int btrfs_mksubvol(const struct path *parent,
 963                                    const char *name, int namelen,
 964                                    struct btrfs_root *snap_src,
 965                                    bool readonly,
 966                                    struct btrfs_qgroup_inherit *inherit)
 967 {
 968         struct inode *dir = d_inode(parent->dentry);
 969         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 970         struct dentry *dentry;
 971         int error;
 972
 973         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 974         if (error == -EINTR)
 975                 return error;
 976
 977         dentry = lookup_one_len(name, parent->dentry, namelen);
 978         error = PTR_ERR(dentry);
 979         if (IS_ERR(dentry))
 980                 goto out_unlock;
 981
 982         error = btrfs_may_create(dir, dentry);
 983         if (error)
 984                 goto out_dput;
 985
 986         /*
 987          * even if this name doesn't exist, we may get hash collisions.
 988          * check for them now when we can safely fail
 989          */
 990         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 991                                                dir->i_ino, name,
 992                                                namelen);
 993         if (error)
 994                 goto out_dput;
 995
 996         down_read(&fs_info->subvol_sem);
 997
 998         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 999                 goto out_up_read;
1000
1001         if (snap_src)
1002                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
1003         else
1004                 error = create_subvol(dir, dentry, name, namelen, inherit);
1005
1006         if (!error)
1007                 fsnotify_mkdir(dir, dentry);
1008 out_up_read:
1009         up_read(&fs_info->subvol_sem);
1010 out_dput:
1011         dput(dentry);
1012 out_unlock:
1013         inode_unlock(dir);
1014         return error;
1015 }
1016
1017 static noinline int btrfs_mksnapshot(const struct path *parent,
1018                                    const char *name, int namelen,
1019                                    struct btrfs_root *root,
1020                                    bool readonly,
1021                                    struct btrfs_qgroup_inherit *inherit)
1022 {
1023         int ret;
1024         bool snapshot_force_cow = false;
1025
1026         /*
1027          * Force new buffered writes to reserve space even when NOCOW is
1028          * possible. This is to avoid later writeback (running dealloc) to
1029          * fallback to COW mode and unexpectedly fail with ENOSPC.
1030          */
1031         btrfs_drew_read_lock(&root->snapshot_lock);
1032
1033         ret = btrfs_start_delalloc_snapshot(root);
1034         if (ret)
1035                 goto out;
1036
1037         /*
1038          * All previous writes have started writeback in NOCOW mode, so now
1039          * we force future writes to fallback to COW mode during snapshot
1040          * creation.
1041          */
1042         atomic_inc(&root->snapshot_force_cow);
1043         snapshot_force_cow = true;
1044
1045         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1046
1047         ret = btrfs_mksubvol(parent, name, namelen,
1048                              root, readonly, inherit);
1049 out:
1050         if (snapshot_force_cow)
1051                 atomic_dec(&root->snapshot_force_cow);
1052         btrfs_drew_read_unlock(&root->snapshot_lock);
1053         return ret;
1054 }
1055
1056 /*
1057  * When we're defragging a range, we don't want to kick it off again
1058  * if it is really just waiting for delalloc to send it down.
1059  * If we find a nice big extent or delalloc range for the bytes in the
1060  * file you want to defrag, we return 0 to let you know to skip this
1061  * part of the file
1062  */
1063 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1064 {
1065         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1066         struct extent_map *em = NULL;
1067         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1068         u64 end;
1069
1070         read_lock(&em_tree->lock);
1071         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1072         read_unlock(&em_tree->lock);
1073
1074         if (em) {
1075                 end = extent_map_end(em);
1076                 free_extent_map(em);
1077                 if (end - offset > thresh)
1078                         return 0;
1079         }
1080         /* if we already have a nice delalloc here, just stop */
1081         thresh /= 2;
1082         end = count_range_bits(io_tree, &offset, offset + thresh,
1083                                thresh, EXTENT_DELALLOC, 1);
1084         if (end >= thresh)
1085                 return 0;
1086         return 1;
1087 }
1088
1089 /*
1090  * helper function to walk through a file and find extents
1091  * newer than a specific transid, and smaller than thresh.
1092  *
1093  * This is used by the defragging code to find new and small
1094  * extents
1095  */
1096 static int find_new_extents(struct btrfs_root *root,
1097                             struct inode *inode, u64 newer_than,
1098                             u64 *off, u32 thresh)
1099 {
1100         struct btrfs_path *path;
1101         struct btrfs_key min_key;
1102         struct extent_buffer *leaf;
1103         struct btrfs_file_extent_item *extent;
1104         int type;
1105         int ret;
1106         u64 ino = btrfs_ino(BTRFS_I(inode));
1107
1108         path = btrfs_alloc_path();
1109         if (!path)
1110                 return -ENOMEM;
1111
1112         min_key.objectid = ino;
1113         min_key.type = BTRFS_EXTENT_DATA_KEY;
1114         min_key.offset = *off;
1115
1116         while (1) {
1117                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1118                 if (ret != 0)
1119                         goto none;
1120 process_slot:
1121                 if (min_key.objectid != ino)
1122                         goto none;
1123                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1124                         goto none;
1125
1126                 leaf = path->nodes[0];
1127                 extent = btrfs_item_ptr(leaf, path->slots[0],
1128                                         struct btrfs_file_extent_item);
1129
1130                 type = btrfs_file_extent_type(leaf, extent);
1131                 if (type == BTRFS_FILE_EXTENT_REG &&
1132                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1133                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1134                         *off = min_key.offset;
1135                         btrfs_free_path(path);
1136                         return 0;
1137                 }
1138
1139                 path->slots[0]++;
1140                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1141                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1142                         goto process_slot;
1143                 }
1144
1145                 if (min_key.offset == (u64)-1)
1146                         goto none;
1147
1148                 min_key.offset++;
1149                 btrfs_release_path(path);
1150         }
1151 none:
1152         btrfs_free_path(path);
1153         return -ENOENT;
1154 }
1155
1156 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1157 {
1158         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1159         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1160         struct extent_map *em;
1161         u64 len = PAGE_SIZE;
1162
1163         /*
1164          * hopefully we have this extent in the tree already, try without
1165          * the full extent lock
1166          */
1167         read_lock(&em_tree->lock);
1168         em = lookup_extent_mapping(em_tree, start, len);
1169         read_unlock(&em_tree->lock);
1170
1171         if (!em) {
1172                 struct extent_state *cached = NULL;
1173                 u64 end = start + len - 1;
1174
1175                 /* get the big lock and read metadata off disk */
1176                 lock_extent_bits(io_tree, start, end, &cached);
1177                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1178                 unlock_extent_cached(io_tree, start, end, &cached);
1179
1180                 if (IS_ERR(em))
1181                         return NULL;
1182         }
1183
1184         return em;
1185 }
1186
1187 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1188 {
1189         struct extent_map *next;
1190         bool ret = true;
1191
1192         /* this is the last extent */
1193         if (em->start + em->len >= i_size_read(inode))
1194                 return false;
1195
1196         next = defrag_lookup_extent(inode, em->start + em->len);
1197         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1198                 ret = false;
1199         else if ((em->block_start + em->block_len == next->block_start) &&
1200                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1201                 ret = false;
1202
1203         free_extent_map(next);
1204         return ret;
1205 }
1206
1207 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1208                                u64 *last_len, u64 *skip, u64 *defrag_end,
1209                                int compress)
1210 {
1211         struct extent_map *em;
1212         int ret = 1;
1213         bool next_mergeable = true;
1214         bool prev_mergeable = true;
1215
1216         /*
1217          * make sure that once we start defragging an extent, we keep on
1218          * defragging it
1219          */
1220         if (start < *defrag_end)
1221                 return 1;
1222
1223         *skip = 0;
1224
1225         em = defrag_lookup_extent(inode, start);
1226         if (!em)
1227                 return 0;
1228
1229         /* this will cover holes, and inline extents */
1230         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1231                 ret = 0;
1232                 goto out;
1233         }
1234
1235         if (!*defrag_end)
1236                 prev_mergeable = false;
1237
1238         next_mergeable = defrag_check_next_extent(inode, em);
1239         /*
1240          * we hit a real extent, if it is big or the next extent is not a
1241          * real extent, don't bother defragging it
1242          */
1243         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1244             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1245                 ret = 0;
1246 out:
1247         /*
1248          * last_len ends up being a counter of how many bytes we've defragged.
1249          * every time we choose not to defrag an extent, we reset *last_len
1250          * so that the next tiny extent will force a defrag.
1251          *
1252          * The end result of this is that tiny extents before a single big
1253          * extent will force at least part of that big extent to be defragged.
1254          */
1255         if (ret) {
1256                 *defrag_end = extent_map_end(em);
1257         } else {
1258                 *last_len = 0;
1259                 *skip = extent_map_end(em);
1260                 *defrag_end = 0;
1261         }
1262
1263         free_extent_map(em);
1264         return ret;
1265 }
1266
1267 /*
1268  * it doesn't do much good to defrag one or two pages
1269  * at a time.  This pulls in a nice chunk of pages
1270  * to COW and defrag.
1271  *
1272  * It also makes sure the delalloc code has enough
1273  * dirty data to avoid making new small extents as part
1274  * of the defrag
1275  *
1276  * It's a good idea to start RA on this range
1277  * before calling this.
1278  */
1279 static int cluster_pages_for_defrag(struct inode *inode,
1280                                     struct page **pages,
1281                                     unsigned long start_index,
1282                                     unsigned long num_pages)
1283 {
1284         unsigned long file_end;
1285         u64 isize = i_size_read(inode);
1286         u64 page_start;
1287         u64 page_end;
1288         u64 page_cnt;
1289         u64 start = (u64)start_index << PAGE_SHIFT;
1290         u64 search_start;
1291         int ret;
1292         int i;
1293         int i_done;
1294         struct btrfs_ordered_extent *ordered;
1295         struct extent_state *cached_state = NULL;
1296         struct extent_io_tree *tree;
1297         struct extent_changeset *data_reserved = NULL;
1298         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1299
1300         file_end = (isize - 1) >> PAGE_SHIFT;
1301         if (!isize || start_index > file_end)
1302                 return 0;
1303
1304         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1305
1306         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1307                         start, page_cnt << PAGE_SHIFT);
1308         if (ret)
1309                 return ret;
1310         i_done = 0;
1311         tree = &BTRFS_I(inode)->io_tree;
1312
1313         /* step one, lock all the pages */
1314         for (i = 0; i < page_cnt; i++) {
1315                 struct page *page;
1316 again:
1317                 page = find_or_create_page(inode->i_mapping,
1318                                            start_index + i, mask);
1319                 if (!page)
1320                         break;
1321
1322                 page_start = page_offset(page);
1323                 page_end = page_start + PAGE_SIZE - 1;
1324                 while (1) {
1325                         lock_extent_bits(tree, page_start, page_end,
1326                                          &cached_state);
1327                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1328                                                               page_start);
1329                         unlock_extent_cached(tree, page_start, page_end,
1330                                              &cached_state);
1331                         if (!ordered)
1332                                 break;
1333
1334                         unlock_page(page);
1335                         btrfs_start_ordered_extent(ordered, 1);
1336                         btrfs_put_ordered_extent(ordered);
1337                         lock_page(page);
1338                         /*
1339                          * we unlocked the page above, so we need check if
1340                          * it was released or not.
1341                          */
1342                         if (page->mapping != inode->i_mapping) {
1343                                 unlock_page(page);
1344                                 put_page(page);
1345                                 goto again;
1346                         }
1347                 }
1348
1349                 if (!PageUptodate(page)) {
1350                         btrfs_readpage(NULL, page);
1351                         lock_page(page);
1352                         if (!PageUptodate(page)) {
1353                                 unlock_page(page);
1354                                 put_page(page);
1355                                 ret = -EIO;
1356                                 break;
1357                         }
1358                 }
1359
1360                 if (page->mapping != inode->i_mapping) {
1361                         unlock_page(page);
1362                         put_page(page);
1363                         goto again;
1364                 }
1365
1366                 pages[i] = page;
1367                 i_done++;
1368         }
1369         if (!i_done || ret)
1370                 goto out;
1371
1372         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1373                 goto out;
1374
1375         /*
1376          * so now we have a nice long stream of locked
1377          * and up to date pages, lets wait on them
1378          */
1379         for (i = 0; i < i_done; i++)
1380                 wait_on_page_writeback(pages[i]);
1381
1382         page_start = page_offset(pages[0]);
1383         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1384
1385         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1386                          page_start, page_end - 1, &cached_state);
1387
1388         /*
1389          * When defragmenting we skip ranges that have holes or inline extents,
1390          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1391          * space. At btrfs_defrag_file(), we check if a range should be defragged
1392          * before locking the inode and then, if it should, we trigger a sync
1393          * page cache readahead - we lock the inode only after that to avoid
1394          * blocking for too long other tasks that possibly want to operate on
1395          * other file ranges. But before we were able to get the inode lock,
1396          * some other task may have punched a hole in the range, or we may have
1397          * now an inline extent, in which case we should not defrag. So check
1398          * for that here, where we have the inode and the range locked, and bail
1399          * out if that happened.
1400          */
1401         search_start = page_start;
1402         while (search_start < page_end) {
1403                 struct extent_map *em;
1404
1405                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1406                                       page_end - search_start);
1407                 if (IS_ERR(em)) {
1408                         ret = PTR_ERR(em);
1409                         goto out_unlock_range;
1410                 }
1411                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1412                         free_extent_map(em);
1413                         /* Ok, 0 means we did not defrag anything */
1414                         ret = 0;
1415                         goto out_unlock_range;
1416                 }
1417                 search_start = extent_map_end(em);
1418                 free_extent_map(em);
1419         }
1420
1421         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1422                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1423                           EXTENT_DEFRAG, 0, 0, &cached_state);
1424
1425         if (i_done != page_cnt) {
1426                 spin_lock(&BTRFS_I(inode)->lock);
1427                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1428                 spin_unlock(&BTRFS_I(inode)->lock);
1429                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1430                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1431         }
1432
1433
1434         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1435                           &cached_state);
1436
1437         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1438                              page_start, page_end - 1, &cached_state);
1439
1440         for (i = 0; i < i_done; i++) {
1441                 clear_page_dirty_for_io(pages[i]);
1442                 ClearPageChecked(pages[i]);
1443                 set_page_extent_mapped(pages[i]);
1444                 set_page_dirty(pages[i]);
1445                 unlock_page(pages[i]);
1446                 put_page(pages[i]);
1447         }
1448         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1449         extent_changeset_free(data_reserved);
1450         return i_done;
1451
1452 out_unlock_range:
1453         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1454                              page_start, page_end - 1, &cached_state);
1455 out:
1456         for (i = 0; i < i_done; i++) {
1457                 unlock_page(pages[i]);
1458                 put_page(pages[i]);
1459         }
1460         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1461                         start, page_cnt << PAGE_SHIFT, true);
1462         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1463         extent_changeset_free(data_reserved);
1464         return ret;
1465
1466 }
1467
1468 int btrfs_defrag_file(struct inode *inode, struct file *file,
1469                       struct btrfs_ioctl_defrag_range_args *range,
1470                       u64 newer_than, unsigned long max_to_defrag)
1471 {
1472         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1473         struct btrfs_root *root = BTRFS_I(inode)->root;
1474         struct file_ra_state *ra = NULL;
1475         unsigned long last_index;
1476         u64 isize = i_size_read(inode);
1477         u64 last_len = 0;
1478         u64 skip = 0;
1479         u64 defrag_end = 0;
1480         u64 newer_off = range->start;
1481         unsigned long i;
1482         unsigned long ra_index = 0;
1483         int ret;
1484         int defrag_count = 0;
1485         int compress_type = BTRFS_COMPRESS_ZLIB;
1486         u32 extent_thresh = range->extent_thresh;
1487         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1488         unsigned long cluster = max_cluster;
1489         u64 new_align = ~((u64)SZ_128K - 1);
1490         struct page **pages = NULL;
1491         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1492
1493         if (isize == 0)
1494                 return 0;
1495
1496         if (range->start >= isize)
1497                 return -EINVAL;
1498
1499         if (do_compress) {
1500                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1501                         return -EINVAL;
1502                 if (range->compress_type)
1503                         compress_type = range->compress_type;
1504         }
1505
1506         if (extent_thresh == 0)
1507                 extent_thresh = SZ_256K;
1508
1509         /*
1510          * If we were not given a file, allocate a readahead context. As
1511          * readahead is just an optimization, defrag will work without it so
1512          * we don't error out.
1513          */
1514         if (!file) {
1515                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1516                 if (ra)
1517                         file_ra_state_init(ra, inode->i_mapping);
1518         } else {
1519                 ra = &file->f_ra;
1520         }
1521
1522         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1523         if (!pages) {
1524                 ret = -ENOMEM;
1525                 goto out_ra;
1526         }
1527
1528         /* find the last page to defrag */
1529         if (range->start + range->len > range->start) {
1530                 last_index = min_t(u64, isize - 1,
1531                          range->start + range->len - 1) >> PAGE_SHIFT;
1532         } else {
1533                 last_index = (isize - 1) >> PAGE_SHIFT;
1534         }
1535
1536         if (newer_than) {
1537                 ret = find_new_extents(root, inode, newer_than,
1538                                        &newer_off, SZ_64K);
1539                 if (!ret) {
1540                         range->start = newer_off;
1541                         /*
1542                          * we always align our defrag to help keep
1543                          * the extents in the file evenly spaced
1544                          */
1545                         i = (newer_off & new_align) >> PAGE_SHIFT;
1546                 } else
1547                         goto out_ra;
1548         } else {
1549                 i = range->start >> PAGE_SHIFT;
1550         }
1551         if (!max_to_defrag)
1552                 max_to_defrag = last_index - i + 1;
1553
1554         /*
1555          * make writeback starts from i, so the defrag range can be
1556          * written sequentially.
1557          */
1558         if (i < inode->i_mapping->writeback_index)
1559                 inode->i_mapping->writeback_index = i;
1560
1561         while (i <= last_index && defrag_count < max_to_defrag &&
1562                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1563                 /*
1564                  * make sure we stop running if someone unmounts
1565                  * the FS
1566                  */
1567                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1568                         break;
1569
1570                 if (btrfs_defrag_cancelled(fs_info)) {
1571                         btrfs_debug(fs_info, "defrag_file cancelled");
1572                         ret = -EAGAIN;
1573                         break;
1574                 }
1575
1576                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1577                                          extent_thresh, &last_len, &skip,
1578                                          &defrag_end, do_compress)){
1579                         unsigned long next;
1580                         /*
1581                          * the should_defrag function tells us how much to skip
1582                          * bump our counter by the suggested amount
1583                          */
1584                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1585                         i = max(i + 1, next);
1586                         continue;
1587                 }
1588
1589                 if (!newer_than) {
1590                         cluster = (PAGE_ALIGN(defrag_end) >>
1591                                    PAGE_SHIFT) - i;
1592                         cluster = min(cluster, max_cluster);
1593                 } else {
1594                         cluster = max_cluster;
1595                 }
1596
1597                 if (i + cluster > ra_index) {
1598                         ra_index = max(i, ra_index);
1599                         if (ra)
1600                                 page_cache_sync_readahead(inode->i_mapping, ra,
1601                                                 file, ra_index, cluster);
1602                         ra_index += cluster;
1603                 }
1604
1605                 inode_lock(inode);
1606                 if (IS_SWAPFILE(inode)) {
1607                         ret = -ETXTBSY;
1608                 } else {
1609                         if (do_compress)
1610                                 BTRFS_I(inode)->defrag_compress = compress_type;
1611                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1612                 }
1613                 if (ret < 0) {
1614                         inode_unlock(inode);
1615                         goto out_ra;
1616                 }
1617
1618                 defrag_count += ret;
1619                 balance_dirty_pages_ratelimited(inode->i_mapping);
1620                 inode_unlock(inode);
1621
1622                 if (newer_than) {
1623                         if (newer_off == (u64)-1)
1624                                 break;
1625
1626                         if (ret > 0)
1627                                 i += ret;
1628
1629                         newer_off = max(newer_off + 1,
1630                                         (u64)i << PAGE_SHIFT);
1631
1632                         ret = find_new_extents(root, inode, newer_than,
1633                                                &newer_off, SZ_64K);
1634                         if (!ret) {
1635                                 range->start = newer_off;
1636                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1637                         } else {
1638                                 break;
1639                         }
1640                 } else {
1641                         if (ret > 0) {
1642                                 i += ret;
1643                                 last_len += ret << PAGE_SHIFT;
1644                         } else {
1645                                 i++;
1646                                 last_len = 0;
1647                         }
1648                 }
1649         }
1650
1651         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1652                 filemap_flush(inode->i_mapping);
1653                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1654                              &BTRFS_I(inode)->runtime_flags))
1655                         filemap_flush(inode->i_mapping);
1656         }
1657
1658         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1659                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1660         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1661                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1662         }
1663
1664         ret = defrag_count;
1665
1666 out_ra:
1667         if (do_compress) {
1668                 inode_lock(inode);
1669                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1670                 inode_unlock(inode);
1671         }
1672         if (!file)
1673                 kfree(ra);
1674         kfree(pages);
1675         return ret;
1676 }
1677
1678 static noinline int btrfs_ioctl_resize(struct file *file,
1679                                         void __user *arg)
1680 {
1681         struct inode *inode = file_inode(file);
1682         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1683         u64 new_size;
1684         u64 old_size;
1685         u64 devid = 1;
1686         struct btrfs_root *root = BTRFS_I(inode)->root;
1687         struct btrfs_ioctl_vol_args *vol_args;
1688         struct btrfs_trans_handle *trans;
1689         struct btrfs_device *device = NULL;
1690         char *sizestr;
1691         char *retptr;
1692         char *devstr = NULL;
1693         int ret = 0;
1694         int mod = 0;
1695
1696         if (!capable(CAP_SYS_ADMIN))
1697                 return -EPERM;
1698
1699         ret = mnt_want_write_file(file);
1700         if (ret)
1701                 return ret;
1702
1703         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_RESIZE)) {
1704                 mnt_drop_write_file(file);
1705                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1706         }
1707
1708         vol_args = memdup_user(arg, sizeof(*vol_args));
1709         if (IS_ERR(vol_args)) {
1710                 ret = PTR_ERR(vol_args);
1711                 goto out;
1712         }
1713
1714         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1715
1716         sizestr = vol_args->name;
1717         devstr = strchr(sizestr, ':');
1718         if (devstr) {
1719                 sizestr = devstr + 1;
1720                 *devstr = '\0';
1721                 devstr = vol_args->name;
1722                 ret = kstrtoull(devstr, 10, &devid);
1723                 if (ret)
1724                         goto out_free;
1725                 if (!devid) {
1726                         ret = -EINVAL;
1727                         goto out_free;
1728                 }
1729                 btrfs_info(fs_info, "resizing devid %llu", devid);
1730         }
1731
1732         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
1733         if (!device) {
1734                 btrfs_info(fs_info, "resizer unable to find device %llu",
1735                            devid);
1736                 ret = -ENODEV;
1737                 goto out_free;
1738         }
1739
1740         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1741                 btrfs_info(fs_info,
1742                            "resizer unable to apply on readonly device %llu",
1743                        devid);
1744                 ret = -EPERM;
1745                 goto out_free;
1746         }
1747
1748         if (!strcmp(sizestr, "max"))
1749                 new_size = device->bdev->bd_inode->i_size;
1750         else {
1751                 if (sizestr[0] == '-') {
1752                         mod = -1;
1753                         sizestr++;
1754                 } else if (sizestr[0] == '+') {
1755                         mod = 1;
1756                         sizestr++;
1757                 }
1758                 new_size = memparse(sizestr, &retptr);
1759                 if (*retptr != '\0' || new_size == 0) {
1760                         ret = -EINVAL;
1761                         goto out_free;
1762                 }
1763         }
1764
1765         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1766                 ret = -EPERM;
1767                 goto out_free;
1768         }
1769
1770         old_size = btrfs_device_get_total_bytes(device);
1771
1772         if (mod < 0) {
1773                 if (new_size > old_size) {
1774                         ret = -EINVAL;
1775                         goto out_free;
1776                 }
1777                 new_size = old_size - new_size;
1778         } else if (mod > 0) {
1779                 if (new_size > ULLONG_MAX - old_size) {
1780                         ret = -ERANGE;
1781                         goto out_free;
1782                 }
1783                 new_size = old_size + new_size;
1784         }
1785
1786         if (new_size < SZ_256M) {
1787                 ret = -EINVAL;
1788                 goto out_free;
1789         }
1790         if (new_size > device->bdev->bd_inode->i_size) {
1791                 ret = -EFBIG;
1792                 goto out_free;
1793         }
1794
1795         new_size = round_down(new_size, fs_info->sectorsize);
1796
1797         if (new_size > old_size) {
1798                 trans = btrfs_start_transaction(root, 0);
1799                 if (IS_ERR(trans)) {
1800                         ret = PTR_ERR(trans);
1801                         goto out_free;
1802                 }
1803                 ret = btrfs_grow_device(trans, device, new_size);
1804                 btrfs_commit_transaction(trans);
1805         } else if (new_size < old_size) {
1806                 ret = btrfs_shrink_device(device, new_size);
1807         } /* equal, nothing need to do */
1808
1809         if (ret == 0 && new_size != old_size)
1810                 btrfs_info_in_rcu(fs_info,
1811                         "resize device %s (devid %llu) from %llu to %llu",
1812                         rcu_str_deref(device->name), device->devid,
1813                         old_size, new_size);
1814 out_free:
1815         kfree(vol_args);
1816 out:
1817         btrfs_exclop_finish(fs_info);
1818         mnt_drop_write_file(file);
1819         return ret;
1820 }
1821
1822 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1823                                 const char *name, unsigned long fd, int subvol,
1824                                 bool readonly,
1825                                 struct btrfs_qgroup_inherit *inherit)
1826 {
1827         int namelen;
1828         int ret = 0;
1829
1830         if (!S_ISDIR(file_inode(file)->i_mode))
1831                 return -ENOTDIR;
1832
1833         ret = mnt_want_write_file(file);
1834         if (ret)
1835                 goto out;
1836
1837         namelen = strlen(name);
1838         if (strchr(name, '/')) {
1839                 ret = -EINVAL;
1840                 goto out_drop_write;
1841         }
1842
1843         if (name[0] == '.' &&
1844            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1845                 ret = -EEXIST;
1846                 goto out_drop_write;
1847         }
1848
1849         if (subvol) {
1850                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1851                                      NULL, readonly, inherit);
1852         } else {
1853                 struct fd src = fdget(fd);
1854                 struct inode *src_inode;
1855                 if (!src.file) {
1856                         ret = -EINVAL;
1857                         goto out_drop_write;
1858                 }
1859
1860                 src_inode = file_inode(src.file);
1861                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1862                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1863                                    "Snapshot src from another FS");
1864                         ret = -EXDEV;
1865                 } else if (!inode_owner_or_capable(src_inode)) {
1866                         /*
1867                          * Subvolume creation is not restricted, but snapshots
1868                          * are limited to own subvolumes only
1869                          */
1870                         ret = -EPERM;
1871                 } else {
1872                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1873                                              BTRFS_I(src_inode)->root,
1874                                              readonly, inherit);
1875                 }
1876                 fdput(src);
1877         }
1878 out_drop_write:
1879         mnt_drop_write_file(file);
1880 out:
1881         return ret;
1882 }
1883
1884 static noinline int btrfs_ioctl_snap_create(struct file *file,
1885                                             void __user *arg, int subvol)
1886 {
1887         struct btrfs_ioctl_vol_args *vol_args;
1888         int ret;
1889
1890         if (!S_ISDIR(file_inode(file)->i_mode))
1891                 return -ENOTDIR;
1892
1893         vol_args = memdup_user(arg, sizeof(*vol_args));
1894         if (IS_ERR(vol_args))
1895                 return PTR_ERR(vol_args);
1896         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1897
1898         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1899                                         subvol, false, NULL);
1900
1901         kfree(vol_args);
1902         return ret;
1903 }
1904
1905 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1906                                                void __user *arg, int subvol)
1907 {
1908         struct btrfs_ioctl_vol_args_v2 *vol_args;
1909         int ret;
1910         bool readonly = false;
1911         struct btrfs_qgroup_inherit *inherit = NULL;
1912
1913         if (!S_ISDIR(file_inode(file)->i_mode))
1914                 return -ENOTDIR;
1915
1916         vol_args = memdup_user(arg, sizeof(*vol_args));
1917         if (IS_ERR(vol_args))
1918                 return PTR_ERR(vol_args);
1919         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1920
1921         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1922                 ret = -EOPNOTSUPP;
1923                 goto free_args;
1924         }
1925
1926         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1927                 readonly = true;
1928         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1929                 if (vol_args->size > PAGE_SIZE) {
1930                         ret = -EINVAL;
1931                         goto free_args;
1932                 }
1933                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1934                 if (IS_ERR(inherit)) {
1935                         ret = PTR_ERR(inherit);
1936                         goto free_args;
1937                 }
1938         }
1939
1940         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1941                                         subvol, readonly, inherit);
1942         if (ret)
1943                 goto free_inherit;
1944 free_inherit:
1945         kfree(inherit);
1946 free_args:
1947         kfree(vol_args);
1948         return ret;
1949 }
1950
1951 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1952                                                 void __user *arg)
1953 {
1954         struct inode *inode = file_inode(file);
1955         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1956         struct btrfs_root *root = BTRFS_I(inode)->root;
1957         int ret = 0;
1958         u64 flags = 0;
1959
1960         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1961                 return -EINVAL;
1962
1963         down_read(&fs_info->subvol_sem);
1964         if (btrfs_root_readonly(root))
1965                 flags |= BTRFS_SUBVOL_RDONLY;
1966         up_read(&fs_info->subvol_sem);
1967
1968         if (copy_to_user(arg, &flags, sizeof(flags)))
1969                 ret = -EFAULT;
1970
1971         return ret;
1972 }
1973
1974 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1975                                               void __user *arg)
1976 {
1977         struct inode *inode = file_inode(file);
1978         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1979         struct btrfs_root *root = BTRFS_I(inode)->root;
1980         struct btrfs_trans_handle *trans;
1981         u64 root_flags;
1982         u64 flags;
1983         int ret = 0;
1984
1985         if (!inode_owner_or_capable(inode))
1986                 return -EPERM;
1987
1988         ret = mnt_want_write_file(file);
1989         if (ret)
1990                 goto out;
1991
1992         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1993                 ret = -EINVAL;
1994                 goto out_drop_write;
1995         }
1996
1997         if (copy_from_user(&flags, arg, sizeof(flags))) {
1998                 ret = -EFAULT;
1999                 goto out_drop_write;
2000         }
2001
2002         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2003                 ret = -EOPNOTSUPP;
2004                 goto out_drop_write;
2005         }
2006
2007         down_write(&fs_info->subvol_sem);
2008
2009         /* nothing to do */
2010         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2011                 goto out_drop_sem;
2012
2013         root_flags = btrfs_root_flags(&root->root_item);
2014         if (flags & BTRFS_SUBVOL_RDONLY) {
2015                 btrfs_set_root_flags(&root->root_item,
2016                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2017         } else {
2018                 /*
2019                  * Block RO -> RW transition if this subvolume is involved in
2020                  * send
2021                  */
2022                 spin_lock(&root->root_item_lock);
2023                 if (root->send_in_progress == 0) {
2024                         btrfs_set_root_flags(&root->root_item,
2025                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2026                         spin_unlock(&root->root_item_lock);
2027                 } else {
2028                         spin_unlock(&root->root_item_lock);
2029                         btrfs_warn(fs_info,
2030                                    "Attempt to set subvolume %llu read-write during send",
2031                                    root->root_key.objectid);
2032                         ret = -EPERM;
2033                         goto out_drop_sem;
2034                 }
2035         }
2036
2037         trans = btrfs_start_transaction(root, 1);
2038         if (IS_ERR(trans)) {
2039                 ret = PTR_ERR(trans);
2040                 goto out_reset;
2041         }
2042
2043         ret = btrfs_update_root(trans, fs_info->tree_root,
2044                                 &root->root_key, &root->root_item);
2045         if (ret < 0) {
2046                 btrfs_end_transaction(trans);
2047                 goto out_reset;
2048         }
2049
2050         ret = btrfs_commit_transaction(trans);
2051
2052 out_reset:
2053         if (ret)
2054                 btrfs_set_root_flags(&root->root_item, root_flags);
2055 out_drop_sem:
2056         up_write(&fs_info->subvol_sem);
2057 out_drop_write:
2058         mnt_drop_write_file(file);
2059 out:
2060         return ret;
2061 }
2062
2063 static noinline int key_in_sk(struct btrfs_key *key,
2064                               struct btrfs_ioctl_search_key *sk)
2065 {
2066         struct btrfs_key test;
2067         int ret;
2068
2069         test.objectid = sk->min_objectid;
2070         test.type = sk->min_type;
2071         test.offset = sk->min_offset;
2072
2073         ret = btrfs_comp_cpu_keys(key, &test);
2074         if (ret < 0)
2075                 return 0;
2076
2077         test.objectid = sk->max_objectid;
2078         test.type = sk->max_type;
2079         test.offset = sk->max_offset;
2080
2081         ret = btrfs_comp_cpu_keys(key, &test);
2082         if (ret > 0)
2083                 return 0;
2084         return 1;
2085 }
2086
2087 static noinline int copy_to_sk(struct btrfs_path *path,
2088                                struct btrfs_key *key,
2089                                struct btrfs_ioctl_search_key *sk,
2090                                size_t *buf_size,
2091                                char __user *ubuf,
2092                                unsigned long *sk_offset,
2093                                int *num_found)
2094 {
2095         u64 found_transid;
2096         struct extent_buffer *leaf;
2097         struct btrfs_ioctl_search_header sh;
2098         struct btrfs_key test;
2099         unsigned long item_off;
2100         unsigned long item_len;
2101         int nritems;
2102         int i;
2103         int slot;
2104         int ret = 0;
2105
2106         leaf = path->nodes[0];
2107         slot = path->slots[0];
2108         nritems = btrfs_header_nritems(leaf);
2109
2110         if (btrfs_header_generation(leaf) > sk->max_transid) {
2111                 i = nritems;
2112                 goto advance_key;
2113         }
2114         found_transid = btrfs_header_generation(leaf);
2115
2116         for (i = slot; i < nritems; i++) {
2117                 item_off = btrfs_item_ptr_offset(leaf, i);
2118                 item_len = btrfs_item_size_nr(leaf, i);
2119
2120                 btrfs_item_key_to_cpu(leaf, key, i);
2121                 if (!key_in_sk(key, sk))
2122                         continue;
2123
2124                 if (sizeof(sh) + item_len > *buf_size) {
2125                         if (*num_found) {
2126                                 ret = 1;
2127                                 goto out;
2128                         }
2129
2130                         /*
2131                          * return one empty item back for v1, which does not
2132                          * handle -EOVERFLOW
2133                          */
2134
2135                         *buf_size = sizeof(sh) + item_len;
2136                         item_len = 0;
2137                         ret = -EOVERFLOW;
2138                 }
2139
2140                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2141                         ret = 1;
2142                         goto out;
2143                 }
2144
2145                 sh.objectid = key->objectid;
2146                 sh.offset = key->offset;
2147                 sh.type = key->type;
2148                 sh.len = item_len;
2149                 sh.transid = found_transid;
2150
2151                 /*
2152                  * Copy search result header. If we fault then loop again so we
2153                  * can fault in the pages and -EFAULT there if there's a
2154                  * problem. Otherwise we'll fault and then copy the buffer in
2155                  * properly this next time through
2156                  */
2157                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2158                         ret = 0;
2159                         goto out;
2160                 }
2161
2162                 *sk_offset += sizeof(sh);
2163
2164                 if (item_len) {
2165                         char __user *up = ubuf + *sk_offset;
2166                         /*
2167                          * Copy the item, same behavior as above, but reset the
2168                          * * sk_offset so we copy the full thing again.
2169                          */
2170                         if (read_extent_buffer_to_user_nofault(leaf, up,
2171                                                 item_off, item_len)) {
2172                                 ret = 0;
2173                                 *sk_offset -= sizeof(sh);
2174                                 goto out;
2175                         }
2176
2177                         *sk_offset += item_len;
2178                 }
2179                 (*num_found)++;
2180
2181                 if (ret) /* -EOVERFLOW from above */
2182                         goto out;
2183
2184                 if (*num_found >= sk->nr_items) {
2185                         ret = 1;
2186                         goto out;
2187                 }
2188         }
2189 advance_key:
2190         ret = 0;
2191         test.objectid = sk->max_objectid;
2192         test.type = sk->max_type;
2193         test.offset = sk->max_offset;
2194         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2195                 ret = 1;
2196         else if (key->offset < (u64)-1)
2197                 key->offset++;
2198         else if (key->type < (u8)-1) {
2199                 key->offset = 0;
2200                 key->type++;
2201         } else if (key->objectid < (u64)-1) {
2202                 key->offset = 0;
2203                 key->type = 0;
2204                 key->objectid++;
2205         } else
2206                 ret = 1;
2207 out:
2208         /*
2209          *  0: all items from this leaf copied, continue with next
2210          *  1: * more items can be copied, but unused buffer is too small
2211          *     * all items were found
2212          *     Either way, it will stops the loop which iterates to the next
2213          *     leaf
2214          *  -EOVERFLOW: item was to large for buffer
2215          *  -EFAULT: could not copy extent buffer back to userspace
2216          */
2217         return ret;
2218 }
2219
2220 static noinline int search_ioctl(struct inode *inode,
2221                                  struct btrfs_ioctl_search_key *sk,
2222                                  size_t *buf_size,
2223                                  char __user *ubuf)
2224 {
2225         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2226         struct btrfs_root *root;
2227         struct btrfs_key key;
2228         struct btrfs_path *path;
2229         int ret;
2230         int num_found = 0;
2231         unsigned long sk_offset = 0;
2232
2233         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2234                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2235                 return -EOVERFLOW;
2236         }
2237
2238         path = btrfs_alloc_path();
2239         if (!path)
2240                 return -ENOMEM;
2241
2242         if (sk->tree_id == 0) {
2243                 /* search the root of the inode that was passed */
2244                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2245         } else {
2246                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2247                 if (IS_ERR(root)) {
2248                         btrfs_free_path(path);
2249                         return PTR_ERR(root);
2250                 }
2251         }
2252
2253         key.objectid = sk->min_objectid;
2254         key.type = sk->min_type;
2255         key.offset = sk->min_offset;
2256
2257         while (1) {
2258                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2259                                                *buf_size - sk_offset);
2260                 if (ret)
2261                         break;
2262
2263                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2264                 if (ret != 0) {
2265                         if (ret > 0)
2266                                 ret = 0;
2267                         goto err;
2268                 }
2269                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2270                                  &sk_offset, &num_found);
2271                 btrfs_release_path(path);
2272                 if (ret)
2273                         break;
2274
2275         }
2276         if (ret > 0)
2277                 ret = 0;
2278 err:
2279         sk->nr_items = num_found;
2280         btrfs_put_root(root);
2281         btrfs_free_path(path);
2282         return ret;
2283 }
2284
2285 static noinline int btrfs_ioctl_tree_search(struct file *file,
2286                                            void __user *argp)
2287 {
2288         struct btrfs_ioctl_search_args __user *uargs;
2289         struct btrfs_ioctl_search_key sk;
2290         struct inode *inode;
2291         int ret;
2292         size_t buf_size;
2293
2294         if (!capable(CAP_SYS_ADMIN))
2295                 return -EPERM;
2296
2297         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2298
2299         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2300                 return -EFAULT;
2301
2302         buf_size = sizeof(uargs->buf);
2303
2304         inode = file_inode(file);
2305         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2306
2307         /*
2308          * In the origin implementation an overflow is handled by returning a
2309          * search header with a len of zero, so reset ret.
2310          */
2311         if (ret == -EOVERFLOW)
2312                 ret = 0;
2313
2314         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2315                 ret = -EFAULT;
2316         return ret;
2317 }
2318
2319 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2320                                                void __user *argp)
2321 {
2322         struct btrfs_ioctl_search_args_v2 __user *uarg;
2323         struct btrfs_ioctl_search_args_v2 args;
2324         struct inode *inode;
2325         int ret;
2326         size_t buf_size;
2327         const size_t buf_limit = SZ_16M;
2328
2329         if (!capable(CAP_SYS_ADMIN))
2330                 return -EPERM;
2331
2332         /* copy search header and buffer size */
2333         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2334         if (copy_from_user(&args, uarg, sizeof(args)))
2335                 return -EFAULT;
2336
2337         buf_size = args.buf_size;
2338
2339         /* limit result size to 16MB */
2340         if (buf_size > buf_limit)
2341                 buf_size = buf_limit;
2342
2343         inode = file_inode(file);
2344         ret = search_ioctl(inode, &args.key, &buf_size,
2345                            (char __user *)(&uarg->buf[0]));
2346         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2347                 ret = -EFAULT;
2348         else if (ret == -EOVERFLOW &&
2349                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2350                 ret = -EFAULT;
2351
2352         return ret;
2353 }
2354
2355 /*
2356  * Search INODE_REFs to identify path name of 'dirid' directory
2357  * in a 'tree_id' tree. and sets path name to 'name'.
2358  */
2359 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2360                                 u64 tree_id, u64 dirid, char *name)
2361 {
2362         struct btrfs_root *root;
2363         struct btrfs_key key;
2364         char *ptr;
2365         int ret = -1;
2366         int slot;
2367         int len;
2368         int total_len = 0;
2369         struct btrfs_inode_ref *iref;
2370         struct extent_buffer *l;
2371         struct btrfs_path *path;
2372
2373         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2374                 name[0]='\0';
2375                 return 0;
2376         }
2377
2378         path = btrfs_alloc_path();
2379         if (!path)
2380                 return -ENOMEM;
2381
2382         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2383
2384         root = btrfs_get_fs_root(info, tree_id, true);
2385         if (IS_ERR(root)) {
2386                 ret = PTR_ERR(root);
2387                 root = NULL;
2388                 goto out;
2389         }
2390
2391         key.objectid = dirid;
2392         key.type = BTRFS_INODE_REF_KEY;
2393         key.offset = (u64)-1;
2394
2395         while (1) {
2396                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2397                 if (ret < 0)
2398                         goto out;
2399                 else if (ret > 0) {
2400                         ret = btrfs_previous_item(root, path, dirid,
2401                                                   BTRFS_INODE_REF_KEY);
2402                         if (ret < 0)
2403                                 goto out;
2404                         else if (ret > 0) {
2405                                 ret = -ENOENT;
2406                                 goto out;
2407                         }
2408                 }
2409
2410                 l = path->nodes[0];
2411                 slot = path->slots[0];
2412                 btrfs_item_key_to_cpu(l, &key, slot);
2413
2414                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2415                 len = btrfs_inode_ref_name_len(l, iref);
2416                 ptr -= len + 1;
2417                 total_len += len + 1;
2418                 if (ptr < name) {
2419                         ret = -ENAMETOOLONG;
2420                         goto out;
2421                 }
2422
2423                 *(ptr + len) = '/';
2424                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2425
2426                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2427                         break;
2428
2429                 btrfs_release_path(path);
2430                 key.objectid = key.offset;
2431                 key.offset = (u64)-1;
2432                 dirid = key.objectid;
2433         }
2434         memmove(name, ptr, total_len);
2435         name[total_len] = '\0';
2436         ret = 0;
2437 out:
2438         btrfs_put_root(root);
2439         btrfs_free_path(path);
2440         return ret;
2441 }
2442
2443 static int btrfs_search_path_in_tree_user(struct inode *inode,
2444                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2445 {
2446         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2447         struct super_block *sb = inode->i_sb;
2448         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2449         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2450         u64 dirid = args->dirid;
2451         unsigned long item_off;
2452         unsigned long item_len;
2453         struct btrfs_inode_ref *iref;
2454         struct btrfs_root_ref *rref;
2455         struct btrfs_root *root = NULL;
2456         struct btrfs_path *path;
2457         struct btrfs_key key, key2;
2458         struct extent_buffer *leaf;
2459         struct inode *temp_inode;
2460         char *ptr;
2461         int slot;
2462         int len;
2463         int total_len = 0;
2464         int ret;
2465
2466         path = btrfs_alloc_path();
2467         if (!path)
2468                 return -ENOMEM;
2469
2470         /*
2471          * If the bottom subvolume does not exist directly under upper_limit,
2472          * construct the path in from the bottom up.
2473          */
2474         if (dirid != upper_limit.objectid) {
2475                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2476
2477                 root = btrfs_get_fs_root(fs_info, treeid, true);
2478                 if (IS_ERR(root)) {
2479                         ret = PTR_ERR(root);
2480                         goto out;
2481                 }
2482
2483                 key.objectid = dirid;
2484                 key.type = BTRFS_INODE_REF_KEY;
2485                 key.offset = (u64)-1;
2486                 while (1) {
2487                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2488                         if (ret < 0) {
2489                                 goto out_put;
2490                         } else if (ret > 0) {
2491                                 ret = btrfs_previous_item(root, path, dirid,
2492                                                           BTRFS_INODE_REF_KEY);
2493                                 if (ret < 0) {
2494                                         goto out_put;
2495                                 } else if (ret > 0) {
2496                                         ret = -ENOENT;
2497                                         goto out_put;
2498                                 }
2499                         }
2500
2501                         leaf = path->nodes[0];
2502                         slot = path->slots[0];
2503                         btrfs_item_key_to_cpu(leaf, &key, slot);
2504
2505                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2506                         len = btrfs_inode_ref_name_len(leaf, iref);
2507                         ptr -= len + 1;
2508                         total_len += len + 1;
2509                         if (ptr < args->path) {
2510                                 ret = -ENAMETOOLONG;
2511                                 goto out_put;
2512                         }
2513
2514                         *(ptr + len) = '/';
2515                         read_extent_buffer(leaf, ptr,
2516                                         (unsigned long)(iref + 1), len);
2517
2518                         /* Check the read+exec permission of this directory */
2519                         ret = btrfs_previous_item(root, path, dirid,
2520                                                   BTRFS_INODE_ITEM_KEY);
2521                         if (ret < 0) {
2522                                 goto out_put;
2523                         } else if (ret > 0) {
2524                                 ret = -ENOENT;
2525                                 goto out_put;
2526                         }
2527
2528                         leaf = path->nodes[0];
2529                         slot = path->slots[0];
2530                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2531                         if (key2.objectid != dirid) {
2532                                 ret = -ENOENT;
2533                                 goto out_put;
2534                         }
2535
2536                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2537                         if (IS_ERR(temp_inode)) {
2538                                 ret = PTR_ERR(temp_inode);
2539                                 goto out_put;
2540                         }
2541                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2542                         iput(temp_inode);
2543                         if (ret) {
2544                                 ret = -EACCES;
2545                                 goto out_put;
2546                         }
2547
2548                         if (key.offset == upper_limit.objectid)
2549                                 break;
2550                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2551                                 ret = -EACCES;
2552                                 goto out_put;
2553                         }
2554
2555                         btrfs_release_path(path);
2556                         key.objectid = key.offset;
2557                         key.offset = (u64)-1;
2558                         dirid = key.objectid;
2559                 }
2560
2561                 memmove(args->path, ptr, total_len);
2562                 args->path[total_len] = '\0';
2563                 btrfs_put_root(root);
2564                 root = NULL;
2565                 btrfs_release_path(path);
2566         }
2567
2568         /* Get the bottom subvolume's name from ROOT_REF */
2569         key.objectid = treeid;
2570         key.type = BTRFS_ROOT_REF_KEY;
2571         key.offset = args->treeid;
2572         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2573         if (ret < 0) {
2574                 goto out;
2575         } else if (ret > 0) {
2576                 ret = -ENOENT;
2577                 goto out;
2578         }
2579
2580         leaf = path->nodes[0];
2581         slot = path->slots[0];
2582         btrfs_item_key_to_cpu(leaf, &key, slot);
2583
2584         item_off = btrfs_item_ptr_offset(leaf, slot);
2585         item_len = btrfs_item_size_nr(leaf, slot);
2586         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2587         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2588         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2589                 ret = -EINVAL;
2590                 goto out;
2591         }
2592
2593         /* Copy subvolume's name */
2594         item_off += sizeof(struct btrfs_root_ref);
2595         item_len -= sizeof(struct btrfs_root_ref);
2596         read_extent_buffer(leaf, args->name, item_off, item_len);
2597         args->name[item_len] = 0;
2598
2599 out_put:
2600         btrfs_put_root(root);
2601 out:
2602         btrfs_free_path(path);
2603         return ret;
2604 }
2605
2606 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2607                                            void __user *argp)
2608 {
2609         struct btrfs_ioctl_ino_lookup_args *args;
2610         struct inode *inode;
2611         int ret = 0;
2612
2613         args = memdup_user(argp, sizeof(*args));
2614         if (IS_ERR(args))
2615                 return PTR_ERR(args);
2616
2617         inode = file_inode(file);
2618
2619         /*
2620          * Unprivileged query to obtain the containing subvolume root id. The
2621          * path is reset so it's consistent with btrfs_search_path_in_tree.
2622          */
2623         if (args->treeid == 0)
2624                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2625
2626         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2627                 args->name[0] = 0;
2628                 goto out;
2629         }
2630
2631         if (!capable(CAP_SYS_ADMIN)) {
2632                 ret = -EPERM;
2633                 goto out;
2634         }
2635
2636         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2637                                         args->treeid, args->objectid,
2638                                         args->name);
2639
2640 out:
2641         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2642                 ret = -EFAULT;
2643
2644         kfree(args);
2645         return ret;
2646 }
2647
2648 /*
2649  * Version of ino_lookup ioctl (unprivileged)
2650  *
2651  * The main differences from ino_lookup ioctl are:
2652  *
2653  *   1. Read + Exec permission will be checked using inode_permission() during
2654  *      path construction. -EACCES will be returned in case of failure.
2655  *   2. Path construction will be stopped at the inode number which corresponds
2656  *      to the fd with which this ioctl is called. If constructed path does not
2657  *      exist under fd's inode, -EACCES will be returned.
2658  *   3. The name of bottom subvolume is also searched and filled.
2659  */
2660 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2661 {
2662         struct btrfs_ioctl_ino_lookup_user_args *args;
2663         struct inode *inode;
2664         int ret;
2665
2666         args = memdup_user(argp, sizeof(*args));
2667         if (IS_ERR(args))
2668                 return PTR_ERR(args);
2669
2670         inode = file_inode(file);
2671
2672         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2673             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2674                 /*
2675                  * The subvolume does not exist under fd with which this is
2676                  * called
2677                  */
2678                 kfree(args);
2679                 return -EACCES;
2680         }
2681
2682         ret = btrfs_search_path_in_tree_user(inode, args);
2683
2684         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2685                 ret = -EFAULT;
2686
2687         kfree(args);
2688         return ret;
2689 }
2690
2691 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2692 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2693 {
2694         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2695         struct btrfs_fs_info *fs_info;
2696         struct btrfs_root *root;
2697         struct btrfs_path *path;
2698         struct btrfs_key key;
2699         struct btrfs_root_item *root_item;
2700         struct btrfs_root_ref *rref;
2701         struct extent_buffer *leaf;
2702         unsigned long item_off;
2703         unsigned long item_len;
2704         struct inode *inode;
2705         int slot;
2706         int ret = 0;
2707
2708         path = btrfs_alloc_path();
2709         if (!path)
2710                 return -ENOMEM;
2711
2712         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2713         if (!subvol_info) {
2714                 btrfs_free_path(path);
2715                 return -ENOMEM;
2716         }
2717
2718         inode = file_inode(file);
2719         fs_info = BTRFS_I(inode)->root->fs_info;
2720
2721         /* Get root_item of inode's subvolume */
2722         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2723         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2724         if (IS_ERR(root)) {
2725                 ret = PTR_ERR(root);
2726                 goto out_free;
2727         }
2728         root_item = &root->root_item;
2729
2730         subvol_info->treeid = key.objectid;
2731
2732         subvol_info->generation = btrfs_root_generation(root_item);
2733         subvol_info->flags = btrfs_root_flags(root_item);
2734
2735         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2736         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2737                                                     BTRFS_UUID_SIZE);
2738         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2739                                                     BTRFS_UUID_SIZE);
2740
2741         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2742         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2743         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2744
2745         subvol_info->otransid = btrfs_root_otransid(root_item);
2746         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2747         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2748
2749         subvol_info->stransid = btrfs_root_stransid(root_item);
2750         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2751         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2752
2753         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2754         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2755         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2756
2757         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2758                 /* Search root tree for ROOT_BACKREF of this subvolume */
2759                 key.type = BTRFS_ROOT_BACKREF_KEY;
2760                 key.offset = 0;
2761                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2762                 if (ret < 0) {
2763                         goto out;
2764                 } else if (path->slots[0] >=
2765                            btrfs_header_nritems(path->nodes[0])) {
2766                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2767                         if (ret < 0) {
2768                                 goto out;
2769                         } else if (ret > 0) {
2770                                 ret = -EUCLEAN;
2771                                 goto out;
2772                         }
2773                 }
2774
2775                 leaf = path->nodes[0];
2776                 slot = path->slots[0];
2777                 btrfs_item_key_to_cpu(leaf, &key, slot);
2778                 if (key.objectid == subvol_info->treeid &&
2779                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2780                         subvol_info->parent_id = key.offset;
2781
2782                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2783                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2784
2785                         item_off = btrfs_item_ptr_offset(leaf, slot)
2786                                         + sizeof(struct btrfs_root_ref);
2787                         item_len = btrfs_item_size_nr(leaf, slot)
2788                                         - sizeof(struct btrfs_root_ref);
2789                         read_extent_buffer(leaf, subvol_info->name,
2790                                            item_off, item_len);
2791                 } else {
2792                         ret = -ENOENT;
2793                         goto out;
2794                 }
2795         }
2796
2797         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2798                 ret = -EFAULT;
2799
2800 out:
2801         btrfs_put_root(root);
2802 out_free:
2803         btrfs_free_path(path);
2804         kfree(subvol_info);
2805         return ret;
2806 }
2807
2808 /*
2809  * Return ROOT_REF information of the subvolume containing this inode
2810  * except the subvolume name.
2811  */
2812 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2813 {
2814         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2815         struct btrfs_root_ref *rref;
2816         struct btrfs_root *root;
2817         struct btrfs_path *path;
2818         struct btrfs_key key;
2819         struct extent_buffer *leaf;
2820         struct inode *inode;
2821         u64 objectid;
2822         int slot;
2823         int ret;
2824         u8 found;
2825
2826         path = btrfs_alloc_path();
2827         if (!path)
2828                 return -ENOMEM;
2829
2830         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2831         if (IS_ERR(rootrefs)) {
2832                 btrfs_free_path(path);
2833                 return PTR_ERR(rootrefs);
2834         }
2835
2836         inode = file_inode(file);
2837         root = BTRFS_I(inode)->root->fs_info->tree_root;
2838         objectid = BTRFS_I(inode)->root->root_key.objectid;
2839
2840         key.objectid = objectid;
2841         key.type = BTRFS_ROOT_REF_KEY;
2842         key.offset = rootrefs->min_treeid;
2843         found = 0;
2844
2845         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2846         if (ret < 0) {
2847                 goto out;
2848         } else if (path->slots[0] >=
2849                    btrfs_header_nritems(path->nodes[0])) {
2850                 ret = btrfs_next_leaf(root, path);
2851                 if (ret < 0) {
2852                         goto out;
2853                 } else if (ret > 0) {
2854                         ret = -EUCLEAN;
2855                         goto out;
2856                 }
2857         }
2858         while (1) {
2859                 leaf = path->nodes[0];
2860                 slot = path->slots[0];
2861
2862                 btrfs_item_key_to_cpu(leaf, &key, slot);
2863                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2864                         ret = 0;
2865                         goto out;
2866                 }
2867
2868                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2869                         ret = -EOVERFLOW;
2870                         goto out;
2871                 }
2872
2873                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2874                 rootrefs->rootref[found].treeid = key.offset;
2875                 rootrefs->rootref[found].dirid =
2876                                   btrfs_root_ref_dirid(leaf, rref);
2877                 found++;
2878
2879                 ret = btrfs_next_item(root, path);
2880                 if (ret < 0) {
2881                         goto out;
2882                 } else if (ret > 0) {
2883                         ret = -EUCLEAN;
2884                         goto out;
2885                 }
2886         }
2887
2888 out:
2889         if (!ret || ret == -EOVERFLOW) {
2890                 rootrefs->num_items = found;
2891                 /* update min_treeid for next search */
2892                 if (found)
2893                         rootrefs->min_treeid =
2894                                 rootrefs->rootref[found - 1].treeid + 1;
2895                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2896                         ret = -EFAULT;
2897         }
2898
2899         kfree(rootrefs);
2900         btrfs_free_path(path);
2901
2902         return ret;
2903 }
2904
2905 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2906                                              void __user *arg,
2907                                              bool destroy_v2)
2908 {
2909         struct dentry *parent = file->f_path.dentry;
2910         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2911         struct dentry *dentry;
2912         struct inode *dir = d_inode(parent);
2913         struct inode *inode;
2914         struct btrfs_root *root = BTRFS_I(dir)->root;
2915         struct btrfs_root *dest = NULL;
2916         struct btrfs_ioctl_vol_args *vol_args = NULL;
2917         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2918         char *subvol_name, *subvol_name_ptr = NULL;
2919         int subvol_namelen;
2920         int err = 0;
2921         bool destroy_parent = false;
2922
2923         if (destroy_v2) {
2924                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2925                 if (IS_ERR(vol_args2))
2926                         return PTR_ERR(vol_args2);
2927
2928                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2929                         err = -EOPNOTSUPP;
2930                         goto out;
2931                 }
2932
2933                 /*
2934                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2935                  * name, same as v1 currently does.
2936                  */
2937                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2938                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2939                         subvol_name = vol_args2->name;
2940
2941                         err = mnt_want_write_file(file);
2942                         if (err)
2943                                 goto out;
2944                 } else {
2945                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2946                                 err = -EINVAL;
2947                                 goto out;
2948                         }
2949
2950                         err = mnt_want_write_file(file);
2951                         if (err)
2952                                 goto out;
2953
2954                         dentry = btrfs_get_dentry(fs_info->sb,
2955                                         BTRFS_FIRST_FREE_OBJECTID,
2956                                         vol_args2->subvolid, 0, 0);
2957                         if (IS_ERR(dentry)) {
2958                                 err = PTR_ERR(dentry);
2959                                 goto out_drop_write;
2960                         }
2961
2962                         /*
2963                          * Change the default parent since the subvolume being
2964                          * deleted can be outside of the current mount point.
2965                          */
2966                         parent = btrfs_get_parent(dentry);
2967
2968                         /*
2969                          * At this point dentry->d_name can point to '/' if the
2970                          * subvolume we want to destroy is outsite of the
2971                          * current mount point, so we need to release the
2972                          * current dentry and execute the lookup to return a new
2973                          * one with ->d_name pointing to the
2974                          * <mount point>/subvol_name.
2975                          */
2976                         dput(dentry);
2977                         if (IS_ERR(parent)) {
2978                                 err = PTR_ERR(parent);
2979                                 goto out_drop_write;
2980                         }
2981                         dir = d_inode(parent);
2982
2983                         /*
2984                          * If v2 was used with SPEC_BY_ID, a new parent was
2985                          * allocated since the subvolume can be outside of the
2986                          * current mount point. Later on we need to release this
2987                          * new parent dentry.
2988                          */
2989                         destroy_parent = true;
2990
2991                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2992                                                 fs_info, vol_args2->subvolid);
2993                         if (IS_ERR(subvol_name_ptr)) {
2994                                 err = PTR_ERR(subvol_name_ptr);
2995                                 goto free_parent;
2996                         }
2997                         /* subvol_name_ptr is already NULL termined */
2998                         subvol_name = (char *)kbasename(subvol_name_ptr);
2999                 }
3000         } else {
3001                 vol_args = memdup_user(arg, sizeof(*vol_args));
3002                 if (IS_ERR(vol_args))
3003                         return PTR_ERR(vol_args);
3004
3005                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3006                 subvol_name = vol_args->name;
3007
3008                 err = mnt_want_write_file(file);
3009                 if (err)
3010                         goto out;
3011         }
3012
3013         subvol_namelen = strlen(subvol_name);
3014
3015         if (strchr(subvol_name, '/') ||
3016             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3017                 err = -EINVAL;
3018                 goto free_subvol_name;
3019         }
3020
3021         if (!S_ISDIR(dir->i_mode)) {
3022                 err = -ENOTDIR;
3023                 goto free_subvol_name;
3024         }
3025
3026         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3027         if (err == -EINTR)
3028                 goto free_subvol_name;
3029         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
3030         if (IS_ERR(dentry)) {
3031                 err = PTR_ERR(dentry);
3032                 goto out_unlock_dir;
3033         }
3034
3035         if (d_really_is_negative(dentry)) {
3036                 err = -ENOENT;
3037                 goto out_dput;
3038         }
3039
3040         inode = d_inode(dentry);
3041         dest = BTRFS_I(inode)->root;
3042         if (!capable(CAP_SYS_ADMIN)) {
3043                 /*
3044                  * Regular user.  Only allow this with a special mount
3045                  * option, when the user has write+exec access to the
3046                  * subvol root, and when rmdir(2) would have been
3047                  * allowed.
3048                  *
3049                  * Note that this is _not_ check that the subvol is
3050                  * empty or doesn't contain data that we wouldn't
3051                  * otherwise be able to delete.
3052                  *
3053                  * Users who want to delete empty subvols should try
3054                  * rmdir(2).
3055                  */
3056                 err = -EPERM;
3057                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3058                         goto out_dput;
3059
3060                 /*
3061                  * Do not allow deletion if the parent dir is the same
3062                  * as the dir to be deleted.  That means the ioctl
3063                  * must be called on the dentry referencing the root
3064                  * of the subvol, not a random directory contained
3065                  * within it.
3066                  */
3067                 err = -EINVAL;
3068                 if (root == dest)
3069                         goto out_dput;
3070
3071                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
3072                 if (err)
3073                         goto out_dput;
3074         }
3075
3076         /* check if subvolume may be deleted by a user */
3077         err = btrfs_may_delete(dir, dentry, 1);
3078         if (err)
3079                 goto out_dput;
3080
3081         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3082                 err = -EINVAL;
3083                 goto out_dput;
3084         }
3085
3086         inode_lock(inode);
3087         err = btrfs_delete_subvolume(dir, dentry);
3088         inode_unlock(inode);
3089         if (!err) {
3090                 fsnotify_rmdir(dir, dentry);
3091                 d_delete(dentry);
3092         }
3093
3094 out_dput:
3095         dput(dentry);
3096 out_unlock_dir:
3097         inode_unlock(dir);
3098 free_subvol_name:
3099         kfree(subvol_name_ptr);
3100 free_parent:
3101         if (destroy_parent)
3102                 dput(parent);
3103 out_drop_write:
3104         mnt_drop_write_file(file);
3105 out:
3106         kfree(vol_args2);
3107         kfree(vol_args);
3108         return err;
3109 }
3110
3111 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3112 {
3113         struct inode *inode = file_inode(file);
3114         struct btrfs_root *root = BTRFS_I(inode)->root;
3115         struct btrfs_ioctl_defrag_range_args *range;
3116         int ret;
3117
3118         ret = mnt_want_write_file(file);
3119         if (ret)
3120                 return ret;
3121
3122         if (btrfs_root_readonly(root)) {
3123                 ret = -EROFS;
3124                 goto out;
3125         }
3126
3127         switch (inode->i_mode & S_IFMT) {
3128         case S_IFDIR:
3129                 if (!capable(CAP_SYS_ADMIN)) {
3130                         ret = -EPERM;
3131                         goto out;
3132                 }
3133                 ret = btrfs_defrag_root(root);
3134                 break;
3135         case S_IFREG:
3136                 /*
3137                  * Note that this does not check the file descriptor for write
3138                  * access. This prevents defragmenting executables that are
3139                  * running and allows defrag on files open in read-only mode.
3140                  */
3141                 if (!capable(CAP_SYS_ADMIN) &&
3142                     inode_permission(inode, MAY_WRITE)) {
3143                         ret = -EPERM;
3144                         goto out;
3145                 }
3146
3147                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3148                 if (!range) {
3149                         ret = -ENOMEM;
3150                         goto out;
3151                 }
3152
3153                 if (argp) {
3154                         if (copy_from_user(range, argp,
3155                                            sizeof(*range))) {
3156                                 ret = -EFAULT;
3157                                 kfree(range);
3158                                 goto out;
3159                         }
3160                         /* compression requires us to start the IO */
3161                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3162                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3163                                 range->extent_thresh = (u32)-1;
3164                         }
3165                 } else {
3166                         /* the rest are all set to zero by kzalloc */
3167                         range->len = (u64)-1;
3168                 }
3169                 ret = btrfs_defrag_file(file_inode(file), file,
3170                                         range, BTRFS_OLDEST_GENERATION, 0);
3171                 if (ret > 0)
3172                         ret = 0;
3173                 kfree(range);
3174                 break;
3175         default:
3176                 ret = -EINVAL;
3177         }
3178 out:
3179         mnt_drop_write_file(file);
3180         return ret;
3181 }
3182
3183 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3184 {
3185         struct btrfs_ioctl_vol_args *vol_args;
3186         int ret;
3187
3188         if (!capable(CAP_SYS_ADMIN))
3189                 return -EPERM;
3190
3191         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3192                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3193
3194         vol_args = memdup_user(arg, sizeof(*vol_args));
3195         if (IS_ERR(vol_args)) {
3196                 ret = PTR_ERR(vol_args);
3197                 goto out;
3198         }
3199
3200         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3201         ret = btrfs_init_new_device(fs_info, vol_args->name);
3202
3203         if (!ret)
3204                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3205
3206         kfree(vol_args);
3207 out:
3208         btrfs_exclop_finish(fs_info);
3209         return ret;
3210 }
3211
3212 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3213 {
3214         struct inode *inode = file_inode(file);
3215         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3216         struct btrfs_ioctl_vol_args_v2 *vol_args;
3217         int ret;
3218
3219         if (!capable(CAP_SYS_ADMIN))
3220                 return -EPERM;
3221
3222         ret = mnt_want_write_file(file);
3223         if (ret)
3224                 return ret;
3225
3226         vol_args = memdup_user(arg, sizeof(*vol_args));
3227         if (IS_ERR(vol_args)) {
3228                 ret = PTR_ERR(vol_args);
3229                 goto err_drop;
3230         }
3231
3232         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3233                 ret = -EOPNOTSUPP;
3234                 goto out;
3235         }
3236
3237         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3238                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3239                 goto out;
3240         }
3241
3242         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3243                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3244         } else {
3245                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3246                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3247         }
3248         btrfs_exclop_finish(fs_info);
3249
3250         if (!ret) {
3251                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3252                         btrfs_info(fs_info, "device deleted: id %llu",
3253                                         vol_args->devid);
3254                 else
3255                         btrfs_info(fs_info, "device deleted: %s",
3256                                         vol_args->name);
3257         }
3258 out:
3259         kfree(vol_args);
3260 err_drop:
3261         mnt_drop_write_file(file);
3262         return ret;
3263 }
3264
3265 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3266 {
3267         struct inode *inode = file_inode(file);
3268         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3269         struct btrfs_ioctl_vol_args *vol_args;
3270         int ret;
3271
3272         if (!capable(CAP_SYS_ADMIN))
3273                 return -EPERM;
3274
3275         ret = mnt_want_write_file(file);
3276         if (ret)
3277                 return ret;
3278
3279         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3280                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3281                 goto out_drop_write;
3282         }
3283
3284         vol_args = memdup_user(arg, sizeof(*vol_args));
3285         if (IS_ERR(vol_args)) {
3286                 ret = PTR_ERR(vol_args);
3287                 goto out;
3288         }
3289
3290         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3291         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3292
3293         if (!ret)
3294                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3295         kfree(vol_args);
3296 out:
3297         btrfs_exclop_finish(fs_info);
3298 out_drop_write:
3299         mnt_drop_write_file(file);
3300
3301         return ret;
3302 }
3303
3304 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3305                                 void __user *arg)
3306 {
3307         struct btrfs_ioctl_fs_info_args *fi_args;
3308         struct btrfs_device *device;
3309         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3310         u64 flags_in;
3311         int ret = 0;
3312
3313         fi_args = memdup_user(arg, sizeof(*fi_args));
3314         if (IS_ERR(fi_args))
3315                 return PTR_ERR(fi_args);
3316
3317         flags_in = fi_args->flags;
3318         memset(fi_args, 0, sizeof(*fi_args));
3319
3320         rcu_read_lock();
3321         fi_args->num_devices = fs_devices->num_devices;
3322
3323         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3324                 if (device->devid > fi_args->max_id)
3325                         fi_args->max_id = device->devid;
3326         }
3327         rcu_read_unlock();
3328
3329         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3330         fi_args->nodesize = fs_info->nodesize;
3331         fi_args->sectorsize = fs_info->sectorsize;
3332         fi_args->clone_alignment = fs_info->sectorsize;
3333
3334         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3335                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3336                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3337                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3338         }
3339
3340         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3341                 fi_args->generation = fs_info->generation;
3342                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3343         }
3344
3345         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3346                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3347                        sizeof(fi_args->metadata_uuid));
3348                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3349         }
3350
3351         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3352                 ret = -EFAULT;
3353
3354         kfree(fi_args);
3355         return ret;
3356 }
3357
3358 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3359                                  void __user *arg)
3360 {
3361         struct btrfs_ioctl_dev_info_args *di_args;
3362         struct btrfs_device *dev;
3363         int ret = 0;
3364         char *s_uuid = NULL;
3365
3366         di_args = memdup_user(arg, sizeof(*di_args));
3367         if (IS_ERR(di_args))
3368                 return PTR_ERR(di_args);
3369
3370         if (!btrfs_is_empty_uuid(di_args->uuid))
3371                 s_uuid = di_args->uuid;
3372
3373         rcu_read_lock();
3374         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3375                                 NULL);
3376
3377         if (!dev) {
3378                 ret = -ENODEV;
3379                 goto out;
3380         }
3381
3382         di_args->devid = dev->devid;
3383         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3384         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3385         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3386         if (dev->name) {
3387                 strncpy(di_args->path, rcu_str_deref(dev->name),
3388                                 sizeof(di_args->path) - 1);
3389                 di_args->path[sizeof(di_args->path) - 1] = 0;
3390         } else {
3391                 di_args->path[0] = '\0';
3392         }
3393
3394 out:
3395         rcu_read_unlock();
3396         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3397                 ret = -EFAULT;
3398
3399         kfree(di_args);
3400         return ret;
3401 }
3402
3403 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3404 {
3405         struct inode *inode = file_inode(file);
3406         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3407         struct btrfs_root *root = BTRFS_I(inode)->root;
3408         struct btrfs_root *new_root;
3409         struct btrfs_dir_item *di;
3410         struct btrfs_trans_handle *trans;
3411         struct btrfs_path *path = NULL;
3412         struct btrfs_disk_key disk_key;
3413         u64 objectid = 0;
3414         u64 dir_id;
3415         int ret;
3416
3417         if (!capable(CAP_SYS_ADMIN))
3418                 return -EPERM;
3419
3420         ret = mnt_want_write_file(file);
3421         if (ret)
3422                 return ret;
3423
3424         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3425                 ret = -EFAULT;
3426                 goto out;
3427         }
3428
3429         if (!objectid)
3430                 objectid = BTRFS_FS_TREE_OBJECTID;
3431
3432         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3433         if (IS_ERR(new_root)) {
3434                 ret = PTR_ERR(new_root);
3435                 goto out;
3436         }
3437         if (!is_fstree(new_root->root_key.objectid)) {
3438                 ret = -ENOENT;
3439                 goto out_free;
3440         }
3441
3442         path = btrfs_alloc_path();
3443         if (!path) {
3444                 ret = -ENOMEM;
3445                 goto out_free;
3446         }
3447
3448         trans = btrfs_start_transaction(root, 1);
3449         if (IS_ERR(trans)) {
3450                 ret = PTR_ERR(trans);
3451                 goto out_free;
3452         }
3453
3454         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3455         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3456                                    dir_id, "default", 7, 1);
3457         if (IS_ERR_OR_NULL(di)) {
3458                 btrfs_release_path(path);
3459                 btrfs_end_transaction(trans);
3460                 btrfs_err(fs_info,
3461                           "Umm, you don't have the default diritem, this isn't going to work");
3462                 ret = -ENOENT;
3463                 goto out_free;
3464         }
3465
3466         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3467         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3468         btrfs_mark_buffer_dirty(path->nodes[0]);
3469         btrfs_release_path(path);
3470
3471         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3472         btrfs_end_transaction(trans);
3473 out_free:
3474         btrfs_put_root(new_root);
3475         btrfs_free_path(path);
3476 out:
3477         mnt_drop_write_file(file);
3478         return ret;
3479 }
3480
3481 static void get_block_group_info(struct list_head *groups_list,
3482                                  struct btrfs_ioctl_space_info *space)
3483 {
3484         struct btrfs_block_group *block_group;
3485
3486         space->total_bytes = 0;
3487         space->used_bytes = 0;
3488         space->flags = 0;
3489         list_for_each_entry(block_group, groups_list, list) {
3490                 space->flags = block_group->flags;
3491                 space->total_bytes += block_group->length;
3492                 space->used_bytes += block_group->used;
3493         }
3494 }
3495
3496 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3497                                    void __user *arg)
3498 {
3499         struct btrfs_ioctl_space_args space_args;
3500         struct btrfs_ioctl_space_info space;
3501         struct btrfs_ioctl_space_info *dest;
3502         struct btrfs_ioctl_space_info *dest_orig;
3503         struct btrfs_ioctl_space_info __user *user_dest;
3504         struct btrfs_space_info *info;
3505         static const u64 types[] = {
3506                 BTRFS_BLOCK_GROUP_DATA,
3507                 BTRFS_BLOCK_GROUP_SYSTEM,
3508                 BTRFS_BLOCK_GROUP_METADATA,
3509                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3510         };
3511         int num_types = 4;
3512         int alloc_size;
3513         int ret = 0;
3514         u64 slot_count = 0;
3515         int i, c;
3516
3517         if (copy_from_user(&space_args,
3518                            (struct btrfs_ioctl_space_args __user *)arg,
3519                            sizeof(space_args)))
3520                 return -EFAULT;
3521
3522         for (i = 0; i < num_types; i++) {
3523                 struct btrfs_space_info *tmp;
3524
3525                 info = NULL;
3526                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3527                         if (tmp->flags == types[i]) {
3528                                 info = tmp;
3529                                 break;
3530                         }
3531                 }
3532
3533                 if (!info)
3534                         continue;
3535
3536                 down_read(&info->groups_sem);
3537                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3538                         if (!list_empty(&info->block_groups[c]))
3539                                 slot_count++;
3540                 }
3541                 up_read(&info->groups_sem);
3542         }
3543
3544         /*
3545          * Global block reserve, exported as a space_info
3546          */
3547         slot_count++;
3548
3549         /* space_slots == 0 means they are asking for a count */
3550         if (space_args.space_slots == 0) {
3551                 space_args.total_spaces = slot_count;
3552                 goto out;
3553         }
3554
3555         slot_count = min_t(u64, space_args.space_slots, slot_count);
3556
3557         alloc_size = sizeof(*dest) * slot_count;
3558
3559         /* we generally have at most 6 or so space infos, one for each raid
3560          * level.  So, a whole page should be more than enough for everyone
3561          */
3562         if (alloc_size > PAGE_SIZE)
3563                 return -ENOMEM;
3564
3565         space_args.total_spaces = 0;
3566         dest = kmalloc(alloc_size, GFP_KERNEL);
3567         if (!dest)
3568                 return -ENOMEM;
3569         dest_orig = dest;
3570
3571         /* now we have a buffer to copy into */
3572         for (i = 0; i < num_types; i++) {
3573                 struct btrfs_space_info *tmp;
3574
3575                 if (!slot_count)
3576                         break;
3577
3578                 info = NULL;
3579                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3580                         if (tmp->flags == types[i]) {
3581                                 info = tmp;
3582                                 break;
3583                         }
3584                 }
3585
3586                 if (!info)
3587                         continue;
3588                 down_read(&info->groups_sem);
3589                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3590                         if (!list_empty(&info->block_groups[c])) {
3591                                 get_block_group_info(&info->block_groups[c],
3592                                                      &space);
3593                                 memcpy(dest, &space, sizeof(space));
3594                                 dest++;
3595                                 space_args.total_spaces++;
3596                                 slot_count--;
3597                         }
3598                         if (!slot_count)
3599                                 break;
3600                 }
3601                 up_read(&info->groups_sem);
3602         }
3603
3604         /*
3605          * Add global block reserve
3606          */
3607         if (slot_count) {
3608                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3609
3610                 spin_lock(&block_rsv->lock);
3611                 space.total_bytes = block_rsv->size;
3612                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3613                 spin_unlock(&block_rsv->lock);
3614                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3615                 memcpy(dest, &space, sizeof(space));
3616                 space_args.total_spaces++;
3617         }
3618
3619         user_dest = (struct btrfs_ioctl_space_info __user *)
3620                 (arg + sizeof(struct btrfs_ioctl_space_args));
3621
3622         if (copy_to_user(user_dest, dest_orig, alloc_size))
3623                 ret = -EFAULT;
3624
3625         kfree(dest_orig);
3626 out:
3627         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3628                 ret = -EFAULT;
3629
3630         return ret;
3631 }
3632
3633 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3634                                             void __user *argp)
3635 {
3636         struct btrfs_trans_handle *trans;
3637         u64 transid;
3638         int ret;
3639
3640         trans = btrfs_attach_transaction_barrier(root);
3641         if (IS_ERR(trans)) {
3642                 if (PTR_ERR(trans) != -ENOENT)
3643                         return PTR_ERR(trans);
3644
3645                 /* No running transaction, don't bother */
3646                 transid = root->fs_info->last_trans_committed;
3647                 goto out;
3648         }
3649         transid = trans->transid;
3650         ret = btrfs_commit_transaction_async(trans, 0);
3651         if (ret) {
3652                 btrfs_end_transaction(trans);
3653                 return ret;
3654         }
3655 out:
3656         if (argp)
3657                 if (copy_to_user(argp, &transid, sizeof(transid)))
3658                         return -EFAULT;
3659         return 0;
3660 }
3661
3662 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3663                                            void __user *argp)
3664 {
3665         u64 transid;
3666
3667         if (argp) {
3668                 if (copy_from_user(&transid, argp, sizeof(transid)))
3669                         return -EFAULT;
3670         } else {
3671                 transid = 0;  /* current trans */
3672         }
3673         return btrfs_wait_for_commit(fs_info, transid);
3674 }
3675
3676 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3677 {
3678         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3679         struct btrfs_ioctl_scrub_args *sa;
3680         int ret;
3681
3682         if (!capable(CAP_SYS_ADMIN))
3683                 return -EPERM;
3684
3685         sa = memdup_user(arg, sizeof(*sa));
3686         if (IS_ERR(sa))
3687                 return PTR_ERR(sa);
3688
3689         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3690                 ret = mnt_want_write_file(file);
3691                 if (ret)
3692                         goto out;
3693         }
3694
3695         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3696                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3697                               0);
3698
3699         /*
3700          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3701          * error. This is important as it allows user space to know how much
3702          * progress scrub has done. For example, if scrub is canceled we get
3703          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3704          * space. Later user space can inspect the progress from the structure
3705          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3706          * previously (btrfs-progs does this).
3707          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3708          * then return -EFAULT to signal the structure was not copied or it may
3709          * be corrupt and unreliable due to a partial copy.
3710          */
3711         if (copy_to_user(arg, sa, sizeof(*sa)))
3712                 ret = -EFAULT;
3713
3714         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3715                 mnt_drop_write_file(file);
3716 out:
3717         kfree(sa);
3718         return ret;
3719 }
3720
3721 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3722 {
3723         if (!capable(CAP_SYS_ADMIN))
3724                 return -EPERM;
3725
3726         return btrfs_scrub_cancel(fs_info);
3727 }
3728
3729 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3730                                        void __user *arg)
3731 {
3732         struct btrfs_ioctl_scrub_args *sa;
3733         int ret;
3734
3735         if (!capable(CAP_SYS_ADMIN))
3736                 return -EPERM;
3737
3738         sa = memdup_user(arg, sizeof(*sa));
3739         if (IS_ERR(sa))
3740                 return PTR_ERR(sa);
3741
3742         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3743
3744         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3745                 ret = -EFAULT;
3746
3747         kfree(sa);
3748         return ret;
3749 }
3750
3751 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3752                                       void __user *arg)
3753 {
3754         struct btrfs_ioctl_get_dev_stats *sa;
3755         int ret;
3756
3757         sa = memdup_user(arg, sizeof(*sa));
3758         if (IS_ERR(sa))
3759                 return PTR_ERR(sa);
3760
3761         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3762                 kfree(sa);
3763                 return -EPERM;
3764         }
3765
3766         ret = btrfs_get_dev_stats(fs_info, sa);
3767
3768         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3769                 ret = -EFAULT;
3770
3771         kfree(sa);
3772         return ret;
3773 }
3774
3775 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3776                                     void __user *arg)
3777 {
3778         struct btrfs_ioctl_dev_replace_args *p;
3779         int ret;
3780
3781         if (!capable(CAP_SYS_ADMIN))
3782                 return -EPERM;
3783
3784         p = memdup_user(arg, sizeof(*p));
3785         if (IS_ERR(p))
3786                 return PTR_ERR(p);
3787
3788         switch (p->cmd) {
3789         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3790                 if (sb_rdonly(fs_info->sb)) {
3791                         ret = -EROFS;
3792                         goto out;
3793                 }
3794                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3795                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3796                 } else {
3797                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3798                         btrfs_exclop_finish(fs_info);
3799                 }
3800                 break;
3801         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3802                 btrfs_dev_replace_status(fs_info, p);
3803                 ret = 0;
3804                 break;
3805         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3806                 p->result = btrfs_dev_replace_cancel(fs_info);
3807                 ret = 0;
3808                 break;
3809         default:
3810                 ret = -EINVAL;
3811                 break;
3812         }
3813
3814         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3815                 ret = -EFAULT;
3816 out:
3817         kfree(p);
3818         return ret;
3819 }
3820
3821 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3822 {
3823         int ret = 0;
3824         int i;
3825         u64 rel_ptr;
3826         int size;
3827         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3828         struct inode_fs_paths *ipath = NULL;
3829         struct btrfs_path *path;
3830
3831         if (!capable(CAP_DAC_READ_SEARCH))
3832                 return -EPERM;
3833
3834         path = btrfs_alloc_path();
3835         if (!path) {
3836                 ret = -ENOMEM;
3837                 goto out;
3838         }
3839
3840         ipa = memdup_user(arg, sizeof(*ipa));
3841         if (IS_ERR(ipa)) {
3842                 ret = PTR_ERR(ipa);
3843                 ipa = NULL;
3844                 goto out;
3845         }
3846
3847         size = min_t(u32, ipa->size, 4096);
3848         ipath = init_ipath(size, root, path);
3849         if (IS_ERR(ipath)) {
3850                 ret = PTR_ERR(ipath);
3851                 ipath = NULL;
3852                 goto out;
3853         }
3854
3855         ret = paths_from_inode(ipa->inum, ipath);
3856         if (ret < 0)
3857                 goto out;
3858
3859         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3860                 rel_ptr = ipath->fspath->val[i] -
3861                           (u64)(unsigned long)ipath->fspath->val;
3862                 ipath->fspath->val[i] = rel_ptr;
3863         }
3864
3865         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3866                            ipath->fspath, size);
3867         if (ret) {
3868                 ret = -EFAULT;
3869                 goto out;
3870         }
3871
3872 out:
3873         btrfs_free_path(path);
3874         free_ipath(ipath);
3875         kfree(ipa);
3876
3877         return ret;
3878 }
3879
3880 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3881 {
3882         struct btrfs_data_container *inodes = ctx;
3883         const size_t c = 3 * sizeof(u64);
3884
3885         if (inodes->bytes_left >= c) {
3886                 inodes->bytes_left -= c;
3887                 inodes->val[inodes->elem_cnt] = inum;
3888                 inodes->val[inodes->elem_cnt + 1] = offset;
3889                 inodes->val[inodes->elem_cnt + 2] = root;
3890                 inodes->elem_cnt += 3;
3891         } else {
3892                 inodes->bytes_missing += c - inodes->bytes_left;
3893                 inodes->bytes_left = 0;
3894                 inodes->elem_missed += 3;
3895         }
3896
3897         return 0;
3898 }
3899
3900 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3901                                         void __user *arg, int version)
3902 {
3903         int ret = 0;
3904         int size;
3905         struct btrfs_ioctl_logical_ino_args *loi;
3906         struct btrfs_data_container *inodes = NULL;
3907         struct btrfs_path *path = NULL;
3908         bool ignore_offset;
3909
3910         if (!capable(CAP_SYS_ADMIN))
3911                 return -EPERM;
3912
3913         loi = memdup_user(arg, sizeof(*loi));
3914         if (IS_ERR(loi))
3915                 return PTR_ERR(loi);
3916
3917         if (version == 1) {
3918                 ignore_offset = false;
3919                 size = min_t(u32, loi->size, SZ_64K);
3920         } else {
3921                 /* All reserved bits must be 0 for now */
3922                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3923                         ret = -EINVAL;
3924                         goto out_loi;
3925                 }
3926                 /* Only accept flags we have defined so far */
3927                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3928                         ret = -EINVAL;
3929                         goto out_loi;
3930                 }
3931                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3932                 size = min_t(u32, loi->size, SZ_16M);
3933         }
3934
3935         path = btrfs_alloc_path();
3936         if (!path) {
3937                 ret = -ENOMEM;
3938                 goto out;
3939         }
3940
3941         inodes = init_data_container(size);
3942         if (IS_ERR(inodes)) {
3943                 ret = PTR_ERR(inodes);
3944                 inodes = NULL;
3945                 goto out;
3946         }
3947
3948         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3949                                           build_ino_list, inodes, ignore_offset);
3950         if (ret == -EINVAL)
3951                 ret = -ENOENT;
3952         if (ret < 0)
3953                 goto out;
3954
3955         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3956                            size);
3957         if (ret)
3958                 ret = -EFAULT;
3959
3960 out:
3961         btrfs_free_path(path);
3962         kvfree(inodes);
3963 out_loi:
3964         kfree(loi);
3965
3966         return ret;
3967 }
3968
3969 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3970                                struct btrfs_ioctl_balance_args *bargs)
3971 {
3972         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3973
3974         bargs->flags = bctl->flags;
3975
3976         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3977                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3978         if (atomic_read(&fs_info->balance_pause_req))
3979                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3980         if (atomic_read(&fs_info->balance_cancel_req))
3981                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3982
3983         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3984         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3985         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3986
3987         spin_lock(&fs_info->balance_lock);
3988         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3989         spin_unlock(&fs_info->balance_lock);
3990 }
3991
3992 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3993 {
3994         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3995         struct btrfs_fs_info *fs_info = root->fs_info;
3996         struct btrfs_ioctl_balance_args *bargs;
3997         struct btrfs_balance_control *bctl;
3998         bool need_unlock; /* for mut. excl. ops lock */
3999         int ret;
4000
4001         if (!capable(CAP_SYS_ADMIN))
4002                 return -EPERM;
4003
4004         ret = mnt_want_write_file(file);
4005         if (ret)
4006                 return ret;
4007
4008 again:
4009         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4010                 mutex_lock(&fs_info->balance_mutex);
4011                 need_unlock = true;
4012                 goto locked;
4013         }
4014
4015         /*
4016          * mut. excl. ops lock is locked.  Three possibilities:
4017          *   (1) some other op is running
4018          *   (2) balance is running
4019          *   (3) balance is paused -- special case (think resume)
4020          */
4021         mutex_lock(&fs_info->balance_mutex);
4022         if (fs_info->balance_ctl) {
4023                 /* this is either (2) or (3) */
4024                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4025                         mutex_unlock(&fs_info->balance_mutex);
4026                         /*
4027                          * Lock released to allow other waiters to continue,
4028                          * we'll reexamine the status again.
4029                          */
4030                         mutex_lock(&fs_info->balance_mutex);
4031
4032                         if (fs_info->balance_ctl &&
4033                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4034                                 /* this is (3) */
4035                                 need_unlock = false;
4036                                 goto locked;
4037                         }
4038
4039                         mutex_unlock(&fs_info->balance_mutex);
4040                         goto again;
4041                 } else {
4042                         /* this is (2) */
4043                         mutex_unlock(&fs_info->balance_mutex);
4044                         ret = -EINPROGRESS;
4045                         goto out;
4046                 }
4047         } else {
4048                 /* this is (1) */
4049                 mutex_unlock(&fs_info->balance_mutex);
4050                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4051                 goto out;
4052         }
4053
4054 locked:
4055
4056         if (arg) {
4057                 bargs = memdup_user(arg, sizeof(*bargs));
4058                 if (IS_ERR(bargs)) {
4059                         ret = PTR_ERR(bargs);
4060                         goto out_unlock;
4061                 }
4062
4063                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4064                         if (!fs_info->balance_ctl) {
4065                                 ret = -ENOTCONN;
4066                                 goto out_bargs;
4067                         }
4068
4069                         bctl = fs_info->balance_ctl;
4070                         spin_lock(&fs_info->balance_lock);
4071                         bctl->flags |= BTRFS_BALANCE_RESUME;
4072                         spin_unlock(&fs_info->balance_lock);
4073
4074                         goto do_balance;
4075                 }
4076         } else {
4077                 bargs = NULL;
4078         }
4079
4080         if (fs_info->balance_ctl) {
4081                 ret = -EINPROGRESS;
4082                 goto out_bargs;
4083         }
4084
4085         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4086         if (!bctl) {
4087                 ret = -ENOMEM;
4088                 goto out_bargs;
4089         }
4090
4091         if (arg) {
4092                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4093                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4094                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4095
4096                 bctl->flags = bargs->flags;
4097         } else {
4098                 /* balance everything - no filters */
4099                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4100         }
4101
4102         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4103                 ret = -EINVAL;
4104                 goto out_bctl;
4105         }
4106
4107 do_balance:
4108         /*
4109          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4110          * bctl is freed in reset_balance_state, or, if restriper was paused
4111          * all the way until unmount, in free_fs_info.  The flag should be
4112          * cleared after reset_balance_state.
4113          */
4114         need_unlock = false;
4115
4116         ret = btrfs_balance(fs_info, bctl, bargs);
4117         bctl = NULL;
4118
4119         if ((ret == 0 || ret == -ECANCELED) && arg) {
4120                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4121                         ret = -EFAULT;
4122         }
4123
4124 out_bctl:
4125         kfree(bctl);
4126 out_bargs:
4127         kfree(bargs);
4128 out_unlock:
4129         mutex_unlock(&fs_info->balance_mutex);
4130         if (need_unlock)
4131                 btrfs_exclop_finish(fs_info);
4132 out:
4133         mnt_drop_write_file(file);
4134         return ret;
4135 }
4136
4137 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4138 {
4139         if (!capable(CAP_SYS_ADMIN))
4140                 return -EPERM;
4141
4142         switch (cmd) {
4143         case BTRFS_BALANCE_CTL_PAUSE:
4144                 return btrfs_pause_balance(fs_info);
4145         case BTRFS_BALANCE_CTL_CANCEL:
4146                 return btrfs_cancel_balance(fs_info);
4147         }
4148
4149         return -EINVAL;
4150 }
4151
4152 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4153                                          void __user *arg)
4154 {
4155         struct btrfs_ioctl_balance_args *bargs;
4156         int ret = 0;
4157
4158         if (!capable(CAP_SYS_ADMIN))
4159                 return -EPERM;
4160
4161         mutex_lock(&fs_info->balance_mutex);
4162         if (!fs_info->balance_ctl) {
4163                 ret = -ENOTCONN;
4164                 goto out;
4165         }
4166
4167         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4168         if (!bargs) {
4169                 ret = -ENOMEM;
4170                 goto out;
4171         }
4172
4173         btrfs_update_ioctl_balance_args(fs_info, bargs);
4174
4175         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4176                 ret = -EFAULT;
4177
4178         kfree(bargs);
4179 out:
4180         mutex_unlock(&fs_info->balance_mutex);
4181         return ret;
4182 }
4183
4184 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4185 {
4186         struct inode *inode = file_inode(file);
4187         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4188         struct btrfs_ioctl_quota_ctl_args *sa;
4189         int ret;
4190
4191         if (!capable(CAP_SYS_ADMIN))
4192                 return -EPERM;
4193
4194         ret = mnt_want_write_file(file);
4195         if (ret)
4196                 return ret;
4197
4198         sa = memdup_user(arg, sizeof(*sa));
4199         if (IS_ERR(sa)) {
4200                 ret = PTR_ERR(sa);
4201                 goto drop_write;
4202         }
4203
4204         down_write(&fs_info->subvol_sem);
4205
4206         switch (sa->cmd) {
4207         case BTRFS_QUOTA_CTL_ENABLE:
4208                 ret = btrfs_quota_enable(fs_info);
4209                 break;
4210         case BTRFS_QUOTA_CTL_DISABLE:
4211                 ret = btrfs_quota_disable(fs_info);
4212                 break;
4213         default:
4214                 ret = -EINVAL;
4215                 break;
4216         }
4217
4218         kfree(sa);
4219         up_write(&fs_info->subvol_sem);
4220 drop_write:
4221         mnt_drop_write_file(file);
4222         return ret;
4223 }
4224
4225 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4226 {
4227         struct inode *inode = file_inode(file);
4228         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4229         struct btrfs_root *root = BTRFS_I(inode)->root;
4230         struct btrfs_ioctl_qgroup_assign_args *sa;
4231         struct btrfs_trans_handle *trans;
4232         int ret;
4233         int err;
4234
4235         if (!capable(CAP_SYS_ADMIN))
4236                 return -EPERM;
4237
4238         ret = mnt_want_write_file(file);
4239         if (ret)
4240                 return ret;
4241
4242         sa = memdup_user(arg, sizeof(*sa));
4243         if (IS_ERR(sa)) {
4244                 ret = PTR_ERR(sa);
4245                 goto drop_write;
4246         }
4247
4248         trans = btrfs_join_transaction(root);
4249         if (IS_ERR(trans)) {
4250                 ret = PTR_ERR(trans);
4251                 goto out;
4252         }
4253
4254         if (sa->assign) {
4255                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4256         } else {
4257                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4258         }
4259
4260         /* update qgroup status and info */
4261         err = btrfs_run_qgroups(trans);
4262         if (err < 0)
4263                 btrfs_handle_fs_error(fs_info, err,
4264                                       "failed to update qgroup status and info");
4265         err = btrfs_end_transaction(trans);
4266         if (err && !ret)
4267                 ret = err;
4268
4269 out:
4270         kfree(sa);
4271 drop_write:
4272         mnt_drop_write_file(file);
4273         return ret;
4274 }
4275
4276 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4277 {
4278         struct inode *inode = file_inode(file);
4279         struct btrfs_root *root = BTRFS_I(inode)->root;
4280         struct btrfs_ioctl_qgroup_create_args *sa;
4281         struct btrfs_trans_handle *trans;
4282         int ret;
4283         int err;
4284
4285         if (!capable(CAP_SYS_ADMIN))
4286                 return -EPERM;
4287
4288         ret = mnt_want_write_file(file);
4289         if (ret)
4290                 return ret;
4291
4292         sa = memdup_user(arg, sizeof(*sa));
4293         if (IS_ERR(sa)) {
4294                 ret = PTR_ERR(sa);
4295                 goto drop_write;
4296         }
4297
4298         if (!sa->qgroupid) {
4299                 ret = -EINVAL;
4300                 goto out;
4301         }
4302
4303         trans = btrfs_join_transaction(root);
4304         if (IS_ERR(trans)) {
4305                 ret = PTR_ERR(trans);
4306                 goto out;
4307         }
4308
4309         if (sa->create) {
4310                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4311         } else {
4312                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4313         }
4314
4315         err = btrfs_end_transaction(trans);
4316         if (err && !ret)
4317                 ret = err;
4318
4319 out:
4320         kfree(sa);
4321 drop_write:
4322         mnt_drop_write_file(file);
4323         return ret;
4324 }
4325
4326 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4327 {
4328         struct inode *inode = file_inode(file);
4329         struct btrfs_root *root = BTRFS_I(inode)->root;
4330         struct btrfs_ioctl_qgroup_limit_args *sa;
4331         struct btrfs_trans_handle *trans;
4332         int ret;
4333         int err;
4334         u64 qgroupid;
4335
4336         if (!capable(CAP_SYS_ADMIN))
4337                 return -EPERM;
4338
4339         ret = mnt_want_write_file(file);
4340         if (ret)
4341                 return ret;
4342
4343         sa = memdup_user(arg, sizeof(*sa));
4344         if (IS_ERR(sa)) {
4345                 ret = PTR_ERR(sa);
4346                 goto drop_write;
4347         }
4348
4349         trans = btrfs_join_transaction(root);
4350         if (IS_ERR(trans)) {
4351                 ret = PTR_ERR(trans);
4352                 goto out;
4353         }
4354
4355         qgroupid = sa->qgroupid;
4356         if (!qgroupid) {
4357                 /* take the current subvol as qgroup */
4358                 qgroupid = root->root_key.objectid;
4359         }
4360
4361         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4362
4363         err = btrfs_end_transaction(trans);
4364         if (err && !ret)
4365                 ret = err;
4366
4367 out:
4368         kfree(sa);
4369 drop_write:
4370         mnt_drop_write_file(file);
4371         return ret;
4372 }
4373
4374 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4375 {
4376         struct inode *inode = file_inode(file);
4377         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4378         struct btrfs_ioctl_quota_rescan_args *qsa;
4379         int ret;
4380
4381         if (!capable(CAP_SYS_ADMIN))
4382                 return -EPERM;
4383
4384         ret = mnt_want_write_file(file);
4385         if (ret)
4386                 return ret;
4387
4388         qsa = memdup_user(arg, sizeof(*qsa));
4389         if (IS_ERR(qsa)) {
4390                 ret = PTR_ERR(qsa);
4391                 goto drop_write;
4392         }
4393
4394         if (qsa->flags) {
4395                 ret = -EINVAL;
4396                 goto out;
4397         }
4398
4399         ret = btrfs_qgroup_rescan(fs_info);
4400
4401 out:
4402         kfree(qsa);
4403 drop_write:
4404         mnt_drop_write_file(file);
4405         return ret;
4406 }
4407
4408 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4409                                                 void __user *arg)
4410 {
4411         struct btrfs_ioctl_quota_rescan_args *qsa;
4412         int ret = 0;
4413
4414         if (!capable(CAP_SYS_ADMIN))
4415                 return -EPERM;
4416
4417         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4418         if (!qsa)
4419                 return -ENOMEM;
4420
4421         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4422                 qsa->flags = 1;
4423                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4424         }
4425
4426         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4427                 ret = -EFAULT;
4428
4429         kfree(qsa);
4430         return ret;
4431 }
4432
4433 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4434                                                 void __user *arg)
4435 {
4436         if (!capable(CAP_SYS_ADMIN))
4437                 return -EPERM;
4438
4439         return btrfs_qgroup_wait_for_completion(fs_info, true);
4440 }
4441
4442 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4443                                             struct btrfs_ioctl_received_subvol_args *sa)
4444 {
4445         struct inode *inode = file_inode(file);
4446         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4447         struct btrfs_root *root = BTRFS_I(inode)->root;
4448         struct btrfs_root_item *root_item = &root->root_item;
4449         struct btrfs_trans_handle *trans;
4450         struct timespec64 ct = current_time(inode);
4451         int ret = 0;
4452         int received_uuid_changed;
4453
4454         if (!inode_owner_or_capable(inode))
4455                 return -EPERM;
4456
4457         ret = mnt_want_write_file(file);
4458         if (ret < 0)
4459                 return ret;
4460
4461         down_write(&fs_info->subvol_sem);
4462
4463         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4464                 ret = -EINVAL;
4465                 goto out;
4466         }
4467
4468         if (btrfs_root_readonly(root)) {
4469                 ret = -EROFS;
4470                 goto out;
4471         }
4472
4473         /*
4474          * 1 - root item
4475          * 2 - uuid items (received uuid + subvol uuid)
4476          */
4477         trans = btrfs_start_transaction(root, 3);
4478         if (IS_ERR(trans)) {
4479                 ret = PTR_ERR(trans);
4480                 trans = NULL;
4481                 goto out;
4482         }
4483
4484         sa->rtransid = trans->transid;
4485         sa->rtime.sec = ct.tv_sec;
4486         sa->rtime.nsec = ct.tv_nsec;
4487
4488         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4489                                        BTRFS_UUID_SIZE);
4490         if (received_uuid_changed &&
4491             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4492                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4493                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4494                                           root->root_key.objectid);
4495                 if (ret && ret != -ENOENT) {
4496                         btrfs_abort_transaction(trans, ret);
4497                         btrfs_end_transaction(trans);
4498                         goto out;
4499                 }
4500         }
4501         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4502         btrfs_set_root_stransid(root_item, sa->stransid);
4503         btrfs_set_root_rtransid(root_item, sa->rtransid);
4504         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4505         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4506         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4507         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4508
4509         ret = btrfs_update_root(trans, fs_info->tree_root,
4510                                 &root->root_key, &root->root_item);
4511         if (ret < 0) {
4512                 btrfs_end_transaction(trans);
4513                 goto out;
4514         }
4515         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4516                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4517                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4518                                           root->root_key.objectid);
4519                 if (ret < 0 && ret != -EEXIST) {
4520                         btrfs_abort_transaction(trans, ret);
4521                         btrfs_end_transaction(trans);
4522                         goto out;
4523                 }
4524         }
4525         ret = btrfs_commit_transaction(trans);
4526 out:
4527         up_write(&fs_info->subvol_sem);
4528         mnt_drop_write_file(file);
4529         return ret;
4530 }
4531
4532 #ifdef CONFIG_64BIT
4533 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4534                                                 void __user *arg)
4535 {
4536         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4537         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4538         int ret = 0;
4539
4540         args32 = memdup_user(arg, sizeof(*args32));
4541         if (IS_ERR(args32))
4542                 return PTR_ERR(args32);
4543
4544         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4545         if (!args64) {
4546                 ret = -ENOMEM;
4547                 goto out;
4548         }
4549
4550         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4551         args64->stransid = args32->stransid;
4552         args64->rtransid = args32->rtransid;
4553         args64->stime.sec = args32->stime.sec;
4554         args64->stime.nsec = args32->stime.nsec;
4555         args64->rtime.sec = args32->rtime.sec;
4556         args64->rtime.nsec = args32->rtime.nsec;
4557         args64->flags = args32->flags;
4558
4559         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4560         if (ret)
4561                 goto out;
4562
4563         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4564         args32->stransid = args64->stransid;
4565         args32->rtransid = args64->rtransid;
4566         args32->stime.sec = args64->stime.sec;
4567         args32->stime.nsec = args64->stime.nsec;
4568         args32->rtime.sec = args64->rtime.sec;
4569         args32->rtime.nsec = args64->rtime.nsec;
4570         args32->flags = args64->flags;
4571
4572         ret = copy_to_user(arg, args32, sizeof(*args32));
4573         if (ret)
4574                 ret = -EFAULT;
4575
4576 out:
4577         kfree(args32);
4578         kfree(args64);
4579         return ret;
4580 }
4581 #endif
4582
4583 static long btrfs_ioctl_set_received_subvol(struct file *file,
4584                                             void __user *arg)
4585 {
4586         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4587         int ret = 0;
4588
4589         sa = memdup_user(arg, sizeof(*sa));
4590         if (IS_ERR(sa))
4591                 return PTR_ERR(sa);
4592
4593         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4594
4595         if (ret)
4596                 goto out;
4597
4598         ret = copy_to_user(arg, sa, sizeof(*sa));
4599         if (ret)
4600                 ret = -EFAULT;
4601
4602 out:
4603         kfree(sa);
4604         return ret;
4605 }
4606
4607 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4608                                         void __user *arg)
4609 {
4610         size_t len;
4611         int ret;
4612         char label[BTRFS_LABEL_SIZE];
4613
4614         spin_lock(&fs_info->super_lock);
4615         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4616         spin_unlock(&fs_info->super_lock);
4617
4618         len = strnlen(label, BTRFS_LABEL_SIZE);
4619
4620         if (len == BTRFS_LABEL_SIZE) {
4621                 btrfs_warn(fs_info,
4622                            "label is too long, return the first %zu bytes",
4623                            --len);
4624         }
4625
4626         ret = copy_to_user(arg, label, len);
4627
4628         return ret ? -EFAULT : 0;
4629 }
4630
4631 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4632 {
4633         struct inode *inode = file_inode(file);
4634         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4635         struct btrfs_root *root = BTRFS_I(inode)->root;
4636         struct btrfs_super_block *super_block = fs_info->super_copy;
4637         struct btrfs_trans_handle *trans;
4638         char label[BTRFS_LABEL_SIZE];
4639         int ret;
4640
4641         if (!capable(CAP_SYS_ADMIN))
4642                 return -EPERM;
4643
4644         if (copy_from_user(label, arg, sizeof(label)))
4645                 return -EFAULT;
4646
4647         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4648                 btrfs_err(fs_info,
4649                           "unable to set label with more than %d bytes",
4650                           BTRFS_LABEL_SIZE - 1);
4651                 return -EINVAL;
4652         }
4653
4654         ret = mnt_want_write_file(file);
4655         if (ret)
4656                 return ret;
4657
4658         trans = btrfs_start_transaction(root, 0);
4659         if (IS_ERR(trans)) {
4660                 ret = PTR_ERR(trans);
4661                 goto out_unlock;
4662         }
4663
4664         spin_lock(&fs_info->super_lock);
4665         strcpy(super_block->label, label);
4666         spin_unlock(&fs_info->super_lock);
4667         ret = btrfs_commit_transaction(trans);
4668
4669 out_unlock:
4670         mnt_drop_write_file(file);
4671         return ret;
4672 }
4673
4674 #define INIT_FEATURE_FLAGS(suffix) \
4675         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4676           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4677           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4678
4679 int btrfs_ioctl_get_supported_features(void __user *arg)
4680 {
4681         static const struct btrfs_ioctl_feature_flags features[3] = {
4682                 INIT_FEATURE_FLAGS(SUPP),
4683                 INIT_FEATURE_FLAGS(SAFE_SET),
4684                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4685         };
4686
4687         if (copy_to_user(arg, &features, sizeof(features)))
4688                 return -EFAULT;
4689
4690         return 0;
4691 }
4692
4693 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4694                                         void __user *arg)
4695 {
4696         struct btrfs_super_block *super_block = fs_info->super_copy;
4697         struct btrfs_ioctl_feature_flags features;
4698
4699         features.compat_flags = btrfs_super_compat_flags(super_block);
4700         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4701         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4702
4703         if (copy_to_user(arg, &features, sizeof(features)))
4704                 return -EFAULT;
4705
4706         return 0;
4707 }
4708
4709 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4710                               enum btrfs_feature_set set,
4711                               u64 change_mask, u64 flags, u64 supported_flags,
4712                               u64 safe_set, u64 safe_clear)
4713 {
4714         const char *type = btrfs_feature_set_name(set);
4715         char *names;
4716         u64 disallowed, unsupported;
4717         u64 set_mask = flags & change_mask;
4718         u64 clear_mask = ~flags & change_mask;
4719
4720         unsupported = set_mask & ~supported_flags;
4721         if (unsupported) {
4722                 names = btrfs_printable_features(set, unsupported);
4723                 if (names) {
4724                         btrfs_warn(fs_info,
4725                                    "this kernel does not support the %s feature bit%s",
4726                                    names, strchr(names, ',') ? "s" : "");
4727                         kfree(names);
4728                 } else
4729                         btrfs_warn(fs_info,
4730                                    "this kernel does not support %s bits 0x%llx",
4731                                    type, unsupported);
4732                 return -EOPNOTSUPP;
4733         }
4734
4735         disallowed = set_mask & ~safe_set;
4736         if (disallowed) {
4737                 names = btrfs_printable_features(set, disallowed);
4738                 if (names) {
4739                         btrfs_warn(fs_info,
4740                                    "can't set the %s feature bit%s while mounted",
4741                                    names, strchr(names, ',') ? "s" : "");
4742                         kfree(names);
4743                 } else
4744                         btrfs_warn(fs_info,
4745                                    "can't set %s bits 0x%llx while mounted",
4746                                    type, disallowed);
4747                 return -EPERM;
4748         }
4749
4750         disallowed = clear_mask & ~safe_clear;
4751         if (disallowed) {
4752                 names = btrfs_printable_features(set, disallowed);
4753                 if (names) {
4754                         btrfs_warn(fs_info,
4755                                    "can't clear the %s feature bit%s while mounted",
4756                                    names, strchr(names, ',') ? "s" : "");
4757                         kfree(names);
4758                 } else
4759                         btrfs_warn(fs_info,
4760                                    "can't clear %s bits 0x%llx while mounted",
4761                                    type, disallowed);
4762                 return -EPERM;
4763         }
4764
4765         return 0;
4766 }
4767
4768 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4769 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4770                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4771                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4772                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4773
4774 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4775 {
4776         struct inode *inode = file_inode(file);
4777         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4778         struct btrfs_root *root = BTRFS_I(inode)->root;
4779         struct btrfs_super_block *super_block = fs_info->super_copy;
4780         struct btrfs_ioctl_feature_flags flags[2];
4781         struct btrfs_trans_handle *trans;
4782         u64 newflags;
4783         int ret;
4784
4785         if (!capable(CAP_SYS_ADMIN))
4786                 return -EPERM;
4787
4788         if (copy_from_user(flags, arg, sizeof(flags)))
4789                 return -EFAULT;
4790
4791         /* Nothing to do */
4792         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4793             !flags[0].incompat_flags)
4794                 return 0;
4795
4796         ret = check_feature(fs_info, flags[0].compat_flags,
4797                             flags[1].compat_flags, COMPAT);
4798         if (ret)
4799                 return ret;
4800
4801         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4802                             flags[1].compat_ro_flags, COMPAT_RO);
4803         if (ret)
4804                 return ret;
4805
4806         ret = check_feature(fs_info, flags[0].incompat_flags,
4807                             flags[1].incompat_flags, INCOMPAT);
4808         if (ret)
4809                 return ret;
4810
4811         ret = mnt_want_write_file(file);
4812         if (ret)
4813                 return ret;
4814
4815         trans = btrfs_start_transaction(root, 0);
4816         if (IS_ERR(trans)) {
4817                 ret = PTR_ERR(trans);
4818                 goto out_drop_write;
4819         }
4820
4821         spin_lock(&fs_info->super_lock);
4822         newflags = btrfs_super_compat_flags(super_block);
4823         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4824         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4825         btrfs_set_super_compat_flags(super_block, newflags);
4826
4827         newflags = btrfs_super_compat_ro_flags(super_block);
4828         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4829         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4830         btrfs_set_super_compat_ro_flags(super_block, newflags);
4831
4832         newflags = btrfs_super_incompat_flags(super_block);
4833         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4834         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4835         btrfs_set_super_incompat_flags(super_block, newflags);
4836         spin_unlock(&fs_info->super_lock);
4837
4838         ret = btrfs_commit_transaction(trans);
4839 out_drop_write:
4840         mnt_drop_write_file(file);
4841
4842         return ret;
4843 }
4844
4845 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4846 {
4847         struct btrfs_ioctl_send_args *arg;
4848         int ret;
4849
4850         if (compat) {
4851 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4852                 struct btrfs_ioctl_send_args_32 args32;
4853
4854                 ret = copy_from_user(&args32, argp, sizeof(args32));
4855                 if (ret)
4856                         return -EFAULT;
4857                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4858                 if (!arg)
4859                         return -ENOMEM;
4860                 arg->send_fd = args32.send_fd;
4861                 arg->clone_sources_count = args32.clone_sources_count;
4862                 arg->clone_sources = compat_ptr(args32.clone_sources);
4863                 arg->parent_root = args32.parent_root;
4864                 arg->flags = args32.flags;
4865                 memcpy(arg->reserved, args32.reserved,
4866                        sizeof(args32.reserved));
4867 #else
4868                 return -ENOTTY;
4869 #endif
4870         } else {
4871                 arg = memdup_user(argp, sizeof(*arg));
4872                 if (IS_ERR(arg))
4873                         return PTR_ERR(arg);
4874         }
4875         ret = btrfs_ioctl_send(file, arg);
4876         kfree(arg);
4877         return ret;
4878 }
4879
4880 long btrfs_ioctl(struct file *file, unsigned int
4881                 cmd, unsigned long arg)
4882 {
4883         struct inode *inode = file_inode(file);
4884         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4885         struct btrfs_root *root = BTRFS_I(inode)->root;
4886         void __user *argp = (void __user *)arg;
4887
4888         switch (cmd) {
4889         case FS_IOC_GETFLAGS:
4890                 return btrfs_ioctl_getflags(file, argp);
4891         case FS_IOC_SETFLAGS:
4892                 return btrfs_ioctl_setflags(file, argp);
4893         case FS_IOC_GETVERSION:
4894                 return btrfs_ioctl_getversion(file, argp);
4895         case FS_IOC_GETFSLABEL:
4896                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4897         case FS_IOC_SETFSLABEL:
4898                 return btrfs_ioctl_set_fslabel(file, argp);
4899         case FITRIM:
4900                 return btrfs_ioctl_fitrim(fs_info, argp);
4901         case BTRFS_IOC_SNAP_CREATE:
4902                 return btrfs_ioctl_snap_create(file, argp, 0);
4903         case BTRFS_IOC_SNAP_CREATE_V2:
4904                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4905         case BTRFS_IOC_SUBVOL_CREATE:
4906                 return btrfs_ioctl_snap_create(file, argp, 1);
4907         case BTRFS_IOC_SUBVOL_CREATE_V2:
4908                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4909         case BTRFS_IOC_SNAP_DESTROY:
4910                 return btrfs_ioctl_snap_destroy(file, argp, false);
4911         case BTRFS_IOC_SNAP_DESTROY_V2:
4912                 return btrfs_ioctl_snap_destroy(file, argp, true);
4913         case BTRFS_IOC_SUBVOL_GETFLAGS:
4914                 return btrfs_ioctl_subvol_getflags(file, argp);
4915         case BTRFS_IOC_SUBVOL_SETFLAGS:
4916                 return btrfs_ioctl_subvol_setflags(file, argp);
4917         case BTRFS_IOC_DEFAULT_SUBVOL:
4918                 return btrfs_ioctl_default_subvol(file, argp);
4919         case BTRFS_IOC_DEFRAG:
4920                 return btrfs_ioctl_defrag(file, NULL);
4921         case BTRFS_IOC_DEFRAG_RANGE:
4922                 return btrfs_ioctl_defrag(file, argp);
4923         case BTRFS_IOC_RESIZE:
4924                 return btrfs_ioctl_resize(file, argp);
4925         case BTRFS_IOC_ADD_DEV:
4926                 return btrfs_ioctl_add_dev(fs_info, argp);
4927         case BTRFS_IOC_RM_DEV:
4928                 return btrfs_ioctl_rm_dev(file, argp);
4929         case BTRFS_IOC_RM_DEV_V2:
4930                 return btrfs_ioctl_rm_dev_v2(file, argp);
4931         case BTRFS_IOC_FS_INFO:
4932                 return btrfs_ioctl_fs_info(fs_info, argp);
4933         case BTRFS_IOC_DEV_INFO:
4934                 return btrfs_ioctl_dev_info(fs_info, argp);
4935         case BTRFS_IOC_BALANCE:
4936                 return btrfs_ioctl_balance(file, NULL);
4937         case BTRFS_IOC_TREE_SEARCH:
4938                 return btrfs_ioctl_tree_search(file, argp);
4939         case BTRFS_IOC_TREE_SEARCH_V2:
4940                 return btrfs_ioctl_tree_search_v2(file, argp);
4941         case BTRFS_IOC_INO_LOOKUP:
4942                 return btrfs_ioctl_ino_lookup(file, argp);
4943         case BTRFS_IOC_INO_PATHS:
4944                 return btrfs_ioctl_ino_to_path(root, argp);
4945         case BTRFS_IOC_LOGICAL_INO:
4946                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4947         case BTRFS_IOC_LOGICAL_INO_V2:
4948                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4949         case BTRFS_IOC_SPACE_INFO:
4950                 return btrfs_ioctl_space_info(fs_info, argp);
4951         case BTRFS_IOC_SYNC: {
4952                 int ret;
4953
4954                 ret = btrfs_start_delalloc_roots(fs_info, U64_MAX);
4955                 if (ret)
4956                         return ret;
4957                 ret = btrfs_sync_fs(inode->i_sb, 1);
4958                 /*
4959                  * The transaction thread may want to do more work,
4960                  * namely it pokes the cleaner kthread that will start
4961                  * processing uncleaned subvols.
4962                  */
4963                 wake_up_process(fs_info->transaction_kthread);
4964                 return ret;
4965         }
4966         case BTRFS_IOC_START_SYNC:
4967                 return btrfs_ioctl_start_sync(root, argp);
4968         case BTRFS_IOC_WAIT_SYNC:
4969                 return btrfs_ioctl_wait_sync(fs_info, argp);
4970         case BTRFS_IOC_SCRUB:
4971                 return btrfs_ioctl_scrub(file, argp);
4972         case BTRFS_IOC_SCRUB_CANCEL:
4973                 return btrfs_ioctl_scrub_cancel(fs_info);
4974         case BTRFS_IOC_SCRUB_PROGRESS:
4975                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4976         case BTRFS_IOC_BALANCE_V2:
4977                 return btrfs_ioctl_balance(file, argp);
4978         case BTRFS_IOC_BALANCE_CTL:
4979                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4980         case BTRFS_IOC_BALANCE_PROGRESS:
4981                 return btrfs_ioctl_balance_progress(fs_info, argp);
4982         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4983                 return btrfs_ioctl_set_received_subvol(file, argp);
4984 #ifdef CONFIG_64BIT
4985         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4986                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4987 #endif
4988         case BTRFS_IOC_SEND:
4989                 return _btrfs_ioctl_send(file, argp, false);
4990 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4991         case BTRFS_IOC_SEND_32:
4992                 return _btrfs_ioctl_send(file, argp, true);
4993 #endif
4994         case BTRFS_IOC_GET_DEV_STATS:
4995                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4996         case BTRFS_IOC_QUOTA_CTL:
4997                 return btrfs_ioctl_quota_ctl(file, argp);
4998         case BTRFS_IOC_QGROUP_ASSIGN:
4999                 return btrfs_ioctl_qgroup_assign(file, argp);
5000         case BTRFS_IOC_QGROUP_CREATE:
5001                 return btrfs_ioctl_qgroup_create(file, argp);
5002         case BTRFS_IOC_QGROUP_LIMIT:
5003                 return btrfs_ioctl_qgroup_limit(file, argp);
5004         case BTRFS_IOC_QUOTA_RESCAN:
5005                 return btrfs_ioctl_quota_rescan(file, argp);
5006         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5007                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5008         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5009                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5010         case BTRFS_IOC_DEV_REPLACE:
5011                 return btrfs_ioctl_dev_replace(fs_info, argp);
5012         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5013                 return btrfs_ioctl_get_supported_features(argp);
5014         case BTRFS_IOC_GET_FEATURES:
5015                 return btrfs_ioctl_get_features(fs_info, argp);
5016         case BTRFS_IOC_SET_FEATURES:
5017                 return btrfs_ioctl_set_features(file, argp);
5018         case FS_IOC_FSGETXATTR:
5019                 return btrfs_ioctl_fsgetxattr(file, argp);
5020         case FS_IOC_FSSETXATTR:
5021                 return btrfs_ioctl_fssetxattr(file, argp);
5022         case BTRFS_IOC_GET_SUBVOL_INFO:
5023                 return btrfs_ioctl_get_subvol_info(file, argp);
5024         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5025                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5026         case BTRFS_IOC_INO_LOOKUP_USER:
5027                 return btrfs_ioctl_ino_lookup_user(file, argp);
5028         }
5029
5030         return -ENOTTY;
5031 }
5032
5033 #ifdef CONFIG_COMPAT
5034 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5035 {
5036         /*
5037          * These all access 32-bit values anyway so no further
5038          * handling is necessary.
5039          */
5040         switch (cmd) {
5041         case FS_IOC32_GETFLAGS:
5042                 cmd = FS_IOC_GETFLAGS;
5043                 break;
5044         case FS_IOC32_SETFLAGS:
5045                 cmd = FS_IOC_SETFLAGS;
5046                 break;
5047         case FS_IOC32_GETVERSION:
5048                 cmd = FS_IOC_GETVERSION;
5049                 break;
5050         }
5051
5052         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5053 }
5054 #endif