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 "inode-map.h"
  38 #include "backref.h"
  39 #include "rcu-string.h"
  40 #include "send.h"
  41 #include "dev-replace.h"
  42 #include "props.h"
  43 #include "sysfs.h"
  44 #include "qgroup.h"
  45 #include "tree-log.h"
  46 #include "compression.h"
  47 #include "space-info.h"
  48 #include "delalloc-space.h"
  49 #include "block-group.h"
  50
  51 #ifdef CONFIG_64BIT
  52 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  53  * structures are incorrect, as the timespec structure from userspace
  54  * is 4 bytes too small. We define these alternatives here to teach
  55  * the kernel about the 32-bit struct packing.
  56  */
  57 struct btrfs_ioctl_timespec_32 {
  58         __u64 sec;
  59         __u32 nsec;
  60 } __attribute__ ((__packed__));
  61
  62 struct btrfs_ioctl_received_subvol_args_32 {
  63         char    uuid[BTRFS_UUID_SIZE];  /* in */
  64         __u64   stransid;               /* in */
  65         __u64   rtransid;               /* out */
  66         struct btrfs_ioctl_timespec_32 stime; /* in */
  67         struct btrfs_ioctl_timespec_32 rtime; /* out */
  68         __u64   flags;                  /* in */
  69         __u64   reserved[16];           /* in */
  70 } __attribute__ ((__packed__));
  71
  72 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  73                                 struct btrfs_ioctl_received_subvol_args_32)
  74 #endif
  75
  76 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  77 struct btrfs_ioctl_send_args_32 {
  78         __s64 send_fd;                  /* in */
  79         __u64 clone_sources_count;      /* in */
  80         compat_uptr_t clone_sources;    /* in */
  81         __u64 parent_root;              /* in */
  82         __u64 flags;                    /* in */
  83         __u64 reserved[4];              /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  87                                struct btrfs_ioctl_send_args_32)
  88 #endif
  89
  90 /* Mask out flags that are inappropriate for the given type of inode. */
  91 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  92                 unsigned int flags)
  93 {
  94         if (S_ISDIR(inode->i_mode))
  95                 return flags;
  96         else if (S_ISREG(inode->i_mode))
  97                 return flags & ~FS_DIRSYNC_FL;
  98         else
  99                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 100 }
 101
 102 /*
 103  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 104  * ioctl.
 105  */
 106 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if (flags & BTRFS_INODE_NOCOMPRESS)
 126                 iflags |= FS_NOCOMP_FL;
 127         else if (flags & BTRFS_INODE_COMPRESS)
 128                 iflags |= FS_COMPR_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 137 {
 138         struct btrfs_inode *binode = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (binode->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (binode->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (binode->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (binode->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 158 {
 159         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 160         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 161
 162         if (copy_to_user(arg, &flags, sizeof(flags)))
 163                 return -EFAULT;
 164         return 0;
 165 }
 166
 167 /*
 168  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 169  * the old and new flags are not conflicting
 170  */
 171 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 172 {
 173         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 174                       FS_NOATIME_FL | FS_NODUMP_FL | \
 175                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 176                       FS_NOCOMP_FL | FS_COMPR_FL |
 177                       FS_NOCOW_FL))
 178                 return -EOPNOTSUPP;
 179
 180         /* COMPR and NOCOMP on new/old are valid */
 181         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 182                 return -EINVAL;
 183
 184         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 185                 return -EINVAL;
 186
 187         /* NOCOW and compression options are mutually exclusive */
 188         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 189                 return -EINVAL;
 190         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 191                 return -EINVAL;
 192
 193         return 0;
 194 }
 195
 196 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 197 {
 198         struct inode *inode = file_inode(file);
 199         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 200         struct btrfs_inode *binode = BTRFS_I(inode);
 201         struct btrfs_root *root = binode->root;
 202         struct btrfs_trans_handle *trans;
 203         unsigned int fsflags, old_fsflags;
 204         int ret;
 205         const char *comp = NULL;
 206         u32 binode_flags;
 207
 208         if (!inode_owner_or_capable(inode))
 209                 return -EPERM;
 210
 211         if (btrfs_root_readonly(root))
 212                 return -EROFS;
 213
 214         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 215                 return -EFAULT;
 216
 217         ret = mnt_want_write_file(file);
 218         if (ret)
 219                 return ret;
 220
 221         inode_lock(inode);
 222         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 223         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 224
 225         ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 226         if (ret)
 227                 goto out_unlock;
 228
 229         ret = check_fsflags(old_fsflags, fsflags);
 230         if (ret)
 231                 goto out_unlock;
 232
 233         binode_flags = binode->flags;
 234         if (fsflags & FS_SYNC_FL)
 235                 binode_flags |= BTRFS_INODE_SYNC;
 236         else
 237                 binode_flags &= ~BTRFS_INODE_SYNC;
 238         if (fsflags & FS_IMMUTABLE_FL)
 239                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 240         else
 241                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 242         if (fsflags & FS_APPEND_FL)
 243                 binode_flags |= BTRFS_INODE_APPEND;
 244         else
 245                 binode_flags &= ~BTRFS_INODE_APPEND;
 246         if (fsflags & FS_NODUMP_FL)
 247                 binode_flags |= BTRFS_INODE_NODUMP;
 248         else
 249                 binode_flags &= ~BTRFS_INODE_NODUMP;
 250         if (fsflags & FS_NOATIME_FL)
 251                 binode_flags |= BTRFS_INODE_NOATIME;
 252         else
 253                 binode_flags &= ~BTRFS_INODE_NOATIME;
 254         if (fsflags & FS_DIRSYNC_FL)
 255                 binode_flags |= BTRFS_INODE_DIRSYNC;
 256         else
 257                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 258         if (fsflags & FS_NOCOW_FL) {
 259                 if (S_ISREG(inode->i_mode)) {
 260                         /*
 261                          * It's safe to turn csums off here, no extents exist.
 262                          * Otherwise we want the flag to reflect the real COW
 263                          * status of the file and will not set it.
 264                          */
 265                         if (inode->i_size == 0)
 266                                 binode_flags |= BTRFS_INODE_NODATACOW |
 267                                                 BTRFS_INODE_NODATASUM;
 268                 } else {
 269                         binode_flags |= BTRFS_INODE_NODATACOW;
 270                 }
 271         } else {
 272                 /*
 273                  * Revert back under same assumptions as above
 274                  */
 275                 if (S_ISREG(inode->i_mode)) {
 276                         if (inode->i_size == 0)
 277                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 278                                                   BTRFS_INODE_NODATASUM);
 279                 } else {
 280                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 281                 }
 282         }
 283
 284         /*
 285          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 286          * flag may be changed automatically if compression code won't make
 287          * things smaller.
 288          */
 289         if (fsflags & FS_NOCOMP_FL) {
 290                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 291                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 292         } else if (fsflags & FS_COMPR_FL) {
 293
 294                 if (IS_SWAPFILE(inode)) {
 295                         ret = -ETXTBSY;
 296                         goto out_unlock;
 297                 }
 298
 299                 binode_flags |= BTRFS_INODE_COMPRESS;
 300                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 301
 302                 comp = btrfs_compress_type2str(fs_info->compress_type);
 303                 if (!comp || comp[0] == 0)
 304                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 305         } else {
 306                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 307         }
 308
 309         /*
 310          * 1 for inode item
 311          * 2 for properties
 312          */
 313         trans = btrfs_start_transaction(root, 3);
 314         if (IS_ERR(trans)) {
 315                 ret = PTR_ERR(trans);
 316                 goto out_unlock;
 317         }
 318
 319         if (comp) {
 320                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 321                                      strlen(comp), 0);
 322                 if (ret) {
 323                         btrfs_abort_transaction(trans, ret);
 324                         goto out_end_trans;
 325                 }
 326         } else {
 327                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 328                                      0, 0);
 329                 if (ret && ret != -ENODATA) {
 330                         btrfs_abort_transaction(trans, ret);
 331                         goto out_end_trans;
 332                 }
 333         }
 334
 335         binode->flags = binode_flags;
 336         btrfs_sync_inode_flags_to_i_flags(inode);
 337         inode_inc_iversion(inode);
 338         inode->i_ctime = current_time(inode);
 339         ret = btrfs_update_inode(trans, root, inode);
 340
 341  out_end_trans:
 342         btrfs_end_transaction(trans);
 343  out_unlock:
 344         inode_unlock(inode);
 345         mnt_drop_write_file(file);
 346         return ret;
 347 }
 348
 349 /*
 350  * Translate btrfs internal inode flags to xflags as expected by the
 351  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 352  * silently dropped.
 353  */
 354 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 355 {
 356         unsigned int xflags = 0;
 357
 358         if (flags & BTRFS_INODE_APPEND)
 359                 xflags |= FS_XFLAG_APPEND;
 360         if (flags & BTRFS_INODE_IMMUTABLE)
 361                 xflags |= FS_XFLAG_IMMUTABLE;
 362         if (flags & BTRFS_INODE_NOATIME)
 363                 xflags |= FS_XFLAG_NOATIME;
 364         if (flags & BTRFS_INODE_NODUMP)
 365                 xflags |= FS_XFLAG_NODUMP;
 366         if (flags & BTRFS_INODE_SYNC)
 367                 xflags |= FS_XFLAG_SYNC;
 368
 369         return xflags;
 370 }
 371
 372 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 373 static int check_xflags(unsigned int flags)
 374 {
 375         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 376                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 377                 return -EOPNOTSUPP;
 378         return 0;
 379 }
 380
 381 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 382                         enum btrfs_exclusive_operation type)
 383 {
 384         return !cmpxchg(&fs_info->exclusive_operation, BTRFS_EXCLOP_NONE, type);
 385 }
 386
 387 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 388 {
 389         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 390         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 391 }
 392
 393 /*
 394  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 395  * are zeroed.
 396  */
 397 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 398 {
 399         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 400         struct fsxattr fa;
 401
 402         simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 403         if (copy_to_user(arg, &fa, sizeof(fa)))
 404                 return -EFAULT;
 405
 406         return 0;
 407 }
 408
 409 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 410 {
 411         struct inode *inode = file_inode(file);
 412         struct btrfs_inode *binode = BTRFS_I(inode);
 413         struct btrfs_root *root = binode->root;
 414         struct btrfs_trans_handle *trans;
 415         struct fsxattr fa, old_fa;
 416         unsigned old_flags;
 417         unsigned old_i_flags;
 418         int ret = 0;
 419
 420         if (!inode_owner_or_capable(inode))
 421                 return -EPERM;
 422
 423         if (btrfs_root_readonly(root))
 424                 return -EROFS;
 425
 426         if (copy_from_user(&fa, arg, sizeof(fa)))
 427                 return -EFAULT;
 428
 429         ret = check_xflags(fa.fsx_xflags);
 430         if (ret)
 431                 return ret;
 432
 433         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 434                 return -EOPNOTSUPP;
 435
 436         ret = mnt_want_write_file(file);
 437         if (ret)
 438                 return ret;
 439
 440         inode_lock(inode);
 441
 442         old_flags = binode->flags;
 443         old_i_flags = inode->i_flags;
 444
 445         simple_fill_fsxattr(&old_fa,
 446                             btrfs_inode_flags_to_xflags(binode->flags));
 447         ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 448         if (ret)
 449                 goto out_unlock;
 450
 451         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 452                 binode->flags |= BTRFS_INODE_SYNC;
 453         else
 454                 binode->flags &= ~BTRFS_INODE_SYNC;
 455         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 456                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 457         else
 458                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 459         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 460                 binode->flags |= BTRFS_INODE_APPEND;
 461         else
 462                 binode->flags &= ~BTRFS_INODE_APPEND;
 463         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 464                 binode->flags |= BTRFS_INODE_NODUMP;
 465         else
 466                 binode->flags &= ~BTRFS_INODE_NODUMP;
 467         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 468                 binode->flags |= BTRFS_INODE_NOATIME;
 469         else
 470                 binode->flags &= ~BTRFS_INODE_NOATIME;
 471
 472         /* 1 item for the inode */
 473         trans = btrfs_start_transaction(root, 1);
 474         if (IS_ERR(trans)) {
 475                 ret = PTR_ERR(trans);
 476                 goto out_unlock;
 477         }
 478
 479         btrfs_sync_inode_flags_to_i_flags(inode);
 480         inode_inc_iversion(inode);
 481         inode->i_ctime = current_time(inode);
 482         ret = btrfs_update_inode(trans, root, inode);
 483
 484         btrfs_end_transaction(trans);
 485
 486 out_unlock:
 487         if (ret) {
 488                 binode->flags = old_flags;
 489                 inode->i_flags = old_i_flags;
 490         }
 491
 492         inode_unlock(inode);
 493         mnt_drop_write_file(file);
 494
 495         return ret;
 496 }
 497
 498 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 499 {
 500         struct inode *inode = file_inode(file);
 501
 502         return put_user(inode->i_generation, arg);
 503 }
 504
 505 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 506                                         void __user *arg)
 507 {
 508         struct btrfs_device *device;
 509         struct request_queue *q;
 510         struct fstrim_range range;
 511         u64 minlen = ULLONG_MAX;
 512         u64 num_devices = 0;
 513         int ret;
 514
 515         if (!capable(CAP_SYS_ADMIN))
 516                 return -EPERM;
 517
 518         /*
 519          * If the fs is mounted with nologreplay, which requires it to be
 520          * mounted in RO mode as well, we can not allow discard on free space
 521          * inside block groups, because log trees refer to extents that are not
 522          * pinned in a block group's free space cache (pinning the extents is
 523          * precisely the first phase of replaying a log tree).
 524          */
 525         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 526                 return -EROFS;
 527
 528         rcu_read_lock();
 529         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 530                                 dev_list) {
 531                 if (!device->bdev)
 532                         continue;
 533                 q = bdev_get_queue(device->bdev);
 534                 if (blk_queue_discard(q)) {
 535                         num_devices++;
 536                         minlen = min_t(u64, q->limits.discard_granularity,
 537                                      minlen);
 538                 }
 539         }
 540         rcu_read_unlock();
 541
 542         if (!num_devices)
 543                 return -EOPNOTSUPP;
 544         if (copy_from_user(&range, arg, sizeof(range)))
 545                 return -EFAULT;
 546
 547         /*
 548          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 549          * block group is in the logical address space, which can be any
 550          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 551          */
 552         if (range.len < fs_info->sb->s_blocksize)
 553                 return -EINVAL;
 554
 555         range.minlen = max(range.minlen, minlen);
 556         ret = btrfs_trim_fs(fs_info, &range);
 557         if (ret < 0)
 558                 return ret;
 559
 560         if (copy_to_user(arg, &range, sizeof(range)))
 561                 return -EFAULT;
 562
 563         return 0;
 564 }
 565
 566 int __pure btrfs_is_empty_uuid(u8 *uuid)
 567 {
 568         int i;
 569
 570         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 571                 if (uuid[i])
 572                         return 0;
 573         }
 574         return 1;
 575 }
 576
 577 static noinline int create_subvol(struct inode *dir,
 578                                   struct dentry *dentry,
 579                                   const char *name, int namelen,
 580                                   struct btrfs_qgroup_inherit *inherit)
 581 {
 582         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 583         struct btrfs_trans_handle *trans;
 584         struct btrfs_key key;
 585         struct btrfs_root_item *root_item;
 586         struct btrfs_inode_item *inode_item;
 587         struct extent_buffer *leaf;
 588         struct btrfs_root *root = BTRFS_I(dir)->root;
 589         struct btrfs_root *new_root;
 590         struct btrfs_block_rsv block_rsv;
 591         struct timespec64 cur_time = current_time(dir);
 592         struct inode *inode;
 593         int ret;
 594         int err;
 595         dev_t anon_dev = 0;
 596         u64 objectid;
 597         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 598         u64 index = 0;
 599
 600         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 601         if (!root_item)
 602                 return -ENOMEM;
 603
 604         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 605         if (ret)
 606                 goto fail_free;
 607
 608         ret = get_anon_bdev(&anon_dev);
 609         if (ret < 0)
 610                 goto fail_free;
 611
 612         /*
 613          * Don't create subvolume whose level is not zero. Or qgroup will be
 614          * screwed up since it assumes subvolume qgroup's level to be 0.
 615          */
 616         if (btrfs_qgroup_level(objectid)) {
 617                 ret = -ENOSPC;
 618                 goto fail_free;
 619         }
 620
 621         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 622         /*
 623          * The same as the snapshot creation, please see the comment
 624          * of create_snapshot().
 625          */
 626         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 627         if (ret)
 628                 goto fail_free;
 629
 630         trans = btrfs_start_transaction(root, 0);
 631         if (IS_ERR(trans)) {
 632                 ret = PTR_ERR(trans);
 633                 btrfs_subvolume_release_metadata(root, &block_rsv);
 634                 goto fail_free;
 635         }
 636         trans->block_rsv = &block_rsv;
 637         trans->bytes_reserved = block_rsv.size;
 638
 639         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 640         if (ret)
 641                 goto fail;
 642
 643         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 644                                       BTRFS_NESTING_NORMAL);
 645         if (IS_ERR(leaf)) {
 646                 ret = PTR_ERR(leaf);
 647                 goto fail;
 648         }
 649
 650         btrfs_mark_buffer_dirty(leaf);
 651
 652         inode_item = &root_item->inode;
 653         btrfs_set_stack_inode_generation(inode_item, 1);
 654         btrfs_set_stack_inode_size(inode_item, 3);
 655         btrfs_set_stack_inode_nlink(inode_item, 1);
 656         btrfs_set_stack_inode_nbytes(inode_item,
 657                                      fs_info->nodesize);
 658         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 659
 660         btrfs_set_root_flags(root_item, 0);
 661         btrfs_set_root_limit(root_item, 0);
 662         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 663
 664         btrfs_set_root_bytenr(root_item, leaf->start);
 665         btrfs_set_root_generation(root_item, trans->transid);
 666         btrfs_set_root_level(root_item, 0);
 667         btrfs_set_root_refs(root_item, 1);
 668         btrfs_set_root_used(root_item, leaf->len);
 669         btrfs_set_root_last_snapshot(root_item, 0);
 670
 671         btrfs_set_root_generation_v2(root_item,
 672                         btrfs_root_generation(root_item));
 673         generate_random_guid(root_item->uuid);
 674         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 675         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 676         root_item->ctime = root_item->otime;
 677         btrfs_set_root_ctransid(root_item, trans->transid);
 678         btrfs_set_root_otransid(root_item, trans->transid);
 679
 680         btrfs_tree_unlock(leaf);
 681         free_extent_buffer(leaf);
 682         leaf = NULL;
 683
 684         btrfs_set_root_dirid(root_item, new_dirid);
 685
 686         key.objectid = objectid;
 687         key.offset = 0;
 688         key.type = BTRFS_ROOT_ITEM_KEY;
 689         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 690                                 root_item);
 691         if (ret)
 692                 goto fail;
 693
 694         key.offset = (u64)-1;
 695         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 696         if (IS_ERR(new_root)) {
 697                 free_anon_bdev(anon_dev);
 698                 ret = PTR_ERR(new_root);
 699                 btrfs_abort_transaction(trans, ret);
 700                 goto fail;
 701         }
 702         /* Freeing will be done in btrfs_put_root() of new_root */
 703         anon_dev = 0;
 704
 705         btrfs_record_root_in_trans(trans, new_root);
 706
 707         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 708         btrfs_put_root(new_root);
 709         if (ret) {
 710                 /* We potentially lose an unused inode item here */
 711                 btrfs_abort_transaction(trans, ret);
 712                 goto fail;
 713         }
 714
 715         mutex_lock(&new_root->objectid_mutex);
 716         new_root->highest_objectid = new_dirid;
 717         mutex_unlock(&new_root->objectid_mutex);
 718
 719         /*
 720          * insert the directory item
 721          */
 722         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 723         if (ret) {
 724                 btrfs_abort_transaction(trans, ret);
 725                 goto fail;
 726         }
 727
 728         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 729                                     BTRFS_FT_DIR, index);
 730         if (ret) {
 731                 btrfs_abort_transaction(trans, ret);
 732                 goto fail;
 733         }
 734
 735         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 736         ret = btrfs_update_inode(trans, root, dir);
 737         if (ret) {
 738                 btrfs_abort_transaction(trans, ret);
 739                 goto fail;
 740         }
 741
 742         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 743                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 744         if (ret) {
 745                 btrfs_abort_transaction(trans, ret);
 746                 goto fail;
 747         }
 748
 749         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 750                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 751         if (ret)
 752                 btrfs_abort_transaction(trans, ret);
 753
 754 fail:
 755         kfree(root_item);
 756         trans->block_rsv = NULL;
 757         trans->bytes_reserved = 0;
 758         btrfs_subvolume_release_metadata(root, &block_rsv);
 759
 760         err = btrfs_commit_transaction(trans);
 761         if (err && !ret)
 762                 ret = err;
 763
 764         if (!ret) {
 765                 inode = btrfs_lookup_dentry(dir, dentry);
 766                 if (IS_ERR(inode))
 767                         return PTR_ERR(inode);
 768                 d_instantiate(dentry, inode);
 769         }
 770         return ret;
 771
 772 fail_free:
 773         if (anon_dev)
 774                 free_anon_bdev(anon_dev);
 775         kfree(root_item);
 776         return ret;
 777 }
 778
 779 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 780                            struct dentry *dentry, bool readonly,
 781                            struct btrfs_qgroup_inherit *inherit)
 782 {
 783         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 784         struct inode *inode;
 785         struct btrfs_pending_snapshot *pending_snapshot;
 786         struct btrfs_trans_handle *trans;
 787         int ret;
 788
 789         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 790                 return -EINVAL;
 791
 792         if (atomic_read(&root->nr_swapfiles)) {
 793                 btrfs_warn(fs_info,
 794                            "cannot snapshot subvolume with active swapfile");
 795                 return -ETXTBSY;
 796         }
 797
 798         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 799         if (!pending_snapshot)
 800                 return -ENOMEM;
 801
 802         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 803         if (ret < 0)
 804                 goto free_pending;
 805         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 806                         GFP_KERNEL);
 807         pending_snapshot->path = btrfs_alloc_path();
 808         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 809                 ret = -ENOMEM;
 810                 goto free_pending;
 811         }
 812
 813         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 814                              BTRFS_BLOCK_RSV_TEMP);
 815         /*
 816          * 1 - parent dir inode
 817          * 2 - dir entries
 818          * 1 - root item
 819          * 2 - root ref/backref
 820          * 1 - root of snapshot
 821          * 1 - UUID item
 822          */
 823         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 824                                         &pending_snapshot->block_rsv, 8,
 825                                         false);
 826         if (ret)
 827                 goto free_pending;
 828
 829         pending_snapshot->dentry = dentry;
 830         pending_snapshot->root = root;
 831         pending_snapshot->readonly = readonly;
 832         pending_snapshot->dir = dir;
 833         pending_snapshot->inherit = inherit;
 834
 835         trans = btrfs_start_transaction(root, 0);
 836         if (IS_ERR(trans)) {
 837                 ret = PTR_ERR(trans);
 838                 goto fail;
 839         }
 840
 841         spin_lock(&fs_info->trans_lock);
 842         list_add(&pending_snapshot->list,
 843                  &trans->transaction->pending_snapshots);
 844         spin_unlock(&fs_info->trans_lock);
 845
 846         ret = btrfs_commit_transaction(trans);
 847         if (ret)
 848                 goto fail;
 849
 850         ret = pending_snapshot->error;
 851         if (ret)
 852                 goto fail;
 853
 854         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 855         if (ret)
 856                 goto fail;
 857
 858         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 859         if (IS_ERR(inode)) {
 860                 ret = PTR_ERR(inode);
 861                 goto fail;
 862         }
 863
 864         d_instantiate(dentry, inode);
 865         ret = 0;
 866         pending_snapshot->anon_dev = 0;
 867 fail:
 868         /* Prevent double freeing of anon_dev */
 869         if (ret && pending_snapshot->snap)
 870                 pending_snapshot->snap->anon_dev = 0;
 871         btrfs_put_root(pending_snapshot->snap);
 872         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 873 free_pending:
 874         if (pending_snapshot->anon_dev)
 875                 free_anon_bdev(pending_snapshot->anon_dev);
 876         kfree(pending_snapshot->root_item);
 877         btrfs_free_path(pending_snapshot->path);
 878         kfree(pending_snapshot);
 879
 880         return ret;
 881 }
 882
 883 /*  copy of may_delete in fs/namei.c()
 884  *      Check whether we can remove a link victim from directory dir, check
 885  *  whether the type of victim is right.
 886  *  1. We can't do it if dir is read-only (done in permission())
 887  *  2. We should have write and exec permissions on dir
 888  *  3. We can't remove anything from append-only dir
 889  *  4. We can't do anything with immutable dir (done in permission())
 890  *  5. If the sticky bit on dir is set we should either
 891  *      a. be owner of dir, or
 892  *      b. be owner of victim, or
 893  *      c. have CAP_FOWNER capability
 894  *  6. If the victim is append-only or immutable we can't do anything with
 895  *     links pointing to it.
 896  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 897  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 898  *  9. We can't remove a root or mountpoint.
 899  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 900  *     nfs_async_unlink().
 901  */
 902
 903 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 904 {
 905         int error;
 906
 907         if (d_really_is_negative(victim))
 908                 return -ENOENT;
 909
 910         BUG_ON(d_inode(victim->d_parent) != dir);
 911         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 912
 913         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 914         if (error)
 915                 return error;
 916         if (IS_APPEND(dir))
 917                 return -EPERM;
 918         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 919             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 920                 return -EPERM;
 921         if (isdir) {
 922                 if (!d_is_dir(victim))
 923                         return -ENOTDIR;
 924                 if (IS_ROOT(victim))
 925                         return -EBUSY;
 926         } else if (d_is_dir(victim))
 927                 return -EISDIR;
 928         if (IS_DEADDIR(dir))
 929                 return -ENOENT;
 930         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 931                 return -EBUSY;
 932         return 0;
 933 }
 934
 935 /* copy of may_create in fs/namei.c() */
 936 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 937 {
 938         if (d_really_is_positive(child))
 939                 return -EEXIST;
 940         if (IS_DEADDIR(dir))
 941                 return -ENOENT;
 942         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 943 }
 944
 945 /*
 946  * Create a new subvolume below @parent.  This is largely modeled after
 947  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 948  * inside this filesystem so it's quite a bit simpler.
 949  */
 950 static noinline int btrfs_mksubvol(const struct path *parent,
 951                                    const char *name, int namelen,
 952                                    struct btrfs_root *snap_src,
 953                                    bool readonly,
 954                                    struct btrfs_qgroup_inherit *inherit)
 955 {
 956         struct inode *dir = d_inode(parent->dentry);
 957         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 958         struct dentry *dentry;
 959         int error;
 960
 961         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 962         if (error == -EINTR)
 963                 return error;
 964
 965         dentry = lookup_one_len(name, parent->dentry, namelen);
 966         error = PTR_ERR(dentry);
 967         if (IS_ERR(dentry))
 968                 goto out_unlock;
 969
 970         error = btrfs_may_create(dir, dentry);
 971         if (error)
 972                 goto out_dput;
 973
 974         /*
 975          * even if this name doesn't exist, we may get hash collisions.
 976          * check for them now when we can safely fail
 977          */
 978         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 979                                                dir->i_ino, name,
 980                                                namelen);
 981         if (error)
 982                 goto out_dput;
 983
 984         down_read(&fs_info->subvol_sem);
 985
 986         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 987                 goto out_up_read;
 988
 989         if (snap_src)
 990                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 991         else
 992                 error = create_subvol(dir, dentry, name, namelen, inherit);
 993
 994         if (!error)
 995                 fsnotify_mkdir(dir, dentry);
 996 out_up_read:
 997         up_read(&fs_info->subvol_sem);
 998 out_dput:
 999         dput(dentry);
1000 out_unlock:
1001         inode_unlock(dir);
1002         return error;
1003 }
1004
1005 static noinline int btrfs_mksnapshot(const struct path *parent,
1006                                    const char *name, int namelen,
1007                                    struct btrfs_root *root,
1008                                    bool readonly,
1009                                    struct btrfs_qgroup_inherit *inherit)
1010 {
1011         int ret;
1012         bool snapshot_force_cow = false;
1013
1014         /*
1015          * Force new buffered writes to reserve space even when NOCOW is
1016          * possible. This is to avoid later writeback (running dealloc) to
1017          * fallback to COW mode and unexpectedly fail with ENOSPC.
1018          */
1019         btrfs_drew_read_lock(&root->snapshot_lock);
1020
1021         ret = btrfs_start_delalloc_snapshot(root);
1022         if (ret)
1023                 goto out;
1024
1025         /*
1026          * All previous writes have started writeback in NOCOW mode, so now
1027          * we force future writes to fallback to COW mode during snapshot
1028          * creation.
1029          */
1030         atomic_inc(&root->snapshot_force_cow);
1031         snapshot_force_cow = true;
1032
1033         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1034
1035         ret = btrfs_mksubvol(parent, name, namelen,
1036                              root, readonly, inherit);
1037 out:
1038         if (snapshot_force_cow)
1039                 atomic_dec(&root->snapshot_force_cow);
1040         btrfs_drew_read_unlock(&root->snapshot_lock);
1041         return ret;
1042 }
1043
1044 /*
1045  * When we're defragging a range, we don't want to kick it off again
1046  * if it is really just waiting for delalloc to send it down.
1047  * If we find a nice big extent or delalloc range for the bytes in the
1048  * file you want to defrag, we return 0 to let you know to skip this
1049  * part of the file
1050  */
1051 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1052 {
1053         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1054         struct extent_map *em = NULL;
1055         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1056         u64 end;
1057
1058         read_lock(&em_tree->lock);
1059         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1060         read_unlock(&em_tree->lock);
1061
1062         if (em) {
1063                 end = extent_map_end(em);
1064                 free_extent_map(em);
1065                 if (end - offset > thresh)
1066                         return 0;
1067         }
1068         /* if we already have a nice delalloc here, just stop */
1069         thresh /= 2;
1070         end = count_range_bits(io_tree, &offset, offset + thresh,
1071                                thresh, EXTENT_DELALLOC, 1);
1072         if (end >= thresh)
1073                 return 0;
1074         return 1;
1075 }
1076
1077 /*
1078  * helper function to walk through a file and find extents
1079  * newer than a specific transid, and smaller than thresh.
1080  *
1081  * This is used by the defragging code to find new and small
1082  * extents
1083  */
1084 static int find_new_extents(struct btrfs_root *root,
1085                             struct inode *inode, u64 newer_than,
1086                             u64 *off, u32 thresh)
1087 {
1088         struct btrfs_path *path;
1089         struct btrfs_key min_key;
1090         struct extent_buffer *leaf;
1091         struct btrfs_file_extent_item *extent;
1092         int type;
1093         int ret;
1094         u64 ino = btrfs_ino(BTRFS_I(inode));
1095
1096         path = btrfs_alloc_path();
1097         if (!path)
1098                 return -ENOMEM;
1099
1100         min_key.objectid = ino;
1101         min_key.type = BTRFS_EXTENT_DATA_KEY;
1102         min_key.offset = *off;
1103
1104         while (1) {
1105                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1106                 if (ret != 0)
1107                         goto none;
1108 process_slot:
1109                 if (min_key.objectid != ino)
1110                         goto none;
1111                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1112                         goto none;
1113
1114                 leaf = path->nodes[0];
1115                 extent = btrfs_item_ptr(leaf, path->slots[0],
1116                                         struct btrfs_file_extent_item);
1117
1118                 type = btrfs_file_extent_type(leaf, extent);
1119                 if (type == BTRFS_FILE_EXTENT_REG &&
1120                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1121                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1122                         *off = min_key.offset;
1123                         btrfs_free_path(path);
1124                         return 0;
1125                 }
1126
1127                 path->slots[0]++;
1128                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1129                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1130                         goto process_slot;
1131                 }
1132
1133                 if (min_key.offset == (u64)-1)
1134                         goto none;
1135
1136                 min_key.offset++;
1137                 btrfs_release_path(path);
1138         }
1139 none:
1140         btrfs_free_path(path);
1141         return -ENOENT;
1142 }
1143
1144 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1145 {
1146         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1147         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1148         struct extent_map *em;
1149         u64 len = PAGE_SIZE;
1150
1151         /*
1152          * hopefully we have this extent in the tree already, try without
1153          * the full extent lock
1154          */
1155         read_lock(&em_tree->lock);
1156         em = lookup_extent_mapping(em_tree, start, len);
1157         read_unlock(&em_tree->lock);
1158
1159         if (!em) {
1160                 struct extent_state *cached = NULL;
1161                 u64 end = start + len - 1;
1162
1163                 /* get the big lock and read metadata off disk */
1164                 lock_extent_bits(io_tree, start, end, &cached);
1165                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1166                 unlock_extent_cached(io_tree, start, end, &cached);
1167
1168                 if (IS_ERR(em))
1169                         return NULL;
1170         }
1171
1172         return em;
1173 }
1174
1175 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1176 {
1177         struct extent_map *next;
1178         bool ret = true;
1179
1180         /* this is the last extent */
1181         if (em->start + em->len >= i_size_read(inode))
1182                 return false;
1183
1184         next = defrag_lookup_extent(inode, em->start + em->len);
1185         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1186                 ret = false;
1187         else if ((em->block_start + em->block_len == next->block_start) &&
1188                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1189                 ret = false;
1190
1191         free_extent_map(next);
1192         return ret;
1193 }
1194
1195 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1196                                u64 *last_len, u64 *skip, u64 *defrag_end,
1197                                int compress)
1198 {
1199         struct extent_map *em;
1200         int ret = 1;
1201         bool next_mergeable = true;
1202         bool prev_mergeable = true;
1203
1204         /*
1205          * make sure that once we start defragging an extent, we keep on
1206          * defragging it
1207          */
1208         if (start < *defrag_end)
1209                 return 1;
1210
1211         *skip = 0;
1212
1213         em = defrag_lookup_extent(inode, start);
1214         if (!em)
1215                 return 0;
1216
1217         /* this will cover holes, and inline extents */
1218         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1219                 ret = 0;
1220                 goto out;
1221         }
1222
1223         if (!*defrag_end)
1224                 prev_mergeable = false;
1225
1226         next_mergeable = defrag_check_next_extent(inode, em);
1227         /*
1228          * we hit a real extent, if it is big or the next extent is not a
1229          * real extent, don't bother defragging it
1230          */
1231         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1232             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1233                 ret = 0;
1234 out:
1235         /*
1236          * last_len ends up being a counter of how many bytes we've defragged.
1237          * every time we choose not to defrag an extent, we reset *last_len
1238          * so that the next tiny extent will force a defrag.
1239          *
1240          * The end result of this is that tiny extents before a single big
1241          * extent will force at least part of that big extent to be defragged.
1242          */
1243         if (ret) {
1244                 *defrag_end = extent_map_end(em);
1245         } else {
1246                 *last_len = 0;
1247                 *skip = extent_map_end(em);
1248                 *defrag_end = 0;
1249         }
1250
1251         free_extent_map(em);
1252         return ret;
1253 }
1254
1255 /*
1256  * it doesn't do much good to defrag one or two pages
1257  * at a time.  This pulls in a nice chunk of pages
1258  * to COW and defrag.
1259  *
1260  * It also makes sure the delalloc code has enough
1261  * dirty data to avoid making new small extents as part
1262  * of the defrag
1263  *
1264  * It's a good idea to start RA on this range
1265  * before calling this.
1266  */
1267 static int cluster_pages_for_defrag(struct inode *inode,
1268                                     struct page **pages,
1269                                     unsigned long start_index,
1270                                     unsigned long num_pages)
1271 {
1272         unsigned long file_end;
1273         u64 isize = i_size_read(inode);
1274         u64 page_start;
1275         u64 page_end;
1276         u64 page_cnt;
1277         u64 start = (u64)start_index << PAGE_SHIFT;
1278         int ret;
1279         int i;
1280         int i_done;
1281         struct btrfs_ordered_extent *ordered;
1282         struct extent_state *cached_state = NULL;
1283         struct extent_io_tree *tree;
1284         struct extent_changeset *data_reserved = NULL;
1285         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1286
1287         file_end = (isize - 1) >> PAGE_SHIFT;
1288         if (!isize || start_index > file_end)
1289                 return 0;
1290
1291         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1292
1293         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1294                         start, page_cnt << PAGE_SHIFT);
1295         if (ret)
1296                 return ret;
1297         i_done = 0;
1298         tree = &BTRFS_I(inode)->io_tree;
1299
1300         /* step one, lock all the pages */
1301         for (i = 0; i < page_cnt; i++) {
1302                 struct page *page;
1303 again:
1304                 page = find_or_create_page(inode->i_mapping,
1305                                            start_index + i, mask);
1306                 if (!page)
1307                         break;
1308
1309                 page_start = page_offset(page);
1310                 page_end = page_start + PAGE_SIZE - 1;
1311                 while (1) {
1312                         lock_extent_bits(tree, page_start, page_end,
1313                                          &cached_state);
1314                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1315                                                               page_start);
1316                         unlock_extent_cached(tree, page_start, page_end,
1317                                              &cached_state);
1318                         if (!ordered)
1319                                 break;
1320
1321                         unlock_page(page);
1322                         btrfs_start_ordered_extent(ordered, 1);
1323                         btrfs_put_ordered_extent(ordered);
1324                         lock_page(page);
1325                         /*
1326                          * we unlocked the page above, so we need check if
1327                          * it was released or not.
1328                          */
1329                         if (page->mapping != inode->i_mapping) {
1330                                 unlock_page(page);
1331                                 put_page(page);
1332                                 goto again;
1333                         }
1334                 }
1335
1336                 if (!PageUptodate(page)) {
1337                         btrfs_readpage(NULL, page);
1338                         lock_page(page);
1339                         if (!PageUptodate(page)) {
1340                                 unlock_page(page);
1341                                 put_page(page);
1342                                 ret = -EIO;
1343                                 break;
1344                         }
1345                 }
1346
1347                 if (page->mapping != inode->i_mapping) {
1348                         unlock_page(page);
1349                         put_page(page);
1350                         goto again;
1351                 }
1352
1353                 pages[i] = page;
1354                 i_done++;
1355         }
1356         if (!i_done || ret)
1357                 goto out;
1358
1359         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1360                 goto out;
1361
1362         /*
1363          * so now we have a nice long stream of locked
1364          * and up to date pages, lets wait on them
1365          */
1366         for (i = 0; i < i_done; i++)
1367                 wait_on_page_writeback(pages[i]);
1368
1369         page_start = page_offset(pages[0]);
1370         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1371
1372         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1373                          page_start, page_end - 1, &cached_state);
1374         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1375                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1376                           EXTENT_DEFRAG, 0, 0, &cached_state);
1377
1378         if (i_done != page_cnt) {
1379                 spin_lock(&BTRFS_I(inode)->lock);
1380                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1381                 spin_unlock(&BTRFS_I(inode)->lock);
1382                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1383                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1384         }
1385
1386
1387         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1388                           &cached_state);
1389
1390         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1391                              page_start, page_end - 1, &cached_state);
1392
1393         for (i = 0; i < i_done; i++) {
1394                 clear_page_dirty_for_io(pages[i]);
1395                 ClearPageChecked(pages[i]);
1396                 set_page_extent_mapped(pages[i]);
1397                 set_page_dirty(pages[i]);
1398                 unlock_page(pages[i]);
1399                 put_page(pages[i]);
1400         }
1401         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1402         extent_changeset_free(data_reserved);
1403         return i_done;
1404 out:
1405         for (i = 0; i < i_done; i++) {
1406                 unlock_page(pages[i]);
1407                 put_page(pages[i]);
1408         }
1409         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1410                         start, page_cnt << PAGE_SHIFT, true);
1411         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1412         extent_changeset_free(data_reserved);
1413         return ret;
1414
1415 }
1416
1417 int btrfs_defrag_file(struct inode *inode, struct file *file,
1418                       struct btrfs_ioctl_defrag_range_args *range,
1419                       u64 newer_than, unsigned long max_to_defrag)
1420 {
1421         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1422         struct btrfs_root *root = BTRFS_I(inode)->root;
1423         struct file_ra_state *ra = NULL;
1424         unsigned long last_index;
1425         u64 isize = i_size_read(inode);
1426         u64 last_len = 0;
1427         u64 skip = 0;
1428         u64 defrag_end = 0;
1429         u64 newer_off = range->start;
1430         unsigned long i;
1431         unsigned long ra_index = 0;
1432         int ret;
1433         int defrag_count = 0;
1434         int compress_type = BTRFS_COMPRESS_ZLIB;
1435         u32 extent_thresh = range->extent_thresh;
1436         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1437         unsigned long cluster = max_cluster;
1438         u64 new_align = ~((u64)SZ_128K - 1);
1439         struct page **pages = NULL;
1440         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1441
1442         if (isize == 0)
1443                 return 0;
1444
1445         if (range->start >= isize)
1446                 return -EINVAL;
1447
1448         if (do_compress) {
1449                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1450                         return -EINVAL;
1451                 if (range->compress_type)
1452                         compress_type = range->compress_type;
1453         }
1454
1455         if (extent_thresh == 0)
1456                 extent_thresh = SZ_256K;
1457
1458         /*
1459          * If we were not given a file, allocate a readahead context. As
1460          * readahead is just an optimization, defrag will work without it so
1461          * we don't error out.
1462          */
1463         if (!file) {
1464                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1465                 if (ra)
1466                         file_ra_state_init(ra, inode->i_mapping);
1467         } else {
1468                 ra = &file->f_ra;
1469         }
1470
1471         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1472         if (!pages) {
1473                 ret = -ENOMEM;
1474                 goto out_ra;
1475         }
1476
1477         /* find the last page to defrag */
1478         if (range->start + range->len > range->start) {
1479                 last_index = min_t(u64, isize - 1,
1480                          range->start + range->len - 1) >> PAGE_SHIFT;
1481         } else {
1482                 last_index = (isize - 1) >> PAGE_SHIFT;
1483         }
1484
1485         if (newer_than) {
1486                 ret = find_new_extents(root, inode, newer_than,
1487                                        &newer_off, SZ_64K);
1488                 if (!ret) {
1489                         range->start = newer_off;
1490                         /*
1491                          * we always align our defrag to help keep
1492                          * the extents in the file evenly spaced
1493                          */
1494                         i = (newer_off & new_align) >> PAGE_SHIFT;
1495                 } else
1496                         goto out_ra;
1497         } else {
1498                 i = range->start >> PAGE_SHIFT;
1499         }
1500         if (!max_to_defrag)
1501                 max_to_defrag = last_index - i + 1;
1502
1503         /*
1504          * make writeback starts from i, so the defrag range can be
1505          * written sequentially.
1506          */
1507         if (i < inode->i_mapping->writeback_index)
1508                 inode->i_mapping->writeback_index = i;
1509
1510         while (i <= last_index && defrag_count < max_to_defrag &&
1511                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1512                 /*
1513                  * make sure we stop running if someone unmounts
1514                  * the FS
1515                  */
1516                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1517                         break;
1518
1519                 if (btrfs_defrag_cancelled(fs_info)) {
1520                         btrfs_debug(fs_info, "defrag_file cancelled");
1521                         ret = -EAGAIN;
1522                         break;
1523                 }
1524
1525                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1526                                          extent_thresh, &last_len, &skip,
1527                                          &defrag_end, do_compress)){
1528                         unsigned long next;
1529                         /*
1530                          * the should_defrag function tells us how much to skip
1531                          * bump our counter by the suggested amount
1532                          */
1533                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1534                         i = max(i + 1, next);
1535                         continue;
1536                 }
1537
1538                 if (!newer_than) {
1539                         cluster = (PAGE_ALIGN(defrag_end) >>
1540                                    PAGE_SHIFT) - i;
1541                         cluster = min(cluster, max_cluster);
1542                 } else {
1543                         cluster = max_cluster;
1544                 }
1545
1546                 if (i + cluster > ra_index) {
1547                         ra_index = max(i, ra_index);
1548                         if (ra)
1549                                 page_cache_sync_readahead(inode->i_mapping, ra,
1550                                                 file, ra_index, cluster);
1551                         ra_index += cluster;
1552                 }
1553
1554                 inode_lock(inode);
1555                 if (IS_SWAPFILE(inode)) {
1556                         ret = -ETXTBSY;
1557                 } else {
1558                         if (do_compress)
1559                                 BTRFS_I(inode)->defrag_compress = compress_type;
1560                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1561                 }
1562                 if (ret < 0) {
1563                         inode_unlock(inode);
1564                         goto out_ra;
1565                 }
1566
1567                 defrag_count += ret;
1568                 balance_dirty_pages_ratelimited(inode->i_mapping);
1569                 inode_unlock(inode);
1570
1571                 if (newer_than) {
1572                         if (newer_off == (u64)-1)
1573                                 break;
1574
1575                         if (ret > 0)
1576                                 i += ret;
1577
1578                         newer_off = max(newer_off + 1,
1579                                         (u64)i << PAGE_SHIFT);
1580
1581                         ret = find_new_extents(root, inode, newer_than,
1582                                                &newer_off, SZ_64K);
1583                         if (!ret) {
1584                                 range->start = newer_off;
1585                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1586                         } else {
1587                                 break;
1588                         }
1589                 } else {
1590                         if (ret > 0) {
1591                                 i += ret;
1592                                 last_len += ret << PAGE_SHIFT;
1593                         } else {
1594                                 i++;
1595                                 last_len = 0;
1596                         }
1597                 }
1598         }
1599
1600         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1601                 filemap_flush(inode->i_mapping);
1602                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1603                              &BTRFS_I(inode)->runtime_flags))
1604                         filemap_flush(inode->i_mapping);
1605         }
1606
1607         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1608                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1609         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1610                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1611         }
1612
1613         ret = defrag_count;
1614
1615 out_ra:
1616         if (do_compress) {
1617                 inode_lock(inode);
1618                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1619                 inode_unlock(inode);
1620         }
1621         if (!file)
1622                 kfree(ra);
1623         kfree(pages);
1624         return ret;
1625 }
1626
1627 static noinline int btrfs_ioctl_resize(struct file *file,
1628                                         void __user *arg)
1629 {
1630         struct inode *inode = file_inode(file);
1631         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1632         u64 new_size;
1633         u64 old_size;
1634         u64 devid = 1;
1635         struct btrfs_root *root = BTRFS_I(inode)->root;
1636         struct btrfs_ioctl_vol_args *vol_args;
1637         struct btrfs_trans_handle *trans;
1638         struct btrfs_device *device = NULL;
1639         char *sizestr;
1640         char *retptr;
1641         char *devstr = NULL;
1642         int ret = 0;
1643         int mod = 0;
1644
1645         if (!capable(CAP_SYS_ADMIN))
1646                 return -EPERM;
1647
1648         ret = mnt_want_write_file(file);
1649         if (ret)
1650                 return ret;
1651
1652         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_RESIZE)) {
1653                 mnt_drop_write_file(file);
1654                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1655         }
1656
1657         vol_args = memdup_user(arg, sizeof(*vol_args));
1658         if (IS_ERR(vol_args)) {
1659                 ret = PTR_ERR(vol_args);
1660                 goto out;
1661         }
1662
1663         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1664
1665         sizestr = vol_args->name;
1666         devstr = strchr(sizestr, ':');
1667         if (devstr) {
1668                 sizestr = devstr + 1;
1669                 *devstr = '\0';
1670                 devstr = vol_args->name;
1671                 ret = kstrtoull(devstr, 10, &devid);
1672                 if (ret)
1673                         goto out_free;
1674                 if (!devid) {
1675                         ret = -EINVAL;
1676                         goto out_free;
1677                 }
1678                 btrfs_info(fs_info, "resizing devid %llu", devid);
1679         }
1680
1681         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1682         if (!device) {
1683                 btrfs_info(fs_info, "resizer unable to find device %llu",
1684                            devid);
1685                 ret = -ENODEV;
1686                 goto out_free;
1687         }
1688
1689         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1690                 btrfs_info(fs_info,
1691                            "resizer unable to apply on readonly device %llu",
1692                        devid);
1693                 ret = -EPERM;
1694                 goto out_free;
1695         }
1696
1697         if (!strcmp(sizestr, "max"))
1698                 new_size = device->bdev->bd_inode->i_size;
1699         else {
1700                 if (sizestr[0] == '-') {
1701                         mod = -1;
1702                         sizestr++;
1703                 } else if (sizestr[0] == '+') {
1704                         mod = 1;
1705                         sizestr++;
1706                 }
1707                 new_size = memparse(sizestr, &retptr);
1708                 if (*retptr != '\0' || new_size == 0) {
1709                         ret = -EINVAL;
1710                         goto out_free;
1711                 }
1712         }
1713
1714         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1715                 ret = -EPERM;
1716                 goto out_free;
1717         }
1718
1719         old_size = btrfs_device_get_total_bytes(device);
1720
1721         if (mod < 0) {
1722                 if (new_size > old_size) {
1723                         ret = -EINVAL;
1724                         goto out_free;
1725                 }
1726                 new_size = old_size - new_size;
1727         } else if (mod > 0) {
1728                 if (new_size > ULLONG_MAX - old_size) {
1729                         ret = -ERANGE;
1730                         goto out_free;
1731                 }
1732                 new_size = old_size + new_size;
1733         }
1734
1735         if (new_size < SZ_256M) {
1736                 ret = -EINVAL;
1737                 goto out_free;
1738         }
1739         if (new_size > device->bdev->bd_inode->i_size) {
1740                 ret = -EFBIG;
1741                 goto out_free;
1742         }
1743
1744         new_size = round_down(new_size, fs_info->sectorsize);
1745
1746         if (new_size > old_size) {
1747                 trans = btrfs_start_transaction(root, 0);
1748                 if (IS_ERR(trans)) {
1749                         ret = PTR_ERR(trans);
1750                         goto out_free;
1751                 }
1752                 ret = btrfs_grow_device(trans, device, new_size);
1753                 btrfs_commit_transaction(trans);
1754         } else if (new_size < old_size) {
1755                 ret = btrfs_shrink_device(device, new_size);
1756         } /* equal, nothing need to do */
1757
1758         if (ret == 0 && new_size != old_size)
1759                 btrfs_info_in_rcu(fs_info,
1760                         "resize device %s (devid %llu) from %llu to %llu",
1761                         rcu_str_deref(device->name), device->devid,
1762                         old_size, new_size);
1763 out_free:
1764         kfree(vol_args);
1765 out:
1766         btrfs_exclop_finish(fs_info);
1767         mnt_drop_write_file(file);
1768         return ret;
1769 }
1770
1771 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1772                                 const char *name, unsigned long fd, int subvol,
1773                                 bool readonly,
1774                                 struct btrfs_qgroup_inherit *inherit)
1775 {
1776         int namelen;
1777         int ret = 0;
1778
1779         if (!S_ISDIR(file_inode(file)->i_mode))
1780                 return -ENOTDIR;
1781
1782         ret = mnt_want_write_file(file);
1783         if (ret)
1784                 goto out;
1785
1786         namelen = strlen(name);
1787         if (strchr(name, '/')) {
1788                 ret = -EINVAL;
1789                 goto out_drop_write;
1790         }
1791
1792         if (name[0] == '.' &&
1793            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1794                 ret = -EEXIST;
1795                 goto out_drop_write;
1796         }
1797
1798         if (subvol) {
1799                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1800                                      NULL, readonly, inherit);
1801         } else {
1802                 struct fd src = fdget(fd);
1803                 struct inode *src_inode;
1804                 if (!src.file) {
1805                         ret = -EINVAL;
1806                         goto out_drop_write;
1807                 }
1808
1809                 src_inode = file_inode(src.file);
1810                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1811                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1812                                    "Snapshot src from another FS");
1813                         ret = -EXDEV;
1814                 } else if (!inode_owner_or_capable(src_inode)) {
1815                         /*
1816                          * Subvolume creation is not restricted, but snapshots
1817                          * are limited to own subvolumes only
1818                          */
1819                         ret = -EPERM;
1820                 } else {
1821                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1822                                              BTRFS_I(src_inode)->root,
1823                                              readonly, inherit);
1824                 }
1825                 fdput(src);
1826         }
1827 out_drop_write:
1828         mnt_drop_write_file(file);
1829 out:
1830         return ret;
1831 }
1832
1833 static noinline int btrfs_ioctl_snap_create(struct file *file,
1834                                             void __user *arg, int subvol)
1835 {
1836         struct btrfs_ioctl_vol_args *vol_args;
1837         int ret;
1838
1839         if (!S_ISDIR(file_inode(file)->i_mode))
1840                 return -ENOTDIR;
1841
1842         vol_args = memdup_user(arg, sizeof(*vol_args));
1843         if (IS_ERR(vol_args))
1844                 return PTR_ERR(vol_args);
1845         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1846
1847         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1848                                         subvol, false, NULL);
1849
1850         kfree(vol_args);
1851         return ret;
1852 }
1853
1854 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1855                                                void __user *arg, int subvol)
1856 {
1857         struct btrfs_ioctl_vol_args_v2 *vol_args;
1858         int ret;
1859         bool readonly = false;
1860         struct btrfs_qgroup_inherit *inherit = NULL;
1861
1862         if (!S_ISDIR(file_inode(file)->i_mode))
1863                 return -ENOTDIR;
1864
1865         vol_args = memdup_user(arg, sizeof(*vol_args));
1866         if (IS_ERR(vol_args))
1867                 return PTR_ERR(vol_args);
1868         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1869
1870         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1871                 ret = -EOPNOTSUPP;
1872                 goto free_args;
1873         }
1874
1875         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1876                 readonly = true;
1877         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1878                 if (vol_args->size > PAGE_SIZE) {
1879                         ret = -EINVAL;
1880                         goto free_args;
1881                 }
1882                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1883                 if (IS_ERR(inherit)) {
1884                         ret = PTR_ERR(inherit);
1885                         goto free_args;
1886                 }
1887         }
1888
1889         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1890                                         subvol, readonly, inherit);
1891         if (ret)
1892                 goto free_inherit;
1893 free_inherit:
1894         kfree(inherit);
1895 free_args:
1896         kfree(vol_args);
1897         return ret;
1898 }
1899
1900 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1901                                                 void __user *arg)
1902 {
1903         struct inode *inode = file_inode(file);
1904         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1905         struct btrfs_root *root = BTRFS_I(inode)->root;
1906         int ret = 0;
1907         u64 flags = 0;
1908
1909         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1910                 return -EINVAL;
1911
1912         down_read(&fs_info->subvol_sem);
1913         if (btrfs_root_readonly(root))
1914                 flags |= BTRFS_SUBVOL_RDONLY;
1915         up_read(&fs_info->subvol_sem);
1916
1917         if (copy_to_user(arg, &flags, sizeof(flags)))
1918                 ret = -EFAULT;
1919
1920         return ret;
1921 }
1922
1923 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1924                                               void __user *arg)
1925 {
1926         struct inode *inode = file_inode(file);
1927         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1928         struct btrfs_root *root = BTRFS_I(inode)->root;
1929         struct btrfs_trans_handle *trans;
1930         u64 root_flags;
1931         u64 flags;
1932         int ret = 0;
1933
1934         if (!inode_owner_or_capable(inode))
1935                 return -EPERM;
1936
1937         ret = mnt_want_write_file(file);
1938         if (ret)
1939                 goto out;
1940
1941         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1942                 ret = -EINVAL;
1943                 goto out_drop_write;
1944         }
1945
1946         if (copy_from_user(&flags, arg, sizeof(flags))) {
1947                 ret = -EFAULT;
1948                 goto out_drop_write;
1949         }
1950
1951         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1952                 ret = -EOPNOTSUPP;
1953                 goto out_drop_write;
1954         }
1955
1956         down_write(&fs_info->subvol_sem);
1957
1958         /* nothing to do */
1959         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1960                 goto out_drop_sem;
1961
1962         root_flags = btrfs_root_flags(&root->root_item);
1963         if (flags & BTRFS_SUBVOL_RDONLY) {
1964                 btrfs_set_root_flags(&root->root_item,
1965                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1966         } else {
1967                 /*
1968                  * Block RO -> RW transition if this subvolume is involved in
1969                  * send
1970                  */
1971                 spin_lock(&root->root_item_lock);
1972                 if (root->send_in_progress == 0) {
1973                         btrfs_set_root_flags(&root->root_item,
1974                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1975                         spin_unlock(&root->root_item_lock);
1976                 } else {
1977                         spin_unlock(&root->root_item_lock);
1978                         btrfs_warn(fs_info,
1979                                    "Attempt to set subvolume %llu read-write during send",
1980                                    root->root_key.objectid);
1981                         ret = -EPERM;
1982                         goto out_drop_sem;
1983                 }
1984         }
1985
1986         trans = btrfs_start_transaction(root, 1);
1987         if (IS_ERR(trans)) {
1988                 ret = PTR_ERR(trans);
1989                 goto out_reset;
1990         }
1991
1992         ret = btrfs_update_root(trans, fs_info->tree_root,
1993                                 &root->root_key, &root->root_item);
1994         if (ret < 0) {
1995                 btrfs_end_transaction(trans);
1996                 goto out_reset;
1997         }
1998
1999         ret = btrfs_commit_transaction(trans);
2000
2001 out_reset:
2002         if (ret)
2003                 btrfs_set_root_flags(&root->root_item, root_flags);
2004 out_drop_sem:
2005         up_write(&fs_info->subvol_sem);
2006 out_drop_write:
2007         mnt_drop_write_file(file);
2008 out:
2009         return ret;
2010 }
2011
2012 static noinline int key_in_sk(struct btrfs_key *key,
2013                               struct btrfs_ioctl_search_key *sk)
2014 {
2015         struct btrfs_key test;
2016         int ret;
2017
2018         test.objectid = sk->min_objectid;
2019         test.type = sk->min_type;
2020         test.offset = sk->min_offset;
2021
2022         ret = btrfs_comp_cpu_keys(key, &test);
2023         if (ret < 0)
2024                 return 0;
2025
2026         test.objectid = sk->max_objectid;
2027         test.type = sk->max_type;
2028         test.offset = sk->max_offset;
2029
2030         ret = btrfs_comp_cpu_keys(key, &test);
2031         if (ret > 0)
2032                 return 0;
2033         return 1;
2034 }
2035
2036 static noinline int copy_to_sk(struct btrfs_path *path,
2037                                struct btrfs_key *key,
2038                                struct btrfs_ioctl_search_key *sk,
2039                                size_t *buf_size,
2040                                char __user *ubuf,
2041                                unsigned long *sk_offset,
2042                                int *num_found)
2043 {
2044         u64 found_transid;
2045         struct extent_buffer *leaf;
2046         struct btrfs_ioctl_search_header sh;
2047         struct btrfs_key test;
2048         unsigned long item_off;
2049         unsigned long item_len;
2050         int nritems;
2051         int i;
2052         int slot;
2053         int ret = 0;
2054
2055         leaf = path->nodes[0];
2056         slot = path->slots[0];
2057         nritems = btrfs_header_nritems(leaf);
2058
2059         if (btrfs_header_generation(leaf) > sk->max_transid) {
2060                 i = nritems;
2061                 goto advance_key;
2062         }
2063         found_transid = btrfs_header_generation(leaf);
2064
2065         for (i = slot; i < nritems; i++) {
2066                 item_off = btrfs_item_ptr_offset(leaf, i);
2067                 item_len = btrfs_item_size_nr(leaf, i);
2068
2069                 btrfs_item_key_to_cpu(leaf, key, i);
2070                 if (!key_in_sk(key, sk))
2071                         continue;
2072
2073                 if (sizeof(sh) + item_len > *buf_size) {
2074                         if (*num_found) {
2075                                 ret = 1;
2076                                 goto out;
2077                         }
2078
2079                         /*
2080                          * return one empty item back for v1, which does not
2081                          * handle -EOVERFLOW
2082                          */
2083
2084                         *buf_size = sizeof(sh) + item_len;
2085                         item_len = 0;
2086                         ret = -EOVERFLOW;
2087                 }
2088
2089                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2090                         ret = 1;
2091                         goto out;
2092                 }
2093
2094                 sh.objectid = key->objectid;
2095                 sh.offset = key->offset;
2096                 sh.type = key->type;
2097                 sh.len = item_len;
2098                 sh.transid = found_transid;
2099
2100                 /*
2101                  * Copy search result header. If we fault then loop again so we
2102                  * can fault in the pages and -EFAULT there if there's a
2103                  * problem. Otherwise we'll fault and then copy the buffer in
2104                  * properly this next time through
2105                  */
2106                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2107                         ret = 0;
2108                         goto out;
2109                 }
2110
2111                 *sk_offset += sizeof(sh);
2112
2113                 if (item_len) {
2114                         char __user *up = ubuf + *sk_offset;
2115                         /*
2116                          * Copy the item, same behavior as above, but reset the
2117                          * * sk_offset so we copy the full thing again.
2118                          */
2119                         if (read_extent_buffer_to_user_nofault(leaf, up,
2120                                                 item_off, item_len)) {
2121                                 ret = 0;
2122                                 *sk_offset -= sizeof(sh);
2123                                 goto out;
2124                         }
2125
2126                         *sk_offset += item_len;
2127                 }
2128                 (*num_found)++;
2129
2130                 if (ret) /* -EOVERFLOW from above */
2131                         goto out;
2132
2133                 if (*num_found >= sk->nr_items) {
2134                         ret = 1;
2135                         goto out;
2136                 }
2137         }
2138 advance_key:
2139         ret = 0;
2140         test.objectid = sk->max_objectid;
2141         test.type = sk->max_type;
2142         test.offset = sk->max_offset;
2143         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2144                 ret = 1;
2145         else if (key->offset < (u64)-1)
2146                 key->offset++;
2147         else if (key->type < (u8)-1) {
2148                 key->offset = 0;
2149                 key->type++;
2150         } else if (key->objectid < (u64)-1) {
2151                 key->offset = 0;
2152                 key->type = 0;
2153                 key->objectid++;
2154         } else
2155                 ret = 1;
2156 out:
2157         /*
2158          *  0: all items from this leaf copied, continue with next
2159          *  1: * more items can be copied, but unused buffer is too small
2160          *     * all items were found
2161          *     Either way, it will stops the loop which iterates to the next
2162          *     leaf
2163          *  -EOVERFLOW: item was to large for buffer
2164          *  -EFAULT: could not copy extent buffer back to userspace
2165          */
2166         return ret;
2167 }
2168
2169 static noinline int search_ioctl(struct inode *inode,
2170                                  struct btrfs_ioctl_search_key *sk,
2171                                  size_t *buf_size,
2172                                  char __user *ubuf)
2173 {
2174         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2175         struct btrfs_root *root;
2176         struct btrfs_key key;
2177         struct btrfs_path *path;
2178         int ret;
2179         int num_found = 0;
2180         unsigned long sk_offset = 0;
2181
2182         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2183                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2184                 return -EOVERFLOW;
2185         }
2186
2187         path = btrfs_alloc_path();
2188         if (!path)
2189                 return -ENOMEM;
2190
2191         if (sk->tree_id == 0) {
2192                 /* search the root of the inode that was passed */
2193                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2194         } else {
2195                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2196                 if (IS_ERR(root)) {
2197                         btrfs_free_path(path);
2198                         return PTR_ERR(root);
2199                 }
2200         }
2201
2202         key.objectid = sk->min_objectid;
2203         key.type = sk->min_type;
2204         key.offset = sk->min_offset;
2205
2206         while (1) {
2207                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2208                                                *buf_size - sk_offset);
2209                 if (ret)
2210                         break;
2211
2212                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2213                 if (ret != 0) {
2214                         if (ret > 0)
2215                                 ret = 0;
2216                         goto err;
2217                 }
2218                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2219                                  &sk_offset, &num_found);
2220                 btrfs_release_path(path);
2221                 if (ret)
2222                         break;
2223
2224         }
2225         if (ret > 0)
2226                 ret = 0;
2227 err:
2228         sk->nr_items = num_found;
2229         btrfs_put_root(root);
2230         btrfs_free_path(path);
2231         return ret;
2232 }
2233
2234 static noinline int btrfs_ioctl_tree_search(struct file *file,
2235                                            void __user *argp)
2236 {
2237         struct btrfs_ioctl_search_args __user *uargs;
2238         struct btrfs_ioctl_search_key sk;
2239         struct inode *inode;
2240         int ret;
2241         size_t buf_size;
2242
2243         if (!capable(CAP_SYS_ADMIN))
2244                 return -EPERM;
2245
2246         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2247
2248         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2249                 return -EFAULT;
2250
2251         buf_size = sizeof(uargs->buf);
2252
2253         inode = file_inode(file);
2254         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2255
2256         /*
2257          * In the origin implementation an overflow is handled by returning a
2258          * search header with a len of zero, so reset ret.
2259          */
2260         if (ret == -EOVERFLOW)
2261                 ret = 0;
2262
2263         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2264                 ret = -EFAULT;
2265         return ret;
2266 }
2267
2268 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2269                                                void __user *argp)
2270 {
2271         struct btrfs_ioctl_search_args_v2 __user *uarg;
2272         struct btrfs_ioctl_search_args_v2 args;
2273         struct inode *inode;
2274         int ret;
2275         size_t buf_size;
2276         const size_t buf_limit = SZ_16M;
2277
2278         if (!capable(CAP_SYS_ADMIN))
2279                 return -EPERM;
2280
2281         /* copy search header and buffer size */
2282         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2283         if (copy_from_user(&args, uarg, sizeof(args)))
2284                 return -EFAULT;
2285
2286         buf_size = args.buf_size;
2287
2288         /* limit result size to 16MB */
2289         if (buf_size > buf_limit)
2290                 buf_size = buf_limit;
2291
2292         inode = file_inode(file);
2293         ret = search_ioctl(inode, &args.key, &buf_size,
2294                            (char __user *)(&uarg->buf[0]));
2295         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2296                 ret = -EFAULT;
2297         else if (ret == -EOVERFLOW &&
2298                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2299                 ret = -EFAULT;
2300
2301         return ret;
2302 }
2303
2304 /*
2305  * Search INODE_REFs to identify path name of 'dirid' directory
2306  * in a 'tree_id' tree. and sets path name to 'name'.
2307  */
2308 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2309                                 u64 tree_id, u64 dirid, char *name)
2310 {
2311         struct btrfs_root *root;
2312         struct btrfs_key key;
2313         char *ptr;
2314         int ret = -1;
2315         int slot;
2316         int len;
2317         int total_len = 0;
2318         struct btrfs_inode_ref *iref;
2319         struct extent_buffer *l;
2320         struct btrfs_path *path;
2321
2322         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2323                 name[0]='\0';
2324                 return 0;
2325         }
2326
2327         path = btrfs_alloc_path();
2328         if (!path)
2329                 return -ENOMEM;
2330
2331         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2332
2333         root = btrfs_get_fs_root(info, tree_id, true);
2334         if (IS_ERR(root)) {
2335                 ret = PTR_ERR(root);
2336                 root = NULL;
2337                 goto out;
2338         }
2339
2340         key.objectid = dirid;
2341         key.type = BTRFS_INODE_REF_KEY;
2342         key.offset = (u64)-1;
2343
2344         while (1) {
2345                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2346                 if (ret < 0)
2347                         goto out;
2348                 else if (ret > 0) {
2349                         ret = btrfs_previous_item(root, path, dirid,
2350                                                   BTRFS_INODE_REF_KEY);
2351                         if (ret < 0)
2352                                 goto out;
2353                         else if (ret > 0) {
2354                                 ret = -ENOENT;
2355                                 goto out;
2356                         }
2357                 }
2358
2359                 l = path->nodes[0];
2360                 slot = path->slots[0];
2361                 btrfs_item_key_to_cpu(l, &key, slot);
2362
2363                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2364                 len = btrfs_inode_ref_name_len(l, iref);
2365                 ptr -= len + 1;
2366                 total_len += len + 1;
2367                 if (ptr < name) {
2368                         ret = -ENAMETOOLONG;
2369                         goto out;
2370                 }
2371
2372                 *(ptr + len) = '/';
2373                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2374
2375                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2376                         break;
2377
2378                 btrfs_release_path(path);
2379                 key.objectid = key.offset;
2380                 key.offset = (u64)-1;
2381                 dirid = key.objectid;
2382         }
2383         memmove(name, ptr, total_len);
2384         name[total_len] = '\0';
2385         ret = 0;
2386 out:
2387         btrfs_put_root(root);
2388         btrfs_free_path(path);
2389         return ret;
2390 }
2391
2392 static int btrfs_search_path_in_tree_user(struct inode *inode,
2393                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2394 {
2395         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2396         struct super_block *sb = inode->i_sb;
2397         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2398         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2399         u64 dirid = args->dirid;
2400         unsigned long item_off;
2401         unsigned long item_len;
2402         struct btrfs_inode_ref *iref;
2403         struct btrfs_root_ref *rref;
2404         struct btrfs_root *root = NULL;
2405         struct btrfs_path *path;
2406         struct btrfs_key key, key2;
2407         struct extent_buffer *leaf;
2408         struct inode *temp_inode;
2409         char *ptr;
2410         int slot;
2411         int len;
2412         int total_len = 0;
2413         int ret;
2414
2415         path = btrfs_alloc_path();
2416         if (!path)
2417                 return -ENOMEM;
2418
2419         /*
2420          * If the bottom subvolume does not exist directly under upper_limit,
2421          * construct the path in from the bottom up.
2422          */
2423         if (dirid != upper_limit.objectid) {
2424                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2425
2426                 root = btrfs_get_fs_root(fs_info, treeid, true);
2427                 if (IS_ERR(root)) {
2428                         ret = PTR_ERR(root);
2429                         goto out;
2430                 }
2431
2432                 key.objectid = dirid;
2433                 key.type = BTRFS_INODE_REF_KEY;
2434                 key.offset = (u64)-1;
2435                 while (1) {
2436                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2437                         if (ret < 0) {
2438                                 goto out_put;
2439                         } else if (ret > 0) {
2440                                 ret = btrfs_previous_item(root, path, dirid,
2441                                                           BTRFS_INODE_REF_KEY);
2442                                 if (ret < 0) {
2443                                         goto out_put;
2444                                 } else if (ret > 0) {
2445                                         ret = -ENOENT;
2446                                         goto out_put;
2447                                 }
2448                         }
2449
2450                         leaf = path->nodes[0];
2451                         slot = path->slots[0];
2452                         btrfs_item_key_to_cpu(leaf, &key, slot);
2453
2454                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2455                         len = btrfs_inode_ref_name_len(leaf, iref);
2456                         ptr -= len + 1;
2457                         total_len += len + 1;
2458                         if (ptr < args->path) {
2459                                 ret = -ENAMETOOLONG;
2460                                 goto out_put;
2461                         }
2462
2463                         *(ptr + len) = '/';
2464                         read_extent_buffer(leaf, ptr,
2465                                         (unsigned long)(iref + 1), len);
2466
2467                         /* Check the read+exec permission of this directory */
2468                         ret = btrfs_previous_item(root, path, dirid,
2469                                                   BTRFS_INODE_ITEM_KEY);
2470                         if (ret < 0) {
2471                                 goto out_put;
2472                         } else if (ret > 0) {
2473                                 ret = -ENOENT;
2474                                 goto out_put;
2475                         }
2476
2477                         leaf = path->nodes[0];
2478                         slot = path->slots[0];
2479                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2480                         if (key2.objectid != dirid) {
2481                                 ret = -ENOENT;
2482                                 goto out_put;
2483                         }
2484
2485                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2486                         if (IS_ERR(temp_inode)) {
2487                                 ret = PTR_ERR(temp_inode);
2488                                 goto out_put;
2489                         }
2490                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2491                         iput(temp_inode);
2492                         if (ret) {
2493                                 ret = -EACCES;
2494                                 goto out_put;
2495                         }
2496
2497                         if (key.offset == upper_limit.objectid)
2498                                 break;
2499                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2500                                 ret = -EACCES;
2501                                 goto out_put;
2502                         }
2503
2504                         btrfs_release_path(path);
2505                         key.objectid = key.offset;
2506                         key.offset = (u64)-1;
2507                         dirid = key.objectid;
2508                 }
2509
2510                 memmove(args->path, ptr, total_len);
2511                 args->path[total_len] = '\0';
2512                 btrfs_put_root(root);
2513                 root = NULL;
2514                 btrfs_release_path(path);
2515         }
2516
2517         /* Get the bottom subvolume's name from ROOT_REF */
2518         key.objectid = treeid;
2519         key.type = BTRFS_ROOT_REF_KEY;
2520         key.offset = args->treeid;
2521         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2522         if (ret < 0) {
2523                 goto out;
2524         } else if (ret > 0) {
2525                 ret = -ENOENT;
2526                 goto out;
2527         }
2528
2529         leaf = path->nodes[0];
2530         slot = path->slots[0];
2531         btrfs_item_key_to_cpu(leaf, &key, slot);
2532
2533         item_off = btrfs_item_ptr_offset(leaf, slot);
2534         item_len = btrfs_item_size_nr(leaf, slot);
2535         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2536         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2537         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2538                 ret = -EINVAL;
2539                 goto out;
2540         }
2541
2542         /* Copy subvolume's name */
2543         item_off += sizeof(struct btrfs_root_ref);
2544         item_len -= sizeof(struct btrfs_root_ref);
2545         read_extent_buffer(leaf, args->name, item_off, item_len);
2546         args->name[item_len] = 0;
2547
2548 out_put:
2549         btrfs_put_root(root);
2550 out:
2551         btrfs_free_path(path);
2552         return ret;
2553 }
2554
2555 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2556                                            void __user *argp)
2557 {
2558         struct btrfs_ioctl_ino_lookup_args *args;
2559         struct inode *inode;
2560         int ret = 0;
2561
2562         args = memdup_user(argp, sizeof(*args));
2563         if (IS_ERR(args))
2564                 return PTR_ERR(args);
2565
2566         inode = file_inode(file);
2567
2568         /*
2569          * Unprivileged query to obtain the containing subvolume root id. The
2570          * path is reset so it's consistent with btrfs_search_path_in_tree.
2571          */
2572         if (args->treeid == 0)
2573                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2574
2575         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2576                 args->name[0] = 0;
2577                 goto out;
2578         }
2579
2580         if (!capable(CAP_SYS_ADMIN)) {
2581                 ret = -EPERM;
2582                 goto out;
2583         }
2584
2585         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2586                                         args->treeid, args->objectid,
2587                                         args->name);
2588
2589 out:
2590         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2591                 ret = -EFAULT;
2592
2593         kfree(args);
2594         return ret;
2595 }
2596
2597 /*
2598  * Version of ino_lookup ioctl (unprivileged)
2599  *
2600  * The main differences from ino_lookup ioctl are:
2601  *
2602  *   1. Read + Exec permission will be checked using inode_permission() during
2603  *      path construction. -EACCES will be returned in case of failure.
2604  *   2. Path construction will be stopped at the inode number which corresponds
2605  *      to the fd with which this ioctl is called. If constructed path does not
2606  *      exist under fd's inode, -EACCES will be returned.
2607  *   3. The name of bottom subvolume is also searched and filled.
2608  */
2609 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2610 {
2611         struct btrfs_ioctl_ino_lookup_user_args *args;
2612         struct inode *inode;
2613         int ret;
2614
2615         args = memdup_user(argp, sizeof(*args));
2616         if (IS_ERR(args))
2617                 return PTR_ERR(args);
2618
2619         inode = file_inode(file);
2620
2621         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2622             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2623                 /*
2624                  * The subvolume does not exist under fd with which this is
2625                  * called
2626                  */
2627                 kfree(args);
2628                 return -EACCES;
2629         }
2630
2631         ret = btrfs_search_path_in_tree_user(inode, args);
2632
2633         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2634                 ret = -EFAULT;
2635
2636         kfree(args);
2637         return ret;
2638 }
2639
2640 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2641 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2642 {
2643         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2644         struct btrfs_fs_info *fs_info;
2645         struct btrfs_root *root;
2646         struct btrfs_path *path;
2647         struct btrfs_key key;
2648         struct btrfs_root_item *root_item;
2649         struct btrfs_root_ref *rref;
2650         struct extent_buffer *leaf;
2651         unsigned long item_off;
2652         unsigned long item_len;
2653         struct inode *inode;
2654         int slot;
2655         int ret = 0;
2656
2657         path = btrfs_alloc_path();
2658         if (!path)
2659                 return -ENOMEM;
2660
2661         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2662         if (!subvol_info) {
2663                 btrfs_free_path(path);
2664                 return -ENOMEM;
2665         }
2666
2667         inode = file_inode(file);
2668         fs_info = BTRFS_I(inode)->root->fs_info;
2669
2670         /* Get root_item of inode's subvolume */
2671         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2672         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2673         if (IS_ERR(root)) {
2674                 ret = PTR_ERR(root);
2675                 goto out_free;
2676         }
2677         root_item = &root->root_item;
2678
2679         subvol_info->treeid = key.objectid;
2680
2681         subvol_info->generation = btrfs_root_generation(root_item);
2682         subvol_info->flags = btrfs_root_flags(root_item);
2683
2684         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2685         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2686                                                     BTRFS_UUID_SIZE);
2687         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2688                                                     BTRFS_UUID_SIZE);
2689
2690         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2691         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2692         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2693
2694         subvol_info->otransid = btrfs_root_otransid(root_item);
2695         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2696         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2697
2698         subvol_info->stransid = btrfs_root_stransid(root_item);
2699         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2700         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2701
2702         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2703         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2704         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2705
2706         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2707                 /* Search root tree for ROOT_BACKREF of this subvolume */
2708                 key.type = BTRFS_ROOT_BACKREF_KEY;
2709                 key.offset = 0;
2710                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2711                 if (ret < 0) {
2712                         goto out;
2713                 } else if (path->slots[0] >=
2714                            btrfs_header_nritems(path->nodes[0])) {
2715                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2716                         if (ret < 0) {
2717                                 goto out;
2718                         } else if (ret > 0) {
2719                                 ret = -EUCLEAN;
2720                                 goto out;
2721                         }
2722                 }
2723
2724                 leaf = path->nodes[0];
2725                 slot = path->slots[0];
2726                 btrfs_item_key_to_cpu(leaf, &key, slot);
2727                 if (key.objectid == subvol_info->treeid &&
2728                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2729                         subvol_info->parent_id = key.offset;
2730
2731                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2732                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2733
2734                         item_off = btrfs_item_ptr_offset(leaf, slot)
2735                                         + sizeof(struct btrfs_root_ref);
2736                         item_len = btrfs_item_size_nr(leaf, slot)
2737                                         - sizeof(struct btrfs_root_ref);
2738                         read_extent_buffer(leaf, subvol_info->name,
2739                                            item_off, item_len);
2740                 } else {
2741                         ret = -ENOENT;
2742                         goto out;
2743                 }
2744         }
2745
2746         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2747                 ret = -EFAULT;
2748
2749 out:
2750         btrfs_put_root(root);
2751 out_free:
2752         btrfs_free_path(path);
2753         kfree(subvol_info);
2754         return ret;
2755 }
2756
2757 /*
2758  * Return ROOT_REF information of the subvolume containing this inode
2759  * except the subvolume name.
2760  */
2761 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2762 {
2763         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2764         struct btrfs_root_ref *rref;
2765         struct btrfs_root *root;
2766         struct btrfs_path *path;
2767         struct btrfs_key key;
2768         struct extent_buffer *leaf;
2769         struct inode *inode;
2770         u64 objectid;
2771         int slot;
2772         int ret;
2773         u8 found;
2774
2775         path = btrfs_alloc_path();
2776         if (!path)
2777                 return -ENOMEM;
2778
2779         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2780         if (IS_ERR(rootrefs)) {
2781                 btrfs_free_path(path);
2782                 return PTR_ERR(rootrefs);
2783         }
2784
2785         inode = file_inode(file);
2786         root = BTRFS_I(inode)->root->fs_info->tree_root;
2787         objectid = BTRFS_I(inode)->root->root_key.objectid;
2788
2789         key.objectid = objectid;
2790         key.type = BTRFS_ROOT_REF_KEY;
2791         key.offset = rootrefs->min_treeid;
2792         found = 0;
2793
2794         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2795         if (ret < 0) {
2796                 goto out;
2797         } else if (path->slots[0] >=
2798                    btrfs_header_nritems(path->nodes[0])) {
2799                 ret = btrfs_next_leaf(root, path);
2800                 if (ret < 0) {
2801                         goto out;
2802                 } else if (ret > 0) {
2803                         ret = -EUCLEAN;
2804                         goto out;
2805                 }
2806         }
2807         while (1) {
2808                 leaf = path->nodes[0];
2809                 slot = path->slots[0];
2810
2811                 btrfs_item_key_to_cpu(leaf, &key, slot);
2812                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2813                         ret = 0;
2814                         goto out;
2815                 }
2816
2817                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2818                         ret = -EOVERFLOW;
2819                         goto out;
2820                 }
2821
2822                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2823                 rootrefs->rootref[found].treeid = key.offset;
2824                 rootrefs->rootref[found].dirid =
2825                                   btrfs_root_ref_dirid(leaf, rref);
2826                 found++;
2827
2828                 ret = btrfs_next_item(root, path);
2829                 if (ret < 0) {
2830                         goto out;
2831                 } else if (ret > 0) {
2832                         ret = -EUCLEAN;
2833                         goto out;
2834                 }
2835         }
2836
2837 out:
2838         if (!ret || ret == -EOVERFLOW) {
2839                 rootrefs->num_items = found;
2840                 /* update min_treeid for next search */
2841                 if (found)
2842                         rootrefs->min_treeid =
2843                                 rootrefs->rootref[found - 1].treeid + 1;
2844                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2845                         ret = -EFAULT;
2846         }
2847
2848         kfree(rootrefs);
2849         btrfs_free_path(path);
2850
2851         return ret;
2852 }
2853
2854 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2855                                              void __user *arg,
2856                                              bool destroy_v2)
2857 {
2858         struct dentry *parent = file->f_path.dentry;
2859         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2860         struct dentry *dentry;
2861         struct inode *dir = d_inode(parent);
2862         struct inode *inode;
2863         struct btrfs_root *root = BTRFS_I(dir)->root;
2864         struct btrfs_root *dest = NULL;
2865         struct btrfs_ioctl_vol_args *vol_args = NULL;
2866         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2867         char *subvol_name, *subvol_name_ptr = NULL;
2868         int subvol_namelen;
2869         int err = 0;
2870         bool destroy_parent = false;
2871
2872         if (destroy_v2) {
2873                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2874                 if (IS_ERR(vol_args2))
2875                         return PTR_ERR(vol_args2);
2876
2877                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2878                         err = -EOPNOTSUPP;
2879                         goto out;
2880                 }
2881
2882                 /*
2883                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2884                  * name, same as v1 currently does.
2885                  */
2886                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2887                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2888                         subvol_name = vol_args2->name;
2889
2890                         err = mnt_want_write_file(file);
2891                         if (err)
2892                                 goto out;
2893                 } else {
2894                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2895                                 err = -EINVAL;
2896                                 goto out;
2897                         }
2898
2899                         err = mnt_want_write_file(file);
2900                         if (err)
2901                                 goto out;
2902
2903                         dentry = btrfs_get_dentry(fs_info->sb,
2904                                         BTRFS_FIRST_FREE_OBJECTID,
2905                                         vol_args2->subvolid, 0, 0);
2906                         if (IS_ERR(dentry)) {
2907                                 err = PTR_ERR(dentry);
2908                                 goto out_drop_write;
2909                         }
2910
2911                         /*
2912                          * Change the default parent since the subvolume being
2913                          * deleted can be outside of the current mount point.
2914                          */
2915                         parent = btrfs_get_parent(dentry);
2916
2917                         /*
2918                          * At this point dentry->d_name can point to '/' if the
2919                          * subvolume we want to destroy is outsite of the
2920                          * current mount point, so we need to release the
2921                          * current dentry and execute the lookup to return a new
2922                          * one with ->d_name pointing to the
2923                          * <mount point>/subvol_name.
2924                          */
2925                         dput(dentry);
2926                         if (IS_ERR(parent)) {
2927                                 err = PTR_ERR(parent);
2928                                 goto out_drop_write;
2929                         }
2930                         dir = d_inode(parent);
2931
2932                         /*
2933                          * If v2 was used with SPEC_BY_ID, a new parent was
2934                          * allocated since the subvolume can be outside of the
2935                          * current mount point. Later on we need to release this
2936                          * new parent dentry.
2937                          */
2938                         destroy_parent = true;
2939
2940                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2941                                                 fs_info, vol_args2->subvolid);
2942                         if (IS_ERR(subvol_name_ptr)) {
2943                                 err = PTR_ERR(subvol_name_ptr);
2944                                 goto free_parent;
2945                         }
2946                         /* subvol_name_ptr is already NULL termined */
2947                         subvol_name = (char *)kbasename(subvol_name_ptr);
2948                 }
2949         } else {
2950                 vol_args = memdup_user(arg, sizeof(*vol_args));
2951                 if (IS_ERR(vol_args))
2952                         return PTR_ERR(vol_args);
2953
2954                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2955                 subvol_name = vol_args->name;
2956
2957                 err = mnt_want_write_file(file);
2958                 if (err)
2959                         goto out;
2960         }
2961
2962         subvol_namelen = strlen(subvol_name);
2963
2964         if (strchr(subvol_name, '/') ||
2965             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2966                 err = -EINVAL;
2967                 goto free_subvol_name;
2968         }
2969
2970         if (!S_ISDIR(dir->i_mode)) {
2971                 err = -ENOTDIR;
2972                 goto free_subvol_name;
2973         }
2974
2975         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2976         if (err == -EINTR)
2977                 goto free_subvol_name;
2978         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
2979         if (IS_ERR(dentry)) {
2980                 err = PTR_ERR(dentry);
2981                 goto out_unlock_dir;
2982         }
2983
2984         if (d_really_is_negative(dentry)) {
2985                 err = -ENOENT;
2986                 goto out_dput;
2987         }
2988
2989         inode = d_inode(dentry);
2990         dest = BTRFS_I(inode)->root;
2991         if (!capable(CAP_SYS_ADMIN)) {
2992                 /*
2993                  * Regular user.  Only allow this with a special mount
2994                  * option, when the user has write+exec access to the
2995                  * subvol root, and when rmdir(2) would have been
2996                  * allowed.
2997                  *
2998                  * Note that this is _not_ check that the subvol is
2999                  * empty or doesn't contain data that we wouldn't
3000                  * otherwise be able to delete.
3001                  *
3002                  * Users who want to delete empty subvols should try
3003                  * rmdir(2).
3004                  */
3005                 err = -EPERM;
3006                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3007                         goto out_dput;
3008
3009                 /*
3010                  * Do not allow deletion if the parent dir is the same
3011                  * as the dir to be deleted.  That means the ioctl
3012                  * must be called on the dentry referencing the root
3013                  * of the subvol, not a random directory contained
3014                  * within it.
3015                  */
3016                 err = -EINVAL;
3017                 if (root == dest)
3018                         goto out_dput;
3019
3020                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
3021                 if (err)
3022                         goto out_dput;
3023         }
3024
3025         /* check if subvolume may be deleted by a user */
3026         err = btrfs_may_delete(dir, dentry, 1);
3027         if (err)
3028                 goto out_dput;
3029
3030         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3031                 err = -EINVAL;
3032                 goto out_dput;
3033         }
3034
3035         inode_lock(inode);
3036         err = btrfs_delete_subvolume(dir, dentry);
3037         inode_unlock(inode);
3038         if (!err) {
3039                 fsnotify_rmdir(dir, dentry);
3040                 d_delete(dentry);
3041         }
3042
3043 out_dput:
3044         dput(dentry);
3045 out_unlock_dir:
3046         inode_unlock(dir);
3047 free_subvol_name:
3048         kfree(subvol_name_ptr);
3049 free_parent:
3050         if (destroy_parent)
3051                 dput(parent);
3052 out_drop_write:
3053         mnt_drop_write_file(file);
3054 out:
3055         kfree(vol_args2);
3056         kfree(vol_args);
3057         return err;
3058 }
3059
3060 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3061 {
3062         struct inode *inode = file_inode(file);
3063         struct btrfs_root *root = BTRFS_I(inode)->root;
3064         struct btrfs_ioctl_defrag_range_args *range;
3065         int ret;
3066
3067         ret = mnt_want_write_file(file);
3068         if (ret)
3069                 return ret;
3070
3071         if (btrfs_root_readonly(root)) {
3072                 ret = -EROFS;
3073                 goto out;
3074         }
3075
3076         switch (inode->i_mode & S_IFMT) {
3077         case S_IFDIR:
3078                 if (!capable(CAP_SYS_ADMIN)) {
3079                         ret = -EPERM;
3080                         goto out;
3081                 }
3082                 ret = btrfs_defrag_root(root);
3083                 break;
3084         case S_IFREG:
3085                 /*
3086                  * Note that this does not check the file descriptor for write
3087                  * access. This prevents defragmenting executables that are
3088                  * running and allows defrag on files open in read-only mode.
3089                  */
3090                 if (!capable(CAP_SYS_ADMIN) &&
3091                     inode_permission(inode, MAY_WRITE)) {
3092                         ret = -EPERM;
3093                         goto out;
3094                 }
3095
3096                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3097                 if (!range) {
3098                         ret = -ENOMEM;
3099                         goto out;
3100                 }
3101
3102                 if (argp) {
3103                         if (copy_from_user(range, argp,
3104                                            sizeof(*range))) {
3105                                 ret = -EFAULT;
3106                                 kfree(range);
3107                                 goto out;
3108                         }
3109                         /* compression requires us to start the IO */
3110                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3111                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3112                                 range->extent_thresh = (u32)-1;
3113                         }
3114                 } else {
3115                         /* the rest are all set to zero by kzalloc */
3116                         range->len = (u64)-1;
3117                 }
3118                 ret = btrfs_defrag_file(file_inode(file), file,
3119                                         range, BTRFS_OLDEST_GENERATION, 0);
3120                 if (ret > 0)
3121                         ret = 0;
3122                 kfree(range);
3123                 break;
3124         default:
3125                 ret = -EINVAL;
3126         }
3127 out:
3128         mnt_drop_write_file(file);
3129         return ret;
3130 }
3131
3132 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3133 {
3134         struct btrfs_ioctl_vol_args *vol_args;
3135         int ret;
3136
3137         if (!capable(CAP_SYS_ADMIN))
3138                 return -EPERM;
3139
3140         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3141                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3142
3143         vol_args = memdup_user(arg, sizeof(*vol_args));
3144         if (IS_ERR(vol_args)) {
3145                 ret = PTR_ERR(vol_args);
3146                 goto out;
3147         }
3148
3149         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3150         ret = btrfs_init_new_device(fs_info, vol_args->name);
3151
3152         if (!ret)
3153                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3154
3155         kfree(vol_args);
3156 out:
3157         btrfs_exclop_finish(fs_info);
3158         return ret;
3159 }
3160
3161 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3162 {
3163         struct inode *inode = file_inode(file);
3164         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3165         struct btrfs_ioctl_vol_args_v2 *vol_args;
3166         int ret;
3167
3168         if (!capable(CAP_SYS_ADMIN))
3169                 return -EPERM;
3170
3171         ret = mnt_want_write_file(file);
3172         if (ret)
3173                 return ret;
3174
3175         vol_args = memdup_user(arg, sizeof(*vol_args));
3176         if (IS_ERR(vol_args)) {
3177                 ret = PTR_ERR(vol_args);
3178                 goto err_drop;
3179         }
3180
3181         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3182                 ret = -EOPNOTSUPP;
3183                 goto out;
3184         }
3185
3186         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3187                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3188                 goto out;
3189         }
3190
3191         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3192                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3193         } else {
3194                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3195                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3196         }
3197         btrfs_exclop_finish(fs_info);
3198
3199         if (!ret) {
3200                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3201                         btrfs_info(fs_info, "device deleted: id %llu",
3202                                         vol_args->devid);
3203                 else
3204                         btrfs_info(fs_info, "device deleted: %s",
3205                                         vol_args->name);
3206         }
3207 out:
3208         kfree(vol_args);
3209 err_drop:
3210         mnt_drop_write_file(file);
3211         return ret;
3212 }
3213
3214 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3215 {
3216         struct inode *inode = file_inode(file);
3217         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3218         struct btrfs_ioctl_vol_args *vol_args;
3219         int ret;
3220
3221         if (!capable(CAP_SYS_ADMIN))
3222                 return -EPERM;
3223
3224         ret = mnt_want_write_file(file);
3225         if (ret)
3226                 return ret;
3227
3228         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3229                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3230                 goto out_drop_write;
3231         }
3232
3233         vol_args = memdup_user(arg, sizeof(*vol_args));
3234         if (IS_ERR(vol_args)) {
3235                 ret = PTR_ERR(vol_args);
3236                 goto out;
3237         }
3238
3239         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3240         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3241
3242         if (!ret)
3243                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3244         kfree(vol_args);
3245 out:
3246         btrfs_exclop_finish(fs_info);
3247 out_drop_write:
3248         mnt_drop_write_file(file);
3249
3250         return ret;
3251 }
3252
3253 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3254                                 void __user *arg)
3255 {
3256         struct btrfs_ioctl_fs_info_args *fi_args;
3257         struct btrfs_device *device;
3258         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3259         u64 flags_in;
3260         int ret = 0;
3261
3262         fi_args = memdup_user(arg, sizeof(*fi_args));
3263         if (IS_ERR(fi_args))
3264                 return PTR_ERR(fi_args);
3265
3266         flags_in = fi_args->flags;
3267         memset(fi_args, 0, sizeof(*fi_args));
3268
3269         rcu_read_lock();
3270         fi_args->num_devices = fs_devices->num_devices;
3271
3272         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3273                 if (device->devid > fi_args->max_id)
3274                         fi_args->max_id = device->devid;
3275         }
3276         rcu_read_unlock();
3277
3278         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3279         fi_args->nodesize = fs_info->nodesize;
3280         fi_args->sectorsize = fs_info->sectorsize;
3281         fi_args->clone_alignment = fs_info->sectorsize;
3282
3283         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3284                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3285                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3286                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3287         }
3288
3289         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3290                 fi_args->generation = fs_info->generation;
3291                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3292         }
3293
3294         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3295                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3296                        sizeof(fi_args->metadata_uuid));
3297                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3298         }
3299
3300         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3301                 ret = -EFAULT;
3302
3303         kfree(fi_args);
3304         return ret;
3305 }
3306
3307 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3308                                  void __user *arg)
3309 {
3310         struct btrfs_ioctl_dev_info_args *di_args;
3311         struct btrfs_device *dev;
3312         int ret = 0;
3313         char *s_uuid = NULL;
3314
3315         di_args = memdup_user(arg, sizeof(*di_args));
3316         if (IS_ERR(di_args))
3317                 return PTR_ERR(di_args);
3318
3319         if (!btrfs_is_empty_uuid(di_args->uuid))
3320                 s_uuid = di_args->uuid;
3321
3322         rcu_read_lock();
3323         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3324                                 NULL, true);
3325
3326         if (!dev) {
3327                 ret = -ENODEV;
3328                 goto out;
3329         }
3330
3331         di_args->devid = dev->devid;
3332         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3333         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3334         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3335         if (dev->name) {
3336                 strncpy(di_args->path, rcu_str_deref(dev->name),
3337                                 sizeof(di_args->path) - 1);
3338                 di_args->path[sizeof(di_args->path) - 1] = 0;
3339         } else {
3340                 di_args->path[0] = '\0';
3341         }
3342
3343 out:
3344         rcu_read_unlock();
3345         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3346                 ret = -EFAULT;
3347
3348         kfree(di_args);
3349         return ret;
3350 }
3351
3352 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3353 {
3354         struct inode *inode = file_inode(file);
3355         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3356         struct btrfs_root *root = BTRFS_I(inode)->root;
3357         struct btrfs_root *new_root;
3358         struct btrfs_dir_item *di;
3359         struct btrfs_trans_handle *trans;
3360         struct btrfs_path *path = NULL;
3361         struct btrfs_disk_key disk_key;
3362         u64 objectid = 0;
3363         u64 dir_id;
3364         int ret;
3365
3366         if (!capable(CAP_SYS_ADMIN))
3367                 return -EPERM;
3368
3369         ret = mnt_want_write_file(file);
3370         if (ret)
3371                 return ret;
3372
3373         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3374                 ret = -EFAULT;
3375                 goto out;
3376         }
3377
3378         if (!objectid)
3379                 objectid = BTRFS_FS_TREE_OBJECTID;
3380
3381         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3382         if (IS_ERR(new_root)) {
3383                 ret = PTR_ERR(new_root);
3384                 goto out;
3385         }
3386         if (!is_fstree(new_root->root_key.objectid)) {
3387                 ret = -ENOENT;
3388                 goto out_free;
3389         }
3390
3391         path = btrfs_alloc_path();
3392         if (!path) {
3393                 ret = -ENOMEM;
3394                 goto out_free;
3395         }
3396         path->leave_spinning = 1;
3397
3398         trans = btrfs_start_transaction(root, 1);
3399         if (IS_ERR(trans)) {
3400                 ret = PTR_ERR(trans);
3401                 goto out_free;
3402         }
3403
3404         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3405         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3406                                    dir_id, "default", 7, 1);
3407         if (IS_ERR_OR_NULL(di)) {
3408                 btrfs_release_path(path);
3409                 btrfs_end_transaction(trans);
3410                 btrfs_err(fs_info,
3411                           "Umm, you don't have the default diritem, this isn't going to work");
3412                 ret = -ENOENT;
3413                 goto out_free;
3414         }
3415
3416         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3417         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3418         btrfs_mark_buffer_dirty(path->nodes[0]);
3419         btrfs_release_path(path);
3420
3421         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3422         btrfs_end_transaction(trans);
3423 out_free:
3424         btrfs_put_root(new_root);
3425         btrfs_free_path(path);
3426 out:
3427         mnt_drop_write_file(file);
3428         return ret;
3429 }
3430
3431 static void get_block_group_info(struct list_head *groups_list,
3432                                  struct btrfs_ioctl_space_info *space)
3433 {
3434         struct btrfs_block_group *block_group;
3435
3436         space->total_bytes = 0;
3437         space->used_bytes = 0;
3438         space->flags = 0;
3439         list_for_each_entry(block_group, groups_list, list) {
3440                 space->flags = block_group->flags;
3441                 space->total_bytes += block_group->length;
3442                 space->used_bytes += block_group->used;
3443         }
3444 }
3445
3446 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3447                                    void __user *arg)
3448 {
3449         struct btrfs_ioctl_space_args space_args;
3450         struct btrfs_ioctl_space_info space;
3451         struct btrfs_ioctl_space_info *dest;
3452         struct btrfs_ioctl_space_info *dest_orig;
3453         struct btrfs_ioctl_space_info __user *user_dest;
3454         struct btrfs_space_info *info;
3455         static const u64 types[] = {
3456                 BTRFS_BLOCK_GROUP_DATA,
3457                 BTRFS_BLOCK_GROUP_SYSTEM,
3458                 BTRFS_BLOCK_GROUP_METADATA,
3459                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3460         };
3461         int num_types = 4;
3462         int alloc_size;
3463         int ret = 0;
3464         u64 slot_count = 0;
3465         int i, c;
3466
3467         if (copy_from_user(&space_args,
3468                            (struct btrfs_ioctl_space_args __user *)arg,
3469                            sizeof(space_args)))
3470                 return -EFAULT;
3471
3472         for (i = 0; i < num_types; i++) {
3473                 struct btrfs_space_info *tmp;
3474
3475                 info = NULL;
3476                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3477                         if (tmp->flags == types[i]) {
3478                                 info = tmp;
3479                                 break;
3480                         }
3481                 }
3482
3483                 if (!info)
3484                         continue;
3485
3486                 down_read(&info->groups_sem);
3487                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3488                         if (!list_empty(&info->block_groups[c]))
3489                                 slot_count++;
3490                 }
3491                 up_read(&info->groups_sem);
3492         }
3493
3494         /*
3495          * Global block reserve, exported as a space_info
3496          */
3497         slot_count++;
3498
3499         /* space_slots == 0 means they are asking for a count */
3500         if (space_args.space_slots == 0) {
3501                 space_args.total_spaces = slot_count;
3502                 goto out;
3503         }
3504
3505         slot_count = min_t(u64, space_args.space_slots, slot_count);
3506
3507         alloc_size = sizeof(*dest) * slot_count;
3508
3509         /* we generally have at most 6 or so space infos, one for each raid
3510          * level.  So, a whole page should be more than enough for everyone
3511          */
3512         if (alloc_size > PAGE_SIZE)
3513                 return -ENOMEM;
3514
3515         space_args.total_spaces = 0;
3516         dest = kmalloc(alloc_size, GFP_KERNEL);
3517         if (!dest)
3518                 return -ENOMEM;
3519         dest_orig = dest;
3520
3521         /* now we have a buffer to copy into */
3522         for (i = 0; i < num_types; i++) {
3523                 struct btrfs_space_info *tmp;
3524
3525                 if (!slot_count)
3526                         break;
3527
3528                 info = NULL;
3529                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3530                         if (tmp->flags == types[i]) {
3531                                 info = tmp;
3532                                 break;
3533                         }
3534                 }
3535
3536                 if (!info)
3537                         continue;
3538                 down_read(&info->groups_sem);
3539                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3540                         if (!list_empty(&info->block_groups[c])) {
3541                                 get_block_group_info(&info->block_groups[c],
3542                                                      &space);
3543                                 memcpy(dest, &space, sizeof(space));
3544                                 dest++;
3545                                 space_args.total_spaces++;
3546                                 slot_count--;
3547                         }
3548                         if (!slot_count)
3549                                 break;
3550                 }
3551                 up_read(&info->groups_sem);
3552         }
3553
3554         /*
3555          * Add global block reserve
3556          */
3557         if (slot_count) {
3558                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3559
3560                 spin_lock(&block_rsv->lock);
3561                 space.total_bytes = block_rsv->size;
3562                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3563                 spin_unlock(&block_rsv->lock);
3564                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3565                 memcpy(dest, &space, sizeof(space));
3566                 space_args.total_spaces++;
3567         }
3568
3569         user_dest = (struct btrfs_ioctl_space_info __user *)
3570                 (arg + sizeof(struct btrfs_ioctl_space_args));
3571
3572         if (copy_to_user(user_dest, dest_orig, alloc_size))
3573                 ret = -EFAULT;
3574
3575         kfree(dest_orig);
3576 out:
3577         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3578                 ret = -EFAULT;
3579
3580         return ret;
3581 }
3582
3583 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3584                                             void __user *argp)
3585 {
3586         struct btrfs_trans_handle *trans;
3587         u64 transid;
3588         int ret;
3589
3590         trans = btrfs_attach_transaction_barrier(root);
3591         if (IS_ERR(trans)) {
3592                 if (PTR_ERR(trans) != -ENOENT)
3593                         return PTR_ERR(trans);
3594
3595                 /* No running transaction, don't bother */
3596                 transid = root->fs_info->last_trans_committed;
3597                 goto out;
3598         }
3599         transid = trans->transid;
3600         ret = btrfs_commit_transaction_async(trans, 0);
3601         if (ret) {
3602                 btrfs_end_transaction(trans);
3603                 return ret;
3604         }
3605 out:
3606         if (argp)
3607                 if (copy_to_user(argp, &transid, sizeof(transid)))
3608                         return -EFAULT;
3609         return 0;
3610 }
3611
3612 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3613                                            void __user *argp)
3614 {
3615         u64 transid;
3616
3617         if (argp) {
3618                 if (copy_from_user(&transid, argp, sizeof(transid)))
3619                         return -EFAULT;
3620         } else {
3621                 transid = 0;  /* current trans */
3622         }
3623         return btrfs_wait_for_commit(fs_info, transid);
3624 }
3625
3626 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3627 {
3628         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3629         struct btrfs_ioctl_scrub_args *sa;
3630         int ret;
3631
3632         if (!capable(CAP_SYS_ADMIN))
3633                 return -EPERM;
3634
3635         sa = memdup_user(arg, sizeof(*sa));
3636         if (IS_ERR(sa))
3637                 return PTR_ERR(sa);
3638
3639         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3640                 ret = mnt_want_write_file(file);
3641                 if (ret)
3642                         goto out;
3643         }
3644
3645         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3646                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3647                               0);
3648
3649         /*
3650          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3651          * error. This is important as it allows user space to know how much
3652          * progress scrub has done. For example, if scrub is canceled we get
3653          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3654          * space. Later user space can inspect the progress from the structure
3655          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3656          * previously (btrfs-progs does this).
3657          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3658          * then return -EFAULT to signal the structure was not copied or it may
3659          * be corrupt and unreliable due to a partial copy.
3660          */
3661         if (copy_to_user(arg, sa, sizeof(*sa)))
3662                 ret = -EFAULT;
3663
3664         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3665                 mnt_drop_write_file(file);
3666 out:
3667         kfree(sa);
3668         return ret;
3669 }
3670
3671 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3672 {
3673         if (!capable(CAP_SYS_ADMIN))
3674                 return -EPERM;
3675
3676         return btrfs_scrub_cancel(fs_info);
3677 }
3678
3679 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3680                                        void __user *arg)
3681 {
3682         struct btrfs_ioctl_scrub_args *sa;
3683         int ret;
3684
3685         if (!capable(CAP_SYS_ADMIN))
3686                 return -EPERM;
3687
3688         sa = memdup_user(arg, sizeof(*sa));
3689         if (IS_ERR(sa))
3690                 return PTR_ERR(sa);
3691
3692         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3693
3694         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3695                 ret = -EFAULT;
3696
3697         kfree(sa);
3698         return ret;
3699 }
3700
3701 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3702                                       void __user *arg)
3703 {
3704         struct btrfs_ioctl_get_dev_stats *sa;
3705         int ret;
3706
3707         sa = memdup_user(arg, sizeof(*sa));
3708         if (IS_ERR(sa))
3709                 return PTR_ERR(sa);
3710
3711         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3712                 kfree(sa);
3713                 return -EPERM;
3714         }
3715
3716         ret = btrfs_get_dev_stats(fs_info, sa);
3717
3718         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3719                 ret = -EFAULT;
3720
3721         kfree(sa);
3722         return ret;
3723 }
3724
3725 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3726                                     void __user *arg)
3727 {
3728         struct btrfs_ioctl_dev_replace_args *p;
3729         int ret;
3730
3731         if (!capable(CAP_SYS_ADMIN))
3732                 return -EPERM;
3733
3734         p = memdup_user(arg, sizeof(*p));
3735         if (IS_ERR(p))
3736                 return PTR_ERR(p);
3737
3738         switch (p->cmd) {
3739         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3740                 if (sb_rdonly(fs_info->sb)) {
3741                         ret = -EROFS;
3742                         goto out;
3743                 }
3744                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3745                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3746                 } else {
3747                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3748                         btrfs_exclop_finish(fs_info);
3749                 }
3750                 break;
3751         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3752                 btrfs_dev_replace_status(fs_info, p);
3753                 ret = 0;
3754                 break;
3755         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3756                 p->result = btrfs_dev_replace_cancel(fs_info);
3757                 ret = 0;
3758                 break;
3759         default:
3760                 ret = -EINVAL;
3761                 break;
3762         }
3763
3764         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3765                 ret = -EFAULT;
3766 out:
3767         kfree(p);
3768         return ret;
3769 }
3770
3771 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3772 {
3773         int ret = 0;
3774         int i;
3775         u64 rel_ptr;
3776         int size;
3777         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3778         struct inode_fs_paths *ipath = NULL;
3779         struct btrfs_path *path;
3780
3781         if (!capable(CAP_DAC_READ_SEARCH))
3782                 return -EPERM;
3783
3784         path = btrfs_alloc_path();
3785         if (!path) {
3786                 ret = -ENOMEM;
3787                 goto out;
3788         }
3789
3790         ipa = memdup_user(arg, sizeof(*ipa));
3791         if (IS_ERR(ipa)) {
3792                 ret = PTR_ERR(ipa);
3793                 ipa = NULL;
3794                 goto out;
3795         }
3796
3797         size = min_t(u32, ipa->size, 4096);
3798         ipath = init_ipath(size, root, path);
3799         if (IS_ERR(ipath)) {
3800                 ret = PTR_ERR(ipath);
3801                 ipath = NULL;
3802                 goto out;
3803         }
3804
3805         ret = paths_from_inode(ipa->inum, ipath);
3806         if (ret < 0)
3807                 goto out;
3808
3809         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3810                 rel_ptr = ipath->fspath->val[i] -
3811                           (u64)(unsigned long)ipath->fspath->val;
3812                 ipath->fspath->val[i] = rel_ptr;
3813         }
3814
3815         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3816                            ipath->fspath, size);
3817         if (ret) {
3818                 ret = -EFAULT;
3819                 goto out;
3820         }
3821
3822 out:
3823         btrfs_free_path(path);
3824         free_ipath(ipath);
3825         kfree(ipa);
3826
3827         return ret;
3828 }
3829
3830 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3831 {
3832         struct btrfs_data_container *inodes = ctx;
3833         const size_t c = 3 * sizeof(u64);
3834
3835         if (inodes->bytes_left >= c) {
3836                 inodes->bytes_left -= c;
3837                 inodes->val[inodes->elem_cnt] = inum;
3838                 inodes->val[inodes->elem_cnt + 1] = offset;
3839                 inodes->val[inodes->elem_cnt + 2] = root;
3840                 inodes->elem_cnt += 3;
3841         } else {
3842                 inodes->bytes_missing += c - inodes->bytes_left;
3843                 inodes->bytes_left = 0;
3844                 inodes->elem_missed += 3;
3845         }
3846
3847         return 0;
3848 }
3849
3850 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3851                                         void __user *arg, int version)
3852 {
3853         int ret = 0;
3854         int size;
3855         struct btrfs_ioctl_logical_ino_args *loi;
3856         struct btrfs_data_container *inodes = NULL;
3857         struct btrfs_path *path = NULL;
3858         bool ignore_offset;
3859
3860         if (!capable(CAP_SYS_ADMIN))
3861                 return -EPERM;
3862
3863         loi = memdup_user(arg, sizeof(*loi));
3864         if (IS_ERR(loi))
3865                 return PTR_ERR(loi);
3866
3867         if (version == 1) {
3868                 ignore_offset = false;
3869                 size = min_t(u32, loi->size, SZ_64K);
3870         } else {
3871                 /* All reserved bits must be 0 for now */
3872                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3873                         ret = -EINVAL;
3874                         goto out_loi;
3875                 }
3876                 /* Only accept flags we have defined so far */
3877                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3878                         ret = -EINVAL;
3879                         goto out_loi;
3880                 }
3881                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3882                 size = min_t(u32, loi->size, SZ_16M);
3883         }
3884
3885         path = btrfs_alloc_path();
3886         if (!path) {
3887                 ret = -ENOMEM;
3888                 goto out;
3889         }
3890
3891         inodes = init_data_container(size);
3892         if (IS_ERR(inodes)) {
3893                 ret = PTR_ERR(inodes);
3894                 inodes = NULL;
3895                 goto out;
3896         }
3897
3898         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3899                                           build_ino_list, inodes, ignore_offset);
3900         if (ret == -EINVAL)
3901                 ret = -ENOENT;
3902         if (ret < 0)
3903                 goto out;
3904
3905         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3906                            size);
3907         if (ret)
3908                 ret = -EFAULT;
3909
3910 out:
3911         btrfs_free_path(path);
3912         kvfree(inodes);
3913 out_loi:
3914         kfree(loi);
3915
3916         return ret;
3917 }
3918
3919 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3920                                struct btrfs_ioctl_balance_args *bargs)
3921 {
3922         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3923
3924         bargs->flags = bctl->flags;
3925
3926         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3927                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3928         if (atomic_read(&fs_info->balance_pause_req))
3929                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3930         if (atomic_read(&fs_info->balance_cancel_req))
3931                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3932
3933         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3934         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3935         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3936
3937         spin_lock(&fs_info->balance_lock);
3938         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3939         spin_unlock(&fs_info->balance_lock);
3940 }
3941
3942 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3943 {
3944         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3945         struct btrfs_fs_info *fs_info = root->fs_info;
3946         struct btrfs_ioctl_balance_args *bargs;
3947         struct btrfs_balance_control *bctl;
3948         bool need_unlock; /* for mut. excl. ops lock */
3949         int ret;
3950
3951         if (!capable(CAP_SYS_ADMIN))
3952                 return -EPERM;
3953
3954         ret = mnt_want_write_file(file);
3955         if (ret)
3956                 return ret;
3957
3958 again:
3959         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
3960                 mutex_lock(&fs_info->balance_mutex);
3961                 need_unlock = true;
3962                 goto locked;
3963         }
3964
3965         /*
3966          * mut. excl. ops lock is locked.  Three possibilities:
3967          *   (1) some other op is running
3968          *   (2) balance is running
3969          *   (3) balance is paused -- special case (think resume)
3970          */
3971         mutex_lock(&fs_info->balance_mutex);
3972         if (fs_info->balance_ctl) {
3973                 /* this is either (2) or (3) */
3974                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3975                         mutex_unlock(&fs_info->balance_mutex);
3976                         /*
3977                          * Lock released to allow other waiters to continue,
3978                          * we'll reexamine the status again.
3979                          */
3980                         mutex_lock(&fs_info->balance_mutex);
3981
3982                         if (fs_info->balance_ctl &&
3983                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3984                                 /* this is (3) */
3985                                 need_unlock = false;
3986                                 goto locked;
3987                         }
3988
3989                         mutex_unlock(&fs_info->balance_mutex);
3990                         goto again;
3991                 } else {
3992                         /* this is (2) */
3993                         mutex_unlock(&fs_info->balance_mutex);
3994                         ret = -EINPROGRESS;
3995                         goto out;
3996                 }
3997         } else {
3998                 /* this is (1) */
3999                 mutex_unlock(&fs_info->balance_mutex);
4000                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4001                 goto out;
4002         }
4003
4004 locked:
4005
4006         if (arg) {
4007                 bargs = memdup_user(arg, sizeof(*bargs));
4008                 if (IS_ERR(bargs)) {
4009                         ret = PTR_ERR(bargs);
4010                         goto out_unlock;
4011                 }
4012
4013                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4014                         if (!fs_info->balance_ctl) {
4015                                 ret = -ENOTCONN;
4016                                 goto out_bargs;
4017                         }
4018
4019                         bctl = fs_info->balance_ctl;
4020                         spin_lock(&fs_info->balance_lock);
4021                         bctl->flags |= BTRFS_BALANCE_RESUME;
4022                         spin_unlock(&fs_info->balance_lock);
4023
4024                         goto do_balance;
4025                 }
4026         } else {
4027                 bargs = NULL;
4028         }
4029
4030         if (fs_info->balance_ctl) {
4031                 ret = -EINPROGRESS;
4032                 goto out_bargs;
4033         }
4034
4035         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4036         if (!bctl) {
4037                 ret = -ENOMEM;
4038                 goto out_bargs;
4039         }
4040
4041         if (arg) {
4042                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4043                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4044                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4045
4046                 bctl->flags = bargs->flags;
4047         } else {
4048                 /* balance everything - no filters */
4049                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4050         }
4051
4052         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4053                 ret = -EINVAL;
4054                 goto out_bctl;
4055         }
4056
4057 do_balance:
4058         /*
4059          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4060          * bctl is freed in reset_balance_state, or, if restriper was paused
4061          * all the way until unmount, in free_fs_info.  The flag should be
4062          * cleared after reset_balance_state.
4063          */
4064         need_unlock = false;
4065
4066         ret = btrfs_balance(fs_info, bctl, bargs);
4067         bctl = NULL;
4068
4069         if ((ret == 0 || ret == -ECANCELED) && arg) {
4070                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4071                         ret = -EFAULT;
4072         }
4073
4074 out_bctl:
4075         kfree(bctl);
4076 out_bargs:
4077         kfree(bargs);
4078 out_unlock:
4079         mutex_unlock(&fs_info->balance_mutex);
4080         if (need_unlock)
4081                 btrfs_exclop_finish(fs_info);
4082 out:
4083         mnt_drop_write_file(file);
4084         return ret;
4085 }
4086
4087 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4088 {
4089         if (!capable(CAP_SYS_ADMIN))
4090                 return -EPERM;
4091
4092         switch (cmd) {
4093         case BTRFS_BALANCE_CTL_PAUSE:
4094                 return btrfs_pause_balance(fs_info);
4095         case BTRFS_BALANCE_CTL_CANCEL:
4096                 return btrfs_cancel_balance(fs_info);
4097         }
4098
4099         return -EINVAL;
4100 }
4101
4102 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4103                                          void __user *arg)
4104 {
4105         struct btrfs_ioctl_balance_args *bargs;
4106         int ret = 0;
4107
4108         if (!capable(CAP_SYS_ADMIN))
4109                 return -EPERM;
4110
4111         mutex_lock(&fs_info->balance_mutex);
4112         if (!fs_info->balance_ctl) {
4113                 ret = -ENOTCONN;
4114                 goto out;
4115         }
4116
4117         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4118         if (!bargs) {
4119                 ret = -ENOMEM;
4120                 goto out;
4121         }
4122
4123         btrfs_update_ioctl_balance_args(fs_info, bargs);
4124
4125         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4126                 ret = -EFAULT;
4127
4128         kfree(bargs);
4129 out:
4130         mutex_unlock(&fs_info->balance_mutex);
4131         return ret;
4132 }
4133
4134 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4135 {
4136         struct inode *inode = file_inode(file);
4137         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4138         struct btrfs_ioctl_quota_ctl_args *sa;
4139         int ret;
4140
4141         if (!capable(CAP_SYS_ADMIN))
4142                 return -EPERM;
4143
4144         ret = mnt_want_write_file(file);
4145         if (ret)
4146                 return ret;
4147
4148         sa = memdup_user(arg, sizeof(*sa));
4149         if (IS_ERR(sa)) {
4150                 ret = PTR_ERR(sa);
4151                 goto drop_write;
4152         }
4153
4154         down_write(&fs_info->subvol_sem);
4155
4156         switch (sa->cmd) {
4157         case BTRFS_QUOTA_CTL_ENABLE:
4158                 ret = btrfs_quota_enable(fs_info);
4159                 break;
4160         case BTRFS_QUOTA_CTL_DISABLE:
4161                 ret = btrfs_quota_disable(fs_info);
4162                 break;
4163         default:
4164                 ret = -EINVAL;
4165                 break;
4166         }
4167
4168         kfree(sa);
4169         up_write(&fs_info->subvol_sem);
4170 drop_write:
4171         mnt_drop_write_file(file);
4172         return ret;
4173 }
4174
4175 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4176 {
4177         struct inode *inode = file_inode(file);
4178         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4179         struct btrfs_root *root = BTRFS_I(inode)->root;
4180         struct btrfs_ioctl_qgroup_assign_args *sa;
4181         struct btrfs_trans_handle *trans;
4182         int ret;
4183         int err;
4184
4185         if (!capable(CAP_SYS_ADMIN))
4186                 return -EPERM;
4187
4188         ret = mnt_want_write_file(file);
4189         if (ret)
4190                 return ret;
4191
4192         sa = memdup_user(arg, sizeof(*sa));
4193         if (IS_ERR(sa)) {
4194                 ret = PTR_ERR(sa);
4195                 goto drop_write;
4196         }
4197
4198         trans = btrfs_join_transaction(root);
4199         if (IS_ERR(trans)) {
4200                 ret = PTR_ERR(trans);
4201                 goto out;
4202         }
4203
4204         if (sa->assign) {
4205                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4206         } else {
4207                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4208         }
4209
4210         /* update qgroup status and info */
4211         err = btrfs_run_qgroups(trans);
4212         if (err < 0)
4213                 btrfs_handle_fs_error(fs_info, err,
4214                                       "failed to update qgroup status and info");
4215         err = btrfs_end_transaction(trans);
4216         if (err && !ret)
4217                 ret = err;
4218
4219 out:
4220         kfree(sa);
4221 drop_write:
4222         mnt_drop_write_file(file);
4223         return ret;
4224 }
4225
4226 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4227 {
4228         struct inode *inode = file_inode(file);
4229         struct btrfs_root *root = BTRFS_I(inode)->root;
4230         struct btrfs_ioctl_qgroup_create_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         if (!sa->qgroupid) {
4249                 ret = -EINVAL;
4250                 goto out;
4251         }
4252
4253         trans = btrfs_join_transaction(root);
4254         if (IS_ERR(trans)) {
4255                 ret = PTR_ERR(trans);
4256                 goto out;
4257         }
4258
4259         if (sa->create) {
4260                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4261         } else {
4262                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4263         }
4264
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_limit(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_limit_args *sa;
4281         struct btrfs_trans_handle *trans;
4282         int ret;
4283         int err;
4284         u64 qgroupid;
4285
4286         if (!capable(CAP_SYS_ADMIN))
4287                 return -EPERM;
4288
4289         ret = mnt_want_write_file(file);
4290         if (ret)
4291                 return ret;
4292
4293         sa = memdup_user(arg, sizeof(*sa));
4294         if (IS_ERR(sa)) {
4295                 ret = PTR_ERR(sa);
4296                 goto drop_write;
4297         }
4298
4299         trans = btrfs_join_transaction(root);
4300         if (IS_ERR(trans)) {
4301                 ret = PTR_ERR(trans);
4302                 goto out;
4303         }
4304
4305         qgroupid = sa->qgroupid;
4306         if (!qgroupid) {
4307                 /* take the current subvol as qgroup */
4308                 qgroupid = root->root_key.objectid;
4309         }
4310
4311         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4312
4313         err = btrfs_end_transaction(trans);
4314         if (err && !ret)
4315                 ret = err;
4316
4317 out:
4318         kfree(sa);
4319 drop_write:
4320         mnt_drop_write_file(file);
4321         return ret;
4322 }
4323
4324 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4325 {
4326         struct inode *inode = file_inode(file);
4327         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4328         struct btrfs_ioctl_quota_rescan_args *qsa;
4329         int ret;
4330
4331         if (!capable(CAP_SYS_ADMIN))
4332                 return -EPERM;
4333
4334         ret = mnt_want_write_file(file);
4335         if (ret)
4336                 return ret;
4337
4338         qsa = memdup_user(arg, sizeof(*qsa));
4339         if (IS_ERR(qsa)) {
4340                 ret = PTR_ERR(qsa);
4341                 goto drop_write;
4342         }
4343
4344         if (qsa->flags) {
4345                 ret = -EINVAL;
4346                 goto out;
4347         }
4348
4349         ret = btrfs_qgroup_rescan(fs_info);
4350
4351 out:
4352         kfree(qsa);
4353 drop_write:
4354         mnt_drop_write_file(file);
4355         return ret;
4356 }
4357
4358 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4359                                                 void __user *arg)
4360 {
4361         struct btrfs_ioctl_quota_rescan_args *qsa;
4362         int ret = 0;
4363
4364         if (!capable(CAP_SYS_ADMIN))
4365                 return -EPERM;
4366
4367         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4368         if (!qsa)
4369                 return -ENOMEM;
4370
4371         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4372                 qsa->flags = 1;
4373                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4374         }
4375
4376         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4377                 ret = -EFAULT;
4378
4379         kfree(qsa);
4380         return ret;
4381 }
4382
4383 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4384                                                 void __user *arg)
4385 {
4386         if (!capable(CAP_SYS_ADMIN))
4387                 return -EPERM;
4388
4389         return btrfs_qgroup_wait_for_completion(fs_info, true);
4390 }
4391
4392 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4393                                             struct btrfs_ioctl_received_subvol_args *sa)
4394 {
4395         struct inode *inode = file_inode(file);
4396         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4397         struct btrfs_root *root = BTRFS_I(inode)->root;
4398         struct btrfs_root_item *root_item = &root->root_item;
4399         struct btrfs_trans_handle *trans;
4400         struct timespec64 ct = current_time(inode);
4401         int ret = 0;
4402         int received_uuid_changed;
4403
4404         if (!inode_owner_or_capable(inode))
4405                 return -EPERM;
4406
4407         ret = mnt_want_write_file(file);
4408         if (ret < 0)
4409                 return ret;
4410
4411         down_write(&fs_info->subvol_sem);
4412
4413         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4414                 ret = -EINVAL;
4415                 goto out;
4416         }
4417
4418         if (btrfs_root_readonly(root)) {
4419                 ret = -EROFS;
4420                 goto out;
4421         }
4422
4423         /*
4424          * 1 - root item
4425          * 2 - uuid items (received uuid + subvol uuid)
4426          */
4427         trans = btrfs_start_transaction(root, 3);
4428         if (IS_ERR(trans)) {
4429                 ret = PTR_ERR(trans);
4430                 trans = NULL;
4431                 goto out;
4432         }
4433
4434         sa->rtransid = trans->transid;
4435         sa->rtime.sec = ct.tv_sec;
4436         sa->rtime.nsec = ct.tv_nsec;
4437
4438         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4439                                        BTRFS_UUID_SIZE);
4440         if (received_uuid_changed &&
4441             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4442                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4443                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4444                                           root->root_key.objectid);
4445                 if (ret && ret != -ENOENT) {
4446                         btrfs_abort_transaction(trans, ret);
4447                         btrfs_end_transaction(trans);
4448                         goto out;
4449                 }
4450         }
4451         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4452         btrfs_set_root_stransid(root_item, sa->stransid);
4453         btrfs_set_root_rtransid(root_item, sa->rtransid);
4454         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4455         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4456         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4457         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4458
4459         ret = btrfs_update_root(trans, fs_info->tree_root,
4460                                 &root->root_key, &root->root_item);
4461         if (ret < 0) {
4462                 btrfs_end_transaction(trans);
4463                 goto out;
4464         }
4465         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4466                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4467                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4468                                           root->root_key.objectid);
4469                 if (ret < 0 && ret != -EEXIST) {
4470                         btrfs_abort_transaction(trans, ret);
4471                         btrfs_end_transaction(trans);
4472                         goto out;
4473                 }
4474         }
4475         ret = btrfs_commit_transaction(trans);
4476 out:
4477         up_write(&fs_info->subvol_sem);
4478         mnt_drop_write_file(file);
4479         return ret;
4480 }
4481
4482 #ifdef CONFIG_64BIT
4483 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4484                                                 void __user *arg)
4485 {
4486         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4487         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4488         int ret = 0;
4489
4490         args32 = memdup_user(arg, sizeof(*args32));
4491         if (IS_ERR(args32))
4492                 return PTR_ERR(args32);
4493
4494         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4495         if (!args64) {
4496                 ret = -ENOMEM;
4497                 goto out;
4498         }
4499
4500         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4501         args64->stransid = args32->stransid;
4502         args64->rtransid = args32->rtransid;
4503         args64->stime.sec = args32->stime.sec;
4504         args64->stime.nsec = args32->stime.nsec;
4505         args64->rtime.sec = args32->rtime.sec;
4506         args64->rtime.nsec = args32->rtime.nsec;
4507         args64->flags = args32->flags;
4508
4509         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4510         if (ret)
4511                 goto out;
4512
4513         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4514         args32->stransid = args64->stransid;
4515         args32->rtransid = args64->rtransid;
4516         args32->stime.sec = args64->stime.sec;
4517         args32->stime.nsec = args64->stime.nsec;
4518         args32->rtime.sec = args64->rtime.sec;
4519         args32->rtime.nsec = args64->rtime.nsec;
4520         args32->flags = args64->flags;
4521
4522         ret = copy_to_user(arg, args32, sizeof(*args32));
4523         if (ret)
4524                 ret = -EFAULT;
4525
4526 out:
4527         kfree(args32);
4528         kfree(args64);
4529         return ret;
4530 }
4531 #endif
4532
4533 static long btrfs_ioctl_set_received_subvol(struct file *file,
4534                                             void __user *arg)
4535 {
4536         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4537         int ret = 0;
4538
4539         sa = memdup_user(arg, sizeof(*sa));
4540         if (IS_ERR(sa))
4541                 return PTR_ERR(sa);
4542
4543         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4544
4545         if (ret)
4546                 goto out;
4547
4548         ret = copy_to_user(arg, sa, sizeof(*sa));
4549         if (ret)
4550                 ret = -EFAULT;
4551
4552 out:
4553         kfree(sa);
4554         return ret;
4555 }
4556
4557 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4558                                         void __user *arg)
4559 {
4560         size_t len;
4561         int ret;
4562         char label[BTRFS_LABEL_SIZE];
4563
4564         spin_lock(&fs_info->super_lock);
4565         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4566         spin_unlock(&fs_info->super_lock);
4567
4568         len = strnlen(label, BTRFS_LABEL_SIZE);
4569
4570         if (len == BTRFS_LABEL_SIZE) {
4571                 btrfs_warn(fs_info,
4572                            "label is too long, return the first %zu bytes",
4573                            --len);
4574         }
4575
4576         ret = copy_to_user(arg, label, len);
4577
4578         return ret ? -EFAULT : 0;
4579 }
4580
4581 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4582 {
4583         struct inode *inode = file_inode(file);
4584         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4585         struct btrfs_root *root = BTRFS_I(inode)->root;
4586         struct btrfs_super_block *super_block = fs_info->super_copy;
4587         struct btrfs_trans_handle *trans;
4588         char label[BTRFS_LABEL_SIZE];
4589         int ret;
4590
4591         if (!capable(CAP_SYS_ADMIN))
4592                 return -EPERM;
4593
4594         if (copy_from_user(label, arg, sizeof(label)))
4595                 return -EFAULT;
4596
4597         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4598                 btrfs_err(fs_info,
4599                           "unable to set label with more than %d bytes",
4600                           BTRFS_LABEL_SIZE - 1);
4601                 return -EINVAL;
4602         }
4603
4604         ret = mnt_want_write_file(file);
4605         if (ret)
4606                 return ret;
4607
4608         trans = btrfs_start_transaction(root, 0);
4609         if (IS_ERR(trans)) {
4610                 ret = PTR_ERR(trans);
4611                 goto out_unlock;
4612         }
4613
4614         spin_lock(&fs_info->super_lock);
4615         strcpy(super_block->label, label);
4616         spin_unlock(&fs_info->super_lock);
4617         ret = btrfs_commit_transaction(trans);
4618
4619 out_unlock:
4620         mnt_drop_write_file(file);
4621         return ret;
4622 }
4623
4624 #define INIT_FEATURE_FLAGS(suffix) \
4625         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4626           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4627           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4628
4629 int btrfs_ioctl_get_supported_features(void __user *arg)
4630 {
4631         static const struct btrfs_ioctl_feature_flags features[3] = {
4632                 INIT_FEATURE_FLAGS(SUPP),
4633                 INIT_FEATURE_FLAGS(SAFE_SET),
4634                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4635         };
4636
4637         if (copy_to_user(arg, &features, sizeof(features)))
4638                 return -EFAULT;
4639
4640         return 0;
4641 }
4642
4643 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4644                                         void __user *arg)
4645 {
4646         struct btrfs_super_block *super_block = fs_info->super_copy;
4647         struct btrfs_ioctl_feature_flags features;
4648
4649         features.compat_flags = btrfs_super_compat_flags(super_block);
4650         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4651         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4652
4653         if (copy_to_user(arg, &features, sizeof(features)))
4654                 return -EFAULT;
4655
4656         return 0;
4657 }
4658
4659 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4660                               enum btrfs_feature_set set,
4661                               u64 change_mask, u64 flags, u64 supported_flags,
4662                               u64 safe_set, u64 safe_clear)
4663 {
4664         const char *type = btrfs_feature_set_name(set);
4665         char *names;
4666         u64 disallowed, unsupported;
4667         u64 set_mask = flags & change_mask;
4668         u64 clear_mask = ~flags & change_mask;
4669
4670         unsupported = set_mask & ~supported_flags;
4671         if (unsupported) {
4672                 names = btrfs_printable_features(set, unsupported);
4673                 if (names) {
4674                         btrfs_warn(fs_info,
4675                                    "this kernel does not support the %s feature bit%s",
4676                                    names, strchr(names, ',') ? "s" : "");
4677                         kfree(names);
4678                 } else
4679                         btrfs_warn(fs_info,
4680                                    "this kernel does not support %s bits 0x%llx",
4681                                    type, unsupported);
4682                 return -EOPNOTSUPP;
4683         }
4684
4685         disallowed = set_mask & ~safe_set;
4686         if (disallowed) {
4687                 names = btrfs_printable_features(set, disallowed);
4688                 if (names) {
4689                         btrfs_warn(fs_info,
4690                                    "can't set the %s feature bit%s while mounted",
4691                                    names, strchr(names, ',') ? "s" : "");
4692                         kfree(names);
4693                 } else
4694                         btrfs_warn(fs_info,
4695                                    "can't set %s bits 0x%llx while mounted",
4696                                    type, disallowed);
4697                 return -EPERM;
4698         }
4699
4700         disallowed = clear_mask & ~safe_clear;
4701         if (disallowed) {
4702                 names = btrfs_printable_features(set, disallowed);
4703                 if (names) {
4704                         btrfs_warn(fs_info,
4705                                    "can't clear the %s feature bit%s while mounted",
4706                                    names, strchr(names, ',') ? "s" : "");
4707                         kfree(names);
4708                 } else
4709                         btrfs_warn(fs_info,
4710                                    "can't clear %s bits 0x%llx while mounted",
4711                                    type, disallowed);
4712                 return -EPERM;
4713         }
4714
4715         return 0;
4716 }
4717
4718 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4719 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4720                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4721                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4722                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4723
4724 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4725 {
4726         struct inode *inode = file_inode(file);
4727         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4728         struct btrfs_root *root = BTRFS_I(inode)->root;
4729         struct btrfs_super_block *super_block = fs_info->super_copy;
4730         struct btrfs_ioctl_feature_flags flags[2];
4731         struct btrfs_trans_handle *trans;
4732         u64 newflags;
4733         int ret;
4734
4735         if (!capable(CAP_SYS_ADMIN))
4736                 return -EPERM;
4737
4738         if (copy_from_user(flags, arg, sizeof(flags)))
4739                 return -EFAULT;
4740
4741         /* Nothing to do */
4742         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4743             !flags[0].incompat_flags)
4744                 return 0;
4745
4746         ret = check_feature(fs_info, flags[0].compat_flags,
4747                             flags[1].compat_flags, COMPAT);
4748         if (ret)
4749                 return ret;
4750
4751         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4752                             flags[1].compat_ro_flags, COMPAT_RO);
4753         if (ret)
4754                 return ret;
4755
4756         ret = check_feature(fs_info, flags[0].incompat_flags,
4757                             flags[1].incompat_flags, INCOMPAT);
4758         if (ret)
4759                 return ret;
4760
4761         ret = mnt_want_write_file(file);
4762         if (ret)
4763                 return ret;
4764
4765         trans = btrfs_start_transaction(root, 0);
4766         if (IS_ERR(trans)) {
4767                 ret = PTR_ERR(trans);
4768                 goto out_drop_write;
4769         }
4770
4771         spin_lock(&fs_info->super_lock);
4772         newflags = btrfs_super_compat_flags(super_block);
4773         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4774         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4775         btrfs_set_super_compat_flags(super_block, newflags);
4776
4777         newflags = btrfs_super_compat_ro_flags(super_block);
4778         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4779         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4780         btrfs_set_super_compat_ro_flags(super_block, newflags);
4781
4782         newflags = btrfs_super_incompat_flags(super_block);
4783         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4784         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4785         btrfs_set_super_incompat_flags(super_block, newflags);
4786         spin_unlock(&fs_info->super_lock);
4787
4788         ret = btrfs_commit_transaction(trans);
4789 out_drop_write:
4790         mnt_drop_write_file(file);
4791
4792         return ret;
4793 }
4794
4795 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4796 {
4797         struct btrfs_ioctl_send_args *arg;
4798         int ret;
4799
4800         if (compat) {
4801 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4802                 struct btrfs_ioctl_send_args_32 args32;
4803
4804                 ret = copy_from_user(&args32, argp, sizeof(args32));
4805                 if (ret)
4806                         return -EFAULT;
4807                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4808                 if (!arg)
4809                         return -ENOMEM;
4810                 arg->send_fd = args32.send_fd;
4811                 arg->clone_sources_count = args32.clone_sources_count;
4812                 arg->clone_sources = compat_ptr(args32.clone_sources);
4813                 arg->parent_root = args32.parent_root;
4814                 arg->flags = args32.flags;
4815                 memcpy(arg->reserved, args32.reserved,
4816                        sizeof(args32.reserved));
4817 #else
4818                 return -ENOTTY;
4819 #endif
4820         } else {
4821                 arg = memdup_user(argp, sizeof(*arg));
4822                 if (IS_ERR(arg))
4823                         return PTR_ERR(arg);
4824         }
4825         ret = btrfs_ioctl_send(file, arg);
4826         kfree(arg);
4827         return ret;
4828 }
4829
4830 long btrfs_ioctl(struct file *file, unsigned int
4831                 cmd, unsigned long arg)
4832 {
4833         struct inode *inode = file_inode(file);
4834         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4835         struct btrfs_root *root = BTRFS_I(inode)->root;
4836         void __user *argp = (void __user *)arg;
4837
4838         switch (cmd) {
4839         case FS_IOC_GETFLAGS:
4840                 return btrfs_ioctl_getflags(file, argp);
4841         case FS_IOC_SETFLAGS:
4842                 return btrfs_ioctl_setflags(file, argp);
4843         case FS_IOC_GETVERSION:
4844                 return btrfs_ioctl_getversion(file, argp);
4845         case FS_IOC_GETFSLABEL:
4846                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4847         case FS_IOC_SETFSLABEL:
4848                 return btrfs_ioctl_set_fslabel(file, argp);
4849         case FITRIM:
4850                 return btrfs_ioctl_fitrim(fs_info, argp);
4851         case BTRFS_IOC_SNAP_CREATE:
4852                 return btrfs_ioctl_snap_create(file, argp, 0);
4853         case BTRFS_IOC_SNAP_CREATE_V2:
4854                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4855         case BTRFS_IOC_SUBVOL_CREATE:
4856                 return btrfs_ioctl_snap_create(file, argp, 1);
4857         case BTRFS_IOC_SUBVOL_CREATE_V2:
4858                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4859         case BTRFS_IOC_SNAP_DESTROY:
4860                 return btrfs_ioctl_snap_destroy(file, argp, false);
4861         case BTRFS_IOC_SNAP_DESTROY_V2:
4862                 return btrfs_ioctl_snap_destroy(file, argp, true);
4863         case BTRFS_IOC_SUBVOL_GETFLAGS:
4864                 return btrfs_ioctl_subvol_getflags(file, argp);
4865         case BTRFS_IOC_SUBVOL_SETFLAGS:
4866                 return btrfs_ioctl_subvol_setflags(file, argp);
4867         case BTRFS_IOC_DEFAULT_SUBVOL:
4868                 return btrfs_ioctl_default_subvol(file, argp);
4869         case BTRFS_IOC_DEFRAG:
4870                 return btrfs_ioctl_defrag(file, NULL);
4871         case BTRFS_IOC_DEFRAG_RANGE:
4872                 return btrfs_ioctl_defrag(file, argp);
4873         case BTRFS_IOC_RESIZE:
4874                 return btrfs_ioctl_resize(file, argp);
4875         case BTRFS_IOC_ADD_DEV:
4876                 return btrfs_ioctl_add_dev(fs_info, argp);
4877         case BTRFS_IOC_RM_DEV:
4878                 return btrfs_ioctl_rm_dev(file, argp);
4879         case BTRFS_IOC_RM_DEV_V2:
4880                 return btrfs_ioctl_rm_dev_v2(file, argp);
4881         case BTRFS_IOC_FS_INFO:
4882                 return btrfs_ioctl_fs_info(fs_info, argp);
4883         case BTRFS_IOC_DEV_INFO:
4884                 return btrfs_ioctl_dev_info(fs_info, argp);
4885         case BTRFS_IOC_BALANCE:
4886                 return btrfs_ioctl_balance(file, NULL);
4887         case BTRFS_IOC_TREE_SEARCH:
4888                 return btrfs_ioctl_tree_search(file, argp);
4889         case BTRFS_IOC_TREE_SEARCH_V2:
4890                 return btrfs_ioctl_tree_search_v2(file, argp);
4891         case BTRFS_IOC_INO_LOOKUP:
4892                 return btrfs_ioctl_ino_lookup(file, argp);
4893         case BTRFS_IOC_INO_PATHS:
4894                 return btrfs_ioctl_ino_to_path(root, argp);
4895         case BTRFS_IOC_LOGICAL_INO:
4896                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4897         case BTRFS_IOC_LOGICAL_INO_V2:
4898                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4899         case BTRFS_IOC_SPACE_INFO:
4900                 return btrfs_ioctl_space_info(fs_info, argp);
4901         case BTRFS_IOC_SYNC: {
4902                 int ret;
4903
4904                 ret = btrfs_start_delalloc_roots(fs_info, U64_MAX);
4905                 if (ret)
4906                         return ret;
4907                 ret = btrfs_sync_fs(inode->i_sb, 1);
4908                 /*
4909                  * The transaction thread may want to do more work,
4910                  * namely it pokes the cleaner kthread that will start
4911                  * processing uncleaned subvols.
4912                  */
4913                 wake_up_process(fs_info->transaction_kthread);
4914                 return ret;
4915         }
4916         case BTRFS_IOC_START_SYNC:
4917                 return btrfs_ioctl_start_sync(root, argp);
4918         case BTRFS_IOC_WAIT_SYNC:
4919                 return btrfs_ioctl_wait_sync(fs_info, argp);
4920         case BTRFS_IOC_SCRUB:
4921                 return btrfs_ioctl_scrub(file, argp);
4922         case BTRFS_IOC_SCRUB_CANCEL:
4923                 return btrfs_ioctl_scrub_cancel(fs_info);
4924         case BTRFS_IOC_SCRUB_PROGRESS:
4925                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4926         case BTRFS_IOC_BALANCE_V2:
4927                 return btrfs_ioctl_balance(file, argp);
4928         case BTRFS_IOC_BALANCE_CTL:
4929                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4930         case BTRFS_IOC_BALANCE_PROGRESS:
4931                 return btrfs_ioctl_balance_progress(fs_info, argp);
4932         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4933                 return btrfs_ioctl_set_received_subvol(file, argp);
4934 #ifdef CONFIG_64BIT
4935         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4936                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4937 #endif
4938         case BTRFS_IOC_SEND:
4939                 return _btrfs_ioctl_send(file, argp, false);
4940 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4941         case BTRFS_IOC_SEND_32:
4942                 return _btrfs_ioctl_send(file, argp, true);
4943 #endif
4944         case BTRFS_IOC_GET_DEV_STATS:
4945                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4946         case BTRFS_IOC_QUOTA_CTL:
4947                 return btrfs_ioctl_quota_ctl(file, argp);
4948         case BTRFS_IOC_QGROUP_ASSIGN:
4949                 return btrfs_ioctl_qgroup_assign(file, argp);
4950         case BTRFS_IOC_QGROUP_CREATE:
4951                 return btrfs_ioctl_qgroup_create(file, argp);
4952         case BTRFS_IOC_QGROUP_LIMIT:
4953                 return btrfs_ioctl_qgroup_limit(file, argp);
4954         case BTRFS_IOC_QUOTA_RESCAN:
4955                 return btrfs_ioctl_quota_rescan(file, argp);
4956         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4957                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4958         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4959                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4960         case BTRFS_IOC_DEV_REPLACE:
4961                 return btrfs_ioctl_dev_replace(fs_info, argp);
4962         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4963                 return btrfs_ioctl_get_supported_features(argp);
4964         case BTRFS_IOC_GET_FEATURES:
4965                 return btrfs_ioctl_get_features(fs_info, argp);
4966         case BTRFS_IOC_SET_FEATURES:
4967                 return btrfs_ioctl_set_features(file, argp);
4968         case FS_IOC_FSGETXATTR:
4969                 return btrfs_ioctl_fsgetxattr(file, argp);
4970         case FS_IOC_FSSETXATTR:
4971                 return btrfs_ioctl_fssetxattr(file, argp);
4972         case BTRFS_IOC_GET_SUBVOL_INFO:
4973                 return btrfs_ioctl_get_subvol_info(file, argp);
4974         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4975                 return btrfs_ioctl_get_subvol_rootref(file, argp);
4976         case BTRFS_IOC_INO_LOOKUP_USER:
4977                 return btrfs_ioctl_ino_lookup_user(file, argp);
4978         }
4979
4980         return -ENOTTY;
4981 }
4982
4983 #ifdef CONFIG_COMPAT
4984 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4985 {
4986         /*
4987          * These all access 32-bit values anyway so no further
4988          * handling is necessary.
4989          */
4990         switch (cmd) {
4991         case FS_IOC32_GETFLAGS:
4992                 cmd = FS_IOC_GETFLAGS;
4993                 break;
4994         case FS_IOC32_SETFLAGS:
4995                 cmd = FS_IOC_SETFLAGS;
4996                 break;
4997         case FS_IOC32_GETVERSION:
4998                 cmd = FS_IOC_GETVERSION;
4999                 break;
5000         }
5001
5002         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5003 }
5004 #endif