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