fs/btrfs/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/file.h>
   9 #include <linux/fs.h>
  10 #include <linux/fsnotify.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/writeback.h>
  19 #include <linux/compat.h>
  20 #include <linux/security.h>
  21 #include <linux/xattr.h>
  22 #include <linux/mm.h>
  23 #include <linux/slab.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/uuid.h>
  26 #include <linux/btrfs.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/iversion.h>
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "export.h"
  32 #include "transaction.h"
  33 #include "btrfs_inode.h"
  34 #include "print-tree.h"
  35 #include "volumes.h"
  36 #include "locking.h"
  37 #include "backref.h"
  38 #include "rcu-string.h"
  39 #include "send.h"
  40 #include "dev-replace.h"
  41 #include "props.h"
  42 #include "sysfs.h"
  43 #include "qgroup.h"
  44 #include "tree-log.h"
  45 #include "compression.h"
  46 #include "space-info.h"
  47 #include "delalloc-space.h"
  48 #include "block-group.h"
  49
  50 #ifdef CONFIG_64BIT
  51 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  52  * structures are incorrect, as the timespec structure from userspace
  53  * is 4 bytes too small. We define these alternatives here to teach
  54  * the kernel about the 32-bit struct packing.
  55  */
  56 struct btrfs_ioctl_timespec_32 {
  57         __u64 sec;
  58         __u32 nsec;
  59 } __attribute__ ((__packed__));
  60
  61 struct btrfs_ioctl_received_subvol_args_32 {
  62         char    uuid[BTRFS_UUID_SIZE];  /* in */
  63         __u64   stransid;               /* in */
  64         __u64   rtransid;               /* out */
  65         struct btrfs_ioctl_timespec_32 stime; /* in */
  66         struct btrfs_ioctl_timespec_32 rtime; /* out */
  67         __u64   flags;                  /* in */
  68         __u64   reserved[16];           /* in */
  69 } __attribute__ ((__packed__));
  70
  71 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  72                                 struct btrfs_ioctl_received_subvol_args_32)
  73 #endif
  74
  75 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  76 struct btrfs_ioctl_send_args_32 {
  77         __s64 send_fd;                  /* in */
  78         __u64 clone_sources_count;      /* in */
  79         compat_uptr_t clone_sources;    /* in */
  80         __u64 parent_root;              /* in */
  81         __u64 flags;                    /* in */
  82         __u64 reserved[4];              /* in */
  83 } __attribute__ ((__packed__));
  84
  85 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  86                                struct btrfs_ioctl_send_args_32)
  87 #endif
  88
  89 /* Mask out flags that are inappropriate for the given type of inode. */
  90 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  91                 unsigned int flags)
  92 {
  93         if (S_ISDIR(inode->i_mode))
  94                 return flags;
  95         else if (S_ISREG(inode->i_mode))
  96                 return flags & ~FS_DIRSYNC_FL;
  97         else
  98                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  99 }
 100
 101 /*
 102  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 103  * ioctl.
 104  */
 105 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 106 {
 107         unsigned int iflags = 0;
 108
 109         if (flags & BTRFS_INODE_SYNC)
 110                 iflags |= FS_SYNC_FL;
 111         if (flags & BTRFS_INODE_IMMUTABLE)
 112                 iflags |= FS_IMMUTABLE_FL;
 113         if (flags & BTRFS_INODE_APPEND)
 114                 iflags |= FS_APPEND_FL;
 115         if (flags & BTRFS_INODE_NODUMP)
 116                 iflags |= FS_NODUMP_FL;
 117         if (flags & BTRFS_INODE_NOATIME)
 118                 iflags |= FS_NOATIME_FL;
 119         if (flags & BTRFS_INODE_DIRSYNC)
 120                 iflags |= FS_DIRSYNC_FL;
 121         if (flags & BTRFS_INODE_NODATACOW)
 122                 iflags |= FS_NOCOW_FL;
 123
 124         if (flags & BTRFS_INODE_NOCOMPRESS)
 125                 iflags |= FS_NOCOMP_FL;
 126         else if (flags & BTRFS_INODE_COMPRESS)
 127                 iflags |= FS_COMPR_FL;
 128
 129         return iflags;
 130 }
 131
 132 /*
 133  * Update inode->i_flags based on the btrfs internal flags.
 134  */
 135 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 136 {
 137         struct btrfs_inode *binode = BTRFS_I(inode);
 138         unsigned int new_fl = 0;
 139
 140         if (binode->flags & BTRFS_INODE_SYNC)
 141                 new_fl |= S_SYNC;
 142         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 143                 new_fl |= S_IMMUTABLE;
 144         if (binode->flags & BTRFS_INODE_APPEND)
 145                 new_fl |= S_APPEND;
 146         if (binode->flags & BTRFS_INODE_NOATIME)
 147                 new_fl |= S_NOATIME;
 148         if (binode->flags & BTRFS_INODE_DIRSYNC)
 149                 new_fl |= S_DIRSYNC;
 150
 151         set_mask_bits(&inode->i_flags,
 152                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 153                       new_fl);
 154 }
 155
 156 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 157 {
 158         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 159         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 160
 161         if (copy_to_user(arg, &flags, sizeof(flags)))
 162                 return -EFAULT;
 163         return 0;
 164 }
 165
 166 /*
 167  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 168  * the old and new flags are not conflicting
 169  */
 170 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 171 {
 172         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 173                       FS_NOATIME_FL | FS_NODUMP_FL | \
 174                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 175                       FS_NOCOMP_FL | FS_COMPR_FL |
 176                       FS_NOCOW_FL))
 177                 return -EOPNOTSUPP;
 178
 179         /* COMPR and NOCOMP on new/old are valid */
 180         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 181                 return -EINVAL;
 182
 183         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 184                 return -EINVAL;
 185
 186         /* NOCOW and compression options are mutually exclusive */
 187         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 188                 return -EINVAL;
 189         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 190                 return -EINVAL;
 191
 192         return 0;
 193 }
 194
 195 static int check_fsflags_compatible(struct btrfs_fs_info *fs_info,
 196                                     unsigned int flags)
 197 {
 198         if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL))
 199                 return -EPERM;
 200
 201         return 0;
 202 }
 203
 204 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 205 {
 206         struct inode *inode = file_inode(file);
 207         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 208         struct btrfs_inode *binode = BTRFS_I(inode);
 209         struct btrfs_root *root = binode->root;
 210         struct btrfs_trans_handle *trans;
 211         unsigned int fsflags, old_fsflags;
 212         int ret;
 213         const char *comp = NULL;
 214         u32 binode_flags;
 215
 216         if (!inode_owner_or_capable(&init_user_ns, inode))
 217                 return -EPERM;
 218
 219         if (btrfs_root_readonly(root))
 220                 return -EROFS;
 221
 222         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 223                 return -EFAULT;
 224
 225         ret = mnt_want_write_file(file);
 226         if (ret)
 227                 return ret;
 228
 229         inode_lock(inode);
 230         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 231         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 232
 233         ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 234         if (ret)
 235                 goto out_unlock;
 236
 237         ret = check_fsflags(old_fsflags, fsflags);
 238         if (ret)
 239                 goto out_unlock;
 240
 241         ret = check_fsflags_compatible(fs_info, fsflags);
 242         if (ret)
 243                 goto out_unlock;
 244
 245         binode_flags = binode->flags;
 246         if (fsflags & FS_SYNC_FL)
 247                 binode_flags |= BTRFS_INODE_SYNC;
 248         else
 249                 binode_flags &= ~BTRFS_INODE_SYNC;
 250         if (fsflags & FS_IMMUTABLE_FL)
 251                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 252         else
 253                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 254         if (fsflags & FS_APPEND_FL)
 255                 binode_flags |= BTRFS_INODE_APPEND;
 256         else
 257                 binode_flags &= ~BTRFS_INODE_APPEND;
 258         if (fsflags & FS_NODUMP_FL)
 259                 binode_flags |= BTRFS_INODE_NODUMP;
 260         else
 261                 binode_flags &= ~BTRFS_INODE_NODUMP;
 262         if (fsflags & FS_NOATIME_FL)
 263                 binode_flags |= BTRFS_INODE_NOATIME;
 264         else
 265                 binode_flags &= ~BTRFS_INODE_NOATIME;
 266         if (fsflags & FS_DIRSYNC_FL)
 267                 binode_flags |= BTRFS_INODE_DIRSYNC;
 268         else
 269                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 270         if (fsflags & FS_NOCOW_FL) {
 271                 if (S_ISREG(inode->i_mode)) {
 272                         /*
 273                          * It's safe to turn csums off here, no extents exist.
 274                          * Otherwise we want the flag to reflect the real COW
 275                          * status of the file and will not set it.
 276                          */
 277                         if (inode->i_size == 0)
 278                                 binode_flags |= BTRFS_INODE_NODATACOW |
 279                                                 BTRFS_INODE_NODATASUM;
 280                 } else {
 281                         binode_flags |= BTRFS_INODE_NODATACOW;
 282                 }
 283         } else {
 284                 /*
 285                  * Revert back under same assumptions as above
 286                  */
 287                 if (S_ISREG(inode->i_mode)) {
 288                         if (inode->i_size == 0)
 289                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 290                                                   BTRFS_INODE_NODATASUM);
 291                 } else {
 292                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 293                 }
 294         }
 295
 296         /*
 297          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 298          * flag may be changed automatically if compression code won't make
 299          * things smaller.
 300          */
 301         if (fsflags & FS_NOCOMP_FL) {
 302                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 303                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 304         } else if (fsflags & FS_COMPR_FL) {
 305
 306                 if (IS_SWAPFILE(inode)) {
 307                         ret = -ETXTBSY;
 308                         goto out_unlock;
 309                 }
 310
 311                 binode_flags |= BTRFS_INODE_COMPRESS;
 312                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 313
 314                 comp = btrfs_compress_type2str(fs_info->compress_type);
 315                 if (!comp || comp[0] == 0)
 316                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 317         } else {
 318                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 319         }
 320
 321         /*
 322          * 1 for inode item
 323          * 2 for properties
 324          */
 325         trans = btrfs_start_transaction(root, 3);
 326         if (IS_ERR(trans)) {
 327                 ret = PTR_ERR(trans);
 328                 goto out_unlock;
 329         }
 330
 331         if (comp) {
 332                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 333                                      strlen(comp), 0);
 334                 if (ret) {
 335                         btrfs_abort_transaction(trans, ret);
 336                         goto out_end_trans;
 337                 }
 338         } else {
 339                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 340                                      0, 0);
 341                 if (ret && ret != -ENODATA) {
 342                         btrfs_abort_transaction(trans, ret);
 343                         goto out_end_trans;
 344                 }
 345         }
 346
 347         binode->flags = binode_flags;
 348         btrfs_sync_inode_flags_to_i_flags(inode);
 349         inode_inc_iversion(inode);
 350         inode->i_ctime = current_time(inode);
 351         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 352
 353  out_end_trans:
 354         btrfs_end_transaction(trans);
 355  out_unlock:
 356         inode_unlock(inode);
 357         mnt_drop_write_file(file);
 358         return ret;
 359 }
 360
 361 /*
 362  * Translate btrfs internal inode flags to xflags as expected by the
 363  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 364  * silently dropped.
 365  */
 366 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 367 {
 368         unsigned int xflags = 0;
 369
 370         if (flags & BTRFS_INODE_APPEND)
 371                 xflags |= FS_XFLAG_APPEND;
 372         if (flags & BTRFS_INODE_IMMUTABLE)
 373                 xflags |= FS_XFLAG_IMMUTABLE;
 374         if (flags & BTRFS_INODE_NOATIME)
 375                 xflags |= FS_XFLAG_NOATIME;
 376         if (flags & BTRFS_INODE_NODUMP)
 377                 xflags |= FS_XFLAG_NODUMP;
 378         if (flags & BTRFS_INODE_SYNC)
 379                 xflags |= FS_XFLAG_SYNC;
 380
 381         return xflags;
 382 }
 383
 384 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 385 static int check_xflags(unsigned int flags)
 386 {
 387         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 388                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 389                 return -EOPNOTSUPP;
 390         return 0;
 391 }
 392
 393 bool btrfs_exclop_start(struct btrfs_fs_info *fs_info,
 394                         enum btrfs_exclusive_operation type)
 395 {
 396         return !cmpxchg(&fs_info->exclusive_operation, BTRFS_EXCLOP_NONE, type);
 397 }
 398
 399 void btrfs_exclop_finish(struct btrfs_fs_info *fs_info)
 400 {
 401         WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE);
 402         sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation");
 403 }
 404
 405 /*
 406  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 407  * are zeroed.
 408  */
 409 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 410 {
 411         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 412         struct fsxattr fa;
 413
 414         simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 415         if (copy_to_user(arg, &fa, sizeof(fa)))
 416                 return -EFAULT;
 417
 418         return 0;
 419 }
 420
 421 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 422 {
 423         struct inode *inode = file_inode(file);
 424         struct btrfs_inode *binode = BTRFS_I(inode);
 425         struct btrfs_root *root = binode->root;
 426         struct btrfs_trans_handle *trans;
 427         struct fsxattr fa, old_fa;
 428         unsigned old_flags;
 429         unsigned old_i_flags;
 430         int ret = 0;
 431
 432         if (!inode_owner_or_capable(&init_user_ns, inode))
 433                 return -EPERM;
 434
 435         if (btrfs_root_readonly(root))
 436                 return -EROFS;
 437
 438         if (copy_from_user(&fa, arg, sizeof(fa)))
 439                 return -EFAULT;
 440
 441         ret = check_xflags(fa.fsx_xflags);
 442         if (ret)
 443                 return ret;
 444
 445         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 446                 return -EOPNOTSUPP;
 447
 448         ret = mnt_want_write_file(file);
 449         if (ret)
 450                 return ret;
 451
 452         inode_lock(inode);
 453
 454         old_flags = binode->flags;
 455         old_i_flags = inode->i_flags;
 456
 457         simple_fill_fsxattr(&old_fa,
 458                             btrfs_inode_flags_to_xflags(binode->flags));
 459         ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 460         if (ret)
 461                 goto out_unlock;
 462
 463         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 464                 binode->flags |= BTRFS_INODE_SYNC;
 465         else
 466                 binode->flags &= ~BTRFS_INODE_SYNC;
 467         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 468                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 469         else
 470                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 471         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 472                 binode->flags |= BTRFS_INODE_APPEND;
 473         else
 474                 binode->flags &= ~BTRFS_INODE_APPEND;
 475         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 476                 binode->flags |= BTRFS_INODE_NODUMP;
 477         else
 478                 binode->flags &= ~BTRFS_INODE_NODUMP;
 479         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 480                 binode->flags |= BTRFS_INODE_NOATIME;
 481         else
 482                 binode->flags &= ~BTRFS_INODE_NOATIME;
 483
 484         /* 1 item for the inode */
 485         trans = btrfs_start_transaction(root, 1);
 486         if (IS_ERR(trans)) {
 487                 ret = PTR_ERR(trans);
 488                 goto out_unlock;
 489         }
 490
 491         btrfs_sync_inode_flags_to_i_flags(inode);
 492         inode_inc_iversion(inode);
 493         inode->i_ctime = current_time(inode);
 494         ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
 495
 496         btrfs_end_transaction(trans);
 497
 498 out_unlock:
 499         if (ret) {
 500                 binode->flags = old_flags;
 501                 inode->i_flags = old_i_flags;
 502         }
 503
 504         inode_unlock(inode);
 505         mnt_drop_write_file(file);
 506
 507         return ret;
 508 }
 509
 510 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 511 {
 512         struct inode *inode = file_inode(file);
 513
 514         return put_user(inode->i_generation, arg);
 515 }
 516
 517 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 518                                         void __user *arg)
 519 {
 520         struct btrfs_device *device;
 521         struct request_queue *q;
 522         struct fstrim_range range;
 523         u64 minlen = ULLONG_MAX;
 524         u64 num_devices = 0;
 525         int ret;
 526
 527         if (!capable(CAP_SYS_ADMIN))
 528                 return -EPERM;
 529
 530         /*
 531          * btrfs_trim_block_group() depends on space cache, which is not
 532          * available in zoned filesystem. So, disallow fitrim on a zoned
 533          * filesystem for now.
 534          */
 535         if (btrfs_is_zoned(fs_info))
 536                 return -EOPNOTSUPP;
 537
 538         /*
 539          * If the fs is mounted with nologreplay, which requires it to be
 540          * mounted in RO mode as well, we can not allow discard on free space
 541          * inside block groups, because log trees refer to extents that are not
 542          * pinned in a block group's free space cache (pinning the extents is
 543          * precisely the first phase of replaying a log tree).
 544          */
 545         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 546                 return -EROFS;
 547
 548         rcu_read_lock();
 549         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 550                                 dev_list) {
 551                 if (!device->bdev)
 552                         continue;
 553                 q = bdev_get_queue(device->bdev);
 554                 if (blk_queue_discard(q)) {
 555                         num_devices++;
 556                         minlen = min_t(u64, q->limits.discard_granularity,
 557                                      minlen);
 558                 }
 559         }
 560         rcu_read_unlock();
 561
 562         if (!num_devices)
 563                 return -EOPNOTSUPP;
 564         if (copy_from_user(&range, arg, sizeof(range)))
 565                 return -EFAULT;
 566
 567         /*
 568          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 569          * block group is in the logical address space, which can be any
 570          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 571          */
 572         if (range.len < fs_info->sb->s_blocksize)
 573                 return -EINVAL;
 574
 575         range.minlen = max(range.minlen, minlen);
 576         ret = btrfs_trim_fs(fs_info, &range);
 577         if (ret < 0)
 578                 return ret;
 579
 580         if (copy_to_user(arg, &range, sizeof(range)))
 581                 return -EFAULT;
 582
 583         return 0;
 584 }
 585
 586 int __pure btrfs_is_empty_uuid(u8 *uuid)
 587 {
 588         int i;
 589
 590         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 591                 if (uuid[i])
 592                         return 0;
 593         }
 594         return 1;
 595 }
 596
 597 static noinline int create_subvol(struct inode *dir,
 598                                   struct dentry *dentry,
 599                                   const char *name, int namelen,
 600                                   struct btrfs_qgroup_inherit *inherit)
 601 {
 602         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 603         struct btrfs_trans_handle *trans;
 604         struct btrfs_key key;
 605         struct btrfs_root_item *root_item;
 606         struct btrfs_inode_item *inode_item;
 607         struct extent_buffer *leaf;
 608         struct btrfs_root *root = BTRFS_I(dir)->root;
 609         struct btrfs_root *new_root;
 610         struct btrfs_block_rsv block_rsv;
 611         struct timespec64 cur_time = current_time(dir);
 612         struct inode *inode;
 613         int ret;
 614         int err;
 615         dev_t anon_dev = 0;
 616         u64 objectid;
 617         u64 index = 0;
 618
 619         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 620         if (!root_item)
 621                 return -ENOMEM;
 622
 623         ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid);
 624         if (ret)
 625                 goto fail_free;
 626
 627         ret = get_anon_bdev(&anon_dev);
 628         if (ret < 0)
 629                 goto fail_free;
 630
 631         /*
 632          * Don't create subvolume whose level is not zero. Or qgroup will be
 633          * screwed up since it assumes subvolume qgroup's level to be 0.
 634          */
 635         if (btrfs_qgroup_level(objectid)) {
 636                 ret = -ENOSPC;
 637                 goto fail_free;
 638         }
 639
 640         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 641         /*
 642          * The same as the snapshot creation, please see the comment
 643          * of create_snapshot().
 644          */
 645         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 646         if (ret)
 647                 goto fail_free;
 648
 649         trans = btrfs_start_transaction(root, 0);
 650         if (IS_ERR(trans)) {
 651                 ret = PTR_ERR(trans);
 652                 btrfs_subvolume_release_metadata(root, &block_rsv);
 653                 goto fail_free;
 654         }
 655         trans->block_rsv = &block_rsv;
 656         trans->bytes_reserved = block_rsv.size;
 657
 658         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 659         if (ret)
 660                 goto fail;
 661
 662         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 663                                       BTRFS_NESTING_NORMAL);
 664         if (IS_ERR(leaf)) {
 665                 ret = PTR_ERR(leaf);
 666                 goto fail;
 667         }
 668
 669         btrfs_mark_buffer_dirty(leaf);
 670
 671         inode_item = &root_item->inode;
 672         btrfs_set_stack_inode_generation(inode_item, 1);
 673         btrfs_set_stack_inode_size(inode_item, 3);
 674         btrfs_set_stack_inode_nlink(inode_item, 1);
 675         btrfs_set_stack_inode_nbytes(inode_item,
 676                                      fs_info->nodesize);
 677         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 678
 679         btrfs_set_root_flags(root_item, 0);
 680         btrfs_set_root_limit(root_item, 0);
 681         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 682
 683         btrfs_set_root_bytenr(root_item, leaf->start);
 684         btrfs_set_root_generation(root_item, trans->transid);
 685         btrfs_set_root_level(root_item, 0);
 686         btrfs_set_root_refs(root_item, 1);
 687         btrfs_set_root_used(root_item, leaf->len);
 688         btrfs_set_root_last_snapshot(root_item, 0);
 689
 690         btrfs_set_root_generation_v2(root_item,
 691                         btrfs_root_generation(root_item));
 692         generate_random_guid(root_item->uuid);
 693         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 694         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 695         root_item->ctime = root_item->otime;
 696         btrfs_set_root_ctransid(root_item, trans->transid);
 697         btrfs_set_root_otransid(root_item, trans->transid);
 698
 699         btrfs_tree_unlock(leaf);
 700         free_extent_buffer(leaf);
 701         leaf = NULL;
 702
 703         btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID);
 704
 705         key.objectid = objectid;
 706         key.offset = 0;
 707         key.type = BTRFS_ROOT_ITEM_KEY;
 708         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 709                                 root_item);
 710         if (ret)
 711                 goto fail;
 712
 713         key.offset = (u64)-1;
 714         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 715         if (IS_ERR(new_root)) {
 716                 free_anon_bdev(anon_dev);
 717                 ret = PTR_ERR(new_root);
 718                 btrfs_abort_transaction(trans, ret);
 719                 goto fail;
 720         }
 721         /* Freeing will be done in btrfs_put_root() of new_root */
 722         anon_dev = 0;
 723
 724         btrfs_record_root_in_trans(trans, new_root);
 725
 726         ret = btrfs_create_subvol_root(trans, new_root, root);
 727         btrfs_put_root(new_root);
 728         if (ret) {
 729                 /* We potentially lose an unused inode item here */
 730                 btrfs_abort_transaction(trans, ret);
 731                 goto fail;
 732         }
 733
 734         /*
 735          * insert the directory item
 736          */
 737         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 738         if (ret) {
 739                 btrfs_abort_transaction(trans, ret);
 740                 goto fail;
 741         }
 742
 743         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 744                                     BTRFS_FT_DIR, index);
 745         if (ret) {
 746                 btrfs_abort_transaction(trans, ret);
 747                 goto fail;
 748         }
 749
 750         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 751         ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
 752         if (ret) {
 753                 btrfs_abort_transaction(trans, ret);
 754                 goto fail;
 755         }
 756
 757         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 758                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 759         if (ret) {
 760                 btrfs_abort_transaction(trans, ret);
 761                 goto fail;
 762         }
 763
 764         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 765                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 766         if (ret)
 767                 btrfs_abort_transaction(trans, ret);
 768
 769 fail:
 770         kfree(root_item);
 771         trans->block_rsv = NULL;
 772         trans->bytes_reserved = 0;
 773         btrfs_subvolume_release_metadata(root, &block_rsv);
 774
 775         err = btrfs_commit_transaction(trans);
 776         if (err && !ret)
 777                 ret = err;
 778
 779         if (!ret) {
 780                 inode = btrfs_lookup_dentry(dir, dentry);
 781                 if (IS_ERR(inode))
 782                         return PTR_ERR(inode);
 783                 d_instantiate(dentry, inode);
 784         }
 785         return ret;
 786
 787 fail_free:
 788         if (anon_dev)
 789                 free_anon_bdev(anon_dev);
 790         kfree(root_item);
 791         return ret;
 792 }
 793
 794 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 795                            struct dentry *dentry, bool readonly,
 796                            struct btrfs_qgroup_inherit *inherit)
 797 {
 798         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 799         struct inode *inode;
 800         struct btrfs_pending_snapshot *pending_snapshot;
 801         struct btrfs_trans_handle *trans;
 802         int ret;
 803
 804         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 805                 return -EINVAL;
 806
 807         if (atomic_read(&root->nr_swapfiles)) {
 808                 btrfs_warn(fs_info,
 809                            "cannot snapshot subvolume with active swapfile");
 810                 return -ETXTBSY;
 811         }
 812
 813         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 814         if (!pending_snapshot)
 815                 return -ENOMEM;
 816
 817         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 818         if (ret < 0)
 819                 goto free_pending;
 820         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 821                         GFP_KERNEL);
 822         pending_snapshot->path = btrfs_alloc_path();
 823         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 824                 ret = -ENOMEM;
 825                 goto free_pending;
 826         }
 827
 828         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 829                              BTRFS_BLOCK_RSV_TEMP);
 830         /*
 831          * 1 - parent dir inode
 832          * 2 - dir entries
 833          * 1 - root item
 834          * 2 - root ref/backref
 835          * 1 - root of snapshot
 836          * 1 - UUID item
 837          */
 838         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 839                                         &pending_snapshot->block_rsv, 8,
 840                                         false);
 841         if (ret)
 842                 goto free_pending;
 843
 844         pending_snapshot->dentry = dentry;
 845         pending_snapshot->root = root;
 846         pending_snapshot->readonly = readonly;
 847         pending_snapshot->dir = dir;
 848         pending_snapshot->inherit = inherit;
 849
 850         trans = btrfs_start_transaction(root, 0);
 851         if (IS_ERR(trans)) {
 852                 ret = PTR_ERR(trans);
 853                 goto fail;
 854         }
 855
 856         spin_lock(&fs_info->trans_lock);
 857         list_add(&pending_snapshot->list,
 858                  &trans->transaction->pending_snapshots);
 859         spin_unlock(&fs_info->trans_lock);
 860
 861         ret = btrfs_commit_transaction(trans);
 862         if (ret)
 863                 goto fail;
 864
 865         ret = pending_snapshot->error;
 866         if (ret)
 867                 goto fail;
 868
 869         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 870         if (ret)
 871                 goto fail;
 872
 873         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 874         if (IS_ERR(inode)) {
 875                 ret = PTR_ERR(inode);
 876                 goto fail;
 877         }
 878
 879         d_instantiate(dentry, inode);
 880         ret = 0;
 881         pending_snapshot->anon_dev = 0;
 882 fail:
 883         /* Prevent double freeing of anon_dev */
 884         if (ret && pending_snapshot->snap)
 885                 pending_snapshot->snap->anon_dev = 0;
 886         btrfs_put_root(pending_snapshot->snap);
 887         btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv);
 888 free_pending:
 889         if (pending_snapshot->anon_dev)
 890                 free_anon_bdev(pending_snapshot->anon_dev);
 891         kfree(pending_snapshot->root_item);
 892         btrfs_free_path(pending_snapshot->path);
 893         kfree(pending_snapshot);
 894
 895         return ret;
 896 }
 897
 898 /*  copy of may_delete in fs/namei.c()
 899  *      Check whether we can remove a link victim from directory dir, check
 900  *  whether the type of victim is right.
 901  *  1. We can't do it if dir is read-only (done in permission())
 902  *  2. We should have write and exec permissions on dir
 903  *  3. We can't remove anything from append-only dir
 904  *  4. We can't do anything with immutable dir (done in permission())
 905  *  5. If the sticky bit on dir is set we should either
 906  *      a. be owner of dir, or
 907  *      b. be owner of victim, or
 908  *      c. have CAP_FOWNER capability
 909  *  6. If the victim is append-only or immutable we can't do anything with
 910  *     links pointing to it.
 911  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 912  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 913  *  9. We can't remove a root or mountpoint.
 914  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 915  *     nfs_async_unlink().
 916  */
 917
 918 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 919 {
 920         int error;
 921
 922         if (d_really_is_negative(victim))
 923                 return -ENOENT;
 924
 925         BUG_ON(d_inode(victim->d_parent) != dir);
 926         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 927
 928         error = inode_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
 929         if (error)
 930                 return error;
 931         if (IS_APPEND(dir))
 932                 return -EPERM;
 933         if (check_sticky(&init_user_ns, dir, d_inode(victim)) ||
 934             IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) ||
 935             IS_SWAPFILE(d_inode(victim)))
 936                 return -EPERM;
 937         if (isdir) {
 938                 if (!d_is_dir(victim))
 939                         return -ENOTDIR;
 940                 if (IS_ROOT(victim))
 941                         return -EBUSY;
 942         } else if (d_is_dir(victim))
 943                 return -EISDIR;
 944         if (IS_DEADDIR(dir))
 945                 return -ENOENT;
 946         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 947                 return -EBUSY;
 948         return 0;
 949 }
 950
 951 /* copy of may_create in fs/namei.c() */
 952 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 953 {
 954         if (d_really_is_positive(child))
 955                 return -EEXIST;
 956         if (IS_DEADDIR(dir))
 957                 return -ENOENT;
 958         return inode_permission(&init_user_ns, dir, MAY_WRITE | MAY_EXEC);
 959 }
 960
 961 /*
 962  * Create a new subvolume below @parent.  This is largely modeled after
 963  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 964  * inside this filesystem so it's quite a bit simpler.
 965  */
 966 static noinline int btrfs_mksubvol(const struct path *parent,
 967                                    const char *name, int namelen,
 968                                    struct btrfs_root *snap_src,
 969                                    bool readonly,
 970                                    struct btrfs_qgroup_inherit *inherit)
 971 {
 972         struct inode *dir = d_inode(parent->dentry);
 973         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 974         struct dentry *dentry;
 975         int error;
 976
 977         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 978         if (error == -EINTR)
 979                 return error;
 980
 981         dentry = lookup_one_len(name, parent->dentry, namelen);
 982         error = PTR_ERR(dentry);
 983         if (IS_ERR(dentry))
 984                 goto out_unlock;
 985
 986         error = btrfs_may_create(dir, dentry);
 987         if (error)
 988                 goto out_dput;
 989
 990         /*
 991          * even if this name doesn't exist, we may get hash collisions.
 992          * check for them now when we can safely fail
 993          */
 994         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 995                                                dir->i_ino, name,
 996                                                namelen);
 997         if (error)
 998                 goto out_dput;
 999
1000         down_read(&fs_info->subvol_sem);
1001
1002         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
1003                 goto out_up_read;
1004
1005         if (snap_src)
1006                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
1007         else
1008                 error = create_subvol(dir, dentry, name, namelen, inherit);
1009
1010         if (!error)
1011                 fsnotify_mkdir(dir, dentry);
1012 out_up_read:
1013         up_read(&fs_info->subvol_sem);
1014 out_dput:
1015         dput(dentry);
1016 out_unlock:
1017         inode_unlock(dir);
1018         return error;
1019 }
1020
1021 static noinline int btrfs_mksnapshot(const struct path *parent,
1022                                    const char *name, int namelen,
1023                                    struct btrfs_root *root,
1024                                    bool readonly,
1025                                    struct btrfs_qgroup_inherit *inherit)
1026 {
1027         int ret;
1028         bool snapshot_force_cow = false;
1029
1030         /*
1031          * Force new buffered writes to reserve space even when NOCOW is
1032          * possible. This is to avoid later writeback (running dealloc) to
1033          * fallback to COW mode and unexpectedly fail with ENOSPC.
1034          */
1035         btrfs_drew_read_lock(&root->snapshot_lock);
1036
1037         ret = btrfs_start_delalloc_snapshot(root);
1038         if (ret)
1039                 goto out;
1040
1041         /*
1042          * All previous writes have started writeback in NOCOW mode, so now
1043          * we force future writes to fallback to COW mode during snapshot
1044          * creation.
1045          */
1046         atomic_inc(&root->snapshot_force_cow);
1047         snapshot_force_cow = true;
1048
1049         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1050
1051         ret = btrfs_mksubvol(parent, name, namelen,
1052                              root, readonly, inherit);
1053 out:
1054         if (snapshot_force_cow)
1055                 atomic_dec(&root->snapshot_force_cow);
1056         btrfs_drew_read_unlock(&root->snapshot_lock);
1057         return ret;
1058 }
1059
1060 /*
1061  * When we're defragging a range, we don't want to kick it off again
1062  * if it is really just waiting for delalloc to send it down.
1063  * If we find a nice big extent or delalloc range for the bytes in the
1064  * file you want to defrag, we return 0 to let you know to skip this
1065  * part of the file
1066  */
1067 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1068 {
1069         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1070         struct extent_map *em = NULL;
1071         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1072         u64 end;
1073
1074         read_lock(&em_tree->lock);
1075         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1076         read_unlock(&em_tree->lock);
1077
1078         if (em) {
1079                 end = extent_map_end(em);
1080                 free_extent_map(em);
1081                 if (end - offset > thresh)
1082                         return 0;
1083         }
1084         /* if we already have a nice delalloc here, just stop */
1085         thresh /= 2;
1086         end = count_range_bits(io_tree, &offset, offset + thresh,
1087                                thresh, EXTENT_DELALLOC, 1);
1088         if (end >= thresh)
1089                 return 0;
1090         return 1;
1091 }
1092
1093 /*
1094  * helper function to walk through a file and find extents
1095  * newer than a specific transid, and smaller than thresh.
1096  *
1097  * This is used by the defragging code to find new and small
1098  * extents
1099  */
1100 static int find_new_extents(struct btrfs_root *root,
1101                             struct inode *inode, u64 newer_than,
1102                             u64 *off, u32 thresh)
1103 {
1104         struct btrfs_path *path;
1105         struct btrfs_key min_key;
1106         struct extent_buffer *leaf;
1107         struct btrfs_file_extent_item *extent;
1108         int type;
1109         int ret;
1110         u64 ino = btrfs_ino(BTRFS_I(inode));
1111
1112         path = btrfs_alloc_path();
1113         if (!path)
1114                 return -ENOMEM;
1115
1116         min_key.objectid = ino;
1117         min_key.type = BTRFS_EXTENT_DATA_KEY;
1118         min_key.offset = *off;
1119
1120         while (1) {
1121                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1122                 if (ret != 0)
1123                         goto none;
1124 process_slot:
1125                 if (min_key.objectid != ino)
1126                         goto none;
1127                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1128                         goto none;
1129
1130                 leaf = path->nodes[0];
1131                 extent = btrfs_item_ptr(leaf, path->slots[0],
1132                                         struct btrfs_file_extent_item);
1133
1134                 type = btrfs_file_extent_type(leaf, extent);
1135                 if (type == BTRFS_FILE_EXTENT_REG &&
1136                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1137                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1138                         *off = min_key.offset;
1139                         btrfs_free_path(path);
1140                         return 0;
1141                 }
1142
1143                 path->slots[0]++;
1144                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1145                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1146                         goto process_slot;
1147                 }
1148
1149                 if (min_key.offset == (u64)-1)
1150                         goto none;
1151
1152                 min_key.offset++;
1153                 btrfs_release_path(path);
1154         }
1155 none:
1156         btrfs_free_path(path);
1157         return -ENOENT;
1158 }
1159
1160 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1161 {
1162         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1163         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1164         struct extent_map *em;
1165         u64 len = PAGE_SIZE;
1166
1167         /*
1168          * hopefully we have this extent in the tree already, try without
1169          * the full extent lock
1170          */
1171         read_lock(&em_tree->lock);
1172         em = lookup_extent_mapping(em_tree, start, len);
1173         read_unlock(&em_tree->lock);
1174
1175         if (!em) {
1176                 struct extent_state *cached = NULL;
1177                 u64 end = start + len - 1;
1178
1179                 /* get the big lock and read metadata off disk */
1180                 lock_extent_bits(io_tree, start, end, &cached);
1181                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1182                 unlock_extent_cached(io_tree, start, end, &cached);
1183
1184                 if (IS_ERR(em))
1185                         return NULL;
1186         }
1187
1188         return em;
1189 }
1190
1191 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1192 {
1193         struct extent_map *next;
1194         bool ret = true;
1195
1196         /* this is the last extent */
1197         if (em->start + em->len >= i_size_read(inode))
1198                 return false;
1199
1200         next = defrag_lookup_extent(inode, em->start + em->len);
1201         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1202                 ret = false;
1203         else if ((em->block_start + em->block_len == next->block_start) &&
1204                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1205                 ret = false;
1206
1207         free_extent_map(next);
1208         return ret;
1209 }
1210
1211 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1212                                u64 *last_len, u64 *skip, u64 *defrag_end,
1213                                int compress)
1214 {
1215         struct extent_map *em;
1216         int ret = 1;
1217         bool next_mergeable = true;
1218         bool prev_mergeable = true;
1219
1220         /*
1221          * make sure that once we start defragging an extent, we keep on
1222          * defragging it
1223          */
1224         if (start < *defrag_end)
1225                 return 1;
1226
1227         *skip = 0;
1228
1229         em = defrag_lookup_extent(inode, start);
1230         if (!em)
1231                 return 0;
1232
1233         /* this will cover holes, and inline extents */
1234         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1235                 ret = 0;
1236                 goto out;
1237         }
1238
1239         if (!*defrag_end)
1240                 prev_mergeable = false;
1241
1242         next_mergeable = defrag_check_next_extent(inode, em);
1243         /*
1244          * we hit a real extent, if it is big or the next extent is not a
1245          * real extent, don't bother defragging it
1246          */
1247         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1248             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1249                 ret = 0;
1250 out:
1251         /*
1252          * last_len ends up being a counter of how many bytes we've defragged.
1253          * every time we choose not to defrag an extent, we reset *last_len
1254          * so that the next tiny extent will force a defrag.
1255          *
1256          * The end result of this is that tiny extents before a single big
1257          * extent will force at least part of that big extent to be defragged.
1258          */
1259         if (ret) {
1260                 *defrag_end = extent_map_end(em);
1261         } else {
1262                 *last_len = 0;
1263                 *skip = extent_map_end(em);
1264                 *defrag_end = 0;
1265         }
1266
1267         free_extent_map(em);
1268         return ret;
1269 }
1270
1271 /*
1272  * it doesn't do much good to defrag one or two pages
1273  * at a time.  This pulls in a nice chunk of pages
1274  * to COW and defrag.
1275  *
1276  * It also makes sure the delalloc code has enough
1277  * dirty data to avoid making new small extents as part
1278  * of the defrag
1279  *
1280  * It's a good idea to start RA on this range
1281  * before calling this.
1282  */
1283 static int cluster_pages_for_defrag(struct inode *inode,
1284                                     struct page **pages,
1285                                     unsigned long start_index,
1286                                     unsigned long num_pages)
1287 {
1288         unsigned long file_end;
1289         u64 isize = i_size_read(inode);
1290         u64 page_start;
1291         u64 page_end;
1292         u64 page_cnt;
1293         u64 start = (u64)start_index << PAGE_SHIFT;
1294         u64 search_start;
1295         int ret;
1296         int i;
1297         int i_done;
1298         struct btrfs_ordered_extent *ordered;
1299         struct extent_state *cached_state = NULL;
1300         struct extent_io_tree *tree;
1301         struct extent_changeset *data_reserved = NULL;
1302         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1303
1304         file_end = (isize - 1) >> PAGE_SHIFT;
1305         if (!isize || start_index > file_end)
1306                 return 0;
1307
1308         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1309
1310         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1311                         start, page_cnt << PAGE_SHIFT);
1312         if (ret)
1313                 return ret;
1314         i_done = 0;
1315         tree = &BTRFS_I(inode)->io_tree;
1316
1317         /* step one, lock all the pages */
1318         for (i = 0; i < page_cnt; i++) {
1319                 struct page *page;
1320 again:
1321                 page = find_or_create_page(inode->i_mapping,
1322                                            start_index + i, mask);
1323                 if (!page)
1324                         break;
1325
1326                 ret = set_page_extent_mapped(page);
1327                 if (ret < 0) {
1328                         unlock_page(page);
1329                         put_page(page);
1330                         break;
1331                 }
1332
1333                 page_start = page_offset(page);
1334                 page_end = page_start + PAGE_SIZE - 1;
1335                 while (1) {
1336                         lock_extent_bits(tree, page_start, page_end,
1337                                          &cached_state);
1338                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1339                                                               page_start);
1340                         unlock_extent_cached(tree, page_start, page_end,
1341                                              &cached_state);
1342                         if (!ordered)
1343                                 break;
1344
1345                         unlock_page(page);
1346                         btrfs_start_ordered_extent(ordered, 1);
1347                         btrfs_put_ordered_extent(ordered);
1348                         lock_page(page);
1349                         /*
1350                          * we unlocked the page above, so we need check if
1351                          * it was released or not.
1352                          */
1353                         if (page->mapping != inode->i_mapping) {
1354                                 unlock_page(page);
1355                                 put_page(page);
1356                                 goto again;
1357                         }
1358                 }
1359
1360                 if (!PageUptodate(page)) {
1361                         btrfs_readpage(NULL, page);
1362                         lock_page(page);
1363                         if (!PageUptodate(page)) {
1364                                 unlock_page(page);
1365                                 put_page(page);
1366                                 ret = -EIO;
1367                                 break;
1368                         }
1369                 }
1370
1371                 if (page->mapping != inode->i_mapping) {
1372                         unlock_page(page);
1373                         put_page(page);
1374                         goto again;
1375                 }
1376
1377                 pages[i] = page;
1378                 i_done++;
1379         }
1380         if (!i_done || ret)
1381                 goto out;
1382
1383         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1384                 goto out;
1385
1386         /*
1387          * so now we have a nice long stream of locked
1388          * and up to date pages, lets wait on them
1389          */
1390         for (i = 0; i < i_done; i++)
1391                 wait_on_page_writeback(pages[i]);
1392
1393         page_start = page_offset(pages[0]);
1394         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1395
1396         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1397                          page_start, page_end - 1, &cached_state);
1398
1399         /*
1400          * When defragmenting we skip ranges that have holes or inline extents,
1401          * (check should_defrag_range()), to avoid unnecessary IO and wasting
1402          * space. At btrfs_defrag_file(), we check if a range should be defragged
1403          * before locking the inode and then, if it should, we trigger a sync
1404          * page cache readahead - we lock the inode only after that to avoid
1405          * blocking for too long other tasks that possibly want to operate on
1406          * other file ranges. But before we were able to get the inode lock,
1407          * some other task may have punched a hole in the range, or we may have
1408          * now an inline extent, in which case we should not defrag. So check
1409          * for that here, where we have the inode and the range locked, and bail
1410          * out if that happened.
1411          */
1412         search_start = page_start;
1413         while (search_start < page_end) {
1414                 struct extent_map *em;
1415
1416                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, search_start,
1417                                       page_end - search_start);
1418                 if (IS_ERR(em)) {
1419                         ret = PTR_ERR(em);
1420                         goto out_unlock_range;
1421                 }
1422                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1423                         free_extent_map(em);
1424                         /* Ok, 0 means we did not defrag anything */
1425                         ret = 0;
1426                         goto out_unlock_range;
1427                 }
1428                 search_start = extent_map_end(em);
1429                 free_extent_map(em);
1430         }
1431
1432         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1433                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1434                           EXTENT_DEFRAG, 0, 0, &cached_state);
1435
1436         if (i_done != page_cnt) {
1437                 spin_lock(&BTRFS_I(inode)->lock);
1438                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1439                 spin_unlock(&BTRFS_I(inode)->lock);
1440                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1441                                 start, (page_cnt - i_done) << PAGE_SHIFT, true);
1442         }
1443
1444
1445         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1446                           &cached_state);
1447
1448         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1449                              page_start, page_end - 1, &cached_state);
1450
1451         for (i = 0; i < i_done; i++) {
1452                 clear_page_dirty_for_io(pages[i]);
1453                 ClearPageChecked(pages[i]);
1454                 set_page_dirty(pages[i]);
1455                 unlock_page(pages[i]);
1456                 put_page(pages[i]);
1457         }
1458         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1459         extent_changeset_free(data_reserved);
1460         return i_done;
1461
1462 out_unlock_range:
1463         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1464                              page_start, page_end - 1, &cached_state);
1465 out:
1466         for (i = 0; i < i_done; i++) {
1467                 unlock_page(pages[i]);
1468                 put_page(pages[i]);
1469         }
1470         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1471                         start, page_cnt << PAGE_SHIFT, true);
1472         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1473         extent_changeset_free(data_reserved);
1474         return ret;
1475
1476 }
1477
1478 int btrfs_defrag_file(struct inode *inode, struct file *file,
1479                       struct btrfs_ioctl_defrag_range_args *range,
1480                       u64 newer_than, unsigned long max_to_defrag)
1481 {
1482         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1483         struct btrfs_root *root = BTRFS_I(inode)->root;
1484         struct file_ra_state *ra = NULL;
1485         unsigned long last_index;
1486         u64 isize = i_size_read(inode);
1487         u64 last_len = 0;
1488         u64 skip = 0;
1489         u64 defrag_end = 0;
1490         u64 newer_off = range->start;
1491         unsigned long i;
1492         unsigned long ra_index = 0;
1493         int ret;
1494         int defrag_count = 0;
1495         int compress_type = BTRFS_COMPRESS_ZLIB;
1496         u32 extent_thresh = range->extent_thresh;
1497         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1498         unsigned long cluster = max_cluster;
1499         u64 new_align = ~((u64)SZ_128K - 1);
1500         struct page **pages = NULL;
1501         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1502
1503         if (isize == 0)
1504                 return 0;
1505
1506         if (range->start >= isize)
1507                 return -EINVAL;
1508
1509         if (do_compress) {
1510                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1511                         return -EINVAL;
1512                 if (range->compress_type)
1513                         compress_type = range->compress_type;
1514         }
1515
1516         if (extent_thresh == 0)
1517                 extent_thresh = SZ_256K;
1518
1519         /*
1520          * If we were not given a file, allocate a readahead context. As
1521          * readahead is just an optimization, defrag will work without it so
1522          * we don't error out.
1523          */
1524         if (!file) {
1525                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1526                 if (ra)
1527                         file_ra_state_init(ra, inode->i_mapping);
1528         } else {
1529                 ra = &file->f_ra;
1530         }
1531
1532         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1533         if (!pages) {
1534                 ret = -ENOMEM;
1535                 goto out_ra;
1536         }
1537
1538         /* find the last page to defrag */
1539         if (range->start + range->len > range->start) {
1540                 last_index = min_t(u64, isize - 1,
1541                          range->start + range->len - 1) >> PAGE_SHIFT;
1542         } else {
1543                 last_index = (isize - 1) >> PAGE_SHIFT;
1544         }
1545
1546         if (newer_than) {
1547                 ret = find_new_extents(root, inode, newer_than,
1548                                        &newer_off, SZ_64K);
1549                 if (!ret) {
1550                         range->start = newer_off;
1551                         /*
1552                          * we always align our defrag to help keep
1553                          * the extents in the file evenly spaced
1554                          */
1555                         i = (newer_off & new_align) >> PAGE_SHIFT;
1556                 } else
1557                         goto out_ra;
1558         } else {
1559                 i = range->start >> PAGE_SHIFT;
1560         }
1561         if (!max_to_defrag)
1562                 max_to_defrag = last_index - i + 1;
1563
1564         /*
1565          * make writeback starts from i, so the defrag range can be
1566          * written sequentially.
1567          */
1568         if (i < inode->i_mapping->writeback_index)
1569                 inode->i_mapping->writeback_index = i;
1570
1571         while (i <= last_index && defrag_count < max_to_defrag &&
1572                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1573                 /*
1574                  * make sure we stop running if someone unmounts
1575                  * the FS
1576                  */
1577                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1578                         break;
1579
1580                 if (btrfs_defrag_cancelled(fs_info)) {
1581                         btrfs_debug(fs_info, "defrag_file cancelled");
1582                         ret = -EAGAIN;
1583                         break;
1584                 }
1585
1586                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1587                                          extent_thresh, &last_len, &skip,
1588                                          &defrag_end, do_compress)){
1589                         unsigned long next;
1590                         /*
1591                          * the should_defrag function tells us how much to skip
1592                          * bump our counter by the suggested amount
1593                          */
1594                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1595                         i = max(i + 1, next);
1596                         continue;
1597                 }
1598
1599                 if (!newer_than) {
1600                         cluster = (PAGE_ALIGN(defrag_end) >>
1601                                    PAGE_SHIFT) - i;
1602                         cluster = min(cluster, max_cluster);
1603                 } else {
1604                         cluster = max_cluster;
1605                 }
1606
1607                 if (i + cluster > ra_index) {
1608                         ra_index = max(i, ra_index);
1609                         if (ra)
1610                                 page_cache_sync_readahead(inode->i_mapping, ra,
1611                                                 file, ra_index, cluster);
1612                         ra_index += cluster;
1613                 }
1614
1615                 inode_lock(inode);
1616                 if (IS_SWAPFILE(inode)) {
1617                         ret = -ETXTBSY;
1618                 } else {
1619                         if (do_compress)
1620                                 BTRFS_I(inode)->defrag_compress = compress_type;
1621                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1622                 }
1623                 if (ret < 0) {
1624                         inode_unlock(inode);
1625                         goto out_ra;
1626                 }
1627
1628                 defrag_count += ret;
1629                 balance_dirty_pages_ratelimited(inode->i_mapping);
1630                 inode_unlock(inode);
1631
1632                 if (newer_than) {
1633                         if (newer_off == (u64)-1)
1634                                 break;
1635
1636                         if (ret > 0)
1637                                 i += ret;
1638
1639                         newer_off = max(newer_off + 1,
1640                                         (u64)i << PAGE_SHIFT);
1641
1642                         ret = find_new_extents(root, inode, newer_than,
1643                                                &newer_off, SZ_64K);
1644                         if (!ret) {
1645                                 range->start = newer_off;
1646                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1647                         } else {
1648                                 break;
1649                         }
1650                 } else {
1651                         if (ret > 0) {
1652                                 i += ret;
1653                                 last_len += ret << PAGE_SHIFT;
1654                         } else {
1655                                 i++;
1656                                 last_len = 0;
1657                         }
1658                 }
1659         }
1660
1661         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1662                 filemap_flush(inode->i_mapping);
1663                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1664                              &BTRFS_I(inode)->runtime_flags))
1665                         filemap_flush(inode->i_mapping);
1666         }
1667
1668         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1669                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1670         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1671                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1672         }
1673
1674         ret = defrag_count;
1675
1676 out_ra:
1677         if (do_compress) {
1678                 inode_lock(inode);
1679                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1680                 inode_unlock(inode);
1681         }
1682         if (!file)
1683                 kfree(ra);
1684         kfree(pages);
1685         return ret;
1686 }
1687
1688 static noinline int btrfs_ioctl_resize(struct file *file,
1689                                         void __user *arg)
1690 {
1691         struct inode *inode = file_inode(file);
1692         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1693         u64 new_size;
1694         u64 old_size;
1695         u64 devid = 1;
1696         struct btrfs_root *root = BTRFS_I(inode)->root;
1697         struct btrfs_ioctl_vol_args *vol_args;
1698         struct btrfs_trans_handle *trans;
1699         struct btrfs_device *device = NULL;
1700         char *sizestr;
1701         char *retptr;
1702         char *devstr = NULL;
1703         int ret = 0;
1704         int mod = 0;
1705
1706         if (!capable(CAP_SYS_ADMIN))
1707                 return -EPERM;
1708
1709         ret = mnt_want_write_file(file);
1710         if (ret)
1711                 return ret;
1712
1713         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_RESIZE)) {
1714                 mnt_drop_write_file(file);
1715                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1716         }
1717
1718         vol_args = memdup_user(arg, sizeof(*vol_args));
1719         if (IS_ERR(vol_args)) {
1720                 ret = PTR_ERR(vol_args);
1721                 goto out;
1722         }
1723
1724         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1725
1726         sizestr = vol_args->name;
1727         devstr = strchr(sizestr, ':');
1728         if (devstr) {
1729                 sizestr = devstr + 1;
1730                 *devstr = '\0';
1731                 devstr = vol_args->name;
1732                 ret = kstrtoull(devstr, 10, &devid);
1733                 if (ret)
1734                         goto out_free;
1735                 if (!devid) {
1736                         ret = -EINVAL;
1737                         goto out_free;
1738                 }
1739                 btrfs_info(fs_info, "resizing devid %llu", devid);
1740         }
1741
1742         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
1743         if (!device) {
1744                 btrfs_info(fs_info, "resizer unable to find device %llu",
1745                            devid);
1746                 ret = -ENODEV;
1747                 goto out_free;
1748         }
1749
1750         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1751                 btrfs_info(fs_info,
1752                            "resizer unable to apply on readonly device %llu",
1753                        devid);
1754                 ret = -EPERM;
1755                 goto out_free;
1756         }
1757
1758         if (!strcmp(sizestr, "max"))
1759                 new_size = device->bdev->bd_inode->i_size;
1760         else {
1761                 if (sizestr[0] == '-') {
1762                         mod = -1;
1763                         sizestr++;
1764                 } else if (sizestr[0] == '+') {
1765                         mod = 1;
1766                         sizestr++;
1767                 }
1768                 new_size = memparse(sizestr, &retptr);
1769                 if (*retptr != '\0' || new_size == 0) {
1770                         ret = -EINVAL;
1771                         goto out_free;
1772                 }
1773         }
1774
1775         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1776                 ret = -EPERM;
1777                 goto out_free;
1778         }
1779
1780         old_size = btrfs_device_get_total_bytes(device);
1781
1782         if (mod < 0) {
1783                 if (new_size > old_size) {
1784                         ret = -EINVAL;
1785                         goto out_free;
1786                 }
1787                 new_size = old_size - new_size;
1788         } else if (mod > 0) {
1789                 if (new_size > ULLONG_MAX - old_size) {
1790                         ret = -ERANGE;
1791                         goto out_free;
1792                 }
1793                 new_size = old_size + new_size;
1794         }
1795
1796         if (new_size < SZ_256M) {
1797                 ret = -EINVAL;
1798                 goto out_free;
1799         }
1800         if (new_size > device->bdev->bd_inode->i_size) {
1801                 ret = -EFBIG;
1802                 goto out_free;
1803         }
1804
1805         new_size = round_down(new_size, fs_info->sectorsize);
1806
1807         if (new_size > old_size) {
1808                 trans = btrfs_start_transaction(root, 0);
1809                 if (IS_ERR(trans)) {
1810                         ret = PTR_ERR(trans);
1811                         goto out_free;
1812                 }
1813                 ret = btrfs_grow_device(trans, device, new_size);
1814                 btrfs_commit_transaction(trans);
1815         } else if (new_size < old_size) {
1816                 ret = btrfs_shrink_device(device, new_size);
1817         } /* equal, nothing need to do */
1818
1819         if (ret == 0 && new_size != old_size)
1820                 btrfs_info_in_rcu(fs_info,
1821                         "resize device %s (devid %llu) from %llu to %llu",
1822                         rcu_str_deref(device->name), device->devid,
1823                         old_size, new_size);
1824 out_free:
1825         kfree(vol_args);
1826 out:
1827         btrfs_exclop_finish(fs_info);
1828         mnt_drop_write_file(file);
1829         return ret;
1830 }
1831
1832 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1833                                 const char *name, unsigned long fd, int subvol,
1834                                 bool readonly,
1835                                 struct btrfs_qgroup_inherit *inherit)
1836 {
1837         int namelen;
1838         int ret = 0;
1839
1840         if (!S_ISDIR(file_inode(file)->i_mode))
1841                 return -ENOTDIR;
1842
1843         ret = mnt_want_write_file(file);
1844         if (ret)
1845                 goto out;
1846
1847         namelen = strlen(name);
1848         if (strchr(name, '/')) {
1849                 ret = -EINVAL;
1850                 goto out_drop_write;
1851         }
1852
1853         if (name[0] == '.' &&
1854            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1855                 ret = -EEXIST;
1856                 goto out_drop_write;
1857         }
1858
1859         if (subvol) {
1860                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1861                                      NULL, readonly, inherit);
1862         } else {
1863                 struct fd src = fdget(fd);
1864                 struct inode *src_inode;
1865                 if (!src.file) {
1866                         ret = -EINVAL;
1867                         goto out_drop_write;
1868                 }
1869
1870                 src_inode = file_inode(src.file);
1871                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1872                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1873                                    "Snapshot src from another FS");
1874                         ret = -EXDEV;
1875                 } else if (!inode_owner_or_capable(&init_user_ns, src_inode)) {
1876                         /*
1877                          * Subvolume creation is not restricted, but snapshots
1878                          * are limited to own subvolumes only
1879                          */
1880                         ret = -EPERM;
1881                 } else {
1882                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1883                                              BTRFS_I(src_inode)->root,
1884                                              readonly, inherit);
1885                 }
1886                 fdput(src);
1887         }
1888 out_drop_write:
1889         mnt_drop_write_file(file);
1890 out:
1891         return ret;
1892 }
1893
1894 static noinline int btrfs_ioctl_snap_create(struct file *file,
1895                                             void __user *arg, int subvol)
1896 {
1897         struct btrfs_ioctl_vol_args *vol_args;
1898         int ret;
1899
1900         if (!S_ISDIR(file_inode(file)->i_mode))
1901                 return -ENOTDIR;
1902
1903         vol_args = memdup_user(arg, sizeof(*vol_args));
1904         if (IS_ERR(vol_args))
1905                 return PTR_ERR(vol_args);
1906         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1907
1908         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1909                                         subvol, false, NULL);
1910
1911         kfree(vol_args);
1912         return ret;
1913 }
1914
1915 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1916                                                void __user *arg, int subvol)
1917 {
1918         struct btrfs_ioctl_vol_args_v2 *vol_args;
1919         int ret;
1920         bool readonly = false;
1921         struct btrfs_qgroup_inherit *inherit = NULL;
1922
1923         if (!S_ISDIR(file_inode(file)->i_mode))
1924                 return -ENOTDIR;
1925
1926         vol_args = memdup_user(arg, sizeof(*vol_args));
1927         if (IS_ERR(vol_args))
1928                 return PTR_ERR(vol_args);
1929         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1930
1931         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1932                 ret = -EOPNOTSUPP;
1933                 goto free_args;
1934         }
1935
1936         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1937                 readonly = true;
1938         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1939                 u64 nums;
1940
1941                 if (vol_args->size < sizeof(*inherit) ||
1942                     vol_args->size > PAGE_SIZE) {
1943                         ret = -EINVAL;
1944                         goto free_args;
1945                 }
1946                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1947                 if (IS_ERR(inherit)) {
1948                         ret = PTR_ERR(inherit);
1949                         goto free_args;
1950                 }
1951
1952                 if (inherit->num_qgroups > PAGE_SIZE ||
1953                     inherit->num_ref_copies > PAGE_SIZE ||
1954                     inherit->num_excl_copies > PAGE_SIZE) {
1955                         ret = -EINVAL;
1956                         goto free_inherit;
1957                 }
1958
1959                 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1960                        2 * inherit->num_excl_copies;
1961                 if (vol_args->size != struct_size(inherit, qgroups, nums)) {
1962                         ret = -EINVAL;
1963                         goto free_inherit;
1964                 }
1965         }
1966
1967         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1968                                         subvol, readonly, inherit);
1969         if (ret)
1970                 goto free_inherit;
1971 free_inherit:
1972         kfree(inherit);
1973 free_args:
1974         kfree(vol_args);
1975         return ret;
1976 }
1977
1978 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1979                                                 void __user *arg)
1980 {
1981         struct inode *inode = file_inode(file);
1982         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1983         struct btrfs_root *root = BTRFS_I(inode)->root;
1984         int ret = 0;
1985         u64 flags = 0;
1986
1987         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1988                 return -EINVAL;
1989
1990         down_read(&fs_info->subvol_sem);
1991         if (btrfs_root_readonly(root))
1992                 flags |= BTRFS_SUBVOL_RDONLY;
1993         up_read(&fs_info->subvol_sem);
1994
1995         if (copy_to_user(arg, &flags, sizeof(flags)))
1996                 ret = -EFAULT;
1997
1998         return ret;
1999 }
2000
2001 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
2002                                               void __user *arg)
2003 {
2004         struct inode *inode = file_inode(file);
2005         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2006         struct btrfs_root *root = BTRFS_I(inode)->root;
2007         struct btrfs_trans_handle *trans;
2008         u64 root_flags;
2009         u64 flags;
2010         int ret = 0;
2011
2012         if (!inode_owner_or_capable(&init_user_ns, inode))
2013                 return -EPERM;
2014
2015         ret = mnt_want_write_file(file);
2016         if (ret)
2017                 goto out;
2018
2019         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
2020                 ret = -EINVAL;
2021                 goto out_drop_write;
2022         }
2023
2024         if (copy_from_user(&flags, arg, sizeof(flags))) {
2025                 ret = -EFAULT;
2026                 goto out_drop_write;
2027         }
2028
2029         if (flags & ~BTRFS_SUBVOL_RDONLY) {
2030                 ret = -EOPNOTSUPP;
2031                 goto out_drop_write;
2032         }
2033
2034         down_write(&fs_info->subvol_sem);
2035
2036         /* nothing to do */
2037         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
2038                 goto out_drop_sem;
2039
2040         root_flags = btrfs_root_flags(&root->root_item);
2041         if (flags & BTRFS_SUBVOL_RDONLY) {
2042                 btrfs_set_root_flags(&root->root_item,
2043                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
2044         } else {
2045                 /*
2046                  * Block RO -> RW transition if this subvolume is involved in
2047                  * send
2048                  */
2049                 spin_lock(&root->root_item_lock);
2050                 if (root->send_in_progress == 0) {
2051                         btrfs_set_root_flags(&root->root_item,
2052                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
2053                         spin_unlock(&root->root_item_lock);
2054                 } else {
2055                         spin_unlock(&root->root_item_lock);
2056                         btrfs_warn(fs_info,
2057                                    "Attempt to set subvolume %llu read-write during send",
2058                                    root->root_key.objectid);
2059                         ret = -EPERM;
2060                         goto out_drop_sem;
2061                 }
2062         }
2063
2064         trans = btrfs_start_transaction(root, 1);
2065         if (IS_ERR(trans)) {
2066                 ret = PTR_ERR(trans);
2067                 goto out_reset;
2068         }
2069
2070         ret = btrfs_update_root(trans, fs_info->tree_root,
2071                                 &root->root_key, &root->root_item);
2072         if (ret < 0) {
2073                 btrfs_end_transaction(trans);
2074                 goto out_reset;
2075         }
2076
2077         ret = btrfs_commit_transaction(trans);
2078
2079 out_reset:
2080         if (ret)
2081                 btrfs_set_root_flags(&root->root_item, root_flags);
2082 out_drop_sem:
2083         up_write(&fs_info->subvol_sem);
2084 out_drop_write:
2085         mnt_drop_write_file(file);
2086 out:
2087         return ret;
2088 }
2089
2090 static noinline int key_in_sk(struct btrfs_key *key,
2091                               struct btrfs_ioctl_search_key *sk)
2092 {
2093         struct btrfs_key test;
2094         int ret;
2095
2096         test.objectid = sk->min_objectid;
2097         test.type = sk->min_type;
2098         test.offset = sk->min_offset;
2099
2100         ret = btrfs_comp_cpu_keys(key, &test);
2101         if (ret < 0)
2102                 return 0;
2103
2104         test.objectid = sk->max_objectid;
2105         test.type = sk->max_type;
2106         test.offset = sk->max_offset;
2107
2108         ret = btrfs_comp_cpu_keys(key, &test);
2109         if (ret > 0)
2110                 return 0;
2111         return 1;
2112 }
2113
2114 static noinline int copy_to_sk(struct btrfs_path *path,
2115                                struct btrfs_key *key,
2116                                struct btrfs_ioctl_search_key *sk,
2117                                size_t *buf_size,
2118                                char __user *ubuf,
2119                                unsigned long *sk_offset,
2120                                int *num_found)
2121 {
2122         u64 found_transid;
2123         struct extent_buffer *leaf;
2124         struct btrfs_ioctl_search_header sh;
2125         struct btrfs_key test;
2126         unsigned long item_off;
2127         unsigned long item_len;
2128         int nritems;
2129         int i;
2130         int slot;
2131         int ret = 0;
2132
2133         leaf = path->nodes[0];
2134         slot = path->slots[0];
2135         nritems = btrfs_header_nritems(leaf);
2136
2137         if (btrfs_header_generation(leaf) > sk->max_transid) {
2138                 i = nritems;
2139                 goto advance_key;
2140         }
2141         found_transid = btrfs_header_generation(leaf);
2142
2143         for (i = slot; i < nritems; i++) {
2144                 item_off = btrfs_item_ptr_offset(leaf, i);
2145                 item_len = btrfs_item_size_nr(leaf, i);
2146
2147                 btrfs_item_key_to_cpu(leaf, key, i);
2148                 if (!key_in_sk(key, sk))
2149                         continue;
2150
2151                 if (sizeof(sh) + item_len > *buf_size) {
2152                         if (*num_found) {
2153                                 ret = 1;
2154                                 goto out;
2155                         }
2156
2157                         /*
2158                          * return one empty item back for v1, which does not
2159                          * handle -EOVERFLOW
2160                          */
2161
2162                         *buf_size = sizeof(sh) + item_len;
2163                         item_len = 0;
2164                         ret = -EOVERFLOW;
2165                 }
2166
2167                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2168                         ret = 1;
2169                         goto out;
2170                 }
2171
2172                 sh.objectid = key->objectid;
2173                 sh.offset = key->offset;
2174                 sh.type = key->type;
2175                 sh.len = item_len;
2176                 sh.transid = found_transid;
2177
2178                 /*
2179                  * Copy search result header. If we fault then loop again so we
2180                  * can fault in the pages and -EFAULT there if there's a
2181                  * problem. Otherwise we'll fault and then copy the buffer in
2182                  * properly this next time through
2183                  */
2184                 if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) {
2185                         ret = 0;
2186                         goto out;
2187                 }
2188
2189                 *sk_offset += sizeof(sh);
2190
2191                 if (item_len) {
2192                         char __user *up = ubuf + *sk_offset;
2193                         /*
2194                          * Copy the item, same behavior as above, but reset the
2195                          * * sk_offset so we copy the full thing again.
2196                          */
2197                         if (read_extent_buffer_to_user_nofault(leaf, up,
2198                                                 item_off, item_len)) {
2199                                 ret = 0;
2200                                 *sk_offset -= sizeof(sh);
2201                                 goto out;
2202                         }
2203
2204                         *sk_offset += item_len;
2205                 }
2206                 (*num_found)++;
2207
2208                 if (ret) /* -EOVERFLOW from above */
2209                         goto out;
2210
2211                 if (*num_found >= sk->nr_items) {
2212                         ret = 1;
2213                         goto out;
2214                 }
2215         }
2216 advance_key:
2217         ret = 0;
2218         test.objectid = sk->max_objectid;
2219         test.type = sk->max_type;
2220         test.offset = sk->max_offset;
2221         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2222                 ret = 1;
2223         else if (key->offset < (u64)-1)
2224                 key->offset++;
2225         else if (key->type < (u8)-1) {
2226                 key->offset = 0;
2227                 key->type++;
2228         } else if (key->objectid < (u64)-1) {
2229                 key->offset = 0;
2230                 key->type = 0;
2231                 key->objectid++;
2232         } else
2233                 ret = 1;
2234 out:
2235         /*
2236          *  0: all items from this leaf copied, continue with next
2237          *  1: * more items can be copied, but unused buffer is too small
2238          *     * all items were found
2239          *     Either way, it will stops the loop which iterates to the next
2240          *     leaf
2241          *  -EOVERFLOW: item was to large for buffer
2242          *  -EFAULT: could not copy extent buffer back to userspace
2243          */
2244         return ret;
2245 }
2246
2247 static noinline int search_ioctl(struct inode *inode,
2248                                  struct btrfs_ioctl_search_key *sk,
2249                                  size_t *buf_size,
2250                                  char __user *ubuf)
2251 {
2252         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2253         struct btrfs_root *root;
2254         struct btrfs_key key;
2255         struct btrfs_path *path;
2256         int ret;
2257         int num_found = 0;
2258         unsigned long sk_offset = 0;
2259
2260         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2261                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2262                 return -EOVERFLOW;
2263         }
2264
2265         path = btrfs_alloc_path();
2266         if (!path)
2267                 return -ENOMEM;
2268
2269         if (sk->tree_id == 0) {
2270                 /* search the root of the inode that was passed */
2271                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2272         } else {
2273                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2274                 if (IS_ERR(root)) {
2275                         btrfs_free_path(path);
2276                         return PTR_ERR(root);
2277                 }
2278         }
2279
2280         key.objectid = sk->min_objectid;
2281         key.type = sk->min_type;
2282         key.offset = sk->min_offset;
2283
2284         while (1) {
2285                 ret = fault_in_pages_writeable(ubuf + sk_offset,
2286                                                *buf_size - sk_offset);
2287                 if (ret)
2288                         break;
2289
2290                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2291                 if (ret != 0) {
2292                         if (ret > 0)
2293                                 ret = 0;
2294                         goto err;
2295                 }
2296                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2297                                  &sk_offset, &num_found);
2298                 btrfs_release_path(path);
2299                 if (ret)
2300                         break;
2301
2302         }
2303         if (ret > 0)
2304                 ret = 0;
2305 err:
2306         sk->nr_items = num_found;
2307         btrfs_put_root(root);
2308         btrfs_free_path(path);
2309         return ret;
2310 }
2311
2312 static noinline int btrfs_ioctl_tree_search(struct file *file,
2313                                            void __user *argp)
2314 {
2315         struct btrfs_ioctl_search_args __user *uargs;
2316         struct btrfs_ioctl_search_key sk;
2317         struct inode *inode;
2318         int ret;
2319         size_t buf_size;
2320
2321         if (!capable(CAP_SYS_ADMIN))
2322                 return -EPERM;
2323
2324         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2325
2326         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2327                 return -EFAULT;
2328
2329         buf_size = sizeof(uargs->buf);
2330
2331         inode = file_inode(file);
2332         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2333
2334         /*
2335          * In the origin implementation an overflow is handled by returning a
2336          * search header with a len of zero, so reset ret.
2337          */
2338         if (ret == -EOVERFLOW)
2339                 ret = 0;
2340
2341         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2342                 ret = -EFAULT;
2343         return ret;
2344 }
2345
2346 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2347                                                void __user *argp)
2348 {
2349         struct btrfs_ioctl_search_args_v2 __user *uarg;
2350         struct btrfs_ioctl_search_args_v2 args;
2351         struct inode *inode;
2352         int ret;
2353         size_t buf_size;
2354         const size_t buf_limit = SZ_16M;
2355
2356         if (!capable(CAP_SYS_ADMIN))
2357                 return -EPERM;
2358
2359         /* copy search header and buffer size */
2360         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2361         if (copy_from_user(&args, uarg, sizeof(args)))
2362                 return -EFAULT;
2363
2364         buf_size = args.buf_size;
2365
2366         /* limit result size to 16MB */
2367         if (buf_size > buf_limit)
2368                 buf_size = buf_limit;
2369
2370         inode = file_inode(file);
2371         ret = search_ioctl(inode, &args.key, &buf_size,
2372                            (char __user *)(&uarg->buf[0]));
2373         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2374                 ret = -EFAULT;
2375         else if (ret == -EOVERFLOW &&
2376                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2377                 ret = -EFAULT;
2378
2379         return ret;
2380 }
2381
2382 /*
2383  * Search INODE_REFs to identify path name of 'dirid' directory
2384  * in a 'tree_id' tree. and sets path name to 'name'.
2385  */
2386 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2387                                 u64 tree_id, u64 dirid, char *name)
2388 {
2389         struct btrfs_root *root;
2390         struct btrfs_key key;
2391         char *ptr;
2392         int ret = -1;
2393         int slot;
2394         int len;
2395         int total_len = 0;
2396         struct btrfs_inode_ref *iref;
2397         struct extent_buffer *l;
2398         struct btrfs_path *path;
2399
2400         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2401                 name[0]='\0';
2402                 return 0;
2403         }
2404
2405         path = btrfs_alloc_path();
2406         if (!path)
2407                 return -ENOMEM;
2408
2409         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2410
2411         root = btrfs_get_fs_root(info, tree_id, true);
2412         if (IS_ERR(root)) {
2413                 ret = PTR_ERR(root);
2414                 root = NULL;
2415                 goto out;
2416         }
2417
2418         key.objectid = dirid;
2419         key.type = BTRFS_INODE_REF_KEY;
2420         key.offset = (u64)-1;
2421
2422         while (1) {
2423                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2424                 if (ret < 0)
2425                         goto out;
2426                 else if (ret > 0) {
2427                         ret = btrfs_previous_item(root, path, dirid,
2428                                                   BTRFS_INODE_REF_KEY);
2429                         if (ret < 0)
2430                                 goto out;
2431                         else if (ret > 0) {
2432                                 ret = -ENOENT;
2433                                 goto out;
2434                         }
2435                 }
2436
2437                 l = path->nodes[0];
2438                 slot = path->slots[0];
2439                 btrfs_item_key_to_cpu(l, &key, slot);
2440
2441                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2442                 len = btrfs_inode_ref_name_len(l, iref);
2443                 ptr -= len + 1;
2444                 total_len += len + 1;
2445                 if (ptr < name) {
2446                         ret = -ENAMETOOLONG;
2447                         goto out;
2448                 }
2449
2450                 *(ptr + len) = '/';
2451                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2452
2453                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2454                         break;
2455
2456                 btrfs_release_path(path);
2457                 key.objectid = key.offset;
2458                 key.offset = (u64)-1;
2459                 dirid = key.objectid;
2460         }
2461         memmove(name, ptr, total_len);
2462         name[total_len] = '\0';
2463         ret = 0;
2464 out:
2465         btrfs_put_root(root);
2466         btrfs_free_path(path);
2467         return ret;
2468 }
2469
2470 static int btrfs_search_path_in_tree_user(struct inode *inode,
2471                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2472 {
2473         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2474         struct super_block *sb = inode->i_sb;
2475         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2476         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2477         u64 dirid = args->dirid;
2478         unsigned long item_off;
2479         unsigned long item_len;
2480         struct btrfs_inode_ref *iref;
2481         struct btrfs_root_ref *rref;
2482         struct btrfs_root *root = NULL;
2483         struct btrfs_path *path;
2484         struct btrfs_key key, key2;
2485         struct extent_buffer *leaf;
2486         struct inode *temp_inode;
2487         char *ptr;
2488         int slot;
2489         int len;
2490         int total_len = 0;
2491         int ret;
2492
2493         path = btrfs_alloc_path();
2494         if (!path)
2495                 return -ENOMEM;
2496
2497         /*
2498          * If the bottom subvolume does not exist directly under upper_limit,
2499          * construct the path in from the bottom up.
2500          */
2501         if (dirid != upper_limit.objectid) {
2502                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2503
2504                 root = btrfs_get_fs_root(fs_info, treeid, true);
2505                 if (IS_ERR(root)) {
2506                         ret = PTR_ERR(root);
2507                         goto out;
2508                 }
2509
2510                 key.objectid = dirid;
2511                 key.type = BTRFS_INODE_REF_KEY;
2512                 key.offset = (u64)-1;
2513                 while (1) {
2514                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2515                         if (ret < 0) {
2516                                 goto out_put;
2517                         } else if (ret > 0) {
2518                                 ret = btrfs_previous_item(root, path, dirid,
2519                                                           BTRFS_INODE_REF_KEY);
2520                                 if (ret < 0) {
2521                                         goto out_put;
2522                                 } else if (ret > 0) {
2523                                         ret = -ENOENT;
2524                                         goto out_put;
2525                                 }
2526                         }
2527
2528                         leaf = path->nodes[0];
2529                         slot = path->slots[0];
2530                         btrfs_item_key_to_cpu(leaf, &key, slot);
2531
2532                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2533                         len = btrfs_inode_ref_name_len(leaf, iref);
2534                         ptr -= len + 1;
2535                         total_len += len + 1;
2536                         if (ptr < args->path) {
2537                                 ret = -ENAMETOOLONG;
2538                                 goto out_put;
2539                         }
2540
2541                         *(ptr + len) = '/';
2542                         read_extent_buffer(leaf, ptr,
2543                                         (unsigned long)(iref + 1), len);
2544
2545                         /* Check the read+exec permission of this directory */
2546                         ret = btrfs_previous_item(root, path, dirid,
2547                                                   BTRFS_INODE_ITEM_KEY);
2548                         if (ret < 0) {
2549                                 goto out_put;
2550                         } else if (ret > 0) {
2551                                 ret = -ENOENT;
2552                                 goto out_put;
2553                         }
2554
2555                         leaf = path->nodes[0];
2556                         slot = path->slots[0];
2557                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2558                         if (key2.objectid != dirid) {
2559                                 ret = -ENOENT;
2560                                 goto out_put;
2561                         }
2562
2563                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2564                         if (IS_ERR(temp_inode)) {
2565                                 ret = PTR_ERR(temp_inode);
2566                                 goto out_put;
2567                         }
2568                         ret = inode_permission(&init_user_ns, temp_inode,
2569                                                MAY_READ | MAY_EXEC);
2570                         iput(temp_inode);
2571                         if (ret) {
2572                                 ret = -EACCES;
2573                                 goto out_put;
2574                         }
2575
2576                         if (key.offset == upper_limit.objectid)
2577                                 break;
2578                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2579                                 ret = -EACCES;
2580                                 goto out_put;
2581                         }
2582
2583                         btrfs_release_path(path);
2584                         key.objectid = key.offset;
2585                         key.offset = (u64)-1;
2586                         dirid = key.objectid;
2587                 }
2588
2589                 memmove(args->path, ptr, total_len);
2590                 args->path[total_len] = '\0';
2591                 btrfs_put_root(root);
2592                 root = NULL;
2593                 btrfs_release_path(path);
2594         }
2595
2596         /* Get the bottom subvolume's name from ROOT_REF */
2597         key.objectid = treeid;
2598         key.type = BTRFS_ROOT_REF_KEY;
2599         key.offset = args->treeid;
2600         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2601         if (ret < 0) {
2602                 goto out;
2603         } else if (ret > 0) {
2604                 ret = -ENOENT;
2605                 goto out;
2606         }
2607
2608         leaf = path->nodes[0];
2609         slot = path->slots[0];
2610         btrfs_item_key_to_cpu(leaf, &key, slot);
2611
2612         item_off = btrfs_item_ptr_offset(leaf, slot);
2613         item_len = btrfs_item_size_nr(leaf, slot);
2614         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2615         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2616         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2617                 ret = -EINVAL;
2618                 goto out;
2619         }
2620
2621         /* Copy subvolume's name */
2622         item_off += sizeof(struct btrfs_root_ref);
2623         item_len -= sizeof(struct btrfs_root_ref);
2624         read_extent_buffer(leaf, args->name, item_off, item_len);
2625         args->name[item_len] = 0;
2626
2627 out_put:
2628         btrfs_put_root(root);
2629 out:
2630         btrfs_free_path(path);
2631         return ret;
2632 }
2633
2634 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2635                                            void __user *argp)
2636 {
2637         struct btrfs_ioctl_ino_lookup_args *args;
2638         struct inode *inode;
2639         int ret = 0;
2640
2641         args = memdup_user(argp, sizeof(*args));
2642         if (IS_ERR(args))
2643                 return PTR_ERR(args);
2644
2645         inode = file_inode(file);
2646
2647         /*
2648          * Unprivileged query to obtain the containing subvolume root id. The
2649          * path is reset so it's consistent with btrfs_search_path_in_tree.
2650          */
2651         if (args->treeid == 0)
2652                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2653
2654         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2655                 args->name[0] = 0;
2656                 goto out;
2657         }
2658
2659         if (!capable(CAP_SYS_ADMIN)) {
2660                 ret = -EPERM;
2661                 goto out;
2662         }
2663
2664         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2665                                         args->treeid, args->objectid,
2666                                         args->name);
2667
2668 out:
2669         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2670                 ret = -EFAULT;
2671
2672         kfree(args);
2673         return ret;
2674 }
2675
2676 /*
2677  * Version of ino_lookup ioctl (unprivileged)
2678  *
2679  * The main differences from ino_lookup ioctl are:
2680  *
2681  *   1. Read + Exec permission will be checked using inode_permission() during
2682  *      path construction. -EACCES will be returned in case of failure.
2683  *   2. Path construction will be stopped at the inode number which corresponds
2684  *      to the fd with which this ioctl is called. If constructed path does not
2685  *      exist under fd's inode, -EACCES will be returned.
2686  *   3. The name of bottom subvolume is also searched and filled.
2687  */
2688 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2689 {
2690         struct btrfs_ioctl_ino_lookup_user_args *args;
2691         struct inode *inode;
2692         int ret;
2693
2694         args = memdup_user(argp, sizeof(*args));
2695         if (IS_ERR(args))
2696                 return PTR_ERR(args);
2697
2698         inode = file_inode(file);
2699
2700         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2701             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2702                 /*
2703                  * The subvolume does not exist under fd with which this is
2704                  * called
2705                  */
2706                 kfree(args);
2707                 return -EACCES;
2708         }
2709
2710         ret = btrfs_search_path_in_tree_user(inode, args);
2711
2712         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2713                 ret = -EFAULT;
2714
2715         kfree(args);
2716         return ret;
2717 }
2718
2719 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2720 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2721 {
2722         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2723         struct btrfs_fs_info *fs_info;
2724         struct btrfs_root *root;
2725         struct btrfs_path *path;
2726         struct btrfs_key key;
2727         struct btrfs_root_item *root_item;
2728         struct btrfs_root_ref *rref;
2729         struct extent_buffer *leaf;
2730         unsigned long item_off;
2731         unsigned long item_len;
2732         struct inode *inode;
2733         int slot;
2734         int ret = 0;
2735
2736         path = btrfs_alloc_path();
2737         if (!path)
2738                 return -ENOMEM;
2739
2740         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2741         if (!subvol_info) {
2742                 btrfs_free_path(path);
2743                 return -ENOMEM;
2744         }
2745
2746         inode = file_inode(file);
2747         fs_info = BTRFS_I(inode)->root->fs_info;
2748
2749         /* Get root_item of inode's subvolume */
2750         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2751         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2752         if (IS_ERR(root)) {
2753                 ret = PTR_ERR(root);
2754                 goto out_free;
2755         }
2756         root_item = &root->root_item;
2757
2758         subvol_info->treeid = key.objectid;
2759
2760         subvol_info->generation = btrfs_root_generation(root_item);
2761         subvol_info->flags = btrfs_root_flags(root_item);
2762
2763         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2764         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2765                                                     BTRFS_UUID_SIZE);
2766         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2767                                                     BTRFS_UUID_SIZE);
2768
2769         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2770         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2771         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2772
2773         subvol_info->otransid = btrfs_root_otransid(root_item);
2774         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2775         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2776
2777         subvol_info->stransid = btrfs_root_stransid(root_item);
2778         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2779         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2780
2781         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2782         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2783         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2784
2785         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2786                 /* Search root tree for ROOT_BACKREF of this subvolume */
2787                 key.type = BTRFS_ROOT_BACKREF_KEY;
2788                 key.offset = 0;
2789                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2790                 if (ret < 0) {
2791                         goto out;
2792                 } else if (path->slots[0] >=
2793                            btrfs_header_nritems(path->nodes[0])) {
2794                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2795                         if (ret < 0) {
2796                                 goto out;
2797                         } else if (ret > 0) {
2798                                 ret = -EUCLEAN;
2799                                 goto out;
2800                         }
2801                 }
2802
2803                 leaf = path->nodes[0];
2804                 slot = path->slots[0];
2805                 btrfs_item_key_to_cpu(leaf, &key, slot);
2806                 if (key.objectid == subvol_info->treeid &&
2807                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2808                         subvol_info->parent_id = key.offset;
2809
2810                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2811                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2812
2813                         item_off = btrfs_item_ptr_offset(leaf, slot)
2814                                         + sizeof(struct btrfs_root_ref);
2815                         item_len = btrfs_item_size_nr(leaf, slot)
2816                                         - sizeof(struct btrfs_root_ref);
2817                         read_extent_buffer(leaf, subvol_info->name,
2818                                            item_off, item_len);
2819                 } else {
2820                         ret = -ENOENT;
2821                         goto out;
2822                 }
2823         }
2824
2825         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2826                 ret = -EFAULT;
2827
2828 out:
2829         btrfs_put_root(root);
2830 out_free:
2831         btrfs_free_path(path);
2832         kfree(subvol_info);
2833         return ret;
2834 }
2835
2836 /*
2837  * Return ROOT_REF information of the subvolume containing this inode
2838  * except the subvolume name.
2839  */
2840 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2841 {
2842         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2843         struct btrfs_root_ref *rref;
2844         struct btrfs_root *root;
2845         struct btrfs_path *path;
2846         struct btrfs_key key;
2847         struct extent_buffer *leaf;
2848         struct inode *inode;
2849         u64 objectid;
2850         int slot;
2851         int ret;
2852         u8 found;
2853
2854         path = btrfs_alloc_path();
2855         if (!path)
2856                 return -ENOMEM;
2857
2858         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2859         if (IS_ERR(rootrefs)) {
2860                 btrfs_free_path(path);
2861                 return PTR_ERR(rootrefs);
2862         }
2863
2864         inode = file_inode(file);
2865         root = BTRFS_I(inode)->root->fs_info->tree_root;
2866         objectid = BTRFS_I(inode)->root->root_key.objectid;
2867
2868         key.objectid = objectid;
2869         key.type = BTRFS_ROOT_REF_KEY;
2870         key.offset = rootrefs->min_treeid;
2871         found = 0;
2872
2873         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2874         if (ret < 0) {
2875                 goto out;
2876         } else if (path->slots[0] >=
2877                    btrfs_header_nritems(path->nodes[0])) {
2878                 ret = btrfs_next_leaf(root, path);
2879                 if (ret < 0) {
2880                         goto out;
2881                 } else if (ret > 0) {
2882                         ret = -EUCLEAN;
2883                         goto out;
2884                 }
2885         }
2886         while (1) {
2887                 leaf = path->nodes[0];
2888                 slot = path->slots[0];
2889
2890                 btrfs_item_key_to_cpu(leaf, &key, slot);
2891                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2892                         ret = 0;
2893                         goto out;
2894                 }
2895
2896                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2897                         ret = -EOVERFLOW;
2898                         goto out;
2899                 }
2900
2901                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2902                 rootrefs->rootref[found].treeid = key.offset;
2903                 rootrefs->rootref[found].dirid =
2904                                   btrfs_root_ref_dirid(leaf, rref);
2905                 found++;
2906
2907                 ret = btrfs_next_item(root, path);
2908                 if (ret < 0) {
2909                         goto out;
2910                 } else if (ret > 0) {
2911                         ret = -EUCLEAN;
2912                         goto out;
2913                 }
2914         }
2915
2916 out:
2917         if (!ret || ret == -EOVERFLOW) {
2918                 rootrefs->num_items = found;
2919                 /* update min_treeid for next search */
2920                 if (found)
2921                         rootrefs->min_treeid =
2922                                 rootrefs->rootref[found - 1].treeid + 1;
2923                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2924                         ret = -EFAULT;
2925         }
2926
2927         kfree(rootrefs);
2928         btrfs_free_path(path);
2929
2930         return ret;
2931 }
2932
2933 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2934                                              void __user *arg,
2935                                              bool destroy_v2)
2936 {
2937         struct dentry *parent = file->f_path.dentry;
2938         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2939         struct dentry *dentry;
2940         struct inode *dir = d_inode(parent);
2941         struct inode *inode;
2942         struct btrfs_root *root = BTRFS_I(dir)->root;
2943         struct btrfs_root *dest = NULL;
2944         struct btrfs_ioctl_vol_args *vol_args = NULL;
2945         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2946         char *subvol_name, *subvol_name_ptr = NULL;
2947         int subvol_namelen;
2948         int err = 0;
2949         bool destroy_parent = false;
2950
2951         if (destroy_v2) {
2952                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2953                 if (IS_ERR(vol_args2))
2954                         return PTR_ERR(vol_args2);
2955
2956                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2957                         err = -EOPNOTSUPP;
2958                         goto out;
2959                 }
2960
2961                 /*
2962                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2963                  * name, same as v1 currently does.
2964                  */
2965                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2966                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2967                         subvol_name = vol_args2->name;
2968
2969                         err = mnt_want_write_file(file);
2970                         if (err)
2971                                 goto out;
2972                 } else {
2973                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2974                                 err = -EINVAL;
2975                                 goto out;
2976                         }
2977
2978                         err = mnt_want_write_file(file);
2979                         if (err)
2980                                 goto out;
2981
2982                         dentry = btrfs_get_dentry(fs_info->sb,
2983                                         BTRFS_FIRST_FREE_OBJECTID,
2984                                         vol_args2->subvolid, 0, 0);
2985                         if (IS_ERR(dentry)) {
2986                                 err = PTR_ERR(dentry);
2987                                 goto out_drop_write;
2988                         }
2989
2990                         /*
2991                          * Change the default parent since the subvolume being
2992                          * deleted can be outside of the current mount point.
2993                          */
2994                         parent = btrfs_get_parent(dentry);
2995
2996                         /*
2997                          * At this point dentry->d_name can point to '/' if the
2998                          * subvolume we want to destroy is outsite of the
2999                          * current mount point, so we need to release the
3000                          * current dentry and execute the lookup to return a new
3001                          * one with ->d_name pointing to the
3002                          * <mount point>/subvol_name.
3003                          */
3004                         dput(dentry);
3005                         if (IS_ERR(parent)) {
3006                                 err = PTR_ERR(parent);
3007                                 goto out_drop_write;
3008                         }
3009                         dir = d_inode(parent);
3010
3011                         /*
3012                          * If v2 was used with SPEC_BY_ID, a new parent was
3013                          * allocated since the subvolume can be outside of the
3014                          * current mount point. Later on we need to release this
3015                          * new parent dentry.
3016                          */
3017                         destroy_parent = true;
3018
3019                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
3020                                                 fs_info, vol_args2->subvolid);
3021                         if (IS_ERR(subvol_name_ptr)) {
3022                                 err = PTR_ERR(subvol_name_ptr);
3023                                 goto free_parent;
3024                         }
3025                         /* subvol_name_ptr is already NULL termined */
3026                         subvol_name = (char *)kbasename(subvol_name_ptr);
3027                 }
3028         } else {
3029                 vol_args = memdup_user(arg, sizeof(*vol_args));
3030                 if (IS_ERR(vol_args))
3031                         return PTR_ERR(vol_args);
3032
3033                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
3034                 subvol_name = vol_args->name;
3035
3036                 err = mnt_want_write_file(file);
3037                 if (err)
3038                         goto out;
3039         }
3040
3041         subvol_namelen = strlen(subvol_name);
3042
3043         if (strchr(subvol_name, '/') ||
3044             strncmp(subvol_name, "..", subvol_namelen) == 0) {
3045                 err = -EINVAL;
3046                 goto free_subvol_name;
3047         }
3048
3049         if (!S_ISDIR(dir->i_mode)) {
3050                 err = -ENOTDIR;
3051                 goto free_subvol_name;
3052         }
3053
3054         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
3055         if (err == -EINTR)
3056                 goto free_subvol_name;
3057         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
3058         if (IS_ERR(dentry)) {
3059                 err = PTR_ERR(dentry);
3060                 goto out_unlock_dir;
3061         }
3062
3063         if (d_really_is_negative(dentry)) {
3064                 err = -ENOENT;
3065                 goto out_dput;
3066         }
3067
3068         inode = d_inode(dentry);
3069         dest = BTRFS_I(inode)->root;
3070         if (!capable(CAP_SYS_ADMIN)) {
3071                 /*
3072                  * Regular user.  Only allow this with a special mount
3073                  * option, when the user has write+exec access to the
3074                  * subvol root, and when rmdir(2) would have been
3075                  * allowed.
3076                  *
3077                  * Note that this is _not_ check that the subvol is
3078                  * empty or doesn't contain data that we wouldn't
3079                  * otherwise be able to delete.
3080                  *
3081                  * Users who want to delete empty subvols should try
3082                  * rmdir(2).
3083                  */
3084                 err = -EPERM;
3085                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
3086                         goto out_dput;
3087
3088                 /*
3089                  * Do not allow deletion if the parent dir is the same
3090                  * as the dir to be deleted.  That means the ioctl
3091                  * must be called on the dentry referencing the root
3092                  * of the subvol, not a random directory contained
3093                  * within it.
3094                  */
3095                 err = -EINVAL;
3096                 if (root == dest)
3097                         goto out_dput;
3098
3099                 err = inode_permission(&init_user_ns, inode,
3100                                        MAY_WRITE | MAY_EXEC);
3101                 if (err)
3102                         goto out_dput;
3103         }
3104
3105         /* check if subvolume may be deleted by a user */
3106         err = btrfs_may_delete(dir, dentry, 1);
3107         if (err)
3108                 goto out_dput;
3109
3110         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3111                 err = -EINVAL;
3112                 goto out_dput;
3113         }
3114
3115         inode_lock(inode);
3116         err = btrfs_delete_subvolume(dir, dentry);
3117         inode_unlock(inode);
3118         if (!err) {
3119                 fsnotify_rmdir(dir, dentry);
3120                 d_delete(dentry);
3121         }
3122
3123 out_dput:
3124         dput(dentry);
3125 out_unlock_dir:
3126         inode_unlock(dir);
3127 free_subvol_name:
3128         kfree(subvol_name_ptr);
3129 free_parent:
3130         if (destroy_parent)
3131                 dput(parent);
3132 out_drop_write:
3133         mnt_drop_write_file(file);
3134 out:
3135         kfree(vol_args2);
3136         kfree(vol_args);
3137         return err;
3138 }
3139
3140 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3141 {
3142         struct inode *inode = file_inode(file);
3143         struct btrfs_root *root = BTRFS_I(inode)->root;
3144         struct btrfs_ioctl_defrag_range_args *range;
3145         int ret;
3146
3147         ret = mnt_want_write_file(file);
3148         if (ret)
3149                 return ret;
3150
3151         if (btrfs_root_readonly(root)) {
3152                 ret = -EROFS;
3153                 goto out;
3154         }
3155
3156         switch (inode->i_mode & S_IFMT) {
3157         case S_IFDIR:
3158                 if (!capable(CAP_SYS_ADMIN)) {
3159                         ret = -EPERM;
3160                         goto out;
3161                 }
3162                 ret = btrfs_defrag_root(root);
3163                 break;
3164         case S_IFREG:
3165                 /*
3166                  * Note that this does not check the file descriptor for write
3167                  * access. This prevents defragmenting executables that are
3168                  * running and allows defrag on files open in read-only mode.
3169                  */
3170                 if (!capable(CAP_SYS_ADMIN) &&
3171                     inode_permission(&init_user_ns, inode, MAY_WRITE)) {
3172                         ret = -EPERM;
3173                         goto out;
3174                 }
3175
3176                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3177                 if (!range) {
3178                         ret = -ENOMEM;
3179                         goto out;
3180                 }
3181
3182                 if (argp) {
3183                         if (copy_from_user(range, argp,
3184                                            sizeof(*range))) {
3185                                 ret = -EFAULT;
3186                                 kfree(range);
3187                                 goto out;
3188                         }
3189                         /* compression requires us to start the IO */
3190                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3191                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3192                                 range->extent_thresh = (u32)-1;
3193                         }
3194                 } else {
3195                         /* the rest are all set to zero by kzalloc */
3196                         range->len = (u64)-1;
3197                 }
3198                 ret = btrfs_defrag_file(file_inode(file), file,
3199                                         range, BTRFS_OLDEST_GENERATION, 0);
3200                 if (ret > 0)
3201                         ret = 0;
3202                 kfree(range);
3203                 break;
3204         default:
3205                 ret = -EINVAL;
3206         }
3207 out:
3208         mnt_drop_write_file(file);
3209         return ret;
3210 }
3211
3212 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3213 {
3214         struct btrfs_ioctl_vol_args *vol_args;
3215         int ret;
3216
3217         if (!capable(CAP_SYS_ADMIN))
3218                 return -EPERM;
3219
3220         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD))
3221                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3222
3223         vol_args = memdup_user(arg, sizeof(*vol_args));
3224         if (IS_ERR(vol_args)) {
3225                 ret = PTR_ERR(vol_args);
3226                 goto out;
3227         }
3228
3229         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3230         ret = btrfs_init_new_device(fs_info, vol_args->name);
3231
3232         if (!ret)
3233                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3234
3235         kfree(vol_args);
3236 out:
3237         btrfs_exclop_finish(fs_info);
3238         return ret;
3239 }
3240
3241 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3242 {
3243         struct inode *inode = file_inode(file);
3244         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3245         struct btrfs_ioctl_vol_args_v2 *vol_args;
3246         int ret;
3247
3248         if (!capable(CAP_SYS_ADMIN))
3249                 return -EPERM;
3250
3251         ret = mnt_want_write_file(file);
3252         if (ret)
3253                 return ret;
3254
3255         vol_args = memdup_user(arg, sizeof(*vol_args));
3256         if (IS_ERR(vol_args)) {
3257                 ret = PTR_ERR(vol_args);
3258                 goto err_drop;
3259         }
3260
3261         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3262                 ret = -EOPNOTSUPP;
3263                 goto out;
3264         }
3265
3266         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3267                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3268                 goto out;
3269         }
3270
3271         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3272                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3273         } else {
3274                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3275                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3276         }
3277         btrfs_exclop_finish(fs_info);
3278
3279         if (!ret) {
3280                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3281                         btrfs_info(fs_info, "device deleted: id %llu",
3282                                         vol_args->devid);
3283                 else
3284                         btrfs_info(fs_info, "device deleted: %s",
3285                                         vol_args->name);
3286         }
3287 out:
3288         kfree(vol_args);
3289 err_drop:
3290         mnt_drop_write_file(file);
3291         return ret;
3292 }
3293
3294 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3295 {
3296         struct inode *inode = file_inode(file);
3297         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3298         struct btrfs_ioctl_vol_args *vol_args;
3299         int ret;
3300
3301         if (!capable(CAP_SYS_ADMIN))
3302                 return -EPERM;
3303
3304         ret = mnt_want_write_file(file);
3305         if (ret)
3306                 return ret;
3307
3308         if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REMOVE)) {
3309                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3310                 goto out_drop_write;
3311         }
3312
3313         vol_args = memdup_user(arg, sizeof(*vol_args));
3314         if (IS_ERR(vol_args)) {
3315                 ret = PTR_ERR(vol_args);
3316                 goto out;
3317         }
3318
3319         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3320         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3321
3322         if (!ret)
3323                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3324         kfree(vol_args);
3325 out:
3326         btrfs_exclop_finish(fs_info);
3327 out_drop_write:
3328         mnt_drop_write_file(file);
3329
3330         return ret;
3331 }
3332
3333 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3334                                 void __user *arg)
3335 {
3336         struct btrfs_ioctl_fs_info_args *fi_args;
3337         struct btrfs_device *device;
3338         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3339         u64 flags_in;
3340         int ret = 0;
3341
3342         fi_args = memdup_user(arg, sizeof(*fi_args));
3343         if (IS_ERR(fi_args))
3344                 return PTR_ERR(fi_args);
3345
3346         flags_in = fi_args->flags;
3347         memset(fi_args, 0, sizeof(*fi_args));
3348
3349         rcu_read_lock();
3350         fi_args->num_devices = fs_devices->num_devices;
3351
3352         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3353                 if (device->devid > fi_args->max_id)
3354                         fi_args->max_id = device->devid;
3355         }
3356         rcu_read_unlock();
3357
3358         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3359         fi_args->nodesize = fs_info->nodesize;
3360         fi_args->sectorsize = fs_info->sectorsize;
3361         fi_args->clone_alignment = fs_info->sectorsize;
3362
3363         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3364                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3365                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3366                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3367         }
3368
3369         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3370                 fi_args->generation = fs_info->generation;
3371                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3372         }
3373
3374         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3375                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3376                        sizeof(fi_args->metadata_uuid));
3377                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3378         }
3379
3380         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3381                 ret = -EFAULT;
3382
3383         kfree(fi_args);
3384         return ret;
3385 }
3386
3387 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3388                                  void __user *arg)
3389 {
3390         struct btrfs_ioctl_dev_info_args *di_args;
3391         struct btrfs_device *dev;
3392         int ret = 0;
3393         char *s_uuid = NULL;
3394
3395         di_args = memdup_user(arg, sizeof(*di_args));
3396         if (IS_ERR(di_args))
3397                 return PTR_ERR(di_args);
3398
3399         if (!btrfs_is_empty_uuid(di_args->uuid))
3400                 s_uuid = di_args->uuid;
3401
3402         rcu_read_lock();
3403         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3404                                 NULL);
3405
3406         if (!dev) {
3407                 ret = -ENODEV;
3408                 goto out;
3409         }
3410
3411         di_args->devid = dev->devid;
3412         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3413         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3414         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3415         if (dev->name) {
3416                 strncpy(di_args->path, rcu_str_deref(dev->name),
3417                                 sizeof(di_args->path) - 1);
3418                 di_args->path[sizeof(di_args->path) - 1] = 0;
3419         } else {
3420                 di_args->path[0] = '\0';
3421         }
3422
3423 out:
3424         rcu_read_unlock();
3425         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3426                 ret = -EFAULT;
3427
3428         kfree(di_args);
3429         return ret;
3430 }
3431
3432 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3433 {
3434         struct inode *inode = file_inode(file);
3435         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3436         struct btrfs_root *root = BTRFS_I(inode)->root;
3437         struct btrfs_root *new_root;
3438         struct btrfs_dir_item *di;
3439         struct btrfs_trans_handle *trans;
3440         struct btrfs_path *path = NULL;
3441         struct btrfs_disk_key disk_key;
3442         u64 objectid = 0;
3443         u64 dir_id;
3444         int ret;
3445
3446         if (!capable(CAP_SYS_ADMIN))
3447                 return -EPERM;
3448
3449         ret = mnt_want_write_file(file);
3450         if (ret)
3451                 return ret;
3452
3453         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3454                 ret = -EFAULT;
3455                 goto out;
3456         }
3457
3458         if (!objectid)
3459                 objectid = BTRFS_FS_TREE_OBJECTID;
3460
3461         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3462         if (IS_ERR(new_root)) {
3463                 ret = PTR_ERR(new_root);
3464                 goto out;
3465         }
3466         if (!is_fstree(new_root->root_key.objectid)) {
3467                 ret = -ENOENT;
3468                 goto out_free;
3469         }
3470
3471         path = btrfs_alloc_path();
3472         if (!path) {
3473                 ret = -ENOMEM;
3474                 goto out_free;
3475         }
3476
3477         trans = btrfs_start_transaction(root, 1);
3478         if (IS_ERR(trans)) {
3479                 ret = PTR_ERR(trans);
3480                 goto out_free;
3481         }
3482
3483         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3484         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3485                                    dir_id, "default", 7, 1);
3486         if (IS_ERR_OR_NULL(di)) {
3487                 btrfs_release_path(path);
3488                 btrfs_end_transaction(trans);
3489                 btrfs_err(fs_info,
3490                           "Umm, you don't have the default diritem, this isn't going to work");
3491                 ret = -ENOENT;
3492                 goto out_free;
3493         }
3494
3495         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3496         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3497         btrfs_mark_buffer_dirty(path->nodes[0]);
3498         btrfs_release_path(path);
3499
3500         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3501         btrfs_end_transaction(trans);
3502 out_free:
3503         btrfs_put_root(new_root);
3504         btrfs_free_path(path);
3505 out:
3506         mnt_drop_write_file(file);
3507         return ret;
3508 }
3509
3510 static void get_block_group_info(struct list_head *groups_list,
3511                                  struct btrfs_ioctl_space_info *space)
3512 {
3513         struct btrfs_block_group *block_group;
3514
3515         space->total_bytes = 0;
3516         space->used_bytes = 0;
3517         space->flags = 0;
3518         list_for_each_entry(block_group, groups_list, list) {
3519                 space->flags = block_group->flags;
3520                 space->total_bytes += block_group->length;
3521                 space->used_bytes += block_group->used;
3522         }
3523 }
3524
3525 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3526                                    void __user *arg)
3527 {
3528         struct btrfs_ioctl_space_args space_args;
3529         struct btrfs_ioctl_space_info space;
3530         struct btrfs_ioctl_space_info *dest;
3531         struct btrfs_ioctl_space_info *dest_orig;
3532         struct btrfs_ioctl_space_info __user *user_dest;
3533         struct btrfs_space_info *info;
3534         static const u64 types[] = {
3535                 BTRFS_BLOCK_GROUP_DATA,
3536                 BTRFS_BLOCK_GROUP_SYSTEM,
3537                 BTRFS_BLOCK_GROUP_METADATA,
3538                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3539         };
3540         int num_types = 4;
3541         int alloc_size;
3542         int ret = 0;
3543         u64 slot_count = 0;
3544         int i, c;
3545
3546         if (copy_from_user(&space_args,
3547                            (struct btrfs_ioctl_space_args __user *)arg,
3548                            sizeof(space_args)))
3549                 return -EFAULT;
3550
3551         for (i = 0; i < num_types; i++) {
3552                 struct btrfs_space_info *tmp;
3553
3554                 info = NULL;
3555                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3556                         if (tmp->flags == types[i]) {
3557                                 info = tmp;
3558                                 break;
3559                         }
3560                 }
3561
3562                 if (!info)
3563                         continue;
3564
3565                 down_read(&info->groups_sem);
3566                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3567                         if (!list_empty(&info->block_groups[c]))
3568                                 slot_count++;
3569                 }
3570                 up_read(&info->groups_sem);
3571         }
3572
3573         /*
3574          * Global block reserve, exported as a space_info
3575          */
3576         slot_count++;
3577
3578         /* space_slots == 0 means they are asking for a count */
3579         if (space_args.space_slots == 0) {
3580                 space_args.total_spaces = slot_count;
3581                 goto out;
3582         }
3583
3584         slot_count = min_t(u64, space_args.space_slots, slot_count);
3585
3586         alloc_size = sizeof(*dest) * slot_count;
3587
3588         /* we generally have at most 6 or so space infos, one for each raid
3589          * level.  So, a whole page should be more than enough for everyone
3590          */
3591         if (alloc_size > PAGE_SIZE)
3592                 return -ENOMEM;
3593
3594         space_args.total_spaces = 0;
3595         dest = kmalloc(alloc_size, GFP_KERNEL);
3596         if (!dest)
3597                 return -ENOMEM;
3598         dest_orig = dest;
3599
3600         /* now we have a buffer to copy into */
3601         for (i = 0; i < num_types; i++) {
3602                 struct btrfs_space_info *tmp;
3603
3604                 if (!slot_count)
3605                         break;
3606
3607                 info = NULL;
3608                 list_for_each_entry(tmp, &fs_info->space_info, list) {
3609                         if (tmp->flags == types[i]) {
3610                                 info = tmp;
3611                                 break;
3612                         }
3613                 }
3614
3615                 if (!info)
3616                         continue;
3617                 down_read(&info->groups_sem);
3618                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3619                         if (!list_empty(&info->block_groups[c])) {
3620                                 get_block_group_info(&info->block_groups[c],
3621                                                      &space);
3622                                 memcpy(dest, &space, sizeof(space));
3623                                 dest++;
3624                                 space_args.total_spaces++;
3625                                 slot_count--;
3626                         }
3627                         if (!slot_count)
3628                                 break;
3629                 }
3630                 up_read(&info->groups_sem);
3631         }
3632
3633         /*
3634          * Add global block reserve
3635          */
3636         if (slot_count) {
3637                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3638
3639                 spin_lock(&block_rsv->lock);
3640                 space.total_bytes = block_rsv->size;
3641                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3642                 spin_unlock(&block_rsv->lock);
3643                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3644                 memcpy(dest, &space, sizeof(space));
3645                 space_args.total_spaces++;
3646         }
3647
3648         user_dest = (struct btrfs_ioctl_space_info __user *)
3649                 (arg + sizeof(struct btrfs_ioctl_space_args));
3650
3651         if (copy_to_user(user_dest, dest_orig, alloc_size))
3652                 ret = -EFAULT;
3653
3654         kfree(dest_orig);
3655 out:
3656         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3657                 ret = -EFAULT;
3658
3659         return ret;
3660 }
3661
3662 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3663                                             void __user *argp)
3664 {
3665         struct btrfs_trans_handle *trans;
3666         u64 transid;
3667         int ret;
3668
3669         trans = btrfs_attach_transaction_barrier(root);
3670         if (IS_ERR(trans)) {
3671                 if (PTR_ERR(trans) != -ENOENT)
3672                         return PTR_ERR(trans);
3673
3674                 /* No running transaction, don't bother */
3675                 transid = root->fs_info->last_trans_committed;
3676                 goto out;
3677         }
3678         transid = trans->transid;
3679         ret = btrfs_commit_transaction_async(trans, 0);
3680         if (ret) {
3681                 btrfs_end_transaction(trans);
3682                 return ret;
3683         }
3684 out:
3685         if (argp)
3686                 if (copy_to_user(argp, &transid, sizeof(transid)))
3687                         return -EFAULT;
3688         return 0;
3689 }
3690
3691 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3692                                            void __user *argp)
3693 {
3694         u64 transid;
3695
3696         if (argp) {
3697                 if (copy_from_user(&transid, argp, sizeof(transid)))
3698                         return -EFAULT;
3699         } else {
3700                 transid = 0;  /* current trans */
3701         }
3702         return btrfs_wait_for_commit(fs_info, transid);
3703 }
3704
3705 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3706 {
3707         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3708         struct btrfs_ioctl_scrub_args *sa;
3709         int ret;
3710
3711         if (!capable(CAP_SYS_ADMIN))
3712                 return -EPERM;
3713
3714         sa = memdup_user(arg, sizeof(*sa));
3715         if (IS_ERR(sa))
3716                 return PTR_ERR(sa);
3717
3718         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3719                 ret = mnt_want_write_file(file);
3720                 if (ret)
3721                         goto out;
3722         }
3723
3724         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3725                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3726                               0);
3727
3728         /*
3729          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3730          * error. This is important as it allows user space to know how much
3731          * progress scrub has done. For example, if scrub is canceled we get
3732          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3733          * space. Later user space can inspect the progress from the structure
3734          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3735          * previously (btrfs-progs does this).
3736          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3737          * then return -EFAULT to signal the structure was not copied or it may
3738          * be corrupt and unreliable due to a partial copy.
3739          */
3740         if (copy_to_user(arg, sa, sizeof(*sa)))
3741                 ret = -EFAULT;
3742
3743         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3744                 mnt_drop_write_file(file);
3745 out:
3746         kfree(sa);
3747         return ret;
3748 }
3749
3750 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3751 {
3752         if (!capable(CAP_SYS_ADMIN))
3753                 return -EPERM;
3754
3755         return btrfs_scrub_cancel(fs_info);
3756 }
3757
3758 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3759                                        void __user *arg)
3760 {
3761         struct btrfs_ioctl_scrub_args *sa;
3762         int ret;
3763
3764         if (!capable(CAP_SYS_ADMIN))
3765                 return -EPERM;
3766
3767         sa = memdup_user(arg, sizeof(*sa));
3768         if (IS_ERR(sa))
3769                 return PTR_ERR(sa);
3770
3771         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3772
3773         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3774                 ret = -EFAULT;
3775
3776         kfree(sa);
3777         return ret;
3778 }
3779
3780 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3781                                       void __user *arg)
3782 {
3783         struct btrfs_ioctl_get_dev_stats *sa;
3784         int ret;
3785
3786         sa = memdup_user(arg, sizeof(*sa));
3787         if (IS_ERR(sa))
3788                 return PTR_ERR(sa);
3789
3790         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3791                 kfree(sa);
3792                 return -EPERM;
3793         }
3794
3795         ret = btrfs_get_dev_stats(fs_info, sa);
3796
3797         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3798                 ret = -EFAULT;
3799
3800         kfree(sa);
3801         return ret;
3802 }
3803
3804 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3805                                     void __user *arg)
3806 {
3807         struct btrfs_ioctl_dev_replace_args *p;
3808         int ret;
3809
3810         if (!capable(CAP_SYS_ADMIN))
3811                 return -EPERM;
3812
3813         p = memdup_user(arg, sizeof(*p));
3814         if (IS_ERR(p))
3815                 return PTR_ERR(p);
3816
3817         switch (p->cmd) {
3818         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3819                 if (sb_rdonly(fs_info->sb)) {
3820                         ret = -EROFS;
3821                         goto out;
3822                 }
3823                 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
3824                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3825                 } else {
3826                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3827                         btrfs_exclop_finish(fs_info);
3828                 }
3829                 break;
3830         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3831                 btrfs_dev_replace_status(fs_info, p);
3832                 ret = 0;
3833                 break;
3834         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3835                 p->result = btrfs_dev_replace_cancel(fs_info);
3836                 ret = 0;
3837                 break;
3838         default:
3839                 ret = -EINVAL;
3840                 break;
3841         }
3842
3843         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3844                 ret = -EFAULT;
3845 out:
3846         kfree(p);
3847         return ret;
3848 }
3849
3850 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3851 {
3852         int ret = 0;
3853         int i;
3854         u64 rel_ptr;
3855         int size;
3856         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3857         struct inode_fs_paths *ipath = NULL;
3858         struct btrfs_path *path;
3859
3860         if (!capable(CAP_DAC_READ_SEARCH))
3861                 return -EPERM;
3862
3863         path = btrfs_alloc_path();
3864         if (!path) {
3865                 ret = -ENOMEM;
3866                 goto out;
3867         }
3868
3869         ipa = memdup_user(arg, sizeof(*ipa));
3870         if (IS_ERR(ipa)) {
3871                 ret = PTR_ERR(ipa);
3872                 ipa = NULL;
3873                 goto out;
3874         }
3875
3876         size = min_t(u32, ipa->size, 4096);
3877         ipath = init_ipath(size, root, path);
3878         if (IS_ERR(ipath)) {
3879                 ret = PTR_ERR(ipath);
3880                 ipath = NULL;
3881                 goto out;
3882         }
3883
3884         ret = paths_from_inode(ipa->inum, ipath);
3885         if (ret < 0)
3886                 goto out;
3887
3888         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3889                 rel_ptr = ipath->fspath->val[i] -
3890                           (u64)(unsigned long)ipath->fspath->val;
3891                 ipath->fspath->val[i] = rel_ptr;
3892         }
3893
3894         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3895                            ipath->fspath, size);
3896         if (ret) {
3897                 ret = -EFAULT;
3898                 goto out;
3899         }
3900
3901 out:
3902         btrfs_free_path(path);
3903         free_ipath(ipath);
3904         kfree(ipa);
3905
3906         return ret;
3907 }
3908
3909 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3910 {
3911         struct btrfs_data_container *inodes = ctx;
3912         const size_t c = 3 * sizeof(u64);
3913
3914         if (inodes->bytes_left >= c) {
3915                 inodes->bytes_left -= c;
3916                 inodes->val[inodes->elem_cnt] = inum;
3917                 inodes->val[inodes->elem_cnt + 1] = offset;
3918                 inodes->val[inodes->elem_cnt + 2] = root;
3919                 inodes->elem_cnt += 3;
3920         } else {
3921                 inodes->bytes_missing += c - inodes->bytes_left;
3922                 inodes->bytes_left = 0;
3923                 inodes->elem_missed += 3;
3924         }
3925
3926         return 0;
3927 }
3928
3929 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3930                                         void __user *arg, int version)
3931 {
3932         int ret = 0;
3933         int size;
3934         struct btrfs_ioctl_logical_ino_args *loi;
3935         struct btrfs_data_container *inodes = NULL;
3936         struct btrfs_path *path = NULL;
3937         bool ignore_offset;
3938
3939         if (!capable(CAP_SYS_ADMIN))
3940                 return -EPERM;
3941
3942         loi = memdup_user(arg, sizeof(*loi));
3943         if (IS_ERR(loi))
3944                 return PTR_ERR(loi);
3945
3946         if (version == 1) {
3947                 ignore_offset = false;
3948                 size = min_t(u32, loi->size, SZ_64K);
3949         } else {
3950                 /* All reserved bits must be 0 for now */
3951                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3952                         ret = -EINVAL;
3953                         goto out_loi;
3954                 }
3955                 /* Only accept flags we have defined so far */
3956                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3957                         ret = -EINVAL;
3958                         goto out_loi;
3959                 }
3960                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3961                 size = min_t(u32, loi->size, SZ_16M);
3962         }
3963
3964         path = btrfs_alloc_path();
3965         if (!path) {
3966                 ret = -ENOMEM;
3967                 goto out;
3968         }
3969
3970         inodes = init_data_container(size);
3971         if (IS_ERR(inodes)) {
3972                 ret = PTR_ERR(inodes);
3973                 inodes = NULL;
3974                 goto out;
3975         }
3976
3977         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3978                                           build_ino_list, inodes, ignore_offset);
3979         if (ret == -EINVAL)
3980                 ret = -ENOENT;
3981         if (ret < 0)
3982                 goto out;
3983
3984         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3985                            size);
3986         if (ret)
3987                 ret = -EFAULT;
3988
3989 out:
3990         btrfs_free_path(path);
3991         kvfree(inodes);
3992 out_loi:
3993         kfree(loi);
3994
3995         return ret;
3996 }
3997
3998 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3999                                struct btrfs_ioctl_balance_args *bargs)
4000 {
4001         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4002
4003         bargs->flags = bctl->flags;
4004
4005         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
4006                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
4007         if (atomic_read(&fs_info->balance_pause_req))
4008                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
4009         if (atomic_read(&fs_info->balance_cancel_req))
4010                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
4011
4012         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
4013         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
4014         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
4015
4016         spin_lock(&fs_info->balance_lock);
4017         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
4018         spin_unlock(&fs_info->balance_lock);
4019 }
4020
4021 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
4022 {
4023         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
4024         struct btrfs_fs_info *fs_info = root->fs_info;
4025         struct btrfs_ioctl_balance_args *bargs;
4026         struct btrfs_balance_control *bctl;
4027         bool need_unlock; /* for mut. excl. ops lock */
4028         int ret;
4029
4030         if (!capable(CAP_SYS_ADMIN))
4031                 return -EPERM;
4032
4033         ret = mnt_want_write_file(file);
4034         if (ret)
4035                 return ret;
4036
4037 again:
4038         if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
4039                 mutex_lock(&fs_info->balance_mutex);
4040                 need_unlock = true;
4041                 goto locked;
4042         }
4043
4044         /*
4045          * mut. excl. ops lock is locked.  Three possibilities:
4046          *   (1) some other op is running
4047          *   (2) balance is running
4048          *   (3) balance is paused -- special case (think resume)
4049          */
4050         mutex_lock(&fs_info->balance_mutex);
4051         if (fs_info->balance_ctl) {
4052                 /* this is either (2) or (3) */
4053                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4054                         mutex_unlock(&fs_info->balance_mutex);
4055                         /*
4056                          * Lock released to allow other waiters to continue,
4057                          * we'll reexamine the status again.
4058                          */
4059                         mutex_lock(&fs_info->balance_mutex);
4060
4061                         if (fs_info->balance_ctl &&
4062                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4063                                 /* this is (3) */
4064                                 need_unlock = false;
4065                                 goto locked;
4066                         }
4067
4068                         mutex_unlock(&fs_info->balance_mutex);
4069                         goto again;
4070                 } else {
4071                         /* this is (2) */
4072                         mutex_unlock(&fs_info->balance_mutex);
4073                         ret = -EINPROGRESS;
4074                         goto out;
4075                 }
4076         } else {
4077                 /* this is (1) */
4078                 mutex_unlock(&fs_info->balance_mutex);
4079                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
4080                 goto out;
4081         }
4082
4083 locked:
4084
4085         if (arg) {
4086                 bargs = memdup_user(arg, sizeof(*bargs));
4087                 if (IS_ERR(bargs)) {
4088                         ret = PTR_ERR(bargs);
4089                         goto out_unlock;
4090                 }
4091
4092                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
4093                         if (!fs_info->balance_ctl) {
4094                                 ret = -ENOTCONN;
4095                                 goto out_bargs;
4096                         }
4097
4098                         bctl = fs_info->balance_ctl;
4099                         spin_lock(&fs_info->balance_lock);
4100                         bctl->flags |= BTRFS_BALANCE_RESUME;
4101                         spin_unlock(&fs_info->balance_lock);
4102
4103                         goto do_balance;
4104                 }
4105         } else {
4106                 bargs = NULL;
4107         }
4108
4109         if (fs_info->balance_ctl) {
4110                 ret = -EINPROGRESS;
4111                 goto out_bargs;
4112         }
4113
4114         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4115         if (!bctl) {
4116                 ret = -ENOMEM;
4117                 goto out_bargs;
4118         }
4119
4120         if (arg) {
4121                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4122                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4123                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4124
4125                 bctl->flags = bargs->flags;
4126         } else {
4127                 /* balance everything - no filters */
4128                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4129         }
4130
4131         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4132                 ret = -EINVAL;
4133                 goto out_bctl;
4134         }
4135
4136 do_balance:
4137         /*
4138          * Ownership of bctl and exclusive operation goes to btrfs_balance.
4139          * bctl is freed in reset_balance_state, or, if restriper was paused
4140          * all the way until unmount, in free_fs_info.  The flag should be
4141          * cleared after reset_balance_state.
4142          */
4143         need_unlock = false;
4144
4145         ret = btrfs_balance(fs_info, bctl, bargs);
4146         bctl = NULL;
4147
4148         if ((ret == 0 || ret == -ECANCELED) && arg) {
4149                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4150                         ret = -EFAULT;
4151         }
4152
4153 out_bctl:
4154         kfree(bctl);
4155 out_bargs:
4156         kfree(bargs);
4157 out_unlock:
4158         mutex_unlock(&fs_info->balance_mutex);
4159         if (need_unlock)
4160                 btrfs_exclop_finish(fs_info);
4161 out:
4162         mnt_drop_write_file(file);
4163         return ret;
4164 }
4165
4166 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4167 {
4168         if (!capable(CAP_SYS_ADMIN))
4169                 return -EPERM;
4170
4171         switch (cmd) {
4172         case BTRFS_BALANCE_CTL_PAUSE:
4173                 return btrfs_pause_balance(fs_info);
4174         case BTRFS_BALANCE_CTL_CANCEL:
4175                 return btrfs_cancel_balance(fs_info);
4176         }
4177
4178         return -EINVAL;
4179 }
4180
4181 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4182                                          void __user *arg)
4183 {
4184         struct btrfs_ioctl_balance_args *bargs;
4185         int ret = 0;
4186
4187         if (!capable(CAP_SYS_ADMIN))
4188                 return -EPERM;
4189
4190         mutex_lock(&fs_info->balance_mutex);
4191         if (!fs_info->balance_ctl) {
4192                 ret = -ENOTCONN;
4193                 goto out;
4194         }
4195
4196         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4197         if (!bargs) {
4198                 ret = -ENOMEM;
4199                 goto out;
4200         }
4201
4202         btrfs_update_ioctl_balance_args(fs_info, bargs);
4203
4204         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4205                 ret = -EFAULT;
4206
4207         kfree(bargs);
4208 out:
4209         mutex_unlock(&fs_info->balance_mutex);
4210         return ret;
4211 }
4212
4213 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4214 {
4215         struct inode *inode = file_inode(file);
4216         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4217         struct btrfs_ioctl_quota_ctl_args *sa;
4218         int ret;
4219
4220         if (!capable(CAP_SYS_ADMIN))
4221                 return -EPERM;
4222
4223         ret = mnt_want_write_file(file);
4224         if (ret)
4225                 return ret;
4226
4227         sa = memdup_user(arg, sizeof(*sa));
4228         if (IS_ERR(sa)) {
4229                 ret = PTR_ERR(sa);
4230                 goto drop_write;
4231         }
4232
4233         down_write(&fs_info->subvol_sem);
4234
4235         switch (sa->cmd) {
4236         case BTRFS_QUOTA_CTL_ENABLE:
4237                 ret = btrfs_quota_enable(fs_info);
4238                 break;
4239         case BTRFS_QUOTA_CTL_DISABLE:
4240                 ret = btrfs_quota_disable(fs_info);
4241                 break;
4242         default:
4243                 ret = -EINVAL;
4244                 break;
4245         }
4246
4247         kfree(sa);
4248         up_write(&fs_info->subvol_sem);
4249 drop_write:
4250         mnt_drop_write_file(file);
4251         return ret;
4252 }
4253
4254 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4255 {
4256         struct inode *inode = file_inode(file);
4257         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4258         struct btrfs_root *root = BTRFS_I(inode)->root;
4259         struct btrfs_ioctl_qgroup_assign_args *sa;
4260         struct btrfs_trans_handle *trans;
4261         int ret;
4262         int err;
4263
4264         if (!capable(CAP_SYS_ADMIN))
4265                 return -EPERM;
4266
4267         ret = mnt_want_write_file(file);
4268         if (ret)
4269                 return ret;
4270
4271         sa = memdup_user(arg, sizeof(*sa));
4272         if (IS_ERR(sa)) {
4273                 ret = PTR_ERR(sa);
4274                 goto drop_write;
4275         }
4276
4277         trans = btrfs_join_transaction(root);
4278         if (IS_ERR(trans)) {
4279                 ret = PTR_ERR(trans);
4280                 goto out;
4281         }
4282
4283         if (sa->assign) {
4284                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4285         } else {
4286                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4287         }
4288
4289         /* update qgroup status and info */
4290         err = btrfs_run_qgroups(trans);
4291         if (err < 0)
4292                 btrfs_handle_fs_error(fs_info, err,
4293                                       "failed to update qgroup status and info");
4294         err = btrfs_end_transaction(trans);
4295         if (err && !ret)
4296                 ret = err;
4297
4298 out:
4299         kfree(sa);
4300 drop_write:
4301         mnt_drop_write_file(file);
4302         return ret;
4303 }
4304
4305 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4306 {
4307         struct inode *inode = file_inode(file);
4308         struct btrfs_root *root = BTRFS_I(inode)->root;
4309         struct btrfs_ioctl_qgroup_create_args *sa;
4310         struct btrfs_trans_handle *trans;
4311         int ret;
4312         int err;
4313
4314         if (!capable(CAP_SYS_ADMIN))
4315                 return -EPERM;
4316
4317         ret = mnt_want_write_file(file);
4318         if (ret)
4319                 return ret;
4320
4321         sa = memdup_user(arg, sizeof(*sa));
4322         if (IS_ERR(sa)) {
4323                 ret = PTR_ERR(sa);
4324                 goto drop_write;
4325         }
4326
4327         if (!sa->qgroupid) {
4328                 ret = -EINVAL;
4329                 goto out;
4330         }
4331
4332         trans = btrfs_join_transaction(root);
4333         if (IS_ERR(trans)) {
4334                 ret = PTR_ERR(trans);
4335                 goto out;
4336         }
4337
4338         if (sa->create) {
4339                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4340         } else {
4341                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4342         }
4343
4344         err = btrfs_end_transaction(trans);
4345         if (err && !ret)
4346                 ret = err;
4347
4348 out:
4349         kfree(sa);
4350 drop_write:
4351         mnt_drop_write_file(file);
4352         return ret;
4353 }
4354
4355 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4356 {
4357         struct inode *inode = file_inode(file);
4358         struct btrfs_root *root = BTRFS_I(inode)->root;
4359         struct btrfs_ioctl_qgroup_limit_args *sa;
4360         struct btrfs_trans_handle *trans;
4361         int ret;
4362         int err;
4363         u64 qgroupid;
4364
4365         if (!capable(CAP_SYS_ADMIN))
4366                 return -EPERM;
4367
4368         ret = mnt_want_write_file(file);
4369         if (ret)
4370                 return ret;
4371
4372         sa = memdup_user(arg, sizeof(*sa));
4373         if (IS_ERR(sa)) {
4374                 ret = PTR_ERR(sa);
4375                 goto drop_write;
4376         }
4377
4378         trans = btrfs_join_transaction(root);
4379         if (IS_ERR(trans)) {
4380                 ret = PTR_ERR(trans);
4381                 goto out;
4382         }
4383
4384         qgroupid = sa->qgroupid;
4385         if (!qgroupid) {
4386                 /* take the current subvol as qgroup */
4387                 qgroupid = root->root_key.objectid;
4388         }
4389
4390         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4391
4392         err = btrfs_end_transaction(trans);
4393         if (err && !ret)
4394                 ret = err;
4395
4396 out:
4397         kfree(sa);
4398 drop_write:
4399         mnt_drop_write_file(file);
4400         return ret;
4401 }
4402
4403 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4404 {
4405         struct inode *inode = file_inode(file);
4406         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4407         struct btrfs_ioctl_quota_rescan_args *qsa;
4408         int ret;
4409
4410         if (!capable(CAP_SYS_ADMIN))
4411                 return -EPERM;
4412
4413         ret = mnt_want_write_file(file);
4414         if (ret)
4415                 return ret;
4416
4417         qsa = memdup_user(arg, sizeof(*qsa));
4418         if (IS_ERR(qsa)) {
4419                 ret = PTR_ERR(qsa);
4420                 goto drop_write;
4421         }
4422
4423         if (qsa->flags) {
4424                 ret = -EINVAL;
4425                 goto out;
4426         }
4427
4428         ret = btrfs_qgroup_rescan(fs_info);
4429
4430 out:
4431         kfree(qsa);
4432 drop_write:
4433         mnt_drop_write_file(file);
4434         return ret;
4435 }
4436
4437 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4438                                                 void __user *arg)
4439 {
4440         struct btrfs_ioctl_quota_rescan_args *qsa;
4441         int ret = 0;
4442
4443         if (!capable(CAP_SYS_ADMIN))
4444                 return -EPERM;
4445
4446         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4447         if (!qsa)
4448                 return -ENOMEM;
4449
4450         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4451                 qsa->flags = 1;
4452                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4453         }
4454
4455         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4456                 ret = -EFAULT;
4457
4458         kfree(qsa);
4459         return ret;
4460 }
4461
4462 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4463                                                 void __user *arg)
4464 {
4465         if (!capable(CAP_SYS_ADMIN))
4466                 return -EPERM;
4467
4468         return btrfs_qgroup_wait_for_completion(fs_info, true);
4469 }
4470
4471 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4472                                             struct btrfs_ioctl_received_subvol_args *sa)
4473 {
4474         struct inode *inode = file_inode(file);
4475         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4476         struct btrfs_root *root = BTRFS_I(inode)->root;
4477         struct btrfs_root_item *root_item = &root->root_item;
4478         struct btrfs_trans_handle *trans;
4479         struct timespec64 ct = current_time(inode);
4480         int ret = 0;
4481         int received_uuid_changed;
4482
4483         if (!inode_owner_or_capable(&init_user_ns, inode))
4484                 return -EPERM;
4485
4486         ret = mnt_want_write_file(file);
4487         if (ret < 0)
4488                 return ret;
4489
4490         down_write(&fs_info->subvol_sem);
4491
4492         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4493                 ret = -EINVAL;
4494                 goto out;
4495         }
4496
4497         if (btrfs_root_readonly(root)) {
4498                 ret = -EROFS;
4499                 goto out;
4500         }
4501
4502         /*
4503          * 1 - root item
4504          * 2 - uuid items (received uuid + subvol uuid)
4505          */
4506         trans = btrfs_start_transaction(root, 3);
4507         if (IS_ERR(trans)) {
4508                 ret = PTR_ERR(trans);
4509                 trans = NULL;
4510                 goto out;
4511         }
4512
4513         sa->rtransid = trans->transid;
4514         sa->rtime.sec = ct.tv_sec;
4515         sa->rtime.nsec = ct.tv_nsec;
4516
4517         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4518                                        BTRFS_UUID_SIZE);
4519         if (received_uuid_changed &&
4520             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4521                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4522                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4523                                           root->root_key.objectid);
4524                 if (ret && ret != -ENOENT) {
4525                         btrfs_abort_transaction(trans, ret);
4526                         btrfs_end_transaction(trans);
4527                         goto out;
4528                 }
4529         }
4530         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4531         btrfs_set_root_stransid(root_item, sa->stransid);
4532         btrfs_set_root_rtransid(root_item, sa->rtransid);
4533         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4534         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4535         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4536         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4537
4538         ret = btrfs_update_root(trans, fs_info->tree_root,
4539                                 &root->root_key, &root->root_item);
4540         if (ret < 0) {
4541                 btrfs_end_transaction(trans);
4542                 goto out;
4543         }
4544         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4545                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4546                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4547                                           root->root_key.objectid);
4548                 if (ret < 0 && ret != -EEXIST) {
4549                         btrfs_abort_transaction(trans, ret);
4550                         btrfs_end_transaction(trans);
4551                         goto out;
4552                 }
4553         }
4554         ret = btrfs_commit_transaction(trans);
4555 out:
4556         up_write(&fs_info->subvol_sem);
4557         mnt_drop_write_file(file);
4558         return ret;
4559 }
4560
4561 #ifdef CONFIG_64BIT
4562 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4563                                                 void __user *arg)
4564 {
4565         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4566         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4567         int ret = 0;
4568
4569         args32 = memdup_user(arg, sizeof(*args32));
4570         if (IS_ERR(args32))
4571                 return PTR_ERR(args32);
4572
4573         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4574         if (!args64) {
4575                 ret = -ENOMEM;
4576                 goto out;
4577         }
4578
4579         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4580         args64->stransid = args32->stransid;
4581         args64->rtransid = args32->rtransid;
4582         args64->stime.sec = args32->stime.sec;
4583         args64->stime.nsec = args32->stime.nsec;
4584         args64->rtime.sec = args32->rtime.sec;
4585         args64->rtime.nsec = args32->rtime.nsec;
4586         args64->flags = args32->flags;
4587
4588         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4589         if (ret)
4590                 goto out;
4591
4592         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4593         args32->stransid = args64->stransid;
4594         args32->rtransid = args64->rtransid;
4595         args32->stime.sec = args64->stime.sec;
4596         args32->stime.nsec = args64->stime.nsec;
4597         args32->rtime.sec = args64->rtime.sec;
4598         args32->rtime.nsec = args64->rtime.nsec;
4599         args32->flags = args64->flags;
4600
4601         ret = copy_to_user(arg, args32, sizeof(*args32));
4602         if (ret)
4603                 ret = -EFAULT;
4604
4605 out:
4606         kfree(args32);
4607         kfree(args64);
4608         return ret;
4609 }
4610 #endif
4611
4612 static long btrfs_ioctl_set_received_subvol(struct file *file,
4613                                             void __user *arg)
4614 {
4615         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4616         int ret = 0;
4617
4618         sa = memdup_user(arg, sizeof(*sa));
4619         if (IS_ERR(sa))
4620                 return PTR_ERR(sa);
4621
4622         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4623
4624         if (ret)
4625                 goto out;
4626
4627         ret = copy_to_user(arg, sa, sizeof(*sa));
4628         if (ret)
4629                 ret = -EFAULT;
4630
4631 out:
4632         kfree(sa);
4633         return ret;
4634 }
4635
4636 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4637                                         void __user *arg)
4638 {
4639         size_t len;
4640         int ret;
4641         char label[BTRFS_LABEL_SIZE];
4642
4643         spin_lock(&fs_info->super_lock);
4644         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4645         spin_unlock(&fs_info->super_lock);
4646
4647         len = strnlen(label, BTRFS_LABEL_SIZE);
4648
4649         if (len == BTRFS_LABEL_SIZE) {
4650                 btrfs_warn(fs_info,
4651                            "label is too long, return the first %zu bytes",
4652                            --len);
4653         }
4654
4655         ret = copy_to_user(arg, label, len);
4656
4657         return ret ? -EFAULT : 0;
4658 }
4659
4660 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4661 {
4662         struct inode *inode = file_inode(file);
4663         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4664         struct btrfs_root *root = BTRFS_I(inode)->root;
4665         struct btrfs_super_block *super_block = fs_info->super_copy;
4666         struct btrfs_trans_handle *trans;
4667         char label[BTRFS_LABEL_SIZE];
4668         int ret;
4669
4670         if (!capable(CAP_SYS_ADMIN))
4671                 return -EPERM;
4672
4673         if (copy_from_user(label, arg, sizeof(label)))
4674                 return -EFAULT;
4675
4676         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4677                 btrfs_err(fs_info,
4678                           "unable to set label with more than %d bytes",
4679                           BTRFS_LABEL_SIZE - 1);
4680                 return -EINVAL;
4681         }
4682
4683         ret = mnt_want_write_file(file);
4684         if (ret)
4685                 return ret;
4686
4687         trans = btrfs_start_transaction(root, 0);
4688         if (IS_ERR(trans)) {
4689                 ret = PTR_ERR(trans);
4690                 goto out_unlock;
4691         }
4692
4693         spin_lock(&fs_info->super_lock);
4694         strcpy(super_block->label, label);
4695         spin_unlock(&fs_info->super_lock);
4696         ret = btrfs_commit_transaction(trans);
4697
4698 out_unlock:
4699         mnt_drop_write_file(file);
4700         return ret;
4701 }
4702
4703 #define INIT_FEATURE_FLAGS(suffix) \
4704         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4705           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4706           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4707
4708 int btrfs_ioctl_get_supported_features(void __user *arg)
4709 {
4710         static const struct btrfs_ioctl_feature_flags features[3] = {
4711                 INIT_FEATURE_FLAGS(SUPP),
4712                 INIT_FEATURE_FLAGS(SAFE_SET),
4713                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4714         };
4715
4716         if (copy_to_user(arg, &features, sizeof(features)))
4717                 return -EFAULT;
4718
4719         return 0;
4720 }
4721
4722 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4723                                         void __user *arg)
4724 {
4725         struct btrfs_super_block *super_block = fs_info->super_copy;
4726         struct btrfs_ioctl_feature_flags features;
4727
4728         features.compat_flags = btrfs_super_compat_flags(super_block);
4729         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4730         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4731
4732         if (copy_to_user(arg, &features, sizeof(features)))
4733                 return -EFAULT;
4734
4735         return 0;
4736 }
4737
4738 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4739                               enum btrfs_feature_set set,
4740                               u64 change_mask, u64 flags, u64 supported_flags,
4741                               u64 safe_set, u64 safe_clear)
4742 {
4743         const char *type = btrfs_feature_set_name(set);
4744         char *names;
4745         u64 disallowed, unsupported;
4746         u64 set_mask = flags & change_mask;
4747         u64 clear_mask = ~flags & change_mask;
4748
4749         unsupported = set_mask & ~supported_flags;
4750         if (unsupported) {
4751                 names = btrfs_printable_features(set, unsupported);
4752                 if (names) {
4753                         btrfs_warn(fs_info,
4754                                    "this kernel does not support the %s feature bit%s",
4755                                    names, strchr(names, ',') ? "s" : "");
4756                         kfree(names);
4757                 } else
4758                         btrfs_warn(fs_info,
4759                                    "this kernel does not support %s bits 0x%llx",
4760                                    type, unsupported);
4761                 return -EOPNOTSUPP;
4762         }
4763
4764         disallowed = set_mask & ~safe_set;
4765         if (disallowed) {
4766                 names = btrfs_printable_features(set, disallowed);
4767                 if (names) {
4768                         btrfs_warn(fs_info,
4769                                    "can't set the %s feature bit%s while mounted",
4770                                    names, strchr(names, ',') ? "s" : "");
4771                         kfree(names);
4772                 } else
4773                         btrfs_warn(fs_info,
4774                                    "can't set %s bits 0x%llx while mounted",
4775                                    type, disallowed);
4776                 return -EPERM;
4777         }
4778
4779         disallowed = clear_mask & ~safe_clear;
4780         if (disallowed) {
4781                 names = btrfs_printable_features(set, disallowed);
4782                 if (names) {
4783                         btrfs_warn(fs_info,
4784                                    "can't clear the %s feature bit%s while mounted",
4785                                    names, strchr(names, ',') ? "s" : "");
4786                         kfree(names);
4787                 } else
4788                         btrfs_warn(fs_info,
4789                                    "can't clear %s bits 0x%llx while mounted",
4790                                    type, disallowed);
4791                 return -EPERM;
4792         }
4793
4794         return 0;
4795 }
4796
4797 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4798 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4799                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4800                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4801                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4802
4803 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4804 {
4805         struct inode *inode = file_inode(file);
4806         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4807         struct btrfs_root *root = BTRFS_I(inode)->root;
4808         struct btrfs_super_block *super_block = fs_info->super_copy;
4809         struct btrfs_ioctl_feature_flags flags[2];
4810         struct btrfs_trans_handle *trans;
4811         u64 newflags;
4812         int ret;
4813
4814         if (!capable(CAP_SYS_ADMIN))
4815                 return -EPERM;
4816
4817         if (copy_from_user(flags, arg, sizeof(flags)))
4818                 return -EFAULT;
4819
4820         /* Nothing to do */
4821         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4822             !flags[0].incompat_flags)
4823                 return 0;
4824
4825         ret = check_feature(fs_info, flags[0].compat_flags,
4826                             flags[1].compat_flags, COMPAT);
4827         if (ret)
4828                 return ret;
4829
4830         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4831                             flags[1].compat_ro_flags, COMPAT_RO);
4832         if (ret)
4833                 return ret;
4834
4835         ret = check_feature(fs_info, flags[0].incompat_flags,
4836                             flags[1].incompat_flags, INCOMPAT);
4837         if (ret)
4838                 return ret;
4839
4840         ret = mnt_want_write_file(file);
4841         if (ret)
4842                 return ret;
4843
4844         trans = btrfs_start_transaction(root, 0);
4845         if (IS_ERR(trans)) {
4846                 ret = PTR_ERR(trans);
4847                 goto out_drop_write;
4848         }
4849
4850         spin_lock(&fs_info->super_lock);
4851         newflags = btrfs_super_compat_flags(super_block);
4852         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4853         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4854         btrfs_set_super_compat_flags(super_block, newflags);
4855
4856         newflags = btrfs_super_compat_ro_flags(super_block);
4857         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4858         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4859         btrfs_set_super_compat_ro_flags(super_block, newflags);
4860
4861         newflags = btrfs_super_incompat_flags(super_block);
4862         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4863         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4864         btrfs_set_super_incompat_flags(super_block, newflags);
4865         spin_unlock(&fs_info->super_lock);
4866
4867         ret = btrfs_commit_transaction(trans);
4868 out_drop_write:
4869         mnt_drop_write_file(file);
4870
4871         return ret;
4872 }
4873
4874 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4875 {
4876         struct btrfs_ioctl_send_args *arg;
4877         int ret;
4878
4879         if (compat) {
4880 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4881                 struct btrfs_ioctl_send_args_32 args32;
4882
4883                 ret = copy_from_user(&args32, argp, sizeof(args32));
4884                 if (ret)
4885                         return -EFAULT;
4886                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4887                 if (!arg)
4888                         return -ENOMEM;
4889                 arg->send_fd = args32.send_fd;
4890                 arg->clone_sources_count = args32.clone_sources_count;
4891                 arg->clone_sources = compat_ptr(args32.clone_sources);
4892                 arg->parent_root = args32.parent_root;
4893                 arg->flags = args32.flags;
4894                 memcpy(arg->reserved, args32.reserved,
4895                        sizeof(args32.reserved));
4896 #else
4897                 return -ENOTTY;
4898 #endif
4899         } else {
4900                 arg = memdup_user(argp, sizeof(*arg));
4901                 if (IS_ERR(arg))
4902                         return PTR_ERR(arg);
4903         }
4904         ret = btrfs_ioctl_send(file, arg);
4905         kfree(arg);
4906         return ret;
4907 }
4908
4909 long btrfs_ioctl(struct file *file, unsigned int
4910                 cmd, unsigned long arg)
4911 {
4912         struct inode *inode = file_inode(file);
4913         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4914         struct btrfs_root *root = BTRFS_I(inode)->root;
4915         void __user *argp = (void __user *)arg;
4916
4917         switch (cmd) {
4918         case FS_IOC_GETFLAGS:
4919                 return btrfs_ioctl_getflags(file, argp);
4920         case FS_IOC_SETFLAGS:
4921                 return btrfs_ioctl_setflags(file, argp);
4922         case FS_IOC_GETVERSION:
4923                 return btrfs_ioctl_getversion(file, argp);
4924         case FS_IOC_GETFSLABEL:
4925                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4926         case FS_IOC_SETFSLABEL:
4927                 return btrfs_ioctl_set_fslabel(file, argp);
4928         case FITRIM:
4929                 return btrfs_ioctl_fitrim(fs_info, argp);
4930         case BTRFS_IOC_SNAP_CREATE:
4931                 return btrfs_ioctl_snap_create(file, argp, 0);
4932         case BTRFS_IOC_SNAP_CREATE_V2:
4933                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4934         case BTRFS_IOC_SUBVOL_CREATE:
4935                 return btrfs_ioctl_snap_create(file, argp, 1);
4936         case BTRFS_IOC_SUBVOL_CREATE_V2:
4937                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4938         case BTRFS_IOC_SNAP_DESTROY:
4939                 return btrfs_ioctl_snap_destroy(file, argp, false);
4940         case BTRFS_IOC_SNAP_DESTROY_V2:
4941                 return btrfs_ioctl_snap_destroy(file, argp, true);
4942         case BTRFS_IOC_SUBVOL_GETFLAGS:
4943                 return btrfs_ioctl_subvol_getflags(file, argp);
4944         case BTRFS_IOC_SUBVOL_SETFLAGS:
4945                 return btrfs_ioctl_subvol_setflags(file, argp);
4946         case BTRFS_IOC_DEFAULT_SUBVOL:
4947                 return btrfs_ioctl_default_subvol(file, argp);
4948         case BTRFS_IOC_DEFRAG:
4949                 return btrfs_ioctl_defrag(file, NULL);
4950         case BTRFS_IOC_DEFRAG_RANGE:
4951                 return btrfs_ioctl_defrag(file, argp);
4952         case BTRFS_IOC_RESIZE:
4953                 return btrfs_ioctl_resize(file, argp);
4954         case BTRFS_IOC_ADD_DEV:
4955                 return btrfs_ioctl_add_dev(fs_info, argp);
4956         case BTRFS_IOC_RM_DEV:
4957                 return btrfs_ioctl_rm_dev(file, argp);
4958         case BTRFS_IOC_RM_DEV_V2:
4959                 return btrfs_ioctl_rm_dev_v2(file, argp);
4960         case BTRFS_IOC_FS_INFO:
4961                 return btrfs_ioctl_fs_info(fs_info, argp);
4962         case BTRFS_IOC_DEV_INFO:
4963                 return btrfs_ioctl_dev_info(fs_info, argp);
4964         case BTRFS_IOC_BALANCE:
4965                 return btrfs_ioctl_balance(file, NULL);
4966         case BTRFS_IOC_TREE_SEARCH:
4967                 return btrfs_ioctl_tree_search(file, argp);
4968         case BTRFS_IOC_TREE_SEARCH_V2:
4969                 return btrfs_ioctl_tree_search_v2(file, argp);
4970         case BTRFS_IOC_INO_LOOKUP:
4971                 return btrfs_ioctl_ino_lookup(file, argp);
4972         case BTRFS_IOC_INO_PATHS:
4973                 return btrfs_ioctl_ino_to_path(root, argp);
4974         case BTRFS_IOC_LOGICAL_INO:
4975                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4976         case BTRFS_IOC_LOGICAL_INO_V2:
4977                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4978         case BTRFS_IOC_SPACE_INFO:
4979                 return btrfs_ioctl_space_info(fs_info, argp);
4980         case BTRFS_IOC_SYNC: {
4981                 int ret;
4982
4983                 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
4984                 if (ret)
4985                         return ret;
4986                 ret = btrfs_sync_fs(inode->i_sb, 1);
4987                 /*
4988                  * The transaction thread may want to do more work,
4989                  * namely it pokes the cleaner kthread that will start
4990                  * processing uncleaned subvols.
4991                  */
4992                 wake_up_process(fs_info->transaction_kthread);
4993                 return ret;
4994         }
4995         case BTRFS_IOC_START_SYNC:
4996                 return btrfs_ioctl_start_sync(root, argp);
4997         case BTRFS_IOC_WAIT_SYNC:
4998                 return btrfs_ioctl_wait_sync(fs_info, argp);
4999         case BTRFS_IOC_SCRUB:
5000                 return btrfs_ioctl_scrub(file, argp);
5001         case BTRFS_IOC_SCRUB_CANCEL:
5002                 return btrfs_ioctl_scrub_cancel(fs_info);
5003         case BTRFS_IOC_SCRUB_PROGRESS:
5004                 return btrfs_ioctl_scrub_progress(fs_info, argp);
5005         case BTRFS_IOC_BALANCE_V2:
5006                 return btrfs_ioctl_balance(file, argp);
5007         case BTRFS_IOC_BALANCE_CTL:
5008                 return btrfs_ioctl_balance_ctl(fs_info, arg);
5009         case BTRFS_IOC_BALANCE_PROGRESS:
5010                 return btrfs_ioctl_balance_progress(fs_info, argp);
5011         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
5012                 return btrfs_ioctl_set_received_subvol(file, argp);
5013 #ifdef CONFIG_64BIT
5014         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
5015                 return btrfs_ioctl_set_received_subvol_32(file, argp);
5016 #endif
5017         case BTRFS_IOC_SEND:
5018                 return _btrfs_ioctl_send(file, argp, false);
5019 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
5020         case BTRFS_IOC_SEND_32:
5021                 return _btrfs_ioctl_send(file, argp, true);
5022 #endif
5023         case BTRFS_IOC_GET_DEV_STATS:
5024                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
5025         case BTRFS_IOC_QUOTA_CTL:
5026                 return btrfs_ioctl_quota_ctl(file, argp);
5027         case BTRFS_IOC_QGROUP_ASSIGN:
5028                 return btrfs_ioctl_qgroup_assign(file, argp);
5029         case BTRFS_IOC_QGROUP_CREATE:
5030                 return btrfs_ioctl_qgroup_create(file, argp);
5031         case BTRFS_IOC_QGROUP_LIMIT:
5032                 return btrfs_ioctl_qgroup_limit(file, argp);
5033         case BTRFS_IOC_QUOTA_RESCAN:
5034                 return btrfs_ioctl_quota_rescan(file, argp);
5035         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
5036                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
5037         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
5038                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
5039         case BTRFS_IOC_DEV_REPLACE:
5040                 return btrfs_ioctl_dev_replace(fs_info, argp);
5041         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
5042                 return btrfs_ioctl_get_supported_features(argp);
5043         case BTRFS_IOC_GET_FEATURES:
5044                 return btrfs_ioctl_get_features(fs_info, argp);
5045         case BTRFS_IOC_SET_FEATURES:
5046                 return btrfs_ioctl_set_features(file, argp);
5047         case FS_IOC_FSGETXATTR:
5048                 return btrfs_ioctl_fsgetxattr(file, argp);
5049         case FS_IOC_FSSETXATTR:
5050                 return btrfs_ioctl_fssetxattr(file, argp);
5051         case BTRFS_IOC_GET_SUBVOL_INFO:
5052                 return btrfs_ioctl_get_subvol_info(file, argp);
5053         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
5054                 return btrfs_ioctl_get_subvol_rootref(file, argp);
5055         case BTRFS_IOC_INO_LOOKUP_USER:
5056                 return btrfs_ioctl_ino_lookup_user(file, argp);
5057         }
5058
5059         return -ENOTTY;
5060 }
5061
5062 #ifdef CONFIG_COMPAT
5063 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
5064 {
5065         /*
5066          * These all access 32-bit values anyway so no further
5067          * handling is necessary.
5068          */
5069         switch (cmd) {
5070         case FS_IOC32_GETFLAGS:
5071                 cmd = FS_IOC_GETFLAGS;
5072                 break;
5073         case FS_IOC32_SETFLAGS:
5074                 cmd = FS_IOC_SETFLAGS;
5075                 break;
5076         case FS_IOC32_GETVERSION:
5077                 cmd = FS_IOC_GETVERSION;
5078                 break;
5079         }
5080
5081         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
5082 }
5083 #endif