fs/btrfs/ioctl.c

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