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