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