fs/btrfs/ioctl.c

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