fs/btrfs/ioctl.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/highmem.h>
  27 #include <linux/time.h>
  28 #include <linux/init.h>
  29 #include <linux/string.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mount.h>
  32 #include <linux/mpage.h>
  33 #include <linux/namei.h>
  34 #include <linux/swap.h>
  35 #include <linux/writeback.h>
  36 #include <linux/statfs.h>
  37 #include <linux/compat.h>
  38 #include <linux/bit_spinlock.h>
  39 #include <linux/security.h>
  40 #include <linux/xattr.h>
  41 #include <linux/vmalloc.h>
  42 #include <linux/slab.h>
  43 #include <linux/blkdev.h>
  44 #include "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "ioctl.h"
  50 #include "print-tree.h"
  51 #include "volumes.h"
  52 #include "locking.h"
  53 #include "inode-map.h"
  54 #include "backref.h"
  55 #include "rcu-string.h"
  56
  57 /* Mask out flags that are inappropriate for the given type of inode. */
  58 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
  59 {
  60         if (S_ISDIR(mode))
  61                 return flags;
  62         else if (S_ISREG(mode))
  63                 return flags & ~FS_DIRSYNC_FL;
  64         else
  65                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
  66 }
  67
  68 /*
  69  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
  70  */
  71 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
  72 {
  73         unsigned int iflags = 0;
  74
  75         if (flags & BTRFS_INODE_SYNC)
  76                 iflags |= FS_SYNC_FL;
  77         if (flags & BTRFS_INODE_IMMUTABLE)
  78                 iflags |= FS_IMMUTABLE_FL;
  79         if (flags & BTRFS_INODE_APPEND)
  80                 iflags |= FS_APPEND_FL;
  81         if (flags & BTRFS_INODE_NODUMP)
  82                 iflags |= FS_NODUMP_FL;
  83         if (flags & BTRFS_INODE_NOATIME)
  84                 iflags |= FS_NOATIME_FL;
  85         if (flags & BTRFS_INODE_DIRSYNC)
  86                 iflags |= FS_DIRSYNC_FL;
  87         if (flags & BTRFS_INODE_NODATACOW)
  88                 iflags |= FS_NOCOW_FL;
  89
  90         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
  91                 iflags |= FS_COMPR_FL;
  92         else if (flags & BTRFS_INODE_NOCOMPRESS)
  93                 iflags |= FS_NOCOMP_FL;
  94
  95         return iflags;
  96 }
  97
  98 /*
  99  * Update inode->i_flags based on the btrfs internal flags.
 100  */
 101 void btrfs_update_iflags(struct inode *inode)
 102 {
 103         struct btrfs_inode *ip = BTRFS_I(inode);
 104
 105         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
 106
 107         if (ip->flags & BTRFS_INODE_SYNC)
 108                 inode->i_flags |= S_SYNC;
 109         if (ip->flags & BTRFS_INODE_IMMUTABLE)
 110                 inode->i_flags |= S_IMMUTABLE;
 111         if (ip->flags & BTRFS_INODE_APPEND)
 112                 inode->i_flags |= S_APPEND;
 113         if (ip->flags & BTRFS_INODE_NOATIME)
 114                 inode->i_flags |= S_NOATIME;
 115         if (ip->flags & BTRFS_INODE_DIRSYNC)
 116                 inode->i_flags |= S_DIRSYNC;
 117 }
 118
 119 /*
 120  * Inherit flags from the parent inode.
 121  *
 122  * Currently only the compression flags and the cow flags are inherited.
 123  */
 124 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
 125 {
 126         unsigned int flags;
 127
 128         if (!dir)
 129                 return;
 130
 131         flags = BTRFS_I(dir)->flags;
 132
 133         if (flags & BTRFS_INODE_NOCOMPRESS) {
 134                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
 135                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 136         } else if (flags & BTRFS_INODE_COMPRESS) {
 137                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
 138                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
 139         }
 140
 141         if (flags & BTRFS_INODE_NODATACOW)
 142                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
 143
 144         btrfs_update_iflags(inode);
 145 }
 146
 147 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 148 {
 149         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
 150         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
 151
 152         if (copy_to_user(arg, &flags, sizeof(flags)))
 153                 return -EFAULT;
 154         return 0;
 155 }
 156
 157 static int check_flags(unsigned int flags)
 158 {
 159         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 160                       FS_NOATIME_FL | FS_NODUMP_FL | \
 161                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 162                       FS_NOCOMP_FL | FS_COMPR_FL |
 163                       FS_NOCOW_FL))
 164                 return -EOPNOTSUPP;
 165
 166         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 167                 return -EINVAL;
 168
 169         return 0;
 170 }
 171
 172 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 173 {
 174         struct inode *inode = file->f_path.dentry->d_inode;
 175         struct btrfs_inode *ip = BTRFS_I(inode);
 176         struct btrfs_root *root = ip->root;
 177         struct btrfs_trans_handle *trans;
 178         unsigned int flags, oldflags;
 179         int ret;
 180         u64 ip_oldflags;
 181         unsigned int i_oldflags;
 182
 183         if (btrfs_root_readonly(root))
 184                 return -EROFS;
 185
 186         if (copy_from_user(&flags, arg, sizeof(flags)))
 187                 return -EFAULT;
 188
 189         ret = check_flags(flags);
 190         if (ret)
 191                 return ret;
 192
 193         if (!inode_owner_or_capable(inode))
 194                 return -EACCES;
 195
 196         mutex_lock(&inode->i_mutex);
 197
 198         ip_oldflags = ip->flags;
 199         i_oldflags = inode->i_flags;
 200
 201         flags = btrfs_mask_flags(inode->i_mode, flags);
 202         oldflags = btrfs_flags_to_ioctl(ip->flags);
 203         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
 204                 if (!capable(CAP_LINUX_IMMUTABLE)) {
 205                         ret = -EPERM;
 206                         goto out_unlock;
 207                 }
 208         }
 209
 210         ret = mnt_want_write_file(file);
 211         if (ret)
 212                 goto out_unlock;
 213
 214         if (flags & FS_SYNC_FL)
 215                 ip->flags |= BTRFS_INODE_SYNC;
 216         else
 217                 ip->flags &= ~BTRFS_INODE_SYNC;
 218         if (flags & FS_IMMUTABLE_FL)
 219                 ip->flags |= BTRFS_INODE_IMMUTABLE;
 220         else
 221                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
 222         if (flags & FS_APPEND_FL)
 223                 ip->flags |= BTRFS_INODE_APPEND;
 224         else
 225                 ip->flags &= ~BTRFS_INODE_APPEND;
 226         if (flags & FS_NODUMP_FL)
 227                 ip->flags |= BTRFS_INODE_NODUMP;
 228         else
 229                 ip->flags &= ~BTRFS_INODE_NODUMP;
 230         if (flags & FS_NOATIME_FL)
 231                 ip->flags |= BTRFS_INODE_NOATIME;
 232         else
 233                 ip->flags &= ~BTRFS_INODE_NOATIME;
 234         if (flags & FS_DIRSYNC_FL)
 235                 ip->flags |= BTRFS_INODE_DIRSYNC;
 236         else
 237                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
 238         if (flags & FS_NOCOW_FL)
 239                 ip->flags |= BTRFS_INODE_NODATACOW;
 240         else
 241                 ip->flags &= ~BTRFS_INODE_NODATACOW;
 242
 243         /*
 244          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 245          * flag may be changed automatically if compression code won't make
 246          * things smaller.
 247          */
 248         if (flags & FS_NOCOMP_FL) {
 249                 ip->flags &= ~BTRFS_INODE_COMPRESS;
 250                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
 251         } else if (flags & FS_COMPR_FL) {
 252                 ip->flags |= BTRFS_INODE_COMPRESS;
 253                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
 254         } else {
 255                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 256         }
 257
 258         trans = btrfs_start_transaction(root, 1);
 259         if (IS_ERR(trans)) {
 260                 ret = PTR_ERR(trans);
 261                 goto out_drop;
 262         }
 263
 264         btrfs_update_iflags(inode);
 265         inode_inc_iversion(inode);
 266         inode->i_ctime = CURRENT_TIME;
 267         ret = btrfs_update_inode(trans, root, inode);
 268
 269         btrfs_end_transaction(trans, root);
 270  out_drop:
 271         if (ret) {
 272                 ip->flags = ip_oldflags;
 273                 inode->i_flags = i_oldflags;
 274         }
 275
 276         mnt_drop_write_file(file);
 277  out_unlock:
 278         mutex_unlock(&inode->i_mutex);
 279         return ret;
 280 }
 281
 282 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 283 {
 284         struct inode *inode = file->f_path.dentry->d_inode;
 285
 286         return put_user(inode->i_generation, arg);
 287 }
 288
 289 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
 290 {
 291         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
 292         struct btrfs_device *device;
 293         struct request_queue *q;
 294         struct fstrim_range range;
 295         u64 minlen = ULLONG_MAX;
 296         u64 num_devices = 0;
 297         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
 298         int ret;
 299
 300         if (!capable(CAP_SYS_ADMIN))
 301                 return -EPERM;
 302
 303         rcu_read_lock();
 304         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 305                                 dev_list) {
 306                 if (!device->bdev)
 307                         continue;
 308                 q = bdev_get_queue(device->bdev);
 309                 if (blk_queue_discard(q)) {
 310                         num_devices++;
 311                         minlen = min((u64)q->limits.discard_granularity,
 312                                      minlen);
 313                 }
 314         }
 315         rcu_read_unlock();
 316
 317         if (!num_devices)
 318                 return -EOPNOTSUPP;
 319         if (copy_from_user(&range, arg, sizeof(range)))
 320                 return -EFAULT;
 321         if (range.start > total_bytes)
 322                 return -EINVAL;
 323
 324         range.len = min(range.len, total_bytes - range.start);
 325         range.minlen = max(range.minlen, minlen);
 326         ret = btrfs_trim_fs(fs_info->tree_root, &range);
 327         if (ret < 0)
 328                 return ret;
 329
 330         if (copy_to_user(arg, &range, sizeof(range)))
 331                 return -EFAULT;
 332
 333         return 0;
 334 }
 335
 336 static noinline int create_subvol(struct btrfs_root *root,
 337                                   struct dentry *dentry,
 338                                   char *name, int namelen,
 339                                   u64 *async_transid)
 340 {
 341         struct btrfs_trans_handle *trans;
 342         struct btrfs_key key;
 343         struct btrfs_root_item root_item;
 344         struct btrfs_inode_item *inode_item;
 345         struct extent_buffer *leaf;
 346         struct btrfs_root *new_root;
 347         struct dentry *parent = dentry->d_parent;
 348         struct inode *dir;
 349         int ret;
 350         int err;
 351         u64 objectid;
 352         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 353         u64 index = 0;
 354
 355         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
 356         if (ret)
 357                 return ret;
 358
 359         dir = parent->d_inode;
 360
 361         /*
 362          * 1 - inode item
 363          * 2 - refs
 364          * 1 - root item
 365          * 2 - dir items
 366          */
 367         trans = btrfs_start_transaction(root, 6);
 368         if (IS_ERR(trans))
 369                 return PTR_ERR(trans);
 370
 371         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
 372                                       0, objectid, NULL, 0, 0, 0);
 373         if (IS_ERR(leaf)) {
 374                 ret = PTR_ERR(leaf);
 375                 goto fail;
 376         }
 377
 378         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
 379         btrfs_set_header_bytenr(leaf, leaf->start);
 380         btrfs_set_header_generation(leaf, trans->transid);
 381         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
 382         btrfs_set_header_owner(leaf, objectid);
 383
 384         write_extent_buffer(leaf, root->fs_info->fsid,
 385                             (unsigned long)btrfs_header_fsid(leaf),
 386                             BTRFS_FSID_SIZE);
 387         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
 388                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
 389                             BTRFS_UUID_SIZE);
 390         btrfs_mark_buffer_dirty(leaf);
 391
 392         inode_item = &root_item.inode;
 393         memset(inode_item, 0, sizeof(*inode_item));
 394         inode_item->generation = cpu_to_le64(1);
 395         inode_item->size = cpu_to_le64(3);
 396         inode_item->nlink = cpu_to_le32(1);
 397         inode_item->nbytes = cpu_to_le64(root->leafsize);
 398         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
 399
 400         root_item.flags = 0;
 401         root_item.byte_limit = 0;
 402         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
 403
 404         btrfs_set_root_bytenr(&root_item, leaf->start);
 405         btrfs_set_root_generation(&root_item, trans->transid);
 406         btrfs_set_root_level(&root_item, 0);
 407         btrfs_set_root_refs(&root_item, 1);
 408         btrfs_set_root_used(&root_item, leaf->len);
 409         btrfs_set_root_last_snapshot(&root_item, 0);
 410
 411         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
 412         root_item.drop_level = 0;
 413
 414         btrfs_tree_unlock(leaf);
 415         free_extent_buffer(leaf);
 416         leaf = NULL;
 417
 418         btrfs_set_root_dirid(&root_item, new_dirid);
 419
 420         key.objectid = objectid;
 421         key.offset = 0;
 422         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 423         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 424                                 &root_item);
 425         if (ret)
 426                 goto fail;
 427
 428         key.offset = (u64)-1;
 429         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
 430         if (IS_ERR(new_root)) {
 431                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
 432                 ret = PTR_ERR(new_root);
 433                 goto fail;
 434         }
 435
 436         btrfs_record_root_in_trans(trans, new_root);
 437
 438         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
 439         if (ret) {
 440                 /* We potentially lose an unused inode item here */
 441                 btrfs_abort_transaction(trans, root, ret);
 442                 goto fail;
 443         }
 444
 445         /*
 446          * insert the directory item
 447          */
 448         ret = btrfs_set_inode_index(dir, &index);
 449         if (ret) {
 450                 btrfs_abort_transaction(trans, root, ret);
 451                 goto fail;
 452         }
 453
 454         ret = btrfs_insert_dir_item(trans, root,
 455                                     name, namelen, dir, &key,
 456                                     BTRFS_FT_DIR, index);
 457         if (ret) {
 458                 btrfs_abort_transaction(trans, root, ret);
 459                 goto fail;
 460         }
 461
 462         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
 463         ret = btrfs_update_inode(trans, root, dir);
 464         BUG_ON(ret);
 465
 466         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
 467                                  objectid, root->root_key.objectid,
 468                                  btrfs_ino(dir), index, name, namelen);
 469
 470         BUG_ON(ret);
 471
 472         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
 473 fail:
 474         if (async_transid) {
 475                 *async_transid = trans->transid;
 476                 err = btrfs_commit_transaction_async(trans, root, 1);
 477         } else {
 478                 err = btrfs_commit_transaction(trans, root);
 479         }
 480         if (err && !ret)
 481                 ret = err;
 482         return ret;
 483 }
 484
 485 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
 486                            char *name, int namelen, u64 *async_transid,
 487                            bool readonly)
 488 {
 489         struct inode *inode;
 490         struct btrfs_pending_snapshot *pending_snapshot;
 491         struct btrfs_trans_handle *trans;
 492         int ret;
 493
 494         if (!root->ref_cows)
 495                 return -EINVAL;
 496
 497         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
 498         if (!pending_snapshot)
 499                 return -ENOMEM;
 500
 501         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
 502         pending_snapshot->dentry = dentry;
 503         pending_snapshot->root = root;
 504         pending_snapshot->readonly = readonly;
 505
 506         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
 507         if (IS_ERR(trans)) {
 508                 ret = PTR_ERR(trans);
 509                 goto fail;
 510         }
 511
 512         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
 513         BUG_ON(ret);
 514
 515         spin_lock(&root->fs_info->trans_lock);
 516         list_add(&pending_snapshot->list,
 517                  &trans->transaction->pending_snapshots);
 518         spin_unlock(&root->fs_info->trans_lock);
 519         if (async_transid) {
 520                 *async_transid = trans->transid;
 521                 ret = btrfs_commit_transaction_async(trans,
 522                                      root->fs_info->extent_root, 1);
 523         } else {
 524                 ret = btrfs_commit_transaction(trans,
 525                                                root->fs_info->extent_root);
 526         }
 527         BUG_ON(ret);
 528
 529         ret = pending_snapshot->error;
 530         if (ret)
 531                 goto fail;
 532
 533         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 534         if (ret)
 535                 goto fail;
 536
 537         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
 538         if (IS_ERR(inode)) {
 539                 ret = PTR_ERR(inode);
 540                 goto fail;
 541         }
 542         BUG_ON(!inode);
 543         d_instantiate(dentry, inode);
 544         ret = 0;
 545 fail:
 546         kfree(pending_snapshot);
 547         return ret;
 548 }
 549
 550 /*  copy of check_sticky in fs/namei.c()
 551 * It's inline, so penalty for filesystems that don't use sticky bit is
 552 * minimal.
 553 */
 554 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
 555 {
 556         uid_t fsuid = current_fsuid();
 557
 558         if (!(dir->i_mode & S_ISVTX))
 559                 return 0;
 560         if (inode->i_uid == fsuid)
 561                 return 0;
 562         if (dir->i_uid == fsuid)
 563                 return 0;
 564         return !capable(CAP_FOWNER);
 565 }
 566
 567 /*  copy of may_delete in fs/namei.c()
 568  *      Check whether we can remove a link victim from directory dir, check
 569  *  whether the type of victim is right.
 570  *  1. We can't do it if dir is read-only (done in permission())
 571  *  2. We should have write and exec permissions on dir
 572  *  3. We can't remove anything from append-only dir
 573  *  4. We can't do anything with immutable dir (done in permission())
 574  *  5. If the sticky bit on dir is set we should either
 575  *      a. be owner of dir, or
 576  *      b. be owner of victim, or
 577  *      c. have CAP_FOWNER capability
 578  *  6. If the victim is append-only or immutable we can't do antyhing with
 579  *     links pointing to it.
 580  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 581  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 582  *  9. We can't remove a root or mountpoint.
 583  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 584  *     nfs_async_unlink().
 585  */
 586
 587 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
 588 {
 589         int error;
 590
 591         if (!victim->d_inode)
 592                 return -ENOENT;
 593
 594         BUG_ON(victim->d_parent->d_inode != dir);
 595         audit_inode_child(victim, dir);
 596
 597         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 598         if (error)
 599                 return error;
 600         if (IS_APPEND(dir))
 601                 return -EPERM;
 602         if (btrfs_check_sticky(dir, victim->d_inode)||
 603                 IS_APPEND(victim->d_inode)||
 604             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
 605                 return -EPERM;
 606         if (isdir) {
 607                 if (!S_ISDIR(victim->d_inode->i_mode))
 608                         return -ENOTDIR;
 609                 if (IS_ROOT(victim))
 610                         return -EBUSY;
 611         } else if (S_ISDIR(victim->d_inode->i_mode))
 612                 return -EISDIR;
 613         if (IS_DEADDIR(dir))
 614                 return -ENOENT;
 615         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 616                 return -EBUSY;
 617         return 0;
 618 }
 619
 620 /* copy of may_create in fs/namei.c() */
 621 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 622 {
 623         if (child->d_inode)
 624                 return -EEXIST;
 625         if (IS_DEADDIR(dir))
 626                 return -ENOENT;
 627         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 628 }
 629
 630 /*
 631  * Create a new subvolume below @parent.  This is largely modeled after
 632  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 633  * inside this filesystem so it's quite a bit simpler.
 634  */
 635 static noinline int btrfs_mksubvol(struct path *parent,
 636                                    char *name, int namelen,
 637                                    struct btrfs_root *snap_src,
 638                                    u64 *async_transid, bool readonly)
 639 {
 640         struct inode *dir  = parent->dentry->d_inode;
 641         struct dentry *dentry;
 642         int error;
 643
 644         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
 645
 646         dentry = lookup_one_len(name, parent->dentry, namelen);
 647         error = PTR_ERR(dentry);
 648         if (IS_ERR(dentry))
 649                 goto out_unlock;
 650
 651         error = -EEXIST;
 652         if (dentry->d_inode)
 653                 goto out_dput;
 654
 655         error = btrfs_may_create(dir, dentry);
 656         if (error)
 657                 goto out_dput;
 658
 659         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 660
 661         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 662                 goto out_up_read;
 663
 664         if (snap_src) {
 665                 error = create_snapshot(snap_src, dentry,
 666                                         name, namelen, async_transid, readonly);
 667         } else {
 668                 error = create_subvol(BTRFS_I(dir)->root, dentry,
 669                                       name, namelen, async_transid);
 670         }
 671         if (!error)
 672                 fsnotify_mkdir(dir, dentry);
 673 out_up_read:
 674         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
 675 out_dput:
 676         dput(dentry);
 677 out_unlock:
 678         mutex_unlock(&dir->i_mutex);
 679         return error;
 680 }
 681
 682 /*
 683  * When we're defragging a range, we don't want to kick it off again
 684  * if it is really just waiting for delalloc to send it down.
 685  * If we find a nice big extent or delalloc range for the bytes in the
 686  * file you want to defrag, we return 0 to let you know to skip this
 687  * part of the file
 688  */
 689 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
 690 {
 691         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 692         struct extent_map *em = NULL;
 693         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 694         u64 end;
 695
 696         read_lock(&em_tree->lock);
 697         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 698         read_unlock(&em_tree->lock);
 699
 700         if (em) {
 701                 end = extent_map_end(em);
 702                 free_extent_map(em);
 703                 if (end - offset > thresh)
 704                         return 0;
 705         }
 706         /* if we already have a nice delalloc here, just stop */
 707         thresh /= 2;
 708         end = count_range_bits(io_tree, &offset, offset + thresh,
 709                                thresh, EXTENT_DELALLOC, 1);
 710         if (end >= thresh)
 711                 return 0;
 712         return 1;
 713 }
 714
 715 /*
 716  * helper function to walk through a file and find extents
 717  * newer than a specific transid, and smaller than thresh.
 718  *
 719  * This is used by the defragging code to find new and small
 720  * extents
 721  */
 722 static int find_new_extents(struct btrfs_root *root,
 723                             struct inode *inode, u64 newer_than,
 724                             u64 *off, int thresh)
 725 {
 726         struct btrfs_path *path;
 727         struct btrfs_key min_key;
 728         struct btrfs_key max_key;
 729         struct extent_buffer *leaf;
 730         struct btrfs_file_extent_item *extent;
 731         int type;
 732         int ret;
 733         u64 ino = btrfs_ino(inode);
 734
 735         path = btrfs_alloc_path();
 736         if (!path)
 737                 return -ENOMEM;
 738
 739         min_key.objectid = ino;
 740         min_key.type = BTRFS_EXTENT_DATA_KEY;
 741         min_key.offset = *off;
 742
 743         max_key.objectid = ino;
 744         max_key.type = (u8)-1;
 745         max_key.offset = (u64)-1;
 746
 747         path->keep_locks = 1;
 748
 749         while(1) {
 750                 ret = btrfs_search_forward(root, &min_key, &max_key,
 751                                            path, 0, newer_than);
 752                 if (ret != 0)
 753                         goto none;
 754                 if (min_key.objectid != ino)
 755                         goto none;
 756                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
 757                         goto none;
 758
 759                 leaf = path->nodes[0];
 760                 extent = btrfs_item_ptr(leaf, path->slots[0],
 761                                         struct btrfs_file_extent_item);
 762
 763                 type = btrfs_file_extent_type(leaf, extent);
 764                 if (type == BTRFS_FILE_EXTENT_REG &&
 765                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
 766                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
 767                         *off = min_key.offset;
 768                         btrfs_free_path(path);
 769                         return 0;
 770                 }
 771
 772                 if (min_key.offset == (u64)-1)
 773                         goto none;
 774
 775                 min_key.offset++;
 776                 btrfs_release_path(path);
 777         }
 778 none:
 779         btrfs_free_path(path);
 780         return -ENOENT;
 781 }
 782
 783 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
 784 {
 785         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 786         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 787         struct extent_map *em;
 788         u64 len = PAGE_CACHE_SIZE;
 789
 790         /*
 791          * hopefully we have this extent in the tree already, try without
 792          * the full extent lock
 793          */
 794         read_lock(&em_tree->lock);
 795         em = lookup_extent_mapping(em_tree, start, len);
 796         read_unlock(&em_tree->lock);
 797
 798         if (!em) {
 799                 /* get the big lock and read metadata off disk */
 800                 lock_extent(io_tree, start, start + len - 1);
 801                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
 802                 unlock_extent(io_tree, start, start + len - 1);
 803
 804                 if (IS_ERR(em))
 805                         return NULL;
 806         }
 807
 808         return em;
 809 }
 810
 811 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
 812 {
 813         struct extent_map *next;
 814         bool ret = true;
 815
 816         /* this is the last extent */
 817         if (em->start + em->len >= i_size_read(inode))
 818                 return false;
 819
 820         next = defrag_lookup_extent(inode, em->start + em->len);
 821         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
 822                 ret = false;
 823
 824         free_extent_map(next);
 825         return ret;
 826 }
 827
 828 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
 829                                u64 *last_len, u64 *skip, u64 *defrag_end,
 830                                int compress)
 831 {
 832         struct extent_map *em;
 833         int ret = 1;
 834         bool next_mergeable = true;
 835
 836         /*
 837          * make sure that once we start defragging an extent, we keep on
 838          * defragging it
 839          */
 840         if (start < *defrag_end)
 841                 return 1;
 842
 843         *skip = 0;
 844
 845         em = defrag_lookup_extent(inode, start);
 846         if (!em)
 847                 return 0;
 848
 849         /* this will cover holes, and inline extents */
 850         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 851                 ret = 0;
 852                 goto out;
 853         }
 854
 855         next_mergeable = defrag_check_next_extent(inode, em);
 856
 857         /*
 858          * we hit a real extent, if it is big or the next extent is not a
 859          * real extent, don't bother defragging it
 860          */
 861         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
 862             (em->len >= thresh || !next_mergeable))
 863                 ret = 0;
 864 out:
 865         /*
 866          * last_len ends up being a counter of how many bytes we've defragged.
 867          * every time we choose not to defrag an extent, we reset *last_len
 868          * so that the next tiny extent will force a defrag.
 869          *
 870          * The end result of this is that tiny extents before a single big
 871          * extent will force at least part of that big extent to be defragged.
 872          */
 873         if (ret) {
 874                 *defrag_end = extent_map_end(em);
 875         } else {
 876                 *last_len = 0;
 877                 *skip = extent_map_end(em);
 878                 *defrag_end = 0;
 879         }
 880
 881         free_extent_map(em);
 882         return ret;
 883 }
 884
 885 /*
 886  * it doesn't do much good to defrag one or two pages
 887  * at a time.  This pulls in a nice chunk of pages
 888  * to COW and defrag.
 889  *
 890  * It also makes sure the delalloc code has enough
 891  * dirty data to avoid making new small extents as part
 892  * of the defrag
 893  *
 894  * It's a good idea to start RA on this range
 895  * before calling this.
 896  */
 897 static int cluster_pages_for_defrag(struct inode *inode,
 898                                     struct page **pages,
 899                                     unsigned long start_index,
 900                                     int num_pages)
 901 {
 902         unsigned long file_end;
 903         u64 isize = i_size_read(inode);
 904         u64 page_start;
 905         u64 page_end;
 906         u64 page_cnt;
 907         int ret;
 908         int i;
 909         int i_done;
 910         struct btrfs_ordered_extent *ordered;
 911         struct extent_state *cached_state = NULL;
 912         struct extent_io_tree *tree;
 913         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 914
 915         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
 916         if (!isize || start_index > file_end)
 917                 return 0;
 918
 919         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
 920
 921         ret = btrfs_delalloc_reserve_space(inode,
 922                                            page_cnt << PAGE_CACHE_SHIFT);
 923         if (ret)
 924                 return ret;
 925         i_done = 0;
 926         tree = &BTRFS_I(inode)->io_tree;
 927
 928         /* step one, lock all the pages */
 929         for (i = 0; i < page_cnt; i++) {
 930                 struct page *page;
 931 again:
 932                 page = find_or_create_page(inode->i_mapping,
 933                                            start_index + i, mask);
 934                 if (!page)
 935                         break;
 936
 937                 page_start = page_offset(page);
 938                 page_end = page_start + PAGE_CACHE_SIZE - 1;
 939                 while (1) {
 940                         lock_extent(tree, page_start, page_end);
 941                         ordered = btrfs_lookup_ordered_extent(inode,
 942                                                               page_start);
 943                         unlock_extent(tree, page_start, page_end);
 944                         if (!ordered)
 945                                 break;
 946
 947                         unlock_page(page);
 948                         btrfs_start_ordered_extent(inode, ordered, 1);
 949                         btrfs_put_ordered_extent(ordered);
 950                         lock_page(page);
 951                         /*
 952                          * we unlocked the page above, so we need check if
 953                          * it was released or not.
 954                          */
 955                         if (page->mapping != inode->i_mapping) {
 956                                 unlock_page(page);
 957                                 page_cache_release(page);
 958                                 goto again;
 959                         }
 960                 }
 961
 962                 if (!PageUptodate(page)) {
 963                         btrfs_readpage(NULL, page);
 964                         lock_page(page);
 965                         if (!PageUptodate(page)) {
 966                                 unlock_page(page);
 967                                 page_cache_release(page);
 968                                 ret = -EIO;
 969                                 break;
 970                         }
 971                 }
 972
 973                 if (page->mapping != inode->i_mapping) {
 974                         unlock_page(page);
 975                         page_cache_release(page);
 976                         goto again;
 977                 }
 978
 979                 pages[i] = page;
 980                 i_done++;
 981         }
 982         if (!i_done || ret)
 983                 goto out;
 984
 985         if (!(inode->i_sb->s_flags & MS_ACTIVE))
 986                 goto out;
 987
 988         /*
 989          * so now we have a nice long stream of locked
 990          * and up to date pages, lets wait on them
 991          */
 992         for (i = 0; i < i_done; i++)
 993                 wait_on_page_writeback(pages[i]);
 994
 995         page_start = page_offset(pages[0]);
 996         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
 997
 998         lock_extent_bits(&BTRFS_I(inode)->io_tree,
 999                          page_start, page_end - 1, 0, &cached_state);
1000         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1001                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1002                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
1003                           GFP_NOFS);
1004
1005         if (i_done != page_cnt) {
1006                 spin_lock(&BTRFS_I(inode)->lock);
1007                 BTRFS_I(inode)->outstanding_extents++;
1008                 spin_unlock(&BTRFS_I(inode)->lock);
1009                 btrfs_delalloc_release_space(inode,
1010                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1011         }
1012
1013
1014         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
1015                                   &cached_state);
1016
1017         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1018                              page_start, page_end - 1, &cached_state,
1019                              GFP_NOFS);
1020
1021         for (i = 0; i < i_done; i++) {
1022                 clear_page_dirty_for_io(pages[i]);
1023                 ClearPageChecked(pages[i]);
1024                 set_page_extent_mapped(pages[i]);
1025                 set_page_dirty(pages[i]);
1026                 unlock_page(pages[i]);
1027                 page_cache_release(pages[i]);
1028         }
1029         return i_done;
1030 out:
1031         for (i = 0; i < i_done; i++) {
1032                 unlock_page(pages[i]);
1033                 page_cache_release(pages[i]);
1034         }
1035         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1036         return ret;
1037
1038 }
1039
1040 int btrfs_defrag_file(struct inode *inode, struct file *file,
1041                       struct btrfs_ioctl_defrag_range_args *range,
1042                       u64 newer_than, unsigned long max_to_defrag)
1043 {
1044         struct btrfs_root *root = BTRFS_I(inode)->root;
1045         struct btrfs_super_block *disk_super;
1046         struct file_ra_state *ra = NULL;
1047         unsigned long last_index;
1048         u64 isize = i_size_read(inode);
1049         u64 features;
1050         u64 last_len = 0;
1051         u64 skip = 0;
1052         u64 defrag_end = 0;
1053         u64 newer_off = range->start;
1054         unsigned long i;
1055         unsigned long ra_index = 0;
1056         int ret;
1057         int defrag_count = 0;
1058         int compress_type = BTRFS_COMPRESS_ZLIB;
1059         int extent_thresh = range->extent_thresh;
1060         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1061         int cluster = max_cluster;
1062         u64 new_align = ~((u64)128 * 1024 - 1);
1063         struct page **pages = NULL;
1064
1065         if (extent_thresh == 0)
1066                 extent_thresh = 256 * 1024;
1067
1068         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1069                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1070                         return -EINVAL;
1071                 if (range->compress_type)
1072                         compress_type = range->compress_type;
1073         }
1074
1075         if (isize == 0)
1076                 return 0;
1077
1078         /*
1079          * if we were not given a file, allocate a readahead
1080          * context
1081          */
1082         if (!file) {
1083                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1084                 if (!ra)
1085                         return -ENOMEM;
1086                 file_ra_state_init(ra, inode->i_mapping);
1087         } else {
1088                 ra = &file->f_ra;
1089         }
1090
1091         pages = kmalloc(sizeof(struct page *) * max_cluster,
1092                         GFP_NOFS);
1093         if (!pages) {
1094                 ret = -ENOMEM;
1095                 goto out_ra;
1096         }
1097
1098         /* find the last page to defrag */
1099         if (range->start + range->len > range->start) {
1100                 last_index = min_t(u64, isize - 1,
1101                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1102         } else {
1103                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1104         }
1105
1106         if (newer_than) {
1107                 ret = find_new_extents(root, inode, newer_than,
1108                                        &newer_off, 64 * 1024);
1109                 if (!ret) {
1110                         range->start = newer_off;
1111                         /*
1112                          * we always align our defrag to help keep
1113                          * the extents in the file evenly spaced
1114                          */
1115                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1116                 } else
1117                         goto out_ra;
1118         } else {
1119                 i = range->start >> PAGE_CACHE_SHIFT;
1120         }
1121         if (!max_to_defrag)
1122                 max_to_defrag = last_index + 1;
1123
1124         /*
1125          * make writeback starts from i, so the defrag range can be
1126          * written sequentially.
1127          */
1128         if (i < inode->i_mapping->writeback_index)
1129                 inode->i_mapping->writeback_index = i;
1130
1131         while (i <= last_index && defrag_count < max_to_defrag &&
1132                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1133                 PAGE_CACHE_SHIFT)) {
1134                 /*
1135                  * make sure we stop running if someone unmounts
1136                  * the FS
1137                  */
1138                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1139                         break;
1140
1141                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1142                                          extent_thresh, &last_len, &skip,
1143                                          &defrag_end, range->flags &
1144                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1145                         unsigned long next;
1146                         /*
1147                          * the should_defrag function tells us how much to skip
1148                          * bump our counter by the suggested amount
1149                          */
1150                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1151                         i = max(i + 1, next);
1152                         continue;
1153                 }
1154
1155                 if (!newer_than) {
1156                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1157                                    PAGE_CACHE_SHIFT) - i;
1158                         cluster = min(cluster, max_cluster);
1159                 } else {
1160                         cluster = max_cluster;
1161                 }
1162
1163                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1164                         BTRFS_I(inode)->force_compress = compress_type;
1165
1166                 if (i + cluster > ra_index) {
1167                         ra_index = max(i, ra_index);
1168                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1169                                        cluster);
1170                         ra_index += max_cluster;
1171                 }
1172
1173                 mutex_lock(&inode->i_mutex);
1174                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1175                 if (ret < 0) {
1176                         mutex_unlock(&inode->i_mutex);
1177                         goto out_ra;
1178                 }
1179
1180                 defrag_count += ret;
1181                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1182                 mutex_unlock(&inode->i_mutex);
1183
1184                 if (newer_than) {
1185                         if (newer_off == (u64)-1)
1186                                 break;
1187
1188                         if (ret > 0)
1189                                 i += ret;
1190
1191                         newer_off = max(newer_off + 1,
1192                                         (u64)i << PAGE_CACHE_SHIFT);
1193
1194                         ret = find_new_extents(root, inode,
1195                                                newer_than, &newer_off,
1196                                                64 * 1024);
1197                         if (!ret) {
1198                                 range->start = newer_off;
1199                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1200                         } else {
1201                                 break;
1202                         }
1203                 } else {
1204                         if (ret > 0) {
1205                                 i += ret;
1206                                 last_len += ret << PAGE_CACHE_SHIFT;
1207                         } else {
1208                                 i++;
1209                                 last_len = 0;
1210                         }
1211                 }
1212         }
1213
1214         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1215                 filemap_flush(inode->i_mapping);
1216
1217         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1218                 /* the filemap_flush will queue IO into the worker threads, but
1219                  * we have to make sure the IO is actually started and that
1220                  * ordered extents get created before we return
1221                  */
1222                 atomic_inc(&root->fs_info->async_submit_draining);
1223                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1224                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1225                         wait_event(root->fs_info->async_submit_wait,
1226                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1227                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1228                 }
1229                 atomic_dec(&root->fs_info->async_submit_draining);
1230
1231                 mutex_lock(&inode->i_mutex);
1232                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1233                 mutex_unlock(&inode->i_mutex);
1234         }
1235
1236         disk_super = root->fs_info->super_copy;
1237         features = btrfs_super_incompat_flags(disk_super);
1238         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1239                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1240                 btrfs_set_super_incompat_flags(disk_super, features);
1241         }
1242
1243         ret = defrag_count;
1244
1245 out_ra:
1246         if (!file)
1247                 kfree(ra);
1248         kfree(pages);
1249         return ret;
1250 }
1251
1252 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1253                                         void __user *arg)
1254 {
1255         u64 new_size;
1256         u64 old_size;
1257         u64 devid = 1;
1258         struct btrfs_ioctl_vol_args *vol_args;
1259         struct btrfs_trans_handle *trans;
1260         struct btrfs_device *device = NULL;
1261         char *sizestr;
1262         char *devstr = NULL;
1263         int ret = 0;
1264         int mod = 0;
1265
1266         if (root->fs_info->sb->s_flags & MS_RDONLY)
1267                 return -EROFS;
1268
1269         if (!capable(CAP_SYS_ADMIN))
1270                 return -EPERM;
1271
1272         mutex_lock(&root->fs_info->volume_mutex);
1273         if (root->fs_info->balance_ctl) {
1274                 printk(KERN_INFO "btrfs: balance in progress\n");
1275                 ret = -EINVAL;
1276                 goto out;
1277         }
1278
1279         vol_args = memdup_user(arg, sizeof(*vol_args));
1280         if (IS_ERR(vol_args)) {
1281                 ret = PTR_ERR(vol_args);
1282                 goto out;
1283         }
1284
1285         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1286
1287         sizestr = vol_args->name;
1288         devstr = strchr(sizestr, ':');
1289         if (devstr) {
1290                 char *end;
1291                 sizestr = devstr + 1;
1292                 *devstr = '\0';
1293                 devstr = vol_args->name;
1294                 devid = simple_strtoull(devstr, &end, 10);
1295                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1296                        (unsigned long long)devid);
1297         }
1298         device = btrfs_find_device(root, devid, NULL, NULL);
1299         if (!device) {
1300                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1301                        (unsigned long long)devid);
1302                 ret = -EINVAL;
1303                 goto out_free;
1304         }
1305         if (device->fs_devices && device->fs_devices->seeding) {
1306                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1307                        "seeding device %llu\n",
1308                        (unsigned long long)devid);
1309                 ret = -EINVAL;
1310                 goto out_free;
1311         }
1312
1313         if (!strcmp(sizestr, "max"))
1314                 new_size = device->bdev->bd_inode->i_size;
1315         else {
1316                 if (sizestr[0] == '-') {
1317                         mod = -1;
1318                         sizestr++;
1319                 } else if (sizestr[0] == '+') {
1320                         mod = 1;
1321                         sizestr++;
1322                 }
1323                 new_size = memparse(sizestr, NULL);
1324                 if (new_size == 0) {
1325                         ret = -EINVAL;
1326                         goto out_free;
1327                 }
1328         }
1329
1330         old_size = device->total_bytes;
1331
1332         if (mod < 0) {
1333                 if (new_size > old_size) {
1334                         ret = -EINVAL;
1335                         goto out_free;
1336                 }
1337                 new_size = old_size - new_size;
1338         } else if (mod > 0) {
1339                 new_size = old_size + new_size;
1340         }
1341
1342         if (new_size < 256 * 1024 * 1024) {
1343                 ret = -EINVAL;
1344                 goto out_free;
1345         }
1346         if (new_size > device->bdev->bd_inode->i_size) {
1347                 ret = -EFBIG;
1348                 goto out_free;
1349         }
1350
1351         do_div(new_size, root->sectorsize);
1352         new_size *= root->sectorsize;
1353
1354         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1355                       rcu_str_deref(device->name),
1356                       (unsigned long long)new_size);
1357
1358         if (new_size > old_size) {
1359                 trans = btrfs_start_transaction(root, 0);
1360                 if (IS_ERR(trans)) {
1361                         ret = PTR_ERR(trans);
1362                         goto out_free;
1363                 }
1364                 ret = btrfs_grow_device(trans, device, new_size);
1365                 btrfs_commit_transaction(trans, root);
1366         } else if (new_size < old_size) {
1367                 ret = btrfs_shrink_device(device, new_size);
1368         }
1369
1370 out_free:
1371         kfree(vol_args);
1372 out:
1373         mutex_unlock(&root->fs_info->volume_mutex);
1374         return ret;
1375 }
1376
1377 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1378                                                     char *name,
1379                                                     unsigned long fd,
1380                                                     int subvol,
1381                                                     u64 *transid,
1382                                                     bool readonly)
1383 {
1384         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1385         struct file *src_file;
1386         int namelen;
1387         int ret = 0;
1388
1389         if (root->fs_info->sb->s_flags & MS_RDONLY)
1390                 return -EROFS;
1391
1392         ret = mnt_want_write_file(file);
1393         if (ret)
1394                 goto out;
1395
1396         namelen = strlen(name);
1397         if (strchr(name, '/')) {
1398                 ret = -EINVAL;
1399                 goto out_drop_write;
1400         }
1401
1402         if (name[0] == '.' &&
1403            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1404                 ret = -EEXIST;
1405                 goto out_drop_write;
1406         }
1407
1408         if (subvol) {
1409                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1410                                      NULL, transid, readonly);
1411         } else {
1412                 struct inode *src_inode;
1413                 src_file = fget(fd);
1414                 if (!src_file) {
1415                         ret = -EINVAL;
1416                         goto out_drop_write;
1417                 }
1418
1419                 src_inode = src_file->f_path.dentry->d_inode;
1420                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1421                         printk(KERN_INFO "btrfs: Snapshot src from "
1422                                "another FS\n");
1423                         ret = -EINVAL;
1424                         fput(src_file);
1425                         goto out_drop_write;
1426                 }
1427                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1428                                      BTRFS_I(src_inode)->root,
1429                                      transid, readonly);
1430                 fput(src_file);
1431         }
1432 out_drop_write:
1433         mnt_drop_write_file(file);
1434 out:
1435         return ret;
1436 }
1437
1438 static noinline int btrfs_ioctl_snap_create(struct file *file,
1439                                             void __user *arg, int subvol)
1440 {
1441         struct btrfs_ioctl_vol_args *vol_args;
1442         int ret;
1443
1444         vol_args = memdup_user(arg, sizeof(*vol_args));
1445         if (IS_ERR(vol_args))
1446                 return PTR_ERR(vol_args);
1447         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1448
1449         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1450                                               vol_args->fd, subvol,
1451                                               NULL, false);
1452
1453         kfree(vol_args);
1454         return ret;
1455 }
1456
1457 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1458                                                void __user *arg, int subvol)
1459 {
1460         struct btrfs_ioctl_vol_args_v2 *vol_args;
1461         int ret;
1462         u64 transid = 0;
1463         u64 *ptr = NULL;
1464         bool readonly = false;
1465
1466         vol_args = memdup_user(arg, sizeof(*vol_args));
1467         if (IS_ERR(vol_args))
1468                 return PTR_ERR(vol_args);
1469         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1470
1471         if (vol_args->flags &
1472             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1473                 ret = -EOPNOTSUPP;
1474                 goto out;
1475         }
1476
1477         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1478                 ptr = &transid;
1479         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1480                 readonly = true;
1481
1482         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1483                                               vol_args->fd, subvol,
1484                                               ptr, readonly);
1485
1486         if (ret == 0 && ptr &&
1487             copy_to_user(arg +
1488                          offsetof(struct btrfs_ioctl_vol_args_v2,
1489                                   transid), ptr, sizeof(*ptr)))
1490                 ret = -EFAULT;
1491 out:
1492         kfree(vol_args);
1493         return ret;
1494 }
1495
1496 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1497                                                 void __user *arg)
1498 {
1499         struct inode *inode = fdentry(file)->d_inode;
1500         struct btrfs_root *root = BTRFS_I(inode)->root;
1501         int ret = 0;
1502         u64 flags = 0;
1503
1504         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1505                 return -EINVAL;
1506
1507         down_read(&root->fs_info->subvol_sem);
1508         if (btrfs_root_readonly(root))
1509                 flags |= BTRFS_SUBVOL_RDONLY;
1510         up_read(&root->fs_info->subvol_sem);
1511
1512         if (copy_to_user(arg, &flags, sizeof(flags)))
1513                 ret = -EFAULT;
1514
1515         return ret;
1516 }
1517
1518 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1519                                               void __user *arg)
1520 {
1521         struct inode *inode = fdentry(file)->d_inode;
1522         struct btrfs_root *root = BTRFS_I(inode)->root;
1523         struct btrfs_trans_handle *trans;
1524         u64 root_flags;
1525         u64 flags;
1526         int ret = 0;
1527
1528         if (root->fs_info->sb->s_flags & MS_RDONLY)
1529                 return -EROFS;
1530
1531         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1532                 return -EINVAL;
1533
1534         if (copy_from_user(&flags, arg, sizeof(flags)))
1535                 return -EFAULT;
1536
1537         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1538                 return -EINVAL;
1539
1540         if (flags & ~BTRFS_SUBVOL_RDONLY)
1541                 return -EOPNOTSUPP;
1542
1543         if (!inode_owner_or_capable(inode))
1544                 return -EACCES;
1545
1546         down_write(&root->fs_info->subvol_sem);
1547
1548         /* nothing to do */
1549         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1550                 goto out;
1551
1552         root_flags = btrfs_root_flags(&root->root_item);
1553         if (flags & BTRFS_SUBVOL_RDONLY)
1554                 btrfs_set_root_flags(&root->root_item,
1555                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1556         else
1557                 btrfs_set_root_flags(&root->root_item,
1558                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1559
1560         trans = btrfs_start_transaction(root, 1);
1561         if (IS_ERR(trans)) {
1562                 ret = PTR_ERR(trans);
1563                 goto out_reset;
1564         }
1565
1566         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1567                                 &root->root_key, &root->root_item);
1568
1569         btrfs_commit_transaction(trans, root);
1570 out_reset:
1571         if (ret)
1572                 btrfs_set_root_flags(&root->root_item, root_flags);
1573 out:
1574         up_write(&root->fs_info->subvol_sem);
1575         return ret;
1576 }
1577
1578 /*
1579  * helper to check if the subvolume references other subvolumes
1580  */
1581 static noinline int may_destroy_subvol(struct btrfs_root *root)
1582 {
1583         struct btrfs_path *path;
1584         struct btrfs_key key;
1585         int ret;
1586
1587         path = btrfs_alloc_path();
1588         if (!path)
1589                 return -ENOMEM;
1590
1591         key.objectid = root->root_key.objectid;
1592         key.type = BTRFS_ROOT_REF_KEY;
1593         key.offset = (u64)-1;
1594
1595         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1596                                 &key, path, 0, 0);
1597         if (ret < 0)
1598                 goto out;
1599         BUG_ON(ret == 0);
1600
1601         ret = 0;
1602         if (path->slots[0] > 0) {
1603                 path->slots[0]--;
1604                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1605                 if (key.objectid == root->root_key.objectid &&
1606                     key.type == BTRFS_ROOT_REF_KEY)
1607                         ret = -ENOTEMPTY;
1608         }
1609 out:
1610         btrfs_free_path(path);
1611         return ret;
1612 }
1613
1614 static noinline int key_in_sk(struct btrfs_key *key,
1615                               struct btrfs_ioctl_search_key *sk)
1616 {
1617         struct btrfs_key test;
1618         int ret;
1619
1620         test.objectid = sk->min_objectid;
1621         test.type = sk->min_type;
1622         test.offset = sk->min_offset;
1623
1624         ret = btrfs_comp_cpu_keys(key, &test);
1625         if (ret < 0)
1626                 return 0;
1627
1628         test.objectid = sk->max_objectid;
1629         test.type = sk->max_type;
1630         test.offset = sk->max_offset;
1631
1632         ret = btrfs_comp_cpu_keys(key, &test);
1633         if (ret > 0)
1634                 return 0;
1635         return 1;
1636 }
1637
1638 static noinline int copy_to_sk(struct btrfs_root *root,
1639                                struct btrfs_path *path,
1640                                struct btrfs_key *key,
1641                                struct btrfs_ioctl_search_key *sk,
1642                                char *buf,
1643                                unsigned long *sk_offset,
1644                                int *num_found)
1645 {
1646         u64 found_transid;
1647         struct extent_buffer *leaf;
1648         struct btrfs_ioctl_search_header sh;
1649         unsigned long item_off;
1650         unsigned long item_len;
1651         int nritems;
1652         int i;
1653         int slot;
1654         int ret = 0;
1655
1656         leaf = path->nodes[0];
1657         slot = path->slots[0];
1658         nritems = btrfs_header_nritems(leaf);
1659
1660         if (btrfs_header_generation(leaf) > sk->max_transid) {
1661                 i = nritems;
1662                 goto advance_key;
1663         }
1664         found_transid = btrfs_header_generation(leaf);
1665
1666         for (i = slot; i < nritems; i++) {
1667                 item_off = btrfs_item_ptr_offset(leaf, i);
1668                 item_len = btrfs_item_size_nr(leaf, i);
1669
1670                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1671                         item_len = 0;
1672
1673                 if (sizeof(sh) + item_len + *sk_offset >
1674                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1675                         ret = 1;
1676                         goto overflow;
1677                 }
1678
1679                 btrfs_item_key_to_cpu(leaf, key, i);
1680                 if (!key_in_sk(key, sk))
1681                         continue;
1682
1683                 sh.objectid = key->objectid;
1684                 sh.offset = key->offset;
1685                 sh.type = key->type;
1686                 sh.len = item_len;
1687                 sh.transid = found_transid;
1688
1689                 /* copy search result header */
1690                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1691                 *sk_offset += sizeof(sh);
1692
1693                 if (item_len) {
1694                         char *p = buf + *sk_offset;
1695                         /* copy the item */
1696                         read_extent_buffer(leaf, p,
1697                                            item_off, item_len);
1698                         *sk_offset += item_len;
1699                 }
1700                 (*num_found)++;
1701
1702                 if (*num_found >= sk->nr_items)
1703                         break;
1704         }
1705 advance_key:
1706         ret = 0;
1707         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1708                 key->offset++;
1709         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1710                 key->offset = 0;
1711                 key->type++;
1712         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1713                 key->offset = 0;
1714                 key->type = 0;
1715                 key->objectid++;
1716         } else
1717                 ret = 1;
1718 overflow:
1719         return ret;
1720 }
1721
1722 static noinline int search_ioctl(struct inode *inode,
1723                                  struct btrfs_ioctl_search_args *args)
1724 {
1725         struct btrfs_root *root;
1726         struct btrfs_key key;
1727         struct btrfs_key max_key;
1728         struct btrfs_path *path;
1729         struct btrfs_ioctl_search_key *sk = &args->key;
1730         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1731         int ret;
1732         int num_found = 0;
1733         unsigned long sk_offset = 0;
1734
1735         path = btrfs_alloc_path();
1736         if (!path)
1737                 return -ENOMEM;
1738
1739         if (sk->tree_id == 0) {
1740                 /* search the root of the inode that was passed */
1741                 root = BTRFS_I(inode)->root;
1742         } else {
1743                 key.objectid = sk->tree_id;
1744                 key.type = BTRFS_ROOT_ITEM_KEY;
1745                 key.offset = (u64)-1;
1746                 root = btrfs_read_fs_root_no_name(info, &key);
1747                 if (IS_ERR(root)) {
1748                         printk(KERN_ERR "could not find root %llu\n",
1749                                sk->tree_id);
1750                         btrfs_free_path(path);
1751                         return -ENOENT;
1752                 }
1753         }
1754
1755         key.objectid = sk->min_objectid;
1756         key.type = sk->min_type;
1757         key.offset = sk->min_offset;
1758
1759         max_key.objectid = sk->max_objectid;
1760         max_key.type = sk->max_type;
1761         max_key.offset = sk->max_offset;
1762
1763         path->keep_locks = 1;
1764
1765         while(1) {
1766                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1767                                            sk->min_transid);
1768                 if (ret != 0) {
1769                         if (ret > 0)
1770                                 ret = 0;
1771                         goto err;
1772                 }
1773                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1774                                  &sk_offset, &num_found);
1775                 btrfs_release_path(path);
1776                 if (ret || num_found >= sk->nr_items)
1777                         break;
1778
1779         }
1780         ret = 0;
1781 err:
1782         sk->nr_items = num_found;
1783         btrfs_free_path(path);
1784         return ret;
1785 }
1786
1787 static noinline int btrfs_ioctl_tree_search(struct file *file,
1788                                            void __user *argp)
1789 {
1790          struct btrfs_ioctl_search_args *args;
1791          struct inode *inode;
1792          int ret;
1793
1794         if (!capable(CAP_SYS_ADMIN))
1795                 return -EPERM;
1796
1797         args = memdup_user(argp, sizeof(*args));
1798         if (IS_ERR(args))
1799                 return PTR_ERR(args);
1800
1801         inode = fdentry(file)->d_inode;
1802         ret = search_ioctl(inode, args);
1803         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1804                 ret = -EFAULT;
1805         kfree(args);
1806         return ret;
1807 }
1808
1809 /*
1810  * Search INODE_REFs to identify path name of 'dirid' directory
1811  * in a 'tree_id' tree. and sets path name to 'name'.
1812  */
1813 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1814                                 u64 tree_id, u64 dirid, char *name)
1815 {
1816         struct btrfs_root *root;
1817         struct btrfs_key key;
1818         char *ptr;
1819         int ret = -1;
1820         int slot;
1821         int len;
1822         int total_len = 0;
1823         struct btrfs_inode_ref *iref;
1824         struct extent_buffer *l;
1825         struct btrfs_path *path;
1826
1827         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1828                 name[0]='\0';
1829                 return 0;
1830         }
1831
1832         path = btrfs_alloc_path();
1833         if (!path)
1834                 return -ENOMEM;
1835
1836         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1837
1838         key.objectid = tree_id;
1839         key.type = BTRFS_ROOT_ITEM_KEY;
1840         key.offset = (u64)-1;
1841         root = btrfs_read_fs_root_no_name(info, &key);
1842         if (IS_ERR(root)) {
1843                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1844                 ret = -ENOENT;
1845                 goto out;
1846         }
1847
1848         key.objectid = dirid;
1849         key.type = BTRFS_INODE_REF_KEY;
1850         key.offset = (u64)-1;
1851
1852         while(1) {
1853                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1854                 if (ret < 0)
1855                         goto out;
1856
1857                 l = path->nodes[0];
1858                 slot = path->slots[0];
1859                 if (ret > 0 && slot > 0)
1860                         slot--;
1861                 btrfs_item_key_to_cpu(l, &key, slot);
1862
1863                 if (ret > 0 && (key.objectid != dirid ||
1864                                 key.type != BTRFS_INODE_REF_KEY)) {
1865                         ret = -ENOENT;
1866                         goto out;
1867                 }
1868
1869                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1870                 len = btrfs_inode_ref_name_len(l, iref);
1871                 ptr -= len + 1;
1872                 total_len += len + 1;
1873                 if (ptr < name)
1874                         goto out;
1875
1876                 *(ptr + len) = '/';
1877                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1878
1879                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1880                         break;
1881
1882                 btrfs_release_path(path);
1883                 key.objectid = key.offset;
1884                 key.offset = (u64)-1;
1885                 dirid = key.objectid;
1886         }
1887         if (ptr < name)
1888                 goto out;
1889         memmove(name, ptr, total_len);
1890         name[total_len]='\0';
1891         ret = 0;
1892 out:
1893         btrfs_free_path(path);
1894         return ret;
1895 }
1896
1897 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1898                                            void __user *argp)
1899 {
1900          struct btrfs_ioctl_ino_lookup_args *args;
1901          struct inode *inode;
1902          int ret;
1903
1904         if (!capable(CAP_SYS_ADMIN))
1905                 return -EPERM;
1906
1907         args = memdup_user(argp, sizeof(*args));
1908         if (IS_ERR(args))
1909                 return PTR_ERR(args);
1910
1911         inode = fdentry(file)->d_inode;
1912
1913         if (args->treeid == 0)
1914                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1915
1916         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1917                                         args->treeid, args->objectid,
1918                                         args->name);
1919
1920         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1921                 ret = -EFAULT;
1922
1923         kfree(args);
1924         return ret;
1925 }
1926
1927 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1928                                              void __user *arg)
1929 {
1930         struct dentry *parent = fdentry(file);
1931         struct dentry *dentry;
1932         struct inode *dir = parent->d_inode;
1933         struct inode *inode;
1934         struct btrfs_root *root = BTRFS_I(dir)->root;
1935         struct btrfs_root *dest = NULL;
1936         struct btrfs_ioctl_vol_args *vol_args;
1937         struct btrfs_trans_handle *trans;
1938         int namelen;
1939         int ret;
1940         int err = 0;
1941
1942         vol_args = memdup_user(arg, sizeof(*vol_args));
1943         if (IS_ERR(vol_args))
1944                 return PTR_ERR(vol_args);
1945
1946         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1947         namelen = strlen(vol_args->name);
1948         if (strchr(vol_args->name, '/') ||
1949             strncmp(vol_args->name, "..", namelen) == 0) {
1950                 err = -EINVAL;
1951                 goto out;
1952         }
1953
1954         err = mnt_want_write_file(file);
1955         if (err)
1956                 goto out;
1957
1958         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1959         dentry = lookup_one_len(vol_args->name, parent, namelen);
1960         if (IS_ERR(dentry)) {
1961                 err = PTR_ERR(dentry);
1962                 goto out_unlock_dir;
1963         }
1964
1965         if (!dentry->d_inode) {
1966                 err = -ENOENT;
1967                 goto out_dput;
1968         }
1969
1970         inode = dentry->d_inode;
1971         dest = BTRFS_I(inode)->root;
1972         if (!capable(CAP_SYS_ADMIN)){
1973                 /*
1974                  * Regular user.  Only allow this with a special mount
1975                  * option, when the user has write+exec access to the
1976                  * subvol root, and when rmdir(2) would have been
1977                  * allowed.
1978                  *
1979                  * Note that this is _not_ check that the subvol is
1980                  * empty or doesn't contain data that we wouldn't
1981                  * otherwise be able to delete.
1982                  *
1983                  * Users who want to delete empty subvols should try
1984                  * rmdir(2).
1985                  */
1986                 err = -EPERM;
1987                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1988                         goto out_dput;
1989
1990                 /*
1991                  * Do not allow deletion if the parent dir is the same
1992                  * as the dir to be deleted.  That means the ioctl
1993                  * must be called on the dentry referencing the root
1994                  * of the subvol, not a random directory contained
1995                  * within it.
1996                  */
1997                 err = -EINVAL;
1998                 if (root == dest)
1999                         goto out_dput;
2000
2001                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2002                 if (err)
2003                         goto out_dput;
2004
2005                 /* check if subvolume may be deleted by a non-root user */
2006                 err = btrfs_may_delete(dir, dentry, 1);
2007                 if (err)
2008                         goto out_dput;
2009         }
2010
2011         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2012                 err = -EINVAL;
2013                 goto out_dput;
2014         }
2015
2016         mutex_lock(&inode->i_mutex);
2017         err = d_invalidate(dentry);
2018         if (err)
2019                 goto out_unlock;
2020
2021         down_write(&root->fs_info->subvol_sem);
2022
2023         err = may_destroy_subvol(dest);
2024         if (err)
2025                 goto out_up_write;
2026
2027         trans = btrfs_start_transaction(root, 0);
2028         if (IS_ERR(trans)) {
2029                 err = PTR_ERR(trans);
2030                 goto out_up_write;
2031         }
2032         trans->block_rsv = &root->fs_info->global_block_rsv;
2033
2034         ret = btrfs_unlink_subvol(trans, root, dir,
2035                                 dest->root_key.objectid,
2036                                 dentry->d_name.name,
2037                                 dentry->d_name.len);
2038         if (ret) {
2039                 err = ret;
2040                 btrfs_abort_transaction(trans, root, ret);
2041                 goto out_end_trans;
2042         }
2043
2044         btrfs_record_root_in_trans(trans, dest);
2045
2046         memset(&dest->root_item.drop_progress, 0,
2047                 sizeof(dest->root_item.drop_progress));
2048         dest->root_item.drop_level = 0;
2049         btrfs_set_root_refs(&dest->root_item, 0);
2050
2051         if (!xchg(&dest->orphan_item_inserted, 1)) {
2052                 ret = btrfs_insert_orphan_item(trans,
2053                                         root->fs_info->tree_root,
2054                                         dest->root_key.objectid);
2055                 if (ret) {
2056                         btrfs_abort_transaction(trans, root, ret);
2057                         err = ret;
2058                         goto out_end_trans;
2059                 }
2060         }
2061 out_end_trans:
2062         ret = btrfs_end_transaction(trans, root);
2063         if (ret && !err)
2064                 err = ret;
2065         inode->i_flags |= S_DEAD;
2066 out_up_write:
2067         up_write(&root->fs_info->subvol_sem);
2068 out_unlock:
2069         mutex_unlock(&inode->i_mutex);
2070         if (!err) {
2071                 shrink_dcache_sb(root->fs_info->sb);
2072                 btrfs_invalidate_inodes(dest);
2073                 d_delete(dentry);
2074         }
2075 out_dput:
2076         dput(dentry);
2077 out_unlock_dir:
2078         mutex_unlock(&dir->i_mutex);
2079         mnt_drop_write_file(file);
2080 out:
2081         kfree(vol_args);
2082         return err;
2083 }
2084
2085 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2086 {
2087         struct inode *inode = fdentry(file)->d_inode;
2088         struct btrfs_root *root = BTRFS_I(inode)->root;
2089         struct btrfs_ioctl_defrag_range_args *range;
2090         int ret;
2091
2092         if (btrfs_root_readonly(root))
2093                 return -EROFS;
2094
2095         ret = mnt_want_write_file(file);
2096         if (ret)
2097                 return ret;
2098
2099         switch (inode->i_mode & S_IFMT) {
2100         case S_IFDIR:
2101                 if (!capable(CAP_SYS_ADMIN)) {
2102                         ret = -EPERM;
2103                         goto out;
2104                 }
2105                 ret = btrfs_defrag_root(root, 0);
2106                 if (ret)
2107                         goto out;
2108                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2109                 break;
2110         case S_IFREG:
2111                 if (!(file->f_mode & FMODE_WRITE)) {
2112                         ret = -EINVAL;
2113                         goto out;
2114                 }
2115
2116                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2117                 if (!range) {
2118                         ret = -ENOMEM;
2119                         goto out;
2120                 }
2121
2122                 if (argp) {
2123                         if (copy_from_user(range, argp,
2124                                            sizeof(*range))) {
2125                                 ret = -EFAULT;
2126                                 kfree(range);
2127                                 goto out;
2128                         }
2129                         /* compression requires us to start the IO */
2130                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2131                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2132                                 range->extent_thresh = (u32)-1;
2133                         }
2134                 } else {
2135                         /* the rest are all set to zero by kzalloc */
2136                         range->len = (u64)-1;
2137                 }
2138                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2139                                         range, 0, 0);
2140                 if (ret > 0)
2141                         ret = 0;
2142                 kfree(range);
2143                 break;
2144         default:
2145                 ret = -EINVAL;
2146         }
2147 out:
2148         mnt_drop_write_file(file);
2149         return ret;
2150 }
2151
2152 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2153 {
2154         struct btrfs_ioctl_vol_args *vol_args;
2155         int ret;
2156
2157         if (!capable(CAP_SYS_ADMIN))
2158                 return -EPERM;
2159
2160         mutex_lock(&root->fs_info->volume_mutex);
2161         if (root->fs_info->balance_ctl) {
2162                 printk(KERN_INFO "btrfs: balance in progress\n");
2163                 ret = -EINVAL;
2164                 goto out;
2165         }
2166
2167         vol_args = memdup_user(arg, sizeof(*vol_args));
2168         if (IS_ERR(vol_args)) {
2169                 ret = PTR_ERR(vol_args);
2170                 goto out;
2171         }
2172
2173         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2174         ret = btrfs_init_new_device(root, vol_args->name);
2175
2176         kfree(vol_args);
2177 out:
2178         mutex_unlock(&root->fs_info->volume_mutex);
2179         return ret;
2180 }
2181
2182 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2183 {
2184         struct btrfs_ioctl_vol_args *vol_args;
2185         int ret;
2186
2187         if (!capable(CAP_SYS_ADMIN))
2188                 return -EPERM;
2189
2190         if (root->fs_info->sb->s_flags & MS_RDONLY)
2191                 return -EROFS;
2192
2193         mutex_lock(&root->fs_info->volume_mutex);
2194         if (root->fs_info->balance_ctl) {
2195                 printk(KERN_INFO "btrfs: balance in progress\n");
2196                 ret = -EINVAL;
2197                 goto out;
2198         }
2199
2200         vol_args = memdup_user(arg, sizeof(*vol_args));
2201         if (IS_ERR(vol_args)) {
2202                 ret = PTR_ERR(vol_args);
2203                 goto out;
2204         }
2205
2206         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2207         ret = btrfs_rm_device(root, vol_args->name);
2208
2209         kfree(vol_args);
2210 out:
2211         mutex_unlock(&root->fs_info->volume_mutex);
2212         return ret;
2213 }
2214
2215 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2216 {
2217         struct btrfs_ioctl_fs_info_args *fi_args;
2218         struct btrfs_device *device;
2219         struct btrfs_device *next;
2220         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2221         int ret = 0;
2222
2223         if (!capable(CAP_SYS_ADMIN))
2224                 return -EPERM;
2225
2226         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2227         if (!fi_args)
2228                 return -ENOMEM;
2229
2230         fi_args->num_devices = fs_devices->num_devices;
2231         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2232
2233         mutex_lock(&fs_devices->device_list_mutex);
2234         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2235                 if (device->devid > fi_args->max_id)
2236                         fi_args->max_id = device->devid;
2237         }
2238         mutex_unlock(&fs_devices->device_list_mutex);
2239
2240         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2241                 ret = -EFAULT;
2242
2243         kfree(fi_args);
2244         return ret;
2245 }
2246
2247 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2248 {
2249         struct btrfs_ioctl_dev_info_args *di_args;
2250         struct btrfs_device *dev;
2251         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2252         int ret = 0;
2253         char *s_uuid = NULL;
2254         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2255
2256         if (!capable(CAP_SYS_ADMIN))
2257                 return -EPERM;
2258
2259         di_args = memdup_user(arg, sizeof(*di_args));
2260         if (IS_ERR(di_args))
2261                 return PTR_ERR(di_args);
2262
2263         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2264                 s_uuid = di_args->uuid;
2265
2266         mutex_lock(&fs_devices->device_list_mutex);
2267         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2268         mutex_unlock(&fs_devices->device_list_mutex);
2269
2270         if (!dev) {
2271                 ret = -ENODEV;
2272                 goto out;
2273         }
2274
2275         di_args->devid = dev->devid;
2276         di_args->bytes_used = dev->bytes_used;
2277         di_args->total_bytes = dev->total_bytes;
2278         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2279         if (dev->name) {
2280                 struct rcu_string *name;
2281
2282                 rcu_read_lock();
2283                 name = rcu_dereference(dev->name);
2284                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2285                 rcu_read_unlock();
2286                 di_args->path[sizeof(di_args->path) - 1] = 0;
2287         } else {
2288                 di_args->path[0] = '\0';
2289         }
2290
2291 out:
2292         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2293                 ret = -EFAULT;
2294
2295         kfree(di_args);
2296         return ret;
2297 }
2298
2299 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2300                                        u64 off, u64 olen, u64 destoff)
2301 {
2302         struct inode *inode = fdentry(file)->d_inode;
2303         struct btrfs_root *root = BTRFS_I(inode)->root;
2304         struct file *src_file;
2305         struct inode *src;
2306         struct btrfs_trans_handle *trans;
2307         struct btrfs_path *path;
2308         struct extent_buffer *leaf;
2309         char *buf;
2310         struct btrfs_key key;
2311         u32 nritems;
2312         int slot;
2313         int ret;
2314         u64 len = olen;
2315         u64 bs = root->fs_info->sb->s_blocksize;
2316         u64 hint_byte;
2317
2318         /*
2319          * TODO:
2320          * - split compressed inline extents.  annoying: we need to
2321          *   decompress into destination's address_space (the file offset
2322          *   may change, so source mapping won't do), then recompress (or
2323          *   otherwise reinsert) a subrange.
2324          * - allow ranges within the same file to be cloned (provided
2325          *   they don't overlap)?
2326          */
2327
2328         /* the destination must be opened for writing */
2329         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2330                 return -EINVAL;
2331
2332         if (btrfs_root_readonly(root))
2333                 return -EROFS;
2334
2335         ret = mnt_want_write_file(file);
2336         if (ret)
2337                 return ret;
2338
2339         src_file = fget(srcfd);
2340         if (!src_file) {
2341                 ret = -EBADF;
2342                 goto out_drop_write;
2343         }
2344
2345         src = src_file->f_dentry->d_inode;
2346
2347         ret = -EINVAL;
2348         if (src == inode)
2349                 goto out_fput;
2350
2351         /* the src must be open for reading */
2352         if (!(src_file->f_mode & FMODE_READ))
2353                 goto out_fput;
2354
2355         /* don't make the dst file partly checksummed */
2356         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2357             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2358                 goto out_fput;
2359
2360         ret = -EISDIR;
2361         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2362                 goto out_fput;
2363
2364         ret = -EXDEV;
2365         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2366                 goto out_fput;
2367
2368         ret = -ENOMEM;
2369         buf = vmalloc(btrfs_level_size(root, 0));
2370         if (!buf)
2371                 goto out_fput;
2372
2373         path = btrfs_alloc_path();
2374         if (!path) {
2375                 vfree(buf);
2376                 goto out_fput;
2377         }
2378         path->reada = 2;
2379
2380         if (inode < src) {
2381                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2382                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2383         } else {
2384                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2385                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2386         }
2387
2388         /* determine range to clone */
2389         ret = -EINVAL;
2390         if (off + len > src->i_size || off + len < off)
2391                 goto out_unlock;
2392         if (len == 0)
2393                 olen = len = src->i_size - off;
2394         /* if we extend to eof, continue to block boundary */
2395         if (off + len == src->i_size)
2396                 len = ALIGN(src->i_size, bs) - off;
2397
2398         /* verify the end result is block aligned */
2399         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2400             !IS_ALIGNED(destoff, bs))
2401                 goto out_unlock;
2402
2403         if (destoff > inode->i_size) {
2404                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2405                 if (ret)
2406                         goto out_unlock;
2407         }
2408
2409         /* truncate page cache pages from target inode range */
2410         truncate_inode_pages_range(&inode->i_data, destoff,
2411                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2412
2413         /* do any pending delalloc/csum calc on src, one way or
2414            another, and lock file content */
2415         while (1) {
2416                 struct btrfs_ordered_extent *ordered;
2417                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len);
2418                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2419                 if (!ordered &&
2420                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2421                                    EXTENT_DELALLOC, 0, NULL))
2422                         break;
2423                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len);
2424                 if (ordered)
2425                         btrfs_put_ordered_extent(ordered);
2426                 btrfs_wait_ordered_range(src, off, len);
2427         }
2428
2429         /* clone data */
2430         key.objectid = btrfs_ino(src);
2431         key.type = BTRFS_EXTENT_DATA_KEY;
2432         key.offset = 0;
2433
2434         while (1) {
2435                 /*
2436                  * note the key will change type as we walk through the
2437                  * tree.
2438                  */
2439                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2440                 if (ret < 0)
2441                         goto out;
2442
2443                 nritems = btrfs_header_nritems(path->nodes[0]);
2444                 if (path->slots[0] >= nritems) {
2445                         ret = btrfs_next_leaf(root, path);
2446                         if (ret < 0)
2447                                 goto out;
2448                         if (ret > 0)
2449                                 break;
2450                         nritems = btrfs_header_nritems(path->nodes[0]);
2451                 }
2452                 leaf = path->nodes[0];
2453                 slot = path->slots[0];
2454
2455                 btrfs_item_key_to_cpu(leaf, &key, slot);
2456                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2457                     key.objectid != btrfs_ino(src))
2458                         break;
2459
2460                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2461                         struct btrfs_file_extent_item *extent;
2462                         int type;
2463                         u32 size;
2464                         struct btrfs_key new_key;
2465                         u64 disko = 0, diskl = 0;
2466                         u64 datao = 0, datal = 0;
2467                         u8 comp;
2468                         u64 endoff;
2469
2470                         size = btrfs_item_size_nr(leaf, slot);
2471                         read_extent_buffer(leaf, buf,
2472                                            btrfs_item_ptr_offset(leaf, slot),
2473                                            size);
2474
2475                         extent = btrfs_item_ptr(leaf, slot,
2476                                                 struct btrfs_file_extent_item);
2477                         comp = btrfs_file_extent_compression(leaf, extent);
2478                         type = btrfs_file_extent_type(leaf, extent);
2479                         if (type == BTRFS_FILE_EXTENT_REG ||
2480                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2481                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2482                                                                       extent);
2483                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2484                                                                  extent);
2485                                 datao = btrfs_file_extent_offset(leaf, extent);
2486                                 datal = btrfs_file_extent_num_bytes(leaf,
2487                                                                     extent);
2488                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2489                                 /* take upper bound, may be compressed */
2490                                 datal = btrfs_file_extent_ram_bytes(leaf,
2491                                                                     extent);
2492                         }
2493                         btrfs_release_path(path);
2494
2495                         if (key.offset + datal <= off ||
2496                             key.offset >= off+len)
2497                                 goto next;
2498
2499                         memcpy(&new_key, &key, sizeof(new_key));
2500                         new_key.objectid = btrfs_ino(inode);
2501                         if (off <= key.offset)
2502                                 new_key.offset = key.offset + destoff - off;
2503                         else
2504                                 new_key.offset = destoff;
2505
2506                         /*
2507                          * 1 - adjusting old extent (we may have to split it)
2508                          * 1 - add new extent
2509                          * 1 - inode update
2510                          */
2511                         trans = btrfs_start_transaction(root, 3);
2512                         if (IS_ERR(trans)) {
2513                                 ret = PTR_ERR(trans);
2514                                 goto out;
2515                         }
2516
2517                         if (type == BTRFS_FILE_EXTENT_REG ||
2518                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2519                                 /*
2520                                  *    a  | --- range to clone ---|  b
2521                                  * | ------------- extent ------------- |
2522                                  */
2523
2524                                 /* substract range b */
2525                                 if (key.offset + datal > off + len)
2526                                         datal = off + len - key.offset;
2527
2528                                 /* substract range a */
2529                                 if (off > key.offset) {
2530                                         datao += off - key.offset;
2531                                         datal -= off - key.offset;
2532                                 }
2533
2534                                 ret = btrfs_drop_extents(trans, inode,
2535                                                          new_key.offset,
2536                                                          new_key.offset + datal,
2537                                                          &hint_byte, 1);
2538                                 if (ret) {
2539                                         btrfs_abort_transaction(trans, root,
2540                                                                 ret);
2541                                         btrfs_end_transaction(trans, root);
2542                                         goto out;
2543                                 }
2544
2545                                 ret = btrfs_insert_empty_item(trans, root, path,
2546                                                               &new_key, size);
2547                                 if (ret) {
2548                                         btrfs_abort_transaction(trans, root,
2549                                                                 ret);
2550                                         btrfs_end_transaction(trans, root);
2551                                         goto out;
2552                                 }
2553
2554                                 leaf = path->nodes[0];
2555                                 slot = path->slots[0];
2556                                 write_extent_buffer(leaf, buf,
2557                                             btrfs_item_ptr_offset(leaf, slot),
2558                                             size);
2559
2560                                 extent = btrfs_item_ptr(leaf, slot,
2561                                                 struct btrfs_file_extent_item);
2562
2563                                 /* disko == 0 means it's a hole */
2564                                 if (!disko)
2565                                         datao = 0;
2566
2567                                 btrfs_set_file_extent_offset(leaf, extent,
2568                                                              datao);
2569                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2570                                                                 datal);
2571                                 if (disko) {
2572                                         inode_add_bytes(inode, datal);
2573                                         ret = btrfs_inc_extent_ref(trans, root,
2574                                                         disko, diskl, 0,
2575                                                         root->root_key.objectid,
2576                                                         btrfs_ino(inode),
2577                                                         new_key.offset - datao,
2578                                                         0);
2579                                         if (ret) {
2580                                                 btrfs_abort_transaction(trans,
2581                                                                         root,
2582                                                                         ret);
2583                                                 btrfs_end_transaction(trans,
2584                                                                       root);
2585                                                 goto out;
2586
2587                                         }
2588                                 }
2589                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2590                                 u64 skip = 0;
2591                                 u64 trim = 0;
2592                                 if (off > key.offset) {
2593                                         skip = off - key.offset;
2594                                         new_key.offset += skip;
2595                                 }
2596
2597                                 if (key.offset + datal > off+len)
2598                                         trim = key.offset + datal - (off+len);
2599
2600                                 if (comp && (skip || trim)) {
2601                                         ret = -EINVAL;
2602                                         btrfs_end_transaction(trans, root);
2603                                         goto out;
2604                                 }
2605                                 size -= skip + trim;
2606                                 datal -= skip + trim;
2607
2608                                 ret = btrfs_drop_extents(trans, inode,
2609                                                          new_key.offset,
2610                                                          new_key.offset + datal,
2611                                                          &hint_byte, 1);
2612                                 if (ret) {
2613                                         btrfs_abort_transaction(trans, root,
2614                                                                 ret);
2615                                         btrfs_end_transaction(trans, root);
2616                                         goto out;
2617                                 }
2618
2619                                 ret = btrfs_insert_empty_item(trans, root, path,
2620                                                               &new_key, size);
2621                                 if (ret) {
2622                                         btrfs_abort_transaction(trans, root,
2623                                                                 ret);
2624                                         btrfs_end_transaction(trans, root);
2625                                         goto out;
2626                                 }
2627
2628                                 if (skip) {
2629                                         u32 start =
2630                                           btrfs_file_extent_calc_inline_size(0);
2631                                         memmove(buf+start, buf+start+skip,
2632                                                 datal);
2633                                 }
2634
2635                                 leaf = path->nodes[0];
2636                                 slot = path->slots[0];
2637                                 write_extent_buffer(leaf, buf,
2638                                             btrfs_item_ptr_offset(leaf, slot),
2639                                             size);
2640                                 inode_add_bytes(inode, datal);
2641                         }
2642
2643                         btrfs_mark_buffer_dirty(leaf);
2644                         btrfs_release_path(path);
2645
2646                         inode_inc_iversion(inode);
2647                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2648
2649                         /*
2650                          * we round up to the block size at eof when
2651                          * determining which extents to clone above,
2652                          * but shouldn't round up the file size
2653                          */
2654                         endoff = new_key.offset + datal;
2655                         if (endoff > destoff+olen)
2656                                 endoff = destoff+olen;
2657                         if (endoff > inode->i_size)
2658                                 btrfs_i_size_write(inode, endoff);
2659
2660                         ret = btrfs_update_inode(trans, root, inode);
2661                         if (ret) {
2662                                 btrfs_abort_transaction(trans, root, ret);
2663                                 btrfs_end_transaction(trans, root);
2664                                 goto out;
2665                         }
2666                         ret = btrfs_end_transaction(trans, root);
2667                 }
2668 next:
2669                 btrfs_release_path(path);
2670                 key.offset++;
2671         }
2672         ret = 0;
2673 out:
2674         btrfs_release_path(path);
2675         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len);
2676 out_unlock:
2677         mutex_unlock(&src->i_mutex);
2678         mutex_unlock(&inode->i_mutex);
2679         vfree(buf);
2680         btrfs_free_path(path);
2681 out_fput:
2682         fput(src_file);
2683 out_drop_write:
2684         mnt_drop_write_file(file);
2685         return ret;
2686 }
2687
2688 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2689 {
2690         struct btrfs_ioctl_clone_range_args args;
2691
2692         if (copy_from_user(&args, argp, sizeof(args)))
2693                 return -EFAULT;
2694         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2695                                  args.src_length, args.dest_offset);
2696 }
2697
2698 /*
2699  * there are many ways the trans_start and trans_end ioctls can lead
2700  * to deadlocks.  They should only be used by applications that
2701  * basically own the machine, and have a very in depth understanding
2702  * of all the possible deadlocks and enospc problems.
2703  */
2704 static long btrfs_ioctl_trans_start(struct file *file)
2705 {
2706         struct inode *inode = fdentry(file)->d_inode;
2707         struct btrfs_root *root = BTRFS_I(inode)->root;
2708         struct btrfs_trans_handle *trans;
2709         int ret;
2710
2711         ret = -EPERM;
2712         if (!capable(CAP_SYS_ADMIN))
2713                 goto out;
2714
2715         ret = -EINPROGRESS;
2716         if (file->private_data)
2717                 goto out;
2718
2719         ret = -EROFS;
2720         if (btrfs_root_readonly(root))
2721                 goto out;
2722
2723         ret = mnt_want_write_file(file);
2724         if (ret)
2725                 goto out;
2726
2727         atomic_inc(&root->fs_info->open_ioctl_trans);
2728
2729         ret = -ENOMEM;
2730         trans = btrfs_start_ioctl_transaction(root);
2731         if (IS_ERR(trans))
2732                 goto out_drop;
2733
2734         file->private_data = trans;
2735         return 0;
2736
2737 out_drop:
2738         atomic_dec(&root->fs_info->open_ioctl_trans);
2739         mnt_drop_write_file(file);
2740 out:
2741         return ret;
2742 }
2743
2744 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2745 {
2746         struct inode *inode = fdentry(file)->d_inode;
2747         struct btrfs_root *root = BTRFS_I(inode)->root;
2748         struct btrfs_root *new_root;
2749         struct btrfs_dir_item *di;
2750         struct btrfs_trans_handle *trans;
2751         struct btrfs_path *path;
2752         struct btrfs_key location;
2753         struct btrfs_disk_key disk_key;
2754         struct btrfs_super_block *disk_super;
2755         u64 features;
2756         u64 objectid = 0;
2757         u64 dir_id;
2758
2759         if (!capable(CAP_SYS_ADMIN))
2760                 return -EPERM;
2761
2762         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2763                 return -EFAULT;
2764
2765         if (!objectid)
2766                 objectid = root->root_key.objectid;
2767
2768         location.objectid = objectid;
2769         location.type = BTRFS_ROOT_ITEM_KEY;
2770         location.offset = (u64)-1;
2771
2772         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2773         if (IS_ERR(new_root))
2774                 return PTR_ERR(new_root);
2775
2776         if (btrfs_root_refs(&new_root->root_item) == 0)
2777                 return -ENOENT;
2778
2779         path = btrfs_alloc_path();
2780         if (!path)
2781                 return -ENOMEM;
2782         path->leave_spinning = 1;
2783
2784         trans = btrfs_start_transaction(root, 1);
2785         if (IS_ERR(trans)) {
2786                 btrfs_free_path(path);
2787                 return PTR_ERR(trans);
2788         }
2789
2790         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2791         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2792                                    dir_id, "default", 7, 1);
2793         if (IS_ERR_OR_NULL(di)) {
2794                 btrfs_free_path(path);
2795                 btrfs_end_transaction(trans, root);
2796                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2797                        "this isn't going to work\n");
2798                 return -ENOENT;
2799         }
2800
2801         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2802         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2803         btrfs_mark_buffer_dirty(path->nodes[0]);
2804         btrfs_free_path(path);
2805
2806         disk_super = root->fs_info->super_copy;
2807         features = btrfs_super_incompat_flags(disk_super);
2808         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2809                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2810                 btrfs_set_super_incompat_flags(disk_super, features);
2811         }
2812         btrfs_end_transaction(trans, root);
2813
2814         return 0;
2815 }
2816
2817 static void get_block_group_info(struct list_head *groups_list,
2818                                  struct btrfs_ioctl_space_info *space)
2819 {
2820         struct btrfs_block_group_cache *block_group;
2821
2822         space->total_bytes = 0;
2823         space->used_bytes = 0;
2824         space->flags = 0;
2825         list_for_each_entry(block_group, groups_list, list) {
2826                 space->flags = block_group->flags;
2827                 space->total_bytes += block_group->key.offset;
2828                 space->used_bytes +=
2829                         btrfs_block_group_used(&block_group->item);
2830         }
2831 }
2832
2833 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2834 {
2835         struct btrfs_ioctl_space_args space_args;
2836         struct btrfs_ioctl_space_info space;
2837         struct btrfs_ioctl_space_info *dest;
2838         struct btrfs_ioctl_space_info *dest_orig;
2839         struct btrfs_ioctl_space_info __user *user_dest;
2840         struct btrfs_space_info *info;
2841         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2842                        BTRFS_BLOCK_GROUP_SYSTEM,
2843                        BTRFS_BLOCK_GROUP_METADATA,
2844                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2845         int num_types = 4;
2846         int alloc_size;
2847         int ret = 0;
2848         u64 slot_count = 0;
2849         int i, c;
2850
2851         if (copy_from_user(&space_args,
2852                            (struct btrfs_ioctl_space_args __user *)arg,
2853                            sizeof(space_args)))
2854                 return -EFAULT;
2855
2856         for (i = 0; i < num_types; i++) {
2857                 struct btrfs_space_info *tmp;
2858
2859                 info = NULL;
2860                 rcu_read_lock();
2861                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2862                                         list) {
2863                         if (tmp->flags == types[i]) {
2864                                 info = tmp;
2865                                 break;
2866                         }
2867                 }
2868                 rcu_read_unlock();
2869
2870                 if (!info)
2871                         continue;
2872
2873                 down_read(&info->groups_sem);
2874                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2875                         if (!list_empty(&info->block_groups[c]))
2876                                 slot_count++;
2877                 }
2878                 up_read(&info->groups_sem);
2879         }
2880
2881         /* space_slots == 0 means they are asking for a count */
2882         if (space_args.space_slots == 0) {
2883                 space_args.total_spaces = slot_count;
2884                 goto out;
2885         }
2886
2887         slot_count = min_t(u64, space_args.space_slots, slot_count);
2888
2889         alloc_size = sizeof(*dest) * slot_count;
2890
2891         /* we generally have at most 6 or so space infos, one for each raid
2892          * level.  So, a whole page should be more than enough for everyone
2893          */
2894         if (alloc_size > PAGE_CACHE_SIZE)
2895                 return -ENOMEM;
2896
2897         space_args.total_spaces = 0;
2898         dest = kmalloc(alloc_size, GFP_NOFS);
2899         if (!dest)
2900                 return -ENOMEM;
2901         dest_orig = dest;
2902
2903         /* now we have a buffer to copy into */
2904         for (i = 0; i < num_types; i++) {
2905                 struct btrfs_space_info *tmp;
2906
2907                 if (!slot_count)
2908                         break;
2909
2910                 info = NULL;
2911                 rcu_read_lock();
2912                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2913                                         list) {
2914                         if (tmp->flags == types[i]) {
2915                                 info = tmp;
2916                                 break;
2917                         }
2918                 }
2919                 rcu_read_unlock();
2920
2921                 if (!info)
2922                         continue;
2923                 down_read(&info->groups_sem);
2924                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2925                         if (!list_empty(&info->block_groups[c])) {
2926                                 get_block_group_info(&info->block_groups[c],
2927                                                      &space);
2928                                 memcpy(dest, &space, sizeof(space));
2929                                 dest++;
2930                                 space_args.total_spaces++;
2931                                 slot_count--;
2932                         }
2933                         if (!slot_count)
2934                                 break;
2935                 }
2936                 up_read(&info->groups_sem);
2937         }
2938
2939         user_dest = (struct btrfs_ioctl_space_info __user *)
2940                 (arg + sizeof(struct btrfs_ioctl_space_args));
2941
2942         if (copy_to_user(user_dest, dest_orig, alloc_size))
2943                 ret = -EFAULT;
2944
2945         kfree(dest_orig);
2946 out:
2947         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2948                 ret = -EFAULT;
2949
2950         return ret;
2951 }
2952
2953 /*
2954  * there are many ways the trans_start and trans_end ioctls can lead
2955  * to deadlocks.  They should only be used by applications that
2956  * basically own the machine, and have a very in depth understanding
2957  * of all the possible deadlocks and enospc problems.
2958  */
2959 long btrfs_ioctl_trans_end(struct file *file)
2960 {
2961         struct inode *inode = fdentry(file)->d_inode;
2962         struct btrfs_root *root = BTRFS_I(inode)->root;
2963         struct btrfs_trans_handle *trans;
2964
2965         trans = file->private_data;
2966         if (!trans)
2967                 return -EINVAL;
2968         file->private_data = NULL;
2969
2970         btrfs_end_transaction(trans, root);
2971
2972         atomic_dec(&root->fs_info->open_ioctl_trans);
2973
2974         mnt_drop_write_file(file);
2975         return 0;
2976 }
2977
2978 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2979 {
2980         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2981         struct btrfs_trans_handle *trans;
2982         u64 transid;
2983         int ret;
2984
2985         trans = btrfs_start_transaction(root, 0);
2986         if (IS_ERR(trans))
2987                 return PTR_ERR(trans);
2988         transid = trans->transid;
2989         ret = btrfs_commit_transaction_async(trans, root, 0);
2990         if (ret) {
2991                 btrfs_end_transaction(trans, root);
2992                 return ret;
2993         }
2994
2995         if (argp)
2996                 if (copy_to_user(argp, &transid, sizeof(transid)))
2997                         return -EFAULT;
2998         return 0;
2999 }
3000
3001 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
3002 {
3003         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
3004         u64 transid;
3005
3006         if (argp) {
3007                 if (copy_from_user(&transid, argp, sizeof(transid)))
3008                         return -EFAULT;
3009         } else {
3010                 transid = 0;  /* current trans */
3011         }
3012         return btrfs_wait_for_commit(root, transid);
3013 }
3014
3015 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
3016 {
3017         int ret;
3018         struct btrfs_ioctl_scrub_args *sa;
3019
3020         if (!capable(CAP_SYS_ADMIN))
3021                 return -EPERM;
3022
3023         sa = memdup_user(arg, sizeof(*sa));
3024         if (IS_ERR(sa))
3025                 return PTR_ERR(sa);
3026
3027         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
3028                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
3029
3030         if (copy_to_user(arg, sa, sizeof(*sa)))
3031                 ret = -EFAULT;
3032
3033         kfree(sa);
3034         return ret;
3035 }
3036
3037 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3038 {
3039         if (!capable(CAP_SYS_ADMIN))
3040                 return -EPERM;
3041
3042         return btrfs_scrub_cancel(root);
3043 }
3044
3045 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3046                                        void __user *arg)
3047 {
3048         struct btrfs_ioctl_scrub_args *sa;
3049         int ret;
3050
3051         if (!capable(CAP_SYS_ADMIN))
3052                 return -EPERM;
3053
3054         sa = memdup_user(arg, sizeof(*sa));
3055         if (IS_ERR(sa))
3056                 return PTR_ERR(sa);
3057
3058         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3059
3060         if (copy_to_user(arg, sa, sizeof(*sa)))
3061                 ret = -EFAULT;
3062
3063         kfree(sa);
3064         return ret;
3065 }
3066
3067 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3068                                       void __user *arg)
3069 {
3070         struct btrfs_ioctl_get_dev_stats *sa;
3071         int ret;
3072
3073         sa = memdup_user(arg, sizeof(*sa));
3074         if (IS_ERR(sa))
3075                 return PTR_ERR(sa);
3076
3077         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3078                 kfree(sa);
3079                 return -EPERM;
3080         }
3081
3082         ret = btrfs_get_dev_stats(root, sa);
3083
3084         if (copy_to_user(arg, sa, sizeof(*sa)))
3085                 ret = -EFAULT;
3086
3087         kfree(sa);
3088         return ret;
3089 }
3090
3091 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3092 {
3093         int ret = 0;
3094         int i;
3095         u64 rel_ptr;
3096         int size;
3097         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3098         struct inode_fs_paths *ipath = NULL;
3099         struct btrfs_path *path;
3100
3101         if (!capable(CAP_SYS_ADMIN))
3102                 return -EPERM;
3103
3104         path = btrfs_alloc_path();
3105         if (!path) {
3106                 ret = -ENOMEM;
3107                 goto out;
3108         }
3109
3110         ipa = memdup_user(arg, sizeof(*ipa));
3111         if (IS_ERR(ipa)) {
3112                 ret = PTR_ERR(ipa);
3113                 ipa = NULL;
3114                 goto out;
3115         }
3116
3117         size = min_t(u32, ipa->size, 4096);
3118         ipath = init_ipath(size, root, path);
3119         if (IS_ERR(ipath)) {
3120                 ret = PTR_ERR(ipath);
3121                 ipath = NULL;
3122                 goto out;
3123         }
3124
3125         ret = paths_from_inode(ipa->inum, ipath);
3126         if (ret < 0)
3127                 goto out;
3128
3129         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3130                 rel_ptr = ipath->fspath->val[i] -
3131                           (u64)(unsigned long)ipath->fspath->val;
3132                 ipath->fspath->val[i] = rel_ptr;
3133         }
3134
3135         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3136                            (void *)(unsigned long)ipath->fspath, size);
3137         if (ret) {
3138                 ret = -EFAULT;
3139                 goto out;
3140         }
3141
3142 out:
3143         btrfs_free_path(path);
3144         free_ipath(ipath);
3145         kfree(ipa);
3146
3147         return ret;
3148 }
3149
3150 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3151 {
3152         struct btrfs_data_container *inodes = ctx;
3153         const size_t c = 3 * sizeof(u64);
3154
3155         if (inodes->bytes_left >= c) {
3156                 inodes->bytes_left -= c;
3157                 inodes->val[inodes->elem_cnt] = inum;
3158                 inodes->val[inodes->elem_cnt + 1] = offset;
3159                 inodes->val[inodes->elem_cnt + 2] = root;
3160                 inodes->elem_cnt += 3;
3161         } else {
3162                 inodes->bytes_missing += c - inodes->bytes_left;
3163                 inodes->bytes_left = 0;
3164                 inodes->elem_missed += 3;
3165         }
3166
3167         return 0;
3168 }
3169
3170 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3171                                         void __user *arg)
3172 {
3173         int ret = 0;
3174         int size;
3175         u64 extent_item_pos;
3176         struct btrfs_ioctl_logical_ino_args *loi;
3177         struct btrfs_data_container *inodes = NULL;
3178         struct btrfs_path *path = NULL;
3179         struct btrfs_key key;
3180
3181         if (!capable(CAP_SYS_ADMIN))
3182                 return -EPERM;
3183
3184         loi = memdup_user(arg, sizeof(*loi));
3185         if (IS_ERR(loi)) {
3186                 ret = PTR_ERR(loi);
3187                 loi = NULL;
3188                 goto out;
3189         }
3190
3191         path = btrfs_alloc_path();
3192         if (!path) {
3193                 ret = -ENOMEM;
3194                 goto out;
3195         }
3196
3197         size = min_t(u32, loi->size, 4096);
3198         inodes = init_data_container(size);
3199         if (IS_ERR(inodes)) {
3200                 ret = PTR_ERR(inodes);
3201                 inodes = NULL;
3202                 goto out;
3203         }
3204
3205         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
3206         btrfs_release_path(path);
3207
3208         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
3209                 ret = -ENOENT;
3210         if (ret < 0)
3211                 goto out;
3212
3213         extent_item_pos = loi->logical - key.objectid;
3214         ret = iterate_extent_inodes(root->fs_info, key.objectid,
3215                                         extent_item_pos, 0, build_ino_list,
3216                                         inodes);
3217
3218         if (ret < 0)
3219                 goto out;
3220
3221         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3222                            (void *)(unsigned long)inodes, size);
3223         if (ret)
3224                 ret = -EFAULT;
3225
3226 out:
3227         btrfs_free_path(path);
3228         kfree(inodes);
3229         kfree(loi);
3230
3231         return ret;
3232 }
3233
3234 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3235                                struct btrfs_ioctl_balance_args *bargs)
3236 {
3237         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3238
3239         bargs->flags = bctl->flags;
3240
3241         if (atomic_read(&fs_info->balance_running))
3242                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3243         if (atomic_read(&fs_info->balance_pause_req))
3244                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3245         if (atomic_read(&fs_info->balance_cancel_req))
3246                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3247
3248         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3249         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3250         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3251
3252         if (lock) {
3253                 spin_lock(&fs_info->balance_lock);
3254                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3255                 spin_unlock(&fs_info->balance_lock);
3256         } else {
3257                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3258         }
3259 }
3260
3261 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3262 {
3263         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3264         struct btrfs_fs_info *fs_info = root->fs_info;
3265         struct btrfs_ioctl_balance_args *bargs;
3266         struct btrfs_balance_control *bctl;
3267         int ret;
3268
3269         if (!capable(CAP_SYS_ADMIN))
3270                 return -EPERM;
3271
3272         if (fs_info->sb->s_flags & MS_RDONLY)
3273                 return -EROFS;
3274
3275         ret = mnt_want_write(file->f_path.mnt);
3276         if (ret)
3277                 return ret;
3278
3279         mutex_lock(&fs_info->volume_mutex);
3280         mutex_lock(&fs_info->balance_mutex);
3281
3282         if (arg) {
3283                 bargs = memdup_user(arg, sizeof(*bargs));
3284                 if (IS_ERR(bargs)) {
3285                         ret = PTR_ERR(bargs);
3286                         goto out;
3287                 }
3288
3289                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3290                         if (!fs_info->balance_ctl) {
3291                                 ret = -ENOTCONN;
3292                                 goto out_bargs;
3293                         }
3294
3295                         bctl = fs_info->balance_ctl;
3296                         spin_lock(&fs_info->balance_lock);
3297                         bctl->flags |= BTRFS_BALANCE_RESUME;
3298                         spin_unlock(&fs_info->balance_lock);
3299
3300                         goto do_balance;
3301                 }
3302         } else {
3303                 bargs = NULL;
3304         }
3305
3306         if (fs_info->balance_ctl) {
3307                 ret = -EINPROGRESS;
3308                 goto out_bargs;
3309         }
3310
3311         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3312         if (!bctl) {
3313                 ret = -ENOMEM;
3314                 goto out_bargs;
3315         }
3316
3317         bctl->fs_info = fs_info;
3318         if (arg) {
3319                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3320                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3321                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3322
3323                 bctl->flags = bargs->flags;
3324         } else {
3325                 /* balance everything - no filters */
3326                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3327         }
3328
3329 do_balance:
3330         ret = btrfs_balance(bctl, bargs);
3331         /*
3332          * bctl is freed in __cancel_balance or in free_fs_info if
3333          * restriper was paused all the way until unmount
3334          */
3335         if (arg) {
3336                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3337                         ret = -EFAULT;
3338         }
3339
3340 out_bargs:
3341         kfree(bargs);
3342 out:
3343         mutex_unlock(&fs_info->balance_mutex);
3344         mutex_unlock(&fs_info->volume_mutex);
3345         mnt_drop_write(file->f_path.mnt);
3346         return ret;
3347 }
3348
3349 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3350 {
3351         if (!capable(CAP_SYS_ADMIN))
3352                 return -EPERM;
3353
3354         switch (cmd) {
3355         case BTRFS_BALANCE_CTL_PAUSE:
3356                 return btrfs_pause_balance(root->fs_info);
3357         case BTRFS_BALANCE_CTL_CANCEL:
3358                 return btrfs_cancel_balance(root->fs_info);
3359         }
3360
3361         return -EINVAL;
3362 }
3363
3364 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3365                                          void __user *arg)
3366 {
3367         struct btrfs_fs_info *fs_info = root->fs_info;
3368         struct btrfs_ioctl_balance_args *bargs;
3369         int ret = 0;
3370
3371         if (!capable(CAP_SYS_ADMIN))
3372                 return -EPERM;
3373
3374         mutex_lock(&fs_info->balance_mutex);
3375         if (!fs_info->balance_ctl) {
3376                 ret = -ENOTCONN;
3377                 goto out;
3378         }
3379
3380         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3381         if (!bargs) {
3382                 ret = -ENOMEM;
3383                 goto out;
3384         }
3385
3386         update_ioctl_balance_args(fs_info, 1, bargs);
3387
3388         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3389                 ret = -EFAULT;
3390
3391         kfree(bargs);
3392 out:
3393         mutex_unlock(&fs_info->balance_mutex);
3394         return ret;
3395 }
3396
3397 long btrfs_ioctl(struct file *file, unsigned int
3398                 cmd, unsigned long arg)
3399 {
3400         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3401         void __user *argp = (void __user *)arg;
3402
3403         switch (cmd) {
3404         case FS_IOC_GETFLAGS:
3405                 return btrfs_ioctl_getflags(file, argp);
3406         case FS_IOC_SETFLAGS:
3407                 return btrfs_ioctl_setflags(file, argp);
3408         case FS_IOC_GETVERSION:
3409                 return btrfs_ioctl_getversion(file, argp);
3410         case FITRIM:
3411                 return btrfs_ioctl_fitrim(file, argp);
3412         case BTRFS_IOC_SNAP_CREATE:
3413                 return btrfs_ioctl_snap_create(file, argp, 0);
3414         case BTRFS_IOC_SNAP_CREATE_V2:
3415                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3416         case BTRFS_IOC_SUBVOL_CREATE:
3417                 return btrfs_ioctl_snap_create(file, argp, 1);
3418         case BTRFS_IOC_SNAP_DESTROY:
3419                 return btrfs_ioctl_snap_destroy(file, argp);
3420         case BTRFS_IOC_SUBVOL_GETFLAGS:
3421                 return btrfs_ioctl_subvol_getflags(file, argp);
3422         case BTRFS_IOC_SUBVOL_SETFLAGS:
3423                 return btrfs_ioctl_subvol_setflags(file, argp);
3424         case BTRFS_IOC_DEFAULT_SUBVOL:
3425                 return btrfs_ioctl_default_subvol(file, argp);
3426         case BTRFS_IOC_DEFRAG:
3427                 return btrfs_ioctl_defrag(file, NULL);
3428         case BTRFS_IOC_DEFRAG_RANGE:
3429                 return btrfs_ioctl_defrag(file, argp);
3430         case BTRFS_IOC_RESIZE:
3431                 return btrfs_ioctl_resize(root, argp);
3432         case BTRFS_IOC_ADD_DEV:
3433                 return btrfs_ioctl_add_dev(root, argp);
3434         case BTRFS_IOC_RM_DEV:
3435                 return btrfs_ioctl_rm_dev(root, argp);
3436         case BTRFS_IOC_FS_INFO:
3437                 return btrfs_ioctl_fs_info(root, argp);
3438         case BTRFS_IOC_DEV_INFO:
3439                 return btrfs_ioctl_dev_info(root, argp);
3440         case BTRFS_IOC_BALANCE:
3441                 return btrfs_ioctl_balance(file, NULL);
3442         case BTRFS_IOC_CLONE:
3443                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3444         case BTRFS_IOC_CLONE_RANGE:
3445                 return btrfs_ioctl_clone_range(file, argp);
3446         case BTRFS_IOC_TRANS_START:
3447                 return btrfs_ioctl_trans_start(file);
3448         case BTRFS_IOC_TRANS_END:
3449                 return btrfs_ioctl_trans_end(file);
3450         case BTRFS_IOC_TREE_SEARCH:
3451                 return btrfs_ioctl_tree_search(file, argp);
3452         case BTRFS_IOC_INO_LOOKUP:
3453                 return btrfs_ioctl_ino_lookup(file, argp);
3454         case BTRFS_IOC_INO_PATHS:
3455                 return btrfs_ioctl_ino_to_path(root, argp);
3456         case BTRFS_IOC_LOGICAL_INO:
3457                 return btrfs_ioctl_logical_to_ino(root, argp);
3458         case BTRFS_IOC_SPACE_INFO:
3459                 return btrfs_ioctl_space_info(root, argp);
3460         case BTRFS_IOC_SYNC:
3461                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3462                 return 0;
3463         case BTRFS_IOC_START_SYNC:
3464                 return btrfs_ioctl_start_sync(file, argp);
3465         case BTRFS_IOC_WAIT_SYNC:
3466                 return btrfs_ioctl_wait_sync(file, argp);
3467         case BTRFS_IOC_SCRUB:
3468                 return btrfs_ioctl_scrub(root, argp);
3469         case BTRFS_IOC_SCRUB_CANCEL:
3470                 return btrfs_ioctl_scrub_cancel(root, argp);
3471         case BTRFS_IOC_SCRUB_PROGRESS:
3472                 return btrfs_ioctl_scrub_progress(root, argp);
3473         case BTRFS_IOC_BALANCE_V2:
3474                 return btrfs_ioctl_balance(file, argp);
3475         case BTRFS_IOC_BALANCE_CTL:
3476                 return btrfs_ioctl_balance_ctl(root, arg);
3477         case BTRFS_IOC_BALANCE_PROGRESS:
3478                 return btrfs_ioctl_balance_progress(root, argp);
3479         case BTRFS_IOC_GET_DEV_STATS:
3480                 return btrfs_ioctl_get_dev_stats(root, argp);
3481         }
3482
3483         return -ENOTTY;
3484 }