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