fd252fff4c6666d19ebc53553c96d17d8567d37a
[pandora-kernel.git] / 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  * Unlike extN we don't have any flags we don't want to inherit currently.
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 (S_ISREG(inode->i_mode))
132                 flags &= ~BTRFS_INODE_DIRSYNC;
133         else if (!S_ISDIR(inode->i_mode))
134                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
135
136         BTRFS_I(inode)->flags = flags;
137         btrfs_update_iflags(inode);
138 }
139
140 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
141 {
142         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
143         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
144
145         if (copy_to_user(arg, &flags, sizeof(flags)))
146                 return -EFAULT;
147         return 0;
148 }
149
150 static int check_flags(unsigned int flags)
151 {
152         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
153                       FS_NOATIME_FL | FS_NODUMP_FL | \
154                       FS_SYNC_FL | FS_DIRSYNC_FL | \
155                       FS_NOCOMP_FL | FS_COMPR_FL |
156                       FS_NOCOW_FL))
157                 return -EOPNOTSUPP;
158
159         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
160                 return -EINVAL;
161
162         return 0;
163 }
164
165 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
166 {
167         struct inode *inode = file->f_path.dentry->d_inode;
168         struct btrfs_inode *ip = BTRFS_I(inode);
169         struct btrfs_root *root = ip->root;
170         struct btrfs_trans_handle *trans;
171         unsigned int flags, oldflags;
172         int ret;
173
174         if (btrfs_root_readonly(root))
175                 return -EROFS;
176
177         if (copy_from_user(&flags, arg, sizeof(flags)))
178                 return -EFAULT;
179
180         ret = check_flags(flags);
181         if (ret)
182                 return ret;
183
184         if (!inode_owner_or_capable(inode))
185                 return -EACCES;
186
187         mutex_lock(&inode->i_mutex);
188
189         flags = btrfs_mask_flags(inode->i_mode, flags);
190         oldflags = btrfs_flags_to_ioctl(ip->flags);
191         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
192                 if (!capable(CAP_LINUX_IMMUTABLE)) {
193                         ret = -EPERM;
194                         goto out_unlock;
195                 }
196         }
197
198         ret = mnt_want_write(file->f_path.mnt);
199         if (ret)
200                 goto out_unlock;
201
202         if (flags & FS_SYNC_FL)
203                 ip->flags |= BTRFS_INODE_SYNC;
204         else
205                 ip->flags &= ~BTRFS_INODE_SYNC;
206         if (flags & FS_IMMUTABLE_FL)
207                 ip->flags |= BTRFS_INODE_IMMUTABLE;
208         else
209                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
210         if (flags & FS_APPEND_FL)
211                 ip->flags |= BTRFS_INODE_APPEND;
212         else
213                 ip->flags &= ~BTRFS_INODE_APPEND;
214         if (flags & FS_NODUMP_FL)
215                 ip->flags |= BTRFS_INODE_NODUMP;
216         else
217                 ip->flags &= ~BTRFS_INODE_NODUMP;
218         if (flags & FS_NOATIME_FL)
219                 ip->flags |= BTRFS_INODE_NOATIME;
220         else
221                 ip->flags &= ~BTRFS_INODE_NOATIME;
222         if (flags & FS_DIRSYNC_FL)
223                 ip->flags |= BTRFS_INODE_DIRSYNC;
224         else
225                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
226         if (flags & FS_NOCOW_FL)
227                 ip->flags |= BTRFS_INODE_NODATACOW;
228         else
229                 ip->flags &= ~BTRFS_INODE_NODATACOW;
230
231         /*
232          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
233          * flag may be changed automatically if compression code won't make
234          * things smaller.
235          */
236         if (flags & FS_NOCOMP_FL) {
237                 ip->flags &= ~BTRFS_INODE_COMPRESS;
238                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
239         } else if (flags & FS_COMPR_FL) {
240                 ip->flags |= BTRFS_INODE_COMPRESS;
241                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
242         } else {
243                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
244         }
245
246         trans = btrfs_join_transaction(root);
247         BUG_ON(IS_ERR(trans));
248
249         ret = btrfs_update_inode(trans, root, inode);
250         BUG_ON(ret);
251
252         btrfs_update_iflags(inode);
253         inode->i_ctime = CURRENT_TIME;
254         btrfs_end_transaction(trans, root);
255
256         mnt_drop_write(file->f_path.mnt);
257
258         ret = 0;
259  out_unlock:
260         mutex_unlock(&inode->i_mutex);
261         return ret;
262 }
263
264 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
265 {
266         struct inode *inode = file->f_path.dentry->d_inode;
267
268         return put_user(inode->i_generation, arg);
269 }
270
271 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
272 {
273         struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info;
274         struct btrfs_fs_info *fs_info = root->fs_info;
275         struct btrfs_device *device;
276         struct request_queue *q;
277         struct fstrim_range range;
278         u64 minlen = ULLONG_MAX;
279         u64 num_devices = 0;
280         int ret;
281
282         if (!capable(CAP_SYS_ADMIN))
283                 return -EPERM;
284
285         rcu_read_lock();
286         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
287                                 dev_list) {
288                 if (!device->bdev)
289                         continue;
290                 q = bdev_get_queue(device->bdev);
291                 if (blk_queue_discard(q)) {
292                         num_devices++;
293                         minlen = min((u64)q->limits.discard_granularity,
294                                      minlen);
295                 }
296         }
297         rcu_read_unlock();
298         if (!num_devices)
299                 return -EOPNOTSUPP;
300
301         if (copy_from_user(&range, arg, sizeof(range)))
302                 return -EFAULT;
303
304         range.minlen = max(range.minlen, minlen);
305         ret = btrfs_trim_fs(root, &range);
306         if (ret < 0)
307                 return ret;
308
309         if (copy_to_user(arg, &range, sizeof(range)))
310                 return -EFAULT;
311
312         return 0;
313 }
314
315 static noinline int create_subvol(struct btrfs_root *root,
316                                   struct dentry *dentry,
317                                   char *name, int namelen,
318                                   u64 *async_transid)
319 {
320         struct btrfs_trans_handle *trans;
321         struct btrfs_key key;
322         struct btrfs_root_item root_item;
323         struct btrfs_inode_item *inode_item;
324         struct extent_buffer *leaf;
325         struct btrfs_root *new_root;
326         struct dentry *parent = dget_parent(dentry);
327         struct inode *dir;
328         int ret;
329         int err;
330         u64 objectid;
331         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
332         u64 index = 0;
333
334         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
335         if (ret) {
336                 dput(parent);
337                 return ret;
338         }
339
340         dir = parent->d_inode;
341
342         /*
343          * 1 - inode item
344          * 2 - refs
345          * 1 - root item
346          * 2 - dir items
347          */
348         trans = btrfs_start_transaction(root, 6);
349         if (IS_ERR(trans)) {
350                 dput(parent);
351                 return PTR_ERR(trans);
352         }
353
354         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
355                                       0, objectid, NULL, 0, 0, 0);
356         if (IS_ERR(leaf)) {
357                 ret = PTR_ERR(leaf);
358                 goto fail;
359         }
360
361         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
362         btrfs_set_header_bytenr(leaf, leaf->start);
363         btrfs_set_header_generation(leaf, trans->transid);
364         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
365         btrfs_set_header_owner(leaf, objectid);
366
367         write_extent_buffer(leaf, root->fs_info->fsid,
368                             (unsigned long)btrfs_header_fsid(leaf),
369                             BTRFS_FSID_SIZE);
370         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
371                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
372                             BTRFS_UUID_SIZE);
373         btrfs_mark_buffer_dirty(leaf);
374
375         inode_item = &root_item.inode;
376         memset(inode_item, 0, sizeof(*inode_item));
377         inode_item->generation = cpu_to_le64(1);
378         inode_item->size = cpu_to_le64(3);
379         inode_item->nlink = cpu_to_le32(1);
380         inode_item->nbytes = cpu_to_le64(root->leafsize);
381         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
382
383         root_item.flags = 0;
384         root_item.byte_limit = 0;
385         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
386
387         btrfs_set_root_bytenr(&root_item, leaf->start);
388         btrfs_set_root_generation(&root_item, trans->transid);
389         btrfs_set_root_level(&root_item, 0);
390         btrfs_set_root_refs(&root_item, 1);
391         btrfs_set_root_used(&root_item, leaf->len);
392         btrfs_set_root_last_snapshot(&root_item, 0);
393
394         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
395         root_item.drop_level = 0;
396
397         btrfs_tree_unlock(leaf);
398         free_extent_buffer(leaf);
399         leaf = NULL;
400
401         btrfs_set_root_dirid(&root_item, new_dirid);
402
403         key.objectid = objectid;
404         key.offset = 0;
405         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
406         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
407                                 &root_item);
408         if (ret)
409                 goto fail;
410
411         key.offset = (u64)-1;
412         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
413         BUG_ON(IS_ERR(new_root));
414
415         btrfs_record_root_in_trans(trans, new_root);
416
417         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
418         /*
419          * insert the directory item
420          */
421         ret = btrfs_set_inode_index(dir, &index);
422         BUG_ON(ret);
423
424         ret = btrfs_insert_dir_item(trans, root,
425                                     name, namelen, dir, &key,
426                                     BTRFS_FT_DIR, index);
427         if (ret)
428                 goto fail;
429
430         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
431         ret = btrfs_update_inode(trans, root, dir);
432         BUG_ON(ret);
433
434         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
435                                  objectid, root->root_key.objectid,
436                                  btrfs_ino(dir), index, name, namelen);
437
438         BUG_ON(ret);
439
440         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
441 fail:
442         dput(parent);
443         if (async_transid) {
444                 *async_transid = trans->transid;
445                 err = btrfs_commit_transaction_async(trans, root, 1);
446         } else {
447                 err = btrfs_commit_transaction(trans, root);
448         }
449         if (err && !ret)
450                 ret = err;
451         return ret;
452 }
453
454 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
455                            char *name, int namelen, u64 *async_transid,
456                            bool readonly)
457 {
458         struct inode *inode;
459         struct dentry *parent;
460         struct btrfs_pending_snapshot *pending_snapshot;
461         struct btrfs_trans_handle *trans;
462         int ret;
463
464         if (!root->ref_cows)
465                 return -EINVAL;
466
467         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
468         if (!pending_snapshot)
469                 return -ENOMEM;
470
471         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
472         pending_snapshot->dentry = dentry;
473         pending_snapshot->root = root;
474         pending_snapshot->readonly = readonly;
475
476         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
477         if (IS_ERR(trans)) {
478                 ret = PTR_ERR(trans);
479                 goto fail;
480         }
481
482         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
483         BUG_ON(ret);
484
485         spin_lock(&root->fs_info->trans_lock);
486         list_add(&pending_snapshot->list,
487                  &trans->transaction->pending_snapshots);
488         spin_unlock(&root->fs_info->trans_lock);
489         if (async_transid) {
490                 *async_transid = trans->transid;
491                 ret = btrfs_commit_transaction_async(trans,
492                                      root->fs_info->extent_root, 1);
493         } else {
494                 ret = btrfs_commit_transaction(trans,
495                                                root->fs_info->extent_root);
496         }
497         BUG_ON(ret);
498
499         ret = pending_snapshot->error;
500         if (ret)
501                 goto fail;
502
503         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
504         if (ret)
505                 goto fail;
506
507         parent = dget_parent(dentry);
508         inode = btrfs_lookup_dentry(parent->d_inode, dentry);
509         dput(parent);
510         if (IS_ERR(inode)) {
511                 ret = PTR_ERR(inode);
512                 goto fail;
513         }
514         BUG_ON(!inode);
515         d_instantiate(dentry, inode);
516         ret = 0;
517 fail:
518         kfree(pending_snapshot);
519         return ret;
520 }
521
522 /*  copy of check_sticky in fs/namei.c()
523 * It's inline, so penalty for filesystems that don't use sticky bit is
524 * minimal.
525 */
526 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
527 {
528         uid_t fsuid = current_fsuid();
529
530         if (!(dir->i_mode & S_ISVTX))
531                 return 0;
532         if (inode->i_uid == fsuid)
533                 return 0;
534         if (dir->i_uid == fsuid)
535                 return 0;
536         return !capable(CAP_FOWNER);
537 }
538
539 /*  copy of may_delete in fs/namei.c()
540  *      Check whether we can remove a link victim from directory dir, check
541  *  whether the type of victim is right.
542  *  1. We can't do it if dir is read-only (done in permission())
543  *  2. We should have write and exec permissions on dir
544  *  3. We can't remove anything from append-only dir
545  *  4. We can't do anything with immutable dir (done in permission())
546  *  5. If the sticky bit on dir is set we should either
547  *      a. be owner of dir, or
548  *      b. be owner of victim, or
549  *      c. have CAP_FOWNER capability
550  *  6. If the victim is append-only or immutable we can't do antyhing with
551  *     links pointing to it.
552  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
553  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
554  *  9. We can't remove a root or mountpoint.
555  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
556  *     nfs_async_unlink().
557  */
558
559 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
560 {
561         int error;
562
563         if (!victim->d_inode)
564                 return -ENOENT;
565
566         BUG_ON(victim->d_parent->d_inode != dir);
567         audit_inode_child(victim, dir);
568
569         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
570         if (error)
571                 return error;
572         if (IS_APPEND(dir))
573                 return -EPERM;
574         if (btrfs_check_sticky(dir, victim->d_inode)||
575                 IS_APPEND(victim->d_inode)||
576             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
577                 return -EPERM;
578         if (isdir) {
579                 if (!S_ISDIR(victim->d_inode->i_mode))
580                         return -ENOTDIR;
581                 if (IS_ROOT(victim))
582                         return -EBUSY;
583         } else if (S_ISDIR(victim->d_inode->i_mode))
584                 return -EISDIR;
585         if (IS_DEADDIR(dir))
586                 return -ENOENT;
587         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
588                 return -EBUSY;
589         return 0;
590 }
591
592 /* copy of may_create in fs/namei.c() */
593 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
594 {
595         if (child->d_inode)
596                 return -EEXIST;
597         if (IS_DEADDIR(dir))
598                 return -ENOENT;
599         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
600 }
601
602 /*
603  * Create a new subvolume below @parent.  This is largely modeled after
604  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
605  * inside this filesystem so it's quite a bit simpler.
606  */
607 static noinline int btrfs_mksubvol(struct path *parent,
608                                    char *name, int namelen,
609                                    struct btrfs_root *snap_src,
610                                    u64 *async_transid, bool readonly)
611 {
612         struct inode *dir  = parent->dentry->d_inode;
613         struct dentry *dentry;
614         int error;
615
616         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
617
618         dentry = lookup_one_len(name, parent->dentry, namelen);
619         error = PTR_ERR(dentry);
620         if (IS_ERR(dentry))
621                 goto out_unlock;
622
623         error = -EEXIST;
624         if (dentry->d_inode)
625                 goto out_dput;
626
627         error = mnt_want_write(parent->mnt);
628         if (error)
629                 goto out_dput;
630
631         error = btrfs_may_create(dir, dentry);
632         if (error)
633                 goto out_drop_write;
634
635         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
636
637         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
638                 goto out_up_read;
639
640         if (snap_src) {
641                 error = create_snapshot(snap_src, dentry,
642                                         name, namelen, async_transid, readonly);
643         } else {
644                 error = create_subvol(BTRFS_I(dir)->root, dentry,
645                                       name, namelen, async_transid);
646         }
647         if (!error)
648                 fsnotify_mkdir(dir, dentry);
649 out_up_read:
650         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
651 out_drop_write:
652         mnt_drop_write(parent->mnt);
653 out_dput:
654         dput(dentry);
655 out_unlock:
656         mutex_unlock(&dir->i_mutex);
657         return error;
658 }
659
660 /*
661  * When we're defragging a range, we don't want to kick it off again
662  * if it is really just waiting for delalloc to send it down.
663  * If we find a nice big extent or delalloc range for the bytes in the
664  * file you want to defrag, we return 0 to let you know to skip this
665  * part of the file
666  */
667 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
668 {
669         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
670         struct extent_map *em = NULL;
671         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
672         u64 end;
673
674         read_lock(&em_tree->lock);
675         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
676         read_unlock(&em_tree->lock);
677
678         if (em) {
679                 end = extent_map_end(em);
680                 free_extent_map(em);
681                 if (end - offset > thresh)
682                         return 0;
683         }
684         /* if we already have a nice delalloc here, just stop */
685         thresh /= 2;
686         end = count_range_bits(io_tree, &offset, offset + thresh,
687                                thresh, EXTENT_DELALLOC, 1);
688         if (end >= thresh)
689                 return 0;
690         return 1;
691 }
692
693 /*
694  * helper function to walk through a file and find extents
695  * newer than a specific transid, and smaller than thresh.
696  *
697  * This is used by the defragging code to find new and small
698  * extents
699  */
700 static int find_new_extents(struct btrfs_root *root,
701                             struct inode *inode, u64 newer_than,
702                             u64 *off, int thresh)
703 {
704         struct btrfs_path *path;
705         struct btrfs_key min_key;
706         struct btrfs_key max_key;
707         struct extent_buffer *leaf;
708         struct btrfs_file_extent_item *extent;
709         int type;
710         int ret;
711         u64 ino = btrfs_ino(inode);
712
713         path = btrfs_alloc_path();
714         if (!path)
715                 return -ENOMEM;
716
717         min_key.objectid = ino;
718         min_key.type = BTRFS_EXTENT_DATA_KEY;
719         min_key.offset = *off;
720
721         max_key.objectid = ino;
722         max_key.type = (u8)-1;
723         max_key.offset = (u64)-1;
724
725         path->keep_locks = 1;
726
727         while(1) {
728                 ret = btrfs_search_forward(root, &min_key, &max_key,
729                                            path, 0, newer_than);
730                 if (ret != 0)
731                         goto none;
732                 if (min_key.objectid != ino)
733                         goto none;
734                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
735                         goto none;
736
737                 leaf = path->nodes[0];
738                 extent = btrfs_item_ptr(leaf, path->slots[0],
739                                         struct btrfs_file_extent_item);
740
741                 type = btrfs_file_extent_type(leaf, extent);
742                 if (type == BTRFS_FILE_EXTENT_REG &&
743                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
744                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
745                         *off = min_key.offset;
746                         btrfs_free_path(path);
747                         return 0;
748                 }
749
750                 if (min_key.offset == (u64)-1)
751                         goto none;
752
753                 min_key.offset++;
754                 btrfs_release_path(path);
755         }
756 none:
757         btrfs_free_path(path);
758         return -ENOENT;
759 }
760
761 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
762                                int thresh, u64 *last_len, u64 *skip,
763                                u64 *defrag_end)
764 {
765         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
766         struct extent_map *em = NULL;
767         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
768         int ret = 1;
769
770         /*
771          * make sure that once we start defragging and extent, we keep on
772          * defragging it
773          */
774         if (start < *defrag_end)
775                 return 1;
776
777         *skip = 0;
778
779         /*
780          * hopefully we have this extent in the tree already, try without
781          * the full extent lock
782          */
783         read_lock(&em_tree->lock);
784         em = lookup_extent_mapping(em_tree, start, len);
785         read_unlock(&em_tree->lock);
786
787         if (!em) {
788                 /* get the big lock and read metadata off disk */
789                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
790                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
791                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
792
793                 if (IS_ERR(em))
794                         return 0;
795         }
796
797         /* this will cover holes, and inline extents */
798         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
799                 ret = 0;
800
801         /*
802          * we hit a real extent, if it is big don't bother defragging it again
803          */
804         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
805                 ret = 0;
806
807         /*
808          * last_len ends up being a counter of how many bytes we've defragged.
809          * every time we choose not to defrag an extent, we reset *last_len
810          * so that the next tiny extent will force a defrag.
811          *
812          * The end result of this is that tiny extents before a single big
813          * extent will force at least part of that big extent to be defragged.
814          */
815         if (ret) {
816                 *last_len += len;
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
855         if (isize == 0)
856                 return 0;
857         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
858
859         ret = btrfs_delalloc_reserve_space(inode,
860                                            num_pages << PAGE_CACHE_SHIFT);
861         if (ret)
862                 return ret;
863 again:
864         ret = 0;
865         i_done = 0;
866
867         /* step one, lock all the pages */
868         for (i = 0; i < num_pages; i++) {
869                 struct page *page;
870                 page = find_or_create_page(inode->i_mapping,
871                                             start_index + i, GFP_NOFS);
872                 if (!page)
873                         break;
874
875                 if (!PageUptodate(page)) {
876                         btrfs_readpage(NULL, page);
877                         lock_page(page);
878                         if (!PageUptodate(page)) {
879                                 unlock_page(page);
880                                 page_cache_release(page);
881                                 ret = -EIO;
882                                 break;
883                         }
884                 }
885                 isize = i_size_read(inode);
886                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
887                 if (!isize || page->index > file_end ||
888                     page->mapping != inode->i_mapping) {
889                         /* whoops, we blew past eof, skip this page */
890                         unlock_page(page);
891                         page_cache_release(page);
892                         break;
893                 }
894                 pages[i] = page;
895                 i_done++;
896         }
897         if (!i_done || ret)
898                 goto out;
899
900         if (!(inode->i_sb->s_flags & MS_ACTIVE))
901                 goto out;
902
903         /*
904          * so now we have a nice long stream of locked
905          * and up to date pages, lets wait on them
906          */
907         for (i = 0; i < i_done; i++)
908                 wait_on_page_writeback(pages[i]);
909
910         page_start = page_offset(pages[0]);
911         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
912
913         lock_extent_bits(&BTRFS_I(inode)->io_tree,
914                          page_start, page_end - 1, 0, &cached_state,
915                          GFP_NOFS);
916         ordered = btrfs_lookup_first_ordered_extent(inode, page_end - 1);
917         if (ordered &&
918             ordered->file_offset + ordered->len > page_start &&
919             ordered->file_offset < page_end) {
920                 btrfs_put_ordered_extent(ordered);
921                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
922                                      page_start, page_end - 1,
923                                      &cached_state, GFP_NOFS);
924                 for (i = 0; i < i_done; i++) {
925                         unlock_page(pages[i]);
926                         page_cache_release(pages[i]);
927                 }
928                 btrfs_wait_ordered_range(inode, page_start,
929                                          page_end - page_start);
930                 goto again;
931         }
932         if (ordered)
933                 btrfs_put_ordered_extent(ordered);
934
935         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
936                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
937                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
938                           GFP_NOFS);
939
940         if (i_done != num_pages) {
941                 spin_lock(&BTRFS_I(inode)->lock);
942                 BTRFS_I(inode)->outstanding_extents++;
943                 spin_unlock(&BTRFS_I(inode)->lock);
944                 btrfs_delalloc_release_space(inode,
945                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
946         }
947
948
949         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
950                                   &cached_state);
951
952         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
953                              page_start, page_end - 1, &cached_state,
954                              GFP_NOFS);
955
956         for (i = 0; i < i_done; i++) {
957                 clear_page_dirty_for_io(pages[i]);
958                 ClearPageChecked(pages[i]);
959                 set_page_extent_mapped(pages[i]);
960                 set_page_dirty(pages[i]);
961                 unlock_page(pages[i]);
962                 page_cache_release(pages[i]);
963         }
964         return i_done;
965 out:
966         for (i = 0; i < i_done; i++) {
967                 unlock_page(pages[i]);
968                 page_cache_release(pages[i]);
969         }
970         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
971         return ret;
972
973 }
974
975 int btrfs_defrag_file(struct inode *inode, struct file *file,
976                       struct btrfs_ioctl_defrag_range_args *range,
977                       u64 newer_than, unsigned long max_to_defrag)
978 {
979         struct btrfs_root *root = BTRFS_I(inode)->root;
980         struct btrfs_super_block *disk_super;
981         struct file_ra_state *ra = NULL;
982         unsigned long last_index;
983         u64 features;
984         u64 last_len = 0;
985         u64 skip = 0;
986         u64 defrag_end = 0;
987         u64 newer_off = range->start;
988         int newer_left = 0;
989         unsigned long i;
990         int ret;
991         int defrag_count = 0;
992         int compress_type = BTRFS_COMPRESS_ZLIB;
993         int extent_thresh = range->extent_thresh;
994         int newer_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
995         u64 new_align = ~((u64)128 * 1024 - 1);
996         struct page **pages = NULL;
997
998         if (extent_thresh == 0)
999                 extent_thresh = 256 * 1024;
1000
1001         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1002                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1003                         return -EINVAL;
1004                 if (range->compress_type)
1005                         compress_type = range->compress_type;
1006         }
1007
1008         if (inode->i_size == 0)
1009                 return 0;
1010
1011         /*
1012          * if we were not given a file, allocate a readahead
1013          * context
1014          */
1015         if (!file) {
1016                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1017                 if (!ra)
1018                         return -ENOMEM;
1019                 file_ra_state_init(ra, inode->i_mapping);
1020         } else {
1021                 ra = &file->f_ra;
1022         }
1023
1024         pages = kmalloc(sizeof(struct page *) * newer_cluster,
1025                         GFP_NOFS);
1026         if (!pages) {
1027                 ret = -ENOMEM;
1028                 goto out_ra;
1029         }
1030
1031         /* find the last page to defrag */
1032         if (range->start + range->len > range->start) {
1033                 last_index = min_t(u64, inode->i_size - 1,
1034                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1035         } else {
1036                 last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
1037         }
1038
1039         if (newer_than) {
1040                 ret = find_new_extents(root, inode, newer_than,
1041                                        &newer_off, 64 * 1024);
1042                 if (!ret) {
1043                         range->start = newer_off;
1044                         /*
1045                          * we always align our defrag to help keep
1046                          * the extents in the file evenly spaced
1047                          */
1048                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1049                         newer_left = newer_cluster;
1050                 } else
1051                         goto out_ra;
1052         } else {
1053                 i = range->start >> PAGE_CACHE_SHIFT;
1054         }
1055         if (!max_to_defrag)
1056                 max_to_defrag = last_index - 1;
1057
1058         while (i <= last_index && defrag_count < max_to_defrag) {
1059                 /*
1060                  * make sure we stop running if someone unmounts
1061                  * the FS
1062                  */
1063                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1064                         break;
1065
1066                 if (!newer_than &&
1067                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1068                                         PAGE_CACHE_SIZE,
1069                                         extent_thresh,
1070                                         &last_len, &skip,
1071                                         &defrag_end)) {
1072                         unsigned long next;
1073                         /*
1074                          * the should_defrag function tells us how much to skip
1075                          * bump our counter by the suggested amount
1076                          */
1077                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1078                         i = max(i + 1, next);
1079                         continue;
1080                 }
1081                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1082                         BTRFS_I(inode)->force_compress = compress_type;
1083
1084                 btrfs_force_ra(inode->i_mapping, ra, file, i, newer_cluster);
1085
1086                 ret = cluster_pages_for_defrag(inode, pages, i, newer_cluster);
1087                 if (ret < 0)
1088                         goto out_ra;
1089
1090                 defrag_count += ret;
1091                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1092                 i += ret;
1093
1094                 if (newer_than) {
1095                         if (newer_off == (u64)-1)
1096                                 break;
1097
1098                         newer_off = max(newer_off + 1,
1099                                         (u64)i << PAGE_CACHE_SHIFT);
1100
1101                         ret = find_new_extents(root, inode,
1102                                                newer_than, &newer_off,
1103                                                64 * 1024);
1104                         if (!ret) {
1105                                 range->start = newer_off;
1106                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1107                                 newer_left = newer_cluster;
1108                         } else {
1109                                 break;
1110                         }
1111                 } else {
1112                         i++;
1113                 }
1114         }
1115
1116         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1117                 filemap_flush(inode->i_mapping);
1118
1119         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1120                 /* the filemap_flush will queue IO into the worker threads, but
1121                  * we have to make sure the IO is actually started and that
1122                  * ordered extents get created before we return
1123                  */
1124                 atomic_inc(&root->fs_info->async_submit_draining);
1125                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1126                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1127                         wait_event(root->fs_info->async_submit_wait,
1128                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1129                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1130                 }
1131                 atomic_dec(&root->fs_info->async_submit_draining);
1132
1133                 mutex_lock(&inode->i_mutex);
1134                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1135                 mutex_unlock(&inode->i_mutex);
1136         }
1137
1138         disk_super = &root->fs_info->super_copy;
1139         features = btrfs_super_incompat_flags(disk_super);
1140         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1141                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1142                 btrfs_set_super_incompat_flags(disk_super, features);
1143         }
1144
1145         if (!file)
1146                 kfree(ra);
1147         return defrag_count;
1148
1149 out_ra:
1150         if (!file)
1151                 kfree(ra);
1152         kfree(pages);
1153         return ret;
1154 }
1155
1156 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1157                                         void __user *arg)
1158 {
1159         u64 new_size;
1160         u64 old_size;
1161         u64 devid = 1;
1162         struct btrfs_ioctl_vol_args *vol_args;
1163         struct btrfs_trans_handle *trans;
1164         struct btrfs_device *device = NULL;
1165         char *sizestr;
1166         char *devstr = NULL;
1167         int ret = 0;
1168         int mod = 0;
1169
1170         if (root->fs_info->sb->s_flags & MS_RDONLY)
1171                 return -EROFS;
1172
1173         if (!capable(CAP_SYS_ADMIN))
1174                 return -EPERM;
1175
1176         vol_args = memdup_user(arg, sizeof(*vol_args));
1177         if (IS_ERR(vol_args))
1178                 return PTR_ERR(vol_args);
1179
1180         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1181
1182         mutex_lock(&root->fs_info->volume_mutex);
1183         sizestr = vol_args->name;
1184         devstr = strchr(sizestr, ':');
1185         if (devstr) {
1186                 char *end;
1187                 sizestr = devstr + 1;
1188                 *devstr = '\0';
1189                 devstr = vol_args->name;
1190                 devid = simple_strtoull(devstr, &end, 10);
1191                 printk(KERN_INFO "resizing devid %llu\n",
1192                        (unsigned long long)devid);
1193         }
1194         device = btrfs_find_device(root, devid, NULL, NULL);
1195         if (!device) {
1196                 printk(KERN_INFO "resizer unable to find device %llu\n",
1197                        (unsigned long long)devid);
1198                 ret = -EINVAL;
1199                 goto out_unlock;
1200         }
1201         if (!strcmp(sizestr, "max"))
1202                 new_size = device->bdev->bd_inode->i_size;
1203         else {
1204                 if (sizestr[0] == '-') {
1205                         mod = -1;
1206                         sizestr++;
1207                 } else if (sizestr[0] == '+') {
1208                         mod = 1;
1209                         sizestr++;
1210                 }
1211                 new_size = memparse(sizestr, NULL);
1212                 if (new_size == 0) {
1213                         ret = -EINVAL;
1214                         goto out_unlock;
1215                 }
1216         }
1217
1218         old_size = device->total_bytes;
1219
1220         if (mod < 0) {
1221                 if (new_size > old_size) {
1222                         ret = -EINVAL;
1223                         goto out_unlock;
1224                 }
1225                 new_size = old_size - new_size;
1226         } else if (mod > 0) {
1227                 new_size = old_size + new_size;
1228         }
1229
1230         if (new_size < 256 * 1024 * 1024) {
1231                 ret = -EINVAL;
1232                 goto out_unlock;
1233         }
1234         if (new_size > device->bdev->bd_inode->i_size) {
1235                 ret = -EFBIG;
1236                 goto out_unlock;
1237         }
1238
1239         do_div(new_size, root->sectorsize);
1240         new_size *= root->sectorsize;
1241
1242         printk(KERN_INFO "new size for %s is %llu\n",
1243                 device->name, (unsigned long long)new_size);
1244
1245         if (new_size > old_size) {
1246                 trans = btrfs_start_transaction(root, 0);
1247                 if (IS_ERR(trans)) {
1248                         ret = PTR_ERR(trans);
1249                         goto out_unlock;
1250                 }
1251                 ret = btrfs_grow_device(trans, device, new_size);
1252                 btrfs_commit_transaction(trans, root);
1253         } else {
1254                 ret = btrfs_shrink_device(device, new_size);
1255         }
1256
1257 out_unlock:
1258         mutex_unlock(&root->fs_info->volume_mutex);
1259         kfree(vol_args);
1260         return ret;
1261 }
1262
1263 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1264                                                     char *name,
1265                                                     unsigned long fd,
1266                                                     int subvol,
1267                                                     u64 *transid,
1268                                                     bool readonly)
1269 {
1270         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1271         struct file *src_file;
1272         int namelen;
1273         int ret = 0;
1274
1275         if (root->fs_info->sb->s_flags & MS_RDONLY)
1276                 return -EROFS;
1277
1278         namelen = strlen(name);
1279         if (strchr(name, '/')) {
1280                 ret = -EINVAL;
1281                 goto out;
1282         }
1283
1284         if (subvol) {
1285                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1286                                      NULL, transid, readonly);
1287         } else {
1288                 struct inode *src_inode;
1289                 src_file = fget(fd);
1290                 if (!src_file) {
1291                         ret = -EINVAL;
1292                         goto out;
1293                 }
1294
1295                 src_inode = src_file->f_path.dentry->d_inode;
1296                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1297                         printk(KERN_INFO "btrfs: Snapshot src from "
1298                                "another FS\n");
1299                         ret = -EINVAL;
1300                         fput(src_file);
1301                         goto out;
1302                 }
1303                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1304                                      BTRFS_I(src_inode)->root,
1305                                      transid, readonly);
1306                 fput(src_file);
1307         }
1308 out:
1309         return ret;
1310 }
1311
1312 static noinline int btrfs_ioctl_snap_create(struct file *file,
1313                                             void __user *arg, int subvol)
1314 {
1315         struct btrfs_ioctl_vol_args *vol_args;
1316         int ret;
1317
1318         vol_args = memdup_user(arg, sizeof(*vol_args));
1319         if (IS_ERR(vol_args))
1320                 return PTR_ERR(vol_args);
1321         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1322
1323         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1324                                               vol_args->fd, subvol,
1325                                               NULL, false);
1326
1327         kfree(vol_args);
1328         return ret;
1329 }
1330
1331 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1332                                                void __user *arg, int subvol)
1333 {
1334         struct btrfs_ioctl_vol_args_v2 *vol_args;
1335         int ret;
1336         u64 transid = 0;
1337         u64 *ptr = NULL;
1338         bool readonly = false;
1339
1340         vol_args = memdup_user(arg, sizeof(*vol_args));
1341         if (IS_ERR(vol_args))
1342                 return PTR_ERR(vol_args);
1343         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1344
1345         if (vol_args->flags &
1346             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1347                 ret = -EOPNOTSUPP;
1348                 goto out;
1349         }
1350
1351         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1352                 ptr = &transid;
1353         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1354                 readonly = true;
1355
1356         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1357                                               vol_args->fd, subvol,
1358                                               ptr, readonly);
1359
1360         if (ret == 0 && ptr &&
1361             copy_to_user(arg +
1362                          offsetof(struct btrfs_ioctl_vol_args_v2,
1363                                   transid), ptr, sizeof(*ptr)))
1364                 ret = -EFAULT;
1365 out:
1366         kfree(vol_args);
1367         return ret;
1368 }
1369
1370 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1371                                                 void __user *arg)
1372 {
1373         struct inode *inode = fdentry(file)->d_inode;
1374         struct btrfs_root *root = BTRFS_I(inode)->root;
1375         int ret = 0;
1376         u64 flags = 0;
1377
1378         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1379                 return -EINVAL;
1380
1381         down_read(&root->fs_info->subvol_sem);
1382         if (btrfs_root_readonly(root))
1383                 flags |= BTRFS_SUBVOL_RDONLY;
1384         up_read(&root->fs_info->subvol_sem);
1385
1386         if (copy_to_user(arg, &flags, sizeof(flags)))
1387                 ret = -EFAULT;
1388
1389         return ret;
1390 }
1391
1392 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1393                                               void __user *arg)
1394 {
1395         struct inode *inode = fdentry(file)->d_inode;
1396         struct btrfs_root *root = BTRFS_I(inode)->root;
1397         struct btrfs_trans_handle *trans;
1398         u64 root_flags;
1399         u64 flags;
1400         int ret = 0;
1401
1402         if (root->fs_info->sb->s_flags & MS_RDONLY)
1403                 return -EROFS;
1404
1405         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1406                 return -EINVAL;
1407
1408         if (copy_from_user(&flags, arg, sizeof(flags)))
1409                 return -EFAULT;
1410
1411         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1412                 return -EINVAL;
1413
1414         if (flags & ~BTRFS_SUBVOL_RDONLY)
1415                 return -EOPNOTSUPP;
1416
1417         if (!inode_owner_or_capable(inode))
1418                 return -EACCES;
1419
1420         down_write(&root->fs_info->subvol_sem);
1421
1422         /* nothing to do */
1423         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1424                 goto out;
1425
1426         root_flags = btrfs_root_flags(&root->root_item);
1427         if (flags & BTRFS_SUBVOL_RDONLY)
1428                 btrfs_set_root_flags(&root->root_item,
1429                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1430         else
1431                 btrfs_set_root_flags(&root->root_item,
1432                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1433
1434         trans = btrfs_start_transaction(root, 1);
1435         if (IS_ERR(trans)) {
1436                 ret = PTR_ERR(trans);
1437                 goto out_reset;
1438         }
1439
1440         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1441                                 &root->root_key, &root->root_item);
1442
1443         btrfs_commit_transaction(trans, root);
1444 out_reset:
1445         if (ret)
1446                 btrfs_set_root_flags(&root->root_item, root_flags);
1447 out:
1448         up_write(&root->fs_info->subvol_sem);
1449         return ret;
1450 }
1451
1452 /*
1453  * helper to check if the subvolume references other subvolumes
1454  */
1455 static noinline int may_destroy_subvol(struct btrfs_root *root)
1456 {
1457         struct btrfs_path *path;
1458         struct btrfs_key key;
1459         int ret;
1460
1461         path = btrfs_alloc_path();
1462         if (!path)
1463                 return -ENOMEM;
1464
1465         key.objectid = root->root_key.objectid;
1466         key.type = BTRFS_ROOT_REF_KEY;
1467         key.offset = (u64)-1;
1468
1469         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1470                                 &key, path, 0, 0);
1471         if (ret < 0)
1472                 goto out;
1473         BUG_ON(ret == 0);
1474
1475         ret = 0;
1476         if (path->slots[0] > 0) {
1477                 path->slots[0]--;
1478                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1479                 if (key.objectid == root->root_key.objectid &&
1480                     key.type == BTRFS_ROOT_REF_KEY)
1481                         ret = -ENOTEMPTY;
1482         }
1483 out:
1484         btrfs_free_path(path);
1485         return ret;
1486 }
1487
1488 static noinline int key_in_sk(struct btrfs_key *key,
1489                               struct btrfs_ioctl_search_key *sk)
1490 {
1491         struct btrfs_key test;
1492         int ret;
1493
1494         test.objectid = sk->min_objectid;
1495         test.type = sk->min_type;
1496         test.offset = sk->min_offset;
1497
1498         ret = btrfs_comp_cpu_keys(key, &test);
1499         if (ret < 0)
1500                 return 0;
1501
1502         test.objectid = sk->max_objectid;
1503         test.type = sk->max_type;
1504         test.offset = sk->max_offset;
1505
1506         ret = btrfs_comp_cpu_keys(key, &test);
1507         if (ret > 0)
1508                 return 0;
1509         return 1;
1510 }
1511
1512 static noinline int copy_to_sk(struct btrfs_root *root,
1513                                struct btrfs_path *path,
1514                                struct btrfs_key *key,
1515                                struct btrfs_ioctl_search_key *sk,
1516                                char *buf,
1517                                unsigned long *sk_offset,
1518                                int *num_found)
1519 {
1520         u64 found_transid;
1521         struct extent_buffer *leaf;
1522         struct btrfs_ioctl_search_header sh;
1523         unsigned long item_off;
1524         unsigned long item_len;
1525         int nritems;
1526         int i;
1527         int slot;
1528         int ret = 0;
1529
1530         leaf = path->nodes[0];
1531         slot = path->slots[0];
1532         nritems = btrfs_header_nritems(leaf);
1533
1534         if (btrfs_header_generation(leaf) > sk->max_transid) {
1535                 i = nritems;
1536                 goto advance_key;
1537         }
1538         found_transid = btrfs_header_generation(leaf);
1539
1540         for (i = slot; i < nritems; i++) {
1541                 item_off = btrfs_item_ptr_offset(leaf, i);
1542                 item_len = btrfs_item_size_nr(leaf, i);
1543
1544                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1545                         item_len = 0;
1546
1547                 if (sizeof(sh) + item_len + *sk_offset >
1548                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1549                         ret = 1;
1550                         goto overflow;
1551                 }
1552
1553                 btrfs_item_key_to_cpu(leaf, key, i);
1554                 if (!key_in_sk(key, sk))
1555                         continue;
1556
1557                 sh.objectid = key->objectid;
1558                 sh.offset = key->offset;
1559                 sh.type = key->type;
1560                 sh.len = item_len;
1561                 sh.transid = found_transid;
1562
1563                 /* copy search result header */
1564                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1565                 *sk_offset += sizeof(sh);
1566
1567                 if (item_len) {
1568                         char *p = buf + *sk_offset;
1569                         /* copy the item */
1570                         read_extent_buffer(leaf, p,
1571                                            item_off, item_len);
1572                         *sk_offset += item_len;
1573                 }
1574                 (*num_found)++;
1575
1576                 if (*num_found >= sk->nr_items)
1577                         break;
1578         }
1579 advance_key:
1580         ret = 0;
1581         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1582                 key->offset++;
1583         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1584                 key->offset = 0;
1585                 key->type++;
1586         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1587                 key->offset = 0;
1588                 key->type = 0;
1589                 key->objectid++;
1590         } else
1591                 ret = 1;
1592 overflow:
1593         return ret;
1594 }
1595
1596 static noinline int search_ioctl(struct inode *inode,
1597                                  struct btrfs_ioctl_search_args *args)
1598 {
1599         struct btrfs_root *root;
1600         struct btrfs_key key;
1601         struct btrfs_key max_key;
1602         struct btrfs_path *path;
1603         struct btrfs_ioctl_search_key *sk = &args->key;
1604         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1605         int ret;
1606         int num_found = 0;
1607         unsigned long sk_offset = 0;
1608
1609         path = btrfs_alloc_path();
1610         if (!path)
1611                 return -ENOMEM;
1612
1613         if (sk->tree_id == 0) {
1614                 /* search the root of the inode that was passed */
1615                 root = BTRFS_I(inode)->root;
1616         } else {
1617                 key.objectid = sk->tree_id;
1618                 key.type = BTRFS_ROOT_ITEM_KEY;
1619                 key.offset = (u64)-1;
1620                 root = btrfs_read_fs_root_no_name(info, &key);
1621                 if (IS_ERR(root)) {
1622                         printk(KERN_ERR "could not find root %llu\n",
1623                                sk->tree_id);
1624                         btrfs_free_path(path);
1625                         return -ENOENT;
1626                 }
1627         }
1628
1629         key.objectid = sk->min_objectid;
1630         key.type = sk->min_type;
1631         key.offset = sk->min_offset;
1632
1633         max_key.objectid = sk->max_objectid;
1634         max_key.type = sk->max_type;
1635         max_key.offset = sk->max_offset;
1636
1637         path->keep_locks = 1;
1638
1639         while(1) {
1640                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1641                                            sk->min_transid);
1642                 if (ret != 0) {
1643                         if (ret > 0)
1644                                 ret = 0;
1645                         goto err;
1646                 }
1647                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1648                                  &sk_offset, &num_found);
1649                 btrfs_release_path(path);
1650                 if (ret || num_found >= sk->nr_items)
1651                         break;
1652
1653         }
1654         ret = 0;
1655 err:
1656         sk->nr_items = num_found;
1657         btrfs_free_path(path);
1658         return ret;
1659 }
1660
1661 static noinline int btrfs_ioctl_tree_search(struct file *file,
1662                                            void __user *argp)
1663 {
1664          struct btrfs_ioctl_search_args *args;
1665          struct inode *inode;
1666          int ret;
1667
1668         if (!capable(CAP_SYS_ADMIN))
1669                 return -EPERM;
1670
1671         args = memdup_user(argp, sizeof(*args));
1672         if (IS_ERR(args))
1673                 return PTR_ERR(args);
1674
1675         inode = fdentry(file)->d_inode;
1676         ret = search_ioctl(inode, args);
1677         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1678                 ret = -EFAULT;
1679         kfree(args);
1680         return ret;
1681 }
1682
1683 /*
1684  * Search INODE_REFs to identify path name of 'dirid' directory
1685  * in a 'tree_id' tree. and sets path name to 'name'.
1686  */
1687 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1688                                 u64 tree_id, u64 dirid, char *name)
1689 {
1690         struct btrfs_root *root;
1691         struct btrfs_key key;
1692         char *ptr;
1693         int ret = -1;
1694         int slot;
1695         int len;
1696         int total_len = 0;
1697         struct btrfs_inode_ref *iref;
1698         struct extent_buffer *l;
1699         struct btrfs_path *path;
1700
1701         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1702                 name[0]='\0';
1703                 return 0;
1704         }
1705
1706         path = btrfs_alloc_path();
1707         if (!path)
1708                 return -ENOMEM;
1709
1710         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1711
1712         key.objectid = tree_id;
1713         key.type = BTRFS_ROOT_ITEM_KEY;
1714         key.offset = (u64)-1;
1715         root = btrfs_read_fs_root_no_name(info, &key);
1716         if (IS_ERR(root)) {
1717                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1718                 ret = -ENOENT;
1719                 goto out;
1720         }
1721
1722         key.objectid = dirid;
1723         key.type = BTRFS_INODE_REF_KEY;
1724         key.offset = (u64)-1;
1725
1726         while(1) {
1727                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1728                 if (ret < 0)
1729                         goto out;
1730
1731                 l = path->nodes[0];
1732                 slot = path->slots[0];
1733                 if (ret > 0 && slot > 0)
1734                         slot--;
1735                 btrfs_item_key_to_cpu(l, &key, slot);
1736
1737                 if (ret > 0 && (key.objectid != dirid ||
1738                                 key.type != BTRFS_INODE_REF_KEY)) {
1739                         ret = -ENOENT;
1740                         goto out;
1741                 }
1742
1743                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1744                 len = btrfs_inode_ref_name_len(l, iref);
1745                 ptr -= len + 1;
1746                 total_len += len + 1;
1747                 if (ptr < name)
1748                         goto out;
1749
1750                 *(ptr + len) = '/';
1751                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1752
1753                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1754                         break;
1755
1756                 btrfs_release_path(path);
1757                 key.objectid = key.offset;
1758                 key.offset = (u64)-1;
1759                 dirid = key.objectid;
1760
1761         }
1762         if (ptr < name)
1763                 goto out;
1764         memcpy(name, ptr, total_len);
1765         name[total_len]='\0';
1766         ret = 0;
1767 out:
1768         btrfs_free_path(path);
1769         return ret;
1770 }
1771
1772 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1773                                            void __user *argp)
1774 {
1775          struct btrfs_ioctl_ino_lookup_args *args;
1776          struct inode *inode;
1777          int ret;
1778
1779         if (!capable(CAP_SYS_ADMIN))
1780                 return -EPERM;
1781
1782         args = memdup_user(argp, sizeof(*args));
1783         if (IS_ERR(args))
1784                 return PTR_ERR(args);
1785
1786         inode = fdentry(file)->d_inode;
1787
1788         if (args->treeid == 0)
1789                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1790
1791         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1792                                         args->treeid, args->objectid,
1793                                         args->name);
1794
1795         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1796                 ret = -EFAULT;
1797
1798         kfree(args);
1799         return ret;
1800 }
1801
1802 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1803                                              void __user *arg)
1804 {
1805         struct dentry *parent = fdentry(file);
1806         struct dentry *dentry;
1807         struct inode *dir = parent->d_inode;
1808         struct inode *inode;
1809         struct btrfs_root *root = BTRFS_I(dir)->root;
1810         struct btrfs_root *dest = NULL;
1811         struct btrfs_ioctl_vol_args *vol_args;
1812         struct btrfs_trans_handle *trans;
1813         int namelen;
1814         int ret;
1815         int err = 0;
1816
1817         vol_args = memdup_user(arg, sizeof(*vol_args));
1818         if (IS_ERR(vol_args))
1819                 return PTR_ERR(vol_args);
1820
1821         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1822         namelen = strlen(vol_args->name);
1823         if (strchr(vol_args->name, '/') ||
1824             strncmp(vol_args->name, "..", namelen) == 0) {
1825                 err = -EINVAL;
1826                 goto out;
1827         }
1828
1829         err = mnt_want_write(file->f_path.mnt);
1830         if (err)
1831                 goto out;
1832
1833         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1834         dentry = lookup_one_len(vol_args->name, parent, namelen);
1835         if (IS_ERR(dentry)) {
1836                 err = PTR_ERR(dentry);
1837                 goto out_unlock_dir;
1838         }
1839
1840         if (!dentry->d_inode) {
1841                 err = -ENOENT;
1842                 goto out_dput;
1843         }
1844
1845         inode = dentry->d_inode;
1846         dest = BTRFS_I(inode)->root;
1847         if (!capable(CAP_SYS_ADMIN)){
1848                 /*
1849                  * Regular user.  Only allow this with a special mount
1850                  * option, when the user has write+exec access to the
1851                  * subvol root, and when rmdir(2) would have been
1852                  * allowed.
1853                  *
1854                  * Note that this is _not_ check that the subvol is
1855                  * empty or doesn't contain data that we wouldn't
1856                  * otherwise be able to delete.
1857                  *
1858                  * Users who want to delete empty subvols should try
1859                  * rmdir(2).
1860                  */
1861                 err = -EPERM;
1862                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1863                         goto out_dput;
1864
1865                 /*
1866                  * Do not allow deletion if the parent dir is the same
1867                  * as the dir to be deleted.  That means the ioctl
1868                  * must be called on the dentry referencing the root
1869                  * of the subvol, not a random directory contained
1870                  * within it.
1871                  */
1872                 err = -EINVAL;
1873                 if (root == dest)
1874                         goto out_dput;
1875
1876                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1877                 if (err)
1878                         goto out_dput;
1879
1880                 /* check if subvolume may be deleted by a non-root user */
1881                 err = btrfs_may_delete(dir, dentry, 1);
1882                 if (err)
1883                         goto out_dput;
1884         }
1885
1886         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1887                 err = -EINVAL;
1888                 goto out_dput;
1889         }
1890
1891         mutex_lock(&inode->i_mutex);
1892         err = d_invalidate(dentry);
1893         if (err)
1894                 goto out_unlock;
1895
1896         down_write(&root->fs_info->subvol_sem);
1897
1898         err = may_destroy_subvol(dest);
1899         if (err)
1900                 goto out_up_write;
1901
1902         trans = btrfs_start_transaction(root, 0);
1903         if (IS_ERR(trans)) {
1904                 err = PTR_ERR(trans);
1905                 goto out_up_write;
1906         }
1907         trans->block_rsv = &root->fs_info->global_block_rsv;
1908
1909         ret = btrfs_unlink_subvol(trans, root, dir,
1910                                 dest->root_key.objectid,
1911                                 dentry->d_name.name,
1912                                 dentry->d_name.len);
1913         BUG_ON(ret);
1914
1915         btrfs_record_root_in_trans(trans, dest);
1916
1917         memset(&dest->root_item.drop_progress, 0,
1918                 sizeof(dest->root_item.drop_progress));
1919         dest->root_item.drop_level = 0;
1920         btrfs_set_root_refs(&dest->root_item, 0);
1921
1922         if (!xchg(&dest->orphan_item_inserted, 1)) {
1923                 ret = btrfs_insert_orphan_item(trans,
1924                                         root->fs_info->tree_root,
1925                                         dest->root_key.objectid);
1926                 BUG_ON(ret);
1927         }
1928
1929         ret = btrfs_end_transaction(trans, root);
1930         BUG_ON(ret);
1931         inode->i_flags |= S_DEAD;
1932 out_up_write:
1933         up_write(&root->fs_info->subvol_sem);
1934 out_unlock:
1935         mutex_unlock(&inode->i_mutex);
1936         if (!err) {
1937                 shrink_dcache_sb(root->fs_info->sb);
1938                 btrfs_invalidate_inodes(dest);
1939                 d_delete(dentry);
1940         }
1941 out_dput:
1942         dput(dentry);
1943 out_unlock_dir:
1944         mutex_unlock(&dir->i_mutex);
1945         mnt_drop_write(file->f_path.mnt);
1946 out:
1947         kfree(vol_args);
1948         return err;
1949 }
1950
1951 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1952 {
1953         struct inode *inode = fdentry(file)->d_inode;
1954         struct btrfs_root *root = BTRFS_I(inode)->root;
1955         struct btrfs_ioctl_defrag_range_args *range;
1956         int ret;
1957
1958         if (btrfs_root_readonly(root))
1959                 return -EROFS;
1960
1961         ret = mnt_want_write(file->f_path.mnt);
1962         if (ret)
1963                 return ret;
1964
1965         switch (inode->i_mode & S_IFMT) {
1966         case S_IFDIR:
1967                 if (!capable(CAP_SYS_ADMIN)) {
1968                         ret = -EPERM;
1969                         goto out;
1970                 }
1971                 ret = btrfs_defrag_root(root, 0);
1972                 if (ret)
1973                         goto out;
1974                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1975                 break;
1976         case S_IFREG:
1977                 if (!(file->f_mode & FMODE_WRITE)) {
1978                         ret = -EINVAL;
1979                         goto out;
1980                 }
1981
1982                 range = kzalloc(sizeof(*range), GFP_KERNEL);
1983                 if (!range) {
1984                         ret = -ENOMEM;
1985                         goto out;
1986                 }
1987
1988                 if (argp) {
1989                         if (copy_from_user(range, argp,
1990                                            sizeof(*range))) {
1991                                 ret = -EFAULT;
1992                                 kfree(range);
1993                                 goto out;
1994                         }
1995                         /* compression requires us to start the IO */
1996                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1997                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1998                                 range->extent_thresh = (u32)-1;
1999                         }
2000                 } else {
2001                         /* the rest are all set to zero by kzalloc */
2002                         range->len = (u64)-1;
2003                 }
2004                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2005                                         range, 0, 0);
2006                 if (ret > 0)
2007                         ret = 0;
2008                 kfree(range);
2009                 break;
2010         default:
2011                 ret = -EINVAL;
2012         }
2013 out:
2014         mnt_drop_write(file->f_path.mnt);
2015         return ret;
2016 }
2017
2018 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2019 {
2020         struct btrfs_ioctl_vol_args *vol_args;
2021         int ret;
2022
2023         if (!capable(CAP_SYS_ADMIN))
2024                 return -EPERM;
2025
2026         vol_args = memdup_user(arg, sizeof(*vol_args));
2027         if (IS_ERR(vol_args))
2028                 return PTR_ERR(vol_args);
2029
2030         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2031         ret = btrfs_init_new_device(root, vol_args->name);
2032
2033         kfree(vol_args);
2034         return ret;
2035 }
2036
2037 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2038 {
2039         struct btrfs_ioctl_vol_args *vol_args;
2040         int ret;
2041
2042         if (!capable(CAP_SYS_ADMIN))
2043                 return -EPERM;
2044
2045         if (root->fs_info->sb->s_flags & MS_RDONLY)
2046                 return -EROFS;
2047
2048         vol_args = memdup_user(arg, sizeof(*vol_args));
2049         if (IS_ERR(vol_args))
2050                 return PTR_ERR(vol_args);
2051
2052         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2053         ret = btrfs_rm_device(root, vol_args->name);
2054
2055         kfree(vol_args);
2056         return ret;
2057 }
2058
2059 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2060 {
2061         struct btrfs_ioctl_fs_info_args *fi_args;
2062         struct btrfs_device *device;
2063         struct btrfs_device *next;
2064         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2065         int ret = 0;
2066
2067         if (!capable(CAP_SYS_ADMIN))
2068                 return -EPERM;
2069
2070         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2071         if (!fi_args)
2072                 return -ENOMEM;
2073
2074         fi_args->num_devices = fs_devices->num_devices;
2075         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2076
2077         mutex_lock(&fs_devices->device_list_mutex);
2078         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2079                 if (device->devid > fi_args->max_id)
2080                         fi_args->max_id = device->devid;
2081         }
2082         mutex_unlock(&fs_devices->device_list_mutex);
2083
2084         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2085                 ret = -EFAULT;
2086
2087         kfree(fi_args);
2088         return ret;
2089 }
2090
2091 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2092 {
2093         struct btrfs_ioctl_dev_info_args *di_args;
2094         struct btrfs_device *dev;
2095         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2096         int ret = 0;
2097         char *s_uuid = NULL;
2098         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2099
2100         if (!capable(CAP_SYS_ADMIN))
2101                 return -EPERM;
2102
2103         di_args = memdup_user(arg, sizeof(*di_args));
2104         if (IS_ERR(di_args))
2105                 return PTR_ERR(di_args);
2106
2107         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2108                 s_uuid = di_args->uuid;
2109
2110         mutex_lock(&fs_devices->device_list_mutex);
2111         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2112         mutex_unlock(&fs_devices->device_list_mutex);
2113
2114         if (!dev) {
2115                 ret = -ENODEV;
2116                 goto out;
2117         }
2118
2119         di_args->devid = dev->devid;
2120         di_args->bytes_used = dev->bytes_used;
2121         di_args->total_bytes = dev->total_bytes;
2122         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2123         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2124
2125 out:
2126         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2127                 ret = -EFAULT;
2128
2129         kfree(di_args);
2130         return ret;
2131 }
2132
2133 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2134                                        u64 off, u64 olen, u64 destoff)
2135 {
2136         struct inode *inode = fdentry(file)->d_inode;
2137         struct btrfs_root *root = BTRFS_I(inode)->root;
2138         struct file *src_file;
2139         struct inode *src;
2140         struct btrfs_trans_handle *trans;
2141         struct btrfs_path *path;
2142         struct extent_buffer *leaf;
2143         char *buf;
2144         struct btrfs_key key;
2145         u32 nritems;
2146         int slot;
2147         int ret;
2148         u64 len = olen;
2149         u64 bs = root->fs_info->sb->s_blocksize;
2150         u64 hint_byte;
2151
2152         /*
2153          * TODO:
2154          * - split compressed inline extents.  annoying: we need to
2155          *   decompress into destination's address_space (the file offset
2156          *   may change, so source mapping won't do), then recompress (or
2157          *   otherwise reinsert) a subrange.
2158          * - allow ranges within the same file to be cloned (provided
2159          *   they don't overlap)?
2160          */
2161
2162         /* the destination must be opened for writing */
2163         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2164                 return -EINVAL;
2165
2166         if (btrfs_root_readonly(root))
2167                 return -EROFS;
2168
2169         ret = mnt_want_write(file->f_path.mnt);
2170         if (ret)
2171                 return ret;
2172
2173         src_file = fget(srcfd);
2174         if (!src_file) {
2175                 ret = -EBADF;
2176                 goto out_drop_write;
2177         }
2178
2179         src = src_file->f_dentry->d_inode;
2180
2181         ret = -EINVAL;
2182         if (src == inode)
2183                 goto out_fput;
2184
2185         /* the src must be open for reading */
2186         if (!(src_file->f_mode & FMODE_READ))
2187                 goto out_fput;
2188
2189         ret = -EISDIR;
2190         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2191                 goto out_fput;
2192
2193         ret = -EXDEV;
2194         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2195                 goto out_fput;
2196
2197         ret = -ENOMEM;
2198         buf = vmalloc(btrfs_level_size(root, 0));
2199         if (!buf)
2200                 goto out_fput;
2201
2202         path = btrfs_alloc_path();
2203         if (!path) {
2204                 vfree(buf);
2205                 goto out_fput;
2206         }
2207         path->reada = 2;
2208
2209         if (inode < src) {
2210                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2211                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2212         } else {
2213                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2214                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2215         }
2216
2217         /* determine range to clone */
2218         ret = -EINVAL;
2219         if (off + len > src->i_size || off + len < off)
2220                 goto out_unlock;
2221         if (len == 0)
2222                 olen = len = src->i_size - off;
2223         /* if we extend to eof, continue to block boundary */
2224         if (off + len == src->i_size)
2225                 len = ALIGN(src->i_size, bs) - off;
2226
2227         /* verify the end result is block aligned */
2228         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2229             !IS_ALIGNED(destoff, bs))
2230                 goto out_unlock;
2231
2232         /* do any pending delalloc/csum calc on src, one way or
2233            another, and lock file content */
2234         while (1) {
2235                 struct btrfs_ordered_extent *ordered;
2236                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2237                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2238                 if (!ordered &&
2239                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2240                                    EXTENT_DELALLOC, 0, NULL))
2241                         break;
2242                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2243                 if (ordered)
2244                         btrfs_put_ordered_extent(ordered);
2245                 btrfs_wait_ordered_range(src, off, len);
2246         }
2247
2248         /* clone data */
2249         key.objectid = btrfs_ino(src);
2250         key.type = BTRFS_EXTENT_DATA_KEY;
2251         key.offset = 0;
2252
2253         while (1) {
2254                 /*
2255                  * note the key will change type as we walk through the
2256                  * tree.
2257                  */
2258                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2259                 if (ret < 0)
2260                         goto out;
2261
2262                 nritems = btrfs_header_nritems(path->nodes[0]);
2263                 if (path->slots[0] >= nritems) {
2264                         ret = btrfs_next_leaf(root, path);
2265                         if (ret < 0)
2266                                 goto out;
2267                         if (ret > 0)
2268                                 break;
2269                         nritems = btrfs_header_nritems(path->nodes[0]);
2270                 }
2271                 leaf = path->nodes[0];
2272                 slot = path->slots[0];
2273
2274                 btrfs_item_key_to_cpu(leaf, &key, slot);
2275                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2276                     key.objectid != btrfs_ino(src))
2277                         break;
2278
2279                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2280                         struct btrfs_file_extent_item *extent;
2281                         int type;
2282                         u32 size;
2283                         struct btrfs_key new_key;
2284                         u64 disko = 0, diskl = 0;
2285                         u64 datao = 0, datal = 0;
2286                         u8 comp;
2287                         u64 endoff;
2288
2289                         size = btrfs_item_size_nr(leaf, slot);
2290                         read_extent_buffer(leaf, buf,
2291                                            btrfs_item_ptr_offset(leaf, slot),
2292                                            size);
2293
2294                         extent = btrfs_item_ptr(leaf, slot,
2295                                                 struct btrfs_file_extent_item);
2296                         comp = btrfs_file_extent_compression(leaf, extent);
2297                         type = btrfs_file_extent_type(leaf, extent);
2298                         if (type == BTRFS_FILE_EXTENT_REG ||
2299                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2300                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2301                                                                       extent);
2302                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2303                                                                  extent);
2304                                 datao = btrfs_file_extent_offset(leaf, extent);
2305                                 datal = btrfs_file_extent_num_bytes(leaf,
2306                                                                     extent);
2307                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2308                                 /* take upper bound, may be compressed */
2309                                 datal = btrfs_file_extent_ram_bytes(leaf,
2310                                                                     extent);
2311                         }
2312                         btrfs_release_path(path);
2313
2314                         if (key.offset + datal <= off ||
2315                             key.offset >= off+len)
2316                                 goto next;
2317
2318                         memcpy(&new_key, &key, sizeof(new_key));
2319                         new_key.objectid = btrfs_ino(inode);
2320                         if (off <= key.offset)
2321                                 new_key.offset = key.offset + destoff - off;
2322                         else
2323                                 new_key.offset = destoff;
2324
2325                         trans = btrfs_start_transaction(root, 1);
2326                         if (IS_ERR(trans)) {
2327                                 ret = PTR_ERR(trans);
2328                                 goto out;
2329                         }
2330
2331                         if (type == BTRFS_FILE_EXTENT_REG ||
2332                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2333                                 if (off > key.offset) {
2334                                         datao += off - key.offset;
2335                                         datal -= off - key.offset;
2336                                 }
2337
2338                                 if (key.offset + datal > off + len)
2339                                         datal = off + len - key.offset;
2340
2341                                 ret = btrfs_drop_extents(trans, inode,
2342                                                          new_key.offset,
2343                                                          new_key.offset + datal,
2344                                                          &hint_byte, 1);
2345                                 BUG_ON(ret);
2346
2347                                 ret = btrfs_insert_empty_item(trans, root, path,
2348                                                               &new_key, size);
2349                                 BUG_ON(ret);
2350
2351                                 leaf = path->nodes[0];
2352                                 slot = path->slots[0];
2353                                 write_extent_buffer(leaf, buf,
2354                                             btrfs_item_ptr_offset(leaf, slot),
2355                                             size);
2356
2357                                 extent = btrfs_item_ptr(leaf, slot,
2358                                                 struct btrfs_file_extent_item);
2359
2360                                 /* disko == 0 means it's a hole */
2361                                 if (!disko)
2362                                         datao = 0;
2363
2364                                 btrfs_set_file_extent_offset(leaf, extent,
2365                                                              datao);
2366                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2367                                                                 datal);
2368                                 if (disko) {
2369                                         inode_add_bytes(inode, datal);
2370                                         ret = btrfs_inc_extent_ref(trans, root,
2371                                                         disko, diskl, 0,
2372                                                         root->root_key.objectid,
2373                                                         btrfs_ino(inode),
2374                                                         new_key.offset - datao);
2375                                         BUG_ON(ret);
2376                                 }
2377                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2378                                 u64 skip = 0;
2379                                 u64 trim = 0;
2380                                 if (off > key.offset) {
2381                                         skip = off - key.offset;
2382                                         new_key.offset += skip;
2383                                 }
2384
2385                                 if (key.offset + datal > off+len)
2386                                         trim = key.offset + datal - (off+len);
2387
2388                                 if (comp && (skip || trim)) {
2389                                         ret = -EINVAL;
2390                                         btrfs_end_transaction(trans, root);
2391                                         goto out;
2392                                 }
2393                                 size -= skip + trim;
2394                                 datal -= skip + trim;
2395
2396                                 ret = btrfs_drop_extents(trans, inode,
2397                                                          new_key.offset,
2398                                                          new_key.offset + datal,
2399                                                          &hint_byte, 1);
2400                                 BUG_ON(ret);
2401
2402                                 ret = btrfs_insert_empty_item(trans, root, path,
2403                                                               &new_key, size);
2404                                 BUG_ON(ret);
2405
2406                                 if (skip) {
2407                                         u32 start =
2408                                           btrfs_file_extent_calc_inline_size(0);
2409                                         memmove(buf+start, buf+start+skip,
2410                                                 datal);
2411                                 }
2412
2413                                 leaf = path->nodes[0];
2414                                 slot = path->slots[0];
2415                                 write_extent_buffer(leaf, buf,
2416                                             btrfs_item_ptr_offset(leaf, slot),
2417                                             size);
2418                                 inode_add_bytes(inode, datal);
2419                         }
2420
2421                         btrfs_mark_buffer_dirty(leaf);
2422                         btrfs_release_path(path);
2423
2424                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2425
2426                         /*
2427                          * we round up to the block size at eof when
2428                          * determining which extents to clone above,
2429                          * but shouldn't round up the file size
2430                          */
2431                         endoff = new_key.offset + datal;
2432                         if (endoff > destoff+olen)
2433                                 endoff = destoff+olen;
2434                         if (endoff > inode->i_size)
2435                                 btrfs_i_size_write(inode, endoff);
2436
2437                         BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
2438                         ret = btrfs_update_inode(trans, root, inode);
2439                         BUG_ON(ret);
2440                         btrfs_end_transaction(trans, root);
2441                 }
2442 next:
2443                 btrfs_release_path(path);
2444                 key.offset++;
2445         }
2446         ret = 0;
2447 out:
2448         btrfs_release_path(path);
2449         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2450 out_unlock:
2451         mutex_unlock(&src->i_mutex);
2452         mutex_unlock(&inode->i_mutex);
2453         vfree(buf);
2454         btrfs_free_path(path);
2455 out_fput:
2456         fput(src_file);
2457 out_drop_write:
2458         mnt_drop_write(file->f_path.mnt);
2459         return ret;
2460 }
2461
2462 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2463 {
2464         struct btrfs_ioctl_clone_range_args args;
2465
2466         if (copy_from_user(&args, argp, sizeof(args)))
2467                 return -EFAULT;
2468         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2469                                  args.src_length, args.dest_offset);
2470 }
2471
2472 /*
2473  * there are many ways the trans_start and trans_end ioctls can lead
2474  * to deadlocks.  They should only be used by applications that
2475  * basically own the machine, and have a very in depth understanding
2476  * of all the possible deadlocks and enospc problems.
2477  */
2478 static long btrfs_ioctl_trans_start(struct file *file)
2479 {
2480         struct inode *inode = fdentry(file)->d_inode;
2481         struct btrfs_root *root = BTRFS_I(inode)->root;
2482         struct btrfs_trans_handle *trans;
2483         int ret;
2484
2485         ret = -EPERM;
2486         if (!capable(CAP_SYS_ADMIN))
2487                 goto out;
2488
2489         ret = -EINPROGRESS;
2490         if (file->private_data)
2491                 goto out;
2492
2493         ret = -EROFS;
2494         if (btrfs_root_readonly(root))
2495                 goto out;
2496
2497         ret = mnt_want_write(file->f_path.mnt);
2498         if (ret)
2499                 goto out;
2500
2501         atomic_inc(&root->fs_info->open_ioctl_trans);
2502
2503         ret = -ENOMEM;
2504         trans = btrfs_start_ioctl_transaction(root);
2505         if (IS_ERR(trans))
2506                 goto out_drop;
2507
2508         file->private_data = trans;
2509         return 0;
2510
2511 out_drop:
2512         atomic_dec(&root->fs_info->open_ioctl_trans);
2513         mnt_drop_write(file->f_path.mnt);
2514 out:
2515         return ret;
2516 }
2517
2518 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2519 {
2520         struct inode *inode = fdentry(file)->d_inode;
2521         struct btrfs_root *root = BTRFS_I(inode)->root;
2522         struct btrfs_root *new_root;
2523         struct btrfs_dir_item *di;
2524         struct btrfs_trans_handle *trans;
2525         struct btrfs_path *path;
2526         struct btrfs_key location;
2527         struct btrfs_disk_key disk_key;
2528         struct btrfs_super_block *disk_super;
2529         u64 features;
2530         u64 objectid = 0;
2531         u64 dir_id;
2532
2533         if (!capable(CAP_SYS_ADMIN))
2534                 return -EPERM;
2535
2536         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2537                 return -EFAULT;
2538
2539         if (!objectid)
2540                 objectid = root->root_key.objectid;
2541
2542         location.objectid = objectid;
2543         location.type = BTRFS_ROOT_ITEM_KEY;
2544         location.offset = (u64)-1;
2545
2546         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2547         if (IS_ERR(new_root))
2548                 return PTR_ERR(new_root);
2549
2550         if (btrfs_root_refs(&new_root->root_item) == 0)
2551                 return -ENOENT;
2552
2553         path = btrfs_alloc_path();
2554         if (!path)
2555                 return -ENOMEM;
2556         path->leave_spinning = 1;
2557
2558         trans = btrfs_start_transaction(root, 1);
2559         if (IS_ERR(trans)) {
2560                 btrfs_free_path(path);
2561                 return PTR_ERR(trans);
2562         }
2563
2564         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
2565         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2566                                    dir_id, "default", 7, 1);
2567         if (IS_ERR_OR_NULL(di)) {
2568                 btrfs_free_path(path);
2569                 btrfs_end_transaction(trans, root);
2570                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2571                        "this isn't going to work\n");
2572                 return -ENOENT;
2573         }
2574
2575         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2576         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2577         btrfs_mark_buffer_dirty(path->nodes[0]);
2578         btrfs_free_path(path);
2579
2580         disk_super = &root->fs_info->super_copy;
2581         features = btrfs_super_incompat_flags(disk_super);
2582         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2583                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2584                 btrfs_set_super_incompat_flags(disk_super, features);
2585         }
2586         btrfs_end_transaction(trans, root);
2587
2588         return 0;
2589 }
2590
2591 static void get_block_group_info(struct list_head *groups_list,
2592                                  struct btrfs_ioctl_space_info *space)
2593 {
2594         struct btrfs_block_group_cache *block_group;
2595
2596         space->total_bytes = 0;
2597         space->used_bytes = 0;
2598         space->flags = 0;
2599         list_for_each_entry(block_group, groups_list, list) {
2600                 space->flags = block_group->flags;
2601                 space->total_bytes += block_group->key.offset;
2602                 space->used_bytes +=
2603                         btrfs_block_group_used(&block_group->item);
2604         }
2605 }
2606
2607 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2608 {
2609         struct btrfs_ioctl_space_args space_args;
2610         struct btrfs_ioctl_space_info space;
2611         struct btrfs_ioctl_space_info *dest;
2612         struct btrfs_ioctl_space_info *dest_orig;
2613         struct btrfs_ioctl_space_info __user *user_dest;
2614         struct btrfs_space_info *info;
2615         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2616                        BTRFS_BLOCK_GROUP_SYSTEM,
2617                        BTRFS_BLOCK_GROUP_METADATA,
2618                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2619         int num_types = 4;
2620         int alloc_size;
2621         int ret = 0;
2622         u64 slot_count = 0;
2623         int i, c;
2624
2625         if (copy_from_user(&space_args,
2626                            (struct btrfs_ioctl_space_args __user *)arg,
2627                            sizeof(space_args)))
2628                 return -EFAULT;
2629
2630         for (i = 0; i < num_types; i++) {
2631                 struct btrfs_space_info *tmp;
2632
2633                 info = NULL;
2634                 rcu_read_lock();
2635                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2636                                         list) {
2637                         if (tmp->flags == types[i]) {
2638                                 info = tmp;
2639                                 break;
2640                         }
2641                 }
2642                 rcu_read_unlock();
2643
2644                 if (!info)
2645                         continue;
2646
2647                 down_read(&info->groups_sem);
2648                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2649                         if (!list_empty(&info->block_groups[c]))
2650                                 slot_count++;
2651                 }
2652                 up_read(&info->groups_sem);
2653         }
2654
2655         /* space_slots == 0 means they are asking for a count */
2656         if (space_args.space_slots == 0) {
2657                 space_args.total_spaces = slot_count;
2658                 goto out;
2659         }
2660
2661         slot_count = min_t(u64, space_args.space_slots, slot_count);
2662
2663         alloc_size = sizeof(*dest) * slot_count;
2664
2665         /* we generally have at most 6 or so space infos, one for each raid
2666          * level.  So, a whole page should be more than enough for everyone
2667          */
2668         if (alloc_size > PAGE_CACHE_SIZE)
2669                 return -ENOMEM;
2670
2671         space_args.total_spaces = 0;
2672         dest = kmalloc(alloc_size, GFP_NOFS);
2673         if (!dest)
2674                 return -ENOMEM;
2675         dest_orig = dest;
2676
2677         /* now we have a buffer to copy into */
2678         for (i = 0; i < num_types; i++) {
2679                 struct btrfs_space_info *tmp;
2680
2681                 if (!slot_count)
2682                         break;
2683
2684                 info = NULL;
2685                 rcu_read_lock();
2686                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2687                                         list) {
2688                         if (tmp->flags == types[i]) {
2689                                 info = tmp;
2690                                 break;
2691                         }
2692                 }
2693                 rcu_read_unlock();
2694
2695                 if (!info)
2696                         continue;
2697                 down_read(&info->groups_sem);
2698                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2699                         if (!list_empty(&info->block_groups[c])) {
2700                                 get_block_group_info(&info->block_groups[c],
2701                                                      &space);
2702                                 memcpy(dest, &space, sizeof(space));
2703                                 dest++;
2704                                 space_args.total_spaces++;
2705                                 slot_count--;
2706                         }
2707                         if (!slot_count)
2708                                 break;
2709                 }
2710                 up_read(&info->groups_sem);
2711         }
2712
2713         user_dest = (struct btrfs_ioctl_space_info *)
2714                 (arg + sizeof(struct btrfs_ioctl_space_args));
2715
2716         if (copy_to_user(user_dest, dest_orig, alloc_size))
2717                 ret = -EFAULT;
2718
2719         kfree(dest_orig);
2720 out:
2721         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2722                 ret = -EFAULT;
2723
2724         return ret;
2725 }
2726
2727 /*
2728  * there are many ways the trans_start and trans_end ioctls can lead
2729  * to deadlocks.  They should only be used by applications that
2730  * basically own the machine, and have a very in depth understanding
2731  * of all the possible deadlocks and enospc problems.
2732  */
2733 long btrfs_ioctl_trans_end(struct file *file)
2734 {
2735         struct inode *inode = fdentry(file)->d_inode;
2736         struct btrfs_root *root = BTRFS_I(inode)->root;
2737         struct btrfs_trans_handle *trans;
2738
2739         trans = file->private_data;
2740         if (!trans)
2741                 return -EINVAL;
2742         file->private_data = NULL;
2743
2744         btrfs_end_transaction(trans, root);
2745
2746         atomic_dec(&root->fs_info->open_ioctl_trans);
2747
2748         mnt_drop_write(file->f_path.mnt);
2749         return 0;
2750 }
2751
2752 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2753 {
2754         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2755         struct btrfs_trans_handle *trans;
2756         u64 transid;
2757         int ret;
2758
2759         trans = btrfs_start_transaction(root, 0);
2760         if (IS_ERR(trans))
2761                 return PTR_ERR(trans);
2762         transid = trans->transid;
2763         ret = btrfs_commit_transaction_async(trans, root, 0);
2764         if (ret) {
2765                 btrfs_end_transaction(trans, root);
2766                 return ret;
2767         }
2768
2769         if (argp)
2770                 if (copy_to_user(argp, &transid, sizeof(transid)))
2771                         return -EFAULT;
2772         return 0;
2773 }
2774
2775 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2776 {
2777         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2778         u64 transid;
2779
2780         if (argp) {
2781                 if (copy_from_user(&transid, argp, sizeof(transid)))
2782                         return -EFAULT;
2783         } else {
2784                 transid = 0;  /* current trans */
2785         }
2786         return btrfs_wait_for_commit(root, transid);
2787 }
2788
2789 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2790 {
2791         int ret;
2792         struct btrfs_ioctl_scrub_args *sa;
2793
2794         if (!capable(CAP_SYS_ADMIN))
2795                 return -EPERM;
2796
2797         sa = memdup_user(arg, sizeof(*sa));
2798         if (IS_ERR(sa))
2799                 return PTR_ERR(sa);
2800
2801         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2802                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2803
2804         if (copy_to_user(arg, sa, sizeof(*sa)))
2805                 ret = -EFAULT;
2806
2807         kfree(sa);
2808         return ret;
2809 }
2810
2811 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2812 {
2813         if (!capable(CAP_SYS_ADMIN))
2814                 return -EPERM;
2815
2816         return btrfs_scrub_cancel(root);
2817 }
2818
2819 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2820                                        void __user *arg)
2821 {
2822         struct btrfs_ioctl_scrub_args *sa;
2823         int ret;
2824
2825         if (!capable(CAP_SYS_ADMIN))
2826                 return -EPERM;
2827
2828         sa = memdup_user(arg, sizeof(*sa));
2829         if (IS_ERR(sa))
2830                 return PTR_ERR(sa);
2831
2832         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2833
2834         if (copy_to_user(arg, sa, sizeof(*sa)))
2835                 ret = -EFAULT;
2836
2837         kfree(sa);
2838         return ret;
2839 }
2840
2841 long btrfs_ioctl(struct file *file, unsigned int
2842                 cmd, unsigned long arg)
2843 {
2844         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2845         void __user *argp = (void __user *)arg;
2846
2847         switch (cmd) {
2848         case FS_IOC_GETFLAGS:
2849                 return btrfs_ioctl_getflags(file, argp);
2850         case FS_IOC_SETFLAGS:
2851                 return btrfs_ioctl_setflags(file, argp);
2852         case FS_IOC_GETVERSION:
2853                 return btrfs_ioctl_getversion(file, argp);
2854         case FITRIM:
2855                 return btrfs_ioctl_fitrim(file, argp);
2856         case BTRFS_IOC_SNAP_CREATE:
2857                 return btrfs_ioctl_snap_create(file, argp, 0);
2858         case BTRFS_IOC_SNAP_CREATE_V2:
2859                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
2860         case BTRFS_IOC_SUBVOL_CREATE:
2861                 return btrfs_ioctl_snap_create(file, argp, 1);
2862         case BTRFS_IOC_SNAP_DESTROY:
2863                 return btrfs_ioctl_snap_destroy(file, argp);
2864         case BTRFS_IOC_SUBVOL_GETFLAGS:
2865                 return btrfs_ioctl_subvol_getflags(file, argp);
2866         case BTRFS_IOC_SUBVOL_SETFLAGS:
2867                 return btrfs_ioctl_subvol_setflags(file, argp);
2868         case BTRFS_IOC_DEFAULT_SUBVOL:
2869                 return btrfs_ioctl_default_subvol(file, argp);
2870         case BTRFS_IOC_DEFRAG:
2871                 return btrfs_ioctl_defrag(file, NULL);
2872         case BTRFS_IOC_DEFRAG_RANGE:
2873                 return btrfs_ioctl_defrag(file, argp);
2874         case BTRFS_IOC_RESIZE:
2875                 return btrfs_ioctl_resize(root, argp);
2876         case BTRFS_IOC_ADD_DEV:
2877                 return btrfs_ioctl_add_dev(root, argp);
2878         case BTRFS_IOC_RM_DEV:
2879                 return btrfs_ioctl_rm_dev(root, argp);
2880         case BTRFS_IOC_FS_INFO:
2881                 return btrfs_ioctl_fs_info(root, argp);
2882         case BTRFS_IOC_DEV_INFO:
2883                 return btrfs_ioctl_dev_info(root, argp);
2884         case BTRFS_IOC_BALANCE:
2885                 return btrfs_balance(root->fs_info->dev_root);
2886         case BTRFS_IOC_CLONE:
2887                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
2888         case BTRFS_IOC_CLONE_RANGE:
2889                 return btrfs_ioctl_clone_range(file, argp);
2890         case BTRFS_IOC_TRANS_START:
2891                 return btrfs_ioctl_trans_start(file);
2892         case BTRFS_IOC_TRANS_END:
2893                 return btrfs_ioctl_trans_end(file);
2894         case BTRFS_IOC_TREE_SEARCH:
2895                 return btrfs_ioctl_tree_search(file, argp);
2896         case BTRFS_IOC_INO_LOOKUP:
2897                 return btrfs_ioctl_ino_lookup(file, argp);
2898         case BTRFS_IOC_SPACE_INFO:
2899                 return btrfs_ioctl_space_info(root, argp);
2900         case BTRFS_IOC_SYNC:
2901                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
2902                 return 0;
2903         case BTRFS_IOC_START_SYNC:
2904                 return btrfs_ioctl_start_sync(file, argp);
2905         case BTRFS_IOC_WAIT_SYNC:
2906                 return btrfs_ioctl_wait_sync(file, argp);
2907         case BTRFS_IOC_SCRUB:
2908                 return btrfs_ioctl_scrub(root, argp);
2909         case BTRFS_IOC_SCRUB_CANCEL:
2910                 return btrfs_ioctl_scrub_cancel(root, argp);
2911         case BTRFS_IOC_SCRUB_PROGRESS:
2912                 return btrfs_ioctl_scrub_progress(root, argp);
2913         }
2914
2915         return -ENOTTY;
2916 }