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