Merge branch 'btrfs-3.0' into for-linus
[linux-2.6.git] / fs / btrfs / ioctl.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include "compat.h"
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "ioctl.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54
55 /* Mask out flags that are inappropriate for the given type of inode. */
56 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
57 {
58         if (S_ISDIR(mode))
59                 return flags;
60         else if (S_ISREG(mode))
61                 return flags & ~FS_DIRSYNC_FL;
62         else
63                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
64 }
65
66 /*
67  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
68  */
69 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
70 {
71         unsigned int iflags = 0;
72
73         if (flags & BTRFS_INODE_SYNC)
74                 iflags |= FS_SYNC_FL;
75         if (flags & BTRFS_INODE_IMMUTABLE)
76                 iflags |= FS_IMMUTABLE_FL;
77         if (flags & BTRFS_INODE_APPEND)
78                 iflags |= FS_APPEND_FL;
79         if (flags & BTRFS_INODE_NODUMP)
80                 iflags |= FS_NODUMP_FL;
81         if (flags & BTRFS_INODE_NOATIME)
82                 iflags |= FS_NOATIME_FL;
83         if (flags & BTRFS_INODE_DIRSYNC)
84                 iflags |= FS_DIRSYNC_FL;
85         if (flags & BTRFS_INODE_NODATACOW)
86                 iflags |= FS_NOCOW_FL;
87
88         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
89                 iflags |= FS_COMPR_FL;
90         else if (flags & BTRFS_INODE_NOCOMPRESS)
91                 iflags |= FS_NOCOMP_FL;
92
93         return iflags;
94 }
95
96 /*
97  * Update inode->i_flags based on the btrfs internal flags.
98  */
99 void btrfs_update_iflags(struct inode *inode)
100 {
101         struct btrfs_inode *ip = BTRFS_I(inode);
102
103         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
104
105         if (ip->flags & BTRFS_INODE_SYNC)
106                 inode->i_flags |= S_SYNC;
107         if (ip->flags & BTRFS_INODE_IMMUTABLE)
108                 inode->i_flags |= S_IMMUTABLE;
109         if (ip->flags & BTRFS_INODE_APPEND)
110                 inode->i_flags |= S_APPEND;
111         if (ip->flags & BTRFS_INODE_NOATIME)
112                 inode->i_flags |= S_NOATIME;
113         if (ip->flags & BTRFS_INODE_DIRSYNC)
114                 inode->i_flags |= S_DIRSYNC;
115 }
116
117 /*
118  * Inherit flags from the parent inode.
119  *
120  * Unlike extN we don't have any flags we don't want to inherit currently.
121  */
122 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
123 {
124         unsigned int flags;
125
126         if (!dir)
127                 return;
128
129         flags = BTRFS_I(dir)->flags;
130
131         if (S_ISREG(inode->i_mode))
132                 flags &= ~BTRFS_INODE_DIRSYNC;
133         else if (!S_ISDIR(inode->i_mode))
134                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
135
136         BTRFS_I(inode)->flags = flags;
137         btrfs_update_iflags(inode);
138 }
139
140 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
141 {
142         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
143         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
144
145         if (copy_to_user(arg, &flags, sizeof(flags)))
146                 return -EFAULT;
147         return 0;
148 }
149
150 static int check_flags(unsigned int flags)
151 {
152         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
153                       FS_NOATIME_FL | FS_NODUMP_FL | \
154                       FS_SYNC_FL | FS_DIRSYNC_FL | \
155                       FS_NOCOMP_FL | FS_COMPR_FL |
156                       FS_NOCOW_FL))
157                 return -EOPNOTSUPP;
158
159         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
160                 return -EINVAL;
161
162         return 0;
163 }
164
165 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
166 {
167         struct inode *inode = file->f_path.dentry->d_inode;
168         struct btrfs_inode *ip = BTRFS_I(inode);
169         struct btrfs_root *root = ip->root;
170         struct btrfs_trans_handle *trans;
171         unsigned int flags, oldflags;
172         int ret;
173
174         if (btrfs_root_readonly(root))
175                 return -EROFS;
176
177         if (copy_from_user(&flags, arg, sizeof(flags)))
178                 return -EFAULT;
179
180         ret = check_flags(flags);
181         if (ret)
182                 return ret;
183
184         if (!inode_owner_or_capable(inode))
185                 return -EACCES;
186
187         mutex_lock(&inode->i_mutex);
188
189         flags = btrfs_mask_flags(inode->i_mode, flags);
190         oldflags = btrfs_flags_to_ioctl(ip->flags);
191         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
192                 if (!capable(CAP_LINUX_IMMUTABLE)) {
193                         ret = -EPERM;
194                         goto out_unlock;
195                 }
196         }
197
198         ret = mnt_want_write(file->f_path.mnt);
199         if (ret)
200                 goto out_unlock;
201
202         if (flags & FS_SYNC_FL)
203                 ip->flags |= BTRFS_INODE_SYNC;
204         else
205                 ip->flags &= ~BTRFS_INODE_SYNC;
206         if (flags & FS_IMMUTABLE_FL)
207                 ip->flags |= BTRFS_INODE_IMMUTABLE;
208         else
209                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
210         if (flags & FS_APPEND_FL)
211                 ip->flags |= BTRFS_INODE_APPEND;
212         else
213                 ip->flags &= ~BTRFS_INODE_APPEND;
214         if (flags & FS_NODUMP_FL)
215                 ip->flags |= BTRFS_INODE_NODUMP;
216         else
217                 ip->flags &= ~BTRFS_INODE_NODUMP;
218         if (flags & FS_NOATIME_FL)
219                 ip->flags |= BTRFS_INODE_NOATIME;
220         else
221                 ip->flags &= ~BTRFS_INODE_NOATIME;
222         if (flags & FS_DIRSYNC_FL)
223                 ip->flags |= BTRFS_INODE_DIRSYNC;
224         else
225                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
226         if (flags & FS_NOCOW_FL)
227                 ip->flags |= BTRFS_INODE_NODATACOW;
228         else
229                 ip->flags &= ~BTRFS_INODE_NODATACOW;
230
231         /*
232          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
233          * flag may be changed automatically if compression code won't make
234          * things smaller.
235          */
236         if (flags & FS_NOCOMP_FL) {
237                 ip->flags &= ~BTRFS_INODE_COMPRESS;
238                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
239         } else if (flags & FS_COMPR_FL) {
240                 ip->flags |= BTRFS_INODE_COMPRESS;
241                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
242         } else {
243                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
244         }
245
246         trans = btrfs_join_transaction(root);
247         BUG_ON(IS_ERR(trans));
248
249         ret = btrfs_update_inode(trans, root, inode);
250         BUG_ON(ret);
251
252         btrfs_update_iflags(inode);
253         inode->i_ctime = CURRENT_TIME;
254         btrfs_end_transaction(trans, root);
255
256         mnt_drop_write(file->f_path.mnt);
257
258         ret = 0;
259  out_unlock:
260         mutex_unlock(&inode->i_mutex);
261         return ret;
262 }
263
264 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
265 {
266         struct inode *inode = file->f_path.dentry->d_inode;
267
268         return put_user(inode->i_generation, arg);
269 }
270
271 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
272 {
273         struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info;
274         struct btrfs_fs_info *fs_info = root->fs_info;
275         struct btrfs_device *device;
276         struct request_queue *q;
277         struct fstrim_range range;
278         u64 minlen = ULLONG_MAX;
279         u64 num_devices = 0;
280         int ret;
281
282         if (!capable(CAP_SYS_ADMIN))
283                 return -EPERM;
284
285         rcu_read_lock();
286         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
287                                 dev_list) {
288                 if (!device->bdev)
289                         continue;
290                 q = bdev_get_queue(device->bdev);
291                 if (blk_queue_discard(q)) {
292                         num_devices++;
293                         minlen = min((u64)q->limits.discard_granularity,
294                                      minlen);
295                 }
296         }
297         rcu_read_unlock();
298         if (!num_devices)
299                 return -EOPNOTSUPP;
300
301         if (copy_from_user(&range, arg, sizeof(range)))
302                 return -EFAULT;
303
304         range.minlen = max(range.minlen, minlen);
305         ret = btrfs_trim_fs(root, &range);
306         if (ret < 0)
307                 return ret;
308
309         if (copy_to_user(arg, &range, sizeof(range)))
310                 return -EFAULT;
311
312         return 0;
313 }
314
315 static noinline int create_subvol(struct btrfs_root *root,
316                                   struct dentry *dentry,
317                                   char *name, int namelen,
318                                   u64 *async_transid)
319 {
320         struct btrfs_trans_handle *trans;
321         struct btrfs_key key;
322         struct btrfs_root_item root_item;
323         struct btrfs_inode_item *inode_item;
324         struct extent_buffer *leaf;
325         struct btrfs_root *new_root;
326         struct dentry *parent = dentry->d_parent;
327         struct inode *dir;
328         int ret;
329         int err;
330         u64 objectid;
331         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
332         u64 index = 0;
333
334         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
335         if (ret)
336                 return ret;
337
338         dir = parent->d_inode;
339
340         /*
341          * 1 - inode item
342          * 2 - refs
343          * 1 - root item
344          * 2 - dir items
345          */
346         trans = btrfs_start_transaction(root, 6);
347         if (IS_ERR(trans))
348                 return PTR_ERR(trans);
349
350         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
351                                       0, objectid, NULL, 0, 0, 0);
352         if (IS_ERR(leaf)) {
353                 ret = PTR_ERR(leaf);
354                 goto fail;
355         }
356
357         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
358         btrfs_set_header_bytenr(leaf, leaf->start);
359         btrfs_set_header_generation(leaf, trans->transid);
360         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
361         btrfs_set_header_owner(leaf, objectid);
362
363         write_extent_buffer(leaf, root->fs_info->fsid,
364                             (unsigned long)btrfs_header_fsid(leaf),
365                             BTRFS_FSID_SIZE);
366         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
367                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
368                             BTRFS_UUID_SIZE);
369         btrfs_mark_buffer_dirty(leaf);
370
371         inode_item = &root_item.inode;
372         memset(inode_item, 0, sizeof(*inode_item));
373         inode_item->generation = cpu_to_le64(1);
374         inode_item->size = cpu_to_le64(3);
375         inode_item->nlink = cpu_to_le32(1);
376         inode_item->nbytes = cpu_to_le64(root->leafsize);
377         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
378
379         root_item.flags = 0;
380         root_item.byte_limit = 0;
381         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
382
383         btrfs_set_root_bytenr(&root_item, leaf->start);
384         btrfs_set_root_generation(&root_item, trans->transid);
385         btrfs_set_root_level(&root_item, 0);
386         btrfs_set_root_refs(&root_item, 1);
387         btrfs_set_root_used(&root_item, leaf->len);
388         btrfs_set_root_last_snapshot(&root_item, 0);
389
390         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
391         root_item.drop_level = 0;
392
393         btrfs_tree_unlock(leaf);
394         free_extent_buffer(leaf);
395         leaf = NULL;
396
397         btrfs_set_root_dirid(&root_item, new_dirid);
398
399         key.objectid = objectid;
400         key.offset = 0;
401         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
402         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
403                                 &root_item);
404         if (ret)
405                 goto fail;
406
407         key.offset = (u64)-1;
408         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
409         BUG_ON(IS_ERR(new_root));
410
411         btrfs_record_root_in_trans(trans, new_root);
412
413         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
414         /*
415          * insert the directory item
416          */
417         ret = btrfs_set_inode_index(dir, &index);
418         BUG_ON(ret);
419
420         ret = btrfs_insert_dir_item(trans, root,
421                                     name, namelen, dir, &key,
422                                     BTRFS_FT_DIR, index);
423         if (ret)
424                 goto fail;
425
426         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
427         ret = btrfs_update_inode(trans, root, dir);
428         BUG_ON(ret);
429
430         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
431                                  objectid, root->root_key.objectid,
432                                  btrfs_ino(dir), index, name, namelen);
433
434         BUG_ON(ret);
435
436         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
437 fail:
438         if (async_transid) {
439                 *async_transid = trans->transid;
440                 err = btrfs_commit_transaction_async(trans, root, 1);
441         } else {
442                 err = btrfs_commit_transaction(trans, root);
443         }
444         if (err && !ret)
445                 ret = err;
446         return ret;
447 }
448
449 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
450                            char *name, int namelen, u64 *async_transid,
451                            bool readonly)
452 {
453         struct inode *inode;
454         struct btrfs_pending_snapshot *pending_snapshot;
455         struct btrfs_trans_handle *trans;
456         int ret;
457
458         if (!root->ref_cows)
459                 return -EINVAL;
460
461         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
462         if (!pending_snapshot)
463                 return -ENOMEM;
464
465         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
466         pending_snapshot->dentry = dentry;
467         pending_snapshot->root = root;
468         pending_snapshot->readonly = readonly;
469
470         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
471         if (IS_ERR(trans)) {
472                 ret = PTR_ERR(trans);
473                 goto fail;
474         }
475
476         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
477         BUG_ON(ret);
478
479         spin_lock(&root->fs_info->trans_lock);
480         list_add(&pending_snapshot->list,
481                  &trans->transaction->pending_snapshots);
482         spin_unlock(&root->fs_info->trans_lock);
483         if (async_transid) {
484                 *async_transid = trans->transid;
485                 ret = btrfs_commit_transaction_async(trans,
486                                      root->fs_info->extent_root, 1);
487         } else {
488                 ret = btrfs_commit_transaction(trans,
489                                                root->fs_info->extent_root);
490         }
491         BUG_ON(ret);
492
493         ret = pending_snapshot->error;
494         if (ret)
495                 goto fail;
496
497         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
498         if (ret)
499                 goto fail;
500
501         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
502         if (IS_ERR(inode)) {
503                 ret = PTR_ERR(inode);
504                 goto fail;
505         }
506         BUG_ON(!inode);
507         d_instantiate(dentry, inode);
508         ret = 0;
509 fail:
510         kfree(pending_snapshot);
511         return ret;
512 }
513
514 /*  copy of check_sticky in fs/namei.c()
515 * It's inline, so penalty for filesystems that don't use sticky bit is
516 * minimal.
517 */
518 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
519 {
520         uid_t fsuid = current_fsuid();
521
522         if (!(dir->i_mode & S_ISVTX))
523                 return 0;
524         if (inode->i_uid == fsuid)
525                 return 0;
526         if (dir->i_uid == fsuid)
527                 return 0;
528         return !capable(CAP_FOWNER);
529 }
530
531 /*  copy of may_delete in fs/namei.c()
532  *      Check whether we can remove a link victim from directory dir, check
533  *  whether the type of victim is right.
534  *  1. We can't do it if dir is read-only (done in permission())
535  *  2. We should have write and exec permissions on dir
536  *  3. We can't remove anything from append-only dir
537  *  4. We can't do anything with immutable dir (done in permission())
538  *  5. If the sticky bit on dir is set we should either
539  *      a. be owner of dir, or
540  *      b. be owner of victim, or
541  *      c. have CAP_FOWNER capability
542  *  6. If the victim is append-only or immutable we can't do antyhing with
543  *     links pointing to it.
544  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
545  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
546  *  9. We can't remove a root or mountpoint.
547  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
548  *     nfs_async_unlink().
549  */
550
551 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
552 {
553         int error;
554
555         if (!victim->d_inode)
556                 return -ENOENT;
557
558         BUG_ON(victim->d_parent->d_inode != dir);
559         audit_inode_child(victim, dir);
560
561         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
562         if (error)
563                 return error;
564         if (IS_APPEND(dir))
565                 return -EPERM;
566         if (btrfs_check_sticky(dir, victim->d_inode)||
567                 IS_APPEND(victim->d_inode)||
568             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
569                 return -EPERM;
570         if (isdir) {
571                 if (!S_ISDIR(victim->d_inode->i_mode))
572                         return -ENOTDIR;
573                 if (IS_ROOT(victim))
574                         return -EBUSY;
575         } else if (S_ISDIR(victim->d_inode->i_mode))
576                 return -EISDIR;
577         if (IS_DEADDIR(dir))
578                 return -ENOENT;
579         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
580                 return -EBUSY;
581         return 0;
582 }
583
584 /* copy of may_create in fs/namei.c() */
585 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
586 {
587         if (child->d_inode)
588                 return -EEXIST;
589         if (IS_DEADDIR(dir))
590                 return -ENOENT;
591         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
592 }
593
594 /*
595  * Create a new subvolume below @parent.  This is largely modeled after
596  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
597  * inside this filesystem so it's quite a bit simpler.
598  */
599 static noinline int btrfs_mksubvol(struct path *parent,
600                                    char *name, int namelen,
601                                    struct btrfs_root *snap_src,
602                                    u64 *async_transid, bool readonly)
603 {
604         struct inode *dir  = parent->dentry->d_inode;
605         struct dentry *dentry;
606         int error;
607
608         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
609
610         dentry = lookup_one_len(name, parent->dentry, namelen);
611         error = PTR_ERR(dentry);
612         if (IS_ERR(dentry))
613                 goto out_unlock;
614
615         error = -EEXIST;
616         if (dentry->d_inode)
617                 goto out_dput;
618
619         error = mnt_want_write(parent->mnt);
620         if (error)
621                 goto out_dput;
622
623         error = btrfs_may_create(dir, dentry);
624         if (error)
625                 goto out_drop_write;
626
627         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
628
629         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
630                 goto out_up_read;
631
632         if (snap_src) {
633                 error = create_snapshot(snap_src, dentry,
634                                         name, namelen, async_transid, readonly);
635         } else {
636                 error = create_subvol(BTRFS_I(dir)->root, dentry,
637                                       name, namelen, async_transid);
638         }
639         if (!error)
640                 fsnotify_mkdir(dir, dentry);
641 out_up_read:
642         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
643 out_drop_write:
644         mnt_drop_write(parent->mnt);
645 out_dput:
646         dput(dentry);
647 out_unlock:
648         mutex_unlock(&dir->i_mutex);
649         return error;
650 }
651
652 /*
653  * When we're defragging a range, we don't want to kick it off again
654  * if it is really just waiting for delalloc to send it down.
655  * If we find a nice big extent or delalloc range for the bytes in the
656  * file you want to defrag, we return 0 to let you know to skip this
657  * part of the file
658  */
659 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
660 {
661         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
662         struct extent_map *em = NULL;
663         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
664         u64 end;
665
666         read_lock(&em_tree->lock);
667         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
668         read_unlock(&em_tree->lock);
669
670         if (em) {
671                 end = extent_map_end(em);
672                 free_extent_map(em);
673                 if (end - offset > thresh)
674                         return 0;
675         }
676         /* if we already have a nice delalloc here, just stop */
677         thresh /= 2;
678         end = count_range_bits(io_tree, &offset, offset + thresh,
679                                thresh, EXTENT_DELALLOC, 1);
680         if (end >= thresh)
681                 return 0;
682         return 1;
683 }
684
685 /*
686  * helper function to walk through a file and find extents
687  * newer than a specific transid, and smaller than thresh.
688  *
689  * This is used by the defragging code to find new and small
690  * extents
691  */
692 static int find_new_extents(struct btrfs_root *root,
693                             struct inode *inode, u64 newer_than,
694                             u64 *off, int thresh)
695 {
696         struct btrfs_path *path;
697         struct btrfs_key min_key;
698         struct btrfs_key max_key;
699         struct extent_buffer *leaf;
700         struct btrfs_file_extent_item *extent;
701         int type;
702         int ret;
703         u64 ino = btrfs_ino(inode);
704
705         path = btrfs_alloc_path();
706         if (!path)
707                 return -ENOMEM;
708
709         min_key.objectid = ino;
710         min_key.type = BTRFS_EXTENT_DATA_KEY;
711         min_key.offset = *off;
712
713         max_key.objectid = ino;
714         max_key.type = (u8)-1;
715         max_key.offset = (u64)-1;
716
717         path->keep_locks = 1;
718
719         while(1) {
720                 ret = btrfs_search_forward(root, &min_key, &max_key,
721                                            path, 0, newer_than);
722                 if (ret != 0)
723                         goto none;
724                 if (min_key.objectid != ino)
725                         goto none;
726                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
727                         goto none;
728
729                 leaf = path->nodes[0];
730                 extent = btrfs_item_ptr(leaf, path->slots[0],
731                                         struct btrfs_file_extent_item);
732
733                 type = btrfs_file_extent_type(leaf, extent);
734                 if (type == BTRFS_FILE_EXTENT_REG &&
735                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
736                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
737                         *off = min_key.offset;
738                         btrfs_free_path(path);
739                         return 0;
740                 }
741
742                 if (min_key.offset == (u64)-1)
743                         goto none;
744
745                 min_key.offset++;
746                 btrfs_release_path(path);
747         }
748 none:
749         btrfs_free_path(path);
750         return -ENOENT;
751 }
752
753 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
754                                int thresh, u64 *last_len, u64 *skip,
755                                u64 *defrag_end)
756 {
757         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
758         struct extent_map *em = NULL;
759         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
760         int ret = 1;
761
762         /*
763          * make sure that once we start defragging and extent, we keep on
764          * defragging it
765          */
766         if (start < *defrag_end)
767                 return 1;
768
769         *skip = 0;
770
771         /*
772          * hopefully we have this extent in the tree already, try without
773          * the full extent lock
774          */
775         read_lock(&em_tree->lock);
776         em = lookup_extent_mapping(em_tree, start, len);
777         read_unlock(&em_tree->lock);
778
779         if (!em) {
780                 /* get the big lock and read metadata off disk */
781                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
782                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
783                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
784
785                 if (IS_ERR(em))
786                         return 0;
787         }
788
789         /* this will cover holes, and inline extents */
790         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
791                 ret = 0;
792
793         /*
794          * we hit a real extent, if it is big don't bother defragging it again
795          */
796         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
797                 ret = 0;
798
799         /*
800          * last_len ends up being a counter of how many bytes we've defragged.
801          * every time we choose not to defrag an extent, we reset *last_len
802          * so that the next tiny extent will force a defrag.
803          *
804          * The end result of this is that tiny extents before a single big
805          * extent will force at least part of that big extent to be defragged.
806          */
807         if (ret) {
808                 *last_len += len;
809                 *defrag_end = extent_map_end(em);
810         } else {
811                 *last_len = 0;
812                 *skip = extent_map_end(em);
813                 *defrag_end = 0;
814         }
815
816         free_extent_map(em);
817         return ret;
818 }
819
820 /*
821  * it doesn't do much good to defrag one or two pages
822  * at a time.  This pulls in a nice chunk of pages
823  * to COW and defrag.
824  *
825  * It also makes sure the delalloc code has enough
826  * dirty data to avoid making new small extents as part
827  * of the defrag
828  *
829  * It's a good idea to start RA on this range
830  * before calling this.
831  */
832 static int cluster_pages_for_defrag(struct inode *inode,
833                                     struct page **pages,
834                                     unsigned long start_index,
835                                     int num_pages)
836 {
837         unsigned long file_end;
838         u64 isize = i_size_read(inode);
839         u64 page_start;
840         u64 page_end;
841         int ret;
842         int i;
843         int i_done;
844         struct btrfs_ordered_extent *ordered;
845         struct extent_state *cached_state = NULL;
846
847         if (isize == 0)
848                 return 0;
849         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
850
851         ret = btrfs_delalloc_reserve_space(inode,
852                                            num_pages << PAGE_CACHE_SHIFT);
853         if (ret)
854                 return ret;
855 again:
856         ret = 0;
857         i_done = 0;
858
859         /* step one, lock all the pages */
860         for (i = 0; i < num_pages; i++) {
861                 struct page *page;
862                 page = find_or_create_page(inode->i_mapping,
863                                             start_index + i, GFP_NOFS);
864                 if (!page)
865                         break;
866
867                 if (!PageUptodate(page)) {
868                         btrfs_readpage(NULL, page);
869                         lock_page(page);
870                         if (!PageUptodate(page)) {
871                                 unlock_page(page);
872                                 page_cache_release(page);
873                                 ret = -EIO;
874                                 break;
875                         }
876                 }
877                 isize = i_size_read(inode);
878                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
879                 if (!isize || page->index > file_end ||
880                     page->mapping != inode->i_mapping) {
881                         /* whoops, we blew past eof, skip this page */
882                         unlock_page(page);
883                         page_cache_release(page);
884                         break;
885                 }
886                 pages[i] = page;
887                 i_done++;
888         }
889         if (!i_done || ret)
890                 goto out;
891
892         if (!(inode->i_sb->s_flags & MS_ACTIVE))
893                 goto out;
894
895         /*
896          * so now we have a nice long stream of locked
897          * and up to date pages, lets wait on them
898          */
899         for (i = 0; i < i_done; i++)
900                 wait_on_page_writeback(pages[i]);
901
902         page_start = page_offset(pages[0]);
903         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
904
905         lock_extent_bits(&BTRFS_I(inode)->io_tree,
906                          page_start, page_end - 1, 0, &cached_state,
907                          GFP_NOFS);
908         ordered = btrfs_lookup_first_ordered_extent(inode, page_end - 1);
909         if (ordered &&
910             ordered->file_offset + ordered->len > page_start &&
911             ordered->file_offset < page_end) {
912                 btrfs_put_ordered_extent(ordered);
913                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
914                                      page_start, page_end - 1,
915                                      &cached_state, GFP_NOFS);
916                 for (i = 0; i < i_done; i++) {
917                         unlock_page(pages[i]);
918                         page_cache_release(pages[i]);
919                 }
920                 btrfs_wait_ordered_range(inode, page_start,
921                                          page_end - page_start);
922                 goto again;
923         }
924         if (ordered)
925                 btrfs_put_ordered_extent(ordered);
926
927         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
928                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
929                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
930                           GFP_NOFS);
931
932         if (i_done != num_pages) {
933                 spin_lock(&BTRFS_I(inode)->lock);
934                 BTRFS_I(inode)->outstanding_extents++;
935                 spin_unlock(&BTRFS_I(inode)->lock);
936                 btrfs_delalloc_release_space(inode,
937                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
938         }
939
940
941         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
942                                   &cached_state);
943
944         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
945                              page_start, page_end - 1, &cached_state,
946                              GFP_NOFS);
947
948         for (i = 0; i < i_done; i++) {
949                 clear_page_dirty_for_io(pages[i]);
950                 ClearPageChecked(pages[i]);
951                 set_page_extent_mapped(pages[i]);
952                 set_page_dirty(pages[i]);
953                 unlock_page(pages[i]);
954                 page_cache_release(pages[i]);
955         }
956         return i_done;
957 out:
958         for (i = 0; i < i_done; i++) {
959                 unlock_page(pages[i]);
960                 page_cache_release(pages[i]);
961         }
962         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
963         return ret;
964
965 }
966
967 int btrfs_defrag_file(struct inode *inode, struct file *file,
968                       struct btrfs_ioctl_defrag_range_args *range,
969                       u64 newer_than, unsigned long max_to_defrag)
970 {
971         struct btrfs_root *root = BTRFS_I(inode)->root;
972         struct btrfs_super_block *disk_super;
973         struct file_ra_state *ra = NULL;
974         unsigned long last_index;
975         u64 features;
976         u64 last_len = 0;
977         u64 skip = 0;
978         u64 defrag_end = 0;
979         u64 newer_off = range->start;
980         int newer_left = 0;
981         unsigned long i;
982         int ret;
983         int defrag_count = 0;
984         int compress_type = BTRFS_COMPRESS_ZLIB;
985         int extent_thresh = range->extent_thresh;
986         int newer_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
987         u64 new_align = ~((u64)128 * 1024 - 1);
988         struct page **pages = NULL;
989
990         if (extent_thresh == 0)
991                 extent_thresh = 256 * 1024;
992
993         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
994                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
995                         return -EINVAL;
996                 if (range->compress_type)
997                         compress_type = range->compress_type;
998         }
999
1000         if (inode->i_size == 0)
1001                 return 0;
1002
1003         /*
1004          * if we were not given a file, allocate a readahead
1005          * context
1006          */
1007         if (!file) {
1008                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1009                 if (!ra)
1010                         return -ENOMEM;
1011                 file_ra_state_init(ra, inode->i_mapping);
1012         } else {
1013                 ra = &file->f_ra;
1014         }
1015
1016         pages = kmalloc(sizeof(struct page *) * newer_cluster,
1017                         GFP_NOFS);
1018         if (!pages) {
1019                 ret = -ENOMEM;
1020                 goto out_ra;
1021         }
1022
1023         /* find the last page to defrag */
1024         if (range->start + range->len > range->start) {
1025                 last_index = min_t(u64, inode->i_size - 1,
1026                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1027         } else {
1028                 last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
1029         }
1030
1031         if (newer_than) {
1032                 ret = find_new_extents(root, inode, newer_than,
1033                                        &newer_off, 64 * 1024);
1034                 if (!ret) {
1035                         range->start = newer_off;
1036                         /*
1037                          * we always align our defrag to help keep
1038                          * the extents in the file evenly spaced
1039                          */
1040                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1041                         newer_left = newer_cluster;
1042                 } else
1043                         goto out_ra;
1044         } else {
1045                 i = range->start >> PAGE_CACHE_SHIFT;
1046         }
1047         if (!max_to_defrag)
1048                 max_to_defrag = last_index - 1;
1049
1050         while (i <= last_index && defrag_count < max_to_defrag) {
1051                 /*
1052                  * make sure we stop running if someone unmounts
1053                  * the FS
1054                  */
1055                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1056                         break;
1057
1058                 if (!newer_than &&
1059                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1060                                         PAGE_CACHE_SIZE,
1061                                         extent_thresh,
1062                                         &last_len, &skip,
1063                                         &defrag_end)) {
1064                         unsigned long next;
1065                         /*
1066                          * the should_defrag function tells us how much to skip
1067                          * bump our counter by the suggested amount
1068                          */
1069                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1070                         i = max(i + 1, next);
1071                         continue;
1072                 }
1073                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1074                         BTRFS_I(inode)->force_compress = compress_type;
1075
1076                 btrfs_force_ra(inode->i_mapping, ra, file, i, newer_cluster);
1077
1078                 ret = cluster_pages_for_defrag(inode, pages, i, newer_cluster);
1079                 if (ret < 0)
1080                         goto out_ra;
1081
1082                 defrag_count += ret;
1083                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1084                 i += ret;
1085
1086                 if (newer_than) {
1087                         if (newer_off == (u64)-1)
1088                                 break;
1089
1090                         newer_off = max(newer_off + 1,
1091                                         (u64)i << PAGE_CACHE_SHIFT);
1092
1093                         ret = find_new_extents(root, inode,
1094                                                newer_than, &newer_off,
1095                                                64 * 1024);
1096                         if (!ret) {
1097                                 range->start = newer_off;
1098                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1099                                 newer_left = newer_cluster;
1100                         } else {
1101                                 break;
1102                         }
1103                 } else {
1104                         i++;
1105                 }
1106         }
1107
1108         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1109                 filemap_flush(inode->i_mapping);
1110
1111         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1112                 /* the filemap_flush will queue IO into the worker threads, but
1113                  * we have to make sure the IO is actually started and that
1114                  * ordered extents get created before we return
1115                  */
1116                 atomic_inc(&root->fs_info->async_submit_draining);
1117                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1118                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1119                         wait_event(root->fs_info->async_submit_wait,
1120                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1121                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1122                 }
1123                 atomic_dec(&root->fs_info->async_submit_draining);
1124
1125                 mutex_lock(&inode->i_mutex);
1126                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1127                 mutex_unlock(&inode->i_mutex);
1128         }
1129
1130         disk_super = &root->fs_info->super_copy;
1131         features = btrfs_super_incompat_flags(disk_super);
1132         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1133                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1134                 btrfs_set_super_incompat_flags(disk_super, features);
1135         }
1136
1137         if (!file)
1138                 kfree(ra);
1139         return defrag_count;
1140
1141 out_ra:
1142         if (!file)
1143                 kfree(ra);
1144         kfree(pages);
1145         return ret;
1146 }
1147
1148 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1149                                         void __user *arg)
1150 {
1151         u64 new_size;
1152         u64 old_size;
1153         u64 devid = 1;
1154         struct btrfs_ioctl_vol_args *vol_args;
1155         struct btrfs_trans_handle *trans;
1156         struct btrfs_device *device = NULL;
1157         char *sizestr;
1158         char *devstr = NULL;
1159         int ret = 0;
1160         int mod = 0;
1161
1162         if (root->fs_info->sb->s_flags & MS_RDONLY)
1163                 return -EROFS;
1164
1165         if (!capable(CAP_SYS_ADMIN))
1166                 return -EPERM;
1167
1168         vol_args = memdup_user(arg, sizeof(*vol_args));
1169         if (IS_ERR(vol_args))
1170                 return PTR_ERR(vol_args);
1171
1172         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1173
1174         mutex_lock(&root->fs_info->volume_mutex);
1175         sizestr = vol_args->name;
1176         devstr = strchr(sizestr, ':');
1177         if (devstr) {
1178                 char *end;
1179                 sizestr = devstr + 1;
1180                 *devstr = '\0';
1181                 devstr = vol_args->name;
1182                 devid = simple_strtoull(devstr, &end, 10);
1183                 printk(KERN_INFO "resizing devid %llu\n",
1184                        (unsigned long long)devid);
1185         }
1186         device = btrfs_find_device(root, devid, NULL, NULL);
1187         if (!device) {
1188                 printk(KERN_INFO "resizer unable to find device %llu\n",
1189                        (unsigned long long)devid);
1190                 ret = -EINVAL;
1191                 goto out_unlock;
1192         }
1193         if (!strcmp(sizestr, "max"))
1194                 new_size = device->bdev->bd_inode->i_size;
1195         else {
1196                 if (sizestr[0] == '-') {
1197                         mod = -1;
1198                         sizestr++;
1199                 } else if (sizestr[0] == '+') {
1200                         mod = 1;
1201                         sizestr++;
1202                 }
1203                 new_size = memparse(sizestr, NULL);
1204                 if (new_size == 0) {
1205                         ret = -EINVAL;
1206                         goto out_unlock;
1207                 }
1208         }
1209
1210         old_size = device->total_bytes;
1211
1212         if (mod < 0) {
1213                 if (new_size > old_size) {
1214                         ret = -EINVAL;
1215                         goto out_unlock;
1216                 }
1217                 new_size = old_size - new_size;
1218         } else if (mod > 0) {
1219                 new_size = old_size + new_size;
1220         }
1221
1222         if (new_size < 256 * 1024 * 1024) {
1223                 ret = -EINVAL;
1224                 goto out_unlock;
1225         }
1226         if (new_size > device->bdev->bd_inode->i_size) {
1227                 ret = -EFBIG;
1228                 goto out_unlock;
1229         }
1230
1231         do_div(new_size, root->sectorsize);
1232         new_size *= root->sectorsize;
1233
1234         printk(KERN_INFO "new size for %s is %llu\n",
1235                 device->name, (unsigned long long)new_size);
1236
1237         if (new_size > old_size) {
1238                 trans = btrfs_start_transaction(root, 0);
1239                 if (IS_ERR(trans)) {
1240                         ret = PTR_ERR(trans);
1241                         goto out_unlock;
1242                 }
1243                 ret = btrfs_grow_device(trans, device, new_size);
1244                 btrfs_commit_transaction(trans, root);
1245         } else {
1246                 ret = btrfs_shrink_device(device, new_size);
1247         }
1248
1249 out_unlock:
1250         mutex_unlock(&root->fs_info->volume_mutex);
1251         kfree(vol_args);
1252         return ret;
1253 }
1254
1255 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1256                                                     char *name,
1257                                                     unsigned long fd,
1258                                                     int subvol,
1259                                                     u64 *transid,
1260                                                     bool readonly)
1261 {
1262         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1263         struct file *src_file;
1264         int namelen;
1265         int ret = 0;
1266
1267         if (root->fs_info->sb->s_flags & MS_RDONLY)
1268                 return -EROFS;
1269
1270         namelen = strlen(name);
1271         if (strchr(name, '/')) {
1272                 ret = -EINVAL;
1273                 goto out;
1274         }
1275
1276         if (subvol) {
1277                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1278                                      NULL, transid, readonly);
1279         } else {
1280                 struct inode *src_inode;
1281                 src_file = fget(fd);
1282                 if (!src_file) {
1283                         ret = -EINVAL;
1284                         goto out;
1285                 }
1286
1287                 src_inode = src_file->f_path.dentry->d_inode;
1288                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1289                         printk(KERN_INFO "btrfs: Snapshot src from "
1290                                "another FS\n");
1291                         ret = -EINVAL;
1292                         fput(src_file);
1293                         goto out;
1294                 }
1295                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1296                                      BTRFS_I(src_inode)->root,
1297                                      transid, readonly);
1298                 fput(src_file);
1299         }
1300 out:
1301         return ret;
1302 }
1303
1304 static noinline int btrfs_ioctl_snap_create(struct file *file,
1305                                             void __user *arg, int subvol)
1306 {
1307         struct btrfs_ioctl_vol_args *vol_args;
1308         int ret;
1309
1310         vol_args = memdup_user(arg, sizeof(*vol_args));
1311         if (IS_ERR(vol_args))
1312                 return PTR_ERR(vol_args);
1313         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1314
1315         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1316                                               vol_args->fd, subvol,
1317                                               NULL, false);
1318
1319         kfree(vol_args);
1320         return ret;
1321 }
1322
1323 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1324                                                void __user *arg, int subvol)
1325 {
1326         struct btrfs_ioctl_vol_args_v2 *vol_args;
1327         int ret;
1328         u64 transid = 0;
1329         u64 *ptr = NULL;
1330         bool readonly = false;
1331
1332         vol_args = memdup_user(arg, sizeof(*vol_args));
1333         if (IS_ERR(vol_args))
1334                 return PTR_ERR(vol_args);
1335         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1336
1337         if (vol_args->flags &
1338             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1339                 ret = -EOPNOTSUPP;
1340                 goto out;
1341         }
1342
1343         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1344                 ptr = &transid;
1345         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1346                 readonly = true;
1347
1348         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1349                                               vol_args->fd, subvol,
1350                                               ptr, readonly);
1351
1352         if (ret == 0 && ptr &&
1353             copy_to_user(arg +
1354                          offsetof(struct btrfs_ioctl_vol_args_v2,
1355                                   transid), ptr, sizeof(*ptr)))
1356                 ret = -EFAULT;
1357 out:
1358         kfree(vol_args);
1359         return ret;
1360 }
1361
1362 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1363                                                 void __user *arg)
1364 {
1365         struct inode *inode = fdentry(file)->d_inode;
1366         struct btrfs_root *root = BTRFS_I(inode)->root;
1367         int ret = 0;
1368         u64 flags = 0;
1369
1370         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1371                 return -EINVAL;
1372
1373         down_read(&root->fs_info->subvol_sem);
1374         if (btrfs_root_readonly(root))
1375                 flags |= BTRFS_SUBVOL_RDONLY;
1376         up_read(&root->fs_info->subvol_sem);
1377
1378         if (copy_to_user(arg, &flags, sizeof(flags)))
1379                 ret = -EFAULT;
1380
1381         return ret;
1382 }
1383
1384 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1385                                               void __user *arg)
1386 {
1387         struct inode *inode = fdentry(file)->d_inode;
1388         struct btrfs_root *root = BTRFS_I(inode)->root;
1389         struct btrfs_trans_handle *trans;
1390         u64 root_flags;
1391         u64 flags;
1392         int ret = 0;
1393
1394         if (root->fs_info->sb->s_flags & MS_RDONLY)
1395                 return -EROFS;
1396
1397         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1398                 return -EINVAL;
1399
1400         if (copy_from_user(&flags, arg, sizeof(flags)))
1401                 return -EFAULT;
1402
1403         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1404                 return -EINVAL;
1405
1406         if (flags & ~BTRFS_SUBVOL_RDONLY)
1407                 return -EOPNOTSUPP;
1408
1409         if (!inode_owner_or_capable(inode))
1410                 return -EACCES;
1411
1412         down_write(&root->fs_info->subvol_sem);
1413
1414         /* nothing to do */
1415         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1416                 goto out;
1417
1418         root_flags = btrfs_root_flags(&root->root_item);
1419         if (flags & BTRFS_SUBVOL_RDONLY)
1420                 btrfs_set_root_flags(&root->root_item,
1421                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1422         else
1423                 btrfs_set_root_flags(&root->root_item,
1424                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1425
1426         trans = btrfs_start_transaction(root, 1);
1427         if (IS_ERR(trans)) {
1428                 ret = PTR_ERR(trans);
1429                 goto out_reset;
1430         }
1431
1432         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1433                                 &root->root_key, &root->root_item);
1434
1435         btrfs_commit_transaction(trans, root);
1436 out_reset:
1437         if (ret)
1438                 btrfs_set_root_flags(&root->root_item, root_flags);
1439 out:
1440         up_write(&root->fs_info->subvol_sem);
1441         return ret;
1442 }
1443
1444 /*
1445  * helper to check if the subvolume references other subvolumes
1446  */
1447 static noinline int may_destroy_subvol(struct btrfs_root *root)
1448 {
1449         struct btrfs_path *path;
1450         struct btrfs_key key;
1451         int ret;
1452
1453         path = btrfs_alloc_path();
1454         if (!path)
1455                 return -ENOMEM;
1456
1457         key.objectid = root->root_key.objectid;
1458         key.type = BTRFS_ROOT_REF_KEY;
1459         key.offset = (u64)-1;
1460
1461         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1462                                 &key, path, 0, 0);
1463         if (ret < 0)
1464                 goto out;
1465         BUG_ON(ret == 0);
1466
1467         ret = 0;
1468         if (path->slots[0] > 0) {
1469                 path->slots[0]--;
1470                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1471                 if (key.objectid == root->root_key.objectid &&
1472                     key.type == BTRFS_ROOT_REF_KEY)
1473                         ret = -ENOTEMPTY;
1474         }
1475 out:
1476         btrfs_free_path(path);
1477         return ret;
1478 }
1479
1480 static noinline int key_in_sk(struct btrfs_key *key,
1481                               struct btrfs_ioctl_search_key *sk)
1482 {
1483         struct btrfs_key test;
1484         int ret;
1485
1486         test.objectid = sk->min_objectid;
1487         test.type = sk->min_type;
1488         test.offset = sk->min_offset;
1489
1490         ret = btrfs_comp_cpu_keys(key, &test);
1491         if (ret < 0)
1492                 return 0;
1493
1494         test.objectid = sk->max_objectid;
1495         test.type = sk->max_type;
1496         test.offset = sk->max_offset;
1497
1498         ret = btrfs_comp_cpu_keys(key, &test);
1499         if (ret > 0)
1500                 return 0;
1501         return 1;
1502 }
1503
1504 static noinline int copy_to_sk(struct btrfs_root *root,
1505                                struct btrfs_path *path,
1506                                struct btrfs_key *key,
1507                                struct btrfs_ioctl_search_key *sk,
1508                                char *buf,
1509                                unsigned long *sk_offset,
1510                                int *num_found)
1511 {
1512         u64 found_transid;
1513         struct extent_buffer *leaf;
1514         struct btrfs_ioctl_search_header sh;
1515         unsigned long item_off;
1516         unsigned long item_len;
1517         int nritems;
1518         int i;
1519         int slot;
1520         int ret = 0;
1521
1522         leaf = path->nodes[0];
1523         slot = path->slots[0];
1524         nritems = btrfs_header_nritems(leaf);
1525
1526         if (btrfs_header_generation(leaf) > sk->max_transid) {
1527                 i = nritems;
1528                 goto advance_key;
1529         }
1530         found_transid = btrfs_header_generation(leaf);
1531
1532         for (i = slot; i < nritems; i++) {
1533                 item_off = btrfs_item_ptr_offset(leaf, i);
1534                 item_len = btrfs_item_size_nr(leaf, i);
1535
1536                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1537                         item_len = 0;
1538
1539                 if (sizeof(sh) + item_len + *sk_offset >
1540                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1541                         ret = 1;
1542                         goto overflow;
1543                 }
1544
1545                 btrfs_item_key_to_cpu(leaf, key, i);
1546                 if (!key_in_sk(key, sk))
1547                         continue;
1548
1549                 sh.objectid = key->objectid;
1550                 sh.offset = key->offset;
1551                 sh.type = key->type;
1552                 sh.len = item_len;
1553                 sh.transid = found_transid;
1554
1555                 /* copy search result header */
1556                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1557                 *sk_offset += sizeof(sh);
1558
1559                 if (item_len) {
1560                         char *p = buf + *sk_offset;
1561                         /* copy the item */
1562                         read_extent_buffer(leaf, p,
1563                                            item_off, item_len);
1564                         *sk_offset += item_len;
1565                 }
1566                 (*num_found)++;
1567
1568                 if (*num_found >= sk->nr_items)
1569                         break;
1570         }
1571 advance_key:
1572         ret = 0;
1573         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1574                 key->offset++;
1575         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1576                 key->offset = 0;
1577                 key->type++;
1578         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1579                 key->offset = 0;
1580                 key->type = 0;
1581                 key->objectid++;
1582         } else
1583                 ret = 1;
1584 overflow:
1585         return ret;
1586 }
1587
1588 static noinline int search_ioctl(struct inode *inode,
1589                                  struct btrfs_ioctl_search_args *args)
1590 {
1591         struct btrfs_root *root;
1592         struct btrfs_key key;
1593         struct btrfs_key max_key;
1594         struct btrfs_path *path;
1595         struct btrfs_ioctl_search_key *sk = &args->key;
1596         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1597         int ret;
1598         int num_found = 0;
1599         unsigned long sk_offset = 0;
1600
1601         path = btrfs_alloc_path();
1602         if (!path)
1603                 return -ENOMEM;
1604
1605         if (sk->tree_id == 0) {
1606                 /* search the root of the inode that was passed */
1607                 root = BTRFS_I(inode)->root;
1608         } else {
1609                 key.objectid = sk->tree_id;
1610                 key.type = BTRFS_ROOT_ITEM_KEY;
1611                 key.offset = (u64)-1;
1612                 root = btrfs_read_fs_root_no_name(info, &key);
1613                 if (IS_ERR(root)) {
1614                         printk(KERN_ERR "could not find root %llu\n",
1615                                sk->tree_id);
1616                         btrfs_free_path(path);
1617                         return -ENOENT;
1618                 }
1619         }
1620
1621         key.objectid = sk->min_objectid;
1622         key.type = sk->min_type;
1623         key.offset = sk->min_offset;
1624
1625         max_key.objectid = sk->max_objectid;
1626         max_key.type = sk->max_type;
1627         max_key.offset = sk->max_offset;
1628
1629         path->keep_locks = 1;
1630
1631         while(1) {
1632                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1633                                            sk->min_transid);
1634                 if (ret != 0) {
1635                         if (ret > 0)
1636                                 ret = 0;
1637                         goto err;
1638                 }
1639                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1640                                  &sk_offset, &num_found);
1641                 btrfs_release_path(path);
1642                 if (ret || num_found >= sk->nr_items)
1643                         break;
1644
1645         }
1646         ret = 0;
1647 err:
1648         sk->nr_items = num_found;
1649         btrfs_free_path(path);
1650         return ret;
1651 }
1652
1653 static noinline int btrfs_ioctl_tree_search(struct file *file,
1654                                            void __user *argp)
1655 {
1656          struct btrfs_ioctl_search_args *args;
1657          struct inode *inode;
1658          int ret;
1659
1660         if (!capable(CAP_SYS_ADMIN))
1661                 return -EPERM;
1662
1663         args = memdup_user(argp, sizeof(*args));
1664         if (IS_ERR(args))
1665                 return PTR_ERR(args);
1666
1667         inode = fdentry(file)->d_inode;
1668         ret = search_ioctl(inode, args);
1669         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1670                 ret = -EFAULT;
1671         kfree(args);
1672         return ret;
1673 }
1674
1675 /*
1676  * Search INODE_REFs to identify path name of 'dirid' directory
1677  * in a 'tree_id' tree. and sets path name to 'name'.
1678  */
1679 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1680                                 u64 tree_id, u64 dirid, char *name)
1681 {
1682         struct btrfs_root *root;
1683         struct btrfs_key key;
1684         char *ptr;
1685         int ret = -1;
1686         int slot;
1687         int len;
1688         int total_len = 0;
1689         struct btrfs_inode_ref *iref;
1690         struct extent_buffer *l;
1691         struct btrfs_path *path;
1692
1693         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1694                 name[0]='\0';
1695                 return 0;
1696         }
1697
1698         path = btrfs_alloc_path();
1699         if (!path)
1700                 return -ENOMEM;
1701
1702         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1703
1704         key.objectid = tree_id;
1705         key.type = BTRFS_ROOT_ITEM_KEY;
1706         key.offset = (u64)-1;
1707         root = btrfs_read_fs_root_no_name(info, &key);
1708         if (IS_ERR(root)) {
1709                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1710                 ret = -ENOENT;
1711                 goto out;
1712         }
1713
1714         key.objectid = dirid;
1715         key.type = BTRFS_INODE_REF_KEY;
1716         key.offset = (u64)-1;
1717
1718         while(1) {
1719                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1720                 if (ret < 0)
1721                         goto out;
1722
1723                 l = path->nodes[0];
1724                 slot = path->slots[0];
1725                 if (ret > 0 && slot > 0)
1726                         slot--;
1727                 btrfs_item_key_to_cpu(l, &key, slot);
1728
1729                 if (ret > 0 && (key.objectid != dirid ||
1730                                 key.type != BTRFS_INODE_REF_KEY)) {
1731                         ret = -ENOENT;
1732                         goto out;
1733                 }
1734
1735                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1736                 len = btrfs_inode_ref_name_len(l, iref);
1737                 ptr -= len + 1;
1738                 total_len += len + 1;
1739                 if (ptr < name)
1740                         goto out;
1741
1742                 *(ptr + len) = '/';
1743                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1744
1745                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1746                         break;
1747
1748                 btrfs_release_path(path);
1749                 key.objectid = key.offset;
1750                 key.offset = (u64)-1;
1751                 dirid = key.objectid;
1752         }
1753         if (ptr < name)
1754                 goto out;
1755         memmove(name, ptr, total_len);
1756         name[total_len]='\0';
1757         ret = 0;
1758 out:
1759         btrfs_free_path(path);
1760         return ret;
1761 }
1762
1763 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1764                                            void __user *argp)
1765 {
1766          struct btrfs_ioctl_ino_lookup_args *args;
1767          struct inode *inode;
1768          int ret;
1769
1770         if (!capable(CAP_SYS_ADMIN))
1771                 return -EPERM;
1772
1773         args = memdup_user(argp, sizeof(*args));
1774         if (IS_ERR(args))
1775                 return PTR_ERR(args);
1776
1777         inode = fdentry(file)->d_inode;
1778
1779         if (args->treeid == 0)
1780                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1781
1782         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1783                                         args->treeid, args->objectid,
1784                                         args->name);
1785
1786         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1787                 ret = -EFAULT;
1788
1789         kfree(args);
1790         return ret;
1791 }
1792
1793 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1794                                              void __user *arg)
1795 {
1796         struct dentry *parent = fdentry(file);
1797         struct dentry *dentry;
1798         struct inode *dir = parent->d_inode;
1799         struct inode *inode;
1800         struct btrfs_root *root = BTRFS_I(dir)->root;
1801         struct btrfs_root *dest = NULL;
1802         struct btrfs_ioctl_vol_args *vol_args;
1803         struct btrfs_trans_handle *trans;
1804         int namelen;
1805         int ret;
1806         int err = 0;
1807
1808         vol_args = memdup_user(arg, sizeof(*vol_args));
1809         if (IS_ERR(vol_args))
1810                 return PTR_ERR(vol_args);
1811
1812         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1813         namelen = strlen(vol_args->name);
1814         if (strchr(vol_args->name, '/') ||
1815             strncmp(vol_args->name, "..", namelen) == 0) {
1816                 err = -EINVAL;
1817                 goto out;
1818         }
1819
1820         err = mnt_want_write(file->f_path.mnt);
1821         if (err)
1822                 goto out;
1823
1824         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1825         dentry = lookup_one_len(vol_args->name, parent, namelen);
1826         if (IS_ERR(dentry)) {
1827                 err = PTR_ERR(dentry);
1828                 goto out_unlock_dir;
1829         }
1830
1831         if (!dentry->d_inode) {
1832                 err = -ENOENT;
1833                 goto out_dput;
1834         }
1835
1836         inode = dentry->d_inode;
1837         dest = BTRFS_I(inode)->root;
1838         if (!capable(CAP_SYS_ADMIN)){
1839                 /*
1840                  * Regular user.  Only allow this with a special mount
1841                  * option, when the user has write+exec access to the
1842                  * subvol root, and when rmdir(2) would have been
1843                  * allowed.
1844                  *
1845                  * Note that this is _not_ check that the subvol is
1846                  * empty or doesn't contain data that we wouldn't
1847                  * otherwise be able to delete.
1848                  *
1849                  * Users who want to delete empty subvols should try
1850                  * rmdir(2).
1851                  */
1852                 err = -EPERM;
1853                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1854                         goto out_dput;
1855
1856                 /*
1857                  * Do not allow deletion if the parent dir is the same
1858                  * as the dir to be deleted.  That means the ioctl
1859                  * must be called on the dentry referencing the root
1860                  * of the subvol, not a random directory contained
1861                  * within it.
1862                  */
1863                 err = -EINVAL;
1864                 if (root == dest)
1865                         goto out_dput;
1866
1867                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1868                 if (err)
1869                         goto out_dput;
1870
1871                 /* check if subvolume may be deleted by a non-root user */
1872                 err = btrfs_may_delete(dir, dentry, 1);
1873                 if (err)
1874                         goto out_dput;
1875         }
1876
1877         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1878                 err = -EINVAL;
1879                 goto out_dput;
1880         }
1881
1882         mutex_lock(&inode->i_mutex);
1883         err = d_invalidate(dentry);
1884         if (err)
1885                 goto out_unlock;
1886
1887         down_write(&root->fs_info->subvol_sem);
1888
1889         err = may_destroy_subvol(dest);
1890         if (err)
1891                 goto out_up_write;
1892
1893         trans = btrfs_start_transaction(root, 0);
1894         if (IS_ERR(trans)) {
1895                 err = PTR_ERR(trans);
1896                 goto out_up_write;
1897         }
1898         trans->block_rsv = &root->fs_info->global_block_rsv;
1899
1900         ret = btrfs_unlink_subvol(trans, root, dir,
1901                                 dest->root_key.objectid,
1902                                 dentry->d_name.name,
1903                                 dentry->d_name.len);
1904         BUG_ON(ret);
1905
1906         btrfs_record_root_in_trans(trans, dest);
1907
1908         memset(&dest->root_item.drop_progress, 0,
1909                 sizeof(dest->root_item.drop_progress));
1910         dest->root_item.drop_level = 0;
1911         btrfs_set_root_refs(&dest->root_item, 0);
1912
1913         if (!xchg(&dest->orphan_item_inserted, 1)) {
1914                 ret = btrfs_insert_orphan_item(trans,
1915                                         root->fs_info->tree_root,
1916                                         dest->root_key.objectid);
1917                 BUG_ON(ret);
1918         }
1919
1920         ret = btrfs_end_transaction(trans, root);
1921         BUG_ON(ret);
1922         inode->i_flags |= S_DEAD;
1923 out_up_write:
1924         up_write(&root->fs_info->subvol_sem);
1925 out_unlock:
1926         mutex_unlock(&inode->i_mutex);
1927         if (!err) {
1928                 shrink_dcache_sb(root->fs_info->sb);
1929                 btrfs_invalidate_inodes(dest);
1930                 d_delete(dentry);
1931         }
1932 out_dput:
1933         dput(dentry);
1934 out_unlock_dir:
1935         mutex_unlock(&dir->i_mutex);
1936         mnt_drop_write(file->f_path.mnt);
1937 out:
1938         kfree(vol_args);
1939         return err;
1940 }
1941
1942 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1943 {
1944         struct inode *inode = fdentry(file)->d_inode;
1945         struct btrfs_root *root = BTRFS_I(inode)->root;
1946         struct btrfs_ioctl_defrag_range_args *range;
1947         int ret;
1948
1949         if (btrfs_root_readonly(root))
1950                 return -EROFS;
1951
1952         ret = mnt_want_write(file->f_path.mnt);
1953         if (ret)
1954                 return ret;
1955
1956         switch (inode->i_mode & S_IFMT) {
1957         case S_IFDIR:
1958                 if (!capable(CAP_SYS_ADMIN)) {
1959                         ret = -EPERM;
1960                         goto out;
1961                 }
1962                 ret = btrfs_defrag_root(root, 0);
1963                 if (ret)
1964                         goto out;
1965                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1966                 break;
1967         case S_IFREG:
1968                 if (!(file->f_mode & FMODE_WRITE)) {
1969                         ret = -EINVAL;
1970                         goto out;
1971                 }
1972
1973                 range = kzalloc(sizeof(*range), GFP_KERNEL);
1974                 if (!range) {
1975                         ret = -ENOMEM;
1976                         goto out;
1977                 }
1978
1979                 if (argp) {
1980                         if (copy_from_user(range, argp,
1981                                            sizeof(*range))) {
1982                                 ret = -EFAULT;
1983                                 kfree(range);
1984                                 goto out;
1985                         }
1986                         /* compression requires us to start the IO */
1987                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1988                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1989                                 range->extent_thresh = (u32)-1;
1990                         }
1991                 } else {
1992                         /* the rest are all set to zero by kzalloc */
1993                         range->len = (u64)-1;
1994                 }
1995                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
1996                                         range, 0, 0);
1997                 if (ret > 0)
1998                         ret = 0;
1999                 kfree(range);
2000                 break;
2001         default:
2002                 ret = -EINVAL;
2003         }
2004 out:
2005         mnt_drop_write(file->f_path.mnt);
2006         return ret;
2007 }
2008
2009 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2010 {
2011         struct btrfs_ioctl_vol_args *vol_args;
2012         int ret;
2013
2014         if (!capable(CAP_SYS_ADMIN))
2015                 return -EPERM;
2016
2017         vol_args = memdup_user(arg, sizeof(*vol_args));
2018         if (IS_ERR(vol_args))
2019                 return PTR_ERR(vol_args);
2020
2021         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2022         ret = btrfs_init_new_device(root, vol_args->name);
2023
2024         kfree(vol_args);
2025         return ret;
2026 }
2027
2028 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2029 {
2030         struct btrfs_ioctl_vol_args *vol_args;
2031         int ret;
2032
2033         if (!capable(CAP_SYS_ADMIN))
2034                 return -EPERM;
2035
2036         if (root->fs_info->sb->s_flags & MS_RDONLY)
2037                 return -EROFS;
2038
2039         vol_args = memdup_user(arg, sizeof(*vol_args));
2040         if (IS_ERR(vol_args))
2041                 return PTR_ERR(vol_args);
2042
2043         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2044         ret = btrfs_rm_device(root, vol_args->name);
2045
2046         kfree(vol_args);
2047         return ret;
2048 }
2049
2050 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2051 {
2052         struct btrfs_ioctl_fs_info_args *fi_args;
2053         struct btrfs_device *device;
2054         struct btrfs_device *next;
2055         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2056         int ret = 0;
2057
2058         if (!capable(CAP_SYS_ADMIN))
2059                 return -EPERM;
2060
2061         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2062         if (!fi_args)
2063                 return -ENOMEM;
2064
2065         fi_args->num_devices = fs_devices->num_devices;
2066         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2067
2068         mutex_lock(&fs_devices->device_list_mutex);
2069         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2070                 if (device->devid > fi_args->max_id)
2071                         fi_args->max_id = device->devid;
2072         }
2073         mutex_unlock(&fs_devices->device_list_mutex);
2074
2075         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2076                 ret = -EFAULT;
2077
2078         kfree(fi_args);
2079         return ret;
2080 }
2081
2082 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2083 {
2084         struct btrfs_ioctl_dev_info_args *di_args;
2085         struct btrfs_device *dev;
2086         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2087         int ret = 0;
2088         char *s_uuid = NULL;
2089         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2090
2091         if (!capable(CAP_SYS_ADMIN))
2092                 return -EPERM;
2093
2094         di_args = memdup_user(arg, sizeof(*di_args));
2095         if (IS_ERR(di_args))
2096                 return PTR_ERR(di_args);
2097
2098         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2099                 s_uuid = di_args->uuid;
2100
2101         mutex_lock(&fs_devices->device_list_mutex);
2102         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2103         mutex_unlock(&fs_devices->device_list_mutex);
2104
2105         if (!dev) {
2106                 ret = -ENODEV;
2107                 goto out;
2108         }
2109
2110         di_args->devid = dev->devid;
2111         di_args->bytes_used = dev->bytes_used;
2112         di_args->total_bytes = dev->total_bytes;
2113         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2114         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2115
2116 out:
2117         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2118                 ret = -EFAULT;
2119
2120         kfree(di_args);
2121         return ret;
2122 }
2123
2124 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2125                                        u64 off, u64 olen, u64 destoff)
2126 {
2127         struct inode *inode = fdentry(file)->d_inode;
2128         struct btrfs_root *root = BTRFS_I(inode)->root;
2129         struct file *src_file;
2130         struct inode *src;
2131         struct btrfs_trans_handle *trans;
2132         struct btrfs_path *path;
2133         struct extent_buffer *leaf;
2134         char *buf;
2135         struct btrfs_key key;
2136         u32 nritems;
2137         int slot;
2138         int ret;
2139         u64 len = olen;
2140         u64 bs = root->fs_info->sb->s_blocksize;
2141         u64 hint_byte;
2142
2143         /*
2144          * TODO:
2145          * - split compressed inline extents.  annoying: we need to
2146          *   decompress into destination's address_space (the file offset
2147          *   may change, so source mapping won't do), then recompress (or
2148          *   otherwise reinsert) a subrange.
2149          * - allow ranges within the same file to be cloned (provided
2150          *   they don't overlap)?
2151          */
2152
2153         /* the destination must be opened for writing */
2154         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2155                 return -EINVAL;
2156
2157         if (btrfs_root_readonly(root))
2158                 return -EROFS;
2159
2160         ret = mnt_want_write(file->f_path.mnt);
2161         if (ret)
2162                 return ret;
2163
2164         src_file = fget(srcfd);
2165         if (!src_file) {
2166                 ret = -EBADF;
2167                 goto out_drop_write;
2168         }
2169
2170         src = src_file->f_dentry->d_inode;
2171
2172         ret = -EINVAL;
2173         if (src == inode)
2174                 goto out_fput;
2175
2176         /* the src must be open for reading */
2177         if (!(src_file->f_mode & FMODE_READ))
2178                 goto out_fput;
2179
2180         ret = -EISDIR;
2181         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2182                 goto out_fput;
2183
2184         ret = -EXDEV;
2185         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2186                 goto out_fput;
2187
2188         ret = -ENOMEM;
2189         buf = vmalloc(btrfs_level_size(root, 0));
2190         if (!buf)
2191                 goto out_fput;
2192
2193         path = btrfs_alloc_path();
2194         if (!path) {
2195                 vfree(buf);
2196                 goto out_fput;
2197         }
2198         path->reada = 2;
2199
2200         if (inode < src) {
2201                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2202                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2203         } else {
2204                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2205                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2206         }
2207
2208         /* determine range to clone */
2209         ret = -EINVAL;
2210         if (off + len > src->i_size || off + len < off)
2211                 goto out_unlock;
2212         if (len == 0)
2213                 olen = len = src->i_size - off;
2214         /* if we extend to eof, continue to block boundary */
2215         if (off + len == src->i_size)
2216                 len = ALIGN(src->i_size, bs) - off;
2217
2218         /* verify the end result is block aligned */
2219         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2220             !IS_ALIGNED(destoff, bs))
2221                 goto out_unlock;
2222
2223         /* do any pending delalloc/csum calc on src, one way or
2224            another, and lock file content */
2225         while (1) {
2226                 struct btrfs_ordered_extent *ordered;
2227                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2228                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2229                 if (!ordered &&
2230                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2231                                    EXTENT_DELALLOC, 0, NULL))
2232                         break;
2233                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2234                 if (ordered)
2235                         btrfs_put_ordered_extent(ordered);
2236                 btrfs_wait_ordered_range(src, off, len);
2237         }
2238
2239         /* truncate page cache pages from target inode range */
2240         truncate_inode_pages_range(&inode->i_data, off,
2241                                    ALIGN(off + len, PAGE_CACHE_SIZE) - 1);
2242
2243         /* clone data */
2244         key.objectid = btrfs_ino(src);
2245         key.type = BTRFS_EXTENT_DATA_KEY;
2246         key.offset = 0;
2247
2248         while (1) {
2249                 /*
2250                  * note the key will change type as we walk through the
2251                  * tree.
2252                  */
2253                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2254                 if (ret < 0)
2255                         goto out;
2256
2257                 nritems = btrfs_header_nritems(path->nodes[0]);
2258                 if (path->slots[0] >= nritems) {
2259                         ret = btrfs_next_leaf(root, path);
2260                         if (ret < 0)
2261                                 goto out;
2262                         if (ret > 0)
2263                                 break;
2264                         nritems = btrfs_header_nritems(path->nodes[0]);
2265                 }
2266                 leaf = path->nodes[0];
2267                 slot = path->slots[0];
2268
2269                 btrfs_item_key_to_cpu(leaf, &key, slot);
2270                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2271                     key.objectid != btrfs_ino(src))
2272                         break;
2273
2274                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2275                         struct btrfs_file_extent_item *extent;
2276                         int type;
2277                         u32 size;
2278                         struct btrfs_key new_key;
2279                         u64 disko = 0, diskl = 0;
2280                         u64 datao = 0, datal = 0;
2281                         u8 comp;
2282                         u64 endoff;
2283
2284                         size = btrfs_item_size_nr(leaf, slot);
2285                         read_extent_buffer(leaf, buf,
2286                                            btrfs_item_ptr_offset(leaf, slot),
2287                                            size);
2288
2289                         extent = btrfs_item_ptr(leaf, slot,
2290                                                 struct btrfs_file_extent_item);
2291                         comp = btrfs_file_extent_compression(leaf, extent);
2292                         type = btrfs_file_extent_type(leaf, extent);
2293                         if (type == BTRFS_FILE_EXTENT_REG ||
2294                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2295                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2296                                                                       extent);
2297                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2298                                                                  extent);
2299                                 datao = btrfs_file_extent_offset(leaf, extent);
2300                                 datal = btrfs_file_extent_num_bytes(leaf,
2301                                                                     extent);
2302                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2303                                 /* take upper bound, may be compressed */
2304                                 datal = btrfs_file_extent_ram_bytes(leaf,
2305                                                                     extent);
2306                         }
2307                         btrfs_release_path(path);
2308
2309                         if (key.offset + datal <= off ||
2310                             key.offset >= off+len)
2311                                 goto next;
2312
2313                         memcpy(&new_key, &key, sizeof(new_key));
2314                         new_key.objectid = btrfs_ino(inode);
2315                         if (off <= key.offset)
2316                                 new_key.offset = key.offset + destoff - off;
2317                         else
2318                                 new_key.offset = destoff;
2319
2320                         trans = btrfs_start_transaction(root, 1);
2321                         if (IS_ERR(trans)) {
2322                                 ret = PTR_ERR(trans);
2323                                 goto out;
2324                         }
2325
2326                         if (type == BTRFS_FILE_EXTENT_REG ||
2327                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2328                                 if (off > key.offset) {
2329                                         datao += off - key.offset;
2330                                         datal -= off - key.offset;
2331                                 }
2332
2333                                 if (key.offset + datal > off + len)
2334                                         datal = off + len - key.offset;
2335
2336                                 ret = btrfs_drop_extents(trans, inode,
2337                                                          new_key.offset,
2338                                                          new_key.offset + datal,
2339                                                          &hint_byte, 1);
2340                                 BUG_ON(ret);
2341
2342                                 ret = btrfs_insert_empty_item(trans, root, path,
2343                                                               &new_key, size);
2344                                 BUG_ON(ret);
2345
2346                                 leaf = path->nodes[0];
2347                                 slot = path->slots[0];
2348                                 write_extent_buffer(leaf, buf,
2349                                             btrfs_item_ptr_offset(leaf, slot),
2350                                             size);
2351
2352                                 extent = btrfs_item_ptr(leaf, slot,
2353                                                 struct btrfs_file_extent_item);
2354
2355                                 /* disko == 0 means it's a hole */
2356                                 if (!disko)
2357                                         datao = 0;
2358
2359                                 btrfs_set_file_extent_offset(leaf, extent,
2360                                                              datao);
2361                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2362                                                                 datal);
2363                                 if (disko) {
2364                                         inode_add_bytes(inode, datal);
2365                                         ret = btrfs_inc_extent_ref(trans, root,
2366                                                         disko, diskl, 0,
2367                                                         root->root_key.objectid,
2368                                                         btrfs_ino(inode),
2369                                                         new_key.offset - datao);
2370                                         BUG_ON(ret);
2371                                 }
2372                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2373                                 u64 skip = 0;
2374                                 u64 trim = 0;
2375                                 if (off > key.offset) {
2376                                         skip = off - key.offset;
2377                                         new_key.offset += skip;
2378                                 }
2379
2380                                 if (key.offset + datal > off+len)
2381                                         trim = key.offset + datal - (off+len);
2382
2383                                 if (comp && (skip || trim)) {
2384                                         ret = -EINVAL;
2385                                         btrfs_end_transaction(trans, root);
2386                                         goto out;
2387                                 }
2388                                 size -= skip + trim;
2389                                 datal -= skip + trim;
2390
2391                                 ret = btrfs_drop_extents(trans, inode,
2392                                                          new_key.offset,
2393                                                          new_key.offset + datal,
2394                                                          &hint_byte, 1);
2395                                 BUG_ON(ret);
2396
2397                                 ret = btrfs_insert_empty_item(trans, root, path,
2398                                                               &new_key, size);
2399                                 BUG_ON(ret);
2400
2401                                 if (skip) {
2402                                         u32 start =
2403                                           btrfs_file_extent_calc_inline_size(0);
2404                                         memmove(buf+start, buf+start+skip,
2405                                                 datal);
2406                                 }
2407
2408                                 leaf = path->nodes[0];
2409                                 slot = path->slots[0];
2410                                 write_extent_buffer(leaf, buf,
2411                                             btrfs_item_ptr_offset(leaf, slot),
2412                                             size);
2413                                 inode_add_bytes(inode, datal);
2414                         }
2415
2416                         btrfs_mark_buffer_dirty(leaf);
2417                         btrfs_release_path(path);
2418
2419                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2420
2421                         /*
2422                          * we round up to the block size at eof when
2423                          * determining which extents to clone above,
2424                          * but shouldn't round up the file size
2425                          */
2426                         endoff = new_key.offset + datal;
2427                         if (endoff > destoff+olen)
2428                                 endoff = destoff+olen;
2429                         if (endoff > inode->i_size)
2430                                 btrfs_i_size_write(inode, endoff);
2431
2432                         BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
2433                         ret = btrfs_update_inode(trans, root, inode);
2434                         BUG_ON(ret);
2435                         btrfs_end_transaction(trans, root);
2436                 }
2437 next:
2438                 btrfs_release_path(path);
2439                 key.offset++;
2440         }
2441         ret = 0;
2442 out:
2443         btrfs_release_path(path);
2444         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2445 out_unlock:
2446         mutex_unlock(&src->i_mutex);
2447         mutex_unlock(&inode->i_mutex);
2448         vfree(buf);
2449         btrfs_free_path(path);
2450 out_fput:
2451         fput(src_file);
2452 out_drop_write:
2453         mnt_drop_write(file->f_path.mnt);
2454         return ret;
2455 }
2456
2457 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2458 {
2459         struct btrfs_ioctl_clone_range_args args;
2460
2461         if (copy_from_user(&args, argp, sizeof(args)))
2462                 return -EFAULT;
2463         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2464                                  args.src_length, args.dest_offset);
2465 }
2466
2467 /*
2468  * there are many ways the trans_start and trans_end ioctls can lead
2469  * to deadlocks.  They should only be used by applications that
2470  * basically own the machine, and have a very in depth understanding
2471  * of all the possible deadlocks and enospc problems.
2472  */
2473 static long btrfs_ioctl_trans_start(struct file *file)
2474 {
2475         struct inode *inode = fdentry(file)->d_inode;
2476         struct btrfs_root *root = BTRFS_I(inode)->root;
2477         struct btrfs_trans_handle *trans;
2478         int ret;
2479
2480         ret = -EPERM;
2481         if (!capable(CAP_SYS_ADMIN))
2482                 goto out;
2483
2484         ret = -EINPROGRESS;
2485         if (file->private_data)
2486                 goto out;
2487
2488         ret = -EROFS;
2489         if (btrfs_root_readonly(root))
2490                 goto out;
2491
2492         ret = mnt_want_write(file->f_path.mnt);
2493         if (ret)
2494                 goto out;
2495
2496         atomic_inc(&root->fs_info->open_ioctl_trans);
2497
2498         ret = -ENOMEM;
2499         trans = btrfs_start_ioctl_transaction(root);
2500         if (IS_ERR(trans))
2501                 goto out_drop;
2502
2503         file->private_data = trans;
2504         return 0;
2505
2506 out_drop:
2507         atomic_dec(&root->fs_info->open_ioctl_trans);
2508         mnt_drop_write(file->f_path.mnt);
2509 out:
2510         return ret;
2511 }
2512
2513 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2514 {
2515         struct inode *inode = fdentry(file)->d_inode;
2516         struct btrfs_root *root = BTRFS_I(inode)->root;
2517         struct btrfs_root *new_root;
2518         struct btrfs_dir_item *di;
2519         struct btrfs_trans_handle *trans;
2520         struct btrfs_path *path;
2521         struct btrfs_key location;
2522         struct btrfs_disk_key disk_key;
2523         struct btrfs_super_block *disk_super;
2524         u64 features;
2525         u64 objectid = 0;
2526         u64 dir_id;
2527
2528         if (!capable(CAP_SYS_ADMIN))
2529                 return -EPERM;
2530
2531         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2532                 return -EFAULT;
2533
2534         if (!objectid)
2535                 objectid = root->root_key.objectid;
2536
2537         location.objectid = objectid;
2538         location.type = BTRFS_ROOT_ITEM_KEY;
2539         location.offset = (u64)-1;
2540
2541         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2542         if (IS_ERR(new_root))
2543                 return PTR_ERR(new_root);
2544
2545         if (btrfs_root_refs(&new_root->root_item) == 0)
2546                 return -ENOENT;
2547
2548         path = btrfs_alloc_path();
2549         if (!path)
2550                 return -ENOMEM;
2551         path->leave_spinning = 1;
2552
2553         trans = btrfs_start_transaction(root, 1);
2554         if (IS_ERR(trans)) {
2555                 btrfs_free_path(path);
2556                 return PTR_ERR(trans);
2557         }
2558
2559         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
2560         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2561                                    dir_id, "default", 7, 1);
2562         if (IS_ERR_OR_NULL(di)) {
2563                 btrfs_free_path(path);
2564                 btrfs_end_transaction(trans, root);
2565                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2566                        "this isn't going to work\n");
2567                 return -ENOENT;
2568         }
2569
2570         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2571         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2572         btrfs_mark_buffer_dirty(path->nodes[0]);
2573         btrfs_free_path(path);
2574
2575         disk_super = &root->fs_info->super_copy;
2576         features = btrfs_super_incompat_flags(disk_super);
2577         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2578                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2579                 btrfs_set_super_incompat_flags(disk_super, features);
2580         }
2581         btrfs_end_transaction(trans, root);
2582
2583         return 0;
2584 }
2585
2586 static void get_block_group_info(struct list_head *groups_list,
2587                                  struct btrfs_ioctl_space_info *space)
2588 {
2589         struct btrfs_block_group_cache *block_group;
2590
2591         space->total_bytes = 0;
2592         space->used_bytes = 0;
2593         space->flags = 0;
2594         list_for_each_entry(block_group, groups_list, list) {
2595                 space->flags = block_group->flags;
2596                 space->total_bytes += block_group->key.offset;
2597                 space->used_bytes +=
2598                         btrfs_block_group_used(&block_group->item);
2599         }
2600 }
2601
2602 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2603 {
2604         struct btrfs_ioctl_space_args space_args;
2605         struct btrfs_ioctl_space_info space;
2606         struct btrfs_ioctl_space_info *dest;
2607         struct btrfs_ioctl_space_info *dest_orig;
2608         struct btrfs_ioctl_space_info __user *user_dest;
2609         struct btrfs_space_info *info;
2610         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2611                        BTRFS_BLOCK_GROUP_SYSTEM,
2612                        BTRFS_BLOCK_GROUP_METADATA,
2613                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2614         int num_types = 4;
2615         int alloc_size;
2616         int ret = 0;
2617         u64 slot_count = 0;
2618         int i, c;
2619
2620         if (copy_from_user(&space_args,
2621                            (struct btrfs_ioctl_space_args __user *)arg,
2622                            sizeof(space_args)))
2623                 return -EFAULT;
2624
2625         for (i = 0; i < num_types; i++) {
2626                 struct btrfs_space_info *tmp;
2627
2628                 info = NULL;
2629                 rcu_read_lock();
2630                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2631                                         list) {
2632                         if (tmp->flags == types[i]) {
2633                                 info = tmp;
2634                                 break;
2635                         }
2636                 }
2637                 rcu_read_unlock();
2638
2639                 if (!info)
2640                         continue;
2641
2642                 down_read(&info->groups_sem);
2643                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2644                         if (!list_empty(&info->block_groups[c]))
2645                                 slot_count++;
2646                 }
2647                 up_read(&info->groups_sem);
2648         }
2649
2650         /* space_slots == 0 means they are asking for a count */
2651         if (space_args.space_slots == 0) {
2652                 space_args.total_spaces = slot_count;
2653                 goto out;
2654         }
2655
2656         slot_count = min_t(u64, space_args.space_slots, slot_count);
2657
2658         alloc_size = sizeof(*dest) * slot_count;
2659
2660         /* we generally have at most 6 or so space infos, one for each raid
2661          * level.  So, a whole page should be more than enough for everyone
2662          */
2663         if (alloc_size > PAGE_CACHE_SIZE)
2664                 return -ENOMEM;
2665
2666         space_args.total_spaces = 0;
2667         dest = kmalloc(alloc_size, GFP_NOFS);
2668         if (!dest)
2669                 return -ENOMEM;
2670         dest_orig = dest;
2671
2672         /* now we have a buffer to copy into */
2673         for (i = 0; i < num_types; i++) {
2674                 struct btrfs_space_info *tmp;
2675
2676                 if (!slot_count)
2677                         break;
2678
2679                 info = NULL;
2680                 rcu_read_lock();
2681                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2682                                         list) {
2683                         if (tmp->flags == types[i]) {
2684                                 info = tmp;
2685                                 break;
2686                         }
2687                 }
2688                 rcu_read_unlock();
2689
2690                 if (!info)
2691                         continue;
2692                 down_read(&info->groups_sem);
2693                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2694                         if (!list_empty(&info->block_groups[c])) {
2695                                 get_block_group_info(&info->block_groups[c],
2696                                                      &space);
2697                                 memcpy(dest, &space, sizeof(space));
2698                                 dest++;
2699                                 space_args.total_spaces++;
2700                                 slot_count--;
2701                         }
2702                         if (!slot_count)
2703                                 break;
2704                 }
2705                 up_read(&info->groups_sem);
2706         }
2707
2708         user_dest = (struct btrfs_ioctl_space_info *)
2709                 (arg + sizeof(struct btrfs_ioctl_space_args));
2710
2711         if (copy_to_user(user_dest, dest_orig, alloc_size))
2712                 ret = -EFAULT;
2713
2714         kfree(dest_orig);
2715 out:
2716         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2717                 ret = -EFAULT;
2718
2719         return ret;
2720 }
2721
2722 /*
2723  * there are many ways the trans_start and trans_end ioctls can lead
2724  * to deadlocks.  They should only be used by applications that
2725  * basically own the machine, and have a very in depth understanding
2726  * of all the possible deadlocks and enospc problems.
2727  */
2728 long btrfs_ioctl_trans_end(struct file *file)
2729 {
2730         struct inode *inode = fdentry(file)->d_inode;
2731         struct btrfs_root *root = BTRFS_I(inode)->root;
2732         struct btrfs_trans_handle *trans;
2733
2734         trans = file->private_data;
2735         if (!trans)
2736                 return -EINVAL;
2737         file->private_data = NULL;
2738
2739         btrfs_end_transaction(trans, root);
2740
2741         atomic_dec(&root->fs_info->open_ioctl_trans);
2742
2743         mnt_drop_write(file->f_path.mnt);
2744         return 0;
2745 }
2746
2747 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2748 {
2749         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2750         struct btrfs_trans_handle *trans;
2751         u64 transid;
2752         int ret;
2753
2754         trans = btrfs_start_transaction(root, 0);
2755         if (IS_ERR(trans))
2756                 return PTR_ERR(trans);
2757         transid = trans->transid;
2758         ret = btrfs_commit_transaction_async(trans, root, 0);
2759         if (ret) {
2760                 btrfs_end_transaction(trans, root);
2761                 return ret;
2762         }
2763
2764         if (argp)
2765                 if (copy_to_user(argp, &transid, sizeof(transid)))
2766                         return -EFAULT;
2767         return 0;
2768 }
2769
2770 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2771 {
2772         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2773         u64 transid;
2774
2775         if (argp) {
2776                 if (copy_from_user(&transid, argp, sizeof(transid)))
2777                         return -EFAULT;
2778         } else {
2779                 transid = 0;  /* current trans */
2780         }
2781         return btrfs_wait_for_commit(root, transid);
2782 }
2783
2784 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2785 {
2786         int ret;
2787         struct btrfs_ioctl_scrub_args *sa;
2788
2789         if (!capable(CAP_SYS_ADMIN))
2790                 return -EPERM;
2791
2792         sa = memdup_user(arg, sizeof(*sa));
2793         if (IS_ERR(sa))
2794                 return PTR_ERR(sa);
2795
2796         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2797                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2798
2799         if (copy_to_user(arg, sa, sizeof(*sa)))
2800                 ret = -EFAULT;
2801
2802         kfree(sa);
2803         return ret;
2804 }
2805
2806 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2807 {
2808         if (!capable(CAP_SYS_ADMIN))
2809                 return -EPERM;
2810
2811         return btrfs_scrub_cancel(root);
2812 }
2813
2814 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2815                                        void __user *arg)
2816 {
2817         struct btrfs_ioctl_scrub_args *sa;
2818         int ret;
2819
2820         if (!capable(CAP_SYS_ADMIN))
2821                 return -EPERM;
2822
2823         sa = memdup_user(arg, sizeof(*sa));
2824         if (IS_ERR(sa))
2825                 return PTR_ERR(sa);
2826
2827         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2828
2829         if (copy_to_user(arg, sa, sizeof(*sa)))
2830                 ret = -EFAULT;
2831
2832         kfree(sa);
2833         return ret;
2834 }
2835
2836 long btrfs_ioctl(struct file *file, unsigned int
2837                 cmd, unsigned long arg)
2838 {
2839         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2840         void __user *argp = (void __user *)arg;
2841
2842         switch (cmd) {
2843         case FS_IOC_GETFLAGS:
2844                 return btrfs_ioctl_getflags(file, argp);
2845         case FS_IOC_SETFLAGS:
2846                 return btrfs_ioctl_setflags(file, argp);
2847         case FS_IOC_GETVERSION:
2848                 return btrfs_ioctl_getversion(file, argp);
2849         case FITRIM:
2850                 return btrfs_ioctl_fitrim(file, argp);
2851         case BTRFS_IOC_SNAP_CREATE:
2852                 return btrfs_ioctl_snap_create(file, argp, 0);
2853         case BTRFS_IOC_SNAP_CREATE_V2:
2854                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
2855         case BTRFS_IOC_SUBVOL_CREATE:
2856                 return btrfs_ioctl_snap_create(file, argp, 1);
2857         case BTRFS_IOC_SNAP_DESTROY:
2858                 return btrfs_ioctl_snap_destroy(file, argp);
2859         case BTRFS_IOC_SUBVOL_GETFLAGS:
2860                 return btrfs_ioctl_subvol_getflags(file, argp);
2861         case BTRFS_IOC_SUBVOL_SETFLAGS:
2862                 return btrfs_ioctl_subvol_setflags(file, argp);
2863         case BTRFS_IOC_DEFAULT_SUBVOL:
2864                 return btrfs_ioctl_default_subvol(file, argp);
2865         case BTRFS_IOC_DEFRAG:
2866                 return btrfs_ioctl_defrag(file, NULL);
2867         case BTRFS_IOC_DEFRAG_RANGE:
2868                 return btrfs_ioctl_defrag(file, argp);
2869         case BTRFS_IOC_RESIZE:
2870                 return btrfs_ioctl_resize(root, argp);
2871         case BTRFS_IOC_ADD_DEV:
2872                 return btrfs_ioctl_add_dev(root, argp);
2873         case BTRFS_IOC_RM_DEV:
2874                 return btrfs_ioctl_rm_dev(root, argp);
2875         case BTRFS_IOC_FS_INFO:
2876                 return btrfs_ioctl_fs_info(root, argp);
2877         case BTRFS_IOC_DEV_INFO:
2878                 return btrfs_ioctl_dev_info(root, argp);
2879         case BTRFS_IOC_BALANCE:
2880                 return btrfs_balance(root->fs_info->dev_root);
2881         case BTRFS_IOC_CLONE:
2882                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
2883         case BTRFS_IOC_CLONE_RANGE:
2884                 return btrfs_ioctl_clone_range(file, argp);
2885         case BTRFS_IOC_TRANS_START:
2886                 return btrfs_ioctl_trans_start(file);
2887         case BTRFS_IOC_TRANS_END:
2888                 return btrfs_ioctl_trans_end(file);
2889         case BTRFS_IOC_TREE_SEARCH:
2890                 return btrfs_ioctl_tree_search(file, argp);
2891         case BTRFS_IOC_INO_LOOKUP:
2892                 return btrfs_ioctl_ino_lookup(file, argp);
2893         case BTRFS_IOC_SPACE_INFO:
2894                 return btrfs_ioctl_space_info(root, argp);
2895         case BTRFS_IOC_SYNC:
2896                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
2897                 return 0;
2898         case BTRFS_IOC_START_SYNC:
2899                 return btrfs_ioctl_start_sync(file, argp);
2900         case BTRFS_IOC_WAIT_SYNC:
2901                 return btrfs_ioctl_wait_sync(file, argp);
2902         case BTRFS_IOC_SCRUB:
2903                 return btrfs_ioctl_scrub(root, argp);
2904         case BTRFS_IOC_SCRUB_CANCEL:
2905                 return btrfs_ioctl_scrub_cancel(root, argp);
2906         case BTRFS_IOC_SCRUB_PROGRESS:
2907                 return btrfs_ioctl_scrub_progress(root, argp);
2908         }
2909
2910         return -ENOTTY;
2911 }