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