]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/btrfs/ioctl.c
Btrfs: add qgroup inheritance
[~shefty/rdma-dev.git] / fs / btrfs / ioctl.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include "compat.h"
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "ioctl.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54 #include "backref.h"
55 #include "rcu-string.h"
56
57 /* Mask out flags that are inappropriate for the given type of inode. */
58 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
59 {
60         if (S_ISDIR(mode))
61                 return flags;
62         else if (S_ISREG(mode))
63                 return flags & ~FS_DIRSYNC_FL;
64         else
65                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
66 }
67
68 /*
69  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
70  */
71 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
72 {
73         unsigned int iflags = 0;
74
75         if (flags & BTRFS_INODE_SYNC)
76                 iflags |= FS_SYNC_FL;
77         if (flags & BTRFS_INODE_IMMUTABLE)
78                 iflags |= FS_IMMUTABLE_FL;
79         if (flags & BTRFS_INODE_APPEND)
80                 iflags |= FS_APPEND_FL;
81         if (flags & BTRFS_INODE_NODUMP)
82                 iflags |= FS_NODUMP_FL;
83         if (flags & BTRFS_INODE_NOATIME)
84                 iflags |= FS_NOATIME_FL;
85         if (flags & BTRFS_INODE_DIRSYNC)
86                 iflags |= FS_DIRSYNC_FL;
87         if (flags & BTRFS_INODE_NODATACOW)
88                 iflags |= FS_NOCOW_FL;
89
90         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
91                 iflags |= FS_COMPR_FL;
92         else if (flags & BTRFS_INODE_NOCOMPRESS)
93                 iflags |= FS_NOCOMP_FL;
94
95         return iflags;
96 }
97
98 /*
99  * Update inode->i_flags based on the btrfs internal flags.
100  */
101 void btrfs_update_iflags(struct inode *inode)
102 {
103         struct btrfs_inode *ip = BTRFS_I(inode);
104
105         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
106
107         if (ip->flags & BTRFS_INODE_SYNC)
108                 inode->i_flags |= S_SYNC;
109         if (ip->flags & BTRFS_INODE_IMMUTABLE)
110                 inode->i_flags |= S_IMMUTABLE;
111         if (ip->flags & BTRFS_INODE_APPEND)
112                 inode->i_flags |= S_APPEND;
113         if (ip->flags & BTRFS_INODE_NOATIME)
114                 inode->i_flags |= S_NOATIME;
115         if (ip->flags & BTRFS_INODE_DIRSYNC)
116                 inode->i_flags |= S_DIRSYNC;
117 }
118
119 /*
120  * Inherit flags from the parent inode.
121  *
122  * Currently only the compression flags and the cow flags are inherited.
123  */
124 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
125 {
126         unsigned int flags;
127
128         if (!dir)
129                 return;
130
131         flags = BTRFS_I(dir)->flags;
132
133         if (flags & BTRFS_INODE_NOCOMPRESS) {
134                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
135                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
136         } else if (flags & BTRFS_INODE_COMPRESS) {
137                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
138                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
139         }
140
141         if (flags & BTRFS_INODE_NODATACOW)
142                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
143
144         btrfs_update_iflags(inode);
145 }
146
147 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
148 {
149         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
150         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
151
152         if (copy_to_user(arg, &flags, sizeof(flags)))
153                 return -EFAULT;
154         return 0;
155 }
156
157 static int check_flags(unsigned int flags)
158 {
159         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
160                       FS_NOATIME_FL | FS_NODUMP_FL | \
161                       FS_SYNC_FL | FS_DIRSYNC_FL | \
162                       FS_NOCOMP_FL | FS_COMPR_FL |
163                       FS_NOCOW_FL))
164                 return -EOPNOTSUPP;
165
166         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
167                 return -EINVAL;
168
169         return 0;
170 }
171
172 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
173 {
174         struct inode *inode = file->f_path.dentry->d_inode;
175         struct btrfs_inode *ip = BTRFS_I(inode);
176         struct btrfs_root *root = ip->root;
177         struct btrfs_trans_handle *trans;
178         unsigned int flags, oldflags;
179         int ret;
180         u64 ip_oldflags;
181         unsigned int i_oldflags;
182
183         if (btrfs_root_readonly(root))
184                 return -EROFS;
185
186         if (copy_from_user(&flags, arg, sizeof(flags)))
187                 return -EFAULT;
188
189         ret = check_flags(flags);
190         if (ret)
191                 return ret;
192
193         if (!inode_owner_or_capable(inode))
194                 return -EACCES;
195
196         mutex_lock(&inode->i_mutex);
197
198         ip_oldflags = ip->flags;
199         i_oldflags = inode->i_flags;
200
201         flags = btrfs_mask_flags(inode->i_mode, flags);
202         oldflags = btrfs_flags_to_ioctl(ip->flags);
203         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
204                 if (!capable(CAP_LINUX_IMMUTABLE)) {
205                         ret = -EPERM;
206                         goto out_unlock;
207                 }
208         }
209
210         ret = mnt_want_write_file(file);
211         if (ret)
212                 goto out_unlock;
213
214         if (flags & FS_SYNC_FL)
215                 ip->flags |= BTRFS_INODE_SYNC;
216         else
217                 ip->flags &= ~BTRFS_INODE_SYNC;
218         if (flags & FS_IMMUTABLE_FL)
219                 ip->flags |= BTRFS_INODE_IMMUTABLE;
220         else
221                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
222         if (flags & FS_APPEND_FL)
223                 ip->flags |= BTRFS_INODE_APPEND;
224         else
225                 ip->flags &= ~BTRFS_INODE_APPEND;
226         if (flags & FS_NODUMP_FL)
227                 ip->flags |= BTRFS_INODE_NODUMP;
228         else
229                 ip->flags &= ~BTRFS_INODE_NODUMP;
230         if (flags & FS_NOATIME_FL)
231                 ip->flags |= BTRFS_INODE_NOATIME;
232         else
233                 ip->flags &= ~BTRFS_INODE_NOATIME;
234         if (flags & FS_DIRSYNC_FL)
235                 ip->flags |= BTRFS_INODE_DIRSYNC;
236         else
237                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
238         if (flags & FS_NOCOW_FL)
239                 ip->flags |= BTRFS_INODE_NODATACOW;
240         else
241                 ip->flags &= ~BTRFS_INODE_NODATACOW;
242
243         /*
244          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
245          * flag may be changed automatically if compression code won't make
246          * things smaller.
247          */
248         if (flags & FS_NOCOMP_FL) {
249                 ip->flags &= ~BTRFS_INODE_COMPRESS;
250                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
251         } else if (flags & FS_COMPR_FL) {
252                 ip->flags |= BTRFS_INODE_COMPRESS;
253                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
254         } else {
255                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
256         }
257
258         trans = btrfs_start_transaction(root, 1);
259         if (IS_ERR(trans)) {
260                 ret = PTR_ERR(trans);
261                 goto out_drop;
262         }
263
264         btrfs_update_iflags(inode);
265         inode_inc_iversion(inode);
266         inode->i_ctime = CURRENT_TIME;
267         ret = btrfs_update_inode(trans, root, inode);
268
269         btrfs_end_transaction(trans, root);
270  out_drop:
271         if (ret) {
272                 ip->flags = ip_oldflags;
273                 inode->i_flags = i_oldflags;
274         }
275
276         mnt_drop_write_file(file);
277  out_unlock:
278         mutex_unlock(&inode->i_mutex);
279         return ret;
280 }
281
282 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
283 {
284         struct inode *inode = file->f_path.dentry->d_inode;
285
286         return put_user(inode->i_generation, arg);
287 }
288
289 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
290 {
291         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
292         struct btrfs_device *device;
293         struct request_queue *q;
294         struct fstrim_range range;
295         u64 minlen = ULLONG_MAX;
296         u64 num_devices = 0;
297         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
298         int ret;
299
300         if (!capable(CAP_SYS_ADMIN))
301                 return -EPERM;
302
303         rcu_read_lock();
304         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
305                                 dev_list) {
306                 if (!device->bdev)
307                         continue;
308                 q = bdev_get_queue(device->bdev);
309                 if (blk_queue_discard(q)) {
310                         num_devices++;
311                         minlen = min((u64)q->limits.discard_granularity,
312                                      minlen);
313                 }
314         }
315         rcu_read_unlock();
316
317         if (!num_devices)
318                 return -EOPNOTSUPP;
319         if (copy_from_user(&range, arg, sizeof(range)))
320                 return -EFAULT;
321         if (range.start > total_bytes)
322                 return -EINVAL;
323
324         range.len = min(range.len, total_bytes - range.start);
325         range.minlen = max(range.minlen, minlen);
326         ret = btrfs_trim_fs(fs_info->tree_root, &range);
327         if (ret < 0)
328                 return ret;
329
330         if (copy_to_user(arg, &range, sizeof(range)))
331                 return -EFAULT;
332
333         return 0;
334 }
335
336 static noinline int create_subvol(struct btrfs_root *root,
337                                   struct dentry *dentry,
338                                   char *name, int namelen,
339                                   u64 *async_transid,
340                                   struct btrfs_qgroup_inherit **inherit)
341 {
342         struct btrfs_trans_handle *trans;
343         struct btrfs_key key;
344         struct btrfs_root_item root_item;
345         struct btrfs_inode_item *inode_item;
346         struct extent_buffer *leaf;
347         struct btrfs_root *new_root;
348         struct dentry *parent = dentry->d_parent;
349         struct inode *dir;
350         int ret;
351         int err;
352         u64 objectid;
353         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
354         u64 index = 0;
355
356         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
357         if (ret)
358                 return ret;
359
360         dir = parent->d_inode;
361
362         /*
363          * 1 - inode item
364          * 2 - refs
365          * 1 - root item
366          * 2 - dir items
367          */
368         trans = btrfs_start_transaction(root, 6);
369         if (IS_ERR(trans))
370                 return PTR_ERR(trans);
371
372         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid,
373                                    inherit ? *inherit : NULL);
374         if (ret)
375                 goto fail;
376
377         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
378                                       0, objectid, NULL, 0, 0, 0);
379         if (IS_ERR(leaf)) {
380                 ret = PTR_ERR(leaf);
381                 goto fail;
382         }
383
384         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
385         btrfs_set_header_bytenr(leaf, leaf->start);
386         btrfs_set_header_generation(leaf, trans->transid);
387         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
388         btrfs_set_header_owner(leaf, objectid);
389
390         write_extent_buffer(leaf, root->fs_info->fsid,
391                             (unsigned long)btrfs_header_fsid(leaf),
392                             BTRFS_FSID_SIZE);
393         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
394                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
395                             BTRFS_UUID_SIZE);
396         btrfs_mark_buffer_dirty(leaf);
397
398         inode_item = &root_item.inode;
399         memset(inode_item, 0, sizeof(*inode_item));
400         inode_item->generation = cpu_to_le64(1);
401         inode_item->size = cpu_to_le64(3);
402         inode_item->nlink = cpu_to_le32(1);
403         inode_item->nbytes = cpu_to_le64(root->leafsize);
404         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
405
406         root_item.flags = 0;
407         root_item.byte_limit = 0;
408         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
409
410         btrfs_set_root_bytenr(&root_item, leaf->start);
411         btrfs_set_root_generation(&root_item, trans->transid);
412         btrfs_set_root_level(&root_item, 0);
413         btrfs_set_root_refs(&root_item, 1);
414         btrfs_set_root_used(&root_item, leaf->len);
415         btrfs_set_root_last_snapshot(&root_item, 0);
416
417         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
418         root_item.drop_level = 0;
419
420         btrfs_tree_unlock(leaf);
421         free_extent_buffer(leaf);
422         leaf = NULL;
423
424         btrfs_set_root_dirid(&root_item, new_dirid);
425
426         key.objectid = objectid;
427         key.offset = 0;
428         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
429         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
430                                 &root_item);
431         if (ret)
432                 goto fail;
433
434         key.offset = (u64)-1;
435         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
436         if (IS_ERR(new_root)) {
437                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
438                 ret = PTR_ERR(new_root);
439                 goto fail;
440         }
441
442         btrfs_record_root_in_trans(trans, new_root);
443
444         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
445         if (ret) {
446                 /* We potentially lose an unused inode item here */
447                 btrfs_abort_transaction(trans, root, ret);
448                 goto fail;
449         }
450
451         /*
452          * insert the directory item
453          */
454         ret = btrfs_set_inode_index(dir, &index);
455         if (ret) {
456                 btrfs_abort_transaction(trans, root, ret);
457                 goto fail;
458         }
459
460         ret = btrfs_insert_dir_item(trans, root,
461                                     name, namelen, dir, &key,
462                                     BTRFS_FT_DIR, index);
463         if (ret) {
464                 btrfs_abort_transaction(trans, root, ret);
465                 goto fail;
466         }
467
468         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
469         ret = btrfs_update_inode(trans, root, dir);
470         BUG_ON(ret);
471
472         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
473                                  objectid, root->root_key.objectid,
474                                  btrfs_ino(dir), index, name, namelen);
475
476         BUG_ON(ret);
477
478         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
479 fail:
480         if (async_transid) {
481                 *async_transid = trans->transid;
482                 err = btrfs_commit_transaction_async(trans, root, 1);
483         } else {
484                 err = btrfs_commit_transaction(trans, root);
485         }
486         if (err && !ret)
487                 ret = err;
488         return ret;
489 }
490
491 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
492                            char *name, int namelen, u64 *async_transid,
493                            bool readonly, struct btrfs_qgroup_inherit **inherit)
494 {
495         struct inode *inode;
496         struct btrfs_pending_snapshot *pending_snapshot;
497         struct btrfs_trans_handle *trans;
498         int ret;
499
500         if (!root->ref_cows)
501                 return -EINVAL;
502
503         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
504         if (!pending_snapshot)
505                 return -ENOMEM;
506
507         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
508         pending_snapshot->dentry = dentry;
509         pending_snapshot->root = root;
510         pending_snapshot->readonly = readonly;
511         if (inherit) {
512                 pending_snapshot->inherit = *inherit;
513                 *inherit = NULL;        /* take responsibility to free it */
514         }
515
516         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
517         if (IS_ERR(trans)) {
518                 ret = PTR_ERR(trans);
519                 goto fail;
520         }
521
522         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
523         BUG_ON(ret);
524
525         spin_lock(&root->fs_info->trans_lock);
526         list_add(&pending_snapshot->list,
527                  &trans->transaction->pending_snapshots);
528         spin_unlock(&root->fs_info->trans_lock);
529         if (async_transid) {
530                 *async_transid = trans->transid;
531                 ret = btrfs_commit_transaction_async(trans,
532                                      root->fs_info->extent_root, 1);
533         } else {
534                 ret = btrfs_commit_transaction(trans,
535                                                root->fs_info->extent_root);
536         }
537         BUG_ON(ret);
538
539         ret = pending_snapshot->error;
540         if (ret)
541                 goto fail;
542
543         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
544         if (ret)
545                 goto fail;
546
547         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
548         if (IS_ERR(inode)) {
549                 ret = PTR_ERR(inode);
550                 goto fail;
551         }
552         BUG_ON(!inode);
553         d_instantiate(dentry, inode);
554         ret = 0;
555 fail:
556         kfree(pending_snapshot);
557         return ret;
558 }
559
560 /*  copy of check_sticky in fs/namei.c()
561 * It's inline, so penalty for filesystems that don't use sticky bit is
562 * minimal.
563 */
564 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
565 {
566         uid_t fsuid = current_fsuid();
567
568         if (!(dir->i_mode & S_ISVTX))
569                 return 0;
570         if (inode->i_uid == fsuid)
571                 return 0;
572         if (dir->i_uid == fsuid)
573                 return 0;
574         return !capable(CAP_FOWNER);
575 }
576
577 /*  copy of may_delete in fs/namei.c()
578  *      Check whether we can remove a link victim from directory dir, check
579  *  whether the type of victim is right.
580  *  1. We can't do it if dir is read-only (done in permission())
581  *  2. We should have write and exec permissions on dir
582  *  3. We can't remove anything from append-only dir
583  *  4. We can't do anything with immutable dir (done in permission())
584  *  5. If the sticky bit on dir is set we should either
585  *      a. be owner of dir, or
586  *      b. be owner of victim, or
587  *      c. have CAP_FOWNER capability
588  *  6. If the victim is append-only or immutable we can't do antyhing with
589  *     links pointing to it.
590  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
591  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
592  *  9. We can't remove a root or mountpoint.
593  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
594  *     nfs_async_unlink().
595  */
596
597 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
598 {
599         int error;
600
601         if (!victim->d_inode)
602                 return -ENOENT;
603
604         BUG_ON(victim->d_parent->d_inode != dir);
605         audit_inode_child(victim, dir);
606
607         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
608         if (error)
609                 return error;
610         if (IS_APPEND(dir))
611                 return -EPERM;
612         if (btrfs_check_sticky(dir, victim->d_inode)||
613                 IS_APPEND(victim->d_inode)||
614             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
615                 return -EPERM;
616         if (isdir) {
617                 if (!S_ISDIR(victim->d_inode->i_mode))
618                         return -ENOTDIR;
619                 if (IS_ROOT(victim))
620                         return -EBUSY;
621         } else if (S_ISDIR(victim->d_inode->i_mode))
622                 return -EISDIR;
623         if (IS_DEADDIR(dir))
624                 return -ENOENT;
625         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
626                 return -EBUSY;
627         return 0;
628 }
629
630 /* copy of may_create in fs/namei.c() */
631 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
632 {
633         if (child->d_inode)
634                 return -EEXIST;
635         if (IS_DEADDIR(dir))
636                 return -ENOENT;
637         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
638 }
639
640 /*
641  * Create a new subvolume below @parent.  This is largely modeled after
642  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
643  * inside this filesystem so it's quite a bit simpler.
644  */
645 static noinline int btrfs_mksubvol(struct path *parent,
646                                    char *name, int namelen,
647                                    struct btrfs_root *snap_src,
648                                    u64 *async_transid, bool readonly,
649                                    struct btrfs_qgroup_inherit **inherit)
650 {
651         struct inode *dir  = parent->dentry->d_inode;
652         struct dentry *dentry;
653         int error;
654
655         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
656
657         dentry = lookup_one_len(name, parent->dentry, namelen);
658         error = PTR_ERR(dentry);
659         if (IS_ERR(dentry))
660                 goto out_unlock;
661
662         error = -EEXIST;
663         if (dentry->d_inode)
664                 goto out_dput;
665
666         error = mnt_want_write(parent->mnt);
667         if (error)
668                 goto out_dput;
669
670         error = btrfs_may_create(dir, dentry);
671         if (error)
672                 goto out_drop_write;
673
674         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
675
676         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
677                 goto out_up_read;
678
679         if (snap_src) {
680                 error = create_snapshot(snap_src, dentry, name, namelen,
681                                         async_transid, readonly, inherit);
682         } else {
683                 error = create_subvol(BTRFS_I(dir)->root, dentry,
684                                       name, namelen, async_transid, inherit);
685         }
686         if (!error)
687                 fsnotify_mkdir(dir, dentry);
688 out_up_read:
689         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
690 out_drop_write:
691         mnt_drop_write(parent->mnt);
692 out_dput:
693         dput(dentry);
694 out_unlock:
695         mutex_unlock(&dir->i_mutex);
696         return error;
697 }
698
699 /*
700  * When we're defragging a range, we don't want to kick it off again
701  * if it is really just waiting for delalloc to send it down.
702  * If we find a nice big extent or delalloc range for the bytes in the
703  * file you want to defrag, we return 0 to let you know to skip this
704  * part of the file
705  */
706 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
707 {
708         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
709         struct extent_map *em = NULL;
710         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
711         u64 end;
712
713         read_lock(&em_tree->lock);
714         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
715         read_unlock(&em_tree->lock);
716
717         if (em) {
718                 end = extent_map_end(em);
719                 free_extent_map(em);
720                 if (end - offset > thresh)
721                         return 0;
722         }
723         /* if we already have a nice delalloc here, just stop */
724         thresh /= 2;
725         end = count_range_bits(io_tree, &offset, offset + thresh,
726                                thresh, EXTENT_DELALLOC, 1);
727         if (end >= thresh)
728                 return 0;
729         return 1;
730 }
731
732 /*
733  * helper function to walk through a file and find extents
734  * newer than a specific transid, and smaller than thresh.
735  *
736  * This is used by the defragging code to find new and small
737  * extents
738  */
739 static int find_new_extents(struct btrfs_root *root,
740                             struct inode *inode, u64 newer_than,
741                             u64 *off, int thresh)
742 {
743         struct btrfs_path *path;
744         struct btrfs_key min_key;
745         struct btrfs_key max_key;
746         struct extent_buffer *leaf;
747         struct btrfs_file_extent_item *extent;
748         int type;
749         int ret;
750         u64 ino = btrfs_ino(inode);
751
752         path = btrfs_alloc_path();
753         if (!path)
754                 return -ENOMEM;
755
756         min_key.objectid = ino;
757         min_key.type = BTRFS_EXTENT_DATA_KEY;
758         min_key.offset = *off;
759
760         max_key.objectid = ino;
761         max_key.type = (u8)-1;
762         max_key.offset = (u64)-1;
763
764         path->keep_locks = 1;
765
766         while(1) {
767                 ret = btrfs_search_forward(root, &min_key, &max_key,
768                                            path, 0, newer_than);
769                 if (ret != 0)
770                         goto none;
771                 if (min_key.objectid != ino)
772                         goto none;
773                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
774                         goto none;
775
776                 leaf = path->nodes[0];
777                 extent = btrfs_item_ptr(leaf, path->slots[0],
778                                         struct btrfs_file_extent_item);
779
780                 type = btrfs_file_extent_type(leaf, extent);
781                 if (type == BTRFS_FILE_EXTENT_REG &&
782                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
783                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
784                         *off = min_key.offset;
785                         btrfs_free_path(path);
786                         return 0;
787                 }
788
789                 if (min_key.offset == (u64)-1)
790                         goto none;
791
792                 min_key.offset++;
793                 btrfs_release_path(path);
794         }
795 none:
796         btrfs_free_path(path);
797         return -ENOENT;
798 }
799
800 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
801 {
802         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
803         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
804         struct extent_map *em;
805         u64 len = PAGE_CACHE_SIZE;
806
807         /*
808          * hopefully we have this extent in the tree already, try without
809          * the full extent lock
810          */
811         read_lock(&em_tree->lock);
812         em = lookup_extent_mapping(em_tree, start, len);
813         read_unlock(&em_tree->lock);
814
815         if (!em) {
816                 /* get the big lock and read metadata off disk */
817                 lock_extent(io_tree, start, start + len - 1);
818                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
819                 unlock_extent(io_tree, start, start + len - 1);
820
821                 if (IS_ERR(em))
822                         return NULL;
823         }
824
825         return em;
826 }
827
828 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
829 {
830         struct extent_map *next;
831         bool ret = true;
832
833         /* this is the last extent */
834         if (em->start + em->len >= i_size_read(inode))
835                 return false;
836
837         next = defrag_lookup_extent(inode, em->start + em->len);
838         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
839                 ret = false;
840
841         free_extent_map(next);
842         return ret;
843 }
844
845 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
846                                u64 *last_len, u64 *skip, u64 *defrag_end)
847 {
848         struct extent_map *em;
849         int ret = 1;
850         bool next_mergeable = true;
851
852         /*
853          * make sure that once we start defragging an extent, we keep on
854          * defragging it
855          */
856         if (start < *defrag_end)
857                 return 1;
858
859         *skip = 0;
860
861         em = defrag_lookup_extent(inode, start);
862         if (!em)
863                 return 0;
864
865         /* this will cover holes, and inline extents */
866         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
867                 ret = 0;
868                 goto out;
869         }
870
871         next_mergeable = defrag_check_next_extent(inode, em);
872
873         /*
874          * we hit a real extent, if it is big or the next extent is not a
875          * real extent, don't bother defragging it
876          */
877         if ((*last_len == 0 || *last_len >= thresh) &&
878             (em->len >= thresh || !next_mergeable))
879                 ret = 0;
880 out:
881         /*
882          * last_len ends up being a counter of how many bytes we've defragged.
883          * every time we choose not to defrag an extent, we reset *last_len
884          * so that the next tiny extent will force a defrag.
885          *
886          * The end result of this is that tiny extents before a single big
887          * extent will force at least part of that big extent to be defragged.
888          */
889         if (ret) {
890                 *defrag_end = extent_map_end(em);
891         } else {
892                 *last_len = 0;
893                 *skip = extent_map_end(em);
894                 *defrag_end = 0;
895         }
896
897         free_extent_map(em);
898         return ret;
899 }
900
901 /*
902  * it doesn't do much good to defrag one or two pages
903  * at a time.  This pulls in a nice chunk of pages
904  * to COW and defrag.
905  *
906  * It also makes sure the delalloc code has enough
907  * dirty data to avoid making new small extents as part
908  * of the defrag
909  *
910  * It's a good idea to start RA on this range
911  * before calling this.
912  */
913 static int cluster_pages_for_defrag(struct inode *inode,
914                                     struct page **pages,
915                                     unsigned long start_index,
916                                     int num_pages)
917 {
918         unsigned long file_end;
919         u64 isize = i_size_read(inode);
920         u64 page_start;
921         u64 page_end;
922         u64 page_cnt;
923         int ret;
924         int i;
925         int i_done;
926         struct btrfs_ordered_extent *ordered;
927         struct extent_state *cached_state = NULL;
928         struct extent_io_tree *tree;
929         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
930
931         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
932         if (!isize || start_index > file_end)
933                 return 0;
934
935         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
936
937         ret = btrfs_delalloc_reserve_space(inode,
938                                            page_cnt << PAGE_CACHE_SHIFT);
939         if (ret)
940                 return ret;
941         i_done = 0;
942         tree = &BTRFS_I(inode)->io_tree;
943
944         /* step one, lock all the pages */
945         for (i = 0; i < page_cnt; i++) {
946                 struct page *page;
947 again:
948                 page = find_or_create_page(inode->i_mapping,
949                                            start_index + i, mask);
950                 if (!page)
951                         break;
952
953                 page_start = page_offset(page);
954                 page_end = page_start + PAGE_CACHE_SIZE - 1;
955                 while (1) {
956                         lock_extent(tree, page_start, page_end);
957                         ordered = btrfs_lookup_ordered_extent(inode,
958                                                               page_start);
959                         unlock_extent(tree, page_start, page_end);
960                         if (!ordered)
961                                 break;
962
963                         unlock_page(page);
964                         btrfs_start_ordered_extent(inode, ordered, 1);
965                         btrfs_put_ordered_extent(ordered);
966                         lock_page(page);
967                         /*
968                          * we unlocked the page above, so we need check if
969                          * it was released or not.
970                          */
971                         if (page->mapping != inode->i_mapping) {
972                                 unlock_page(page);
973                                 page_cache_release(page);
974                                 goto again;
975                         }
976                 }
977
978                 if (!PageUptodate(page)) {
979                         btrfs_readpage(NULL, page);
980                         lock_page(page);
981                         if (!PageUptodate(page)) {
982                                 unlock_page(page);
983                                 page_cache_release(page);
984                                 ret = -EIO;
985                                 break;
986                         }
987                 }
988
989                 if (page->mapping != inode->i_mapping) {
990                         unlock_page(page);
991                         page_cache_release(page);
992                         goto again;
993                 }
994
995                 pages[i] = page;
996                 i_done++;
997         }
998         if (!i_done || ret)
999                 goto out;
1000
1001         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1002                 goto out;
1003
1004         /*
1005          * so now we have a nice long stream of locked
1006          * and up to date pages, lets wait on them
1007          */
1008         for (i = 0; i < i_done; i++)
1009                 wait_on_page_writeback(pages[i]);
1010
1011         page_start = page_offset(pages[0]);
1012         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1013
1014         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1015                          page_start, page_end - 1, 0, &cached_state);
1016         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1017                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1018                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
1019                           GFP_NOFS);
1020
1021         if (i_done != page_cnt) {
1022                 spin_lock(&BTRFS_I(inode)->lock);
1023                 BTRFS_I(inode)->outstanding_extents++;
1024                 spin_unlock(&BTRFS_I(inode)->lock);
1025                 btrfs_delalloc_release_space(inode,
1026                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1027         }
1028
1029
1030         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
1031                                   &cached_state);
1032
1033         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1034                              page_start, page_end - 1, &cached_state,
1035                              GFP_NOFS);
1036
1037         for (i = 0; i < i_done; i++) {
1038                 clear_page_dirty_for_io(pages[i]);
1039                 ClearPageChecked(pages[i]);
1040                 set_page_extent_mapped(pages[i]);
1041                 set_page_dirty(pages[i]);
1042                 unlock_page(pages[i]);
1043                 page_cache_release(pages[i]);
1044         }
1045         return i_done;
1046 out:
1047         for (i = 0; i < i_done; i++) {
1048                 unlock_page(pages[i]);
1049                 page_cache_release(pages[i]);
1050         }
1051         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1052         return ret;
1053
1054 }
1055
1056 int btrfs_defrag_file(struct inode *inode, struct file *file,
1057                       struct btrfs_ioctl_defrag_range_args *range,
1058                       u64 newer_than, unsigned long max_to_defrag)
1059 {
1060         struct btrfs_root *root = BTRFS_I(inode)->root;
1061         struct btrfs_super_block *disk_super;
1062         struct file_ra_state *ra = NULL;
1063         unsigned long last_index;
1064         u64 isize = i_size_read(inode);
1065         u64 features;
1066         u64 last_len = 0;
1067         u64 skip = 0;
1068         u64 defrag_end = 0;
1069         u64 newer_off = range->start;
1070         unsigned long i;
1071         unsigned long ra_index = 0;
1072         int ret;
1073         int defrag_count = 0;
1074         int compress_type = BTRFS_COMPRESS_ZLIB;
1075         int extent_thresh = range->extent_thresh;
1076         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1077         int cluster = max_cluster;
1078         u64 new_align = ~((u64)128 * 1024 - 1);
1079         struct page **pages = NULL;
1080
1081         if (extent_thresh == 0)
1082                 extent_thresh = 256 * 1024;
1083
1084         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1085                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1086                         return -EINVAL;
1087                 if (range->compress_type)
1088                         compress_type = range->compress_type;
1089         }
1090
1091         if (isize == 0)
1092                 return 0;
1093
1094         /*
1095          * if we were not given a file, allocate a readahead
1096          * context
1097          */
1098         if (!file) {
1099                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1100                 if (!ra)
1101                         return -ENOMEM;
1102                 file_ra_state_init(ra, inode->i_mapping);
1103         } else {
1104                 ra = &file->f_ra;
1105         }
1106
1107         pages = kmalloc(sizeof(struct page *) * max_cluster,
1108                         GFP_NOFS);
1109         if (!pages) {
1110                 ret = -ENOMEM;
1111                 goto out_ra;
1112         }
1113
1114         /* find the last page to defrag */
1115         if (range->start + range->len > range->start) {
1116                 last_index = min_t(u64, isize - 1,
1117                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1118         } else {
1119                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1120         }
1121
1122         if (newer_than) {
1123                 ret = find_new_extents(root, inode, newer_than,
1124                                        &newer_off, 64 * 1024);
1125                 if (!ret) {
1126                         range->start = newer_off;
1127                         /*
1128                          * we always align our defrag to help keep
1129                          * the extents in the file evenly spaced
1130                          */
1131                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1132                 } else
1133                         goto out_ra;
1134         } else {
1135                 i = range->start >> PAGE_CACHE_SHIFT;
1136         }
1137         if (!max_to_defrag)
1138                 max_to_defrag = last_index + 1;
1139
1140         /*
1141          * make writeback starts from i, so the defrag range can be
1142          * written sequentially.
1143          */
1144         if (i < inode->i_mapping->writeback_index)
1145                 inode->i_mapping->writeback_index = i;
1146
1147         while (i <= last_index && defrag_count < max_to_defrag &&
1148                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1149                 PAGE_CACHE_SHIFT)) {
1150                 /*
1151                  * make sure we stop running if someone unmounts
1152                  * the FS
1153                  */
1154                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1155                         break;
1156
1157                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1158                                          extent_thresh, &last_len, &skip,
1159                                          &defrag_end)) {
1160                         unsigned long next;
1161                         /*
1162                          * the should_defrag function tells us how much to skip
1163                          * bump our counter by the suggested amount
1164                          */
1165                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1166                         i = max(i + 1, next);
1167                         continue;
1168                 }
1169
1170                 if (!newer_than) {
1171                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1172                                    PAGE_CACHE_SHIFT) - i;
1173                         cluster = min(cluster, max_cluster);
1174                 } else {
1175                         cluster = max_cluster;
1176                 }
1177
1178                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1179                         BTRFS_I(inode)->force_compress = compress_type;
1180
1181                 if (i + cluster > ra_index) {
1182                         ra_index = max(i, ra_index);
1183                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1184                                        cluster);
1185                         ra_index += max_cluster;
1186                 }
1187
1188                 mutex_lock(&inode->i_mutex);
1189                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1190                 if (ret < 0) {
1191                         mutex_unlock(&inode->i_mutex);
1192                         goto out_ra;
1193                 }
1194
1195                 defrag_count += ret;
1196                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1197                 mutex_unlock(&inode->i_mutex);
1198
1199                 if (newer_than) {
1200                         if (newer_off == (u64)-1)
1201                                 break;
1202
1203                         if (ret > 0)
1204                                 i += ret;
1205
1206                         newer_off = max(newer_off + 1,
1207                                         (u64)i << PAGE_CACHE_SHIFT);
1208
1209                         ret = find_new_extents(root, inode,
1210                                                newer_than, &newer_off,
1211                                                64 * 1024);
1212                         if (!ret) {
1213                                 range->start = newer_off;
1214                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1215                         } else {
1216                                 break;
1217                         }
1218                 } else {
1219                         if (ret > 0) {
1220                                 i += ret;
1221                                 last_len += ret << PAGE_CACHE_SHIFT;
1222                         } else {
1223                                 i++;
1224                                 last_len = 0;
1225                         }
1226                 }
1227         }
1228
1229         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1230                 filemap_flush(inode->i_mapping);
1231
1232         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1233                 /* the filemap_flush will queue IO into the worker threads, but
1234                  * we have to make sure the IO is actually started and that
1235                  * ordered extents get created before we return
1236                  */
1237                 atomic_inc(&root->fs_info->async_submit_draining);
1238                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1239                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1240                         wait_event(root->fs_info->async_submit_wait,
1241                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1242                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1243                 }
1244                 atomic_dec(&root->fs_info->async_submit_draining);
1245
1246                 mutex_lock(&inode->i_mutex);
1247                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1248                 mutex_unlock(&inode->i_mutex);
1249         }
1250
1251         disk_super = root->fs_info->super_copy;
1252         features = btrfs_super_incompat_flags(disk_super);
1253         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1254                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1255                 btrfs_set_super_incompat_flags(disk_super, features);
1256         }
1257
1258         ret = defrag_count;
1259
1260 out_ra:
1261         if (!file)
1262                 kfree(ra);
1263         kfree(pages);
1264         return ret;
1265 }
1266
1267 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1268                                         void __user *arg)
1269 {
1270         u64 new_size;
1271         u64 old_size;
1272         u64 devid = 1;
1273         struct btrfs_ioctl_vol_args *vol_args;
1274         struct btrfs_trans_handle *trans;
1275         struct btrfs_device *device = NULL;
1276         char *sizestr;
1277         char *devstr = NULL;
1278         int ret = 0;
1279         int mod = 0;
1280
1281         if (root->fs_info->sb->s_flags & MS_RDONLY)
1282                 return -EROFS;
1283
1284         if (!capable(CAP_SYS_ADMIN))
1285                 return -EPERM;
1286
1287         mutex_lock(&root->fs_info->volume_mutex);
1288         if (root->fs_info->balance_ctl) {
1289                 printk(KERN_INFO "btrfs: balance in progress\n");
1290                 ret = -EINVAL;
1291                 goto out;
1292         }
1293
1294         vol_args = memdup_user(arg, sizeof(*vol_args));
1295         if (IS_ERR(vol_args)) {
1296                 ret = PTR_ERR(vol_args);
1297                 goto out;
1298         }
1299
1300         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1301
1302         sizestr = vol_args->name;
1303         devstr = strchr(sizestr, ':');
1304         if (devstr) {
1305                 char *end;
1306                 sizestr = devstr + 1;
1307                 *devstr = '\0';
1308                 devstr = vol_args->name;
1309                 devid = simple_strtoull(devstr, &end, 10);
1310                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1311                        (unsigned long long)devid);
1312         }
1313         device = btrfs_find_device(root, devid, NULL, NULL);
1314         if (!device) {
1315                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1316                        (unsigned long long)devid);
1317                 ret = -EINVAL;
1318                 goto out_free;
1319         }
1320         if (device->fs_devices && device->fs_devices->seeding) {
1321                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1322                        "seeding device %llu\n",
1323                        (unsigned long long)devid);
1324                 ret = -EINVAL;
1325                 goto out_free;
1326         }
1327
1328         if (!strcmp(sizestr, "max"))
1329                 new_size = device->bdev->bd_inode->i_size;
1330         else {
1331                 if (sizestr[0] == '-') {
1332                         mod = -1;
1333                         sizestr++;
1334                 } else if (sizestr[0] == '+') {
1335                         mod = 1;
1336                         sizestr++;
1337                 }
1338                 new_size = memparse(sizestr, NULL);
1339                 if (new_size == 0) {
1340                         ret = -EINVAL;
1341                         goto out_free;
1342                 }
1343         }
1344
1345         old_size = device->total_bytes;
1346
1347         if (mod < 0) {
1348                 if (new_size > old_size) {
1349                         ret = -EINVAL;
1350                         goto out_free;
1351                 }
1352                 new_size = old_size - new_size;
1353         } else if (mod > 0) {
1354                 new_size = old_size + new_size;
1355         }
1356
1357         if (new_size < 256 * 1024 * 1024) {
1358                 ret = -EINVAL;
1359                 goto out_free;
1360         }
1361         if (new_size > device->bdev->bd_inode->i_size) {
1362                 ret = -EFBIG;
1363                 goto out_free;
1364         }
1365
1366         do_div(new_size, root->sectorsize);
1367         new_size *= root->sectorsize;
1368
1369         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1370                       rcu_str_deref(device->name),
1371                       (unsigned long long)new_size);
1372
1373         if (new_size > old_size) {
1374                 trans = btrfs_start_transaction(root, 0);
1375                 if (IS_ERR(trans)) {
1376                         ret = PTR_ERR(trans);
1377                         goto out_free;
1378                 }
1379                 ret = btrfs_grow_device(trans, device, new_size);
1380                 btrfs_commit_transaction(trans, root);
1381         } else if (new_size < old_size) {
1382                 ret = btrfs_shrink_device(device, new_size);
1383         }
1384
1385 out_free:
1386         kfree(vol_args);
1387 out:
1388         mutex_unlock(&root->fs_info->volume_mutex);
1389         return ret;
1390 }
1391
1392 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1393                                 char *name, unsigned long fd, int subvol,
1394                                 u64 *transid, bool readonly,
1395                                 struct btrfs_qgroup_inherit **inherit)
1396 {
1397         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1398         struct file *src_file;
1399         int namelen;
1400         int ret = 0;
1401
1402         if (root->fs_info->sb->s_flags & MS_RDONLY)
1403                 return -EROFS;
1404
1405         namelen = strlen(name);
1406         if (strchr(name, '/')) {
1407                 ret = -EINVAL;
1408                 goto out;
1409         }
1410
1411         if (name[0] == '.' &&
1412            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1413                 ret = -EEXIST;
1414                 goto out;
1415         }
1416
1417         if (subvol) {
1418                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1419                                      NULL, transid, readonly, inherit);
1420         } else {
1421                 struct inode *src_inode;
1422                 src_file = fget(fd);
1423                 if (!src_file) {
1424                         ret = -EINVAL;
1425                         goto out;
1426                 }
1427
1428                 src_inode = src_file->f_path.dentry->d_inode;
1429                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1430                         printk(KERN_INFO "btrfs: Snapshot src from "
1431                                "another FS\n");
1432                         ret = -EINVAL;
1433                         fput(src_file);
1434                         goto out;
1435                 }
1436                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1437                                      BTRFS_I(src_inode)->root,
1438                                      transid, readonly, inherit);
1439                 fput(src_file);
1440         }
1441 out:
1442         return ret;
1443 }
1444
1445 static noinline int btrfs_ioctl_snap_create(struct file *file,
1446                                             void __user *arg, int subvol)
1447 {
1448         struct btrfs_ioctl_vol_args *vol_args;
1449         int ret;
1450
1451         vol_args = memdup_user(arg, sizeof(*vol_args));
1452         if (IS_ERR(vol_args))
1453                 return PTR_ERR(vol_args);
1454         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1455
1456         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1457                                               vol_args->fd, subvol,
1458                                               NULL, false, NULL);
1459
1460         kfree(vol_args);
1461         return ret;
1462 }
1463
1464 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1465                                                void __user *arg, int subvol)
1466 {
1467         struct btrfs_ioctl_vol_args_v2 *vol_args;
1468         int ret;
1469         u64 transid = 0;
1470         u64 *ptr = NULL;
1471         bool readonly = false;
1472         struct btrfs_qgroup_inherit *inherit = NULL;
1473
1474         vol_args = memdup_user(arg, sizeof(*vol_args));
1475         if (IS_ERR(vol_args))
1476                 return PTR_ERR(vol_args);
1477         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1478
1479         if (vol_args->flags &
1480             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1481               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1482                 ret = -EOPNOTSUPP;
1483                 goto out;
1484         }
1485
1486         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1487                 ptr = &transid;
1488         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1489                 readonly = true;
1490         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1491                 if (vol_args->size > PAGE_CACHE_SIZE) {
1492                         ret = -EINVAL;
1493                         goto out;
1494                 }
1495                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1496                 if (IS_ERR(inherit)) {
1497                         ret = PTR_ERR(inherit);
1498                         goto out;
1499                 }
1500         }
1501
1502         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1503                                               vol_args->fd, subvol, ptr,
1504                                               readonly, &inherit);
1505
1506         if (ret == 0 && ptr &&
1507             copy_to_user(arg +
1508                          offsetof(struct btrfs_ioctl_vol_args_v2,
1509                                   transid), ptr, sizeof(*ptr)))
1510                 ret = -EFAULT;
1511 out:
1512         kfree(vol_args);
1513         kfree(inherit);
1514         return ret;
1515 }
1516
1517 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1518                                                 void __user *arg)
1519 {
1520         struct inode *inode = fdentry(file)->d_inode;
1521         struct btrfs_root *root = BTRFS_I(inode)->root;
1522         int ret = 0;
1523         u64 flags = 0;
1524
1525         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1526                 return -EINVAL;
1527
1528         down_read(&root->fs_info->subvol_sem);
1529         if (btrfs_root_readonly(root))
1530                 flags |= BTRFS_SUBVOL_RDONLY;
1531         up_read(&root->fs_info->subvol_sem);
1532
1533         if (copy_to_user(arg, &flags, sizeof(flags)))
1534                 ret = -EFAULT;
1535
1536         return ret;
1537 }
1538
1539 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1540                                               void __user *arg)
1541 {
1542         struct inode *inode = fdentry(file)->d_inode;
1543         struct btrfs_root *root = BTRFS_I(inode)->root;
1544         struct btrfs_trans_handle *trans;
1545         u64 root_flags;
1546         u64 flags;
1547         int ret = 0;
1548
1549         if (root->fs_info->sb->s_flags & MS_RDONLY)
1550                 return -EROFS;
1551
1552         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1553                 return -EINVAL;
1554
1555         if (copy_from_user(&flags, arg, sizeof(flags)))
1556                 return -EFAULT;
1557
1558         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1559                 return -EINVAL;
1560
1561         if (flags & ~BTRFS_SUBVOL_RDONLY)
1562                 return -EOPNOTSUPP;
1563
1564         if (!inode_owner_or_capable(inode))
1565                 return -EACCES;
1566
1567         down_write(&root->fs_info->subvol_sem);
1568
1569         /* nothing to do */
1570         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1571                 goto out;
1572
1573         root_flags = btrfs_root_flags(&root->root_item);
1574         if (flags & BTRFS_SUBVOL_RDONLY)
1575                 btrfs_set_root_flags(&root->root_item,
1576                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1577         else
1578                 btrfs_set_root_flags(&root->root_item,
1579                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1580
1581         trans = btrfs_start_transaction(root, 1);
1582         if (IS_ERR(trans)) {
1583                 ret = PTR_ERR(trans);
1584                 goto out_reset;
1585         }
1586
1587         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1588                                 &root->root_key, &root->root_item);
1589
1590         btrfs_commit_transaction(trans, root);
1591 out_reset:
1592         if (ret)
1593                 btrfs_set_root_flags(&root->root_item, root_flags);
1594 out:
1595         up_write(&root->fs_info->subvol_sem);
1596         return ret;
1597 }
1598
1599 /*
1600  * helper to check if the subvolume references other subvolumes
1601  */
1602 static noinline int may_destroy_subvol(struct btrfs_root *root)
1603 {
1604         struct btrfs_path *path;
1605         struct btrfs_key key;
1606         int ret;
1607
1608         path = btrfs_alloc_path();
1609         if (!path)
1610                 return -ENOMEM;
1611
1612         key.objectid = root->root_key.objectid;
1613         key.type = BTRFS_ROOT_REF_KEY;
1614         key.offset = (u64)-1;
1615
1616         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1617                                 &key, path, 0, 0);
1618         if (ret < 0)
1619                 goto out;
1620         BUG_ON(ret == 0);
1621
1622         ret = 0;
1623         if (path->slots[0] > 0) {
1624                 path->slots[0]--;
1625                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1626                 if (key.objectid == root->root_key.objectid &&
1627                     key.type == BTRFS_ROOT_REF_KEY)
1628                         ret = -ENOTEMPTY;
1629         }
1630 out:
1631         btrfs_free_path(path);
1632         return ret;
1633 }
1634
1635 static noinline int key_in_sk(struct btrfs_key *key,
1636                               struct btrfs_ioctl_search_key *sk)
1637 {
1638         struct btrfs_key test;
1639         int ret;
1640
1641         test.objectid = sk->min_objectid;
1642         test.type = sk->min_type;
1643         test.offset = sk->min_offset;
1644
1645         ret = btrfs_comp_cpu_keys(key, &test);
1646         if (ret < 0)
1647                 return 0;
1648
1649         test.objectid = sk->max_objectid;
1650         test.type = sk->max_type;
1651         test.offset = sk->max_offset;
1652
1653         ret = btrfs_comp_cpu_keys(key, &test);
1654         if (ret > 0)
1655                 return 0;
1656         return 1;
1657 }
1658
1659 static noinline int copy_to_sk(struct btrfs_root *root,
1660                                struct btrfs_path *path,
1661                                struct btrfs_key *key,
1662                                struct btrfs_ioctl_search_key *sk,
1663                                char *buf,
1664                                unsigned long *sk_offset,
1665                                int *num_found)
1666 {
1667         u64 found_transid;
1668         struct extent_buffer *leaf;
1669         struct btrfs_ioctl_search_header sh;
1670         unsigned long item_off;
1671         unsigned long item_len;
1672         int nritems;
1673         int i;
1674         int slot;
1675         int ret = 0;
1676
1677         leaf = path->nodes[0];
1678         slot = path->slots[0];
1679         nritems = btrfs_header_nritems(leaf);
1680
1681         if (btrfs_header_generation(leaf) > sk->max_transid) {
1682                 i = nritems;
1683                 goto advance_key;
1684         }
1685         found_transid = btrfs_header_generation(leaf);
1686
1687         for (i = slot; i < nritems; i++) {
1688                 item_off = btrfs_item_ptr_offset(leaf, i);
1689                 item_len = btrfs_item_size_nr(leaf, i);
1690
1691                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1692                         item_len = 0;
1693
1694                 if (sizeof(sh) + item_len + *sk_offset >
1695                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1696                         ret = 1;
1697                         goto overflow;
1698                 }
1699
1700                 btrfs_item_key_to_cpu(leaf, key, i);
1701                 if (!key_in_sk(key, sk))
1702                         continue;
1703
1704                 sh.objectid = key->objectid;
1705                 sh.offset = key->offset;
1706                 sh.type = key->type;
1707                 sh.len = item_len;
1708                 sh.transid = found_transid;
1709
1710                 /* copy search result header */
1711                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1712                 *sk_offset += sizeof(sh);
1713
1714                 if (item_len) {
1715                         char *p = buf + *sk_offset;
1716                         /* copy the item */
1717                         read_extent_buffer(leaf, p,
1718                                            item_off, item_len);
1719                         *sk_offset += item_len;
1720                 }
1721                 (*num_found)++;
1722
1723                 if (*num_found >= sk->nr_items)
1724                         break;
1725         }
1726 advance_key:
1727         ret = 0;
1728         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1729                 key->offset++;
1730         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1731                 key->offset = 0;
1732                 key->type++;
1733         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1734                 key->offset = 0;
1735                 key->type = 0;
1736                 key->objectid++;
1737         } else
1738                 ret = 1;
1739 overflow:
1740         return ret;
1741 }
1742
1743 static noinline int search_ioctl(struct inode *inode,
1744                                  struct btrfs_ioctl_search_args *args)
1745 {
1746         struct btrfs_root *root;
1747         struct btrfs_key key;
1748         struct btrfs_key max_key;
1749         struct btrfs_path *path;
1750         struct btrfs_ioctl_search_key *sk = &args->key;
1751         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1752         int ret;
1753         int num_found = 0;
1754         unsigned long sk_offset = 0;
1755
1756         path = btrfs_alloc_path();
1757         if (!path)
1758                 return -ENOMEM;
1759
1760         if (sk->tree_id == 0) {
1761                 /* search the root of the inode that was passed */
1762                 root = BTRFS_I(inode)->root;
1763         } else {
1764                 key.objectid = sk->tree_id;
1765                 key.type = BTRFS_ROOT_ITEM_KEY;
1766                 key.offset = (u64)-1;
1767                 root = btrfs_read_fs_root_no_name(info, &key);
1768                 if (IS_ERR(root)) {
1769                         printk(KERN_ERR "could not find root %llu\n",
1770                                sk->tree_id);
1771                         btrfs_free_path(path);
1772                         return -ENOENT;
1773                 }
1774         }
1775
1776         key.objectid = sk->min_objectid;
1777         key.type = sk->min_type;
1778         key.offset = sk->min_offset;
1779
1780         max_key.objectid = sk->max_objectid;
1781         max_key.type = sk->max_type;
1782         max_key.offset = sk->max_offset;
1783
1784         path->keep_locks = 1;
1785
1786         while(1) {
1787                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1788                                            sk->min_transid);
1789                 if (ret != 0) {
1790                         if (ret > 0)
1791                                 ret = 0;
1792                         goto err;
1793                 }
1794                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1795                                  &sk_offset, &num_found);
1796                 btrfs_release_path(path);
1797                 if (ret || num_found >= sk->nr_items)
1798                         break;
1799
1800         }
1801         ret = 0;
1802 err:
1803         sk->nr_items = num_found;
1804         btrfs_free_path(path);
1805         return ret;
1806 }
1807
1808 static noinline int btrfs_ioctl_tree_search(struct file *file,
1809                                            void __user *argp)
1810 {
1811          struct btrfs_ioctl_search_args *args;
1812          struct inode *inode;
1813          int ret;
1814
1815         if (!capable(CAP_SYS_ADMIN))
1816                 return -EPERM;
1817
1818         args = memdup_user(argp, sizeof(*args));
1819         if (IS_ERR(args))
1820                 return PTR_ERR(args);
1821
1822         inode = fdentry(file)->d_inode;
1823         ret = search_ioctl(inode, args);
1824         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1825                 ret = -EFAULT;
1826         kfree(args);
1827         return ret;
1828 }
1829
1830 /*
1831  * Search INODE_REFs to identify path name of 'dirid' directory
1832  * in a 'tree_id' tree. and sets path name to 'name'.
1833  */
1834 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1835                                 u64 tree_id, u64 dirid, char *name)
1836 {
1837         struct btrfs_root *root;
1838         struct btrfs_key key;
1839         char *ptr;
1840         int ret = -1;
1841         int slot;
1842         int len;
1843         int total_len = 0;
1844         struct btrfs_inode_ref *iref;
1845         struct extent_buffer *l;
1846         struct btrfs_path *path;
1847
1848         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1849                 name[0]='\0';
1850                 return 0;
1851         }
1852
1853         path = btrfs_alloc_path();
1854         if (!path)
1855                 return -ENOMEM;
1856
1857         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1858
1859         key.objectid = tree_id;
1860         key.type = BTRFS_ROOT_ITEM_KEY;
1861         key.offset = (u64)-1;
1862         root = btrfs_read_fs_root_no_name(info, &key);
1863         if (IS_ERR(root)) {
1864                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1865                 ret = -ENOENT;
1866                 goto out;
1867         }
1868
1869         key.objectid = dirid;
1870         key.type = BTRFS_INODE_REF_KEY;
1871         key.offset = (u64)-1;
1872
1873         while(1) {
1874                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1875                 if (ret < 0)
1876                         goto out;
1877
1878                 l = path->nodes[0];
1879                 slot = path->slots[0];
1880                 if (ret > 0 && slot > 0)
1881                         slot--;
1882                 btrfs_item_key_to_cpu(l, &key, slot);
1883
1884                 if (ret > 0 && (key.objectid != dirid ||
1885                                 key.type != BTRFS_INODE_REF_KEY)) {
1886                         ret = -ENOENT;
1887                         goto out;
1888                 }
1889
1890                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1891                 len = btrfs_inode_ref_name_len(l, iref);
1892                 ptr -= len + 1;
1893                 total_len += len + 1;
1894                 if (ptr < name)
1895                         goto out;
1896
1897                 *(ptr + len) = '/';
1898                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1899
1900                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1901                         break;
1902
1903                 btrfs_release_path(path);
1904                 key.objectid = key.offset;
1905                 key.offset = (u64)-1;
1906                 dirid = key.objectid;
1907         }
1908         if (ptr < name)
1909                 goto out;
1910         memmove(name, ptr, total_len);
1911         name[total_len]='\0';
1912         ret = 0;
1913 out:
1914         btrfs_free_path(path);
1915         return ret;
1916 }
1917
1918 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1919                                            void __user *argp)
1920 {
1921          struct btrfs_ioctl_ino_lookup_args *args;
1922          struct inode *inode;
1923          int ret;
1924
1925         if (!capable(CAP_SYS_ADMIN))
1926                 return -EPERM;
1927
1928         args = memdup_user(argp, sizeof(*args));
1929         if (IS_ERR(args))
1930                 return PTR_ERR(args);
1931
1932         inode = fdentry(file)->d_inode;
1933
1934         if (args->treeid == 0)
1935                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1936
1937         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1938                                         args->treeid, args->objectid,
1939                                         args->name);
1940
1941         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1942                 ret = -EFAULT;
1943
1944         kfree(args);
1945         return ret;
1946 }
1947
1948 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1949                                              void __user *arg)
1950 {
1951         struct dentry *parent = fdentry(file);
1952         struct dentry *dentry;
1953         struct inode *dir = parent->d_inode;
1954         struct inode *inode;
1955         struct btrfs_root *root = BTRFS_I(dir)->root;
1956         struct btrfs_root *dest = NULL;
1957         struct btrfs_ioctl_vol_args *vol_args;
1958         struct btrfs_trans_handle *trans;
1959         int namelen;
1960         int ret;
1961         int err = 0;
1962
1963         vol_args = memdup_user(arg, sizeof(*vol_args));
1964         if (IS_ERR(vol_args))
1965                 return PTR_ERR(vol_args);
1966
1967         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1968         namelen = strlen(vol_args->name);
1969         if (strchr(vol_args->name, '/') ||
1970             strncmp(vol_args->name, "..", namelen) == 0) {
1971                 err = -EINVAL;
1972                 goto out;
1973         }
1974
1975         err = mnt_want_write_file(file);
1976         if (err)
1977                 goto out;
1978
1979         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1980         dentry = lookup_one_len(vol_args->name, parent, namelen);
1981         if (IS_ERR(dentry)) {
1982                 err = PTR_ERR(dentry);
1983                 goto out_unlock_dir;
1984         }
1985
1986         if (!dentry->d_inode) {
1987                 err = -ENOENT;
1988                 goto out_dput;
1989         }
1990
1991         inode = dentry->d_inode;
1992         dest = BTRFS_I(inode)->root;
1993         if (!capable(CAP_SYS_ADMIN)){
1994                 /*
1995                  * Regular user.  Only allow this with a special mount
1996                  * option, when the user has write+exec access to the
1997                  * subvol root, and when rmdir(2) would have been
1998                  * allowed.
1999                  *
2000                  * Note that this is _not_ check that the subvol is
2001                  * empty or doesn't contain data that we wouldn't
2002                  * otherwise be able to delete.
2003                  *
2004                  * Users who want to delete empty subvols should try
2005                  * rmdir(2).
2006                  */
2007                 err = -EPERM;
2008                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2009                         goto out_dput;
2010
2011                 /*
2012                  * Do not allow deletion if the parent dir is the same
2013                  * as the dir to be deleted.  That means the ioctl
2014                  * must be called on the dentry referencing the root
2015                  * of the subvol, not a random directory contained
2016                  * within it.
2017                  */
2018                 err = -EINVAL;
2019                 if (root == dest)
2020                         goto out_dput;
2021
2022                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2023                 if (err)
2024                         goto out_dput;
2025
2026                 /* check if subvolume may be deleted by a non-root user */
2027                 err = btrfs_may_delete(dir, dentry, 1);
2028                 if (err)
2029                         goto out_dput;
2030         }
2031
2032         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2033                 err = -EINVAL;
2034                 goto out_dput;
2035         }
2036
2037         mutex_lock(&inode->i_mutex);
2038         err = d_invalidate(dentry);
2039         if (err)
2040                 goto out_unlock;
2041
2042         down_write(&root->fs_info->subvol_sem);
2043
2044         err = may_destroy_subvol(dest);
2045         if (err)
2046                 goto out_up_write;
2047
2048         trans = btrfs_start_transaction(root, 0);
2049         if (IS_ERR(trans)) {
2050                 err = PTR_ERR(trans);
2051                 goto out_up_write;
2052         }
2053         trans->block_rsv = &root->fs_info->global_block_rsv;
2054
2055         ret = btrfs_unlink_subvol(trans, root, dir,
2056                                 dest->root_key.objectid,
2057                                 dentry->d_name.name,
2058                                 dentry->d_name.len);
2059         if (ret) {
2060                 err = ret;
2061                 btrfs_abort_transaction(trans, root, ret);
2062                 goto out_end_trans;
2063         }
2064
2065         btrfs_record_root_in_trans(trans, dest);
2066
2067         memset(&dest->root_item.drop_progress, 0,
2068                 sizeof(dest->root_item.drop_progress));
2069         dest->root_item.drop_level = 0;
2070         btrfs_set_root_refs(&dest->root_item, 0);
2071
2072         if (!xchg(&dest->orphan_item_inserted, 1)) {
2073                 ret = btrfs_insert_orphan_item(trans,
2074                                         root->fs_info->tree_root,
2075                                         dest->root_key.objectid);
2076                 if (ret) {
2077                         btrfs_abort_transaction(trans, root, ret);
2078                         err = ret;
2079                         goto out_end_trans;
2080                 }
2081         }
2082 out_end_trans:
2083         ret = btrfs_end_transaction(trans, root);
2084         if (ret && !err)
2085                 err = ret;
2086         inode->i_flags |= S_DEAD;
2087 out_up_write:
2088         up_write(&root->fs_info->subvol_sem);
2089 out_unlock:
2090         mutex_unlock(&inode->i_mutex);
2091         if (!err) {
2092                 shrink_dcache_sb(root->fs_info->sb);
2093                 btrfs_invalidate_inodes(dest);
2094                 d_delete(dentry);
2095         }
2096 out_dput:
2097         dput(dentry);
2098 out_unlock_dir:
2099         mutex_unlock(&dir->i_mutex);
2100         mnt_drop_write_file(file);
2101 out:
2102         kfree(vol_args);
2103         return err;
2104 }
2105
2106 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2107 {
2108         struct inode *inode = fdentry(file)->d_inode;
2109         struct btrfs_root *root = BTRFS_I(inode)->root;
2110         struct btrfs_ioctl_defrag_range_args *range;
2111         int ret;
2112
2113         if (btrfs_root_readonly(root))
2114                 return -EROFS;
2115
2116         ret = mnt_want_write_file(file);
2117         if (ret)
2118                 return ret;
2119
2120         switch (inode->i_mode & S_IFMT) {
2121         case S_IFDIR:
2122                 if (!capable(CAP_SYS_ADMIN)) {
2123                         ret = -EPERM;
2124                         goto out;
2125                 }
2126                 ret = btrfs_defrag_root(root, 0);
2127                 if (ret)
2128                         goto out;
2129                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2130                 break;
2131         case S_IFREG:
2132                 if (!(file->f_mode & FMODE_WRITE)) {
2133                         ret = -EINVAL;
2134                         goto out;
2135                 }
2136
2137                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2138                 if (!range) {
2139                         ret = -ENOMEM;
2140                         goto out;
2141                 }
2142
2143                 if (argp) {
2144                         if (copy_from_user(range, argp,
2145                                            sizeof(*range))) {
2146                                 ret = -EFAULT;
2147                                 kfree(range);
2148                                 goto out;
2149                         }
2150                         /* compression requires us to start the IO */
2151                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2152                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2153                                 range->extent_thresh = (u32)-1;
2154                         }
2155                 } else {
2156                         /* the rest are all set to zero by kzalloc */
2157                         range->len = (u64)-1;
2158                 }
2159                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2160                                         range, 0, 0);
2161                 if (ret > 0)
2162                         ret = 0;
2163                 kfree(range);
2164                 break;
2165         default:
2166                 ret = -EINVAL;
2167         }
2168 out:
2169         mnt_drop_write_file(file);
2170         return ret;
2171 }
2172
2173 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2174 {
2175         struct btrfs_ioctl_vol_args *vol_args;
2176         int ret;
2177
2178         if (!capable(CAP_SYS_ADMIN))
2179                 return -EPERM;
2180
2181         mutex_lock(&root->fs_info->volume_mutex);
2182         if (root->fs_info->balance_ctl) {
2183                 printk(KERN_INFO "btrfs: balance in progress\n");
2184                 ret = -EINVAL;
2185                 goto out;
2186         }
2187
2188         vol_args = memdup_user(arg, sizeof(*vol_args));
2189         if (IS_ERR(vol_args)) {
2190                 ret = PTR_ERR(vol_args);
2191                 goto out;
2192         }
2193
2194         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2195         ret = btrfs_init_new_device(root, vol_args->name);
2196
2197         kfree(vol_args);
2198 out:
2199         mutex_unlock(&root->fs_info->volume_mutex);
2200         return ret;
2201 }
2202
2203 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2204 {
2205         struct btrfs_ioctl_vol_args *vol_args;
2206         int ret;
2207
2208         if (!capable(CAP_SYS_ADMIN))
2209                 return -EPERM;
2210
2211         if (root->fs_info->sb->s_flags & MS_RDONLY)
2212                 return -EROFS;
2213
2214         mutex_lock(&root->fs_info->volume_mutex);
2215         if (root->fs_info->balance_ctl) {
2216                 printk(KERN_INFO "btrfs: balance in progress\n");
2217                 ret = -EINVAL;
2218                 goto out;
2219         }
2220
2221         vol_args = memdup_user(arg, sizeof(*vol_args));
2222         if (IS_ERR(vol_args)) {
2223                 ret = PTR_ERR(vol_args);
2224                 goto out;
2225         }
2226
2227         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2228         ret = btrfs_rm_device(root, vol_args->name);
2229
2230         kfree(vol_args);
2231 out:
2232         mutex_unlock(&root->fs_info->volume_mutex);
2233         return ret;
2234 }
2235
2236 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2237 {
2238         struct btrfs_ioctl_fs_info_args *fi_args;
2239         struct btrfs_device *device;
2240         struct btrfs_device *next;
2241         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2242         int ret = 0;
2243
2244         if (!capable(CAP_SYS_ADMIN))
2245                 return -EPERM;
2246
2247         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2248         if (!fi_args)
2249                 return -ENOMEM;
2250
2251         fi_args->num_devices = fs_devices->num_devices;
2252         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2253
2254         mutex_lock(&fs_devices->device_list_mutex);
2255         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2256                 if (device->devid > fi_args->max_id)
2257                         fi_args->max_id = device->devid;
2258         }
2259         mutex_unlock(&fs_devices->device_list_mutex);
2260
2261         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2262                 ret = -EFAULT;
2263
2264         kfree(fi_args);
2265         return ret;
2266 }
2267
2268 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2269 {
2270         struct btrfs_ioctl_dev_info_args *di_args;
2271         struct btrfs_device *dev;
2272         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2273         int ret = 0;
2274         char *s_uuid = NULL;
2275         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2276
2277         if (!capable(CAP_SYS_ADMIN))
2278                 return -EPERM;
2279
2280         di_args = memdup_user(arg, sizeof(*di_args));
2281         if (IS_ERR(di_args))
2282                 return PTR_ERR(di_args);
2283
2284         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2285                 s_uuid = di_args->uuid;
2286
2287         mutex_lock(&fs_devices->device_list_mutex);
2288         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2289         mutex_unlock(&fs_devices->device_list_mutex);
2290
2291         if (!dev) {
2292                 ret = -ENODEV;
2293                 goto out;
2294         }
2295
2296         di_args->devid = dev->devid;
2297         di_args->bytes_used = dev->bytes_used;
2298         di_args->total_bytes = dev->total_bytes;
2299         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2300         if (dev->name) {
2301                 struct rcu_string *name;
2302
2303                 rcu_read_lock();
2304                 name = rcu_dereference(dev->name);
2305                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2306                 rcu_read_unlock();
2307                 di_args->path[sizeof(di_args->path) - 1] = 0;
2308         } else {
2309                 di_args->path[0] = '\0';
2310         }
2311
2312 out:
2313         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2314                 ret = -EFAULT;
2315
2316         kfree(di_args);
2317         return ret;
2318 }
2319
2320 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2321                                        u64 off, u64 olen, u64 destoff)
2322 {
2323         struct inode *inode = fdentry(file)->d_inode;
2324         struct btrfs_root *root = BTRFS_I(inode)->root;
2325         struct file *src_file;
2326         struct inode *src;
2327         struct btrfs_trans_handle *trans;
2328         struct btrfs_path *path;
2329         struct extent_buffer *leaf;
2330         char *buf;
2331         struct btrfs_key key;
2332         u32 nritems;
2333         int slot;
2334         int ret;
2335         u64 len = olen;
2336         u64 bs = root->fs_info->sb->s_blocksize;
2337         u64 hint_byte;
2338
2339         /*
2340          * TODO:
2341          * - split compressed inline extents.  annoying: we need to
2342          *   decompress into destination's address_space (the file offset
2343          *   may change, so source mapping won't do), then recompress (or
2344          *   otherwise reinsert) a subrange.
2345          * - allow ranges within the same file to be cloned (provided
2346          *   they don't overlap)?
2347          */
2348
2349         /* the destination must be opened for writing */
2350         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2351                 return -EINVAL;
2352
2353         if (btrfs_root_readonly(root))
2354                 return -EROFS;
2355
2356         ret = mnt_want_write_file(file);
2357         if (ret)
2358                 return ret;
2359
2360         src_file = fget(srcfd);
2361         if (!src_file) {
2362                 ret = -EBADF;
2363                 goto out_drop_write;
2364         }
2365
2366         src = src_file->f_dentry->d_inode;
2367
2368         ret = -EINVAL;
2369         if (src == inode)
2370                 goto out_fput;
2371
2372         /* the src must be open for reading */
2373         if (!(src_file->f_mode & FMODE_READ))
2374                 goto out_fput;
2375
2376         /* don't make the dst file partly checksummed */
2377         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2378             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2379                 goto out_fput;
2380
2381         ret = -EISDIR;
2382         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2383                 goto out_fput;
2384
2385         ret = -EXDEV;
2386         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2387                 goto out_fput;
2388
2389         ret = -ENOMEM;
2390         buf = vmalloc(btrfs_level_size(root, 0));
2391         if (!buf)
2392                 goto out_fput;
2393
2394         path = btrfs_alloc_path();
2395         if (!path) {
2396                 vfree(buf);
2397                 goto out_fput;
2398         }
2399         path->reada = 2;
2400
2401         if (inode < src) {
2402                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2403                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2404         } else {
2405                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2406                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2407         }
2408
2409         /* determine range to clone */
2410         ret = -EINVAL;
2411         if (off + len > src->i_size || off + len < off)
2412                 goto out_unlock;
2413         if (len == 0)
2414                 olen = len = src->i_size - off;
2415         /* if we extend to eof, continue to block boundary */
2416         if (off + len == src->i_size)
2417                 len = ALIGN(src->i_size, bs) - off;
2418
2419         /* verify the end result is block aligned */
2420         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2421             !IS_ALIGNED(destoff, bs))
2422                 goto out_unlock;
2423
2424         if (destoff > inode->i_size) {
2425                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2426                 if (ret)
2427                         goto out_unlock;
2428         }
2429
2430         /* truncate page cache pages from target inode range */
2431         truncate_inode_pages_range(&inode->i_data, destoff,
2432                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2433
2434         /* do any pending delalloc/csum calc on src, one way or
2435            another, and lock file content */
2436         while (1) {
2437                 struct btrfs_ordered_extent *ordered;
2438                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len);
2439                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2440                 if (!ordered &&
2441                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2442                                    EXTENT_DELALLOC, 0, NULL))
2443                         break;
2444                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len);
2445                 if (ordered)
2446                         btrfs_put_ordered_extent(ordered);
2447                 btrfs_wait_ordered_range(src, off, len);
2448         }
2449
2450         /* clone data */
2451         key.objectid = btrfs_ino(src);
2452         key.type = BTRFS_EXTENT_DATA_KEY;
2453         key.offset = 0;
2454
2455         while (1) {
2456                 /*
2457                  * note the key will change type as we walk through the
2458                  * tree.
2459                  */
2460                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2461                 if (ret < 0)
2462                         goto out;
2463
2464                 nritems = btrfs_header_nritems(path->nodes[0]);
2465                 if (path->slots[0] >= nritems) {
2466                         ret = btrfs_next_leaf(root, path);
2467                         if (ret < 0)
2468                                 goto out;
2469                         if (ret > 0)
2470                                 break;
2471                         nritems = btrfs_header_nritems(path->nodes[0]);
2472                 }
2473                 leaf = path->nodes[0];
2474                 slot = path->slots[0];
2475
2476                 btrfs_item_key_to_cpu(leaf, &key, slot);
2477                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2478                     key.objectid != btrfs_ino(src))
2479                         break;
2480
2481                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2482                         struct btrfs_file_extent_item *extent;
2483                         int type;
2484                         u32 size;
2485                         struct btrfs_key new_key;
2486                         u64 disko = 0, diskl = 0;
2487                         u64 datao = 0, datal = 0;
2488                         u8 comp;
2489                         u64 endoff;
2490
2491                         size = btrfs_item_size_nr(leaf, slot);
2492                         read_extent_buffer(leaf, buf,
2493                                            btrfs_item_ptr_offset(leaf, slot),
2494                                            size);
2495
2496                         extent = btrfs_item_ptr(leaf, slot,
2497                                                 struct btrfs_file_extent_item);
2498                         comp = btrfs_file_extent_compression(leaf, extent);
2499                         type = btrfs_file_extent_type(leaf, extent);
2500                         if (type == BTRFS_FILE_EXTENT_REG ||
2501                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2502                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2503                                                                       extent);
2504                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2505                                                                  extent);
2506                                 datao = btrfs_file_extent_offset(leaf, extent);
2507                                 datal = btrfs_file_extent_num_bytes(leaf,
2508                                                                     extent);
2509                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2510                                 /* take upper bound, may be compressed */
2511                                 datal = btrfs_file_extent_ram_bytes(leaf,
2512                                                                     extent);
2513                         }
2514                         btrfs_release_path(path);
2515
2516                         if (key.offset + datal <= off ||
2517                             key.offset >= off+len)
2518                                 goto next;
2519
2520                         memcpy(&new_key, &key, sizeof(new_key));
2521                         new_key.objectid = btrfs_ino(inode);
2522                         if (off <= key.offset)
2523                                 new_key.offset = key.offset + destoff - off;
2524                         else
2525                                 new_key.offset = destoff;
2526
2527                         /*
2528                          * 1 - adjusting old extent (we may have to split it)
2529                          * 1 - add new extent
2530                          * 1 - inode update
2531                          */
2532                         trans = btrfs_start_transaction(root, 3);
2533                         if (IS_ERR(trans)) {
2534                                 ret = PTR_ERR(trans);
2535                                 goto out;
2536                         }
2537
2538                         if (type == BTRFS_FILE_EXTENT_REG ||
2539                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2540                                 /*
2541                                  *    a  | --- range to clone ---|  b
2542                                  * | ------------- extent ------------- |
2543                                  */
2544
2545                                 /* substract range b */
2546                                 if (key.offset + datal > off + len)
2547                                         datal = off + len - key.offset;
2548
2549                                 /* substract range a */
2550                                 if (off > key.offset) {
2551                                         datao += off - key.offset;
2552                                         datal -= off - key.offset;
2553                                 }
2554
2555                                 ret = btrfs_drop_extents(trans, inode,
2556                                                          new_key.offset,
2557                                                          new_key.offset + datal,
2558                                                          &hint_byte, 1);
2559                                 if (ret) {
2560                                         btrfs_abort_transaction(trans, root,
2561                                                                 ret);
2562                                         btrfs_end_transaction(trans, root);
2563                                         goto out;
2564                                 }
2565
2566                                 ret = btrfs_insert_empty_item(trans, root, path,
2567                                                               &new_key, size);
2568                                 if (ret) {
2569                                         btrfs_abort_transaction(trans, root,
2570                                                                 ret);
2571                                         btrfs_end_transaction(trans, root);
2572                                         goto out;
2573                                 }
2574
2575                                 leaf = path->nodes[0];
2576                                 slot = path->slots[0];
2577                                 write_extent_buffer(leaf, buf,
2578                                             btrfs_item_ptr_offset(leaf, slot),
2579                                             size);
2580
2581                                 extent = btrfs_item_ptr(leaf, slot,
2582                                                 struct btrfs_file_extent_item);
2583
2584                                 /* disko == 0 means it's a hole */
2585                                 if (!disko)
2586                                         datao = 0;
2587
2588                                 btrfs_set_file_extent_offset(leaf, extent,
2589                                                              datao);
2590                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2591                                                                 datal);
2592                                 if (disko) {
2593                                         inode_add_bytes(inode, datal);
2594                                         ret = btrfs_inc_extent_ref(trans, root,
2595                                                         disko, diskl, 0,
2596                                                         root->root_key.objectid,
2597                                                         btrfs_ino(inode),
2598                                                         new_key.offset - datao,
2599                                                         0);
2600                                         if (ret) {
2601                                                 btrfs_abort_transaction(trans,
2602                                                                         root,
2603                                                                         ret);
2604                                                 btrfs_end_transaction(trans,
2605                                                                       root);
2606                                                 goto out;
2607
2608                                         }
2609                                 }
2610                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2611                                 u64 skip = 0;
2612                                 u64 trim = 0;
2613                                 if (off > key.offset) {
2614                                         skip = off - key.offset;
2615                                         new_key.offset += skip;
2616                                 }
2617
2618                                 if (key.offset + datal > off+len)
2619                                         trim = key.offset + datal - (off+len);
2620
2621                                 if (comp && (skip || trim)) {
2622                                         ret = -EINVAL;
2623                                         btrfs_end_transaction(trans, root);
2624                                         goto out;
2625                                 }
2626                                 size -= skip + trim;
2627                                 datal -= skip + trim;
2628
2629                                 ret = btrfs_drop_extents(trans, inode,
2630                                                          new_key.offset,
2631                                                          new_key.offset + datal,
2632                                                          &hint_byte, 1);
2633                                 if (ret) {
2634                                         btrfs_abort_transaction(trans, root,
2635                                                                 ret);
2636                                         btrfs_end_transaction(trans, root);
2637                                         goto out;
2638                                 }
2639
2640                                 ret = btrfs_insert_empty_item(trans, root, path,
2641                                                               &new_key, size);
2642                                 if (ret) {
2643                                         btrfs_abort_transaction(trans, root,
2644                                                                 ret);
2645                                         btrfs_end_transaction(trans, root);
2646                                         goto out;
2647                                 }
2648
2649                                 if (skip) {
2650                                         u32 start =
2651                                           btrfs_file_extent_calc_inline_size(0);
2652                                         memmove(buf+start, buf+start+skip,
2653                                                 datal);
2654                                 }
2655
2656                                 leaf = path->nodes[0];
2657                                 slot = path->slots[0];
2658                                 write_extent_buffer(leaf, buf,
2659                                             btrfs_item_ptr_offset(leaf, slot),
2660                                             size);
2661                                 inode_add_bytes(inode, datal);
2662                         }
2663
2664                         btrfs_mark_buffer_dirty(leaf);
2665                         btrfs_release_path(path);
2666
2667                         inode_inc_iversion(inode);
2668                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2669
2670                         /*
2671                          * we round up to the block size at eof when
2672                          * determining which extents to clone above,
2673                          * but shouldn't round up the file size
2674                          */
2675                         endoff = new_key.offset + datal;
2676                         if (endoff > destoff+olen)
2677                                 endoff = destoff+olen;
2678                         if (endoff > inode->i_size)
2679                                 btrfs_i_size_write(inode, endoff);
2680
2681                         ret = btrfs_update_inode(trans, root, inode);
2682                         if (ret) {
2683                                 btrfs_abort_transaction(trans, root, ret);
2684                                 btrfs_end_transaction(trans, root);
2685                                 goto out;
2686                         }
2687                         ret = btrfs_end_transaction(trans, root);
2688                 }
2689 next:
2690                 btrfs_release_path(path);
2691                 key.offset++;
2692         }
2693         ret = 0;
2694 out:
2695         btrfs_release_path(path);
2696         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len);
2697 out_unlock:
2698         mutex_unlock(&src->i_mutex);
2699         mutex_unlock(&inode->i_mutex);
2700         vfree(buf);
2701         btrfs_free_path(path);
2702 out_fput:
2703         fput(src_file);
2704 out_drop_write:
2705         mnt_drop_write_file(file);
2706         return ret;
2707 }
2708
2709 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2710 {
2711         struct btrfs_ioctl_clone_range_args args;
2712
2713         if (copy_from_user(&args, argp, sizeof(args)))
2714                 return -EFAULT;
2715         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2716                                  args.src_length, args.dest_offset);
2717 }
2718
2719 /*
2720  * there are many ways the trans_start and trans_end ioctls can lead
2721  * to deadlocks.  They should only be used by applications that
2722  * basically own the machine, and have a very in depth understanding
2723  * of all the possible deadlocks and enospc problems.
2724  */
2725 static long btrfs_ioctl_trans_start(struct file *file)
2726 {
2727         struct inode *inode = fdentry(file)->d_inode;
2728         struct btrfs_root *root = BTRFS_I(inode)->root;
2729         struct btrfs_trans_handle *trans;
2730         int ret;
2731
2732         ret = -EPERM;
2733         if (!capable(CAP_SYS_ADMIN))
2734                 goto out;
2735
2736         ret = -EINPROGRESS;
2737         if (file->private_data)
2738                 goto out;
2739
2740         ret = -EROFS;
2741         if (btrfs_root_readonly(root))
2742                 goto out;
2743
2744         ret = mnt_want_write_file(file);
2745         if (ret)
2746                 goto out;
2747
2748         atomic_inc(&root->fs_info->open_ioctl_trans);
2749
2750         ret = -ENOMEM;
2751         trans = btrfs_start_ioctl_transaction(root);
2752         if (IS_ERR(trans))
2753                 goto out_drop;
2754
2755         file->private_data = trans;
2756         return 0;
2757
2758 out_drop:
2759         atomic_dec(&root->fs_info->open_ioctl_trans);
2760         mnt_drop_write_file(file);
2761 out:
2762         return ret;
2763 }
2764
2765 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2766 {
2767         struct inode *inode = fdentry(file)->d_inode;
2768         struct btrfs_root *root = BTRFS_I(inode)->root;
2769         struct btrfs_root *new_root;
2770         struct btrfs_dir_item *di;
2771         struct btrfs_trans_handle *trans;
2772         struct btrfs_path *path;
2773         struct btrfs_key location;
2774         struct btrfs_disk_key disk_key;
2775         struct btrfs_super_block *disk_super;
2776         u64 features;
2777         u64 objectid = 0;
2778         u64 dir_id;
2779
2780         if (!capable(CAP_SYS_ADMIN))
2781                 return -EPERM;
2782
2783         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2784                 return -EFAULT;
2785
2786         if (!objectid)
2787                 objectid = root->root_key.objectid;
2788
2789         location.objectid = objectid;
2790         location.type = BTRFS_ROOT_ITEM_KEY;
2791         location.offset = (u64)-1;
2792
2793         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2794         if (IS_ERR(new_root))
2795                 return PTR_ERR(new_root);
2796
2797         if (btrfs_root_refs(&new_root->root_item) == 0)
2798                 return -ENOENT;
2799
2800         path = btrfs_alloc_path();
2801         if (!path)
2802                 return -ENOMEM;
2803         path->leave_spinning = 1;
2804
2805         trans = btrfs_start_transaction(root, 1);
2806         if (IS_ERR(trans)) {
2807                 btrfs_free_path(path);
2808                 return PTR_ERR(trans);
2809         }
2810
2811         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2812         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2813                                    dir_id, "default", 7, 1);
2814         if (IS_ERR_OR_NULL(di)) {
2815                 btrfs_free_path(path);
2816                 btrfs_end_transaction(trans, root);
2817                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2818                        "this isn't going to work\n");
2819                 return -ENOENT;
2820         }
2821
2822         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2823         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2824         btrfs_mark_buffer_dirty(path->nodes[0]);
2825         btrfs_free_path(path);
2826
2827         disk_super = root->fs_info->super_copy;
2828         features = btrfs_super_incompat_flags(disk_super);
2829         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2830                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2831                 btrfs_set_super_incompat_flags(disk_super, features);
2832         }
2833         btrfs_end_transaction(trans, root);
2834
2835         return 0;
2836 }
2837
2838 static void get_block_group_info(struct list_head *groups_list,
2839                                  struct btrfs_ioctl_space_info *space)
2840 {
2841         struct btrfs_block_group_cache *block_group;
2842
2843         space->total_bytes = 0;
2844         space->used_bytes = 0;
2845         space->flags = 0;
2846         list_for_each_entry(block_group, groups_list, list) {
2847                 space->flags = block_group->flags;
2848                 space->total_bytes += block_group->key.offset;
2849                 space->used_bytes +=
2850                         btrfs_block_group_used(&block_group->item);
2851         }
2852 }
2853
2854 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2855 {
2856         struct btrfs_ioctl_space_args space_args;
2857         struct btrfs_ioctl_space_info space;
2858         struct btrfs_ioctl_space_info *dest;
2859         struct btrfs_ioctl_space_info *dest_orig;
2860         struct btrfs_ioctl_space_info __user *user_dest;
2861         struct btrfs_space_info *info;
2862         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2863                        BTRFS_BLOCK_GROUP_SYSTEM,
2864                        BTRFS_BLOCK_GROUP_METADATA,
2865                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2866         int num_types = 4;
2867         int alloc_size;
2868         int ret = 0;
2869         u64 slot_count = 0;
2870         int i, c;
2871
2872         if (copy_from_user(&space_args,
2873                            (struct btrfs_ioctl_space_args __user *)arg,
2874                            sizeof(space_args)))
2875                 return -EFAULT;
2876
2877         for (i = 0; i < num_types; i++) {
2878                 struct btrfs_space_info *tmp;
2879
2880                 info = NULL;
2881                 rcu_read_lock();
2882                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2883                                         list) {
2884                         if (tmp->flags == types[i]) {
2885                                 info = tmp;
2886                                 break;
2887                         }
2888                 }
2889                 rcu_read_unlock();
2890
2891                 if (!info)
2892                         continue;
2893
2894                 down_read(&info->groups_sem);
2895                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2896                         if (!list_empty(&info->block_groups[c]))
2897                                 slot_count++;
2898                 }
2899                 up_read(&info->groups_sem);
2900         }
2901
2902         /* space_slots == 0 means they are asking for a count */
2903         if (space_args.space_slots == 0) {
2904                 space_args.total_spaces = slot_count;
2905                 goto out;
2906         }
2907
2908         slot_count = min_t(u64, space_args.space_slots, slot_count);
2909
2910         alloc_size = sizeof(*dest) * slot_count;
2911
2912         /* we generally have at most 6 or so space infos, one for each raid
2913          * level.  So, a whole page should be more than enough for everyone
2914          */
2915         if (alloc_size > PAGE_CACHE_SIZE)
2916                 return -ENOMEM;
2917
2918         space_args.total_spaces = 0;
2919         dest = kmalloc(alloc_size, GFP_NOFS);
2920         if (!dest)
2921                 return -ENOMEM;
2922         dest_orig = dest;
2923
2924         /* now we have a buffer to copy into */
2925         for (i = 0; i < num_types; i++) {
2926                 struct btrfs_space_info *tmp;
2927
2928                 if (!slot_count)
2929                         break;
2930
2931                 info = NULL;
2932                 rcu_read_lock();
2933                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2934                                         list) {
2935                         if (tmp->flags == types[i]) {
2936                                 info = tmp;
2937                                 break;
2938                         }
2939                 }
2940                 rcu_read_unlock();
2941
2942                 if (!info)
2943                         continue;
2944                 down_read(&info->groups_sem);
2945                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2946                         if (!list_empty(&info->block_groups[c])) {
2947                                 get_block_group_info(&info->block_groups[c],
2948                                                      &space);
2949                                 memcpy(dest, &space, sizeof(space));
2950                                 dest++;
2951                                 space_args.total_spaces++;
2952                                 slot_count--;
2953                         }
2954                         if (!slot_count)
2955                                 break;
2956                 }
2957                 up_read(&info->groups_sem);
2958         }
2959
2960         user_dest = (struct btrfs_ioctl_space_info __user *)
2961                 (arg + sizeof(struct btrfs_ioctl_space_args));
2962
2963         if (copy_to_user(user_dest, dest_orig, alloc_size))
2964                 ret = -EFAULT;
2965
2966         kfree(dest_orig);
2967 out:
2968         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2969                 ret = -EFAULT;
2970
2971         return ret;
2972 }
2973
2974 /*
2975  * there are many ways the trans_start and trans_end ioctls can lead
2976  * to deadlocks.  They should only be used by applications that
2977  * basically own the machine, and have a very in depth understanding
2978  * of all the possible deadlocks and enospc problems.
2979  */
2980 long btrfs_ioctl_trans_end(struct file *file)
2981 {
2982         struct inode *inode = fdentry(file)->d_inode;
2983         struct btrfs_root *root = BTRFS_I(inode)->root;
2984         struct btrfs_trans_handle *trans;
2985
2986         trans = file->private_data;
2987         if (!trans)
2988                 return -EINVAL;
2989         file->private_data = NULL;
2990
2991         btrfs_end_transaction(trans, root);
2992
2993         atomic_dec(&root->fs_info->open_ioctl_trans);
2994
2995         mnt_drop_write_file(file);
2996         return 0;
2997 }
2998
2999 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
3000 {
3001         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
3002         struct btrfs_trans_handle *trans;
3003         u64 transid;
3004         int ret;
3005
3006         trans = btrfs_start_transaction(root, 0);
3007         if (IS_ERR(trans))
3008                 return PTR_ERR(trans);
3009         transid = trans->transid;
3010         ret = btrfs_commit_transaction_async(trans, root, 0);
3011         if (ret) {
3012                 btrfs_end_transaction(trans, root);
3013                 return ret;
3014         }
3015
3016         if (argp)
3017                 if (copy_to_user(argp, &transid, sizeof(transid)))
3018                         return -EFAULT;
3019         return 0;
3020 }
3021
3022 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
3023 {
3024         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
3025         u64 transid;
3026
3027         if (argp) {
3028                 if (copy_from_user(&transid, argp, sizeof(transid)))
3029                         return -EFAULT;
3030         } else {
3031                 transid = 0;  /* current trans */
3032         }
3033         return btrfs_wait_for_commit(root, transid);
3034 }
3035
3036 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
3037 {
3038         int ret;
3039         struct btrfs_ioctl_scrub_args *sa;
3040
3041         if (!capable(CAP_SYS_ADMIN))
3042                 return -EPERM;
3043
3044         sa = memdup_user(arg, sizeof(*sa));
3045         if (IS_ERR(sa))
3046                 return PTR_ERR(sa);
3047
3048         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
3049                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
3050
3051         if (copy_to_user(arg, sa, sizeof(*sa)))
3052                 ret = -EFAULT;
3053
3054         kfree(sa);
3055         return ret;
3056 }
3057
3058 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3059 {
3060         if (!capable(CAP_SYS_ADMIN))
3061                 return -EPERM;
3062
3063         return btrfs_scrub_cancel(root);
3064 }
3065
3066 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3067                                        void __user *arg)
3068 {
3069         struct btrfs_ioctl_scrub_args *sa;
3070         int ret;
3071
3072         if (!capable(CAP_SYS_ADMIN))
3073                 return -EPERM;
3074
3075         sa = memdup_user(arg, sizeof(*sa));
3076         if (IS_ERR(sa))
3077                 return PTR_ERR(sa);
3078
3079         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3080
3081         if (copy_to_user(arg, sa, sizeof(*sa)))
3082                 ret = -EFAULT;
3083
3084         kfree(sa);
3085         return ret;
3086 }
3087
3088 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3089                                       void __user *arg, int reset_after_read)
3090 {
3091         struct btrfs_ioctl_get_dev_stats *sa;
3092         int ret;
3093
3094         if (reset_after_read && !capable(CAP_SYS_ADMIN))
3095                 return -EPERM;
3096
3097         sa = memdup_user(arg, sizeof(*sa));
3098         if (IS_ERR(sa))
3099                 return PTR_ERR(sa);
3100
3101         ret = btrfs_get_dev_stats(root, sa, reset_after_read);
3102
3103         if (copy_to_user(arg, sa, sizeof(*sa)))
3104                 ret = -EFAULT;
3105
3106         kfree(sa);
3107         return ret;
3108 }
3109
3110 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3111 {
3112         int ret = 0;
3113         int i;
3114         u64 rel_ptr;
3115         int size;
3116         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3117         struct inode_fs_paths *ipath = NULL;
3118         struct btrfs_path *path;
3119
3120         if (!capable(CAP_SYS_ADMIN))
3121                 return -EPERM;
3122
3123         path = btrfs_alloc_path();
3124         if (!path) {
3125                 ret = -ENOMEM;
3126                 goto out;
3127         }
3128
3129         ipa = memdup_user(arg, sizeof(*ipa));
3130         if (IS_ERR(ipa)) {
3131                 ret = PTR_ERR(ipa);
3132                 ipa = NULL;
3133                 goto out;
3134         }
3135
3136         size = min_t(u32, ipa->size, 4096);
3137         ipath = init_ipath(size, root, path);
3138         if (IS_ERR(ipath)) {
3139                 ret = PTR_ERR(ipath);
3140                 ipath = NULL;
3141                 goto out;
3142         }
3143
3144         ret = paths_from_inode(ipa->inum, ipath);
3145         if (ret < 0)
3146                 goto out;
3147
3148         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3149                 rel_ptr = ipath->fspath->val[i] -
3150                           (u64)(unsigned long)ipath->fspath->val;
3151                 ipath->fspath->val[i] = rel_ptr;
3152         }
3153
3154         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3155                            (void *)(unsigned long)ipath->fspath, size);
3156         if (ret) {
3157                 ret = -EFAULT;
3158                 goto out;
3159         }
3160
3161 out:
3162         btrfs_free_path(path);
3163         free_ipath(ipath);
3164         kfree(ipa);
3165
3166         return ret;
3167 }
3168
3169 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3170 {
3171         struct btrfs_data_container *inodes = ctx;
3172         const size_t c = 3 * sizeof(u64);
3173
3174         if (inodes->bytes_left >= c) {
3175                 inodes->bytes_left -= c;
3176                 inodes->val[inodes->elem_cnt] = inum;
3177                 inodes->val[inodes->elem_cnt + 1] = offset;
3178                 inodes->val[inodes->elem_cnt + 2] = root;
3179                 inodes->elem_cnt += 3;
3180         } else {
3181                 inodes->bytes_missing += c - inodes->bytes_left;
3182                 inodes->bytes_left = 0;
3183                 inodes->elem_missed += 3;
3184         }
3185
3186         return 0;
3187 }
3188
3189 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3190                                         void __user *arg)
3191 {
3192         int ret = 0;
3193         int size;
3194         u64 extent_item_pos;
3195         struct btrfs_ioctl_logical_ino_args *loi;
3196         struct btrfs_data_container *inodes = NULL;
3197         struct btrfs_path *path = NULL;
3198         struct btrfs_key key;
3199
3200         if (!capable(CAP_SYS_ADMIN))
3201                 return -EPERM;
3202
3203         loi = memdup_user(arg, sizeof(*loi));
3204         if (IS_ERR(loi)) {
3205                 ret = PTR_ERR(loi);
3206                 loi = NULL;
3207                 goto out;
3208         }
3209
3210         path = btrfs_alloc_path();
3211         if (!path) {
3212                 ret = -ENOMEM;
3213                 goto out;
3214         }
3215
3216         size = min_t(u32, loi->size, 4096);
3217         inodes = init_data_container(size);
3218         if (IS_ERR(inodes)) {
3219                 ret = PTR_ERR(inodes);
3220                 inodes = NULL;
3221                 goto out;
3222         }
3223
3224         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
3225         btrfs_release_path(path);
3226
3227         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
3228                 ret = -ENOENT;
3229         if (ret < 0)
3230                 goto out;
3231
3232         extent_item_pos = loi->logical - key.objectid;
3233         ret = iterate_extent_inodes(root->fs_info, key.objectid,
3234                                         extent_item_pos, 0, build_ino_list,
3235                                         inodes);
3236
3237         if (ret < 0)
3238                 goto out;
3239
3240         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3241                            (void *)(unsigned long)inodes, size);
3242         if (ret)
3243                 ret = -EFAULT;
3244
3245 out:
3246         btrfs_free_path(path);
3247         kfree(inodes);
3248         kfree(loi);
3249
3250         return ret;
3251 }
3252
3253 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3254                                struct btrfs_ioctl_balance_args *bargs)
3255 {
3256         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3257
3258         bargs->flags = bctl->flags;
3259
3260         if (atomic_read(&fs_info->balance_running))
3261                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3262         if (atomic_read(&fs_info->balance_pause_req))
3263                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3264         if (atomic_read(&fs_info->balance_cancel_req))
3265                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3266
3267         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3268         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3269         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3270
3271         if (lock) {
3272                 spin_lock(&fs_info->balance_lock);
3273                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3274                 spin_unlock(&fs_info->balance_lock);
3275         } else {
3276                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3277         }
3278 }
3279
3280 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3281 {
3282         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3283         struct btrfs_fs_info *fs_info = root->fs_info;
3284         struct btrfs_ioctl_balance_args *bargs;
3285         struct btrfs_balance_control *bctl;
3286         int ret;
3287
3288         if (!capable(CAP_SYS_ADMIN))
3289                 return -EPERM;
3290
3291         if (fs_info->sb->s_flags & MS_RDONLY)
3292                 return -EROFS;
3293
3294         ret = mnt_want_write(file->f_path.mnt);
3295         if (ret)
3296                 return ret;
3297
3298         mutex_lock(&fs_info->volume_mutex);
3299         mutex_lock(&fs_info->balance_mutex);
3300
3301         if (arg) {
3302                 bargs = memdup_user(arg, sizeof(*bargs));
3303                 if (IS_ERR(bargs)) {
3304                         ret = PTR_ERR(bargs);
3305                         goto out;
3306                 }
3307
3308                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3309                         if (!fs_info->balance_ctl) {
3310                                 ret = -ENOTCONN;
3311                                 goto out_bargs;
3312                         }
3313
3314                         bctl = fs_info->balance_ctl;
3315                         spin_lock(&fs_info->balance_lock);
3316                         bctl->flags |= BTRFS_BALANCE_RESUME;
3317                         spin_unlock(&fs_info->balance_lock);
3318
3319                         goto do_balance;
3320                 }
3321         } else {
3322                 bargs = NULL;
3323         }
3324
3325         if (fs_info->balance_ctl) {
3326                 ret = -EINPROGRESS;
3327                 goto out_bargs;
3328         }
3329
3330         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3331         if (!bctl) {
3332                 ret = -ENOMEM;
3333                 goto out_bargs;
3334         }