2db7c1455c7f8e2f0c526662edd921679fa189b5
[~shefty/rdma-dev.git] / fs / btrfs / ioctl.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include "compat.h"
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "ioctl.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54 #include "backref.h"
55
56 /* Mask out flags that are inappropriate for the given type of inode. */
57 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
58 {
59         if (S_ISDIR(mode))
60                 return flags;
61         else if (S_ISREG(mode))
62                 return flags & ~FS_DIRSYNC_FL;
63         else
64                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
65 }
66
67 /*
68  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
69  */
70 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
71 {
72         unsigned int iflags = 0;
73
74         if (flags & BTRFS_INODE_SYNC)
75                 iflags |= FS_SYNC_FL;
76         if (flags & BTRFS_INODE_IMMUTABLE)
77                 iflags |= FS_IMMUTABLE_FL;
78         if (flags & BTRFS_INODE_APPEND)
79                 iflags |= FS_APPEND_FL;
80         if (flags & BTRFS_INODE_NODUMP)
81                 iflags |= FS_NODUMP_FL;
82         if (flags & BTRFS_INODE_NOATIME)
83                 iflags |= FS_NOATIME_FL;
84         if (flags & BTRFS_INODE_DIRSYNC)
85                 iflags |= FS_DIRSYNC_FL;
86         if (flags & BTRFS_INODE_NODATACOW)
87                 iflags |= FS_NOCOW_FL;
88
89         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
90                 iflags |= FS_COMPR_FL;
91         else if (flags & BTRFS_INODE_NOCOMPRESS)
92                 iflags |= FS_NOCOMP_FL;
93
94         return iflags;
95 }
96
97 /*
98  * Update inode->i_flags based on the btrfs internal flags.
99  */
100 void btrfs_update_iflags(struct inode *inode)
101 {
102         struct btrfs_inode *ip = BTRFS_I(inode);
103
104         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
105
106         if (ip->flags & BTRFS_INODE_SYNC)
107                 inode->i_flags |= S_SYNC;
108         if (ip->flags & BTRFS_INODE_IMMUTABLE)
109                 inode->i_flags |= S_IMMUTABLE;
110         if (ip->flags & BTRFS_INODE_APPEND)
111                 inode->i_flags |= S_APPEND;
112         if (ip->flags & BTRFS_INODE_NOATIME)
113                 inode->i_flags |= S_NOATIME;
114         if (ip->flags & BTRFS_INODE_DIRSYNC)
115                 inode->i_flags |= S_DIRSYNC;
116 }
117
118 /*
119  * Inherit flags from the parent inode.
120  *
121  * Currently only the compression flags and the cow flags are inherited.
122  */
123 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
124 {
125         unsigned int flags;
126
127         if (!dir)
128                 return;
129
130         flags = BTRFS_I(dir)->flags;
131
132         if (flags & BTRFS_INODE_NOCOMPRESS) {
133                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
134                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
135         } else if (flags & BTRFS_INODE_COMPRESS) {
136                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
137                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
138         }
139
140         if (flags & BTRFS_INODE_NODATACOW)
141                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
142
143         btrfs_update_iflags(inode);
144 }
145
146 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
147 {
148         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
149         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
150
151         if (copy_to_user(arg, &flags, sizeof(flags)))
152                 return -EFAULT;
153         return 0;
154 }
155
156 static int check_flags(unsigned int flags)
157 {
158         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
159                       FS_NOATIME_FL | FS_NODUMP_FL | \
160                       FS_SYNC_FL | FS_DIRSYNC_FL | \
161                       FS_NOCOMP_FL | FS_COMPR_FL |
162                       FS_NOCOW_FL))
163                 return -EOPNOTSUPP;
164
165         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
166                 return -EINVAL;
167
168         return 0;
169 }
170
171 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
172 {
173         struct inode *inode = file->f_path.dentry->d_inode;
174         struct btrfs_inode *ip = BTRFS_I(inode);
175         struct btrfs_root *root = ip->root;
176         struct btrfs_trans_handle *trans;
177         unsigned int flags, oldflags;
178         int ret;
179         u64 ip_oldflags;
180         unsigned int i_oldflags;
181
182         if (btrfs_root_readonly(root))
183                 return -EROFS;
184
185         if (copy_from_user(&flags, arg, sizeof(flags)))
186                 return -EFAULT;
187
188         ret = check_flags(flags);
189         if (ret)
190                 return ret;
191
192         if (!inode_owner_or_capable(inode))
193                 return -EACCES;
194
195         mutex_lock(&inode->i_mutex);
196
197         ip_oldflags = ip->flags;
198         i_oldflags = inode->i_flags;
199
200         flags = btrfs_mask_flags(inode->i_mode, flags);
201         oldflags = btrfs_flags_to_ioctl(ip->flags);
202         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
203                 if (!capable(CAP_LINUX_IMMUTABLE)) {
204                         ret = -EPERM;
205                         goto out_unlock;
206                 }
207         }
208
209         ret = mnt_want_write_file(file);
210         if (ret)
211                 goto out_unlock;
212
213         if (flags & FS_SYNC_FL)
214                 ip->flags |= BTRFS_INODE_SYNC;
215         else
216                 ip->flags &= ~BTRFS_INODE_SYNC;
217         if (flags & FS_IMMUTABLE_FL)
218                 ip->flags |= BTRFS_INODE_IMMUTABLE;
219         else
220                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
221         if (flags & FS_APPEND_FL)
222                 ip->flags |= BTRFS_INODE_APPEND;
223         else
224                 ip->flags &= ~BTRFS_INODE_APPEND;
225         if (flags & FS_NODUMP_FL)
226                 ip->flags |= BTRFS_INODE_NODUMP;
227         else
228                 ip->flags &= ~BTRFS_INODE_NODUMP;
229         if (flags & FS_NOATIME_FL)
230                 ip->flags |= BTRFS_INODE_NOATIME;
231         else
232                 ip->flags &= ~BTRFS_INODE_NOATIME;
233         if (flags & FS_DIRSYNC_FL)
234                 ip->flags |= BTRFS_INODE_DIRSYNC;
235         else
236                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
237         if (flags & FS_NOCOW_FL)
238                 ip->flags |= BTRFS_INODE_NODATACOW;
239         else
240                 ip->flags &= ~BTRFS_INODE_NODATACOW;
241
242         /*
243          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
244          * flag may be changed automatically if compression code won't make
245          * things smaller.
246          */
247         if (flags & FS_NOCOMP_FL) {
248                 ip->flags &= ~BTRFS_INODE_COMPRESS;
249                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
250         } else if (flags & FS_COMPR_FL) {
251                 ip->flags |= BTRFS_INODE_COMPRESS;
252                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
253         } else {
254                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
255         }
256
257         trans = btrfs_start_transaction(root, 1);
258         if (IS_ERR(trans)) {
259                 ret = PTR_ERR(trans);
260                 goto out_drop;
261         }
262
263         btrfs_update_iflags(inode);
264         inode->i_ctime = CURRENT_TIME;
265         ret = btrfs_update_inode(trans, root, inode);
266
267         btrfs_end_transaction(trans, root);
268  out_drop:
269         if (ret) {
270                 ip->flags = ip_oldflags;
271                 inode->i_flags = i_oldflags;
272         }
273
274         mnt_drop_write_file(file);
275  out_unlock:
276         mutex_unlock(&inode->i_mutex);
277         return ret;
278 }
279
280 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
281 {
282         struct inode *inode = file->f_path.dentry->d_inode;
283
284         return put_user(inode->i_generation, arg);
285 }
286
287 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
288 {
289         struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info;
290         struct btrfs_fs_info *fs_info = root->fs_info;
291         struct btrfs_device *device;
292         struct request_queue *q;
293         struct fstrim_range range;
294         u64 minlen = ULLONG_MAX;
295         u64 num_devices = 0;
296         u64 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
297         int ret;
298
299         if (!capable(CAP_SYS_ADMIN))
300                 return -EPERM;
301
302         rcu_read_lock();
303         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
304                                 dev_list) {
305                 if (!device->bdev)
306                         continue;
307                 q = bdev_get_queue(device->bdev);
308                 if (blk_queue_discard(q)) {
309                         num_devices++;
310                         minlen = min((u64)q->limits.discard_granularity,
311                                      minlen);
312                 }
313         }
314         rcu_read_unlock();
315
316         if (!num_devices)
317                 return -EOPNOTSUPP;
318         if (copy_from_user(&range, arg, sizeof(range)))
319                 return -EFAULT;
320         if (range.start > total_bytes)
321                 return -EINVAL;
322
323         range.len = min(range.len, total_bytes - range.start);
324         range.minlen = max(range.minlen, minlen);
325         ret = btrfs_trim_fs(root, &range);
326         if (ret < 0)
327                 return ret;
328
329         if (copy_to_user(arg, &range, sizeof(range)))
330                 return -EFAULT;
331
332         return 0;
333 }
334
335 static noinline int create_subvol(struct btrfs_root *root,
336                                   struct dentry *dentry,
337                                   char *name, int namelen,
338                                   u64 *async_transid)
339 {
340         struct btrfs_trans_handle *trans;
341         struct btrfs_key key;
342         struct btrfs_root_item root_item;
343         struct btrfs_inode_item *inode_item;
344         struct extent_buffer *leaf;
345         struct btrfs_root *new_root;
346         struct dentry *parent = dentry->d_parent;
347         struct inode *dir;
348         int ret;
349         int err;
350         u64 objectid;
351         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
352         u64 index = 0;
353
354         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
355         if (ret)
356                 return ret;
357
358         dir = parent->d_inode;
359
360         /*
361          * 1 - inode item
362          * 2 - refs
363          * 1 - root item
364          * 2 - dir items
365          */
366         trans = btrfs_start_transaction(root, 6);
367         if (IS_ERR(trans))
368                 return PTR_ERR(trans);
369
370         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
371                                       0, objectid, NULL, 0, 0, 0, 0);
372         if (IS_ERR(leaf)) {
373                 ret = PTR_ERR(leaf);
374                 goto fail;
375         }
376
377         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
378         btrfs_set_header_bytenr(leaf, leaf->start);
379         btrfs_set_header_generation(leaf, trans->transid);
380         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
381         btrfs_set_header_owner(leaf, objectid);
382
383         write_extent_buffer(leaf, root->fs_info->fsid,
384                             (unsigned long)btrfs_header_fsid(leaf),
385                             BTRFS_FSID_SIZE);
386         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
387                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
388                             BTRFS_UUID_SIZE);
389         btrfs_mark_buffer_dirty(leaf);
390
391         inode_item = &root_item.inode;
392         memset(inode_item, 0, sizeof(*inode_item));
393         inode_item->generation = cpu_to_le64(1);
394         inode_item->size = cpu_to_le64(3);
395         inode_item->nlink = cpu_to_le32(1);
396         inode_item->nbytes = cpu_to_le64(root->leafsize);
397         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
398
399         root_item.flags = 0;
400         root_item.byte_limit = 0;
401         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
402
403         btrfs_set_root_bytenr(&root_item, leaf->start);
404         btrfs_set_root_generation(&root_item, trans->transid);
405         btrfs_set_root_level(&root_item, 0);
406         btrfs_set_root_refs(&root_item, 1);
407         btrfs_set_root_used(&root_item, leaf->len);
408         btrfs_set_root_last_snapshot(&root_item, 0);
409
410         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
411         root_item.drop_level = 0;
412
413         btrfs_tree_unlock(leaf);
414         free_extent_buffer(leaf);
415         leaf = NULL;
416
417         btrfs_set_root_dirid(&root_item, new_dirid);
418
419         key.objectid = objectid;
420         key.offset = 0;
421         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
422         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
423                                 &root_item);
424         if (ret)
425                 goto fail;
426
427         key.offset = (u64)-1;
428         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
429         BUG_ON(IS_ERR(new_root));
430
431         btrfs_record_root_in_trans(trans, new_root);
432
433         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
434         /*
435          * insert the directory item
436          */
437         ret = btrfs_set_inode_index(dir, &index);
438         BUG_ON(ret);
439
440         ret = btrfs_insert_dir_item(trans, root,
441                                     name, namelen, dir, &key,
442                                     BTRFS_FT_DIR, index);
443         if (ret)
444                 goto fail;
445
446         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
447         ret = btrfs_update_inode(trans, root, dir);
448         BUG_ON(ret);
449
450         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
451                                  objectid, root->root_key.objectid,
452                                  btrfs_ino(dir), index, name, namelen);
453
454         BUG_ON(ret);
455
456         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
457 fail:
458         if (async_transid) {
459                 *async_transid = trans->transid;
460                 err = btrfs_commit_transaction_async(trans, root, 1);
461         } else {
462                 err = btrfs_commit_transaction(trans, root);
463         }
464         if (err && !ret)
465                 ret = err;
466         return ret;
467 }
468
469 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
470                            char *name, int namelen, u64 *async_transid,
471                            bool readonly)
472 {
473         struct inode *inode;
474         struct btrfs_pending_snapshot *pending_snapshot;
475         struct btrfs_trans_handle *trans;
476         int ret;
477
478         if (!root->ref_cows)
479                 return -EINVAL;
480
481         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
482         if (!pending_snapshot)
483                 return -ENOMEM;
484
485         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
486         pending_snapshot->dentry = dentry;
487         pending_snapshot->root = root;
488         pending_snapshot->readonly = readonly;
489
490         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
491         if (IS_ERR(trans)) {
492                 ret = PTR_ERR(trans);
493                 goto fail;
494         }
495
496         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
497         BUG_ON(ret);
498
499         spin_lock(&root->fs_info->trans_lock);
500         list_add(&pending_snapshot->list,
501                  &trans->transaction->pending_snapshots);
502         spin_unlock(&root->fs_info->trans_lock);
503         if (async_transid) {
504                 *async_transid = trans->transid;
505                 ret = btrfs_commit_transaction_async(trans,
506                                      root->fs_info->extent_root, 1);
507         } else {
508                 ret = btrfs_commit_transaction(trans,
509                                                root->fs_info->extent_root);
510         }
511         BUG_ON(ret);
512
513         ret = pending_snapshot->error;
514         if (ret)
515                 goto fail;
516
517         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
518         if (ret)
519                 goto fail;
520
521         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
522         if (IS_ERR(inode)) {
523                 ret = PTR_ERR(inode);
524                 goto fail;
525         }
526         BUG_ON(!inode);
527         d_instantiate(dentry, inode);
528         ret = 0;
529 fail:
530         kfree(pending_snapshot);
531         return ret;
532 }
533
534 /*  copy of check_sticky in fs/namei.c()
535 * It's inline, so penalty for filesystems that don't use sticky bit is
536 * minimal.
537 */
538 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
539 {
540         uid_t fsuid = current_fsuid();
541
542         if (!(dir->i_mode & S_ISVTX))
543                 return 0;
544         if (inode->i_uid == fsuid)
545                 return 0;
546         if (dir->i_uid == fsuid)
547                 return 0;
548         return !capable(CAP_FOWNER);
549 }
550
551 /*  copy of may_delete in fs/namei.c()
552  *      Check whether we can remove a link victim from directory dir, check
553  *  whether the type of victim is right.
554  *  1. We can't do it if dir is read-only (done in permission())
555  *  2. We should have write and exec permissions on dir
556  *  3. We can't remove anything from append-only dir
557  *  4. We can't do anything with immutable dir (done in permission())
558  *  5. If the sticky bit on dir is set we should either
559  *      a. be owner of dir, or
560  *      b. be owner of victim, or
561  *      c. have CAP_FOWNER capability
562  *  6. If the victim is append-only or immutable we can't do antyhing with
563  *     links pointing to it.
564  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
565  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
566  *  9. We can't remove a root or mountpoint.
567  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
568  *     nfs_async_unlink().
569  */
570
571 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
572 {
573         int error;
574
575         if (!victim->d_inode)
576                 return -ENOENT;
577
578         BUG_ON(victim->d_parent->d_inode != dir);
579         audit_inode_child(victim, dir);
580
581         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
582         if (error)
583                 return error;
584         if (IS_APPEND(dir))
585                 return -EPERM;
586         if (btrfs_check_sticky(dir, victim->d_inode)||
587                 IS_APPEND(victim->d_inode)||
588             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
589                 return -EPERM;
590         if (isdir) {
591                 if (!S_ISDIR(victim->d_inode->i_mode))
592                         return -ENOTDIR;
593                 if (IS_ROOT(victim))
594                         return -EBUSY;
595         } else if (S_ISDIR(victim->d_inode->i_mode))
596                 return -EISDIR;
597         if (IS_DEADDIR(dir))
598                 return -ENOENT;
599         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
600                 return -EBUSY;
601         return 0;
602 }
603
604 /* copy of may_create in fs/namei.c() */
605 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
606 {
607         if (child->d_inode)
608                 return -EEXIST;
609         if (IS_DEADDIR(dir))
610                 return -ENOENT;
611         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
612 }
613
614 /*
615  * Create a new subvolume below @parent.  This is largely modeled after
616  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
617  * inside this filesystem so it's quite a bit simpler.
618  */
619 static noinline int btrfs_mksubvol(struct path *parent,
620                                    char *name, int namelen,
621                                    struct btrfs_root *snap_src,
622                                    u64 *async_transid, bool readonly)
623 {
624         struct inode *dir  = parent->dentry->d_inode;
625         struct dentry *dentry;
626         int error;
627
628         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
629
630         dentry = lookup_one_len(name, parent->dentry, namelen);
631         error = PTR_ERR(dentry);
632         if (IS_ERR(dentry))
633                 goto out_unlock;
634
635         error = -EEXIST;
636         if (dentry->d_inode)
637                 goto out_dput;
638
639         error = mnt_want_write(parent->mnt);
640         if (error)
641                 goto out_dput;
642
643         error = btrfs_may_create(dir, dentry);
644         if (error)
645                 goto out_drop_write;
646
647         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
648
649         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
650                 goto out_up_read;
651
652         if (snap_src) {
653                 error = create_snapshot(snap_src, dentry,
654                                         name, namelen, async_transid, readonly);
655         } else {
656                 error = create_subvol(BTRFS_I(dir)->root, dentry,
657                                       name, namelen, async_transid);
658         }
659         if (!error)
660                 fsnotify_mkdir(dir, dentry);
661 out_up_read:
662         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
663 out_drop_write:
664         mnt_drop_write(parent->mnt);
665 out_dput:
666         dput(dentry);
667 out_unlock:
668         mutex_unlock(&dir->i_mutex);
669         return error;
670 }
671
672 /*
673  * When we're defragging a range, we don't want to kick it off again
674  * if it is really just waiting for delalloc to send it down.
675  * If we find a nice big extent or delalloc range for the bytes in the
676  * file you want to defrag, we return 0 to let you know to skip this
677  * part of the file
678  */
679 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
680 {
681         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
682         struct extent_map *em = NULL;
683         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
684         u64 end;
685
686         read_lock(&em_tree->lock);
687         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
688         read_unlock(&em_tree->lock);
689
690         if (em) {
691                 end = extent_map_end(em);
692                 free_extent_map(em);
693                 if (end - offset > thresh)
694                         return 0;
695         }
696         /* if we already have a nice delalloc here, just stop */
697         thresh /= 2;
698         end = count_range_bits(io_tree, &offset, offset + thresh,
699                                thresh, EXTENT_DELALLOC, 1);
700         if (end >= thresh)
701                 return 0;
702         return 1;
703 }
704
705 /*
706  * helper function to walk through a file and find extents
707  * newer than a specific transid, and smaller than thresh.
708  *
709  * This is used by the defragging code to find new and small
710  * extents
711  */
712 static int find_new_extents(struct btrfs_root *root,
713                             struct inode *inode, u64 newer_than,
714                             u64 *off, int thresh)
715 {
716         struct btrfs_path *path;
717         struct btrfs_key min_key;
718         struct btrfs_key max_key;
719         struct extent_buffer *leaf;
720         struct btrfs_file_extent_item *extent;
721         int type;
722         int ret;
723         u64 ino = btrfs_ino(inode);
724
725         path = btrfs_alloc_path();
726         if (!path)
727                 return -ENOMEM;
728
729         min_key.objectid = ino;
730         min_key.type = BTRFS_EXTENT_DATA_KEY;
731         min_key.offset = *off;
732
733         max_key.objectid = ino;
734         max_key.type = (u8)-1;
735         max_key.offset = (u64)-1;
736
737         path->keep_locks = 1;
738
739         while(1) {
740                 ret = btrfs_search_forward(root, &min_key, &max_key,
741                                            path, 0, newer_than);
742                 if (ret != 0)
743                         goto none;
744                 if (min_key.objectid != ino)
745                         goto none;
746                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
747                         goto none;
748
749                 leaf = path->nodes[0];
750                 extent = btrfs_item_ptr(leaf, path->slots[0],
751                                         struct btrfs_file_extent_item);
752
753                 type = btrfs_file_extent_type(leaf, extent);
754                 if (type == BTRFS_FILE_EXTENT_REG &&
755                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
756                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
757                         *off = min_key.offset;
758                         btrfs_free_path(path);
759                         return 0;
760                 }
761
762                 if (min_key.offset == (u64)-1)
763                         goto none;
764
765                 min_key.offset++;
766                 btrfs_release_path(path);
767         }
768 none:
769         btrfs_free_path(path);
770         return -ENOENT;
771 }
772
773 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
774                                int thresh, u64 *last_len, u64 *skip,
775                                u64 *defrag_end)
776 {
777         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
778         struct extent_map *em = NULL;
779         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
780         int ret = 1;
781
782         /*
783          * make sure that once we start defragging an extent, we keep on
784          * defragging it
785          */
786         if (start < *defrag_end)
787                 return 1;
788
789         *skip = 0;
790
791         /*
792          * hopefully we have this extent in the tree already, try without
793          * the full extent lock
794          */
795         read_lock(&em_tree->lock);
796         em = lookup_extent_mapping(em_tree, start, len);
797         read_unlock(&em_tree->lock);
798
799         if (!em) {
800                 /* get the big lock and read metadata off disk */
801                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
802                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
803                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
804
805                 if (IS_ERR(em))
806                         return 0;
807         }
808
809         /* this will cover holes, and inline extents */
810         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
811                 ret = 0;
812
813         /*
814          * we hit a real extent, if it is big don't bother defragging it again
815          */
816         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
817                 ret = 0;
818
819         /*
820          * last_len ends up being a counter of how many bytes we've defragged.
821          * every time we choose not to defrag an extent, we reset *last_len
822          * so that the next tiny extent will force a defrag.
823          *
824          * The end result of this is that tiny extents before a single big
825          * extent will force at least part of that big extent to be defragged.
826          */
827         if (ret) {
828                 *defrag_end = extent_map_end(em);
829         } else {
830                 *last_len = 0;
831                 *skip = extent_map_end(em);
832                 *defrag_end = 0;
833         }
834
835         free_extent_map(em);
836         return ret;
837 }
838
839 /*
840  * it doesn't do much good to defrag one or two pages
841  * at a time.  This pulls in a nice chunk of pages
842  * to COW and defrag.
843  *
844  * It also makes sure the delalloc code has enough
845  * dirty data to avoid making new small extents as part
846  * of the defrag
847  *
848  * It's a good idea to start RA on this range
849  * before calling this.
850  */
851 static int cluster_pages_for_defrag(struct inode *inode,
852                                     struct page **pages,
853                                     unsigned long start_index,
854                                     int num_pages)
855 {
856         unsigned long file_end;
857         u64 isize = i_size_read(inode);
858         u64 page_start;
859         u64 page_end;
860         int ret;
861         int i;
862         int i_done;
863         struct btrfs_ordered_extent *ordered;
864         struct extent_state *cached_state = NULL;
865         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
866
867         if (isize == 0)
868                 return 0;
869         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
870
871         ret = btrfs_delalloc_reserve_space(inode,
872                                            num_pages << PAGE_CACHE_SHIFT);
873         if (ret)
874                 return ret;
875 again:
876         ret = 0;
877         i_done = 0;
878
879         /* step one, lock all the pages */
880         for (i = 0; i < num_pages; i++) {
881                 struct page *page;
882                 page = find_or_create_page(inode->i_mapping,
883                                             start_index + i, mask);
884                 if (!page)
885                         break;
886
887                 if (!PageUptodate(page)) {
888                         btrfs_readpage(NULL, page);
889                         lock_page(page);
890                         if (!PageUptodate(page)) {
891                                 unlock_page(page);
892                                 page_cache_release(page);
893                                 ret = -EIO;
894                                 break;
895                         }
896                 }
897                 isize = i_size_read(inode);
898                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
899                 if (!isize || page->index > file_end ||
900                     page->mapping != inode->i_mapping) {
901                         /* whoops, we blew past eof, skip this page */
902                         unlock_page(page);
903                         page_cache_release(page);
904                         break;
905                 }
906                 pages[i] = page;
907                 i_done++;
908         }
909         if (!i_done || ret)
910                 goto out;
911
912         if (!(inode->i_sb->s_flags & MS_ACTIVE))
913                 goto out;
914
915         /*
916          * so now we have a nice long stream of locked
917          * and up to date pages, lets wait on them
918          */
919         for (i = 0; i < i_done; i++)
920                 wait_on_page_writeback(pages[i]);
921
922         page_start = page_offset(pages[0]);
923         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
924
925         lock_extent_bits(&BTRFS_I(inode)->io_tree,
926                          page_start, page_end - 1, 0, &cached_state,
927                          GFP_NOFS);
928         ordered = btrfs_lookup_first_ordered_extent(inode, page_end - 1);
929         if (ordered &&
930             ordered->file_offset + ordered->len > page_start &&
931             ordered->file_offset < page_end) {
932                 btrfs_put_ordered_extent(ordered);
933                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
934                                      page_start, page_end - 1,
935                                      &cached_state, GFP_NOFS);
936                 for (i = 0; i < i_done; i++) {
937                         unlock_page(pages[i]);
938                         page_cache_release(pages[i]);
939                 }
940                 btrfs_wait_ordered_range(inode, page_start,
941                                          page_end - page_start);
942                 goto again;
943         }
944         if (ordered)
945                 btrfs_put_ordered_extent(ordered);
946
947         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
948                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
949                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
950                           GFP_NOFS);
951
952         if (i_done != num_pages) {
953                 spin_lock(&BTRFS_I(inode)->lock);
954                 BTRFS_I(inode)->outstanding_extents++;
955                 spin_unlock(&BTRFS_I(inode)->lock);
956                 btrfs_delalloc_release_space(inode,
957                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
958         }
959
960
961         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
962                                   &cached_state);
963
964         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
965                              page_start, page_end - 1, &cached_state,
966                              GFP_NOFS);
967
968         for (i = 0; i < i_done; i++) {
969                 clear_page_dirty_for_io(pages[i]);
970                 ClearPageChecked(pages[i]);
971                 set_page_extent_mapped(pages[i]);
972                 set_page_dirty(pages[i]);
973                 unlock_page(pages[i]);
974                 page_cache_release(pages[i]);
975         }
976         return i_done;
977 out:
978         for (i = 0; i < i_done; i++) {
979                 unlock_page(pages[i]);
980                 page_cache_release(pages[i]);
981         }
982         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
983         return ret;
984
985 }
986
987 int btrfs_defrag_file(struct inode *inode, struct file *file,
988                       struct btrfs_ioctl_defrag_range_args *range,
989                       u64 newer_than, unsigned long max_to_defrag)
990 {
991         struct btrfs_root *root = BTRFS_I(inode)->root;
992         struct btrfs_super_block *disk_super;
993         struct file_ra_state *ra = NULL;
994         unsigned long last_index;
995         u64 isize = i_size_read(inode);
996         u64 features;
997         u64 last_len = 0;
998         u64 skip = 0;
999         u64 defrag_end = 0;
1000         u64 newer_off = range->start;
1001         unsigned long i;
1002         unsigned long ra_index = 0;
1003         int ret;
1004         int defrag_count = 0;
1005         int compress_type = BTRFS_COMPRESS_ZLIB;
1006         int extent_thresh = range->extent_thresh;
1007         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1008         int cluster = max_cluster;
1009         u64 new_align = ~((u64)128 * 1024 - 1);
1010         struct page **pages = NULL;
1011
1012         if (extent_thresh == 0)
1013                 extent_thresh = 256 * 1024;
1014
1015         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1016                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1017                         return -EINVAL;
1018                 if (range->compress_type)
1019                         compress_type = range->compress_type;
1020         }
1021
1022         if (isize == 0)
1023                 return 0;
1024
1025         /*
1026          * if we were not given a file, allocate a readahead
1027          * context
1028          */
1029         if (!file) {
1030                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1031                 if (!ra)
1032                         return -ENOMEM;
1033                 file_ra_state_init(ra, inode->i_mapping);
1034         } else {
1035                 ra = &file->f_ra;
1036         }
1037
1038         pages = kmalloc(sizeof(struct page *) * max_cluster,
1039                         GFP_NOFS);
1040         if (!pages) {
1041                 ret = -ENOMEM;
1042                 goto out_ra;
1043         }
1044
1045         /* find the last page to defrag */
1046         if (range->start + range->len > range->start) {
1047                 last_index = min_t(u64, isize - 1,
1048                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1049         } else {
1050                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1051         }
1052
1053         if (newer_than) {
1054                 ret = find_new_extents(root, inode, newer_than,
1055                                        &newer_off, 64 * 1024);
1056                 if (!ret) {
1057                         range->start = newer_off;
1058                         /*
1059                          * we always align our defrag to help keep
1060                          * the extents in the file evenly spaced
1061                          */
1062                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1063                 } else
1064                         goto out_ra;
1065         } else {
1066                 i = range->start >> PAGE_CACHE_SHIFT;
1067         }
1068         if (!max_to_defrag)
1069                 max_to_defrag = last_index;
1070
1071         /*
1072          * make writeback starts from i, so the defrag range can be
1073          * written sequentially.
1074          */
1075         if (i < inode->i_mapping->writeback_index)
1076                 inode->i_mapping->writeback_index = i;
1077
1078         while (i <= last_index && defrag_count < max_to_defrag &&
1079                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1080                 PAGE_CACHE_SHIFT)) {
1081                 /*
1082                  * make sure we stop running if someone unmounts
1083                  * the FS
1084                  */
1085                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1086                         break;
1087
1088                 if (!newer_than &&
1089                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1090                                         PAGE_CACHE_SIZE,
1091                                         extent_thresh,
1092                                         &last_len, &skip,
1093                                         &defrag_end)) {
1094                         unsigned long next;
1095                         /*
1096                          * the should_defrag function tells us how much to skip
1097                          * bump our counter by the suggested amount
1098                          */
1099                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1100                         i = max(i + 1, next);
1101                         continue;
1102                 }
1103
1104                 if (!newer_than) {
1105                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1106                                    PAGE_CACHE_SHIFT) - i;
1107                         cluster = min(cluster, max_cluster);
1108                 } else {
1109                         cluster = max_cluster;
1110                 }
1111
1112                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1113                         BTRFS_I(inode)->force_compress = compress_type;
1114
1115                 if (i + cluster > ra_index) {
1116                         ra_index = max(i, ra_index);
1117                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1118                                        cluster);
1119                         ra_index += max_cluster;
1120                 }
1121
1122                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1123                 if (ret < 0)
1124                         goto out_ra;
1125
1126                 defrag_count += ret;
1127                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1128
1129                 if (newer_than) {
1130                         if (newer_off == (u64)-1)
1131                                 break;
1132
1133                         newer_off = max(newer_off + 1,
1134                                         (u64)i << PAGE_CACHE_SHIFT);
1135
1136                         ret = find_new_extents(root, inode,
1137                                                newer_than, &newer_off,
1138                                                64 * 1024);
1139                         if (!ret) {
1140                                 range->start = newer_off;
1141                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1142                         } else {
1143                                 break;
1144                         }
1145                 } else {
1146                         if (ret > 0) {
1147                                 i += ret;
1148                                 last_len += ret << PAGE_CACHE_SHIFT;
1149                         } else {
1150                                 i++;
1151                                 last_len = 0;
1152                         }
1153                 }
1154         }
1155
1156         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1157                 filemap_flush(inode->i_mapping);
1158
1159         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1160                 /* the filemap_flush will queue IO into the worker threads, but
1161                  * we have to make sure the IO is actually started and that
1162                  * ordered extents get created before we return
1163                  */
1164                 atomic_inc(&root->fs_info->async_submit_draining);
1165                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1166                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1167                         wait_event(root->fs_info->async_submit_wait,
1168                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1169                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1170                 }
1171                 atomic_dec(&root->fs_info->async_submit_draining);
1172
1173                 mutex_lock(&inode->i_mutex);
1174                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1175                 mutex_unlock(&inode->i_mutex);
1176         }
1177
1178         disk_super = root->fs_info->super_copy;
1179         features = btrfs_super_incompat_flags(disk_super);
1180         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1181                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1182                 btrfs_set_super_incompat_flags(disk_super, features);
1183         }
1184
1185         ret = defrag_count;
1186
1187 out_ra:
1188         if (!file)
1189                 kfree(ra);
1190         kfree(pages);
1191         return ret;
1192 }
1193
1194 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1195                                         void __user *arg)
1196 {
1197         u64 new_size;
1198         u64 old_size;
1199         u64 devid = 1;
1200         struct btrfs_ioctl_vol_args *vol_args;
1201         struct btrfs_trans_handle *trans;
1202         struct btrfs_device *device = NULL;
1203         char *sizestr;
1204         char *devstr = NULL;
1205         int ret = 0;
1206         int mod = 0;
1207
1208         if (root->fs_info->sb->s_flags & MS_RDONLY)
1209                 return -EROFS;
1210
1211         if (!capable(CAP_SYS_ADMIN))
1212                 return -EPERM;
1213
1214         mutex_lock(&root->fs_info->volume_mutex);
1215         if (root->fs_info->balance_ctl) {
1216                 printk(KERN_INFO "btrfs: balance in progress\n");
1217                 ret = -EINVAL;
1218                 goto out;
1219         }
1220
1221         vol_args = memdup_user(arg, sizeof(*vol_args));
1222         if (IS_ERR(vol_args)) {
1223                 ret = PTR_ERR(vol_args);
1224                 goto out;
1225         }
1226
1227         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1228
1229         sizestr = vol_args->name;
1230         devstr = strchr(sizestr, ':');
1231         if (devstr) {
1232                 char *end;
1233                 sizestr = devstr + 1;
1234                 *devstr = '\0';
1235                 devstr = vol_args->name;
1236                 devid = simple_strtoull(devstr, &end, 10);
1237                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1238                        (unsigned long long)devid);
1239         }
1240         device = btrfs_find_device(root, devid, NULL, NULL);
1241         if (!device) {
1242                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1243                        (unsigned long long)devid);
1244                 ret = -EINVAL;
1245                 goto out_free;
1246         }
1247         if (!strcmp(sizestr, "max"))
1248                 new_size = device->bdev->bd_inode->i_size;
1249         else {
1250                 if (sizestr[0] == '-') {
1251                         mod = -1;
1252                         sizestr++;
1253                 } else if (sizestr[0] == '+') {
1254                         mod = 1;
1255                         sizestr++;
1256                 }
1257                 new_size = memparse(sizestr, NULL);
1258                 if (new_size == 0) {
1259                         ret = -EINVAL;
1260                         goto out_free;
1261                 }
1262         }
1263
1264         old_size = device->total_bytes;
1265
1266         if (mod < 0) {
1267                 if (new_size > old_size) {
1268                         ret = -EINVAL;
1269                         goto out_free;
1270                 }
1271                 new_size = old_size - new_size;
1272         } else if (mod > 0) {
1273                 new_size = old_size + new_size;
1274         }
1275
1276         if (new_size < 256 * 1024 * 1024) {
1277                 ret = -EINVAL;
1278                 goto out_free;
1279         }
1280         if (new_size > device->bdev->bd_inode->i_size) {
1281                 ret = -EFBIG;
1282                 goto out_free;
1283         }
1284
1285         do_div(new_size, root->sectorsize);
1286         new_size *= root->sectorsize;
1287
1288         printk(KERN_INFO "btrfs: new size for %s is %llu\n",
1289                 device->name, (unsigned long long)new_size);
1290
1291         if (new_size > old_size) {
1292                 trans = btrfs_start_transaction(root, 0);
1293                 if (IS_ERR(trans)) {
1294                         ret = PTR_ERR(trans);
1295                         goto out_free;
1296                 }
1297                 ret = btrfs_grow_device(trans, device, new_size);
1298                 btrfs_commit_transaction(trans, root);
1299         } else if (new_size < old_size) {
1300                 ret = btrfs_shrink_device(device, new_size);
1301         }
1302
1303 out_free:
1304         kfree(vol_args);
1305 out:
1306         mutex_unlock(&root->fs_info->volume_mutex);
1307         return ret;
1308 }
1309
1310 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1311                                                     char *name,
1312                                                     unsigned long fd,
1313                                                     int subvol,
1314                                                     u64 *transid,
1315                                                     bool readonly)
1316 {
1317         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1318         struct file *src_file;
1319         int namelen;
1320         int ret = 0;
1321
1322         if (root->fs_info->sb->s_flags & MS_RDONLY)
1323                 return -EROFS;
1324
1325         namelen = strlen(name);
1326         if (strchr(name, '/')) {
1327                 ret = -EINVAL;
1328                 goto out;
1329         }
1330
1331         if (subvol) {
1332                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1333                                      NULL, transid, readonly);
1334         } else {
1335                 struct inode *src_inode;
1336                 src_file = fget(fd);
1337                 if (!src_file) {
1338                         ret = -EINVAL;
1339                         goto out;
1340                 }
1341
1342                 src_inode = src_file->f_path.dentry->d_inode;
1343                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1344                         printk(KERN_INFO "btrfs: Snapshot src from "
1345                                "another FS\n");
1346                         ret = -EINVAL;
1347                         fput(src_file);
1348                         goto out;
1349                 }
1350                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1351                                      BTRFS_I(src_inode)->root,
1352                                      transid, readonly);
1353                 fput(src_file);
1354         }
1355 out:
1356         return ret;
1357 }
1358
1359 static noinline int btrfs_ioctl_snap_create(struct file *file,
1360                                             void __user *arg, int subvol)
1361 {
1362         struct btrfs_ioctl_vol_args *vol_args;
1363         int ret;
1364
1365         vol_args = memdup_user(arg, sizeof(*vol_args));
1366         if (IS_ERR(vol_args))
1367                 return PTR_ERR(vol_args);
1368         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1369
1370         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1371                                               vol_args->fd, subvol,
1372                                               NULL, false);
1373
1374         kfree(vol_args);
1375         return ret;
1376 }
1377
1378 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1379                                                void __user *arg, int subvol)
1380 {
1381         struct btrfs_ioctl_vol_args_v2 *vol_args;
1382         int ret;
1383         u64 transid = 0;
1384         u64 *ptr = NULL;
1385         bool readonly = false;
1386
1387         vol_args = memdup_user(arg, sizeof(*vol_args));
1388         if (IS_ERR(vol_args))
1389                 return PTR_ERR(vol_args);
1390         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1391
1392         if (vol_args->flags &
1393             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1394                 ret = -EOPNOTSUPP;
1395                 goto out;
1396         }
1397
1398         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1399                 ptr = &transid;
1400         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1401                 readonly = true;
1402
1403         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1404                                               vol_args->fd, subvol,
1405                                               ptr, readonly);
1406
1407         if (ret == 0 && ptr &&
1408             copy_to_user(arg +
1409                          offsetof(struct btrfs_ioctl_vol_args_v2,
1410                                   transid), ptr, sizeof(*ptr)))
1411                 ret = -EFAULT;
1412 out:
1413         kfree(vol_args);
1414         return ret;
1415 }
1416
1417 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1418                                                 void __user *arg)
1419 {
1420         struct inode *inode = fdentry(file)->d_inode;
1421         struct btrfs_root *root = BTRFS_I(inode)->root;
1422         int ret = 0;
1423         u64 flags = 0;
1424
1425         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1426                 return -EINVAL;
1427
1428         down_read(&root->fs_info->subvol_sem);
1429         if (btrfs_root_readonly(root))
1430                 flags |= BTRFS_SUBVOL_RDONLY;
1431         up_read(&root->fs_info->subvol_sem);
1432
1433         if (copy_to_user(arg, &flags, sizeof(flags)))
1434                 ret = -EFAULT;
1435
1436         return ret;
1437 }
1438
1439 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1440                                               void __user *arg)
1441 {
1442         struct inode *inode = fdentry(file)->d_inode;
1443         struct btrfs_root *root = BTRFS_I(inode)->root;
1444         struct btrfs_trans_handle *trans;
1445         u64 root_flags;
1446         u64 flags;
1447         int ret = 0;
1448
1449         if (root->fs_info->sb->s_flags & MS_RDONLY)
1450                 return -EROFS;
1451
1452         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1453                 return -EINVAL;
1454
1455         if (copy_from_user(&flags, arg, sizeof(flags)))
1456                 return -EFAULT;
1457
1458         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1459                 return -EINVAL;
1460
1461         if (flags & ~BTRFS_SUBVOL_RDONLY)
1462                 return -EOPNOTSUPP;
1463
1464         if (!inode_owner_or_capable(inode))
1465                 return -EACCES;
1466
1467         down_write(&root->fs_info->subvol_sem);
1468
1469         /* nothing to do */
1470         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1471                 goto out;
1472
1473         root_flags = btrfs_root_flags(&root->root_item);
1474         if (flags & BTRFS_SUBVOL_RDONLY)
1475                 btrfs_set_root_flags(&root->root_item,
1476                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1477         else
1478                 btrfs_set_root_flags(&root->root_item,
1479                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1480
1481         trans = btrfs_start_transaction(root, 1);
1482         if (IS_ERR(trans)) {
1483                 ret = PTR_ERR(trans);
1484                 goto out_reset;
1485         }
1486
1487         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1488                                 &root->root_key, &root->root_item);
1489
1490         btrfs_commit_transaction(trans, root);
1491 out_reset:
1492         if (ret)
1493                 btrfs_set_root_flags(&root->root_item, root_flags);
1494 out:
1495         up_write(&root->fs_info->subvol_sem);
1496         return ret;
1497 }
1498
1499 /*
1500  * helper to check if the subvolume references other subvolumes
1501  */
1502 static noinline int may_destroy_subvol(struct btrfs_root *root)
1503 {
1504         struct btrfs_path *path;
1505         struct btrfs_key key;
1506         int ret;
1507
1508         path = btrfs_alloc_path();
1509         if (!path)
1510                 return -ENOMEM;
1511
1512         key.objectid = root->root_key.objectid;
1513         key.type = BTRFS_ROOT_REF_KEY;
1514         key.offset = (u64)-1;
1515
1516         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1517                                 &key, path, 0, 0);
1518         if (ret < 0)
1519                 goto out;
1520         BUG_ON(ret == 0);
1521
1522         ret = 0;
1523         if (path->slots[0] > 0) {
1524                 path->slots[0]--;
1525                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1526                 if (key.objectid == root->root_key.objectid &&
1527                     key.type == BTRFS_ROOT_REF_KEY)
1528                         ret = -ENOTEMPTY;
1529         }
1530 out:
1531         btrfs_free_path(path);
1532         return ret;
1533 }
1534
1535 static noinline int key_in_sk(struct btrfs_key *key,
1536                               struct btrfs_ioctl_search_key *sk)
1537 {
1538         struct btrfs_key test;
1539         int ret;
1540
1541         test.objectid = sk->min_objectid;
1542         test.type = sk->min_type;
1543         test.offset = sk->min_offset;
1544
1545         ret = btrfs_comp_cpu_keys(key, &test);
1546         if (ret < 0)
1547                 return 0;
1548
1549         test.objectid = sk->max_objectid;
1550         test.type = sk->max_type;
1551         test.offset = sk->max_offset;
1552
1553         ret = btrfs_comp_cpu_keys(key, &test);
1554         if (ret > 0)
1555                 return 0;
1556         return 1;
1557 }
1558
1559 static noinline int copy_to_sk(struct btrfs_root *root,
1560                                struct btrfs_path *path,
1561                                struct btrfs_key *key,
1562                                struct btrfs_ioctl_search_key *sk,
1563                                char *buf,
1564                                unsigned long *sk_offset,
1565                                int *num_found)
1566 {
1567         u64 found_transid;
1568         struct extent_buffer *leaf;
1569         struct btrfs_ioctl_search_header sh;
1570         unsigned long item_off;
1571         unsigned long item_len;
1572         int nritems;
1573         int i;
1574         int slot;
1575         int ret = 0;
1576
1577         leaf = path->nodes[0];
1578         slot = path->slots[0];
1579         nritems = btrfs_header_nritems(leaf);
1580
1581         if (btrfs_header_generation(leaf) > sk->max_transid) {
1582                 i = nritems;
1583                 goto advance_key;
1584         }
1585         found_transid = btrfs_header_generation(leaf);
1586
1587         for (i = slot; i < nritems; i++) {
1588                 item_off = btrfs_item_ptr_offset(leaf, i);
1589                 item_len = btrfs_item_size_nr(leaf, i);
1590
1591                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1592                         item_len = 0;
1593
1594                 if (sizeof(sh) + item_len + *sk_offset >
1595                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1596                         ret = 1;
1597                         goto overflow;
1598                 }
1599
1600                 btrfs_item_key_to_cpu(leaf, key, i);
1601                 if (!key_in_sk(key, sk))
1602                         continue;
1603
1604                 sh.objectid = key->objectid;
1605                 sh.offset = key->offset;
1606                 sh.type = key->type;
1607                 sh.len = item_len;
1608                 sh.transid = found_transid;
1609
1610                 /* copy search result header */
1611                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1612                 *sk_offset += sizeof(sh);
1613
1614                 if (item_len) {
1615                         char *p = buf + *sk_offset;
1616                         /* copy the item */
1617                         read_extent_buffer(leaf, p,
1618                                            item_off, item_len);
1619                         *sk_offset += item_len;
1620                 }
1621                 (*num_found)++;
1622
1623                 if (*num_found >= sk->nr_items)
1624                         break;
1625         }
1626 advance_key:
1627         ret = 0;
1628         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1629                 key->offset++;
1630         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1631                 key->offset = 0;
1632                 key->type++;
1633         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1634                 key->offset = 0;
1635                 key->type = 0;
1636                 key->objectid++;
1637         } else
1638                 ret = 1;
1639 overflow:
1640         return ret;
1641 }
1642
1643 static noinline int search_ioctl(struct inode *inode,
1644                                  struct btrfs_ioctl_search_args *args)
1645 {
1646         struct btrfs_root *root;
1647         struct btrfs_key key;
1648         struct btrfs_key max_key;
1649         struct btrfs_path *path;
1650         struct btrfs_ioctl_search_key *sk = &args->key;
1651         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1652         int ret;
1653         int num_found = 0;
1654         unsigned long sk_offset = 0;
1655
1656         path = btrfs_alloc_path();
1657         if (!path)
1658                 return -ENOMEM;
1659
1660         if (sk->tree_id == 0) {
1661                 /* search the root of the inode that was passed */
1662                 root = BTRFS_I(inode)->root;
1663         } else {
1664                 key.objectid = sk->tree_id;
1665                 key.type = BTRFS_ROOT_ITEM_KEY;
1666                 key.offset = (u64)-1;
1667                 root = btrfs_read_fs_root_no_name(info, &key);
1668                 if (IS_ERR(root)) {
1669                         printk(KERN_ERR "could not find root %llu\n",
1670                                sk->tree_id);
1671                         btrfs_free_path(path);
1672                         return -ENOENT;
1673                 }
1674         }
1675
1676         key.objectid = sk->min_objectid;
1677         key.type = sk->min_type;
1678         key.offset = sk->min_offset;
1679
1680         max_key.objectid = sk->max_objectid;
1681         max_key.type = sk->max_type;
1682         max_key.offset = sk->max_offset;
1683
1684         path->keep_locks = 1;
1685
1686         while(1) {
1687                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1688                                            sk->min_transid);
1689                 if (ret != 0) {
1690                         if (ret > 0)
1691                                 ret = 0;
1692                         goto err;
1693                 }
1694                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1695                                  &sk_offset, &num_found);
1696                 btrfs_release_path(path);
1697                 if (ret || num_found >= sk->nr_items)
1698                         break;
1699
1700         }
1701         ret = 0;
1702 err:
1703         sk->nr_items = num_found;
1704         btrfs_free_path(path);
1705         return ret;
1706 }
1707
1708 static noinline int btrfs_ioctl_tree_search(struct file *file,
1709                                            void __user *argp)
1710 {
1711          struct btrfs_ioctl_search_args *args;
1712          struct inode *inode;
1713          int ret;
1714
1715         if (!capable(CAP_SYS_ADMIN))
1716                 return -EPERM;
1717
1718         args = memdup_user(argp, sizeof(*args));
1719         if (IS_ERR(args))
1720                 return PTR_ERR(args);
1721
1722         inode = fdentry(file)->d_inode;
1723         ret = search_ioctl(inode, args);
1724         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1725                 ret = -EFAULT;
1726         kfree(args);
1727         return ret;
1728 }
1729
1730 /*
1731  * Search INODE_REFs to identify path name of 'dirid' directory
1732  * in a 'tree_id' tree. and sets path name to 'name'.
1733  */
1734 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1735                                 u64 tree_id, u64 dirid, char *name)
1736 {
1737         struct btrfs_root *root;
1738         struct btrfs_key key;
1739         char *ptr;
1740         int ret = -1;
1741         int slot;
1742         int len;
1743         int total_len = 0;
1744         struct btrfs_inode_ref *iref;
1745         struct extent_buffer *l;
1746         struct btrfs_path *path;
1747
1748         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1749                 name[0]='\0';
1750                 return 0;
1751         }
1752
1753         path = btrfs_alloc_path();
1754         if (!path)
1755                 return -ENOMEM;
1756
1757         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1758
1759         key.objectid = tree_id;
1760         key.type = BTRFS_ROOT_ITEM_KEY;
1761         key.offset = (u64)-1;
1762         root = btrfs_read_fs_root_no_name(info, &key);
1763         if (IS_ERR(root)) {
1764                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1765                 ret = -ENOENT;
1766                 goto out;
1767         }
1768
1769         key.objectid = dirid;
1770         key.type = BTRFS_INODE_REF_KEY;
1771         key.offset = (u64)-1;
1772
1773         while(1) {
1774                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1775                 if (ret < 0)
1776                         goto out;
1777
1778                 l = path->nodes[0];
1779                 slot = path->slots[0];
1780                 if (ret > 0 && slot > 0)
1781                         slot--;
1782                 btrfs_item_key_to_cpu(l, &key, slot);
1783
1784                 if (ret > 0 && (key.objectid != dirid ||
1785                                 key.type != BTRFS_INODE_REF_KEY)) {
1786                         ret = -ENOENT;
1787                         goto out;
1788                 }
1789
1790                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1791                 len = btrfs_inode_ref_name_len(l, iref);
1792                 ptr -= len + 1;
1793                 total_len += len + 1;
1794                 if (ptr < name)
1795                         goto out;
1796
1797                 *(ptr + len) = '/';
1798                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1799
1800                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1801                         break;
1802
1803                 btrfs_release_path(path);
1804                 key.objectid = key.offset;
1805                 key.offset = (u64)-1;
1806                 dirid = key.objectid;
1807         }
1808         if (ptr < name)
1809                 goto out;
1810         memmove(name, ptr, total_len);
1811         name[total_len]='\0';
1812         ret = 0;
1813 out:
1814         btrfs_free_path(path);
1815         return ret;
1816 }
1817
1818 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1819                                            void __user *argp)
1820 {
1821          struct btrfs_ioctl_ino_lookup_args *args;
1822          struct inode *inode;
1823          int ret;
1824
1825         if (!capable(CAP_SYS_ADMIN))
1826                 return -EPERM;
1827
1828         args = memdup_user(argp, sizeof(*args));
1829         if (IS_ERR(args))
1830                 return PTR_ERR(args);
1831
1832         inode = fdentry(file)->d_inode;
1833
1834         if (args->treeid == 0)
1835                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1836
1837         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1838                                         args->treeid, args->objectid,
1839                                         args->name);
1840
1841         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1842                 ret = -EFAULT;
1843
1844         kfree(args);
1845         return ret;
1846 }
1847
1848 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1849                                              void __user *arg)
1850 {
1851         struct dentry *parent = fdentry(file);
1852         struct dentry *dentry;
1853         struct inode *dir = parent->d_inode;
1854         struct inode *inode;
1855         struct btrfs_root *root = BTRFS_I(dir)->root;
1856         struct btrfs_root *dest = NULL;
1857         struct btrfs_ioctl_vol_args *vol_args;
1858         struct btrfs_trans_handle *trans;
1859         int namelen;
1860         int ret;
1861         int err = 0;
1862
1863         vol_args = memdup_user(arg, sizeof(*vol_args));
1864         if (IS_ERR(vol_args))
1865                 return PTR_ERR(vol_args);
1866
1867         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1868         namelen = strlen(vol_args->name);
1869         if (strchr(vol_args->name, '/') ||
1870             strncmp(vol_args->name, "..", namelen) == 0) {
1871                 err = -EINVAL;
1872                 goto out;
1873         }
1874
1875         err = mnt_want_write_file(file);
1876         if (err)
1877                 goto out;
1878
1879         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1880         dentry = lookup_one_len(vol_args->name, parent, namelen);
1881         if (IS_ERR(dentry)) {
1882                 err = PTR_ERR(dentry);
1883                 goto out_unlock_dir;
1884         }
1885
1886         if (!dentry->d_inode) {
1887                 err = -ENOENT;
1888                 goto out_dput;
1889         }
1890
1891         inode = dentry->d_inode;
1892         dest = BTRFS_I(inode)->root;
1893         if (!capable(CAP_SYS_ADMIN)){
1894                 /*
1895                  * Regular user.  Only allow this with a special mount
1896                  * option, when the user has write+exec access to the
1897                  * subvol root, and when rmdir(2) would have been
1898                  * allowed.
1899                  *
1900                  * Note that this is _not_ check that the subvol is
1901                  * empty or doesn't contain data that we wouldn't
1902                  * otherwise be able to delete.
1903                  *
1904                  * Users who want to delete empty subvols should try
1905                  * rmdir(2).
1906                  */
1907                 err = -EPERM;
1908                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1909                         goto out_dput;
1910
1911                 /*
1912                  * Do not allow deletion if the parent dir is the same
1913                  * as the dir to be deleted.  That means the ioctl
1914                  * must be called on the dentry referencing the root
1915                  * of the subvol, not a random directory contained
1916                  * within it.
1917                  */
1918                 err = -EINVAL;
1919                 if (root == dest)
1920                         goto out_dput;
1921
1922                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1923                 if (err)
1924                         goto out_dput;
1925
1926                 /* check if subvolume may be deleted by a non-root user */
1927                 err = btrfs_may_delete(dir, dentry, 1);
1928                 if (err)
1929                         goto out_dput;
1930         }
1931
1932         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1933                 err = -EINVAL;
1934                 goto out_dput;
1935         }
1936
1937         mutex_lock(&inode->i_mutex);
1938         err = d_invalidate(dentry);
1939         if (err)
1940                 goto out_unlock;
1941
1942         down_write(&root->fs_info->subvol_sem);
1943
1944         err = may_destroy_subvol(dest);
1945         if (err)
1946                 goto out_up_write;
1947
1948         trans = btrfs_start_transaction(root, 0);
1949         if (IS_ERR(trans)) {
1950                 err = PTR_ERR(trans);
1951                 goto out_up_write;
1952         }
1953         trans->block_rsv = &root->fs_info->global_block_rsv;
1954
1955         ret = btrfs_unlink_subvol(trans, root, dir,
1956                                 dest->root_key.objectid,
1957                                 dentry->d_name.name,
1958                                 dentry->d_name.len);
1959         BUG_ON(ret);
1960
1961         btrfs_record_root_in_trans(trans, dest);
1962
1963         memset(&dest->root_item.drop_progress, 0,
1964                 sizeof(dest->root_item.drop_progress));
1965         dest->root_item.drop_level = 0;
1966         btrfs_set_root_refs(&dest->root_item, 0);
1967
1968         if (!xchg(&dest->orphan_item_inserted, 1)) {
1969                 ret = btrfs_insert_orphan_item(trans,
1970                                         root->fs_info->tree_root,
1971                                         dest->root_key.objectid);
1972                 BUG_ON(ret);
1973         }
1974
1975         ret = btrfs_end_transaction(trans, root);
1976         BUG_ON(ret);
1977         inode->i_flags |= S_DEAD;
1978 out_up_write:
1979         up_write(&root->fs_info->subvol_sem);
1980 out_unlock:
1981         mutex_unlock(&inode->i_mutex);
1982         if (!err) {
1983                 shrink_dcache_sb(root->fs_info->sb);
1984                 btrfs_invalidate_inodes(dest);
1985                 d_delete(dentry);
1986         }
1987 out_dput:
1988         dput(dentry);
1989 out_unlock_dir:
1990         mutex_unlock(&dir->i_mutex);
1991         mnt_drop_write_file(file);
1992 out:
1993         kfree(vol_args);
1994         return err;
1995 }
1996
1997 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1998 {
1999         struct inode *inode = fdentry(file)->d_inode;
2000         struct btrfs_root *root = BTRFS_I(inode)->root;
2001         struct btrfs_ioctl_defrag_range_args *range;
2002         int ret;
2003
2004         if (btrfs_root_readonly(root))
2005                 return -EROFS;
2006
2007         ret = mnt_want_write_file(file);
2008         if (ret)
2009                 return ret;
2010
2011         switch (inode->i_mode & S_IFMT) {
2012         case S_IFDIR:
2013                 if (!capable(CAP_SYS_ADMIN)) {
2014                         ret = -EPERM;
2015                         goto out;
2016                 }
2017                 ret = btrfs_defrag_root(root, 0);
2018                 if (ret)
2019                         goto out;
2020                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2021                 break;
2022         case S_IFREG:
2023                 if (!(file->f_mode & FMODE_WRITE)) {
2024                         ret = -EINVAL;
2025                         goto out;
2026                 }
2027
2028                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2029                 if (!range) {
2030                         ret = -ENOMEM;
2031                         goto out;
2032                 }
2033
2034                 if (argp) {
2035                         if (copy_from_user(range, argp,
2036                                            sizeof(*range))) {
2037                                 ret = -EFAULT;
2038                                 kfree(range);
2039                                 goto out;
2040                         }
2041                         /* compression requires us to start the IO */
2042                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2043                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2044                                 range->extent_thresh = (u32)-1;
2045                         }
2046                 } else {
2047                         /* the rest are all set to zero by kzalloc */
2048                         range->len = (u64)-1;
2049                 }
2050                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2051                                         range, 0, 0);
2052                 if (ret > 0)
2053                         ret = 0;
2054                 kfree(range);
2055                 break;
2056         default:
2057                 ret = -EINVAL;
2058         }
2059 out:
2060         mnt_drop_write_file(file);
2061         return ret;
2062 }
2063
2064 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2065 {
2066         struct btrfs_ioctl_vol_args *vol_args;
2067         int ret;
2068
2069         if (!capable(CAP_SYS_ADMIN))
2070                 return -EPERM;
2071
2072         mutex_lock(&root->fs_info->volume_mutex);
2073         if (root->fs_info->balance_ctl) {
2074                 printk(KERN_INFO "btrfs: balance in progress\n");
2075                 ret = -EINVAL;
2076                 goto out;
2077         }
2078
2079         vol_args = memdup_user(arg, sizeof(*vol_args));
2080         if (IS_ERR(vol_args)) {
2081                 ret = PTR_ERR(vol_args);
2082                 goto out;
2083         }
2084
2085         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2086         ret = btrfs_init_new_device(root, vol_args->name);
2087
2088         kfree(vol_args);
2089 out:
2090         mutex_unlock(&root->fs_info->volume_mutex);
2091         return ret;
2092 }
2093
2094 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2095 {
2096         struct btrfs_ioctl_vol_args *vol_args;
2097         int ret;
2098
2099         if (!capable(CAP_SYS_ADMIN))
2100                 return -EPERM;
2101
2102         if (root->fs_info->sb->s_flags & MS_RDONLY)
2103                 return -EROFS;
2104
2105         mutex_lock(&root->fs_info->volume_mutex);
2106         if (root->fs_info->balance_ctl) {
2107                 printk(KERN_INFO "btrfs: balance in progress\n");
2108                 ret = -EINVAL;
2109                 goto out;
2110         }
2111
2112         vol_args = memdup_user(arg, sizeof(*vol_args));
2113         if (IS_ERR(vol_args)) {
2114                 ret = PTR_ERR(vol_args);
2115                 goto out;
2116         }
2117
2118         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2119         ret = btrfs_rm_device(root, vol_args->name);
2120
2121         kfree(vol_args);
2122 out:
2123         mutex_unlock(&root->fs_info->volume_mutex);
2124         return ret;
2125 }
2126
2127 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2128 {
2129         struct btrfs_ioctl_fs_info_args *fi_args;
2130         struct btrfs_device *device;
2131         struct btrfs_device *next;
2132         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2133         int ret = 0;
2134
2135         if (!capable(CAP_SYS_ADMIN))
2136                 return -EPERM;
2137
2138         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2139         if (!fi_args)
2140                 return -ENOMEM;
2141
2142         fi_args->num_devices = fs_devices->num_devices;
2143         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2144
2145         mutex_lock(&fs_devices->device_list_mutex);
2146         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2147                 if (device->devid > fi_args->max_id)
2148                         fi_args->max_id = device->devid;
2149         }
2150         mutex_unlock(&fs_devices->device_list_mutex);
2151
2152         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2153                 ret = -EFAULT;
2154
2155         kfree(fi_args);
2156         return ret;
2157 }
2158
2159 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2160 {
2161         struct btrfs_ioctl_dev_info_args *di_args;
2162         struct btrfs_device *dev;
2163         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2164         int ret = 0;
2165         char *s_uuid = NULL;
2166         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2167
2168         if (!capable(CAP_SYS_ADMIN))
2169                 return -EPERM;
2170
2171         di_args = memdup_user(arg, sizeof(*di_args));
2172         if (IS_ERR(di_args))
2173                 return PTR_ERR(di_args);
2174
2175         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2176                 s_uuid = di_args->uuid;
2177
2178         mutex_lock(&fs_devices->device_list_mutex);
2179         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2180         mutex_unlock(&fs_devices->device_list_mutex);
2181
2182         if (!dev) {
2183                 ret = -ENODEV;
2184                 goto out;
2185         }
2186
2187         di_args->devid = dev->devid;
2188         di_args->bytes_used = dev->bytes_used;
2189         di_args->total_bytes = dev->total_bytes;
2190         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2191         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2192
2193 out:
2194         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2195                 ret = -EFAULT;
2196
2197         kfree(di_args);
2198         return ret;
2199 }
2200
2201 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2202                                        u64 off, u64 olen, u64 destoff)
2203 {
2204         struct inode *inode = fdentry(file)->d_inode;
2205         struct btrfs_root *root = BTRFS_I(inode)->root;
2206         struct file *src_file;
2207         struct inode *src;
2208         struct btrfs_trans_handle *trans;
2209         struct btrfs_path *path;
2210         struct extent_buffer *leaf;
2211         char *buf;
2212         struct btrfs_key key;
2213         u32 nritems;
2214         int slot;
2215         int ret;
2216         u64 len = olen;
2217         u64 bs = root->fs_info->sb->s_blocksize;
2218         u64 hint_byte;
2219
2220         /*
2221          * TODO:
2222          * - split compressed inline extents.  annoying: we need to
2223          *   decompress into destination's address_space (the file offset
2224          *   may change, so source mapping won't do), then recompress (or
2225          *   otherwise reinsert) a subrange.
2226          * - allow ranges within the same file to be cloned (provided
2227          *   they don't overlap)?
2228          */
2229
2230         /* the destination must be opened for writing */
2231         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2232                 return -EINVAL;
2233
2234         if (btrfs_root_readonly(root))
2235                 return -EROFS;
2236
2237         ret = mnt_want_write_file(file);
2238         if (ret)
2239                 return ret;
2240
2241         src_file = fget(srcfd);
2242         if (!src_file) {
2243                 ret = -EBADF;
2244                 goto out_drop_write;
2245         }
2246
2247         src = src_file->f_dentry->d_inode;
2248
2249         ret = -EINVAL;
2250         if (src == inode)
2251                 goto out_fput;
2252
2253         /* the src must be open for reading */
2254         if (!(src_file->f_mode & FMODE_READ))
2255                 goto out_fput;
2256
2257         /* don't make the dst file partly checksummed */
2258         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2259             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2260                 goto out_fput;
2261
2262         ret = -EISDIR;
2263         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2264                 goto out_fput;
2265
2266         ret = -EXDEV;
2267         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2268                 goto out_fput;
2269
2270         ret = -ENOMEM;
2271         buf = vmalloc(btrfs_level_size(root, 0));
2272         if (!buf)
2273                 goto out_fput;
2274
2275         path = btrfs_alloc_path();
2276         if (!path) {
2277                 vfree(buf);
2278                 goto out_fput;
2279         }
2280         path->reada = 2;
2281
2282         if (inode < src) {
2283                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2284                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2285         } else {
2286                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2287                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2288         }
2289
2290         /* determine range to clone */
2291         ret = -EINVAL;
2292         if (off + len > src->i_size || off + len < off)
2293                 goto out_unlock;
2294         if (len == 0)
2295                 olen = len = src->i_size - off;
2296         /* if we extend to eof, continue to block boundary */
2297         if (off + len == src->i_size)
2298                 len = ALIGN(src->i_size, bs) - off;
2299
2300         /* verify the end result is block aligned */
2301         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2302             !IS_ALIGNED(destoff, bs))
2303                 goto out_unlock;
2304
2305         if (destoff > inode->i_size) {
2306                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2307                 if (ret)
2308                         goto out_unlock;
2309         }
2310
2311         /* truncate page cache pages from target inode range */
2312         truncate_inode_pages_range(&inode->i_data, destoff,
2313                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2314
2315         /* do any pending delalloc/csum calc on src, one way or
2316            another, and lock file content */
2317         while (1) {
2318                 struct btrfs_ordered_extent *ordered;
2319                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2320                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2321                 if (!ordered &&
2322                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2323                                    EXTENT_DELALLOC, 0, NULL))
2324                         break;
2325                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2326                 if (ordered)
2327                         btrfs_put_ordered_extent(ordered);
2328                 btrfs_wait_ordered_range(src, off, len);
2329         }
2330
2331         /* clone data */
2332         key.objectid = btrfs_ino(src);
2333         key.type = BTRFS_EXTENT_DATA_KEY;
2334         key.offset = 0;
2335
2336         while (1) {
2337                 /*
2338                  * note the key will change type as we walk through the
2339                  * tree.
2340                  */
2341                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2342                 if (ret < 0)
2343                         goto out;
2344
2345                 nritems = btrfs_header_nritems(path->nodes[0]);
2346                 if (path->slots[0] >= nritems) {
2347                         ret = btrfs_next_leaf(root, path);
2348                         if (ret < 0)
2349                                 goto out;
2350                         if (ret > 0)
2351                                 break;
2352                         nritems = btrfs_header_nritems(path->nodes[0]);
2353                 }
2354                 leaf = path->nodes[0];
2355                 slot = path->slots[0];
2356
2357                 btrfs_item_key_to_cpu(leaf, &key, slot);
2358                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2359                     key.objectid != btrfs_ino(src))
2360                         break;
2361
2362                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2363                         struct btrfs_file_extent_item *extent;
2364                         int type;
2365                         u32 size;
2366                         struct btrfs_key new_key;
2367                         u64 disko = 0, diskl = 0;
2368                         u64 datao = 0, datal = 0;
2369                         u8 comp;
2370                         u64 endoff;
2371
2372                         size = btrfs_item_size_nr(leaf, slot);
2373                         read_extent_buffer(leaf, buf,
2374                                            btrfs_item_ptr_offset(leaf, slot),
2375                                            size);
2376
2377                         extent = btrfs_item_ptr(leaf, slot,
2378                                                 struct btrfs_file_extent_item);
2379                         comp = btrfs_file_extent_compression(leaf, extent);
2380                         type = btrfs_file_extent_type(leaf, extent);
2381                         if (type == BTRFS_FILE_EXTENT_REG ||
2382                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2383                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2384                                                                       extent);
2385                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2386                                                                  extent);
2387                                 datao = btrfs_file_extent_offset(leaf, extent);
2388                                 datal = btrfs_file_extent_num_bytes(leaf,
2389                                                                     extent);
2390                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2391                                 /* take upper bound, may be compressed */
2392                                 datal = btrfs_file_extent_ram_bytes(leaf,
2393                                                                     extent);
2394                         }
2395                         btrfs_release_path(path);
2396
2397                         if (key.offset + datal <= off ||
2398                             key.offset >= off+len)
2399                                 goto next;
2400
2401                         memcpy(&new_key, &key, sizeof(new_key));
2402                         new_key.objectid = btrfs_ino(inode);
2403                         if (off <= key.offset)
2404                                 new_key.offset = key.offset + destoff - off;
2405                         else
2406                                 new_key.offset = destoff;
2407
2408                         /*
2409                          * 1 - adjusting old extent (we may have to split it)
2410                          * 1 - add new extent
2411                          * 1 - inode update
2412                          */
2413                         trans = btrfs_start_transaction(root, 3);
2414                         if (IS_ERR(trans)) {
2415                                 ret = PTR_ERR(trans);
2416                                 goto out;
2417                         }
2418
2419                         if (type == BTRFS_FILE_EXTENT_REG ||
2420                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2421                                 /*
2422                                  *    a  | --- range to clone ---|  b
2423                                  * | ------------- extent ------------- |
2424                                  */
2425
2426                                 /* substract range b */
2427                                 if (key.offset + datal > off + len)
2428                                         datal = off + len - key.offset;
2429
2430                                 /* substract range a */
2431                                 if (off > key.offset) {
2432                                         datao += off - key.offset;
2433                                         datal -= off - key.offset;
2434                                 }
2435
2436                                 ret = btrfs_drop_extents(trans, inode,
2437                                                          new_key.offset,
2438                                                          new_key.offset + datal,
2439                                                          &hint_byte, 1);
2440                                 BUG_ON(ret);
2441
2442                                 ret = btrfs_insert_empty_item(trans, root, path,
2443                                                               &new_key, size);
2444                                 BUG_ON(ret);
2445
2446                                 leaf = path->nodes[0];
2447                                 slot = path->slots[0];
2448                                 write_extent_buffer(leaf, buf,
2449                                             btrfs_item_ptr_offset(leaf, slot),
2450                                             size);
2451
2452                                 extent = btrfs_item_ptr(leaf, slot,
2453                                                 struct btrfs_file_extent_item);
2454
2455                                 /* disko == 0 means it's a hole */
2456                                 if (!disko)
2457                                         datao = 0;
2458
2459                                 btrfs_set_file_extent_offset(leaf, extent,
2460                                                              datao);
2461                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2462                                                                 datal);
2463                                 if (disko) {
2464                                         inode_add_bytes(inode, datal);
2465                                         ret = btrfs_inc_extent_ref(trans, root,
2466                                                         disko, diskl, 0,
2467                                                         root->root_key.objectid,
2468                                                         btrfs_ino(inode),
2469                                                         new_key.offset - datao,
2470                                                         0);
2471                                         BUG_ON(ret);
2472                                 }
2473                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2474                                 u64 skip = 0;
2475                                 u64 trim = 0;
2476                                 if (off > key.offset) {
2477                                         skip = off - key.offset;
2478                                         new_key.offset += skip;
2479                                 }
2480
2481                                 if (key.offset + datal > off+len)
2482                                         trim = key.offset + datal - (off+len);
2483
2484                                 if (comp && (skip || trim)) {
2485                                         ret = -EINVAL;
2486                                         btrfs_end_transaction(trans, root);
2487                                         goto out;
2488                                 }
2489                                 size -= skip + trim;
2490                                 datal -= skip + trim;
2491
2492                                 ret = btrfs_drop_extents(trans, inode,
2493                                                          new_key.offset,
2494                                                          new_key.offset + datal,
2495                                                          &hint_byte, 1);
2496                                 BUG_ON(ret);
2497
2498                                 ret = btrfs_insert_empty_item(trans, root, path,
2499                                                               &new_key, size);
2500                                 BUG_ON(ret);
2501
2502                                 if (skip) {
2503                                         u32 start =
2504                                           btrfs_file_extent_calc_inline_size(0);
2505                                         memmove(buf+start, buf+start+skip,
2506                                                 datal);
2507                                 }
2508
2509                                 leaf = path->nodes[0];
2510                                 slot = path->slots[0];
2511                                 write_extent_buffer(leaf, buf,
2512                                             btrfs_item_ptr_offset(leaf, slot),
2513                                             size);
2514                                 inode_add_bytes(inode, datal);
2515                         }
2516
2517                         btrfs_mark_buffer_dirty(leaf);
2518                         btrfs_release_path(path);
2519
2520                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2521
2522                         /*
2523                          * we round up to the block size at eof when
2524                          * determining which extents to clone above,
2525                          * but shouldn't round up the file size
2526                          */
2527                         endoff = new_key.offset + datal;
2528                         if (endoff > destoff+olen)
2529                                 endoff = destoff+olen;
2530                         if (endoff > inode->i_size)
2531                                 btrfs_i_size_write(inode, endoff);
2532
2533                         ret = btrfs_update_inode(trans, root, inode);
2534                         BUG_ON(ret);
2535                         btrfs_end_transaction(trans, root);
2536                 }
2537 next:
2538                 btrfs_release_path(path);
2539                 key.offset++;
2540         }
2541         ret = 0;
2542 out:
2543         btrfs_release_path(path);
2544         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2545 out_unlock:
2546         mutex_unlock(&src->i_mutex);
2547         mutex_unlock(&inode->i_mutex);
2548         vfree(buf);
2549         btrfs_free_path(path);
2550 out_fput:
2551         fput(src_file);
2552 out_drop_write:
2553         mnt_drop_write_file(file);
2554         return ret;
2555 }
2556
2557 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2558 {
2559         struct btrfs_ioctl_clone_range_args args;
2560
2561         if (copy_from_user(&args, argp, sizeof(args)))
2562                 return -EFAULT;
2563         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2564                                  args.src_length, args.dest_offset);
2565 }
2566
2567 /*
2568  * there are many ways the trans_start and trans_end ioctls can lead
2569  * to deadlocks.  They should only be used by applications that
2570  * basically own the machine, and have a very in depth understanding
2571  * of all the possible deadlocks and enospc problems.
2572  */
2573 static long btrfs_ioctl_trans_start(struct file *file)
2574 {
2575         struct inode *inode = fdentry(file)->d_inode;
2576         struct btrfs_root *root = BTRFS_I(inode)->root;
2577         struct btrfs_trans_handle *trans;
2578         int ret;
2579
2580         ret = -EPERM;
2581         if (!capable(CAP_SYS_ADMIN))
2582                 goto out;
2583
2584         ret = -EINPROGRESS;
2585         if (file->private_data)
2586                 goto out;
2587
2588         ret = -EROFS;
2589         if (btrfs_root_readonly(root))
2590                 goto out;
2591
2592         ret = mnt_want_write_file(file);
2593         if (ret)
2594                 goto out;
2595
2596         atomic_inc(&root->fs_info->open_ioctl_trans);
2597
2598         ret = -ENOMEM;
2599         trans = btrfs_start_ioctl_transaction(root);
2600         if (IS_ERR(trans))
2601                 goto out_drop;
2602
2603         file->private_data = trans;
2604         return 0;
2605
2606 out_drop:
2607         atomic_dec(&root->fs_info->open_ioctl_trans);
2608         mnt_drop_write_file(file);
2609 out:
2610         return ret;
2611 }
2612
2613 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2614 {
2615         struct inode *inode = fdentry(file)->d_inode;
2616         struct btrfs_root *root = BTRFS_I(inode)->root;
2617         struct btrfs_root *new_root;
2618         struct btrfs_dir_item *di;
2619         struct btrfs_trans_handle *trans;
2620         struct btrfs_path *path;
2621         struct btrfs_key location;
2622         struct btrfs_disk_key disk_key;
2623         struct btrfs_super_block *disk_super;
2624         u64 features;
2625         u64 objectid = 0;
2626         u64 dir_id;
2627
2628         if (!capable(CAP_SYS_ADMIN))
2629                 return -EPERM;
2630
2631         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2632                 return -EFAULT;
2633
2634         if (!objectid)
2635                 objectid = root->root_key.objectid;
2636
2637         location.objectid = objectid;
2638         location.type = BTRFS_ROOT_ITEM_KEY;
2639         location.offset = (u64)-1;
2640
2641         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2642         if (IS_ERR(new_root))
2643                 return PTR_ERR(new_root);
2644
2645         if (btrfs_root_refs(&new_root->root_item) == 0)
2646                 return -ENOENT;
2647
2648         path = btrfs_alloc_path();
2649         if (!path)
2650                 return -ENOMEM;
2651         path->leave_spinning = 1;
2652
2653         trans = btrfs_start_transaction(root, 1);
2654         if (IS_ERR(trans)) {
2655                 btrfs_free_path(path);
2656                 return PTR_ERR(trans);
2657         }
2658
2659         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2660         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2661                                    dir_id, "default", 7, 1);
2662         if (IS_ERR_OR_NULL(di)) {
2663                 btrfs_free_path(path);
2664                 btrfs_end_transaction(trans, root);
2665                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2666                        "this isn't going to work\n");
2667                 return -ENOENT;
2668         }
2669
2670         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2671         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2672         btrfs_mark_buffer_dirty(path->nodes[0]);
2673         btrfs_free_path(path);
2674
2675         disk_super = root->fs_info->super_copy;
2676         features = btrfs_super_incompat_flags(disk_super);
2677         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2678                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2679                 btrfs_set_super_incompat_flags(disk_super, features);
2680         }
2681         btrfs_end_transaction(trans, root);
2682
2683         return 0;
2684 }
2685
2686 static void get_block_group_info(struct list_head *groups_list,
2687                                  struct btrfs_ioctl_space_info *space)
2688 {
2689         struct btrfs_block_group_cache *block_group;
2690
2691         space->total_bytes = 0;
2692         space->used_bytes = 0;
2693         space->flags = 0;
2694         list_for_each_entry(block_group, groups_list, list) {
2695                 space->flags = block_group->flags;
2696                 space->total_bytes += block_group->key.offset;
2697                 space->used_bytes +=
2698                         btrfs_block_group_used(&block_group->item);
2699         }
2700 }
2701
2702 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2703 {
2704         struct btrfs_ioctl_space_args space_args;
2705         struct btrfs_ioctl_space_info space;
2706         struct btrfs_ioctl_space_info *dest;
2707         struct btrfs_ioctl_space_info *dest_orig;
2708         struct btrfs_ioctl_space_info __user *user_dest;
2709         struct btrfs_space_info *info;
2710         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2711                        BTRFS_BLOCK_GROUP_SYSTEM,
2712                        BTRFS_BLOCK_GROUP_METADATA,
2713                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2714         int num_types = 4;
2715         int alloc_size;
2716         int ret = 0;
2717         u64 slot_count = 0;
2718         int i, c;
2719
2720         if (copy_from_user(&space_args,
2721                            (struct btrfs_ioctl_space_args __user *)arg,
2722                            sizeof(space_args)))
2723                 return -EFAULT;
2724
2725         for (i = 0; i < num_types; i++) {
2726                 struct btrfs_space_info *tmp;
2727
2728                 info = NULL;
2729                 rcu_read_lock();
2730                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2731                                         list) {
2732                         if (tmp->flags == types[i]) {
2733                                 info = tmp;
2734                                 break;
2735                         }
2736                 }
2737                 rcu_read_unlock();
2738
2739                 if (!info)
2740                         continue;
2741
2742                 down_read(&info->groups_sem);
2743                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2744                         if (!list_empty(&info->block_groups[c]))
2745                                 slot_count++;
2746                 }
2747                 up_read(&info->groups_sem);
2748         }
2749
2750         /* space_slots == 0 means they are asking for a count */
2751         if (space_args.space_slots == 0) {
2752                 space_args.total_spaces = slot_count;
2753                 goto out;
2754         }
2755
2756         slot_count = min_t(u64, space_args.space_slots, slot_count);
2757
2758         alloc_size = sizeof(*dest) * slot_count;
2759
2760         /* we generally have at most 6 or so space infos, one for each raid
2761          * level.  So, a whole page should be more than enough for everyone
2762          */
2763         if (alloc_size > PAGE_CACHE_SIZE)
2764                 return -ENOMEM;
2765
2766         space_args.total_spaces = 0;
2767         dest = kmalloc(alloc_size, GFP_NOFS);
2768         if (!dest)
2769                 return -ENOMEM;
2770         dest_orig = dest;
2771
2772         /* now we have a buffer to copy into */
2773         for (i = 0; i < num_types; i++) {
2774                 struct btrfs_space_info *tmp;
2775
2776                 if (!slot_count)
2777                         break;
2778
2779                 info = NULL;
2780                 rcu_read_lock();
2781                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2782                                         list) {
2783                         if (tmp->flags == types[i]) {
2784                                 info = tmp;
2785                                 break;
2786                         }
2787                 }
2788                 rcu_read_unlock();
2789
2790                 if (!info)
2791                         continue;
2792                 down_read(&info->groups_sem);
2793                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2794                         if (!list_empty(&info->block_groups[c])) {
2795                                 get_block_group_info(&info->block_groups[c],
2796                                                      &space);
2797                                 memcpy(dest, &space, sizeof(space));
2798                                 dest++;
2799                                 space_args.total_spaces++;
2800                                 slot_count--;
2801                         }
2802                         if (!slot_count)
2803                                 break;
2804                 }
2805                 up_read(&info->groups_sem);
2806         }
2807
2808         user_dest = (struct btrfs_ioctl_space_info *)
2809                 (arg + sizeof(struct btrfs_ioctl_space_args));
2810
2811         if (copy_to_user(user_dest, dest_orig, alloc_size))
2812                 ret = -EFAULT;
2813
2814         kfree(dest_orig);
2815 out:
2816         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2817                 ret = -EFAULT;
2818
2819         return ret;
2820 }
2821
2822 /*
2823  * there are many ways the trans_start and trans_end ioctls can lead
2824  * to deadlocks.  They should only be used by applications that
2825  * basically own the machine, and have a very in depth understanding
2826  * of all the possible deadlocks and enospc problems.
2827  */
2828 long btrfs_ioctl_trans_end(struct file *file)
2829 {
2830         struct inode *inode = fdentry(file)->d_inode;
2831         struct btrfs_root *root = BTRFS_I(inode)->root;
2832         struct btrfs_trans_handle *trans;
2833
2834         trans = file->private_data;
2835         if (!trans)
2836                 return -EINVAL;
2837         file->private_data = NULL;
2838
2839         btrfs_end_transaction(trans, root);
2840
2841         atomic_dec(&root->fs_info->open_ioctl_trans);
2842
2843         mnt_drop_write_file(file);
2844         return 0;
2845 }
2846
2847 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2848 {
2849         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2850         struct btrfs_trans_handle *trans;
2851         u64 transid;
2852         int ret;
2853
2854         trans = btrfs_start_transaction(root, 0);
2855         if (IS_ERR(trans))
2856                 return PTR_ERR(trans);
2857         transid = trans->transid;
2858         ret = btrfs_commit_transaction_async(trans, root, 0);
2859         if (ret) {
2860                 btrfs_end_transaction(trans, root);
2861                 return ret;
2862         }
2863
2864         if (argp)
2865                 if (copy_to_user(argp, &transid, sizeof(transid)))
2866                         return -EFAULT;
2867         return 0;
2868 }
2869
2870 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2871 {
2872         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2873         u64 transid;
2874
2875         if (argp) {
2876                 if (copy_from_user(&transid, argp, sizeof(transid)))
2877                         return -EFAULT;
2878         } else {
2879                 transid = 0;  /* current trans */
2880         }
2881         return btrfs_wait_for_commit(root, transid);
2882 }
2883
2884 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2885 {
2886         int ret;
2887         struct btrfs_ioctl_scrub_args *sa;
2888
2889         if (!capable(CAP_SYS_ADMIN))
2890                 return -EPERM;
2891
2892         sa = memdup_user(arg, sizeof(*sa));
2893         if (IS_ERR(sa))
2894                 return PTR_ERR(sa);
2895
2896         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2897                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2898
2899         if (copy_to_user(arg, sa, sizeof(*sa)))
2900                 ret = -EFAULT;
2901
2902         kfree(sa);
2903         return ret;
2904 }
2905
2906 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2907 {
2908         if (!capable(CAP_SYS_ADMIN))
2909                 return -EPERM;
2910
2911         return btrfs_scrub_cancel(root);
2912 }
2913
2914 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2915                                        void __user *arg)
2916 {
2917         struct btrfs_ioctl_scrub_args *sa;
2918         int ret;
2919
2920         if (!capable(CAP_SYS_ADMIN))
2921                 return -EPERM;
2922
2923         sa = memdup_user(arg, sizeof(*sa));
2924         if (IS_ERR(sa))
2925                 return PTR_ERR(sa);
2926
2927         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2928
2929         if (copy_to_user(arg, sa, sizeof(*sa)))
2930                 ret = -EFAULT;
2931
2932         kfree(sa);
2933         return ret;
2934 }
2935
2936 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
2937 {
2938         int ret = 0;
2939         int i;
2940         u64 rel_ptr;
2941         int size;
2942         struct btrfs_ioctl_ino_path_args *ipa = NULL;
2943         struct inode_fs_paths *ipath = NULL;
2944         struct btrfs_path *path;
2945
2946         if (!capable(CAP_SYS_ADMIN))
2947                 return -EPERM;
2948
2949         path = btrfs_alloc_path();
2950         if (!path) {
2951                 ret = -ENOMEM;
2952                 goto out;
2953         }
2954
2955         ipa = memdup_user(arg, sizeof(*ipa));
2956         if (IS_ERR(ipa)) {
2957                 ret = PTR_ERR(ipa);
2958                 ipa = NULL;
2959                 goto out;
2960         }
2961
2962         size = min_t(u32, ipa->size, 4096);
2963         ipath = init_ipath(size, root, path);
2964         if (IS_ERR(ipath)) {
2965                 ret = PTR_ERR(ipath);
2966                 ipath = NULL;
2967                 goto out;
2968         }
2969
2970         ret = paths_from_inode(ipa->inum, ipath);
2971         if (ret < 0)
2972                 goto out;
2973
2974         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
2975                 rel_ptr = ipath->fspath->val[i] -
2976                           (u64)(unsigned long)ipath->fspath->val;
2977                 ipath->fspath->val[i] = rel_ptr;
2978         }
2979
2980         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
2981                            (void *)(unsigned long)ipath->fspath, size);
2982         if (ret) {
2983                 ret = -EFAULT;
2984                 goto out;
2985         }
2986
2987 out:
2988         btrfs_free_path(path);
2989         free_ipath(ipath);
2990         kfree(ipa);
2991
2992         return ret;
2993 }
2994
2995 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2996 {
2997         struct btrfs_data_container *inodes = ctx;
2998         const size_t c = 3 * sizeof(u64);
2999
3000         if (inodes->bytes_left >= c) {
3001                 inodes->bytes_left -= c;
3002                 inodes->val[inodes->elem_cnt] = inum;
3003                 inodes->val[inodes->elem_cnt + 1] = offset;
3004                 inodes->val[inodes->elem_cnt + 2] = root;
3005                 inodes->elem_cnt += 3;
3006         } else {
3007                 inodes->bytes_missing += c - inodes->bytes_left;
3008                 inodes->bytes_left = 0;
3009                 inodes->elem_missed += 3;
3010         }
3011
3012         return 0;
3013 }
3014
3015 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3016                                         void __user *arg)
3017 {
3018         int ret = 0;
3019         int size;
3020         u64 extent_item_pos;
3021         struct btrfs_ioctl_logical_ino_args *loi;
3022         struct btrfs_data_container *inodes = NULL;
3023         struct btrfs_path *path = NULL;
3024         struct btrfs_key key;
3025
3026         if (!capable(CAP_SYS_ADMIN))
3027                 return -EPERM;
3028
3029         loi = memdup_user(arg, sizeof(*loi));
3030         if (IS_ERR(loi)) {
3031                 ret = PTR_ERR(loi);
3032                 loi = NULL;
3033                 goto out;
3034         }
3035
3036         path = btrfs_alloc_path();
3037         if (!path) {
3038                 ret = -ENOMEM;
3039                 goto out;
3040         }
3041
3042         size = min_t(u32, loi->size, 4096);
3043         inodes = init_data_container(size);
3044         if (IS_ERR(inodes)) {
3045                 ret = PTR_ERR(inodes);
3046                 inodes = NULL;
3047                 goto out;
3048         }
3049
3050         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
3051         btrfs_release_path(path);
3052
3053         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
3054                 ret = -ENOENT;
3055         if (ret < 0)
3056                 goto out;
3057
3058         extent_item_pos = loi->logical - key.objectid;
3059         ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
3060                                         extent_item_pos, build_ino_list,
3061                                         inodes);
3062
3063         if (ret < 0)
3064                 goto out;
3065
3066         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3067                            (void *)(unsigned long)inodes, size);
3068         if (ret)
3069                 ret = -EFAULT;
3070
3071 out:
3072         btrfs_free_path(path);
3073         kfree(inodes);
3074         kfree(loi);
3075
3076         return ret;
3077 }
3078
3079 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3080                                struct btrfs_ioctl_balance_args *bargs)
3081 {
3082         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3083
3084         bargs->flags = bctl->flags;
3085
3086         if (atomic_read(&fs_info->balance_running))
3087                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3088         if (atomic_read(&fs_info->balance_pause_req))
3089                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3090         if (atomic_read(&fs_info->balance_cancel_req))
3091                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3092
3093         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3094         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3095         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3096
3097         if (lock) {
3098                 spin_lock(&fs_info->balance_lock);
3099                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3100                 spin_unlock(&fs_info->balance_lock);
3101         } else {
3102                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3103         }
3104 }
3105
3106 static long btrfs_ioctl_balance(struct btrfs_root *root, void __user *arg)
3107 {
3108         struct btrfs_fs_info *fs_info = root->fs_info;
3109         struct btrfs_ioctl_balance_args *bargs;
3110         struct btrfs_balance_control *bctl;
3111         int ret;
3112
3113         if (!capable(CAP_SYS_ADMIN))
3114                 return -EPERM;
3115
3116         if (fs_info->sb->s_flags & MS_RDONLY)
3117                 return -EROFS;
3118
3119         mutex_lock(&fs_info->volume_mutex);
3120         mutex_lock(&fs_info->balance_mutex);
3121
3122         if (arg) {
3123                 bargs = memdup_user(arg, sizeof(*bargs));
3124                 if (IS_ERR(bargs)) {
3125                         ret = PTR_ERR(bargs);
3126                         goto out;
3127                 }
3128
3129                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3130                         if (!fs_info->balance_ctl) {
3131                                 ret = -ENOTCONN;
3132                                 goto out_bargs;
3133                         }
3134
3135                         bctl = fs_info->balance_ctl;
3136                         spin_lock(&fs_info->balance_lock);
3137                         bctl->flags |= BTRFS_BALANCE_RESUME;
3138                         spin_unlock(&fs_info->balance_lock);
3139
3140                         goto do_balance;
3141                 }
3142         } else {
3143                 bargs = NULL;
3144         }
3145
3146         if (fs_info->balance_ctl) {
3147                 ret = -EINPROGRESS;
3148                 goto out_bargs;
3149         }
3150
3151         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3152         if (!bctl) {
3153                 ret = -ENOMEM;
3154                 goto out_bargs;
3155         }
3156
3157         bctl->fs_info = fs_info;
3158         if (arg) {
3159                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3160                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3161                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3162
3163                 bctl->flags = bargs->flags;
3164         } else {
3165                 /* balance everything - no filters */
3166                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3167         }
3168
3169 do_balance:
3170         ret = btrfs_balance(bctl, bargs);
3171         /*
3172          * bctl is freed in __cancel_balance or in free_fs_info if
3173          * restriper was paused all the way until unmount
3174          */
3175         if (arg) {
3176                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3177                         ret = -EFAULT;
3178         }
3179
3180 out_bargs:
3181         kfree(bargs);
3182 out:
3183         mutex_unlock(&fs_info->balance_mutex);
3184         mutex_unlock(&fs_info->volume_mutex);
3185         return ret;
3186 }
3187
3188 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3189 {
3190         if (!capable(CAP_SYS_ADMIN))
3191                 return -EPERM;
3192
3193         switch (cmd) {
3194         case BTRFS_BALANCE_CTL_PAUSE:
3195                 return btrfs_pause_balance(root->fs_info);
3196         case BTRFS_BALANCE_CTL_CANCEL:
3197                 return btrfs_cancel_balance(root->fs_info);
3198         }
3199
3200         return -EINVAL;
3201 }
3202
3203 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3204                                          void __user *arg)
3205 {
3206         struct btrfs_fs_info *fs_info = root->fs_info;
3207         struct btrfs_ioctl_balance_args *bargs;
3208         int ret = 0;
3209
3210         if (!capable(CAP_SYS_ADMIN))
3211                 return -EPERM;
3212
3213         mutex_lock(&fs_info->balance_mutex);
3214         if (!fs_info->balance_ctl) {
3215                 ret = -ENOTCONN;
3216                 goto out;
3217         }
3218
3219         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3220         if (!bargs) {
3221                 ret = -ENOMEM;
3222                 goto out;
3223         }
3224
3225         update_ioctl_balance_args(fs_info, 1, bargs);
3226
3227         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3228                 ret = -EFAULT;
3229
3230         kfree(bargs);
3231 out:
3232         mutex_unlock(&fs_info->balance_mutex);
3233         return ret;
3234 }
3235
3236 long btrfs_ioctl(struct file *file, unsigned int
3237                 cmd, unsigned long arg)
3238 {
3239         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3240         void __user *argp = (void __user *)arg;
3241
3242         switch (cmd) {
3243         case FS_IOC_GETFLAGS:
3244                 return btrfs_ioctl_getflags(file, argp);
3245         case FS_IOC_SETFLAGS:
3246                 return btrfs_ioctl_setflags(file, argp);
3247         case FS_IOC_GETVERSION:
3248                 return btrfs_ioctl_getversion(file, argp);
3249         case FITRIM:
3250                 return btrfs_ioctl_fitrim(file, argp);
3251         case BTRFS_IOC_SNAP_CREATE:
3252                 return btrfs_ioctl_snap_create(file, argp, 0);
3253         case BTRFS_IOC_SNAP_CREATE_V2:
3254                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3255         case BTRFS_IOC_SUBVOL_CREATE:
3256                 return btrfs_ioctl_snap_create(file, argp, 1);
3257         case BTRFS_IOC_SNAP_DESTROY:
3258                 return btrfs_ioctl_snap_destroy(file, argp);
3259         case BTRFS_IOC_SUBVOL_GETFLAGS:
3260                 return btrfs_ioctl_subvol_getflags(file, argp);
3261         case BTRFS_IOC_SUBVOL_SETFLAGS:
3262                 return btrfs_ioctl_subvol_setflags(file, argp);
3263         case BTRFS_IOC_DEFAULT_SUBVOL:
3264                 return btrfs_ioctl_default_subvol(file, argp);
3265         case BTRFS_IOC_DEFRAG:
3266                 return btrfs_ioctl_defrag(file, NULL);
3267         case BTRFS_IOC_DEFRAG_RANGE:
3268                 return btrfs_ioctl_defrag(file, argp);
3269         case BTRFS_IOC_RESIZE:
3270                 return btrfs_ioctl_resize(root, argp);
3271         case BTRFS_IOC_ADD_DEV:
3272                 return btrfs_ioctl_add_dev(root, argp);
3273         case BTRFS_IOC_RM_DEV:
3274                 return btrfs_ioctl_rm_dev(root, argp);
3275         case BTRFS_IOC_FS_INFO:
3276                 return btrfs_ioctl_fs_info(root, argp);
3277         case BTRFS_IOC_DEV_INFO:
3278                 return btrfs_ioctl_dev_info(root, argp);
3279         case BTRFS_IOC_BALANCE:
3280                 return btrfs_ioctl_balance(root, NULL);
3281         case BTRFS_IOC_CLONE:
3282                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3283         case BTRFS_IOC_CLONE_RANGE:
3284                 return btrfs_ioctl_clone_range(file, argp);
3285         case BTRFS_IOC_TRANS_START:
3286                 return btrfs_ioctl_trans_start(file);
3287         case BTRFS_IOC_TRANS_END:
3288                 return btrfs_ioctl_trans_end(file);
3289         case BTRFS_IOC_TREE_SEARCH:
3290                 return btrfs_ioctl_tree_search(file, argp);
3291         case BTRFS_IOC_INO_LOOKUP:
3292                 return btrfs_ioctl_ino_lookup(file, argp);
3293         case BTRFS_IOC_INO_PATHS:
3294                 return btrfs_ioctl_ino_to_path(root, argp);
3295         case BTRFS_IOC_LOGICAL_INO:
3296                 return btrfs_ioctl_logical_to_ino(root, argp);
3297         case BTRFS_IOC_SPACE_INFO:
3298                 return btrfs_ioctl_space_info(root, argp);
3299         case BTRFS_IOC_SYNC:
3300                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3301                 return 0;
3302         case BTRFS_IOC_START_SYNC:
3303                 return btrfs_ioctl_start_sync(file, argp);
3304         case BTRFS_IOC_WAIT_SYNC:
3305                 return btrfs_ioctl_wait_sync(file, argp);
3306         case BTRFS_IOC_SCRUB:
3307                 return btrfs_ioctl_scrub(root, argp);
3308         case BTRFS_IOC_SCRUB_CANCEL:
3309                 return btrfs_ioctl_scrub_cancel(root, argp);
3310         case BTRFS_IOC_SCRUB_PROGRESS:
3311                 return btrfs_ioctl_scrub_progress(root, argp);
3312         case BTRFS_IOC_BALANCE_V2:
3313                 return btrfs_ioctl_balance(root, argp);
3314         case BTRFS_IOC_BALANCE_CTL:
3315                 return btrfs_ioctl_balance_ctl(root, arg);
3316         case BTRFS_IOC_BALANCE_PROGRESS:
3317                 return btrfs_ioctl_balance_progress(root, argp);
3318         }
3319
3320         return -ENOTTY;
3321 }