]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/btrfs/ioctl.c
1e7a9bac31ab9f6774e98e76264893752921ff36
[~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
180         if (btrfs_root_readonly(root))
181                 return -EROFS;
182
183         if (copy_from_user(&flags, arg, sizeof(flags)))
184                 return -EFAULT;
185
186         ret = check_flags(flags);
187         if (ret)
188                 return ret;
189
190         if (!inode_owner_or_capable(inode))
191                 return -EACCES;
192
193         mutex_lock(&inode->i_mutex);
194
195         flags = btrfs_mask_flags(inode->i_mode, flags);
196         oldflags = btrfs_flags_to_ioctl(ip->flags);
197         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
198                 if (!capable(CAP_LINUX_IMMUTABLE)) {
199                         ret = -EPERM;
200                         goto out_unlock;
201                 }
202         }
203
204         ret = mnt_want_write(file->f_path.mnt);
205         if (ret)
206                 goto out_unlock;
207
208         if (flags & FS_SYNC_FL)
209                 ip->flags |= BTRFS_INODE_SYNC;
210         else
211                 ip->flags &= ~BTRFS_INODE_SYNC;
212         if (flags & FS_IMMUTABLE_FL)
213                 ip->flags |= BTRFS_INODE_IMMUTABLE;
214         else
215                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
216         if (flags & FS_APPEND_FL)
217                 ip->flags |= BTRFS_INODE_APPEND;
218         else
219                 ip->flags &= ~BTRFS_INODE_APPEND;
220         if (flags & FS_NODUMP_FL)
221                 ip->flags |= BTRFS_INODE_NODUMP;
222         else
223                 ip->flags &= ~BTRFS_INODE_NODUMP;
224         if (flags & FS_NOATIME_FL)
225                 ip->flags |= BTRFS_INODE_NOATIME;
226         else
227                 ip->flags &= ~BTRFS_INODE_NOATIME;
228         if (flags & FS_DIRSYNC_FL)
229                 ip->flags |= BTRFS_INODE_DIRSYNC;
230         else
231                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
232         if (flags & FS_NOCOW_FL)
233                 ip->flags |= BTRFS_INODE_NODATACOW;
234         else
235                 ip->flags &= ~BTRFS_INODE_NODATACOW;
236
237         /*
238          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
239          * flag may be changed automatically if compression code won't make
240          * things smaller.
241          */
242         if (flags & FS_NOCOMP_FL) {
243                 ip->flags &= ~BTRFS_INODE_COMPRESS;
244                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
245         } else if (flags & FS_COMPR_FL) {
246                 ip->flags |= BTRFS_INODE_COMPRESS;
247                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
248         } else {
249                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
250         }
251
252         trans = btrfs_join_transaction(root);
253         BUG_ON(IS_ERR(trans));
254
255         btrfs_update_iflags(inode);
256         inode->i_ctime = CURRENT_TIME;
257         ret = btrfs_update_inode(trans, root, inode);
258         BUG_ON(ret);
259
260         btrfs_end_transaction(trans, root);
261
262         mnt_drop_write(file->f_path.mnt);
263
264         ret = 0;
265  out_unlock:
266         mutex_unlock(&inode->i_mutex);
267         return ret;
268 }
269
270 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
271 {
272         struct inode *inode = file->f_path.dentry->d_inode;
273
274         return put_user(inode->i_generation, arg);
275 }
276
277 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
278 {
279         struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info;
280         struct btrfs_fs_info *fs_info = root->fs_info;
281         struct btrfs_device *device;
282         struct request_queue *q;
283         struct fstrim_range range;
284         u64 minlen = ULLONG_MAX;
285         u64 num_devices = 0;
286         u64 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
287         int ret;
288
289         if (!capable(CAP_SYS_ADMIN))
290                 return -EPERM;
291
292         rcu_read_lock();
293         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
294                                 dev_list) {
295                 if (!device->bdev)
296                         continue;
297                 q = bdev_get_queue(device->bdev);
298                 if (blk_queue_discard(q)) {
299                         num_devices++;
300                         minlen = min((u64)q->limits.discard_granularity,
301                                      minlen);
302                 }
303         }
304         rcu_read_unlock();
305
306         if (!num_devices)
307                 return -EOPNOTSUPP;
308         if (copy_from_user(&range, arg, sizeof(range)))
309                 return -EFAULT;
310         if (range.start > total_bytes)
311                 return -EINVAL;
312
313         range.len = min(range.len, total_bytes - range.start);
314         range.minlen = max(range.minlen, minlen);
315         ret = btrfs_trim_fs(root, &range);
316         if (ret < 0)
317                 return ret;
318
319         if (copy_to_user(arg, &range, sizeof(range)))
320                 return -EFAULT;
321
322         return 0;
323 }
324
325 static noinline int create_subvol(struct btrfs_root *root,
326                                   struct dentry *dentry,
327                                   char *name, int namelen,
328                                   u64 *async_transid)
329 {
330         struct btrfs_trans_handle *trans;
331         struct btrfs_key key;
332         struct btrfs_root_item root_item;
333         struct btrfs_inode_item *inode_item;
334         struct extent_buffer *leaf;
335         struct btrfs_root *new_root;
336         struct dentry *parent = dentry->d_parent;
337         struct inode *dir;
338         int ret;
339         int err;
340         u64 objectid;
341         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
342         u64 index = 0;
343
344         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
345         if (ret)
346                 return ret;
347
348         dir = parent->d_inode;
349
350         /*
351          * 1 - inode item
352          * 2 - refs
353          * 1 - root item
354          * 2 - dir items
355          */
356         trans = btrfs_start_transaction(root, 6);
357         if (IS_ERR(trans))
358                 return PTR_ERR(trans);
359
360         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
361                                       0, objectid, NULL, 0, 0, 0);
362         if (IS_ERR(leaf)) {
363                 ret = PTR_ERR(leaf);
364                 goto fail;
365         }
366
367         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
368         btrfs_set_header_bytenr(leaf, leaf->start);
369         btrfs_set_header_generation(leaf, trans->transid);
370         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
371         btrfs_set_header_owner(leaf, objectid);
372
373         write_extent_buffer(leaf, root->fs_info->fsid,
374                             (unsigned long)btrfs_header_fsid(leaf),
375                             BTRFS_FSID_SIZE);
376         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
377                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
378                             BTRFS_UUID_SIZE);
379         btrfs_mark_buffer_dirty(leaf);
380
381         inode_item = &root_item.inode;
382         memset(inode_item, 0, sizeof(*inode_item));
383         inode_item->generation = cpu_to_le64(1);
384         inode_item->size = cpu_to_le64(3);
385         inode_item->nlink = cpu_to_le32(1);
386         inode_item->nbytes = cpu_to_le64(root->leafsize);
387         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
388
389         root_item.flags = 0;
390         root_item.byte_limit = 0;
391         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
392
393         btrfs_set_root_bytenr(&root_item, leaf->start);
394         btrfs_set_root_generation(&root_item, trans->transid);
395         btrfs_set_root_level(&root_item, 0);
396         btrfs_set_root_refs(&root_item, 1);
397         btrfs_set_root_used(&root_item, leaf->len);
398         btrfs_set_root_last_snapshot(&root_item, 0);
399
400         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
401         root_item.drop_level = 0;
402
403         btrfs_tree_unlock(leaf);
404         free_extent_buffer(leaf);
405         leaf = NULL;
406
407         btrfs_set_root_dirid(&root_item, new_dirid);
408
409         key.objectid = objectid;
410         key.offset = 0;
411         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
412         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
413                                 &root_item);
414         if (ret)
415                 goto fail;
416
417         key.offset = (u64)-1;
418         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
419         BUG_ON(IS_ERR(new_root));
420
421         btrfs_record_root_in_trans(trans, new_root);
422
423         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
424         /*
425          * insert the directory item
426          */
427         ret = btrfs_set_inode_index(dir, &index);
428         BUG_ON(ret);
429
430         ret = btrfs_insert_dir_item(trans, root,
431                                     name, namelen, dir, &key,
432                                     BTRFS_FT_DIR, index);
433         if (ret)
434                 goto fail;
435
436         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
437         ret = btrfs_update_inode(trans, root, dir);
438         BUG_ON(ret);
439
440         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
441                                  objectid, root->root_key.objectid,
442                                  btrfs_ino(dir), index, name, namelen);
443
444         BUG_ON(ret);
445
446         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
447 fail:
448         if (async_transid) {
449                 *async_transid = trans->transid;
450                 err = btrfs_commit_transaction_async(trans, root, 1);
451         } else {
452                 err = btrfs_commit_transaction(trans, root);
453         }
454         if (err && !ret)
455                 ret = err;
456         return ret;
457 }
458
459 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
460                            char *name, int namelen, u64 *async_transid,
461                            bool readonly)
462 {
463         struct inode *inode;
464         struct btrfs_pending_snapshot *pending_snapshot;
465         struct btrfs_trans_handle *trans;
466         int ret;
467
468         if (!root->ref_cows)
469                 return -EINVAL;
470
471         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
472         if (!pending_snapshot)
473                 return -ENOMEM;
474
475         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
476         pending_snapshot->dentry = dentry;
477         pending_snapshot->root = root;
478         pending_snapshot->readonly = readonly;
479
480         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
481         if (IS_ERR(trans)) {
482                 ret = PTR_ERR(trans);
483                 goto fail;
484         }
485
486         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
487         BUG_ON(ret);
488
489         spin_lock(&root->fs_info->trans_lock);
490         list_add(&pending_snapshot->list,
491                  &trans->transaction->pending_snapshots);
492         spin_unlock(&root->fs_info->trans_lock);
493         if (async_transid) {
494                 *async_transid = trans->transid;
495                 ret = btrfs_commit_transaction_async(trans,
496                                      root->fs_info->extent_root, 1);
497         } else {
498                 ret = btrfs_commit_transaction(trans,
499                                                root->fs_info->extent_root);
500         }
501         BUG_ON(ret);
502
503         ret = pending_snapshot->error;
504         if (ret)
505                 goto fail;
506
507         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
508         if (ret)
509                 goto fail;
510
511         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
512         if (IS_ERR(inode)) {
513                 ret = PTR_ERR(inode);
514                 goto fail;
515         }
516         BUG_ON(!inode);
517         d_instantiate(dentry, inode);
518         ret = 0;
519 fail:
520         kfree(pending_snapshot);
521         return ret;
522 }
523
524 /*  copy of check_sticky in fs/namei.c()
525 * It's inline, so penalty for filesystems that don't use sticky bit is
526 * minimal.
527 */
528 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
529 {
530         uid_t fsuid = current_fsuid();
531
532         if (!(dir->i_mode & S_ISVTX))
533                 return 0;
534         if (inode->i_uid == fsuid)
535                 return 0;
536         if (dir->i_uid == fsuid)
537                 return 0;
538         return !capable(CAP_FOWNER);
539 }
540
541 /*  copy of may_delete in fs/namei.c()
542  *      Check whether we can remove a link victim from directory dir, check
543  *  whether the type of victim is right.
544  *  1. We can't do it if dir is read-only (done in permission())
545  *  2. We should have write and exec permissions on dir
546  *  3. We can't remove anything from append-only dir
547  *  4. We can't do anything with immutable dir (done in permission())
548  *  5. If the sticky bit on dir is set we should either
549  *      a. be owner of dir, or
550  *      b. be owner of victim, or
551  *      c. have CAP_FOWNER capability
552  *  6. If the victim is append-only or immutable we can't do antyhing with
553  *     links pointing to it.
554  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
555  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
556  *  9. We can't remove a root or mountpoint.
557  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
558  *     nfs_async_unlink().
559  */
560
561 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
562 {
563         int error;
564
565         if (!victim->d_inode)
566                 return -ENOENT;
567
568         BUG_ON(victim->d_parent->d_inode != dir);
569         audit_inode_child(victim, dir);
570
571         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
572         if (error)
573                 return error;
574         if (IS_APPEND(dir))
575                 return -EPERM;
576         if (btrfs_check_sticky(dir, victim->d_inode)||
577                 IS_APPEND(victim->d_inode)||
578             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
579                 return -EPERM;
580         if (isdir) {
581                 if (!S_ISDIR(victim->d_inode->i_mode))
582                         return -ENOTDIR;
583                 if (IS_ROOT(victim))
584                         return -EBUSY;
585         } else if (S_ISDIR(victim->d_inode->i_mode))
586                 return -EISDIR;
587         if (IS_DEADDIR(dir))
588                 return -ENOENT;
589         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
590                 return -EBUSY;
591         return 0;
592 }
593
594 /* copy of may_create in fs/namei.c() */
595 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
596 {
597         if (child->d_inode)
598                 return -EEXIST;
599         if (IS_DEADDIR(dir))
600                 return -ENOENT;
601         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
602 }
603
604 /*
605  * Create a new subvolume below @parent.  This is largely modeled after
606  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
607  * inside this filesystem so it's quite a bit simpler.
608  */
609 static noinline int btrfs_mksubvol(struct path *parent,
610                                    char *name, int namelen,
611                                    struct btrfs_root *snap_src,
612                                    u64 *async_transid, bool readonly)
613 {
614         struct inode *dir  = parent->dentry->d_inode;
615         struct dentry *dentry;
616         int error;
617
618         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
619
620         dentry = lookup_one_len(name, parent->dentry, namelen);
621         error = PTR_ERR(dentry);
622         if (IS_ERR(dentry))
623                 goto out_unlock;
624
625         error = -EEXIST;
626         if (dentry->d_inode)
627                 goto out_dput;
628
629         error = mnt_want_write(parent->mnt);
630         if (error)
631                 goto out_dput;
632
633         error = btrfs_may_create(dir, dentry);
634         if (error)
635                 goto out_drop_write;
636
637         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
638
639         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
640                 goto out_up_read;
641
642         if (snap_src) {
643                 error = create_snapshot(snap_src, dentry,
644                                         name, namelen, async_transid, readonly);
645         } else {
646                 error = create_subvol(BTRFS_I(dir)->root, dentry,
647                                       name, namelen, async_transid);
648         }
649         if (!error)
650                 fsnotify_mkdir(dir, dentry);
651 out_up_read:
652         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
653 out_drop_write:
654         mnt_drop_write(parent->mnt);
655 out_dput:
656         dput(dentry);
657 out_unlock:
658         mutex_unlock(&dir->i_mutex);
659         return error;
660 }
661
662 /*
663  * When we're defragging a range, we don't want to kick it off again
664  * if it is really just waiting for delalloc to send it down.
665  * If we find a nice big extent or delalloc range for the bytes in the
666  * file you want to defrag, we return 0 to let you know to skip this
667  * part of the file
668  */
669 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
670 {
671         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
672         struct extent_map *em = NULL;
673         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
674         u64 end;
675
676         read_lock(&em_tree->lock);
677         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
678         read_unlock(&em_tree->lock);
679
680         if (em) {
681                 end = extent_map_end(em);
682                 free_extent_map(em);
683                 if (end - offset > thresh)
684                         return 0;
685         }
686         /* if we already have a nice delalloc here, just stop */
687         thresh /= 2;
688         end = count_range_bits(io_tree, &offset, offset + thresh,
689                                thresh, EXTENT_DELALLOC, 1);
690         if (end >= thresh)
691                 return 0;
692         return 1;
693 }
694
695 /*
696  * helper function to walk through a file and find extents
697  * newer than a specific transid, and smaller than thresh.
698  *
699  * This is used by the defragging code to find new and small
700  * extents
701  */
702 static int find_new_extents(struct btrfs_root *root,
703                             struct inode *inode, u64 newer_than,
704                             u64 *off, int thresh)
705 {
706         struct btrfs_path *path;
707         struct btrfs_key min_key;
708         struct btrfs_key max_key;
709         struct extent_buffer *leaf;
710         struct btrfs_file_extent_item *extent;
711         int type;
712         int ret;
713         u64 ino = btrfs_ino(inode);
714
715         path = btrfs_alloc_path();
716         if (!path)
717                 return -ENOMEM;
718
719         min_key.objectid = ino;
720         min_key.type = BTRFS_EXTENT_DATA_KEY;
721         min_key.offset = *off;
722
723         max_key.objectid = ino;
724         max_key.type = (u8)-1;
725         max_key.offset = (u64)-1;
726
727         path->keep_locks = 1;
728
729         while(1) {
730                 ret = btrfs_search_forward(root, &min_key, &max_key,
731                                            path, 0, newer_than);
732                 if (ret != 0)
733                         goto none;
734                 if (min_key.objectid != ino)
735                         goto none;
736                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
737                         goto none;
738
739                 leaf = path->nodes[0];
740                 extent = btrfs_item_ptr(leaf, path->slots[0],
741                                         struct btrfs_file_extent_item);
742
743                 type = btrfs_file_extent_type(leaf, extent);
744                 if (type == BTRFS_FILE_EXTENT_REG &&
745                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
746                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
747                         *off = min_key.offset;
748                         btrfs_free_path(path);
749                         return 0;
750                 }
751
752                 if (min_key.offset == (u64)-1)
753                         goto none;
754
755                 min_key.offset++;
756                 btrfs_release_path(path);
757         }
758 none:
759         btrfs_free_path(path);
760         return -ENOENT;
761 }
762
763 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
764                                int thresh, u64 *last_len, u64 *skip,
765                                u64 *defrag_end)
766 {
767         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
768         struct extent_map *em = NULL;
769         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
770         int ret = 1;
771
772         /*
773          * make sure that once we start defragging an extent, we keep on
774          * defragging it
775          */
776         if (start < *defrag_end)
777                 return 1;
778
779         *skip = 0;
780
781         /*
782          * hopefully we have this extent in the tree already, try without
783          * the full extent lock
784          */
785         read_lock(&em_tree->lock);
786         em = lookup_extent_mapping(em_tree, start, len);
787         read_unlock(&em_tree->lock);
788
789         if (!em) {
790                 /* get the big lock and read metadata off disk */
791                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
792                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
793                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
794
795                 if (IS_ERR(em))
796                         return 0;
797         }
798
799         /* this will cover holes, and inline extents */
800         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
801                 ret = 0;
802
803         /*
804          * we hit a real extent, if it is big don't bother defragging it again
805          */
806         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
807                 ret = 0;
808
809         /*
810          * last_len ends up being a counter of how many bytes we've defragged.
811          * every time we choose not to defrag an extent, we reset *last_len
812          * so that the next tiny extent will force a defrag.
813          *
814          * The end result of this is that tiny extents before a single big
815          * extent will force at least part of that big extent to be defragged.
816          */
817         if (ret) {
818                 *defrag_end = extent_map_end(em);
819         } else {
820                 *last_len = 0;
821                 *skip = extent_map_end(em);
822                 *defrag_end = 0;
823         }
824
825         free_extent_map(em);
826         return ret;
827 }
828
829 /*
830  * it doesn't do much good to defrag one or two pages
831  * at a time.  This pulls in a nice chunk of pages
832  * to COW and defrag.
833  *
834  * It also makes sure the delalloc code has enough
835  * dirty data to avoid making new small extents as part
836  * of the defrag
837  *
838  * It's a good idea to start RA on this range
839  * before calling this.
840  */
841 static int cluster_pages_for_defrag(struct inode *inode,
842                                     struct page **pages,
843                                     unsigned long start_index,
844                                     int num_pages)
845 {
846         unsigned long file_end;
847         u64 isize = i_size_read(inode);
848         u64 page_start;
849         u64 page_end;
850         int ret;
851         int i;
852         int i_done;
853         struct btrfs_ordered_extent *ordered;
854         struct extent_state *cached_state = NULL;
855         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
856
857         if (isize == 0)
858                 return 0;
859         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
860
861         mutex_lock(&inode->i_mutex);
862         ret = btrfs_delalloc_reserve_space(inode,
863                                            num_pages << PAGE_CACHE_SHIFT);
864         mutex_unlock(&inode->i_mutex);
865         if (ret)
866                 return ret;
867 again:
868         ret = 0;
869         i_done = 0;
870
871         /* step one, lock all the pages */
872         for (i = 0; i < num_pages; i++) {
873                 struct page *page;
874                 page = find_or_create_page(inode->i_mapping,
875                                             start_index + i, mask);
876                 if (!page)
877                         break;
878
879                 if (!PageUptodate(page)) {
880                         btrfs_readpage(NULL, page);
881                         lock_page(page);
882                         if (!PageUptodate(page)) {
883                                 unlock_page(page);
884                                 page_cache_release(page);
885                                 ret = -EIO;
886                                 break;
887                         }
888                 }
889                 isize = i_size_read(inode);
890                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
891                 if (!isize || page->index > file_end ||
892                     page->mapping != inode->i_mapping) {
893                         /* whoops, we blew past eof, skip this page */
894                         unlock_page(page);
895                         page_cache_release(page);
896                         break;
897                 }
898                 pages[i] = page;
899                 i_done++;
900         }
901         if (!i_done || ret)
902                 goto out;
903
904         if (!(inode->i_sb->s_flags & MS_ACTIVE))
905                 goto out;
906
907         /*
908          * so now we have a nice long stream of locked
909          * and up to date pages, lets wait on them
910          */
911         for (i = 0; i < i_done; i++)
912                 wait_on_page_writeback(pages[i]);
913
914         page_start = page_offset(pages[0]);
915         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
916
917         lock_extent_bits(&BTRFS_I(inode)->io_tree,
918                          page_start, page_end - 1, 0, &cached_state,
919                          GFP_NOFS);
920         ordered = btrfs_lookup_first_ordered_extent(inode, page_end - 1);
921         if (ordered &&
922             ordered->file_offset + ordered->len > page_start &&
923             ordered->file_offset < page_end) {
924                 btrfs_put_ordered_extent(ordered);
925                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
926                                      page_start, page_end - 1,
927                                      &cached_state, GFP_NOFS);
928                 for (i = 0; i < i_done; i++) {
929                         unlock_page(pages[i]);
930                         page_cache_release(pages[i]);
931                 }
932                 btrfs_wait_ordered_range(inode, page_start,
933                                          page_end - page_start);
934                 goto again;
935         }
936         if (ordered)
937                 btrfs_put_ordered_extent(ordered);
938
939         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
940                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
941                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
942                           GFP_NOFS);
943
944         if (i_done != num_pages) {
945                 spin_lock(&BTRFS_I(inode)->lock);
946                 BTRFS_I(inode)->outstanding_extents++;
947                 spin_unlock(&BTRFS_I(inode)->lock);
948                 btrfs_delalloc_release_space(inode,
949                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
950         }
951
952
953         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
954                                   &cached_state);
955
956         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
957                              page_start, page_end - 1, &cached_state,
958                              GFP_NOFS);
959
960         for (i = 0; i < i_done; i++) {
961                 clear_page_dirty_for_io(pages[i]);
962                 ClearPageChecked(pages[i]);
963                 set_page_extent_mapped(pages[i]);
964                 set_page_dirty(pages[i]);
965                 unlock_page(pages[i]);
966                 page_cache_release(pages[i]);
967         }
968         return i_done;
969 out:
970         for (i = 0; i < i_done; i++) {
971                 unlock_page(pages[i]);
972                 page_cache_release(pages[i]);
973         }
974         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
975         return ret;
976
977 }
978
979 int btrfs_defrag_file(struct inode *inode, struct file *file,
980                       struct btrfs_ioctl_defrag_range_args *range,
981                       u64 newer_than, unsigned long max_to_defrag)
982 {
983         struct btrfs_root *root = BTRFS_I(inode)->root;
984         struct btrfs_super_block *disk_super;
985         struct file_ra_state *ra = NULL;
986         unsigned long last_index;
987         u64 isize = i_size_read(inode);
988         u64 features;
989         u64 last_len = 0;
990         u64 skip = 0;
991         u64 defrag_end = 0;
992         u64 newer_off = range->start;
993         unsigned long i;
994         unsigned long ra_index = 0;
995         int ret;
996         int defrag_count = 0;
997         int compress_type = BTRFS_COMPRESS_ZLIB;
998         int extent_thresh = range->extent_thresh;
999         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1000         int cluster = max_cluster;
1001         u64 new_align = ~((u64)128 * 1024 - 1);
1002         struct page **pages = NULL;
1003
1004         if (extent_thresh == 0)
1005                 extent_thresh = 256 * 1024;
1006
1007         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1008                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1009                         return -EINVAL;
1010                 if (range->compress_type)
1011                         compress_type = range->compress_type;
1012         }
1013
1014         if (isize == 0)
1015                 return 0;
1016
1017         /*
1018          * if we were not given a file, allocate a readahead
1019          * context
1020          */
1021         if (!file) {
1022                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1023                 if (!ra)
1024                         return -ENOMEM;
1025                 file_ra_state_init(ra, inode->i_mapping);
1026         } else {
1027                 ra = &file->f_ra;
1028         }
1029
1030         pages = kmalloc(sizeof(struct page *) * max_cluster,
1031                         GFP_NOFS);
1032         if (!pages) {
1033                 ret = -ENOMEM;
1034                 goto out_ra;
1035         }
1036
1037         /* find the last page to defrag */
1038         if (range->start + range->len > range->start) {
1039                 last_index = min_t(u64, isize - 1,
1040                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1041         } else {
1042                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1043         }
1044
1045         if (newer_than) {
1046                 ret = find_new_extents(root, inode, newer_than,
1047                                        &newer_off, 64 * 1024);
1048                 if (!ret) {
1049                         range->start = newer_off;
1050                         /*
1051                          * we always align our defrag to help keep
1052                          * the extents in the file evenly spaced
1053                          */
1054                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1055                 } else
1056                         goto out_ra;
1057         } else {
1058                 i = range->start >> PAGE_CACHE_SHIFT;
1059         }
1060         if (!max_to_defrag)
1061                 max_to_defrag = last_index;
1062
1063         /*
1064          * make writeback starts from i, so the defrag range can be
1065          * written sequentially.
1066          */
1067         if (i < inode->i_mapping->writeback_index)
1068                 inode->i_mapping->writeback_index = i;
1069
1070         while (i <= last_index && defrag_count < max_to_defrag &&
1071                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1072                 PAGE_CACHE_SHIFT)) {
1073                 /*
1074                  * make sure we stop running if someone unmounts
1075                  * the FS
1076                  */
1077                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1078                         break;
1079
1080                 if (!newer_than &&
1081                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1082                                         PAGE_CACHE_SIZE,
1083                                         extent_thresh,
1084                                         &last_len, &skip,
1085                                         &defrag_end)) {
1086                         unsigned long next;
1087                         /*
1088                          * the should_defrag function tells us how much to skip
1089                          * bump our counter by the suggested amount
1090                          */
1091                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1092                         i = max(i + 1, next);
1093                         continue;
1094                 }
1095
1096                 if (!newer_than) {
1097                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1098                                    PAGE_CACHE_SHIFT) - i;
1099                         cluster = min(cluster, max_cluster);
1100                 } else {
1101                         cluster = max_cluster;
1102                 }
1103
1104                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1105                         BTRFS_I(inode)->force_compress = compress_type;
1106
1107                 if (i + cluster > ra_index) {
1108                         ra_index = max(i, ra_index);
1109                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1110                                        cluster);
1111                         ra_index += max_cluster;
1112                 }
1113
1114                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1115                 if (ret < 0)
1116                         goto out_ra;
1117
1118                 defrag_count += ret;
1119                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1120
1121                 if (newer_than) {
1122                         if (newer_off == (u64)-1)
1123                                 break;
1124
1125                         newer_off = max(newer_off + 1,
1126                                         (u64)i << PAGE_CACHE_SHIFT);
1127
1128                         ret = find_new_extents(root, inode,
1129                                                newer_than, &newer_off,
1130                                                64 * 1024);
1131                         if (!ret) {
1132                                 range->start = newer_off;
1133                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1134                         } else {
1135                                 break;
1136                         }
1137                 } else {
1138                         if (ret > 0) {
1139                                 i += ret;
1140                                 last_len += ret << PAGE_CACHE_SHIFT;
1141                         } else {
1142                                 i++;
1143                                 last_len = 0;
1144                         }
1145                 }
1146         }
1147
1148         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1149                 filemap_flush(inode->i_mapping);
1150
1151         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1152                 /* the filemap_flush will queue IO into the worker threads, but
1153                  * we have to make sure the IO is actually started and that
1154                  * ordered extents get created before we return
1155                  */
1156                 atomic_inc(&root->fs_info->async_submit_draining);
1157                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1158                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1159                         wait_event(root->fs_info->async_submit_wait,
1160                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1161                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1162                 }
1163                 atomic_dec(&root->fs_info->async_submit_draining);
1164
1165                 mutex_lock(&inode->i_mutex);
1166                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1167                 mutex_unlock(&inode->i_mutex);
1168         }
1169
1170         disk_super = root->fs_info->super_copy;
1171         features = btrfs_super_incompat_flags(disk_super);
1172         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1173                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1174                 btrfs_set_super_incompat_flags(disk_super, features);
1175         }
1176
1177         ret = defrag_count;
1178
1179 out_ra:
1180         if (!file)
1181                 kfree(ra);
1182         kfree(pages);
1183         return ret;
1184 }
1185
1186 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1187                                         void __user *arg)
1188 {
1189         u64 new_size;
1190         u64 old_size;
1191         u64 devid = 1;
1192         struct btrfs_ioctl_vol_args *vol_args;
1193         struct btrfs_trans_handle *trans;
1194         struct btrfs_device *device = NULL;
1195         char *sizestr;
1196         char *devstr = NULL;
1197         int ret = 0;
1198         int mod = 0;
1199
1200         if (root->fs_info->sb->s_flags & MS_RDONLY)
1201                 return -EROFS;
1202
1203         if (!capable(CAP_SYS_ADMIN))
1204                 return -EPERM;
1205
1206         mutex_lock(&root->fs_info->volume_mutex);
1207         if (root->fs_info->balance_ctl) {
1208                 printk(KERN_INFO "btrfs: balance in progress\n");
1209                 ret = -EINVAL;
1210                 goto out;
1211         }
1212
1213         vol_args = memdup_user(arg, sizeof(*vol_args));
1214         if (IS_ERR(vol_args)) {
1215                 ret = PTR_ERR(vol_args);
1216                 goto out;
1217         }
1218
1219         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1220
1221         sizestr = vol_args->name;
1222         devstr = strchr(sizestr, ':');
1223         if (devstr) {
1224                 char *end;
1225                 sizestr = devstr + 1;
1226                 *devstr = '\0';
1227                 devstr = vol_args->name;
1228                 devid = simple_strtoull(devstr, &end, 10);
1229                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1230                        (unsigned long long)devid);
1231         }
1232         device = btrfs_find_device(root, devid, NULL, NULL);
1233         if (!device) {
1234                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1235                        (unsigned long long)devid);
1236                 ret = -EINVAL;
1237                 goto out_free;
1238         }
1239         if (!strcmp(sizestr, "max"))
1240                 new_size = device->bdev->bd_inode->i_size;
1241         else {
1242                 if (sizestr[0] == '-') {
1243                         mod = -1;
1244                         sizestr++;
1245                 } else if (sizestr[0] == '+') {
1246                         mod = 1;
1247                         sizestr++;
1248                 }
1249                 new_size = memparse(sizestr, NULL);
1250                 if (new_size == 0) {
1251                         ret = -EINVAL;
1252                         goto out_free;
1253                 }
1254         }
1255
1256         old_size = device->total_bytes;
1257
1258         if (mod < 0) {
1259                 if (new_size > old_size) {
1260                         ret = -EINVAL;
1261                         goto out_free;
1262                 }
1263                 new_size = old_size - new_size;
1264         } else if (mod > 0) {
1265                 new_size = old_size + new_size;
1266         }
1267
1268         if (new_size < 256 * 1024 * 1024) {
1269                 ret = -EINVAL;
1270                 goto out_free;
1271         }
1272         if (new_size > device->bdev->bd_inode->i_size) {
1273                 ret = -EFBIG;
1274                 goto out_free;
1275         }
1276
1277         do_div(new_size, root->sectorsize);
1278         new_size *= root->sectorsize;
1279
1280         printk(KERN_INFO "btrfs: new size for %s is %llu\n",
1281                 device->name, (unsigned long long)new_size);
1282
1283         if (new_size > old_size) {
1284                 trans = btrfs_start_transaction(root, 0);
1285                 if (IS_ERR(trans)) {
1286                         ret = PTR_ERR(trans);
1287                         goto out_free;
1288                 }
1289                 ret = btrfs_grow_device(trans, device, new_size);
1290                 btrfs_commit_transaction(trans, root);
1291         } else if (new_size < old_size) {
1292                 ret = btrfs_shrink_device(device, new_size);
1293         }
1294
1295 out_free:
1296         kfree(vol_args);
1297 out:
1298         mutex_unlock(&root->fs_info->volume_mutex);
1299         return ret;
1300 }
1301
1302 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1303                                                     char *name,
1304                                                     unsigned long fd,
1305                                                     int subvol,
1306                                                     u64 *transid,
1307                                                     bool readonly)
1308 {
1309         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1310         struct file *src_file;
1311         int namelen;
1312         int ret = 0;
1313
1314         if (root->fs_info->sb->s_flags & MS_RDONLY)
1315                 return -EROFS;
1316
1317         namelen = strlen(name);
1318         if (strchr(name, '/')) {
1319                 ret = -EINVAL;
1320                 goto out;
1321         }
1322
1323         if (subvol) {
1324                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1325                                      NULL, transid, readonly);
1326         } else {
1327                 struct inode *src_inode;
1328                 src_file = fget(fd);
1329                 if (!src_file) {
1330                         ret = -EINVAL;
1331                         goto out;
1332                 }
1333
1334                 src_inode = src_file->f_path.dentry->d_inode;
1335                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1336                         printk(KERN_INFO "btrfs: Snapshot src from "
1337                                "another FS\n");
1338                         ret = -EINVAL;
1339                         fput(src_file);
1340                         goto out;
1341                 }
1342                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1343                                      BTRFS_I(src_inode)->root,
1344                                      transid, readonly);
1345                 fput(src_file);
1346         }
1347 out:
1348         return ret;
1349 }
1350
1351 static noinline int btrfs_ioctl_snap_create(struct file *file,
1352                                             void __user *arg, int subvol)
1353 {
1354         struct btrfs_ioctl_vol_args *vol_args;
1355         int ret;
1356
1357         vol_args = memdup_user(arg, sizeof(*vol_args));
1358         if (IS_ERR(vol_args))
1359                 return PTR_ERR(vol_args);
1360         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1361
1362         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1363                                               vol_args->fd, subvol,
1364                                               NULL, false);
1365
1366         kfree(vol_args);
1367         return ret;
1368 }
1369
1370 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1371                                                void __user *arg, int subvol)
1372 {
1373         struct btrfs_ioctl_vol_args_v2 *vol_args;
1374         int ret;
1375         u64 transid = 0;
1376         u64 *ptr = NULL;
1377         bool readonly = false;
1378
1379         vol_args = memdup_user(arg, sizeof(*vol_args));
1380         if (IS_ERR(vol_args))
1381                 return PTR_ERR(vol_args);
1382         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1383
1384         if (vol_args->flags &
1385             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1386                 ret = -EOPNOTSUPP;
1387                 goto out;
1388         }
1389
1390         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1391                 ptr = &transid;
1392         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1393                 readonly = true;
1394
1395         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1396                                               vol_args->fd, subvol,
1397                                               ptr, readonly);
1398
1399         if (ret == 0 && ptr &&
1400             copy_to_user(arg +
1401                          offsetof(struct btrfs_ioctl_vol_args_v2,
1402                                   transid), ptr, sizeof(*ptr)))
1403                 ret = -EFAULT;
1404 out:
1405         kfree(vol_args);
1406         return ret;
1407 }
1408
1409 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1410                                                 void __user *arg)
1411 {
1412         struct inode *inode = fdentry(file)->d_inode;
1413         struct btrfs_root *root = BTRFS_I(inode)->root;
1414         int ret = 0;
1415         u64 flags = 0;
1416
1417         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1418                 return -EINVAL;
1419
1420         down_read(&root->fs_info->subvol_sem);
1421         if (btrfs_root_readonly(root))
1422                 flags |= BTRFS_SUBVOL_RDONLY;
1423         up_read(&root->fs_info->subvol_sem);
1424
1425         if (copy_to_user(arg, &flags, sizeof(flags)))
1426                 ret = -EFAULT;
1427
1428         return ret;
1429 }
1430
1431 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1432                                               void __user *arg)
1433 {
1434         struct inode *inode = fdentry(file)->d_inode;
1435         struct btrfs_root *root = BTRFS_I(inode)->root;
1436         struct btrfs_trans_handle *trans;
1437         u64 root_flags;
1438         u64 flags;
1439         int ret = 0;
1440
1441         if (root->fs_info->sb->s_flags & MS_RDONLY)
1442                 return -EROFS;
1443
1444         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1445                 return -EINVAL;
1446
1447         if (copy_from_user(&flags, arg, sizeof(flags)))
1448                 return -EFAULT;
1449
1450         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1451                 return -EINVAL;
1452
1453         if (flags & ~BTRFS_SUBVOL_RDONLY)
1454                 return -EOPNOTSUPP;
1455
1456         if (!inode_owner_or_capable(inode))
1457                 return -EACCES;
1458
1459         down_write(&root->fs_info->subvol_sem);
1460
1461         /* nothing to do */
1462         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1463                 goto out;
1464
1465         root_flags = btrfs_root_flags(&root->root_item);
1466         if (flags & BTRFS_SUBVOL_RDONLY)
1467                 btrfs_set_root_flags(&root->root_item,
1468                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1469         else
1470                 btrfs_set_root_flags(&root->root_item,
1471                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1472
1473         trans = btrfs_start_transaction(root, 1);
1474         if (IS_ERR(trans)) {
1475                 ret = PTR_ERR(trans);
1476                 goto out_reset;
1477         }
1478
1479         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1480                                 &root->root_key, &root->root_item);
1481
1482         btrfs_commit_transaction(trans, root);
1483 out_reset:
1484         if (ret)
1485                 btrfs_set_root_flags(&root->root_item, root_flags);
1486 out:
1487         up_write(&root->fs_info->subvol_sem);
1488         return ret;
1489 }
1490
1491 /*
1492  * helper to check if the subvolume references other subvolumes
1493  */
1494 static noinline int may_destroy_subvol(struct btrfs_root *root)
1495 {
1496         struct btrfs_path *path;
1497         struct btrfs_key key;
1498         int ret;
1499
1500         path = btrfs_alloc_path();
1501         if (!path)
1502                 return -ENOMEM;
1503
1504         key.objectid = root->root_key.objectid;
1505         key.type = BTRFS_ROOT_REF_KEY;
1506         key.offset = (u64)-1;
1507
1508         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1509                                 &key, path, 0, 0);
1510         if (ret < 0)
1511                 goto out;
1512         BUG_ON(ret == 0);
1513
1514         ret = 0;
1515         if (path->slots[0] > 0) {
1516                 path->slots[0]--;
1517                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1518                 if (key.objectid == root->root_key.objectid &&
1519                     key.type == BTRFS_ROOT_REF_KEY)
1520                         ret = -ENOTEMPTY;
1521         }
1522 out:
1523         btrfs_free_path(path);
1524         return ret;
1525 }
1526
1527 static noinline int key_in_sk(struct btrfs_key *key,
1528                               struct btrfs_ioctl_search_key *sk)
1529 {
1530         struct btrfs_key test;
1531         int ret;
1532
1533         test.objectid = sk->min_objectid;
1534         test.type = sk->min_type;
1535         test.offset = sk->min_offset;
1536
1537         ret = btrfs_comp_cpu_keys(key, &test);
1538         if (ret < 0)
1539                 return 0;
1540
1541         test.objectid = sk->max_objectid;
1542         test.type = sk->max_type;
1543         test.offset = sk->max_offset;
1544
1545         ret = btrfs_comp_cpu_keys(key, &test);
1546         if (ret > 0)
1547                 return 0;
1548         return 1;
1549 }
1550
1551 static noinline int copy_to_sk(struct btrfs_root *root,
1552                                struct btrfs_path *path,
1553                                struct btrfs_key *key,
1554                                struct btrfs_ioctl_search_key *sk,
1555                                char *buf,
1556                                unsigned long *sk_offset,
1557                                int *num_found)
1558 {
1559         u64 found_transid;
1560         struct extent_buffer *leaf;
1561         struct btrfs_ioctl_search_header sh;
1562         unsigned long item_off;
1563         unsigned long item_len;
1564         int nritems;
1565         int i;
1566         int slot;
1567         int ret = 0;
1568
1569         leaf = path->nodes[0];
1570         slot = path->slots[0];
1571         nritems = btrfs_header_nritems(leaf);
1572
1573         if (btrfs_header_generation(leaf) > sk->max_transid) {
1574                 i = nritems;
1575                 goto advance_key;
1576         }
1577         found_transid = btrfs_header_generation(leaf);
1578
1579         for (i = slot; i < nritems; i++) {
1580                 item_off = btrfs_item_ptr_offset(leaf, i);
1581                 item_len = btrfs_item_size_nr(leaf, i);
1582
1583                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1584                         item_len = 0;
1585
1586                 if (sizeof(sh) + item_len + *sk_offset >
1587                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1588                         ret = 1;
1589                         goto overflow;
1590                 }
1591
1592                 btrfs_item_key_to_cpu(leaf, key, i);
1593                 if (!key_in_sk(key, sk))
1594                         continue;
1595
1596                 sh.objectid = key->objectid;
1597                 sh.offset = key->offset;
1598                 sh.type = key->type;
1599                 sh.len = item_len;
1600                 sh.transid = found_transid;
1601
1602                 /* copy search result header */
1603                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1604                 *sk_offset += sizeof(sh);
1605
1606                 if (item_len) {
1607                         char *p = buf + *sk_offset;
1608                         /* copy the item */
1609                         read_extent_buffer(leaf, p,
1610                                            item_off, item_len);
1611                         *sk_offset += item_len;
1612                 }
1613                 (*num_found)++;
1614
1615                 if (*num_found >= sk->nr_items)
1616                         break;
1617         }
1618 advance_key:
1619         ret = 0;
1620         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1621                 key->offset++;
1622         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1623                 key->offset = 0;
1624                 key->type++;
1625         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1626                 key->offset = 0;
1627                 key->type = 0;
1628                 key->objectid++;
1629         } else
1630                 ret = 1;
1631 overflow:
1632         return ret;
1633 }
1634
1635 static noinline int search_ioctl(struct inode *inode,
1636                                  struct btrfs_ioctl_search_args *args)
1637 {
1638         struct btrfs_root *root;
1639         struct btrfs_key key;
1640         struct btrfs_key max_key;
1641         struct btrfs_path *path;
1642         struct btrfs_ioctl_search_key *sk = &args->key;
1643         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1644         int ret;
1645         int num_found = 0;
1646         unsigned long sk_offset = 0;
1647
1648         path = btrfs_alloc_path();
1649         if (!path)
1650                 return -ENOMEM;
1651
1652         if (sk->tree_id == 0) {
1653                 /* search the root of the inode that was passed */
1654                 root = BTRFS_I(inode)->root;
1655         } else {
1656                 key.objectid = sk->tree_id;
1657                 key.type = BTRFS_ROOT_ITEM_KEY;
1658                 key.offset = (u64)-1;
1659                 root = btrfs_read_fs_root_no_name(info, &key);
1660                 if (IS_ERR(root)) {
1661                         printk(KERN_ERR "could not find root %llu\n",
1662                                sk->tree_id);
1663                         btrfs_free_path(path);
1664                         return -ENOENT;
1665                 }
1666         }
1667
1668         key.objectid = sk->min_objectid;
1669         key.type = sk->min_type;
1670         key.offset = sk->min_offset;
1671
1672         max_key.objectid = sk->max_objectid;
1673         max_key.type = sk->max_type;
1674         max_key.offset = sk->max_offset;
1675
1676         path->keep_locks = 1;
1677
1678         while(1) {
1679                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1680                                            sk->min_transid);
1681                 if (ret != 0) {
1682                         if (ret > 0)
1683                                 ret = 0;
1684                         goto err;
1685                 }
1686                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1687                                  &sk_offset, &num_found);
1688                 btrfs_release_path(path);
1689                 if (ret || num_found >= sk->nr_items)
1690                         break;
1691
1692         }
1693         ret = 0;
1694 err:
1695         sk->nr_items = num_found;
1696         btrfs_free_path(path);
1697         return ret;
1698 }
1699
1700 static noinline int btrfs_ioctl_tree_search(struct file *file,
1701                                            void __user *argp)
1702 {
1703          struct btrfs_ioctl_search_args *args;
1704          struct inode *inode;
1705          int ret;
1706
1707         if (!capable(CAP_SYS_ADMIN))
1708                 return -EPERM;
1709
1710         args = memdup_user(argp, sizeof(*args));
1711         if (IS_ERR(args))
1712                 return PTR_ERR(args);
1713
1714         inode = fdentry(file)->d_inode;
1715         ret = search_ioctl(inode, args);
1716         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1717                 ret = -EFAULT;
1718         kfree(args);
1719         return ret;
1720 }
1721
1722 /*
1723  * Search INODE_REFs to identify path name of 'dirid' directory
1724  * in a 'tree_id' tree. and sets path name to 'name'.
1725  */
1726 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1727                                 u64 tree_id, u64 dirid, char *name)
1728 {
1729         struct btrfs_root *root;
1730         struct btrfs_key key;
1731         char *ptr;
1732         int ret = -1;
1733         int slot;
1734         int len;
1735         int total_len = 0;
1736         struct btrfs_inode_ref *iref;
1737         struct extent_buffer *l;
1738         struct btrfs_path *path;
1739
1740         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1741                 name[0]='\0';
1742                 return 0;
1743         }
1744
1745         path = btrfs_alloc_path();
1746         if (!path)
1747                 return -ENOMEM;
1748
1749         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1750
1751         key.objectid = tree_id;
1752         key.type = BTRFS_ROOT_ITEM_KEY;
1753         key.offset = (u64)-1;
1754         root = btrfs_read_fs_root_no_name(info, &key);
1755         if (IS_ERR(root)) {
1756                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1757                 ret = -ENOENT;
1758                 goto out;
1759         }
1760
1761         key.objectid = dirid;
1762         key.type = BTRFS_INODE_REF_KEY;
1763         key.offset = (u64)-1;
1764
1765         while(1) {
1766                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1767                 if (ret < 0)
1768                         goto out;
1769
1770                 l = path->nodes[0];
1771                 slot = path->slots[0];
1772                 if (ret > 0 && slot > 0)
1773                         slot--;
1774                 btrfs_item_key_to_cpu(l, &key, slot);
1775
1776                 if (ret > 0 && (key.objectid != dirid ||
1777                                 key.type != BTRFS_INODE_REF_KEY)) {
1778                         ret = -ENOENT;
1779                         goto out;
1780                 }
1781
1782                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1783                 len = btrfs_inode_ref_name_len(l, iref);
1784                 ptr -= len + 1;
1785                 total_len += len + 1;
1786                 if (ptr < name)
1787                         goto out;
1788
1789                 *(ptr + len) = '/';
1790                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1791
1792                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1793                         break;
1794
1795                 btrfs_release_path(path);
1796                 key.objectid = key.offset;
1797                 key.offset = (u64)-1;
1798                 dirid = key.objectid;
1799         }
1800         if (ptr < name)
1801                 goto out;
1802         memmove(name, ptr, total_len);
1803         name[total_len]='\0';
1804         ret = 0;
1805 out:
1806         btrfs_free_path(path);
1807         return ret;
1808 }
1809
1810 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1811                                            void __user *argp)
1812 {
1813          struct btrfs_ioctl_ino_lookup_args *args;
1814          struct inode *inode;
1815          int ret;
1816
1817         if (!capable(CAP_SYS_ADMIN))
1818                 return -EPERM;
1819
1820         args = memdup_user(argp, sizeof(*args));
1821         if (IS_ERR(args))
1822                 return PTR_ERR(args);
1823
1824         inode = fdentry(file)->d_inode;
1825
1826         if (args->treeid == 0)
1827                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1828
1829         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1830                                         args->treeid, args->objectid,
1831                                         args->name);
1832
1833         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1834                 ret = -EFAULT;
1835
1836         kfree(args);
1837         return ret;
1838 }
1839
1840 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1841                                              void __user *arg)
1842 {
1843         struct dentry *parent = fdentry(file);
1844         struct dentry *dentry;
1845         struct inode *dir = parent->d_inode;
1846         struct inode *inode;
1847         struct btrfs_root *root = BTRFS_I(dir)->root;
1848         struct btrfs_root *dest = NULL;
1849         struct btrfs_ioctl_vol_args *vol_args;
1850         struct btrfs_trans_handle *trans;
1851         int namelen;
1852         int ret;
1853         int err = 0;
1854
1855         vol_args = memdup_user(arg, sizeof(*vol_args));
1856         if (IS_ERR(vol_args))
1857                 return PTR_ERR(vol_args);
1858
1859         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1860         namelen = strlen(vol_args->name);
1861         if (strchr(vol_args->name, '/') ||
1862             strncmp(vol_args->name, "..", namelen) == 0) {
1863                 err = -EINVAL;
1864                 goto out;
1865         }
1866
1867         err = mnt_want_write(file->f_path.mnt);
1868         if (err)
1869                 goto out;
1870
1871         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1872         dentry = lookup_one_len(vol_args->name, parent, namelen);
1873         if (IS_ERR(dentry)) {
1874                 err = PTR_ERR(dentry);
1875                 goto out_unlock_dir;
1876         }
1877
1878         if (!dentry->d_inode) {
1879                 err = -ENOENT;
1880                 goto out_dput;
1881         }
1882
1883         inode = dentry->d_inode;
1884         dest = BTRFS_I(inode)->root;
1885         if (!capable(CAP_SYS_ADMIN)){
1886                 /*
1887                  * Regular user.  Only allow this with a special mount
1888                  * option, when the user has write+exec access to the
1889                  * subvol root, and when rmdir(2) would have been
1890                  * allowed.
1891                  *
1892                  * Note that this is _not_ check that the subvol is
1893                  * empty or doesn't contain data that we wouldn't
1894                  * otherwise be able to delete.
1895                  *
1896                  * Users who want to delete empty subvols should try
1897                  * rmdir(2).
1898                  */
1899                 err = -EPERM;
1900                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1901                         goto out_dput;
1902
1903                 /*
1904                  * Do not allow deletion if the parent dir is the same
1905                  * as the dir to be deleted.  That means the ioctl
1906                  * must be called on the dentry referencing the root
1907                  * of the subvol, not a random directory contained
1908                  * within it.
1909                  */
1910                 err = -EINVAL;
1911                 if (root == dest)
1912                         goto out_dput;
1913
1914                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1915                 if (err)
1916                         goto out_dput;
1917
1918                 /* check if subvolume may be deleted by a non-root user */
1919                 err = btrfs_may_delete(dir, dentry, 1);
1920                 if (err)
1921                         goto out_dput;
1922         }
1923
1924         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1925                 err = -EINVAL;
1926                 goto out_dput;
1927         }
1928
1929         mutex_lock(&inode->i_mutex);
1930         err = d_invalidate(dentry);
1931         if (err)
1932                 goto out_unlock;
1933
1934         down_write(&root->fs_info->subvol_sem);
1935
1936         err = may_destroy_subvol(dest);
1937         if (err)
1938                 goto out_up_write;
1939
1940         trans = btrfs_start_transaction(root, 0);
1941         if (IS_ERR(trans)) {
1942                 err = PTR_ERR(trans);
1943                 goto out_up_write;
1944         }
1945         trans->block_rsv = &root->fs_info->global_block_rsv;
1946
1947         ret = btrfs_unlink_subvol(trans, root, dir,
1948                                 dest->root_key.objectid,
1949                                 dentry->d_name.name,
1950                                 dentry->d_name.len);
1951         BUG_ON(ret);
1952
1953         btrfs_record_root_in_trans(trans, dest);
1954
1955         memset(&dest->root_item.drop_progress, 0,
1956                 sizeof(dest->root_item.drop_progress));
1957         dest->root_item.drop_level = 0;
1958         btrfs_set_root_refs(&dest->root_item, 0);
1959
1960         if (!xchg(&dest->orphan_item_inserted, 1)) {
1961                 ret = btrfs_insert_orphan_item(trans,
1962                                         root->fs_info->tree_root,
1963                                         dest->root_key.objectid);
1964                 BUG_ON(ret);
1965         }
1966
1967         ret = btrfs_end_transaction(trans, root);
1968         BUG_ON(ret);
1969         inode->i_flags |= S_DEAD;
1970 out_up_write:
1971         up_write(&root->fs_info->subvol_sem);
1972 out_unlock:
1973         mutex_unlock(&inode->i_mutex);
1974         if (!err) {
1975                 shrink_dcache_sb(root->fs_info->sb);
1976                 btrfs_invalidate_inodes(dest);
1977                 d_delete(dentry);
1978         }
1979 out_dput:
1980         dput(dentry);
1981 out_unlock_dir:
1982         mutex_unlock(&dir->i_mutex);
1983         mnt_drop_write(file->f_path.mnt);
1984 out:
1985         kfree(vol_args);
1986         return err;
1987 }
1988
1989 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1990 {
1991         struct inode *inode = fdentry(file)->d_inode;
1992         struct btrfs_root *root = BTRFS_I(inode)->root;
1993         struct btrfs_ioctl_defrag_range_args *range;
1994         int ret;
1995
1996         if (btrfs_root_readonly(root))
1997                 return -EROFS;
1998
1999         ret = mnt_want_write(file->f_path.mnt);
2000         if (ret)
2001                 return ret;
2002
2003         switch (inode->i_mode & S_IFMT) {
2004         case S_IFDIR:
2005                 if (!capable(CAP_SYS_ADMIN)) {
2006                         ret = -EPERM;
2007                         goto out;
2008                 }
2009                 ret = btrfs_defrag_root(root, 0);
2010                 if (ret)
2011                         goto out;
2012                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2013                 break;
2014         case S_IFREG:
2015                 if (!(file->f_mode & FMODE_WRITE)) {
2016                         ret = -EINVAL;
2017                         goto out;
2018                 }
2019
2020                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2021                 if (!range) {
2022                         ret = -ENOMEM;
2023                         goto out;
2024                 }
2025
2026                 if (argp) {
2027                         if (copy_from_user(range, argp,
2028                                            sizeof(*range))) {
2029                                 ret = -EFAULT;
2030                                 kfree(range);
2031                                 goto out;
2032                         }
2033                         /* compression requires us to start the IO */
2034                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2035                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2036                                 range->extent_thresh = (u32)-1;
2037                         }
2038                 } else {
2039                         /* the rest are all set to zero by kzalloc */
2040                         range->len = (u64)-1;
2041                 }
2042                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2043                                         range, 0, 0);
2044                 if (ret > 0)
2045                         ret = 0;
2046                 kfree(range);
2047                 break;
2048         default:
2049                 ret = -EINVAL;
2050         }
2051 out:
2052         mnt_drop_write(file->f_path.mnt);
2053         return ret;
2054 }
2055
2056 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2057 {
2058         struct btrfs_ioctl_vol_args *vol_args;
2059         int ret;
2060
2061         if (!capable(CAP_SYS_ADMIN))
2062                 return -EPERM;
2063
2064         mutex_lock(&root->fs_info->volume_mutex);
2065         if (root->fs_info->balance_ctl) {
2066                 printk(KERN_INFO "btrfs: balance in progress\n");
2067                 ret = -EINVAL;
2068                 goto out;
2069         }
2070
2071         vol_args = memdup_user(arg, sizeof(*vol_args));
2072         if (IS_ERR(vol_args)) {
2073                 ret = PTR_ERR(vol_args);
2074                 goto out;
2075         }
2076
2077         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2078         ret = btrfs_init_new_device(root, vol_args->name);
2079
2080         kfree(vol_args);
2081 out:
2082         mutex_unlock(&root->fs_info->volume_mutex);
2083         return ret;
2084 }
2085
2086 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2087 {
2088         struct btrfs_ioctl_vol_args *vol_args;
2089         int ret;
2090
2091         if (!capable(CAP_SYS_ADMIN))
2092                 return -EPERM;
2093
2094         if (root->fs_info->sb->s_flags & MS_RDONLY)
2095                 return -EROFS;
2096
2097         mutex_lock(&root->fs_info->volume_mutex);
2098         if (root->fs_info->balance_ctl) {
2099                 printk(KERN_INFO "btrfs: balance in progress\n");
2100                 ret = -EINVAL;
2101                 goto out;
2102         }
2103
2104         vol_args = memdup_user(arg, sizeof(*vol_args));
2105         if (IS_ERR(vol_args)) {
2106                 ret = PTR_ERR(vol_args);
2107                 goto out;
2108         }
2109
2110         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2111         ret = btrfs_rm_device(root, vol_args->name);
2112
2113         kfree(vol_args);
2114 out:
2115         mutex_unlock(&root->fs_info->volume_mutex);
2116         return ret;
2117 }
2118
2119 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2120 {
2121         struct btrfs_ioctl_fs_info_args *fi_args;
2122         struct btrfs_device *device;
2123         struct btrfs_device *next;
2124         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2125         int ret = 0;
2126
2127         if (!capable(CAP_SYS_ADMIN))
2128                 return -EPERM;
2129
2130         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2131         if (!fi_args)
2132                 return -ENOMEM;
2133
2134         fi_args->num_devices = fs_devices->num_devices;
2135         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2136
2137         mutex_lock(&fs_devices->device_list_mutex);
2138         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2139                 if (device->devid > fi_args->max_id)
2140                         fi_args->max_id = device->devid;
2141         }
2142         mutex_unlock(&fs_devices->device_list_mutex);
2143
2144         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2145                 ret = -EFAULT;
2146
2147         kfree(fi_args);
2148         return ret;
2149 }
2150
2151 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2152 {
2153         struct btrfs_ioctl_dev_info_args *di_args;
2154         struct btrfs_device *dev;
2155         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2156         int ret = 0;
2157         char *s_uuid = NULL;
2158         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2159
2160         if (!capable(CAP_SYS_ADMIN))
2161                 return -EPERM;
2162
2163         di_args = memdup_user(arg, sizeof(*di_args));
2164         if (IS_ERR(di_args))
2165                 return PTR_ERR(di_args);
2166
2167         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2168                 s_uuid = di_args->uuid;
2169
2170         mutex_lock(&fs_devices->device_list_mutex);
2171         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2172         mutex_unlock(&fs_devices->device_list_mutex);
2173
2174         if (!dev) {
2175                 ret = -ENODEV;
2176                 goto out;
2177         }
2178
2179         di_args->devid = dev->devid;
2180         di_args->bytes_used = dev->bytes_used;
2181         di_args->total_bytes = dev->total_bytes;
2182         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2183         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2184
2185 out:
2186         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2187                 ret = -EFAULT;
2188
2189         kfree(di_args);
2190         return ret;
2191 }
2192
2193 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2194                                        u64 off, u64 olen, u64 destoff)
2195 {
2196         struct inode *inode = fdentry(file)->d_inode;
2197         struct btrfs_root *root = BTRFS_I(inode)->root;
2198         struct file *src_file;
2199         struct inode *src;
2200         struct btrfs_trans_handle *trans;
2201         struct btrfs_path *path;
2202         struct extent_buffer *leaf;
2203         char *buf;
2204         struct btrfs_key key;
2205         u32 nritems;
2206         int slot;
2207         int ret;
2208         u64 len = olen;
2209         u64 bs = root->fs_info->sb->s_blocksize;
2210         u64 hint_byte;
2211
2212         /*
2213          * TODO:
2214          * - split compressed inline extents.  annoying: we need to
2215          *   decompress into destination's address_space (the file offset
2216          *   may change, so source mapping won't do), then recompress (or
2217          *   otherwise reinsert) a subrange.
2218          * - allow ranges within the same file to be cloned (provided
2219          *   they don't overlap)?
2220          */
2221
2222         /* the destination must be opened for writing */
2223         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2224                 return -EINVAL;
2225
2226         if (btrfs_root_readonly(root))
2227                 return -EROFS;
2228
2229         ret = mnt_want_write(file->f_path.mnt);
2230         if (ret)
2231                 return ret;
2232
2233         src_file = fget(srcfd);
2234         if (!src_file) {
2235                 ret = -EBADF;
2236                 goto out_drop_write;
2237         }
2238
2239         src = src_file->f_dentry->d_inode;
2240
2241         ret = -EINVAL;
2242         if (src == inode)
2243                 goto out_fput;
2244
2245         /* the src must be open for reading */
2246         if (!(src_file->f_mode & FMODE_READ))
2247                 goto out_fput;
2248
2249         /* don't make the dst file partly checksummed */
2250         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2251             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2252                 goto out_fput;
2253
2254         ret = -EISDIR;
2255         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2256                 goto out_fput;
2257
2258         ret = -EXDEV;
2259         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2260                 goto out_fput;
2261
2262         ret = -ENOMEM;
2263         buf = vmalloc(btrfs_level_size(root, 0));
2264         if (!buf)
2265                 goto out_fput;
2266
2267         path = btrfs_alloc_path();
2268         if (!path) {
2269                 vfree(buf);
2270                 goto out_fput;
2271         }
2272         path->reada = 2;
2273
2274         if (inode < src) {
2275                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2276                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2277         } else {
2278                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2279                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2280         }
2281
2282         /* determine range to clone */
2283         ret = -EINVAL;
2284         if (off + len > src->i_size || off + len < off)
2285                 goto out_unlock;
2286         if (len == 0)
2287                 olen = len = src->i_size - off;
2288         /* if we extend to eof, continue to block boundary */
2289         if (off + len == src->i_size)
2290                 len = ALIGN(src->i_size, bs) - off;
2291
2292         /* verify the end result is block aligned */
2293         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2294             !IS_ALIGNED(destoff, bs))
2295                 goto out_unlock;
2296
2297         if (destoff > inode->i_size) {
2298                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2299                 if (ret)
2300                         goto out_unlock;
2301         }
2302
2303         /* truncate page cache pages from target inode range */
2304         truncate_inode_pages_range(&inode->i_data, destoff,
2305                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2306
2307         /* do any pending delalloc/csum calc on src, one way or
2308            another, and lock file content */
2309         while (1) {
2310                 struct btrfs_ordered_extent *ordered;
2311                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2312                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2313                 if (!ordered &&
2314                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2315                                    EXTENT_DELALLOC, 0, NULL))
2316                         break;
2317                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2318                 if (ordered)
2319                         btrfs_put_ordered_extent(ordered);
2320                 btrfs_wait_ordered_range(src, off, len);
2321         }
2322
2323         /* clone data */
2324         key.objectid = btrfs_ino(src);
2325         key.type = BTRFS_EXTENT_DATA_KEY;
2326         key.offset = 0;
2327
2328         while (1) {
2329                 /*
2330                  * note the key will change type as we walk through the
2331                  * tree.
2332                  */
2333                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2334                 if (ret < 0)
2335                         goto out;
2336
2337                 nritems = btrfs_header_nritems(path->nodes[0]);
2338                 if (path->slots[0] >= nritems) {
2339                         ret = btrfs_next_leaf(root, path);
2340                         if (ret < 0)
2341                                 goto out;
2342                         if (ret > 0)
2343                                 break;
2344                         nritems = btrfs_header_nritems(path->nodes[0]);
2345                 }
2346                 leaf = path->nodes[0];
2347                 slot = path->slots[0];
2348
2349                 btrfs_item_key_to_cpu(leaf, &key, slot);
2350                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2351                     key.objectid != btrfs_ino(src))
2352                         break;
2353
2354                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2355                         struct btrfs_file_extent_item *extent;
2356                         int type;
2357                         u32 size;
2358                         struct btrfs_key new_key;
2359                         u64 disko = 0, diskl = 0;
2360                         u64 datao = 0, datal = 0;
2361                         u8 comp;
2362                         u64 endoff;
2363
2364                         size = btrfs_item_size_nr(leaf, slot);
2365                         read_extent_buffer(leaf, buf,
2366                                            btrfs_item_ptr_offset(leaf, slot),
2367                                            size);
2368
2369                         extent = btrfs_item_ptr(leaf, slot,
2370                                                 struct btrfs_file_extent_item);
2371                         comp = btrfs_file_extent_compression(leaf, extent);
2372                         type = btrfs_file_extent_type(leaf, extent);
2373                         if (type == BTRFS_FILE_EXTENT_REG ||
2374                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2375                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2376                                                                       extent);
2377                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2378                                                                  extent);
2379                                 datao = btrfs_file_extent_offset(leaf, extent);
2380                                 datal = btrfs_file_extent_num_bytes(leaf,
2381                                                                     extent);
2382                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2383                                 /* take upper bound, may be compressed */
2384                                 datal = btrfs_file_extent_ram_bytes(leaf,
2385                                                                     extent);
2386                         }
2387                         btrfs_release_path(path);
2388
2389                         if (key.offset + datal <= off ||
2390                             key.offset >= off+len)
2391                                 goto next;
2392
2393                         memcpy(&new_key, &key, sizeof(new_key));
2394                         new_key.objectid = btrfs_ino(inode);
2395                         if (off <= key.offset)
2396                                 new_key.offset = key.offset + destoff - off;
2397                         else
2398                                 new_key.offset = destoff;
2399
2400                         /*
2401                          * 1 - adjusting old extent (we may have to split it)
2402                          * 1 - add new extent
2403                          * 1 - inode update
2404                          */
2405                         trans = btrfs_start_transaction(root, 3);
2406                         if (IS_ERR(trans)) {
2407                                 ret = PTR_ERR(trans);
2408                                 goto out;
2409                         }
2410
2411                         if (type == BTRFS_FILE_EXTENT_REG ||
2412                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2413                                 /*
2414                                  *    a  | --- range to clone ---|  b
2415                                  * | ------------- extent ------------- |
2416                                  */
2417
2418                                 /* substract range b */
2419                                 if (key.offset + datal > off + len)
2420                                         datal = off + len - key.offset;
2421
2422                                 /* substract range a */
2423                                 if (off > key.offset) {
2424                                         datao += off - key.offset;
2425                                         datal -= off - key.offset;
2426                                 }
2427
2428                                 ret = btrfs_drop_extents(trans, inode,
2429                                                          new_key.offset,
2430                                                          new_key.offset + datal,
2431                                                          &hint_byte, 1);
2432                                 BUG_ON(ret);
2433
2434                                 ret = btrfs_insert_empty_item(trans, root, path,
2435                                                               &new_key, size);
2436                                 BUG_ON(ret);
2437
2438                                 leaf = path->nodes[0];
2439                                 slot = path->slots[0];
2440                                 write_extent_buffer(leaf, buf,
2441                                             btrfs_item_ptr_offset(leaf, slot),
2442                                             size);
2443
2444                                 extent = btrfs_item_ptr(leaf, slot,
2445                                                 struct btrfs_file_extent_item);
2446
2447                                 /* disko == 0 means it's a hole */
2448                                 if (!disko)
2449                                         datao = 0;
2450
2451                                 btrfs_set_file_extent_offset(leaf, extent,
2452                                                              datao);
2453                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2454                                                                 datal);
2455                                 if (disko) {
2456                                         inode_add_bytes(inode, datal);
2457                                         ret = btrfs_inc_extent_ref(trans, root,
2458                                                         disko, diskl, 0,
2459                                                         root->root_key.objectid,
2460                                                         btrfs_ino(inode),
2461                                                         new_key.offset - datao);
2462                                         BUG_ON(ret);
2463                                 }
2464                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2465                                 u64 skip = 0;
2466                                 u64 trim = 0;
2467                                 if (off > key.offset) {
2468                                         skip = off - key.offset;
2469                                         new_key.offset += skip;
2470                                 }
2471
2472                                 if (key.offset + datal > off+len)
2473                                         trim = key.offset + datal - (off+len);
2474
2475                                 if (comp && (skip || trim)) {
2476                                         ret = -EINVAL;
2477                                         btrfs_end_transaction(trans, root);
2478                                         goto out;
2479                                 }
2480                                 size -= skip + trim;
2481                                 datal -= skip + trim;
2482
2483                                 ret = btrfs_drop_extents(trans, inode,
2484                                                          new_key.offset,
2485                                                          new_key.offset + datal,
2486                                                          &hint_byte, 1);
2487                                 BUG_ON(ret);
2488
2489                                 ret = btrfs_insert_empty_item(trans, root, path,
2490                                                               &new_key, size);
2491                                 BUG_ON(ret);
2492
2493                                 if (skip) {
2494                                         u32 start =
2495                                           btrfs_file_extent_calc_inline_size(0);
2496                                         memmove(buf+start, buf+start+skip,
2497                                                 datal);
2498                                 }
2499
2500                                 leaf = path->nodes[0];
2501                                 slot = path->slots[0];
2502                                 write_extent_buffer(leaf, buf,
2503                                             btrfs_item_ptr_offset(leaf, slot),
2504                                             size);
2505                                 inode_add_bytes(inode, datal);
2506                         }
2507
2508                         btrfs_mark_buffer_dirty(leaf);
2509                         btrfs_release_path(path);
2510
2511                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2512
2513                         /*
2514                          * we round up to the block size at eof when
2515                          * determining which extents to clone above,
2516                          * but shouldn't round up the file size
2517                          */
2518                         endoff = new_key.offset + datal;
2519                         if (endoff > destoff+olen)
2520                                 endoff = destoff+olen;
2521                         if (endoff > inode->i_size)
2522                                 btrfs_i_size_write(inode, endoff);
2523
2524                         ret = btrfs_update_inode(trans, root, inode);
2525                         BUG_ON(ret);
2526                         btrfs_end_transaction(trans, root);
2527                 }
2528 next:
2529                 btrfs_release_path(path);
2530                 key.offset++;
2531         }
2532         ret = 0;
2533 out:
2534         btrfs_release_path(path);
2535         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2536 out_unlock:
2537         mutex_unlock(&src->i_mutex);
2538         mutex_unlock(&inode->i_mutex);
2539         vfree(buf);
2540         btrfs_free_path(path);
2541 out_fput:
2542         fput(src_file);
2543 out_drop_write:
2544         mnt_drop_write(file->f_path.mnt);
2545         return ret;
2546 }
2547
2548 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2549 {
2550         struct btrfs_ioctl_clone_range_args args;
2551
2552         if (copy_from_user(&args, argp, sizeof(args)))
2553                 return -EFAULT;
2554         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2555                                  args.src_length, args.dest_offset);
2556 }
2557
2558 /*
2559  * there are many ways the trans_start and trans_end ioctls can lead
2560  * to deadlocks.  They should only be used by applications that
2561  * basically own the machine, and have a very in depth understanding
2562  * of all the possible deadlocks and enospc problems.
2563  */
2564 static long btrfs_ioctl_trans_start(struct file *file)
2565 {
2566         struct inode *inode = fdentry(file)->d_inode;
2567         struct btrfs_root *root = BTRFS_I(inode)->root;
2568         struct btrfs_trans_handle *trans;
2569         int ret;
2570
2571         ret = -EPERM;
2572         if (!capable(CAP_SYS_ADMIN))
2573                 goto out;
2574
2575         ret = -EINPROGRESS;
2576         if (file->private_data)
2577                 goto out;
2578
2579         ret = -EROFS;
2580         if (btrfs_root_readonly(root))
2581                 goto out;
2582
2583         ret = mnt_want_write(file->f_path.mnt);
2584         if (ret)
2585                 goto out;
2586
2587         atomic_inc(&root->fs_info->open_ioctl_trans);
2588
2589         ret = -ENOMEM;
2590         trans = btrfs_start_ioctl_transaction(root);
2591         if (IS_ERR(trans))
2592                 goto out_drop;
2593
2594         file->private_data = trans;
2595         return 0;
2596
2597 out_drop:
2598         atomic_dec(&root->fs_info->open_ioctl_trans);
2599         mnt_drop_write(file->f_path.mnt);
2600 out:
2601         return ret;
2602 }
2603
2604 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2605 {
2606         struct inode *inode = fdentry(file)->d_inode;
2607         struct btrfs_root *root = BTRFS_I(inode)->root;
2608         struct btrfs_root *new_root;
2609         struct btrfs_dir_item *di;
2610         struct btrfs_trans_handle *trans;
2611         struct btrfs_path *path;
2612         struct btrfs_key location;
2613         struct btrfs_disk_key disk_key;
2614         struct btrfs_super_block *disk_super;
2615         u64 features;
2616         u64 objectid = 0;
2617         u64 dir_id;
2618
2619         if (!capable(CAP_SYS_ADMIN))
2620                 return -EPERM;
2621
2622         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2623                 return -EFAULT;
2624
2625         if (!objectid)
2626                 objectid = root->root_key.objectid;
2627
2628         location.objectid = objectid;
2629         location.type = BTRFS_ROOT_ITEM_KEY;
2630         location.offset = (u64)-1;
2631
2632         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2633         if (IS_ERR(new_root))
2634                 return PTR_ERR(new_root);
2635
2636         if (btrfs_root_refs(&new_root->root_item) == 0)
2637                 return -ENOENT;
2638
2639         path = btrfs_alloc_path();
2640         if (!path)
2641                 return -ENOMEM;
2642         path->leave_spinning = 1;
2643
2644         trans = btrfs_start_transaction(root, 1);
2645         if (IS_ERR(trans)) {
2646                 btrfs_free_path(path);
2647                 return PTR_ERR(trans);
2648         }
2649
2650         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2651         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2652                                    dir_id, "default", 7, 1);
2653         if (IS_ERR_OR_NULL(di)) {
2654                 btrfs_free_path(path);
2655                 btrfs_end_transaction(trans, root);
2656                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2657                        "this isn't going to work\n");
2658                 return -ENOENT;
2659         }
2660
2661         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2662         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2663         btrfs_mark_buffer_dirty(path->nodes[0]);
2664         btrfs_free_path(path);
2665
2666         disk_super = root->fs_info->super_copy;
2667         features = btrfs_super_incompat_flags(disk_super);
2668         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2669                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2670                 btrfs_set_super_incompat_flags(disk_super, features);
2671         }
2672         btrfs_end_transaction(trans, root);
2673
2674         return 0;
2675 }
2676
2677 static void get_block_group_info(struct list_head *groups_list,
2678                                  struct btrfs_ioctl_space_info *space)
2679 {
2680         struct btrfs_block_group_cache *block_group;
2681
2682         space->total_bytes = 0;
2683         space->used_bytes = 0;
2684         space->flags = 0;
2685         list_for_each_entry(block_group, groups_list, list) {
2686                 space->flags = block_group->flags;
2687                 space->total_bytes += block_group->key.offset;
2688                 space->used_bytes +=
2689                         btrfs_block_group_used(&block_group->item);
2690         }
2691 }
2692
2693 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2694 {
2695         struct btrfs_ioctl_space_args space_args;
2696         struct btrfs_ioctl_space_info space;
2697         struct btrfs_ioctl_space_info *dest;
2698         struct btrfs_ioctl_space_info *dest_orig;
2699         struct btrfs_ioctl_space_info __user *user_dest;
2700         struct btrfs_space_info *info;
2701         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2702                        BTRFS_BLOCK_GROUP_SYSTEM,
2703                        BTRFS_BLOCK_GROUP_METADATA,
2704                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2705         int num_types = 4;
2706         int alloc_size;
2707         int ret = 0;
2708         u64 slot_count = 0;
2709         int i, c;
2710
2711         if (copy_from_user(&space_args,
2712                            (struct btrfs_ioctl_space_args __user *)arg,
2713                            sizeof(space_args)))
2714                 return -EFAULT;
2715
2716         for (i = 0; i < num_types; i++) {
2717                 struct btrfs_space_info *tmp;
2718
2719                 info = NULL;
2720                 rcu_read_lock();
2721                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2722                                         list) {
2723                         if (tmp->flags == types[i]) {
2724                                 info = tmp;
2725                                 break;
2726                         }
2727                 }
2728                 rcu_read_unlock();
2729
2730                 if (!info)
2731                         continue;
2732
2733                 down_read(&info->groups_sem);
2734                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2735                         if (!list_empty(&info->block_groups[c]))
2736                                 slot_count++;
2737                 }
2738                 up_read(&info->groups_sem);
2739         }
2740
2741         /* space_slots == 0 means they are asking for a count */
2742         if (space_args.space_slots == 0) {
2743                 space_args.total_spaces = slot_count;
2744                 goto out;
2745         }
2746
2747         slot_count = min_t(u64, space_args.space_slots, slot_count);
2748
2749         alloc_size = sizeof(*dest) * slot_count;
2750
2751         /* we generally have at most 6 or so space infos, one for each raid
2752          * level.  So, a whole page should be more than enough for everyone
2753          */
2754         if (alloc_size > PAGE_CACHE_SIZE)
2755                 return -ENOMEM;
2756
2757         space_args.total_spaces = 0;
2758         dest = kmalloc(alloc_size, GFP_NOFS);
2759         if (!dest)
2760                 return -ENOMEM;
2761         dest_orig = dest;
2762
2763         /* now we have a buffer to copy into */
2764         for (i = 0; i < num_types; i++) {
2765                 struct btrfs_space_info *tmp;
2766
2767                 if (!slot_count)
2768                         break;
2769
2770                 info = NULL;
2771                 rcu_read_lock();
2772                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2773                                         list) {
2774                         if (tmp->flags == types[i]) {
2775                                 info = tmp;
2776                                 break;
2777                         }
2778                 }
2779                 rcu_read_unlock();
2780
2781                 if (!info)
2782                         continue;
2783                 down_read(&info->groups_sem);
2784                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2785                         if (!list_empty(&info->block_groups[c])) {
2786                                 get_block_group_info(&info->block_groups[c],
2787                                                      &space);
2788                                 memcpy(dest, &space, sizeof(space));
2789                                 dest++;
2790                                 space_args.total_spaces++;
2791                                 slot_count--;
2792                         }
2793                         if (!slot_count)
2794                                 break;
2795                 }
2796                 up_read(&info->groups_sem);
2797         }
2798
2799         user_dest = (struct btrfs_ioctl_space_info *)
2800                 (arg + sizeof(struct btrfs_ioctl_space_args));
2801
2802         if (copy_to_user(user_dest, dest_orig, alloc_size))
2803                 ret = -EFAULT;
2804
2805         kfree(dest_orig);
2806 out:
2807         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2808                 ret = -EFAULT;
2809
2810         return ret;
2811 }
2812
2813 /*
2814  * there are many ways the trans_start and trans_end ioctls can lead
2815  * to deadlocks.  They should only be used by applications that
2816  * basically own the machine, and have a very in depth understanding
2817  * of all the possible deadlocks and enospc problems.
2818  */
2819 long btrfs_ioctl_trans_end(struct file *file)
2820 {
2821         struct inode *inode = fdentry(file)->d_inode;
2822         struct btrfs_root *root = BTRFS_I(inode)->root;
2823         struct btrfs_trans_handle *trans;
2824
2825         trans = file->private_data;
2826         if (!trans)
2827                 return -EINVAL;
2828         file->private_data = NULL;
2829
2830         btrfs_end_transaction(trans, root);
2831
2832         atomic_dec(&root->fs_info->open_ioctl_trans);
2833
2834         mnt_drop_write(file->f_path.mnt);
2835         return 0;
2836 }
2837
2838 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2839 {
2840         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2841         struct btrfs_trans_handle *trans;
2842         u64 transid;
2843         int ret;
2844
2845         trans = btrfs_start_transaction(root, 0);
2846         if (IS_ERR(trans))
2847                 return PTR_ERR(trans);
2848         transid = trans->transid;
2849         ret = btrfs_commit_transaction_async(trans, root, 0);
2850         if (ret) {
2851                 btrfs_end_transaction(trans, root);
2852                 return ret;
2853         }
2854
2855         if (argp)
2856                 if (copy_to_user(argp, &transid, sizeof(transid)))
2857                         return -EFAULT;
2858         return 0;
2859 }
2860
2861 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2862 {
2863         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2864         u64 transid;
2865
2866         if (argp) {
2867                 if (copy_from_user(&transid, argp, sizeof(transid)))
2868                         return -EFAULT;
2869         } else {
2870                 transid = 0;  /* current trans */
2871         }
2872         return btrfs_wait_for_commit(root, transid);
2873 }
2874
2875 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2876 {
2877         int ret;
2878         struct btrfs_ioctl_scrub_args *sa;
2879
2880         if (!capable(CAP_SYS_ADMIN))
2881                 return -EPERM;
2882
2883         sa = memdup_user(arg, sizeof(*sa));
2884         if (IS_ERR(sa))
2885                 return PTR_ERR(sa);
2886
2887         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2888                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2889
2890         if (copy_to_user(arg, sa, sizeof(*sa)))
2891                 ret = -EFAULT;
2892
2893         kfree(sa);
2894         return ret;
2895 }
2896
2897 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2898 {
2899         if (!capable(CAP_SYS_ADMIN))
2900                 return -EPERM;
2901
2902         return btrfs_scrub_cancel(root);
2903 }
2904
2905 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2906                                        void __user *arg)
2907 {
2908         struct btrfs_ioctl_scrub_args *sa;
2909         int ret;
2910
2911         if (!capable(CAP_SYS_ADMIN))
2912                 return -EPERM;
2913
2914         sa = memdup_user(arg, sizeof(*sa));
2915         if (IS_ERR(sa))
2916                 return PTR_ERR(sa);
2917
2918         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2919
2920         if (copy_to_user(arg, sa, sizeof(*sa)))
2921                 ret = -EFAULT;
2922
2923         kfree(sa);
2924         return ret;
2925 }
2926
2927 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
2928 {
2929         int ret = 0;
2930         int i;
2931         u64 rel_ptr;
2932         int size;
2933         struct btrfs_ioctl_ino_path_args *ipa = NULL;
2934         struct inode_fs_paths *ipath = NULL;
2935         struct btrfs_path *path;
2936
2937         if (!capable(CAP_SYS_ADMIN))
2938                 return -EPERM;
2939
2940         path = btrfs_alloc_path();
2941         if (!path) {
2942                 ret = -ENOMEM;
2943                 goto out;
2944         }
2945
2946         ipa = memdup_user(arg, sizeof(*ipa));
2947         if (IS_ERR(ipa)) {
2948                 ret = PTR_ERR(ipa);
2949                 ipa = NULL;
2950                 goto out;
2951         }
2952
2953         size = min_t(u32, ipa->size, 4096);
2954         ipath = init_ipath(size, root, path);
2955         if (IS_ERR(ipath)) {
2956                 ret = PTR_ERR(ipath);
2957                 ipath = NULL;
2958                 goto out;
2959         }
2960
2961         ret = paths_from_inode(ipa->inum, ipath);
2962         if (ret < 0)
2963                 goto out;
2964
2965         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
2966                 rel_ptr = ipath->fspath->val[i] -
2967                           (u64)(unsigned long)ipath->fspath->val;
2968                 ipath->fspath->val[i] = rel_ptr;
2969         }
2970
2971         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
2972                            (void *)(unsigned long)ipath->fspath, size);
2973         if (ret) {
2974                 ret = -EFAULT;
2975                 goto out;
2976         }
2977
2978 out:
2979         btrfs_free_path(path);
2980         free_ipath(ipath);
2981         kfree(ipa);
2982
2983         return ret;
2984 }
2985
2986 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2987 {
2988         struct btrfs_data_container *inodes = ctx;
2989         const size_t c = 3 * sizeof(u64);
2990
2991         if (inodes->bytes_left >= c) {
2992                 inodes->bytes_left -= c;
2993                 inodes->val[inodes->elem_cnt] = inum;
2994                 inodes->val[inodes->elem_cnt + 1] = offset;
2995                 inodes->val[inodes->elem_cnt + 2] = root;
2996                 inodes->elem_cnt += 3;
2997         } else {
2998                 inodes->bytes_missing += c - inodes->bytes_left;
2999                 inodes->bytes_left = 0;
3000                 inodes->elem_missed += 3;
3001         }
3002
3003         return 0;
3004 }
3005
3006 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3007                                         void __user *arg)
3008 {
3009         int ret = 0;
3010         int size;
3011         u64 extent_offset;
3012         struct btrfs_ioctl_logical_ino_args *loi;
3013         struct btrfs_data_container *inodes = NULL;
3014         struct btrfs_path *path = NULL;
3015         struct btrfs_key key;
3016
3017         if (!capable(CAP_SYS_ADMIN))
3018                 return -EPERM;
3019
3020         loi = memdup_user(arg, sizeof(*loi));
3021         if (IS_ERR(loi)) {
3022                 ret = PTR_ERR(loi);
3023                 loi = NULL;
3024                 goto out;
3025         }
3026
3027         path = btrfs_alloc_path();
3028         if (!path) {
3029                 ret = -ENOMEM;
3030                 goto out;
3031         }
3032
3033         size = min_t(u32, loi->size, 4096);
3034         inodes = init_data_container(size);
3035         if (IS_ERR(inodes)) {
3036                 ret = PTR_ERR(inodes);
3037                 inodes = NULL;
3038                 goto out;
3039         }
3040
3041         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
3042
3043         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
3044                 ret = -ENOENT;
3045         if (ret < 0)
3046                 goto out;
3047
3048         extent_offset = loi->logical - key.objectid;
3049         ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
3050                                         extent_offset, build_ino_list, inodes);
3051
3052         if (ret < 0)
3053                 goto out;
3054
3055         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3056                            (void *)(unsigned long)inodes, size);
3057         if (ret)
3058                 ret = -EFAULT;
3059
3060 out:
3061         btrfs_free_path(path);
3062         kfree(inodes);
3063         kfree(loi);
3064
3065         return ret;
3066 }
3067
3068 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3069                                struct btrfs_ioctl_balance_args *bargs)
3070 {
3071         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3072
3073         bargs->flags = bctl->flags;
3074
3075         if (atomic_read(&fs_info->balance_running))
3076                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3077         if (atomic_read(&fs_info->balance_pause_req))
3078                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3079         if (atomic_read(&fs_info->balance_cancel_req))
3080                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3081
3082         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3083         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3084         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3085
3086         if (lock) {
3087                 spin_lock(&fs_info->balance_lock);
3088                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3089                 spin_unlock(&fs_info->balance_lock);
3090         } else {
3091                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3092         }
3093 }
3094
3095 static long btrfs_ioctl_balance(struct btrfs_root *root, void __user *arg)
3096 {
3097         struct btrfs_fs_info *fs_info = root->fs_info;
3098         struct btrfs_ioctl_balance_args *bargs;
3099         struct btrfs_balance_control *bctl;
3100         int ret;
3101
3102         if (!capable(CAP_SYS_ADMIN))
3103                 return -EPERM;
3104
3105         if (fs_info->sb->s_flags & MS_RDONLY)
3106                 return -EROFS;
3107
3108         mutex_lock(&fs_info->volume_mutex);
3109         mutex_lock(&fs_info->balance_mutex);
3110
3111         if (arg) {
3112                 bargs = memdup_user(arg, sizeof(*bargs));
3113                 if (IS_ERR(bargs)) {
3114                         ret = PTR_ERR(bargs);
3115                         goto out;
3116                 }
3117
3118                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3119                         if (!fs_info->balance_ctl) {
3120                                 ret = -ENOTCONN;
3121                                 goto out_bargs;
3122                         }
3123
3124                         bctl = fs_info->balance_ctl;
3125                         spin_lock(&fs_info->balance_lock);
3126                         bctl->flags |= BTRFS_BALANCE_RESUME;
3127                         spin_unlock(&fs_info->balance_lock);
3128
3129                         goto do_balance;
3130                 }
3131         } else {
3132                 bargs = NULL;
3133         }
3134
3135         if (fs_info->balance_ctl) {
3136                 ret = -EINPROGRESS;
3137                 goto out_bargs;
3138         }
3139
3140         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3141         if (!bctl) {
3142                 ret = -ENOMEM;
3143                 goto out_bargs;
3144         }
3145
3146         bctl->fs_info = fs_info;
3147         if (arg) {
3148                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3149                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3150                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3151
3152                 bctl->flags = bargs->flags;
3153         } else {
3154                 /* balance everything - no filters */
3155                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3156         }
3157
3158 do_balance:
3159         ret = btrfs_balance(bctl, bargs);
3160         /*
3161          * bctl is freed in __cancel_balance or in free_fs_info if
3162          * restriper was paused all the way until unmount
3163          */
3164         if (arg) {
3165                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3166                         ret = -EFAULT;
3167         }
3168
3169 out_bargs:
3170         kfree(bargs);
3171 out:
3172         mutex_unlock(&fs_info->balance_mutex);
3173         mutex_unlock(&fs_info->volume_mutex);
3174         return ret;
3175 }
3176
3177 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3178 {
3179         if (!capable(CAP_SYS_ADMIN))
3180                 return -EPERM;
3181
3182         switch (cmd) {
3183         case BTRFS_BALANCE_CTL_PAUSE:
3184                 return btrfs_pause_balance(root->fs_info);
3185         case BTRFS_BALANCE_CTL_CANCEL:
3186                 return btrfs_cancel_balance(root->fs_info);
3187         }
3188
3189         return -EINVAL;
3190 }
3191
3192 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3193                                          void __user *arg)
3194 {
3195         struct btrfs_fs_info *fs_info = root->fs_info;
3196         struct btrfs_ioctl_balance_args *bargs;
3197         int ret = 0;
3198
3199         if (!capable(CAP_SYS_ADMIN))
3200                 return -EPERM;
3201
3202         mutex_lock(&fs_info->balance_mutex);
3203         if (!fs_info->balance_ctl) {
3204                 ret = -ENOTCONN;
3205                 goto out;
3206         }
3207
3208         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3209         if (!bargs) {
3210                 ret = -ENOMEM;
3211                 goto out;
3212         }
3213
3214         update_ioctl_balance_args(fs_info, 1, bargs);
3215
3216         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3217                 ret = -EFAULT;
3218
3219         kfree(bargs);
3220 out:
3221         mutex_unlock(&fs_info->balance_mutex);
3222         return ret;
3223 }
3224
3225 long btrfs_ioctl(struct file *file, unsigned int
3226                 cmd, unsigned long arg)
3227 {
3228         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3229         void __user *argp = (void __user *)arg;
3230
3231         switch (cmd) {
3232         case FS_IOC_GETFLAGS:
3233                 return btrfs_ioctl_getflags(file, argp);
3234         case FS_IOC_SETFLAGS:
3235                 return btrfs_ioctl_setflags(file, argp);
3236         case FS_IOC_GETVERSION:
3237                 return btrfs_ioctl_getversion(file, argp);
3238         case FITRIM:
3239                 return btrfs_ioctl_fitrim(file, argp);
3240         case BTRFS_IOC_SNAP_CREATE:
3241                 return btrfs_ioctl_snap_create(file, argp, 0);
3242         case BTRFS_IOC_SNAP_CREATE_V2:
3243                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3244         case BTRFS_IOC_SUBVOL_CREATE:
3245                 return btrfs_ioctl_snap_create(file, argp, 1);
3246         case BTRFS_IOC_SNAP_DESTROY:
3247                 return btrfs_ioctl_snap_destroy(file, argp);
3248         case BTRFS_IOC_SUBVOL_GETFLAGS:
3249                 return btrfs_ioctl_subvol_getflags(file, argp);
3250         case BTRFS_IOC_SUBVOL_SETFLAGS:
3251                 return btrfs_ioctl_subvol_setflags(file, argp);
3252         case BTRFS_IOC_DEFAULT_SUBVOL:
3253                 return btrfs_ioctl_default_subvol(file, argp);
3254         case BTRFS_IOC_DEFRAG:
3255                 return btrfs_ioctl_defrag(file, NULL);
3256         case BTRFS_IOC_DEFRAG_RANGE:
3257                 return btrfs_ioctl_defrag(file, argp);
3258         case BTRFS_IOC_RESIZE:
3259                 return btrfs_ioctl_resize(root, argp);
3260         case BTRFS_IOC_ADD_DEV:
3261                 return btrfs_ioctl_add_dev(root, argp);
3262         case BTRFS_IOC_RM_DEV:
3263                 return btrfs_ioctl_rm_dev(root, argp);
3264         case BTRFS_IOC_FS_INFO:
3265                 return btrfs_ioctl_fs_info(root, argp);
3266         case BTRFS_IOC_DEV_INFO:
3267                 return btrfs_ioctl_dev_info(root, argp);
3268         case BTRFS_IOC_BALANCE:
3269                 return btrfs_ioctl_balance(root, NULL);
3270         case BTRFS_IOC_CLONE:
3271                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3272         case BTRFS_IOC_CLONE_RANGE:
3273                 return btrfs_ioctl_clone_range(file, argp);
3274         case BTRFS_IOC_TRANS_START:
3275                 return btrfs_ioctl_trans_start(file);
3276         case BTRFS_IOC_TRANS_END:
3277                 return btrfs_ioctl_trans_end(file);
3278         case BTRFS_IOC_TREE_SEARCH:
3279                 return btrfs_ioctl_tree_search(file, argp);
3280         case BTRFS_IOC_INO_LOOKUP:
3281                 return btrfs_ioctl_ino_lookup(file, argp);
3282         case BTRFS_IOC_INO_PATHS:
3283                 return btrfs_ioctl_ino_to_path(root, argp);
3284         case BTRFS_IOC_LOGICAL_INO:
3285                 return btrfs_ioctl_logical_to_ino(root, argp);
3286         case BTRFS_IOC_SPACE_INFO:
3287                 return btrfs_ioctl_space_info(root, argp);
3288         case BTRFS_IOC_SYNC:
3289                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3290                 return 0;
3291         case BTRFS_IOC_START_SYNC:
3292                 return btrfs_ioctl_start_sync(file, argp);
3293         case BTRFS_IOC_WAIT_SYNC:
3294                 return btrfs_ioctl_wait_sync(file, argp);
3295         case BTRFS_IOC_SCRUB:
3296                 return btrfs_ioctl_scrub(root, argp);
3297         case BTRFS_IOC_SCRUB_CANCEL:
3298                 return btrfs_ioctl_scrub_cancel(root, argp);
3299         case BTRFS_IOC_SCRUB_PROGRESS:
3300                 return btrfs_ioctl_scrub_progress(root, argp);
3301         case BTRFS_IOC_BALANCE_V2:
3302                 return btrfs_ioctl_balance(root, argp);
3303         case BTRFS_IOC_BALANCE_CTL:
3304                 return btrfs_ioctl_balance_ctl(root, arg);
3305         case BTRFS_IOC_BALANCE_PROGRESS:
3306                 return btrfs_ioctl_balance_progress(root, argp);
3307         }
3308
3309         return -ENOTTY;
3310 }