Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux...
[~shefty/rdma-dev.git] / fs / btrfs / ioctl.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/fsnotify.h>
25 #include <linux/pagemap.h>
26 #include <linux/highmem.h>
27 #include <linux/time.h>
28 #include <linux/init.h>
29 #include <linux/string.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include "compat.h"
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "ioctl.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 #include "inode-map.h"
54 #include "backref.h"
55
56 /* Mask out flags that are inappropriate for the given type of inode. */
57 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
58 {
59         if (S_ISDIR(mode))
60                 return flags;
61         else if (S_ISREG(mode))
62                 return flags & ~FS_DIRSYNC_FL;
63         else
64                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
65 }
66
67 /*
68  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
69  */
70 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
71 {
72         unsigned int iflags = 0;
73
74         if (flags & BTRFS_INODE_SYNC)
75                 iflags |= FS_SYNC_FL;
76         if (flags & BTRFS_INODE_IMMUTABLE)
77                 iflags |= FS_IMMUTABLE_FL;
78         if (flags & BTRFS_INODE_APPEND)
79                 iflags |= FS_APPEND_FL;
80         if (flags & BTRFS_INODE_NODUMP)
81                 iflags |= FS_NODUMP_FL;
82         if (flags & BTRFS_INODE_NOATIME)
83                 iflags |= FS_NOATIME_FL;
84         if (flags & BTRFS_INODE_DIRSYNC)
85                 iflags |= FS_DIRSYNC_FL;
86         if (flags & BTRFS_INODE_NODATACOW)
87                 iflags |= FS_NOCOW_FL;
88
89         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
90                 iflags |= FS_COMPR_FL;
91         else if (flags & BTRFS_INODE_NOCOMPRESS)
92                 iflags |= FS_NOCOMP_FL;
93
94         return iflags;
95 }
96
97 /*
98  * Update inode->i_flags based on the btrfs internal flags.
99  */
100 void btrfs_update_iflags(struct inode *inode)
101 {
102         struct btrfs_inode *ip = BTRFS_I(inode);
103
104         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
105
106         if (ip->flags & BTRFS_INODE_SYNC)
107                 inode->i_flags |= S_SYNC;
108         if (ip->flags & BTRFS_INODE_IMMUTABLE)
109                 inode->i_flags |= S_IMMUTABLE;
110         if (ip->flags & BTRFS_INODE_APPEND)
111                 inode->i_flags |= S_APPEND;
112         if (ip->flags & BTRFS_INODE_NOATIME)
113                 inode->i_flags |= S_NOATIME;
114         if (ip->flags & BTRFS_INODE_DIRSYNC)
115                 inode->i_flags |= S_DIRSYNC;
116 }
117
118 /*
119  * Inherit flags from the parent inode.
120  *
121  * Currently only the compression flags and the cow flags are inherited.
122  */
123 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
124 {
125         unsigned int flags;
126
127         if (!dir)
128                 return;
129
130         flags = BTRFS_I(dir)->flags;
131
132         if (flags & BTRFS_INODE_NOCOMPRESS) {
133                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
134                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
135         } else if (flags & BTRFS_INODE_COMPRESS) {
136                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
137                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
138         }
139
140         if (flags & BTRFS_INODE_NODATACOW)
141                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
142
143         btrfs_update_iflags(inode);
144 }
145
146 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
147 {
148         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
149         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
150
151         if (copy_to_user(arg, &flags, sizeof(flags)))
152                 return -EFAULT;
153         return 0;
154 }
155
156 static int check_flags(unsigned int flags)
157 {
158         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
159                       FS_NOATIME_FL | FS_NODUMP_FL | \
160                       FS_SYNC_FL | FS_DIRSYNC_FL | \
161                       FS_NOCOMP_FL | FS_COMPR_FL |
162                       FS_NOCOW_FL))
163                 return -EOPNOTSUPP;
164
165         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
166                 return -EINVAL;
167
168         return 0;
169 }
170
171 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
172 {
173         struct inode *inode = file->f_path.dentry->d_inode;
174         struct btrfs_inode *ip = BTRFS_I(inode);
175         struct btrfs_root *root = ip->root;
176         struct btrfs_trans_handle *trans;
177         unsigned int flags, oldflags;
178         int ret;
179         u64 ip_oldflags;
180         unsigned int i_oldflags;
181
182         if (btrfs_root_readonly(root))
183                 return -EROFS;
184
185         if (copy_from_user(&flags, arg, sizeof(flags)))
186                 return -EFAULT;
187
188         ret = check_flags(flags);
189         if (ret)
190                 return ret;
191
192         if (!inode_owner_or_capable(inode))
193                 return -EACCES;
194
195         mutex_lock(&inode->i_mutex);
196
197         ip_oldflags = ip->flags;
198         i_oldflags = inode->i_flags;
199
200         flags = btrfs_mask_flags(inode->i_mode, flags);
201         oldflags = btrfs_flags_to_ioctl(ip->flags);
202         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
203                 if (!capable(CAP_LINUX_IMMUTABLE)) {
204                         ret = -EPERM;
205                         goto out_unlock;
206                 }
207         }
208
209         ret = mnt_want_write_file(file);
210         if (ret)
211                 goto out_unlock;
212
213         if (flags & FS_SYNC_FL)
214                 ip->flags |= BTRFS_INODE_SYNC;
215         else
216                 ip->flags &= ~BTRFS_INODE_SYNC;
217         if (flags & FS_IMMUTABLE_FL)
218                 ip->flags |= BTRFS_INODE_IMMUTABLE;
219         else
220                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
221         if (flags & FS_APPEND_FL)
222                 ip->flags |= BTRFS_INODE_APPEND;
223         else
224                 ip->flags &= ~BTRFS_INODE_APPEND;
225         if (flags & FS_NODUMP_FL)
226                 ip->flags |= BTRFS_INODE_NODUMP;
227         else
228                 ip->flags &= ~BTRFS_INODE_NODUMP;
229         if (flags & FS_NOATIME_FL)
230                 ip->flags |= BTRFS_INODE_NOATIME;
231         else
232                 ip->flags &= ~BTRFS_INODE_NOATIME;
233         if (flags & FS_DIRSYNC_FL)
234                 ip->flags |= BTRFS_INODE_DIRSYNC;
235         else
236                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
237         if (flags & FS_NOCOW_FL)
238                 ip->flags |= BTRFS_INODE_NODATACOW;
239         else
240                 ip->flags &= ~BTRFS_INODE_NODATACOW;
241
242         /*
243          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
244          * flag may be changed automatically if compression code won't make
245          * things smaller.
246          */
247         if (flags & FS_NOCOMP_FL) {
248                 ip->flags &= ~BTRFS_INODE_COMPRESS;
249                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
250         } else if (flags & FS_COMPR_FL) {
251                 ip->flags |= BTRFS_INODE_COMPRESS;
252                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
253         } else {
254                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
255         }
256
257         trans = btrfs_start_transaction(root, 1);
258         if (IS_ERR(trans)) {
259                 ret = PTR_ERR(trans);
260                 goto out_drop;
261         }
262
263         btrfs_update_iflags(inode);
264         inode->i_ctime = CURRENT_TIME;
265         ret = btrfs_update_inode(trans, root, inode);
266
267         btrfs_end_transaction(trans, root);
268  out_drop:
269         if (ret) {
270                 ip->flags = ip_oldflags;
271                 inode->i_flags = i_oldflags;
272         }
273
274         mnt_drop_write_file(file);
275  out_unlock:
276         mutex_unlock(&inode->i_mutex);
277         return ret;
278 }
279
280 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
281 {
282         struct inode *inode = file->f_path.dentry->d_inode;
283
284         return put_user(inode->i_generation, arg);
285 }
286
287 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
288 {
289         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
290         struct btrfs_device *device;
291         struct request_queue *q;
292         struct fstrim_range range;
293         u64 minlen = ULLONG_MAX;
294         u64 num_devices = 0;
295         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
296         int ret;
297
298         if (!capable(CAP_SYS_ADMIN))
299                 return -EPERM;
300
301         rcu_read_lock();
302         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
303                                 dev_list) {
304                 if (!device->bdev)
305                         continue;
306                 q = bdev_get_queue(device->bdev);
307                 if (blk_queue_discard(q)) {
308                         num_devices++;
309                         minlen = min((u64)q->limits.discard_granularity,
310                                      minlen);
311                 }
312         }
313         rcu_read_unlock();
314
315         if (!num_devices)
316                 return -EOPNOTSUPP;
317         if (copy_from_user(&range, arg, sizeof(range)))
318                 return -EFAULT;
319         if (range.start > total_bytes)
320                 return -EINVAL;
321
322         range.len = min(range.len, total_bytes - range.start);
323         range.minlen = max(range.minlen, minlen);
324         ret = btrfs_trim_fs(fs_info->tree_root, &range);
325         if (ret < 0)
326                 return ret;
327
328         if (copy_to_user(arg, &range, sizeof(range)))
329                 return -EFAULT;
330
331         return 0;
332 }
333
334 static noinline int create_subvol(struct btrfs_root *root,
335                                   struct dentry *dentry,
336                                   char *name, int namelen,
337                                   u64 *async_transid)
338 {
339         struct btrfs_trans_handle *trans;
340         struct btrfs_key key;
341         struct btrfs_root_item root_item;
342         struct btrfs_inode_item *inode_item;
343         struct extent_buffer *leaf;
344         struct btrfs_root *new_root;
345         struct dentry *parent = dentry->d_parent;
346         struct inode *dir;
347         int ret;
348         int err;
349         u64 objectid;
350         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
351         u64 index = 0;
352
353         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
354         if (ret)
355                 return ret;
356
357         dir = parent->d_inode;
358
359         /*
360          * 1 - inode item
361          * 2 - refs
362          * 1 - root item
363          * 2 - dir items
364          */
365         trans = btrfs_start_transaction(root, 6);
366         if (IS_ERR(trans))
367                 return PTR_ERR(trans);
368
369         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
370                                       0, objectid, NULL, 0, 0, 0, 0);
371         if (IS_ERR(leaf)) {
372                 ret = PTR_ERR(leaf);
373                 goto fail;
374         }
375
376         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
377         btrfs_set_header_bytenr(leaf, leaf->start);
378         btrfs_set_header_generation(leaf, trans->transid);
379         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
380         btrfs_set_header_owner(leaf, objectid);
381
382         write_extent_buffer(leaf, root->fs_info->fsid,
383                             (unsigned long)btrfs_header_fsid(leaf),
384                             BTRFS_FSID_SIZE);
385         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
386                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
387                             BTRFS_UUID_SIZE);
388         btrfs_mark_buffer_dirty(leaf);
389
390         inode_item = &root_item.inode;
391         memset(inode_item, 0, sizeof(*inode_item));
392         inode_item->generation = cpu_to_le64(1);
393         inode_item->size = cpu_to_le64(3);
394         inode_item->nlink = cpu_to_le32(1);
395         inode_item->nbytes = cpu_to_le64(root->leafsize);
396         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
397
398         root_item.flags = 0;
399         root_item.byte_limit = 0;
400         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
401
402         btrfs_set_root_bytenr(&root_item, leaf->start);
403         btrfs_set_root_generation(&root_item, trans->transid);
404         btrfs_set_root_level(&root_item, 0);
405         btrfs_set_root_refs(&root_item, 1);
406         btrfs_set_root_used(&root_item, leaf->len);
407         btrfs_set_root_last_snapshot(&root_item, 0);
408
409         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
410         root_item.drop_level = 0;
411
412         btrfs_tree_unlock(leaf);
413         free_extent_buffer(leaf);
414         leaf = NULL;
415
416         btrfs_set_root_dirid(&root_item, new_dirid);
417
418         key.objectid = objectid;
419         key.offset = 0;
420         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
421         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
422                                 &root_item);
423         if (ret)
424                 goto fail;
425
426         key.offset = (u64)-1;
427         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
428         BUG_ON(IS_ERR(new_root));
429
430         btrfs_record_root_in_trans(trans, new_root);
431
432         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
433         /*
434          * insert the directory item
435          */
436         ret = btrfs_set_inode_index(dir, &index);
437         BUG_ON(ret);
438
439         ret = btrfs_insert_dir_item(trans, root,
440                                     name, namelen, dir, &key,
441                                     BTRFS_FT_DIR, index);
442         if (ret)
443                 goto fail;
444
445         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
446         ret = btrfs_update_inode(trans, root, dir);
447         BUG_ON(ret);
448
449         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
450                                  objectid, root->root_key.objectid,
451                                  btrfs_ino(dir), index, name, namelen);
452
453         BUG_ON(ret);
454
455         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
456 fail:
457         if (async_transid) {
458                 *async_transid = trans->transid;
459                 err = btrfs_commit_transaction_async(trans, root, 1);
460         } else {
461                 err = btrfs_commit_transaction(trans, root);
462         }
463         if (err && !ret)
464                 ret = err;
465         return ret;
466 }
467
468 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
469                            char *name, int namelen, u64 *async_transid,
470                            bool readonly)
471 {
472         struct inode *inode;
473         struct btrfs_pending_snapshot *pending_snapshot;
474         struct btrfs_trans_handle *trans;
475         int ret;
476
477         if (!root->ref_cows)
478                 return -EINVAL;
479
480         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
481         if (!pending_snapshot)
482                 return -ENOMEM;
483
484         btrfs_init_block_rsv(&pending_snapshot->block_rsv);
485         pending_snapshot->dentry = dentry;
486         pending_snapshot->root = root;
487         pending_snapshot->readonly = readonly;
488
489         trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
490         if (IS_ERR(trans)) {
491                 ret = PTR_ERR(trans);
492                 goto fail;
493         }
494
495         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
496         BUG_ON(ret);
497
498         spin_lock(&root->fs_info->trans_lock);
499         list_add(&pending_snapshot->list,
500                  &trans->transaction->pending_snapshots);
501         spin_unlock(&root->fs_info->trans_lock);
502         if (async_transid) {
503                 *async_transid = trans->transid;
504                 ret = btrfs_commit_transaction_async(trans,
505                                      root->fs_info->extent_root, 1);
506         } else {
507                 ret = btrfs_commit_transaction(trans,
508                                                root->fs_info->extent_root);
509         }
510         BUG_ON(ret);
511
512         ret = pending_snapshot->error;
513         if (ret)
514                 goto fail;
515
516         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
517         if (ret)
518                 goto fail;
519
520         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
521         if (IS_ERR(inode)) {
522                 ret = PTR_ERR(inode);
523                 goto fail;
524         }
525         BUG_ON(!inode);
526         d_instantiate(dentry, inode);
527         ret = 0;
528 fail:
529         kfree(pending_snapshot);
530         return ret;
531 }
532
533 /*  copy of check_sticky in fs/namei.c()
534 * It's inline, so penalty for filesystems that don't use sticky bit is
535 * minimal.
536 */
537 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
538 {
539         uid_t fsuid = current_fsuid();
540
541         if (!(dir->i_mode & S_ISVTX))
542                 return 0;
543         if (inode->i_uid == fsuid)
544                 return 0;
545         if (dir->i_uid == fsuid)
546                 return 0;
547         return !capable(CAP_FOWNER);
548 }
549
550 /*  copy of may_delete in fs/namei.c()
551  *      Check whether we can remove a link victim from directory dir, check
552  *  whether the type of victim is right.
553  *  1. We can't do it if dir is read-only (done in permission())
554  *  2. We should have write and exec permissions on dir
555  *  3. We can't remove anything from append-only dir
556  *  4. We can't do anything with immutable dir (done in permission())
557  *  5. If the sticky bit on dir is set we should either
558  *      a. be owner of dir, or
559  *      b. be owner of victim, or
560  *      c. have CAP_FOWNER capability
561  *  6. If the victim is append-only or immutable we can't do antyhing with
562  *     links pointing to it.
563  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
564  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
565  *  9. We can't remove a root or mountpoint.
566  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
567  *     nfs_async_unlink().
568  */
569
570 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
571 {
572         int error;
573
574         if (!victim->d_inode)
575                 return -ENOENT;
576
577         BUG_ON(victim->d_parent->d_inode != dir);
578         audit_inode_child(victim, dir);
579
580         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
581         if (error)
582                 return error;
583         if (IS_APPEND(dir))
584                 return -EPERM;
585         if (btrfs_check_sticky(dir, victim->d_inode)||
586                 IS_APPEND(victim->d_inode)||
587             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
588                 return -EPERM;
589         if (isdir) {
590                 if (!S_ISDIR(victim->d_inode->i_mode))
591                         return -ENOTDIR;
592                 if (IS_ROOT(victim))
593                         return -EBUSY;
594         } else if (S_ISDIR(victim->d_inode->i_mode))
595                 return -EISDIR;
596         if (IS_DEADDIR(dir))
597                 return -ENOENT;
598         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
599                 return -EBUSY;
600         return 0;
601 }
602
603 /* copy of may_create in fs/namei.c() */
604 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
605 {
606         if (child->d_inode)
607                 return -EEXIST;
608         if (IS_DEADDIR(dir))
609                 return -ENOENT;
610         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
611 }
612
613 /*
614  * Create a new subvolume below @parent.  This is largely modeled after
615  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
616  * inside this filesystem so it's quite a bit simpler.
617  */
618 static noinline int btrfs_mksubvol(struct path *parent,
619                                    char *name, int namelen,
620                                    struct btrfs_root *snap_src,
621                                    u64 *async_transid, bool readonly)
622 {
623         struct inode *dir  = parent->dentry->d_inode;
624         struct dentry *dentry;
625         int error;
626
627         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
628
629         dentry = lookup_one_len(name, parent->dentry, namelen);
630         error = PTR_ERR(dentry);
631         if (IS_ERR(dentry))
632                 goto out_unlock;
633
634         error = -EEXIST;
635         if (dentry->d_inode)
636                 goto out_dput;
637
638         error = mnt_want_write(parent->mnt);
639         if (error)
640                 goto out_dput;
641
642         error = btrfs_may_create(dir, dentry);
643         if (error)
644                 goto out_drop_write;
645
646         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
647
648         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
649                 goto out_up_read;
650
651         if (snap_src) {
652                 error = create_snapshot(snap_src, dentry,
653                                         name, namelen, async_transid, readonly);
654         } else {
655                 error = create_subvol(BTRFS_I(dir)->root, dentry,
656                                       name, namelen, async_transid);
657         }
658         if (!error)
659                 fsnotify_mkdir(dir, dentry);
660 out_up_read:
661         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
662 out_drop_write:
663         mnt_drop_write(parent->mnt);
664 out_dput:
665         dput(dentry);
666 out_unlock:
667         mutex_unlock(&dir->i_mutex);
668         return error;
669 }
670
671 /*
672  * When we're defragging a range, we don't want to kick it off again
673  * if it is really just waiting for delalloc to send it down.
674  * If we find a nice big extent or delalloc range for the bytes in the
675  * file you want to defrag, we return 0 to let you know to skip this
676  * part of the file
677  */
678 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
679 {
680         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
681         struct extent_map *em = NULL;
682         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
683         u64 end;
684
685         read_lock(&em_tree->lock);
686         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
687         read_unlock(&em_tree->lock);
688
689         if (em) {
690                 end = extent_map_end(em);
691                 free_extent_map(em);
692                 if (end - offset > thresh)
693                         return 0;
694         }
695         /* if we already have a nice delalloc here, just stop */
696         thresh /= 2;
697         end = count_range_bits(io_tree, &offset, offset + thresh,
698                                thresh, EXTENT_DELALLOC, 1);
699         if (end >= thresh)
700                 return 0;
701         return 1;
702 }
703
704 /*
705  * helper function to walk through a file and find extents
706  * newer than a specific transid, and smaller than thresh.
707  *
708  * This is used by the defragging code to find new and small
709  * extents
710  */
711 static int find_new_extents(struct btrfs_root *root,
712                             struct inode *inode, u64 newer_than,
713                             u64 *off, int thresh)
714 {
715         struct btrfs_path *path;
716         struct btrfs_key min_key;
717         struct btrfs_key max_key;
718         struct extent_buffer *leaf;
719         struct btrfs_file_extent_item *extent;
720         int type;
721         int ret;
722         u64 ino = btrfs_ino(inode);
723
724         path = btrfs_alloc_path();
725         if (!path)
726                 return -ENOMEM;
727
728         min_key.objectid = ino;
729         min_key.type = BTRFS_EXTENT_DATA_KEY;
730         min_key.offset = *off;
731
732         max_key.objectid = ino;
733         max_key.type = (u8)-1;
734         max_key.offset = (u64)-1;
735
736         path->keep_locks = 1;
737
738         while(1) {
739                 ret = btrfs_search_forward(root, &min_key, &max_key,
740                                            path, 0, newer_than);
741                 if (ret != 0)
742                         goto none;
743                 if (min_key.objectid != ino)
744                         goto none;
745                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
746                         goto none;
747
748                 leaf = path->nodes[0];
749                 extent = btrfs_item_ptr(leaf, path->slots[0],
750                                         struct btrfs_file_extent_item);
751
752                 type = btrfs_file_extent_type(leaf, extent);
753                 if (type == BTRFS_FILE_EXTENT_REG &&
754                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
755                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
756                         *off = min_key.offset;
757                         btrfs_free_path(path);
758                         return 0;
759                 }
760
761                 if (min_key.offset == (u64)-1)
762                         goto none;
763
764                 min_key.offset++;
765                 btrfs_release_path(path);
766         }
767 none:
768         btrfs_free_path(path);
769         return -ENOENT;
770 }
771
772 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
773                                int thresh, u64 *last_len, u64 *skip,
774                                u64 *defrag_end)
775 {
776         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
777         struct extent_map *em = NULL;
778         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
779         int ret = 1;
780
781         /*
782          * make sure that once we start defragging an extent, we keep on
783          * defragging it
784          */
785         if (start < *defrag_end)
786                 return 1;
787
788         *skip = 0;
789
790         /*
791          * hopefully we have this extent in the tree already, try without
792          * the full extent lock
793          */
794         read_lock(&em_tree->lock);
795         em = lookup_extent_mapping(em_tree, start, len);
796         read_unlock(&em_tree->lock);
797
798         if (!em) {
799                 /* get the big lock and read metadata off disk */
800                 lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
801                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
802                 unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
803
804                 if (IS_ERR(em))
805                         return 0;
806         }
807
808         /* this will cover holes, and inline extents */
809         if (em->block_start >= EXTENT_MAP_LAST_BYTE)
810                 ret = 0;
811
812         /*
813          * we hit a real extent, if it is big don't bother defragging it again
814          */
815         if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
816                 ret = 0;
817
818         /*
819          * last_len ends up being a counter of how many bytes we've defragged.
820          * every time we choose not to defrag an extent, we reset *last_len
821          * so that the next tiny extent will force a defrag.
822          *
823          * The end result of this is that tiny extents before a single big
824          * extent will force at least part of that big extent to be defragged.
825          */
826         if (ret) {
827                 *defrag_end = extent_map_end(em);
828         } else {
829                 *last_len = 0;
830                 *skip = extent_map_end(em);
831                 *defrag_end = 0;
832         }
833
834         free_extent_map(em);
835         return ret;
836 }
837
838 /*
839  * it doesn't do much good to defrag one or two pages
840  * at a time.  This pulls in a nice chunk of pages
841  * to COW and defrag.
842  *
843  * It also makes sure the delalloc code has enough
844  * dirty data to avoid making new small extents as part
845  * of the defrag
846  *
847  * It's a good idea to start RA on this range
848  * before calling this.
849  */
850 static int cluster_pages_for_defrag(struct inode *inode,
851                                     struct page **pages,
852                                     unsigned long start_index,
853                                     int num_pages)
854 {
855         unsigned long file_end;
856         u64 isize = i_size_read(inode);
857         u64 page_start;
858         u64 page_end;
859         int ret;
860         int i;
861         int i_done;
862         struct btrfs_ordered_extent *ordered;
863         struct extent_state *cached_state = NULL;
864         struct extent_io_tree *tree;
865         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
866
867         if (isize == 0)
868                 return 0;
869         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
870
871         ret = btrfs_delalloc_reserve_space(inode,
872                                            num_pages << PAGE_CACHE_SHIFT);
873         if (ret)
874                 return ret;
875         i_done = 0;
876         tree = &BTRFS_I(inode)->io_tree;
877
878         /* step one, lock all the pages */
879         for (i = 0; i < num_pages; i++) {
880                 struct page *page;
881 again:
882                 page = find_or_create_page(inode->i_mapping,
883                                            start_index + i, mask);
884                 if (!page)
885                         break;
886
887                 page_start = page_offset(page);
888                 page_end = page_start + PAGE_CACHE_SIZE - 1;
889                 while (1) {
890                         lock_extent(tree, page_start, page_end, GFP_NOFS);
891                         ordered = btrfs_lookup_ordered_extent(inode,
892                                                               page_start);
893                         unlock_extent(tree, page_start, page_end, GFP_NOFS);
894                         if (!ordered)
895                                 break;
896
897                         unlock_page(page);
898                         btrfs_start_ordered_extent(inode, ordered, 1);
899                         btrfs_put_ordered_extent(ordered);
900                         lock_page(page);
901                 }
902
903                 if (!PageUptodate(page)) {
904                         btrfs_readpage(NULL, page);
905                         lock_page(page);
906                         if (!PageUptodate(page)) {
907                                 unlock_page(page);
908                                 page_cache_release(page);
909                                 ret = -EIO;
910                                 break;
911                         }
912                 }
913
914                 isize = i_size_read(inode);
915                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
916                 if (!isize || page->index > file_end) {
917                         /* whoops, we blew past eof, skip this page */
918                         unlock_page(page);
919                         page_cache_release(page);
920                         break;
921                 }
922
923                 if (page->mapping != inode->i_mapping) {
924                         unlock_page(page);
925                         page_cache_release(page);
926                         goto again;
927                 }
928
929                 pages[i] = page;
930                 i_done++;
931         }
932         if (!i_done || ret)
933                 goto out;
934
935         if (!(inode->i_sb->s_flags & MS_ACTIVE))
936                 goto out;
937
938         /*
939          * so now we have a nice long stream of locked
940          * and up to date pages, lets wait on them
941          */
942         for (i = 0; i < i_done; i++)
943                 wait_on_page_writeback(pages[i]);
944
945         page_start = page_offset(pages[0]);
946         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
947
948         lock_extent_bits(&BTRFS_I(inode)->io_tree,
949                          page_start, page_end - 1, 0, &cached_state,
950                          GFP_NOFS);
951         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
952                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
953                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
954                           GFP_NOFS);
955
956         if (i_done != num_pages) {
957                 spin_lock(&BTRFS_I(inode)->lock);
958                 BTRFS_I(inode)->outstanding_extents++;
959                 spin_unlock(&BTRFS_I(inode)->lock);
960                 btrfs_delalloc_release_space(inode,
961                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
962         }
963
964
965         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
966                                   &cached_state);
967
968         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
969                              page_start, page_end - 1, &cached_state,
970                              GFP_NOFS);
971
972         for (i = 0; i < i_done; i++) {
973                 clear_page_dirty_for_io(pages[i]);
974                 ClearPageChecked(pages[i]);
975                 set_page_extent_mapped(pages[i]);
976                 set_page_dirty(pages[i]);
977                 unlock_page(pages[i]);
978                 page_cache_release(pages[i]);
979         }
980         return i_done;
981 out:
982         for (i = 0; i < i_done; i++) {
983                 unlock_page(pages[i]);
984                 page_cache_release(pages[i]);
985         }
986         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
987         return ret;
988
989 }
990
991 int btrfs_defrag_file(struct inode *inode, struct file *file,
992                       struct btrfs_ioctl_defrag_range_args *range,
993                       u64 newer_than, unsigned long max_to_defrag)
994 {
995         struct btrfs_root *root = BTRFS_I(inode)->root;
996         struct btrfs_super_block *disk_super;
997         struct file_ra_state *ra = NULL;
998         unsigned long last_index;
999         u64 isize = i_size_read(inode);
1000         u64 features;
1001         u64 last_len = 0;
1002         u64 skip = 0;
1003         u64 defrag_end = 0;
1004         u64 newer_off = range->start;
1005         unsigned long i;
1006         unsigned long ra_index = 0;
1007         int ret;
1008         int defrag_count = 0;
1009         int compress_type = BTRFS_COMPRESS_ZLIB;
1010         int extent_thresh = range->extent_thresh;
1011         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1012         int cluster = max_cluster;
1013         u64 new_align = ~((u64)128 * 1024 - 1);
1014         struct page **pages = NULL;
1015
1016         if (extent_thresh == 0)
1017                 extent_thresh = 256 * 1024;
1018
1019         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1020                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1021                         return -EINVAL;
1022                 if (range->compress_type)
1023                         compress_type = range->compress_type;
1024         }
1025
1026         if (isize == 0)
1027                 return 0;
1028
1029         /*
1030          * if we were not given a file, allocate a readahead
1031          * context
1032          */
1033         if (!file) {
1034                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1035                 if (!ra)
1036                         return -ENOMEM;
1037                 file_ra_state_init(ra, inode->i_mapping);
1038         } else {
1039                 ra = &file->f_ra;
1040         }
1041
1042         pages = kmalloc(sizeof(struct page *) * max_cluster,
1043                         GFP_NOFS);
1044         if (!pages) {
1045                 ret = -ENOMEM;
1046                 goto out_ra;
1047         }
1048
1049         /* find the last page to defrag */
1050         if (range->start + range->len > range->start) {
1051                 last_index = min_t(u64, isize - 1,
1052                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1053         } else {
1054                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1055         }
1056
1057         if (newer_than) {
1058                 ret = find_new_extents(root, inode, newer_than,
1059                                        &newer_off, 64 * 1024);
1060                 if (!ret) {
1061                         range->start = newer_off;
1062                         /*
1063                          * we always align our defrag to help keep
1064                          * the extents in the file evenly spaced
1065                          */
1066                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1067                 } else
1068                         goto out_ra;
1069         } else {
1070                 i = range->start >> PAGE_CACHE_SHIFT;
1071         }
1072         if (!max_to_defrag)
1073                 max_to_defrag = last_index + 1;
1074
1075         /*
1076          * make writeback starts from i, so the defrag range can be
1077          * written sequentially.
1078          */
1079         if (i < inode->i_mapping->writeback_index)
1080                 inode->i_mapping->writeback_index = i;
1081
1082         while (i <= last_index && defrag_count < max_to_defrag &&
1083                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1084                 PAGE_CACHE_SHIFT)) {
1085                 /*
1086                  * make sure we stop running if someone unmounts
1087                  * the FS
1088                  */
1089                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1090                         break;
1091
1092                 if (!newer_than &&
1093                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1094                                         PAGE_CACHE_SIZE,
1095                                         extent_thresh,
1096                                         &last_len, &skip,
1097                                         &defrag_end)) {
1098                         unsigned long next;
1099                         /*
1100                          * the should_defrag function tells us how much to skip
1101                          * bump our counter by the suggested amount
1102                          */
1103                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1104                         i = max(i + 1, next);
1105                         continue;
1106                 }
1107
1108                 if (!newer_than) {
1109                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1110                                    PAGE_CACHE_SHIFT) - i;
1111                         cluster = min(cluster, max_cluster);
1112                 } else {
1113                         cluster = max_cluster;
1114                 }
1115
1116                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1117                         BTRFS_I(inode)->force_compress = compress_type;
1118
1119                 if (i + cluster > ra_index) {
1120                         ra_index = max(i, ra_index);
1121                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1122                                        cluster);
1123                         ra_index += max_cluster;
1124                 }
1125
1126                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1127                 if (ret < 0)
1128                         goto out_ra;
1129
1130                 defrag_count += ret;
1131                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1132
1133                 if (newer_than) {
1134                         if (newer_off == (u64)-1)
1135                                 break;
1136
1137                         newer_off = max(newer_off + 1,
1138                                         (u64)i << PAGE_CACHE_SHIFT);
1139
1140                         ret = find_new_extents(root, inode,
1141                                                newer_than, &newer_off,
1142                                                64 * 1024);
1143                         if (!ret) {
1144                                 range->start = newer_off;
1145                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1146                         } else {
1147                                 break;
1148                         }
1149                 } else {
1150                         if (ret > 0) {
1151                                 i += ret;
1152                                 last_len += ret << PAGE_CACHE_SHIFT;
1153                         } else {
1154                                 i++;
1155                                 last_len = 0;
1156                         }
1157                 }
1158         }
1159
1160         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1161                 filemap_flush(inode->i_mapping);
1162
1163         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1164                 /* the filemap_flush will queue IO into the worker threads, but
1165                  * we have to make sure the IO is actually started and that
1166                  * ordered extents get created before we return
1167                  */
1168                 atomic_inc(&root->fs_info->async_submit_draining);
1169                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1170                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1171                         wait_event(root->fs_info->async_submit_wait,
1172                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1173                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1174                 }
1175                 atomic_dec(&root->fs_info->async_submit_draining);
1176
1177                 mutex_lock(&inode->i_mutex);
1178                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1179                 mutex_unlock(&inode->i_mutex);
1180         }
1181
1182         disk_super = root->fs_info->super_copy;
1183         features = btrfs_super_incompat_flags(disk_super);
1184         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1185                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1186                 btrfs_set_super_incompat_flags(disk_super, features);
1187         }
1188
1189         ret = defrag_count;
1190
1191 out_ra:
1192         if (!file)
1193                 kfree(ra);
1194         kfree(pages);
1195         return ret;
1196 }
1197
1198 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1199                                         void __user *arg)
1200 {
1201         u64 new_size;
1202         u64 old_size;
1203         u64 devid = 1;
1204         struct btrfs_ioctl_vol_args *vol_args;
1205         struct btrfs_trans_handle *trans;
1206         struct btrfs_device *device = NULL;
1207         char *sizestr;
1208         char *devstr = NULL;
1209         int ret = 0;
1210         int mod = 0;
1211
1212         if (root->fs_info->sb->s_flags & MS_RDONLY)
1213                 return -EROFS;
1214
1215         if (!capable(CAP_SYS_ADMIN))
1216                 return -EPERM;
1217
1218         mutex_lock(&root->fs_info->volume_mutex);
1219         if (root->fs_info->balance_ctl) {
1220                 printk(KERN_INFO "btrfs: balance in progress\n");
1221                 ret = -EINVAL;
1222                 goto out;
1223         }
1224
1225         vol_args = memdup_user(arg, sizeof(*vol_args));
1226         if (IS_ERR(vol_args)) {
1227                 ret = PTR_ERR(vol_args);
1228                 goto out;
1229         }
1230
1231         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1232
1233         sizestr = vol_args->name;
1234         devstr = strchr(sizestr, ':');
1235         if (devstr) {
1236                 char *end;
1237                 sizestr = devstr + 1;
1238                 *devstr = '\0';
1239                 devstr = vol_args->name;
1240                 devid = simple_strtoull(devstr, &end, 10);
1241                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1242                        (unsigned long long)devid);
1243         }
1244         device = btrfs_find_device(root, devid, NULL, NULL);
1245         if (!device) {
1246                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1247                        (unsigned long long)devid);
1248                 ret = -EINVAL;
1249                 goto out_free;
1250         }
1251         if (!strcmp(sizestr, "max"))
1252                 new_size = device->bdev->bd_inode->i_size;
1253         else {
1254                 if (sizestr[0] == '-') {
1255                         mod = -1;
1256                         sizestr++;
1257                 } else if (sizestr[0] == '+') {
1258                         mod = 1;
1259                         sizestr++;
1260                 }
1261                 new_size = memparse(sizestr, NULL);
1262                 if (new_size == 0) {
1263                         ret = -EINVAL;
1264                         goto out_free;
1265                 }
1266         }
1267
1268         old_size = device->total_bytes;
1269
1270         if (mod < 0) {
1271                 if (new_size > old_size) {
1272                         ret = -EINVAL;
1273                         goto out_free;
1274                 }
1275                 new_size = old_size - new_size;
1276         } else if (mod > 0) {
1277                 new_size = old_size + new_size;
1278         }
1279
1280         if (new_size < 256 * 1024 * 1024) {
1281                 ret = -EINVAL;
1282                 goto out_free;
1283         }
1284         if (new_size > device->bdev->bd_inode->i_size) {
1285                 ret = -EFBIG;
1286                 goto out_free;
1287         }
1288
1289         do_div(new_size, root->sectorsize);
1290         new_size *= root->sectorsize;
1291
1292         printk(KERN_INFO "btrfs: new size for %s is %llu\n",
1293                 device->name, (unsigned long long)new_size);
1294
1295         if (new_size > old_size) {
1296                 trans = btrfs_start_transaction(root, 0);
1297                 if (IS_ERR(trans)) {
1298                         ret = PTR_ERR(trans);
1299                         goto out_free;
1300                 }
1301                 ret = btrfs_grow_device(trans, device, new_size);
1302                 btrfs_commit_transaction(trans, root);
1303         } else if (new_size < old_size) {
1304                 ret = btrfs_shrink_device(device, new_size);
1305         }
1306
1307 out_free:
1308         kfree(vol_args);
1309 out:
1310         mutex_unlock(&root->fs_info->volume_mutex);
1311         return ret;
1312 }
1313
1314 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1315                                                     char *name,
1316                                                     unsigned long fd,
1317                                                     int subvol,
1318                                                     u64 *transid,
1319                                                     bool readonly)
1320 {
1321         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1322         struct file *src_file;
1323         int namelen;
1324         int ret = 0;
1325
1326         if (root->fs_info->sb->s_flags & MS_RDONLY)
1327                 return -EROFS;
1328
1329         namelen = strlen(name);
1330         if (strchr(name, '/')) {
1331                 ret = -EINVAL;
1332                 goto out;
1333         }
1334
1335         if (name[0] == '.' &&
1336            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1337                 ret = -EEXIST;
1338                 goto out;
1339         }
1340
1341         if (subvol) {
1342                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1343                                      NULL, transid, readonly);
1344         } else {
1345                 struct inode *src_inode;
1346                 src_file = fget(fd);
1347                 if (!src_file) {
1348                         ret = -EINVAL;
1349                         goto out;
1350                 }
1351
1352                 src_inode = src_file->f_path.dentry->d_inode;
1353                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1354                         printk(KERN_INFO "btrfs: Snapshot src from "
1355                                "another FS\n");
1356                         ret = -EINVAL;
1357                         fput(src_file);
1358                         goto out;
1359                 }
1360                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1361                                      BTRFS_I(src_inode)->root,
1362                                      transid, readonly);
1363                 fput(src_file);
1364         }
1365 out:
1366         return ret;
1367 }
1368
1369 static noinline int btrfs_ioctl_snap_create(struct file *file,
1370                                             void __user *arg, int subvol)
1371 {
1372         struct btrfs_ioctl_vol_args *vol_args;
1373         int ret;
1374
1375         vol_args = memdup_user(arg, sizeof(*vol_args));
1376         if (IS_ERR(vol_args))
1377                 return PTR_ERR(vol_args);
1378         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1379
1380         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1381                                               vol_args->fd, subvol,
1382                                               NULL, false);
1383
1384         kfree(vol_args);
1385         return ret;
1386 }
1387
1388 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1389                                                void __user *arg, int subvol)
1390 {
1391         struct btrfs_ioctl_vol_args_v2 *vol_args;
1392         int ret;
1393         u64 transid = 0;
1394         u64 *ptr = NULL;
1395         bool readonly = false;
1396
1397         vol_args = memdup_user(arg, sizeof(*vol_args));
1398         if (IS_ERR(vol_args))
1399                 return PTR_ERR(vol_args);
1400         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1401
1402         if (vol_args->flags &
1403             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1404                 ret = -EOPNOTSUPP;
1405                 goto out;
1406         }
1407
1408         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1409                 ptr = &transid;
1410         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1411                 readonly = true;
1412
1413         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1414                                               vol_args->fd, subvol,
1415                                               ptr, readonly);
1416
1417         if (ret == 0 && ptr &&
1418             copy_to_user(arg +
1419                          offsetof(struct btrfs_ioctl_vol_args_v2,
1420                                   transid), ptr, sizeof(*ptr)))
1421                 ret = -EFAULT;
1422 out:
1423         kfree(vol_args);
1424         return ret;
1425 }
1426
1427 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1428                                                 void __user *arg)
1429 {
1430         struct inode *inode = fdentry(file)->d_inode;
1431         struct btrfs_root *root = BTRFS_I(inode)->root;
1432         int ret = 0;
1433         u64 flags = 0;
1434
1435         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1436                 return -EINVAL;
1437
1438         down_read(&root->fs_info->subvol_sem);
1439         if (btrfs_root_readonly(root))
1440                 flags |= BTRFS_SUBVOL_RDONLY;
1441         up_read(&root->fs_info->subvol_sem);
1442
1443         if (copy_to_user(arg, &flags, sizeof(flags)))
1444                 ret = -EFAULT;
1445
1446         return ret;
1447 }
1448
1449 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1450                                               void __user *arg)
1451 {
1452         struct inode *inode = fdentry(file)->d_inode;
1453         struct btrfs_root *root = BTRFS_I(inode)->root;
1454         struct btrfs_trans_handle *trans;
1455         u64 root_flags;
1456         u64 flags;
1457         int ret = 0;
1458
1459         if (root->fs_info->sb->s_flags & MS_RDONLY)
1460                 return -EROFS;
1461
1462         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1463                 return -EINVAL;
1464
1465         if (copy_from_user(&flags, arg, sizeof(flags)))
1466                 return -EFAULT;
1467
1468         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1469                 return -EINVAL;
1470
1471         if (flags & ~BTRFS_SUBVOL_RDONLY)
1472                 return -EOPNOTSUPP;
1473
1474         if (!inode_owner_or_capable(inode))
1475                 return -EACCES;
1476
1477         down_write(&root->fs_info->subvol_sem);
1478
1479         /* nothing to do */
1480         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1481                 goto out;
1482
1483         root_flags = btrfs_root_flags(&root->root_item);
1484         if (flags & BTRFS_SUBVOL_RDONLY)
1485                 btrfs_set_root_flags(&root->root_item,
1486                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1487         else
1488                 btrfs_set_root_flags(&root->root_item,
1489                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1490
1491         trans = btrfs_start_transaction(root, 1);
1492         if (IS_ERR(trans)) {
1493                 ret = PTR_ERR(trans);
1494                 goto out_reset;
1495         }
1496
1497         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1498                                 &root->root_key, &root->root_item);
1499
1500         btrfs_commit_transaction(trans, root);
1501 out_reset:
1502         if (ret)
1503                 btrfs_set_root_flags(&root->root_item, root_flags);
1504 out:
1505         up_write(&root->fs_info->subvol_sem);
1506         return ret;
1507 }
1508
1509 /*
1510  * helper to check if the subvolume references other subvolumes
1511  */
1512 static noinline int may_destroy_subvol(struct btrfs_root *root)
1513 {
1514         struct btrfs_path *path;
1515         struct btrfs_key key;
1516         int ret;
1517
1518         path = btrfs_alloc_path();
1519         if (!path)
1520                 return -ENOMEM;
1521
1522         key.objectid = root->root_key.objectid;
1523         key.type = BTRFS_ROOT_REF_KEY;
1524         key.offset = (u64)-1;
1525
1526         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1527                                 &key, path, 0, 0);
1528         if (ret < 0)
1529                 goto out;
1530         BUG_ON(ret == 0);
1531
1532         ret = 0;
1533         if (path->slots[0] > 0) {
1534                 path->slots[0]--;
1535                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1536                 if (key.objectid == root->root_key.objectid &&
1537                     key.type == BTRFS_ROOT_REF_KEY)
1538                         ret = -ENOTEMPTY;
1539         }
1540 out:
1541         btrfs_free_path(path);
1542         return ret;
1543 }
1544
1545 static noinline int key_in_sk(struct btrfs_key *key,
1546                               struct btrfs_ioctl_search_key *sk)
1547 {
1548         struct btrfs_key test;
1549         int ret;
1550
1551         test.objectid = sk->min_objectid;
1552         test.type = sk->min_type;
1553         test.offset = sk->min_offset;
1554
1555         ret = btrfs_comp_cpu_keys(key, &test);
1556         if (ret < 0)
1557                 return 0;
1558
1559         test.objectid = sk->max_objectid;
1560         test.type = sk->max_type;
1561         test.offset = sk->max_offset;
1562
1563         ret = btrfs_comp_cpu_keys(key, &test);
1564         if (ret > 0)
1565                 return 0;
1566         return 1;
1567 }
1568
1569 static noinline int copy_to_sk(struct btrfs_root *root,
1570                                struct btrfs_path *path,
1571                                struct btrfs_key *key,
1572                                struct btrfs_ioctl_search_key *sk,
1573                                char *buf,
1574                                unsigned long *sk_offset,
1575                                int *num_found)
1576 {
1577         u64 found_transid;
1578         struct extent_buffer *leaf;
1579         struct btrfs_ioctl_search_header sh;
1580         unsigned long item_off;
1581         unsigned long item_len;
1582         int nritems;
1583         int i;
1584         int slot;
1585         int ret = 0;
1586
1587         leaf = path->nodes[0];
1588         slot = path->slots[0];
1589         nritems = btrfs_header_nritems(leaf);
1590
1591         if (btrfs_header_generation(leaf) > sk->max_transid) {
1592                 i = nritems;
1593                 goto advance_key;
1594         }
1595         found_transid = btrfs_header_generation(leaf);
1596
1597         for (i = slot; i < nritems; i++) {
1598                 item_off = btrfs_item_ptr_offset(leaf, i);
1599                 item_len = btrfs_item_size_nr(leaf, i);
1600
1601                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1602                         item_len = 0;
1603
1604                 if (sizeof(sh) + item_len + *sk_offset >
1605                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1606                         ret = 1;
1607                         goto overflow;
1608                 }
1609
1610                 btrfs_item_key_to_cpu(leaf, key, i);
1611                 if (!key_in_sk(key, sk))
1612                         continue;
1613
1614                 sh.objectid = key->objectid;
1615                 sh.offset = key->offset;
1616                 sh.type = key->type;
1617                 sh.len = item_len;
1618                 sh.transid = found_transid;
1619
1620                 /* copy search result header */
1621                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1622                 *sk_offset += sizeof(sh);
1623
1624                 if (item_len) {
1625                         char *p = buf + *sk_offset;
1626                         /* copy the item */
1627                         read_extent_buffer(leaf, p,
1628                                            item_off, item_len);
1629                         *sk_offset += item_len;
1630                 }
1631                 (*num_found)++;
1632
1633                 if (*num_found >= sk->nr_items)
1634                         break;
1635         }
1636 advance_key:
1637         ret = 0;
1638         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1639                 key->offset++;
1640         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1641                 key->offset = 0;
1642                 key->type++;
1643         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1644                 key->offset = 0;
1645                 key->type = 0;
1646                 key->objectid++;
1647         } else
1648                 ret = 1;
1649 overflow:
1650         return ret;
1651 }
1652
1653 static noinline int search_ioctl(struct inode *inode,
1654                                  struct btrfs_ioctl_search_args *args)
1655 {
1656         struct btrfs_root *root;
1657         struct btrfs_key key;
1658         struct btrfs_key max_key;
1659         struct btrfs_path *path;
1660         struct btrfs_ioctl_search_key *sk = &args->key;
1661         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1662         int ret;
1663         int num_found = 0;
1664         unsigned long sk_offset = 0;
1665
1666         path = btrfs_alloc_path();
1667         if (!path)
1668                 return -ENOMEM;
1669
1670         if (sk->tree_id == 0) {
1671                 /* search the root of the inode that was passed */
1672                 root = BTRFS_I(inode)->root;
1673         } else {
1674                 key.objectid = sk->tree_id;
1675                 key.type = BTRFS_ROOT_ITEM_KEY;
1676                 key.offset = (u64)-1;
1677                 root = btrfs_read_fs_root_no_name(info, &key);
1678                 if (IS_ERR(root)) {
1679                         printk(KERN_ERR "could not find root %llu\n",
1680                                sk->tree_id);
1681                         btrfs_free_path(path);
1682                         return -ENOENT;
1683                 }
1684         }
1685
1686         key.objectid = sk->min_objectid;
1687         key.type = sk->min_type;
1688         key.offset = sk->min_offset;
1689
1690         max_key.objectid = sk->max_objectid;
1691         max_key.type = sk->max_type;
1692         max_key.offset = sk->max_offset;
1693
1694         path->keep_locks = 1;
1695
1696         while(1) {
1697                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1698                                            sk->min_transid);
1699                 if (ret != 0) {
1700                         if (ret > 0)
1701                                 ret = 0;
1702                         goto err;
1703                 }
1704                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1705                                  &sk_offset, &num_found);
1706                 btrfs_release_path(path);
1707                 if (ret || num_found >= sk->nr_items)
1708                         break;
1709
1710         }
1711         ret = 0;
1712 err:
1713         sk->nr_items = num_found;
1714         btrfs_free_path(path);
1715         return ret;
1716 }
1717
1718 static noinline int btrfs_ioctl_tree_search(struct file *file,
1719                                            void __user *argp)
1720 {
1721          struct btrfs_ioctl_search_args *args;
1722          struct inode *inode;
1723          int ret;
1724
1725         if (!capable(CAP_SYS_ADMIN))
1726                 return -EPERM;
1727
1728         args = memdup_user(argp, sizeof(*args));
1729         if (IS_ERR(args))
1730                 return PTR_ERR(args);
1731
1732         inode = fdentry(file)->d_inode;
1733         ret = search_ioctl(inode, args);
1734         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1735                 ret = -EFAULT;
1736         kfree(args);
1737         return ret;
1738 }
1739
1740 /*
1741  * Search INODE_REFs to identify path name of 'dirid' directory
1742  * in a 'tree_id' tree. and sets path name to 'name'.
1743  */
1744 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1745                                 u64 tree_id, u64 dirid, char *name)
1746 {
1747         struct btrfs_root *root;
1748         struct btrfs_key key;
1749         char *ptr;
1750         int ret = -1;
1751         int slot;
1752         int len;
1753         int total_len = 0;
1754         struct btrfs_inode_ref *iref;
1755         struct extent_buffer *l;
1756         struct btrfs_path *path;
1757
1758         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1759                 name[0]='\0';
1760                 return 0;
1761         }
1762
1763         path = btrfs_alloc_path();
1764         if (!path)
1765                 return -ENOMEM;
1766
1767         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1768
1769         key.objectid = tree_id;
1770         key.type = BTRFS_ROOT_ITEM_KEY;
1771         key.offset = (u64)-1;
1772         root = btrfs_read_fs_root_no_name(info, &key);
1773         if (IS_ERR(root)) {
1774                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1775                 ret = -ENOENT;
1776                 goto out;
1777         }
1778
1779         key.objectid = dirid;
1780         key.type = BTRFS_INODE_REF_KEY;
1781         key.offset = (u64)-1;
1782
1783         while(1) {
1784                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1785                 if (ret < 0)
1786                         goto out;
1787
1788                 l = path->nodes[0];
1789                 slot = path->slots[0];
1790                 if (ret > 0 && slot > 0)
1791                         slot--;
1792                 btrfs_item_key_to_cpu(l, &key, slot);
1793
1794                 if (ret > 0 && (key.objectid != dirid ||
1795                                 key.type != BTRFS_INODE_REF_KEY)) {
1796                         ret = -ENOENT;
1797                         goto out;
1798                 }
1799
1800                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1801                 len = btrfs_inode_ref_name_len(l, iref);
1802                 ptr -= len + 1;
1803                 total_len += len + 1;
1804                 if (ptr < name)
1805                         goto out;
1806
1807                 *(ptr + len) = '/';
1808                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1809
1810                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1811                         break;
1812
1813                 btrfs_release_path(path);
1814                 key.objectid = key.offset;
1815                 key.offset = (u64)-1;
1816                 dirid = key.objectid;
1817         }
1818         if (ptr < name)
1819                 goto out;
1820         memmove(name, ptr, total_len);
1821         name[total_len]='\0';
1822         ret = 0;
1823 out:
1824         btrfs_free_path(path);
1825         return ret;
1826 }
1827
1828 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1829                                            void __user *argp)
1830 {
1831          struct btrfs_ioctl_ino_lookup_args *args;
1832          struct inode *inode;
1833          int ret;
1834
1835         if (!capable(CAP_SYS_ADMIN))
1836                 return -EPERM;
1837
1838         args = memdup_user(argp, sizeof(*args));
1839         if (IS_ERR(args))
1840                 return PTR_ERR(args);
1841
1842         inode = fdentry(file)->d_inode;
1843
1844         if (args->treeid == 0)
1845                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1846
1847         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1848                                         args->treeid, args->objectid,
1849                                         args->name);
1850
1851         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1852                 ret = -EFAULT;
1853
1854         kfree(args);
1855         return ret;
1856 }
1857
1858 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1859                                              void __user *arg)
1860 {
1861         struct dentry *parent = fdentry(file);
1862         struct dentry *dentry;
1863         struct inode *dir = parent->d_inode;
1864         struct inode *inode;
1865         struct btrfs_root *root = BTRFS_I(dir)->root;
1866         struct btrfs_root *dest = NULL;
1867         struct btrfs_ioctl_vol_args *vol_args;
1868         struct btrfs_trans_handle *trans;
1869         int namelen;
1870         int ret;
1871         int err = 0;
1872
1873         vol_args = memdup_user(arg, sizeof(*vol_args));
1874         if (IS_ERR(vol_args))
1875                 return PTR_ERR(vol_args);
1876
1877         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1878         namelen = strlen(vol_args->name);
1879         if (strchr(vol_args->name, '/') ||
1880             strncmp(vol_args->name, "..", namelen) == 0) {
1881                 err = -EINVAL;
1882                 goto out;
1883         }
1884
1885         err = mnt_want_write_file(file);
1886         if (err)
1887                 goto out;
1888
1889         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1890         dentry = lookup_one_len(vol_args->name, parent, namelen);
1891         if (IS_ERR(dentry)) {
1892                 err = PTR_ERR(dentry);
1893                 goto out_unlock_dir;
1894         }
1895
1896         if (!dentry->d_inode) {
1897                 err = -ENOENT;
1898                 goto out_dput;
1899         }
1900
1901         inode = dentry->d_inode;
1902         dest = BTRFS_I(inode)->root;
1903         if (!capable(CAP_SYS_ADMIN)){
1904                 /*
1905                  * Regular user.  Only allow this with a special mount
1906                  * option, when the user has write+exec access to the
1907                  * subvol root, and when rmdir(2) would have been
1908                  * allowed.
1909                  *
1910                  * Note that this is _not_ check that the subvol is
1911                  * empty or doesn't contain data that we wouldn't
1912                  * otherwise be able to delete.
1913                  *
1914                  * Users who want to delete empty subvols should try
1915                  * rmdir(2).
1916                  */
1917                 err = -EPERM;
1918                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1919                         goto out_dput;
1920
1921                 /*
1922                  * Do not allow deletion if the parent dir is the same
1923                  * as the dir to be deleted.  That means the ioctl
1924                  * must be called on the dentry referencing the root
1925                  * of the subvol, not a random directory contained
1926                  * within it.
1927                  */
1928                 err = -EINVAL;
1929                 if (root == dest)
1930                         goto out_dput;
1931
1932                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1933                 if (err)
1934                         goto out_dput;
1935
1936                 /* check if subvolume may be deleted by a non-root user */
1937                 err = btrfs_may_delete(dir, dentry, 1);
1938                 if (err)
1939                         goto out_dput;
1940         }
1941
1942         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1943                 err = -EINVAL;
1944                 goto out_dput;
1945         }
1946
1947         mutex_lock(&inode->i_mutex);
1948         err = d_invalidate(dentry);
1949         if (err)
1950                 goto out_unlock;
1951
1952         down_write(&root->fs_info->subvol_sem);
1953
1954         err = may_destroy_subvol(dest);
1955         if (err)
1956                 goto out_up_write;
1957
1958         trans = btrfs_start_transaction(root, 0);
1959         if (IS_ERR(trans)) {
1960                 err = PTR_ERR(trans);
1961                 goto out_up_write;
1962         }
1963         trans->block_rsv = &root->fs_info->global_block_rsv;
1964
1965         ret = btrfs_unlink_subvol(trans, root, dir,
1966                                 dest->root_key.objectid,
1967                                 dentry->d_name.name,
1968                                 dentry->d_name.len);
1969         BUG_ON(ret);
1970
1971         btrfs_record_root_in_trans(trans, dest);
1972
1973         memset(&dest->root_item.drop_progress, 0,
1974                 sizeof(dest->root_item.drop_progress));
1975         dest->root_item.drop_level = 0;
1976         btrfs_set_root_refs(&dest->root_item, 0);
1977
1978         if (!xchg(&dest->orphan_item_inserted, 1)) {
1979                 ret = btrfs_insert_orphan_item(trans,
1980                                         root->fs_info->tree_root,
1981                                         dest->root_key.objectid);
1982                 BUG_ON(ret);
1983         }
1984
1985         ret = btrfs_end_transaction(trans, root);
1986         BUG_ON(ret);
1987         inode->i_flags |= S_DEAD;
1988 out_up_write:
1989         up_write(&root->fs_info->subvol_sem);
1990 out_unlock:
1991         mutex_unlock(&inode->i_mutex);
1992         if (!err) {
1993                 shrink_dcache_sb(root->fs_info->sb);
1994                 btrfs_invalidate_inodes(dest);
1995                 d_delete(dentry);
1996         }
1997 out_dput:
1998         dput(dentry);
1999 out_unlock_dir:
2000         mutex_unlock(&dir->i_mutex);
2001         mnt_drop_write_file(file);
2002 out:
2003         kfree(vol_args);
2004         return err;
2005 }
2006
2007 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2008 {
2009         struct inode *inode = fdentry(file)->d_inode;
2010         struct btrfs_root *root = BTRFS_I(inode)->root;
2011         struct btrfs_ioctl_defrag_range_args *range;
2012         int ret;
2013
2014         if (btrfs_root_readonly(root))
2015                 return -EROFS;
2016
2017         ret = mnt_want_write_file(file);
2018         if (ret)
2019                 return ret;
2020
2021         switch (inode->i_mode & S_IFMT) {
2022         case S_IFDIR:
2023                 if (!capable(CAP_SYS_ADMIN)) {
2024                         ret = -EPERM;
2025                         goto out;
2026                 }
2027                 ret = btrfs_defrag_root(root, 0);
2028                 if (ret)
2029                         goto out;
2030                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2031                 break;
2032         case S_IFREG:
2033                 if (!(file->f_mode & FMODE_WRITE)) {
2034                         ret = -EINVAL;
2035                         goto out;
2036                 }
2037
2038                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2039                 if (!range) {
2040                         ret = -ENOMEM;
2041                         goto out;
2042                 }
2043
2044                 if (argp) {
2045                         if (copy_from_user(range, argp,
2046                                            sizeof(*range))) {
2047                                 ret = -EFAULT;
2048                                 kfree(range);
2049                                 goto out;
2050                         }
2051                         /* compression requires us to start the IO */
2052                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2053                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2054                                 range->extent_thresh = (u32)-1;
2055                         }
2056                 } else {
2057                         /* the rest are all set to zero by kzalloc */
2058                         range->len = (u64)-1;
2059                 }
2060                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2061                                         range, 0, 0);
2062                 if (ret > 0)
2063                         ret = 0;
2064                 kfree(range);
2065                 break;
2066         default:
2067                 ret = -EINVAL;
2068         }
2069 out:
2070         mnt_drop_write_file(file);
2071         return ret;
2072 }
2073
2074 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2075 {
2076         struct btrfs_ioctl_vol_args *vol_args;
2077         int ret;
2078
2079         if (!capable(CAP_SYS_ADMIN))
2080                 return -EPERM;
2081
2082         mutex_lock(&root->fs_info->volume_mutex);
2083         if (root->fs_info->balance_ctl) {
2084                 printk(KERN_INFO "btrfs: balance in progress\n");
2085                 ret = -EINVAL;
2086                 goto out;
2087         }
2088
2089         vol_args = memdup_user(arg, sizeof(*vol_args));
2090         if (IS_ERR(vol_args)) {
2091                 ret = PTR_ERR(vol_args);
2092                 goto out;
2093         }
2094
2095         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2096         ret = btrfs_init_new_device(root, vol_args->name);
2097
2098         kfree(vol_args);
2099 out:
2100         mutex_unlock(&root->fs_info->volume_mutex);
2101         return ret;
2102 }
2103
2104 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2105 {
2106         struct btrfs_ioctl_vol_args *vol_args;
2107         int ret;
2108
2109         if (!capable(CAP_SYS_ADMIN))
2110                 return -EPERM;
2111
2112         if (root->fs_info->sb->s_flags & MS_RDONLY)
2113                 return -EROFS;
2114
2115         mutex_lock(&root->fs_info->volume_mutex);
2116         if (root->fs_info->balance_ctl) {
2117                 printk(KERN_INFO "btrfs: balance in progress\n");
2118                 ret = -EINVAL;
2119                 goto out;
2120         }
2121
2122         vol_args = memdup_user(arg, sizeof(*vol_args));
2123         if (IS_ERR(vol_args)) {
2124                 ret = PTR_ERR(vol_args);
2125                 goto out;
2126         }
2127
2128         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2129         ret = btrfs_rm_device(root, vol_args->name);
2130
2131         kfree(vol_args);
2132 out:
2133         mutex_unlock(&root->fs_info->volume_mutex);
2134         return ret;
2135 }
2136
2137 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2138 {
2139         struct btrfs_ioctl_fs_info_args *fi_args;
2140         struct btrfs_device *device;
2141         struct btrfs_device *next;
2142         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2143         int ret = 0;
2144
2145         if (!capable(CAP_SYS_ADMIN))
2146                 return -EPERM;
2147
2148         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2149         if (!fi_args)
2150                 return -ENOMEM;
2151
2152         fi_args->num_devices = fs_devices->num_devices;
2153         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2154
2155         mutex_lock(&fs_devices->device_list_mutex);
2156         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2157                 if (device->devid > fi_args->max_id)
2158                         fi_args->max_id = device->devid;
2159         }
2160         mutex_unlock(&fs_devices->device_list_mutex);
2161
2162         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2163                 ret = -EFAULT;
2164
2165         kfree(fi_args);
2166         return ret;
2167 }
2168
2169 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2170 {
2171         struct btrfs_ioctl_dev_info_args *di_args;
2172         struct btrfs_device *dev;
2173         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2174         int ret = 0;
2175         char *s_uuid = NULL;
2176         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2177
2178         if (!capable(CAP_SYS_ADMIN))
2179                 return -EPERM;
2180
2181         di_args = memdup_user(arg, sizeof(*di_args));
2182         if (IS_ERR(di_args))
2183                 return PTR_ERR(di_args);
2184
2185         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2186                 s_uuid = di_args->uuid;
2187
2188         mutex_lock(&fs_devices->device_list_mutex);
2189         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2190         mutex_unlock(&fs_devices->device_list_mutex);
2191
2192         if (!dev) {
2193                 ret = -ENODEV;
2194                 goto out;
2195         }
2196
2197         di_args->devid = dev->devid;
2198         di_args->bytes_used = dev->bytes_used;
2199         di_args->total_bytes = dev->total_bytes;
2200         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2201         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2202
2203 out:
2204         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2205                 ret = -EFAULT;
2206
2207         kfree(di_args);
2208         return ret;
2209 }
2210
2211 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2212                                        u64 off, u64 olen, u64 destoff)
2213 {
2214         struct inode *inode = fdentry(file)->d_inode;
2215         struct btrfs_root *root = BTRFS_I(inode)->root;
2216         struct file *src_file;
2217         struct inode *src;
2218         struct btrfs_trans_handle *trans;
2219         struct btrfs_path *path;
2220         struct extent_buffer *leaf;
2221         char *buf;
2222         struct btrfs_key key;
2223         u32 nritems;
2224         int slot;
2225         int ret;
2226         u64 len = olen;
2227         u64 bs = root->fs_info->sb->s_blocksize;
2228         u64 hint_byte;
2229
2230         /*
2231          * TODO:
2232          * - split compressed inline extents.  annoying: we need to
2233          *   decompress into destination's address_space (the file offset
2234          *   may change, so source mapping won't do), then recompress (or
2235          *   otherwise reinsert) a subrange.
2236          * - allow ranges within the same file to be cloned (provided
2237          *   they don't overlap)?
2238          */
2239
2240         /* the destination must be opened for writing */
2241         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2242                 return -EINVAL;
2243
2244         if (btrfs_root_readonly(root))
2245                 return -EROFS;
2246
2247         ret = mnt_want_write_file(file);
2248         if (ret)
2249                 return ret;
2250
2251         src_file = fget(srcfd);
2252         if (!src_file) {
2253                 ret = -EBADF;
2254                 goto out_drop_write;
2255         }
2256
2257         src = src_file->f_dentry->d_inode;
2258
2259         ret = -EINVAL;
2260         if (src == inode)
2261                 goto out_fput;
2262
2263         /* the src must be open for reading */
2264         if (!(src_file->f_mode & FMODE_READ))
2265                 goto out_fput;
2266
2267         /* don't make the dst file partly checksummed */
2268         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2269             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2270                 goto out_fput;
2271
2272         ret = -EISDIR;
2273         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2274                 goto out_fput;
2275
2276         ret = -EXDEV;
2277         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2278                 goto out_fput;
2279
2280         ret = -ENOMEM;
2281         buf = vmalloc(btrfs_level_size(root, 0));
2282         if (!buf)
2283                 goto out_fput;
2284
2285         path = btrfs_alloc_path();
2286         if (!path) {
2287                 vfree(buf);
2288                 goto out_fput;
2289         }
2290         path->reada = 2;
2291
2292         if (inode < src) {
2293                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2294                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2295         } else {
2296                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2297                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2298         }
2299
2300         /* determine range to clone */
2301         ret = -EINVAL;
2302         if (off + len > src->i_size || off + len < off)
2303                 goto out_unlock;
2304         if (len == 0)
2305                 olen = len = src->i_size - off;
2306         /* if we extend to eof, continue to block boundary */
2307         if (off + len == src->i_size)
2308                 len = ALIGN(src->i_size, bs) - off;
2309
2310         /* verify the end result is block aligned */
2311         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2312             !IS_ALIGNED(destoff, bs))
2313                 goto out_unlock;
2314
2315         if (destoff > inode->i_size) {
2316                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2317                 if (ret)
2318                         goto out_unlock;
2319         }
2320
2321         /* truncate page cache pages from target inode range */
2322         truncate_inode_pages_range(&inode->i_data, destoff,
2323                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2324
2325         /* do any pending delalloc/csum calc on src, one way or
2326            another, and lock file content */
2327         while (1) {
2328                 struct btrfs_ordered_extent *ordered;
2329                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2330                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2331                 if (!ordered &&
2332                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2333                                    EXTENT_DELALLOC, 0, NULL))
2334                         break;
2335                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2336                 if (ordered)
2337                         btrfs_put_ordered_extent(ordered);
2338                 btrfs_wait_ordered_range(src, off, len);
2339         }
2340
2341         /* clone data */
2342         key.objectid = btrfs_ino(src);
2343         key.type = BTRFS_EXTENT_DATA_KEY;
2344         key.offset = 0;
2345
2346         while (1) {
2347                 /*
2348                  * note the key will change type as we walk through the
2349                  * tree.
2350                  */
2351                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2352                 if (ret < 0)
2353                         goto out;
2354
2355                 nritems = btrfs_header_nritems(path->nodes[0]);
2356                 if (path->slots[0] >= nritems) {
2357                         ret = btrfs_next_leaf(root, path);
2358                         if (ret < 0)
2359                                 goto out;
2360                         if (ret > 0)
2361                                 break;
2362                         nritems = btrfs_header_nritems(path->nodes[0]);
2363                 }
2364                 leaf = path->nodes[0];
2365                 slot = path->slots[0];
2366
2367                 btrfs_item_key_to_cpu(leaf, &key, slot);
2368                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2369                     key.objectid != btrfs_ino(src))
2370                         break;
2371
2372                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2373                         struct btrfs_file_extent_item *extent;
2374                         int type;
2375                         u32 size;
2376                         struct btrfs_key new_key;
2377                         u64 disko = 0, diskl = 0;
2378                         u64 datao = 0, datal = 0;
2379                         u8 comp;
2380                         u64 endoff;
2381
2382                         size = btrfs_item_size_nr(leaf, slot);
2383                         read_extent_buffer(leaf, buf,
2384                                            btrfs_item_ptr_offset(leaf, slot),
2385                                            size);
2386
2387                         extent = btrfs_item_ptr(leaf, slot,
2388                                                 struct btrfs_file_extent_item);
2389                         comp = btrfs_file_extent_compression(leaf, extent);
2390                         type = btrfs_file_extent_type(leaf, extent);
2391                         if (type == BTRFS_FILE_EXTENT_REG ||
2392                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2393                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2394                                                                       extent);
2395                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2396                                                                  extent);
2397                                 datao = btrfs_file_extent_offset(leaf, extent);
2398                                 datal = btrfs_file_extent_num_bytes(leaf,
2399                                                                     extent);
2400                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2401                                 /* take upper bound, may be compressed */
2402                                 datal = btrfs_file_extent_ram_bytes(leaf,
2403                                                                     extent);
2404                         }
2405                         btrfs_release_path(path);
2406
2407                         if (key.offset + datal <= off ||
2408                             key.offset >= off+len)
2409                                 goto next;
2410
2411                         memcpy(&new_key, &key, sizeof(new_key));
2412                         new_key.objectid = btrfs_ino(inode);
2413                         if (off <= key.offset)
2414                                 new_key.offset = key.offset + destoff - off;
2415                         else
2416                                 new_key.offset = destoff;
2417
2418                         /*
2419                          * 1 - adjusting old extent (we may have to split it)
2420                          * 1 - add new extent
2421                          * 1 - inode update
2422                          */
2423                         trans = btrfs_start_transaction(root, 3);
2424                         if (IS_ERR(trans)) {
2425                                 ret = PTR_ERR(trans);
2426                                 goto out;
2427                         }
2428
2429                         if (type == BTRFS_FILE_EXTENT_REG ||
2430                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2431                                 /*
2432                                  *    a  | --- range to clone ---|  b
2433                                  * | ------------- extent ------------- |
2434                                  */
2435
2436                                 /* substract range b */
2437                                 if (key.offset + datal > off + len)
2438                                         datal = off + len - key.offset;
2439
2440                                 /* substract range a */
2441                                 if (off > key.offset) {
2442                                         datao += off - key.offset;
2443                                         datal -= off - key.offset;
2444                                 }
2445
2446                                 ret = btrfs_drop_extents(trans, inode,
2447                                                          new_key.offset,
2448                                                          new_key.offset + datal,
2449                                                          &hint_byte, 1);
2450                                 BUG_ON(ret);
2451
2452                                 ret = btrfs_insert_empty_item(trans, root, path,
2453                                                               &new_key, size);
2454                                 BUG_ON(ret);
2455
2456                                 leaf = path->nodes[0];
2457                                 slot = path->slots[0];
2458                                 write_extent_buffer(leaf, buf,
2459                                             btrfs_item_ptr_offset(leaf, slot),
2460                                             size);
2461
2462                                 extent = btrfs_item_ptr(leaf, slot,
2463                                                 struct btrfs_file_extent_item);
2464
2465                                 /* disko == 0 means it's a hole */
2466                                 if (!disko)
2467                                         datao = 0;
2468
2469                                 btrfs_set_file_extent_offset(leaf, extent,
2470                                                              datao);
2471                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2472                                                                 datal);
2473                                 if (disko) {
2474                                         inode_add_bytes(inode, datal);
2475                                         ret = btrfs_inc_extent_ref(trans, root,
2476                                                         disko, diskl, 0,
2477                                                         root->root_key.objectid,
2478                                                         btrfs_ino(inode),
2479                                                         new_key.offset - datao,
2480                                                         0);
2481                                         BUG_ON(ret);
2482                                 }
2483                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2484                                 u64 skip = 0;
2485                                 u64 trim = 0;
2486                                 if (off > key.offset) {
2487                                         skip = off - key.offset;
2488                                         new_key.offset += skip;
2489                                 }
2490
2491                                 if (key.offset + datal > off+len)
2492                                         trim = key.offset + datal - (off+len);
2493
2494                                 if (comp && (skip || trim)) {
2495                                         ret = -EINVAL;
2496                                         btrfs_end_transaction(trans, root);
2497                                         goto out;
2498                                 }
2499                                 size -= skip + trim;
2500                                 datal -= skip + trim;
2501
2502                                 ret = btrfs_drop_extents(trans, inode,
2503                                                          new_key.offset,
2504                                                          new_key.offset + datal,
2505                                                          &hint_byte, 1);
2506                                 BUG_ON(ret);
2507
2508                                 ret = btrfs_insert_empty_item(trans, root, path,
2509                                                               &new_key, size);
2510                                 BUG_ON(ret);
2511
2512                                 if (skip) {
2513                                         u32 start =
2514                                           btrfs_file_extent_calc_inline_size(0);
2515                                         memmove(buf+start, buf+start+skip,
2516                                                 datal);
2517                                 }
2518
2519                                 leaf = path->nodes[0];
2520                                 slot = path->slots[0];
2521                                 write_extent_buffer(leaf, buf,
2522                                             btrfs_item_ptr_offset(leaf, slot),
2523                                             size);
2524                                 inode_add_bytes(inode, datal);
2525                         }
2526
2527                         btrfs_mark_buffer_dirty(leaf);
2528                         btrfs_release_path(path);
2529
2530                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2531
2532                         /*
2533                          * we round up to the block size at eof when
2534                          * determining which extents to clone above,
2535                          * but shouldn't round up the file size
2536                          */
2537                         endoff = new_key.offset + datal;
2538                         if (endoff > destoff+olen)
2539                                 endoff = destoff+olen;
2540                         if (endoff > inode->i_size)
2541                                 btrfs_i_size_write(inode, endoff);
2542
2543                         ret = btrfs_update_inode(trans, root, inode);
2544                         BUG_ON(ret);
2545                         btrfs_end_transaction(trans, root);
2546                 }
2547 next:
2548                 btrfs_release_path(path);
2549                 key.offset++;
2550         }
2551         ret = 0;
2552 out:
2553         btrfs_release_path(path);
2554         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2555 out_unlock:
2556         mutex_unlock(&src->i_mutex);
2557         mutex_unlock(&inode->i_mutex);
2558         vfree(buf);
2559         btrfs_free_path(path);
2560 out_fput:
2561         fput(src_file);
2562 out_drop_write:
2563         mnt_drop_write_file(file);
2564         return ret;
2565 }
2566
2567 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2568 {
2569         struct btrfs_ioctl_clone_range_args args;
2570
2571         if (copy_from_user(&args, argp, sizeof(args)))
2572                 return -EFAULT;
2573         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2574                                  args.src_length, args.dest_offset);
2575 }
2576
2577 /*
2578  * there are many ways the trans_start and trans_end ioctls can lead
2579  * to deadlocks.  They should only be used by applications that
2580  * basically own the machine, and have a very in depth understanding
2581  * of all the possible deadlocks and enospc problems.
2582  */
2583 static long btrfs_ioctl_trans_start(struct file *file)
2584 {
2585         struct inode *inode = fdentry(file)->d_inode;
2586         struct btrfs_root *root = BTRFS_I(inode)->root;
2587         struct btrfs_trans_handle *trans;
2588         int ret;
2589
2590         ret = -EPERM;
2591         if (!capable(CAP_SYS_ADMIN))
2592                 goto out;
2593
2594         ret = -EINPROGRESS;
2595         if (file->private_data)
2596                 goto out;
2597
2598         ret = -EROFS;
2599         if (btrfs_root_readonly(root))
2600                 goto out;
2601
2602         ret = mnt_want_write_file(file);
2603         if (ret)
2604                 goto out;
2605
2606         atomic_inc(&root->fs_info->open_ioctl_trans);
2607
2608         ret = -ENOMEM;
2609         trans = btrfs_start_ioctl_transaction(root);
2610         if (IS_ERR(trans))
2611                 goto out_drop;
2612
2613         file->private_data = trans;
2614         return 0;
2615
2616 out_drop:
2617         atomic_dec(&root->fs_info->open_ioctl_trans);
2618         mnt_drop_write_file(file);
2619 out:
2620         return ret;
2621 }
2622
2623 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2624 {
2625         struct inode *inode = fdentry(file)->d_inode;
2626         struct btrfs_root *root = BTRFS_I(inode)->root;
2627         struct btrfs_root *new_root;
2628         struct btrfs_dir_item *di;
2629         struct btrfs_trans_handle *trans;
2630         struct btrfs_path *path;
2631         struct btrfs_key location;
2632         struct btrfs_disk_key disk_key;
2633         struct btrfs_super_block *disk_super;
2634         u64 features;
2635         u64 objectid = 0;
2636         u64 dir_id;
2637
2638         if (!capable(CAP_SYS_ADMIN))
2639                 return -EPERM;
2640
2641         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2642                 return -EFAULT;
2643
2644         if (!objectid)
2645                 objectid = root->root_key.objectid;
2646
2647         location.objectid = objectid;
2648         location.type = BTRFS_ROOT_ITEM_KEY;
2649         location.offset = (u64)-1;
2650
2651         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2652         if (IS_ERR(new_root))
2653                 return PTR_ERR(new_root);
2654
2655         if (btrfs_root_refs(&new_root->root_item) == 0)
2656                 return -ENOENT;
2657
2658         path = btrfs_alloc_path();
2659         if (!path)
2660                 return -ENOMEM;
2661         path->leave_spinning = 1;
2662
2663         trans = btrfs_start_transaction(root, 1);
2664         if (IS_ERR(trans)) {
2665                 btrfs_free_path(path);
2666                 return PTR_ERR(trans);
2667         }
2668
2669         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2670         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2671                                    dir_id, "default", 7, 1);
2672         if (IS_ERR_OR_NULL(di)) {
2673                 btrfs_free_path(path);
2674                 btrfs_end_transaction(trans, root);
2675                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2676                        "this isn't going to work\n");
2677                 return -ENOENT;
2678         }
2679
2680         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2681         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2682         btrfs_mark_buffer_dirty(path->nodes[0]);
2683         btrfs_free_path(path);
2684
2685         disk_super = root->fs_info->super_copy;
2686         features = btrfs_super_incompat_flags(disk_super);
2687         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2688                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2689                 btrfs_set_super_incompat_flags(disk_super, features);
2690         }
2691         btrfs_end_transaction(trans, root);
2692
2693         return 0;
2694 }
2695
2696 static void get_block_group_info(struct list_head *groups_list,
2697                                  struct btrfs_ioctl_space_info *space)
2698 {
2699         struct btrfs_block_group_cache *block_group;
2700
2701         space->total_bytes = 0;
2702         space->used_bytes = 0;
2703         space->flags = 0;
2704         list_for_each_entry(block_group, groups_list, list) {
2705                 space->flags = block_group->flags;
2706                 space->total_bytes += block_group->key.offset;
2707                 space->used_bytes +=
2708                         btrfs_block_group_used(&block_group->item);
2709         }
2710 }
2711
2712 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2713 {
2714         struct btrfs_ioctl_space_args space_args;
2715         struct btrfs_ioctl_space_info space;
2716         struct btrfs_ioctl_space_info *dest;
2717         struct btrfs_ioctl_space_info *dest_orig;
2718         struct btrfs_ioctl_space_info __user *user_dest;
2719         struct btrfs_space_info *info;
2720         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2721                        BTRFS_BLOCK_GROUP_SYSTEM,
2722                        BTRFS_BLOCK_GROUP_METADATA,
2723                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2724         int num_types = 4;
2725         int alloc_size;
2726         int ret = 0;
2727         u64 slot_count = 0;
2728         int i, c;
2729
2730         if (copy_from_user(&space_args,
2731                            (struct btrfs_ioctl_space_args __user *)arg,
2732                            sizeof(space_args)))
2733                 return -EFAULT;
2734
2735         for (i = 0; i < num_types; i++) {
2736                 struct btrfs_space_info *tmp;
2737
2738                 info = NULL;
2739                 rcu_read_lock();
2740                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2741                                         list) {
2742                         if (tmp->flags == types[i]) {
2743                                 info = tmp;
2744                                 break;
2745                         }
2746                 }
2747                 rcu_read_unlock();
2748
2749                 if (!info)
2750                         continue;
2751
2752                 down_read(&info->groups_sem);
2753                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2754                         if (!list_empty(&info->block_groups[c]))
2755                                 slot_count++;
2756                 }
2757                 up_read(&info->groups_sem);
2758         }
2759
2760         /* space_slots == 0 means they are asking for a count */
2761         if (space_args.space_slots == 0) {
2762                 space_args.total_spaces = slot_count;
2763                 goto out;
2764         }
2765
2766         slot_count = min_t(u64, space_args.space_slots, slot_count);
2767
2768         alloc_size = sizeof(*dest) * slot_count;
2769
2770         /* we generally have at most 6 or so space infos, one for each raid
2771          * level.  So, a whole page should be more than enough for everyone
2772          */
2773         if (alloc_size > PAGE_CACHE_SIZE)
2774                 return -ENOMEM;
2775
2776         space_args.total_spaces = 0;
2777         dest = kmalloc(alloc_size, GFP_NOFS);
2778         if (!dest)
2779                 return -ENOMEM;
2780         dest_orig = dest;
2781
2782         /* now we have a buffer to copy into */
2783         for (i = 0; i < num_types; i++) {
2784                 struct btrfs_space_info *tmp;
2785
2786                 if (!slot_count)
2787                         break;
2788
2789                 info = NULL;
2790                 rcu_read_lock();
2791                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2792                                         list) {
2793                         if (tmp->flags == types[i]) {
2794                                 info = tmp;
2795                                 break;
2796                         }
2797                 }
2798                 rcu_read_unlock();
2799
2800                 if (!info)
2801                         continue;
2802                 down_read(&info->groups_sem);
2803                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2804                         if (!list_empty(&info->block_groups[c])) {
2805                                 get_block_group_info(&info->block_groups[c],
2806                                                      &space);
2807                                 memcpy(dest, &space, sizeof(space));
2808                                 dest++;
2809                                 space_args.total_spaces++;
2810                                 slot_count--;
2811                         }
2812                         if (!slot_count)
2813                                 break;
2814                 }
2815                 up_read(&info->groups_sem);
2816         }
2817
2818         user_dest = (struct btrfs_ioctl_space_info *)
2819                 (arg + sizeof(struct btrfs_ioctl_space_args));
2820
2821         if (copy_to_user(user_dest, dest_orig, alloc_size))
2822                 ret = -EFAULT;
2823
2824         kfree(dest_orig);
2825 out:
2826         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2827                 ret = -EFAULT;
2828
2829         return ret;
2830 }
2831
2832 /*
2833  * there are many ways the trans_start and trans_end ioctls can lead
2834  * to deadlocks.  They should only be used by applications that
2835  * basically own the machine, and have a very in depth understanding
2836  * of all the possible deadlocks and enospc problems.
2837  */
2838 long btrfs_ioctl_trans_end(struct file *file)
2839 {
2840         struct inode *inode = fdentry(file)->d_inode;
2841         struct btrfs_root *root = BTRFS_I(inode)->root;
2842         struct btrfs_trans_handle *trans;
2843
2844         trans = file->private_data;
2845         if (!trans)
2846                 return -EINVAL;
2847         file->private_data = NULL;
2848
2849         btrfs_end_transaction(trans, root);
2850
2851         atomic_dec(&root->fs_info->open_ioctl_trans);
2852
2853         mnt_drop_write_file(file);
2854         return 0;
2855 }
2856
2857 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2858 {
2859         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2860         struct btrfs_trans_handle *trans;
2861         u64 transid;
2862         int ret;
2863
2864         trans = btrfs_start_transaction(root, 0);
2865         if (IS_ERR(trans))
2866                 return PTR_ERR(trans);
2867         transid = trans->transid;
2868         ret = btrfs_commit_transaction_async(trans, root, 0);
2869         if (ret) {
2870                 btrfs_end_transaction(trans, root);
2871                 return ret;
2872         }
2873
2874         if (argp)
2875                 if (copy_to_user(argp, &transid, sizeof(transid)))
2876                         return -EFAULT;
2877         return 0;
2878 }
2879
2880 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2881 {
2882         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2883         u64 transid;
2884
2885         if (argp) {
2886                 if (copy_from_user(&transid, argp, sizeof(transid)))
2887                         return -EFAULT;
2888         } else {
2889                 transid = 0;  /* current trans */
2890         }
2891         return btrfs_wait_for_commit(root, transid);
2892 }
2893
2894 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2895 {
2896         int ret;
2897         struct btrfs_ioctl_scrub_args *sa;
2898
2899         if (!capable(CAP_SYS_ADMIN))
2900                 return -EPERM;
2901
2902         sa = memdup_user(arg, sizeof(*sa));
2903         if (IS_ERR(sa))
2904                 return PTR_ERR(sa);
2905
2906         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2907                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2908
2909         if (copy_to_user(arg, sa, sizeof(*sa)))
2910                 ret = -EFAULT;
2911
2912         kfree(sa);
2913         return ret;
2914 }
2915
2916 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2917 {
2918         if (!capable(CAP_SYS_ADMIN))
2919                 return -EPERM;
2920
2921         return btrfs_scrub_cancel(root);
2922 }
2923
2924 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2925                                        void __user *arg)
2926 {
2927         struct btrfs_ioctl_scrub_args *sa;
2928         int ret;
2929
2930         if (!capable(CAP_SYS_ADMIN))
2931                 return -EPERM;
2932
2933         sa = memdup_user(arg, sizeof(*sa));
2934         if (IS_ERR(sa))
2935                 return PTR_ERR(sa);
2936
2937         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2938
2939         if (copy_to_user(arg, sa, sizeof(*sa)))
2940                 ret = -EFAULT;
2941
2942         kfree(sa);
2943         return ret;
2944 }
2945
2946 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
2947 {
2948         int ret = 0;
2949         int i;
2950         u64 rel_ptr;
2951         int size;
2952         struct btrfs_ioctl_ino_path_args *ipa = NULL;
2953         struct inode_fs_paths *ipath = NULL;
2954         struct btrfs_path *path;
2955
2956         if (!capable(CAP_SYS_ADMIN))
2957                 return -EPERM;
2958
2959         path = btrfs_alloc_path();
2960         if (!path) {
2961                 ret = -ENOMEM;
2962                 goto out;
2963         }
2964
2965         ipa = memdup_user(arg, sizeof(*ipa));
2966         if (IS_ERR(ipa)) {
2967                 ret = PTR_ERR(ipa);
2968                 ipa = NULL;
2969                 goto out;
2970         }
2971
2972         size = min_t(u32, ipa->size, 4096);
2973         ipath = init_ipath(size, root, path);
2974         if (IS_ERR(ipath)) {
2975                 ret = PTR_ERR(ipath);
2976                 ipath = NULL;
2977                 goto out;
2978         }
2979
2980         ret = paths_from_inode(ipa->inum, ipath);
2981         if (ret < 0)
2982                 goto out;
2983
2984         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
2985                 rel_ptr = ipath->fspath->val[i] -
2986                           (u64)(unsigned long)ipath->fspath->val;
2987                 ipath->fspath->val[i] = rel_ptr;
2988         }
2989
2990         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
2991                            (void *)(unsigned long)ipath->fspath, size);
2992         if (ret) {
2993                 ret = -EFAULT;
2994                 goto out;
2995         }
2996
2997 out:
2998         btrfs_free_path(path);
2999         free_ipath(ipath);
3000         kfree(ipa);
3001
3002         return ret;
3003 }
3004
3005 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3006 {
3007         struct btrfs_data_container *inodes = ctx;
3008         const size_t c = 3 * sizeof(u64);
3009
3010         if (inodes->bytes_left >= c) {
3011                 inodes->bytes_left -= c;
3012                 inodes->val[inodes->elem_cnt] = inum;
3013                 inodes->val[inodes->elem_cnt + 1] = offset;
3014                 inodes->val[inodes->elem_cnt + 2] = root;
3015                 inodes->elem_cnt += 3;
3016         } else {
3017                 inodes->bytes_missing += c - inodes->bytes_left;
3018                 inodes->bytes_left = 0;
3019                 inodes->elem_missed += 3;
3020         }
3021
3022         return 0;
3023 }
3024
3025 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3026                                         void __user *arg)
3027 {
3028         int ret = 0;
3029         int size;
3030         u64 extent_item_pos;
3031         struct btrfs_ioctl_logical_ino_args *loi;
3032         struct btrfs_data_container *inodes = NULL;
3033         struct btrfs_path *path = NULL;
3034         struct btrfs_key key;
3035
3036         if (!capable(CAP_SYS_ADMIN))
3037                 return -EPERM;
3038
3039         loi = memdup_user(arg, sizeof(*loi));
3040         if (IS_ERR(loi)) {
3041                 ret = PTR_ERR(loi);
3042                 loi = NULL;
3043                 goto out;
3044         }
3045
3046         path = btrfs_alloc_path();
3047         if (!path) {
3048                 ret = -ENOMEM;
3049                 goto out;
3050         }
3051
3052         size = min_t(u32, loi->size, 4096);
3053         inodes = init_data_container(size);
3054         if (IS_ERR(inodes)) {
3055                 ret = PTR_ERR(inodes);
3056                 inodes = NULL;
3057                 goto out;
3058         }
3059
3060         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
3061         btrfs_release_path(path);
3062
3063         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
3064                 ret = -ENOENT;
3065         if (ret < 0)
3066                 goto out;
3067
3068         extent_item_pos = loi->logical - key.objectid;
3069         ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
3070                                         extent_item_pos, build_ino_list,
3071                                         inodes);
3072
3073         if (ret < 0)
3074                 goto out;
3075
3076         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3077                            (void *)(unsigned long)inodes, size);
3078         if (ret)
3079                 ret = -EFAULT;
3080
3081 out:
3082         btrfs_free_path(path);
3083         kfree(inodes);
3084         kfree(loi);
3085
3086         return ret;
3087 }
3088
3089 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3090                                struct btrfs_ioctl_balance_args *bargs)
3091 {
3092         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3093
3094         bargs->flags = bctl->flags;
3095
3096         if (atomic_read(&fs_info->balance_running))
3097                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3098         if (atomic_read(&fs_info->balance_pause_req))
3099                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3100         if (atomic_read(&fs_info->balance_cancel_req))
3101                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3102
3103         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3104         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3105         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3106
3107         if (lock) {
3108                 spin_lock(&fs_info->balance_lock);
3109                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3110                 spin_unlock(&fs_info->balance_lock);
3111         } else {
3112                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3113         }
3114 }
3115
3116 static long btrfs_ioctl_balance(struct btrfs_root *root, void __user *arg)
3117 {
3118         struct btrfs_fs_info *fs_info = root->fs_info;
3119         struct btrfs_ioctl_balance_args *bargs;
3120         struct btrfs_balance_control *bctl;
3121         int ret;
3122
3123         if (!capable(CAP_SYS_ADMIN))
3124                 return -EPERM;
3125
3126         if (fs_info->sb->s_flags & MS_RDONLY)
3127                 return -EROFS;
3128
3129         mutex_lock(&fs_info->volume_mutex);
3130         mutex_lock(&fs_info->balance_mutex);
3131
3132         if (arg) {
3133                 bargs = memdup_user(arg, sizeof(*bargs));
3134                 if (IS_ERR(bargs)) {
3135                         ret = PTR_ERR(bargs);
3136                         goto out;
3137                 }
3138
3139                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3140                         if (!fs_info->balance_ctl) {
3141                                 ret = -ENOTCONN;
3142                                 goto out_bargs;
3143                         }
3144
3145                         bctl = fs_info->balance_ctl;
3146                         spin_lock(&fs_info->balance_lock);
3147                         bctl->flags |= BTRFS_BALANCE_RESUME;
3148                         spin_unlock(&fs_info->balance_lock);
3149
3150                         goto do_balance;
3151                 }
3152         } else {
3153                 bargs = NULL;
3154         }
3155
3156         if (fs_info->balance_ctl) {
3157                 ret = -EINPROGRESS;
3158                 goto out_bargs;
3159         }
3160
3161         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3162         if (!bctl) {
3163                 ret = -ENOMEM;
3164                 goto out_bargs;
3165         }
3166
3167         bctl->fs_info = fs_info;
3168         if (arg) {
3169                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3170                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3171                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3172
3173                 bctl->flags = bargs->flags;
3174         } else {
3175                 /* balance everything - no filters */
3176                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3177         }
3178
3179 do_balance:
3180         ret = btrfs_balance(bctl, bargs);
3181         /*
3182          * bctl is freed in __cancel_balance or in free_fs_info if
3183          * restriper was paused all the way until unmount
3184          */
3185         if (arg) {
3186                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3187                         ret = -EFAULT;
3188         }
3189
3190 out_bargs:
3191         kfree(bargs);
3192 out:
3193         mutex_unlock(&fs_info->balance_mutex);
3194         mutex_unlock(&fs_info->volume_mutex);
3195         return ret;
3196 }
3197
3198 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3199 {
3200         if (!capable(CAP_SYS_ADMIN))
3201                 return -EPERM;
3202
3203         switch (cmd) {
3204         case BTRFS_BALANCE_CTL_PAUSE:
3205                 return btrfs_pause_balance(root->fs_info);
3206         case BTRFS_BALANCE_CTL_CANCEL:
3207                 return btrfs_cancel_balance(root->fs_info);
3208         }
3209
3210         return -EINVAL;
3211 }
3212
3213 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3214                                          void __user *arg)
3215 {
3216         struct btrfs_fs_info *fs_info = root->fs_info;
3217         struct btrfs_ioctl_balance_args *bargs;
3218         int ret = 0;
3219
3220         if (!capable(CAP_SYS_ADMIN))
3221                 return -EPERM;
3222
3223         mutex_lock(&fs_info->balance_mutex);
3224         if (!fs_info->balance_ctl) {
3225                 ret = -ENOTCONN;
3226                 goto out;
3227         }
3228
3229         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3230         if (!bargs) {
3231                 ret = -ENOMEM;
3232                 goto out;
3233         }
3234
3235         update_ioctl_balance_args(fs_info, 1, bargs);
3236
3237         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3238                 ret = -EFAULT;
3239
3240         kfree(bargs);
3241 out:
3242         mutex_unlock(&fs_info->balance_mutex);
3243         return ret;
3244 }
3245
3246 long btrfs_ioctl(struct file *file, unsigned int
3247                 cmd, unsigned long arg)
3248 {
3249         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3250         void __user *argp = (void __user *)arg;
3251
3252         switch (cmd) {
3253         case FS_IOC_GETFLAGS:
3254                 return btrfs_ioctl_getflags(file, argp);
3255         case FS_IOC_SETFLAGS:
3256                 return btrfs_ioctl_setflags(file, argp);
3257         case FS_IOC_GETVERSION:
3258                 return btrfs_ioctl_getversion(file, argp);
3259         case FITRIM:
3260                 return btrfs_ioctl_fitrim(file, argp);
3261         case BTRFS_IOC_SNAP_CREATE:
3262                 return btrfs_ioctl_snap_create(file, argp, 0);
3263         case BTRFS_IOC_SNAP_CREATE_V2:
3264                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3265         case BTRFS_IOC_SUBVOL_CREATE:
3266                 return btrfs_ioctl_snap_create(file, argp, 1);
3267         case BTRFS_IOC_SNAP_DESTROY:
3268                 return btrfs_ioctl_snap_destroy(file, argp);
3269         case BTRFS_IOC_SUBVOL_GETFLAGS:
3270                 return btrfs_ioctl_subvol_getflags(file, argp);
3271         case BTRFS_IOC_SUBVOL_SETFLAGS:
3272                 return btrfs_ioctl_subvol_setflags(file, argp);
3273         case BTRFS_IOC_DEFAULT_SUBVOL:
3274                 return btrfs_ioctl_default_subvol(file, argp);
3275         case BTRFS_IOC_DEFRAG:
3276                 return btrfs_ioctl_defrag(file, NULL);
3277         case BTRFS_IOC_DEFRAG_RANGE:
3278                 return btrfs_ioctl_defrag(file, argp);
3279         case BTRFS_IOC_RESIZE:
3280                 return btrfs_ioctl_resize(root, argp);
3281         case BTRFS_IOC_ADD_DEV:
3282                 return btrfs_ioctl_add_dev(root, argp);
3283         case BTRFS_IOC_RM_DEV:
3284                 return btrfs_ioctl_rm_dev(root, argp);
3285         case BTRFS_IOC_FS_INFO:
3286                 return btrfs_ioctl_fs_info(root, argp);
3287         case BTRFS_IOC_DEV_INFO:
3288                 return btrfs_ioctl_dev_info(root, argp);
3289         case BTRFS_IOC_BALANCE:
3290                 return btrfs_ioctl_balance(root, NULL);
3291         case BTRFS_IOC_CLONE:
3292                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3293         case BTRFS_IOC_CLONE_RANGE:
3294                 return btrfs_ioctl_clone_range(file, argp);
3295         case BTRFS_IOC_TRANS_START:
3296                 return btrfs_ioctl_trans_start(file);
3297         case BTRFS_IOC_TRANS_END:
3298                 return btrfs_ioctl_trans_end(file);
3299         case BTRFS_IOC_TREE_SEARCH:
3300                 return btrfs_ioctl_tree_search(file, argp);
3301         case BTRFS_IOC_INO_LOOKUP:
3302                 return btrfs_ioctl_ino_lookup(file, argp);
3303         case BTRFS_IOC_INO_PATHS:
3304                 return btrfs_ioctl_ino_to_path(root, argp);
3305         case BTRFS_IOC_LOGICAL_INO:
3306                 return btrfs_ioctl_logical_to_ino(root, argp);
3307         case BTRFS_IOC_SPACE_INFO:
3308                 return btrfs_ioctl_space_info(root, argp);
3309         case BTRFS_IOC_SYNC:
3310                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3311                 return 0;
3312         case BTRFS_IOC_START_SYNC:
3313                 return btrfs_ioctl_start_sync(file, argp);
3314         case BTRFS_IOC_WAIT_SYNC:
3315                 return btrfs_ioctl_wait_sync(file, argp);
3316         case BTRFS_IOC_SCRUB:
3317                 return btrfs_ioctl_scrub(root, argp);
3318         case BTRFS_IOC_SCRUB_CANCEL:
3319                 return btrfs_ioctl_scrub_cancel(root, argp);
3320         case BTRFS_IOC_SCRUB_PROGRESS:
3321                 return btrfs_ioctl_scrub_progress(root, argp);
3322         case BTRFS_IOC_BALANCE_V2:
3323                 return btrfs_ioctl_balance(root, argp);
3324         case BTRFS_IOC_BALANCE_CTL:
3325                 return btrfs_ioctl_balance_ctl(root, arg);
3326         case BTRFS_IOC_BALANCE_PROGRESS:
3327                 return btrfs_ioctl_balance_progress(root, argp);
3328         }
3329
3330         return -ENOTTY;
3331 }