ab620014bcc3a5bd4782f948251bb25a166b6a78
[~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         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
865
866         if (isize == 0)
867                 return 0;
868         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
869
870         ret = btrfs_delalloc_reserve_space(inode,
871                                            num_pages << PAGE_CACHE_SHIFT);
872         if (ret)
873                 return ret;
874 again:
875         ret = 0;
876         i_done = 0;
877
878         /* step one, lock all the pages */
879         for (i = 0; i < num_pages; i++) {
880                 struct page *page;
881                 page = find_or_create_page(inode->i_mapping,
882                                             start_index + i, mask);
883                 if (!page)
884                         break;
885
886                 if (!PageUptodate(page)) {
887                         btrfs_readpage(NULL, page);
888                         lock_page(page);
889                         if (!PageUptodate(page)) {
890                                 unlock_page(page);
891                                 page_cache_release(page);
892                                 ret = -EIO;
893                                 break;
894                         }
895                 }
896                 isize = i_size_read(inode);
897                 file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
898                 if (!isize || page->index > file_end ||
899                     page->mapping != inode->i_mapping) {
900                         /* whoops, we blew past eof, skip this page */
901                         unlock_page(page);
902                         page_cache_release(page);
903                         break;
904                 }
905                 pages[i] = page;
906                 i_done++;
907         }
908         if (!i_done || ret)
909                 goto out;
910
911         if (!(inode->i_sb->s_flags & MS_ACTIVE))
912                 goto out;
913
914         /*
915          * so now we have a nice long stream of locked
916          * and up to date pages, lets wait on them
917          */
918         for (i = 0; i < i_done; i++)
919                 wait_on_page_writeback(pages[i]);
920
921         page_start = page_offset(pages[0]);
922         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
923
924         lock_extent_bits(&BTRFS_I(inode)->io_tree,
925                          page_start, page_end - 1, 0, &cached_state,
926                          GFP_NOFS);
927         ordered = btrfs_lookup_first_ordered_extent(inode, page_end - 1);
928         if (ordered &&
929             ordered->file_offset + ordered->len > page_start &&
930             ordered->file_offset < page_end) {
931                 btrfs_put_ordered_extent(ordered);
932                 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
933                                      page_start, page_end - 1,
934                                      &cached_state, GFP_NOFS);
935                 for (i = 0; i < i_done; i++) {
936                         unlock_page(pages[i]);
937                         page_cache_release(pages[i]);
938                 }
939                 btrfs_wait_ordered_range(inode, page_start,
940                                          page_end - page_start);
941                 goto again;
942         }
943         if (ordered)
944                 btrfs_put_ordered_extent(ordered);
945
946         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
947                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
948                           EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
949                           GFP_NOFS);
950
951         if (i_done != num_pages) {
952                 spin_lock(&BTRFS_I(inode)->lock);
953                 BTRFS_I(inode)->outstanding_extents++;
954                 spin_unlock(&BTRFS_I(inode)->lock);
955                 btrfs_delalloc_release_space(inode,
956                                      (num_pages - i_done) << PAGE_CACHE_SHIFT);
957         }
958
959
960         btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
961                                   &cached_state);
962
963         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
964                              page_start, page_end - 1, &cached_state,
965                              GFP_NOFS);
966
967         for (i = 0; i < i_done; i++) {
968                 clear_page_dirty_for_io(pages[i]);
969                 ClearPageChecked(pages[i]);
970                 set_page_extent_mapped(pages[i]);
971                 set_page_dirty(pages[i]);
972                 unlock_page(pages[i]);
973                 page_cache_release(pages[i]);
974         }
975         return i_done;
976 out:
977         for (i = 0; i < i_done; i++) {
978                 unlock_page(pages[i]);
979                 page_cache_release(pages[i]);
980         }
981         btrfs_delalloc_release_space(inode, num_pages << PAGE_CACHE_SHIFT);
982         return ret;
983
984 }
985
986 int btrfs_defrag_file(struct inode *inode, struct file *file,
987                       struct btrfs_ioctl_defrag_range_args *range,
988                       u64 newer_than, unsigned long max_to_defrag)
989 {
990         struct btrfs_root *root = BTRFS_I(inode)->root;
991         struct btrfs_super_block *disk_super;
992         struct file_ra_state *ra = NULL;
993         unsigned long last_index;
994         u64 isize = i_size_read(inode);
995         u64 features;
996         u64 last_len = 0;
997         u64 skip = 0;
998         u64 defrag_end = 0;
999         u64 newer_off = range->start;
1000         unsigned long i;
1001         unsigned long ra_index = 0;
1002         int ret;
1003         int defrag_count = 0;
1004         int compress_type = BTRFS_COMPRESS_ZLIB;
1005         int extent_thresh = range->extent_thresh;
1006         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1007         int cluster = max_cluster;
1008         u64 new_align = ~((u64)128 * 1024 - 1);
1009         struct page **pages = NULL;
1010
1011         if (extent_thresh == 0)
1012                 extent_thresh = 256 * 1024;
1013
1014         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1015                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1016                         return -EINVAL;
1017                 if (range->compress_type)
1018                         compress_type = range->compress_type;
1019         }
1020
1021         if (isize == 0)
1022                 return 0;
1023
1024         /*
1025          * if we were not given a file, allocate a readahead
1026          * context
1027          */
1028         if (!file) {
1029                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1030                 if (!ra)
1031                         return -ENOMEM;
1032                 file_ra_state_init(ra, inode->i_mapping);
1033         } else {
1034                 ra = &file->f_ra;
1035         }
1036
1037         pages = kmalloc(sizeof(struct page *) * max_cluster,
1038                         GFP_NOFS);
1039         if (!pages) {
1040                 ret = -ENOMEM;
1041                 goto out_ra;
1042         }
1043
1044         /* find the last page to defrag */
1045         if (range->start + range->len > range->start) {
1046                 last_index = min_t(u64, isize - 1,
1047                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1048         } else {
1049                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1050         }
1051
1052         if (newer_than) {
1053                 ret = find_new_extents(root, inode, newer_than,
1054                                        &newer_off, 64 * 1024);
1055                 if (!ret) {
1056                         range->start = newer_off;
1057                         /*
1058                          * we always align our defrag to help keep
1059                          * the extents in the file evenly spaced
1060                          */
1061                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1062                 } else
1063                         goto out_ra;
1064         } else {
1065                 i = range->start >> PAGE_CACHE_SHIFT;
1066         }
1067         if (!max_to_defrag)
1068                 max_to_defrag = last_index;
1069
1070         /*
1071          * make writeback starts from i, so the defrag range can be
1072          * written sequentially.
1073          */
1074         if (i < inode->i_mapping->writeback_index)
1075                 inode->i_mapping->writeback_index = i;
1076
1077         while (i <= last_index && defrag_count < max_to_defrag &&
1078                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1079                 PAGE_CACHE_SHIFT)) {
1080                 /*
1081                  * make sure we stop running if someone unmounts
1082                  * the FS
1083                  */
1084                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1085                         break;
1086
1087                 if (!newer_than &&
1088                     !should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1089                                         PAGE_CACHE_SIZE,
1090                                         extent_thresh,
1091                                         &last_len, &skip,
1092                                         &defrag_end)) {
1093                         unsigned long next;
1094                         /*
1095                          * the should_defrag function tells us how much to skip
1096                          * bump our counter by the suggested amount
1097                          */
1098                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1099                         i = max(i + 1, next);
1100                         continue;
1101                 }
1102
1103                 if (!newer_than) {
1104                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1105                                    PAGE_CACHE_SHIFT) - i;
1106                         cluster = min(cluster, max_cluster);
1107                 } else {
1108                         cluster = max_cluster;
1109                 }
1110
1111                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1112                         BTRFS_I(inode)->force_compress = compress_type;
1113
1114                 if (i + cluster > ra_index) {
1115                         ra_index = max(i, ra_index);
1116                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1117                                        cluster);
1118                         ra_index += max_cluster;
1119                 }
1120
1121                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1122                 if (ret < 0)
1123                         goto out_ra;
1124
1125                 defrag_count += ret;
1126                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1127
1128                 if (newer_than) {
1129                         if (newer_off == (u64)-1)
1130                                 break;
1131
1132                         newer_off = max(newer_off + 1,
1133                                         (u64)i << PAGE_CACHE_SHIFT);
1134
1135                         ret = find_new_extents(root, inode,
1136                                                newer_than, &newer_off,
1137                                                64 * 1024);
1138                         if (!ret) {
1139                                 range->start = newer_off;
1140                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1141                         } else {
1142                                 break;
1143                         }
1144                 } else {
1145                         if (ret > 0) {
1146                                 i += ret;
1147                                 last_len += ret << PAGE_CACHE_SHIFT;
1148                         } else {
1149                                 i++;
1150                                 last_len = 0;
1151                         }
1152                 }
1153         }
1154
1155         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1156                 filemap_flush(inode->i_mapping);
1157
1158         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1159                 /* the filemap_flush will queue IO into the worker threads, but
1160                  * we have to make sure the IO is actually started and that
1161                  * ordered extents get created before we return
1162                  */
1163                 atomic_inc(&root->fs_info->async_submit_draining);
1164                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1165                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1166                         wait_event(root->fs_info->async_submit_wait,
1167                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1168                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1169                 }
1170                 atomic_dec(&root->fs_info->async_submit_draining);
1171
1172                 mutex_lock(&inode->i_mutex);
1173                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1174                 mutex_unlock(&inode->i_mutex);
1175         }
1176
1177         disk_super = root->fs_info->super_copy;
1178         features = btrfs_super_incompat_flags(disk_super);
1179         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1180                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1181                 btrfs_set_super_incompat_flags(disk_super, features);
1182         }
1183
1184         ret = defrag_count;
1185
1186 out_ra:
1187         if (!file)
1188                 kfree(ra);
1189         kfree(pages);
1190         return ret;
1191 }
1192
1193 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
1194                                         void __user *arg)
1195 {
1196         u64 new_size;
1197         u64 old_size;
1198         u64 devid = 1;
1199         struct btrfs_ioctl_vol_args *vol_args;
1200         struct btrfs_trans_handle *trans;
1201         struct btrfs_device *device = NULL;
1202         char *sizestr;
1203         char *devstr = NULL;
1204         int ret = 0;
1205         int mod = 0;
1206
1207         if (root->fs_info->sb->s_flags & MS_RDONLY)
1208                 return -EROFS;
1209
1210         if (!capable(CAP_SYS_ADMIN))
1211                 return -EPERM;
1212
1213         mutex_lock(&root->fs_info->volume_mutex);
1214         if (root->fs_info->balance_ctl) {
1215                 printk(KERN_INFO "btrfs: balance in progress\n");
1216                 ret = -EINVAL;
1217                 goto out;
1218         }
1219
1220         vol_args = memdup_user(arg, sizeof(*vol_args));
1221         if (IS_ERR(vol_args)) {
1222                 ret = PTR_ERR(vol_args);
1223                 goto out;
1224         }
1225
1226         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1227
1228         sizestr = vol_args->name;
1229         devstr = strchr(sizestr, ':');
1230         if (devstr) {
1231                 char *end;
1232                 sizestr = devstr + 1;
1233                 *devstr = '\0';
1234                 devstr = vol_args->name;
1235                 devid = simple_strtoull(devstr, &end, 10);
1236                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1237                        (unsigned long long)devid);
1238         }
1239         device = btrfs_find_device(root, devid, NULL, NULL);
1240         if (!device) {
1241                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1242                        (unsigned long long)devid);
1243                 ret = -EINVAL;
1244                 goto out_free;
1245         }
1246         if (!strcmp(sizestr, "max"))
1247                 new_size = device->bdev->bd_inode->i_size;
1248         else {
1249                 if (sizestr[0] == '-') {
1250                         mod = -1;
1251                         sizestr++;
1252                 } else if (sizestr[0] == '+') {
1253                         mod = 1;
1254                         sizestr++;
1255                 }
1256                 new_size = memparse(sizestr, NULL);
1257                 if (new_size == 0) {
1258                         ret = -EINVAL;
1259                         goto out_free;
1260                 }
1261         }
1262
1263         old_size = device->total_bytes;
1264
1265         if (mod < 0) {
1266                 if (new_size > old_size) {
1267                         ret = -EINVAL;
1268                         goto out_free;
1269                 }
1270                 new_size = old_size - new_size;
1271         } else if (mod > 0) {
1272                 new_size = old_size + new_size;
1273         }
1274
1275         if (new_size < 256 * 1024 * 1024) {
1276                 ret = -EINVAL;
1277                 goto out_free;
1278         }
1279         if (new_size > device->bdev->bd_inode->i_size) {
1280                 ret = -EFBIG;
1281                 goto out_free;
1282         }
1283
1284         do_div(new_size, root->sectorsize);
1285         new_size *= root->sectorsize;
1286
1287         printk(KERN_INFO "btrfs: new size for %s is %llu\n",
1288                 device->name, (unsigned long long)new_size);
1289
1290         if (new_size > old_size) {
1291                 trans = btrfs_start_transaction(root, 0);
1292                 if (IS_ERR(trans)) {
1293                         ret = PTR_ERR(trans);
1294                         goto out_free;
1295                 }
1296                 ret = btrfs_grow_device(trans, device, new_size);
1297                 btrfs_commit_transaction(trans, root);
1298         } else if (new_size < old_size) {
1299                 ret = btrfs_shrink_device(device, new_size);
1300         }
1301
1302 out_free:
1303         kfree(vol_args);
1304 out:
1305         mutex_unlock(&root->fs_info->volume_mutex);
1306         return ret;
1307 }
1308
1309 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1310                                                     char *name,
1311                                                     unsigned long fd,
1312                                                     int subvol,
1313                                                     u64 *transid,
1314                                                     bool readonly)
1315 {
1316         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1317         struct file *src_file;
1318         int namelen;
1319         int ret = 0;
1320
1321         if (root->fs_info->sb->s_flags & MS_RDONLY)
1322                 return -EROFS;
1323
1324         namelen = strlen(name);
1325         if (strchr(name, '/')) {
1326                 ret = -EINVAL;
1327                 goto out;
1328         }
1329
1330         if (subvol) {
1331                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1332                                      NULL, transid, readonly);
1333         } else {
1334                 struct inode *src_inode;
1335                 src_file = fget(fd);
1336                 if (!src_file) {
1337                         ret = -EINVAL;
1338                         goto out;
1339                 }
1340
1341                 src_inode = src_file->f_path.dentry->d_inode;
1342                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1343                         printk(KERN_INFO "btrfs: Snapshot src from "
1344                                "another FS\n");
1345                         ret = -EINVAL;
1346                         fput(src_file);
1347                         goto out;
1348                 }
1349                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1350                                      BTRFS_I(src_inode)->root,
1351                                      transid, readonly);
1352                 fput(src_file);
1353         }
1354 out:
1355         return ret;
1356 }
1357
1358 static noinline int btrfs_ioctl_snap_create(struct file *file,
1359                                             void __user *arg, int subvol)
1360 {
1361         struct btrfs_ioctl_vol_args *vol_args;
1362         int ret;
1363
1364         vol_args = memdup_user(arg, sizeof(*vol_args));
1365         if (IS_ERR(vol_args))
1366                 return PTR_ERR(vol_args);
1367         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1368
1369         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1370                                               vol_args->fd, subvol,
1371                                               NULL, false);
1372
1373         kfree(vol_args);
1374         return ret;
1375 }
1376
1377 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1378                                                void __user *arg, int subvol)
1379 {
1380         struct btrfs_ioctl_vol_args_v2 *vol_args;
1381         int ret;
1382         u64 transid = 0;
1383         u64 *ptr = NULL;
1384         bool readonly = false;
1385
1386         vol_args = memdup_user(arg, sizeof(*vol_args));
1387         if (IS_ERR(vol_args))
1388                 return PTR_ERR(vol_args);
1389         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1390
1391         if (vol_args->flags &
1392             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1393                 ret = -EOPNOTSUPP;
1394                 goto out;
1395         }
1396
1397         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1398                 ptr = &transid;
1399         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1400                 readonly = true;
1401
1402         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1403                                               vol_args->fd, subvol,
1404                                               ptr, readonly);
1405
1406         if (ret == 0 && ptr &&
1407             copy_to_user(arg +
1408                          offsetof(struct btrfs_ioctl_vol_args_v2,
1409                                   transid), ptr, sizeof(*ptr)))
1410                 ret = -EFAULT;
1411 out:
1412         kfree(vol_args);
1413         return ret;
1414 }
1415
1416 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1417                                                 void __user *arg)
1418 {
1419         struct inode *inode = fdentry(file)->d_inode;
1420         struct btrfs_root *root = BTRFS_I(inode)->root;
1421         int ret = 0;
1422         u64 flags = 0;
1423
1424         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1425                 return -EINVAL;
1426
1427         down_read(&root->fs_info->subvol_sem);
1428         if (btrfs_root_readonly(root))
1429                 flags |= BTRFS_SUBVOL_RDONLY;
1430         up_read(&root->fs_info->subvol_sem);
1431
1432         if (copy_to_user(arg, &flags, sizeof(flags)))
1433                 ret = -EFAULT;
1434
1435         return ret;
1436 }
1437
1438 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1439                                               void __user *arg)
1440 {
1441         struct inode *inode = fdentry(file)->d_inode;
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         struct btrfs_trans_handle *trans;
1444         u64 root_flags;
1445         u64 flags;
1446         int ret = 0;
1447
1448         if (root->fs_info->sb->s_flags & MS_RDONLY)
1449                 return -EROFS;
1450
1451         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1452                 return -EINVAL;
1453
1454         if (copy_from_user(&flags, arg, sizeof(flags)))
1455                 return -EFAULT;
1456
1457         if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1458                 return -EINVAL;
1459
1460         if (flags & ~BTRFS_SUBVOL_RDONLY)
1461                 return -EOPNOTSUPP;
1462
1463         if (!inode_owner_or_capable(inode))
1464                 return -EACCES;
1465
1466         down_write(&root->fs_info->subvol_sem);
1467
1468         /* nothing to do */
1469         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1470                 goto out;
1471
1472         root_flags = btrfs_root_flags(&root->root_item);
1473         if (flags & BTRFS_SUBVOL_RDONLY)
1474                 btrfs_set_root_flags(&root->root_item,
1475                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1476         else
1477                 btrfs_set_root_flags(&root->root_item,
1478                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1479
1480         trans = btrfs_start_transaction(root, 1);
1481         if (IS_ERR(trans)) {
1482                 ret = PTR_ERR(trans);
1483                 goto out_reset;
1484         }
1485
1486         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1487                                 &root->root_key, &root->root_item);
1488
1489         btrfs_commit_transaction(trans, root);
1490 out_reset:
1491         if (ret)
1492                 btrfs_set_root_flags(&root->root_item, root_flags);
1493 out:
1494         up_write(&root->fs_info->subvol_sem);
1495         return ret;
1496 }
1497
1498 /*
1499  * helper to check if the subvolume references other subvolumes
1500  */
1501 static noinline int may_destroy_subvol(struct btrfs_root *root)
1502 {
1503         struct btrfs_path *path;
1504         struct btrfs_key key;
1505         int ret;
1506
1507         path = btrfs_alloc_path();
1508         if (!path)
1509                 return -ENOMEM;
1510
1511         key.objectid = root->root_key.objectid;
1512         key.type = BTRFS_ROOT_REF_KEY;
1513         key.offset = (u64)-1;
1514
1515         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1516                                 &key, path, 0, 0);
1517         if (ret < 0)
1518                 goto out;
1519         BUG_ON(ret == 0);
1520
1521         ret = 0;
1522         if (path->slots[0] > 0) {
1523                 path->slots[0]--;
1524                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1525                 if (key.objectid == root->root_key.objectid &&
1526                     key.type == BTRFS_ROOT_REF_KEY)
1527                         ret = -ENOTEMPTY;
1528         }
1529 out:
1530         btrfs_free_path(path);
1531         return ret;
1532 }
1533
1534 static noinline int key_in_sk(struct btrfs_key *key,
1535                               struct btrfs_ioctl_search_key *sk)
1536 {
1537         struct btrfs_key test;
1538         int ret;
1539
1540         test.objectid = sk->min_objectid;
1541         test.type = sk->min_type;
1542         test.offset = sk->min_offset;
1543
1544         ret = btrfs_comp_cpu_keys(key, &test);
1545         if (ret < 0)
1546                 return 0;
1547
1548         test.objectid = sk->max_objectid;
1549         test.type = sk->max_type;
1550         test.offset = sk->max_offset;
1551
1552         ret = btrfs_comp_cpu_keys(key, &test);
1553         if (ret > 0)
1554                 return 0;
1555         return 1;
1556 }
1557
1558 static noinline int copy_to_sk(struct btrfs_root *root,
1559                                struct btrfs_path *path,
1560                                struct btrfs_key *key,
1561                                struct btrfs_ioctl_search_key *sk,
1562                                char *buf,
1563                                unsigned long *sk_offset,
1564                                int *num_found)
1565 {
1566         u64 found_transid;
1567         struct extent_buffer *leaf;
1568         struct btrfs_ioctl_search_header sh;
1569         unsigned long item_off;
1570         unsigned long item_len;
1571         int nritems;
1572         int i;
1573         int slot;
1574         int ret = 0;
1575
1576         leaf = path->nodes[0];
1577         slot = path->slots[0];
1578         nritems = btrfs_header_nritems(leaf);
1579
1580         if (btrfs_header_generation(leaf) > sk->max_transid) {
1581                 i = nritems;
1582                 goto advance_key;
1583         }
1584         found_transid = btrfs_header_generation(leaf);
1585
1586         for (i = slot; i < nritems; i++) {
1587                 item_off = btrfs_item_ptr_offset(leaf, i);
1588                 item_len = btrfs_item_size_nr(leaf, i);
1589
1590                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1591                         item_len = 0;
1592
1593                 if (sizeof(sh) + item_len + *sk_offset >
1594                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1595                         ret = 1;
1596                         goto overflow;
1597                 }
1598
1599                 btrfs_item_key_to_cpu(leaf, key, i);
1600                 if (!key_in_sk(key, sk))
1601                         continue;
1602
1603                 sh.objectid = key->objectid;
1604                 sh.offset = key->offset;
1605                 sh.type = key->type;
1606                 sh.len = item_len;
1607                 sh.transid = found_transid;
1608
1609                 /* copy search result header */
1610                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1611                 *sk_offset += sizeof(sh);
1612
1613                 if (item_len) {
1614                         char *p = buf + *sk_offset;
1615                         /* copy the item */
1616                         read_extent_buffer(leaf, p,
1617                                            item_off, item_len);
1618                         *sk_offset += item_len;
1619                 }
1620                 (*num_found)++;
1621
1622                 if (*num_found >= sk->nr_items)
1623                         break;
1624         }
1625 advance_key:
1626         ret = 0;
1627         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1628                 key->offset++;
1629         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1630                 key->offset = 0;
1631                 key->type++;
1632         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1633                 key->offset = 0;
1634                 key->type = 0;
1635                 key->objectid++;
1636         } else
1637                 ret = 1;
1638 overflow:
1639         return ret;
1640 }
1641
1642 static noinline int search_ioctl(struct inode *inode,
1643                                  struct btrfs_ioctl_search_args *args)
1644 {
1645         struct btrfs_root *root;
1646         struct btrfs_key key;
1647         struct btrfs_key max_key;
1648         struct btrfs_path *path;
1649         struct btrfs_ioctl_search_key *sk = &args->key;
1650         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1651         int ret;
1652         int num_found = 0;
1653         unsigned long sk_offset = 0;
1654
1655         path = btrfs_alloc_path();
1656         if (!path)
1657                 return -ENOMEM;
1658
1659         if (sk->tree_id == 0) {
1660                 /* search the root of the inode that was passed */
1661                 root = BTRFS_I(inode)->root;
1662         } else {
1663                 key.objectid = sk->tree_id;
1664                 key.type = BTRFS_ROOT_ITEM_KEY;
1665                 key.offset = (u64)-1;
1666                 root = btrfs_read_fs_root_no_name(info, &key);
1667                 if (IS_ERR(root)) {
1668                         printk(KERN_ERR "could not find root %llu\n",
1669                                sk->tree_id);
1670                         btrfs_free_path(path);
1671                         return -ENOENT;
1672                 }
1673         }
1674
1675         key.objectid = sk->min_objectid;
1676         key.type = sk->min_type;
1677         key.offset = sk->min_offset;
1678
1679         max_key.objectid = sk->max_objectid;
1680         max_key.type = sk->max_type;
1681         max_key.offset = sk->max_offset;
1682
1683         path->keep_locks = 1;
1684
1685         while(1) {
1686                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1687                                            sk->min_transid);
1688                 if (ret != 0) {
1689                         if (ret > 0)
1690                                 ret = 0;
1691                         goto err;
1692                 }
1693                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1694                                  &sk_offset, &num_found);
1695                 btrfs_release_path(path);
1696                 if (ret || num_found >= sk->nr_items)
1697                         break;
1698
1699         }
1700         ret = 0;
1701 err:
1702         sk->nr_items = num_found;
1703         btrfs_free_path(path);
1704         return ret;
1705 }
1706
1707 static noinline int btrfs_ioctl_tree_search(struct file *file,
1708                                            void __user *argp)
1709 {
1710          struct btrfs_ioctl_search_args *args;
1711          struct inode *inode;
1712          int ret;
1713
1714         if (!capable(CAP_SYS_ADMIN))
1715                 return -EPERM;
1716
1717         args = memdup_user(argp, sizeof(*args));
1718         if (IS_ERR(args))
1719                 return PTR_ERR(args);
1720
1721         inode = fdentry(file)->d_inode;
1722         ret = search_ioctl(inode, args);
1723         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1724                 ret = -EFAULT;
1725         kfree(args);
1726         return ret;
1727 }
1728
1729 /*
1730  * Search INODE_REFs to identify path name of 'dirid' directory
1731  * in a 'tree_id' tree. and sets path name to 'name'.
1732  */
1733 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1734                                 u64 tree_id, u64 dirid, char *name)
1735 {
1736         struct btrfs_root *root;
1737         struct btrfs_key key;
1738         char *ptr;
1739         int ret = -1;
1740         int slot;
1741         int len;
1742         int total_len = 0;
1743         struct btrfs_inode_ref *iref;
1744         struct extent_buffer *l;
1745         struct btrfs_path *path;
1746
1747         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1748                 name[0]='\0';
1749                 return 0;
1750         }
1751
1752         path = btrfs_alloc_path();
1753         if (!path)
1754                 return -ENOMEM;
1755
1756         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1757
1758         key.objectid = tree_id;
1759         key.type = BTRFS_ROOT_ITEM_KEY;
1760         key.offset = (u64)-1;
1761         root = btrfs_read_fs_root_no_name(info, &key);
1762         if (IS_ERR(root)) {
1763                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1764                 ret = -ENOENT;
1765                 goto out;
1766         }
1767
1768         key.objectid = dirid;
1769         key.type = BTRFS_INODE_REF_KEY;
1770         key.offset = (u64)-1;
1771
1772         while(1) {
1773                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1774                 if (ret < 0)
1775                         goto out;
1776
1777                 l = path->nodes[0];
1778                 slot = path->slots[0];
1779                 if (ret > 0 && slot > 0)
1780                         slot--;
1781                 btrfs_item_key_to_cpu(l, &key, slot);
1782
1783                 if (ret > 0 && (key.objectid != dirid ||
1784                                 key.type != BTRFS_INODE_REF_KEY)) {
1785                         ret = -ENOENT;
1786                         goto out;
1787                 }
1788
1789                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1790                 len = btrfs_inode_ref_name_len(l, iref);
1791                 ptr -= len + 1;
1792                 total_len += len + 1;
1793                 if (ptr < name)
1794                         goto out;
1795
1796                 *(ptr + len) = '/';
1797                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1798
1799                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1800                         break;
1801
1802                 btrfs_release_path(path);
1803                 key.objectid = key.offset;
1804                 key.offset = (u64)-1;
1805                 dirid = key.objectid;
1806         }
1807         if (ptr < name)
1808                 goto out;
1809         memmove(name, ptr, total_len);
1810         name[total_len]='\0';
1811         ret = 0;
1812 out:
1813         btrfs_free_path(path);
1814         return ret;
1815 }
1816
1817 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1818                                            void __user *argp)
1819 {
1820          struct btrfs_ioctl_ino_lookup_args *args;
1821          struct inode *inode;
1822          int ret;
1823
1824         if (!capable(CAP_SYS_ADMIN))
1825                 return -EPERM;
1826
1827         args = memdup_user(argp, sizeof(*args));
1828         if (IS_ERR(args))
1829                 return PTR_ERR(args);
1830
1831         inode = fdentry(file)->d_inode;
1832
1833         if (args->treeid == 0)
1834                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1835
1836         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1837                                         args->treeid, args->objectid,
1838                                         args->name);
1839
1840         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1841                 ret = -EFAULT;
1842
1843         kfree(args);
1844         return ret;
1845 }
1846
1847 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1848                                              void __user *arg)
1849 {
1850         struct dentry *parent = fdentry(file);
1851         struct dentry *dentry;
1852         struct inode *dir = parent->d_inode;
1853         struct inode *inode;
1854         struct btrfs_root *root = BTRFS_I(dir)->root;
1855         struct btrfs_root *dest = NULL;
1856         struct btrfs_ioctl_vol_args *vol_args;
1857         struct btrfs_trans_handle *trans;
1858         int namelen;
1859         int ret;
1860         int err = 0;
1861
1862         vol_args = memdup_user(arg, sizeof(*vol_args));
1863         if (IS_ERR(vol_args))
1864                 return PTR_ERR(vol_args);
1865
1866         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1867         namelen = strlen(vol_args->name);
1868         if (strchr(vol_args->name, '/') ||
1869             strncmp(vol_args->name, "..", namelen) == 0) {
1870                 err = -EINVAL;
1871                 goto out;
1872         }
1873
1874         err = mnt_want_write_file(file);
1875         if (err)
1876                 goto out;
1877
1878         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1879         dentry = lookup_one_len(vol_args->name, parent, namelen);
1880         if (IS_ERR(dentry)) {
1881                 err = PTR_ERR(dentry);
1882                 goto out_unlock_dir;
1883         }
1884
1885         if (!dentry->d_inode) {
1886                 err = -ENOENT;
1887                 goto out_dput;
1888         }
1889
1890         inode = dentry->d_inode;
1891         dest = BTRFS_I(inode)->root;
1892         if (!capable(CAP_SYS_ADMIN)){
1893                 /*
1894                  * Regular user.  Only allow this with a special mount
1895                  * option, when the user has write+exec access to the
1896                  * subvol root, and when rmdir(2) would have been
1897                  * allowed.
1898                  *
1899                  * Note that this is _not_ check that the subvol is
1900                  * empty or doesn't contain data that we wouldn't
1901                  * otherwise be able to delete.
1902                  *
1903                  * Users who want to delete empty subvols should try
1904                  * rmdir(2).
1905                  */
1906                 err = -EPERM;
1907                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1908                         goto out_dput;
1909
1910                 /*
1911                  * Do not allow deletion if the parent dir is the same
1912                  * as the dir to be deleted.  That means the ioctl
1913                  * must be called on the dentry referencing the root
1914                  * of the subvol, not a random directory contained
1915                  * within it.
1916                  */
1917                 err = -EINVAL;
1918                 if (root == dest)
1919                         goto out_dput;
1920
1921                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1922                 if (err)
1923                         goto out_dput;
1924
1925                 /* check if subvolume may be deleted by a non-root user */
1926                 err = btrfs_may_delete(dir, dentry, 1);
1927                 if (err)
1928                         goto out_dput;
1929         }
1930
1931         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1932                 err = -EINVAL;
1933                 goto out_dput;
1934         }
1935
1936         mutex_lock(&inode->i_mutex);
1937         err = d_invalidate(dentry);
1938         if (err)
1939                 goto out_unlock;
1940
1941         down_write(&root->fs_info->subvol_sem);
1942
1943         err = may_destroy_subvol(dest);
1944         if (err)
1945                 goto out_up_write;
1946
1947         trans = btrfs_start_transaction(root, 0);
1948         if (IS_ERR(trans)) {
1949                 err = PTR_ERR(trans);
1950                 goto out_up_write;
1951         }
1952         trans->block_rsv = &root->fs_info->global_block_rsv;
1953
1954         ret = btrfs_unlink_subvol(trans, root, dir,
1955                                 dest->root_key.objectid,
1956                                 dentry->d_name.name,
1957                                 dentry->d_name.len);
1958         BUG_ON(ret);
1959
1960         btrfs_record_root_in_trans(trans, dest);
1961
1962         memset(&dest->root_item.drop_progress, 0,
1963                 sizeof(dest->root_item.drop_progress));
1964         dest->root_item.drop_level = 0;
1965         btrfs_set_root_refs(&dest->root_item, 0);
1966
1967         if (!xchg(&dest->orphan_item_inserted, 1)) {
1968                 ret = btrfs_insert_orphan_item(trans,
1969                                         root->fs_info->tree_root,
1970                                         dest->root_key.objectid);
1971                 BUG_ON(ret);
1972         }
1973
1974         ret = btrfs_end_transaction(trans, root);
1975         BUG_ON(ret);
1976         inode->i_flags |= S_DEAD;
1977 out_up_write:
1978         up_write(&root->fs_info->subvol_sem);
1979 out_unlock:
1980         mutex_unlock(&inode->i_mutex);
1981         if (!err) {
1982                 shrink_dcache_sb(root->fs_info->sb);
1983                 btrfs_invalidate_inodes(dest);
1984                 d_delete(dentry);
1985         }
1986 out_dput:
1987         dput(dentry);
1988 out_unlock_dir:
1989         mutex_unlock(&dir->i_mutex);
1990         mnt_drop_write_file(file);
1991 out:
1992         kfree(vol_args);
1993         return err;
1994 }
1995
1996 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1997 {
1998         struct inode *inode = fdentry(file)->d_inode;
1999         struct btrfs_root *root = BTRFS_I(inode)->root;
2000         struct btrfs_ioctl_defrag_range_args *range;
2001         int ret;
2002
2003         if (btrfs_root_readonly(root))
2004                 return -EROFS;
2005
2006         ret = mnt_want_write_file(file);
2007         if (ret)
2008                 return ret;
2009
2010         switch (inode->i_mode & S_IFMT) {
2011         case S_IFDIR:
2012                 if (!capable(CAP_SYS_ADMIN)) {
2013                         ret = -EPERM;
2014                         goto out;
2015                 }
2016                 ret = btrfs_defrag_root(root, 0);
2017                 if (ret)
2018                         goto out;
2019                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2020                 break;
2021         case S_IFREG:
2022                 if (!(file->f_mode & FMODE_WRITE)) {
2023                         ret = -EINVAL;
2024                         goto out;
2025                 }
2026
2027                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2028                 if (!range) {
2029                         ret = -ENOMEM;
2030                         goto out;
2031                 }
2032
2033                 if (argp) {
2034                         if (copy_from_user(range, argp,
2035                                            sizeof(*range))) {
2036                                 ret = -EFAULT;
2037                                 kfree(range);
2038                                 goto out;
2039                         }
2040                         /* compression requires us to start the IO */
2041                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2042                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2043                                 range->extent_thresh = (u32)-1;
2044                         }
2045                 } else {
2046                         /* the rest are all set to zero by kzalloc */
2047                         range->len = (u64)-1;
2048                 }
2049                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2050                                         range, 0, 0);
2051                 if (ret > 0)
2052                         ret = 0;
2053                 kfree(range);
2054                 break;
2055         default:
2056                 ret = -EINVAL;
2057         }
2058 out:
2059         mnt_drop_write_file(file);
2060         return ret;
2061 }
2062
2063 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2064 {
2065         struct btrfs_ioctl_vol_args *vol_args;
2066         int ret;
2067
2068         if (!capable(CAP_SYS_ADMIN))
2069                 return -EPERM;
2070
2071         mutex_lock(&root->fs_info->volume_mutex);
2072         if (root->fs_info->balance_ctl) {
2073                 printk(KERN_INFO "btrfs: balance in progress\n");
2074                 ret = -EINVAL;
2075                 goto out;
2076         }
2077
2078         vol_args = memdup_user(arg, sizeof(*vol_args));
2079         if (IS_ERR(vol_args)) {
2080                 ret = PTR_ERR(vol_args);
2081                 goto out;
2082         }
2083
2084         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2085         ret = btrfs_init_new_device(root, vol_args->name);
2086
2087         kfree(vol_args);
2088 out:
2089         mutex_unlock(&root->fs_info->volume_mutex);
2090         return ret;
2091 }
2092
2093 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
2094 {
2095         struct btrfs_ioctl_vol_args *vol_args;
2096         int ret;
2097
2098         if (!capable(CAP_SYS_ADMIN))
2099                 return -EPERM;
2100
2101         if (root->fs_info->sb->s_flags & MS_RDONLY)
2102                 return -EROFS;
2103
2104         mutex_lock(&root->fs_info->volume_mutex);
2105         if (root->fs_info->balance_ctl) {
2106                 printk(KERN_INFO "btrfs: balance in progress\n");
2107                 ret = -EINVAL;
2108                 goto out;
2109         }
2110
2111         vol_args = memdup_user(arg, sizeof(*vol_args));
2112         if (IS_ERR(vol_args)) {
2113                 ret = PTR_ERR(vol_args);
2114                 goto out;
2115         }
2116
2117         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2118         ret = btrfs_rm_device(root, vol_args->name);
2119
2120         kfree(vol_args);
2121 out:
2122         mutex_unlock(&root->fs_info->volume_mutex);
2123         return ret;
2124 }
2125
2126 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2127 {
2128         struct btrfs_ioctl_fs_info_args *fi_args;
2129         struct btrfs_device *device;
2130         struct btrfs_device *next;
2131         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2132         int ret = 0;
2133
2134         if (!capable(CAP_SYS_ADMIN))
2135                 return -EPERM;
2136
2137         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2138         if (!fi_args)
2139                 return -ENOMEM;
2140
2141         fi_args->num_devices = fs_devices->num_devices;
2142         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2143
2144         mutex_lock(&fs_devices->device_list_mutex);
2145         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2146                 if (device->devid > fi_args->max_id)
2147                         fi_args->max_id = device->devid;
2148         }
2149         mutex_unlock(&fs_devices->device_list_mutex);
2150
2151         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2152                 ret = -EFAULT;
2153
2154         kfree(fi_args);
2155         return ret;
2156 }
2157
2158 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2159 {
2160         struct btrfs_ioctl_dev_info_args *di_args;
2161         struct btrfs_device *dev;
2162         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2163         int ret = 0;
2164         char *s_uuid = NULL;
2165         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2166
2167         if (!capable(CAP_SYS_ADMIN))
2168                 return -EPERM;
2169
2170         di_args = memdup_user(arg, sizeof(*di_args));
2171         if (IS_ERR(di_args))
2172                 return PTR_ERR(di_args);
2173
2174         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2175                 s_uuid = di_args->uuid;
2176
2177         mutex_lock(&fs_devices->device_list_mutex);
2178         dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL);
2179         mutex_unlock(&fs_devices->device_list_mutex);
2180
2181         if (!dev) {
2182                 ret = -ENODEV;
2183                 goto out;
2184         }
2185
2186         di_args->devid = dev->devid;
2187         di_args->bytes_used = dev->bytes_used;
2188         di_args->total_bytes = dev->total_bytes;
2189         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2190         strncpy(di_args->path, dev->name, sizeof(di_args->path));
2191
2192 out:
2193         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2194                 ret = -EFAULT;
2195
2196         kfree(di_args);
2197         return ret;
2198 }
2199
2200 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2201                                        u64 off, u64 olen, u64 destoff)
2202 {
2203         struct inode *inode = fdentry(file)->d_inode;
2204         struct btrfs_root *root = BTRFS_I(inode)->root;
2205         struct file *src_file;
2206         struct inode *src;
2207         struct btrfs_trans_handle *trans;
2208         struct btrfs_path *path;
2209         struct extent_buffer *leaf;
2210         char *buf;
2211         struct btrfs_key key;
2212         u32 nritems;
2213         int slot;
2214         int ret;
2215         u64 len = olen;
2216         u64 bs = root->fs_info->sb->s_blocksize;
2217         u64 hint_byte;
2218
2219         /*
2220          * TODO:
2221          * - split compressed inline extents.  annoying: we need to
2222          *   decompress into destination's address_space (the file offset
2223          *   may change, so source mapping won't do), then recompress (or
2224          *   otherwise reinsert) a subrange.
2225          * - allow ranges within the same file to be cloned (provided
2226          *   they don't overlap)?
2227          */
2228
2229         /* the destination must be opened for writing */
2230         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2231                 return -EINVAL;
2232
2233         if (btrfs_root_readonly(root))
2234                 return -EROFS;
2235
2236         ret = mnt_want_write_file(file);
2237         if (ret)
2238                 return ret;
2239
2240         src_file = fget(srcfd);
2241         if (!src_file) {
2242                 ret = -EBADF;
2243                 goto out_drop_write;
2244         }
2245
2246         src = src_file->f_dentry->d_inode;
2247
2248         ret = -EINVAL;
2249         if (src == inode)
2250                 goto out_fput;
2251
2252         /* the src must be open for reading */
2253         if (!(src_file->f_mode & FMODE_READ))
2254                 goto out_fput;
2255
2256         /* don't make the dst file partly checksummed */
2257         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2258             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2259                 goto out_fput;
2260
2261         ret = -EISDIR;
2262         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2263                 goto out_fput;
2264
2265         ret = -EXDEV;
2266         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
2267                 goto out_fput;
2268
2269         ret = -ENOMEM;
2270         buf = vmalloc(btrfs_level_size(root, 0));
2271         if (!buf)
2272                 goto out_fput;
2273
2274         path = btrfs_alloc_path();
2275         if (!path) {
2276                 vfree(buf);
2277                 goto out_fput;
2278         }
2279         path->reada = 2;
2280
2281         if (inode < src) {
2282                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2283                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2284         } else {
2285                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2286                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2287         }
2288
2289         /* determine range to clone */
2290         ret = -EINVAL;
2291         if (off + len > src->i_size || off + len < off)
2292                 goto out_unlock;
2293         if (len == 0)
2294                 olen = len = src->i_size - off;
2295         /* if we extend to eof, continue to block boundary */
2296         if (off + len == src->i_size)
2297                 len = ALIGN(src->i_size, bs) - off;
2298
2299         /* verify the end result is block aligned */
2300         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2301             !IS_ALIGNED(destoff, bs))
2302                 goto out_unlock;
2303
2304         if (destoff > inode->i_size) {
2305                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2306                 if (ret)
2307                         goto out_unlock;
2308         }
2309
2310         /* truncate page cache pages from target inode range */
2311         truncate_inode_pages_range(&inode->i_data, destoff,
2312                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2313
2314         /* do any pending delalloc/csum calc on src, one way or
2315            another, and lock file content */
2316         while (1) {
2317                 struct btrfs_ordered_extent *ordered;
2318                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2319                 ordered = btrfs_lookup_first_ordered_extent(src, off+len);
2320                 if (!ordered &&
2321                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
2322                                    EXTENT_DELALLOC, 0, NULL))
2323                         break;
2324                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2325                 if (ordered)
2326                         btrfs_put_ordered_extent(ordered);
2327                 btrfs_wait_ordered_range(src, off, len);
2328         }
2329
2330         /* clone data */
2331         key.objectid = btrfs_ino(src);
2332         key.type = BTRFS_EXTENT_DATA_KEY;
2333         key.offset = 0;
2334
2335         while (1) {
2336                 /*
2337                  * note the key will change type as we walk through the
2338                  * tree.
2339                  */
2340                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2341                 if (ret < 0)
2342                         goto out;
2343
2344                 nritems = btrfs_header_nritems(path->nodes[0]);
2345                 if (path->slots[0] >= nritems) {
2346                         ret = btrfs_next_leaf(root, path);
2347                         if (ret < 0)
2348                                 goto out;
2349                         if (ret > 0)
2350                                 break;
2351                         nritems = btrfs_header_nritems(path->nodes[0]);
2352                 }
2353                 leaf = path->nodes[0];
2354                 slot = path->slots[0];
2355
2356                 btrfs_item_key_to_cpu(leaf, &key, slot);
2357                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2358                     key.objectid != btrfs_ino(src))
2359                         break;
2360
2361                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2362                         struct btrfs_file_extent_item *extent;
2363                         int type;
2364                         u32 size;
2365                         struct btrfs_key new_key;
2366                         u64 disko = 0, diskl = 0;
2367                         u64 datao = 0, datal = 0;
2368                         u8 comp;
2369                         u64 endoff;
2370
2371                         size = btrfs_item_size_nr(leaf, slot);
2372                         read_extent_buffer(leaf, buf,
2373                                            btrfs_item_ptr_offset(leaf, slot),
2374                                            size);
2375
2376                         extent = btrfs_item_ptr(leaf, slot,
2377                                                 struct btrfs_file_extent_item);
2378                         comp = btrfs_file_extent_compression(leaf, extent);
2379                         type = btrfs_file_extent_type(leaf, extent);
2380                         if (type == BTRFS_FILE_EXTENT_REG ||
2381                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2382                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2383                                                                       extent);
2384                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2385                                                                  extent);
2386                                 datao = btrfs_file_extent_offset(leaf, extent);
2387                                 datal = btrfs_file_extent_num_bytes(leaf,
2388                                                                     extent);
2389                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2390                                 /* take upper bound, may be compressed */
2391                                 datal = btrfs_file_extent_ram_bytes(leaf,
2392                                                                     extent);
2393                         }
2394                         btrfs_release_path(path);
2395
2396                         if (key.offset + datal <= off ||
2397                             key.offset >= off+len)
2398                                 goto next;
2399
2400                         memcpy(&new_key, &key, sizeof(new_key));
2401                         new_key.objectid = btrfs_ino(inode);
2402                         if (off <= key.offset)
2403                                 new_key.offset = key.offset + destoff - off;
2404                         else
2405                                 new_key.offset = destoff;
2406
2407                         /*
2408                          * 1 - adjusting old extent (we may have to split it)
2409                          * 1 - add new extent
2410                          * 1 - inode update
2411                          */
2412                         trans = btrfs_start_transaction(root, 3);
2413                         if (IS_ERR(trans)) {
2414                                 ret = PTR_ERR(trans);
2415                                 goto out;
2416                         }
2417
2418                         if (type == BTRFS_FILE_EXTENT_REG ||
2419                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2420                                 /*
2421                                  *    a  | --- range to clone ---|  b
2422                                  * | ------------- extent ------------- |
2423                                  */
2424
2425                                 /* substract range b */
2426                                 if (key.offset + datal > off + len)
2427                                         datal = off + len - key.offset;
2428
2429                                 /* substract range a */
2430                                 if (off > key.offset) {
2431                                         datao += off - key.offset;
2432                                         datal -= off - key.offset;
2433                                 }
2434
2435                                 ret = btrfs_drop_extents(trans, inode,
2436                                                          new_key.offset,
2437                                                          new_key.offset + datal,
2438                                                          &hint_byte, 1);
2439                                 BUG_ON(ret);
2440
2441                                 ret = btrfs_insert_empty_item(trans, root, path,
2442                                                               &new_key, size);
2443                                 BUG_ON(ret);
2444
2445                                 leaf = path->nodes[0];
2446                                 slot = path->slots[0];
2447                                 write_extent_buffer(leaf, buf,
2448                                             btrfs_item_ptr_offset(leaf, slot),
2449                                             size);
2450
2451                                 extent = btrfs_item_ptr(leaf, slot,
2452                                                 struct btrfs_file_extent_item);
2453
2454                                 /* disko == 0 means it's a hole */
2455                                 if (!disko)
2456                                         datao = 0;
2457
2458                                 btrfs_set_file_extent_offset(leaf, extent,
2459                                                              datao);
2460                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2461                                                                 datal);
2462                                 if (disko) {
2463                                         inode_add_bytes(inode, datal);
2464                                         ret = btrfs_inc_extent_ref(trans, root,
2465                                                         disko, diskl, 0,
2466                                                         root->root_key.objectid,
2467                                                         btrfs_ino(inode),
2468                                                         new_key.offset - datao,
2469                                                         0);
2470                                         BUG_ON(ret);
2471                                 }
2472                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2473                                 u64 skip = 0;
2474                                 u64 trim = 0;
2475                                 if (off > key.offset) {
2476                                         skip = off - key.offset;
2477                                         new_key.offset += skip;
2478                                 }
2479
2480                                 if (key.offset + datal > off+len)
2481                                         trim = key.offset + datal - (off+len);
2482
2483                                 if (comp && (skip || trim)) {
2484                                         ret = -EINVAL;
2485                                         btrfs_end_transaction(trans, root);
2486                                         goto out;
2487                                 }
2488                                 size -= skip + trim;
2489                                 datal -= skip + trim;
2490
2491                                 ret = btrfs_drop_extents(trans, inode,
2492                                                          new_key.offset,
2493                                                          new_key.offset + datal,
2494                                                          &hint_byte, 1);
2495                                 BUG_ON(ret);
2496
2497                                 ret = btrfs_insert_empty_item(trans, root, path,
2498                                                               &new_key, size);
2499                                 BUG_ON(ret);
2500
2501                                 if (skip) {
2502                                         u32 start =
2503                                           btrfs_file_extent_calc_inline_size(0);
2504                                         memmove(buf+start, buf+start+skip,
2505                                                 datal);
2506                                 }
2507
2508                                 leaf = path->nodes[0];
2509                                 slot = path->slots[0];
2510                                 write_extent_buffer(leaf, buf,
2511                                             btrfs_item_ptr_offset(leaf, slot),
2512                                             size);
2513                                 inode_add_bytes(inode, datal);
2514                         }
2515
2516                         btrfs_mark_buffer_dirty(leaf);
2517                         btrfs_release_path(path);
2518
2519                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2520
2521                         /*
2522                          * we round up to the block size at eof when
2523                          * determining which extents to clone above,
2524                          * but shouldn't round up the file size
2525                          */
2526                         endoff = new_key.offset + datal;
2527                         if (endoff > destoff+olen)
2528                                 endoff = destoff+olen;
2529                         if (endoff > inode->i_size)
2530                                 btrfs_i_size_write(inode, endoff);
2531
2532                         ret = btrfs_update_inode(trans, root, inode);
2533                         BUG_ON(ret);
2534                         btrfs_end_transaction(trans, root);
2535                 }
2536 next:
2537                 btrfs_release_path(path);
2538                 key.offset++;
2539         }
2540         ret = 0;
2541 out:
2542         btrfs_release_path(path);
2543         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2544 out_unlock:
2545         mutex_unlock(&src->i_mutex);
2546         mutex_unlock(&inode->i_mutex);
2547         vfree(buf);
2548         btrfs_free_path(path);
2549 out_fput:
2550         fput(src_file);
2551 out_drop_write:
2552         mnt_drop_write_file(file);
2553         return ret;
2554 }
2555
2556 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2557 {
2558         struct btrfs_ioctl_clone_range_args args;
2559
2560         if (copy_from_user(&args, argp, sizeof(args)))
2561                 return -EFAULT;
2562         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2563                                  args.src_length, args.dest_offset);
2564 }
2565
2566 /*
2567  * there are many ways the trans_start and trans_end ioctls can lead
2568  * to deadlocks.  They should only be used by applications that
2569  * basically own the machine, and have a very in depth understanding
2570  * of all the possible deadlocks and enospc problems.
2571  */
2572 static long btrfs_ioctl_trans_start(struct file *file)
2573 {
2574         struct inode *inode = fdentry(file)->d_inode;
2575         struct btrfs_root *root = BTRFS_I(inode)->root;
2576         struct btrfs_trans_handle *trans;
2577         int ret;
2578
2579         ret = -EPERM;
2580         if (!capable(CAP_SYS_ADMIN))
2581                 goto out;
2582
2583         ret = -EINPROGRESS;
2584         if (file->private_data)
2585                 goto out;
2586
2587         ret = -EROFS;
2588         if (btrfs_root_readonly(root))
2589                 goto out;
2590
2591         ret = mnt_want_write_file(file);
2592         if (ret)
2593                 goto out;
2594
2595         atomic_inc(&root->fs_info->open_ioctl_trans);
2596
2597         ret = -ENOMEM;
2598         trans = btrfs_start_ioctl_transaction(root);
2599         if (IS_ERR(trans))
2600                 goto out_drop;
2601
2602         file->private_data = trans;
2603         return 0;
2604
2605 out_drop:
2606         atomic_dec(&root->fs_info->open_ioctl_trans);
2607         mnt_drop_write_file(file);
2608 out:
2609         return ret;
2610 }
2611
2612 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2613 {
2614         struct inode *inode = fdentry(file)->d_inode;
2615         struct btrfs_root *root = BTRFS_I(inode)->root;
2616         struct btrfs_root *new_root;
2617         struct btrfs_dir_item *di;
2618         struct btrfs_trans_handle *trans;
2619         struct btrfs_path *path;
2620         struct btrfs_key location;
2621         struct btrfs_disk_key disk_key;
2622         struct btrfs_super_block *disk_super;
2623         u64 features;
2624         u64 objectid = 0;
2625         u64 dir_id;
2626
2627         if (!capable(CAP_SYS_ADMIN))
2628                 return -EPERM;
2629
2630         if (copy_from_user(&objectid, argp, sizeof(objectid)))
2631                 return -EFAULT;
2632
2633         if (!objectid)
2634                 objectid = root->root_key.objectid;
2635
2636         location.objectid = objectid;
2637         location.type = BTRFS_ROOT_ITEM_KEY;
2638         location.offset = (u64)-1;
2639
2640         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2641         if (IS_ERR(new_root))
2642                 return PTR_ERR(new_root);
2643
2644         if (btrfs_root_refs(&new_root->root_item) == 0)
2645                 return -ENOENT;
2646
2647         path = btrfs_alloc_path();
2648         if (!path)
2649                 return -ENOMEM;
2650         path->leave_spinning = 1;
2651
2652         trans = btrfs_start_transaction(root, 1);
2653         if (IS_ERR(trans)) {
2654                 btrfs_free_path(path);
2655                 return PTR_ERR(trans);
2656         }
2657
2658         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2659         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2660                                    dir_id, "default", 7, 1);
2661         if (IS_ERR_OR_NULL(di)) {
2662                 btrfs_free_path(path);
2663                 btrfs_end_transaction(trans, root);
2664                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2665                        "this isn't going to work\n");
2666                 return -ENOENT;
2667         }
2668
2669         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2670         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2671         btrfs_mark_buffer_dirty(path->nodes[0]);
2672         btrfs_free_path(path);
2673
2674         disk_super = root->fs_info->super_copy;
2675         features = btrfs_super_incompat_flags(disk_super);
2676         if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2677                 features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2678                 btrfs_set_super_incompat_flags(disk_super, features);
2679         }
2680         btrfs_end_transaction(trans, root);
2681
2682         return 0;
2683 }
2684
2685 static void get_block_group_info(struct list_head *groups_list,
2686                                  struct btrfs_ioctl_space_info *space)
2687 {
2688         struct btrfs_block_group_cache *block_group;
2689
2690         space->total_bytes = 0;
2691         space->used_bytes = 0;
2692         space->flags = 0;
2693         list_for_each_entry(block_group, groups_list, list) {
2694                 space->flags = block_group->flags;
2695                 space->total_bytes += block_group->key.offset;
2696                 space->used_bytes +=
2697                         btrfs_block_group_used(&block_group->item);
2698         }
2699 }
2700
2701 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2702 {
2703         struct btrfs_ioctl_space_args space_args;
2704         struct btrfs_ioctl_space_info space;
2705         struct btrfs_ioctl_space_info *dest;
2706         struct btrfs_ioctl_space_info *dest_orig;
2707         struct btrfs_ioctl_space_info __user *user_dest;
2708         struct btrfs_space_info *info;
2709         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2710                        BTRFS_BLOCK_GROUP_SYSTEM,
2711                        BTRFS_BLOCK_GROUP_METADATA,
2712                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2713         int num_types = 4;
2714         int alloc_size;
2715         int ret = 0;
2716         u64 slot_count = 0;
2717         int i, c;
2718
2719         if (copy_from_user(&space_args,
2720                            (struct btrfs_ioctl_space_args __user *)arg,
2721                            sizeof(space_args)))
2722                 return -EFAULT;
2723
2724         for (i = 0; i < num_types; i++) {
2725                 struct btrfs_space_info *tmp;
2726
2727                 info = NULL;
2728                 rcu_read_lock();
2729                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2730                                         list) {
2731                         if (tmp->flags == types[i]) {
2732                                 info = tmp;
2733                                 break;
2734                         }
2735                 }
2736                 rcu_read_unlock();
2737
2738                 if (!info)
2739                         continue;
2740
2741                 down_read(&info->groups_sem);
2742                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2743                         if (!list_empty(&info->block_groups[c]))
2744                                 slot_count++;
2745                 }
2746                 up_read(&info->groups_sem);
2747         }
2748
2749         /* space_slots == 0 means they are asking for a count */
2750         if (space_args.space_slots == 0) {
2751                 space_args.total_spaces = slot_count;
2752                 goto out;
2753         }
2754
2755         slot_count = min_t(u64, space_args.space_slots, slot_count);
2756
2757         alloc_size = sizeof(*dest) * slot_count;
2758
2759         /* we generally have at most 6 or so space infos, one for each raid
2760          * level.  So, a whole page should be more than enough for everyone
2761          */
2762         if (alloc_size > PAGE_CACHE_SIZE)
2763                 return -ENOMEM;
2764
2765         space_args.total_spaces = 0;
2766         dest = kmalloc(alloc_size, GFP_NOFS);
2767         if (!dest)
2768                 return -ENOMEM;
2769         dest_orig = dest;
2770
2771         /* now we have a buffer to copy into */
2772         for (i = 0; i < num_types; i++) {
2773                 struct btrfs_space_info *tmp;
2774
2775                 if (!slot_count)
2776                         break;
2777
2778                 info = NULL;
2779                 rcu_read_lock();
2780                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2781                                         list) {
2782                         if (tmp->flags == types[i]) {
2783                                 info = tmp;
2784                                 break;
2785                         }
2786                 }
2787                 rcu_read_unlock();
2788
2789                 if (!info)
2790                         continue;
2791                 down_read(&info->groups_sem);
2792                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2793                         if (!list_empty(&info->block_groups[c])) {
2794                                 get_block_group_info(&info->block_groups[c],
2795                                                      &space);
2796                                 memcpy(dest, &space, sizeof(space));
2797                                 dest++;
2798                                 space_args.total_spaces++;
2799                                 slot_count--;
2800                         }
2801                         if (!slot_count)
2802                                 break;
2803                 }
2804                 up_read(&info->groups_sem);
2805         }
2806
2807         user_dest = (struct btrfs_ioctl_space_info *)
2808                 (arg + sizeof(struct btrfs_ioctl_space_args));
2809
2810         if (copy_to_user(user_dest, dest_orig, alloc_size))
2811                 ret = -EFAULT;
2812
2813         kfree(dest_orig);
2814 out:
2815         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2816                 ret = -EFAULT;
2817
2818         return ret;
2819 }
2820
2821 /*
2822  * there are many ways the trans_start and trans_end ioctls can lead
2823  * to deadlocks.  They should only be used by applications that
2824  * basically own the machine, and have a very in depth understanding
2825  * of all the possible deadlocks and enospc problems.
2826  */
2827 long btrfs_ioctl_trans_end(struct file *file)
2828 {
2829         struct inode *inode = fdentry(file)->d_inode;
2830         struct btrfs_root *root = BTRFS_I(inode)->root;
2831         struct btrfs_trans_handle *trans;
2832
2833         trans = file->private_data;
2834         if (!trans)
2835                 return -EINVAL;
2836         file->private_data = NULL;
2837
2838         btrfs_end_transaction(trans, root);
2839
2840         atomic_dec(&root->fs_info->open_ioctl_trans);
2841
2842         mnt_drop_write_file(file);
2843         return 0;
2844 }
2845
2846 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2847 {
2848         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2849         struct btrfs_trans_handle *trans;
2850         u64 transid;
2851         int ret;
2852
2853         trans = btrfs_start_transaction(root, 0);
2854         if (IS_ERR(trans))
2855                 return PTR_ERR(trans);
2856         transid = trans->transid;
2857         ret = btrfs_commit_transaction_async(trans, root, 0);
2858         if (ret) {
2859                 btrfs_end_transaction(trans, root);
2860                 return ret;
2861         }
2862
2863         if (argp)
2864                 if (copy_to_user(argp, &transid, sizeof(transid)))
2865                         return -EFAULT;
2866         return 0;
2867 }
2868
2869 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2870 {
2871         struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2872         u64 transid;
2873
2874         if (argp) {
2875                 if (copy_from_user(&transid, argp, sizeof(transid)))
2876                         return -EFAULT;
2877         } else {
2878                 transid = 0;  /* current trans */
2879         }
2880         return btrfs_wait_for_commit(root, transid);
2881 }
2882
2883 static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg)
2884 {
2885         int ret;
2886         struct btrfs_ioctl_scrub_args *sa;
2887
2888         if (!capable(CAP_SYS_ADMIN))
2889                 return -EPERM;
2890
2891         sa = memdup_user(arg, sizeof(*sa));
2892         if (IS_ERR(sa))
2893                 return PTR_ERR(sa);
2894
2895         ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end,
2896                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY);
2897
2898         if (copy_to_user(arg, sa, sizeof(*sa)))
2899                 ret = -EFAULT;
2900
2901         kfree(sa);
2902         return ret;
2903 }
2904
2905 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
2906 {
2907         if (!capable(CAP_SYS_ADMIN))
2908                 return -EPERM;
2909
2910         return btrfs_scrub_cancel(root);
2911 }
2912
2913 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
2914                                        void __user *arg)
2915 {
2916         struct btrfs_ioctl_scrub_args *sa;
2917         int ret;
2918
2919         if (!capable(CAP_SYS_ADMIN))
2920                 return -EPERM;
2921
2922         sa = memdup_user(arg, sizeof(*sa));
2923         if (IS_ERR(sa))
2924                 return PTR_ERR(sa);
2925
2926         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
2927
2928         if (copy_to_user(arg, sa, sizeof(*sa)))
2929                 ret = -EFAULT;
2930
2931         kfree(sa);
2932         return ret;
2933 }
2934
2935 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
2936 {
2937         int ret = 0;
2938         int i;
2939         u64 rel_ptr;
2940         int size;
2941         struct btrfs_ioctl_ino_path_args *ipa = NULL;
2942         struct inode_fs_paths *ipath = NULL;
2943         struct btrfs_path *path;
2944
2945         if (!capable(CAP_SYS_ADMIN))
2946                 return -EPERM;
2947
2948         path = btrfs_alloc_path();
2949         if (!path) {
2950                 ret = -ENOMEM;
2951                 goto out;
2952         }
2953
2954         ipa = memdup_user(arg, sizeof(*ipa));
2955         if (IS_ERR(ipa)) {
2956                 ret = PTR_ERR(ipa);
2957                 ipa = NULL;
2958                 goto out;
2959         }
2960
2961         size = min_t(u32, ipa->size, 4096);
2962         ipath = init_ipath(size, root, path);
2963         if (IS_ERR(ipath)) {
2964                 ret = PTR_ERR(ipath);
2965                 ipath = NULL;
2966                 goto out;
2967         }
2968
2969         ret = paths_from_inode(ipa->inum, ipath);
2970         if (ret < 0)
2971                 goto out;
2972
2973         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
2974                 rel_ptr = ipath->fspath->val[i] -
2975                           (u64)(unsigned long)ipath->fspath->val;
2976                 ipath->fspath->val[i] = rel_ptr;
2977         }
2978
2979         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
2980                            (void *)(unsigned long)ipath->fspath, size);
2981         if (ret) {
2982                 ret = -EFAULT;
2983                 goto out;
2984         }
2985
2986 out:
2987         btrfs_free_path(path);
2988         free_ipath(ipath);
2989         kfree(ipa);
2990
2991         return ret;
2992 }
2993
2994 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2995 {
2996         struct btrfs_data_container *inodes = ctx;
2997         const size_t c = 3 * sizeof(u64);
2998
2999         if (inodes->bytes_left >= c) {
3000                 inodes->bytes_left -= c;
3001                 inodes->val[inodes->elem_cnt] = inum;
3002                 inodes->val[inodes->elem_cnt + 1] = offset;
3003                 inodes->val[inodes->elem_cnt + 2] = root;
3004                 inodes->elem_cnt += 3;
3005         } else {
3006                 inodes->bytes_missing += c - inodes->bytes_left;
3007                 inodes->bytes_left = 0;
3008                 inodes->elem_missed += 3;
3009         }
3010
3011         return 0;
3012 }
3013
3014 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3015                                         void __user *arg)
3016 {
3017         int ret = 0;
3018         int size;
3019         u64 extent_item_pos;
3020         struct btrfs_ioctl_logical_ino_args *loi;
3021         struct btrfs_data_container *inodes = NULL;
3022         struct btrfs_path *path = NULL;
3023         struct btrfs_key key;
3024
3025         if (!capable(CAP_SYS_ADMIN))
3026                 return -EPERM;
3027
3028         loi = memdup_user(arg, sizeof(*loi));
3029         if (IS_ERR(loi)) {
3030                 ret = PTR_ERR(loi);
3031                 loi = NULL;
3032                 goto out;
3033         }
3034
3035         path = btrfs_alloc_path();
3036         if (!path) {
3037                 ret = -ENOMEM;
3038                 goto out;
3039         }
3040
3041         size = min_t(u32, loi->size, 4096);
3042         inodes = init_data_container(size);
3043         if (IS_ERR(inodes)) {
3044                 ret = PTR_ERR(inodes);
3045                 inodes = NULL;
3046                 goto out;
3047         }
3048
3049         ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
3050         btrfs_release_path(path);
3051
3052         if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
3053                 ret = -ENOENT;
3054         if (ret < 0)
3055                 goto out;
3056
3057         extent_item_pos = loi->logical - key.objectid;
3058         ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
3059                                         extent_item_pos, build_ino_list,
3060                                         inodes);
3061
3062         if (ret < 0)
3063                 goto out;
3064
3065         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3066                            (void *)(unsigned long)inodes, size);
3067         if (ret)
3068                 ret = -EFAULT;
3069
3070 out:
3071         btrfs_free_path(path);
3072         kfree(inodes);
3073         kfree(loi);
3074
3075         return ret;
3076 }
3077
3078 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3079                                struct btrfs_ioctl_balance_args *bargs)
3080 {
3081         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3082
3083         bargs->flags = bctl->flags;
3084
3085         if (atomic_read(&fs_info->balance_running))
3086                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3087         if (atomic_read(&fs_info->balance_pause_req))
3088                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3089         if (atomic_read(&fs_info->balance_cancel_req))
3090                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3091
3092         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3093         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3094         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3095
3096         if (lock) {
3097                 spin_lock(&fs_info->balance_lock);
3098                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3099                 spin_unlock(&fs_info->balance_lock);
3100         } else {
3101                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3102         }
3103 }
3104
3105 static long btrfs_ioctl_balance(struct btrfs_root *root, void __user *arg)
3106 {
3107         struct btrfs_fs_info *fs_info = root->fs_info;
3108         struct btrfs_ioctl_balance_args *bargs;
3109         struct btrfs_balance_control *bctl;
3110         int ret;
3111
3112         if (!capable(CAP_SYS_ADMIN))
3113                 return -EPERM;
3114
3115         if (fs_info->sb->s_flags & MS_RDONLY)
3116                 return -EROFS;
3117
3118         mutex_lock(&fs_info->volume_mutex);
3119         mutex_lock(&fs_info->balance_mutex);
3120
3121         if (arg) {
3122                 bargs = memdup_user(arg, sizeof(*bargs));
3123                 if (IS_ERR(bargs)) {
3124                         ret = PTR_ERR(bargs);
3125                         goto out;
3126                 }
3127
3128                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3129                         if (!fs_info->balance_ctl) {
3130                                 ret = -ENOTCONN;
3131                                 goto out_bargs;
3132                         }
3133
3134                         bctl = fs_info->balance_ctl;
3135                         spin_lock(&fs_info->balance_lock);
3136                         bctl->flags |= BTRFS_BALANCE_RESUME;
3137                         spin_unlock(&fs_info->balance_lock);
3138
3139                         goto do_balance;
3140                 }
3141         } else {
3142                 bargs = NULL;
3143         }
3144
3145         if (fs_info->balance_ctl) {
3146                 ret = -EINPROGRESS;
3147                 goto out_bargs;
3148         }
3149
3150         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3151         if (!bctl) {
3152                 ret = -ENOMEM;
3153                 goto out_bargs;
3154         }
3155
3156         bctl->fs_info = fs_info;
3157         if (arg) {
3158                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3159                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3160                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3161
3162                 bctl->flags = bargs->flags;
3163         } else {
3164                 /* balance everything - no filters */
3165                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3166         }
3167
3168 do_balance:
3169         ret = btrfs_balance(bctl, bargs);
3170         /*
3171          * bctl is freed in __cancel_balance or in free_fs_info if
3172          * restriper was paused all the way until unmount
3173          */
3174         if (arg) {
3175                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3176                         ret = -EFAULT;
3177         }
3178
3179 out_bargs:
3180         kfree(bargs);
3181 out:
3182         mutex_unlock(&fs_info->balance_mutex);
3183         mutex_unlock(&fs_info->volume_mutex);
3184         return ret;
3185 }
3186
3187 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3188 {
3189         if (!capable(CAP_SYS_ADMIN))
3190                 return -EPERM;
3191
3192         switch (cmd) {
3193         case BTRFS_BALANCE_CTL_PAUSE:
3194                 return btrfs_pause_balance(root->fs_info);
3195         case BTRFS_BALANCE_CTL_CANCEL:
3196                 return btrfs_cancel_balance(root->fs_info);
3197         }
3198
3199         return -EINVAL;
3200 }
3201
3202 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3203                                          void __user *arg)
3204 {
3205         struct btrfs_fs_info *fs_info = root->fs_info;
3206         struct btrfs_ioctl_balance_args *bargs;
3207         int ret = 0;
3208
3209         if (!capable(CAP_SYS_ADMIN))
3210                 return -EPERM;
3211
3212         mutex_lock(&fs_info->balance_mutex);
3213         if (!fs_info->balance_ctl) {
3214                 ret = -ENOTCONN;
3215                 goto out;
3216         }
3217
3218         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3219         if (!bargs) {
3220                 ret = -ENOMEM;
3221                 goto out;
3222         }
3223
3224         update_ioctl_balance_args(fs_info, 1, bargs);
3225
3226         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3227                 ret = -EFAULT;
3228
3229         kfree(bargs);
3230 out:
3231         mutex_unlock(&fs_info->balance_mutex);
3232         return ret;
3233 }
3234
3235 long btrfs_ioctl(struct file *file, unsigned int
3236                 cmd, unsigned long arg)
3237 {
3238         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3239         void __user *argp = (void __user *)arg;
3240
3241         switch (cmd) {
3242         case FS_IOC_GETFLAGS:
3243                 return btrfs_ioctl_getflags(file, argp);
3244         case FS_IOC_SETFLAGS:
3245                 return btrfs_ioctl_setflags(file, argp);
3246         case FS_IOC_GETVERSION:
3247                 return btrfs_ioctl_getversion(file, argp);
3248         case FITRIM:
3249                 return btrfs_ioctl_fitrim(file, argp);
3250         case BTRFS_IOC_SNAP_CREATE:
3251                 return btrfs_ioctl_snap_create(file, argp, 0);
3252         case BTRFS_IOC_SNAP_CREATE_V2:
3253                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
3254         case BTRFS_IOC_SUBVOL_CREATE:
3255                 return btrfs_ioctl_snap_create(file, argp, 1);
3256         case BTRFS_IOC_SNAP_DESTROY:
3257                 return btrfs_ioctl_snap_destroy(file, argp);
3258         case BTRFS_IOC_SUBVOL_GETFLAGS:
3259                 return btrfs_ioctl_subvol_getflags(file, argp);
3260         case BTRFS_IOC_SUBVOL_SETFLAGS:
3261                 return btrfs_ioctl_subvol_setflags(file, argp);
3262         case BTRFS_IOC_DEFAULT_SUBVOL:
3263                 return btrfs_ioctl_default_subvol(file, argp);
3264         case BTRFS_IOC_DEFRAG:
3265                 return btrfs_ioctl_defrag(file, NULL);
3266         case BTRFS_IOC_DEFRAG_RANGE:
3267                 return btrfs_ioctl_defrag(file, argp);
3268         case BTRFS_IOC_RESIZE:
3269                 return btrfs_ioctl_resize(root, argp);
3270         case BTRFS_IOC_ADD_DEV:
3271                 return btrfs_ioctl_add_dev(root, argp);
3272         case BTRFS_IOC_RM_DEV:
3273                 return btrfs_ioctl_rm_dev(root, argp);
3274         case BTRFS_IOC_FS_INFO:
3275                 return btrfs_ioctl_fs_info(root, argp);
3276         case BTRFS_IOC_DEV_INFO:
3277                 return btrfs_ioctl_dev_info(root, argp);
3278         case BTRFS_IOC_BALANCE:
3279                 return btrfs_ioctl_balance(root, NULL);
3280         case BTRFS_IOC_CLONE:
3281                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
3282         case BTRFS_IOC_CLONE_RANGE:
3283                 return btrfs_ioctl_clone_range(file, argp);
3284         case BTRFS_IOC_TRANS_START:
3285                 return btrfs_ioctl_trans_start(file);
3286         case BTRFS_IOC_TRANS_END:
3287                 return btrfs_ioctl_trans_end(file);
3288         case BTRFS_IOC_TREE_SEARCH:
3289                 return btrfs_ioctl_tree_search(file, argp);
3290         case BTRFS_IOC_INO_LOOKUP:
3291                 return btrfs_ioctl_ino_lookup(file, argp);
3292         case BTRFS_IOC_INO_PATHS:
3293                 return btrfs_ioctl_ino_to_path(root, argp);
3294         case BTRFS_IOC_LOGICAL_INO:
3295                 return btrfs_ioctl_logical_to_ino(root, argp);
3296         case BTRFS_IOC_SPACE_INFO:
3297                 return btrfs_ioctl_space_info(root, argp);
3298         case BTRFS_IOC_SYNC:
3299                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
3300                 return 0;
3301         case BTRFS_IOC_START_SYNC:
3302                 return btrfs_ioctl_start_sync(file, argp);
3303         case BTRFS_IOC_WAIT_SYNC:
3304                 return btrfs_ioctl_wait_sync(file, argp);
3305         case BTRFS_IOC_SCRUB:
3306                 return btrfs_ioctl_scrub(root, argp);
3307         case BTRFS_IOC_SCRUB_CANCEL:
3308                 return btrfs_ioctl_scrub_cancel(root, argp);
3309         case BTRFS_IOC_SCRUB_PROGRESS:
3310                 return btrfs_ioctl_scrub_progress(root, argp);
3311         case BTRFS_IOC_BALANCE_V2:
3312                 return btrfs_ioctl_balance(root, argp);
3313         case BTRFS_IOC_BALANCE_CTL:
3314                 return btrfs_ioctl_balance_ctl(root, arg);
3315         case BTRFS_IOC_BALANCE_PROGRESS:
3316                 return btrfs_ioctl_balance_progress(root, argp);
3317         }
3318
3319         return -ENOTTY;
3320 }