]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/btrfs/ioctl.c
afbf3ac2079d32df58a2b0912f5f03267bae77bf
[~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 <linux/uuid.h>
45 #include "compat.h"
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "ioctl.h"
51 #include "print-tree.h"
52 #include "volumes.h"
53 #include "locking.h"
54 #include "inode-map.h"
55 #include "backref.h"
56 #include "rcu-string.h"
57 #include "send.h"
58 #include "dev-replace.h"
59
60 /* Mask out flags that are inappropriate for the given type of inode. */
61 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
62 {
63         if (S_ISDIR(mode))
64                 return flags;
65         else if (S_ISREG(mode))
66                 return flags & ~FS_DIRSYNC_FL;
67         else
68                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
69 }
70
71 /*
72  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
73  */
74 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
75 {
76         unsigned int iflags = 0;
77
78         if (flags & BTRFS_INODE_SYNC)
79                 iflags |= FS_SYNC_FL;
80         if (flags & BTRFS_INODE_IMMUTABLE)
81                 iflags |= FS_IMMUTABLE_FL;
82         if (flags & BTRFS_INODE_APPEND)
83                 iflags |= FS_APPEND_FL;
84         if (flags & BTRFS_INODE_NODUMP)
85                 iflags |= FS_NODUMP_FL;
86         if (flags & BTRFS_INODE_NOATIME)
87                 iflags |= FS_NOATIME_FL;
88         if (flags & BTRFS_INODE_DIRSYNC)
89                 iflags |= FS_DIRSYNC_FL;
90         if (flags & BTRFS_INODE_NODATACOW)
91                 iflags |= FS_NOCOW_FL;
92
93         if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS))
94                 iflags |= FS_COMPR_FL;
95         else if (flags & BTRFS_INODE_NOCOMPRESS)
96                 iflags |= FS_NOCOMP_FL;
97
98         return iflags;
99 }
100
101 /*
102  * Update inode->i_flags based on the btrfs internal flags.
103  */
104 void btrfs_update_iflags(struct inode *inode)
105 {
106         struct btrfs_inode *ip = BTRFS_I(inode);
107
108         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
109
110         if (ip->flags & BTRFS_INODE_SYNC)
111                 inode->i_flags |= S_SYNC;
112         if (ip->flags & BTRFS_INODE_IMMUTABLE)
113                 inode->i_flags |= S_IMMUTABLE;
114         if (ip->flags & BTRFS_INODE_APPEND)
115                 inode->i_flags |= S_APPEND;
116         if (ip->flags & BTRFS_INODE_NOATIME)
117                 inode->i_flags |= S_NOATIME;
118         if (ip->flags & BTRFS_INODE_DIRSYNC)
119                 inode->i_flags |= S_DIRSYNC;
120 }
121
122 /*
123  * Inherit flags from the parent inode.
124  *
125  * Currently only the compression flags and the cow flags are inherited.
126  */
127 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
128 {
129         unsigned int flags;
130
131         if (!dir)
132                 return;
133
134         flags = BTRFS_I(dir)->flags;
135
136         if (flags & BTRFS_INODE_NOCOMPRESS) {
137                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
138                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
139         } else if (flags & BTRFS_INODE_COMPRESS) {
140                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
141                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
142         }
143
144         if (flags & BTRFS_INODE_NODATACOW) {
145                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
146                 if (S_ISREG(inode->i_mode))
147                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
148         }
149
150         btrfs_update_iflags(inode);
151 }
152
153 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
154 {
155         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
156         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
157
158         if (copy_to_user(arg, &flags, sizeof(flags)))
159                 return -EFAULT;
160         return 0;
161 }
162
163 static int check_flags(unsigned int flags)
164 {
165         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
166                       FS_NOATIME_FL | FS_NODUMP_FL | \
167                       FS_SYNC_FL | FS_DIRSYNC_FL | \
168                       FS_NOCOMP_FL | FS_COMPR_FL |
169                       FS_NOCOW_FL))
170                 return -EOPNOTSUPP;
171
172         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
173                 return -EINVAL;
174
175         return 0;
176 }
177
178 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
179 {
180         struct inode *inode = file->f_path.dentry->d_inode;
181         struct btrfs_inode *ip = BTRFS_I(inode);
182         struct btrfs_root *root = ip->root;
183         struct btrfs_trans_handle *trans;
184         unsigned int flags, oldflags;
185         int ret;
186         u64 ip_oldflags;
187         unsigned int i_oldflags;
188         umode_t mode;
189
190         if (btrfs_root_readonly(root))
191                 return -EROFS;
192
193         if (copy_from_user(&flags, arg, sizeof(flags)))
194                 return -EFAULT;
195
196         ret = check_flags(flags);
197         if (ret)
198                 return ret;
199
200         if (!inode_owner_or_capable(inode))
201                 return -EACCES;
202
203         ret = mnt_want_write_file(file);
204         if (ret)
205                 return ret;
206
207         mutex_lock(&inode->i_mutex);
208
209         ip_oldflags = ip->flags;
210         i_oldflags = inode->i_flags;
211         mode = inode->i_mode;
212
213         flags = btrfs_mask_flags(inode->i_mode, flags);
214         oldflags = btrfs_flags_to_ioctl(ip->flags);
215         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
216                 if (!capable(CAP_LINUX_IMMUTABLE)) {
217                         ret = -EPERM;
218                         goto out_unlock;
219                 }
220         }
221
222         if (flags & FS_SYNC_FL)
223                 ip->flags |= BTRFS_INODE_SYNC;
224         else
225                 ip->flags &= ~BTRFS_INODE_SYNC;
226         if (flags & FS_IMMUTABLE_FL)
227                 ip->flags |= BTRFS_INODE_IMMUTABLE;
228         else
229                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
230         if (flags & FS_APPEND_FL)
231                 ip->flags |= BTRFS_INODE_APPEND;
232         else
233                 ip->flags &= ~BTRFS_INODE_APPEND;
234         if (flags & FS_NODUMP_FL)
235                 ip->flags |= BTRFS_INODE_NODUMP;
236         else
237                 ip->flags &= ~BTRFS_INODE_NODUMP;
238         if (flags & FS_NOATIME_FL)
239                 ip->flags |= BTRFS_INODE_NOATIME;
240         else
241                 ip->flags &= ~BTRFS_INODE_NOATIME;
242         if (flags & FS_DIRSYNC_FL)
243                 ip->flags |= BTRFS_INODE_DIRSYNC;
244         else
245                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
246         if (flags & FS_NOCOW_FL) {
247                 if (S_ISREG(mode)) {
248                         /*
249                          * It's safe to turn csums off here, no extents exist.
250                          * Otherwise we want the flag to reflect the real COW
251                          * status of the file and will not set it.
252                          */
253                         if (inode->i_size == 0)
254                                 ip->flags |= BTRFS_INODE_NODATACOW
255                                            | BTRFS_INODE_NODATASUM;
256                 } else {
257                         ip->flags |= BTRFS_INODE_NODATACOW;
258                 }
259         } else {
260                 /*
261                  * Revert back under same assuptions as above
262                  */
263                 if (S_ISREG(mode)) {
264                         if (inode->i_size == 0)
265                                 ip->flags &= ~(BTRFS_INODE_NODATACOW
266                                              | BTRFS_INODE_NODATASUM);
267                 } else {
268                         ip->flags &= ~BTRFS_INODE_NODATACOW;
269                 }
270         }
271
272         /*
273          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
274          * flag may be changed automatically if compression code won't make
275          * things smaller.
276          */
277         if (flags & FS_NOCOMP_FL) {
278                 ip->flags &= ~BTRFS_INODE_COMPRESS;
279                 ip->flags |= BTRFS_INODE_NOCOMPRESS;
280         } else if (flags & FS_COMPR_FL) {
281                 ip->flags |= BTRFS_INODE_COMPRESS;
282                 ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
283         } else {
284                 ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
285         }
286
287         trans = btrfs_start_transaction(root, 1);
288         if (IS_ERR(trans)) {
289                 ret = PTR_ERR(trans);
290                 goto out_drop;
291         }
292
293         btrfs_update_iflags(inode);
294         inode_inc_iversion(inode);
295         inode->i_ctime = CURRENT_TIME;
296         ret = btrfs_update_inode(trans, root, inode);
297
298         btrfs_end_transaction(trans, root);
299  out_drop:
300         if (ret) {
301                 ip->flags = ip_oldflags;
302                 inode->i_flags = i_oldflags;
303         }
304
305  out_unlock:
306         mutex_unlock(&inode->i_mutex);
307         mnt_drop_write_file(file);
308         return ret;
309 }
310
311 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
312 {
313         struct inode *inode = file->f_path.dentry->d_inode;
314
315         return put_user(inode->i_generation, arg);
316 }
317
318 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
319 {
320         struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb);
321         struct btrfs_device *device;
322         struct request_queue *q;
323         struct fstrim_range range;
324         u64 minlen = ULLONG_MAX;
325         u64 num_devices = 0;
326         u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
327         int ret;
328
329         if (!capable(CAP_SYS_ADMIN))
330                 return -EPERM;
331
332         rcu_read_lock();
333         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
334                                 dev_list) {
335                 if (!device->bdev)
336                         continue;
337                 q = bdev_get_queue(device->bdev);
338                 if (blk_queue_discard(q)) {
339                         num_devices++;
340                         minlen = min((u64)q->limits.discard_granularity,
341                                      minlen);
342                 }
343         }
344         rcu_read_unlock();
345
346         if (!num_devices)
347                 return -EOPNOTSUPP;
348         if (copy_from_user(&range, arg, sizeof(range)))
349                 return -EFAULT;
350         if (range.start > total_bytes ||
351             range.len < fs_info->sb->s_blocksize)
352                 return -EINVAL;
353
354         range.len = min(range.len, total_bytes - range.start);
355         range.minlen = max(range.minlen, minlen);
356         ret = btrfs_trim_fs(fs_info->tree_root, &range);
357         if (ret < 0)
358                 return ret;
359
360         if (copy_to_user(arg, &range, sizeof(range)))
361                 return -EFAULT;
362
363         return 0;
364 }
365
366 static noinline int create_subvol(struct btrfs_root *root,
367                                   struct dentry *dentry,
368                                   char *name, int namelen,
369                                   u64 *async_transid,
370                                   struct btrfs_qgroup_inherit **inherit)
371 {
372         struct btrfs_trans_handle *trans;
373         struct btrfs_key key;
374         struct btrfs_root_item root_item;
375         struct btrfs_inode_item *inode_item;
376         struct extent_buffer *leaf;
377         struct btrfs_root *new_root;
378         struct dentry *parent = dentry->d_parent;
379         struct inode *dir;
380         struct timespec cur_time = CURRENT_TIME;
381         int ret;
382         int err;
383         u64 objectid;
384         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
385         u64 index = 0;
386         uuid_le new_uuid;
387
388         ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
389         if (ret)
390                 return ret;
391
392         dir = parent->d_inode;
393
394         /*
395          * 1 - inode item
396          * 2 - refs
397          * 1 - root item
398          * 2 - dir items
399          */
400         trans = btrfs_start_transaction(root, 6);
401         if (IS_ERR(trans))
402                 return PTR_ERR(trans);
403
404         ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid,
405                                    inherit ? *inherit : NULL);
406         if (ret)
407                 goto fail;
408
409         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
410                                       0, objectid, NULL, 0, 0, 0);
411         if (IS_ERR(leaf)) {
412                 ret = PTR_ERR(leaf);
413                 goto fail;
414         }
415
416         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
417         btrfs_set_header_bytenr(leaf, leaf->start);
418         btrfs_set_header_generation(leaf, trans->transid);
419         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
420         btrfs_set_header_owner(leaf, objectid);
421
422         write_extent_buffer(leaf, root->fs_info->fsid,
423                             (unsigned long)btrfs_header_fsid(leaf),
424                             BTRFS_FSID_SIZE);
425         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
426                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
427                             BTRFS_UUID_SIZE);
428         btrfs_mark_buffer_dirty(leaf);
429
430         memset(&root_item, 0, sizeof(root_item));
431
432         inode_item = &root_item.inode;
433         inode_item->generation = cpu_to_le64(1);
434         inode_item->size = cpu_to_le64(3);
435         inode_item->nlink = cpu_to_le32(1);
436         inode_item->nbytes = cpu_to_le64(root->leafsize);
437         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
438
439         root_item.flags = 0;
440         root_item.byte_limit = 0;
441         inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
442
443         btrfs_set_root_bytenr(&root_item, leaf->start);
444         btrfs_set_root_generation(&root_item, trans->transid);
445         btrfs_set_root_level(&root_item, 0);
446         btrfs_set_root_refs(&root_item, 1);
447         btrfs_set_root_used(&root_item, leaf->len);
448         btrfs_set_root_last_snapshot(&root_item, 0);
449
450         btrfs_set_root_generation_v2(&root_item,
451                         btrfs_root_generation(&root_item));
452         uuid_le_gen(&new_uuid);
453         memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
454         root_item.otime.sec = cpu_to_le64(cur_time.tv_sec);
455         root_item.otime.nsec = cpu_to_le32(cur_time.tv_nsec);
456         root_item.ctime = root_item.otime;
457         btrfs_set_root_ctransid(&root_item, trans->transid);
458         btrfs_set_root_otransid(&root_item, trans->transid);
459
460         btrfs_tree_unlock(leaf);
461         free_extent_buffer(leaf);
462         leaf = NULL;
463
464         btrfs_set_root_dirid(&root_item, new_dirid);
465
466         key.objectid = objectid;
467         key.offset = 0;
468         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
469         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
470                                 &root_item);
471         if (ret)
472                 goto fail;
473
474         key.offset = (u64)-1;
475         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
476         if (IS_ERR(new_root)) {
477                 btrfs_abort_transaction(trans, root, PTR_ERR(new_root));
478                 ret = PTR_ERR(new_root);
479                 goto fail;
480         }
481
482         btrfs_record_root_in_trans(trans, new_root);
483
484         ret = btrfs_create_subvol_root(trans, new_root, new_dirid);
485         if (ret) {
486                 /* We potentially lose an unused inode item here */
487                 btrfs_abort_transaction(trans, root, ret);
488                 goto fail;
489         }
490
491         /*
492          * insert the directory item
493          */
494         ret = btrfs_set_inode_index(dir, &index);
495         if (ret) {
496                 btrfs_abort_transaction(trans, root, ret);
497                 goto fail;
498         }
499
500         ret = btrfs_insert_dir_item(trans, root,
501                                     name, namelen, dir, &key,
502                                     BTRFS_FT_DIR, index);
503         if (ret) {
504                 btrfs_abort_transaction(trans, root, ret);
505                 goto fail;
506         }
507
508         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
509         ret = btrfs_update_inode(trans, root, dir);
510         BUG_ON(ret);
511
512         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
513                                  objectid, root->root_key.objectid,
514                                  btrfs_ino(dir), index, name, namelen);
515
516         BUG_ON(ret);
517
518         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
519 fail:
520         if (async_transid) {
521                 *async_transid = trans->transid;
522                 err = btrfs_commit_transaction_async(trans, root, 1);
523         } else {
524                 err = btrfs_commit_transaction(trans, root);
525         }
526         if (err && !ret)
527                 ret = err;
528         return ret;
529 }
530
531 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
532                            char *name, int namelen, u64 *async_transid,
533                            bool readonly, struct btrfs_qgroup_inherit **inherit)
534 {
535         struct inode *inode;
536         struct btrfs_pending_snapshot *pending_snapshot;
537         struct btrfs_trans_handle *trans;
538         int ret;
539
540         if (!root->ref_cows)
541                 return -EINVAL;
542
543         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
544         if (!pending_snapshot)
545                 return -ENOMEM;
546
547         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
548                              BTRFS_BLOCK_RSV_TEMP);
549         pending_snapshot->dentry = dentry;
550         pending_snapshot->root = root;
551         pending_snapshot->readonly = readonly;
552         if (inherit) {
553                 pending_snapshot->inherit = *inherit;
554                 *inherit = NULL;        /* take responsibility to free it */
555         }
556
557         trans = btrfs_start_transaction(root->fs_info->extent_root, 6);
558         if (IS_ERR(trans)) {
559                 ret = PTR_ERR(trans);
560                 goto fail;
561         }
562
563         ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
564         BUG_ON(ret);
565
566         spin_lock(&root->fs_info->trans_lock);
567         list_add(&pending_snapshot->list,
568                  &trans->transaction->pending_snapshots);
569         spin_unlock(&root->fs_info->trans_lock);
570         if (async_transid) {
571                 *async_transid = trans->transid;
572                 ret = btrfs_commit_transaction_async(trans,
573                                      root->fs_info->extent_root, 1);
574         } else {
575                 ret = btrfs_commit_transaction(trans,
576                                                root->fs_info->extent_root);
577         }
578         if (ret) {
579                 /* cleanup_transaction has freed this for us */
580                 if (trans->aborted)
581                         pending_snapshot = NULL;
582                 goto fail;
583         }
584
585         ret = pending_snapshot->error;
586         if (ret)
587                 goto fail;
588
589         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
590         if (ret)
591                 goto fail;
592
593         inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry);
594         if (IS_ERR(inode)) {
595                 ret = PTR_ERR(inode);
596                 goto fail;
597         }
598         BUG_ON(!inode);
599         d_instantiate(dentry, inode);
600         ret = 0;
601 fail:
602         kfree(pending_snapshot);
603         return ret;
604 }
605
606 /*  copy of check_sticky in fs/namei.c()
607 * It's inline, so penalty for filesystems that don't use sticky bit is
608 * minimal.
609 */
610 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
611 {
612         kuid_t fsuid = current_fsuid();
613
614         if (!(dir->i_mode & S_ISVTX))
615                 return 0;
616         if (uid_eq(inode->i_uid, fsuid))
617                 return 0;
618         if (uid_eq(dir->i_uid, fsuid))
619                 return 0;
620         return !capable(CAP_FOWNER);
621 }
622
623 /*  copy of may_delete in fs/namei.c()
624  *      Check whether we can remove a link victim from directory dir, check
625  *  whether the type of victim is right.
626  *  1. We can't do it if dir is read-only (done in permission())
627  *  2. We should have write and exec permissions on dir
628  *  3. We can't remove anything from append-only dir
629  *  4. We can't do anything with immutable dir (done in permission())
630  *  5. If the sticky bit on dir is set we should either
631  *      a. be owner of dir, or
632  *      b. be owner of victim, or
633  *      c. have CAP_FOWNER capability
634  *  6. If the victim is append-only or immutable we can't do antyhing with
635  *     links pointing to it.
636  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
637  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
638  *  9. We can't remove a root or mountpoint.
639  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
640  *     nfs_async_unlink().
641  */
642
643 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
644 {
645         int error;
646
647         if (!victim->d_inode)
648                 return -ENOENT;
649
650         BUG_ON(victim->d_parent->d_inode != dir);
651         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
652
653         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
654         if (error)
655                 return error;
656         if (IS_APPEND(dir))
657                 return -EPERM;
658         if (btrfs_check_sticky(dir, victim->d_inode)||
659                 IS_APPEND(victim->d_inode)||
660             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
661                 return -EPERM;
662         if (isdir) {
663                 if (!S_ISDIR(victim->d_inode->i_mode))
664                         return -ENOTDIR;
665                 if (IS_ROOT(victim))
666                         return -EBUSY;
667         } else if (S_ISDIR(victim->d_inode->i_mode))
668                 return -EISDIR;
669         if (IS_DEADDIR(dir))
670                 return -ENOENT;
671         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
672                 return -EBUSY;
673         return 0;
674 }
675
676 /* copy of may_create in fs/namei.c() */
677 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
678 {
679         if (child->d_inode)
680                 return -EEXIST;
681         if (IS_DEADDIR(dir))
682                 return -ENOENT;
683         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
684 }
685
686 /*
687  * Create a new subvolume below @parent.  This is largely modeled after
688  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
689  * inside this filesystem so it's quite a bit simpler.
690  */
691 static noinline int btrfs_mksubvol(struct path *parent,
692                                    char *name, int namelen,
693                                    struct btrfs_root *snap_src,
694                                    u64 *async_transid, bool readonly,
695                                    struct btrfs_qgroup_inherit **inherit)
696 {
697         struct inode *dir  = parent->dentry->d_inode;
698         struct dentry *dentry;
699         int error;
700
701         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
702
703         dentry = lookup_one_len(name, parent->dentry, namelen);
704         error = PTR_ERR(dentry);
705         if (IS_ERR(dentry))
706                 goto out_unlock;
707
708         error = -EEXIST;
709         if (dentry->d_inode)
710                 goto out_dput;
711
712         error = btrfs_may_create(dir, dentry);
713         if (error)
714                 goto out_dput;
715
716         /*
717          * even if this name doesn't exist, we may get hash collisions.
718          * check for them now when we can safely fail
719          */
720         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
721                                                dir->i_ino, name,
722                                                namelen);
723         if (error)
724                 goto out_dput;
725
726         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
727
728         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
729                 goto out_up_read;
730
731         if (snap_src) {
732                 error = create_snapshot(snap_src, dentry, name, namelen,
733                                         async_transid, readonly, inherit);
734         } else {
735                 error = create_subvol(BTRFS_I(dir)->root, dentry,
736                                       name, namelen, async_transid, inherit);
737         }
738         if (!error)
739                 fsnotify_mkdir(dir, dentry);
740 out_up_read:
741         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
742 out_dput:
743         dput(dentry);
744 out_unlock:
745         mutex_unlock(&dir->i_mutex);
746         return error;
747 }
748
749 /*
750  * When we're defragging a range, we don't want to kick it off again
751  * if it is really just waiting for delalloc to send it down.
752  * If we find a nice big extent or delalloc range for the bytes in the
753  * file you want to defrag, we return 0 to let you know to skip this
754  * part of the file
755  */
756 static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh)
757 {
758         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
759         struct extent_map *em = NULL;
760         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
761         u64 end;
762
763         read_lock(&em_tree->lock);
764         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
765         read_unlock(&em_tree->lock);
766
767         if (em) {
768                 end = extent_map_end(em);
769                 free_extent_map(em);
770                 if (end - offset > thresh)
771                         return 0;
772         }
773         /* if we already have a nice delalloc here, just stop */
774         thresh /= 2;
775         end = count_range_bits(io_tree, &offset, offset + thresh,
776                                thresh, EXTENT_DELALLOC, 1);
777         if (end >= thresh)
778                 return 0;
779         return 1;
780 }
781
782 /*
783  * helper function to walk through a file and find extents
784  * newer than a specific transid, and smaller than thresh.
785  *
786  * This is used by the defragging code to find new and small
787  * extents
788  */
789 static int find_new_extents(struct btrfs_root *root,
790                             struct inode *inode, u64 newer_than,
791                             u64 *off, int thresh)
792 {
793         struct btrfs_path *path;
794         struct btrfs_key min_key;
795         struct btrfs_key max_key;
796         struct extent_buffer *leaf;
797         struct btrfs_file_extent_item *extent;
798         int type;
799         int ret;
800         u64 ino = btrfs_ino(inode);
801
802         path = btrfs_alloc_path();
803         if (!path)
804                 return -ENOMEM;
805
806         min_key.objectid = ino;
807         min_key.type = BTRFS_EXTENT_DATA_KEY;
808         min_key.offset = *off;
809
810         max_key.objectid = ino;
811         max_key.type = (u8)-1;
812         max_key.offset = (u64)-1;
813
814         path->keep_locks = 1;
815
816         while(1) {
817                 ret = btrfs_search_forward(root, &min_key, &max_key,
818                                            path, 0, newer_than);
819                 if (ret != 0)
820                         goto none;
821                 if (min_key.objectid != ino)
822                         goto none;
823                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
824                         goto none;
825
826                 leaf = path->nodes[0];
827                 extent = btrfs_item_ptr(leaf, path->slots[0],
828                                         struct btrfs_file_extent_item);
829
830                 type = btrfs_file_extent_type(leaf, extent);
831                 if (type == BTRFS_FILE_EXTENT_REG &&
832                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
833                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
834                         *off = min_key.offset;
835                         btrfs_free_path(path);
836                         return 0;
837                 }
838
839                 if (min_key.offset == (u64)-1)
840                         goto none;
841
842                 min_key.offset++;
843                 btrfs_release_path(path);
844         }
845 none:
846         btrfs_free_path(path);
847         return -ENOENT;
848 }
849
850 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
851 {
852         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
853         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
854         struct extent_map *em;
855         u64 len = PAGE_CACHE_SIZE;
856
857         /*
858          * hopefully we have this extent in the tree already, try without
859          * the full extent lock
860          */
861         read_lock(&em_tree->lock);
862         em = lookup_extent_mapping(em_tree, start, len);
863         read_unlock(&em_tree->lock);
864
865         if (!em) {
866                 /* get the big lock and read metadata off disk */
867                 lock_extent(io_tree, start, start + len - 1);
868                 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
869                 unlock_extent(io_tree, start, start + len - 1);
870
871                 if (IS_ERR(em))
872                         return NULL;
873         }
874
875         return em;
876 }
877
878 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
879 {
880         struct extent_map *next;
881         bool ret = true;
882
883         /* this is the last extent */
884         if (em->start + em->len >= i_size_read(inode))
885                 return false;
886
887         next = defrag_lookup_extent(inode, em->start + em->len);
888         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
889                 ret = false;
890
891         free_extent_map(next);
892         return ret;
893 }
894
895 static int should_defrag_range(struct inode *inode, u64 start, int thresh,
896                                u64 *last_len, u64 *skip, u64 *defrag_end,
897                                int compress)
898 {
899         struct extent_map *em;
900         int ret = 1;
901         bool next_mergeable = true;
902
903         /*
904          * make sure that once we start defragging an extent, we keep on
905          * defragging it
906          */
907         if (start < *defrag_end)
908                 return 1;
909
910         *skip = 0;
911
912         em = defrag_lookup_extent(inode, start);
913         if (!em)
914                 return 0;
915
916         /* this will cover holes, and inline extents */
917         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
918                 ret = 0;
919                 goto out;
920         }
921
922         next_mergeable = defrag_check_next_extent(inode, em);
923
924         /*
925          * we hit a real extent, if it is big or the next extent is not a
926          * real extent, don't bother defragging it
927          */
928         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
929             (em->len >= thresh || !next_mergeable))
930                 ret = 0;
931 out:
932         /*
933          * last_len ends up being a counter of how many bytes we've defragged.
934          * every time we choose not to defrag an extent, we reset *last_len
935          * so that the next tiny extent will force a defrag.
936          *
937          * The end result of this is that tiny extents before a single big
938          * extent will force at least part of that big extent to be defragged.
939          */
940         if (ret) {
941                 *defrag_end = extent_map_end(em);
942         } else {
943                 *last_len = 0;
944                 *skip = extent_map_end(em);
945                 *defrag_end = 0;
946         }
947
948         free_extent_map(em);
949         return ret;
950 }
951
952 /*
953  * it doesn't do much good to defrag one or two pages
954  * at a time.  This pulls in a nice chunk of pages
955  * to COW and defrag.
956  *
957  * It also makes sure the delalloc code has enough
958  * dirty data to avoid making new small extents as part
959  * of the defrag
960  *
961  * It's a good idea to start RA on this range
962  * before calling this.
963  */
964 static int cluster_pages_for_defrag(struct inode *inode,
965                                     struct page **pages,
966                                     unsigned long start_index,
967                                     int num_pages)
968 {
969         unsigned long file_end;
970         u64 isize = i_size_read(inode);
971         u64 page_start;
972         u64 page_end;
973         u64 page_cnt;
974         int ret;
975         int i;
976         int i_done;
977         struct btrfs_ordered_extent *ordered;
978         struct extent_state *cached_state = NULL;
979         struct extent_io_tree *tree;
980         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
981
982         file_end = (isize - 1) >> PAGE_CACHE_SHIFT;
983         if (!isize || start_index > file_end)
984                 return 0;
985
986         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
987
988         ret = btrfs_delalloc_reserve_space(inode,
989                                            page_cnt << PAGE_CACHE_SHIFT);
990         if (ret)
991                 return ret;
992         i_done = 0;
993         tree = &BTRFS_I(inode)->io_tree;
994
995         /* step one, lock all the pages */
996         for (i = 0; i < page_cnt; i++) {
997                 struct page *page;
998 again:
999                 page = find_or_create_page(inode->i_mapping,
1000                                            start_index + i, mask);
1001                 if (!page)
1002                         break;
1003
1004                 page_start = page_offset(page);
1005                 page_end = page_start + PAGE_CACHE_SIZE - 1;
1006                 while (1) {
1007                         lock_extent(tree, page_start, page_end);
1008                         ordered = btrfs_lookup_ordered_extent(inode,
1009                                                               page_start);
1010                         unlock_extent(tree, page_start, page_end);
1011                         if (!ordered)
1012                                 break;
1013
1014                         unlock_page(page);
1015                         btrfs_start_ordered_extent(inode, ordered, 1);
1016                         btrfs_put_ordered_extent(ordered);
1017                         lock_page(page);
1018                         /*
1019                          * we unlocked the page above, so we need check if
1020                          * it was released or not.
1021                          */
1022                         if (page->mapping != inode->i_mapping) {
1023                                 unlock_page(page);
1024                                 page_cache_release(page);
1025                                 goto again;
1026                         }
1027                 }
1028
1029                 if (!PageUptodate(page)) {
1030                         btrfs_readpage(NULL, page);
1031                         lock_page(page);
1032                         if (!PageUptodate(page)) {
1033                                 unlock_page(page);
1034                                 page_cache_release(page);
1035                                 ret = -EIO;
1036                                 break;
1037                         }
1038                 }
1039
1040                 if (page->mapping != inode->i_mapping) {
1041                         unlock_page(page);
1042                         page_cache_release(page);
1043                         goto again;
1044                 }
1045
1046                 pages[i] = page;
1047                 i_done++;
1048         }
1049         if (!i_done || ret)
1050                 goto out;
1051
1052         if (!(inode->i_sb->s_flags & MS_ACTIVE))
1053                 goto out;
1054
1055         /*
1056          * so now we have a nice long stream of locked
1057          * and up to date pages, lets wait on them
1058          */
1059         for (i = 0; i < i_done; i++)
1060                 wait_on_page_writeback(pages[i]);
1061
1062         page_start = page_offset(pages[0]);
1063         page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE;
1064
1065         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1066                          page_start, page_end - 1, 0, &cached_state);
1067         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1068                           page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1069                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
1070                           &cached_state, GFP_NOFS);
1071
1072         if (i_done != page_cnt) {
1073                 spin_lock(&BTRFS_I(inode)->lock);
1074                 BTRFS_I(inode)->outstanding_extents++;
1075                 spin_unlock(&BTRFS_I(inode)->lock);
1076                 btrfs_delalloc_release_space(inode,
1077                                      (page_cnt - i_done) << PAGE_CACHE_SHIFT);
1078         }
1079
1080
1081         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1082                           &cached_state, GFP_NOFS);
1083
1084         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1085                              page_start, page_end - 1, &cached_state,
1086                              GFP_NOFS);
1087
1088         for (i = 0; i < i_done; i++) {
1089                 clear_page_dirty_for_io(pages[i]);
1090                 ClearPageChecked(pages[i]);
1091                 set_page_extent_mapped(pages[i]);
1092                 set_page_dirty(pages[i]);
1093                 unlock_page(pages[i]);
1094                 page_cache_release(pages[i]);
1095         }
1096         return i_done;
1097 out:
1098         for (i = 0; i < i_done; i++) {
1099                 unlock_page(pages[i]);
1100                 page_cache_release(pages[i]);
1101         }
1102         btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT);
1103         return ret;
1104
1105 }
1106
1107 int btrfs_defrag_file(struct inode *inode, struct file *file,
1108                       struct btrfs_ioctl_defrag_range_args *range,
1109                       u64 newer_than, unsigned long max_to_defrag)
1110 {
1111         struct btrfs_root *root = BTRFS_I(inode)->root;
1112         struct file_ra_state *ra = NULL;
1113         unsigned long last_index;
1114         u64 isize = i_size_read(inode);
1115         u64 last_len = 0;
1116         u64 skip = 0;
1117         u64 defrag_end = 0;
1118         u64 newer_off = range->start;
1119         unsigned long i;
1120         unsigned long ra_index = 0;
1121         int ret;
1122         int defrag_count = 0;
1123         int compress_type = BTRFS_COMPRESS_ZLIB;
1124         int extent_thresh = range->extent_thresh;
1125         int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT;
1126         int cluster = max_cluster;
1127         u64 new_align = ~((u64)128 * 1024 - 1);
1128         struct page **pages = NULL;
1129
1130         if (extent_thresh == 0)
1131                 extent_thresh = 256 * 1024;
1132
1133         if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
1134                 if (range->compress_type > BTRFS_COMPRESS_TYPES)
1135                         return -EINVAL;
1136                 if (range->compress_type)
1137                         compress_type = range->compress_type;
1138         }
1139
1140         if (isize == 0)
1141                 return 0;
1142
1143         /*
1144          * if we were not given a file, allocate a readahead
1145          * context
1146          */
1147         if (!file) {
1148                 ra = kzalloc(sizeof(*ra), GFP_NOFS);
1149                 if (!ra)
1150                         return -ENOMEM;
1151                 file_ra_state_init(ra, inode->i_mapping);
1152         } else {
1153                 ra = &file->f_ra;
1154         }
1155
1156         pages = kmalloc(sizeof(struct page *) * max_cluster,
1157                         GFP_NOFS);
1158         if (!pages) {
1159                 ret = -ENOMEM;
1160                 goto out_ra;
1161         }
1162
1163         /* find the last page to defrag */
1164         if (range->start + range->len > range->start) {
1165                 last_index = min_t(u64, isize - 1,
1166                          range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
1167         } else {
1168                 last_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1169         }
1170
1171         if (newer_than) {
1172                 ret = find_new_extents(root, inode, newer_than,
1173                                        &newer_off, 64 * 1024);
1174                 if (!ret) {
1175                         range->start = newer_off;
1176                         /*
1177                          * we always align our defrag to help keep
1178                          * the extents in the file evenly spaced
1179                          */
1180                         i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1181                 } else
1182                         goto out_ra;
1183         } else {
1184                 i = range->start >> PAGE_CACHE_SHIFT;
1185         }
1186         if (!max_to_defrag)
1187                 max_to_defrag = last_index + 1;
1188
1189         /*
1190          * make writeback starts from i, so the defrag range can be
1191          * written sequentially.
1192          */
1193         if (i < inode->i_mapping->writeback_index)
1194                 inode->i_mapping->writeback_index = i;
1195
1196         while (i <= last_index && defrag_count < max_to_defrag &&
1197                (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
1198                 PAGE_CACHE_SHIFT)) {
1199                 /*
1200                  * make sure we stop running if someone unmounts
1201                  * the FS
1202                  */
1203                 if (!(inode->i_sb->s_flags & MS_ACTIVE))
1204                         break;
1205
1206                 if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
1207                                          extent_thresh, &last_len, &skip,
1208                                          &defrag_end, range->flags &
1209                                          BTRFS_DEFRAG_RANGE_COMPRESS)) {
1210                         unsigned long next;
1211                         /*
1212                          * the should_defrag function tells us how much to skip
1213                          * bump our counter by the suggested amount
1214                          */
1215                         next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1216                         i = max(i + 1, next);
1217                         continue;
1218                 }
1219
1220                 if (!newer_than) {
1221                         cluster = (PAGE_CACHE_ALIGN(defrag_end) >>
1222                                    PAGE_CACHE_SHIFT) - i;
1223                         cluster = min(cluster, max_cluster);
1224                 } else {
1225                         cluster = max_cluster;
1226                 }
1227
1228                 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
1229                         BTRFS_I(inode)->force_compress = compress_type;
1230
1231                 if (i + cluster > ra_index) {
1232                         ra_index = max(i, ra_index);
1233                         btrfs_force_ra(inode->i_mapping, ra, file, ra_index,
1234                                        cluster);
1235                         ra_index += max_cluster;
1236                 }
1237
1238                 mutex_lock(&inode->i_mutex);
1239                 ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1240                 if (ret < 0) {
1241                         mutex_unlock(&inode->i_mutex);
1242                         goto out_ra;
1243                 }
1244
1245                 defrag_count += ret;
1246                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret);
1247                 mutex_unlock(&inode->i_mutex);
1248
1249                 if (newer_than) {
1250                         if (newer_off == (u64)-1)
1251                                 break;
1252
1253                         if (ret > 0)
1254                                 i += ret;
1255
1256                         newer_off = max(newer_off + 1,
1257                                         (u64)i << PAGE_CACHE_SHIFT);
1258
1259                         ret = find_new_extents(root, inode,
1260                                                newer_than, &newer_off,
1261                                                64 * 1024);
1262                         if (!ret) {
1263                                 range->start = newer_off;
1264                                 i = (newer_off & new_align) >> PAGE_CACHE_SHIFT;
1265                         } else {
1266                                 break;
1267                         }
1268                 } else {
1269                         if (ret > 0) {
1270                                 i += ret;
1271                                 last_len += ret << PAGE_CACHE_SHIFT;
1272                         } else {
1273                                 i++;
1274                                 last_len = 0;
1275                         }
1276                 }
1277         }
1278
1279         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
1280                 filemap_flush(inode->i_mapping);
1281
1282         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1283                 /* the filemap_flush will queue IO into the worker threads, but
1284                  * we have to make sure the IO is actually started and that
1285                  * ordered extents get created before we return
1286                  */
1287                 atomic_inc(&root->fs_info->async_submit_draining);
1288                 while (atomic_read(&root->fs_info->nr_async_submits) ||
1289                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1290                         wait_event(root->fs_info->async_submit_wait,
1291                            (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
1292                             atomic_read(&root->fs_info->async_delalloc_pages) == 0));
1293                 }
1294                 atomic_dec(&root->fs_info->async_submit_draining);
1295
1296                 mutex_lock(&inode->i_mutex);
1297                 BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
1298                 mutex_unlock(&inode->i_mutex);
1299         }
1300
1301         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1302                 btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO);
1303         }
1304
1305         ret = defrag_count;
1306
1307 out_ra:
1308         if (!file)
1309                 kfree(ra);
1310         kfree(pages);
1311         return ret;
1312 }
1313
1314 static noinline int btrfs_ioctl_resize(struct file *file,
1315                                         void __user *arg)
1316 {
1317         u64 new_size;
1318         u64 old_size;
1319         u64 devid = 1;
1320         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1321         struct btrfs_ioctl_vol_args *vol_args;
1322         struct btrfs_trans_handle *trans;
1323         struct btrfs_device *device = NULL;
1324         char *sizestr;
1325         char *devstr = NULL;
1326         int ret = 0;
1327         int mod = 0;
1328
1329         if (root->fs_info->sb->s_flags & MS_RDONLY)
1330                 return -EROFS;
1331
1332         if (!capable(CAP_SYS_ADMIN))
1333                 return -EPERM;
1334
1335         ret = mnt_want_write_file(file);
1336         if (ret)
1337                 return ret;
1338
1339         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
1340                         1)) {
1341                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
1342                 mnt_drop_write_file(file);
1343                 return -EINVAL;
1344         }
1345
1346         mutex_lock(&root->fs_info->volume_mutex);
1347         vol_args = memdup_user(arg, sizeof(*vol_args));
1348         if (IS_ERR(vol_args)) {
1349                 ret = PTR_ERR(vol_args);
1350                 goto out;
1351         }
1352
1353         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1354
1355         sizestr = vol_args->name;
1356         devstr = strchr(sizestr, ':');
1357         if (devstr) {
1358                 char *end;
1359                 sizestr = devstr + 1;
1360                 *devstr = '\0';
1361                 devstr = vol_args->name;
1362                 devid = simple_strtoull(devstr, &end, 10);
1363                 printk(KERN_INFO "btrfs: resizing devid %llu\n",
1364                        (unsigned long long)devid);
1365         }
1366
1367         device = btrfs_find_device(root->fs_info, devid, NULL, NULL);
1368         if (!device) {
1369                 printk(KERN_INFO "btrfs: resizer unable to find device %llu\n",
1370                        (unsigned long long)devid);
1371                 ret = -EINVAL;
1372                 goto out_free;
1373         }
1374
1375         if (!device->writeable) {
1376                 printk(KERN_INFO "btrfs: resizer unable to apply on "
1377                        "readonly device %llu\n",
1378                        (unsigned long long)devid);
1379                 ret = -EINVAL;
1380                 goto out_free;
1381         }
1382
1383         if (!strcmp(sizestr, "max"))
1384                 new_size = device->bdev->bd_inode->i_size;
1385         else {
1386                 if (sizestr[0] == '-') {
1387                         mod = -1;
1388                         sizestr++;
1389                 } else if (sizestr[0] == '+') {
1390                         mod = 1;
1391                         sizestr++;
1392                 }
1393                 new_size = memparse(sizestr, NULL);
1394                 if (new_size == 0) {
1395                         ret = -EINVAL;
1396                         goto out_free;
1397                 }
1398         }
1399
1400         if (device->is_tgtdev_for_dev_replace) {
1401                 ret = -EINVAL;
1402                 goto out_free;
1403         }
1404
1405         old_size = device->total_bytes;
1406
1407         if (mod < 0) {
1408                 if (new_size > old_size) {
1409                         ret = -EINVAL;
1410                         goto out_free;
1411                 }
1412                 new_size = old_size - new_size;
1413         } else if (mod > 0) {
1414                 new_size = old_size + new_size;
1415         }
1416
1417         if (new_size < 256 * 1024 * 1024) {
1418                 ret = -EINVAL;
1419                 goto out_free;
1420         }
1421         if (new_size > device->bdev->bd_inode->i_size) {
1422                 ret = -EFBIG;
1423                 goto out_free;
1424         }
1425
1426         do_div(new_size, root->sectorsize);
1427         new_size *= root->sectorsize;
1428
1429         printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n",
1430                       rcu_str_deref(device->name),
1431                       (unsigned long long)new_size);
1432
1433         if (new_size > old_size) {
1434                 trans = btrfs_start_transaction(root, 0);
1435                 if (IS_ERR(trans)) {
1436                         ret = PTR_ERR(trans);
1437                         goto out_free;
1438                 }
1439                 ret = btrfs_grow_device(trans, device, new_size);
1440                 btrfs_commit_transaction(trans, root);
1441         } else if (new_size < old_size) {
1442                 ret = btrfs_shrink_device(device, new_size);
1443         } /* equal, nothing need to do */
1444
1445 out_free:
1446         kfree(vol_args);
1447 out:
1448         mutex_unlock(&root->fs_info->volume_mutex);
1449         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
1450         mnt_drop_write_file(file);
1451         return ret;
1452 }
1453
1454 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1455                                 char *name, unsigned long fd, int subvol,
1456                                 u64 *transid, bool readonly,
1457                                 struct btrfs_qgroup_inherit **inherit)
1458 {
1459         int namelen;
1460         int ret = 0;
1461
1462         ret = mnt_want_write_file(file);
1463         if (ret)
1464                 goto out;
1465
1466         namelen = strlen(name);
1467         if (strchr(name, '/')) {
1468                 ret = -EINVAL;
1469                 goto out_drop_write;
1470         }
1471
1472         if (name[0] == '.' &&
1473            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1474                 ret = -EEXIST;
1475                 goto out_drop_write;
1476         }
1477
1478         if (subvol) {
1479                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1480                                      NULL, transid, readonly, inherit);
1481         } else {
1482                 struct fd src = fdget(fd);
1483                 struct inode *src_inode;
1484                 if (!src.file) {
1485                         ret = -EINVAL;
1486                         goto out_drop_write;
1487                 }
1488
1489                 src_inode = src.file->f_path.dentry->d_inode;
1490                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1491                         printk(KERN_INFO "btrfs: Snapshot src from "
1492                                "another FS\n");
1493                         ret = -EINVAL;
1494                 } else {
1495                         ret = btrfs_mksubvol(&file->f_path, name, namelen,
1496                                              BTRFS_I(src_inode)->root,
1497                                              transid, readonly, inherit);
1498                 }
1499                 fdput(src);
1500         }
1501 out_drop_write:
1502         mnt_drop_write_file(file);
1503 out:
1504         return ret;
1505 }
1506
1507 static noinline int btrfs_ioctl_snap_create(struct file *file,
1508                                             void __user *arg, int subvol)
1509 {
1510         struct btrfs_ioctl_vol_args *vol_args;
1511         int ret;
1512
1513         vol_args = memdup_user(arg, sizeof(*vol_args));
1514         if (IS_ERR(vol_args))
1515                 return PTR_ERR(vol_args);
1516         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1517
1518         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1519                                               vol_args->fd, subvol,
1520                                               NULL, false, NULL);
1521
1522         kfree(vol_args);
1523         return ret;
1524 }
1525
1526 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1527                                                void __user *arg, int subvol)
1528 {
1529         struct btrfs_ioctl_vol_args_v2 *vol_args;
1530         int ret;
1531         u64 transid = 0;
1532         u64 *ptr = NULL;
1533         bool readonly = false;
1534         struct btrfs_qgroup_inherit *inherit = NULL;
1535
1536         vol_args = memdup_user(arg, sizeof(*vol_args));
1537         if (IS_ERR(vol_args))
1538                 return PTR_ERR(vol_args);
1539         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1540
1541         if (vol_args->flags &
1542             ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY |
1543               BTRFS_SUBVOL_QGROUP_INHERIT)) {
1544                 ret = -EOPNOTSUPP;
1545                 goto out;
1546         }
1547
1548         if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1549                 ptr = &transid;
1550         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1551                 readonly = true;
1552         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1553                 if (vol_args->size > PAGE_CACHE_SIZE) {
1554                         ret = -EINVAL;
1555                         goto out;
1556                 }
1557                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1558                 if (IS_ERR(inherit)) {
1559                         ret = PTR_ERR(inherit);
1560                         goto out;
1561                 }
1562         }
1563
1564         ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1565                                               vol_args->fd, subvol, ptr,
1566                                               readonly, &inherit);
1567
1568         if (ret == 0 && ptr &&
1569             copy_to_user(arg +
1570                          offsetof(struct btrfs_ioctl_vol_args_v2,
1571                                   transid), ptr, sizeof(*ptr)))
1572                 ret = -EFAULT;
1573 out:
1574         kfree(vol_args);
1575         kfree(inherit);
1576         return ret;
1577 }
1578
1579 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1580                                                 void __user *arg)
1581 {
1582         struct inode *inode = fdentry(file)->d_inode;
1583         struct btrfs_root *root = BTRFS_I(inode)->root;
1584         int ret = 0;
1585         u64 flags = 0;
1586
1587         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID)
1588                 return -EINVAL;
1589
1590         down_read(&root->fs_info->subvol_sem);
1591         if (btrfs_root_readonly(root))
1592                 flags |= BTRFS_SUBVOL_RDONLY;
1593         up_read(&root->fs_info->subvol_sem);
1594
1595         if (copy_to_user(arg, &flags, sizeof(flags)))
1596                 ret = -EFAULT;
1597
1598         return ret;
1599 }
1600
1601 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1602                                               void __user *arg)
1603 {
1604         struct inode *inode = fdentry(file)->d_inode;
1605         struct btrfs_root *root = BTRFS_I(inode)->root;
1606         struct btrfs_trans_handle *trans;
1607         u64 root_flags;
1608         u64 flags;
1609         int ret = 0;
1610
1611         ret = mnt_want_write_file(file);
1612         if (ret)
1613                 goto out;
1614
1615         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
1616                 ret = -EINVAL;
1617                 goto out_drop_write;
1618         }
1619
1620         if (copy_from_user(&flags, arg, sizeof(flags))) {
1621                 ret = -EFAULT;
1622                 goto out_drop_write;
1623         }
1624
1625         if (flags & BTRFS_SUBVOL_CREATE_ASYNC) {
1626                 ret = -EINVAL;
1627                 goto out_drop_write;
1628         }
1629
1630         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1631                 ret = -EOPNOTSUPP;
1632                 goto out_drop_write;
1633         }
1634
1635         if (!inode_owner_or_capable(inode)) {
1636                 ret = -EACCES;
1637                 goto out_drop_write;
1638         }
1639
1640         down_write(&root->fs_info->subvol_sem);
1641
1642         /* nothing to do */
1643         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1644                 goto out_drop_sem;
1645
1646         root_flags = btrfs_root_flags(&root->root_item);
1647         if (flags & BTRFS_SUBVOL_RDONLY)
1648                 btrfs_set_root_flags(&root->root_item,
1649                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1650         else
1651                 btrfs_set_root_flags(&root->root_item,
1652                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1653
1654         trans = btrfs_start_transaction(root, 1);
1655         if (IS_ERR(trans)) {
1656                 ret = PTR_ERR(trans);
1657                 goto out_reset;
1658         }
1659
1660         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1661                                 &root->root_key, &root->root_item);
1662
1663         btrfs_commit_transaction(trans, root);
1664 out_reset:
1665         if (ret)
1666                 btrfs_set_root_flags(&root->root_item, root_flags);
1667 out_drop_sem:
1668         up_write(&root->fs_info->subvol_sem);
1669 out_drop_write:
1670         mnt_drop_write_file(file);
1671 out:
1672         return ret;
1673 }
1674
1675 /*
1676  * helper to check if the subvolume references other subvolumes
1677  */
1678 static noinline int may_destroy_subvol(struct btrfs_root *root)
1679 {
1680         struct btrfs_path *path;
1681         struct btrfs_key key;
1682         int ret;
1683
1684         path = btrfs_alloc_path();
1685         if (!path)
1686                 return -ENOMEM;
1687
1688         key.objectid = root->root_key.objectid;
1689         key.type = BTRFS_ROOT_REF_KEY;
1690         key.offset = (u64)-1;
1691
1692         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1693                                 &key, path, 0, 0);
1694         if (ret < 0)
1695                 goto out;
1696         BUG_ON(ret == 0);
1697
1698         ret = 0;
1699         if (path->slots[0] > 0) {
1700                 path->slots[0]--;
1701                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1702                 if (key.objectid == root->root_key.objectid &&
1703                     key.type == BTRFS_ROOT_REF_KEY)
1704                         ret = -ENOTEMPTY;
1705         }
1706 out:
1707         btrfs_free_path(path);
1708         return ret;
1709 }
1710
1711 static noinline int key_in_sk(struct btrfs_key *key,
1712                               struct btrfs_ioctl_search_key *sk)
1713 {
1714         struct btrfs_key test;
1715         int ret;
1716
1717         test.objectid = sk->min_objectid;
1718         test.type = sk->min_type;
1719         test.offset = sk->min_offset;
1720
1721         ret = btrfs_comp_cpu_keys(key, &test);
1722         if (ret < 0)
1723                 return 0;
1724
1725         test.objectid = sk->max_objectid;
1726         test.type = sk->max_type;
1727         test.offset = sk->max_offset;
1728
1729         ret = btrfs_comp_cpu_keys(key, &test);
1730         if (ret > 0)
1731                 return 0;
1732         return 1;
1733 }
1734
1735 static noinline int copy_to_sk(struct btrfs_root *root,
1736                                struct btrfs_path *path,
1737                                struct btrfs_key *key,
1738                                struct btrfs_ioctl_search_key *sk,
1739                                char *buf,
1740                                unsigned long *sk_offset,
1741                                int *num_found)
1742 {
1743         u64 found_transid;
1744         struct extent_buffer *leaf;
1745         struct btrfs_ioctl_search_header sh;
1746         unsigned long item_off;
1747         unsigned long item_len;
1748         int nritems;
1749         int i;
1750         int slot;
1751         int ret = 0;
1752
1753         leaf = path->nodes[0];
1754         slot = path->slots[0];
1755         nritems = btrfs_header_nritems(leaf);
1756
1757         if (btrfs_header_generation(leaf) > sk->max_transid) {
1758                 i = nritems;
1759                 goto advance_key;
1760         }
1761         found_transid = btrfs_header_generation(leaf);
1762
1763         for (i = slot; i < nritems; i++) {
1764                 item_off = btrfs_item_ptr_offset(leaf, i);
1765                 item_len = btrfs_item_size_nr(leaf, i);
1766
1767                 if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1768                         item_len = 0;
1769
1770                 if (sizeof(sh) + item_len + *sk_offset >
1771                     BTRFS_SEARCH_ARGS_BUFSIZE) {
1772                         ret = 1;
1773                         goto overflow;
1774                 }
1775
1776                 btrfs_item_key_to_cpu(leaf, key, i);
1777                 if (!key_in_sk(key, sk))
1778                         continue;
1779
1780                 sh.objectid = key->objectid;
1781                 sh.offset = key->offset;
1782                 sh.type = key->type;
1783                 sh.len = item_len;
1784                 sh.transid = found_transid;
1785
1786                 /* copy search result header */
1787                 memcpy(buf + *sk_offset, &sh, sizeof(sh));
1788                 *sk_offset += sizeof(sh);
1789
1790                 if (item_len) {
1791                         char *p = buf + *sk_offset;
1792                         /* copy the item */
1793                         read_extent_buffer(leaf, p,
1794                                            item_off, item_len);
1795                         *sk_offset += item_len;
1796                 }
1797                 (*num_found)++;
1798
1799                 if (*num_found >= sk->nr_items)
1800                         break;
1801         }
1802 advance_key:
1803         ret = 0;
1804         if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1805                 key->offset++;
1806         else if (key->type < (u8)-1 && key->type < sk->max_type) {
1807                 key->offset = 0;
1808                 key->type++;
1809         } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1810                 key->offset = 0;
1811                 key->type = 0;
1812                 key->objectid++;
1813         } else
1814                 ret = 1;
1815 overflow:
1816         return ret;
1817 }
1818
1819 static noinline int search_ioctl(struct inode *inode,
1820                                  struct btrfs_ioctl_search_args *args)
1821 {
1822         struct btrfs_root *root;
1823         struct btrfs_key key;
1824         struct btrfs_key max_key;
1825         struct btrfs_path *path;
1826         struct btrfs_ioctl_search_key *sk = &args->key;
1827         struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1828         int ret;
1829         int num_found = 0;
1830         unsigned long sk_offset = 0;
1831
1832         path = btrfs_alloc_path();
1833         if (!path)
1834                 return -ENOMEM;
1835
1836         if (sk->tree_id == 0) {
1837                 /* search the root of the inode that was passed */
1838                 root = BTRFS_I(inode)->root;
1839         } else {
1840                 key.objectid = sk->tree_id;
1841                 key.type = BTRFS_ROOT_ITEM_KEY;
1842                 key.offset = (u64)-1;
1843                 root = btrfs_read_fs_root_no_name(info, &key);
1844                 if (IS_ERR(root)) {
1845                         printk(KERN_ERR "could not find root %llu\n",
1846                                sk->tree_id);
1847                         btrfs_free_path(path);
1848                         return -ENOENT;
1849                 }
1850         }
1851
1852         key.objectid = sk->min_objectid;
1853         key.type = sk->min_type;
1854         key.offset = sk->min_offset;
1855
1856         max_key.objectid = sk->max_objectid;
1857         max_key.type = sk->max_type;
1858         max_key.offset = sk->max_offset;
1859
1860         path->keep_locks = 1;
1861
1862         while(1) {
1863                 ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1864                                            sk->min_transid);
1865                 if (ret != 0) {
1866                         if (ret > 0)
1867                                 ret = 0;
1868                         goto err;
1869                 }
1870                 ret = copy_to_sk(root, path, &key, sk, args->buf,
1871                                  &sk_offset, &num_found);
1872                 btrfs_release_path(path);
1873                 if (ret || num_found >= sk->nr_items)
1874                         break;
1875
1876         }
1877         ret = 0;
1878 err:
1879         sk->nr_items = num_found;
1880         btrfs_free_path(path);
1881         return ret;
1882 }
1883
1884 static noinline int btrfs_ioctl_tree_search(struct file *file,
1885                                            void __user *argp)
1886 {
1887          struct btrfs_ioctl_search_args *args;
1888          struct inode *inode;
1889          int ret;
1890
1891         if (!capable(CAP_SYS_ADMIN))
1892                 return -EPERM;
1893
1894         args = memdup_user(argp, sizeof(*args));
1895         if (IS_ERR(args))
1896                 return PTR_ERR(args);
1897
1898         inode = fdentry(file)->d_inode;
1899         ret = search_ioctl(inode, args);
1900         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1901                 ret = -EFAULT;
1902         kfree(args);
1903         return ret;
1904 }
1905
1906 /*
1907  * Search INODE_REFs to identify path name of 'dirid' directory
1908  * in a 'tree_id' tree. and sets path name to 'name'.
1909  */
1910 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1911                                 u64 tree_id, u64 dirid, char *name)
1912 {
1913         struct btrfs_root *root;
1914         struct btrfs_key key;
1915         char *ptr;
1916         int ret = -1;
1917         int slot;
1918         int len;
1919         int total_len = 0;
1920         struct btrfs_inode_ref *iref;
1921         struct extent_buffer *l;
1922         struct btrfs_path *path;
1923
1924         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1925                 name[0]='\0';
1926                 return 0;
1927         }
1928
1929         path = btrfs_alloc_path();
1930         if (!path)
1931                 return -ENOMEM;
1932
1933         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1934
1935         key.objectid = tree_id;
1936         key.type = BTRFS_ROOT_ITEM_KEY;
1937         key.offset = (u64)-1;
1938         root = btrfs_read_fs_root_no_name(info, &key);
1939         if (IS_ERR(root)) {
1940                 printk(KERN_ERR "could not find root %llu\n", tree_id);
1941                 ret = -ENOENT;
1942                 goto out;
1943         }
1944
1945         key.objectid = dirid;
1946         key.type = BTRFS_INODE_REF_KEY;
1947         key.offset = (u64)-1;
1948
1949         while(1) {
1950                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1951                 if (ret < 0)
1952                         goto out;
1953
1954                 l = path->nodes[0];
1955                 slot = path->slots[0];
1956                 if (ret > 0 && slot > 0)
1957                         slot--;
1958                 btrfs_item_key_to_cpu(l, &key, slot);
1959
1960                 if (ret > 0 && (key.objectid != dirid ||
1961                                 key.type != BTRFS_INODE_REF_KEY)) {
1962                         ret = -ENOENT;
1963                         goto out;
1964                 }
1965
1966                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1967                 len = btrfs_inode_ref_name_len(l, iref);
1968                 ptr -= len + 1;
1969                 total_len += len + 1;
1970                 if (ptr < name)
1971                         goto out;
1972
1973                 *(ptr + len) = '/';
1974                 read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1975
1976                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1977                         break;
1978
1979                 btrfs_release_path(path);
1980                 key.objectid = key.offset;
1981                 key.offset = (u64)-1;
1982                 dirid = key.objectid;
1983         }
1984         if (ptr < name)
1985                 goto out;
1986         memmove(name, ptr, total_len);
1987         name[total_len]='\0';
1988         ret = 0;
1989 out:
1990         btrfs_free_path(path);
1991         return ret;
1992 }
1993
1994 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1995                                            void __user *argp)
1996 {
1997          struct btrfs_ioctl_ino_lookup_args *args;
1998          struct inode *inode;
1999          int ret;
2000
2001         if (!capable(CAP_SYS_ADMIN))
2002                 return -EPERM;
2003
2004         args = memdup_user(argp, sizeof(*args));
2005         if (IS_ERR(args))
2006                 return PTR_ERR(args);
2007
2008         inode = fdentry(file)->d_inode;
2009
2010         if (args->treeid == 0)
2011                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2012
2013         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2014                                         args->treeid, args->objectid,
2015                                         args->name);
2016
2017         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2018                 ret = -EFAULT;
2019
2020         kfree(args);
2021         return ret;
2022 }
2023
2024 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2025                                              void __user *arg)
2026 {
2027         struct dentry *parent = fdentry(file);
2028         struct dentry *dentry;
2029         struct inode *dir = parent->d_inode;
2030         struct inode *inode;
2031         struct btrfs_root *root = BTRFS_I(dir)->root;
2032         struct btrfs_root *dest = NULL;
2033         struct btrfs_ioctl_vol_args *vol_args;
2034         struct btrfs_trans_handle *trans;
2035         int namelen;
2036         int ret;
2037         int err = 0;
2038
2039         vol_args = memdup_user(arg, sizeof(*vol_args));
2040         if (IS_ERR(vol_args))
2041                 return PTR_ERR(vol_args);
2042
2043         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2044         namelen = strlen(vol_args->name);
2045         if (strchr(vol_args->name, '/') ||
2046             strncmp(vol_args->name, "..", namelen) == 0) {
2047                 err = -EINVAL;
2048                 goto out;
2049         }
2050
2051         err = mnt_want_write_file(file);
2052         if (err)
2053                 goto out;
2054
2055         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
2056         dentry = lookup_one_len(vol_args->name, parent, namelen);
2057         if (IS_ERR(dentry)) {
2058                 err = PTR_ERR(dentry);
2059                 goto out_unlock_dir;
2060         }
2061
2062         if (!dentry->d_inode) {
2063                 err = -ENOENT;
2064                 goto out_dput;
2065         }
2066
2067         inode = dentry->d_inode;
2068         dest = BTRFS_I(inode)->root;
2069         if (!capable(CAP_SYS_ADMIN)){
2070                 /*
2071                  * Regular user.  Only allow this with a special mount
2072                  * option, when the user has write+exec access to the
2073                  * subvol root, and when rmdir(2) would have been
2074                  * allowed.
2075                  *
2076                  * Note that this is _not_ check that the subvol is
2077                  * empty or doesn't contain data that we wouldn't
2078                  * otherwise be able to delete.
2079                  *
2080                  * Users who want to delete empty subvols should try
2081                  * rmdir(2).
2082                  */
2083                 err = -EPERM;
2084                 if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
2085                         goto out_dput;
2086
2087                 /*
2088                  * Do not allow deletion if the parent dir is the same
2089                  * as the dir to be deleted.  That means the ioctl
2090                  * must be called on the dentry referencing the root
2091                  * of the subvol, not a random directory contained
2092                  * within it.
2093                  */
2094                 err = -EINVAL;
2095                 if (root == dest)
2096                         goto out_dput;
2097
2098                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2099                 if (err)
2100                         goto out_dput;
2101         }
2102
2103         /* check if subvolume may be deleted by a user */
2104         err = btrfs_may_delete(dir, dentry, 1);
2105         if (err)
2106                 goto out_dput;
2107
2108         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
2109                 err = -EINVAL;
2110                 goto out_dput;
2111         }
2112
2113         mutex_lock(&inode->i_mutex);
2114         err = d_invalidate(dentry);
2115         if (err)
2116                 goto out_unlock;
2117
2118         down_write(&root->fs_info->subvol_sem);
2119
2120         err = may_destroy_subvol(dest);
2121         if (err)
2122                 goto out_up_write;
2123
2124         trans = btrfs_start_transaction(root, 0);
2125         if (IS_ERR(trans)) {
2126                 err = PTR_ERR(trans);
2127                 goto out_up_write;
2128         }
2129         trans->block_rsv = &root->fs_info->global_block_rsv;
2130
2131         ret = btrfs_unlink_subvol(trans, root, dir,
2132                                 dest->root_key.objectid,
2133                                 dentry->d_name.name,
2134                                 dentry->d_name.len);
2135         if (ret) {
2136                 err = ret;
2137                 btrfs_abort_transaction(trans, root, ret);
2138                 goto out_end_trans;
2139         }
2140
2141         btrfs_record_root_in_trans(trans, dest);
2142
2143         memset(&dest->root_item.drop_progress, 0,
2144                 sizeof(dest->root_item.drop_progress));
2145         dest->root_item.drop_level = 0;
2146         btrfs_set_root_refs(&dest->root_item, 0);
2147
2148         if (!xchg(&dest->orphan_item_inserted, 1)) {
2149                 ret = btrfs_insert_orphan_item(trans,
2150                                         root->fs_info->tree_root,
2151                                         dest->root_key.objectid);
2152                 if (ret) {
2153                         btrfs_abort_transaction(trans, root, ret);
2154                         err = ret;
2155                         goto out_end_trans;
2156                 }
2157         }
2158 out_end_trans:
2159         ret = btrfs_end_transaction(trans, root);
2160         if (ret && !err)
2161                 err = ret;
2162         inode->i_flags |= S_DEAD;
2163 out_up_write:
2164         up_write(&root->fs_info->subvol_sem);
2165 out_unlock:
2166         mutex_unlock(&inode->i_mutex);
2167         if (!err) {
2168                 shrink_dcache_sb(root->fs_info->sb);
2169                 btrfs_invalidate_inodes(dest);
2170                 d_delete(dentry);
2171         }
2172 out_dput:
2173         dput(dentry);
2174 out_unlock_dir:
2175         mutex_unlock(&dir->i_mutex);
2176         mnt_drop_write_file(file);
2177 out:
2178         kfree(vol_args);
2179         return err;
2180 }
2181
2182 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
2183 {
2184         struct inode *inode = fdentry(file)->d_inode;
2185         struct btrfs_root *root = BTRFS_I(inode)->root;
2186         struct btrfs_ioctl_defrag_range_args *range;
2187         int ret;
2188
2189         ret = mnt_want_write_file(file);
2190         if (ret)
2191                 return ret;
2192
2193         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2194                         1)) {
2195                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2196                 mnt_drop_write_file(file);
2197                 return -EINVAL;
2198         }
2199
2200         if (btrfs_root_readonly(root)) {
2201                 ret = -EROFS;
2202                 goto out;
2203         }
2204
2205         switch (inode->i_mode & S_IFMT) {
2206         case S_IFDIR:
2207                 if (!capable(CAP_SYS_ADMIN)) {
2208                         ret = -EPERM;
2209                         goto out;
2210                 }
2211                 ret = btrfs_defrag_root(root, 0);
2212                 if (ret)
2213                         goto out;
2214                 ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
2215                 break;
2216         case S_IFREG:
2217                 if (!(file->f_mode & FMODE_WRITE)) {
2218                         ret = -EINVAL;
2219                         goto out;
2220                 }
2221
2222                 range = kzalloc(sizeof(*range), GFP_KERNEL);
2223                 if (!range) {
2224                         ret = -ENOMEM;
2225                         goto out;
2226                 }
2227
2228                 if (argp) {
2229                         if (copy_from_user(range, argp,
2230                                            sizeof(*range))) {
2231                                 ret = -EFAULT;
2232                                 kfree(range);
2233                                 goto out;
2234                         }
2235                         /* compression requires us to start the IO */
2236                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
2237                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
2238                                 range->extent_thresh = (u32)-1;
2239                         }
2240                 } else {
2241                         /* the rest are all set to zero by kzalloc */
2242                         range->len = (u64)-1;
2243                 }
2244                 ret = btrfs_defrag_file(fdentry(file)->d_inode, file,
2245                                         range, 0, 0);
2246                 if (ret > 0)
2247                         ret = 0;
2248                 kfree(range);
2249                 break;
2250         default:
2251                 ret = -EINVAL;
2252         }
2253 out:
2254         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2255         mnt_drop_write_file(file);
2256         return ret;
2257 }
2258
2259 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
2260 {
2261         struct btrfs_ioctl_vol_args *vol_args;
2262         int ret;
2263
2264         if (!capable(CAP_SYS_ADMIN))
2265                 return -EPERM;
2266
2267         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2268                         1)) {
2269                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2270                 return -EINVAL;
2271         }
2272
2273         mutex_lock(&root->fs_info->volume_mutex);
2274         vol_args = memdup_user(arg, sizeof(*vol_args));
2275         if (IS_ERR(vol_args)) {
2276                 ret = PTR_ERR(vol_args);
2277                 goto out;
2278         }
2279
2280         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2281         ret = btrfs_init_new_device(root, vol_args->name);
2282
2283         kfree(vol_args);
2284 out:
2285         mutex_unlock(&root->fs_info->volume_mutex);
2286         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2287         return ret;
2288 }
2289
2290 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
2291 {
2292         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2293         struct btrfs_ioctl_vol_args *vol_args;
2294         int ret;
2295
2296         if (!capable(CAP_SYS_ADMIN))
2297                 return -EPERM;
2298
2299         ret = mnt_want_write_file(file);
2300         if (ret)
2301                 return ret;
2302
2303         if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running,
2304                         1)) {
2305                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
2306                 mnt_drop_write_file(file);
2307                 return -EINVAL;
2308         }
2309
2310         mutex_lock(&root->fs_info->volume_mutex);
2311         vol_args = memdup_user(arg, sizeof(*vol_args));
2312         if (IS_ERR(vol_args)) {
2313                 ret = PTR_ERR(vol_args);
2314                 goto out;
2315         }
2316
2317         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
2318         ret = btrfs_rm_device(root, vol_args->name);
2319
2320         kfree(vol_args);
2321 out:
2322         mutex_unlock(&root->fs_info->volume_mutex);
2323         atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0);
2324         mnt_drop_write_file(file);
2325         return ret;
2326 }
2327
2328 static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg)
2329 {
2330         struct btrfs_ioctl_fs_info_args *fi_args;
2331         struct btrfs_device *device;
2332         struct btrfs_device *next;
2333         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2334         int ret = 0;
2335
2336         if (!capable(CAP_SYS_ADMIN))
2337                 return -EPERM;
2338
2339         fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL);
2340         if (!fi_args)
2341                 return -ENOMEM;
2342
2343         fi_args->num_devices = fs_devices->num_devices;
2344         memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid));
2345
2346         mutex_lock(&fs_devices->device_list_mutex);
2347         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2348                 if (device->devid > fi_args->max_id)
2349                         fi_args->max_id = device->devid;
2350         }
2351         mutex_unlock(&fs_devices->device_list_mutex);
2352
2353         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
2354                 ret = -EFAULT;
2355
2356         kfree(fi_args);
2357         return ret;
2358 }
2359
2360 static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg)
2361 {
2362         struct btrfs_ioctl_dev_info_args *di_args;
2363         struct btrfs_device *dev;
2364         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2365         int ret = 0;
2366         char *s_uuid = NULL;
2367         char empty_uuid[BTRFS_UUID_SIZE] = {0};
2368
2369         if (!capable(CAP_SYS_ADMIN))
2370                 return -EPERM;
2371
2372         di_args = memdup_user(arg, sizeof(*di_args));
2373         if (IS_ERR(di_args))
2374                 return PTR_ERR(di_args);
2375
2376         if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0)
2377                 s_uuid = di_args->uuid;
2378
2379         mutex_lock(&fs_devices->device_list_mutex);
2380         dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL);
2381         mutex_unlock(&fs_devices->device_list_mutex);
2382
2383         if (!dev) {
2384                 ret = -ENODEV;
2385                 goto out;
2386         }
2387
2388         di_args->devid = dev->devid;
2389         di_args->bytes_used = dev->bytes_used;
2390         di_args->total_bytes = dev->total_bytes;
2391         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
2392         if (dev->name) {
2393                 struct rcu_string *name;
2394
2395                 rcu_read_lock();
2396                 name = rcu_dereference(dev->name);
2397                 strncpy(di_args->path, name->str, sizeof(di_args->path));
2398                 rcu_read_unlock();
2399                 di_args->path[sizeof(di_args->path) - 1] = 0;
2400         } else {
2401                 di_args->path[0] = '\0';
2402         }
2403
2404 out:
2405         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
2406                 ret = -EFAULT;
2407
2408         kfree(di_args);
2409         return ret;
2410 }
2411
2412 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2413                                        u64 off, u64 olen, u64 destoff)
2414 {
2415         struct inode *inode = fdentry(file)->d_inode;
2416         struct btrfs_root *root = BTRFS_I(inode)->root;
2417         struct fd src_file;
2418         struct inode *src;
2419         struct btrfs_trans_handle *trans;
2420         struct btrfs_path *path;
2421         struct extent_buffer *leaf;
2422         char *buf;
2423         struct btrfs_key key;
2424         u32 nritems;
2425         int slot;
2426         int ret;
2427         u64 len = olen;
2428         u64 bs = root->fs_info->sb->s_blocksize;
2429
2430         /*
2431          * TODO:
2432          * - split compressed inline extents.  annoying: we need to
2433          *   decompress into destination's address_space (the file offset
2434          *   may change, so source mapping won't do), then recompress (or
2435          *   otherwise reinsert) a subrange.
2436          * - allow ranges within the same file to be cloned (provided
2437          *   they don't overlap)?
2438          */
2439
2440         /* the destination must be opened for writing */
2441         if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
2442                 return -EINVAL;
2443
2444         if (btrfs_root_readonly(root))
2445                 return -EROFS;
2446
2447         ret = mnt_want_write_file(file);
2448         if (ret)
2449                 return ret;
2450
2451         src_file = fdget(srcfd);
2452         if (!src_file.file) {
2453                 ret = -EBADF;
2454                 goto out_drop_write;
2455         }
2456
2457         ret = -EXDEV;
2458         if (src_file.file->f_path.mnt != file->f_path.mnt)
2459                 goto out_fput;
2460
2461         src = src_file.file->f_dentry->d_inode;
2462
2463         ret = -EINVAL;
2464         if (src == inode)
2465                 goto out_fput;
2466
2467         /* the src must be open for reading */
2468         if (!(src_file.file->f_mode & FMODE_READ))
2469                 goto out_fput;
2470
2471         /* don't make the dst file partly checksummed */
2472         if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
2473             (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
2474                 goto out_fput;
2475
2476         ret = -EISDIR;
2477         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
2478                 goto out_fput;
2479
2480         ret = -EXDEV;
2481         if (src->i_sb != inode->i_sb)
2482                 goto out_fput;
2483
2484         ret = -ENOMEM;
2485         buf = vmalloc(btrfs_level_size(root, 0));
2486         if (!buf)
2487                 goto out_fput;
2488
2489         path = btrfs_alloc_path();
2490         if (!path) {
2491                 vfree(buf);
2492                 goto out_fput;
2493         }
2494         path->reada = 2;
2495
2496         if (inode < src) {
2497                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
2498                 mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
2499         } else {
2500                 mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
2501                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2502         }
2503
2504         /* determine range to clone */
2505         ret = -EINVAL;
2506         if (off + len > src->i_size || off + len < off)
2507                 goto out_unlock;
2508         if (len == 0)
2509                 olen = len = src->i_size - off;
2510         /* if we extend to eof, continue to block boundary */
2511         if (off + len == src->i_size)
2512                 len = ALIGN(src->i_size, bs) - off;
2513
2514         /* verify the end result is block aligned */
2515         if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
2516             !IS_ALIGNED(destoff, bs))
2517                 goto out_unlock;
2518
2519         if (destoff > inode->i_size) {
2520                 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
2521                 if (ret)
2522                         goto out_unlock;
2523         }
2524
2525         /* truncate page cache pages from target inode range */
2526         truncate_inode_pages_range(&inode->i_data, destoff,
2527                                    PAGE_CACHE_ALIGN(destoff + len) - 1);
2528
2529         /* do any pending delalloc/csum calc on src, one way or
2530            another, and lock file content */
2531         while (1) {
2532                 struct btrfs_ordered_extent *ordered;
2533                 lock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2534                 ordered = btrfs_lookup_first_ordered_extent(src, off + len - 1);
2535                 if (!ordered &&
2536                     !test_range_bit(&BTRFS_I(src)->io_tree, off, off + len - 1,
2537                                     EXTENT_DELALLOC, 0, NULL))
2538                         break;
2539                 unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2540                 if (ordered)
2541                         btrfs_put_ordered_extent(ordered);
2542                 btrfs_wait_ordered_range(src, off, len);
2543         }
2544
2545         /* clone data */
2546         key.objectid = btrfs_ino(src);
2547         key.type = BTRFS_EXTENT_DATA_KEY;
2548         key.offset = 0;
2549
2550         while (1) {
2551                 /*
2552                  * note the key will change type as we walk through the
2553                  * tree.
2554                  */
2555                 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2556                                 0, 0);
2557                 if (ret < 0)
2558                         goto out;
2559
2560                 nritems = btrfs_header_nritems(path->nodes[0]);
2561                 if (path->slots[0] >= nritems) {
2562                         ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2563                         if (ret < 0)
2564                                 goto out;
2565                         if (ret > 0)
2566                                 break;
2567                         nritems = btrfs_header_nritems(path->nodes[0]);
2568                 }
2569                 leaf = path->nodes[0];
2570                 slot = path->slots[0];
2571
2572                 btrfs_item_key_to_cpu(leaf, &key, slot);
2573                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
2574                     key.objectid != btrfs_ino(src))
2575                         break;
2576
2577                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
2578                         struct btrfs_file_extent_item *extent;
2579                         int type;
2580                         u32 size;
2581                         struct btrfs_key new_key;
2582                         u64 disko = 0, diskl = 0;
2583                         u64 datao = 0, datal = 0;
2584                         u8 comp;
2585                         u64 endoff;
2586
2587                         size = btrfs_item_size_nr(leaf, slot);
2588                         read_extent_buffer(leaf, buf,
2589                                            btrfs_item_ptr_offset(leaf, slot),
2590                                            size);
2591
2592                         extent = btrfs_item_ptr(leaf, slot,
2593                                                 struct btrfs_file_extent_item);
2594                         comp = btrfs_file_extent_compression(leaf, extent);
2595                         type = btrfs_file_extent_type(leaf, extent);
2596                         if (type == BTRFS_FILE_EXTENT_REG ||
2597                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2598                                 disko = btrfs_file_extent_disk_bytenr(leaf,
2599                                                                       extent);
2600                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
2601                                                                  extent);
2602                                 datao = btrfs_file_extent_offset(leaf, extent);
2603                                 datal = btrfs_file_extent_num_bytes(leaf,
2604                                                                     extent);
2605                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2606                                 /* take upper bound, may be compressed */
2607                                 datal = btrfs_file_extent_ram_bytes(leaf,
2608                                                                     extent);
2609                         }
2610                         btrfs_release_path(path);
2611
2612                         if (key.offset + datal <= off ||
2613                             key.offset >= off + len - 1)
2614                                 goto next;
2615
2616                         memcpy(&new_key, &key, sizeof(new_key));
2617                         new_key.objectid = btrfs_ino(inode);
2618                         if (off <= key.offset)
2619                                 new_key.offset = key.offset + destoff - off;
2620                         else
2621                                 new_key.offset = destoff;
2622
2623                         /*
2624                          * 1 - adjusting old extent (we may have to split it)
2625                          * 1 - add new extent
2626                          * 1 - inode update
2627                          */
2628                         trans = btrfs_start_transaction(root, 3);
2629                         if (IS_ERR(trans)) {
2630                                 ret = PTR_ERR(trans);
2631                                 goto out;
2632                         }
2633
2634                         if (type == BTRFS_FILE_EXTENT_REG ||
2635                             type == BTRFS_FILE_EXTENT_PREALLOC) {
2636                                 /*
2637                                  *    a  | --- range to clone ---|  b
2638                                  * | ------------- extent ------------- |
2639                                  */
2640
2641                                 /* substract range b */
2642                                 if (key.offset + datal > off + len)
2643                                         datal = off + len - key.offset;
2644
2645                                 /* substract range a */
2646                                 if (off > key.offset) {
2647                                         datao += off - key.offset;
2648                                         datal -= off - key.offset;
2649                                 }
2650
2651                                 ret = btrfs_drop_extents(trans, root, inode,
2652                                                          new_key.offset,
2653                                                          new_key.offset + datal,
2654                                                          1);
2655                                 if (ret) {
2656                                         btrfs_abort_transaction(trans, root,
2657                                                                 ret);
2658                                         btrfs_end_transaction(trans, root);
2659                                         goto out;
2660                                 }
2661
2662                                 ret = btrfs_insert_empty_item(trans, root, path,
2663                                                               &new_key, size);
2664                                 if (ret) {
2665                                         btrfs_abort_transaction(trans, root,
2666                                                                 ret);
2667                                         btrfs_end_transaction(trans, root);
2668                                         goto out;
2669                                 }
2670
2671                                 leaf = path->nodes[0];
2672                                 slot = path->slots[0];
2673                                 write_extent_buffer(leaf, buf,
2674                                             btrfs_item_ptr_offset(leaf, slot),
2675                                             size);
2676
2677                                 extent = btrfs_item_ptr(leaf, slot,
2678                                                 struct btrfs_file_extent_item);
2679
2680                                 /* disko == 0 means it's a hole */
2681                                 if (!disko)
2682                                         datao = 0;
2683
2684                                 btrfs_set_file_extent_offset(leaf, extent,
2685                                                              datao);
2686                                 btrfs_set_file_extent_num_bytes(leaf, extent,
2687                                                                 datal);
2688                                 if (disko) {
2689                                         inode_add_bytes(inode, datal);
2690                                         ret = btrfs_inc_extent_ref(trans, root,
2691                                                         disko, diskl, 0,
2692                                                         root->root_key.objectid,
2693                                                         btrfs_ino(inode),
2694                                                         new_key.offset - datao,
2695                                                         0);
2696                                         if (ret) {
2697                                                 btrfs_abort_transaction(trans,
2698                                                                         root,
2699                                                                         ret);
2700                                                 btrfs_end_transaction(trans,
2701                                                                       root);
2702                                                 goto out;
2703
2704                                         }
2705                                 }
2706                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
2707                                 u64 skip = 0;
2708                                 u64 trim = 0;
2709                                 if (off > key.offset) {
2710                                         skip = off - key.offset;
2711                                         new_key.offset += skip;
2712                                 }
2713
2714                                 if (key.offset + datal > off + len)
2715                                         trim = key.offset + datal - (off + len);
2716
2717                                 if (comp && (skip || trim)) {
2718                                         ret = -EINVAL;
2719                                         btrfs_end_transaction(trans, root);
2720                                         goto out;
2721                                 }
2722                                 size -= skip + trim;
2723                                 datal -= skip + trim;
2724
2725                                 ret = btrfs_drop_extents(trans, root, inode,
2726                                                          new_key.offset,
2727                                                          new_key.offset + datal,
2728                                                          1);
2729                                 if (ret) {
2730                                         btrfs_abort_transaction(trans, root,
2731                                                                 ret);
2732                                         btrfs_end_transaction(trans, root);
2733                                         goto out;
2734                                 }
2735
2736                                 ret = btrfs_insert_empty_item(trans, root, path,
2737                                                               &new_key, size);
2738                                 if (ret) {
2739                                         btrfs_abort_transaction(trans, root,
2740                                                                 ret);
2741                                         btrfs_end_transaction(trans, root);
2742                                         goto out;
2743                                 }
2744
2745                                 if (skip) {
2746                                         u32 start =
2747                                           btrfs_file_extent_calc_inline_size(0);
2748                                         memmove(buf+start, buf+start+skip,
2749                                                 datal);
2750                                 }
2751
2752                                 leaf = path->nodes[0];
2753                                 slot = path->slots[0];
2754                                 write_extent_buffer(leaf, buf,
2755                                             btrfs_item_ptr_offset(leaf, slot),
2756                                             size);
2757                                 inode_add_bytes(inode, datal);
2758                         }
2759
2760                         btrfs_mark_buffer_dirty(leaf);
2761                         btrfs_release_path(path);
2762
2763                         inode_inc_iversion(inode);
2764                         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2765
2766                         /*
2767                          * we round up to the block size at eof when
2768                          * determining which extents to clone above,
2769                          * but shouldn't round up the file size
2770                          */
2771                         endoff = new_key.offset + datal;
2772                         if (endoff > destoff+olen)
2773                                 endoff = destoff+olen;
2774                         if (endoff > inode->i_size)
2775                                 btrfs_i_size_write(inode, endoff);
2776
2777                         ret = btrfs_update_inode(trans, root, inode);
2778                         if (ret) {
2779                                 btrfs_abort_transaction(trans, root, ret);
2780                                 btrfs_end_transaction(trans, root);
2781                                 goto out;
2782                         }
2783                         ret = btrfs_end_transaction(trans, root);
2784                 }
2785 next:
2786                 btrfs_release_path(path);
2787                 key.offset++;
2788         }
2789         ret = 0;
2790 out:
2791         btrfs_release_path(path);
2792         unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1);
2793 out_unlock:
2794         mutex_unlock(&src->i_mutex);
2795         mutex_unlock(&inode->i_mutex);
2796         vfree(buf);
2797         btrfs_free_path(path);
2798 out_fput:
2799         fdput(src_file);
2800 out_drop_write:
2801         mnt_drop_write_file(file);
2802         return ret;
2803 }
2804
2805 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2806 {
2807         struct btrfs_ioctl_clone_range_args args;
2808
2809         if (copy_from_user(&args, argp, sizeof(args)))
2810                 return -EFAULT;
2811         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2812                                  args.src_length, args.dest_offset);
2813 }
2814
2815 /*
2816  * there are many ways the trans_start and trans_end ioctls can lead
2817  * to deadlocks.  They should only be used by applications that
2818  * basically own the machine, and have a very in depth understanding
2819  * of all the possible deadlocks and enospc problems.
2820  */
2821 static long btrfs_ioctl_trans_start(struct file *file)
2822 {
2823         struct inode *inode = fdentry(file)->d_inode;
2824         struct btrfs_root *root = BTRFS_I(inode)->root;
2825         struct btrfs_trans_handle *trans;
2826         int ret;
2827
2828         ret = -EPERM;
2829         if (!capable(CAP_SYS_ADMIN))
2830                 goto out;
2831
2832         ret = -EINPROGRESS;
2833         if (file->private_data)
2834                 goto out;
2835
2836         ret = -EROFS;
2837         if (btrfs_root_readonly(root))
2838                 goto out;
2839
2840         ret = mnt_want_write_file(file);
2841         if (ret)
2842                 goto out;
2843
2844         atomic_inc(&root->fs_info->open_ioctl_trans);
2845
2846         ret = -ENOMEM;
2847         trans = btrfs_start_ioctl_transaction(root);
2848         if (IS_ERR(trans))
2849                 goto out_drop;
2850
2851         file->private_data = trans;
2852         return 0;
2853
2854 out_drop:
2855         atomic_dec(&root->fs_info->open_ioctl_trans);
2856         mnt_drop_write_file(file);
2857 out:
2858         return ret;
2859 }
2860
2861 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2862 {
2863         struct inode *inode = fdentry(file)->d_inode;
2864         struct btrfs_root *root = BTRFS_I(inode)->root;
2865         struct btrfs_root *new_root;
2866         struct btrfs_dir_item *di;
2867         struct btrfs_trans_handle *trans;
2868         struct btrfs_path *path;
2869         struct btrfs_key location;
2870         struct btrfs_disk_key disk_key;
2871         u64 objectid = 0;
2872         u64 dir_id;
2873         int ret;
2874
2875         if (!capable(CAP_SYS_ADMIN))
2876                 return -EPERM;
2877
2878         ret = mnt_want_write_file(file);
2879         if (ret)
2880                 return ret;
2881
2882         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
2883                 ret = -EFAULT;
2884                 goto out;
2885         }
2886
2887         if (!objectid)
2888                 objectid = root->root_key.objectid;
2889
2890         location.objectid = objectid;
2891         location.type = BTRFS_ROOT_ITEM_KEY;
2892         location.offset = (u64)-1;
2893
2894         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2895         if (IS_ERR(new_root)) {
2896                 ret = PTR_ERR(new_root);
2897                 goto out;
2898         }
2899
2900         if (btrfs_root_refs(&new_root->root_item) == 0) {
2901                 ret = -ENOENT;
2902                 goto out;
2903         }
2904
2905         path = btrfs_alloc_path();
2906         if (!path) {
2907                 ret = -ENOMEM;
2908                 goto out;
2909         }
2910         path->leave_spinning = 1;
2911
2912         trans = btrfs_start_transaction(root, 1);
2913         if (IS_ERR(trans)) {
2914                 btrfs_free_path(path);
2915                 ret = PTR_ERR(trans);
2916                 goto out;
2917         }
2918
2919         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
2920         di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2921                                    dir_id, "default", 7, 1);
2922         if (IS_ERR_OR_NULL(di)) {
2923                 btrfs_free_path(path);
2924                 btrfs_end_transaction(trans, root);
2925                 printk(KERN_ERR "Umm, you don't have the default dir item, "
2926                        "this isn't going to work\n");
2927                 ret = -ENOENT;
2928                 goto out;
2929         }
2930
2931         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2932         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2933         btrfs_mark_buffer_dirty(path->nodes[0]);
2934         btrfs_free_path(path);
2935
2936         btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL);
2937         btrfs_end_transaction(trans, root);
2938 out:
2939         mnt_drop_write_file(file);
2940         return ret;
2941 }
2942
2943 void btrfs_get_block_group_info(struct list_head *groups_list,
2944                                 struct btrfs_ioctl_space_info *space)
2945 {
2946         struct btrfs_block_group_cache *block_group;
2947
2948         space->total_bytes = 0;
2949         space->used_bytes = 0;
2950         space->flags = 0;
2951         list_for_each_entry(block_group, groups_list, list) {
2952                 space->flags = block_group->flags;
2953                 space->total_bytes += block_group->key.offset;
2954                 space->used_bytes +=
2955                         btrfs_block_group_used(&block_group->item);
2956         }
2957 }
2958
2959 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2960 {
2961         struct btrfs_ioctl_space_args space_args;
2962         struct btrfs_ioctl_space_info space;
2963         struct btrfs_ioctl_space_info *dest;
2964         struct btrfs_ioctl_space_info *dest_orig;
2965         struct btrfs_ioctl_space_info __user *user_dest;
2966         struct btrfs_space_info *info;
2967         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2968                        BTRFS_BLOCK_GROUP_SYSTEM,
2969                        BTRFS_BLOCK_GROUP_METADATA,
2970                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2971         int num_types = 4;
2972         int alloc_size;
2973         int ret = 0;
2974         u64 slot_count = 0;
2975         int i, c;
2976
2977         if (copy_from_user(&space_args,
2978                            (struct btrfs_ioctl_space_args __user *)arg,
2979                            sizeof(space_args)))
2980                 return -EFAULT;
2981
2982         for (i = 0; i < num_types; i++) {
2983                 struct btrfs_space_info *tmp;
2984
2985                 info = NULL;
2986                 rcu_read_lock();
2987                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2988                                         list) {
2989                         if (tmp->flags == types[i]) {
2990                                 info = tmp;
2991                                 break;
2992                         }
2993                 }
2994                 rcu_read_unlock();
2995
2996                 if (!info)
2997                         continue;
2998
2999                 down_read(&info->groups_sem);
3000                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3001                         if (!list_empty(&info->block_groups[c]))
3002                                 slot_count++;
3003                 }
3004                 up_read(&info->groups_sem);
3005         }
3006
3007         /* space_slots == 0 means they are asking for a count */
3008         if (space_args.space_slots == 0) {
3009                 space_args.total_spaces = slot_count;
3010                 goto out;
3011         }
3012
3013         slot_count = min_t(u64, space_args.space_slots, slot_count);
3014
3015         alloc_size = sizeof(*dest) * slot_count;
3016
3017         /* we generally have at most 6 or so space infos, one for each raid
3018          * level.  So, a whole page should be more than enough for everyone
3019          */
3020         if (alloc_size > PAGE_CACHE_SIZE)
3021                 return -ENOMEM;
3022
3023         space_args.total_spaces = 0;
3024         dest = kmalloc(alloc_size, GFP_NOFS);
3025         if (!dest)
3026                 return -ENOMEM;
3027         dest_orig = dest;
3028
3029         /* now we have a buffer to copy into */
3030         for (i = 0; i < num_types; i++) {
3031                 struct btrfs_space_info *tmp;
3032
3033                 if (!slot_count)
3034                         break;
3035
3036                 info = NULL;
3037                 rcu_read_lock();
3038                 list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
3039                                         list) {
3040                         if (tmp->flags == types[i]) {
3041                                 info = tmp;
3042                                 break;
3043                         }
3044                 }
3045                 rcu_read_unlock();
3046
3047                 if (!info)
3048                         continue;
3049                 down_read(&info->groups_sem);
3050                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3051                         if (!list_empty(&info->block_groups[c])) {
3052                                 btrfs_get_block_group_info(
3053                                         &info->block_groups[c], &space);
3054                                 memcpy(dest, &space, sizeof(space));
3055                                 dest++;
3056                                 space_args.total_spaces++;
3057                                 slot_count--;
3058                         }
3059                         if (!slot_count)
3060                                 break;
3061                 }
3062                 up_read(&info->groups_sem);
3063         }
3064
3065         user_dest = (struct btrfs_ioctl_space_info __user *)
3066                 (arg + sizeof(struct btrfs_ioctl_space_args));
3067
3068         if (copy_to_user(user_dest, dest_orig, alloc_size))
3069                 ret = -EFAULT;
3070
3071         kfree(dest_orig);
3072 out:
3073         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3074                 ret = -EFAULT;
3075
3076         return ret;
3077 }
3078
3079 /*
3080  * there are many ways the trans_start and trans_end ioctls can lead
3081  * to deadlocks.  They should only be used by applications that
3082  * basically own the machine, and have a very in depth understanding
3083  * of all the possible deadlocks and enospc problems.
3084  */
3085 long btrfs_ioctl_trans_end(struct file *file)
3086 {
3087         struct inode *inode = fdentry(file)->d_inode;
3088         struct btrfs_root *root = BTRFS_I(inode)->root;
3089         struct btrfs_trans_handle *trans;
3090
3091         trans = file->private_data;
3092         if (!trans)
3093                 return -EINVAL;
3094         file->private_data = NULL;
3095
3096         btrfs_end_transaction(trans, root);
3097
3098         atomic_dec(&root->fs_info->open_ioctl_trans);
3099
3100         mnt_drop_write_file(file);
3101         return 0;
3102 }
3103
3104 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3105                                             void __user *argp)
3106 {
3107         struct btrfs_trans_handle *trans;
3108         u64 transid;
3109         int ret;
3110
3111         trans = btrfs_attach_transaction(root);
3112         if (IS_ERR(trans)) {
3113                 if (PTR_ERR(trans) != -ENOENT)
3114                         return PTR_ERR(trans);
3115
3116                 /* No running transaction, don't bother */
3117                 transid = root->fs_info->last_trans_committed;
3118                 goto out;
3119         }
3120         transid = trans->transid;
3121         ret = btrfs_commit_transaction_async(trans, root, 0);
3122         if (ret) {
3123                 btrfs_end_transaction(trans, root);
3124                 return ret;
3125         }
3126 out:
3127         if (argp)
3128                 if (copy_to_user(argp, &transid, sizeof(transid)))
3129                         return -EFAULT;
3130         return 0;
3131 }
3132
3133 static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root,
3134                                            void __user *argp)
3135 {
3136         u64 transid;
3137
3138         if (argp) {
3139                 if (copy_from_user(&transid, argp, sizeof(transid)))
3140                         return -EFAULT;
3141         } else {
3142                 transid = 0;  /* current trans */
3143         }
3144         return btrfs_wait_for_commit(root, transid);
3145 }
3146
3147 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3148 {
3149         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3150         struct btrfs_ioctl_scrub_args *sa;
3151         int ret;
3152
3153         if (!capable(CAP_SYS_ADMIN))
3154                 return -EPERM;
3155
3156         sa = memdup_user(arg, sizeof(*sa));
3157         if (IS_ERR(sa))
3158                 return PTR_ERR(sa);
3159
3160         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3161                 ret = mnt_want_write_file(file);
3162                 if (ret)
3163                         goto out;
3164         }
3165
3166         ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end,
3167                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3168                               0);
3169
3170         if (copy_to_user(arg, sa, sizeof(*sa)))
3171                 ret = -EFAULT;
3172
3173         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3174                 mnt_drop_write_file(file);
3175 out:
3176         kfree(sa);
3177         return ret;
3178 }
3179
3180 static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg)
3181 {
3182         if (!capable(CAP_SYS_ADMIN))
3183                 return -EPERM;
3184
3185         return btrfs_scrub_cancel(root->fs_info);
3186 }
3187
3188 static long btrfs_ioctl_scrub_progress(struct btrfs_root *root,
3189                                        void __user *arg)
3190 {
3191         struct btrfs_ioctl_scrub_args *sa;
3192         int ret;
3193
3194         if (!capable(CAP_SYS_ADMIN))
3195                 return -EPERM;
3196
3197         sa = memdup_user(arg, sizeof(*sa));
3198         if (IS_ERR(sa))
3199                 return PTR_ERR(sa);
3200
3201         ret = btrfs_scrub_progress(root, sa->devid, &sa->progress);
3202
3203         if (copy_to_user(arg, sa, sizeof(*sa)))
3204                 ret = -EFAULT;
3205
3206         kfree(sa);
3207         return ret;
3208 }
3209
3210 static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root,
3211                                       void __user *arg)
3212 {
3213         struct btrfs_ioctl_get_dev_stats *sa;
3214         int ret;
3215
3216         sa = memdup_user(arg, sizeof(*sa));
3217         if (IS_ERR(sa))
3218                 return PTR_ERR(sa);
3219
3220         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3221                 kfree(sa);
3222                 return -EPERM;
3223         }
3224
3225         ret = btrfs_get_dev_stats(root, sa);
3226
3227         if (copy_to_user(arg, sa, sizeof(*sa)))
3228                 ret = -EFAULT;
3229
3230         kfree(sa);
3231         return ret;
3232 }
3233
3234 static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg)
3235 {
3236         struct btrfs_ioctl_dev_replace_args *p;
3237         int ret;
3238
3239         if (!capable(CAP_SYS_ADMIN))
3240                 return -EPERM;
3241
3242         p = memdup_user(arg, sizeof(*p));
3243         if (IS_ERR(p))
3244                 return PTR_ERR(p);
3245
3246         switch (p->cmd) {
3247         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3248                 if (atomic_xchg(
3249                         &root->fs_info->mutually_exclusive_operation_running,
3250                         1)) {
3251                         pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3252                         ret = -EINPROGRESS;
3253                 } else {
3254                         ret = btrfs_dev_replace_start(root, p);
3255                         atomic_set(
3256                          &root->fs_info->mutually_exclusive_operation_running,
3257                          0);
3258                 }
3259                 break;
3260         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3261                 btrfs_dev_replace_status(root->fs_info, p);
3262                 ret = 0;
3263                 break;
3264         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3265                 ret = btrfs_dev_replace_cancel(root->fs_info, p);
3266                 break;
3267         default:
3268                 ret = -EINVAL;
3269                 break;
3270         }
3271
3272         if (copy_to_user(arg, p, sizeof(*p)))
3273                 ret = -EFAULT;
3274
3275         kfree(p);
3276         return ret;
3277 }
3278
3279 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3280 {
3281         int ret = 0;
3282         int i;
3283         u64 rel_ptr;
3284         int size;
3285         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3286         struct inode_fs_paths *ipath = NULL;
3287         struct btrfs_path *path;
3288
3289         if (!capable(CAP_SYS_ADMIN))
3290                 return -EPERM;
3291
3292         path = btrfs_alloc_path();
3293         if (!path) {
3294                 ret = -ENOMEM;
3295                 goto out;
3296         }
3297
3298         ipa = memdup_user(arg, sizeof(*ipa));
3299         if (IS_ERR(ipa)) {
3300                 ret = PTR_ERR(ipa);
3301                 ipa = NULL;
3302                 goto out;
3303         }
3304
3305         size = min_t(u32, ipa->size, 4096);
3306         ipath = init_ipath(size, root, path);
3307         if (IS_ERR(ipath)) {
3308                 ret = PTR_ERR(ipath);
3309                 ipath = NULL;
3310                 goto out;
3311         }
3312
3313         ret = paths_from_inode(ipa->inum, ipath);
3314         if (ret < 0)
3315                 goto out;
3316
3317         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3318                 rel_ptr = ipath->fspath->val[i] -
3319                           (u64)(unsigned long)ipath->fspath->val;
3320                 ipath->fspath->val[i] = rel_ptr;
3321         }
3322
3323         ret = copy_to_user((void *)(unsigned long)ipa->fspath,
3324                            (void *)(unsigned long)ipath->fspath, size);
3325         if (ret) {
3326                 ret = -EFAULT;
3327                 goto out;
3328         }
3329
3330 out:
3331         btrfs_free_path(path);
3332         free_ipath(ipath);
3333         kfree(ipa);
3334
3335         return ret;
3336 }
3337
3338 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3339 {
3340         struct btrfs_data_container *inodes = ctx;
3341         const size_t c = 3 * sizeof(u64);
3342
3343         if (inodes->bytes_left >= c) {
3344                 inodes->bytes_left -= c;
3345                 inodes->val[inodes->elem_cnt] = inum;
3346                 inodes->val[inodes->elem_cnt + 1] = offset;
3347                 inodes->val[inodes->elem_cnt + 2] = root;
3348                 inodes->elem_cnt += 3;
3349         } else {
3350                 inodes->bytes_missing += c - inodes->bytes_left;
3351                 inodes->bytes_left = 0;
3352                 inodes->elem_missed += 3;
3353         }
3354
3355         return 0;
3356 }
3357
3358 static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
3359                                         void __user *arg)
3360 {
3361         int ret = 0;
3362         int size;
3363         struct btrfs_ioctl_logical_ino_args *loi;
3364         struct btrfs_data_container *inodes = NULL;
3365         struct btrfs_path *path = NULL;
3366
3367         if (!capable(CAP_SYS_ADMIN))
3368                 return -EPERM;
3369
3370         loi = memdup_user(arg, sizeof(*loi));
3371         if (IS_ERR(loi)) {
3372                 ret = PTR_ERR(loi);
3373                 loi = NULL;
3374                 goto out;
3375         }
3376
3377         path = btrfs_alloc_path();
3378         if (!path) {
3379                 ret = -ENOMEM;
3380                 goto out;
3381         }
3382
3383         size = min_t(u32, loi->size, 64 * 1024);
3384         inodes = init_data_container(size);
3385         if (IS_ERR(inodes)) {
3386                 ret = PTR_ERR(inodes);
3387                 inodes = NULL;
3388                 goto out;
3389         }
3390
3391         ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path,
3392                                           build_ino_list, inodes);
3393         if (ret == -EINVAL)
3394                 ret = -ENOENT;
3395         if (ret < 0)
3396                 goto out;
3397
3398         ret = copy_to_user((void *)(unsigned long)loi->inodes,
3399                            (void *)(unsigned long)inodes, size);
3400         if (ret)
3401                 ret = -EFAULT;
3402
3403 out:
3404         btrfs_free_path(path);
3405         vfree(inodes);
3406         kfree(loi);
3407
3408         return ret;
3409 }
3410
3411 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3412                                struct btrfs_ioctl_balance_args *bargs)
3413 {
3414         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3415
3416         bargs->flags = bctl->flags;
3417
3418         if (atomic_read(&fs_info->balance_running))
3419                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3420         if (atomic_read(&fs_info->balance_pause_req))
3421                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3422         if (atomic_read(&fs_info->balance_cancel_req))
3423                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3424
3425         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3426         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3427         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3428
3429         if (lock) {
3430                 spin_lock(&fs_info->balance_lock);
3431                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3432                 spin_unlock(&fs_info->balance_lock);
3433         } else {
3434                 memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3435         }
3436 }
3437
3438 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3439 {
3440         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3441         struct btrfs_fs_info *fs_info = root->fs_info;
3442         struct btrfs_ioctl_balance_args *bargs;
3443         struct btrfs_balance_control *bctl;
3444         bool need_unlock; /* for mut. excl. ops lock */
3445         int ret;
3446
3447         if (!capable(CAP_SYS_ADMIN))
3448                 return -EPERM;
3449
3450         ret = mnt_want_write_file(file);
3451         if (ret)
3452                 return ret;
3453
3454 again:
3455         if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) {
3456                 mutex_lock(&fs_info->volume_mutex);
3457                 mutex_lock(&fs_info->balance_mutex);
3458                 need_unlock = true;
3459                 goto locked;
3460         }
3461
3462         /*
3463          * mut. excl. ops lock is locked.  Three possibilites:
3464          *   (1) some other op is running
3465          *   (2) balance is running
3466          *   (3) balance is paused -- special case (think resume)
3467          */
3468         mutex_lock(&fs_info->balance_mutex);
3469         if (fs_info->balance_ctl) {
3470                 /* this is either (2) or (3) */
3471                 if (!atomic_read(&fs_info->balance_running)) {
3472                         mutex_unlock(&fs_info->balance_mutex);
3473                         if (!mutex_trylock(&fs_info->volume_mutex))
3474                                 goto again;
3475                         mutex_lock(&fs_info->balance_mutex);
3476
3477                         if (fs_info->balance_ctl &&
3478                             !atomic_read(&fs_info->balance_running)) {
3479                                 /* this is (3) */
3480                                 need_unlock = false;
3481                                 goto locked;
3482                         }
3483
3484                         mutex_unlock(&fs_info->balance_mutex);
3485                         mutex_unlock(&fs_info->volume_mutex);
3486                         goto again;
3487                 } else {
3488                         /* this is (2) */
3489                         mutex_unlock(&fs_info->balance_mutex);
3490                         ret = -EINPROGRESS;
3491                         goto out;
3492                 }
3493         } else {
3494                 /* this is (1) */
3495                 mutex_unlock(&fs_info->balance_mutex);
3496                 pr_info("btrfs: dev add/delete/balance/replace/resize operation in progress\n");
3497                 ret = -EINVAL;
3498                 goto out;
3499         }
3500
3501 locked:
3502         BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running));
3503
3504         if (arg) {
3505                 bargs = memdup_user(arg, sizeof(*bargs));
3506                 if (IS_ERR(bargs)) {
3507                         ret = PTR_ERR(bargs);
3508                         goto out_unlock;
3509                 }
3510
3511                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3512                         if (!fs_info->balance_ctl) {
3513                                 ret = -ENOTCONN;
3514                                 goto out_bargs;
3515                         }
3516
3517                         bctl = fs_info->balance_ctl;
3518                         spin_lock(&fs_info->balance_lock);
3519                         bctl->flags |= BTRFS_BALANCE_RESUME;
3520                         spin_unlock(&fs_info->balance_lock);
3521
3522                         goto do_balance;
3523                 }
3524         } else {
3525                 bargs = NULL;
3526         }
3527
3528         if (fs_info->balance_ctl) {
3529                 ret = -EINPROGRESS;
3530                 goto out_bargs;
3531         }
3532
3533         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3534         if (!bctl) {
3535                 ret = -ENOMEM;
3536                 goto out_bargs;
3537         }
3538
3539         bctl->fs_info = fs_info;
3540         if (arg) {
3541                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
3542                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
3543                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
3544
3545                 bctl->flags = bargs->flags;
3546         } else {
3547                 /* balance everything - no filters */
3548                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
3549         }
3550
3551 do_balance:
3552         /*
3553          * Ownership of bctl and mutually_exclusive_operation_running
3554          * goes to to btrfs_balance.  bctl is freed in __cancel_balance,
3555          * or, if restriper was paused all the way until unmount, in
3556          * free_fs_info.  mutually_exclusive_operation_running is
3557          * cleared in __cancel_balance.
3558          */
3559         need_unlock = false;
3560
3561         ret = btrfs_balance(bctl, bargs);
3562
3563         if (arg) {
3564                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
3565                         ret = -EFAULT;
3566         }
3567
3568 out_bargs:
3569         kfree(bargs);
3570 out_unlock:
3571         mutex_unlock(&fs_info->balance_mutex);
3572         mutex_unlock(&fs_info->volume_mutex);
3573         if (need_unlock)
3574                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3575 out:
3576         mnt_drop_write_file(file);
3577         return ret;
3578 }
3579
3580 static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd)
3581 {
3582         if (!capable(CAP_SYS_ADMIN))
3583                 return -EPERM;
3584
3585         switch (cmd) {
3586         case BTRFS_BALANCE_CTL_PAUSE:
3587                 return btrfs_pause_balance(root->fs_info);
3588         case BTRFS_BALANCE_CTL_CANCEL:
3589                 return btrfs_cancel_balance(root->fs_info);
3590         }
3591
3592         return -EINVAL;
3593 }
3594
3595 static long btrfs_ioctl_balance_progress(struct btrfs_root *root,
3596                                          void __user *arg)
3597 {
3598         struct btrfs_fs_info *fs_info = root->fs_info;
3599         struct btrfs_ioctl_balance_args *bargs;
3600         int ret = 0;
3601
3602         if (!capable(CAP_SYS_ADMIN))
3603                 return -EPERM;
3604
3605         mutex_lock(&fs_info->balance_mutex);
3606         if (!fs_info->balance_ctl) {
3607                 ret = -ENOTCONN;
3608                 goto out;
3609         }
3610
3611         bargs = kzalloc(sizeof(*bargs), GFP_NOFS);
3612         if (!bargs) {
3613                 ret = -ENOMEM;
3614                 goto out;
3615         }
3616
3617         update_ioctl_balance_args(fs_info, 1, bargs);
3618
3619         if (copy_to_user(arg, bargs, sizeof(*bargs)))
3620                 ret = -EFAULT;
3621
3622         kfree(bargs);
3623 out:
3624         mutex_unlock(&fs_info->balance_mutex);
3625         return ret;
3626 }
3627
3628 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
3629 {
3630         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3631         struct btrfs_ioctl_quota_ctl_args *sa;
3632         struct btrfs_trans_handle *trans = NULL;
3633         int ret;
3634         int err;
3635
3636         if (!capable(CAP_SYS_ADMIN))
3637                 return -EPERM;
3638
3639         ret = mnt_want_write_file(file);
3640         if (ret)
3641                 return ret;
3642
3643         sa = memdup_user(arg, sizeof(*sa));
3644         if (IS_ERR(sa)) {
3645                 ret = PTR_ERR(sa);
3646                 goto drop_write;
3647         }
3648
3649         if (sa->cmd != BTRFS_QUOTA_CTL_RESCAN) {
3650                 trans = btrfs_start_transaction(root, 2);
3651                 if (IS_ERR(trans)) {
3652                         ret = PTR_ERR(trans);
3653                         goto out;
3654                 }
3655         }
3656
3657         switch (sa->cmd) {
3658         case BTRFS_QUOTA_CTL_ENABLE:
3659                 ret = btrfs_quota_enable(trans, root->fs_info);
3660                 break;
3661         case BTRFS_QUOTA_CTL_DISABLE:
3662                 ret = btrfs_quota_disable(trans, root->fs_info);
3663                 break;
3664         case BTRFS_QUOTA_CTL_RESCAN:
3665                 ret = btrfs_quota_rescan(root->fs_info);
3666                 break;
3667         default:
3668                 ret = -EINVAL;
3669                 break;
3670         }
3671
3672         if (copy_to_user(arg, sa, sizeof(*sa)))
3673                 ret = -EFAULT;
3674
3675         if (trans) {
3676                 err = btrfs_commit_transaction(trans, root);
3677                 if (err && !ret)
3678                         ret = err;
3679         }
3680 out:
3681         kfree(sa);
3682 drop_write:
3683         mnt_drop_write_file(file);
3684         return ret;
3685 }
3686
3687 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
3688 {
3689         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3690         struct btrfs_ioctl_qgroup_assign_args *sa;
3691         struct btrfs_trans_handle *trans;
3692         int ret;
3693         int err;
3694
3695         if (!capable(CAP_SYS_ADMIN))
3696                 return -EPERM;
3697
3698         ret = mnt_want_write_file(file);
3699         if (ret)
3700                 return ret;
3701
3702         sa = memdup_user(arg, sizeof(*sa));
3703         if (IS_ERR(sa)) {
3704                 ret = PTR_ERR(sa);
3705                 goto drop_write;
3706         }
3707
3708         trans = btrfs_join_transaction(root);
3709         if (IS_ERR(trans)) {
3710                 ret = PTR_ERR(trans);
3711                 goto out;
3712         }
3713
3714         /* FIXME: check if the IDs really exist */
3715         if (sa->assign) {
3716                 ret = btrfs_add_qgroup_relation(trans, root->fs_info,
3717                                                 sa->src, sa->dst);
3718         } else {
3719                 ret = btrfs_del_qgroup_relation(trans, root->fs_info,
3720                                                 sa->src, sa->dst);
3721         }
3722
3723         err = btrfs_end_transaction(trans, root);
3724         if (err && !ret)
3725                 ret = err;
3726
3727 out:
3728         kfree(sa);
3729 drop_write:
3730         mnt_drop_write_file(file);
3731         return ret;
3732 }
3733
3734 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
3735 {
3736         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3737         struct btrfs_ioctl_qgroup_create_args *sa;
3738         struct btrfs_trans_handle *trans;
3739         int ret;
3740         int err;
3741
3742         if (!capable(CAP_SYS_ADMIN))
3743                 return -EPERM;
3744
3745         ret = mnt_want_write_file(file);
3746         if (ret)
3747                 return ret;
3748
3749         sa = memdup_user(arg, sizeof(*sa));
3750         if (IS_ERR(sa)) {
3751                 ret = PTR_ERR(sa);
3752                 goto drop_write;
3753         }
3754
3755         if (!sa->qgroupid) {
3756                 ret = -EINVAL;
3757                 goto out;
3758         }
3759
3760         trans = btrfs_join_transaction(root);
3761         if (IS_ERR(trans)) {
3762                 ret = PTR_ERR(trans);
3763                 goto out;
3764         }
3765
3766         /* FIXME: check if the IDs really exist */
3767         if (sa->create) {
3768                 ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid,
3769                                           NULL);
3770         } else {
3771                 ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid);
3772         }
3773
3774         err = btrfs_end_transaction(trans, root);
3775         if (err && !ret)
3776                 ret = err;
3777
3778 out:
3779         kfree(sa);
3780 drop_write:
3781         mnt_drop_write_file(file);
3782         return ret;
3783 }
3784
3785 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
3786 {
3787         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3788         struct btrfs_ioctl_qgroup_limit_args *sa;
3789         struct btrfs_trans_handle *trans;
3790         int ret;
3791         int err;
3792         u64 qgroupid;
3793
3794         if (!capable(CAP_SYS_ADMIN))
3795                 return -EPERM;
3796
3797         ret = mnt_want_write_file(file);
3798         if (ret)
3799                 return ret;
3800
3801         sa = memdup_user(arg, sizeof(*sa));
3802         if (IS_ERR(sa)) {
3803                 ret = PTR_ERR(sa);
3804                 goto drop_write;
3805         }
3806
3807         trans = btrfs_join_transaction(root);
3808         if (IS_ERR(trans)) {
3809                 ret = PTR_ERR(trans);
3810                 goto out;
3811         }
3812
3813         qgroupid = sa->qgroupid;
3814         if (!qgroupid) {
3815                 /* take the current subvol as qgroup */
3816                 qgroupid = root->root_key.objectid;
3817         }
3818
3819         /* FIXME: check if the IDs really exist */
3820         ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim);
3821
3822         err = btrfs_end_transaction(trans, root);
3823         if (err && !ret)
3824                 ret = err;
3825
3826 out:
3827         kfree(sa);
3828 drop_write:
3829         mnt_drop_write_file(file);
3830         return ret;
3831 }
3832
3833 static long btrfs_ioctl_set_received_subvol(struct file *file,
3834                                             void __user *arg)
3835 {
3836         struct btrfs_ioctl_received_subvol_args *sa = NULL;
3837         struct inode *inode = fdentry(file)->d_inode;
3838         struct btrfs_root *root = BTRFS_I(inode)->root;
3839         struct btrfs_root_item *root_item = &root->root_item;
3840         struct btrfs_trans_handle *trans;
3841         struct timespec ct = CURRENT_TIME;
3842         int ret = 0;
3843
3844         ret = mnt_want_write_file(file);
3845         if (ret < 0)
3846                 return ret;
3847
3848         down_write(&root->fs_info->subvol_sem);
3849
3850         if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
3851                 ret = -EINVAL;
3852                 goto out;
3853         }
3854
3855         if (btrfs_root_readonly(root)) {
3856                 ret = -EROFS;
3857                 goto out;
3858         }
3859
3860         if (!inode_owner_or_capable(inode)) {
3861                 ret = -EACCES;
3862                 goto out;
3863         }
3864
3865         sa = memdup_user(arg, sizeof(*sa));
3866         if (IS_ERR(sa)) {
3867                 ret = PTR_ERR(sa);
3868                 sa = NULL;
3869                 goto out;
3870         }
3871
3872         trans = btrfs_start_transaction(root, 1);
3873         if (IS_ERR(trans)) {
3874                 ret = PTR_ERR(trans);
3875                 trans = NULL;
3876                 goto out;
3877         }
3878
3879         sa->rtransid = trans->transid;
3880         sa->rtime.sec = ct.tv_sec;
3881         sa->rtime.nsec = ct.tv_nsec;
3882
3883         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
3884         btrfs_set_root_stransid(root_item, sa->stransid);
3885         btrfs_set_root_rtransid(root_item, sa->rtransid);
3886         root_item->stime.sec = cpu_to_le64(sa->stime.sec);
3887         root_item->stime.nsec = cpu_to_le32(sa->stime.nsec);
3888         root_item->rtime.sec = cpu_to_le64(sa->rtime.sec);
3889         root_item->rtime.nsec = cpu_to_le32(sa->rtime.nsec);
3890
3891         ret = btrfs_update_root(trans, root->fs_info->tree_root,
3892                                 &root->root_key, &root->root_item);
3893         if (ret < 0) {
3894                 btrfs_end_transaction(trans, root);
3895                 trans = NULL;
3896                 goto out;
3897         } else {
3898                 ret = btrfs_commit_transaction(trans, root);
3899                 if (ret < 0)
3900                         goto out;