Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs...
[pandora-kernel.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 "compat.h"
43 #include "ctree.h"
44 #include "disk-io.h"
45 #include "transaction.h"
46 #include "btrfs_inode.h"
47 #include "ioctl.h"
48 #include "print-tree.h"
49 #include "volumes.h"
50 #include "locking.h"
51
52 /* Mask out flags that are inappropriate for the given type of inode. */
53 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
54 {
55         if (S_ISDIR(mode))
56                 return flags;
57         else if (S_ISREG(mode))
58                 return flags & ~FS_DIRSYNC_FL;
59         else
60                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
61 }
62
63 /*
64  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
65  */
66 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
67 {
68         unsigned int iflags = 0;
69
70         if (flags & BTRFS_INODE_SYNC)
71                 iflags |= FS_SYNC_FL;
72         if (flags & BTRFS_INODE_IMMUTABLE)
73                 iflags |= FS_IMMUTABLE_FL;
74         if (flags & BTRFS_INODE_APPEND)
75                 iflags |= FS_APPEND_FL;
76         if (flags & BTRFS_INODE_NODUMP)
77                 iflags |= FS_NODUMP_FL;
78         if (flags & BTRFS_INODE_NOATIME)
79                 iflags |= FS_NOATIME_FL;
80         if (flags & BTRFS_INODE_DIRSYNC)
81                 iflags |= FS_DIRSYNC_FL;
82
83         return iflags;
84 }
85
86 /*
87  * Update inode->i_flags based on the btrfs internal flags.
88  */
89 void btrfs_update_iflags(struct inode *inode)
90 {
91         struct btrfs_inode *ip = BTRFS_I(inode);
92
93         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
94
95         if (ip->flags & BTRFS_INODE_SYNC)
96                 inode->i_flags |= S_SYNC;
97         if (ip->flags & BTRFS_INODE_IMMUTABLE)
98                 inode->i_flags |= S_IMMUTABLE;
99         if (ip->flags & BTRFS_INODE_APPEND)
100                 inode->i_flags |= S_APPEND;
101         if (ip->flags & BTRFS_INODE_NOATIME)
102                 inode->i_flags |= S_NOATIME;
103         if (ip->flags & BTRFS_INODE_DIRSYNC)
104                 inode->i_flags |= S_DIRSYNC;
105 }
106
107 /*
108  * Inherit flags from the parent inode.
109  *
110  * Unlike extN we don't have any flags we don't want to inherit currently.
111  */
112 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
113 {
114         unsigned int flags;
115
116         if (!dir)
117                 return;
118
119         flags = BTRFS_I(dir)->flags;
120
121         if (S_ISREG(inode->i_mode))
122                 flags &= ~BTRFS_INODE_DIRSYNC;
123         else if (!S_ISDIR(inode->i_mode))
124                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
125
126         BTRFS_I(inode)->flags = flags;
127         btrfs_update_iflags(inode);
128 }
129
130 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
131 {
132         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
133         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
134
135         if (copy_to_user(arg, &flags, sizeof(flags)))
136                 return -EFAULT;
137         return 0;
138 }
139
140 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
141 {
142         struct inode *inode = file->f_path.dentry->d_inode;
143         struct btrfs_inode *ip = BTRFS_I(inode);
144         struct btrfs_root *root = ip->root;
145         struct btrfs_trans_handle *trans;
146         unsigned int flags, oldflags;
147         int ret;
148
149         if (copy_from_user(&flags, arg, sizeof(flags)))
150                 return -EFAULT;
151
152         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
153                       FS_NOATIME_FL | FS_NODUMP_FL | \
154                       FS_SYNC_FL | FS_DIRSYNC_FL))
155                 return -EOPNOTSUPP;
156
157         if (!is_owner_or_cap(inode))
158                 return -EACCES;
159
160         mutex_lock(&inode->i_mutex);
161
162         flags = btrfs_mask_flags(inode->i_mode, flags);
163         oldflags = btrfs_flags_to_ioctl(ip->flags);
164         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
165                 if (!capable(CAP_LINUX_IMMUTABLE)) {
166                         ret = -EPERM;
167                         goto out_unlock;
168                 }
169         }
170
171         ret = mnt_want_write(file->f_path.mnt);
172         if (ret)
173                 goto out_unlock;
174
175         if (flags & FS_SYNC_FL)
176                 ip->flags |= BTRFS_INODE_SYNC;
177         else
178                 ip->flags &= ~BTRFS_INODE_SYNC;
179         if (flags & FS_IMMUTABLE_FL)
180                 ip->flags |= BTRFS_INODE_IMMUTABLE;
181         else
182                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
183         if (flags & FS_APPEND_FL)
184                 ip->flags |= BTRFS_INODE_APPEND;
185         else
186                 ip->flags &= ~BTRFS_INODE_APPEND;
187         if (flags & FS_NODUMP_FL)
188                 ip->flags |= BTRFS_INODE_NODUMP;
189         else
190                 ip->flags &= ~BTRFS_INODE_NODUMP;
191         if (flags & FS_NOATIME_FL)
192                 ip->flags |= BTRFS_INODE_NOATIME;
193         else
194                 ip->flags &= ~BTRFS_INODE_NOATIME;
195         if (flags & FS_DIRSYNC_FL)
196                 ip->flags |= BTRFS_INODE_DIRSYNC;
197         else
198                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
199
200
201         trans = btrfs_join_transaction(root, 1);
202         BUG_ON(!trans);
203
204         ret = btrfs_update_inode(trans, root, inode);
205         BUG_ON(ret);
206
207         btrfs_update_iflags(inode);
208         inode->i_ctime = CURRENT_TIME;
209         btrfs_end_transaction(trans, root);
210
211         mnt_drop_write(file->f_path.mnt);
212  out_unlock:
213         mutex_unlock(&inode->i_mutex);
214         return 0;
215 }
216
217 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
218 {
219         struct inode *inode = file->f_path.dentry->d_inode;
220
221         return put_user(inode->i_generation, arg);
222 }
223
224 static noinline int create_subvol(struct btrfs_root *root,
225                                   struct dentry *dentry,
226                                   char *name, int namelen)
227 {
228         struct btrfs_trans_handle *trans;
229         struct btrfs_key key;
230         struct btrfs_root_item root_item;
231         struct btrfs_inode_item *inode_item;
232         struct extent_buffer *leaf;
233         struct btrfs_root *new_root = root;
234         struct inode *dir;
235         int ret;
236         int err;
237         u64 objectid;
238         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
239         u64 index = 0;
240         unsigned long nr = 1;
241
242         ret = btrfs_check_metadata_free_space(root);
243         if (ret)
244                 goto fail_commit;
245
246         trans = btrfs_start_transaction(root, 1);
247         BUG_ON(!trans);
248
249         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
250                                        0, &objectid);
251         if (ret)
252                 goto fail;
253
254         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
255                                       0, objectid, NULL, 0, 0, 0);
256         if (IS_ERR(leaf)) {
257                 ret = PTR_ERR(leaf);
258                 goto fail;
259         }
260
261         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
262         btrfs_set_header_bytenr(leaf, leaf->start);
263         btrfs_set_header_generation(leaf, trans->transid);
264         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
265         btrfs_set_header_owner(leaf, objectid);
266
267         write_extent_buffer(leaf, root->fs_info->fsid,
268                             (unsigned long)btrfs_header_fsid(leaf),
269                             BTRFS_FSID_SIZE);
270         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
271                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
272                             BTRFS_UUID_SIZE);
273         btrfs_mark_buffer_dirty(leaf);
274
275         inode_item = &root_item.inode;
276         memset(inode_item, 0, sizeof(*inode_item));
277         inode_item->generation = cpu_to_le64(1);
278         inode_item->size = cpu_to_le64(3);
279         inode_item->nlink = cpu_to_le32(1);
280         inode_item->nbytes = cpu_to_le64(root->leafsize);
281         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
282
283         btrfs_set_root_bytenr(&root_item, leaf->start);
284         btrfs_set_root_generation(&root_item, trans->transid);
285         btrfs_set_root_level(&root_item, 0);
286         btrfs_set_root_refs(&root_item, 1);
287         btrfs_set_root_used(&root_item, 0);
288         btrfs_set_root_last_snapshot(&root_item, 0);
289
290         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
291         root_item.drop_level = 0;
292
293         btrfs_tree_unlock(leaf);
294         free_extent_buffer(leaf);
295         leaf = NULL;
296
297         btrfs_set_root_dirid(&root_item, new_dirid);
298
299         key.objectid = objectid;
300         key.offset = 0;
301         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
302         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
303                                 &root_item);
304         if (ret)
305                 goto fail;
306
307         /*
308          * insert the directory item
309          */
310         key.offset = (u64)-1;
311         dir = dentry->d_parent->d_inode;
312         ret = btrfs_set_inode_index(dir, &index);
313         BUG_ON(ret);
314
315         ret = btrfs_insert_dir_item(trans, root,
316                                     name, namelen, dir->i_ino, &key,
317                                     BTRFS_FT_DIR, index);
318         if (ret)
319                 goto fail;
320
321         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
322         ret = btrfs_update_inode(trans, root, dir);
323         BUG_ON(ret);
324
325         /* add the backref first */
326         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
327                                  objectid, BTRFS_ROOT_BACKREF_KEY,
328                                  root->root_key.objectid,
329                                  dir->i_ino, index, name, namelen);
330
331         BUG_ON(ret);
332
333         /* now add the forward ref */
334         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
335                                  root->root_key.objectid, BTRFS_ROOT_REF_KEY,
336                                  objectid,
337                                  dir->i_ino, index, name, namelen);
338
339         BUG_ON(ret);
340
341         ret = btrfs_commit_transaction(trans, root);
342         if (ret)
343                 goto fail_commit;
344
345         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
346         BUG_ON(!new_root);
347
348         trans = btrfs_start_transaction(new_root, 1);
349         BUG_ON(!trans);
350
351         ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
352                                        BTRFS_I(dir)->block_group);
353         if (ret)
354                 goto fail;
355
356 fail:
357         nr = trans->blocks_used;
358         err = btrfs_commit_transaction(trans, new_root);
359         if (err && !ret)
360                 ret = err;
361 fail_commit:
362         btrfs_btree_balance_dirty(root, nr);
363         return ret;
364 }
365
366 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
367                            char *name, int namelen)
368 {
369         struct btrfs_pending_snapshot *pending_snapshot;
370         struct btrfs_trans_handle *trans;
371         int ret = 0;
372         int err;
373         unsigned long nr = 0;
374
375         if (!root->ref_cows)
376                 return -EINVAL;
377
378         ret = btrfs_check_metadata_free_space(root);
379         if (ret)
380                 goto fail_unlock;
381
382         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
383         if (!pending_snapshot) {
384                 ret = -ENOMEM;
385                 goto fail_unlock;
386         }
387         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
388         if (!pending_snapshot->name) {
389                 ret = -ENOMEM;
390                 kfree(pending_snapshot);
391                 goto fail_unlock;
392         }
393         memcpy(pending_snapshot->name, name, namelen);
394         pending_snapshot->name[namelen] = '\0';
395         pending_snapshot->dentry = dentry;
396         trans = btrfs_start_transaction(root, 1);
397         BUG_ON(!trans);
398         pending_snapshot->root = root;
399         list_add(&pending_snapshot->list,
400                  &trans->transaction->pending_snapshots);
401         err = btrfs_commit_transaction(trans, root);
402
403 fail_unlock:
404         btrfs_btree_balance_dirty(root, nr);
405         return ret;
406 }
407
408 /* copy of may_create in fs/namei.c() */
409 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
410 {
411         if (child->d_inode)
412                 return -EEXIST;
413         if (IS_DEADDIR(dir))
414                 return -ENOENT;
415         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
416 }
417
418 /*
419  * Create a new subvolume below @parent.  This is largely modeled after
420  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
421  * inside this filesystem so it's quite a bit simpler.
422  */
423 static noinline int btrfs_mksubvol(struct path *parent, char *name,
424                                    int mode, int namelen,
425                                    struct btrfs_root *snap_src)
426 {
427         struct dentry *dentry;
428         int error;
429
430         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
431
432         dentry = lookup_one_len(name, parent->dentry, namelen);
433         error = PTR_ERR(dentry);
434         if (IS_ERR(dentry))
435                 goto out_unlock;
436
437         error = -EEXIST;
438         if (dentry->d_inode)
439                 goto out_dput;
440
441         if (!IS_POSIXACL(parent->dentry->d_inode))
442                 mode &= ~current_umask();
443
444         error = mnt_want_write(parent->mnt);
445         if (error)
446                 goto out_dput;
447
448         error = btrfs_may_create(parent->dentry->d_inode, dentry);
449         if (error)
450                 goto out_drop_write;
451
452         /*
453          * Actually perform the low-level subvolume creation after all
454          * this VFS fuzz.
455          *
456          * Eventually we want to pass in an inode under which we create this
457          * subvolume, but for now all are under the filesystem root.
458          *
459          * Also we should pass on the mode eventually to allow creating new
460          * subvolume with specific mode bits.
461          */
462         if (snap_src) {
463                 struct dentry *dir = dentry->d_parent;
464                 struct dentry *test = dir->d_parent;
465                 struct btrfs_path *path = btrfs_alloc_path();
466                 int ret;
467                 u64 test_oid;
468                 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
469
470                 test_oid = snap_src->root_key.objectid;
471
472                 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
473                                           path, parent_oid, test_oid);
474                 if (ret == 0)
475                         goto create;
476                 btrfs_release_path(snap_src->fs_info->tree_root, path);
477
478                 /* we need to make sure we aren't creating a directory loop
479                  * by taking a snapshot of something that has our current
480                  * subvol in its directory tree.  So, this loops through
481                  * the dentries and checks the forward refs for each subvolume
482                  * to see if is references the subvolume where we are
483                  * placing this new snapshot.
484                  */
485                 while (1) {
486                         if (!test ||
487                             dir == snap_src->fs_info->sb->s_root ||
488                             test == snap_src->fs_info->sb->s_root ||
489                             test->d_inode->i_sb != snap_src->fs_info->sb) {
490                                 break;
491                         }
492                         if (S_ISLNK(test->d_inode->i_mode)) {
493                                 printk(KERN_INFO "Btrfs symlink in snapshot "
494                                        "path, failed\n");
495                                 error = -EMLINK;
496                                 btrfs_free_path(path);
497                                 goto out_drop_write;
498                         }
499                         test_oid =
500                                 BTRFS_I(test->d_inode)->root->root_key.objectid;
501                         ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
502                                   path, test_oid, parent_oid);
503                         if (ret == 0) {
504                                 printk(KERN_INFO "Btrfs snapshot creation "
505                                        "failed, looping\n");
506                                 error = -EMLINK;
507                                 btrfs_free_path(path);
508                                 goto out_drop_write;
509                         }
510                         btrfs_release_path(snap_src->fs_info->tree_root, path);
511                         test = test->d_parent;
512                 }
513 create:
514                 btrfs_free_path(path);
515                 error = create_snapshot(snap_src, dentry, name, namelen);
516         } else {
517                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
518                                       dentry, name, namelen);
519         }
520         if (error)
521                 goto out_drop_write;
522
523         fsnotify_mkdir(parent->dentry->d_inode, dentry);
524 out_drop_write:
525         mnt_drop_write(parent->mnt);
526 out_dput:
527         dput(dentry);
528 out_unlock:
529         mutex_unlock(&parent->dentry->d_inode->i_mutex);
530         return error;
531 }
532
533
534 static int btrfs_defrag_file(struct file *file)
535 {
536         struct inode *inode = fdentry(file)->d_inode;
537         struct btrfs_root *root = BTRFS_I(inode)->root;
538         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
539         struct btrfs_ordered_extent *ordered;
540         struct page *page;
541         unsigned long last_index;
542         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
543         unsigned long total_read = 0;
544         u64 page_start;
545         u64 page_end;
546         unsigned long i;
547         int ret;
548
549         ret = btrfs_check_data_free_space(root, inode, inode->i_size);
550         if (ret)
551                 return -ENOSPC;
552
553         mutex_lock(&inode->i_mutex);
554         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
555         for (i = 0; i <= last_index; i++) {
556                 if (total_read % ra_pages == 0) {
557                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
558                                        min(last_index, i + ra_pages - 1));
559                 }
560                 total_read++;
561 again:
562                 page = grab_cache_page(inode->i_mapping, i);
563                 if (!page)
564                         goto out_unlock;
565                 if (!PageUptodate(page)) {
566                         btrfs_readpage(NULL, page);
567                         lock_page(page);
568                         if (!PageUptodate(page)) {
569                                 unlock_page(page);
570                                 page_cache_release(page);
571                                 goto out_unlock;
572                         }
573                 }
574
575                 wait_on_page_writeback(page);
576
577                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
578                 page_end = page_start + PAGE_CACHE_SIZE - 1;
579                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
580
581                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
582                 if (ordered) {
583                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
584                         unlock_page(page);
585                         page_cache_release(page);
586                         btrfs_start_ordered_extent(inode, ordered, 1);
587                         btrfs_put_ordered_extent(ordered);
588                         goto again;
589                 }
590                 set_page_extent_mapped(page);
591
592                 /*
593                  * this makes sure page_mkwrite is called on the
594                  * page if it is dirtied again later
595                  */
596                 clear_page_dirty_for_io(page);
597
598                 btrfs_set_extent_delalloc(inode, page_start, page_end);
599                 set_page_dirty(page);
600                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
601                 unlock_page(page);
602                 page_cache_release(page);
603                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
604         }
605
606 out_unlock:
607         mutex_unlock(&inode->i_mutex);
608         return 0;
609 }
610
611 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
612 {
613         u64 new_size;
614         u64 old_size;
615         u64 devid = 1;
616         struct btrfs_ioctl_vol_args *vol_args;
617         struct btrfs_trans_handle *trans;
618         struct btrfs_device *device = NULL;
619         char *sizestr;
620         char *devstr = NULL;
621         int ret = 0;
622         int namelen;
623         int mod = 0;
624
625         if (root->fs_info->sb->s_flags & MS_RDONLY)
626                 return -EROFS;
627
628         if (!capable(CAP_SYS_ADMIN))
629                 return -EPERM;
630
631         vol_args = memdup_user(arg, sizeof(*vol_args));
632         if (IS_ERR(vol_args))
633                 return PTR_ERR(vol_args);
634
635         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
636         namelen = strlen(vol_args->name);
637
638         mutex_lock(&root->fs_info->volume_mutex);
639         sizestr = vol_args->name;
640         devstr = strchr(sizestr, ':');
641         if (devstr) {
642                 char *end;
643                 sizestr = devstr + 1;
644                 *devstr = '\0';
645                 devstr = vol_args->name;
646                 devid = simple_strtoull(devstr, &end, 10);
647                 printk(KERN_INFO "resizing devid %llu\n",
648                        (unsigned long long)devid);
649         }
650         device = btrfs_find_device(root, devid, NULL, NULL);
651         if (!device) {
652                 printk(KERN_INFO "resizer unable to find device %llu\n",
653                        (unsigned long long)devid);
654                 ret = -EINVAL;
655                 goto out_unlock;
656         }
657         if (!strcmp(sizestr, "max"))
658                 new_size = device->bdev->bd_inode->i_size;
659         else {
660                 if (sizestr[0] == '-') {
661                         mod = -1;
662                         sizestr++;
663                 } else if (sizestr[0] == '+') {
664                         mod = 1;
665                         sizestr++;
666                 }
667                 new_size = btrfs_parse_size(sizestr);
668                 if (new_size == 0) {
669                         ret = -EINVAL;
670                         goto out_unlock;
671                 }
672         }
673
674         old_size = device->total_bytes;
675
676         if (mod < 0) {
677                 if (new_size > old_size) {
678                         ret = -EINVAL;
679                         goto out_unlock;
680                 }
681                 new_size = old_size - new_size;
682         } else if (mod > 0) {
683                 new_size = old_size + new_size;
684         }
685
686         if (new_size < 256 * 1024 * 1024) {
687                 ret = -EINVAL;
688                 goto out_unlock;
689         }
690         if (new_size > device->bdev->bd_inode->i_size) {
691                 ret = -EFBIG;
692                 goto out_unlock;
693         }
694
695         do_div(new_size, root->sectorsize);
696         new_size *= root->sectorsize;
697
698         printk(KERN_INFO "new size for %s is %llu\n",
699                 device->name, (unsigned long long)new_size);
700
701         if (new_size > old_size) {
702                 trans = btrfs_start_transaction(root, 1);
703                 ret = btrfs_grow_device(trans, device, new_size);
704                 btrfs_commit_transaction(trans, root);
705         } else {
706                 ret = btrfs_shrink_device(device, new_size);
707         }
708
709 out_unlock:
710         mutex_unlock(&root->fs_info->volume_mutex);
711         kfree(vol_args);
712         return ret;
713 }
714
715 static noinline int btrfs_ioctl_snap_create(struct file *file,
716                                             void __user *arg, int subvol)
717 {
718         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
719         struct btrfs_ioctl_vol_args *vol_args;
720         struct btrfs_dir_item *di;
721         struct btrfs_path *path;
722         struct file *src_file;
723         u64 root_dirid;
724         int namelen;
725         int ret = 0;
726
727         if (root->fs_info->sb->s_flags & MS_RDONLY)
728                 return -EROFS;
729
730         vol_args = memdup_user(arg, sizeof(*vol_args));
731         if (IS_ERR(vol_args))
732                 return PTR_ERR(vol_args);
733
734         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
735         namelen = strlen(vol_args->name);
736         if (strchr(vol_args->name, '/')) {
737                 ret = -EINVAL;
738                 goto out;
739         }
740
741         path = btrfs_alloc_path();
742         if (!path) {
743                 ret = -ENOMEM;
744                 goto out;
745         }
746
747         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
748         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
749                             path, root_dirid,
750                             vol_args->name, namelen, 0);
751         btrfs_free_path(path);
752
753         if (di && !IS_ERR(di)) {
754                 ret = -EEXIST;
755                 goto out;
756         }
757
758         if (IS_ERR(di)) {
759                 ret = PTR_ERR(di);
760                 goto out;
761         }
762
763         if (subvol) {
764                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
765                                      file->f_path.dentry->d_inode->i_mode,
766                                      namelen, NULL);
767         } else {
768                 struct inode *src_inode;
769                 src_file = fget(vol_args->fd);
770                 if (!src_file) {
771                         ret = -EINVAL;
772                         goto out;
773                 }
774
775                 src_inode = src_file->f_path.dentry->d_inode;
776                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
777                         printk(KERN_INFO "btrfs: Snapshot src from "
778                                "another FS\n");
779                         ret = -EINVAL;
780                         fput(src_file);
781                         goto out;
782                 }
783                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
784                              file->f_path.dentry->d_inode->i_mode,
785                              namelen, BTRFS_I(src_inode)->root);
786                 fput(src_file);
787         }
788
789 out:
790         kfree(vol_args);
791         return ret;
792 }
793
794 static int btrfs_ioctl_defrag(struct file *file)
795 {
796         struct inode *inode = fdentry(file)->d_inode;
797         struct btrfs_root *root = BTRFS_I(inode)->root;
798         int ret;
799
800         ret = mnt_want_write(file->f_path.mnt);
801         if (ret)
802                 return ret;
803
804         switch (inode->i_mode & S_IFMT) {
805         case S_IFDIR:
806                 if (!capable(CAP_SYS_ADMIN)) {
807                         ret = -EPERM;
808                         goto out;
809                 }
810                 btrfs_defrag_root(root, 0);
811                 btrfs_defrag_root(root->fs_info->extent_root, 0);
812                 break;
813         case S_IFREG:
814                 if (!(file->f_mode & FMODE_WRITE)) {
815                         ret = -EINVAL;
816                         goto out;
817                 }
818                 btrfs_defrag_file(file);
819                 break;
820         }
821 out:
822         mnt_drop_write(file->f_path.mnt);
823         return ret;
824 }
825
826 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
827 {
828         struct btrfs_ioctl_vol_args *vol_args;
829         int ret;
830
831         if (!capable(CAP_SYS_ADMIN))
832                 return -EPERM;
833
834         vol_args = memdup_user(arg, sizeof(*vol_args));
835         if (IS_ERR(vol_args))
836                 return PTR_ERR(vol_args);
837
838         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
839         ret = btrfs_init_new_device(root, vol_args->name);
840
841         kfree(vol_args);
842         return ret;
843 }
844
845 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
846 {
847         struct btrfs_ioctl_vol_args *vol_args;
848         int ret;
849
850         if (!capable(CAP_SYS_ADMIN))
851                 return -EPERM;
852
853         if (root->fs_info->sb->s_flags & MS_RDONLY)
854                 return -EROFS;
855
856         vol_args = memdup_user(arg, sizeof(*vol_args));
857         if (IS_ERR(vol_args))
858                 return PTR_ERR(vol_args);
859
860         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
861         ret = btrfs_rm_device(root, vol_args->name);
862
863         kfree(vol_args);
864         return ret;
865 }
866
867 static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
868                 u64 off, u64 olen, u64 destoff)
869 {
870         struct inode *inode = fdentry(file)->d_inode;
871         struct btrfs_root *root = BTRFS_I(inode)->root;
872         struct file *src_file;
873         struct inode *src;
874         struct btrfs_trans_handle *trans;
875         struct btrfs_path *path;
876         struct extent_buffer *leaf;
877         char *buf;
878         struct btrfs_key key;
879         u32 nritems;
880         int slot;
881         int ret;
882         u64 len = olen;
883         u64 bs = root->fs_info->sb->s_blocksize;
884         u64 hint_byte;
885
886         /*
887          * TODO:
888          * - split compressed inline extents.  annoying: we need to
889          *   decompress into destination's address_space (the file offset
890          *   may change, so source mapping won't do), then recompress (or
891          *   otherwise reinsert) a subrange.
892          * - allow ranges within the same file to be cloned (provided
893          *   they don't overlap)?
894          */
895
896         /* the destination must be opened for writing */
897         if (!(file->f_mode & FMODE_WRITE))
898                 return -EINVAL;
899
900         ret = mnt_want_write(file->f_path.mnt);
901         if (ret)
902                 return ret;
903
904         src_file = fget(srcfd);
905         if (!src_file) {
906                 ret = -EBADF;
907                 goto out_drop_write;
908         }
909         src = src_file->f_dentry->d_inode;
910
911         ret = -EINVAL;
912         if (src == inode)
913                 goto out_fput;
914
915         ret = -EISDIR;
916         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
917                 goto out_fput;
918
919         ret = -EXDEV;
920         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
921                 goto out_fput;
922
923         ret = -ENOMEM;
924         buf = vmalloc(btrfs_level_size(root, 0));
925         if (!buf)
926                 goto out_fput;
927
928         path = btrfs_alloc_path();
929         if (!path) {
930                 vfree(buf);
931                 goto out_fput;
932         }
933         path->reada = 2;
934
935         if (inode < src) {
936                 mutex_lock(&inode->i_mutex);
937                 mutex_lock(&src->i_mutex);
938         } else {
939                 mutex_lock(&src->i_mutex);
940                 mutex_lock(&inode->i_mutex);
941         }
942
943         /* determine range to clone */
944         ret = -EINVAL;
945         if (off >= src->i_size || off + len > src->i_size)
946                 goto out_unlock;
947         if (len == 0)
948                 olen = len = src->i_size - off;
949         /* if we extend to eof, continue to block boundary */
950         if (off + len == src->i_size)
951                 len = ((src->i_size + bs-1) & ~(bs-1))
952                         - off;
953
954         /* verify the end result is block aligned */
955         if ((off & (bs-1)) ||
956             ((off + len) & (bs-1)))
957                 goto out_unlock;
958
959         /* do any pending delalloc/csum calc on src, one way or
960            another, and lock file content */
961         while (1) {
962                 struct btrfs_ordered_extent *ordered;
963                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
964                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
965                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
966                         break;
967                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
968                 if (ordered)
969                         btrfs_put_ordered_extent(ordered);
970                 btrfs_wait_ordered_range(src, off, off+len);
971         }
972
973         trans = btrfs_start_transaction(root, 1);
974         BUG_ON(!trans);
975
976         /* punch hole in destination first */
977         btrfs_drop_extents(trans, root, inode, off, off + len,
978                            off + len, 0, &hint_byte, 1);
979
980         /* clone data */
981         key.objectid = src->i_ino;
982         key.type = BTRFS_EXTENT_DATA_KEY;
983         key.offset = 0;
984
985         while (1) {
986                 /*
987                  * note the key will change type as we walk through the
988                  * tree.
989                  */
990                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
991                 if (ret < 0)
992                         goto out;
993
994                 nritems = btrfs_header_nritems(path->nodes[0]);
995                 if (path->slots[0] >= nritems) {
996                         ret = btrfs_next_leaf(root, path);
997                         if (ret < 0)
998                                 goto out;
999                         if (ret > 0)
1000                                 break;
1001                         nritems = btrfs_header_nritems(path->nodes[0]);
1002                 }
1003                 leaf = path->nodes[0];
1004                 slot = path->slots[0];
1005
1006                 btrfs_item_key_to_cpu(leaf, &key, slot);
1007                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1008                     key.objectid != src->i_ino)
1009                         break;
1010
1011                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1012                         struct btrfs_file_extent_item *extent;
1013                         int type;
1014                         u32 size;
1015                         struct btrfs_key new_key;
1016                         u64 disko = 0, diskl = 0;
1017                         u64 datao = 0, datal = 0;
1018                         u8 comp;
1019
1020                         size = btrfs_item_size_nr(leaf, slot);
1021                         read_extent_buffer(leaf, buf,
1022                                            btrfs_item_ptr_offset(leaf, slot),
1023                                            size);
1024
1025                         extent = btrfs_item_ptr(leaf, slot,
1026                                                 struct btrfs_file_extent_item);
1027                         comp = btrfs_file_extent_compression(leaf, extent);
1028                         type = btrfs_file_extent_type(leaf, extent);
1029                         if (type == BTRFS_FILE_EXTENT_REG ||
1030                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1031                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1032                                                                       extent);
1033                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1034                                                                  extent);
1035                                 datao = btrfs_file_extent_offset(leaf, extent);
1036                                 datal = btrfs_file_extent_num_bytes(leaf,
1037                                                                     extent);
1038                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1039                                 /* take upper bound, may be compressed */
1040                                 datal = btrfs_file_extent_ram_bytes(leaf,
1041                                                                     extent);
1042                         }
1043                         btrfs_release_path(root, path);
1044
1045                         if (key.offset + datal < off ||
1046                             key.offset >= off+len)
1047                                 goto next;
1048
1049                         memcpy(&new_key, &key, sizeof(new_key));
1050                         new_key.objectid = inode->i_ino;
1051                         new_key.offset = key.offset + destoff - off;
1052
1053                         if (type == BTRFS_FILE_EXTENT_REG ||
1054                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1055                                 ret = btrfs_insert_empty_item(trans, root, path,
1056                                                               &new_key, size);
1057                                 if (ret)
1058                                         goto out;
1059
1060                                 leaf = path->nodes[0];
1061                                 slot = path->slots[0];
1062                                 write_extent_buffer(leaf, buf,
1063                                             btrfs_item_ptr_offset(leaf, slot),
1064                                             size);
1065
1066                                 extent = btrfs_item_ptr(leaf, slot,
1067                                                 struct btrfs_file_extent_item);
1068
1069                                 if (off > key.offset) {
1070                                         datao += off - key.offset;
1071                                         datal -= off - key.offset;
1072                                 }
1073                                 if (key.offset + datao + datal + key.offset >
1074                                     off + len)
1075                                         datal = off + len - key.offset - datao;
1076                                 /* disko == 0 means it's a hole */
1077                                 if (!disko)
1078                                         datao = 0;
1079
1080                                 btrfs_set_file_extent_offset(leaf, extent,
1081                                                              datao);
1082                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1083                                                                 datal);
1084                                 if (disko) {
1085                                         inode_add_bytes(inode, datal);
1086                                         ret = btrfs_inc_extent_ref(trans, root,
1087                                                         disko, diskl, 0,
1088                                                         root->root_key.objectid,
1089                                                         inode->i_ino,
1090                                                         new_key.offset - datao);
1091                                         BUG_ON(ret);
1092                                 }
1093                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1094                                 u64 skip = 0;
1095                                 u64 trim = 0;
1096                                 if (off > key.offset) {
1097                                         skip = off - key.offset;
1098                                         new_key.offset += skip;
1099                                 }
1100
1101                                 if (key.offset + datal > off+len)
1102                                         trim = key.offset + datal - (off+len);
1103
1104                                 if (comp && (skip || trim)) {
1105                                         ret = -EINVAL;
1106                                         goto out;
1107                                 }
1108                                 size -= skip + trim;
1109                                 datal -= skip + trim;
1110                                 ret = btrfs_insert_empty_item(trans, root, path,
1111                                                               &new_key, size);
1112                                 if (ret)
1113                                         goto out;
1114
1115                                 if (skip) {
1116                                         u32 start =
1117                                           btrfs_file_extent_calc_inline_size(0);
1118                                         memmove(buf+start, buf+start+skip,
1119                                                 datal);
1120                                 }
1121
1122                                 leaf = path->nodes[0];
1123                                 slot = path->slots[0];
1124                                 write_extent_buffer(leaf, buf,
1125                                             btrfs_item_ptr_offset(leaf, slot),
1126                                             size);
1127                                 inode_add_bytes(inode, datal);
1128                         }
1129
1130                         btrfs_mark_buffer_dirty(leaf);
1131                 }
1132
1133 next:
1134                 btrfs_release_path(root, path);
1135                 key.offset++;
1136         }
1137         ret = 0;
1138 out:
1139         btrfs_release_path(root, path);
1140         if (ret == 0) {
1141                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1142                 if (destoff + olen > inode->i_size)
1143                         btrfs_i_size_write(inode, destoff + olen);
1144                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1145                 ret = btrfs_update_inode(trans, root, inode);
1146         }
1147         btrfs_end_transaction(trans, root);
1148         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1149         if (ret)
1150                 vmtruncate(inode, 0);
1151 out_unlock:
1152         mutex_unlock(&src->i_mutex);
1153         mutex_unlock(&inode->i_mutex);
1154         vfree(buf);
1155         btrfs_free_path(path);
1156 out_fput:
1157         fput(src_file);
1158 out_drop_write:
1159         mnt_drop_write(file->f_path.mnt);
1160         return ret;
1161 }
1162
1163 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1164 {
1165         struct btrfs_ioctl_clone_range_args args;
1166
1167         if (copy_from_user(&args, argp, sizeof(args)))
1168                 return -EFAULT;
1169         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1170                                  args.src_length, args.dest_offset);
1171 }
1172
1173 /*
1174  * there are many ways the trans_start and trans_end ioctls can lead
1175  * to deadlocks.  They should only be used by applications that
1176  * basically own the machine, and have a very in depth understanding
1177  * of all the possible deadlocks and enospc problems.
1178  */
1179 static long btrfs_ioctl_trans_start(struct file *file)
1180 {
1181         struct inode *inode = fdentry(file)->d_inode;
1182         struct btrfs_root *root = BTRFS_I(inode)->root;
1183         struct btrfs_trans_handle *trans;
1184         int ret = 0;
1185
1186         if (!capable(CAP_SYS_ADMIN))
1187                 return -EPERM;
1188
1189         if (file->private_data) {
1190                 ret = -EINPROGRESS;
1191                 goto out;
1192         }
1193
1194         ret = mnt_want_write(file->f_path.mnt);
1195         if (ret)
1196                 goto out;
1197
1198         mutex_lock(&root->fs_info->trans_mutex);
1199         root->fs_info->open_ioctl_trans++;
1200         mutex_unlock(&root->fs_info->trans_mutex);
1201
1202         trans = btrfs_start_ioctl_transaction(root, 0);
1203         if (trans)
1204                 file->private_data = trans;
1205         else
1206                 ret = -ENOMEM;
1207         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1208 out:
1209         return ret;
1210 }
1211
1212 /*
1213  * there are many ways the trans_start and trans_end ioctls can lead
1214  * to deadlocks.  They should only be used by applications that
1215  * basically own the machine, and have a very in depth understanding
1216  * of all the possible deadlocks and enospc problems.
1217  */
1218 long btrfs_ioctl_trans_end(struct file *file)
1219 {
1220         struct inode *inode = fdentry(file)->d_inode;
1221         struct btrfs_root *root = BTRFS_I(inode)->root;
1222         struct btrfs_trans_handle *trans;
1223         int ret = 0;
1224
1225         trans = file->private_data;
1226         if (!trans) {
1227                 ret = -EINVAL;
1228                 goto out;
1229         }
1230         btrfs_end_transaction(trans, root);
1231         file->private_data = NULL;
1232
1233         mutex_lock(&root->fs_info->trans_mutex);
1234         root->fs_info->open_ioctl_trans--;
1235         mutex_unlock(&root->fs_info->trans_mutex);
1236
1237         mnt_drop_write(file->f_path.mnt);
1238
1239 out:
1240         return ret;
1241 }
1242
1243 long btrfs_ioctl(struct file *file, unsigned int
1244                 cmd, unsigned long arg)
1245 {
1246         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1247         void __user *argp = (void __user *)arg;
1248
1249         switch (cmd) {
1250         case FS_IOC_GETFLAGS:
1251                 return btrfs_ioctl_getflags(file, argp);
1252         case FS_IOC_SETFLAGS:
1253                 return btrfs_ioctl_setflags(file, argp);
1254         case FS_IOC_GETVERSION:
1255                 return btrfs_ioctl_getversion(file, argp);
1256         case BTRFS_IOC_SNAP_CREATE:
1257                 return btrfs_ioctl_snap_create(file, argp, 0);
1258         case BTRFS_IOC_SUBVOL_CREATE:
1259                 return btrfs_ioctl_snap_create(file, argp, 1);
1260         case BTRFS_IOC_DEFRAG:
1261                 return btrfs_ioctl_defrag(file);
1262         case BTRFS_IOC_RESIZE:
1263                 return btrfs_ioctl_resize(root, argp);
1264         case BTRFS_IOC_ADD_DEV:
1265                 return btrfs_ioctl_add_dev(root, argp);
1266         case BTRFS_IOC_RM_DEV:
1267                 return btrfs_ioctl_rm_dev(root, argp);
1268         case BTRFS_IOC_BALANCE:
1269                 return btrfs_balance(root->fs_info->dev_root);
1270         case BTRFS_IOC_CLONE:
1271                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1272         case BTRFS_IOC_CLONE_RANGE:
1273                 return btrfs_ioctl_clone_range(file, argp);
1274         case BTRFS_IOC_TRANS_START:
1275                 return btrfs_ioctl_trans_start(file);
1276         case BTRFS_IOC_TRANS_END:
1277                 return btrfs_ioctl_trans_end(file);
1278         case BTRFS_IOC_SYNC:
1279                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1280                 return 0;
1281         }
1282
1283         return -ENOTTY;
1284 }