Btrfs: add snapshot/subvolume destroy ioctl
[cascardo/linux.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;
234         struct inode *dir = dentry->d_parent->d_inode;
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                 return ret;
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         key.offset = (u64)-1;
308         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
309         BUG_ON(IS_ERR(new_root));
310
311         btrfs_record_root_in_trans(trans, new_root);
312
313         ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
314                                        BTRFS_I(dir)->block_group);
315         /*
316          * insert the directory item
317          */
318         ret = btrfs_set_inode_index(dir, &index);
319         BUG_ON(ret);
320
321         ret = btrfs_insert_dir_item(trans, root,
322                                     name, namelen, dir->i_ino, &key,
323                                     BTRFS_FT_DIR, index);
324         if (ret)
325                 goto fail;
326
327         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
328         ret = btrfs_update_inode(trans, root, dir);
329         BUG_ON(ret);
330
331         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
332                                  objectid, root->root_key.objectid,
333                                  dir->i_ino, index, name, namelen);
334
335         BUG_ON(ret);
336
337         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
338 fail:
339         nr = trans->blocks_used;
340         err = btrfs_commit_transaction(trans, root);
341         if (err && !ret)
342                 ret = err;
343         return ret;
344 }
345
346 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
347                            char *name, int namelen)
348 {
349         struct btrfs_pending_snapshot *pending_snapshot;
350         struct btrfs_trans_handle *trans;
351         int ret = 0;
352         int err;
353         unsigned long nr = 0;
354
355         if (!root->ref_cows)
356                 return -EINVAL;
357
358         ret = btrfs_check_metadata_free_space(root);
359         if (ret)
360                 goto fail_unlock;
361
362         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
363         if (!pending_snapshot) {
364                 ret = -ENOMEM;
365                 goto fail_unlock;
366         }
367         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
368         if (!pending_snapshot->name) {
369                 ret = -ENOMEM;
370                 kfree(pending_snapshot);
371                 goto fail_unlock;
372         }
373         memcpy(pending_snapshot->name, name, namelen);
374         pending_snapshot->name[namelen] = '\0';
375         pending_snapshot->dentry = dentry;
376         trans = btrfs_start_transaction(root, 1);
377         BUG_ON(!trans);
378         pending_snapshot->root = root;
379         list_add(&pending_snapshot->list,
380                  &trans->transaction->pending_snapshots);
381         err = btrfs_commit_transaction(trans, root);
382
383 fail_unlock:
384         btrfs_btree_balance_dirty(root, nr);
385         return ret;
386 }
387
388 /* copy of may_create in fs/namei.c() */
389 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
390 {
391         if (child->d_inode)
392                 return -EEXIST;
393         if (IS_DEADDIR(dir))
394                 return -ENOENT;
395         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
396 }
397
398 /*
399  * Create a new subvolume below @parent.  This is largely modeled after
400  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
401  * inside this filesystem so it's quite a bit simpler.
402  */
403 static noinline int btrfs_mksubvol(struct path *parent,
404                                    char *name, int namelen,
405                                    struct btrfs_root *snap_src)
406 {
407         struct inode *dir  = parent->dentry->d_inode;
408         struct dentry *dentry;
409         int error;
410
411         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
412
413         dentry = lookup_one_len(name, parent->dentry, namelen);
414         error = PTR_ERR(dentry);
415         if (IS_ERR(dentry))
416                 goto out_unlock;
417
418         error = -EEXIST;
419         if (dentry->d_inode)
420                 goto out_dput;
421
422         error = mnt_want_write(parent->mnt);
423         if (error)
424                 goto out_dput;
425
426         error = btrfs_may_create(dir, dentry);
427         if (error)
428                 goto out_drop_write;
429
430         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
431
432         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
433                 goto out_up_read;
434
435         if (snap_src) {
436                 error = create_snapshot(snap_src, dentry,
437                                         name, namelen);
438         } else {
439                 error = create_subvol(BTRFS_I(dir)->root, dentry,
440                                       name, namelen);
441         }
442         if (!error)
443                 fsnotify_mkdir(dir, dentry);
444 out_up_read:
445         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
446 out_drop_write:
447         mnt_drop_write(parent->mnt);
448 out_dput:
449         dput(dentry);
450 out_unlock:
451         mutex_unlock(&dir->i_mutex);
452         return error;
453 }
454
455 static int btrfs_defrag_file(struct file *file)
456 {
457         struct inode *inode = fdentry(file)->d_inode;
458         struct btrfs_root *root = BTRFS_I(inode)->root;
459         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
460         struct btrfs_ordered_extent *ordered;
461         struct page *page;
462         unsigned long last_index;
463         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
464         unsigned long total_read = 0;
465         u64 page_start;
466         u64 page_end;
467         unsigned long i;
468         int ret;
469
470         ret = btrfs_check_data_free_space(root, inode, inode->i_size);
471         if (ret)
472                 return -ENOSPC;
473
474         mutex_lock(&inode->i_mutex);
475         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
476         for (i = 0; i <= last_index; i++) {
477                 if (total_read % ra_pages == 0) {
478                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
479                                        min(last_index, i + ra_pages - 1));
480                 }
481                 total_read++;
482 again:
483                 page = grab_cache_page(inode->i_mapping, i);
484                 if (!page)
485                         goto out_unlock;
486                 if (!PageUptodate(page)) {
487                         btrfs_readpage(NULL, page);
488                         lock_page(page);
489                         if (!PageUptodate(page)) {
490                                 unlock_page(page);
491                                 page_cache_release(page);
492                                 goto out_unlock;
493                         }
494                 }
495
496                 wait_on_page_writeback(page);
497
498                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
499                 page_end = page_start + PAGE_CACHE_SIZE - 1;
500                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
501
502                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
503                 if (ordered) {
504                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
505                         unlock_page(page);
506                         page_cache_release(page);
507                         btrfs_start_ordered_extent(inode, ordered, 1);
508                         btrfs_put_ordered_extent(ordered);
509                         goto again;
510                 }
511                 set_page_extent_mapped(page);
512
513                 /*
514                  * this makes sure page_mkwrite is called on the
515                  * page if it is dirtied again later
516                  */
517                 clear_page_dirty_for_io(page);
518
519                 btrfs_set_extent_delalloc(inode, page_start, page_end);
520                 set_page_dirty(page);
521                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
522                 unlock_page(page);
523                 page_cache_release(page);
524                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
525         }
526
527 out_unlock:
528         mutex_unlock(&inode->i_mutex);
529         return 0;
530 }
531
532 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
533                                         void __user *arg)
534 {
535         u64 new_size;
536         u64 old_size;
537         u64 devid = 1;
538         struct btrfs_ioctl_vol_args *vol_args;
539         struct btrfs_trans_handle *trans;
540         struct btrfs_device *device = NULL;
541         char *sizestr;
542         char *devstr = NULL;
543         int ret = 0;
544         int namelen;
545         int mod = 0;
546
547         if (root->fs_info->sb->s_flags & MS_RDONLY)
548                 return -EROFS;
549
550         if (!capable(CAP_SYS_ADMIN))
551                 return -EPERM;
552
553         vol_args = memdup_user(arg, sizeof(*vol_args));
554         if (IS_ERR(vol_args))
555                 return PTR_ERR(vol_args);
556
557         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
558         namelen = strlen(vol_args->name);
559
560         mutex_lock(&root->fs_info->volume_mutex);
561         sizestr = vol_args->name;
562         devstr = strchr(sizestr, ':');
563         if (devstr) {
564                 char *end;
565                 sizestr = devstr + 1;
566                 *devstr = '\0';
567                 devstr = vol_args->name;
568                 devid = simple_strtoull(devstr, &end, 10);
569                 printk(KERN_INFO "resizing devid %llu\n",
570                        (unsigned long long)devid);
571         }
572         device = btrfs_find_device(root, devid, NULL, NULL);
573         if (!device) {
574                 printk(KERN_INFO "resizer unable to find device %llu\n",
575                        (unsigned long long)devid);
576                 ret = -EINVAL;
577                 goto out_unlock;
578         }
579         if (!strcmp(sizestr, "max"))
580                 new_size = device->bdev->bd_inode->i_size;
581         else {
582                 if (sizestr[0] == '-') {
583                         mod = -1;
584                         sizestr++;
585                 } else if (sizestr[0] == '+') {
586                         mod = 1;
587                         sizestr++;
588                 }
589                 new_size = btrfs_parse_size(sizestr);
590                 if (new_size == 0) {
591                         ret = -EINVAL;
592                         goto out_unlock;
593                 }
594         }
595
596         old_size = device->total_bytes;
597
598         if (mod < 0) {
599                 if (new_size > old_size) {
600                         ret = -EINVAL;
601                         goto out_unlock;
602                 }
603                 new_size = old_size - new_size;
604         } else if (mod > 0) {
605                 new_size = old_size + new_size;
606         }
607
608         if (new_size < 256 * 1024 * 1024) {
609                 ret = -EINVAL;
610                 goto out_unlock;
611         }
612         if (new_size > device->bdev->bd_inode->i_size) {
613                 ret = -EFBIG;
614                 goto out_unlock;
615         }
616
617         do_div(new_size, root->sectorsize);
618         new_size *= root->sectorsize;
619
620         printk(KERN_INFO "new size for %s is %llu\n",
621                 device->name, (unsigned long long)new_size);
622
623         if (new_size > old_size) {
624                 trans = btrfs_start_transaction(root, 1);
625                 ret = btrfs_grow_device(trans, device, new_size);
626                 btrfs_commit_transaction(trans, root);
627         } else {
628                 ret = btrfs_shrink_device(device, new_size);
629         }
630
631 out_unlock:
632         mutex_unlock(&root->fs_info->volume_mutex);
633         kfree(vol_args);
634         return ret;
635 }
636
637 static noinline int btrfs_ioctl_snap_create(struct file *file,
638                                             void __user *arg, int subvol)
639 {
640         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
641         struct btrfs_ioctl_vol_args *vol_args;
642         struct file *src_file;
643         int namelen;
644         int ret = 0;
645
646         if (root->fs_info->sb->s_flags & MS_RDONLY)
647                 return -EROFS;
648
649         vol_args = memdup_user(arg, sizeof(*vol_args));
650         if (IS_ERR(vol_args))
651                 return PTR_ERR(vol_args);
652
653         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
654         namelen = strlen(vol_args->name);
655         if (strchr(vol_args->name, '/')) {
656                 ret = -EINVAL;
657                 goto out;
658         }
659
660         if (subvol) {
661                 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
662                                      NULL);
663         } else {
664                 struct inode *src_inode;
665                 src_file = fget(vol_args->fd);
666                 if (!src_file) {
667                         ret = -EINVAL;
668                         goto out;
669                 }
670
671                 src_inode = src_file->f_path.dentry->d_inode;
672                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
673                         printk(KERN_INFO "btrfs: Snapshot src from "
674                                "another FS\n");
675                         ret = -EINVAL;
676                         fput(src_file);
677                         goto out;
678                 }
679                 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
680                                      BTRFS_I(src_inode)->root);
681                 fput(src_file);
682         }
683 out:
684         kfree(vol_args);
685         return ret;
686 }
687
688 /*
689  * helper to check if the subvolume references other subvolumes
690  */
691 static noinline int may_destroy_subvol(struct btrfs_root *root)
692 {
693         struct btrfs_path *path;
694         struct btrfs_key key;
695         int ret;
696
697         path = btrfs_alloc_path();
698         if (!path)
699                 return -ENOMEM;
700
701         key.objectid = root->root_key.objectid;
702         key.type = BTRFS_ROOT_REF_KEY;
703         key.offset = (u64)-1;
704
705         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
706                                 &key, path, 0, 0);
707         if (ret < 0)
708                 goto out;
709         BUG_ON(ret == 0);
710
711         ret = 0;
712         if (path->slots[0] > 0) {
713                 path->slots[0]--;
714                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
715                 if (key.objectid == root->root_key.objectid &&
716                     key.type == BTRFS_ROOT_REF_KEY)
717                         ret = -ENOTEMPTY;
718         }
719 out:
720         btrfs_free_path(path);
721         return ret;
722 }
723
724 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
725                                              void __user *arg)
726 {
727         struct dentry *parent = fdentry(file);
728         struct dentry *dentry;
729         struct inode *dir = parent->d_inode;
730         struct inode *inode;
731         struct btrfs_root *root = BTRFS_I(dir)->root;
732         struct btrfs_root *dest = NULL;
733         struct btrfs_ioctl_vol_args *vol_args;
734         struct btrfs_trans_handle *trans;
735         int namelen;
736         int ret;
737         int err = 0;
738
739         if (!capable(CAP_SYS_ADMIN))
740                 return -EPERM;
741
742         vol_args = memdup_user(arg, sizeof(*vol_args));
743         if (IS_ERR(vol_args))
744                 return PTR_ERR(vol_args);
745
746         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
747         namelen = strlen(vol_args->name);
748         if (strchr(vol_args->name, '/') ||
749             strncmp(vol_args->name, "..", namelen) == 0) {
750                 err = -EINVAL;
751                 goto out;
752         }
753
754         err = mnt_want_write(file->f_path.mnt);
755         if (err)
756                 goto out;
757
758         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
759         dentry = lookup_one_len(vol_args->name, parent, namelen);
760         if (IS_ERR(dentry)) {
761                 err = PTR_ERR(dentry);
762                 goto out_unlock_dir;
763         }
764
765         if (!dentry->d_inode) {
766                 err = -ENOENT;
767                 goto out_dput;
768         }
769
770         inode = dentry->d_inode;
771         if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
772                 err = -EINVAL;
773                 goto out_dput;
774         }
775
776         dest = BTRFS_I(inode)->root;
777
778         mutex_lock(&inode->i_mutex);
779         err = d_invalidate(dentry);
780         if (err)
781                 goto out_unlock;
782
783         down_write(&root->fs_info->subvol_sem);
784
785         err = may_destroy_subvol(dest);
786         if (err)
787                 goto out_up_write;
788
789         trans = btrfs_start_transaction(root, 1);
790         ret = btrfs_unlink_subvol(trans, root, dir,
791                                 dest->root_key.objectid,
792                                 dentry->d_name.name,
793                                 dentry->d_name.len);
794         BUG_ON(ret);
795
796         btrfs_record_root_in_trans(trans, dest);
797
798         memset(&dest->root_item.drop_progress, 0,
799                 sizeof(dest->root_item.drop_progress));
800         dest->root_item.drop_level = 0;
801         btrfs_set_root_refs(&dest->root_item, 0);
802
803         ret = btrfs_insert_orphan_item(trans,
804                                 root->fs_info->tree_root,
805                                 dest->root_key.objectid);
806         BUG_ON(ret);
807
808         ret = btrfs_commit_transaction(trans, root);
809         BUG_ON(ret);
810         inode->i_flags |= S_DEAD;
811 out_up_write:
812         up_write(&root->fs_info->subvol_sem);
813 out_unlock:
814         mutex_unlock(&inode->i_mutex);
815         if (!err) {
816                 btrfs_invalidate_inodes(dest);
817                 d_delete(dentry);
818         }
819 out_dput:
820         dput(dentry);
821 out_unlock_dir:
822         mutex_unlock(&dir->i_mutex);
823         mnt_drop_write(file->f_path.mnt);
824 out:
825         kfree(vol_args);
826         return err;
827 }
828
829 static int btrfs_ioctl_defrag(struct file *file)
830 {
831         struct inode *inode = fdentry(file)->d_inode;
832         struct btrfs_root *root = BTRFS_I(inode)->root;
833         int ret;
834
835         ret = mnt_want_write(file->f_path.mnt);
836         if (ret)
837                 return ret;
838
839         switch (inode->i_mode & S_IFMT) {
840         case S_IFDIR:
841                 if (!capable(CAP_SYS_ADMIN)) {
842                         ret = -EPERM;
843                         goto out;
844                 }
845                 btrfs_defrag_root(root, 0);
846                 btrfs_defrag_root(root->fs_info->extent_root, 0);
847                 break;
848         case S_IFREG:
849                 if (!(file->f_mode & FMODE_WRITE)) {
850                         ret = -EINVAL;
851                         goto out;
852                 }
853                 btrfs_defrag_file(file);
854                 break;
855         }
856 out:
857         mnt_drop_write(file->f_path.mnt);
858         return ret;
859 }
860
861 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
862 {
863         struct btrfs_ioctl_vol_args *vol_args;
864         int ret;
865
866         if (!capable(CAP_SYS_ADMIN))
867                 return -EPERM;
868
869         vol_args = memdup_user(arg, sizeof(*vol_args));
870         if (IS_ERR(vol_args))
871                 return PTR_ERR(vol_args);
872
873         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
874         ret = btrfs_init_new_device(root, vol_args->name);
875
876         kfree(vol_args);
877         return ret;
878 }
879
880 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
881 {
882         struct btrfs_ioctl_vol_args *vol_args;
883         int ret;
884
885         if (!capable(CAP_SYS_ADMIN))
886                 return -EPERM;
887
888         if (root->fs_info->sb->s_flags & MS_RDONLY)
889                 return -EROFS;
890
891         vol_args = memdup_user(arg, sizeof(*vol_args));
892         if (IS_ERR(vol_args))
893                 return PTR_ERR(vol_args);
894
895         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
896         ret = btrfs_rm_device(root, vol_args->name);
897
898         kfree(vol_args);
899         return ret;
900 }
901
902 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
903                                        u64 off, u64 olen, u64 destoff)
904 {
905         struct inode *inode = fdentry(file)->d_inode;
906         struct btrfs_root *root = BTRFS_I(inode)->root;
907         struct file *src_file;
908         struct inode *src;
909         struct btrfs_trans_handle *trans;
910         struct btrfs_path *path;
911         struct extent_buffer *leaf;
912         char *buf;
913         struct btrfs_key key;
914         u32 nritems;
915         int slot;
916         int ret;
917         u64 len = olen;
918         u64 bs = root->fs_info->sb->s_blocksize;
919         u64 hint_byte;
920
921         /*
922          * TODO:
923          * - split compressed inline extents.  annoying: we need to
924          *   decompress into destination's address_space (the file offset
925          *   may change, so source mapping won't do), then recompress (or
926          *   otherwise reinsert) a subrange.
927          * - allow ranges within the same file to be cloned (provided
928          *   they don't overlap)?
929          */
930
931         /* the destination must be opened for writing */
932         if (!(file->f_mode & FMODE_WRITE))
933                 return -EINVAL;
934
935         ret = mnt_want_write(file->f_path.mnt);
936         if (ret)
937                 return ret;
938
939         src_file = fget(srcfd);
940         if (!src_file) {
941                 ret = -EBADF;
942                 goto out_drop_write;
943         }
944         src = src_file->f_dentry->d_inode;
945
946         ret = -EINVAL;
947         if (src == inode)
948                 goto out_fput;
949
950         ret = -EISDIR;
951         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
952                 goto out_fput;
953
954         ret = -EXDEV;
955         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
956                 goto out_fput;
957
958         ret = -ENOMEM;
959         buf = vmalloc(btrfs_level_size(root, 0));
960         if (!buf)
961                 goto out_fput;
962
963         path = btrfs_alloc_path();
964         if (!path) {
965                 vfree(buf);
966                 goto out_fput;
967         }
968         path->reada = 2;
969
970         if (inode < src) {
971                 mutex_lock(&inode->i_mutex);
972                 mutex_lock(&src->i_mutex);
973         } else {
974                 mutex_lock(&src->i_mutex);
975                 mutex_lock(&inode->i_mutex);
976         }
977
978         /* determine range to clone */
979         ret = -EINVAL;
980         if (off >= src->i_size || off + len > src->i_size)
981                 goto out_unlock;
982         if (len == 0)
983                 olen = len = src->i_size - off;
984         /* if we extend to eof, continue to block boundary */
985         if (off + len == src->i_size)
986                 len = ((src->i_size + bs-1) & ~(bs-1))
987                         - off;
988
989         /* verify the end result is block aligned */
990         if ((off & (bs-1)) ||
991             ((off + len) & (bs-1)))
992                 goto out_unlock;
993
994         /* do any pending delalloc/csum calc on src, one way or
995            another, and lock file content */
996         while (1) {
997                 struct btrfs_ordered_extent *ordered;
998                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
999                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
1000                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
1001                         break;
1002                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1003                 if (ordered)
1004                         btrfs_put_ordered_extent(ordered);
1005                 btrfs_wait_ordered_range(src, off, off+len);
1006         }
1007
1008         trans = btrfs_start_transaction(root, 1);
1009         BUG_ON(!trans);
1010
1011         /* punch hole in destination first */
1012         btrfs_drop_extents(trans, root, inode, off, off + len,
1013                            off + len, 0, &hint_byte, 1);
1014
1015         /* clone data */
1016         key.objectid = src->i_ino;
1017         key.type = BTRFS_EXTENT_DATA_KEY;
1018         key.offset = 0;
1019
1020         while (1) {
1021                 /*
1022                  * note the key will change type as we walk through the
1023                  * tree.
1024                  */
1025                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
1026                 if (ret < 0)
1027                         goto out;
1028
1029                 nritems = btrfs_header_nritems(path->nodes[0]);
1030                 if (path->slots[0] >= nritems) {
1031                         ret = btrfs_next_leaf(root, path);
1032                         if (ret < 0)
1033                                 goto out;
1034                         if (ret > 0)
1035                                 break;
1036                         nritems = btrfs_header_nritems(path->nodes[0]);
1037                 }
1038                 leaf = path->nodes[0];
1039                 slot = path->slots[0];
1040
1041                 btrfs_item_key_to_cpu(leaf, &key, slot);
1042                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1043                     key.objectid != src->i_ino)
1044                         break;
1045
1046                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1047                         struct btrfs_file_extent_item *extent;
1048                         int type;
1049                         u32 size;
1050                         struct btrfs_key new_key;
1051                         u64 disko = 0, diskl = 0;
1052                         u64 datao = 0, datal = 0;
1053                         u8 comp;
1054
1055                         size = btrfs_item_size_nr(leaf, slot);
1056                         read_extent_buffer(leaf, buf,
1057                                            btrfs_item_ptr_offset(leaf, slot),
1058                                            size);
1059
1060                         extent = btrfs_item_ptr(leaf, slot,
1061                                                 struct btrfs_file_extent_item);
1062                         comp = btrfs_file_extent_compression(leaf, extent);
1063                         type = btrfs_file_extent_type(leaf, extent);
1064                         if (type == BTRFS_FILE_EXTENT_REG ||
1065                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1066                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1067                                                                       extent);
1068                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1069                                                                  extent);
1070                                 datao = btrfs_file_extent_offset(leaf, extent);
1071                                 datal = btrfs_file_extent_num_bytes(leaf,
1072                                                                     extent);
1073                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1074                                 /* take upper bound, may be compressed */
1075                                 datal = btrfs_file_extent_ram_bytes(leaf,
1076                                                                     extent);
1077                         }
1078                         btrfs_release_path(root, path);
1079
1080                         if (key.offset + datal < off ||
1081                             key.offset >= off+len)
1082                                 goto next;
1083
1084                         memcpy(&new_key, &key, sizeof(new_key));
1085                         new_key.objectid = inode->i_ino;
1086                         new_key.offset = key.offset + destoff - off;
1087
1088                         if (type == BTRFS_FILE_EXTENT_REG ||
1089                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1090                                 ret = btrfs_insert_empty_item(trans, root, path,
1091                                                               &new_key, size);
1092                                 if (ret)
1093                                         goto out;
1094
1095                                 leaf = path->nodes[0];
1096                                 slot = path->slots[0];
1097                                 write_extent_buffer(leaf, buf,
1098                                             btrfs_item_ptr_offset(leaf, slot),
1099                                             size);
1100
1101                                 extent = btrfs_item_ptr(leaf, slot,
1102                                                 struct btrfs_file_extent_item);
1103
1104                                 if (off > key.offset) {
1105                                         datao += off - key.offset;
1106                                         datal -= off - key.offset;
1107                                 }
1108                                 if (key.offset + datao + datal + key.offset >
1109                                     off + len)
1110                                         datal = off + len - key.offset - datao;
1111                                 /* disko == 0 means it's a hole */
1112                                 if (!disko)
1113                                         datao = 0;
1114
1115                                 btrfs_set_file_extent_offset(leaf, extent,
1116                                                              datao);
1117                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1118                                                                 datal);
1119                                 if (disko) {
1120                                         inode_add_bytes(inode, datal);
1121                                         ret = btrfs_inc_extent_ref(trans, root,
1122                                                         disko, diskl, 0,
1123                                                         root->root_key.objectid,
1124                                                         inode->i_ino,
1125                                                         new_key.offset - datao);
1126                                         BUG_ON(ret);
1127                                 }
1128                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1129                                 u64 skip = 0;
1130                                 u64 trim = 0;
1131                                 if (off > key.offset) {
1132                                         skip = off - key.offset;
1133                                         new_key.offset += skip;
1134                                 }
1135
1136                                 if (key.offset + datal > off+len)
1137                                         trim = key.offset + datal - (off+len);
1138
1139                                 if (comp && (skip || trim)) {
1140                                         ret = -EINVAL;
1141                                         goto out;
1142                                 }
1143                                 size -= skip + trim;
1144                                 datal -= skip + trim;
1145                                 ret = btrfs_insert_empty_item(trans, root, path,
1146                                                               &new_key, size);
1147                                 if (ret)
1148                                         goto out;
1149
1150                                 if (skip) {
1151                                         u32 start =
1152                                           btrfs_file_extent_calc_inline_size(0);
1153                                         memmove(buf+start, buf+start+skip,
1154                                                 datal);
1155                                 }
1156
1157                                 leaf = path->nodes[0];
1158                                 slot = path->slots[0];
1159                                 write_extent_buffer(leaf, buf,
1160                                             btrfs_item_ptr_offset(leaf, slot),
1161                                             size);
1162                                 inode_add_bytes(inode, datal);
1163                         }
1164
1165                         btrfs_mark_buffer_dirty(leaf);
1166                 }
1167
1168 next:
1169                 btrfs_release_path(root, path);
1170                 key.offset++;
1171         }
1172         ret = 0;
1173 out:
1174         btrfs_release_path(root, path);
1175         if (ret == 0) {
1176                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1177                 if (destoff + olen > inode->i_size)
1178                         btrfs_i_size_write(inode, destoff + olen);
1179                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1180                 ret = btrfs_update_inode(trans, root, inode);
1181         }
1182         btrfs_end_transaction(trans, root);
1183         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1184         if (ret)
1185                 vmtruncate(inode, 0);
1186 out_unlock:
1187         mutex_unlock(&src->i_mutex);
1188         mutex_unlock(&inode->i_mutex);
1189         vfree(buf);
1190         btrfs_free_path(path);
1191 out_fput:
1192         fput(src_file);
1193 out_drop_write:
1194         mnt_drop_write(file->f_path.mnt);
1195         return ret;
1196 }
1197
1198 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1199 {
1200         struct btrfs_ioctl_clone_range_args args;
1201
1202         if (copy_from_user(&args, argp, sizeof(args)))
1203                 return -EFAULT;
1204         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1205                                  args.src_length, args.dest_offset);
1206 }
1207
1208 /*
1209  * there are many ways the trans_start and trans_end ioctls can lead
1210  * to deadlocks.  They should only be used by applications that
1211  * basically own the machine, and have a very in depth understanding
1212  * of all the possible deadlocks and enospc problems.
1213  */
1214 static long btrfs_ioctl_trans_start(struct file *file)
1215 {
1216         struct inode *inode = fdentry(file)->d_inode;
1217         struct btrfs_root *root = BTRFS_I(inode)->root;
1218         struct btrfs_trans_handle *trans;
1219         int ret = 0;
1220
1221         if (!capable(CAP_SYS_ADMIN))
1222                 return -EPERM;
1223
1224         if (file->private_data) {
1225                 ret = -EINPROGRESS;
1226                 goto out;
1227         }
1228
1229         ret = mnt_want_write(file->f_path.mnt);
1230         if (ret)
1231                 goto out;
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         trans = btrfs_start_ioctl_transaction(root, 0);
1238         if (trans)
1239                 file->private_data = trans;
1240         else
1241                 ret = -ENOMEM;
1242         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1243 out:
1244         return ret;
1245 }
1246
1247 /*
1248  * there are many ways the trans_start and trans_end ioctls can lead
1249  * to deadlocks.  They should only be used by applications that
1250  * basically own the machine, and have a very in depth understanding
1251  * of all the possible deadlocks and enospc problems.
1252  */
1253 long btrfs_ioctl_trans_end(struct file *file)
1254 {
1255         struct inode *inode = fdentry(file)->d_inode;
1256         struct btrfs_root *root = BTRFS_I(inode)->root;
1257         struct btrfs_trans_handle *trans;
1258         int ret = 0;
1259
1260         trans = file->private_data;
1261         if (!trans) {
1262                 ret = -EINVAL;
1263                 goto out;
1264         }
1265         btrfs_end_transaction(trans, root);
1266         file->private_data = NULL;
1267
1268         mutex_lock(&root->fs_info->trans_mutex);
1269         root->fs_info->open_ioctl_trans--;
1270         mutex_unlock(&root->fs_info->trans_mutex);
1271
1272         mnt_drop_write(file->f_path.mnt);
1273
1274 out:
1275         return ret;
1276 }
1277
1278 long btrfs_ioctl(struct file *file, unsigned int
1279                 cmd, unsigned long arg)
1280 {
1281         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1282         void __user *argp = (void __user *)arg;
1283
1284         switch (cmd) {
1285         case FS_IOC_GETFLAGS:
1286                 return btrfs_ioctl_getflags(file, argp);
1287         case FS_IOC_SETFLAGS:
1288                 return btrfs_ioctl_setflags(file, argp);
1289         case FS_IOC_GETVERSION:
1290                 return btrfs_ioctl_getversion(file, argp);
1291         case BTRFS_IOC_SNAP_CREATE:
1292                 return btrfs_ioctl_snap_create(file, argp, 0);
1293         case BTRFS_IOC_SUBVOL_CREATE:
1294                 return btrfs_ioctl_snap_create(file, argp, 1);
1295         case BTRFS_IOC_SNAP_DESTROY:
1296                 return btrfs_ioctl_snap_destroy(file, argp);
1297         case BTRFS_IOC_DEFRAG:
1298                 return btrfs_ioctl_defrag(file);
1299         case BTRFS_IOC_RESIZE:
1300                 return btrfs_ioctl_resize(root, argp);
1301         case BTRFS_IOC_ADD_DEV:
1302                 return btrfs_ioctl_add_dev(root, argp);
1303         case BTRFS_IOC_RM_DEV:
1304                 return btrfs_ioctl_rm_dev(root, argp);
1305         case BTRFS_IOC_BALANCE:
1306                 return btrfs_balance(root->fs_info->dev_root);
1307         case BTRFS_IOC_CLONE:
1308                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1309         case BTRFS_IOC_CLONE_RANGE:
1310                 return btrfs_ioctl_clone_range(file, argp);
1311         case BTRFS_IOC_TRANS_START:
1312                 return btrfs_ioctl_trans_start(file);
1313         case BTRFS_IOC_TRANS_END:
1314                 return btrfs_ioctl_trans_end(file);
1315         case BTRFS_IOC_SYNC:
1316                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1317                 return 0;
1318         }
1319
1320         return -ENOTTY;
1321 }