Btrfs: fix error cases for ioctl transactions
[linux-2.6-block.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         /*
243          * 1 - inode item
244          * 2 - refs
245          * 1 - root item
246          * 2 - dir items
247          */
248         ret = btrfs_reserve_metadata_space(root, 6);
249         if (ret)
250                 return ret;
251
252         trans = btrfs_start_transaction(root, 1);
253         BUG_ON(!trans);
254
255         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
256                                        0, &objectid);
257         if (ret)
258                 goto fail;
259
260         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
261                                       0, objectid, NULL, 0, 0, 0);
262         if (IS_ERR(leaf)) {
263                 ret = PTR_ERR(leaf);
264                 goto fail;
265         }
266
267         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
268         btrfs_set_header_bytenr(leaf, leaf->start);
269         btrfs_set_header_generation(leaf, trans->transid);
270         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
271         btrfs_set_header_owner(leaf, objectid);
272
273         write_extent_buffer(leaf, root->fs_info->fsid,
274                             (unsigned long)btrfs_header_fsid(leaf),
275                             BTRFS_FSID_SIZE);
276         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
277                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
278                             BTRFS_UUID_SIZE);
279         btrfs_mark_buffer_dirty(leaf);
280
281         inode_item = &root_item.inode;
282         memset(inode_item, 0, sizeof(*inode_item));
283         inode_item->generation = cpu_to_le64(1);
284         inode_item->size = cpu_to_le64(3);
285         inode_item->nlink = cpu_to_le32(1);
286         inode_item->nbytes = cpu_to_le64(root->leafsize);
287         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
288
289         btrfs_set_root_bytenr(&root_item, leaf->start);
290         btrfs_set_root_generation(&root_item, trans->transid);
291         btrfs_set_root_level(&root_item, 0);
292         btrfs_set_root_refs(&root_item, 1);
293         btrfs_set_root_used(&root_item, 0);
294         btrfs_set_root_last_snapshot(&root_item, 0);
295
296         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
297         root_item.drop_level = 0;
298
299         btrfs_tree_unlock(leaf);
300         free_extent_buffer(leaf);
301         leaf = NULL;
302
303         btrfs_set_root_dirid(&root_item, new_dirid);
304
305         key.objectid = objectid;
306         key.offset = 0;
307         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
308         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
309                                 &root_item);
310         if (ret)
311                 goto fail;
312
313         key.offset = (u64)-1;
314         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
315         BUG_ON(IS_ERR(new_root));
316
317         btrfs_record_root_in_trans(trans, new_root);
318
319         ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
320                                        BTRFS_I(dir)->block_group);
321         /*
322          * insert the directory item
323          */
324         ret = btrfs_set_inode_index(dir, &index);
325         BUG_ON(ret);
326
327         ret = btrfs_insert_dir_item(trans, root,
328                                     name, namelen, dir->i_ino, &key,
329                                     BTRFS_FT_DIR, index);
330         if (ret)
331                 goto fail;
332
333         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
334         ret = btrfs_update_inode(trans, root, dir);
335         BUG_ON(ret);
336
337         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
338                                  objectid, root->root_key.objectid,
339                                  dir->i_ino, index, name, namelen);
340
341         BUG_ON(ret);
342
343         d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
344 fail:
345         nr = trans->blocks_used;
346         err = btrfs_commit_transaction(trans, root);
347         if (err && !ret)
348                 ret = err;
349
350         btrfs_unreserve_metadata_space(root, 6);
351         btrfs_btree_balance_dirty(root, nr);
352         return ret;
353 }
354
355 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
356                            char *name, int namelen)
357 {
358         struct btrfs_pending_snapshot *pending_snapshot;
359         struct btrfs_trans_handle *trans;
360         int ret = 0;
361         int err;
362         unsigned long nr = 0;
363
364         if (!root->ref_cows)
365                 return -EINVAL;
366
367         /*
368          * 1 - inode item
369          * 2 - refs
370          * 1 - root item
371          * 2 - dir items
372          */
373         ret = btrfs_reserve_metadata_space(root, 6);
374         if (ret)
375                 goto fail_unlock;
376
377         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
378         if (!pending_snapshot) {
379                 ret = -ENOMEM;
380                 btrfs_unreserve_metadata_space(root, 6);
381                 goto fail_unlock;
382         }
383         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
384         if (!pending_snapshot->name) {
385                 ret = -ENOMEM;
386                 kfree(pending_snapshot);
387                 btrfs_unreserve_metadata_space(root, 6);
388                 goto fail_unlock;
389         }
390         memcpy(pending_snapshot->name, name, namelen);
391         pending_snapshot->name[namelen] = '\0';
392         pending_snapshot->dentry = dentry;
393         trans = btrfs_start_transaction(root, 1);
394         BUG_ON(!trans);
395         pending_snapshot->root = root;
396         list_add(&pending_snapshot->list,
397                  &trans->transaction->pending_snapshots);
398         err = btrfs_commit_transaction(trans, root);
399
400 fail_unlock:
401         btrfs_btree_balance_dirty(root, nr);
402         return ret;
403 }
404
405 /* copy of may_create in fs/namei.c() */
406 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
407 {
408         if (child->d_inode)
409                 return -EEXIST;
410         if (IS_DEADDIR(dir))
411                 return -ENOENT;
412         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
413 }
414
415 /*
416  * Create a new subvolume below @parent.  This is largely modeled after
417  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
418  * inside this filesystem so it's quite a bit simpler.
419  */
420 static noinline int btrfs_mksubvol(struct path *parent,
421                                    char *name, int namelen,
422                                    struct btrfs_root *snap_src)
423 {
424         struct inode *dir  = parent->dentry->d_inode;
425         struct dentry *dentry;
426         int error;
427
428         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
429
430         dentry = lookup_one_len(name, parent->dentry, namelen);
431         error = PTR_ERR(dentry);
432         if (IS_ERR(dentry))
433                 goto out_unlock;
434
435         error = -EEXIST;
436         if (dentry->d_inode)
437                 goto out_dput;
438
439         error = mnt_want_write(parent->mnt);
440         if (error)
441                 goto out_dput;
442
443         error = btrfs_may_create(dir, dentry);
444         if (error)
445                 goto out_drop_write;
446
447         down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
448
449         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
450                 goto out_up_read;
451
452         if (snap_src) {
453                 error = create_snapshot(snap_src, dentry,
454                                         name, namelen);
455         } else {
456                 error = create_subvol(BTRFS_I(dir)->root, dentry,
457                                       name, namelen);
458         }
459         if (!error)
460                 fsnotify_mkdir(dir, dentry);
461 out_up_read:
462         up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
463 out_drop_write:
464         mnt_drop_write(parent->mnt);
465 out_dput:
466         dput(dentry);
467 out_unlock:
468         mutex_unlock(&dir->i_mutex);
469         return error;
470 }
471
472 static int btrfs_defrag_file(struct file *file)
473 {
474         struct inode *inode = fdentry(file)->d_inode;
475         struct btrfs_root *root = BTRFS_I(inode)->root;
476         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
477         struct btrfs_ordered_extent *ordered;
478         struct page *page;
479         unsigned long last_index;
480         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
481         unsigned long total_read = 0;
482         u64 page_start;
483         u64 page_end;
484         unsigned long i;
485         int ret;
486
487         ret = btrfs_check_data_free_space(root, inode, inode->i_size);
488         if (ret)
489                 return -ENOSPC;
490
491         mutex_lock(&inode->i_mutex);
492         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
493         for (i = 0; i <= last_index; i++) {
494                 if (total_read % ra_pages == 0) {
495                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
496                                        min(last_index, i + ra_pages - 1));
497                 }
498                 total_read++;
499 again:
500                 page = grab_cache_page(inode->i_mapping, i);
501                 if (!page)
502                         goto out_unlock;
503                 if (!PageUptodate(page)) {
504                         btrfs_readpage(NULL, page);
505                         lock_page(page);
506                         if (!PageUptodate(page)) {
507                                 unlock_page(page);
508                                 page_cache_release(page);
509                                 goto out_unlock;
510                         }
511                 }
512
513                 wait_on_page_writeback(page);
514
515                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
516                 page_end = page_start + PAGE_CACHE_SIZE - 1;
517                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
518
519                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
520                 if (ordered) {
521                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
522                         unlock_page(page);
523                         page_cache_release(page);
524                         btrfs_start_ordered_extent(inode, ordered, 1);
525                         btrfs_put_ordered_extent(ordered);
526                         goto again;
527                 }
528                 set_page_extent_mapped(page);
529
530                 /*
531                  * this makes sure page_mkwrite is called on the
532                  * page if it is dirtied again later
533                  */
534                 clear_page_dirty_for_io(page);
535
536                 btrfs_set_extent_delalloc(inode, page_start, page_end);
537                 set_page_dirty(page);
538                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
539                 unlock_page(page);
540                 page_cache_release(page);
541                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
542         }
543
544 out_unlock:
545         mutex_unlock(&inode->i_mutex);
546         return 0;
547 }
548
549 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
550                                         void __user *arg)
551 {
552         u64 new_size;
553         u64 old_size;
554         u64 devid = 1;
555         struct btrfs_ioctl_vol_args *vol_args;
556         struct btrfs_trans_handle *trans;
557         struct btrfs_device *device = NULL;
558         char *sizestr;
559         char *devstr = NULL;
560         int ret = 0;
561         int namelen;
562         int mod = 0;
563
564         if (root->fs_info->sb->s_flags & MS_RDONLY)
565                 return -EROFS;
566
567         if (!capable(CAP_SYS_ADMIN))
568                 return -EPERM;
569
570         vol_args = memdup_user(arg, sizeof(*vol_args));
571         if (IS_ERR(vol_args))
572                 return PTR_ERR(vol_args);
573
574         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
575         namelen = strlen(vol_args->name);
576
577         mutex_lock(&root->fs_info->volume_mutex);
578         sizestr = vol_args->name;
579         devstr = strchr(sizestr, ':');
580         if (devstr) {
581                 char *end;
582                 sizestr = devstr + 1;
583                 *devstr = '\0';
584                 devstr = vol_args->name;
585                 devid = simple_strtoull(devstr, &end, 10);
586                 printk(KERN_INFO "resizing devid %llu\n",
587                        (unsigned long long)devid);
588         }
589         device = btrfs_find_device(root, devid, NULL, NULL);
590         if (!device) {
591                 printk(KERN_INFO "resizer unable to find device %llu\n",
592                        (unsigned long long)devid);
593                 ret = -EINVAL;
594                 goto out_unlock;
595         }
596         if (!strcmp(sizestr, "max"))
597                 new_size = device->bdev->bd_inode->i_size;
598         else {
599                 if (sizestr[0] == '-') {
600                         mod = -1;
601                         sizestr++;
602                 } else if (sizestr[0] == '+') {
603                         mod = 1;
604                         sizestr++;
605                 }
606                 new_size = btrfs_parse_size(sizestr);
607                 if (new_size == 0) {
608                         ret = -EINVAL;
609                         goto out_unlock;
610                 }
611         }
612
613         old_size = device->total_bytes;
614
615         if (mod < 0) {
616                 if (new_size > old_size) {
617                         ret = -EINVAL;
618                         goto out_unlock;
619                 }
620                 new_size = old_size - new_size;
621         } else if (mod > 0) {
622                 new_size = old_size + new_size;
623         }
624
625         if (new_size < 256 * 1024 * 1024) {
626                 ret = -EINVAL;
627                 goto out_unlock;
628         }
629         if (new_size > device->bdev->bd_inode->i_size) {
630                 ret = -EFBIG;
631                 goto out_unlock;
632         }
633
634         do_div(new_size, root->sectorsize);
635         new_size *= root->sectorsize;
636
637         printk(KERN_INFO "new size for %s is %llu\n",
638                 device->name, (unsigned long long)new_size);
639
640         if (new_size > old_size) {
641                 trans = btrfs_start_transaction(root, 1);
642                 ret = btrfs_grow_device(trans, device, new_size);
643                 btrfs_commit_transaction(trans, root);
644         } else {
645                 ret = btrfs_shrink_device(device, new_size);
646         }
647
648 out_unlock:
649         mutex_unlock(&root->fs_info->volume_mutex);
650         kfree(vol_args);
651         return ret;
652 }
653
654 static noinline int btrfs_ioctl_snap_create(struct file *file,
655                                             void __user *arg, int subvol)
656 {
657         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
658         struct btrfs_ioctl_vol_args *vol_args;
659         struct file *src_file;
660         int namelen;
661         int ret = 0;
662
663         if (root->fs_info->sb->s_flags & MS_RDONLY)
664                 return -EROFS;
665
666         vol_args = memdup_user(arg, sizeof(*vol_args));
667         if (IS_ERR(vol_args))
668                 return PTR_ERR(vol_args);
669
670         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
671         namelen = strlen(vol_args->name);
672         if (strchr(vol_args->name, '/')) {
673                 ret = -EINVAL;
674                 goto out;
675         }
676
677         if (subvol) {
678                 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
679                                      NULL);
680         } else {
681                 struct inode *src_inode;
682                 src_file = fget(vol_args->fd);
683                 if (!src_file) {
684                         ret = -EINVAL;
685                         goto out;
686                 }
687
688                 src_inode = src_file->f_path.dentry->d_inode;
689                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
690                         printk(KERN_INFO "btrfs: Snapshot src from "
691                                "another FS\n");
692                         ret = -EINVAL;
693                         fput(src_file);
694                         goto out;
695                 }
696                 ret = btrfs_mksubvol(&file->f_path, vol_args->name, namelen,
697                                      BTRFS_I(src_inode)->root);
698                 fput(src_file);
699         }
700 out:
701         kfree(vol_args);
702         return ret;
703 }
704
705 /*
706  * helper to check if the subvolume references other subvolumes
707  */
708 static noinline int may_destroy_subvol(struct btrfs_root *root)
709 {
710         struct btrfs_path *path;
711         struct btrfs_key key;
712         int ret;
713
714         path = btrfs_alloc_path();
715         if (!path)
716                 return -ENOMEM;
717
718         key.objectid = root->root_key.objectid;
719         key.type = BTRFS_ROOT_REF_KEY;
720         key.offset = (u64)-1;
721
722         ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
723                                 &key, path, 0, 0);
724         if (ret < 0)
725                 goto out;
726         BUG_ON(ret == 0);
727
728         ret = 0;
729         if (path->slots[0] > 0) {
730                 path->slots[0]--;
731                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
732                 if (key.objectid == root->root_key.objectid &&
733                     key.type == BTRFS_ROOT_REF_KEY)
734                         ret = -ENOTEMPTY;
735         }
736 out:
737         btrfs_free_path(path);
738         return ret;
739 }
740
741 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
742                                              void __user *arg)
743 {
744         struct dentry *parent = fdentry(file);
745         struct dentry *dentry;
746         struct inode *dir = parent->d_inode;
747         struct inode *inode;
748         struct btrfs_root *root = BTRFS_I(dir)->root;
749         struct btrfs_root *dest = NULL;
750         struct btrfs_ioctl_vol_args *vol_args;
751         struct btrfs_trans_handle *trans;
752         int namelen;
753         int ret;
754         int err = 0;
755
756         if (!capable(CAP_SYS_ADMIN))
757                 return -EPERM;
758
759         vol_args = memdup_user(arg, sizeof(*vol_args));
760         if (IS_ERR(vol_args))
761                 return PTR_ERR(vol_args);
762
763         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
764         namelen = strlen(vol_args->name);
765         if (strchr(vol_args->name, '/') ||
766             strncmp(vol_args->name, "..", namelen) == 0) {
767                 err = -EINVAL;
768                 goto out;
769         }
770
771         err = mnt_want_write(file->f_path.mnt);
772         if (err)
773                 goto out;
774
775         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
776         dentry = lookup_one_len(vol_args->name, parent, namelen);
777         if (IS_ERR(dentry)) {
778                 err = PTR_ERR(dentry);
779                 goto out_unlock_dir;
780         }
781
782         if (!dentry->d_inode) {
783                 err = -ENOENT;
784                 goto out_dput;
785         }
786
787         inode = dentry->d_inode;
788         if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
789                 err = -EINVAL;
790                 goto out_dput;
791         }
792
793         dest = BTRFS_I(inode)->root;
794
795         mutex_lock(&inode->i_mutex);
796         err = d_invalidate(dentry);
797         if (err)
798                 goto out_unlock;
799
800         down_write(&root->fs_info->subvol_sem);
801
802         err = may_destroy_subvol(dest);
803         if (err)
804                 goto out_up_write;
805
806         trans = btrfs_start_transaction(root, 1);
807         ret = btrfs_unlink_subvol(trans, root, dir,
808                                 dest->root_key.objectid,
809                                 dentry->d_name.name,
810                                 dentry->d_name.len);
811         BUG_ON(ret);
812
813         btrfs_record_root_in_trans(trans, dest);
814
815         memset(&dest->root_item.drop_progress, 0,
816                 sizeof(dest->root_item.drop_progress));
817         dest->root_item.drop_level = 0;
818         btrfs_set_root_refs(&dest->root_item, 0);
819
820         ret = btrfs_insert_orphan_item(trans,
821                                 root->fs_info->tree_root,
822                                 dest->root_key.objectid);
823         BUG_ON(ret);
824
825         ret = btrfs_commit_transaction(trans, root);
826         BUG_ON(ret);
827         inode->i_flags |= S_DEAD;
828 out_up_write:
829         up_write(&root->fs_info->subvol_sem);
830 out_unlock:
831         mutex_unlock(&inode->i_mutex);
832         if (!err) {
833                 btrfs_invalidate_inodes(dest);
834                 d_delete(dentry);
835         }
836 out_dput:
837         dput(dentry);
838 out_unlock_dir:
839         mutex_unlock(&dir->i_mutex);
840         mnt_drop_write(file->f_path.mnt);
841 out:
842         kfree(vol_args);
843         return err;
844 }
845
846 static int btrfs_ioctl_defrag(struct file *file)
847 {
848         struct inode *inode = fdentry(file)->d_inode;
849         struct btrfs_root *root = BTRFS_I(inode)->root;
850         int ret;
851
852         ret = mnt_want_write(file->f_path.mnt);
853         if (ret)
854                 return ret;
855
856         switch (inode->i_mode & S_IFMT) {
857         case S_IFDIR:
858                 if (!capable(CAP_SYS_ADMIN)) {
859                         ret = -EPERM;
860                         goto out;
861                 }
862                 btrfs_defrag_root(root, 0);
863                 btrfs_defrag_root(root->fs_info->extent_root, 0);
864                 break;
865         case S_IFREG:
866                 if (!(file->f_mode & FMODE_WRITE)) {
867                         ret = -EINVAL;
868                         goto out;
869                 }
870                 btrfs_defrag_file(file);
871                 break;
872         }
873 out:
874         mnt_drop_write(file->f_path.mnt);
875         return ret;
876 }
877
878 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
879 {
880         struct btrfs_ioctl_vol_args *vol_args;
881         int ret;
882
883         if (!capable(CAP_SYS_ADMIN))
884                 return -EPERM;
885
886         vol_args = memdup_user(arg, sizeof(*vol_args));
887         if (IS_ERR(vol_args))
888                 return PTR_ERR(vol_args);
889
890         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
891         ret = btrfs_init_new_device(root, vol_args->name);
892
893         kfree(vol_args);
894         return ret;
895 }
896
897 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
898 {
899         struct btrfs_ioctl_vol_args *vol_args;
900         int ret;
901
902         if (!capable(CAP_SYS_ADMIN))
903                 return -EPERM;
904
905         if (root->fs_info->sb->s_flags & MS_RDONLY)
906                 return -EROFS;
907
908         vol_args = memdup_user(arg, sizeof(*vol_args));
909         if (IS_ERR(vol_args))
910                 return PTR_ERR(vol_args);
911
912         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
913         ret = btrfs_rm_device(root, vol_args->name);
914
915         kfree(vol_args);
916         return ret;
917 }
918
919 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
920                                        u64 off, u64 olen, u64 destoff)
921 {
922         struct inode *inode = fdentry(file)->d_inode;
923         struct btrfs_root *root = BTRFS_I(inode)->root;
924         struct file *src_file;
925         struct inode *src;
926         struct btrfs_trans_handle *trans;
927         struct btrfs_path *path;
928         struct extent_buffer *leaf;
929         char *buf;
930         struct btrfs_key key;
931         u32 nritems;
932         int slot;
933         int ret;
934         u64 len = olen;
935         u64 bs = root->fs_info->sb->s_blocksize;
936         u64 hint_byte;
937
938         /*
939          * TODO:
940          * - split compressed inline extents.  annoying: we need to
941          *   decompress into destination's address_space (the file offset
942          *   may change, so source mapping won't do), then recompress (or
943          *   otherwise reinsert) a subrange.
944          * - allow ranges within the same file to be cloned (provided
945          *   they don't overlap)?
946          */
947
948         /* the destination must be opened for writing */
949         if (!(file->f_mode & FMODE_WRITE))
950                 return -EINVAL;
951
952         ret = mnt_want_write(file->f_path.mnt);
953         if (ret)
954                 return ret;
955
956         src_file = fget(srcfd);
957         if (!src_file) {
958                 ret = -EBADF;
959                 goto out_drop_write;
960         }
961         src = src_file->f_dentry->d_inode;
962
963         ret = -EINVAL;
964         if (src == inode)
965                 goto out_fput;
966
967         ret = -EISDIR;
968         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
969                 goto out_fput;
970
971         ret = -EXDEV;
972         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
973                 goto out_fput;
974
975         ret = -ENOMEM;
976         buf = vmalloc(btrfs_level_size(root, 0));
977         if (!buf)
978                 goto out_fput;
979
980         path = btrfs_alloc_path();
981         if (!path) {
982                 vfree(buf);
983                 goto out_fput;
984         }
985         path->reada = 2;
986
987         if (inode < src) {
988                 mutex_lock(&inode->i_mutex);
989                 mutex_lock(&src->i_mutex);
990         } else {
991                 mutex_lock(&src->i_mutex);
992                 mutex_lock(&inode->i_mutex);
993         }
994
995         /* determine range to clone */
996         ret = -EINVAL;
997         if (off >= src->i_size || off + len > src->i_size)
998                 goto out_unlock;
999         if (len == 0)
1000                 olen = len = src->i_size - off;
1001         /* if we extend to eof, continue to block boundary */
1002         if (off + len == src->i_size)
1003                 len = ((src->i_size + bs-1) & ~(bs-1))
1004                         - off;
1005
1006         /* verify the end result is block aligned */
1007         if ((off & (bs-1)) ||
1008             ((off + len) & (bs-1)))
1009                 goto out_unlock;
1010
1011         /* do any pending delalloc/csum calc on src, one way or
1012            another, and lock file content */
1013         while (1) {
1014                 struct btrfs_ordered_extent *ordered;
1015                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1016                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
1017                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
1018                         break;
1019                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1020                 if (ordered)
1021                         btrfs_put_ordered_extent(ordered);
1022                 btrfs_wait_ordered_range(src, off, off+len);
1023         }
1024
1025         trans = btrfs_start_transaction(root, 1);
1026         BUG_ON(!trans);
1027
1028         /* punch hole in destination first */
1029         btrfs_drop_extents(trans, root, inode, off, off + len,
1030                            off + len, 0, &hint_byte, 1);
1031
1032         /* clone data */
1033         key.objectid = src->i_ino;
1034         key.type = BTRFS_EXTENT_DATA_KEY;
1035         key.offset = 0;
1036
1037         while (1) {
1038                 /*
1039                  * note the key will change type as we walk through the
1040                  * tree.
1041                  */
1042                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
1043                 if (ret < 0)
1044                         goto out;
1045
1046                 nritems = btrfs_header_nritems(path->nodes[0]);
1047                 if (path->slots[0] >= nritems) {
1048                         ret = btrfs_next_leaf(root, path);
1049                         if (ret < 0)
1050                                 goto out;
1051                         if (ret > 0)
1052                                 break;
1053                         nritems = btrfs_header_nritems(path->nodes[0]);
1054                 }
1055                 leaf = path->nodes[0];
1056                 slot = path->slots[0];
1057
1058                 btrfs_item_key_to_cpu(leaf, &key, slot);
1059                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1060                     key.objectid != src->i_ino)
1061                         break;
1062
1063                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1064                         struct btrfs_file_extent_item *extent;
1065                         int type;
1066                         u32 size;
1067                         struct btrfs_key new_key;
1068                         u64 disko = 0, diskl = 0;
1069                         u64 datao = 0, datal = 0;
1070                         u8 comp;
1071
1072                         size = btrfs_item_size_nr(leaf, slot);
1073                         read_extent_buffer(leaf, buf,
1074                                            btrfs_item_ptr_offset(leaf, slot),
1075                                            size);
1076
1077                         extent = btrfs_item_ptr(leaf, slot,
1078                                                 struct btrfs_file_extent_item);
1079                         comp = btrfs_file_extent_compression(leaf, extent);
1080                         type = btrfs_file_extent_type(leaf, extent);
1081                         if (type == BTRFS_FILE_EXTENT_REG ||
1082                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1083                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1084                                                                       extent);
1085                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1086                                                                  extent);
1087                                 datao = btrfs_file_extent_offset(leaf, extent);
1088                                 datal = btrfs_file_extent_num_bytes(leaf,
1089                                                                     extent);
1090                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1091                                 /* take upper bound, may be compressed */
1092                                 datal = btrfs_file_extent_ram_bytes(leaf,
1093                                                                     extent);
1094                         }
1095                         btrfs_release_path(root, path);
1096
1097                         if (key.offset + datal < off ||
1098                             key.offset >= off+len)
1099                                 goto next;
1100
1101                         memcpy(&new_key, &key, sizeof(new_key));
1102                         new_key.objectid = inode->i_ino;
1103                         new_key.offset = key.offset + destoff - off;
1104
1105                         if (type == BTRFS_FILE_EXTENT_REG ||
1106                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1107                                 ret = btrfs_insert_empty_item(trans, root, path,
1108                                                               &new_key, size);
1109                                 if (ret)
1110                                         goto out;
1111
1112                                 leaf = path->nodes[0];
1113                                 slot = path->slots[0];
1114                                 write_extent_buffer(leaf, buf,
1115                                             btrfs_item_ptr_offset(leaf, slot),
1116                                             size);
1117
1118                                 extent = btrfs_item_ptr(leaf, slot,
1119                                                 struct btrfs_file_extent_item);
1120
1121                                 if (off > key.offset) {
1122                                         datao += off - key.offset;
1123                                         datal -= off - key.offset;
1124                                 }
1125                                 if (key.offset + datao + datal > off + len)
1126                                         datal = off + len - key.offset - datao;
1127                                 /* disko == 0 means it's a hole */
1128                                 if (!disko)
1129                                         datao = 0;
1130
1131                                 btrfs_set_file_extent_offset(leaf, extent,
1132                                                              datao);
1133                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1134                                                                 datal);
1135                                 if (disko) {
1136                                         inode_add_bytes(inode, datal);
1137                                         ret = btrfs_inc_extent_ref(trans, root,
1138                                                         disko, diskl, 0,
1139                                                         root->root_key.objectid,
1140                                                         inode->i_ino,
1141                                                         new_key.offset - datao);
1142                                         BUG_ON(ret);
1143                                 }
1144                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1145                                 u64 skip = 0;
1146                                 u64 trim = 0;
1147                                 if (off > key.offset) {
1148                                         skip = off - key.offset;
1149                                         new_key.offset += skip;
1150                                 }
1151
1152                                 if (key.offset + datal > off+len)
1153                                         trim = key.offset + datal - (off+len);
1154
1155                                 if (comp && (skip || trim)) {
1156                                         ret = -EINVAL;
1157                                         goto out;
1158                                 }
1159                                 size -= skip + trim;
1160                                 datal -= skip + trim;
1161                                 ret = btrfs_insert_empty_item(trans, root, path,
1162                                                               &new_key, size);
1163                                 if (ret)
1164                                         goto out;
1165
1166                                 if (skip) {
1167                                         u32 start =
1168                                           btrfs_file_extent_calc_inline_size(0);
1169                                         memmove(buf+start, buf+start+skip,
1170                                                 datal);
1171                                 }
1172
1173                                 leaf = path->nodes[0];
1174                                 slot = path->slots[0];
1175                                 write_extent_buffer(leaf, buf,
1176                                             btrfs_item_ptr_offset(leaf, slot),
1177                                             size);
1178                                 inode_add_bytes(inode, datal);
1179                         }
1180
1181                         btrfs_mark_buffer_dirty(leaf);
1182                 }
1183
1184 next:
1185                 btrfs_release_path(root, path);
1186                 key.offset++;
1187         }
1188         ret = 0;
1189 out:
1190         btrfs_release_path(root, path);
1191         if (ret == 0) {
1192                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1193                 if (destoff + olen > inode->i_size)
1194                         btrfs_i_size_write(inode, destoff + olen);
1195                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1196                 ret = btrfs_update_inode(trans, root, inode);
1197         }
1198         btrfs_end_transaction(trans, root);
1199         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1200         if (ret)
1201                 vmtruncate(inode, 0);
1202 out_unlock:
1203         mutex_unlock(&src->i_mutex);
1204         mutex_unlock(&inode->i_mutex);
1205         vfree(buf);
1206         btrfs_free_path(path);
1207 out_fput:
1208         fput(src_file);
1209 out_drop_write:
1210         mnt_drop_write(file->f_path.mnt);
1211         return ret;
1212 }
1213
1214 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1215 {
1216         struct btrfs_ioctl_clone_range_args args;
1217
1218         if (copy_from_user(&args, argp, sizeof(args)))
1219                 return -EFAULT;
1220         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1221                                  args.src_length, args.dest_offset);
1222 }
1223
1224 /*
1225  * there are many ways the trans_start and trans_end ioctls can lead
1226  * to deadlocks.  They should only be used by applications that
1227  * basically own the machine, and have a very in depth understanding
1228  * of all the possible deadlocks and enospc problems.
1229  */
1230 static long btrfs_ioctl_trans_start(struct file *file)
1231 {
1232         struct inode *inode = fdentry(file)->d_inode;
1233         struct btrfs_root *root = BTRFS_I(inode)->root;
1234         struct btrfs_trans_handle *trans;
1235         int ret;
1236
1237         ret = -EPERM;
1238         if (!capable(CAP_SYS_ADMIN))
1239                 goto out;
1240
1241         ret = -EINPROGRESS;
1242         if (file->private_data)
1243                 goto out;
1244
1245         ret = mnt_want_write(file->f_path.mnt);
1246         if (ret)
1247                 goto out;
1248
1249         mutex_lock(&root->fs_info->trans_mutex);
1250         root->fs_info->open_ioctl_trans++;
1251         mutex_unlock(&root->fs_info->trans_mutex);
1252
1253         ret = -ENOMEM;
1254         trans = btrfs_start_ioctl_transaction(root, 0);
1255         if (!trans)
1256                 goto out_drop;
1257
1258         file->private_data = trans;
1259         return 0;
1260
1261 out_drop:
1262         mutex_lock(&root->fs_info->trans_mutex);
1263         root->fs_info->open_ioctl_trans--;
1264         mutex_unlock(&root->fs_info->trans_mutex);
1265         mnt_drop_write(file->f_path.mnt);
1266 out:
1267         return ret;
1268 }
1269
1270 /*
1271  * there are many ways the trans_start and trans_end ioctls can lead
1272  * to deadlocks.  They should only be used by applications that
1273  * basically own the machine, and have a very in depth understanding
1274  * of all the possible deadlocks and enospc problems.
1275  */
1276 long btrfs_ioctl_trans_end(struct file *file)
1277 {
1278         struct inode *inode = fdentry(file)->d_inode;
1279         struct btrfs_root *root = BTRFS_I(inode)->root;
1280         struct btrfs_trans_handle *trans;
1281
1282         trans = file->private_data;
1283         if (!trans)
1284                 return -EINVAL;
1285         file->private_data = NULL;
1286
1287         btrfs_end_transaction(trans, root);
1288
1289         mutex_lock(&root->fs_info->trans_mutex);
1290         root->fs_info->open_ioctl_trans--;
1291         mutex_unlock(&root->fs_info->trans_mutex);
1292
1293         mnt_drop_write(file->f_path.mnt);
1294         return 0;
1295 }
1296
1297 long btrfs_ioctl(struct file *file, unsigned int
1298                 cmd, unsigned long arg)
1299 {
1300         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1301         void __user *argp = (void __user *)arg;
1302
1303         switch (cmd) {
1304         case FS_IOC_GETFLAGS:
1305                 return btrfs_ioctl_getflags(file, argp);
1306         case FS_IOC_SETFLAGS:
1307                 return btrfs_ioctl_setflags(file, argp);
1308         case FS_IOC_GETVERSION:
1309                 return btrfs_ioctl_getversion(file, argp);
1310         case BTRFS_IOC_SNAP_CREATE:
1311                 return btrfs_ioctl_snap_create(file, argp, 0);
1312         case BTRFS_IOC_SUBVOL_CREATE:
1313                 return btrfs_ioctl_snap_create(file, argp, 1);
1314         case BTRFS_IOC_SNAP_DESTROY:
1315                 return btrfs_ioctl_snap_destroy(file, argp);
1316         case BTRFS_IOC_DEFRAG:
1317                 return btrfs_ioctl_defrag(file);
1318         case BTRFS_IOC_RESIZE:
1319                 return btrfs_ioctl_resize(root, argp);
1320         case BTRFS_IOC_ADD_DEV:
1321                 return btrfs_ioctl_add_dev(root, argp);
1322         case BTRFS_IOC_RM_DEV:
1323                 return btrfs_ioctl_rm_dev(root, argp);
1324         case BTRFS_IOC_BALANCE:
1325                 return btrfs_balance(root->fs_info->dev_root);
1326         case BTRFS_IOC_CLONE:
1327                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1328         case BTRFS_IOC_CLONE_RANGE:
1329                 return btrfs_ioctl_clone_range(file, argp);
1330         case BTRFS_IOC_TRANS_START:
1331                 return btrfs_ioctl_trans_start(file);
1332         case BTRFS_IOC_TRANS_END:
1333                 return btrfs_ioctl_trans_end(file);
1334         case BTRFS_IOC_SYNC:
1335                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1336                 return 0;
1337         }
1338
1339         return -ENOTTY;
1340 }