Merge branch 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6] / 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/smp_lock.h>
31 #include <linux/backing-dev.h>
32 #include <linux/mount.h>
33 #include <linux/mpage.h>
34 #include <linux/namei.h>
35 #include <linux/swap.h>
36 #include <linux/writeback.h>
37 #include <linux/statfs.h>
38 #include <linux/compat.h>
39 #include <linux/bit_spinlock.h>
40 #include <linux/security.h>
41 #include <linux/version.h>
42 #include <linux/xattr.h>
43 #include <linux/vmalloc.h>
44 #include "compat.h"
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "ioctl.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53
54
55
56 static noinline int create_subvol(struct btrfs_root *root,
57                                   struct dentry *dentry,
58                                   char *name, int namelen)
59 {
60         struct btrfs_trans_handle *trans;
61         struct btrfs_key key;
62         struct btrfs_root_item root_item;
63         struct btrfs_inode_item *inode_item;
64         struct extent_buffer *leaf;
65         struct btrfs_root *new_root = root;
66         struct inode *dir;
67         int ret;
68         int err;
69         u64 objectid;
70         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
71         u64 index = 0;
72         unsigned long nr = 1;
73
74         ret = btrfs_check_free_space(root, 1, 0);
75         if (ret)
76                 goto fail_commit;
77
78         trans = btrfs_start_transaction(root, 1);
79         BUG_ON(!trans);
80
81         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
82                                        0, &objectid);
83         if (ret)
84                 goto fail;
85
86         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
87                                       objectid, trans->transid, 0, 0, 0);
88         if (IS_ERR(leaf)) {
89                 ret = PTR_ERR(leaf);
90                 goto fail;
91         }
92
93         btrfs_set_header_nritems(leaf, 0);
94         btrfs_set_header_level(leaf, 0);
95         btrfs_set_header_bytenr(leaf, leaf->start);
96         btrfs_set_header_generation(leaf, trans->transid);
97         btrfs_set_header_owner(leaf, objectid);
98
99         write_extent_buffer(leaf, root->fs_info->fsid,
100                             (unsigned long)btrfs_header_fsid(leaf),
101                             BTRFS_FSID_SIZE);
102         btrfs_mark_buffer_dirty(leaf);
103
104         inode_item = &root_item.inode;
105         memset(inode_item, 0, sizeof(*inode_item));
106         inode_item->generation = cpu_to_le64(1);
107         inode_item->size = cpu_to_le64(3);
108         inode_item->nlink = cpu_to_le32(1);
109         inode_item->nbytes = cpu_to_le64(root->leafsize);
110         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
111
112         btrfs_set_root_bytenr(&root_item, leaf->start);
113         btrfs_set_root_generation(&root_item, trans->transid);
114         btrfs_set_root_level(&root_item, 0);
115         btrfs_set_root_refs(&root_item, 1);
116         btrfs_set_root_used(&root_item, 0);
117         btrfs_set_root_last_snapshot(&root_item, 0);
118
119         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
120         root_item.drop_level = 0;
121
122         btrfs_tree_unlock(leaf);
123         free_extent_buffer(leaf);
124         leaf = NULL;
125
126         btrfs_set_root_dirid(&root_item, new_dirid);
127
128         key.objectid = objectid;
129         key.offset = 1;
130         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
131         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
132                                 &root_item);
133         if (ret)
134                 goto fail;
135
136         /*
137          * insert the directory item
138          */
139         key.offset = (u64)-1;
140         dir = dentry->d_parent->d_inode;
141         ret = btrfs_set_inode_index(dir, &index);
142         BUG_ON(ret);
143
144         ret = btrfs_insert_dir_item(trans, root,
145                                     name, namelen, dir->i_ino, &key,
146                                     BTRFS_FT_DIR, index);
147         if (ret)
148                 goto fail;
149
150         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
151         ret = btrfs_update_inode(trans, root, dir);
152         BUG_ON(ret);
153
154         /* add the backref first */
155         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
156                                  objectid, BTRFS_ROOT_BACKREF_KEY,
157                                  root->root_key.objectid,
158                                  dir->i_ino, index, name, namelen);
159
160         BUG_ON(ret);
161
162         /* now add the forward ref */
163         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
164                                  root->root_key.objectid, BTRFS_ROOT_REF_KEY,
165                                  objectid,
166                                  dir->i_ino, index, name, namelen);
167
168         BUG_ON(ret);
169
170         ret = btrfs_commit_transaction(trans, root);
171         if (ret)
172                 goto fail_commit;
173
174         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
175         BUG_ON(!new_root);
176
177         trans = btrfs_start_transaction(new_root, 1);
178         BUG_ON(!trans);
179
180         ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
181                                        BTRFS_I(dir)->block_group);
182         if (ret)
183                 goto fail;
184
185 fail:
186         nr = trans->blocks_used;
187         err = btrfs_commit_transaction(trans, new_root);
188         if (err && !ret)
189                 ret = err;
190 fail_commit:
191         btrfs_btree_balance_dirty(root, nr);
192         return ret;
193 }
194
195 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
196                            char *name, int namelen)
197 {
198         struct btrfs_pending_snapshot *pending_snapshot;
199         struct btrfs_trans_handle *trans;
200         int ret = 0;
201         int err;
202         unsigned long nr = 0;
203
204         if (!root->ref_cows)
205                 return -EINVAL;
206
207         ret = btrfs_check_free_space(root, 1, 0);
208         if (ret)
209                 goto fail_unlock;
210
211         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
212         if (!pending_snapshot) {
213                 ret = -ENOMEM;
214                 goto fail_unlock;
215         }
216         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
217         if (!pending_snapshot->name) {
218                 ret = -ENOMEM;
219                 kfree(pending_snapshot);
220                 goto fail_unlock;
221         }
222         memcpy(pending_snapshot->name, name, namelen);
223         pending_snapshot->name[namelen] = '\0';
224         pending_snapshot->dentry = dentry;
225         trans = btrfs_start_transaction(root, 1);
226         BUG_ON(!trans);
227         pending_snapshot->root = root;
228         list_add(&pending_snapshot->list,
229                  &trans->transaction->pending_snapshots);
230         err = btrfs_commit_transaction(trans, root);
231
232 fail_unlock:
233         btrfs_btree_balance_dirty(root, nr);
234         return ret;
235 }
236
237 /* copy of may_create in fs/namei.c() */
238 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
239 {
240         if (child->d_inode)
241                 return -EEXIST;
242         if (IS_DEADDIR(dir))
243                 return -ENOENT;
244         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
245 }
246
247 /*
248  * Create a new subvolume below @parent.  This is largely modeled after
249  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
250  * inside this filesystem so it's quite a bit simpler.
251  */
252 static noinline int btrfs_mksubvol(struct path *parent, char *name,
253                                    int mode, int namelen,
254                                    struct btrfs_root *snap_src)
255 {
256         struct dentry *dentry;
257         int error;
258
259         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
260
261         dentry = lookup_one_len(name, parent->dentry, namelen);
262         error = PTR_ERR(dentry);
263         if (IS_ERR(dentry))
264                 goto out_unlock;
265
266         error = -EEXIST;
267         if (dentry->d_inode)
268                 goto out_dput;
269
270         if (!IS_POSIXACL(parent->dentry->d_inode))
271                 mode &= ~current->fs->umask;
272
273         error = mnt_want_write(parent->mnt);
274         if (error)
275                 goto out_dput;
276
277         error = btrfs_may_create(parent->dentry->d_inode, dentry);
278         if (error)
279                 goto out_drop_write;
280
281         /*
282          * Actually perform the low-level subvolume creation after all
283          * this VFS fuzz.
284          *
285          * Eventually we want to pass in an inode under which we create this
286          * subvolume, but for now all are under the filesystem root.
287          *
288          * Also we should pass on the mode eventually to allow creating new
289          * subvolume with specific mode bits.
290          */
291         if (snap_src) {
292                 struct dentry *dir = dentry->d_parent;
293                 struct dentry *test = dir->d_parent;
294                 struct btrfs_path *path = btrfs_alloc_path();
295                 int ret;
296                 u64 test_oid;
297                 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
298
299                 test_oid = snap_src->root_key.objectid;
300
301                 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
302                                           path, parent_oid, test_oid);
303                 if (ret == 0)
304                         goto create;
305                 btrfs_release_path(snap_src->fs_info->tree_root, path);
306
307                 /* we need to make sure we aren't creating a directory loop
308                  * by taking a snapshot of something that has our current
309                  * subvol in its directory tree.  So, this loops through
310                  * the dentries and checks the forward refs for each subvolume
311                  * to see if is references the subvolume where we are
312                  * placing this new snapshot.
313                  */
314                 while (1) {
315                         if (!test ||
316                             dir == snap_src->fs_info->sb->s_root ||
317                             test == snap_src->fs_info->sb->s_root ||
318                             test->d_inode->i_sb != snap_src->fs_info->sb) {
319                                 break;
320                         }
321                         if (S_ISLNK(test->d_inode->i_mode)) {
322                                 printk(KERN_INFO "Btrfs symlink in snapshot "
323                                        "path, failed\n");
324                                 error = -EMLINK;
325                                 btrfs_free_path(path);
326                                 goto out_drop_write;
327                         }
328                         test_oid =
329                                 BTRFS_I(test->d_inode)->root->root_key.objectid;
330                         ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
331                                   path, test_oid, parent_oid);
332                         if (ret == 0) {
333                                 printk(KERN_INFO "Btrfs snapshot creation "
334                                        "failed, looping\n");
335                                 error = -EMLINK;
336                                 btrfs_free_path(path);
337                                 goto out_drop_write;
338                         }
339                         btrfs_release_path(snap_src->fs_info->tree_root, path);
340                         test = test->d_parent;
341                 }
342 create:
343                 btrfs_free_path(path);
344                 error = create_snapshot(snap_src, dentry, name, namelen);
345         } else {
346                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
347                                       dentry, name, namelen);
348         }
349         if (error)
350                 goto out_drop_write;
351
352         fsnotify_mkdir(parent->dentry->d_inode, dentry);
353 out_drop_write:
354         mnt_drop_write(parent->mnt);
355 out_dput:
356         dput(dentry);
357 out_unlock:
358         mutex_unlock(&parent->dentry->d_inode->i_mutex);
359         return error;
360 }
361
362
363 static int btrfs_defrag_file(struct file *file)
364 {
365         struct inode *inode = fdentry(file)->d_inode;
366         struct btrfs_root *root = BTRFS_I(inode)->root;
367         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
368         struct btrfs_ordered_extent *ordered;
369         struct page *page;
370         unsigned long last_index;
371         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
372         unsigned long total_read = 0;
373         u64 page_start;
374         u64 page_end;
375         unsigned long i;
376         int ret;
377
378         ret = btrfs_check_free_space(root, inode->i_size, 0);
379         if (ret)
380                 return -ENOSPC;
381
382         mutex_lock(&inode->i_mutex);
383         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
384         for (i = 0; i <= last_index; i++) {
385                 if (total_read % ra_pages == 0) {
386                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
387                                        min(last_index, i + ra_pages - 1));
388                 }
389                 total_read++;
390 again:
391                 page = grab_cache_page(inode->i_mapping, i);
392                 if (!page)
393                         goto out_unlock;
394                 if (!PageUptodate(page)) {
395                         btrfs_readpage(NULL, page);
396                         lock_page(page);
397                         if (!PageUptodate(page)) {
398                                 unlock_page(page);
399                                 page_cache_release(page);
400                                 goto out_unlock;
401                         }
402                 }
403
404                 wait_on_page_writeback(page);
405
406                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
407                 page_end = page_start + PAGE_CACHE_SIZE - 1;
408                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
409
410                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
411                 if (ordered) {
412                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
413                         unlock_page(page);
414                         page_cache_release(page);
415                         btrfs_start_ordered_extent(inode, ordered, 1);
416                         btrfs_put_ordered_extent(ordered);
417                         goto again;
418                 }
419                 set_page_extent_mapped(page);
420
421                 /*
422                  * this makes sure page_mkwrite is called on the
423                  * page if it is dirtied again later
424                  */
425                 clear_page_dirty_for_io(page);
426
427                 btrfs_set_extent_delalloc(inode, page_start, page_end);
428
429                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
430                 set_page_dirty(page);
431                 unlock_page(page);
432                 page_cache_release(page);
433                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
434         }
435
436 out_unlock:
437         mutex_unlock(&inode->i_mutex);
438         return 0;
439 }
440
441 /*
442  * Called inside transaction, so use GFP_NOFS
443  */
444
445 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
446 {
447         u64 new_size;
448         u64 old_size;
449         u64 devid = 1;
450         struct btrfs_ioctl_vol_args *vol_args;
451         struct btrfs_trans_handle *trans;
452         struct btrfs_device *device = NULL;
453         char *sizestr;
454         char *devstr = NULL;
455         int ret = 0;
456         int namelen;
457         int mod = 0;
458
459         if (root->fs_info->sb->s_flags & MS_RDONLY)
460                 return -EROFS;
461
462         if (!capable(CAP_SYS_ADMIN))
463                 return -EPERM;
464
465         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
466
467         if (!vol_args)
468                 return -ENOMEM;
469
470         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
471                 ret = -EFAULT;
472                 goto out;
473         }
474
475         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
476         namelen = strlen(vol_args->name);
477
478         mutex_lock(&root->fs_info->volume_mutex);
479         sizestr = vol_args->name;
480         devstr = strchr(sizestr, ':');
481         if (devstr) {
482                 char *end;
483                 sizestr = devstr + 1;
484                 *devstr = '\0';
485                 devstr = vol_args->name;
486                 devid = simple_strtoull(devstr, &end, 10);
487                 printk(KERN_INFO "resizing devid %llu\n", devid);
488         }
489         device = btrfs_find_device(root, devid, NULL, NULL);
490         if (!device) {
491                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
492                 ret = -EINVAL;
493                 goto out_unlock;
494         }
495         if (!strcmp(sizestr, "max"))
496                 new_size = device->bdev->bd_inode->i_size;
497         else {
498                 if (sizestr[0] == '-') {
499                         mod = -1;
500                         sizestr++;
501                 } else if (sizestr[0] == '+') {
502                         mod = 1;
503                         sizestr++;
504                 }
505                 new_size = btrfs_parse_size(sizestr);
506                 if (new_size == 0) {
507                         ret = -EINVAL;
508                         goto out_unlock;
509                 }
510         }
511
512         old_size = device->total_bytes;
513
514         if (mod < 0) {
515                 if (new_size > old_size) {
516                         ret = -EINVAL;
517                         goto out_unlock;
518                 }
519                 new_size = old_size - new_size;
520         } else if (mod > 0) {
521                 new_size = old_size + new_size;
522         }
523
524         if (new_size < 256 * 1024 * 1024) {
525                 ret = -EINVAL;
526                 goto out_unlock;
527         }
528         if (new_size > device->bdev->bd_inode->i_size) {
529                 ret = -EFBIG;
530                 goto out_unlock;
531         }
532
533         do_div(new_size, root->sectorsize);
534         new_size *= root->sectorsize;
535
536         printk(KERN_INFO "new size for %s is %llu\n",
537                 device->name, (unsigned long long)new_size);
538
539         if (new_size > old_size) {
540                 trans = btrfs_start_transaction(root, 1);
541                 ret = btrfs_grow_device(trans, device, new_size);
542                 btrfs_commit_transaction(trans, root);
543         } else {
544                 ret = btrfs_shrink_device(device, new_size);
545         }
546
547 out_unlock:
548         mutex_unlock(&root->fs_info->volume_mutex);
549 out:
550         kfree(vol_args);
551         return ret;
552 }
553
554 static noinline int btrfs_ioctl_snap_create(struct file *file,
555                                             void __user *arg, int subvol)
556 {
557         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
558         struct btrfs_ioctl_vol_args *vol_args;
559         struct btrfs_dir_item *di;
560         struct btrfs_path *path;
561         struct file *src_file;
562         u64 root_dirid;
563         int namelen;
564         int ret = 0;
565
566         if (root->fs_info->sb->s_flags & MS_RDONLY)
567                 return -EROFS;
568
569         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
570
571         if (!vol_args)
572                 return -ENOMEM;
573
574         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
575                 ret = -EFAULT;
576                 goto out;
577         }
578
579         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
580         namelen = strlen(vol_args->name);
581         if (strchr(vol_args->name, '/')) {
582                 ret = -EINVAL;
583                 goto out;
584         }
585
586         path = btrfs_alloc_path();
587         if (!path) {
588                 ret = -ENOMEM;
589                 goto out;
590         }
591
592         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
593         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
594                             path, root_dirid,
595                             vol_args->name, namelen, 0);
596         btrfs_free_path(path);
597
598         if (di && !IS_ERR(di)) {
599                 ret = -EEXIST;
600                 goto out;
601         }
602
603         if (IS_ERR(di)) {
604                 ret = PTR_ERR(di);
605                 goto out;
606         }
607
608         if (subvol) {
609                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
610                                      file->f_path.dentry->d_inode->i_mode,
611                                      namelen, NULL);
612         } else {
613                 struct inode *src_inode;
614                 src_file = fget(vol_args->fd);
615                 if (!src_file) {
616                         ret = -EINVAL;
617                         goto out;
618                 }
619
620                 src_inode = src_file->f_path.dentry->d_inode;
621                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
622                         printk(KERN_INFO "btrfs: Snapshot src from "
623                                "another FS\n");
624                         ret = -EINVAL;
625                         fput(src_file);
626                         goto out;
627                 }
628                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
629                              file->f_path.dentry->d_inode->i_mode,
630                              namelen, BTRFS_I(src_inode)->root);
631                 fput(src_file);
632         }
633
634 out:
635         kfree(vol_args);
636         return ret;
637 }
638
639 static int btrfs_ioctl_defrag(struct file *file)
640 {
641         struct inode *inode = fdentry(file)->d_inode;
642         struct btrfs_root *root = BTRFS_I(inode)->root;
643         int ret;
644
645         ret = mnt_want_write(file->f_path.mnt);
646         if (ret)
647                 return ret;
648
649         switch (inode->i_mode & S_IFMT) {
650         case S_IFDIR:
651                 if (!capable(CAP_SYS_ADMIN)) {
652                         ret = -EPERM;
653                         goto out;
654                 }
655                 btrfs_defrag_root(root, 0);
656                 btrfs_defrag_root(root->fs_info->extent_root, 0);
657                 break;
658         case S_IFREG:
659                 if (!(file->f_mode & FMODE_WRITE)) {
660                         ret = -EINVAL;
661                         goto out;
662                 }
663                 btrfs_defrag_file(file);
664                 break;
665         }
666 out:
667         mnt_drop_write(file->f_path.mnt);
668         return ret;
669 }
670
671 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
672 {
673         struct btrfs_ioctl_vol_args *vol_args;
674         int ret;
675
676         if (!capable(CAP_SYS_ADMIN))
677                 return -EPERM;
678
679         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
680
681         if (!vol_args)
682                 return -ENOMEM;
683
684         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
685                 ret = -EFAULT;
686                 goto out;
687         }
688         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
689         ret = btrfs_init_new_device(root, vol_args->name);
690
691 out:
692         kfree(vol_args);
693         return ret;
694 }
695
696 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
697 {
698         struct btrfs_ioctl_vol_args *vol_args;
699         int ret;
700
701         if (!capable(CAP_SYS_ADMIN))
702                 return -EPERM;
703
704         if (root->fs_info->sb->s_flags & MS_RDONLY)
705                 return -EROFS;
706
707         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
708
709         if (!vol_args)
710                 return -ENOMEM;
711
712         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
713                 ret = -EFAULT;
714                 goto out;
715         }
716         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
717         ret = btrfs_rm_device(root, vol_args->name);
718
719 out:
720         kfree(vol_args);
721         return ret;
722 }
723
724 static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
725                 u64 off, u64 olen, u64 destoff)
726 {
727         struct inode *inode = fdentry(file)->d_inode;
728         struct btrfs_root *root = BTRFS_I(inode)->root;
729         struct file *src_file;
730         struct inode *src;
731         struct btrfs_trans_handle *trans;
732         struct btrfs_path *path;
733         struct extent_buffer *leaf;
734         char *buf;
735         struct btrfs_key key;
736         u32 nritems;
737         int slot;
738         int ret;
739         u64 len = olen;
740         u64 bs = root->fs_info->sb->s_blocksize;
741         u64 hint_byte;
742
743         /*
744          * TODO:
745          * - split compressed inline extents.  annoying: we need to
746          *   decompress into destination's address_space (the file offset
747          *   may change, so source mapping won't do), then recompress (or
748          *   otherwise reinsert) a subrange.
749          * - allow ranges within the same file to be cloned (provided
750          *   they don't overlap)?
751          */
752
753         /* the destination must be opened for writing */
754         if (!(file->f_mode & FMODE_WRITE))
755                 return -EINVAL;
756
757         ret = mnt_want_write(file->f_path.mnt);
758         if (ret)
759                 return ret;
760
761         src_file = fget(srcfd);
762         if (!src_file) {
763                 ret = -EBADF;
764                 goto out_drop_write;
765         }
766         src = src_file->f_dentry->d_inode;
767
768         ret = -EINVAL;
769         if (src == inode)
770                 goto out_fput;
771
772         ret = -EISDIR;
773         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
774                 goto out_fput;
775
776         ret = -EXDEV;
777         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
778                 goto out_fput;
779
780         ret = -ENOMEM;
781         buf = vmalloc(btrfs_level_size(root, 0));
782         if (!buf)
783                 goto out_fput;
784
785         path = btrfs_alloc_path();
786         if (!path) {
787                 vfree(buf);
788                 goto out_fput;
789         }
790         path->reada = 2;
791
792         if (inode < src) {
793                 mutex_lock(&inode->i_mutex);
794                 mutex_lock(&src->i_mutex);
795         } else {
796                 mutex_lock(&src->i_mutex);
797                 mutex_lock(&inode->i_mutex);
798         }
799
800         /* determine range to clone */
801         ret = -EINVAL;
802         if (off >= src->i_size || off + len > src->i_size)
803                 goto out_unlock;
804         if (len == 0)
805                 olen = len = src->i_size - off;
806         /* if we extend to eof, continue to block boundary */
807         if (off + len == src->i_size)
808                 len = ((src->i_size + bs-1) & ~(bs-1))
809                         - off;
810
811         /* verify the end result is block aligned */
812         if ((off & (bs-1)) ||
813             ((off + len) & (bs-1)))
814                 goto out_unlock;
815
816         /* do any pending delalloc/csum calc on src, one way or
817            another, and lock file content */
818         while (1) {
819                 struct btrfs_ordered_extent *ordered;
820                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
821                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
822                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
823                         break;
824                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
825                 if (ordered)
826                         btrfs_put_ordered_extent(ordered);
827                 btrfs_wait_ordered_range(src, off, off+len);
828         }
829
830         trans = btrfs_start_transaction(root, 1);
831         BUG_ON(!trans);
832
833         /* punch hole in destination first */
834         btrfs_drop_extents(trans, root, inode, off, off+len, 0, &hint_byte);
835
836         /* clone data */
837         key.objectid = src->i_ino;
838         key.type = BTRFS_EXTENT_DATA_KEY;
839         key.offset = 0;
840
841         while (1) {
842                 /*
843                  * note the key will change type as we walk through the
844                  * tree.
845                  */
846                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
847                 if (ret < 0)
848                         goto out;
849
850                 nritems = btrfs_header_nritems(path->nodes[0]);
851                 if (path->slots[0] >= nritems) {
852                         ret = btrfs_next_leaf(root, path);
853                         if (ret < 0)
854                                 goto out;
855                         if (ret > 0)
856                                 break;
857                         nritems = btrfs_header_nritems(path->nodes[0]);
858                 }
859                 leaf = path->nodes[0];
860                 slot = path->slots[0];
861
862                 btrfs_item_key_to_cpu(leaf, &key, slot);
863                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
864                     key.objectid != src->i_ino)
865                         break;
866
867                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
868                         struct btrfs_file_extent_item *extent;
869                         int type;
870                         u32 size;
871                         struct btrfs_key new_key;
872                         u64 disko = 0, diskl = 0;
873                         u64 datao = 0, datal = 0;
874                         u8 comp;
875
876                         size = btrfs_item_size_nr(leaf, slot);
877                         read_extent_buffer(leaf, buf,
878                                            btrfs_item_ptr_offset(leaf, slot),
879                                            size);
880
881                         extent = btrfs_item_ptr(leaf, slot,
882                                                 struct btrfs_file_extent_item);
883                         comp = btrfs_file_extent_compression(leaf, extent);
884                         type = btrfs_file_extent_type(leaf, extent);
885                         if (type == BTRFS_FILE_EXTENT_REG) {
886                                 disko = btrfs_file_extent_disk_bytenr(leaf,
887                                                                       extent);
888                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
889                                                                  extent);
890                                 datao = btrfs_file_extent_offset(leaf, extent);
891                                 datal = btrfs_file_extent_num_bytes(leaf,
892                                                                     extent);
893                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
894                                 /* take upper bound, may be compressed */
895                                 datal = btrfs_file_extent_ram_bytes(leaf,
896                                                                     extent);
897                         }
898                         btrfs_release_path(root, path);
899
900                         if (key.offset + datal < off ||
901                             key.offset >= off+len)
902                                 goto next;
903
904                         memcpy(&new_key, &key, sizeof(new_key));
905                         new_key.objectid = inode->i_ino;
906                         new_key.offset = key.offset + destoff - off;
907
908                         if (type == BTRFS_FILE_EXTENT_REG) {
909                                 ret = btrfs_insert_empty_item(trans, root, path,
910                                                               &new_key, size);
911                                 if (ret)
912                                         goto out;
913
914                                 leaf = path->nodes[0];
915                                 slot = path->slots[0];
916                                 write_extent_buffer(leaf, buf,
917                                             btrfs_item_ptr_offset(leaf, slot),
918                                             size);
919
920                                 extent = btrfs_item_ptr(leaf, slot,
921                                                 struct btrfs_file_extent_item);
922
923                                 if (off > key.offset) {
924                                         datao += off - key.offset;
925                                         datal -= off - key.offset;
926                                 }
927                                 if (key.offset + datao + datal + key.offset >
928                                     off + len)
929                                         datal = off + len - key.offset - datao;
930                                 /* disko == 0 means it's a hole */
931                                 if (!disko)
932                                         datao = 0;
933
934                                 btrfs_set_file_extent_offset(leaf, extent,
935                                                              datao);
936                                 btrfs_set_file_extent_num_bytes(leaf, extent,
937                                                                 datal);
938                                 if (disko) {
939                                         inode_add_bytes(inode, datal);
940                                         ret = btrfs_inc_extent_ref(trans, root,
941                                                    disko, diskl, leaf->start,
942                                                    root->root_key.objectid,
943                                                    trans->transid,
944                                                    inode->i_ino);
945                                         BUG_ON(ret);
946                                 }
947                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
948                                 u64 skip = 0;
949                                 u64 trim = 0;
950                                 if (off > key.offset) {
951                                         skip = off - key.offset;
952                                         new_key.offset += skip;
953                                 }
954
955                                 if (key.offset + datal > off+len)
956                                         trim = key.offset + datal - (off+len);
957
958                                 if (comp && (skip || trim)) {
959                                         ret = -EINVAL;
960                                         goto out;
961                                 }
962                                 size -= skip + trim;
963                                 datal -= skip + trim;
964                                 ret = btrfs_insert_empty_item(trans, root, path,
965                                                               &new_key, size);
966                                 if (ret)
967                                         goto out;
968
969                                 if (skip) {
970                                         u32 start =
971                                           btrfs_file_extent_calc_inline_size(0);
972                                         memmove(buf+start, buf+start+skip,
973                                                 datal);
974                                 }
975
976                                 leaf = path->nodes[0];
977                                 slot = path->slots[0];
978                                 write_extent_buffer(leaf, buf,
979                                             btrfs_item_ptr_offset(leaf, slot),
980                                             size);
981                                 inode_add_bytes(inode, datal);
982                         }
983
984                         btrfs_mark_buffer_dirty(leaf);
985                 }
986
987 next:
988                 btrfs_release_path(root, path);
989                 key.offset++;
990         }
991         ret = 0;
992 out:
993         btrfs_release_path(root, path);
994         if (ret == 0) {
995                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
996                 if (destoff + olen > inode->i_size)
997                         btrfs_i_size_write(inode, destoff + olen);
998                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
999                 ret = btrfs_update_inode(trans, root, inode);
1000         }
1001         btrfs_end_transaction(trans, root);
1002         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1003         if (ret)
1004                 vmtruncate(inode, 0);
1005 out_unlock:
1006         mutex_unlock(&src->i_mutex);
1007         mutex_unlock(&inode->i_mutex);
1008         vfree(buf);
1009         btrfs_free_path(path);
1010 out_fput:
1011         fput(src_file);
1012 out_drop_write:
1013         mnt_drop_write(file->f_path.mnt);
1014         return ret;
1015 }
1016
1017 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1018 {
1019         struct btrfs_ioctl_clone_range_args args;
1020
1021         if (copy_from_user(&args, argp, sizeof(args)))
1022                 return -EFAULT;
1023         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1024                                  args.src_length, args.dest_offset);
1025 }
1026
1027 /*
1028  * there are many ways the trans_start and trans_end ioctls can lead
1029  * to deadlocks.  They should only be used by applications that
1030  * basically own the machine, and have a very in depth understanding
1031  * of all the possible deadlocks and enospc problems.
1032  */
1033 static long btrfs_ioctl_trans_start(struct file *file)
1034 {
1035         struct inode *inode = fdentry(file)->d_inode;
1036         struct btrfs_root *root = BTRFS_I(inode)->root;
1037         struct btrfs_trans_handle *trans;
1038         int ret = 0;
1039
1040         if (!capable(CAP_SYS_ADMIN))
1041                 return -EPERM;
1042
1043         if (file->private_data) {
1044                 ret = -EINPROGRESS;
1045                 goto out;
1046         }
1047
1048         ret = mnt_want_write(file->f_path.mnt);
1049         if (ret)
1050                 goto out;
1051
1052         mutex_lock(&root->fs_info->trans_mutex);
1053         root->fs_info->open_ioctl_trans++;
1054         mutex_unlock(&root->fs_info->trans_mutex);
1055
1056         trans = btrfs_start_ioctl_transaction(root, 0);
1057         if (trans)
1058                 file->private_data = trans;
1059         else
1060                 ret = -ENOMEM;
1061         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1062 out:
1063         return ret;
1064 }
1065
1066 /*
1067  * there are many ways the trans_start and trans_end ioctls can lead
1068  * to deadlocks.  They should only be used by applications that
1069  * basically own the machine, and have a very in depth understanding
1070  * of all the possible deadlocks and enospc problems.
1071  */
1072 long btrfs_ioctl_trans_end(struct file *file)
1073 {
1074         struct inode *inode = fdentry(file)->d_inode;
1075         struct btrfs_root *root = BTRFS_I(inode)->root;
1076         struct btrfs_trans_handle *trans;
1077         int ret = 0;
1078
1079         trans = file->private_data;
1080         if (!trans) {
1081                 ret = -EINVAL;
1082                 goto out;
1083         }
1084         btrfs_end_transaction(trans, root);
1085         file->private_data = NULL;
1086
1087         mutex_lock(&root->fs_info->trans_mutex);
1088         root->fs_info->open_ioctl_trans--;
1089         mutex_unlock(&root->fs_info->trans_mutex);
1090
1091         mnt_drop_write(file->f_path.mnt);
1092
1093 out:
1094         return ret;
1095 }
1096
1097 long btrfs_ioctl(struct file *file, unsigned int
1098                 cmd, unsigned long arg)
1099 {
1100         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1101         void __user *argp = (void __user *)arg;
1102
1103         switch (cmd) {
1104         case BTRFS_IOC_SNAP_CREATE:
1105                 return btrfs_ioctl_snap_create(file, argp, 0);
1106         case BTRFS_IOC_SUBVOL_CREATE:
1107                 return btrfs_ioctl_snap_create(file, argp, 1);
1108         case BTRFS_IOC_DEFRAG:
1109                 return btrfs_ioctl_defrag(file);
1110         case BTRFS_IOC_RESIZE:
1111                 return btrfs_ioctl_resize(root, argp);
1112         case BTRFS_IOC_ADD_DEV:
1113                 return btrfs_ioctl_add_dev(root, argp);
1114         case BTRFS_IOC_RM_DEV:
1115                 return btrfs_ioctl_rm_dev(root, argp);
1116         case BTRFS_IOC_BALANCE:
1117                 return btrfs_balance(root->fs_info->dev_root);
1118         case BTRFS_IOC_CLONE:
1119                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1120         case BTRFS_IOC_CLONE_RANGE:
1121                 return btrfs_ioctl_clone_range(file, argp);
1122         case BTRFS_IOC_TRANS_START:
1123                 return btrfs_ioctl_trans_start(file);
1124         case BTRFS_IOC_TRANS_END:
1125                 return btrfs_ioctl_trans_end(file);
1126         case BTRFS_IOC_SYNC:
1127                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1128                 return 0;
1129         }
1130
1131         return -ENOTTY;
1132 }