Btrfs: Fix nodatacow for the new data=ordered mode
[cascardo/linux.git] / fs / btrfs / ioctl.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include <linux/vmalloc.h>
40 #include "ctree.h"
41 #include "disk-io.h"
42 #include "transaction.h"
43 #include "btrfs_inode.h"
44 #include "ioctl.h"
45 #include "print-tree.h"
46 #include "volumes.h"
47 #include "locking.h"
48
49
50
51 static noinline int create_subvol(struct btrfs_root *root, char *name,
52                                   int namelen)
53 {
54         struct btrfs_trans_handle *trans;
55         struct btrfs_key key;
56         struct btrfs_root_item root_item;
57         struct btrfs_inode_item *inode_item;
58         struct extent_buffer *leaf;
59         struct btrfs_root *new_root = root;
60         struct inode *dir;
61         int ret;
62         int err;
63         u64 objectid;
64         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
65         unsigned long nr = 1;
66
67         ret = btrfs_check_free_space(root, 1, 0);
68         if (ret)
69                 goto fail_commit;
70
71         trans = btrfs_start_transaction(root, 1);
72         BUG_ON(!trans);
73
74         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
75                                        0, &objectid);
76         if (ret)
77                 goto fail;
78
79         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
80                                       objectid, trans->transid, 0, 0,
81                                       0, 0);
82         if (IS_ERR(leaf)) {
83                 ret = PTR_ERR(leaf);
84                 goto fail;
85         }
86
87         btrfs_set_header_nritems(leaf, 0);
88         btrfs_set_header_level(leaf, 0);
89         btrfs_set_header_bytenr(leaf, leaf->start);
90         btrfs_set_header_generation(leaf, trans->transid);
91         btrfs_set_header_owner(leaf, objectid);
92
93         write_extent_buffer(leaf, root->fs_info->fsid,
94                             (unsigned long)btrfs_header_fsid(leaf),
95                             BTRFS_FSID_SIZE);
96         btrfs_mark_buffer_dirty(leaf);
97
98         inode_item = &root_item.inode;
99         memset(inode_item, 0, sizeof(*inode_item));
100         inode_item->generation = cpu_to_le64(1);
101         inode_item->size = cpu_to_le64(3);
102         inode_item->nlink = cpu_to_le32(1);
103         inode_item->nblocks = cpu_to_le64(1);
104         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
105
106         btrfs_set_root_bytenr(&root_item, leaf->start);
107         btrfs_set_root_level(&root_item, 0);
108         btrfs_set_root_refs(&root_item, 1);
109         btrfs_set_root_used(&root_item, 0);
110
111         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
112         root_item.drop_level = 0;
113
114         btrfs_tree_unlock(leaf);
115         free_extent_buffer(leaf);
116         leaf = NULL;
117
118         btrfs_set_root_dirid(&root_item, new_dirid);
119
120         key.objectid = objectid;
121         key.offset = 1;
122         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
123         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
124                                 &root_item);
125         if (ret)
126                 goto fail;
127
128         /*
129          * insert the directory item
130          */
131         key.offset = (u64)-1;
132         dir = root->fs_info->sb->s_root->d_inode;
133         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
134                                     name, namelen, dir->i_ino, &key,
135                                     BTRFS_FT_DIR, 0);
136         if (ret)
137                 goto fail;
138
139         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
140                              name, namelen, objectid,
141                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
142         if (ret)
143                 goto fail;
144
145         ret = btrfs_commit_transaction(trans, root);
146         if (ret)
147                 goto fail_commit;
148
149         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
150         BUG_ON(!new_root);
151
152         trans = btrfs_start_transaction(new_root, 1);
153         BUG_ON(!trans);
154
155         ret = btrfs_create_subvol_root(new_root, trans, new_dirid,
156                                        BTRFS_I(dir)->block_group);
157         if (ret)
158                 goto fail;
159
160         /* Invalidate existing dcache entry for new subvolume. */
161         btrfs_invalidate_dcache_root(root, name, namelen);
162
163 fail:
164         nr = trans->blocks_used;
165         err = btrfs_commit_transaction(trans, new_root);
166         if (err && !ret)
167                 ret = err;
168 fail_commit:
169         btrfs_btree_balance_dirty(root, nr);
170         return ret;
171 }
172
173 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
174 {
175         struct btrfs_pending_snapshot *pending_snapshot;
176         struct btrfs_trans_handle *trans;
177         int ret;
178         int err;
179         unsigned long nr = 0;
180
181         if (!root->ref_cows)
182                 return -EINVAL;
183
184         ret = btrfs_check_free_space(root, 1, 0);
185         if (ret)
186                 goto fail_unlock;
187
188         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
189         if (!pending_snapshot) {
190                 ret = -ENOMEM;
191                 goto fail_unlock;
192         }
193         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
194         if (!pending_snapshot->name) {
195                 ret = -ENOMEM;
196                 kfree(pending_snapshot);
197                 goto fail_unlock;
198         }
199         memcpy(pending_snapshot->name, name, namelen);
200         pending_snapshot->name[namelen] = '\0';
201         trans = btrfs_start_transaction(root, 1);
202         BUG_ON(!trans);
203         pending_snapshot->root = root;
204         list_add(&pending_snapshot->list,
205                  &trans->transaction->pending_snapshots);
206         ret = btrfs_update_inode(trans, root, root->inode);
207         err = btrfs_commit_transaction(trans, root);
208
209 fail_unlock:
210         btrfs_btree_balance_dirty(root, nr);
211         return ret;
212 }
213
214 int btrfs_defrag_file(struct file *file)
215 {
216         struct inode *inode = fdentry(file)->d_inode;
217         struct btrfs_root *root = BTRFS_I(inode)->root;
218         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
219         struct btrfs_ordered_extent *ordered;
220         struct page *page;
221         unsigned long last_index;
222         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
223         unsigned long total_read = 0;
224         u64 page_start;
225         u64 page_end;
226         unsigned long i;
227         int ret;
228
229         ret = btrfs_check_free_space(root, inode->i_size, 0);
230         if (ret)
231                 return -ENOSPC;
232
233         mutex_lock(&inode->i_mutex);
234         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
235         for (i = 0; i <= last_index; i++) {
236                 if (total_read % ra_pages == 0) {
237                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
238                                        min(last_index, i + ra_pages - 1));
239                 }
240                 total_read++;
241 again:
242                 page = grab_cache_page(inode->i_mapping, i);
243                 if (!page)
244                         goto out_unlock;
245                 if (!PageUptodate(page)) {
246                         btrfs_readpage(NULL, page);
247                         lock_page(page);
248                         if (!PageUptodate(page)) {
249                                 unlock_page(page);
250                                 page_cache_release(page);
251                                 goto out_unlock;
252                         }
253                 }
254
255                 wait_on_page_writeback(page);
256
257                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
258                 page_end = page_start + PAGE_CACHE_SIZE - 1;
259                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
260
261                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
262                 if (ordered) {
263                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
264                         unlock_page(page);
265                         page_cache_release(page);
266                         btrfs_start_ordered_extent(inode, ordered, 1);
267                         btrfs_put_ordered_extent(ordered);
268                         goto again;
269                 }
270                 set_page_extent_mapped(page);
271
272                 /*
273                  * this makes sure page_mkwrite is called on the
274                  * page if it is dirtied again later
275                  */
276                 clear_page_dirty_for_io(page);
277
278                 btrfs_set_extent_delalloc(inode, page_start, page_end);
279
280                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
281                 set_page_dirty(page);
282                 unlock_page(page);
283                 page_cache_release(page);
284                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
285         }
286
287 out_unlock:
288         mutex_unlock(&inode->i_mutex);
289         return 0;
290 }
291
292 /*
293  * Called inside transaction, so use GFP_NOFS
294  */
295
296 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
297 {
298         u64 new_size;
299         u64 old_size;
300         u64 devid = 1;
301         struct btrfs_ioctl_vol_args *vol_args;
302         struct btrfs_trans_handle *trans;
303         struct btrfs_device *device = NULL;
304         char *sizestr;
305         char *devstr = NULL;
306         int ret = 0;
307         int namelen;
308         int mod = 0;
309
310         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
311
312         if (!vol_args)
313                 return -ENOMEM;
314
315         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
316                 ret = -EFAULT;
317                 goto out;
318         }
319
320         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
321         namelen = strlen(vol_args->name);
322
323         mutex_lock(&root->fs_info->volume_mutex);
324         sizestr = vol_args->name;
325         devstr = strchr(sizestr, ':');
326         if (devstr) {
327                 char *end;
328                 sizestr = devstr + 1;
329                 *devstr = '\0';
330                 devstr = vol_args->name;
331                 devid = simple_strtoull(devstr, &end, 10);
332                 printk(KERN_INFO "resizing devid %llu\n", devid);
333         }
334         device = btrfs_find_device(root, devid, NULL);
335         if (!device) {
336                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
337                 ret = -EINVAL;
338                 goto out_unlock;
339         }
340         if (!strcmp(sizestr, "max"))
341                 new_size = device->bdev->bd_inode->i_size;
342         else {
343                 if (sizestr[0] == '-') {
344                         mod = -1;
345                         sizestr++;
346                 } else if (sizestr[0] == '+') {
347                         mod = 1;
348                         sizestr++;
349                 }
350                 new_size = btrfs_parse_size(sizestr);
351                 if (new_size == 0) {
352                         ret = -EINVAL;
353                         goto out_unlock;
354                 }
355         }
356
357         old_size = device->total_bytes;
358
359         if (mod < 0) {
360                 if (new_size > old_size) {
361                         ret = -EINVAL;
362                         goto out_unlock;
363                 }
364                 new_size = old_size - new_size;
365         } else if (mod > 0) {
366                 new_size = old_size + new_size;
367         }
368
369         if (new_size < 256 * 1024 * 1024) {
370                 ret = -EINVAL;
371                 goto out_unlock;
372         }
373         if (new_size > device->bdev->bd_inode->i_size) {
374                 ret = -EFBIG;
375                 goto out_unlock;
376         }
377
378         do_div(new_size, root->sectorsize);
379         new_size *= root->sectorsize;
380
381         printk(KERN_INFO "new size for %s is %llu\n",
382                 device->name, (unsigned long long)new_size);
383
384         if (new_size > old_size) {
385                 trans = btrfs_start_transaction(root, 1);
386                 ret = btrfs_grow_device(trans, device, new_size);
387                 btrfs_commit_transaction(trans, root);
388         } else {
389                 ret = btrfs_shrink_device(device, new_size);
390         }
391
392 out_unlock:
393         mutex_unlock(&root->fs_info->volume_mutex);
394 out:
395         kfree(vol_args);
396         return ret;
397 }
398
399 static noinline int btrfs_ioctl_snap_create(struct btrfs_root *root,
400                                             void __user *arg)
401 {
402         struct btrfs_ioctl_vol_args *vol_args;
403         struct btrfs_dir_item *di;
404         struct btrfs_path *path;
405         u64 root_dirid;
406         int namelen;
407         int ret;
408
409         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
410
411         if (!vol_args)
412                 return -ENOMEM;
413
414         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
415                 ret = -EFAULT;
416                 goto out;
417         }
418
419         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
420         namelen = strlen(vol_args->name);
421         if (strchr(vol_args->name, '/')) {
422                 ret = -EINVAL;
423                 goto out;
424         }
425
426         path = btrfs_alloc_path();
427         if (!path) {
428                 ret = -ENOMEM;
429                 goto out;
430         }
431
432         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
433         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
434                             path, root_dirid,
435                             vol_args->name, namelen, 0);
436         btrfs_free_path(path);
437
438         if (di && !IS_ERR(di)) {
439                 ret = -EEXIST;
440                 goto out;
441         }
442
443         if (IS_ERR(di)) {
444                 ret = PTR_ERR(di);
445                 goto out;
446         }
447
448         mutex_lock(&root->fs_info->drop_mutex);
449         if (root == root->fs_info->tree_root)
450                 ret = create_subvol(root, vol_args->name, namelen);
451         else
452                 ret = create_snapshot(root, vol_args->name, namelen);
453         mutex_unlock(&root->fs_info->drop_mutex);
454 out:
455         kfree(vol_args);
456         return ret;
457 }
458
459 static int btrfs_ioctl_defrag(struct file *file)
460 {
461         struct inode *inode = fdentry(file)->d_inode;
462         struct btrfs_root *root = BTRFS_I(inode)->root;
463
464         switch (inode->i_mode & S_IFMT) {
465         case S_IFDIR:
466                 btrfs_defrag_root(root, 0);
467                 btrfs_defrag_root(root->fs_info->extent_root, 0);
468                 break;
469         case S_IFREG:
470                 btrfs_defrag_file(file);
471                 break;
472         }
473
474         return 0;
475 }
476
477 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
478 {
479         struct btrfs_ioctl_vol_args *vol_args;
480         int ret;
481
482         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
483
484         if (!vol_args)
485                 return -ENOMEM;
486
487         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
488                 ret = -EFAULT;
489                 goto out;
490         }
491         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
492         ret = btrfs_init_new_device(root, vol_args->name);
493
494 out:
495         kfree(vol_args);
496         return ret;
497 }
498
499 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
500 {
501         struct btrfs_ioctl_vol_args *vol_args;
502         int ret;
503
504         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
505
506         if (!vol_args)
507                 return -ENOMEM;
508
509         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
510                 ret = -EFAULT;
511                 goto out;
512         }
513         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
514         ret = btrfs_rm_device(root, vol_args->name);
515
516 out:
517         kfree(vol_args);
518         return ret;
519 }
520
521 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
522 {
523         struct inode *inode = fdentry(file)->d_inode;
524         struct btrfs_root *root = BTRFS_I(inode)->root;
525         struct file *src_file;
526         struct inode *src;
527         struct btrfs_trans_handle *trans;
528         struct btrfs_ordered_extent *ordered;
529         struct btrfs_path *path;
530         struct extent_buffer *leaf;
531         char *buf;
532         struct btrfs_key key;
533         struct btrfs_key new_key;
534         u32 size;
535         u32 nritems;
536         int slot;
537         int ret;
538
539         src_file = fget(src_fd);
540         if (!src_file)
541                 return -EBADF;
542         src = src_file->f_dentry->d_inode;
543
544         ret = -EISDIR;
545         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
546                 goto out_fput;
547
548         ret = -EXDEV;
549         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
550                 goto out_fput;
551
552         ret = -ENOMEM;
553         buf = vmalloc(btrfs_level_size(root, 0));
554         if (!buf)
555                 goto out_fput;
556
557         path = btrfs_alloc_path();
558         if (!path) {
559                 vfree(buf);
560                 goto out_fput;
561         }
562         path->reada = 2;
563
564         if (inode < src) {
565                 mutex_lock(&inode->i_mutex);
566                 mutex_lock(&src->i_mutex);
567         } else {
568                 mutex_lock(&src->i_mutex);
569                 mutex_lock(&inode->i_mutex);
570         }
571
572         ret = -ENOTEMPTY;
573         if (inode->i_size)
574                 goto out_unlock;
575
576         /* do any pending delalloc/csum calc on src, one way or
577            another, and lock file content */
578         while (1) {
579                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
580                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
581                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
582                         break;
583                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
584                 if (ordered)
585                         btrfs_put_ordered_extent(ordered);
586                 btrfs_wait_ordered_range(src, 0, (u64)-1);
587         }
588
589         trans = btrfs_start_transaction(root, 1);
590         BUG_ON(!trans);
591
592         key.objectid = src->i_ino;
593         key.type = BTRFS_EXTENT_DATA_KEY;
594         key.offset = 0;
595
596         while (1) {
597                 /*
598                  * note the key will change type as we walk through the
599                  * tree.
600                  */
601                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
602                 if (ret < 0)
603                         goto out;
604
605                 nritems = btrfs_header_nritems(path->nodes[0]);
606                 if (path->slots[0] >= nritems) {
607                         ret = btrfs_next_leaf(root, path);
608                         if (ret < 0)
609                                 goto out;
610                         if (ret > 0)
611                                 break;
612                         nritems = btrfs_header_nritems(path->nodes[0]);
613                 }
614                 leaf = path->nodes[0];
615                 slot = path->slots[0];
616
617                 btrfs_item_key_to_cpu(leaf, &key, slot);
618                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
619                     key.objectid != src->i_ino)
620                         break;
621
622                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
623                         struct btrfs_file_extent_item *extent;
624                         int found_type;
625
626                         extent = btrfs_item_ptr(leaf, slot,
627                                                 struct btrfs_file_extent_item);
628                         found_type = btrfs_file_extent_type(leaf, extent);
629                         if (found_type == BTRFS_FILE_EXTENT_REG) {
630                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
631                                                                        extent);
632                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
633                                                                  extent);
634                                 /* ds == 0 means there's a hole */
635                                 if (ds != 0) {
636                                         ret = btrfs_inc_extent_ref(trans, root,
637                                                      ds, dl,
638                                                      root->root_key.objectid,
639                                                      trans->transid,
640                                                      inode->i_ino, key.offset);
641                                         if (ret)
642                                                 goto out;
643                                 }
644                         }
645                 }
646
647                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY ||
648                     btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
649                         size = btrfs_item_size_nr(leaf, slot);
650                         read_extent_buffer(leaf, buf,
651                                            btrfs_item_ptr_offset(leaf, slot),
652                                            size);
653                         btrfs_release_path(root, path);
654                         memcpy(&new_key, &key, sizeof(new_key));
655                         new_key.objectid = inode->i_ino;
656                         ret = btrfs_insert_item(trans, root, &new_key,
657                                                 buf, size);
658                         BUG_ON(ret);
659                 } else {
660                         btrfs_release_path(root, path);
661                 }
662                 key.offset++;
663         }
664         ret = 0;
665 out:
666         btrfs_release_path(root, path);
667         if (ret == 0) {
668                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
669                 inode->i_blocks = src->i_blocks;
670                 btrfs_i_size_write(inode, src->i_size);
671                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
672                 ret = btrfs_update_inode(trans, root, inode);
673         }
674         btrfs_end_transaction(trans, root);
675         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
676         if (ret)
677                 vmtruncate(inode, 0);
678 out_unlock:
679         mutex_unlock(&src->i_mutex);
680         mutex_unlock(&inode->i_mutex);
681         vfree(buf);
682         btrfs_free_path(path);
683 out_fput:
684         fput(src_file);
685         return ret;
686 }
687
688 /*
689  * there are many ways the trans_start and trans_end ioctls can lead
690  * to deadlocks.  They should only be used by applications that
691  * basically own the machine, and have a very in depth understanding
692  * of all the possible deadlocks and enospc problems.
693  */
694 long btrfs_ioctl_trans_start(struct file *file)
695 {
696         struct inode *inode = fdentry(file)->d_inode;
697         struct btrfs_root *root = BTRFS_I(inode)->root;
698         struct btrfs_trans_handle *trans;
699         int ret = 0;
700
701         if (!capable(CAP_SYS_ADMIN))
702                 return -EPERM;
703
704         if (file->private_data) {
705                 ret = -EINPROGRESS;
706                 goto out;
707         }
708
709         mutex_lock(&root->fs_info->trans_mutex);
710         root->fs_info->open_ioctl_trans++;
711         mutex_unlock(&root->fs_info->trans_mutex);
712
713         trans = btrfs_start_ioctl_transaction(root, 0);
714         if (trans)
715                 file->private_data = trans;
716         else
717                 ret = -ENOMEM;
718         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
719 out:
720         return ret;
721 }
722
723 /*
724  * there are many ways the trans_start and trans_end ioctls can lead
725  * to deadlocks.  They should only be used by applications that
726  * basically own the machine, and have a very in depth understanding
727  * of all the possible deadlocks and enospc problems.
728  */
729 long btrfs_ioctl_trans_end(struct file *file)
730 {
731         struct inode *inode = fdentry(file)->d_inode;
732         struct btrfs_root *root = BTRFS_I(inode)->root;
733         struct btrfs_trans_handle *trans;
734         int ret = 0;
735
736         trans = file->private_data;
737         if (!trans) {
738                 ret = -EINVAL;
739                 goto out;
740         }
741         btrfs_end_transaction(trans, root);
742         file->private_data = 0;
743
744         mutex_lock(&root->fs_info->trans_mutex);
745         root->fs_info->open_ioctl_trans--;
746         mutex_unlock(&root->fs_info->trans_mutex);
747
748 out:
749         return ret;
750 }
751
752 long btrfs_ioctl(struct file *file, unsigned int
753                 cmd, unsigned long arg)
754 {
755         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
756
757         switch (cmd) {
758         case BTRFS_IOC_SNAP_CREATE:
759                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
760         case BTRFS_IOC_DEFRAG:
761                 return btrfs_ioctl_defrag(file);
762         case BTRFS_IOC_RESIZE:
763                 return btrfs_ioctl_resize(root, (void __user *)arg);
764         case BTRFS_IOC_ADD_DEV:
765                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
766         case BTRFS_IOC_RM_DEV:
767                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
768         case BTRFS_IOC_BALANCE:
769                 return btrfs_balance(root->fs_info->dev_root);
770         case BTRFS_IOC_CLONE:
771                 return btrfs_ioctl_clone(file, arg);
772         case BTRFS_IOC_TRANS_START:
773                 return btrfs_ioctl_trans_start(file);
774         case BTRFS_IOC_TRANS_END:
775                 return btrfs_ioctl_trans_end(file);
776         case BTRFS_IOC_SYNC:
777                 btrfs_start_delalloc_inodes(root);
778                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
779                 return 0;
780         }
781
782         return -ENOTTY;
783 }