Btrfs: Start btree concurrency work.
[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 "ctree.h"
40 #include "disk-io.h"
41 #include "transaction.h"
42 #include "btrfs_inode.h"
43 #include "ioctl.h"
44 #include "print-tree.h"
45 #include "volumes.h"
46 #include "locking.h"
47
48
49
50 static noinline int create_subvol(struct btrfs_root *root, char *name,
51                                   int namelen)
52 {
53         struct btrfs_trans_handle *trans;
54         struct btrfs_key key;
55         struct btrfs_root_item root_item;
56         struct btrfs_inode_item *inode_item;
57         struct extent_buffer *leaf;
58         struct btrfs_root *new_root = root;
59         struct inode *dir;
60         int ret;
61         int err;
62         u64 objectid;
63         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
64         unsigned long nr = 1;
65
66         mutex_lock(&root->fs_info->fs_mutex);
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                 return PTR_ERR(leaf);
84
85         btrfs_set_header_nritems(leaf, 0);
86         btrfs_set_header_level(leaf, 0);
87         btrfs_set_header_bytenr(leaf, leaf->start);
88         btrfs_set_header_generation(leaf, trans->transid);
89         btrfs_set_header_owner(leaf, objectid);
90
91         write_extent_buffer(leaf, root->fs_info->fsid,
92                             (unsigned long)btrfs_header_fsid(leaf),
93                             BTRFS_FSID_SIZE);
94         btrfs_mark_buffer_dirty(leaf);
95
96         inode_item = &root_item.inode;
97         memset(inode_item, 0, sizeof(*inode_item));
98         inode_item->generation = cpu_to_le64(1);
99         inode_item->size = cpu_to_le64(3);
100         inode_item->nlink = cpu_to_le32(1);
101         inode_item->nblocks = cpu_to_le64(1);
102         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
103
104         btrfs_set_root_bytenr(&root_item, leaf->start);
105         btrfs_set_root_level(&root_item, 0);
106         btrfs_set_root_refs(&root_item, 1);
107         btrfs_set_root_used(&root_item, 0);
108
109         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
110         root_item.drop_level = 0;
111
112         btrfs_tree_unlock(leaf);
113         free_extent_buffer(leaf);
114         leaf = NULL;
115
116         btrfs_set_root_dirid(&root_item, new_dirid);
117
118         key.objectid = objectid;
119         key.offset = 1;
120         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
121         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
122                                 &root_item);
123         if (ret)
124                 goto fail;
125
126         /*
127          * insert the directory item
128          */
129         key.offset = (u64)-1;
130         dir = root->fs_info->sb->s_root->d_inode;
131         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
132                                     name, namelen, dir->i_ino, &key,
133                                     BTRFS_FT_DIR);
134         if (ret)
135                 goto fail;
136
137         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
138                              name, namelen, objectid,
139                              root->fs_info->sb->s_root->d_inode->i_ino);
140         if (ret)
141                 goto fail;
142
143         ret = btrfs_commit_transaction(trans, root);
144         if (ret)
145                 goto fail_commit;
146
147         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
148         BUG_ON(!new_root);
149
150         trans = btrfs_start_transaction(new_root, 1);
151         BUG_ON(!trans);
152
153         ret = btrfs_create_subvol_root(new_root, trans, new_dirid,
154                                        BTRFS_I(dir)->block_group);
155         if (ret)
156                 goto fail;
157
158         /* Invalidate existing dcache entry for new subvolume. */
159         btrfs_invalidate_dcache_root(root, name, namelen);
160
161 fail:
162         nr = trans->blocks_used;
163         err = btrfs_commit_transaction(trans, new_root);
164         if (err && !ret)
165                 ret = err;
166 fail_commit:
167         mutex_unlock(&root->fs_info->fs_mutex);
168         btrfs_btree_balance_dirty(root, nr);
169         btrfs_throttle(root);
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         mutex_lock(&root->fs_info->fs_mutex);
185         ret = btrfs_check_free_space(root, 1, 0);
186         if (ret)
187                 goto fail_unlock;
188
189         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
190         if (!pending_snapshot) {
191                 ret = -ENOMEM;
192                 goto fail_unlock;
193         }
194         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
195         if (!pending_snapshot->name) {
196                 ret = -ENOMEM;
197                 kfree(pending_snapshot);
198                 goto fail_unlock;
199         }
200         memcpy(pending_snapshot->name, name, namelen);
201         pending_snapshot->name[namelen] = '\0';
202         trans = btrfs_start_transaction(root, 1);
203         BUG_ON(!trans);
204         pending_snapshot->root = root;
205         list_add(&pending_snapshot->list,
206                  &trans->transaction->pending_snapshots);
207         ret = btrfs_update_inode(trans, root, root->inode);
208         err = btrfs_commit_transaction(trans, root);
209
210 fail_unlock:
211         mutex_unlock(&root->fs_info->fs_mutex);
212         btrfs_btree_balance_dirty(root, nr);
213         btrfs_throttle(root);
214         return ret;
215 }
216
217 int btrfs_defrag_file(struct file *file)
218 {
219         struct inode *inode = fdentry(file)->d_inode;
220         struct btrfs_root *root = BTRFS_I(inode)->root;
221         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
222         struct page *page;
223         unsigned long last_index;
224         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
225         unsigned long total_read = 0;
226         u64 page_start;
227         u64 page_end;
228         unsigned long i;
229         int ret;
230
231         mutex_lock(&root->fs_info->fs_mutex);
232         ret = btrfs_check_free_space(root, inode->i_size, 0);
233         mutex_unlock(&root->fs_info->fs_mutex);
234         if (ret)
235                 return -ENOSPC;
236
237         mutex_lock(&inode->i_mutex);
238         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
239         for (i = 0; i <= last_index; i++) {
240                 if (total_read % ra_pages == 0) {
241                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
242                                        min(last_index, i + ra_pages - 1));
243                 }
244                 total_read++;
245                 page = grab_cache_page(inode->i_mapping, i);
246                 if (!page)
247                         goto out_unlock;
248                 if (!PageUptodate(page)) {
249                         btrfs_readpage(NULL, page);
250                         lock_page(page);
251                         if (!PageUptodate(page)) {
252                                 unlock_page(page);
253                                 page_cache_release(page);
254                                 goto out_unlock;
255                         }
256                 }
257
258 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
259                 ClearPageDirty(page);
260 #else
261                 cancel_dirty_page(page, PAGE_CACHE_SIZE);
262 #endif
263                 wait_on_page_writeback(page);
264                 set_page_extent_mapped(page);
265
266                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
267                 page_end = page_start + PAGE_CACHE_SIZE - 1;
268
269                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
270                 set_extent_delalloc(io_tree, page_start,
271                                     page_end, GFP_NOFS);
272
273                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
274                 set_page_dirty(page);
275                 unlock_page(page);
276                 page_cache_release(page);
277                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
278         }
279
280 out_unlock:
281         mutex_unlock(&inode->i_mutex);
282         return 0;
283 }
284
285 /*
286  * Called inside transaction, so use GFP_NOFS
287  */
288
289 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
290 {
291         u64 new_size;
292         u64 old_size;
293         u64 devid = 1;
294         struct btrfs_ioctl_vol_args *vol_args;
295         struct btrfs_trans_handle *trans;
296         struct btrfs_device *device = NULL;
297         char *sizestr;
298         char *devstr = NULL;
299         int ret = 0;
300         int namelen;
301         int mod = 0;
302
303         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
304
305         if (!vol_args)
306                 return -ENOMEM;
307
308         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
309                 ret = -EFAULT;
310                 goto out;
311         }
312         namelen = strlen(vol_args->name);
313         if (namelen > BTRFS_VOL_NAME_MAX) {
314                 ret = -EINVAL;
315                 goto out;
316         }
317
318         mutex_lock(&root->fs_info->fs_mutex);
319         sizestr = vol_args->name;
320         devstr = strchr(sizestr, ':');
321         if (devstr) {
322                 char *end;
323                 sizestr = devstr + 1;
324                 *devstr = '\0';
325                 devstr = vol_args->name;
326                 devid = simple_strtoull(devstr, &end, 10);
327                 printk(KERN_INFO "resizing devid %llu\n", devid);
328         }
329         device = btrfs_find_device(root, devid, NULL);
330         if (!device) {
331                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
332                 ret = -EINVAL;
333                 goto out_unlock;
334         }
335         if (!strcmp(sizestr, "max"))
336                 new_size = device->bdev->bd_inode->i_size;
337         else {
338                 if (sizestr[0] == '-') {
339                         mod = -1;
340                         sizestr++;
341                 } else if (sizestr[0] == '+') {
342                         mod = 1;
343                         sizestr++;
344                 }
345                 new_size = btrfs_parse_size(sizestr);
346                 if (new_size == 0) {
347                         ret = -EINVAL;
348                         goto out_unlock;
349                 }
350         }
351
352         old_size = device->total_bytes;
353
354         if (mod < 0) {
355                 if (new_size > old_size) {
356                         ret = -EINVAL;
357                         goto out_unlock;
358                 }
359                 new_size = old_size - new_size;
360         } else if (mod > 0) {
361                 new_size = old_size + new_size;
362         }
363
364         if (new_size < 256 * 1024 * 1024) {
365                 ret = -EINVAL;
366                 goto out_unlock;
367         }
368         if (new_size > device->bdev->bd_inode->i_size) {
369                 ret = -EFBIG;
370                 goto out_unlock;
371         }
372
373         do_div(new_size, root->sectorsize);
374         new_size *= root->sectorsize;
375
376         printk(KERN_INFO "new size for %s is %llu\n",
377                 device->name, (unsigned long long)new_size);
378
379         if (new_size > old_size) {
380                 trans = btrfs_start_transaction(root, 1);
381                 ret = btrfs_grow_device(trans, device, new_size);
382                 btrfs_commit_transaction(trans, root);
383         } else {
384                 ret = btrfs_shrink_device(device, new_size);
385         }
386
387 out_unlock:
388         mutex_unlock(&root->fs_info->fs_mutex);
389 out:
390         kfree(vol_args);
391         return ret;
392 }
393
394 static noinline int btrfs_ioctl_snap_create(struct btrfs_root *root,
395                                             void __user *arg)
396 {
397         struct btrfs_ioctl_vol_args *vol_args;
398         struct btrfs_dir_item *di;
399         struct btrfs_path *path;
400         u64 root_dirid;
401         int namelen;
402         int ret;
403
404         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
405
406         if (!vol_args)
407                 return -ENOMEM;
408
409         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
410                 ret = -EFAULT;
411                 goto out;
412         }
413
414         namelen = strlen(vol_args->name);
415         if (namelen > BTRFS_VOL_NAME_MAX) {
416                 ret = -EINVAL;
417                 goto out;
418         }
419         if (strchr(vol_args->name, '/')) {
420                 ret = -EINVAL;
421                 goto out;
422         }
423
424         path = btrfs_alloc_path();
425         if (!path) {
426                 ret = -ENOMEM;
427                 goto out;
428         }
429
430         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
431         mutex_lock(&root->fs_info->fs_mutex);
432         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
433                             path, root_dirid,
434                             vol_args->name, namelen, 0);
435         mutex_unlock(&root->fs_info->fs_mutex);
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         if (root == root->fs_info->tree_root)
449                 ret = create_subvol(root, vol_args->name, namelen);
450         else
451                 ret = create_snapshot(root, vol_args->name, namelen);
452 out:
453         kfree(vol_args);
454         return ret;
455 }
456
457 static int btrfs_ioctl_defrag(struct file *file)
458 {
459         struct inode *inode = fdentry(file)->d_inode;
460         struct btrfs_root *root = BTRFS_I(inode)->root;
461
462         switch (inode->i_mode & S_IFMT) {
463         case S_IFDIR:
464                 mutex_lock(&root->fs_info->fs_mutex);
465                 btrfs_defrag_root(root, 0);
466                 btrfs_defrag_root(root->fs_info->extent_root, 0);
467                 mutex_unlock(&root->fs_info->fs_mutex);
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         ret = btrfs_init_new_device(root, vol_args->name);
492
493 out:
494         kfree(vol_args);
495         return ret;
496 }
497
498 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
499 {
500         struct btrfs_ioctl_vol_args *vol_args;
501         int ret;
502
503         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
504
505         if (!vol_args)
506                 return -ENOMEM;
507
508         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
509                 ret = -EFAULT;
510                 goto out;
511         }
512         ret = btrfs_rm_device(root, vol_args->name);
513
514 out:
515         kfree(vol_args);
516         return ret;
517 }
518
519 int dup_item_to_inode(struct btrfs_trans_handle *trans,
520                        struct btrfs_root *root,
521                        struct btrfs_path *path,
522                        struct extent_buffer *leaf,
523                        int slot,
524                        struct btrfs_key *key,
525                        u64 destino)
526 {
527         char *dup;
528         int len = btrfs_item_size_nr(leaf, slot);
529         struct btrfs_key ckey = *key;
530         int ret = 0;
531
532         dup = kmalloc(len, GFP_NOFS);
533         if (!dup)
534                 return -ENOMEM;
535
536         read_extent_buffer(leaf, dup, btrfs_item_ptr_offset(leaf, slot), len);
537         btrfs_release_path(root, path);
538
539         ckey.objectid = destino;
540         ret = btrfs_insert_item(trans, root, &ckey, dup, len);
541         kfree(dup);
542         return ret;
543 }
544
545 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
546 {
547         struct inode *inode = fdentry(file)->d_inode;
548         struct btrfs_root *root = BTRFS_I(inode)->root;
549         struct file *src_file;
550         struct inode *src;
551         struct btrfs_trans_handle *trans;
552         int ret;
553         u64 pos;
554         struct btrfs_path *path;
555         struct btrfs_key key;
556         struct extent_buffer *leaf;
557         u32 nritems;
558         int slot;
559
560         src_file = fget(src_fd);
561         if (!src_file)
562                 return -EBADF;
563         src = src_file->f_dentry->d_inode;
564
565         ret = -EXDEV;
566         if (src->i_sb != inode->i_sb)
567                 goto out_fput;
568
569         if (inode < src) {
570                 mutex_lock(&inode->i_mutex);
571                 mutex_lock(&src->i_mutex);
572         } else {
573                 mutex_lock(&src->i_mutex);
574                 mutex_lock(&inode->i_mutex);
575         }
576
577         ret = -ENOTEMPTY;
578         if (inode->i_size)
579                 goto out_unlock;
580
581         /* do any pending delalloc/csum calc on src, one way or
582            another, and lock file content */
583         while (1) {
584                 filemap_write_and_wait(src->i_mapping);
585                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
586                 if (BTRFS_I(src)->delalloc_bytes == 0)
587                         break;
588                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
589         }
590
591         mutex_lock(&root->fs_info->fs_mutex);
592         trans = btrfs_start_transaction(root, 0);
593         path = btrfs_alloc_path();
594         if (!path) {
595                 ret = -ENOMEM;
596                 goto out;
597         }
598         key.offset = 0;
599         key.type = BTRFS_EXTENT_DATA_KEY;
600         key.objectid = src->i_ino;
601         pos = 0;
602         path->reada = 2;
603
604         while (1) {
605                 /*
606                  * note the key will change type as we walk through the
607                  * tree.
608                  */
609                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
610                 if (ret < 0)
611                         goto out;
612
613                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
614                         ret = btrfs_next_leaf(root, path);
615                         if (ret < 0)
616                                 goto out;
617                         if (ret > 0)
618                                 break;
619                 }
620                 leaf = path->nodes[0];
621                 slot = path->slots[0];
622                 btrfs_item_key_to_cpu(leaf, &key, slot);
623                 nritems = btrfs_header_nritems(leaf);
624
625                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
626                     key.objectid != src->i_ino)
627                         break;
628
629                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
630                         struct btrfs_file_extent_item *extent;
631                         int found_type;
632                         pos = key.offset;
633                         extent = btrfs_item_ptr(leaf, slot,
634                                                 struct btrfs_file_extent_item);
635                         found_type = btrfs_file_extent_type(leaf, extent);
636                         if (found_type == BTRFS_FILE_EXTENT_REG) {
637                                 u64 len = btrfs_file_extent_num_bytes(leaf,
638                                                                       extent);
639                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
640                                                                        extent);
641                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
642                                                                  extent);
643                                 u64 off = btrfs_file_extent_offset(leaf,
644                                                                    extent);
645                                 btrfs_insert_file_extent(trans, root,
646                                                          inode->i_ino, pos,
647                                                          ds, dl, len, off);
648                                 /* ds == 0 means there's a hole */
649                                 if (ds != 0) {
650                                         btrfs_inc_extent_ref(trans, root,
651                                                      ds, dl,
652                                                      root->root_key.objectid,
653                                                      trans->transid,
654                                                      inode->i_ino, pos);
655                                 }
656                                 pos = key.offset + len;
657                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
658                                 ret = dup_item_to_inode(trans, root, path,
659                                                         leaf, slot, &key,
660                                                         inode->i_ino);
661                                 if (ret)
662                                         goto out;
663                                 pos = key.offset + btrfs_item_size_nr(leaf,
664                                                                       slot);
665                         }
666                 } else if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
667                         ret = dup_item_to_inode(trans, root, path, leaf,
668                                                 slot, &key, inode->i_ino);
669
670                         if (ret)
671                                 goto out;
672                 }
673                 key.offset++;
674                 btrfs_release_path(root, path);
675         }
676
677         ret = 0;
678 out:
679         btrfs_free_path(path);
680
681         inode->i_blocks = src->i_blocks;
682         i_size_write(inode, src->i_size);
683         btrfs_update_inode(trans, root, inode);
684
685         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
686
687         btrfs_end_transaction(trans, root);
688         mutex_unlock(&root->fs_info->fs_mutex);
689
690 out_unlock:
691         mutex_unlock(&src->i_mutex);
692         mutex_unlock(&inode->i_mutex);
693 out_fput:
694         fput(src_file);
695         return ret;
696 }
697
698 /*
699  * there are many ways the trans_start and trans_end ioctls can lead
700  * to deadlocks.  They should only be used by applications that
701  * basically own the machine, and have a very in depth understanding
702  * of all the possible deadlocks and enospc problems.
703  */
704 long btrfs_ioctl_trans_start(struct file *file)
705 {
706         struct inode *inode = fdentry(file)->d_inode;
707         struct btrfs_root *root = BTRFS_I(inode)->root;
708         struct btrfs_trans_handle *trans;
709         int ret = 0;
710
711         if (!capable(CAP_SYS_ADMIN))
712                 return -EPERM;
713
714         mutex_lock(&root->fs_info->fs_mutex);
715         if (file->private_data) {
716                 ret = -EINPROGRESS;
717                 goto out;
718         }
719         trans = btrfs_start_transaction(root, 0);
720         if (trans)
721                 file->private_data = trans;
722         else
723                 ret = -ENOMEM;
724         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
725 out:
726         mutex_unlock(&root->fs_info->fs_mutex);
727         return ret;
728 }
729
730 /*
731  * there are many ways the trans_start and trans_end ioctls can lead
732  * to deadlocks.  They should only be used by applications that
733  * basically own the machine, and have a very in depth understanding
734  * of all the possible deadlocks and enospc problems.
735  */
736 long btrfs_ioctl_trans_end(struct file *file)
737 {
738         struct inode *inode = fdentry(file)->d_inode;
739         struct btrfs_root *root = BTRFS_I(inode)->root;
740         struct btrfs_trans_handle *trans;
741         int ret = 0;
742
743         mutex_lock(&root->fs_info->fs_mutex);
744         trans = file->private_data;
745         if (!trans) {
746                 ret = -EINVAL;
747                 goto out;
748         }
749         btrfs_end_transaction(trans, root);
750         file->private_data = 0;
751 out:
752         mutex_unlock(&root->fs_info->fs_mutex);
753         return ret;
754 }
755
756 long btrfs_ioctl(struct file *file, unsigned int
757                 cmd, unsigned long arg)
758 {
759         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
760
761         switch (cmd) {
762         case BTRFS_IOC_SNAP_CREATE:
763                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
764         case BTRFS_IOC_DEFRAG:
765                 return btrfs_ioctl_defrag(file);
766         case BTRFS_IOC_RESIZE:
767                 return btrfs_ioctl_resize(root, (void __user *)arg);
768         case BTRFS_IOC_ADD_DEV:
769                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
770         case BTRFS_IOC_RM_DEV:
771                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
772         case BTRFS_IOC_BALANCE:
773                 return btrfs_balance(root->fs_info->dev_root);
774         case BTRFS_IOC_CLONE:
775                 return btrfs_ioctl_clone(file, arg);
776         case BTRFS_IOC_TRANS_START:
777                 return btrfs_ioctl_trans_start(file);
778         case BTRFS_IOC_TRANS_END:
779                 return btrfs_ioctl_trans_end(file);
780         case BTRFS_IOC_SYNC:
781                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
782                 return 0;
783         }
784
785         return -ENOTTY;
786 }