Btrfs: Add fallocate support v2
[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 "ctree.h"
45 #include "disk-io.h"
46 #include "transaction.h"
47 #include "btrfs_inode.h"
48 #include "ioctl.h"
49 #include "print-tree.h"
50 #include "volumes.h"
51 #include "locking.h"
52
53
54
55 static noinline int create_subvol(struct btrfs_root *root,
56                                   struct dentry *dentry,
57                                   char *name, int namelen)
58 {
59         struct btrfs_trans_handle *trans;
60         struct btrfs_key key;
61         struct btrfs_root_item root_item;
62         struct btrfs_inode_item *inode_item;
63         struct extent_buffer *leaf;
64         struct btrfs_root *new_root = root;
65         struct inode *dir;
66         int ret;
67         int err;
68         u64 objectid;
69         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
70         unsigned long nr = 1;
71
72         ret = btrfs_check_free_space(root, 1, 0);
73         if (ret)
74                 goto fail_commit;
75
76         trans = btrfs_start_transaction(root, 1);
77         BUG_ON(!trans);
78
79         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
80                                        0, &objectid);
81         if (ret)
82                 goto fail;
83
84         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
85                                       objectid, trans->transid, 0, 0, 0);
86         if (IS_ERR(leaf)) {
87                 ret = PTR_ERR(leaf);
88                 goto fail;
89         }
90
91         btrfs_set_header_nritems(leaf, 0);
92         btrfs_set_header_level(leaf, 0);
93         btrfs_set_header_bytenr(leaf, leaf->start);
94         btrfs_set_header_generation(leaf, trans->transid);
95         btrfs_set_header_owner(leaf, objectid);
96
97         write_extent_buffer(leaf, root->fs_info->fsid,
98                             (unsigned long)btrfs_header_fsid(leaf),
99                             BTRFS_FSID_SIZE);
100         btrfs_mark_buffer_dirty(leaf);
101
102         inode_item = &root_item.inode;
103         memset(inode_item, 0, sizeof(*inode_item));
104         inode_item->generation = cpu_to_le64(1);
105         inode_item->size = cpu_to_le64(3);
106         inode_item->nlink = cpu_to_le32(1);
107         inode_item->nbytes = cpu_to_le64(root->leafsize);
108         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
109
110         btrfs_set_root_bytenr(&root_item, leaf->start);
111         btrfs_set_root_generation(&root_item, trans->transid);
112         btrfs_set_root_level(&root_item, 0);
113         btrfs_set_root_refs(&root_item, 1);
114         btrfs_set_root_used(&root_item, 0);
115         btrfs_set_root_last_snapshot(&root_item, 0);
116
117         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
118         root_item.drop_level = 0;
119
120         btrfs_tree_unlock(leaf);
121         free_extent_buffer(leaf);
122         leaf = NULL;
123
124         btrfs_set_root_dirid(&root_item, new_dirid);
125
126         key.objectid = objectid;
127         key.offset = 1;
128         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
129         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
130                                 &root_item);
131         if (ret)
132                 goto fail;
133
134         /*
135          * insert the directory item
136          */
137         key.offset = (u64)-1;
138         dir = root->fs_info->sb->s_root->d_inode;
139         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
140                                     name, namelen, dir->i_ino, &key,
141                                     BTRFS_FT_DIR, 0);
142         if (ret)
143                 goto fail;
144
145         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
146                              name, namelen, objectid,
147                              root->fs_info->sb->s_root->d_inode->i_ino, 0);
148         if (ret)
149                 goto fail;
150
151         ret = btrfs_commit_transaction(trans, root);
152         if (ret)
153                 goto fail_commit;
154
155         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
156         BUG_ON(!new_root);
157
158         trans = btrfs_start_transaction(new_root, 1);
159         BUG_ON(!trans);
160
161         ret = btrfs_create_subvol_root(new_root, dentry, trans, new_dirid,
162                                        BTRFS_I(dir)->block_group);
163         if (ret)
164                 goto fail;
165
166 fail:
167         nr = trans->blocks_used;
168         err = btrfs_commit_transaction(trans, new_root);
169         if (err && !ret)
170                 ret = err;
171 fail_commit:
172         btrfs_btree_balance_dirty(root, nr);
173         return ret;
174 }
175
176 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
177 {
178         struct btrfs_pending_snapshot *pending_snapshot;
179         struct btrfs_trans_handle *trans;
180         int ret;
181         int err;
182         unsigned long nr = 0;
183
184         if (!root->ref_cows)
185                 return -EINVAL;
186
187         ret = btrfs_check_free_space(root, 1, 0);
188         if (ret)
189                 goto fail_unlock;
190
191         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
192         if (!pending_snapshot) {
193                 ret = -ENOMEM;
194                 goto fail_unlock;
195         }
196         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
197         if (!pending_snapshot->name) {
198                 ret = -ENOMEM;
199                 kfree(pending_snapshot);
200                 goto fail_unlock;
201         }
202         memcpy(pending_snapshot->name, name, namelen);
203         pending_snapshot->name[namelen] = '\0';
204         trans = btrfs_start_transaction(root, 1);
205         BUG_ON(!trans);
206         pending_snapshot->root = root;
207         list_add(&pending_snapshot->list,
208                  &trans->transaction->pending_snapshots);
209         ret = btrfs_update_inode(trans, root, root->inode);
210         err = btrfs_commit_transaction(trans, root);
211
212 fail_unlock:
213         btrfs_btree_balance_dirty(root, nr);
214         return ret;
215 }
216
217 /* copy of may_create in fs/namei.c() */
218 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
219 {
220         if (child->d_inode)
221                 return -EEXIST;
222         if (IS_DEADDIR(dir))
223                 return -ENOENT;
224         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
225 }
226
227 /*
228  * Create a new subvolume below @parent.  This is largely modeled after
229  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
230  * inside this filesystem so it's quite a bit simpler.
231  */
232 static noinline int btrfs_mksubvol(struct path *parent, char *name,
233                                    int mode, int namelen)
234 {
235         struct dentry *dentry;
236         int error;
237
238         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
239
240         dentry = lookup_one_len(name, parent->dentry, namelen);
241         error = PTR_ERR(dentry);
242         if (IS_ERR(dentry))
243                 goto out_unlock;
244
245         error = -EEXIST;
246         if (dentry->d_inode)
247                 goto out_dput;
248
249         if (!IS_POSIXACL(parent->dentry->d_inode))
250                 mode &= ~current->fs->umask;
251         error = mnt_want_write(parent->mnt);
252         if (error)
253                 goto out_dput;
254
255         error = btrfs_may_create(parent->dentry->d_inode, dentry);
256         if (error)
257                 goto out_drop_write;
258
259         /*
260          * Actually perform the low-level subvolume creation after all
261          * this VFS fuzz.
262          *
263          * Eventually we want to pass in an inode under which we create this
264          * subvolume, but for now all are under the filesystem root.
265          *
266          * Also we should pass on the mode eventually to allow creating new
267          * subvolume with specific mode bits.
268          */
269         error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root, dentry,
270                               name, namelen);
271         if (error)
272                 goto out_drop_write;
273
274         fsnotify_mkdir(parent->dentry->d_inode, dentry);
275 out_drop_write:
276         mnt_drop_write(parent->mnt);
277 out_dput:
278         dput(dentry);
279 out_unlock:
280         mutex_unlock(&parent->dentry->d_inode->i_mutex);
281         return error;
282 }
283
284
285 int btrfs_defrag_file(struct file *file)
286 {
287         struct inode *inode = fdentry(file)->d_inode;
288         struct btrfs_root *root = BTRFS_I(inode)->root;
289         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
290         struct btrfs_ordered_extent *ordered;
291         struct page *page;
292         unsigned long last_index;
293         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
294         unsigned long total_read = 0;
295         u64 page_start;
296         u64 page_end;
297         unsigned long i;
298         int ret;
299
300         ret = btrfs_check_free_space(root, inode->i_size, 0);
301         if (ret)
302                 return -ENOSPC;
303
304         mutex_lock(&inode->i_mutex);
305         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
306         for (i = 0; i <= last_index; i++) {
307                 if (total_read % ra_pages == 0) {
308                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
309                                        min(last_index, i + ra_pages - 1));
310                 }
311                 total_read++;
312 again:
313                 page = grab_cache_page(inode->i_mapping, i);
314                 if (!page)
315                         goto out_unlock;
316                 if (!PageUptodate(page)) {
317                         btrfs_readpage(NULL, page);
318                         lock_page(page);
319                         if (!PageUptodate(page)) {
320                                 unlock_page(page);
321                                 page_cache_release(page);
322                                 goto out_unlock;
323                         }
324                 }
325
326                 wait_on_page_writeback(page);
327
328                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
329                 page_end = page_start + PAGE_CACHE_SIZE - 1;
330                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
331
332                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
333                 if (ordered) {
334                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
335                         unlock_page(page);
336                         page_cache_release(page);
337                         btrfs_start_ordered_extent(inode, ordered, 1);
338                         btrfs_put_ordered_extent(ordered);
339                         goto again;
340                 }
341                 set_page_extent_mapped(page);
342
343                 /*
344                  * this makes sure page_mkwrite is called on the
345                  * page if it is dirtied again later
346                  */
347                 clear_page_dirty_for_io(page);
348
349                 btrfs_set_extent_delalloc(inode, page_start, page_end);
350
351                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
352                 set_page_dirty(page);
353                 unlock_page(page);
354                 page_cache_release(page);
355                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
356         }
357
358 out_unlock:
359         mutex_unlock(&inode->i_mutex);
360         return 0;
361 }
362
363 /*
364  * Called inside transaction, so use GFP_NOFS
365  */
366
367 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
368 {
369         u64 new_size;
370         u64 old_size;
371         u64 devid = 1;
372         struct btrfs_ioctl_vol_args *vol_args;
373         struct btrfs_trans_handle *trans;
374         struct btrfs_device *device = NULL;
375         char *sizestr;
376         char *devstr = NULL;
377         int ret = 0;
378         int namelen;
379         int mod = 0;
380
381         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
382
383         if (!vol_args)
384                 return -ENOMEM;
385
386         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
387                 ret = -EFAULT;
388                 goto out;
389         }
390
391         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
392         namelen = strlen(vol_args->name);
393
394         mutex_lock(&root->fs_info->volume_mutex);
395         sizestr = vol_args->name;
396         devstr = strchr(sizestr, ':');
397         if (devstr) {
398                 char *end;
399                 sizestr = devstr + 1;
400                 *devstr = '\0';
401                 devstr = vol_args->name;
402                 devid = simple_strtoull(devstr, &end, 10);
403                 printk(KERN_INFO "resizing devid %llu\n", devid);
404         }
405         device = btrfs_find_device(root, devid, NULL);
406         if (!device) {
407                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
408                 ret = -EINVAL;
409                 goto out_unlock;
410         }
411         if (!strcmp(sizestr, "max"))
412                 new_size = device->bdev->bd_inode->i_size;
413         else {
414                 if (sizestr[0] == '-') {
415                         mod = -1;
416                         sizestr++;
417                 } else if (sizestr[0] == '+') {
418                         mod = 1;
419                         sizestr++;
420                 }
421                 new_size = btrfs_parse_size(sizestr);
422                 if (new_size == 0) {
423                         ret = -EINVAL;
424                         goto out_unlock;
425                 }
426         }
427
428         old_size = device->total_bytes;
429
430         if (mod < 0) {
431                 if (new_size > old_size) {
432                         ret = -EINVAL;
433                         goto out_unlock;
434                 }
435                 new_size = old_size - new_size;
436         } else if (mod > 0) {
437                 new_size = old_size + new_size;
438         }
439
440         if (new_size < 256 * 1024 * 1024) {
441                 ret = -EINVAL;
442                 goto out_unlock;
443         }
444         if (new_size > device->bdev->bd_inode->i_size) {
445                 ret = -EFBIG;
446                 goto out_unlock;
447         }
448
449         do_div(new_size, root->sectorsize);
450         new_size *= root->sectorsize;
451
452         printk(KERN_INFO "new size for %s is %llu\n",
453                 device->name, (unsigned long long)new_size);
454
455         if (new_size > old_size) {
456                 trans = btrfs_start_transaction(root, 1);
457                 ret = btrfs_grow_device(trans, device, new_size);
458                 btrfs_commit_transaction(trans, root);
459         } else {
460                 ret = btrfs_shrink_device(device, new_size);
461         }
462
463 out_unlock:
464         mutex_unlock(&root->fs_info->volume_mutex);
465 out:
466         kfree(vol_args);
467         return ret;
468 }
469
470 static noinline int btrfs_ioctl_snap_create(struct file *file,
471                                             void __user *arg)
472 {
473         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
474         struct btrfs_ioctl_vol_args *vol_args;
475         struct btrfs_dir_item *di;
476         struct btrfs_path *path;
477         u64 root_dirid;
478         int namelen;
479         int ret;
480
481         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
482
483         if (!vol_args)
484                 return -ENOMEM;
485
486         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
487                 ret = -EFAULT;
488                 goto out;
489         }
490
491         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
492         namelen = strlen(vol_args->name);
493         if (strchr(vol_args->name, '/')) {
494                 ret = -EINVAL;
495                 goto out;
496         }
497
498         path = btrfs_alloc_path();
499         if (!path) {
500                 ret = -ENOMEM;
501                 goto out;
502         }
503
504         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
505         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
506                             path, root_dirid,
507                             vol_args->name, namelen, 0);
508         btrfs_free_path(path);
509
510         if (di && !IS_ERR(di)) {
511                 ret = -EEXIST;
512                 goto out;
513         }
514
515         if (IS_ERR(di)) {
516                 ret = PTR_ERR(di);
517                 goto out;
518         }
519
520         if (root == root->fs_info->tree_root) {
521                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
522                                      file->f_path.dentry->d_inode->i_mode,
523                                      namelen);
524         } else {
525                 ret = create_snapshot(root, vol_args->name, namelen);
526         }
527
528 out:
529         kfree(vol_args);
530         return ret;
531 }
532
533 static int btrfs_ioctl_defrag(struct file *file)
534 {
535         struct inode *inode = fdentry(file)->d_inode;
536         struct btrfs_root *root = BTRFS_I(inode)->root;
537
538         switch (inode->i_mode & S_IFMT) {
539         case S_IFDIR:
540                 btrfs_defrag_root(root, 0);
541                 btrfs_defrag_root(root->fs_info->extent_root, 0);
542                 break;
543         case S_IFREG:
544                 btrfs_defrag_file(file);
545                 break;
546         }
547
548         return 0;
549 }
550
551 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
552 {
553         struct btrfs_ioctl_vol_args *vol_args;
554         int ret;
555
556         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
557
558         if (!vol_args)
559                 return -ENOMEM;
560
561         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
562                 ret = -EFAULT;
563                 goto out;
564         }
565         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
566         ret = btrfs_init_new_device(root, vol_args->name);
567
568 out:
569         kfree(vol_args);
570         return ret;
571 }
572
573 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
574 {
575         struct btrfs_ioctl_vol_args *vol_args;
576         int ret;
577
578         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
579
580         if (!vol_args)
581                 return -ENOMEM;
582
583         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
584                 ret = -EFAULT;
585                 goto out;
586         }
587         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
588         ret = btrfs_rm_device(root, vol_args->name);
589
590 out:
591         kfree(vol_args);
592         return ret;
593 }
594
595 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
596 {
597         struct inode *inode = fdentry(file)->d_inode;
598         struct btrfs_root *root = BTRFS_I(inode)->root;
599         struct file *src_file;
600         struct inode *src;
601         struct btrfs_trans_handle *trans;
602         struct btrfs_path *path;
603         struct extent_buffer *leaf;
604         char *buf;
605         struct btrfs_key key;
606         u32 nritems;
607         int slot;
608         int ret;
609
610         src_file = fget(src_fd);
611         if (!src_file)
612                 return -EBADF;
613         src = src_file->f_dentry->d_inode;
614
615         ret = -EISDIR;
616         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
617                 goto out_fput;
618
619         ret = -EXDEV;
620         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
621                 goto out_fput;
622
623         ret = -ENOMEM;
624         buf = vmalloc(btrfs_level_size(root, 0));
625         if (!buf)
626                 goto out_fput;
627
628         path = btrfs_alloc_path();
629         if (!path) {
630                 vfree(buf);
631                 goto out_fput;
632         }
633         path->reada = 2;
634
635         if (inode < src) {
636                 mutex_lock(&inode->i_mutex);
637                 mutex_lock(&src->i_mutex);
638         } else {
639                 mutex_lock(&src->i_mutex);
640                 mutex_lock(&inode->i_mutex);
641         }
642
643         ret = -ENOTEMPTY;
644         if (inode->i_size)
645                 goto out_unlock;
646
647         /* do any pending delalloc/csum calc on src, one way or
648            another, and lock file content */
649         while (1) {
650                 struct btrfs_ordered_extent *ordered;
651                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
652                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
653                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
654                         break;
655                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
656                 if (ordered)
657                         btrfs_put_ordered_extent(ordered);
658                 btrfs_wait_ordered_range(src, 0, (u64)-1);
659         }
660
661         trans = btrfs_start_transaction(root, 1);
662         BUG_ON(!trans);
663
664         key.objectid = src->i_ino;
665         key.type = BTRFS_EXTENT_DATA_KEY;
666         key.offset = 0;
667
668         while (1) {
669                 /*
670                  * note the key will change type as we walk through the
671                  * tree.
672                  */
673                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
674                 if (ret < 0)
675                         goto out;
676
677                 nritems = btrfs_header_nritems(path->nodes[0]);
678                 if (path->slots[0] >= nritems) {
679                         ret = btrfs_next_leaf(root, path);
680                         if (ret < 0)
681                                 goto out;
682                         if (ret > 0)
683                                 break;
684                         nritems = btrfs_header_nritems(path->nodes[0]);
685                 }
686                 leaf = path->nodes[0];
687                 slot = path->slots[0];
688
689                 btrfs_item_key_to_cpu(leaf, &key, slot);
690                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
691                     key.objectid != src->i_ino)
692                         break;
693
694                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY ||
695                     btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
696                         u32 size;
697                         struct btrfs_key new_key;
698
699                         size = btrfs_item_size_nr(leaf, slot);
700                         read_extent_buffer(leaf, buf,
701                                            btrfs_item_ptr_offset(leaf, slot),
702                                            size);
703                         btrfs_release_path(root, path);
704
705                         memcpy(&new_key, &key, sizeof(new_key));
706                         new_key.objectid = inode->i_ino;
707                         ret = btrfs_insert_empty_item(trans, root, path,
708                                                       &new_key, size);
709                         if (ret)
710                                 goto out;
711
712                         leaf = path->nodes[0];
713                         slot = path->slots[0];
714                         write_extent_buffer(leaf, buf,
715                                             btrfs_item_ptr_offset(leaf, slot),
716                                             size);
717                         btrfs_mark_buffer_dirty(leaf);
718                 }
719
720                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
721                         struct btrfs_file_extent_item *extent;
722                         int found_type;
723
724                         extent = btrfs_item_ptr(leaf, slot,
725                                                 struct btrfs_file_extent_item);
726                         found_type = btrfs_file_extent_type(leaf, extent);
727                         if (found_type == BTRFS_FILE_EXTENT_REG ||
728                             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
729                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
730                                                                        extent);
731                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
732                                                                  extent);
733                                 /* ds == 0 means there's a hole */
734                                 if (ds != 0) {
735                                         ret = btrfs_inc_extent_ref(trans, root,
736                                                      ds, dl, leaf->start,
737                                                      root->root_key.objectid,
738                                                      trans->transid,
739                                                      inode->i_ino);
740                                         BUG_ON(ret);
741                                 }
742                         }
743                 }
744                 btrfs_release_path(root, path);
745                 key.offset++;
746         }
747         ret = 0;
748 out:
749         btrfs_release_path(root, path);
750         if (ret == 0) {
751                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
752                 inode_set_bytes(inode, inode_get_bytes(src));
753                 btrfs_i_size_write(inode, src->i_size);
754                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
755                 ret = btrfs_update_inode(trans, root, inode);
756         }
757         btrfs_end_transaction(trans, root);
758         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
759         if (ret)
760                 vmtruncate(inode, 0);
761 out_unlock:
762         mutex_unlock(&src->i_mutex);
763         mutex_unlock(&inode->i_mutex);
764         vfree(buf);
765         btrfs_free_path(path);
766 out_fput:
767         fput(src_file);
768         return ret;
769 }
770
771 /*
772  * there are many ways the trans_start and trans_end ioctls can lead
773  * to deadlocks.  They should only be used by applications that
774  * basically own the machine, and have a very in depth understanding
775  * of all the possible deadlocks and enospc problems.
776  */
777 long btrfs_ioctl_trans_start(struct file *file)
778 {
779         struct inode *inode = fdentry(file)->d_inode;
780         struct btrfs_root *root = BTRFS_I(inode)->root;
781         struct btrfs_trans_handle *trans;
782         int ret = 0;
783
784         if (!capable(CAP_SYS_ADMIN))
785                 return -EPERM;
786
787         if (file->private_data) {
788                 ret = -EINPROGRESS;
789                 goto out;
790         }
791
792         mutex_lock(&root->fs_info->trans_mutex);
793         root->fs_info->open_ioctl_trans++;
794         mutex_unlock(&root->fs_info->trans_mutex);
795
796         trans = btrfs_start_ioctl_transaction(root, 0);
797         if (trans)
798                 file->private_data = trans;
799         else
800                 ret = -ENOMEM;
801         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
802 out:
803         return ret;
804 }
805
806 /*
807  * there are many ways the trans_start and trans_end ioctls can lead
808  * to deadlocks.  They should only be used by applications that
809  * basically own the machine, and have a very in depth understanding
810  * of all the possible deadlocks and enospc problems.
811  */
812 long btrfs_ioctl_trans_end(struct file *file)
813 {
814         struct inode *inode = fdentry(file)->d_inode;
815         struct btrfs_root *root = BTRFS_I(inode)->root;
816         struct btrfs_trans_handle *trans;
817         int ret = 0;
818
819         trans = file->private_data;
820         if (!trans) {
821                 ret = -EINVAL;
822                 goto out;
823         }
824         btrfs_end_transaction(trans, root);
825         file->private_data = NULL;
826
827         mutex_lock(&root->fs_info->trans_mutex);
828         root->fs_info->open_ioctl_trans--;
829         mutex_unlock(&root->fs_info->trans_mutex);
830
831 out:
832         return ret;
833 }
834
835 long btrfs_ioctl(struct file *file, unsigned int
836                 cmd, unsigned long arg)
837 {
838         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
839
840         switch (cmd) {
841         case BTRFS_IOC_SNAP_CREATE:
842                 return btrfs_ioctl_snap_create(file, (void __user *)arg);
843         case BTRFS_IOC_DEFRAG:
844                 return btrfs_ioctl_defrag(file);
845         case BTRFS_IOC_RESIZE:
846                 return btrfs_ioctl_resize(root, (void __user *)arg);
847         case BTRFS_IOC_ADD_DEV:
848                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
849         case BTRFS_IOC_RM_DEV:
850                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
851         case BTRFS_IOC_BALANCE:
852                 return btrfs_balance(root->fs_info->dev_root);
853         case BTRFS_IOC_CLONE:
854                 return btrfs_ioctl_clone(file, arg);
855         case BTRFS_IOC_TRANS_START:
856                 return btrfs_ioctl_trans_start(file);
857         case BTRFS_IOC_TRANS_END:
858                 return btrfs_ioctl_trans_end(file);
859         case BTRFS_IOC_SYNC:
860                 btrfs_start_delalloc_inodes(root);
861                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
862                 return 0;
863         }
864
865         return -ENOTTY;
866 }