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