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