Btrfs: fix fallocate deadlock on inode extent lock
[linux-3.10.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 /*
441  * Called inside transaction, so use GFP_NOFS
442  */
443
444 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
445 {
446         u64 new_size;
447         u64 old_size;
448         u64 devid = 1;
449         struct btrfs_ioctl_vol_args *vol_args;
450         struct btrfs_trans_handle *trans;
451         struct btrfs_device *device = NULL;
452         char *sizestr;
453         char *devstr = NULL;
454         int ret = 0;
455         int namelen;
456         int mod = 0;
457
458         if (root->fs_info->sb->s_flags & MS_RDONLY)
459                 return -EROFS;
460
461         if (!capable(CAP_SYS_ADMIN))
462                 return -EPERM;
463
464         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
465
466         if (!vol_args)
467                 return -ENOMEM;
468
469         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
470                 ret = -EFAULT;
471                 goto out;
472         }
473
474         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
475         namelen = strlen(vol_args->name);
476
477         mutex_lock(&root->fs_info->volume_mutex);
478         sizestr = vol_args->name;
479         devstr = strchr(sizestr, ':');
480         if (devstr) {
481                 char *end;
482                 sizestr = devstr + 1;
483                 *devstr = '\0';
484                 devstr = vol_args->name;
485                 devid = simple_strtoull(devstr, &end, 10);
486                 printk(KERN_INFO "resizing devid %llu\n", devid);
487         }
488         device = btrfs_find_device(root, devid, NULL, NULL);
489         if (!device) {
490                 printk(KERN_INFO "resizer unable to find device %llu\n", devid);
491                 ret = -EINVAL;
492                 goto out_unlock;
493         }
494         if (!strcmp(sizestr, "max"))
495                 new_size = device->bdev->bd_inode->i_size;
496         else {
497                 if (sizestr[0] == '-') {
498                         mod = -1;
499                         sizestr++;
500                 } else if (sizestr[0] == '+') {
501                         mod = 1;
502                         sizestr++;
503                 }
504                 new_size = btrfs_parse_size(sizestr);
505                 if (new_size == 0) {
506                         ret = -EINVAL;
507                         goto out_unlock;
508                 }
509         }
510
511         old_size = device->total_bytes;
512
513         if (mod < 0) {
514                 if (new_size > old_size) {
515                         ret = -EINVAL;
516                         goto out_unlock;
517                 }
518                 new_size = old_size - new_size;
519         } else if (mod > 0) {
520                 new_size = old_size + new_size;
521         }
522
523         if (new_size < 256 * 1024 * 1024) {
524                 ret = -EINVAL;
525                 goto out_unlock;
526         }
527         if (new_size > device->bdev->bd_inode->i_size) {
528                 ret = -EFBIG;
529                 goto out_unlock;
530         }
531
532         do_div(new_size, root->sectorsize);
533         new_size *= root->sectorsize;
534
535         printk(KERN_INFO "new size for %s is %llu\n",
536                 device->name, (unsigned long long)new_size);
537
538         if (new_size > old_size) {
539                 trans = btrfs_start_transaction(root, 1);
540                 ret = btrfs_grow_device(trans, device, new_size);
541                 btrfs_commit_transaction(trans, root);
542         } else {
543                 ret = btrfs_shrink_device(device, new_size);
544         }
545
546 out_unlock:
547         mutex_unlock(&root->fs_info->volume_mutex);
548 out:
549         kfree(vol_args);
550         return ret;
551 }
552
553 static noinline int btrfs_ioctl_snap_create(struct file *file,
554                                             void __user *arg, int subvol)
555 {
556         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
557         struct btrfs_ioctl_vol_args *vol_args;
558         struct btrfs_dir_item *di;
559         struct btrfs_path *path;
560         struct file *src_file;
561         u64 root_dirid;
562         int namelen;
563         int ret = 0;
564
565         if (root->fs_info->sb->s_flags & MS_RDONLY)
566                 return -EROFS;
567
568         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
569
570         if (!vol_args)
571                 return -ENOMEM;
572
573         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
574                 ret = -EFAULT;
575                 goto out;
576         }
577
578         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
579         namelen = strlen(vol_args->name);
580         if (strchr(vol_args->name, '/')) {
581                 ret = -EINVAL;
582                 goto out;
583         }
584
585         path = btrfs_alloc_path();
586         if (!path) {
587                 ret = -ENOMEM;
588                 goto out;
589         }
590
591         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
592         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
593                             path, root_dirid,
594                             vol_args->name, namelen, 0);
595         btrfs_free_path(path);
596
597         if (di && !IS_ERR(di)) {
598                 ret = -EEXIST;
599                 goto out;
600         }
601
602         if (IS_ERR(di)) {
603                 ret = PTR_ERR(di);
604                 goto out;
605         }
606
607         if (subvol) {
608                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
609                                      file->f_path.dentry->d_inode->i_mode,
610                                      namelen, NULL);
611         } else {
612                 struct inode *src_inode;
613                 src_file = fget(vol_args->fd);
614                 if (!src_file) {
615                         ret = -EINVAL;
616                         goto out;
617                 }
618
619                 src_inode = src_file->f_path.dentry->d_inode;
620                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
621                         printk(KERN_INFO "btrfs: Snapshot src from "
622                                "another FS\n");
623                         ret = -EINVAL;
624                         fput(src_file);
625                         goto out;
626                 }
627                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
628                              file->f_path.dentry->d_inode->i_mode,
629                              namelen, BTRFS_I(src_inode)->root);
630                 fput(src_file);
631         }
632
633 out:
634         kfree(vol_args);
635         return ret;
636 }
637
638 static int btrfs_ioctl_defrag(struct file *file)
639 {
640         struct inode *inode = fdentry(file)->d_inode;
641         struct btrfs_root *root = BTRFS_I(inode)->root;
642         int ret;
643
644         ret = mnt_want_write(file->f_path.mnt);
645         if (ret)
646                 return ret;
647
648         switch (inode->i_mode & S_IFMT) {
649         case S_IFDIR:
650                 if (!capable(CAP_SYS_ADMIN)) {
651                         ret = -EPERM;
652                         goto out;
653                 }
654                 btrfs_defrag_root(root, 0);
655                 btrfs_defrag_root(root->fs_info->extent_root, 0);
656                 break;
657         case S_IFREG:
658                 if (!(file->f_mode & FMODE_WRITE)) {
659                         ret = -EINVAL;
660                         goto out;
661                 }
662                 btrfs_defrag_file(file);
663                 break;
664         }
665 out:
666         mnt_drop_write(file->f_path.mnt);
667         return ret;
668 }
669
670 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
671 {
672         struct btrfs_ioctl_vol_args *vol_args;
673         int ret;
674
675         if (!capable(CAP_SYS_ADMIN))
676                 return -EPERM;
677
678         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
679
680         if (!vol_args)
681                 return -ENOMEM;
682
683         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
684                 ret = -EFAULT;
685                 goto out;
686         }
687         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
688         ret = btrfs_init_new_device(root, vol_args->name);
689
690 out:
691         kfree(vol_args);
692         return ret;
693 }
694
695 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
696 {
697         struct btrfs_ioctl_vol_args *vol_args;
698         int ret;
699
700         if (!capable(CAP_SYS_ADMIN))
701                 return -EPERM;
702
703         if (root->fs_info->sb->s_flags & MS_RDONLY)
704                 return -EROFS;
705
706         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
707
708         if (!vol_args)
709                 return -ENOMEM;
710
711         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
712                 ret = -EFAULT;
713                 goto out;
714         }
715         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
716         ret = btrfs_rm_device(root, vol_args->name);
717
718 out:
719         kfree(vol_args);
720         return ret;
721 }
722
723 static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
724                 u64 off, u64 olen, u64 destoff)
725 {
726         struct inode *inode = fdentry(file)->d_inode;
727         struct btrfs_root *root = BTRFS_I(inode)->root;
728         struct file *src_file;
729         struct inode *src;
730         struct btrfs_trans_handle *trans;
731         struct btrfs_path *path;
732         struct extent_buffer *leaf;
733         char *buf;
734         struct btrfs_key key;
735         u32 nritems;
736         int slot;
737         int ret;
738         u64 len = olen;
739         u64 bs = root->fs_info->sb->s_blocksize;
740         u64 hint_byte;
741
742         /*
743          * TODO:
744          * - split compressed inline extents.  annoying: we need to
745          *   decompress into destination's address_space (the file offset
746          *   may change, so source mapping won't do), then recompress (or
747          *   otherwise reinsert) a subrange.
748          * - allow ranges within the same file to be cloned (provided
749          *   they don't overlap)?
750          */
751
752         /* the destination must be opened for writing */
753         if (!(file->f_mode & FMODE_WRITE))
754                 return -EINVAL;
755
756         ret = mnt_want_write(file->f_path.mnt);
757         if (ret)
758                 return ret;
759
760         src_file = fget(srcfd);
761         if (!src_file) {
762                 ret = -EBADF;
763                 goto out_drop_write;
764         }
765         src = src_file->f_dentry->d_inode;
766
767         ret = -EINVAL;
768         if (src == inode)
769                 goto out_fput;
770
771         ret = -EISDIR;
772         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
773                 goto out_fput;
774
775         ret = -EXDEV;
776         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
777                 goto out_fput;
778
779         ret = -ENOMEM;
780         buf = vmalloc(btrfs_level_size(root, 0));
781         if (!buf)
782                 goto out_fput;
783
784         path = btrfs_alloc_path();
785         if (!path) {
786                 vfree(buf);
787                 goto out_fput;
788         }
789         path->reada = 2;
790
791         if (inode < src) {
792                 mutex_lock(&inode->i_mutex);
793                 mutex_lock(&src->i_mutex);
794         } else {
795                 mutex_lock(&src->i_mutex);
796                 mutex_lock(&inode->i_mutex);
797         }
798
799         /* determine range to clone */
800         ret = -EINVAL;
801         if (off >= src->i_size || off + len > src->i_size)
802                 goto out_unlock;
803         if (len == 0)
804                 olen = len = src->i_size - off;
805         /* if we extend to eof, continue to block boundary */
806         if (off + len == src->i_size)
807                 len = ((src->i_size + bs-1) & ~(bs-1))
808                         - off;
809
810         /* verify the end result is block aligned */
811         if ((off & (bs-1)) ||
812             ((off + len) & (bs-1)))
813                 goto out_unlock;
814
815         /* do any pending delalloc/csum calc on src, one way or
816            another, and lock file content */
817         while (1) {
818                 struct btrfs_ordered_extent *ordered;
819                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
820                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
821                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
822                         break;
823                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
824                 if (ordered)
825                         btrfs_put_ordered_extent(ordered);
826                 btrfs_wait_ordered_range(src, off, off+len);
827         }
828
829         trans = btrfs_start_transaction(root, 1);
830         BUG_ON(!trans);
831
832         /* punch hole in destination first */
833         btrfs_drop_extents(trans, root, inode, off, off + len,
834                            off + len, 0, &hint_byte);
835
836         /* clone data */
837         key.objectid = src->i_ino;
838         key.type = BTRFS_EXTENT_DATA_KEY;
839         key.offset = 0;
840
841         while (1) {
842                 /*
843                  * note the key will change type as we walk through the
844                  * tree.
845                  */
846                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
847                 if (ret < 0)
848                         goto out;
849
850                 nritems = btrfs_header_nritems(path->nodes[0]);
851                 if (path->slots[0] >= nritems) {
852                         ret = btrfs_next_leaf(root, path);
853                         if (ret < 0)
854                                 goto out;
855                         if (ret > 0)
856                                 break;
857                         nritems = btrfs_header_nritems(path->nodes[0]);
858                 }
859                 leaf = path->nodes[0];
860                 slot = path->slots[0];
861
862                 btrfs_item_key_to_cpu(leaf, &key, slot);
863                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
864                     key.objectid != src->i_ino)
865                         break;
866
867                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
868                         struct btrfs_file_extent_item *extent;
869                         int type;
870                         u32 size;
871                         struct btrfs_key new_key;
872                         u64 disko = 0, diskl = 0;
873                         u64 datao = 0, datal = 0;
874                         u8 comp;
875
876                         size = btrfs_item_size_nr(leaf, slot);
877                         read_extent_buffer(leaf, buf,
878                                            btrfs_item_ptr_offset(leaf, slot),
879                                            size);
880
881                         extent = btrfs_item_ptr(leaf, slot,
882                                                 struct btrfs_file_extent_item);
883                         comp = btrfs_file_extent_compression(leaf, extent);
884                         type = btrfs_file_extent_type(leaf, extent);
885                         if (type == BTRFS_FILE_EXTENT_REG) {
886                                 disko = btrfs_file_extent_disk_bytenr(leaf,
887                                                                       extent);
888                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
889                                                                  extent);
890                                 datao = btrfs_file_extent_offset(leaf, extent);
891                                 datal = btrfs_file_extent_num_bytes(leaf,
892                                                                     extent);
893                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
894                                 /* take upper bound, may be compressed */
895                                 datal = btrfs_file_extent_ram_bytes(leaf,
896                                                                     extent);
897                         }
898                         btrfs_release_path(root, path);
899
900                         if (key.offset + datal < off ||
901                             key.offset >= off+len)
902                                 goto next;
903
904                         memcpy(&new_key, &key, sizeof(new_key));
905                         new_key.objectid = inode->i_ino;
906                         new_key.offset = key.offset + destoff - off;
907
908                         if (type == BTRFS_FILE_EXTENT_REG) {
909                                 ret = btrfs_insert_empty_item(trans, root, path,
910                                                               &new_key, size);
911                                 if (ret)
912                                         goto out;
913
914                                 leaf = path->nodes[0];
915                                 slot = path->slots[0];
916                                 write_extent_buffer(leaf, buf,
917                                             btrfs_item_ptr_offset(leaf, slot),
918                                             size);
919
920                                 extent = btrfs_item_ptr(leaf, slot,
921                                                 struct btrfs_file_extent_item);
922
923                                 if (off > key.offset) {
924                                         datao += off - key.offset;
925                                         datal -= off - key.offset;
926                                 }
927                                 if (key.offset + datao + datal + key.offset >
928                                     off + len)
929                                         datal = off + len - key.offset - datao;
930                                 /* disko == 0 means it's a hole */
931                                 if (!disko)
932                                         datao = 0;
933
934                                 btrfs_set_file_extent_offset(leaf, extent,
935                                                              datao);
936                                 btrfs_set_file_extent_num_bytes(leaf, extent,
937                                                                 datal);
938                                 if (disko) {
939                                         inode_add_bytes(inode, datal);
940                                         ret = btrfs_inc_extent_ref(trans, root,
941                                                    disko, diskl, leaf->start,
942                                                    root->root_key.objectid,
943                                                    trans->transid,
944                                                    inode->i_ino);
945                                         BUG_ON(ret);
946                                 }
947                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
948                                 u64 skip = 0;
949                                 u64 trim = 0;
950                                 if (off > key.offset) {
951                                         skip = off - key.offset;
952                                         new_key.offset += skip;
953                                 }
954
955                                 if (key.offset + datal > off+len)
956                                         trim = key.offset + datal - (off+len);
957
958                                 if (comp && (skip || trim)) {
959                                         ret = -EINVAL;
960                                         goto out;
961                                 }
962                                 size -= skip + trim;
963                                 datal -= skip + trim;
964                                 ret = btrfs_insert_empty_item(trans, root, path,
965                                                               &new_key, size);
966                                 if (ret)
967                                         goto out;
968
969                                 if (skip) {
970                                         u32 start =
971                                           btrfs_file_extent_calc_inline_size(0);
972                                         memmove(buf+start, buf+start+skip,
973                                                 datal);
974                                 }
975
976                                 leaf = path->nodes[0];
977                                 slot = path->slots[0];
978                                 write_extent_buffer(leaf, buf,
979                                             btrfs_item_ptr_offset(leaf, slot),
980                                             size);
981                                 inode_add_bytes(inode, datal);
982                         }
983
984                         btrfs_mark_buffer_dirty(leaf);
985                 }
986
987 next:
988                 btrfs_release_path(root, path);
989                 key.offset++;
990         }
991         ret = 0;
992 out:
993         btrfs_release_path(root, path);
994         if (ret == 0) {
995                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
996                 if (destoff + olen > inode->i_size)
997                         btrfs_i_size_write(inode, destoff + olen);
998                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
999                 ret = btrfs_update_inode(trans, root, inode);
1000         }
1001         btrfs_end_transaction(trans, root);
1002         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1003         if (ret)
1004                 vmtruncate(inode, 0);
1005 out_unlock:
1006         mutex_unlock(&src->i_mutex);
1007         mutex_unlock(&inode->i_mutex);
1008         vfree(buf);
1009         btrfs_free_path(path);
1010 out_fput:
1011         fput(src_file);
1012 out_drop_write:
1013         mnt_drop_write(file->f_path.mnt);
1014         return ret;
1015 }
1016
1017 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1018 {
1019         struct btrfs_ioctl_clone_range_args args;
1020
1021         if (copy_from_user(&args, argp, sizeof(args)))
1022                 return -EFAULT;
1023         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1024                                  args.src_length, args.dest_offset);
1025 }
1026
1027 /*
1028  * there are many ways the trans_start and trans_end ioctls can lead
1029  * to deadlocks.  They should only be used by applications that
1030  * basically own the machine, and have a very in depth understanding
1031  * of all the possible deadlocks and enospc problems.
1032  */
1033 static long btrfs_ioctl_trans_start(struct file *file)
1034 {
1035         struct inode *inode = fdentry(file)->d_inode;
1036         struct btrfs_root *root = BTRFS_I(inode)->root;
1037         struct btrfs_trans_handle *trans;
1038         int ret = 0;
1039
1040         if (!capable(CAP_SYS_ADMIN))
1041                 return -EPERM;
1042
1043         if (file->private_data) {
1044                 ret = -EINPROGRESS;
1045                 goto out;
1046         }
1047
1048         ret = mnt_want_write(file->f_path.mnt);
1049         if (ret)
1050                 goto out;
1051
1052         mutex_lock(&root->fs_info->trans_mutex);
1053         root->fs_info->open_ioctl_trans++;
1054         mutex_unlock(&root->fs_info->trans_mutex);
1055
1056         trans = btrfs_start_ioctl_transaction(root, 0);
1057         if (trans)
1058                 file->private_data = trans;
1059         else
1060                 ret = -ENOMEM;
1061         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1062 out:
1063         return ret;
1064 }
1065
1066 /*
1067  * there are many ways the trans_start and trans_end ioctls can lead
1068  * to deadlocks.  They should only be used by applications that
1069  * basically own the machine, and have a very in depth understanding
1070  * of all the possible deadlocks and enospc problems.
1071  */
1072 long btrfs_ioctl_trans_end(struct file *file)
1073 {
1074         struct inode *inode = fdentry(file)->d_inode;
1075         struct btrfs_root *root = BTRFS_I(inode)->root;
1076         struct btrfs_trans_handle *trans;
1077         int ret = 0;
1078
1079         trans = file->private_data;
1080         if (!trans) {
1081                 ret = -EINVAL;
1082                 goto out;
1083         }
1084         btrfs_end_transaction(trans, root);
1085         file->private_data = NULL;
1086
1087         mutex_lock(&root->fs_info->trans_mutex);
1088         root->fs_info->open_ioctl_trans--;
1089         mutex_unlock(&root->fs_info->trans_mutex);
1090
1091         mnt_drop_write(file->f_path.mnt);
1092
1093 out:
1094         return ret;
1095 }
1096
1097 long btrfs_ioctl(struct file *file, unsigned int
1098                 cmd, unsigned long arg)
1099 {
1100         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1101         void __user *argp = (void __user *)arg;
1102
1103         switch (cmd) {
1104         case BTRFS_IOC_SNAP_CREATE:
1105                 return btrfs_ioctl_snap_create(file, argp, 0);
1106         case BTRFS_IOC_SUBVOL_CREATE:
1107                 return btrfs_ioctl_snap_create(file, argp, 1);
1108         case BTRFS_IOC_DEFRAG:
1109                 return btrfs_ioctl_defrag(file);
1110         case BTRFS_IOC_RESIZE:
1111                 return btrfs_ioctl_resize(root, argp);
1112         case BTRFS_IOC_ADD_DEV:
1113                 return btrfs_ioctl_add_dev(root, argp);
1114         case BTRFS_IOC_RM_DEV:
1115                 return btrfs_ioctl_rm_dev(root, argp);
1116         case BTRFS_IOC_BALANCE:
1117                 return btrfs_balance(root->fs_info->dev_root);
1118         case BTRFS_IOC_CLONE:
1119                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1120         case BTRFS_IOC_CLONE_RANGE:
1121                 return btrfs_ioctl_clone_range(file, argp);
1122         case BTRFS_IOC_TRANS_START:
1123                 return btrfs_ioctl_trans_start(file);
1124         case BTRFS_IOC_TRANS_END:
1125                 return btrfs_ioctl_trans_end(file);
1126         case BTRFS_IOC_SYNC:
1127                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1128                 return 0;
1129         }
1130
1131         return -ENOTTY;
1132 }