]> bbs.cooldavid.org Git - net-next-2.6.git/blob - fs/btrfs/ioctl.c
Btrfs: Fix extent replacment race
[net-next-2.6.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 /* Mask out flags that are inappropriate for the given type of inode. */
54 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
55 {
56         if (S_ISDIR(mode))
57                 return flags;
58         else if (S_ISREG(mode))
59                 return flags & ~FS_DIRSYNC_FL;
60         else
61                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
62 }
63
64 /*
65  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
66  */
67 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
68 {
69         unsigned int iflags = 0;
70
71         if (flags & BTRFS_INODE_SYNC)
72                 iflags |= FS_SYNC_FL;
73         if (flags & BTRFS_INODE_IMMUTABLE)
74                 iflags |= FS_IMMUTABLE_FL;
75         if (flags & BTRFS_INODE_APPEND)
76                 iflags |= FS_APPEND_FL;
77         if (flags & BTRFS_INODE_NODUMP)
78                 iflags |= FS_NODUMP_FL;
79         if (flags & BTRFS_INODE_NOATIME)
80                 iflags |= FS_NOATIME_FL;
81         if (flags & BTRFS_INODE_DIRSYNC)
82                 iflags |= FS_DIRSYNC_FL;
83
84         return iflags;
85 }
86
87 /*
88  * Update inode->i_flags based on the btrfs internal flags.
89  */
90 void btrfs_update_iflags(struct inode *inode)
91 {
92         struct btrfs_inode *ip = BTRFS_I(inode);
93
94         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
95
96         if (ip->flags & BTRFS_INODE_SYNC)
97                 inode->i_flags |= S_SYNC;
98         if (ip->flags & BTRFS_INODE_IMMUTABLE)
99                 inode->i_flags |= S_IMMUTABLE;
100         if (ip->flags & BTRFS_INODE_APPEND)
101                 inode->i_flags |= S_APPEND;
102         if (ip->flags & BTRFS_INODE_NOATIME)
103                 inode->i_flags |= S_NOATIME;
104         if (ip->flags & BTRFS_INODE_DIRSYNC)
105                 inode->i_flags |= S_DIRSYNC;
106 }
107
108 /*
109  * Inherit flags from the parent inode.
110  *
111  * Unlike extN we don't have any flags we don't want to inherit currently.
112  */
113 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
114 {
115         unsigned int flags;
116
117         if (!dir)
118                 return;
119
120         flags = BTRFS_I(dir)->flags;
121
122         if (S_ISREG(inode->i_mode))
123                 flags &= ~BTRFS_INODE_DIRSYNC;
124         else if (!S_ISDIR(inode->i_mode))
125                 flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
126
127         BTRFS_I(inode)->flags = flags;
128         btrfs_update_iflags(inode);
129 }
130
131 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
132 {
133         struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
134         unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
135
136         if (copy_to_user(arg, &flags, sizeof(flags)))
137                 return -EFAULT;
138         return 0;
139 }
140
141 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
142 {
143         struct inode *inode = file->f_path.dentry->d_inode;
144         struct btrfs_inode *ip = BTRFS_I(inode);
145         struct btrfs_root *root = ip->root;
146         struct btrfs_trans_handle *trans;
147         unsigned int flags, oldflags;
148         int ret;
149
150         if (copy_from_user(&flags, arg, sizeof(flags)))
151                 return -EFAULT;
152
153         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
154                       FS_NOATIME_FL | FS_NODUMP_FL | \
155                       FS_SYNC_FL | FS_DIRSYNC_FL))
156                 return -EOPNOTSUPP;
157
158         if (!is_owner_or_cap(inode))
159                 return -EACCES;
160
161         mutex_lock(&inode->i_mutex);
162
163         flags = btrfs_mask_flags(inode->i_mode, flags);
164         oldflags = btrfs_flags_to_ioctl(ip->flags);
165         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
166                 if (!capable(CAP_LINUX_IMMUTABLE)) {
167                         ret = -EPERM;
168                         goto out_unlock;
169                 }
170         }
171
172         ret = mnt_want_write(file->f_path.mnt);
173         if (ret)
174                 goto out_unlock;
175
176         if (flags & FS_SYNC_FL)
177                 ip->flags |= BTRFS_INODE_SYNC;
178         else
179                 ip->flags &= ~BTRFS_INODE_SYNC;
180         if (flags & FS_IMMUTABLE_FL)
181                 ip->flags |= BTRFS_INODE_IMMUTABLE;
182         else
183                 ip->flags &= ~BTRFS_INODE_IMMUTABLE;
184         if (flags & FS_APPEND_FL)
185                 ip->flags |= BTRFS_INODE_APPEND;
186         else
187                 ip->flags &= ~BTRFS_INODE_APPEND;
188         if (flags & FS_NODUMP_FL)
189                 ip->flags |= BTRFS_INODE_NODUMP;
190         else
191                 ip->flags &= ~BTRFS_INODE_NODUMP;
192         if (flags & FS_NOATIME_FL)
193                 ip->flags |= BTRFS_INODE_NOATIME;
194         else
195                 ip->flags &= ~BTRFS_INODE_NOATIME;
196         if (flags & FS_DIRSYNC_FL)
197                 ip->flags |= BTRFS_INODE_DIRSYNC;
198         else
199                 ip->flags &= ~BTRFS_INODE_DIRSYNC;
200
201
202         trans = btrfs_join_transaction(root, 1);
203         BUG_ON(!trans);
204
205         ret = btrfs_update_inode(trans, root, inode);
206         BUG_ON(ret);
207
208         btrfs_update_iflags(inode);
209         inode->i_ctime = CURRENT_TIME;
210         btrfs_end_transaction(trans, root);
211
212         mnt_drop_write(file->f_path.mnt);
213  out_unlock:
214         mutex_unlock(&inode->i_mutex);
215         return 0;
216 }
217
218 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
219 {
220         struct inode *inode = file->f_path.dentry->d_inode;
221
222         return put_user(inode->i_generation, arg);
223 }
224
225 static noinline int create_subvol(struct btrfs_root *root,
226                                   struct dentry *dentry,
227                                   char *name, int namelen)
228 {
229         struct btrfs_trans_handle *trans;
230         struct btrfs_key key;
231         struct btrfs_root_item root_item;
232         struct btrfs_inode_item *inode_item;
233         struct extent_buffer *leaf;
234         struct btrfs_root *new_root = root;
235         struct inode *dir;
236         int ret;
237         int err;
238         u64 objectid;
239         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
240         u64 index = 0;
241         unsigned long nr = 1;
242
243         ret = btrfs_check_metadata_free_space(root);
244         if (ret)
245                 goto fail_commit;
246
247         trans = btrfs_start_transaction(root, 1);
248         BUG_ON(!trans);
249
250         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
251                                        0, &objectid);
252         if (ret)
253                 goto fail;
254
255         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
256                                       0, objectid, NULL, 0, 0, 0);
257         if (IS_ERR(leaf)) {
258                 ret = PTR_ERR(leaf);
259                 goto fail;
260         }
261
262         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
263         btrfs_set_header_bytenr(leaf, leaf->start);
264         btrfs_set_header_generation(leaf, trans->transid);
265         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
266         btrfs_set_header_owner(leaf, objectid);
267
268         write_extent_buffer(leaf, root->fs_info->fsid,
269                             (unsigned long)btrfs_header_fsid(leaf),
270                             BTRFS_FSID_SIZE);
271         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
272                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
273                             BTRFS_UUID_SIZE);
274         btrfs_mark_buffer_dirty(leaf);
275
276         inode_item = &root_item.inode;
277         memset(inode_item, 0, sizeof(*inode_item));
278         inode_item->generation = cpu_to_le64(1);
279         inode_item->size = cpu_to_le64(3);
280         inode_item->nlink = cpu_to_le32(1);
281         inode_item->nbytes = cpu_to_le64(root->leafsize);
282         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
283
284         btrfs_set_root_bytenr(&root_item, leaf->start);
285         btrfs_set_root_generation(&root_item, trans->transid);
286         btrfs_set_root_level(&root_item, 0);
287         btrfs_set_root_refs(&root_item, 1);
288         btrfs_set_root_used(&root_item, 0);
289         btrfs_set_root_last_snapshot(&root_item, 0);
290
291         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
292         root_item.drop_level = 0;
293
294         btrfs_tree_unlock(leaf);
295         free_extent_buffer(leaf);
296         leaf = NULL;
297
298         btrfs_set_root_dirid(&root_item, new_dirid);
299
300         key.objectid = objectid;
301         key.offset = 0;
302         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
303         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
304                                 &root_item);
305         if (ret)
306                 goto fail;
307
308         /*
309          * insert the directory item
310          */
311         key.offset = (u64)-1;
312         dir = dentry->d_parent->d_inode;
313         ret = btrfs_set_inode_index(dir, &index);
314         BUG_ON(ret);
315
316         ret = btrfs_insert_dir_item(trans, root,
317                                     name, namelen, dir->i_ino, &key,
318                                     BTRFS_FT_DIR, index);
319         if (ret)
320                 goto fail;
321
322         btrfs_i_size_write(dir, dir->i_size + namelen * 2);
323         ret = btrfs_update_inode(trans, root, dir);
324         BUG_ON(ret);
325
326         /* add the backref first */
327         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
328                                  objectid, BTRFS_ROOT_BACKREF_KEY,
329                                  root->root_key.objectid,
330                                  dir->i_ino, index, name, namelen);
331
332         BUG_ON(ret);
333
334         /* now add the forward ref */
335         ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
336                                  root->root_key.objectid, BTRFS_ROOT_REF_KEY,
337                                  objectid,
338                                  dir->i_ino, index, name, namelen);
339
340         BUG_ON(ret);
341
342         ret = btrfs_commit_transaction(trans, root);
343         if (ret)
344                 goto fail_commit;
345
346         new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
347         BUG_ON(!new_root);
348
349         trans = btrfs_start_transaction(new_root, 1);
350         BUG_ON(!trans);
351
352         ret = btrfs_create_subvol_root(trans, new_root, dentry, new_dirid,
353                                        BTRFS_I(dir)->block_group);
354         if (ret)
355                 goto fail;
356
357 fail:
358         nr = trans->blocks_used;
359         err = btrfs_commit_transaction(trans, new_root);
360         if (err && !ret)
361                 ret = err;
362 fail_commit:
363         btrfs_btree_balance_dirty(root, nr);
364         return ret;
365 }
366
367 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
368                            char *name, int namelen)
369 {
370         struct btrfs_pending_snapshot *pending_snapshot;
371         struct btrfs_trans_handle *trans;
372         int ret = 0;
373         int err;
374         unsigned long nr = 0;
375
376         if (!root->ref_cows)
377                 return -EINVAL;
378
379         ret = btrfs_check_metadata_free_space(root);
380         if (ret)
381                 goto fail_unlock;
382
383         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
384         if (!pending_snapshot) {
385                 ret = -ENOMEM;
386                 goto fail_unlock;
387         }
388         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
389         if (!pending_snapshot->name) {
390                 ret = -ENOMEM;
391                 kfree(pending_snapshot);
392                 goto fail_unlock;
393         }
394         memcpy(pending_snapshot->name, name, namelen);
395         pending_snapshot->name[namelen] = '\0';
396         pending_snapshot->dentry = dentry;
397         trans = btrfs_start_transaction(root, 1);
398         BUG_ON(!trans);
399         pending_snapshot->root = root;
400         list_add(&pending_snapshot->list,
401                  &trans->transaction->pending_snapshots);
402         err = btrfs_commit_transaction(trans, root);
403
404 fail_unlock:
405         btrfs_btree_balance_dirty(root, nr);
406         return ret;
407 }
408
409 /* copy of may_create in fs/namei.c() */
410 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
411 {
412         if (child->d_inode)
413                 return -EEXIST;
414         if (IS_DEADDIR(dir))
415                 return -ENOENT;
416         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
417 }
418
419 /*
420  * Create a new subvolume below @parent.  This is largely modeled after
421  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
422  * inside this filesystem so it's quite a bit simpler.
423  */
424 static noinline int btrfs_mksubvol(struct path *parent, char *name,
425                                    int mode, int namelen,
426                                    struct btrfs_root *snap_src)
427 {
428         struct dentry *dentry;
429         int error;
430
431         mutex_lock_nested(&parent->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
432
433         dentry = lookup_one_len(name, parent->dentry, namelen);
434         error = PTR_ERR(dentry);
435         if (IS_ERR(dentry))
436                 goto out_unlock;
437
438         error = -EEXIST;
439         if (dentry->d_inode)
440                 goto out_dput;
441
442         if (!IS_POSIXACL(parent->dentry->d_inode))
443                 mode &= ~current_umask();
444
445         error = mnt_want_write(parent->mnt);
446         if (error)
447                 goto out_dput;
448
449         error = btrfs_may_create(parent->dentry->d_inode, dentry);
450         if (error)
451                 goto out_drop_write;
452
453         /*
454          * Actually perform the low-level subvolume creation after all
455          * this VFS fuzz.
456          *
457          * Eventually we want to pass in an inode under which we create this
458          * subvolume, but for now all are under the filesystem root.
459          *
460          * Also we should pass on the mode eventually to allow creating new
461          * subvolume with specific mode bits.
462          */
463         if (snap_src) {
464                 struct dentry *dir = dentry->d_parent;
465                 struct dentry *test = dir->d_parent;
466                 struct btrfs_path *path = btrfs_alloc_path();
467                 int ret;
468                 u64 test_oid;
469                 u64 parent_oid = BTRFS_I(dir->d_inode)->root->root_key.objectid;
470
471                 test_oid = snap_src->root_key.objectid;
472
473                 ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
474                                           path, parent_oid, test_oid);
475                 if (ret == 0)
476                         goto create;
477                 btrfs_release_path(snap_src->fs_info->tree_root, path);
478
479                 /* we need to make sure we aren't creating a directory loop
480                  * by taking a snapshot of something that has our current
481                  * subvol in its directory tree.  So, this loops through
482                  * the dentries and checks the forward refs for each subvolume
483                  * to see if is references the subvolume where we are
484                  * placing this new snapshot.
485                  */
486                 while (1) {
487                         if (!test ||
488                             dir == snap_src->fs_info->sb->s_root ||
489                             test == snap_src->fs_info->sb->s_root ||
490                             test->d_inode->i_sb != snap_src->fs_info->sb) {
491                                 break;
492                         }
493                         if (S_ISLNK(test->d_inode->i_mode)) {
494                                 printk(KERN_INFO "Btrfs symlink in snapshot "
495                                        "path, failed\n");
496                                 error = -EMLINK;
497                                 btrfs_free_path(path);
498                                 goto out_drop_write;
499                         }
500                         test_oid =
501                                 BTRFS_I(test->d_inode)->root->root_key.objectid;
502                         ret = btrfs_find_root_ref(snap_src->fs_info->tree_root,
503                                   path, test_oid, parent_oid);
504                         if (ret == 0) {
505                                 printk(KERN_INFO "Btrfs snapshot creation "
506                                        "failed, looping\n");
507                                 error = -EMLINK;
508                                 btrfs_free_path(path);
509                                 goto out_drop_write;
510                         }
511                         btrfs_release_path(snap_src->fs_info->tree_root, path);
512                         test = test->d_parent;
513                 }
514 create:
515                 btrfs_free_path(path);
516                 error = create_snapshot(snap_src, dentry, name, namelen);
517         } else {
518                 error = create_subvol(BTRFS_I(parent->dentry->d_inode)->root,
519                                       dentry, name, namelen);
520         }
521         if (error)
522                 goto out_drop_write;
523
524         fsnotify_mkdir(parent->dentry->d_inode, dentry);
525 out_drop_write:
526         mnt_drop_write(parent->mnt);
527 out_dput:
528         dput(dentry);
529 out_unlock:
530         mutex_unlock(&parent->dentry->d_inode->i_mutex);
531         return error;
532 }
533
534
535 static int btrfs_defrag_file(struct file *file)
536 {
537         struct inode *inode = fdentry(file)->d_inode;
538         struct btrfs_root *root = BTRFS_I(inode)->root;
539         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
540         struct btrfs_ordered_extent *ordered;
541         struct page *page;
542         unsigned long last_index;
543         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
544         unsigned long total_read = 0;
545         u64 page_start;
546         u64 page_end;
547         unsigned long i;
548         int ret;
549
550         ret = btrfs_check_data_free_space(root, inode, inode->i_size);
551         if (ret)
552                 return -ENOSPC;
553
554         mutex_lock(&inode->i_mutex);
555         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
556         for (i = 0; i <= last_index; i++) {
557                 if (total_read % ra_pages == 0) {
558                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
559                                        min(last_index, i + ra_pages - 1));
560                 }
561                 total_read++;
562 again:
563                 page = grab_cache_page(inode->i_mapping, i);
564                 if (!page)
565                         goto out_unlock;
566                 if (!PageUptodate(page)) {
567                         btrfs_readpage(NULL, page);
568                         lock_page(page);
569                         if (!PageUptodate(page)) {
570                                 unlock_page(page);
571                                 page_cache_release(page);
572                                 goto out_unlock;
573                         }
574                 }
575
576                 wait_on_page_writeback(page);
577
578                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
579                 page_end = page_start + PAGE_CACHE_SIZE - 1;
580                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
581
582                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
583                 if (ordered) {
584                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
585                         unlock_page(page);
586                         page_cache_release(page);
587                         btrfs_start_ordered_extent(inode, ordered, 1);
588                         btrfs_put_ordered_extent(ordered);
589                         goto again;
590                 }
591                 set_page_extent_mapped(page);
592
593                 /*
594                  * this makes sure page_mkwrite is called on the
595                  * page if it is dirtied again later
596                  */
597                 clear_page_dirty_for_io(page);
598
599                 btrfs_set_extent_delalloc(inode, page_start, page_end);
600                 set_page_dirty(page);
601                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
602                 unlock_page(page);
603                 page_cache_release(page);
604                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
605         }
606
607 out_unlock:
608         mutex_unlock(&inode->i_mutex);
609         return 0;
610 }
611
612 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
613 {
614         u64 new_size;
615         u64 old_size;
616         u64 devid = 1;
617         struct btrfs_ioctl_vol_args *vol_args;
618         struct btrfs_trans_handle *trans;
619         struct btrfs_device *device = NULL;
620         char *sizestr;
621         char *devstr = NULL;
622         int ret = 0;
623         int namelen;
624         int mod = 0;
625
626         if (root->fs_info->sb->s_flags & MS_RDONLY)
627                 return -EROFS;
628
629         if (!capable(CAP_SYS_ADMIN))
630                 return -EPERM;
631
632         vol_args = memdup_user(arg, sizeof(*vol_args));
633         if (IS_ERR(vol_args))
634                 return PTR_ERR(vol_args);
635
636         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
637         namelen = strlen(vol_args->name);
638
639         mutex_lock(&root->fs_info->volume_mutex);
640         sizestr = vol_args->name;
641         devstr = strchr(sizestr, ':');
642         if (devstr) {
643                 char *end;
644                 sizestr = devstr + 1;
645                 *devstr = '\0';
646                 devstr = vol_args->name;
647                 devid = simple_strtoull(devstr, &end, 10);
648                 printk(KERN_INFO "resizing devid %llu\n",
649                        (unsigned long long)devid);
650         }
651         device = btrfs_find_device(root, devid, NULL, NULL);
652         if (!device) {
653                 printk(KERN_INFO "resizer unable to find device %llu\n",
654                        (unsigned long long)devid);
655                 ret = -EINVAL;
656                 goto out_unlock;
657         }
658         if (!strcmp(sizestr, "max"))
659                 new_size = device->bdev->bd_inode->i_size;
660         else {
661                 if (sizestr[0] == '-') {
662                         mod = -1;
663                         sizestr++;
664                 } else if (sizestr[0] == '+') {
665                         mod = 1;
666                         sizestr++;
667                 }
668                 new_size = btrfs_parse_size(sizestr);
669                 if (new_size == 0) {
670                         ret = -EINVAL;
671                         goto out_unlock;
672                 }
673         }
674
675         old_size = device->total_bytes;
676
677         if (mod < 0) {
678                 if (new_size > old_size) {
679                         ret = -EINVAL;
680                         goto out_unlock;
681                 }
682                 new_size = old_size - new_size;
683         } else if (mod > 0) {
684                 new_size = old_size + new_size;
685         }
686
687         if (new_size < 256 * 1024 * 1024) {
688                 ret = -EINVAL;
689                 goto out_unlock;
690         }
691         if (new_size > device->bdev->bd_inode->i_size) {
692                 ret = -EFBIG;
693                 goto out_unlock;
694         }
695
696         do_div(new_size, root->sectorsize);
697         new_size *= root->sectorsize;
698
699         printk(KERN_INFO "new size for %s is %llu\n",
700                 device->name, (unsigned long long)new_size);
701
702         if (new_size > old_size) {
703                 trans = btrfs_start_transaction(root, 1);
704                 ret = btrfs_grow_device(trans, device, new_size);
705                 btrfs_commit_transaction(trans, root);
706         } else {
707                 ret = btrfs_shrink_device(device, new_size);
708         }
709
710 out_unlock:
711         mutex_unlock(&root->fs_info->volume_mutex);
712         kfree(vol_args);
713         return ret;
714 }
715
716 static noinline int btrfs_ioctl_snap_create(struct file *file,
717                                             void __user *arg, int subvol)
718 {
719         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
720         struct btrfs_ioctl_vol_args *vol_args;
721         struct btrfs_dir_item *di;
722         struct btrfs_path *path;
723         struct file *src_file;
724         u64 root_dirid;
725         int namelen;
726         int ret = 0;
727
728         if (root->fs_info->sb->s_flags & MS_RDONLY)
729                 return -EROFS;
730
731         vol_args = memdup_user(arg, sizeof(*vol_args));
732         if (IS_ERR(vol_args))
733                 return PTR_ERR(vol_args);
734
735         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
736         namelen = strlen(vol_args->name);
737         if (strchr(vol_args->name, '/')) {
738                 ret = -EINVAL;
739                 goto out;
740         }
741
742         path = btrfs_alloc_path();
743         if (!path) {
744                 ret = -ENOMEM;
745                 goto out;
746         }
747
748         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
749         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
750                             path, root_dirid,
751                             vol_args->name, namelen, 0);
752         btrfs_free_path(path);
753
754         if (di && !IS_ERR(di)) {
755                 ret = -EEXIST;
756                 goto out;
757         }
758
759         if (IS_ERR(di)) {
760                 ret = PTR_ERR(di);
761                 goto out;
762         }
763
764         if (subvol) {
765                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
766                                      file->f_path.dentry->d_inode->i_mode,
767                                      namelen, NULL);
768         } else {
769                 struct inode *src_inode;
770                 src_file = fget(vol_args->fd);
771                 if (!src_file) {
772                         ret = -EINVAL;
773                         goto out;
774                 }
775
776                 src_inode = src_file->f_path.dentry->d_inode;
777                 if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
778                         printk(KERN_INFO "btrfs: Snapshot src from "
779                                "another FS\n");
780                         ret = -EINVAL;
781                         fput(src_file);
782                         goto out;
783                 }
784                 ret = btrfs_mksubvol(&file->f_path, vol_args->name,
785                              file->f_path.dentry->d_inode->i_mode,
786                              namelen, BTRFS_I(src_inode)->root);
787                 fput(src_file);
788         }
789
790 out:
791         kfree(vol_args);
792         return ret;
793 }
794
795 static int btrfs_ioctl_defrag(struct file *file)
796 {
797         struct inode *inode = fdentry(file)->d_inode;
798         struct btrfs_root *root = BTRFS_I(inode)->root;
799         int ret;
800
801         ret = mnt_want_write(file->f_path.mnt);
802         if (ret)
803                 return ret;
804
805         switch (inode->i_mode & S_IFMT) {
806         case S_IFDIR:
807                 if (!capable(CAP_SYS_ADMIN)) {
808                         ret = -EPERM;
809                         goto out;
810                 }
811                 btrfs_defrag_root(root, 0);
812                 btrfs_defrag_root(root->fs_info->extent_root, 0);
813                 break;
814         case S_IFREG:
815                 if (!(file->f_mode & FMODE_WRITE)) {
816                         ret = -EINVAL;
817                         goto out;
818                 }
819                 btrfs_defrag_file(file);
820                 break;
821         }
822 out:
823         mnt_drop_write(file->f_path.mnt);
824         return ret;
825 }
826
827 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
828 {
829         struct btrfs_ioctl_vol_args *vol_args;
830         int ret;
831
832         if (!capable(CAP_SYS_ADMIN))
833                 return -EPERM;
834
835         vol_args = memdup_user(arg, sizeof(*vol_args));
836         if (IS_ERR(vol_args))
837                 return PTR_ERR(vol_args);
838
839         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
840         ret = btrfs_init_new_device(root, vol_args->name);
841
842         kfree(vol_args);
843         return ret;
844 }
845
846 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
847 {
848         struct btrfs_ioctl_vol_args *vol_args;
849         int ret;
850
851         if (!capable(CAP_SYS_ADMIN))
852                 return -EPERM;
853
854         if (root->fs_info->sb->s_flags & MS_RDONLY)
855                 return -EROFS;
856
857         vol_args = memdup_user(arg, sizeof(*vol_args));
858         if (IS_ERR(vol_args))
859                 return PTR_ERR(vol_args);
860
861         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
862         ret = btrfs_rm_device(root, vol_args->name);
863
864         kfree(vol_args);
865         return ret;
866 }
867
868 static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
869                 u64 off, u64 olen, u64 destoff)
870 {
871         struct inode *inode = fdentry(file)->d_inode;
872         struct btrfs_root *root = BTRFS_I(inode)->root;
873         struct file *src_file;
874         struct inode *src;
875         struct btrfs_trans_handle *trans;
876         struct btrfs_path *path;
877         struct extent_buffer *leaf;
878         char *buf;
879         struct btrfs_key key;
880         u32 nritems;
881         int slot;
882         int ret;
883         u64 len = olen;
884         u64 bs = root->fs_info->sb->s_blocksize;
885         u64 hint_byte;
886
887         /*
888          * TODO:
889          * - split compressed inline extents.  annoying: we need to
890          *   decompress into destination's address_space (the file offset
891          *   may change, so source mapping won't do), then recompress (or
892          *   otherwise reinsert) a subrange.
893          * - allow ranges within the same file to be cloned (provided
894          *   they don't overlap)?
895          */
896
897         /* the destination must be opened for writing */
898         if (!(file->f_mode & FMODE_WRITE))
899                 return -EINVAL;
900
901         ret = mnt_want_write(file->f_path.mnt);
902         if (ret)
903                 return ret;
904
905         src_file = fget(srcfd);
906         if (!src_file) {
907                 ret = -EBADF;
908                 goto out_drop_write;
909         }
910         src = src_file->f_dentry->d_inode;
911
912         ret = -EINVAL;
913         if (src == inode)
914                 goto out_fput;
915
916         ret = -EISDIR;
917         if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
918                 goto out_fput;
919
920         ret = -EXDEV;
921         if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
922                 goto out_fput;
923
924         ret = -ENOMEM;
925         buf = vmalloc(btrfs_level_size(root, 0));
926         if (!buf)
927                 goto out_fput;
928
929         path = btrfs_alloc_path();
930         if (!path) {
931                 vfree(buf);
932                 goto out_fput;
933         }
934         path->reada = 2;
935
936         if (inode < src) {
937                 mutex_lock(&inode->i_mutex);
938                 mutex_lock(&src->i_mutex);
939         } else {
940                 mutex_lock(&src->i_mutex);
941                 mutex_lock(&inode->i_mutex);
942         }
943
944         /* determine range to clone */
945         ret = -EINVAL;
946         if (off >= src->i_size || off + len > src->i_size)
947                 goto out_unlock;
948         if (len == 0)
949                 olen = len = src->i_size - off;
950         /* if we extend to eof, continue to block boundary */
951         if (off + len == src->i_size)
952                 len = ((src->i_size + bs-1) & ~(bs-1))
953                         - off;
954
955         /* verify the end result is block aligned */
956         if ((off & (bs-1)) ||
957             ((off + len) & (bs-1)))
958                 goto out_unlock;
959
960         /* do any pending delalloc/csum calc on src, one way or
961            another, and lock file content */
962         while (1) {
963                 struct btrfs_ordered_extent *ordered;
964                 lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
965                 ordered = btrfs_lookup_first_ordered_extent(inode, off+len);
966                 if (BTRFS_I(src)->delalloc_bytes == 0 && !ordered)
967                         break;
968                 unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
969                 if (ordered)
970                         btrfs_put_ordered_extent(ordered);
971                 btrfs_wait_ordered_range(src, off, off+len);
972         }
973
974         trans = btrfs_start_transaction(root, 1);
975         BUG_ON(!trans);
976
977         /* punch hole in destination first */
978         btrfs_drop_extents(trans, root, inode, off, off + len,
979                            off + len, 0, &hint_byte, 1);
980
981         /* clone data */
982         key.objectid = src->i_ino;
983         key.type = BTRFS_EXTENT_DATA_KEY;
984         key.offset = 0;
985
986         while (1) {
987                 /*
988                  * note the key will change type as we walk through the
989                  * tree.
990                  */
991                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
992                 if (ret < 0)
993                         goto out;
994
995                 nritems = btrfs_header_nritems(path->nodes[0]);
996                 if (path->slots[0] >= nritems) {
997                         ret = btrfs_next_leaf(root, path);
998                         if (ret < 0)
999                                 goto out;
1000                         if (ret > 0)
1001                                 break;
1002                         nritems = btrfs_header_nritems(path->nodes[0]);
1003                 }
1004                 leaf = path->nodes[0];
1005                 slot = path->slots[0];
1006
1007                 btrfs_item_key_to_cpu(leaf, &key, slot);
1008                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1009                     key.objectid != src->i_ino)
1010                         break;
1011
1012                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1013                         struct btrfs_file_extent_item *extent;
1014                         int type;
1015                         u32 size;
1016                         struct btrfs_key new_key;
1017                         u64 disko = 0, diskl = 0;
1018                         u64 datao = 0, datal = 0;
1019                         u8 comp;
1020
1021                         size = btrfs_item_size_nr(leaf, slot);
1022                         read_extent_buffer(leaf, buf,
1023                                            btrfs_item_ptr_offset(leaf, slot),
1024                                            size);
1025
1026                         extent = btrfs_item_ptr(leaf, slot,
1027                                                 struct btrfs_file_extent_item);
1028                         comp = btrfs_file_extent_compression(leaf, extent);
1029                         type = btrfs_file_extent_type(leaf, extent);
1030                         if (type == BTRFS_FILE_EXTENT_REG ||
1031                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1032                                 disko = btrfs_file_extent_disk_bytenr(leaf,
1033                                                                       extent);
1034                                 diskl = btrfs_file_extent_disk_num_bytes(leaf,
1035                                                                  extent);
1036                                 datao = btrfs_file_extent_offset(leaf, extent);
1037                                 datal = btrfs_file_extent_num_bytes(leaf,
1038                                                                     extent);
1039                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1040                                 /* take upper bound, may be compressed */
1041                                 datal = btrfs_file_extent_ram_bytes(leaf,
1042                                                                     extent);
1043                         }
1044                         btrfs_release_path(root, path);
1045
1046                         if (key.offset + datal < off ||
1047                             key.offset >= off+len)
1048                                 goto next;
1049
1050                         memcpy(&new_key, &key, sizeof(new_key));
1051                         new_key.objectid = inode->i_ino;
1052                         new_key.offset = key.offset + destoff - off;
1053
1054                         if (type == BTRFS_FILE_EXTENT_REG ||
1055                             type == BTRFS_FILE_EXTENT_PREALLOC) {
1056                                 ret = btrfs_insert_empty_item(trans, root, path,
1057                                                               &new_key, size);
1058                                 if (ret)
1059                                         goto out;
1060
1061                                 leaf = path->nodes[0];
1062                                 slot = path->slots[0];
1063                                 write_extent_buffer(leaf, buf,
1064                                             btrfs_item_ptr_offset(leaf, slot),
1065                                             size);
1066
1067                                 extent = btrfs_item_ptr(leaf, slot,
1068                                                 struct btrfs_file_extent_item);
1069
1070                                 if (off > key.offset) {
1071                                         datao += off - key.offset;
1072                                         datal -= off - key.offset;
1073                                 }
1074                                 if (key.offset + datao + datal + key.offset >
1075                                     off + len)
1076                                         datal = off + len - key.offset - datao;
1077                                 /* disko == 0 means it's a hole */
1078                                 if (!disko)
1079                                         datao = 0;
1080
1081                                 btrfs_set_file_extent_offset(leaf, extent,
1082                                                              datao);
1083                                 btrfs_set_file_extent_num_bytes(leaf, extent,
1084                                                                 datal);
1085                                 if (disko) {
1086                                         inode_add_bytes(inode, datal);
1087                                         ret = btrfs_inc_extent_ref(trans, root,
1088                                                         disko, diskl, 0,
1089                                                         root->root_key.objectid,
1090                                                         inode->i_ino,
1091                                                         new_key.offset - datao);
1092                                         BUG_ON(ret);
1093                                 }
1094                         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1095                                 u64 skip = 0;
1096                                 u64 trim = 0;
1097                                 if (off > key.offset) {
1098                                         skip = off - key.offset;
1099                                         new_key.offset += skip;
1100                                 }
1101
1102                                 if (key.offset + datal > off+len)
1103                                         trim = key.offset + datal - (off+len);
1104
1105                                 if (comp && (skip || trim)) {
1106                                         ret = -EINVAL;
1107                                         goto out;
1108                                 }
1109                                 size -= skip + trim;
1110                                 datal -= skip + trim;
1111                                 ret = btrfs_insert_empty_item(trans, root, path,
1112                                                               &new_key, size);
1113                                 if (ret)
1114                                         goto out;
1115
1116                                 if (skip) {
1117                                         u32 start =
1118                                           btrfs_file_extent_calc_inline_size(0);
1119                                         memmove(buf+start, buf+start+skip,
1120                                                 datal);
1121                                 }
1122
1123                                 leaf = path->nodes[0];
1124                                 slot = path->slots[0];
1125                                 write_extent_buffer(leaf, buf,
1126                                             btrfs_item_ptr_offset(leaf, slot),
1127                                             size);
1128                                 inode_add_bytes(inode, datal);
1129                         }
1130
1131                         btrfs_mark_buffer_dirty(leaf);
1132                 }
1133
1134 next:
1135                 btrfs_release_path(root, path);
1136                 key.offset++;
1137         }
1138         ret = 0;
1139 out:
1140         btrfs_release_path(root, path);
1141         if (ret == 0) {
1142                 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1143                 if (destoff + olen > inode->i_size)
1144                         btrfs_i_size_write(inode, destoff + olen);
1145                 BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
1146                 ret = btrfs_update_inode(trans, root, inode);
1147         }
1148         btrfs_end_transaction(trans, root);
1149         unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1150         if (ret)
1151                 vmtruncate(inode, 0);
1152 out_unlock:
1153         mutex_unlock(&src->i_mutex);
1154         mutex_unlock(&inode->i_mutex);
1155         vfree(buf);
1156         btrfs_free_path(path);
1157 out_fput:
1158         fput(src_file);
1159 out_drop_write:
1160         mnt_drop_write(file->f_path.mnt);
1161         return ret;
1162 }
1163
1164 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
1165 {
1166         struct btrfs_ioctl_clone_range_args args;
1167
1168         if (copy_from_user(&args, argp, sizeof(args)))
1169                 return -EFAULT;
1170         return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
1171                                  args.src_length, args.dest_offset);
1172 }
1173
1174 /*
1175  * there are many ways the trans_start and trans_end ioctls can lead
1176  * to deadlocks.  They should only be used by applications that
1177  * basically own the machine, and have a very in depth understanding
1178  * of all the possible deadlocks and enospc problems.
1179  */
1180 static long btrfs_ioctl_trans_start(struct file *file)
1181 {
1182         struct inode *inode = fdentry(file)->d_inode;
1183         struct btrfs_root *root = BTRFS_I(inode)->root;
1184         struct btrfs_trans_handle *trans;
1185         int ret = 0;
1186
1187         if (!capable(CAP_SYS_ADMIN))
1188                 return -EPERM;
1189
1190         if (file->private_data) {
1191                 ret = -EINPROGRESS;
1192                 goto out;
1193         }
1194
1195         ret = mnt_want_write(file->f_path.mnt);
1196         if (ret)
1197                 goto out;
1198
1199         mutex_lock(&root->fs_info->trans_mutex);
1200         root->fs_info->open_ioctl_trans++;
1201         mutex_unlock(&root->fs_info->trans_mutex);
1202
1203         trans = btrfs_start_ioctl_transaction(root, 0);
1204         if (trans)
1205                 file->private_data = trans;
1206         else
1207                 ret = -ENOMEM;
1208         /*printk(KERN_INFO "btrfs_ioctl_trans_start on %p\n", file);*/
1209 out:
1210         return ret;
1211 }
1212
1213 /*
1214  * there are many ways the trans_start and trans_end ioctls can lead
1215  * to deadlocks.  They should only be used by applications that
1216  * basically own the machine, and have a very in depth understanding
1217  * of all the possible deadlocks and enospc problems.
1218  */
1219 long btrfs_ioctl_trans_end(struct file *file)
1220 {
1221         struct inode *inode = fdentry(file)->d_inode;
1222         struct btrfs_root *root = BTRFS_I(inode)->root;
1223         struct btrfs_trans_handle *trans;
1224         int ret = 0;
1225
1226         trans = file->private_data;
1227         if (!trans) {
1228                 ret = -EINVAL;
1229                 goto out;
1230         }
1231         btrfs_end_transaction(trans, root);
1232         file->private_data = NULL;
1233
1234         mutex_lock(&root->fs_info->trans_mutex);
1235         root->fs_info->open_ioctl_trans--;
1236         mutex_unlock(&root->fs_info->trans_mutex);
1237
1238         mnt_drop_write(file->f_path.mnt);
1239
1240 out:
1241         return ret;
1242 }
1243
1244 long btrfs_ioctl(struct file *file, unsigned int
1245                 cmd, unsigned long arg)
1246 {
1247         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1248         void __user *argp = (void __user *)arg;
1249
1250         switch (cmd) {
1251         case FS_IOC_GETFLAGS:
1252                 return btrfs_ioctl_getflags(file, argp);
1253         case FS_IOC_SETFLAGS:
1254                 return btrfs_ioctl_setflags(file, argp);
1255         case FS_IOC_GETVERSION:
1256                 return btrfs_ioctl_getversion(file, argp);
1257         case BTRFS_IOC_SNAP_CREATE:
1258                 return btrfs_ioctl_snap_create(file, argp, 0);
1259         case BTRFS_IOC_SUBVOL_CREATE:
1260                 return btrfs_ioctl_snap_create(file, argp, 1);
1261         case BTRFS_IOC_DEFRAG:
1262                 return btrfs_ioctl_defrag(file);
1263         case BTRFS_IOC_RESIZE:
1264                 return btrfs_ioctl_resize(root, argp);
1265         case BTRFS_IOC_ADD_DEV:
1266                 return btrfs_ioctl_add_dev(root, argp);
1267         case BTRFS_IOC_RM_DEV:
1268                 return btrfs_ioctl_rm_dev(root, argp);
1269         case BTRFS_IOC_BALANCE:
1270                 return btrfs_balance(root->fs_info->dev_root);
1271         case BTRFS_IOC_CLONE:
1272                 return btrfs_ioctl_clone(file, arg, 0, 0, 0);
1273         case BTRFS_IOC_CLONE_RANGE:
1274                 return btrfs_ioctl_clone_range(file, argp);
1275         case BTRFS_IOC_TRANS_START:
1276                 return btrfs_ioctl_trans_start(file);
1277         case BTRFS_IOC_TRANS_END:
1278                 return btrfs_ioctl_trans_end(file);
1279         case BTRFS_IOC_SYNC:
1280                 btrfs_sync_fs(file->f_dentry->d_sb, 1);
1281                 return 0;
1282         }
1283
1284         return -ENOTTY;
1285 }