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