xref: /linux/fs/btrfs/ioctl.c (revision f850548ef88e5ff9e40bae9e1a7140bef0653e6b)
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/backing-dev.h>
31 #include <linux/mount.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/swap.h>
35 #include <linux/writeback.h>
36 #include <linux/statfs.h>
37 #include <linux/compat.h>
38 #include <linux/bit_spinlock.h>
39 #include <linux/security.h>
40 #include <linux/xattr.h>
41 #include <linux/vmalloc.h>
42 #include <linux/slab.h>
43 #include <linux/blkdev.h>
44 #include "compat.h"
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "ioctl.h"
50 #include "print-tree.h"
51 #include "volumes.h"
52 #include "locking.h"
53 
54 /* Mask out flags that are inappropriate for the given type of inode. */
55 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
56 {
57 	if (S_ISDIR(mode))
58 		return flags;
59 	else if (S_ISREG(mode))
60 		return flags & ~FS_DIRSYNC_FL;
61 	else
62 		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
63 }
64 
65 /*
66  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
67  */
68 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
69 {
70 	unsigned int iflags = 0;
71 
72 	if (flags & BTRFS_INODE_SYNC)
73 		iflags |= FS_SYNC_FL;
74 	if (flags & BTRFS_INODE_IMMUTABLE)
75 		iflags |= FS_IMMUTABLE_FL;
76 	if (flags & BTRFS_INODE_APPEND)
77 		iflags |= FS_APPEND_FL;
78 	if (flags & BTRFS_INODE_NODUMP)
79 		iflags |= FS_NODUMP_FL;
80 	if (flags & BTRFS_INODE_NOATIME)
81 		iflags |= FS_NOATIME_FL;
82 	if (flags & BTRFS_INODE_DIRSYNC)
83 		iflags |= FS_DIRSYNC_FL;
84 
85 	return iflags;
86 }
87 
88 /*
89  * Update inode->i_flags based on the btrfs internal flags.
90  */
91 void btrfs_update_iflags(struct inode *inode)
92 {
93 	struct btrfs_inode *ip = BTRFS_I(inode);
94 
95 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
96 
97 	if (ip->flags & BTRFS_INODE_SYNC)
98 		inode->i_flags |= S_SYNC;
99 	if (ip->flags & BTRFS_INODE_IMMUTABLE)
100 		inode->i_flags |= S_IMMUTABLE;
101 	if (ip->flags & BTRFS_INODE_APPEND)
102 		inode->i_flags |= S_APPEND;
103 	if (ip->flags & BTRFS_INODE_NOATIME)
104 		inode->i_flags |= S_NOATIME;
105 	if (ip->flags & BTRFS_INODE_DIRSYNC)
106 		inode->i_flags |= S_DIRSYNC;
107 }
108 
109 /*
110  * Inherit flags from the parent inode.
111  *
112  * Unlike extN we don't have any flags we don't want to inherit currently.
113  */
114 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
115 {
116 	unsigned int flags;
117 
118 	if (!dir)
119 		return;
120 
121 	flags = BTRFS_I(dir)->flags;
122 
123 	if (S_ISREG(inode->i_mode))
124 		flags &= ~BTRFS_INODE_DIRSYNC;
125 	else if (!S_ISDIR(inode->i_mode))
126 		flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
127 
128 	BTRFS_I(inode)->flags = flags;
129 	btrfs_update_iflags(inode);
130 }
131 
132 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
133 {
134 	struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
135 	unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
136 
137 	if (copy_to_user(arg, &flags, sizeof(flags)))
138 		return -EFAULT;
139 	return 0;
140 }
141 
142 static int check_flags(unsigned int flags)
143 {
144 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
145 		      FS_NOATIME_FL | FS_NODUMP_FL | \
146 		      FS_SYNC_FL | FS_DIRSYNC_FL | \
147 		      FS_NOCOMP_FL | FS_COMPR_FL | \
148 		      FS_NOCOW_FL | FS_COW_FL))
149 		return -EOPNOTSUPP;
150 
151 	if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
152 		return -EINVAL;
153 
154 	if ((flags & FS_NOCOW_FL) && (flags & FS_COW_FL))
155 		return -EINVAL;
156 
157 	return 0;
158 }
159 
160 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
161 {
162 	struct inode *inode = file->f_path.dentry->d_inode;
163 	struct btrfs_inode *ip = BTRFS_I(inode);
164 	struct btrfs_root *root = ip->root;
165 	struct btrfs_trans_handle *trans;
166 	unsigned int flags, oldflags;
167 	int ret;
168 
169 	if (btrfs_root_readonly(root))
170 		return -EROFS;
171 
172 	if (copy_from_user(&flags, arg, sizeof(flags)))
173 		return -EFAULT;
174 
175 	ret = check_flags(flags);
176 	if (ret)
177 		return ret;
178 
179 	if (!inode_owner_or_capable(inode))
180 		return -EACCES;
181 
182 	mutex_lock(&inode->i_mutex);
183 
184 	flags = btrfs_mask_flags(inode->i_mode, flags);
185 	oldflags = btrfs_flags_to_ioctl(ip->flags);
186 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
187 		if (!capable(CAP_LINUX_IMMUTABLE)) {
188 			ret = -EPERM;
189 			goto out_unlock;
190 		}
191 	}
192 
193 	ret = mnt_want_write(file->f_path.mnt);
194 	if (ret)
195 		goto out_unlock;
196 
197 	if (flags & FS_SYNC_FL)
198 		ip->flags |= BTRFS_INODE_SYNC;
199 	else
200 		ip->flags &= ~BTRFS_INODE_SYNC;
201 	if (flags & FS_IMMUTABLE_FL)
202 		ip->flags |= BTRFS_INODE_IMMUTABLE;
203 	else
204 		ip->flags &= ~BTRFS_INODE_IMMUTABLE;
205 	if (flags & FS_APPEND_FL)
206 		ip->flags |= BTRFS_INODE_APPEND;
207 	else
208 		ip->flags &= ~BTRFS_INODE_APPEND;
209 	if (flags & FS_NODUMP_FL)
210 		ip->flags |= BTRFS_INODE_NODUMP;
211 	else
212 		ip->flags &= ~BTRFS_INODE_NODUMP;
213 	if (flags & FS_NOATIME_FL)
214 		ip->flags |= BTRFS_INODE_NOATIME;
215 	else
216 		ip->flags &= ~BTRFS_INODE_NOATIME;
217 	if (flags & FS_DIRSYNC_FL)
218 		ip->flags |= BTRFS_INODE_DIRSYNC;
219 	else
220 		ip->flags &= ~BTRFS_INODE_DIRSYNC;
221 
222 	/*
223 	 * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
224 	 * flag may be changed automatically if compression code won't make
225 	 * things smaller.
226 	 */
227 	if (flags & FS_NOCOMP_FL) {
228 		ip->flags &= ~BTRFS_INODE_COMPRESS;
229 		ip->flags |= BTRFS_INODE_NOCOMPRESS;
230 	} else if (flags & FS_COMPR_FL) {
231 		ip->flags |= BTRFS_INODE_COMPRESS;
232 		ip->flags &= ~BTRFS_INODE_NOCOMPRESS;
233 	}
234 	if (flags & FS_NOCOW_FL)
235 		ip->flags |= BTRFS_INODE_NODATACOW;
236 	else if (flags & FS_COW_FL)
237 		ip->flags &= ~BTRFS_INODE_NODATACOW;
238 
239 	trans = btrfs_join_transaction(root, 1);
240 	BUG_ON(IS_ERR(trans));
241 
242 	ret = btrfs_update_inode(trans, root, inode);
243 	BUG_ON(ret);
244 
245 	btrfs_update_iflags(inode);
246 	inode->i_ctime = CURRENT_TIME;
247 	btrfs_end_transaction(trans, root);
248 
249 	mnt_drop_write(file->f_path.mnt);
250 
251 	ret = 0;
252  out_unlock:
253 	mutex_unlock(&inode->i_mutex);
254 	return ret;
255 }
256 
257 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
258 {
259 	struct inode *inode = file->f_path.dentry->d_inode;
260 
261 	return put_user(inode->i_generation, arg);
262 }
263 
264 static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg)
265 {
266 	struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info;
267 	struct btrfs_fs_info *fs_info = root->fs_info;
268 	struct btrfs_device *device;
269 	struct request_queue *q;
270 	struct fstrim_range range;
271 	u64 minlen = ULLONG_MAX;
272 	u64 num_devices = 0;
273 	int ret;
274 
275 	if (!capable(CAP_SYS_ADMIN))
276 		return -EPERM;
277 
278 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
279 	list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
280 		if (!device->bdev)
281 			continue;
282 		q = bdev_get_queue(device->bdev);
283 		if (blk_queue_discard(q)) {
284 			num_devices++;
285 			minlen = min((u64)q->limits.discard_granularity,
286 				     minlen);
287 		}
288 	}
289 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
290 	if (!num_devices)
291 		return -EOPNOTSUPP;
292 
293 	if (copy_from_user(&range, arg, sizeof(range)))
294 		return -EFAULT;
295 
296 	range.minlen = max(range.minlen, minlen);
297 	ret = btrfs_trim_fs(root, &range);
298 	if (ret < 0)
299 		return ret;
300 
301 	if (copy_to_user(arg, &range, sizeof(range)))
302 		return -EFAULT;
303 
304 	return 0;
305 }
306 
307 static noinline int create_subvol(struct btrfs_root *root,
308 				  struct dentry *dentry,
309 				  char *name, int namelen,
310 				  u64 *async_transid)
311 {
312 	struct btrfs_trans_handle *trans;
313 	struct btrfs_key key;
314 	struct btrfs_root_item root_item;
315 	struct btrfs_inode_item *inode_item;
316 	struct extent_buffer *leaf;
317 	struct btrfs_root *new_root;
318 	struct dentry *parent = dget_parent(dentry);
319 	struct inode *dir;
320 	int ret;
321 	int err;
322 	u64 objectid;
323 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
324 	u64 index = 0;
325 
326 	ret = btrfs_find_free_objectid(NULL, root->fs_info->tree_root,
327 				       0, &objectid);
328 	if (ret) {
329 		dput(parent);
330 		return ret;
331 	}
332 
333 	dir = parent->d_inode;
334 
335 	/*
336 	 * 1 - inode item
337 	 * 2 - refs
338 	 * 1 - root item
339 	 * 2 - dir items
340 	 */
341 	trans = btrfs_start_transaction(root, 6);
342 	if (IS_ERR(trans)) {
343 		dput(parent);
344 		return PTR_ERR(trans);
345 	}
346 
347 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
348 				      0, objectid, NULL, 0, 0, 0);
349 	if (IS_ERR(leaf)) {
350 		ret = PTR_ERR(leaf);
351 		goto fail;
352 	}
353 
354 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
355 	btrfs_set_header_bytenr(leaf, leaf->start);
356 	btrfs_set_header_generation(leaf, trans->transid);
357 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
358 	btrfs_set_header_owner(leaf, objectid);
359 
360 	write_extent_buffer(leaf, root->fs_info->fsid,
361 			    (unsigned long)btrfs_header_fsid(leaf),
362 			    BTRFS_FSID_SIZE);
363 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
364 			    (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
365 			    BTRFS_UUID_SIZE);
366 	btrfs_mark_buffer_dirty(leaf);
367 
368 	inode_item = &root_item.inode;
369 	memset(inode_item, 0, sizeof(*inode_item));
370 	inode_item->generation = cpu_to_le64(1);
371 	inode_item->size = cpu_to_le64(3);
372 	inode_item->nlink = cpu_to_le32(1);
373 	inode_item->nbytes = cpu_to_le64(root->leafsize);
374 	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
375 
376 	root_item.flags = 0;
377 	root_item.byte_limit = 0;
378 	inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT);
379 
380 	btrfs_set_root_bytenr(&root_item, leaf->start);
381 	btrfs_set_root_generation(&root_item, trans->transid);
382 	btrfs_set_root_level(&root_item, 0);
383 	btrfs_set_root_refs(&root_item, 1);
384 	btrfs_set_root_used(&root_item, leaf->len);
385 	btrfs_set_root_last_snapshot(&root_item, 0);
386 
387 	memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
388 	root_item.drop_level = 0;
389 
390 	btrfs_tree_unlock(leaf);
391 	free_extent_buffer(leaf);
392 	leaf = NULL;
393 
394 	btrfs_set_root_dirid(&root_item, new_dirid);
395 
396 	key.objectid = objectid;
397 	key.offset = 0;
398 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
399 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
400 				&root_item);
401 	if (ret)
402 		goto fail;
403 
404 	key.offset = (u64)-1;
405 	new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
406 	BUG_ON(IS_ERR(new_root));
407 
408 	btrfs_record_root_in_trans(trans, new_root);
409 
410 	ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
411 				       BTRFS_I(dir)->block_group);
412 	/*
413 	 * insert the directory item
414 	 */
415 	ret = btrfs_set_inode_index(dir, &index);
416 	BUG_ON(ret);
417 
418 	ret = btrfs_insert_dir_item(trans, root,
419 				    name, namelen, dir->i_ino, &key,
420 				    BTRFS_FT_DIR, index);
421 	if (ret)
422 		goto fail;
423 
424 	btrfs_i_size_write(dir, dir->i_size + namelen * 2);
425 	ret = btrfs_update_inode(trans, root, dir);
426 	BUG_ON(ret);
427 
428 	ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
429 				 objectid, root->root_key.objectid,
430 				 dir->i_ino, index, name, namelen);
431 
432 	BUG_ON(ret);
433 
434 	d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
435 fail:
436 	dput(parent);
437 	if (async_transid) {
438 		*async_transid = trans->transid;
439 		err = btrfs_commit_transaction_async(trans, root, 1);
440 	} else {
441 		err = btrfs_commit_transaction(trans, root);
442 	}
443 	if (err && !ret)
444 		ret = err;
445 	return ret;
446 }
447 
448 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
449 			   char *name, int namelen, u64 *async_transid,
450 			   bool readonly)
451 {
452 	struct inode *inode;
453 	struct dentry *parent;
454 	struct btrfs_pending_snapshot *pending_snapshot;
455 	struct btrfs_trans_handle *trans;
456 	int ret;
457 
458 	if (!root->ref_cows)
459 		return -EINVAL;
460 
461 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
462 	if (!pending_snapshot)
463 		return -ENOMEM;
464 
465 	btrfs_init_block_rsv(&pending_snapshot->block_rsv);
466 	pending_snapshot->dentry = dentry;
467 	pending_snapshot->root = root;
468 	pending_snapshot->readonly = readonly;
469 
470 	trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
471 	if (IS_ERR(trans)) {
472 		ret = PTR_ERR(trans);
473 		goto fail;
474 	}
475 
476 	ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
477 	BUG_ON(ret);
478 
479 	list_add(&pending_snapshot->list,
480 		 &trans->transaction->pending_snapshots);
481 	if (async_transid) {
482 		*async_transid = trans->transid;
483 		ret = btrfs_commit_transaction_async(trans,
484 				     root->fs_info->extent_root, 1);
485 	} else {
486 		ret = btrfs_commit_transaction(trans,
487 					       root->fs_info->extent_root);
488 	}
489 	BUG_ON(ret);
490 
491 	ret = pending_snapshot->error;
492 	if (ret)
493 		goto fail;
494 
495 	ret = btrfs_orphan_cleanup(pending_snapshot->snap);
496 	if (ret)
497 		goto fail;
498 
499 	parent = dget_parent(dentry);
500 	inode = btrfs_lookup_dentry(parent->d_inode, dentry);
501 	dput(parent);
502 	if (IS_ERR(inode)) {
503 		ret = PTR_ERR(inode);
504 		goto fail;
505 	}
506 	BUG_ON(!inode);
507 	d_instantiate(dentry, inode);
508 	ret = 0;
509 fail:
510 	kfree(pending_snapshot);
511 	return ret;
512 }
513 
514 /*  copy of check_sticky in fs/namei.c()
515 * It's inline, so penalty for filesystems that don't use sticky bit is
516 * minimal.
517 */
518 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
519 {
520 	uid_t fsuid = current_fsuid();
521 
522 	if (!(dir->i_mode & S_ISVTX))
523 		return 0;
524 	if (inode->i_uid == fsuid)
525 		return 0;
526 	if (dir->i_uid == fsuid)
527 		return 0;
528 	return !capable(CAP_FOWNER);
529 }
530 
531 /*  copy of may_delete in fs/namei.c()
532  *	Check whether we can remove a link victim from directory dir, check
533  *  whether the type of victim is right.
534  *  1. We can't do it if dir is read-only (done in permission())
535  *  2. We should have write and exec permissions on dir
536  *  3. We can't remove anything from append-only dir
537  *  4. We can't do anything with immutable dir (done in permission())
538  *  5. If the sticky bit on dir is set we should either
539  *	a. be owner of dir, or
540  *	b. be owner of victim, or
541  *	c. have CAP_FOWNER capability
542  *  6. If the victim is append-only or immutable we can't do antyhing with
543  *     links pointing to it.
544  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
545  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
546  *  9. We can't remove a root or mountpoint.
547  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
548  *     nfs_async_unlink().
549  */
550 
551 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
552 {
553 	int error;
554 
555 	if (!victim->d_inode)
556 		return -ENOENT;
557 
558 	BUG_ON(victim->d_parent->d_inode != dir);
559 	audit_inode_child(victim, dir);
560 
561 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
562 	if (error)
563 		return error;
564 	if (IS_APPEND(dir))
565 		return -EPERM;
566 	if (btrfs_check_sticky(dir, victim->d_inode)||
567 		IS_APPEND(victim->d_inode)||
568 	    IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
569 		return -EPERM;
570 	if (isdir) {
571 		if (!S_ISDIR(victim->d_inode->i_mode))
572 			return -ENOTDIR;
573 		if (IS_ROOT(victim))
574 			return -EBUSY;
575 	} else if (S_ISDIR(victim->d_inode->i_mode))
576 		return -EISDIR;
577 	if (IS_DEADDIR(dir))
578 		return -ENOENT;
579 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
580 		return -EBUSY;
581 	return 0;
582 }
583 
584 /* copy of may_create in fs/namei.c() */
585 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
586 {
587 	if (child->d_inode)
588 		return -EEXIST;
589 	if (IS_DEADDIR(dir))
590 		return -ENOENT;
591 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
592 }
593 
594 /*
595  * Create a new subvolume below @parent.  This is largely modeled after
596  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
597  * inside this filesystem so it's quite a bit simpler.
598  */
599 static noinline int btrfs_mksubvol(struct path *parent,
600 				   char *name, int namelen,
601 				   struct btrfs_root *snap_src,
602 				   u64 *async_transid, bool readonly)
603 {
604 	struct inode *dir  = parent->dentry->d_inode;
605 	struct dentry *dentry;
606 	int error;
607 
608 	mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
609 
610 	dentry = lookup_one_len(name, parent->dentry, namelen);
611 	error = PTR_ERR(dentry);
612 	if (IS_ERR(dentry))
613 		goto out_unlock;
614 
615 	error = -EEXIST;
616 	if (dentry->d_inode)
617 		goto out_dput;
618 
619 	error = mnt_want_write(parent->mnt);
620 	if (error)
621 		goto out_dput;
622 
623 	error = btrfs_may_create(dir, dentry);
624 	if (error)
625 		goto out_drop_write;
626 
627 	down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
628 
629 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
630 		goto out_up_read;
631 
632 	if (snap_src) {
633 		error = create_snapshot(snap_src, dentry,
634 					name, namelen, async_transid, readonly);
635 	} else {
636 		error = create_subvol(BTRFS_I(dir)->root, dentry,
637 				      name, namelen, async_transid);
638 	}
639 	if (!error)
640 		fsnotify_mkdir(dir, dentry);
641 out_up_read:
642 	up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
643 out_drop_write:
644 	mnt_drop_write(parent->mnt);
645 out_dput:
646 	dput(dentry);
647 out_unlock:
648 	mutex_unlock(&dir->i_mutex);
649 	return error;
650 }
651 
652 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
653 			       int thresh, u64 *last_len, u64 *skip,
654 			       u64 *defrag_end)
655 {
656 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
657 	struct extent_map *em = NULL;
658 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
659 	int ret = 1;
660 
661 
662 	if (thresh == 0)
663 		thresh = 256 * 1024;
664 
665 	/*
666 	 * make sure that once we start defragging and extent, we keep on
667 	 * defragging it
668 	 */
669 	if (start < *defrag_end)
670 		return 1;
671 
672 	*skip = 0;
673 
674 	/*
675 	 * hopefully we have this extent in the tree already, try without
676 	 * the full extent lock
677 	 */
678 	read_lock(&em_tree->lock);
679 	em = lookup_extent_mapping(em_tree, start, len);
680 	read_unlock(&em_tree->lock);
681 
682 	if (!em) {
683 		/* get the big lock and read metadata off disk */
684 		lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
685 		em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
686 		unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
687 
688 		if (IS_ERR(em))
689 			return 0;
690 	}
691 
692 	/* this will cover holes, and inline extents */
693 	if (em->block_start >= EXTENT_MAP_LAST_BYTE)
694 		ret = 0;
695 
696 	/*
697 	 * we hit a real extent, if it is big don't bother defragging it again
698 	 */
699 	if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
700 		ret = 0;
701 
702 	/*
703 	 * last_len ends up being a counter of how many bytes we've defragged.
704 	 * every time we choose not to defrag an extent, we reset *last_len
705 	 * so that the next tiny extent will force a defrag.
706 	 *
707 	 * The end result of this is that tiny extents before a single big
708 	 * extent will force at least part of that big extent to be defragged.
709 	 */
710 	if (ret) {
711 		*last_len += len;
712 		*defrag_end = extent_map_end(em);
713 	} else {
714 		*last_len = 0;
715 		*skip = extent_map_end(em);
716 		*defrag_end = 0;
717 	}
718 
719 	free_extent_map(em);
720 	return ret;
721 }
722 
723 static int btrfs_defrag_file(struct file *file,
724 			     struct btrfs_ioctl_defrag_range_args *range)
725 {
726 	struct inode *inode = fdentry(file)->d_inode;
727 	struct btrfs_root *root = BTRFS_I(inode)->root;
728 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
729 	struct btrfs_ordered_extent *ordered;
730 	struct page *page;
731 	struct btrfs_super_block *disk_super;
732 	unsigned long last_index;
733 	unsigned long ra_pages = root->fs_info->bdi.ra_pages;
734 	unsigned long total_read = 0;
735 	u64 features;
736 	u64 page_start;
737 	u64 page_end;
738 	u64 last_len = 0;
739 	u64 skip = 0;
740 	u64 defrag_end = 0;
741 	unsigned long i;
742 	int ret;
743 	int compress_type = BTRFS_COMPRESS_ZLIB;
744 
745 	if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
746 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
747 			return -EINVAL;
748 		if (range->compress_type)
749 			compress_type = range->compress_type;
750 	}
751 
752 	if (inode->i_size == 0)
753 		return 0;
754 
755 	if (range->start + range->len > range->start) {
756 		last_index = min_t(u64, inode->i_size - 1,
757 			 range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
758 	} else {
759 		last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
760 	}
761 
762 	i = range->start >> PAGE_CACHE_SHIFT;
763 	while (i <= last_index) {
764 		if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
765 					PAGE_CACHE_SIZE,
766 					range->extent_thresh,
767 					&last_len, &skip,
768 					&defrag_end)) {
769 			unsigned long next;
770 			/*
771 			 * the should_defrag function tells us how much to skip
772 			 * bump our counter by the suggested amount
773 			 */
774 			next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
775 			i = max(i + 1, next);
776 			continue;
777 		}
778 
779 		if (total_read % ra_pages == 0) {
780 			btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
781 				       min(last_index, i + ra_pages - 1));
782 		}
783 		total_read++;
784 		mutex_lock(&inode->i_mutex);
785 		if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
786 			BTRFS_I(inode)->force_compress = compress_type;
787 
788 		ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
789 		if (ret)
790 			goto err_unlock;
791 again:
792 		if (inode->i_size == 0 ||
793 		    i > ((inode->i_size - 1) >> PAGE_CACHE_SHIFT)) {
794 			ret = 0;
795 			goto err_reservations;
796 		}
797 
798 		page = grab_cache_page(inode->i_mapping, i);
799 		if (!page) {
800 			ret = -ENOMEM;
801 			goto err_reservations;
802 		}
803 
804 		if (!PageUptodate(page)) {
805 			btrfs_readpage(NULL, page);
806 			lock_page(page);
807 			if (!PageUptodate(page)) {
808 				unlock_page(page);
809 				page_cache_release(page);
810 				ret = -EIO;
811 				goto err_reservations;
812 			}
813 		}
814 
815 		if (page->mapping != inode->i_mapping) {
816 			unlock_page(page);
817 			page_cache_release(page);
818 			goto again;
819 		}
820 
821 		wait_on_page_writeback(page);
822 
823 		if (PageDirty(page)) {
824 			btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
825 			goto loop_unlock;
826 		}
827 
828 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
829 		page_end = page_start + PAGE_CACHE_SIZE - 1;
830 		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
831 
832 		ordered = btrfs_lookup_ordered_extent(inode, page_start);
833 		if (ordered) {
834 			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
835 			unlock_page(page);
836 			page_cache_release(page);
837 			btrfs_start_ordered_extent(inode, ordered, 1);
838 			btrfs_put_ordered_extent(ordered);
839 			goto again;
840 		}
841 		set_page_extent_mapped(page);
842 
843 		/*
844 		 * this makes sure page_mkwrite is called on the
845 		 * page if it is dirtied again later
846 		 */
847 		clear_page_dirty_for_io(page);
848 		clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start,
849 				  page_end, EXTENT_DIRTY | EXTENT_DELALLOC |
850 				  EXTENT_DO_ACCOUNTING, GFP_NOFS);
851 
852 		btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
853 		ClearPageChecked(page);
854 		set_page_dirty(page);
855 		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
856 
857 loop_unlock:
858 		unlock_page(page);
859 		page_cache_release(page);
860 		mutex_unlock(&inode->i_mutex);
861 
862 		balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
863 		i++;
864 	}
865 
866 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
867 		filemap_flush(inode->i_mapping);
868 
869 	if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
870 		/* the filemap_flush will queue IO into the worker threads, but
871 		 * we have to make sure the IO is actually started and that
872 		 * ordered extents get created before we return
873 		 */
874 		atomic_inc(&root->fs_info->async_submit_draining);
875 		while (atomic_read(&root->fs_info->nr_async_submits) ||
876 		      atomic_read(&root->fs_info->async_delalloc_pages)) {
877 			wait_event(root->fs_info->async_submit_wait,
878 			   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
879 			    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
880 		}
881 		atomic_dec(&root->fs_info->async_submit_draining);
882 
883 		mutex_lock(&inode->i_mutex);
884 		BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
885 		mutex_unlock(&inode->i_mutex);
886 	}
887 
888 	disk_super = &root->fs_info->super_copy;
889 	features = btrfs_super_incompat_flags(disk_super);
890 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
891 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
892 		btrfs_set_super_incompat_flags(disk_super, features);
893 	}
894 
895 	return 0;
896 
897 err_reservations:
898 	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
899 err_unlock:
900 	mutex_unlock(&inode->i_mutex);
901 	return ret;
902 }
903 
904 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
905 					void __user *arg)
906 {
907 	u64 new_size;
908 	u64 old_size;
909 	u64 devid = 1;
910 	struct btrfs_ioctl_vol_args *vol_args;
911 	struct btrfs_trans_handle *trans;
912 	struct btrfs_device *device = NULL;
913 	char *sizestr;
914 	char *devstr = NULL;
915 	int ret = 0;
916 	int mod = 0;
917 
918 	if (root->fs_info->sb->s_flags & MS_RDONLY)
919 		return -EROFS;
920 
921 	if (!capable(CAP_SYS_ADMIN))
922 		return -EPERM;
923 
924 	vol_args = memdup_user(arg, sizeof(*vol_args));
925 	if (IS_ERR(vol_args))
926 		return PTR_ERR(vol_args);
927 
928 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
929 
930 	mutex_lock(&root->fs_info->volume_mutex);
931 	sizestr = vol_args->name;
932 	devstr = strchr(sizestr, ':');
933 	if (devstr) {
934 		char *end;
935 		sizestr = devstr + 1;
936 		*devstr = '\0';
937 		devstr = vol_args->name;
938 		devid = simple_strtoull(devstr, &end, 10);
939 		printk(KERN_INFO "resizing devid %llu\n",
940 		       (unsigned long long)devid);
941 	}
942 	device = btrfs_find_device(root, devid, NULL, NULL);
943 	if (!device) {
944 		printk(KERN_INFO "resizer unable to find device %llu\n",
945 		       (unsigned long long)devid);
946 		ret = -EINVAL;
947 		goto out_unlock;
948 	}
949 	if (!strcmp(sizestr, "max"))
950 		new_size = device->bdev->bd_inode->i_size;
951 	else {
952 		if (sizestr[0] == '-') {
953 			mod = -1;
954 			sizestr++;
955 		} else if (sizestr[0] == '+') {
956 			mod = 1;
957 			sizestr++;
958 		}
959 		new_size = memparse(sizestr, NULL);
960 		if (new_size == 0) {
961 			ret = -EINVAL;
962 			goto out_unlock;
963 		}
964 	}
965 
966 	old_size = device->total_bytes;
967 
968 	if (mod < 0) {
969 		if (new_size > old_size) {
970 			ret = -EINVAL;
971 			goto out_unlock;
972 		}
973 		new_size = old_size - new_size;
974 	} else if (mod > 0) {
975 		new_size = old_size + new_size;
976 	}
977 
978 	if (new_size < 256 * 1024 * 1024) {
979 		ret = -EINVAL;
980 		goto out_unlock;
981 	}
982 	if (new_size > device->bdev->bd_inode->i_size) {
983 		ret = -EFBIG;
984 		goto out_unlock;
985 	}
986 
987 	do_div(new_size, root->sectorsize);
988 	new_size *= root->sectorsize;
989 
990 	printk(KERN_INFO "new size for %s is %llu\n",
991 		device->name, (unsigned long long)new_size);
992 
993 	if (new_size > old_size) {
994 		trans = btrfs_start_transaction(root, 0);
995 		if (IS_ERR(trans)) {
996 			ret = PTR_ERR(trans);
997 			goto out_unlock;
998 		}
999 		ret = btrfs_grow_device(trans, device, new_size);
1000 		btrfs_commit_transaction(trans, root);
1001 	} else {
1002 		ret = btrfs_shrink_device(device, new_size);
1003 	}
1004 
1005 out_unlock:
1006 	mutex_unlock(&root->fs_info->volume_mutex);
1007 	kfree(vol_args);
1008 	return ret;
1009 }
1010 
1011 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
1012 						    char *name,
1013 						    unsigned long fd,
1014 						    int subvol,
1015 						    u64 *transid,
1016 						    bool readonly)
1017 {
1018 	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
1019 	struct file *src_file;
1020 	int namelen;
1021 	int ret = 0;
1022 
1023 	if (root->fs_info->sb->s_flags & MS_RDONLY)
1024 		return -EROFS;
1025 
1026 	namelen = strlen(name);
1027 	if (strchr(name, '/')) {
1028 		ret = -EINVAL;
1029 		goto out;
1030 	}
1031 
1032 	if (subvol) {
1033 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1034 				     NULL, transid, readonly);
1035 	} else {
1036 		struct inode *src_inode;
1037 		src_file = fget(fd);
1038 		if (!src_file) {
1039 			ret = -EINVAL;
1040 			goto out;
1041 		}
1042 
1043 		src_inode = src_file->f_path.dentry->d_inode;
1044 		if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
1045 			printk(KERN_INFO "btrfs: Snapshot src from "
1046 			       "another FS\n");
1047 			ret = -EINVAL;
1048 			fput(src_file);
1049 			goto out;
1050 		}
1051 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
1052 				     BTRFS_I(src_inode)->root,
1053 				     transid, readonly);
1054 		fput(src_file);
1055 	}
1056 out:
1057 	return ret;
1058 }
1059 
1060 static noinline int btrfs_ioctl_snap_create(struct file *file,
1061 					    void __user *arg, int subvol)
1062 {
1063 	struct btrfs_ioctl_vol_args *vol_args;
1064 	int ret;
1065 
1066 	vol_args = memdup_user(arg, sizeof(*vol_args));
1067 	if (IS_ERR(vol_args))
1068 		return PTR_ERR(vol_args);
1069 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1070 
1071 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1072 					      vol_args->fd, subvol,
1073 					      NULL, false);
1074 
1075 	kfree(vol_args);
1076 	return ret;
1077 }
1078 
1079 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1080 					       void __user *arg, int subvol)
1081 {
1082 	struct btrfs_ioctl_vol_args_v2 *vol_args;
1083 	int ret;
1084 	u64 transid = 0;
1085 	u64 *ptr = NULL;
1086 	bool readonly = false;
1087 
1088 	vol_args = memdup_user(arg, sizeof(*vol_args));
1089 	if (IS_ERR(vol_args))
1090 		return PTR_ERR(vol_args);
1091 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1092 
1093 	if (vol_args->flags &
1094 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1095 		ret = -EOPNOTSUPP;
1096 		goto out;
1097 	}
1098 
1099 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1100 		ptr = &transid;
1101 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1102 		readonly = true;
1103 
1104 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1105 					      vol_args->fd, subvol,
1106 					      ptr, readonly);
1107 
1108 	if (ret == 0 && ptr &&
1109 	    copy_to_user(arg +
1110 			 offsetof(struct btrfs_ioctl_vol_args_v2,
1111 				  transid), ptr, sizeof(*ptr)))
1112 		ret = -EFAULT;
1113 out:
1114 	kfree(vol_args);
1115 	return ret;
1116 }
1117 
1118 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1119 						void __user *arg)
1120 {
1121 	struct inode *inode = fdentry(file)->d_inode;
1122 	struct btrfs_root *root = BTRFS_I(inode)->root;
1123 	int ret = 0;
1124 	u64 flags = 0;
1125 
1126 	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID)
1127 		return -EINVAL;
1128 
1129 	down_read(&root->fs_info->subvol_sem);
1130 	if (btrfs_root_readonly(root))
1131 		flags |= BTRFS_SUBVOL_RDONLY;
1132 	up_read(&root->fs_info->subvol_sem);
1133 
1134 	if (copy_to_user(arg, &flags, sizeof(flags)))
1135 		ret = -EFAULT;
1136 
1137 	return ret;
1138 }
1139 
1140 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1141 					      void __user *arg)
1142 {
1143 	struct inode *inode = fdentry(file)->d_inode;
1144 	struct btrfs_root *root = BTRFS_I(inode)->root;
1145 	struct btrfs_trans_handle *trans;
1146 	u64 root_flags;
1147 	u64 flags;
1148 	int ret = 0;
1149 
1150 	if (root->fs_info->sb->s_flags & MS_RDONLY)
1151 		return -EROFS;
1152 
1153 	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID)
1154 		return -EINVAL;
1155 
1156 	if (copy_from_user(&flags, arg, sizeof(flags)))
1157 		return -EFAULT;
1158 
1159 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1160 		return -EINVAL;
1161 
1162 	if (flags & ~BTRFS_SUBVOL_RDONLY)
1163 		return -EOPNOTSUPP;
1164 
1165 	if (!inode_owner_or_capable(inode))
1166 		return -EACCES;
1167 
1168 	down_write(&root->fs_info->subvol_sem);
1169 
1170 	/* nothing to do */
1171 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1172 		goto out;
1173 
1174 	root_flags = btrfs_root_flags(&root->root_item);
1175 	if (flags & BTRFS_SUBVOL_RDONLY)
1176 		btrfs_set_root_flags(&root->root_item,
1177 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1178 	else
1179 		btrfs_set_root_flags(&root->root_item,
1180 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1181 
1182 	trans = btrfs_start_transaction(root, 1);
1183 	if (IS_ERR(trans)) {
1184 		ret = PTR_ERR(trans);
1185 		goto out_reset;
1186 	}
1187 
1188 	ret = btrfs_update_root(trans, root->fs_info->tree_root,
1189 				&root->root_key, &root->root_item);
1190 
1191 	btrfs_commit_transaction(trans, root);
1192 out_reset:
1193 	if (ret)
1194 		btrfs_set_root_flags(&root->root_item, root_flags);
1195 out:
1196 	up_write(&root->fs_info->subvol_sem);
1197 	return ret;
1198 }
1199 
1200 /*
1201  * helper to check if the subvolume references other subvolumes
1202  */
1203 static noinline int may_destroy_subvol(struct btrfs_root *root)
1204 {
1205 	struct btrfs_path *path;
1206 	struct btrfs_key key;
1207 	int ret;
1208 
1209 	path = btrfs_alloc_path();
1210 	if (!path)
1211 		return -ENOMEM;
1212 
1213 	key.objectid = root->root_key.objectid;
1214 	key.type = BTRFS_ROOT_REF_KEY;
1215 	key.offset = (u64)-1;
1216 
1217 	ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1218 				&key, path, 0, 0);
1219 	if (ret < 0)
1220 		goto out;
1221 	BUG_ON(ret == 0);
1222 
1223 	ret = 0;
1224 	if (path->slots[0] > 0) {
1225 		path->slots[0]--;
1226 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1227 		if (key.objectid == root->root_key.objectid &&
1228 		    key.type == BTRFS_ROOT_REF_KEY)
1229 			ret = -ENOTEMPTY;
1230 	}
1231 out:
1232 	btrfs_free_path(path);
1233 	return ret;
1234 }
1235 
1236 static noinline int key_in_sk(struct btrfs_key *key,
1237 			      struct btrfs_ioctl_search_key *sk)
1238 {
1239 	struct btrfs_key test;
1240 	int ret;
1241 
1242 	test.objectid = sk->min_objectid;
1243 	test.type = sk->min_type;
1244 	test.offset = sk->min_offset;
1245 
1246 	ret = btrfs_comp_cpu_keys(key, &test);
1247 	if (ret < 0)
1248 		return 0;
1249 
1250 	test.objectid = sk->max_objectid;
1251 	test.type = sk->max_type;
1252 	test.offset = sk->max_offset;
1253 
1254 	ret = btrfs_comp_cpu_keys(key, &test);
1255 	if (ret > 0)
1256 		return 0;
1257 	return 1;
1258 }
1259 
1260 static noinline int copy_to_sk(struct btrfs_root *root,
1261 			       struct btrfs_path *path,
1262 			       struct btrfs_key *key,
1263 			       struct btrfs_ioctl_search_key *sk,
1264 			       char *buf,
1265 			       unsigned long *sk_offset,
1266 			       int *num_found)
1267 {
1268 	u64 found_transid;
1269 	struct extent_buffer *leaf;
1270 	struct btrfs_ioctl_search_header sh;
1271 	unsigned long item_off;
1272 	unsigned long item_len;
1273 	int nritems;
1274 	int i;
1275 	int slot;
1276 	int found = 0;
1277 	int ret = 0;
1278 
1279 	leaf = path->nodes[0];
1280 	slot = path->slots[0];
1281 	nritems = btrfs_header_nritems(leaf);
1282 
1283 	if (btrfs_header_generation(leaf) > sk->max_transid) {
1284 		i = nritems;
1285 		goto advance_key;
1286 	}
1287 	found_transid = btrfs_header_generation(leaf);
1288 
1289 	for (i = slot; i < nritems; i++) {
1290 		item_off = btrfs_item_ptr_offset(leaf, i);
1291 		item_len = btrfs_item_size_nr(leaf, i);
1292 
1293 		if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1294 			item_len = 0;
1295 
1296 		if (sizeof(sh) + item_len + *sk_offset >
1297 		    BTRFS_SEARCH_ARGS_BUFSIZE) {
1298 			ret = 1;
1299 			goto overflow;
1300 		}
1301 
1302 		btrfs_item_key_to_cpu(leaf, key, i);
1303 		if (!key_in_sk(key, sk))
1304 			continue;
1305 
1306 		sh.objectid = key->objectid;
1307 		sh.offset = key->offset;
1308 		sh.type = key->type;
1309 		sh.len = item_len;
1310 		sh.transid = found_transid;
1311 
1312 		/* copy search result header */
1313 		memcpy(buf + *sk_offset, &sh, sizeof(sh));
1314 		*sk_offset += sizeof(sh);
1315 
1316 		if (item_len) {
1317 			char *p = buf + *sk_offset;
1318 			/* copy the item */
1319 			read_extent_buffer(leaf, p,
1320 					   item_off, item_len);
1321 			*sk_offset += item_len;
1322 		}
1323 		found++;
1324 
1325 		if (*num_found >= sk->nr_items)
1326 			break;
1327 	}
1328 advance_key:
1329 	ret = 0;
1330 	if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1331 		key->offset++;
1332 	else if (key->type < (u8)-1 && key->type < sk->max_type) {
1333 		key->offset = 0;
1334 		key->type++;
1335 	} else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1336 		key->offset = 0;
1337 		key->type = 0;
1338 		key->objectid++;
1339 	} else
1340 		ret = 1;
1341 overflow:
1342 	*num_found += found;
1343 	return ret;
1344 }
1345 
1346 static noinline int search_ioctl(struct inode *inode,
1347 				 struct btrfs_ioctl_search_args *args)
1348 {
1349 	struct btrfs_root *root;
1350 	struct btrfs_key key;
1351 	struct btrfs_key max_key;
1352 	struct btrfs_path *path;
1353 	struct btrfs_ioctl_search_key *sk = &args->key;
1354 	struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1355 	int ret;
1356 	int num_found = 0;
1357 	unsigned long sk_offset = 0;
1358 
1359 	path = btrfs_alloc_path();
1360 	if (!path)
1361 		return -ENOMEM;
1362 
1363 	if (sk->tree_id == 0) {
1364 		/* search the root of the inode that was passed */
1365 		root = BTRFS_I(inode)->root;
1366 	} else {
1367 		key.objectid = sk->tree_id;
1368 		key.type = BTRFS_ROOT_ITEM_KEY;
1369 		key.offset = (u64)-1;
1370 		root = btrfs_read_fs_root_no_name(info, &key);
1371 		if (IS_ERR(root)) {
1372 			printk(KERN_ERR "could not find root %llu\n",
1373 			       sk->tree_id);
1374 			btrfs_free_path(path);
1375 			return -ENOENT;
1376 		}
1377 	}
1378 
1379 	key.objectid = sk->min_objectid;
1380 	key.type = sk->min_type;
1381 	key.offset = sk->min_offset;
1382 
1383 	max_key.objectid = sk->max_objectid;
1384 	max_key.type = sk->max_type;
1385 	max_key.offset = sk->max_offset;
1386 
1387 	path->keep_locks = 1;
1388 
1389 	while(1) {
1390 		ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1391 					   sk->min_transid);
1392 		if (ret != 0) {
1393 			if (ret > 0)
1394 				ret = 0;
1395 			goto err;
1396 		}
1397 		ret = copy_to_sk(root, path, &key, sk, args->buf,
1398 				 &sk_offset, &num_found);
1399 		btrfs_release_path(root, path);
1400 		if (ret || num_found >= sk->nr_items)
1401 			break;
1402 
1403 	}
1404 	ret = 0;
1405 err:
1406 	sk->nr_items = num_found;
1407 	btrfs_free_path(path);
1408 	return ret;
1409 }
1410 
1411 static noinline int btrfs_ioctl_tree_search(struct file *file,
1412 					   void __user *argp)
1413 {
1414 	 struct btrfs_ioctl_search_args *args;
1415 	 struct inode *inode;
1416 	 int ret;
1417 
1418 	if (!capable(CAP_SYS_ADMIN))
1419 		return -EPERM;
1420 
1421 	args = memdup_user(argp, sizeof(*args));
1422 	if (IS_ERR(args))
1423 		return PTR_ERR(args);
1424 
1425 	inode = fdentry(file)->d_inode;
1426 	ret = search_ioctl(inode, args);
1427 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1428 		ret = -EFAULT;
1429 	kfree(args);
1430 	return ret;
1431 }
1432 
1433 /*
1434  * Search INODE_REFs to identify path name of 'dirid' directory
1435  * in a 'tree_id' tree. and sets path name to 'name'.
1436  */
1437 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1438 				u64 tree_id, u64 dirid, char *name)
1439 {
1440 	struct btrfs_root *root;
1441 	struct btrfs_key key;
1442 	char *ptr;
1443 	int ret = -1;
1444 	int slot;
1445 	int len;
1446 	int total_len = 0;
1447 	struct btrfs_inode_ref *iref;
1448 	struct extent_buffer *l;
1449 	struct btrfs_path *path;
1450 
1451 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1452 		name[0]='\0';
1453 		return 0;
1454 	}
1455 
1456 	path = btrfs_alloc_path();
1457 	if (!path)
1458 		return -ENOMEM;
1459 
1460 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1461 
1462 	key.objectid = tree_id;
1463 	key.type = BTRFS_ROOT_ITEM_KEY;
1464 	key.offset = (u64)-1;
1465 	root = btrfs_read_fs_root_no_name(info, &key);
1466 	if (IS_ERR(root)) {
1467 		printk(KERN_ERR "could not find root %llu\n", tree_id);
1468 		ret = -ENOENT;
1469 		goto out;
1470 	}
1471 
1472 	key.objectid = dirid;
1473 	key.type = BTRFS_INODE_REF_KEY;
1474 	key.offset = (u64)-1;
1475 
1476 	while(1) {
1477 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1478 		if (ret < 0)
1479 			goto out;
1480 
1481 		l = path->nodes[0];
1482 		slot = path->slots[0];
1483 		if (ret > 0 && slot > 0)
1484 			slot--;
1485 		btrfs_item_key_to_cpu(l, &key, slot);
1486 
1487 		if (ret > 0 && (key.objectid != dirid ||
1488 				key.type != BTRFS_INODE_REF_KEY)) {
1489 			ret = -ENOENT;
1490 			goto out;
1491 		}
1492 
1493 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1494 		len = btrfs_inode_ref_name_len(l, iref);
1495 		ptr -= len + 1;
1496 		total_len += len + 1;
1497 		if (ptr < name)
1498 			goto out;
1499 
1500 		*(ptr + len) = '/';
1501 		read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1502 
1503 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1504 			break;
1505 
1506 		btrfs_release_path(root, path);
1507 		key.objectid = key.offset;
1508 		key.offset = (u64)-1;
1509 		dirid = key.objectid;
1510 
1511 	}
1512 	if (ptr < name)
1513 		goto out;
1514 	memcpy(name, ptr, total_len);
1515 	name[total_len]='\0';
1516 	ret = 0;
1517 out:
1518 	btrfs_free_path(path);
1519 	return ret;
1520 }
1521 
1522 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1523 					   void __user *argp)
1524 {
1525 	 struct btrfs_ioctl_ino_lookup_args *args;
1526 	 struct inode *inode;
1527 	 int ret;
1528 
1529 	if (!capable(CAP_SYS_ADMIN))
1530 		return -EPERM;
1531 
1532 	args = memdup_user(argp, sizeof(*args));
1533 	if (IS_ERR(args))
1534 		return PTR_ERR(args);
1535 
1536 	inode = fdentry(file)->d_inode;
1537 
1538 	if (args->treeid == 0)
1539 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1540 
1541 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1542 					args->treeid, args->objectid,
1543 					args->name);
1544 
1545 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1546 		ret = -EFAULT;
1547 
1548 	kfree(args);
1549 	return ret;
1550 }
1551 
1552 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1553 					     void __user *arg)
1554 {
1555 	struct dentry *parent = fdentry(file);
1556 	struct dentry *dentry;
1557 	struct inode *dir = parent->d_inode;
1558 	struct inode *inode;
1559 	struct btrfs_root *root = BTRFS_I(dir)->root;
1560 	struct btrfs_root *dest = NULL;
1561 	struct btrfs_ioctl_vol_args *vol_args;
1562 	struct btrfs_trans_handle *trans;
1563 	int namelen;
1564 	int ret;
1565 	int err = 0;
1566 
1567 	vol_args = memdup_user(arg, sizeof(*vol_args));
1568 	if (IS_ERR(vol_args))
1569 		return PTR_ERR(vol_args);
1570 
1571 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1572 	namelen = strlen(vol_args->name);
1573 	if (strchr(vol_args->name, '/') ||
1574 	    strncmp(vol_args->name, "..", namelen) == 0) {
1575 		err = -EINVAL;
1576 		goto out;
1577 	}
1578 
1579 	err = mnt_want_write(file->f_path.mnt);
1580 	if (err)
1581 		goto out;
1582 
1583 	mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1584 	dentry = lookup_one_len(vol_args->name, parent, namelen);
1585 	if (IS_ERR(dentry)) {
1586 		err = PTR_ERR(dentry);
1587 		goto out_unlock_dir;
1588 	}
1589 
1590 	if (!dentry->d_inode) {
1591 		err = -ENOENT;
1592 		goto out_dput;
1593 	}
1594 
1595 	inode = dentry->d_inode;
1596 	dest = BTRFS_I(inode)->root;
1597 	if (!capable(CAP_SYS_ADMIN)){
1598 		/*
1599 		 * Regular user.  Only allow this with a special mount
1600 		 * option, when the user has write+exec access to the
1601 		 * subvol root, and when rmdir(2) would have been
1602 		 * allowed.
1603 		 *
1604 		 * Note that this is _not_ check that the subvol is
1605 		 * empty or doesn't contain data that we wouldn't
1606 		 * otherwise be able to delete.
1607 		 *
1608 		 * Users who want to delete empty subvols should try
1609 		 * rmdir(2).
1610 		 */
1611 		err = -EPERM;
1612 		if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1613 			goto out_dput;
1614 
1615 		/*
1616 		 * Do not allow deletion if the parent dir is the same
1617 		 * as the dir to be deleted.  That means the ioctl
1618 		 * must be called on the dentry referencing the root
1619 		 * of the subvol, not a random directory contained
1620 		 * within it.
1621 		 */
1622 		err = -EINVAL;
1623 		if (root == dest)
1624 			goto out_dput;
1625 
1626 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1627 		if (err)
1628 			goto out_dput;
1629 
1630 		/* check if subvolume may be deleted by a non-root user */
1631 		err = btrfs_may_delete(dir, dentry, 1);
1632 		if (err)
1633 			goto out_dput;
1634 	}
1635 
1636 	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
1637 		err = -EINVAL;
1638 		goto out_dput;
1639 	}
1640 
1641 	mutex_lock(&inode->i_mutex);
1642 	err = d_invalidate(dentry);
1643 	if (err)
1644 		goto out_unlock;
1645 
1646 	down_write(&root->fs_info->subvol_sem);
1647 
1648 	err = may_destroy_subvol(dest);
1649 	if (err)
1650 		goto out_up_write;
1651 
1652 	trans = btrfs_start_transaction(root, 0);
1653 	if (IS_ERR(trans)) {
1654 		err = PTR_ERR(trans);
1655 		goto out_up_write;
1656 	}
1657 	trans->block_rsv = &root->fs_info->global_block_rsv;
1658 
1659 	ret = btrfs_unlink_subvol(trans, root, dir,
1660 				dest->root_key.objectid,
1661 				dentry->d_name.name,
1662 				dentry->d_name.len);
1663 	BUG_ON(ret);
1664 
1665 	btrfs_record_root_in_trans(trans, dest);
1666 
1667 	memset(&dest->root_item.drop_progress, 0,
1668 		sizeof(dest->root_item.drop_progress));
1669 	dest->root_item.drop_level = 0;
1670 	btrfs_set_root_refs(&dest->root_item, 0);
1671 
1672 	if (!xchg(&dest->orphan_item_inserted, 1)) {
1673 		ret = btrfs_insert_orphan_item(trans,
1674 					root->fs_info->tree_root,
1675 					dest->root_key.objectid);
1676 		BUG_ON(ret);
1677 	}
1678 
1679 	ret = btrfs_end_transaction(trans, root);
1680 	BUG_ON(ret);
1681 	inode->i_flags |= S_DEAD;
1682 out_up_write:
1683 	up_write(&root->fs_info->subvol_sem);
1684 out_unlock:
1685 	mutex_unlock(&inode->i_mutex);
1686 	if (!err) {
1687 		shrink_dcache_sb(root->fs_info->sb);
1688 		btrfs_invalidate_inodes(dest);
1689 		d_delete(dentry);
1690 	}
1691 out_dput:
1692 	dput(dentry);
1693 out_unlock_dir:
1694 	mutex_unlock(&dir->i_mutex);
1695 	mnt_drop_write(file->f_path.mnt);
1696 out:
1697 	kfree(vol_args);
1698 	return err;
1699 }
1700 
1701 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1702 {
1703 	struct inode *inode = fdentry(file)->d_inode;
1704 	struct btrfs_root *root = BTRFS_I(inode)->root;
1705 	struct btrfs_ioctl_defrag_range_args *range;
1706 	int ret;
1707 
1708 	if (btrfs_root_readonly(root))
1709 		return -EROFS;
1710 
1711 	ret = mnt_want_write(file->f_path.mnt);
1712 	if (ret)
1713 		return ret;
1714 
1715 	switch (inode->i_mode & S_IFMT) {
1716 	case S_IFDIR:
1717 		if (!capable(CAP_SYS_ADMIN)) {
1718 			ret = -EPERM;
1719 			goto out;
1720 		}
1721 		ret = btrfs_defrag_root(root, 0);
1722 		if (ret)
1723 			goto out;
1724 		ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1725 		break;
1726 	case S_IFREG:
1727 		if (!(file->f_mode & FMODE_WRITE)) {
1728 			ret = -EINVAL;
1729 			goto out;
1730 		}
1731 
1732 		range = kzalloc(sizeof(*range), GFP_KERNEL);
1733 		if (!range) {
1734 			ret = -ENOMEM;
1735 			goto out;
1736 		}
1737 
1738 		if (argp) {
1739 			if (copy_from_user(range, argp,
1740 					   sizeof(*range))) {
1741 				ret = -EFAULT;
1742 				kfree(range);
1743 				goto out;
1744 			}
1745 			/* compression requires us to start the IO */
1746 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1747 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1748 				range->extent_thresh = (u32)-1;
1749 			}
1750 		} else {
1751 			/* the rest are all set to zero by kzalloc */
1752 			range->len = (u64)-1;
1753 		}
1754 		ret = btrfs_defrag_file(file, range);
1755 		kfree(range);
1756 		break;
1757 	default:
1758 		ret = -EINVAL;
1759 	}
1760 out:
1761 	mnt_drop_write(file->f_path.mnt);
1762 	return ret;
1763 }
1764 
1765 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
1766 {
1767 	struct btrfs_ioctl_vol_args *vol_args;
1768 	int ret;
1769 
1770 	if (!capable(CAP_SYS_ADMIN))
1771 		return -EPERM;
1772 
1773 	vol_args = memdup_user(arg, sizeof(*vol_args));
1774 	if (IS_ERR(vol_args))
1775 		return PTR_ERR(vol_args);
1776 
1777 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1778 	ret = btrfs_init_new_device(root, vol_args->name);
1779 
1780 	kfree(vol_args);
1781 	return ret;
1782 }
1783 
1784 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
1785 {
1786 	struct btrfs_ioctl_vol_args *vol_args;
1787 	int ret;
1788 
1789 	if (!capable(CAP_SYS_ADMIN))
1790 		return -EPERM;
1791 
1792 	if (root->fs_info->sb->s_flags & MS_RDONLY)
1793 		return -EROFS;
1794 
1795 	vol_args = memdup_user(arg, sizeof(*vol_args));
1796 	if (IS_ERR(vol_args))
1797 		return PTR_ERR(vol_args);
1798 
1799 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1800 	ret = btrfs_rm_device(root, vol_args->name);
1801 
1802 	kfree(vol_args);
1803 	return ret;
1804 }
1805 
1806 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1807 				       u64 off, u64 olen, u64 destoff)
1808 {
1809 	struct inode *inode = fdentry(file)->d_inode;
1810 	struct btrfs_root *root = BTRFS_I(inode)->root;
1811 	struct file *src_file;
1812 	struct inode *src;
1813 	struct btrfs_trans_handle *trans;
1814 	struct btrfs_path *path;
1815 	struct extent_buffer *leaf;
1816 	char *buf;
1817 	struct btrfs_key key;
1818 	u32 nritems;
1819 	int slot;
1820 	int ret;
1821 	u64 len = olen;
1822 	u64 bs = root->fs_info->sb->s_blocksize;
1823 	u64 hint_byte;
1824 
1825 	/*
1826 	 * TODO:
1827 	 * - split compressed inline extents.  annoying: we need to
1828 	 *   decompress into destination's address_space (the file offset
1829 	 *   may change, so source mapping won't do), then recompress (or
1830 	 *   otherwise reinsert) a subrange.
1831 	 * - allow ranges within the same file to be cloned (provided
1832 	 *   they don't overlap)?
1833 	 */
1834 
1835 	/* the destination must be opened for writing */
1836 	if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
1837 		return -EINVAL;
1838 
1839 	if (btrfs_root_readonly(root))
1840 		return -EROFS;
1841 
1842 	ret = mnt_want_write(file->f_path.mnt);
1843 	if (ret)
1844 		return ret;
1845 
1846 	src_file = fget(srcfd);
1847 	if (!src_file) {
1848 		ret = -EBADF;
1849 		goto out_drop_write;
1850 	}
1851 
1852 	src = src_file->f_dentry->d_inode;
1853 
1854 	ret = -EINVAL;
1855 	if (src == inode)
1856 		goto out_fput;
1857 
1858 	/* the src must be open for reading */
1859 	if (!(src_file->f_mode & FMODE_READ))
1860 		goto out_fput;
1861 
1862 	ret = -EISDIR;
1863 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
1864 		goto out_fput;
1865 
1866 	ret = -EXDEV;
1867 	if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
1868 		goto out_fput;
1869 
1870 	ret = -ENOMEM;
1871 	buf = vmalloc(btrfs_level_size(root, 0));
1872 	if (!buf)
1873 		goto out_fput;
1874 
1875 	path = btrfs_alloc_path();
1876 	if (!path) {
1877 		vfree(buf);
1878 		goto out_fput;
1879 	}
1880 	path->reada = 2;
1881 
1882 	if (inode < src) {
1883 		mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
1884 		mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
1885 	} else {
1886 		mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
1887 		mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1888 	}
1889 
1890 	/* determine range to clone */
1891 	ret = -EINVAL;
1892 	if (off + len > src->i_size || off + len < off)
1893 		goto out_unlock;
1894 	if (len == 0)
1895 		olen = len = src->i_size - off;
1896 	/* if we extend to eof, continue to block boundary */
1897 	if (off + len == src->i_size)
1898 		len = ALIGN(src->i_size, bs) - off;
1899 
1900 	/* verify the end result is block aligned */
1901 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
1902 	    !IS_ALIGNED(destoff, bs))
1903 		goto out_unlock;
1904 
1905 	/* do any pending delalloc/csum calc on src, one way or
1906 	   another, and lock file content */
1907 	while (1) {
1908 		struct btrfs_ordered_extent *ordered;
1909 		lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1910 		ordered = btrfs_lookup_first_ordered_extent(src, off+len);
1911 		if (!ordered &&
1912 		    !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
1913 				   EXTENT_DELALLOC, 0, NULL))
1914 			break;
1915 		unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1916 		if (ordered)
1917 			btrfs_put_ordered_extent(ordered);
1918 		btrfs_wait_ordered_range(src, off, len);
1919 	}
1920 
1921 	/* clone data */
1922 	key.objectid = src->i_ino;
1923 	key.type = BTRFS_EXTENT_DATA_KEY;
1924 	key.offset = 0;
1925 
1926 	while (1) {
1927 		/*
1928 		 * note the key will change type as we walk through the
1929 		 * tree.
1930 		 */
1931 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1932 		if (ret < 0)
1933 			goto out;
1934 
1935 		nritems = btrfs_header_nritems(path->nodes[0]);
1936 		if (path->slots[0] >= nritems) {
1937 			ret = btrfs_next_leaf(root, path);
1938 			if (ret < 0)
1939 				goto out;
1940 			if (ret > 0)
1941 				break;
1942 			nritems = btrfs_header_nritems(path->nodes[0]);
1943 		}
1944 		leaf = path->nodes[0];
1945 		slot = path->slots[0];
1946 
1947 		btrfs_item_key_to_cpu(leaf, &key, slot);
1948 		if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1949 		    key.objectid != src->i_ino)
1950 			break;
1951 
1952 		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1953 			struct btrfs_file_extent_item *extent;
1954 			int type;
1955 			u32 size;
1956 			struct btrfs_key new_key;
1957 			u64 disko = 0, diskl = 0;
1958 			u64 datao = 0, datal = 0;
1959 			u8 comp;
1960 			u64 endoff;
1961 
1962 			size = btrfs_item_size_nr(leaf, slot);
1963 			read_extent_buffer(leaf, buf,
1964 					   btrfs_item_ptr_offset(leaf, slot),
1965 					   size);
1966 
1967 			extent = btrfs_item_ptr(leaf, slot,
1968 						struct btrfs_file_extent_item);
1969 			comp = btrfs_file_extent_compression(leaf, extent);
1970 			type = btrfs_file_extent_type(leaf, extent);
1971 			if (type == BTRFS_FILE_EXTENT_REG ||
1972 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
1973 				disko = btrfs_file_extent_disk_bytenr(leaf,
1974 								      extent);
1975 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
1976 								 extent);
1977 				datao = btrfs_file_extent_offset(leaf, extent);
1978 				datal = btrfs_file_extent_num_bytes(leaf,
1979 								    extent);
1980 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
1981 				/* take upper bound, may be compressed */
1982 				datal = btrfs_file_extent_ram_bytes(leaf,
1983 								    extent);
1984 			}
1985 			btrfs_release_path(root, path);
1986 
1987 			if (key.offset + datal <= off ||
1988 			    key.offset >= off+len)
1989 				goto next;
1990 
1991 			memcpy(&new_key, &key, sizeof(new_key));
1992 			new_key.objectid = inode->i_ino;
1993 			if (off <= key.offset)
1994 				new_key.offset = key.offset + destoff - off;
1995 			else
1996 				new_key.offset = destoff;
1997 
1998 			trans = btrfs_start_transaction(root, 1);
1999 			if (IS_ERR(trans)) {
2000 				ret = PTR_ERR(trans);
2001 				goto out;
2002 			}
2003 
2004 			if (type == BTRFS_FILE_EXTENT_REG ||
2005 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
2006 				if (off > key.offset) {
2007 					datao += off - key.offset;
2008 					datal -= off - key.offset;
2009 				}
2010 
2011 				if (key.offset + datal > off + len)
2012 					datal = off + len - key.offset;
2013 
2014 				ret = btrfs_drop_extents(trans, inode,
2015 							 new_key.offset,
2016 							 new_key.offset + datal,
2017 							 &hint_byte, 1);
2018 				BUG_ON(ret);
2019 
2020 				ret = btrfs_insert_empty_item(trans, root, path,
2021 							      &new_key, size);
2022 				BUG_ON(ret);
2023 
2024 				leaf = path->nodes[0];
2025 				slot = path->slots[0];
2026 				write_extent_buffer(leaf, buf,
2027 					    btrfs_item_ptr_offset(leaf, slot),
2028 					    size);
2029 
2030 				extent = btrfs_item_ptr(leaf, slot,
2031 						struct btrfs_file_extent_item);
2032 
2033 				/* disko == 0 means it's a hole */
2034 				if (!disko)
2035 					datao = 0;
2036 
2037 				btrfs_set_file_extent_offset(leaf, extent,
2038 							     datao);
2039 				btrfs_set_file_extent_num_bytes(leaf, extent,
2040 								datal);
2041 				if (disko) {
2042 					inode_add_bytes(inode, datal);
2043 					ret = btrfs_inc_extent_ref(trans, root,
2044 							disko, diskl, 0,
2045 							root->root_key.objectid,
2046 							inode->i_ino,
2047 							new_key.offset - datao);
2048 					BUG_ON(ret);
2049 				}
2050 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
2051 				u64 skip = 0;
2052 				u64 trim = 0;
2053 				if (off > key.offset) {
2054 					skip = off - key.offset;
2055 					new_key.offset += skip;
2056 				}
2057 
2058 				if (key.offset + datal > off+len)
2059 					trim = key.offset + datal - (off+len);
2060 
2061 				if (comp && (skip || trim)) {
2062 					ret = -EINVAL;
2063 					btrfs_end_transaction(trans, root);
2064 					goto out;
2065 				}
2066 				size -= skip + trim;
2067 				datal -= skip + trim;
2068 
2069 				ret = btrfs_drop_extents(trans, inode,
2070 							 new_key.offset,
2071 							 new_key.offset + datal,
2072 							 &hint_byte, 1);
2073 				BUG_ON(ret);
2074 
2075 				ret = btrfs_insert_empty_item(trans, root, path,
2076 							      &new_key, size);
2077 				BUG_ON(ret);
2078 
2079 				if (skip) {
2080 					u32 start =
2081 					  btrfs_file_extent_calc_inline_size(0);
2082 					memmove(buf+start, buf+start+skip,
2083 						datal);
2084 				}
2085 
2086 				leaf = path->nodes[0];
2087 				slot = path->slots[0];
2088 				write_extent_buffer(leaf, buf,
2089 					    btrfs_item_ptr_offset(leaf, slot),
2090 					    size);
2091 				inode_add_bytes(inode, datal);
2092 			}
2093 
2094 			btrfs_mark_buffer_dirty(leaf);
2095 			btrfs_release_path(root, path);
2096 
2097 			inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2098 
2099 			/*
2100 			 * we round up to the block size at eof when
2101 			 * determining which extents to clone above,
2102 			 * but shouldn't round up the file size
2103 			 */
2104 			endoff = new_key.offset + datal;
2105 			if (endoff > destoff+olen)
2106 				endoff = destoff+olen;
2107 			if (endoff > inode->i_size)
2108 				btrfs_i_size_write(inode, endoff);
2109 
2110 			BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
2111 			ret = btrfs_update_inode(trans, root, inode);
2112 			BUG_ON(ret);
2113 			btrfs_end_transaction(trans, root);
2114 		}
2115 next:
2116 		btrfs_release_path(root, path);
2117 		key.offset++;
2118 	}
2119 	ret = 0;
2120 out:
2121 	btrfs_release_path(root, path);
2122 	unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2123 out_unlock:
2124 	mutex_unlock(&src->i_mutex);
2125 	mutex_unlock(&inode->i_mutex);
2126 	vfree(buf);
2127 	btrfs_free_path(path);
2128 out_fput:
2129 	fput(src_file);
2130 out_drop_write:
2131 	mnt_drop_write(file->f_path.mnt);
2132 	return ret;
2133 }
2134 
2135 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2136 {
2137 	struct btrfs_ioctl_clone_range_args args;
2138 
2139 	if (copy_from_user(&args, argp, sizeof(args)))
2140 		return -EFAULT;
2141 	return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2142 				 args.src_length, args.dest_offset);
2143 }
2144 
2145 /*
2146  * there are many ways the trans_start and trans_end ioctls can lead
2147  * to deadlocks.  They should only be used by applications that
2148  * basically own the machine, and have a very in depth understanding
2149  * of all the possible deadlocks and enospc problems.
2150  */
2151 static long btrfs_ioctl_trans_start(struct file *file)
2152 {
2153 	struct inode *inode = fdentry(file)->d_inode;
2154 	struct btrfs_root *root = BTRFS_I(inode)->root;
2155 	struct btrfs_trans_handle *trans;
2156 	int ret;
2157 
2158 	ret = -EPERM;
2159 	if (!capable(CAP_SYS_ADMIN))
2160 		goto out;
2161 
2162 	ret = -EINPROGRESS;
2163 	if (file->private_data)
2164 		goto out;
2165 
2166 	ret = -EROFS;
2167 	if (btrfs_root_readonly(root))
2168 		goto out;
2169 
2170 	ret = mnt_want_write(file->f_path.mnt);
2171 	if (ret)
2172 		goto out;
2173 
2174 	mutex_lock(&root->fs_info->trans_mutex);
2175 	root->fs_info->open_ioctl_trans++;
2176 	mutex_unlock(&root->fs_info->trans_mutex);
2177 
2178 	ret = -ENOMEM;
2179 	trans = btrfs_start_ioctl_transaction(root, 0);
2180 	if (IS_ERR(trans))
2181 		goto out_drop;
2182 
2183 	file->private_data = trans;
2184 	return 0;
2185 
2186 out_drop:
2187 	mutex_lock(&root->fs_info->trans_mutex);
2188 	root->fs_info->open_ioctl_trans--;
2189 	mutex_unlock(&root->fs_info->trans_mutex);
2190 	mnt_drop_write(file->f_path.mnt);
2191 out:
2192 	return ret;
2193 }
2194 
2195 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2196 {
2197 	struct inode *inode = fdentry(file)->d_inode;
2198 	struct btrfs_root *root = BTRFS_I(inode)->root;
2199 	struct btrfs_root *new_root;
2200 	struct btrfs_dir_item *di;
2201 	struct btrfs_trans_handle *trans;
2202 	struct btrfs_path *path;
2203 	struct btrfs_key location;
2204 	struct btrfs_disk_key disk_key;
2205 	struct btrfs_super_block *disk_super;
2206 	u64 features;
2207 	u64 objectid = 0;
2208 	u64 dir_id;
2209 
2210 	if (!capable(CAP_SYS_ADMIN))
2211 		return -EPERM;
2212 
2213 	if (copy_from_user(&objectid, argp, sizeof(objectid)))
2214 		return -EFAULT;
2215 
2216 	if (!objectid)
2217 		objectid = root->root_key.objectid;
2218 
2219 	location.objectid = objectid;
2220 	location.type = BTRFS_ROOT_ITEM_KEY;
2221 	location.offset = (u64)-1;
2222 
2223 	new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2224 	if (IS_ERR(new_root))
2225 		return PTR_ERR(new_root);
2226 
2227 	if (btrfs_root_refs(&new_root->root_item) == 0)
2228 		return -ENOENT;
2229 
2230 	path = btrfs_alloc_path();
2231 	if (!path)
2232 		return -ENOMEM;
2233 	path->leave_spinning = 1;
2234 
2235 	trans = btrfs_start_transaction(root, 1);
2236 	if (IS_ERR(trans)) {
2237 		btrfs_free_path(path);
2238 		return PTR_ERR(trans);
2239 	}
2240 
2241 	dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
2242 	di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2243 				   dir_id, "default", 7, 1);
2244 	if (IS_ERR_OR_NULL(di)) {
2245 		btrfs_free_path(path);
2246 		btrfs_end_transaction(trans, root);
2247 		printk(KERN_ERR "Umm, you don't have the default dir item, "
2248 		       "this isn't going to work\n");
2249 		return -ENOENT;
2250 	}
2251 
2252 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2253 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2254 	btrfs_mark_buffer_dirty(path->nodes[0]);
2255 	btrfs_free_path(path);
2256 
2257 	disk_super = &root->fs_info->super_copy;
2258 	features = btrfs_super_incompat_flags(disk_super);
2259 	if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2260 		features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2261 		btrfs_set_super_incompat_flags(disk_super, features);
2262 	}
2263 	btrfs_end_transaction(trans, root);
2264 
2265 	return 0;
2266 }
2267 
2268 static void get_block_group_info(struct list_head *groups_list,
2269 				 struct btrfs_ioctl_space_info *space)
2270 {
2271 	struct btrfs_block_group_cache *block_group;
2272 
2273 	space->total_bytes = 0;
2274 	space->used_bytes = 0;
2275 	space->flags = 0;
2276 	list_for_each_entry(block_group, groups_list, list) {
2277 		space->flags = block_group->flags;
2278 		space->total_bytes += block_group->key.offset;
2279 		space->used_bytes +=
2280 			btrfs_block_group_used(&block_group->item);
2281 	}
2282 }
2283 
2284 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2285 {
2286 	struct btrfs_ioctl_space_args space_args;
2287 	struct btrfs_ioctl_space_info space;
2288 	struct btrfs_ioctl_space_info *dest;
2289 	struct btrfs_ioctl_space_info *dest_orig;
2290 	struct btrfs_ioctl_space_info __user *user_dest;
2291 	struct btrfs_space_info *info;
2292 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2293 		       BTRFS_BLOCK_GROUP_SYSTEM,
2294 		       BTRFS_BLOCK_GROUP_METADATA,
2295 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2296 	int num_types = 4;
2297 	int alloc_size;
2298 	int ret = 0;
2299 	u64 slot_count = 0;
2300 	int i, c;
2301 
2302 	if (copy_from_user(&space_args,
2303 			   (struct btrfs_ioctl_space_args __user *)arg,
2304 			   sizeof(space_args)))
2305 		return -EFAULT;
2306 
2307 	for (i = 0; i < num_types; i++) {
2308 		struct btrfs_space_info *tmp;
2309 
2310 		info = NULL;
2311 		rcu_read_lock();
2312 		list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2313 					list) {
2314 			if (tmp->flags == types[i]) {
2315 				info = tmp;
2316 				break;
2317 			}
2318 		}
2319 		rcu_read_unlock();
2320 
2321 		if (!info)
2322 			continue;
2323 
2324 		down_read(&info->groups_sem);
2325 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2326 			if (!list_empty(&info->block_groups[c]))
2327 				slot_count++;
2328 		}
2329 		up_read(&info->groups_sem);
2330 	}
2331 
2332 	/* space_slots == 0 means they are asking for a count */
2333 	if (space_args.space_slots == 0) {
2334 		space_args.total_spaces = slot_count;
2335 		goto out;
2336 	}
2337 
2338 	slot_count = min_t(u64, space_args.space_slots, slot_count);
2339 
2340 	alloc_size = sizeof(*dest) * slot_count;
2341 
2342 	/* we generally have at most 6 or so space infos, one for each raid
2343 	 * level.  So, a whole page should be more than enough for everyone
2344 	 */
2345 	if (alloc_size > PAGE_CACHE_SIZE)
2346 		return -ENOMEM;
2347 
2348 	space_args.total_spaces = 0;
2349 	dest = kmalloc(alloc_size, GFP_NOFS);
2350 	if (!dest)
2351 		return -ENOMEM;
2352 	dest_orig = dest;
2353 
2354 	/* now we have a buffer to copy into */
2355 	for (i = 0; i < num_types; i++) {
2356 		struct btrfs_space_info *tmp;
2357 
2358 		if (!slot_count)
2359 			break;
2360 
2361 		info = NULL;
2362 		rcu_read_lock();
2363 		list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2364 					list) {
2365 			if (tmp->flags == types[i]) {
2366 				info = tmp;
2367 				break;
2368 			}
2369 		}
2370 		rcu_read_unlock();
2371 
2372 		if (!info)
2373 			continue;
2374 		down_read(&info->groups_sem);
2375 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2376 			if (!list_empty(&info->block_groups[c])) {
2377 				get_block_group_info(&info->block_groups[c],
2378 						     &space);
2379 				memcpy(dest, &space, sizeof(space));
2380 				dest++;
2381 				space_args.total_spaces++;
2382 				slot_count--;
2383 			}
2384 			if (!slot_count)
2385 				break;
2386 		}
2387 		up_read(&info->groups_sem);
2388 	}
2389 
2390 	user_dest = (struct btrfs_ioctl_space_info *)
2391 		(arg + sizeof(struct btrfs_ioctl_space_args));
2392 
2393 	if (copy_to_user(user_dest, dest_orig, alloc_size))
2394 		ret = -EFAULT;
2395 
2396 	kfree(dest_orig);
2397 out:
2398 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2399 		ret = -EFAULT;
2400 
2401 	return ret;
2402 }
2403 
2404 /*
2405  * there are many ways the trans_start and trans_end ioctls can lead
2406  * to deadlocks.  They should only be used by applications that
2407  * basically own the machine, and have a very in depth understanding
2408  * of all the possible deadlocks and enospc problems.
2409  */
2410 long btrfs_ioctl_trans_end(struct file *file)
2411 {
2412 	struct inode *inode = fdentry(file)->d_inode;
2413 	struct btrfs_root *root = BTRFS_I(inode)->root;
2414 	struct btrfs_trans_handle *trans;
2415 
2416 	trans = file->private_data;
2417 	if (!trans)
2418 		return -EINVAL;
2419 	file->private_data = NULL;
2420 
2421 	btrfs_end_transaction(trans, root);
2422 
2423 	mutex_lock(&root->fs_info->trans_mutex);
2424 	root->fs_info->open_ioctl_trans--;
2425 	mutex_unlock(&root->fs_info->trans_mutex);
2426 
2427 	mnt_drop_write(file->f_path.mnt);
2428 	return 0;
2429 }
2430 
2431 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2432 {
2433 	struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2434 	struct btrfs_trans_handle *trans;
2435 	u64 transid;
2436 	int ret;
2437 
2438 	trans = btrfs_start_transaction(root, 0);
2439 	if (IS_ERR(trans))
2440 		return PTR_ERR(trans);
2441 	transid = trans->transid;
2442 	ret = btrfs_commit_transaction_async(trans, root, 0);
2443 	if (ret) {
2444 		btrfs_end_transaction(trans, root);
2445 		return ret;
2446 	}
2447 
2448 	if (argp)
2449 		if (copy_to_user(argp, &transid, sizeof(transid)))
2450 			return -EFAULT;
2451 	return 0;
2452 }
2453 
2454 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2455 {
2456 	struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2457 	u64 transid;
2458 
2459 	if (argp) {
2460 		if (copy_from_user(&transid, argp, sizeof(transid)))
2461 			return -EFAULT;
2462 	} else {
2463 		transid = 0;  /* current trans */
2464 	}
2465 	return btrfs_wait_for_commit(root, transid);
2466 }
2467 
2468 long btrfs_ioctl(struct file *file, unsigned int
2469 		cmd, unsigned long arg)
2470 {
2471 	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2472 	void __user *argp = (void __user *)arg;
2473 
2474 	switch (cmd) {
2475 	case FS_IOC_GETFLAGS:
2476 		return btrfs_ioctl_getflags(file, argp);
2477 	case FS_IOC_SETFLAGS:
2478 		return btrfs_ioctl_setflags(file, argp);
2479 	case FS_IOC_GETVERSION:
2480 		return btrfs_ioctl_getversion(file, argp);
2481 	case FITRIM:
2482 		return btrfs_ioctl_fitrim(file, argp);
2483 	case BTRFS_IOC_SNAP_CREATE:
2484 		return btrfs_ioctl_snap_create(file, argp, 0);
2485 	case BTRFS_IOC_SNAP_CREATE_V2:
2486 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
2487 	case BTRFS_IOC_SUBVOL_CREATE:
2488 		return btrfs_ioctl_snap_create(file, argp, 1);
2489 	case BTRFS_IOC_SNAP_DESTROY:
2490 		return btrfs_ioctl_snap_destroy(file, argp);
2491 	case BTRFS_IOC_SUBVOL_GETFLAGS:
2492 		return btrfs_ioctl_subvol_getflags(file, argp);
2493 	case BTRFS_IOC_SUBVOL_SETFLAGS:
2494 		return btrfs_ioctl_subvol_setflags(file, argp);
2495 	case BTRFS_IOC_DEFAULT_SUBVOL:
2496 		return btrfs_ioctl_default_subvol(file, argp);
2497 	case BTRFS_IOC_DEFRAG:
2498 		return btrfs_ioctl_defrag(file, NULL);
2499 	case BTRFS_IOC_DEFRAG_RANGE:
2500 		return btrfs_ioctl_defrag(file, argp);
2501 	case BTRFS_IOC_RESIZE:
2502 		return btrfs_ioctl_resize(root, argp);
2503 	case BTRFS_IOC_ADD_DEV:
2504 		return btrfs_ioctl_add_dev(root, argp);
2505 	case BTRFS_IOC_RM_DEV:
2506 		return btrfs_ioctl_rm_dev(root, argp);
2507 	case BTRFS_IOC_BALANCE:
2508 		return btrfs_balance(root->fs_info->dev_root);
2509 	case BTRFS_IOC_CLONE:
2510 		return btrfs_ioctl_clone(file, arg, 0, 0, 0);
2511 	case BTRFS_IOC_CLONE_RANGE:
2512 		return btrfs_ioctl_clone_range(file, argp);
2513 	case BTRFS_IOC_TRANS_START:
2514 		return btrfs_ioctl_trans_start(file);
2515 	case BTRFS_IOC_TRANS_END:
2516 		return btrfs_ioctl_trans_end(file);
2517 	case BTRFS_IOC_TREE_SEARCH:
2518 		return btrfs_ioctl_tree_search(file, argp);
2519 	case BTRFS_IOC_INO_LOOKUP:
2520 		return btrfs_ioctl_ino_lookup(file, argp);
2521 	case BTRFS_IOC_SPACE_INFO:
2522 		return btrfs_ioctl_space_info(root, argp);
2523 	case BTRFS_IOC_SYNC:
2524 		btrfs_sync_fs(file->f_dentry->d_sb, 1);
2525 		return 0;
2526 	case BTRFS_IOC_START_SYNC:
2527 		return btrfs_ioctl_start_sync(file, argp);
2528 	case BTRFS_IOC_WAIT_SYNC:
2529 		return btrfs_ioctl_wait_sync(file, argp);
2530 	}
2531 
2532 	return -ENOTTY;
2533 }
2534