xref: /openbmc/linux/fs/btrfs/ioctl.c (revision 4800cd83)
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 "compat.h"
44 #include "ctree.h"
45 #include "disk-io.h"
46 #include "transaction.h"
47 #include "btrfs_inode.h"
48 #include "ioctl.h"
49 #include "print-tree.h"
50 #include "volumes.h"
51 #include "locking.h"
52 
53 /* Mask out flags that are inappropriate for the given type of inode. */
54 static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
55 {
56 	if (S_ISDIR(mode))
57 		return flags;
58 	else if (S_ISREG(mode))
59 		return flags & ~FS_DIRSYNC_FL;
60 	else
61 		return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
62 }
63 
64 /*
65  * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl.
66  */
67 static unsigned int btrfs_flags_to_ioctl(unsigned int flags)
68 {
69 	unsigned int iflags = 0;
70 
71 	if (flags & BTRFS_INODE_SYNC)
72 		iflags |= FS_SYNC_FL;
73 	if (flags & BTRFS_INODE_IMMUTABLE)
74 		iflags |= FS_IMMUTABLE_FL;
75 	if (flags & BTRFS_INODE_APPEND)
76 		iflags |= FS_APPEND_FL;
77 	if (flags & BTRFS_INODE_NODUMP)
78 		iflags |= FS_NODUMP_FL;
79 	if (flags & BTRFS_INODE_NOATIME)
80 		iflags |= FS_NOATIME_FL;
81 	if (flags & BTRFS_INODE_DIRSYNC)
82 		iflags |= FS_DIRSYNC_FL;
83 
84 	return iflags;
85 }
86 
87 /*
88  * Update inode->i_flags based on the btrfs internal flags.
89  */
90 void btrfs_update_iflags(struct inode *inode)
91 {
92 	struct btrfs_inode *ip = BTRFS_I(inode);
93 
94 	inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
95 
96 	if (ip->flags & BTRFS_INODE_SYNC)
97 		inode->i_flags |= S_SYNC;
98 	if (ip->flags & BTRFS_INODE_IMMUTABLE)
99 		inode->i_flags |= S_IMMUTABLE;
100 	if (ip->flags & BTRFS_INODE_APPEND)
101 		inode->i_flags |= S_APPEND;
102 	if (ip->flags & BTRFS_INODE_NOATIME)
103 		inode->i_flags |= S_NOATIME;
104 	if (ip->flags & BTRFS_INODE_DIRSYNC)
105 		inode->i_flags |= S_DIRSYNC;
106 }
107 
108 /*
109  * Inherit flags from the parent inode.
110  *
111  * Unlike extN we don't have any flags we don't want to inherit currently.
112  */
113 void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
114 {
115 	unsigned int flags;
116 
117 	if (!dir)
118 		return;
119 
120 	flags = BTRFS_I(dir)->flags;
121 
122 	if (S_ISREG(inode->i_mode))
123 		flags &= ~BTRFS_INODE_DIRSYNC;
124 	else if (!S_ISDIR(inode->i_mode))
125 		flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME);
126 
127 	BTRFS_I(inode)->flags = flags;
128 	btrfs_update_iflags(inode);
129 }
130 
131 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
132 {
133 	struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode);
134 	unsigned int flags = btrfs_flags_to_ioctl(ip->flags);
135 
136 	if (copy_to_user(arg, &flags, sizeof(flags)))
137 		return -EFAULT;
138 	return 0;
139 }
140 
141 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
142 {
143 	struct inode *inode = file->f_path.dentry->d_inode;
144 	struct btrfs_inode *ip = BTRFS_I(inode);
145 	struct btrfs_root *root = ip->root;
146 	struct btrfs_trans_handle *trans;
147 	unsigned int flags, oldflags;
148 	int ret;
149 
150 	if (btrfs_root_readonly(root))
151 		return -EROFS;
152 
153 	if (copy_from_user(&flags, arg, sizeof(flags)))
154 		return -EFAULT;
155 
156 	if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
157 		      FS_NOATIME_FL | FS_NODUMP_FL | \
158 		      FS_SYNC_FL | FS_DIRSYNC_FL))
159 		return -EOPNOTSUPP;
160 
161 	if (!is_owner_or_cap(inode))
162 		return -EACCES;
163 
164 	mutex_lock(&inode->i_mutex);
165 
166 	flags = btrfs_mask_flags(inode->i_mode, flags);
167 	oldflags = btrfs_flags_to_ioctl(ip->flags);
168 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
169 		if (!capable(CAP_LINUX_IMMUTABLE)) {
170 			ret = -EPERM;
171 			goto out_unlock;
172 		}
173 	}
174 
175 	ret = mnt_want_write(file->f_path.mnt);
176 	if (ret)
177 		goto out_unlock;
178 
179 	if (flags & FS_SYNC_FL)
180 		ip->flags |= BTRFS_INODE_SYNC;
181 	else
182 		ip->flags &= ~BTRFS_INODE_SYNC;
183 	if (flags & FS_IMMUTABLE_FL)
184 		ip->flags |= BTRFS_INODE_IMMUTABLE;
185 	else
186 		ip->flags &= ~BTRFS_INODE_IMMUTABLE;
187 	if (flags & FS_APPEND_FL)
188 		ip->flags |= BTRFS_INODE_APPEND;
189 	else
190 		ip->flags &= ~BTRFS_INODE_APPEND;
191 	if (flags & FS_NODUMP_FL)
192 		ip->flags |= BTRFS_INODE_NODUMP;
193 	else
194 		ip->flags &= ~BTRFS_INODE_NODUMP;
195 	if (flags & FS_NOATIME_FL)
196 		ip->flags |= BTRFS_INODE_NOATIME;
197 	else
198 		ip->flags &= ~BTRFS_INODE_NOATIME;
199 	if (flags & FS_DIRSYNC_FL)
200 		ip->flags |= BTRFS_INODE_DIRSYNC;
201 	else
202 		ip->flags &= ~BTRFS_INODE_DIRSYNC;
203 
204 
205 	trans = btrfs_join_transaction(root, 1);
206 	BUG_ON(IS_ERR(trans));
207 
208 	ret = btrfs_update_inode(trans, root, inode);
209 	BUG_ON(ret);
210 
211 	btrfs_update_iflags(inode);
212 	inode->i_ctime = CURRENT_TIME;
213 	btrfs_end_transaction(trans, root);
214 
215 	mnt_drop_write(file->f_path.mnt);
216  out_unlock:
217 	mutex_unlock(&inode->i_mutex);
218 	return 0;
219 }
220 
221 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
222 {
223 	struct inode *inode = file->f_path.dentry->d_inode;
224 
225 	return put_user(inode->i_generation, arg);
226 }
227 
228 static noinline int create_subvol(struct btrfs_root *root,
229 				  struct dentry *dentry,
230 				  char *name, int namelen,
231 				  u64 *async_transid)
232 {
233 	struct btrfs_trans_handle *trans;
234 	struct btrfs_key key;
235 	struct btrfs_root_item root_item;
236 	struct btrfs_inode_item *inode_item;
237 	struct extent_buffer *leaf;
238 	struct btrfs_root *new_root;
239 	struct dentry *parent = dget_parent(dentry);
240 	struct inode *dir;
241 	int ret;
242 	int err;
243 	u64 objectid;
244 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
245 	u64 index = 0;
246 
247 	ret = btrfs_find_free_objectid(NULL, root->fs_info->tree_root,
248 				       0, &objectid);
249 	if (ret) {
250 		dput(parent);
251 		return ret;
252 	}
253 
254 	dir = parent->d_inode;
255 
256 	/*
257 	 * 1 - inode item
258 	 * 2 - refs
259 	 * 1 - root item
260 	 * 2 - dir items
261 	 */
262 	trans = btrfs_start_transaction(root, 6);
263 	if (IS_ERR(trans)) {
264 		dput(parent);
265 		return PTR_ERR(trans);
266 	}
267 
268 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
269 				      0, objectid, NULL, 0, 0, 0);
270 	if (IS_ERR(leaf)) {
271 		ret = PTR_ERR(leaf);
272 		goto fail;
273 	}
274 
275 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
276 	btrfs_set_header_bytenr(leaf, leaf->start);
277 	btrfs_set_header_generation(leaf, trans->transid);
278 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
279 	btrfs_set_header_owner(leaf, objectid);
280 
281 	write_extent_buffer(leaf, root->fs_info->fsid,
282 			    (unsigned long)btrfs_header_fsid(leaf),
283 			    BTRFS_FSID_SIZE);
284 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
285 			    (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
286 			    BTRFS_UUID_SIZE);
287 	btrfs_mark_buffer_dirty(leaf);
288 
289 	inode_item = &root_item.inode;
290 	memset(inode_item, 0, sizeof(*inode_item));
291 	inode_item->generation = cpu_to_le64(1);
292 	inode_item->size = cpu_to_le64(3);
293 	inode_item->nlink = cpu_to_le32(1);
294 	inode_item->nbytes = cpu_to_le64(root->leafsize);
295 	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
296 
297 	btrfs_set_root_bytenr(&root_item, leaf->start);
298 	btrfs_set_root_generation(&root_item, trans->transid);
299 	btrfs_set_root_level(&root_item, 0);
300 	btrfs_set_root_refs(&root_item, 1);
301 	btrfs_set_root_used(&root_item, leaf->len);
302 	btrfs_set_root_last_snapshot(&root_item, 0);
303 
304 	memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
305 	root_item.drop_level = 0;
306 
307 	btrfs_tree_unlock(leaf);
308 	free_extent_buffer(leaf);
309 	leaf = NULL;
310 
311 	btrfs_set_root_dirid(&root_item, new_dirid);
312 
313 	key.objectid = objectid;
314 	key.offset = 0;
315 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
316 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
317 				&root_item);
318 	if (ret)
319 		goto fail;
320 
321 	key.offset = (u64)-1;
322 	new_root = btrfs_read_fs_root_no_name(root->fs_info, &key);
323 	BUG_ON(IS_ERR(new_root));
324 
325 	btrfs_record_root_in_trans(trans, new_root);
326 
327 	ret = btrfs_create_subvol_root(trans, new_root, new_dirid,
328 				       BTRFS_I(dir)->block_group);
329 	/*
330 	 * insert the directory item
331 	 */
332 	ret = btrfs_set_inode_index(dir, &index);
333 	BUG_ON(ret);
334 
335 	ret = btrfs_insert_dir_item(trans, root,
336 				    name, namelen, dir->i_ino, &key,
337 				    BTRFS_FT_DIR, index);
338 	if (ret)
339 		goto fail;
340 
341 	btrfs_i_size_write(dir, dir->i_size + namelen * 2);
342 	ret = btrfs_update_inode(trans, root, dir);
343 	BUG_ON(ret);
344 
345 	ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
346 				 objectid, root->root_key.objectid,
347 				 dir->i_ino, index, name, namelen);
348 
349 	BUG_ON(ret);
350 
351 	d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry));
352 fail:
353 	dput(parent);
354 	if (async_transid) {
355 		*async_transid = trans->transid;
356 		err = btrfs_commit_transaction_async(trans, root, 1);
357 	} else {
358 		err = btrfs_commit_transaction(trans, root);
359 	}
360 	if (err && !ret)
361 		ret = err;
362 	return ret;
363 }
364 
365 static int create_snapshot(struct btrfs_root *root, struct dentry *dentry,
366 			   char *name, int namelen, u64 *async_transid,
367 			   bool readonly)
368 {
369 	struct inode *inode;
370 	struct dentry *parent;
371 	struct btrfs_pending_snapshot *pending_snapshot;
372 	struct btrfs_trans_handle *trans;
373 	int ret;
374 
375 	if (!root->ref_cows)
376 		return -EINVAL;
377 
378 	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
379 	if (!pending_snapshot)
380 		return -ENOMEM;
381 
382 	btrfs_init_block_rsv(&pending_snapshot->block_rsv);
383 	pending_snapshot->dentry = dentry;
384 	pending_snapshot->root = root;
385 	pending_snapshot->readonly = readonly;
386 
387 	trans = btrfs_start_transaction(root->fs_info->extent_root, 5);
388 	if (IS_ERR(trans)) {
389 		ret = PTR_ERR(trans);
390 		goto fail;
391 	}
392 
393 	ret = btrfs_snap_reserve_metadata(trans, pending_snapshot);
394 	BUG_ON(ret);
395 
396 	list_add(&pending_snapshot->list,
397 		 &trans->transaction->pending_snapshots);
398 	if (async_transid) {
399 		*async_transid = trans->transid;
400 		ret = btrfs_commit_transaction_async(trans,
401 				     root->fs_info->extent_root, 1);
402 	} else {
403 		ret = btrfs_commit_transaction(trans,
404 					       root->fs_info->extent_root);
405 	}
406 	BUG_ON(ret);
407 
408 	ret = pending_snapshot->error;
409 	if (ret)
410 		goto fail;
411 
412 	btrfs_orphan_cleanup(pending_snapshot->snap);
413 
414 	parent = dget_parent(dentry);
415 	inode = btrfs_lookup_dentry(parent->d_inode, dentry);
416 	dput(parent);
417 	if (IS_ERR(inode)) {
418 		ret = PTR_ERR(inode);
419 		goto fail;
420 	}
421 	BUG_ON(!inode);
422 	d_instantiate(dentry, inode);
423 	ret = 0;
424 fail:
425 	kfree(pending_snapshot);
426 	return ret;
427 }
428 
429 /*  copy of check_sticky in fs/namei.c()
430 * It's inline, so penalty for filesystems that don't use sticky bit is
431 * minimal.
432 */
433 static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode)
434 {
435 	uid_t fsuid = current_fsuid();
436 
437 	if (!(dir->i_mode & S_ISVTX))
438 		return 0;
439 	if (inode->i_uid == fsuid)
440 		return 0;
441 	if (dir->i_uid == fsuid)
442 		return 0;
443 	return !capable(CAP_FOWNER);
444 }
445 
446 /*  copy of may_delete in fs/namei.c()
447  *	Check whether we can remove a link victim from directory dir, check
448  *  whether the type of victim is right.
449  *  1. We can't do it if dir is read-only (done in permission())
450  *  2. We should have write and exec permissions on dir
451  *  3. We can't remove anything from append-only dir
452  *  4. We can't do anything with immutable dir (done in permission())
453  *  5. If the sticky bit on dir is set we should either
454  *	a. be owner of dir, or
455  *	b. be owner of victim, or
456  *	c. have CAP_FOWNER capability
457  *  6. If the victim is append-only or immutable we can't do antyhing with
458  *     links pointing to it.
459  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
460  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
461  *  9. We can't remove a root or mountpoint.
462  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
463  *     nfs_async_unlink().
464  */
465 
466 static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir)
467 {
468 	int error;
469 
470 	if (!victim->d_inode)
471 		return -ENOENT;
472 
473 	BUG_ON(victim->d_parent->d_inode != dir);
474 	audit_inode_child(victim, dir);
475 
476 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
477 	if (error)
478 		return error;
479 	if (IS_APPEND(dir))
480 		return -EPERM;
481 	if (btrfs_check_sticky(dir, victim->d_inode)||
482 		IS_APPEND(victim->d_inode)||
483 	    IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
484 		return -EPERM;
485 	if (isdir) {
486 		if (!S_ISDIR(victim->d_inode->i_mode))
487 			return -ENOTDIR;
488 		if (IS_ROOT(victim))
489 			return -EBUSY;
490 	} else if (S_ISDIR(victim->d_inode->i_mode))
491 		return -EISDIR;
492 	if (IS_DEADDIR(dir))
493 		return -ENOENT;
494 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
495 		return -EBUSY;
496 	return 0;
497 }
498 
499 /* copy of may_create in fs/namei.c() */
500 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
501 {
502 	if (child->d_inode)
503 		return -EEXIST;
504 	if (IS_DEADDIR(dir))
505 		return -ENOENT;
506 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
507 }
508 
509 /*
510  * Create a new subvolume below @parent.  This is largely modeled after
511  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
512  * inside this filesystem so it's quite a bit simpler.
513  */
514 static noinline int btrfs_mksubvol(struct path *parent,
515 				   char *name, int namelen,
516 				   struct btrfs_root *snap_src,
517 				   u64 *async_transid, bool readonly)
518 {
519 	struct inode *dir  = parent->dentry->d_inode;
520 	struct dentry *dentry;
521 	int error;
522 
523 	mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
524 
525 	dentry = lookup_one_len(name, parent->dentry, namelen);
526 	error = PTR_ERR(dentry);
527 	if (IS_ERR(dentry))
528 		goto out_unlock;
529 
530 	error = -EEXIST;
531 	if (dentry->d_inode)
532 		goto out_dput;
533 
534 	error = mnt_want_write(parent->mnt);
535 	if (error)
536 		goto out_dput;
537 
538 	error = btrfs_may_create(dir, dentry);
539 	if (error)
540 		goto out_drop_write;
541 
542 	down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
543 
544 	if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
545 		goto out_up_read;
546 
547 	if (snap_src) {
548 		error = create_snapshot(snap_src, dentry,
549 					name, namelen, async_transid, readonly);
550 	} else {
551 		error = create_subvol(BTRFS_I(dir)->root, dentry,
552 				      name, namelen, async_transid);
553 	}
554 	if (!error)
555 		fsnotify_mkdir(dir, dentry);
556 out_up_read:
557 	up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem);
558 out_drop_write:
559 	mnt_drop_write(parent->mnt);
560 out_dput:
561 	dput(dentry);
562 out_unlock:
563 	mutex_unlock(&dir->i_mutex);
564 	return error;
565 }
566 
567 static int should_defrag_range(struct inode *inode, u64 start, u64 len,
568 			       int thresh, u64 *last_len, u64 *skip,
569 			       u64 *defrag_end)
570 {
571 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
572 	struct extent_map *em = NULL;
573 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
574 	int ret = 1;
575 
576 
577 	if (thresh == 0)
578 		thresh = 256 * 1024;
579 
580 	/*
581 	 * make sure that once we start defragging and extent, we keep on
582 	 * defragging it
583 	 */
584 	if (start < *defrag_end)
585 		return 1;
586 
587 	*skip = 0;
588 
589 	/*
590 	 * hopefully we have this extent in the tree already, try without
591 	 * the full extent lock
592 	 */
593 	read_lock(&em_tree->lock);
594 	em = lookup_extent_mapping(em_tree, start, len);
595 	read_unlock(&em_tree->lock);
596 
597 	if (!em) {
598 		/* get the big lock and read metadata off disk */
599 		lock_extent(io_tree, start, start + len - 1, GFP_NOFS);
600 		em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
601 		unlock_extent(io_tree, start, start + len - 1, GFP_NOFS);
602 
603 		if (IS_ERR(em))
604 			return 0;
605 	}
606 
607 	/* this will cover holes, and inline extents */
608 	if (em->block_start >= EXTENT_MAP_LAST_BYTE)
609 		ret = 0;
610 
611 	/*
612 	 * we hit a real extent, if it is big don't bother defragging it again
613 	 */
614 	if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh)
615 		ret = 0;
616 
617 	/*
618 	 * last_len ends up being a counter of how many bytes we've defragged.
619 	 * every time we choose not to defrag an extent, we reset *last_len
620 	 * so that the next tiny extent will force a defrag.
621 	 *
622 	 * The end result of this is that tiny extents before a single big
623 	 * extent will force at least part of that big extent to be defragged.
624 	 */
625 	if (ret) {
626 		*last_len += len;
627 		*defrag_end = extent_map_end(em);
628 	} else {
629 		*last_len = 0;
630 		*skip = extent_map_end(em);
631 		*defrag_end = 0;
632 	}
633 
634 	free_extent_map(em);
635 	return ret;
636 }
637 
638 static int btrfs_defrag_file(struct file *file,
639 			     struct btrfs_ioctl_defrag_range_args *range)
640 {
641 	struct inode *inode = fdentry(file)->d_inode;
642 	struct btrfs_root *root = BTRFS_I(inode)->root;
643 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
644 	struct btrfs_ordered_extent *ordered;
645 	struct page *page;
646 	struct btrfs_super_block *disk_super;
647 	unsigned long last_index;
648 	unsigned long ra_pages = root->fs_info->bdi.ra_pages;
649 	unsigned long total_read = 0;
650 	u64 features;
651 	u64 page_start;
652 	u64 page_end;
653 	u64 last_len = 0;
654 	u64 skip = 0;
655 	u64 defrag_end = 0;
656 	unsigned long i;
657 	int ret;
658 	int compress_type = BTRFS_COMPRESS_ZLIB;
659 
660 	if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) {
661 		if (range->compress_type > BTRFS_COMPRESS_TYPES)
662 			return -EINVAL;
663 		if (range->compress_type)
664 			compress_type = range->compress_type;
665 	}
666 
667 	if (inode->i_size == 0)
668 		return 0;
669 
670 	if (range->start + range->len > range->start) {
671 		last_index = min_t(u64, inode->i_size - 1,
672 			 range->start + range->len - 1) >> PAGE_CACHE_SHIFT;
673 	} else {
674 		last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
675 	}
676 
677 	i = range->start >> PAGE_CACHE_SHIFT;
678 	while (i <= last_index) {
679 		if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT,
680 					PAGE_CACHE_SIZE,
681 					range->extent_thresh,
682 					&last_len, &skip,
683 					&defrag_end)) {
684 			unsigned long next;
685 			/*
686 			 * the should_defrag function tells us how much to skip
687 			 * bump our counter by the suggested amount
688 			 */
689 			next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
690 			i = max(i + 1, next);
691 			continue;
692 		}
693 
694 		if (total_read % ra_pages == 0) {
695 			btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
696 				       min(last_index, i + ra_pages - 1));
697 		}
698 		total_read++;
699 		mutex_lock(&inode->i_mutex);
700 		if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)
701 			BTRFS_I(inode)->force_compress = compress_type;
702 
703 		ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
704 		if (ret)
705 			goto err_unlock;
706 again:
707 		if (inode->i_size == 0 ||
708 		    i > ((inode->i_size - 1) >> PAGE_CACHE_SHIFT)) {
709 			ret = 0;
710 			goto err_reservations;
711 		}
712 
713 		page = grab_cache_page(inode->i_mapping, i);
714 		if (!page) {
715 			ret = -ENOMEM;
716 			goto err_reservations;
717 		}
718 
719 		if (!PageUptodate(page)) {
720 			btrfs_readpage(NULL, page);
721 			lock_page(page);
722 			if (!PageUptodate(page)) {
723 				unlock_page(page);
724 				page_cache_release(page);
725 				ret = -EIO;
726 				goto err_reservations;
727 			}
728 		}
729 
730 		if (page->mapping != inode->i_mapping) {
731 			unlock_page(page);
732 			page_cache_release(page);
733 			goto again;
734 		}
735 
736 		wait_on_page_writeback(page);
737 
738 		if (PageDirty(page)) {
739 			btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
740 			goto loop_unlock;
741 		}
742 
743 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
744 		page_end = page_start + PAGE_CACHE_SIZE - 1;
745 		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
746 
747 		ordered = btrfs_lookup_ordered_extent(inode, page_start);
748 		if (ordered) {
749 			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
750 			unlock_page(page);
751 			page_cache_release(page);
752 			btrfs_start_ordered_extent(inode, ordered, 1);
753 			btrfs_put_ordered_extent(ordered);
754 			goto again;
755 		}
756 		set_page_extent_mapped(page);
757 
758 		/*
759 		 * this makes sure page_mkwrite is called on the
760 		 * page if it is dirtied again later
761 		 */
762 		clear_page_dirty_for_io(page);
763 		clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start,
764 				  page_end, EXTENT_DIRTY | EXTENT_DELALLOC |
765 				  EXTENT_DO_ACCOUNTING, GFP_NOFS);
766 
767 		btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
768 		ClearPageChecked(page);
769 		set_page_dirty(page);
770 		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
771 
772 loop_unlock:
773 		unlock_page(page);
774 		page_cache_release(page);
775 		mutex_unlock(&inode->i_mutex);
776 
777 		balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
778 		i++;
779 	}
780 
781 	if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO))
782 		filemap_flush(inode->i_mapping);
783 
784 	if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
785 		/* the filemap_flush will queue IO into the worker threads, but
786 		 * we have to make sure the IO is actually started and that
787 		 * ordered extents get created before we return
788 		 */
789 		atomic_inc(&root->fs_info->async_submit_draining);
790 		while (atomic_read(&root->fs_info->nr_async_submits) ||
791 		      atomic_read(&root->fs_info->async_delalloc_pages)) {
792 			wait_event(root->fs_info->async_submit_wait,
793 			   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
794 			    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
795 		}
796 		atomic_dec(&root->fs_info->async_submit_draining);
797 
798 		mutex_lock(&inode->i_mutex);
799 		BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE;
800 		mutex_unlock(&inode->i_mutex);
801 	}
802 
803 	disk_super = &root->fs_info->super_copy;
804 	features = btrfs_super_incompat_flags(disk_super);
805 	if (range->compress_type == BTRFS_COMPRESS_LZO) {
806 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
807 		btrfs_set_super_incompat_flags(disk_super, features);
808 	}
809 
810 	return 0;
811 
812 err_reservations:
813 	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
814 err_unlock:
815 	mutex_unlock(&inode->i_mutex);
816 	return ret;
817 }
818 
819 static noinline int btrfs_ioctl_resize(struct btrfs_root *root,
820 					void __user *arg)
821 {
822 	u64 new_size;
823 	u64 old_size;
824 	u64 devid = 1;
825 	struct btrfs_ioctl_vol_args *vol_args;
826 	struct btrfs_trans_handle *trans;
827 	struct btrfs_device *device = NULL;
828 	char *sizestr;
829 	char *devstr = NULL;
830 	int ret = 0;
831 	int mod = 0;
832 
833 	if (root->fs_info->sb->s_flags & MS_RDONLY)
834 		return -EROFS;
835 
836 	if (!capable(CAP_SYS_ADMIN))
837 		return -EPERM;
838 
839 	vol_args = memdup_user(arg, sizeof(*vol_args));
840 	if (IS_ERR(vol_args))
841 		return PTR_ERR(vol_args);
842 
843 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
844 
845 	mutex_lock(&root->fs_info->volume_mutex);
846 	sizestr = vol_args->name;
847 	devstr = strchr(sizestr, ':');
848 	if (devstr) {
849 		char *end;
850 		sizestr = devstr + 1;
851 		*devstr = '\0';
852 		devstr = vol_args->name;
853 		devid = simple_strtoull(devstr, &end, 10);
854 		printk(KERN_INFO "resizing devid %llu\n",
855 		       (unsigned long long)devid);
856 	}
857 	device = btrfs_find_device(root, devid, NULL, NULL);
858 	if (!device) {
859 		printk(KERN_INFO "resizer unable to find device %llu\n",
860 		       (unsigned long long)devid);
861 		ret = -EINVAL;
862 		goto out_unlock;
863 	}
864 	if (!strcmp(sizestr, "max"))
865 		new_size = device->bdev->bd_inode->i_size;
866 	else {
867 		if (sizestr[0] == '-') {
868 			mod = -1;
869 			sizestr++;
870 		} else if (sizestr[0] == '+') {
871 			mod = 1;
872 			sizestr++;
873 		}
874 		new_size = memparse(sizestr, NULL);
875 		if (new_size == 0) {
876 			ret = -EINVAL;
877 			goto out_unlock;
878 		}
879 	}
880 
881 	old_size = device->total_bytes;
882 
883 	if (mod < 0) {
884 		if (new_size > old_size) {
885 			ret = -EINVAL;
886 			goto out_unlock;
887 		}
888 		new_size = old_size - new_size;
889 	} else if (mod > 0) {
890 		new_size = old_size + new_size;
891 	}
892 
893 	if (new_size < 256 * 1024 * 1024) {
894 		ret = -EINVAL;
895 		goto out_unlock;
896 	}
897 	if (new_size > device->bdev->bd_inode->i_size) {
898 		ret = -EFBIG;
899 		goto out_unlock;
900 	}
901 
902 	do_div(new_size, root->sectorsize);
903 	new_size *= root->sectorsize;
904 
905 	printk(KERN_INFO "new size for %s is %llu\n",
906 		device->name, (unsigned long long)new_size);
907 
908 	if (new_size > old_size) {
909 		trans = btrfs_start_transaction(root, 0);
910 		if (IS_ERR(trans)) {
911 			ret = PTR_ERR(trans);
912 			goto out_unlock;
913 		}
914 		ret = btrfs_grow_device(trans, device, new_size);
915 		btrfs_commit_transaction(trans, root);
916 	} else {
917 		ret = btrfs_shrink_device(device, new_size);
918 	}
919 
920 out_unlock:
921 	mutex_unlock(&root->fs_info->volume_mutex);
922 	kfree(vol_args);
923 	return ret;
924 }
925 
926 static noinline int btrfs_ioctl_snap_create_transid(struct file *file,
927 						    char *name,
928 						    unsigned long fd,
929 						    int subvol,
930 						    u64 *transid,
931 						    bool readonly)
932 {
933 	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
934 	struct file *src_file;
935 	int namelen;
936 	int ret = 0;
937 
938 	if (root->fs_info->sb->s_flags & MS_RDONLY)
939 		return -EROFS;
940 
941 	namelen = strlen(name);
942 	if (strchr(name, '/')) {
943 		ret = -EINVAL;
944 		goto out;
945 	}
946 
947 	if (subvol) {
948 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
949 				     NULL, transid, readonly);
950 	} else {
951 		struct inode *src_inode;
952 		src_file = fget(fd);
953 		if (!src_file) {
954 			ret = -EINVAL;
955 			goto out;
956 		}
957 
958 		src_inode = src_file->f_path.dentry->d_inode;
959 		if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) {
960 			printk(KERN_INFO "btrfs: Snapshot src from "
961 			       "another FS\n");
962 			ret = -EINVAL;
963 			fput(src_file);
964 			goto out;
965 		}
966 		ret = btrfs_mksubvol(&file->f_path, name, namelen,
967 				     BTRFS_I(src_inode)->root,
968 				     transid, readonly);
969 		fput(src_file);
970 	}
971 out:
972 	return ret;
973 }
974 
975 static noinline int btrfs_ioctl_snap_create(struct file *file,
976 					    void __user *arg, int subvol)
977 {
978 	struct btrfs_ioctl_vol_args *vol_args;
979 	int ret;
980 
981 	vol_args = memdup_user(arg, sizeof(*vol_args));
982 	if (IS_ERR(vol_args))
983 		return PTR_ERR(vol_args);
984 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
985 
986 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
987 					      vol_args->fd, subvol,
988 					      NULL, false);
989 
990 	kfree(vol_args);
991 	return ret;
992 }
993 
994 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
995 					       void __user *arg, int subvol)
996 {
997 	struct btrfs_ioctl_vol_args_v2 *vol_args;
998 	int ret;
999 	u64 transid = 0;
1000 	u64 *ptr = NULL;
1001 	bool readonly = false;
1002 
1003 	vol_args = memdup_user(arg, sizeof(*vol_args));
1004 	if (IS_ERR(vol_args))
1005 		return PTR_ERR(vol_args);
1006 	vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1007 
1008 	if (vol_args->flags &
1009 	    ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) {
1010 		ret = -EOPNOTSUPP;
1011 		goto out;
1012 	}
1013 
1014 	if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC)
1015 		ptr = &transid;
1016 	if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1017 		readonly = true;
1018 
1019 	ret = btrfs_ioctl_snap_create_transid(file, vol_args->name,
1020 					      vol_args->fd, subvol,
1021 					      ptr, readonly);
1022 
1023 	if (ret == 0 && ptr &&
1024 	    copy_to_user(arg +
1025 			 offsetof(struct btrfs_ioctl_vol_args_v2,
1026 				  transid), ptr, sizeof(*ptr)))
1027 		ret = -EFAULT;
1028 out:
1029 	kfree(vol_args);
1030 	return ret;
1031 }
1032 
1033 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1034 						void __user *arg)
1035 {
1036 	struct inode *inode = fdentry(file)->d_inode;
1037 	struct btrfs_root *root = BTRFS_I(inode)->root;
1038 	int ret = 0;
1039 	u64 flags = 0;
1040 
1041 	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID)
1042 		return -EINVAL;
1043 
1044 	down_read(&root->fs_info->subvol_sem);
1045 	if (btrfs_root_readonly(root))
1046 		flags |= BTRFS_SUBVOL_RDONLY;
1047 	up_read(&root->fs_info->subvol_sem);
1048 
1049 	if (copy_to_user(arg, &flags, sizeof(flags)))
1050 		ret = -EFAULT;
1051 
1052 	return ret;
1053 }
1054 
1055 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1056 					      void __user *arg)
1057 {
1058 	struct inode *inode = fdentry(file)->d_inode;
1059 	struct btrfs_root *root = BTRFS_I(inode)->root;
1060 	struct btrfs_trans_handle *trans;
1061 	u64 root_flags;
1062 	u64 flags;
1063 	int ret = 0;
1064 
1065 	if (root->fs_info->sb->s_flags & MS_RDONLY)
1066 		return -EROFS;
1067 
1068 	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID)
1069 		return -EINVAL;
1070 
1071 	if (copy_from_user(&flags, arg, sizeof(flags)))
1072 		return -EFAULT;
1073 
1074 	if (flags & BTRFS_SUBVOL_CREATE_ASYNC)
1075 		return -EINVAL;
1076 
1077 	if (flags & ~BTRFS_SUBVOL_RDONLY)
1078 		return -EOPNOTSUPP;
1079 
1080 	if (!is_owner_or_cap(inode))
1081 		return -EACCES;
1082 
1083 	down_write(&root->fs_info->subvol_sem);
1084 
1085 	/* nothing to do */
1086 	if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1087 		goto out;
1088 
1089 	root_flags = btrfs_root_flags(&root->root_item);
1090 	if (flags & BTRFS_SUBVOL_RDONLY)
1091 		btrfs_set_root_flags(&root->root_item,
1092 				     root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1093 	else
1094 		btrfs_set_root_flags(&root->root_item,
1095 				     root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1096 
1097 	trans = btrfs_start_transaction(root, 1);
1098 	if (IS_ERR(trans)) {
1099 		ret = PTR_ERR(trans);
1100 		goto out_reset;
1101 	}
1102 
1103 	ret = btrfs_update_root(trans, root->fs_info->tree_root,
1104 				&root->root_key, &root->root_item);
1105 
1106 	btrfs_commit_transaction(trans, root);
1107 out_reset:
1108 	if (ret)
1109 		btrfs_set_root_flags(&root->root_item, root_flags);
1110 out:
1111 	up_write(&root->fs_info->subvol_sem);
1112 	return ret;
1113 }
1114 
1115 /*
1116  * helper to check if the subvolume references other subvolumes
1117  */
1118 static noinline int may_destroy_subvol(struct btrfs_root *root)
1119 {
1120 	struct btrfs_path *path;
1121 	struct btrfs_key key;
1122 	int ret;
1123 
1124 	path = btrfs_alloc_path();
1125 	if (!path)
1126 		return -ENOMEM;
1127 
1128 	key.objectid = root->root_key.objectid;
1129 	key.type = BTRFS_ROOT_REF_KEY;
1130 	key.offset = (u64)-1;
1131 
1132 	ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
1133 				&key, path, 0, 0);
1134 	if (ret < 0)
1135 		goto out;
1136 	BUG_ON(ret == 0);
1137 
1138 	ret = 0;
1139 	if (path->slots[0] > 0) {
1140 		path->slots[0]--;
1141 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1142 		if (key.objectid == root->root_key.objectid &&
1143 		    key.type == BTRFS_ROOT_REF_KEY)
1144 			ret = -ENOTEMPTY;
1145 	}
1146 out:
1147 	btrfs_free_path(path);
1148 	return ret;
1149 }
1150 
1151 static noinline int key_in_sk(struct btrfs_key *key,
1152 			      struct btrfs_ioctl_search_key *sk)
1153 {
1154 	struct btrfs_key test;
1155 	int ret;
1156 
1157 	test.objectid = sk->min_objectid;
1158 	test.type = sk->min_type;
1159 	test.offset = sk->min_offset;
1160 
1161 	ret = btrfs_comp_cpu_keys(key, &test);
1162 	if (ret < 0)
1163 		return 0;
1164 
1165 	test.objectid = sk->max_objectid;
1166 	test.type = sk->max_type;
1167 	test.offset = sk->max_offset;
1168 
1169 	ret = btrfs_comp_cpu_keys(key, &test);
1170 	if (ret > 0)
1171 		return 0;
1172 	return 1;
1173 }
1174 
1175 static noinline int copy_to_sk(struct btrfs_root *root,
1176 			       struct btrfs_path *path,
1177 			       struct btrfs_key *key,
1178 			       struct btrfs_ioctl_search_key *sk,
1179 			       char *buf,
1180 			       unsigned long *sk_offset,
1181 			       int *num_found)
1182 {
1183 	u64 found_transid;
1184 	struct extent_buffer *leaf;
1185 	struct btrfs_ioctl_search_header sh;
1186 	unsigned long item_off;
1187 	unsigned long item_len;
1188 	int nritems;
1189 	int i;
1190 	int slot;
1191 	int found = 0;
1192 	int ret = 0;
1193 
1194 	leaf = path->nodes[0];
1195 	slot = path->slots[0];
1196 	nritems = btrfs_header_nritems(leaf);
1197 
1198 	if (btrfs_header_generation(leaf) > sk->max_transid) {
1199 		i = nritems;
1200 		goto advance_key;
1201 	}
1202 	found_transid = btrfs_header_generation(leaf);
1203 
1204 	for (i = slot; i < nritems; i++) {
1205 		item_off = btrfs_item_ptr_offset(leaf, i);
1206 		item_len = btrfs_item_size_nr(leaf, i);
1207 
1208 		if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE)
1209 			item_len = 0;
1210 
1211 		if (sizeof(sh) + item_len + *sk_offset >
1212 		    BTRFS_SEARCH_ARGS_BUFSIZE) {
1213 			ret = 1;
1214 			goto overflow;
1215 		}
1216 
1217 		btrfs_item_key_to_cpu(leaf, key, i);
1218 		if (!key_in_sk(key, sk))
1219 			continue;
1220 
1221 		sh.objectid = key->objectid;
1222 		sh.offset = key->offset;
1223 		sh.type = key->type;
1224 		sh.len = item_len;
1225 		sh.transid = found_transid;
1226 
1227 		/* copy search result header */
1228 		memcpy(buf + *sk_offset, &sh, sizeof(sh));
1229 		*sk_offset += sizeof(sh);
1230 
1231 		if (item_len) {
1232 			char *p = buf + *sk_offset;
1233 			/* copy the item */
1234 			read_extent_buffer(leaf, p,
1235 					   item_off, item_len);
1236 			*sk_offset += item_len;
1237 		}
1238 		found++;
1239 
1240 		if (*num_found >= sk->nr_items)
1241 			break;
1242 	}
1243 advance_key:
1244 	ret = 0;
1245 	if (key->offset < (u64)-1 && key->offset < sk->max_offset)
1246 		key->offset++;
1247 	else if (key->type < (u8)-1 && key->type < sk->max_type) {
1248 		key->offset = 0;
1249 		key->type++;
1250 	} else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) {
1251 		key->offset = 0;
1252 		key->type = 0;
1253 		key->objectid++;
1254 	} else
1255 		ret = 1;
1256 overflow:
1257 	*num_found += found;
1258 	return ret;
1259 }
1260 
1261 static noinline int search_ioctl(struct inode *inode,
1262 				 struct btrfs_ioctl_search_args *args)
1263 {
1264 	struct btrfs_root *root;
1265 	struct btrfs_key key;
1266 	struct btrfs_key max_key;
1267 	struct btrfs_path *path;
1268 	struct btrfs_ioctl_search_key *sk = &args->key;
1269 	struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info;
1270 	int ret;
1271 	int num_found = 0;
1272 	unsigned long sk_offset = 0;
1273 
1274 	path = btrfs_alloc_path();
1275 	if (!path)
1276 		return -ENOMEM;
1277 
1278 	if (sk->tree_id == 0) {
1279 		/* search the root of the inode that was passed */
1280 		root = BTRFS_I(inode)->root;
1281 	} else {
1282 		key.objectid = sk->tree_id;
1283 		key.type = BTRFS_ROOT_ITEM_KEY;
1284 		key.offset = (u64)-1;
1285 		root = btrfs_read_fs_root_no_name(info, &key);
1286 		if (IS_ERR(root)) {
1287 			printk(KERN_ERR "could not find root %llu\n",
1288 			       sk->tree_id);
1289 			btrfs_free_path(path);
1290 			return -ENOENT;
1291 		}
1292 	}
1293 
1294 	key.objectid = sk->min_objectid;
1295 	key.type = sk->min_type;
1296 	key.offset = sk->min_offset;
1297 
1298 	max_key.objectid = sk->max_objectid;
1299 	max_key.type = sk->max_type;
1300 	max_key.offset = sk->max_offset;
1301 
1302 	path->keep_locks = 1;
1303 
1304 	while(1) {
1305 		ret = btrfs_search_forward(root, &key, &max_key, path, 0,
1306 					   sk->min_transid);
1307 		if (ret != 0) {
1308 			if (ret > 0)
1309 				ret = 0;
1310 			goto err;
1311 		}
1312 		ret = copy_to_sk(root, path, &key, sk, args->buf,
1313 				 &sk_offset, &num_found);
1314 		btrfs_release_path(root, path);
1315 		if (ret || num_found >= sk->nr_items)
1316 			break;
1317 
1318 	}
1319 	ret = 0;
1320 err:
1321 	sk->nr_items = num_found;
1322 	btrfs_free_path(path);
1323 	return ret;
1324 }
1325 
1326 static noinline int btrfs_ioctl_tree_search(struct file *file,
1327 					   void __user *argp)
1328 {
1329 	 struct btrfs_ioctl_search_args *args;
1330 	 struct inode *inode;
1331 	 int ret;
1332 
1333 	if (!capable(CAP_SYS_ADMIN))
1334 		return -EPERM;
1335 
1336 	args = memdup_user(argp, sizeof(*args));
1337 	if (IS_ERR(args))
1338 		return PTR_ERR(args);
1339 
1340 	inode = fdentry(file)->d_inode;
1341 	ret = search_ioctl(inode, args);
1342 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1343 		ret = -EFAULT;
1344 	kfree(args);
1345 	return ret;
1346 }
1347 
1348 /*
1349  * Search INODE_REFs to identify path name of 'dirid' directory
1350  * in a 'tree_id' tree. and sets path name to 'name'.
1351  */
1352 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
1353 				u64 tree_id, u64 dirid, char *name)
1354 {
1355 	struct btrfs_root *root;
1356 	struct btrfs_key key;
1357 	char *ptr;
1358 	int ret = -1;
1359 	int slot;
1360 	int len;
1361 	int total_len = 0;
1362 	struct btrfs_inode_ref *iref;
1363 	struct extent_buffer *l;
1364 	struct btrfs_path *path;
1365 
1366 	if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
1367 		name[0]='\0';
1368 		return 0;
1369 	}
1370 
1371 	path = btrfs_alloc_path();
1372 	if (!path)
1373 		return -ENOMEM;
1374 
1375 	ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX];
1376 
1377 	key.objectid = tree_id;
1378 	key.type = BTRFS_ROOT_ITEM_KEY;
1379 	key.offset = (u64)-1;
1380 	root = btrfs_read_fs_root_no_name(info, &key);
1381 	if (IS_ERR(root)) {
1382 		printk(KERN_ERR "could not find root %llu\n", tree_id);
1383 		ret = -ENOENT;
1384 		goto out;
1385 	}
1386 
1387 	key.objectid = dirid;
1388 	key.type = BTRFS_INODE_REF_KEY;
1389 	key.offset = (u64)-1;
1390 
1391 	while(1) {
1392 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1393 		if (ret < 0)
1394 			goto out;
1395 
1396 		l = path->nodes[0];
1397 		slot = path->slots[0];
1398 		if (ret > 0 && slot > 0)
1399 			slot--;
1400 		btrfs_item_key_to_cpu(l, &key, slot);
1401 
1402 		if (ret > 0 && (key.objectid != dirid ||
1403 				key.type != BTRFS_INODE_REF_KEY)) {
1404 			ret = -ENOENT;
1405 			goto out;
1406 		}
1407 
1408 		iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
1409 		len = btrfs_inode_ref_name_len(l, iref);
1410 		ptr -= len + 1;
1411 		total_len += len + 1;
1412 		if (ptr < name)
1413 			goto out;
1414 
1415 		*(ptr + len) = '/';
1416 		read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len);
1417 
1418 		if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
1419 			break;
1420 
1421 		btrfs_release_path(root, path);
1422 		key.objectid = key.offset;
1423 		key.offset = (u64)-1;
1424 		dirid = key.objectid;
1425 
1426 	}
1427 	if (ptr < name)
1428 		goto out;
1429 	memcpy(name, ptr, total_len);
1430 	name[total_len]='\0';
1431 	ret = 0;
1432 out:
1433 	btrfs_free_path(path);
1434 	return ret;
1435 }
1436 
1437 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
1438 					   void __user *argp)
1439 {
1440 	 struct btrfs_ioctl_ino_lookup_args *args;
1441 	 struct inode *inode;
1442 	 int ret;
1443 
1444 	if (!capable(CAP_SYS_ADMIN))
1445 		return -EPERM;
1446 
1447 	args = memdup_user(argp, sizeof(*args));
1448 	if (IS_ERR(args))
1449 		return PTR_ERR(args);
1450 
1451 	inode = fdentry(file)->d_inode;
1452 
1453 	if (args->treeid == 0)
1454 		args->treeid = BTRFS_I(inode)->root->root_key.objectid;
1455 
1456 	ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
1457 					args->treeid, args->objectid,
1458 					args->name);
1459 
1460 	if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
1461 		ret = -EFAULT;
1462 
1463 	kfree(args);
1464 	return ret;
1465 }
1466 
1467 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
1468 					     void __user *arg)
1469 {
1470 	struct dentry *parent = fdentry(file);
1471 	struct dentry *dentry;
1472 	struct inode *dir = parent->d_inode;
1473 	struct inode *inode;
1474 	struct btrfs_root *root = BTRFS_I(dir)->root;
1475 	struct btrfs_root *dest = NULL;
1476 	struct btrfs_ioctl_vol_args *vol_args;
1477 	struct btrfs_trans_handle *trans;
1478 	int namelen;
1479 	int ret;
1480 	int err = 0;
1481 
1482 	vol_args = memdup_user(arg, sizeof(*vol_args));
1483 	if (IS_ERR(vol_args))
1484 		return PTR_ERR(vol_args);
1485 
1486 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1487 	namelen = strlen(vol_args->name);
1488 	if (strchr(vol_args->name, '/') ||
1489 	    strncmp(vol_args->name, "..", namelen) == 0) {
1490 		err = -EINVAL;
1491 		goto out;
1492 	}
1493 
1494 	err = mnt_want_write(file->f_path.mnt);
1495 	if (err)
1496 		goto out;
1497 
1498 	mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
1499 	dentry = lookup_one_len(vol_args->name, parent, namelen);
1500 	if (IS_ERR(dentry)) {
1501 		err = PTR_ERR(dentry);
1502 		goto out_unlock_dir;
1503 	}
1504 
1505 	if (!dentry->d_inode) {
1506 		err = -ENOENT;
1507 		goto out_dput;
1508 	}
1509 
1510 	inode = dentry->d_inode;
1511 	dest = BTRFS_I(inode)->root;
1512 	if (!capable(CAP_SYS_ADMIN)){
1513 		/*
1514 		 * Regular user.  Only allow this with a special mount
1515 		 * option, when the user has write+exec access to the
1516 		 * subvol root, and when rmdir(2) would have been
1517 		 * allowed.
1518 		 *
1519 		 * Note that this is _not_ check that the subvol is
1520 		 * empty or doesn't contain data that we wouldn't
1521 		 * otherwise be able to delete.
1522 		 *
1523 		 * Users who want to delete empty subvols should try
1524 		 * rmdir(2).
1525 		 */
1526 		err = -EPERM;
1527 		if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1528 			goto out_dput;
1529 
1530 		/*
1531 		 * Do not allow deletion if the parent dir is the same
1532 		 * as the dir to be deleted.  That means the ioctl
1533 		 * must be called on the dentry referencing the root
1534 		 * of the subvol, not a random directory contained
1535 		 * within it.
1536 		 */
1537 		err = -EINVAL;
1538 		if (root == dest)
1539 			goto out_dput;
1540 
1541 		err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
1542 		if (err)
1543 			goto out_dput;
1544 
1545 		/* check if subvolume may be deleted by a non-root user */
1546 		err = btrfs_may_delete(dir, dentry, 1);
1547 		if (err)
1548 			goto out_dput;
1549 	}
1550 
1551 	if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
1552 		err = -EINVAL;
1553 		goto out_dput;
1554 	}
1555 
1556 	mutex_lock(&inode->i_mutex);
1557 	err = d_invalidate(dentry);
1558 	if (err)
1559 		goto out_unlock;
1560 
1561 	down_write(&root->fs_info->subvol_sem);
1562 
1563 	err = may_destroy_subvol(dest);
1564 	if (err)
1565 		goto out_up_write;
1566 
1567 	trans = btrfs_start_transaction(root, 0);
1568 	if (IS_ERR(trans)) {
1569 		err = PTR_ERR(trans);
1570 		goto out_up_write;
1571 	}
1572 	trans->block_rsv = &root->fs_info->global_block_rsv;
1573 
1574 	ret = btrfs_unlink_subvol(trans, root, dir,
1575 				dest->root_key.objectid,
1576 				dentry->d_name.name,
1577 				dentry->d_name.len);
1578 	BUG_ON(ret);
1579 
1580 	btrfs_record_root_in_trans(trans, dest);
1581 
1582 	memset(&dest->root_item.drop_progress, 0,
1583 		sizeof(dest->root_item.drop_progress));
1584 	dest->root_item.drop_level = 0;
1585 	btrfs_set_root_refs(&dest->root_item, 0);
1586 
1587 	if (!xchg(&dest->orphan_item_inserted, 1)) {
1588 		ret = btrfs_insert_orphan_item(trans,
1589 					root->fs_info->tree_root,
1590 					dest->root_key.objectid);
1591 		BUG_ON(ret);
1592 	}
1593 
1594 	ret = btrfs_end_transaction(trans, root);
1595 	BUG_ON(ret);
1596 	inode->i_flags |= S_DEAD;
1597 out_up_write:
1598 	up_write(&root->fs_info->subvol_sem);
1599 out_unlock:
1600 	mutex_unlock(&inode->i_mutex);
1601 	if (!err) {
1602 		shrink_dcache_sb(root->fs_info->sb);
1603 		btrfs_invalidate_inodes(dest);
1604 		d_delete(dentry);
1605 	}
1606 out_dput:
1607 	dput(dentry);
1608 out_unlock_dir:
1609 	mutex_unlock(&dir->i_mutex);
1610 	mnt_drop_write(file->f_path.mnt);
1611 out:
1612 	kfree(vol_args);
1613 	return err;
1614 }
1615 
1616 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
1617 {
1618 	struct inode *inode = fdentry(file)->d_inode;
1619 	struct btrfs_root *root = BTRFS_I(inode)->root;
1620 	struct btrfs_ioctl_defrag_range_args *range;
1621 	int ret;
1622 
1623 	if (btrfs_root_readonly(root))
1624 		return -EROFS;
1625 
1626 	ret = mnt_want_write(file->f_path.mnt);
1627 	if (ret)
1628 		return ret;
1629 
1630 	switch (inode->i_mode & S_IFMT) {
1631 	case S_IFDIR:
1632 		if (!capable(CAP_SYS_ADMIN)) {
1633 			ret = -EPERM;
1634 			goto out;
1635 		}
1636 		ret = btrfs_defrag_root(root, 0);
1637 		if (ret)
1638 			goto out;
1639 		ret = btrfs_defrag_root(root->fs_info->extent_root, 0);
1640 		break;
1641 	case S_IFREG:
1642 		if (!(file->f_mode & FMODE_WRITE)) {
1643 			ret = -EINVAL;
1644 			goto out;
1645 		}
1646 
1647 		range = kzalloc(sizeof(*range), GFP_KERNEL);
1648 		if (!range) {
1649 			ret = -ENOMEM;
1650 			goto out;
1651 		}
1652 
1653 		if (argp) {
1654 			if (copy_from_user(range, argp,
1655 					   sizeof(*range))) {
1656 				ret = -EFAULT;
1657 				kfree(range);
1658 				goto out;
1659 			}
1660 			/* compression requires us to start the IO */
1661 			if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
1662 				range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
1663 				range->extent_thresh = (u32)-1;
1664 			}
1665 		} else {
1666 			/* the rest are all set to zero by kzalloc */
1667 			range->len = (u64)-1;
1668 		}
1669 		ret = btrfs_defrag_file(file, range);
1670 		kfree(range);
1671 		break;
1672 	default:
1673 		ret = -EINVAL;
1674 	}
1675 out:
1676 	mnt_drop_write(file->f_path.mnt);
1677 	return ret;
1678 }
1679 
1680 static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
1681 {
1682 	struct btrfs_ioctl_vol_args *vol_args;
1683 	int ret;
1684 
1685 	if (!capable(CAP_SYS_ADMIN))
1686 		return -EPERM;
1687 
1688 	vol_args = memdup_user(arg, sizeof(*vol_args));
1689 	if (IS_ERR(vol_args))
1690 		return PTR_ERR(vol_args);
1691 
1692 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1693 	ret = btrfs_init_new_device(root, vol_args->name);
1694 
1695 	kfree(vol_args);
1696 	return ret;
1697 }
1698 
1699 static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
1700 {
1701 	struct btrfs_ioctl_vol_args *vol_args;
1702 	int ret;
1703 
1704 	if (!capable(CAP_SYS_ADMIN))
1705 		return -EPERM;
1706 
1707 	if (root->fs_info->sb->s_flags & MS_RDONLY)
1708 		return -EROFS;
1709 
1710 	vol_args = memdup_user(arg, sizeof(*vol_args));
1711 	if (IS_ERR(vol_args))
1712 		return PTR_ERR(vol_args);
1713 
1714 	vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1715 	ret = btrfs_rm_device(root, vol_args->name);
1716 
1717 	kfree(vol_args);
1718 	return ret;
1719 }
1720 
1721 static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
1722 				       u64 off, u64 olen, u64 destoff)
1723 {
1724 	struct inode *inode = fdentry(file)->d_inode;
1725 	struct btrfs_root *root = BTRFS_I(inode)->root;
1726 	struct file *src_file;
1727 	struct inode *src;
1728 	struct btrfs_trans_handle *trans;
1729 	struct btrfs_path *path;
1730 	struct extent_buffer *leaf;
1731 	char *buf;
1732 	struct btrfs_key key;
1733 	u32 nritems;
1734 	int slot;
1735 	int ret;
1736 	u64 len = olen;
1737 	u64 bs = root->fs_info->sb->s_blocksize;
1738 	u64 hint_byte;
1739 
1740 	/*
1741 	 * TODO:
1742 	 * - split compressed inline extents.  annoying: we need to
1743 	 *   decompress into destination's address_space (the file offset
1744 	 *   may change, so source mapping won't do), then recompress (or
1745 	 *   otherwise reinsert) a subrange.
1746 	 * - allow ranges within the same file to be cloned (provided
1747 	 *   they don't overlap)?
1748 	 */
1749 
1750 	/* the destination must be opened for writing */
1751 	if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND))
1752 		return -EINVAL;
1753 
1754 	if (btrfs_root_readonly(root))
1755 		return -EROFS;
1756 
1757 	ret = mnt_want_write(file->f_path.mnt);
1758 	if (ret)
1759 		return ret;
1760 
1761 	src_file = fget(srcfd);
1762 	if (!src_file) {
1763 		ret = -EBADF;
1764 		goto out_drop_write;
1765 	}
1766 
1767 	src = src_file->f_dentry->d_inode;
1768 
1769 	ret = -EINVAL;
1770 	if (src == inode)
1771 		goto out_fput;
1772 
1773 	/* the src must be open for reading */
1774 	if (!(src_file->f_mode & FMODE_READ))
1775 		goto out_fput;
1776 
1777 	ret = -EISDIR;
1778 	if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode))
1779 		goto out_fput;
1780 
1781 	ret = -EXDEV;
1782 	if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root)
1783 		goto out_fput;
1784 
1785 	ret = -ENOMEM;
1786 	buf = vmalloc(btrfs_level_size(root, 0));
1787 	if (!buf)
1788 		goto out_fput;
1789 
1790 	path = btrfs_alloc_path();
1791 	if (!path) {
1792 		vfree(buf);
1793 		goto out_fput;
1794 	}
1795 	path->reada = 2;
1796 
1797 	if (inode < src) {
1798 		mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
1799 		mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD);
1800 	} else {
1801 		mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT);
1802 		mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1803 	}
1804 
1805 	/* determine range to clone */
1806 	ret = -EINVAL;
1807 	if (off + len > src->i_size || off + len < off)
1808 		goto out_unlock;
1809 	if (len == 0)
1810 		olen = len = src->i_size - off;
1811 	/* if we extend to eof, continue to block boundary */
1812 	if (off + len == src->i_size)
1813 		len = ALIGN(src->i_size, bs) - off;
1814 
1815 	/* verify the end result is block aligned */
1816 	if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) ||
1817 	    !IS_ALIGNED(destoff, bs))
1818 		goto out_unlock;
1819 
1820 	/* do any pending delalloc/csum calc on src, one way or
1821 	   another, and lock file content */
1822 	while (1) {
1823 		struct btrfs_ordered_extent *ordered;
1824 		lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1825 		ordered = btrfs_lookup_first_ordered_extent(src, off+len);
1826 		if (!ordered &&
1827 		    !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len,
1828 				   EXTENT_DELALLOC, 0, NULL))
1829 			break;
1830 		unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
1831 		if (ordered)
1832 			btrfs_put_ordered_extent(ordered);
1833 		btrfs_wait_ordered_range(src, off, len);
1834 	}
1835 
1836 	/* clone data */
1837 	key.objectid = src->i_ino;
1838 	key.type = BTRFS_EXTENT_DATA_KEY;
1839 	key.offset = 0;
1840 
1841 	while (1) {
1842 		/*
1843 		 * note the key will change type as we walk through the
1844 		 * tree.
1845 		 */
1846 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1847 		if (ret < 0)
1848 			goto out;
1849 
1850 		nritems = btrfs_header_nritems(path->nodes[0]);
1851 		if (path->slots[0] >= nritems) {
1852 			ret = btrfs_next_leaf(root, path);
1853 			if (ret < 0)
1854 				goto out;
1855 			if (ret > 0)
1856 				break;
1857 			nritems = btrfs_header_nritems(path->nodes[0]);
1858 		}
1859 		leaf = path->nodes[0];
1860 		slot = path->slots[0];
1861 
1862 		btrfs_item_key_to_cpu(leaf, &key, slot);
1863 		if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
1864 		    key.objectid != src->i_ino)
1865 			break;
1866 
1867 		if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
1868 			struct btrfs_file_extent_item *extent;
1869 			int type;
1870 			u32 size;
1871 			struct btrfs_key new_key;
1872 			u64 disko = 0, diskl = 0;
1873 			u64 datao = 0, datal = 0;
1874 			u8 comp;
1875 			u64 endoff;
1876 
1877 			size = btrfs_item_size_nr(leaf, slot);
1878 			read_extent_buffer(leaf, buf,
1879 					   btrfs_item_ptr_offset(leaf, slot),
1880 					   size);
1881 
1882 			extent = btrfs_item_ptr(leaf, slot,
1883 						struct btrfs_file_extent_item);
1884 			comp = btrfs_file_extent_compression(leaf, extent);
1885 			type = btrfs_file_extent_type(leaf, extent);
1886 			if (type == BTRFS_FILE_EXTENT_REG ||
1887 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
1888 				disko = btrfs_file_extent_disk_bytenr(leaf,
1889 								      extent);
1890 				diskl = btrfs_file_extent_disk_num_bytes(leaf,
1891 								 extent);
1892 				datao = btrfs_file_extent_offset(leaf, extent);
1893 				datal = btrfs_file_extent_num_bytes(leaf,
1894 								    extent);
1895 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
1896 				/* take upper bound, may be compressed */
1897 				datal = btrfs_file_extent_ram_bytes(leaf,
1898 								    extent);
1899 			}
1900 			btrfs_release_path(root, path);
1901 
1902 			if (key.offset + datal <= off ||
1903 			    key.offset >= off+len)
1904 				goto next;
1905 
1906 			memcpy(&new_key, &key, sizeof(new_key));
1907 			new_key.objectid = inode->i_ino;
1908 			if (off <= key.offset)
1909 				new_key.offset = key.offset + destoff - off;
1910 			else
1911 				new_key.offset = destoff;
1912 
1913 			trans = btrfs_start_transaction(root, 1);
1914 			if (IS_ERR(trans)) {
1915 				ret = PTR_ERR(trans);
1916 				goto out;
1917 			}
1918 
1919 			if (type == BTRFS_FILE_EXTENT_REG ||
1920 			    type == BTRFS_FILE_EXTENT_PREALLOC) {
1921 				if (off > key.offset) {
1922 					datao += off - key.offset;
1923 					datal -= off - key.offset;
1924 				}
1925 
1926 				if (key.offset + datal > off + len)
1927 					datal = off + len - key.offset;
1928 
1929 				ret = btrfs_drop_extents(trans, inode,
1930 							 new_key.offset,
1931 							 new_key.offset + datal,
1932 							 &hint_byte, 1);
1933 				BUG_ON(ret);
1934 
1935 				ret = btrfs_insert_empty_item(trans, root, path,
1936 							      &new_key, size);
1937 				BUG_ON(ret);
1938 
1939 				leaf = path->nodes[0];
1940 				slot = path->slots[0];
1941 				write_extent_buffer(leaf, buf,
1942 					    btrfs_item_ptr_offset(leaf, slot),
1943 					    size);
1944 
1945 				extent = btrfs_item_ptr(leaf, slot,
1946 						struct btrfs_file_extent_item);
1947 
1948 				/* disko == 0 means it's a hole */
1949 				if (!disko)
1950 					datao = 0;
1951 
1952 				btrfs_set_file_extent_offset(leaf, extent,
1953 							     datao);
1954 				btrfs_set_file_extent_num_bytes(leaf, extent,
1955 								datal);
1956 				if (disko) {
1957 					inode_add_bytes(inode, datal);
1958 					ret = btrfs_inc_extent_ref(trans, root,
1959 							disko, diskl, 0,
1960 							root->root_key.objectid,
1961 							inode->i_ino,
1962 							new_key.offset - datao);
1963 					BUG_ON(ret);
1964 				}
1965 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
1966 				u64 skip = 0;
1967 				u64 trim = 0;
1968 				if (off > key.offset) {
1969 					skip = off - key.offset;
1970 					new_key.offset += skip;
1971 				}
1972 
1973 				if (key.offset + datal > off+len)
1974 					trim = key.offset + datal - (off+len);
1975 
1976 				if (comp && (skip || trim)) {
1977 					ret = -EINVAL;
1978 					btrfs_end_transaction(trans, root);
1979 					goto out;
1980 				}
1981 				size -= skip + trim;
1982 				datal -= skip + trim;
1983 
1984 				ret = btrfs_drop_extents(trans, inode,
1985 							 new_key.offset,
1986 							 new_key.offset + datal,
1987 							 &hint_byte, 1);
1988 				BUG_ON(ret);
1989 
1990 				ret = btrfs_insert_empty_item(trans, root, path,
1991 							      &new_key, size);
1992 				BUG_ON(ret);
1993 
1994 				if (skip) {
1995 					u32 start =
1996 					  btrfs_file_extent_calc_inline_size(0);
1997 					memmove(buf+start, buf+start+skip,
1998 						datal);
1999 				}
2000 
2001 				leaf = path->nodes[0];
2002 				slot = path->slots[0];
2003 				write_extent_buffer(leaf, buf,
2004 					    btrfs_item_ptr_offset(leaf, slot),
2005 					    size);
2006 				inode_add_bytes(inode, datal);
2007 			}
2008 
2009 			btrfs_mark_buffer_dirty(leaf);
2010 			btrfs_release_path(root, path);
2011 
2012 			inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2013 
2014 			/*
2015 			 * we round up to the block size at eof when
2016 			 * determining which extents to clone above,
2017 			 * but shouldn't round up the file size
2018 			 */
2019 			endoff = new_key.offset + datal;
2020 			if (endoff > destoff+olen)
2021 				endoff = destoff+olen;
2022 			if (endoff > inode->i_size)
2023 				btrfs_i_size_write(inode, endoff);
2024 
2025 			BTRFS_I(inode)->flags = BTRFS_I(src)->flags;
2026 			ret = btrfs_update_inode(trans, root, inode);
2027 			BUG_ON(ret);
2028 			btrfs_end_transaction(trans, root);
2029 		}
2030 next:
2031 		btrfs_release_path(root, path);
2032 		key.offset++;
2033 	}
2034 	ret = 0;
2035 out:
2036 	btrfs_release_path(root, path);
2037 	unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS);
2038 out_unlock:
2039 	mutex_unlock(&src->i_mutex);
2040 	mutex_unlock(&inode->i_mutex);
2041 	vfree(buf);
2042 	btrfs_free_path(path);
2043 out_fput:
2044 	fput(src_file);
2045 out_drop_write:
2046 	mnt_drop_write(file->f_path.mnt);
2047 	return ret;
2048 }
2049 
2050 static long btrfs_ioctl_clone_range(struct file *file, void __user *argp)
2051 {
2052 	struct btrfs_ioctl_clone_range_args args;
2053 
2054 	if (copy_from_user(&args, argp, sizeof(args)))
2055 		return -EFAULT;
2056 	return btrfs_ioctl_clone(file, args.src_fd, args.src_offset,
2057 				 args.src_length, args.dest_offset);
2058 }
2059 
2060 /*
2061  * there are many ways the trans_start and trans_end ioctls can lead
2062  * to deadlocks.  They should only be used by applications that
2063  * basically own the machine, and have a very in depth understanding
2064  * of all the possible deadlocks and enospc problems.
2065  */
2066 static long btrfs_ioctl_trans_start(struct file *file)
2067 {
2068 	struct inode *inode = fdentry(file)->d_inode;
2069 	struct btrfs_root *root = BTRFS_I(inode)->root;
2070 	struct btrfs_trans_handle *trans;
2071 	int ret;
2072 
2073 	ret = -EPERM;
2074 	if (!capable(CAP_SYS_ADMIN))
2075 		goto out;
2076 
2077 	ret = -EINPROGRESS;
2078 	if (file->private_data)
2079 		goto out;
2080 
2081 	ret = -EROFS;
2082 	if (btrfs_root_readonly(root))
2083 		goto out;
2084 
2085 	ret = mnt_want_write(file->f_path.mnt);
2086 	if (ret)
2087 		goto out;
2088 
2089 	mutex_lock(&root->fs_info->trans_mutex);
2090 	root->fs_info->open_ioctl_trans++;
2091 	mutex_unlock(&root->fs_info->trans_mutex);
2092 
2093 	ret = -ENOMEM;
2094 	trans = btrfs_start_ioctl_transaction(root, 0);
2095 	if (IS_ERR(trans))
2096 		goto out_drop;
2097 
2098 	file->private_data = trans;
2099 	return 0;
2100 
2101 out_drop:
2102 	mutex_lock(&root->fs_info->trans_mutex);
2103 	root->fs_info->open_ioctl_trans--;
2104 	mutex_unlock(&root->fs_info->trans_mutex);
2105 	mnt_drop_write(file->f_path.mnt);
2106 out:
2107 	return ret;
2108 }
2109 
2110 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
2111 {
2112 	struct inode *inode = fdentry(file)->d_inode;
2113 	struct btrfs_root *root = BTRFS_I(inode)->root;
2114 	struct btrfs_root *new_root;
2115 	struct btrfs_dir_item *di;
2116 	struct btrfs_trans_handle *trans;
2117 	struct btrfs_path *path;
2118 	struct btrfs_key location;
2119 	struct btrfs_disk_key disk_key;
2120 	struct btrfs_super_block *disk_super;
2121 	u64 features;
2122 	u64 objectid = 0;
2123 	u64 dir_id;
2124 
2125 	if (!capable(CAP_SYS_ADMIN))
2126 		return -EPERM;
2127 
2128 	if (copy_from_user(&objectid, argp, sizeof(objectid)))
2129 		return -EFAULT;
2130 
2131 	if (!objectid)
2132 		objectid = root->root_key.objectid;
2133 
2134 	location.objectid = objectid;
2135 	location.type = BTRFS_ROOT_ITEM_KEY;
2136 	location.offset = (u64)-1;
2137 
2138 	new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
2139 	if (IS_ERR(new_root))
2140 		return PTR_ERR(new_root);
2141 
2142 	if (btrfs_root_refs(&new_root->root_item) == 0)
2143 		return -ENOENT;
2144 
2145 	path = btrfs_alloc_path();
2146 	if (!path)
2147 		return -ENOMEM;
2148 	path->leave_spinning = 1;
2149 
2150 	trans = btrfs_start_transaction(root, 1);
2151 	if (IS_ERR(trans)) {
2152 		btrfs_free_path(path);
2153 		return PTR_ERR(trans);
2154 	}
2155 
2156 	dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
2157 	di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path,
2158 				   dir_id, "default", 7, 1);
2159 	if (IS_ERR_OR_NULL(di)) {
2160 		btrfs_free_path(path);
2161 		btrfs_end_transaction(trans, root);
2162 		printk(KERN_ERR "Umm, you don't have the default dir item, "
2163 		       "this isn't going to work\n");
2164 		return -ENOENT;
2165 	}
2166 
2167 	btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
2168 	btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
2169 	btrfs_mark_buffer_dirty(path->nodes[0]);
2170 	btrfs_free_path(path);
2171 
2172 	disk_super = &root->fs_info->super_copy;
2173 	features = btrfs_super_incompat_flags(disk_super);
2174 	if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) {
2175 		features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL;
2176 		btrfs_set_super_incompat_flags(disk_super, features);
2177 	}
2178 	btrfs_end_transaction(trans, root);
2179 
2180 	return 0;
2181 }
2182 
2183 static void get_block_group_info(struct list_head *groups_list,
2184 				 struct btrfs_ioctl_space_info *space)
2185 {
2186 	struct btrfs_block_group_cache *block_group;
2187 
2188 	space->total_bytes = 0;
2189 	space->used_bytes = 0;
2190 	space->flags = 0;
2191 	list_for_each_entry(block_group, groups_list, list) {
2192 		space->flags = block_group->flags;
2193 		space->total_bytes += block_group->key.offset;
2194 		space->used_bytes +=
2195 			btrfs_block_group_used(&block_group->item);
2196 	}
2197 }
2198 
2199 long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
2200 {
2201 	struct btrfs_ioctl_space_args space_args;
2202 	struct btrfs_ioctl_space_info space;
2203 	struct btrfs_ioctl_space_info *dest;
2204 	struct btrfs_ioctl_space_info *dest_orig;
2205 	struct btrfs_ioctl_space_info *user_dest;
2206 	struct btrfs_space_info *info;
2207 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
2208 		       BTRFS_BLOCK_GROUP_SYSTEM,
2209 		       BTRFS_BLOCK_GROUP_METADATA,
2210 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
2211 	int num_types = 4;
2212 	int alloc_size;
2213 	int ret = 0;
2214 	u64 slot_count = 0;
2215 	int i, c;
2216 
2217 	if (copy_from_user(&space_args,
2218 			   (struct btrfs_ioctl_space_args __user *)arg,
2219 			   sizeof(space_args)))
2220 		return -EFAULT;
2221 
2222 	for (i = 0; i < num_types; i++) {
2223 		struct btrfs_space_info *tmp;
2224 
2225 		info = NULL;
2226 		rcu_read_lock();
2227 		list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2228 					list) {
2229 			if (tmp->flags == types[i]) {
2230 				info = tmp;
2231 				break;
2232 			}
2233 		}
2234 		rcu_read_unlock();
2235 
2236 		if (!info)
2237 			continue;
2238 
2239 		down_read(&info->groups_sem);
2240 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2241 			if (!list_empty(&info->block_groups[c]))
2242 				slot_count++;
2243 		}
2244 		up_read(&info->groups_sem);
2245 	}
2246 
2247 	/* space_slots == 0 means they are asking for a count */
2248 	if (space_args.space_slots == 0) {
2249 		space_args.total_spaces = slot_count;
2250 		goto out;
2251 	}
2252 
2253 	slot_count = min_t(u64, space_args.space_slots, slot_count);
2254 
2255 	alloc_size = sizeof(*dest) * slot_count;
2256 
2257 	/* we generally have at most 6 or so space infos, one for each raid
2258 	 * level.  So, a whole page should be more than enough for everyone
2259 	 */
2260 	if (alloc_size > PAGE_CACHE_SIZE)
2261 		return -ENOMEM;
2262 
2263 	space_args.total_spaces = 0;
2264 	dest = kmalloc(alloc_size, GFP_NOFS);
2265 	if (!dest)
2266 		return -ENOMEM;
2267 	dest_orig = dest;
2268 
2269 	/* now we have a buffer to copy into */
2270 	for (i = 0; i < num_types; i++) {
2271 		struct btrfs_space_info *tmp;
2272 
2273 		if (!slot_count)
2274 			break;
2275 
2276 		info = NULL;
2277 		rcu_read_lock();
2278 		list_for_each_entry_rcu(tmp, &root->fs_info->space_info,
2279 					list) {
2280 			if (tmp->flags == types[i]) {
2281 				info = tmp;
2282 				break;
2283 			}
2284 		}
2285 		rcu_read_unlock();
2286 
2287 		if (!info)
2288 			continue;
2289 		down_read(&info->groups_sem);
2290 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
2291 			if (!list_empty(&info->block_groups[c])) {
2292 				get_block_group_info(&info->block_groups[c],
2293 						     &space);
2294 				memcpy(dest, &space, sizeof(space));
2295 				dest++;
2296 				space_args.total_spaces++;
2297 				slot_count--;
2298 			}
2299 			if (!slot_count)
2300 				break;
2301 		}
2302 		up_read(&info->groups_sem);
2303 	}
2304 
2305 	user_dest = (struct btrfs_ioctl_space_info *)
2306 		(arg + sizeof(struct btrfs_ioctl_space_args));
2307 
2308 	if (copy_to_user(user_dest, dest_orig, alloc_size))
2309 		ret = -EFAULT;
2310 
2311 	kfree(dest_orig);
2312 out:
2313 	if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
2314 		ret = -EFAULT;
2315 
2316 	return ret;
2317 }
2318 
2319 /*
2320  * there are many ways the trans_start and trans_end ioctls can lead
2321  * to deadlocks.  They should only be used by applications that
2322  * basically own the machine, and have a very in depth understanding
2323  * of all the possible deadlocks and enospc problems.
2324  */
2325 long btrfs_ioctl_trans_end(struct file *file)
2326 {
2327 	struct inode *inode = fdentry(file)->d_inode;
2328 	struct btrfs_root *root = BTRFS_I(inode)->root;
2329 	struct btrfs_trans_handle *trans;
2330 
2331 	trans = file->private_data;
2332 	if (!trans)
2333 		return -EINVAL;
2334 	file->private_data = NULL;
2335 
2336 	btrfs_end_transaction(trans, root);
2337 
2338 	mutex_lock(&root->fs_info->trans_mutex);
2339 	root->fs_info->open_ioctl_trans--;
2340 	mutex_unlock(&root->fs_info->trans_mutex);
2341 
2342 	mnt_drop_write(file->f_path.mnt);
2343 	return 0;
2344 }
2345 
2346 static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp)
2347 {
2348 	struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2349 	struct btrfs_trans_handle *trans;
2350 	u64 transid;
2351 
2352 	trans = btrfs_start_transaction(root, 0);
2353 	if (IS_ERR(trans))
2354 		return PTR_ERR(trans);
2355 	transid = trans->transid;
2356 	btrfs_commit_transaction_async(trans, root, 0);
2357 
2358 	if (argp)
2359 		if (copy_to_user(argp, &transid, sizeof(transid)))
2360 			return -EFAULT;
2361 	return 0;
2362 }
2363 
2364 static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp)
2365 {
2366 	struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root;
2367 	u64 transid;
2368 
2369 	if (argp) {
2370 		if (copy_from_user(&transid, argp, sizeof(transid)))
2371 			return -EFAULT;
2372 	} else {
2373 		transid = 0;  /* current trans */
2374 	}
2375 	return btrfs_wait_for_commit(root, transid);
2376 }
2377 
2378 long btrfs_ioctl(struct file *file, unsigned int
2379 		cmd, unsigned long arg)
2380 {
2381 	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2382 	void __user *argp = (void __user *)arg;
2383 
2384 	switch (cmd) {
2385 	case FS_IOC_GETFLAGS:
2386 		return btrfs_ioctl_getflags(file, argp);
2387 	case FS_IOC_SETFLAGS:
2388 		return btrfs_ioctl_setflags(file, argp);
2389 	case FS_IOC_GETVERSION:
2390 		return btrfs_ioctl_getversion(file, argp);
2391 	case BTRFS_IOC_SNAP_CREATE:
2392 		return btrfs_ioctl_snap_create(file, argp, 0);
2393 	case BTRFS_IOC_SNAP_CREATE_V2:
2394 		return btrfs_ioctl_snap_create_v2(file, argp, 0);
2395 	case BTRFS_IOC_SUBVOL_CREATE:
2396 		return btrfs_ioctl_snap_create(file, argp, 1);
2397 	case BTRFS_IOC_SNAP_DESTROY:
2398 		return btrfs_ioctl_snap_destroy(file, argp);
2399 	case BTRFS_IOC_SUBVOL_GETFLAGS:
2400 		return btrfs_ioctl_subvol_getflags(file, argp);
2401 	case BTRFS_IOC_SUBVOL_SETFLAGS:
2402 		return btrfs_ioctl_subvol_setflags(file, argp);
2403 	case BTRFS_IOC_DEFAULT_SUBVOL:
2404 		return btrfs_ioctl_default_subvol(file, argp);
2405 	case BTRFS_IOC_DEFRAG:
2406 		return btrfs_ioctl_defrag(file, NULL);
2407 	case BTRFS_IOC_DEFRAG_RANGE:
2408 		return btrfs_ioctl_defrag(file, argp);
2409 	case BTRFS_IOC_RESIZE:
2410 		return btrfs_ioctl_resize(root, argp);
2411 	case BTRFS_IOC_ADD_DEV:
2412 		return btrfs_ioctl_add_dev(root, argp);
2413 	case BTRFS_IOC_RM_DEV:
2414 		return btrfs_ioctl_rm_dev(root, argp);
2415 	case BTRFS_IOC_BALANCE:
2416 		return btrfs_balance(root->fs_info->dev_root);
2417 	case BTRFS_IOC_CLONE:
2418 		return btrfs_ioctl_clone(file, arg, 0, 0, 0);
2419 	case BTRFS_IOC_CLONE_RANGE:
2420 		return btrfs_ioctl_clone_range(file, argp);
2421 	case BTRFS_IOC_TRANS_START:
2422 		return btrfs_ioctl_trans_start(file);
2423 	case BTRFS_IOC_TRANS_END:
2424 		return btrfs_ioctl_trans_end(file);
2425 	case BTRFS_IOC_TREE_SEARCH:
2426 		return btrfs_ioctl_tree_search(file, argp);
2427 	case BTRFS_IOC_INO_LOOKUP:
2428 		return btrfs_ioctl_ino_lookup(file, argp);
2429 	case BTRFS_IOC_SPACE_INFO:
2430 		return btrfs_ioctl_space_info(root, argp);
2431 	case BTRFS_IOC_SYNC:
2432 		btrfs_sync_fs(file->f_dentry->d_sb, 1);
2433 		return 0;
2434 	case BTRFS_IOC_START_SYNC:
2435 		return btrfs_ioctl_start_sync(file, argp);
2436 	case BTRFS_IOC_WAIT_SYNC:
2437 		return btrfs_ioctl_wait_sync(file, argp);
2438 	}
2439 
2440 	return -ENOTTY;
2441 }
2442